1fa9e4066Sahrens /* 2fa9e4066Sahrens * CDDL HEADER START 3fa9e4066Sahrens * 4fa9e4066Sahrens * The contents of this file are subject to the terms of the 5ea8dc4b6Seschrock * Common Development and Distribution License (the "License"). 6ea8dc4b6Seschrock * You may not use this file except in compliance with the License. 7fa9e4066Sahrens * 8fa9e4066Sahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9fa9e4066Sahrens * or http://www.opensolaris.org/os/licensing. 10fa9e4066Sahrens * See the License for the specific language governing permissions 11fa9e4066Sahrens * and limitations under the License. 12fa9e4066Sahrens * 13fa9e4066Sahrens * When distributing Covered Code, include this CDDL HEADER in each 14fa9e4066Sahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15fa9e4066Sahrens * If applicable, add the following below this CDDL HEADER, with the 16fa9e4066Sahrens * fields enclosed by brackets "[]" replaced with your own identifying 17fa9e4066Sahrens * information: Portions Copyright [yyyy] [name of copyright owner] 18fa9e4066Sahrens * 19fa9e4066Sahrens * CDDL HEADER END 20fa9e4066Sahrens */ 21fa9e4066Sahrens /* 229842588bSGeorge Wilson * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23be6fd75aSMatthew Ahrens * Copyright (c) 2013 by Delphix. All rights reserved. 24e9103aaeSGarrett D'Amore * Copyright 2011 Nexenta Systems, Inc. All rights reserved. 25fa9e4066Sahrens */ 26fa9e4066Sahrens 27fa9e4066Sahrens #include <sys/zfs_context.h> 28fa9e4066Sahrens #include <sys/spa_impl.h> 29283b8460SGeorge.Wilson #include <sys/spa_boot.h> 30fa9e4066Sahrens #include <sys/zio.h> 31fa9e4066Sahrens #include <sys/zio_checksum.h> 32fa9e4066Sahrens #include <sys/zio_compress.h> 33fa9e4066Sahrens #include <sys/dmu.h> 34fa9e4066Sahrens #include <sys/dmu_tx.h> 35fa9e4066Sahrens #include <sys/zap.h> 36fa9e4066Sahrens #include <sys/zil.h> 37fa9e4066Sahrens #include <sys/vdev_impl.h> 38fa9e4066Sahrens #include <sys/metaslab.h> 39fa9e4066Sahrens #include <sys/uberblock_impl.h> 40fa9e4066Sahrens #include <sys/txg.h> 41fa9e4066Sahrens #include <sys/avl.h> 42fa9e4066Sahrens #include <sys/unique.h> 43fa9e4066Sahrens #include <sys/dsl_pool.h> 44fa9e4066Sahrens #include <sys/dsl_dir.h> 45fa9e4066Sahrens #include <sys/dsl_prop.h> 463f9d6ad7SLin Ling #include <sys/dsl_scan.h> 47fa9e4066Sahrens #include <sys/fs/zfs.h> 486ce0521aSperrin #include <sys/metaslab_impl.h> 49e14bb325SJeff Bonwick #include <sys/arc.h> 50485bbbf5SGeorge Wilson #include <sys/ddt.h> 5191ebeef5Sahrens #include "zfs_prop.h" 52ad135b5dSChristopher Siden #include "zfeature_common.h" 53fa9e4066Sahrens 54fa9e4066Sahrens /* 55fa9e4066Sahrens * SPA locking 56fa9e4066Sahrens * 57fa9e4066Sahrens * There are four basic locks for managing spa_t structures: 58fa9e4066Sahrens * 59fa9e4066Sahrens * spa_namespace_lock (global mutex) 60fa9e4066Sahrens * 6144cd46caSbillm * This lock must be acquired to do any of the following: 62fa9e4066Sahrens * 6344cd46caSbillm * - Lookup a spa_t by name 6444cd46caSbillm * - Add or remove a spa_t from the namespace 6544cd46caSbillm * - Increase spa_refcount from non-zero 6644cd46caSbillm * - Check if spa_refcount is zero 6744cd46caSbillm * - Rename a spa_t 68ea8dc4b6Seschrock * - add/remove/attach/detach devices 6944cd46caSbillm * - Held for the duration of create/destroy/import/export 70fa9e4066Sahrens * 7144cd46caSbillm * It does not need to handle recursion. A create or destroy may 7244cd46caSbillm * reference objects (files or zvols) in other pools, but by 7344cd46caSbillm * definition they must have an existing reference, and will never need 7444cd46caSbillm * to lookup a spa_t by name. 75fa9e4066Sahrens * 76fa9e4066Sahrens * spa_refcount (per-spa refcount_t protected by mutex) 77fa9e4066Sahrens * 7844cd46caSbillm * This reference count keep track of any active users of the spa_t. The 7944cd46caSbillm * spa_t cannot be destroyed or freed while this is non-zero. Internally, 8044cd46caSbillm * the refcount is never really 'zero' - opening a pool implicitly keeps 81088f3894Sahrens * some references in the DMU. Internally we check against spa_minref, but 8244cd46caSbillm * present the image of a zero/non-zero value to consumers. 83fa9e4066Sahrens * 84e14bb325SJeff Bonwick * spa_config_lock[] (per-spa array of rwlocks) 85fa9e4066Sahrens * 8691ebeef5Sahrens * This protects the spa_t from config changes, and must be held in 8791ebeef5Sahrens * the following circumstances: 88fa9e4066Sahrens * 8944cd46caSbillm * - RW_READER to perform I/O to the spa 9044cd46caSbillm * - RW_WRITER to change the vdev config 91fa9e4066Sahrens * 92fa9e4066Sahrens * The locking order is fairly straightforward: 93fa9e4066Sahrens * 9444cd46caSbillm * spa_namespace_lock -> spa_refcount 95fa9e4066Sahrens * 9644cd46caSbillm * The namespace lock must be acquired to increase the refcount from 0 9744cd46caSbillm * or to check if it is zero. 98fa9e4066Sahrens * 99e14bb325SJeff Bonwick * spa_refcount -> spa_config_lock[] 100fa9e4066Sahrens * 10144cd46caSbillm * There must be at least one valid reference on the spa_t to acquire 10244cd46caSbillm * the config lock. 103fa9e4066Sahrens * 104e14bb325SJeff Bonwick * spa_namespace_lock -> spa_config_lock[] 105fa9e4066Sahrens * 10644cd46caSbillm * The namespace lock must always be taken before the config lock. 107fa9e4066Sahrens * 108fa9e4066Sahrens * 109e14bb325SJeff Bonwick * The spa_namespace_lock can be acquired directly and is globally visible. 110fa9e4066Sahrens * 111e14bb325SJeff Bonwick * The namespace is manipulated using the following functions, all of which 112e14bb325SJeff Bonwick * require the spa_namespace_lock to be held. 113fa9e4066Sahrens * 11444cd46caSbillm * spa_lookup() Lookup a spa_t by name. 115fa9e4066Sahrens * 11644cd46caSbillm * spa_add() Create a new spa_t in the namespace. 117fa9e4066Sahrens * 11844cd46caSbillm * spa_remove() Remove a spa_t from the namespace. This also 11944cd46caSbillm * frees up any memory associated with the spa_t. 120fa9e4066Sahrens * 12144cd46caSbillm * spa_next() Returns the next spa_t in the system, or the 12244cd46caSbillm * first if NULL is passed. 123fa9e4066Sahrens * 12444cd46caSbillm * spa_evict_all() Shutdown and remove all spa_t structures in 12544cd46caSbillm * the system. 126fa9e4066Sahrens * 127ea8dc4b6Seschrock * spa_guid_exists() Determine whether a pool/device guid exists. 128fa9e4066Sahrens * 129fa9e4066Sahrens * The spa_refcount is manipulated using the following functions: 130fa9e4066Sahrens * 13144cd46caSbillm * spa_open_ref() Adds a reference to the given spa_t. Must be 13244cd46caSbillm * called with spa_namespace_lock held if the 13344cd46caSbillm * refcount is currently zero. 134fa9e4066Sahrens * 13544cd46caSbillm * spa_close() Remove a reference from the spa_t. This will 13644cd46caSbillm * not free the spa_t or remove it from the 13744cd46caSbillm * namespace. No locking is required. 138fa9e4066Sahrens * 13944cd46caSbillm * spa_refcount_zero() Returns true if the refcount is currently 14044cd46caSbillm * zero. Must be called with spa_namespace_lock 14144cd46caSbillm * held. 142fa9e4066Sahrens * 143e14bb325SJeff Bonwick * The spa_config_lock[] is an array of rwlocks, ordered as follows: 144e14bb325SJeff Bonwick * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV. 145e14bb325SJeff Bonwick * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}(). 146e14bb325SJeff Bonwick * 147e14bb325SJeff Bonwick * To read the configuration, it suffices to hold one of these locks as reader. 148e14bb325SJeff Bonwick * To modify the configuration, you must hold all locks as writer. To modify 149e14bb325SJeff Bonwick * vdev state without altering the vdev tree's topology (e.g. online/offline), 150e14bb325SJeff Bonwick * you must hold SCL_STATE and SCL_ZIO as writer. 151e14bb325SJeff Bonwick * 152e14bb325SJeff Bonwick * We use these distinct config locks to avoid recursive lock entry. 153e14bb325SJeff Bonwick * For example, spa_sync() (which holds SCL_CONFIG as reader) induces 154e14bb325SJeff Bonwick * block allocations (SCL_ALLOC), which may require reading space maps 155e14bb325SJeff Bonwick * from disk (dmu_read() -> zio_read() -> SCL_ZIO). 156e14bb325SJeff Bonwick * 157e14bb325SJeff Bonwick * The spa config locks cannot be normal rwlocks because we need the 158e14bb325SJeff Bonwick * ability to hand off ownership. For example, SCL_ZIO is acquired 159e14bb325SJeff Bonwick * by the issuing thread and later released by an interrupt thread. 160e14bb325SJeff Bonwick * They do, however, obey the usual write-wanted semantics to prevent 161e14bb325SJeff Bonwick * writer (i.e. system administrator) starvation. 162e14bb325SJeff Bonwick * 163e14bb325SJeff Bonwick * The lock acquisition rules are as follows: 164e14bb325SJeff Bonwick * 165e14bb325SJeff Bonwick * SCL_CONFIG 166e14bb325SJeff Bonwick * Protects changes to the vdev tree topology, such as vdev 167e14bb325SJeff Bonwick * add/remove/attach/detach. Protects the dirty config list 168e14bb325SJeff Bonwick * (spa_config_dirty_list) and the set of spares and l2arc devices. 169e14bb325SJeff Bonwick * 170e14bb325SJeff Bonwick * SCL_STATE 171e14bb325SJeff Bonwick * Protects changes to pool state and vdev state, such as vdev 172e14bb325SJeff Bonwick * online/offline/fault/degrade/clear. Protects the dirty state list 173e14bb325SJeff Bonwick * (spa_state_dirty_list) and global pool state (spa_state). 174e14bb325SJeff Bonwick * 175e14bb325SJeff Bonwick * SCL_ALLOC 176e14bb325SJeff Bonwick * Protects changes to metaslab groups and classes. 177e14bb325SJeff Bonwick * Held as reader by metaslab_alloc() and metaslab_claim(). 178e14bb325SJeff Bonwick * 179e14bb325SJeff Bonwick * SCL_ZIO 180e14bb325SJeff Bonwick * Held by bp-level zios (those which have no io_vd upon entry) 181e14bb325SJeff Bonwick * to prevent changes to the vdev tree. The bp-level zio implicitly 182e14bb325SJeff Bonwick * protects all of its vdev child zios, which do not hold SCL_ZIO. 183e14bb325SJeff Bonwick * 184e14bb325SJeff Bonwick * SCL_FREE 185e14bb325SJeff Bonwick * Protects changes to metaslab groups and classes. 186e14bb325SJeff Bonwick * Held as reader by metaslab_free(). SCL_FREE is distinct from 187e14bb325SJeff Bonwick * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free 188e14bb325SJeff Bonwick * blocks in zio_done() while another i/o that holds either 189e14bb325SJeff Bonwick * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete. 190e14bb325SJeff Bonwick * 191e14bb325SJeff Bonwick * SCL_VDEV 192e14bb325SJeff Bonwick * Held as reader to prevent changes to the vdev tree during trivial 193b24ab676SJeff Bonwick * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the 194e14bb325SJeff Bonwick * other locks, and lower than all of them, to ensure that it's safe 195e14bb325SJeff Bonwick * to acquire regardless of caller context. 196e14bb325SJeff Bonwick * 197e14bb325SJeff Bonwick * In addition, the following rules apply: 198e14bb325SJeff Bonwick * 199e14bb325SJeff Bonwick * (a) spa_props_lock protects pool properties, spa_config and spa_config_list. 200e14bb325SJeff Bonwick * The lock ordering is SCL_CONFIG > spa_props_lock. 201e14bb325SJeff Bonwick * 202e14bb325SJeff Bonwick * (b) I/O operations on leaf vdevs. For any zio operation that takes 203e14bb325SJeff Bonwick * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(), 204e14bb325SJeff Bonwick * or zio_write_phys() -- the caller must ensure that the config cannot 205e14bb325SJeff Bonwick * cannot change in the interim, and that the vdev cannot be reopened. 206e14bb325SJeff Bonwick * SCL_STATE as reader suffices for both. 207fa9e4066Sahrens * 208ea8dc4b6Seschrock * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit(). 209fa9e4066Sahrens * 21044cd46caSbillm * spa_vdev_enter() Acquire the namespace lock and the config lock 211ea8dc4b6Seschrock * for writing. 212fa9e4066Sahrens * 21344cd46caSbillm * spa_vdev_exit() Release the config lock, wait for all I/O 21444cd46caSbillm * to complete, sync the updated configs to the 215ea8dc4b6Seschrock * cache, and release the namespace lock. 216fa9e4066Sahrens * 217e14bb325SJeff Bonwick * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit(). 218e14bb325SJeff Bonwick * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual 219e14bb325SJeff Bonwick * locking is, always, based on spa_namespace_lock and spa_config_lock[]. 220e14bb325SJeff Bonwick * 221ad135b5dSChristopher Siden * spa_rename() is also implemented within this file since it requires 222e14bb325SJeff Bonwick * manipulation of the namespace. 223fa9e4066Sahrens */ 224fa9e4066Sahrens 225fa9e4066Sahrens static avl_tree_t spa_namespace_avl; 226fa9e4066Sahrens kmutex_t spa_namespace_lock; 227fa9e4066Sahrens static kcondvar_t spa_namespace_cv; 2280373e76bSbonwick static int spa_active_count; 229416e0cd8Sek int spa_max_replication_override = SPA_DVAS_PER_BP; 230fa9e4066Sahrens 23199653d4eSeschrock static kmutex_t spa_spare_lock; 23239c23413Seschrock static avl_tree_t spa_spare_avl; 233fa94a07fSbrendan static kmutex_t spa_l2cache_lock; 234fa94a07fSbrendan static avl_tree_t spa_l2cache_avl; 23599653d4eSeschrock 236fa9e4066Sahrens kmem_cache_t *spa_buffer_pool; 2378ad4d6ddSJeff Bonwick int spa_mode_global; 238fa9e4066Sahrens 239fa9e4066Sahrens #ifdef ZFS_DEBUG 2403b2aab18SMatthew Ahrens /* Everything except dprintf and spa is on by default in debug builds */ 2413b2aab18SMatthew Ahrens int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA); 242fa9e4066Sahrens #else 243fa9e4066Sahrens int zfs_flags = 0; 244fa9e4066Sahrens #endif 245fa9e4066Sahrens 2460125049cSahrens /* 2470125049cSahrens * zfs_recover can be set to nonzero to attempt to recover from 2480125049cSahrens * otherwise-fatal errors, typically caused by on-disk corruption. When 2490125049cSahrens * set, calls to zfs_panic_recover() will turn into warning messages. 250*8b36997aSMatthew Ahrens * This should only be used as a last resort, as it typically results 251*8b36997aSMatthew Ahrens * in leaked space, or worse. 2520125049cSahrens */ 2530125049cSahrens int zfs_recover = 0; 2540125049cSahrens 25569962b56SMatthew Ahrens /* 25669962b56SMatthew Ahrens * Expiration time in milliseconds. This value has two meanings. First it is 25769962b56SMatthew Ahrens * used to determine when the spa_deadman() logic should fire. By default the 25869962b56SMatthew Ahrens * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds. 25969962b56SMatthew Ahrens * Secondly, the value determines if an I/O is considered "hung". Any I/O that 26069962b56SMatthew Ahrens * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting 26169962b56SMatthew Ahrens * in a system panic. 26269962b56SMatthew Ahrens */ 26369962b56SMatthew Ahrens uint64_t zfs_deadman_synctime_ms = 1000000ULL; 264283b8460SGeorge.Wilson 265283b8460SGeorge.Wilson /* 26669962b56SMatthew Ahrens * Check time in milliseconds. This defines the frequency at which we check 26769962b56SMatthew Ahrens * for hung I/O. 268283b8460SGeorge.Wilson */ 26969962b56SMatthew Ahrens uint64_t zfs_deadman_checktime_ms = 5000ULL; 270283b8460SGeorge.Wilson 271283b8460SGeorge.Wilson /* 272283b8460SGeorge.Wilson * Override the zfs deadman behavior via /etc/system. By default the 273283b8460SGeorge.Wilson * deadman is enabled except on VMware and sparc deployments. 274283b8460SGeorge.Wilson */ 275283b8460SGeorge.Wilson int zfs_deadman_enabled = -1; 276283b8460SGeorge.Wilson 27769962b56SMatthew Ahrens /* 27869962b56SMatthew Ahrens * The worst case is single-sector max-parity RAID-Z blocks, in which 27969962b56SMatthew Ahrens * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1) 28069962b56SMatthew Ahrens * times the size; so just assume that. Add to this the fact that 28169962b56SMatthew Ahrens * we can have up to 3 DVAs per bp, and one more factor of 2 because 28269962b56SMatthew Ahrens * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together, 28369962b56SMatthew Ahrens * the worst case is: 28469962b56SMatthew Ahrens * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24 28569962b56SMatthew Ahrens */ 28669962b56SMatthew Ahrens int spa_asize_inflation = 24; 287fa9e4066Sahrens 288e05725b1Sbonwick /* 289e05725b1Sbonwick * ========================================================================== 290e05725b1Sbonwick * SPA config locking 291e05725b1Sbonwick * ========================================================================== 292e05725b1Sbonwick */ 293e05725b1Sbonwick static void 294e14bb325SJeff Bonwick spa_config_lock_init(spa_t *spa) 295e14bb325SJeff Bonwick { 296e14bb325SJeff Bonwick for (int i = 0; i < SCL_LOCKS; i++) { 297e14bb325SJeff Bonwick spa_config_lock_t *scl = &spa->spa_config_lock[i]; 298e14bb325SJeff Bonwick mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL); 299e14bb325SJeff Bonwick cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL); 3003b2aab18SMatthew Ahrens refcount_create_untracked(&scl->scl_count); 301e14bb325SJeff Bonwick scl->scl_writer = NULL; 302e14bb325SJeff Bonwick scl->scl_write_wanted = 0; 303e14bb325SJeff Bonwick } 304e05725b1Sbonwick } 305e05725b1Sbonwick 306e05725b1Sbonwick static void 307e14bb325SJeff Bonwick spa_config_lock_destroy(spa_t *spa) 308e14bb325SJeff Bonwick { 309e14bb325SJeff Bonwick for (int i = 0; i < SCL_LOCKS; i++) { 310e14bb325SJeff Bonwick spa_config_lock_t *scl = &spa->spa_config_lock[i]; 311e14bb325SJeff Bonwick mutex_destroy(&scl->scl_lock); 312e14bb325SJeff Bonwick cv_destroy(&scl->scl_cv); 313e14bb325SJeff Bonwick refcount_destroy(&scl->scl_count); 314e14bb325SJeff Bonwick ASSERT(scl->scl_writer == NULL); 315e14bb325SJeff Bonwick ASSERT(scl->scl_write_wanted == 0); 316e14bb325SJeff Bonwick } 317e14bb325SJeff Bonwick } 318e14bb325SJeff Bonwick 319e14bb325SJeff Bonwick int 320e14bb325SJeff Bonwick spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw) 321e05725b1Sbonwick { 322e14bb325SJeff Bonwick for (int i = 0; i < SCL_LOCKS; i++) { 323e14bb325SJeff Bonwick spa_config_lock_t *scl = &spa->spa_config_lock[i]; 324e14bb325SJeff Bonwick if (!(locks & (1 << i))) 325e14bb325SJeff Bonwick continue; 326e14bb325SJeff Bonwick mutex_enter(&scl->scl_lock); 327e14bb325SJeff Bonwick if (rw == RW_READER) { 328e14bb325SJeff Bonwick if (scl->scl_writer || scl->scl_write_wanted) { 329e14bb325SJeff Bonwick mutex_exit(&scl->scl_lock); 330e14bb325SJeff Bonwick spa_config_exit(spa, locks ^ (1 << i), tag); 331e14bb325SJeff Bonwick return (0); 332e14bb325SJeff Bonwick } 333e14bb325SJeff Bonwick } else { 334e14bb325SJeff Bonwick ASSERT(scl->scl_writer != curthread); 335e14bb325SJeff Bonwick if (!refcount_is_zero(&scl->scl_count)) { 336e14bb325SJeff Bonwick mutex_exit(&scl->scl_lock); 337e14bb325SJeff Bonwick spa_config_exit(spa, locks ^ (1 << i), tag); 338e14bb325SJeff Bonwick return (0); 339e14bb325SJeff Bonwick } 340e14bb325SJeff Bonwick scl->scl_writer = curthread; 341e14bb325SJeff Bonwick } 342e14bb325SJeff Bonwick (void) refcount_add(&scl->scl_count, tag); 343e14bb325SJeff Bonwick mutex_exit(&scl->scl_lock); 344e14bb325SJeff Bonwick } 345e14bb325SJeff Bonwick return (1); 346e05725b1Sbonwick } 347e05725b1Sbonwick 348e05725b1Sbonwick void 349e14bb325SJeff Bonwick spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw) 350e05725b1Sbonwick { 351f64c0e34SEric Taylor int wlocks_held = 0; 352f64c0e34SEric Taylor 3533b2aab18SMatthew Ahrens ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY); 3543b2aab18SMatthew Ahrens 355e14bb325SJeff Bonwick for (int i = 0; i < SCL_LOCKS; i++) { 356e14bb325SJeff Bonwick spa_config_lock_t *scl = &spa->spa_config_lock[i]; 357f64c0e34SEric Taylor if (scl->scl_writer == curthread) 358f64c0e34SEric Taylor wlocks_held |= (1 << i); 359e14bb325SJeff Bonwick if (!(locks & (1 << i))) 360e14bb325SJeff Bonwick continue; 361e14bb325SJeff Bonwick mutex_enter(&scl->scl_lock); 362e14bb325SJeff Bonwick if (rw == RW_READER) { 363e14bb325SJeff Bonwick while (scl->scl_writer || scl->scl_write_wanted) { 364e14bb325SJeff Bonwick cv_wait(&scl->scl_cv, &scl->scl_lock); 365e14bb325SJeff Bonwick } 366e14bb325SJeff Bonwick } else { 367e14bb325SJeff Bonwick ASSERT(scl->scl_writer != curthread); 368e14bb325SJeff Bonwick while (!refcount_is_zero(&scl->scl_count)) { 369e14bb325SJeff Bonwick scl->scl_write_wanted++; 370e14bb325SJeff Bonwick cv_wait(&scl->scl_cv, &scl->scl_lock); 371e14bb325SJeff Bonwick scl->scl_write_wanted--; 372e14bb325SJeff Bonwick } 373e14bb325SJeff Bonwick scl->scl_writer = curthread; 374e14bb325SJeff Bonwick } 375e14bb325SJeff Bonwick (void) refcount_add(&scl->scl_count, tag); 376e14bb325SJeff Bonwick mutex_exit(&scl->scl_lock); 377e05725b1Sbonwick } 378f64c0e34SEric Taylor ASSERT(wlocks_held <= locks); 379e05725b1Sbonwick } 380e05725b1Sbonwick 381e05725b1Sbonwick void 382e14bb325SJeff Bonwick spa_config_exit(spa_t *spa, int locks, void *tag) 383e05725b1Sbonwick { 384e14bb325SJeff Bonwick for (int i = SCL_LOCKS - 1; i >= 0; i--) { 385e14bb325SJeff Bonwick spa_config_lock_t *scl = &spa->spa_config_lock[i]; 386e14bb325SJeff Bonwick if (!(locks & (1 << i))) 387e14bb325SJeff Bonwick continue; 388e14bb325SJeff Bonwick mutex_enter(&scl->scl_lock); 389e14bb325SJeff Bonwick ASSERT(!refcount_is_zero(&scl->scl_count)); 390e14bb325SJeff Bonwick if (refcount_remove(&scl->scl_count, tag) == 0) { 391e14bb325SJeff Bonwick ASSERT(scl->scl_writer == NULL || 392e14bb325SJeff Bonwick scl->scl_writer == curthread); 393e14bb325SJeff Bonwick scl->scl_writer = NULL; /* OK in either case */ 394e14bb325SJeff Bonwick cv_broadcast(&scl->scl_cv); 395e14bb325SJeff Bonwick } 396e14bb325SJeff Bonwick mutex_exit(&scl->scl_lock); 397e05725b1Sbonwick } 398e05725b1Sbonwick } 399e05725b1Sbonwick 400e14bb325SJeff Bonwick int 401e14bb325SJeff Bonwick spa_config_held(spa_t *spa, int locks, krw_t rw) 402e05725b1Sbonwick { 403e14bb325SJeff Bonwick int locks_held = 0; 404e05725b1Sbonwick 405e14bb325SJeff Bonwick for (int i = 0; i < SCL_LOCKS; i++) { 406e14bb325SJeff Bonwick spa_config_lock_t *scl = &spa->spa_config_lock[i]; 407e14bb325SJeff Bonwick if (!(locks & (1 << i))) 408e14bb325SJeff Bonwick continue; 409e14bb325SJeff Bonwick if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) || 410e14bb325SJeff Bonwick (rw == RW_WRITER && scl->scl_writer == curthread)) 411e14bb325SJeff Bonwick locks_held |= 1 << i; 412e14bb325SJeff Bonwick } 413e14bb325SJeff Bonwick 414e14bb325SJeff Bonwick return (locks_held); 415e05725b1Sbonwick } 416e05725b1Sbonwick 417fa9e4066Sahrens /* 418fa9e4066Sahrens * ========================================================================== 419fa9e4066Sahrens * SPA namespace functions 420fa9e4066Sahrens * ========================================================================== 421fa9e4066Sahrens */ 422fa9e4066Sahrens 423fa9e4066Sahrens /* 424fa9e4066Sahrens * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held. 425fa9e4066Sahrens * Returns NULL if no matching spa_t is found. 426fa9e4066Sahrens */ 427fa9e4066Sahrens spa_t * 428fa9e4066Sahrens spa_lookup(const char *name) 429fa9e4066Sahrens { 430e14bb325SJeff Bonwick static spa_t search; /* spa_t is large; don't allocate on stack */ 431e14bb325SJeff Bonwick spa_t *spa; 432fa9e4066Sahrens avl_index_t where; 43340feaa91Sahrens char *cp; 434fa9e4066Sahrens 435fa9e4066Sahrens ASSERT(MUTEX_HELD(&spa_namespace_lock)); 436fa9e4066Sahrens 4373b2aab18SMatthew Ahrens (void) strlcpy(search.spa_name, name, sizeof (search.spa_name)); 4383b2aab18SMatthew Ahrens 43940feaa91Sahrens /* 44040feaa91Sahrens * If it's a full dataset name, figure out the pool name and 44140feaa91Sahrens * just use that. 44240feaa91Sahrens */ 44378f17100SMatthew Ahrens cp = strpbrk(search.spa_name, "/@#"); 4443b2aab18SMatthew Ahrens if (cp != NULL) 44540feaa91Sahrens *cp = '\0'; 44640feaa91Sahrens 447fa9e4066Sahrens spa = avl_find(&spa_namespace_avl, &search, &where); 448fa9e4066Sahrens 449fa9e4066Sahrens return (spa); 450fa9e4066Sahrens } 451fa9e4066Sahrens 452283b8460SGeorge.Wilson /* 453283b8460SGeorge.Wilson * Fires when spa_sync has not completed within zfs_deadman_synctime_ms. 454283b8460SGeorge.Wilson * If the zfs_deadman_enabled flag is set then it inspects all vdev queues 455283b8460SGeorge.Wilson * looking for potentially hung I/Os. 456283b8460SGeorge.Wilson */ 457283b8460SGeorge.Wilson void 458283b8460SGeorge.Wilson spa_deadman(void *arg) 459283b8460SGeorge.Wilson { 460283b8460SGeorge.Wilson spa_t *spa = arg; 461283b8460SGeorge.Wilson 4620713e232SGeorge Wilson /* 4630713e232SGeorge Wilson * Disable the deadman timer if the pool is suspended. 4640713e232SGeorge Wilson */ 4650713e232SGeorge Wilson if (spa_suspended(spa)) { 4660713e232SGeorge Wilson VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 4670713e232SGeorge Wilson return; 4680713e232SGeorge Wilson } 4690713e232SGeorge Wilson 470283b8460SGeorge.Wilson zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu", 471283b8460SGeorge.Wilson (gethrtime() - spa->spa_sync_starttime) / NANOSEC, 472283b8460SGeorge.Wilson ++spa->spa_deadman_calls); 473283b8460SGeorge.Wilson if (zfs_deadman_enabled) 474283b8460SGeorge.Wilson vdev_deadman(spa->spa_root_vdev); 475283b8460SGeorge.Wilson } 476283b8460SGeorge.Wilson 477fa9e4066Sahrens /* 478fa9e4066Sahrens * Create an uninitialized spa_t with the given name. Requires 479fa9e4066Sahrens * spa_namespace_lock. The caller must ensure that the spa_t doesn't already 480fa9e4066Sahrens * exist by calling spa_lookup() first. 481fa9e4066Sahrens */ 482fa9e4066Sahrens spa_t * 483468c413aSTim Haley spa_add(const char *name, nvlist_t *config, const char *altroot) 484fa9e4066Sahrens { 485fa9e4066Sahrens spa_t *spa; 486c5904d13Seschrock spa_config_dirent_t *dp; 487283b8460SGeorge.Wilson cyc_handler_t hdlr; 488283b8460SGeorge.Wilson cyc_time_t when; 489fa9e4066Sahrens 490fa9e4066Sahrens ASSERT(MUTEX_HELD(&spa_namespace_lock)); 491fa9e4066Sahrens 492fa9e4066Sahrens spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP); 493fa9e4066Sahrens 494c25056deSgw mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL); 495c25056deSgw mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL); 49635a5a358SJonathan Adams mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL); 497c25056deSgw mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL); 49835a5a358SJonathan Adams mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL); 499c25056deSgw mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL); 50035a5a358SJonathan Adams mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL); 501a1521560SJeff Bonwick mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL); 502a1521560SJeff Bonwick mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL); 503c3a66015SMatthew Ahrens mutex_init(&spa->spa_iokstat_lock, NULL, MUTEX_DEFAULT, NULL); 504c25056deSgw 505c25056deSgw cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL); 50635a5a358SJonathan Adams cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL); 507c25056deSgw cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL); 508e14bb325SJeff Bonwick cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL); 509c25056deSgw 510b24ab676SJeff Bonwick for (int t = 0; t < TXG_SIZE; t++) 511cde58dbcSMatthew Ahrens bplist_create(&spa->spa_free_bplist[t]); 512b24ab676SJeff Bonwick 513e14bb325SJeff Bonwick (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name)); 514fa9e4066Sahrens spa->spa_state = POOL_STATE_UNINITIALIZED; 515fa9e4066Sahrens spa->spa_freeze_txg = UINT64_MAX; 5160373e76bSbonwick spa->spa_final_txg = UINT64_MAX; 517468c413aSTim Haley spa->spa_load_max_txg = UINT64_MAX; 51835a5a358SJonathan Adams spa->spa_proc = &p0; 51935a5a358SJonathan Adams spa->spa_proc_state = SPA_PROC_NONE; 520fa9e4066Sahrens 521283b8460SGeorge.Wilson hdlr.cyh_func = spa_deadman; 522283b8460SGeorge.Wilson hdlr.cyh_arg = spa; 523283b8460SGeorge.Wilson hdlr.cyh_level = CY_LOW_LEVEL; 524283b8460SGeorge.Wilson 52569962b56SMatthew Ahrens spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms); 526283b8460SGeorge.Wilson 527283b8460SGeorge.Wilson /* 528283b8460SGeorge.Wilson * This determines how often we need to check for hung I/Os after 529283b8460SGeorge.Wilson * the cyclic has already fired. Since checking for hung I/Os is 530283b8460SGeorge.Wilson * an expensive operation we don't want to check too frequently. 53169962b56SMatthew Ahrens * Instead wait for 5 seconds before checking again. 532283b8460SGeorge.Wilson */ 53369962b56SMatthew Ahrens when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms); 534283b8460SGeorge.Wilson when.cyt_when = CY_INFINITY; 535283b8460SGeorge.Wilson mutex_enter(&cpu_lock); 536283b8460SGeorge.Wilson spa->spa_deadman_cycid = cyclic_add(&hdlr, &when); 537283b8460SGeorge.Wilson mutex_exit(&cpu_lock); 538283b8460SGeorge.Wilson 539fa9e4066Sahrens refcount_create(&spa->spa_refcount); 540e14bb325SJeff Bonwick spa_config_lock_init(spa); 541fa9e4066Sahrens 542fa9e4066Sahrens avl_add(&spa_namespace_avl, spa); 543fa9e4066Sahrens 5440373e76bSbonwick /* 5450373e76bSbonwick * Set the alternate root, if there is one. 5460373e76bSbonwick */ 5470373e76bSbonwick if (altroot) { 5480373e76bSbonwick spa->spa_root = spa_strdup(altroot); 5490373e76bSbonwick spa_active_count++; 5500373e76bSbonwick } 5510373e76bSbonwick 552c5904d13Seschrock /* 553c5904d13Seschrock * Every pool starts with the default cachefile 554c5904d13Seschrock */ 555c5904d13Seschrock list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t), 556c5904d13Seschrock offsetof(spa_config_dirent_t, scd_link)); 557c5904d13Seschrock 558c5904d13Seschrock dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP); 559ef912c80STim Haley dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path); 560c5904d13Seschrock list_insert_head(&spa->spa_config_list, dp); 561c5904d13Seschrock 5624b964adaSGeorge Wilson VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME, 5634b964adaSGeorge Wilson KM_SLEEP) == 0); 5644b964adaSGeorge Wilson 565ad135b5dSChristopher Siden if (config != NULL) { 566ad135b5dSChristopher Siden nvlist_t *features; 567ad135b5dSChristopher Siden 568ad135b5dSChristopher Siden if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ, 569ad135b5dSChristopher Siden &features) == 0) { 570ad135b5dSChristopher Siden VERIFY(nvlist_dup(features, &spa->spa_label_features, 571ad135b5dSChristopher Siden 0) == 0); 572ad135b5dSChristopher Siden } 573ad135b5dSChristopher Siden 574468c413aSTim Haley VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0); 575ad135b5dSChristopher Siden } 576ad135b5dSChristopher Siden 577ad135b5dSChristopher Siden if (spa->spa_label_features == NULL) { 578ad135b5dSChristopher Siden VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME, 579ad135b5dSChristopher Siden KM_SLEEP) == 0); 580ad135b5dSChristopher Siden } 581468c413aSTim Haley 582c3a66015SMatthew Ahrens spa->spa_iokstat = kstat_create("zfs", 0, name, 583c3a66015SMatthew Ahrens "disk", KSTAT_TYPE_IO, 1, 0); 584c3a66015SMatthew Ahrens if (spa->spa_iokstat) { 585c3a66015SMatthew Ahrens spa->spa_iokstat->ks_lock = &spa->spa_iokstat_lock; 586c3a66015SMatthew Ahrens kstat_install(spa->spa_iokstat); 587c3a66015SMatthew Ahrens } 588c3a66015SMatthew Ahrens 5893b2aab18SMatthew Ahrens spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0); 5903b2aab18SMatthew Ahrens 59143466aaeSMax Grossman /* 59243466aaeSMax Grossman * As a pool is being created, treat all features as disabled by 59343466aaeSMax Grossman * setting SPA_FEATURE_DISABLED for all entries in the feature 59443466aaeSMax Grossman * refcount cache. 59543466aaeSMax Grossman */ 59643466aaeSMax Grossman for (int i = 0; i < SPA_FEATURES; i++) { 59743466aaeSMax Grossman spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED; 59843466aaeSMax Grossman } 59943466aaeSMax Grossman 600fa9e4066Sahrens return (spa); 601fa9e4066Sahrens } 602fa9e4066Sahrens 603fa9e4066Sahrens /* 604fa9e4066Sahrens * Removes a spa_t from the namespace, freeing up any memory used. Requires 605fa9e4066Sahrens * spa_namespace_lock. This is called only after the spa_t has been closed and 606fa9e4066Sahrens * deactivated. 607fa9e4066Sahrens */ 608fa9e4066Sahrens void 609fa9e4066Sahrens spa_remove(spa_t *spa) 610fa9e4066Sahrens { 611c5904d13Seschrock spa_config_dirent_t *dp; 612c5904d13Seschrock 613fa9e4066Sahrens ASSERT(MUTEX_HELD(&spa_namespace_lock)); 614fa9e4066Sahrens ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 615fa9e4066Sahrens 6161195e687SMark J Musante nvlist_free(spa->spa_config_splitting); 6171195e687SMark J Musante 618fa9e4066Sahrens avl_remove(&spa_namespace_avl, spa); 619fa9e4066Sahrens cv_broadcast(&spa_namespace_cv); 620fa9e4066Sahrens 6210373e76bSbonwick if (spa->spa_root) { 622fa9e4066Sahrens spa_strfree(spa->spa_root); 6230373e76bSbonwick spa_active_count--; 6240373e76bSbonwick } 625fa9e4066Sahrens 626c5904d13Seschrock while ((dp = list_head(&spa->spa_config_list)) != NULL) { 627c5904d13Seschrock list_remove(&spa->spa_config_list, dp); 628c5904d13Seschrock if (dp->scd_path != NULL) 629c5904d13Seschrock spa_strfree(dp->scd_path); 630c5904d13Seschrock kmem_free(dp, sizeof (spa_config_dirent_t)); 631c5904d13Seschrock } 632c5904d13Seschrock 633c5904d13Seschrock list_destroy(&spa->spa_config_list); 6342f8aaab3Seschrock 635ad135b5dSChristopher Siden nvlist_free(spa->spa_label_features); 6364b964adaSGeorge Wilson nvlist_free(spa->spa_load_info); 637fa9e4066Sahrens spa_config_set(spa, NULL); 638fa9e4066Sahrens 639283b8460SGeorge.Wilson mutex_enter(&cpu_lock); 640283b8460SGeorge.Wilson if (spa->spa_deadman_cycid != CYCLIC_NONE) 641283b8460SGeorge.Wilson cyclic_remove(spa->spa_deadman_cycid); 642283b8460SGeorge.Wilson mutex_exit(&cpu_lock); 643283b8460SGeorge.Wilson spa->spa_deadman_cycid = CYCLIC_NONE; 644283b8460SGeorge.Wilson 645fa9e4066Sahrens refcount_destroy(&spa->spa_refcount); 64691ebeef5Sahrens 647e14bb325SJeff Bonwick spa_config_lock_destroy(spa); 648fa9e4066Sahrens 649c3a66015SMatthew Ahrens kstat_delete(spa->spa_iokstat); 650c3a66015SMatthew Ahrens spa->spa_iokstat = NULL; 651c3a66015SMatthew Ahrens 652b24ab676SJeff Bonwick for (int t = 0; t < TXG_SIZE; t++) 653cde58dbcSMatthew Ahrens bplist_destroy(&spa->spa_free_bplist[t]); 654b24ab676SJeff Bonwick 655c25056deSgw cv_destroy(&spa->spa_async_cv); 65635a5a358SJonathan Adams cv_destroy(&spa->spa_proc_cv); 657c25056deSgw cv_destroy(&spa->spa_scrub_io_cv); 658e14bb325SJeff Bonwick cv_destroy(&spa->spa_suspend_cv); 659c25056deSgw 6605ad82045Snd mutex_destroy(&spa->spa_async_lock); 661c25056deSgw mutex_destroy(&spa->spa_errlist_lock); 66235a5a358SJonathan Adams mutex_destroy(&spa->spa_errlog_lock); 66306eeb2adSek mutex_destroy(&spa->spa_history_lock); 66435a5a358SJonathan Adams mutex_destroy(&spa->spa_proc_lock); 665b1b8ab34Slling mutex_destroy(&spa->spa_props_lock); 66635a5a358SJonathan Adams mutex_destroy(&spa->spa_scrub_lock); 667e14bb325SJeff Bonwick mutex_destroy(&spa->spa_suspend_lock); 668a1521560SJeff Bonwick mutex_destroy(&spa->spa_vdev_top_lock); 669c3a66015SMatthew Ahrens mutex_destroy(&spa->spa_iokstat_lock); 6705ad82045Snd 671fa9e4066Sahrens kmem_free(spa, sizeof (spa_t)); 672fa9e4066Sahrens } 673fa9e4066Sahrens 674fa9e4066Sahrens /* 675fa9e4066Sahrens * Given a pool, return the next pool in the namespace, or NULL if there is 676fa9e4066Sahrens * none. If 'prev' is NULL, return the first pool. 677fa9e4066Sahrens */ 678fa9e4066Sahrens spa_t * 679fa9e4066Sahrens spa_next(spa_t *prev) 680fa9e4066Sahrens { 681fa9e4066Sahrens ASSERT(MUTEX_HELD(&spa_namespace_lock)); 682fa9e4066Sahrens 683fa9e4066Sahrens if (prev) 684fa9e4066Sahrens return (AVL_NEXT(&spa_namespace_avl, prev)); 685fa9e4066Sahrens else 686fa9e4066Sahrens return (avl_first(&spa_namespace_avl)); 687fa9e4066Sahrens } 688fa9e4066Sahrens 689fa9e4066Sahrens /* 690fa9e4066Sahrens * ========================================================================== 691fa9e4066Sahrens * SPA refcount functions 692fa9e4066Sahrens * ========================================================================== 693fa9e4066Sahrens */ 694fa9e4066Sahrens 695fa9e4066Sahrens /* 696fa9e4066Sahrens * Add a reference to the given spa_t. Must have at least one reference, or 697fa9e4066Sahrens * have the namespace lock held. 698fa9e4066Sahrens */ 699fa9e4066Sahrens void 700fa9e4066Sahrens spa_open_ref(spa_t *spa, void *tag) 701fa9e4066Sahrens { 702088f3894Sahrens ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref || 703fa9e4066Sahrens MUTEX_HELD(&spa_namespace_lock)); 704fa9e4066Sahrens (void) refcount_add(&spa->spa_refcount, tag); 705fa9e4066Sahrens } 706fa9e4066Sahrens 707fa9e4066Sahrens /* 708fa9e4066Sahrens * Remove a reference to the given spa_t. Must have at least one reference, or 709fa9e4066Sahrens * have the namespace lock held. 710fa9e4066Sahrens */ 711fa9e4066Sahrens void 712fa9e4066Sahrens spa_close(spa_t *spa, void *tag) 713fa9e4066Sahrens { 714088f3894Sahrens ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref || 715fa9e4066Sahrens MUTEX_HELD(&spa_namespace_lock)); 716fa9e4066Sahrens (void) refcount_remove(&spa->spa_refcount, tag); 717fa9e4066Sahrens } 718fa9e4066Sahrens 719fa9e4066Sahrens /* 720fa9e4066Sahrens * Check to see if the spa refcount is zero. Must be called with 721088f3894Sahrens * spa_namespace_lock held. We really compare against spa_minref, which is the 722fa9e4066Sahrens * number of references acquired when opening a pool 723fa9e4066Sahrens */ 724fa9e4066Sahrens boolean_t 725fa9e4066Sahrens spa_refcount_zero(spa_t *spa) 726fa9e4066Sahrens { 727fa9e4066Sahrens ASSERT(MUTEX_HELD(&spa_namespace_lock)); 728fa9e4066Sahrens 729088f3894Sahrens return (refcount_count(&spa->spa_refcount) == spa->spa_minref); 730fa9e4066Sahrens } 731fa9e4066Sahrens 73299653d4eSeschrock /* 73399653d4eSeschrock * ========================================================================== 734fa94a07fSbrendan * SPA spare and l2cache tracking 73599653d4eSeschrock * ========================================================================== 73699653d4eSeschrock */ 73799653d4eSeschrock 738fa94a07fSbrendan /* 739fa94a07fSbrendan * Hot spares and cache devices are tracked using the same code below, 740fa94a07fSbrendan * for 'auxiliary' devices. 741fa94a07fSbrendan */ 742fa94a07fSbrendan 743fa94a07fSbrendan typedef struct spa_aux { 744fa94a07fSbrendan uint64_t aux_guid; 745fa94a07fSbrendan uint64_t aux_pool; 746fa94a07fSbrendan avl_node_t aux_avl; 747fa94a07fSbrendan int aux_count; 748fa94a07fSbrendan } spa_aux_t; 749fa94a07fSbrendan 750fa94a07fSbrendan static int 751fa94a07fSbrendan spa_aux_compare(const void *a, const void *b) 752fa94a07fSbrendan { 753fa94a07fSbrendan const spa_aux_t *sa = a; 754fa94a07fSbrendan const spa_aux_t *sb = b; 755fa94a07fSbrendan 756fa94a07fSbrendan if (sa->aux_guid < sb->aux_guid) 757fa94a07fSbrendan return (-1); 758fa94a07fSbrendan else if (sa->aux_guid > sb->aux_guid) 759fa94a07fSbrendan return (1); 760fa94a07fSbrendan else 761fa94a07fSbrendan return (0); 762fa94a07fSbrendan } 763fa94a07fSbrendan 764fa94a07fSbrendan void 765fa94a07fSbrendan spa_aux_add(vdev_t *vd, avl_tree_t *avl) 766fa94a07fSbrendan { 767fa94a07fSbrendan avl_index_t where; 768fa94a07fSbrendan spa_aux_t search; 769fa94a07fSbrendan spa_aux_t *aux; 770fa94a07fSbrendan 771fa94a07fSbrendan search.aux_guid = vd->vdev_guid; 772fa94a07fSbrendan if ((aux = avl_find(avl, &search, &where)) != NULL) { 773fa94a07fSbrendan aux->aux_count++; 774fa94a07fSbrendan } else { 775fa94a07fSbrendan aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP); 776fa94a07fSbrendan aux->aux_guid = vd->vdev_guid; 777fa94a07fSbrendan aux->aux_count = 1; 778fa94a07fSbrendan avl_insert(avl, aux, where); 779fa94a07fSbrendan } 780fa94a07fSbrendan } 781fa94a07fSbrendan 782fa94a07fSbrendan void 783fa94a07fSbrendan spa_aux_remove(vdev_t *vd, avl_tree_t *avl) 784fa94a07fSbrendan { 785fa94a07fSbrendan spa_aux_t search; 786fa94a07fSbrendan spa_aux_t *aux; 787fa94a07fSbrendan avl_index_t where; 788fa94a07fSbrendan 789fa94a07fSbrendan search.aux_guid = vd->vdev_guid; 790fa94a07fSbrendan aux = avl_find(avl, &search, &where); 791fa94a07fSbrendan 792fa94a07fSbrendan ASSERT(aux != NULL); 793fa94a07fSbrendan 794fa94a07fSbrendan if (--aux->aux_count == 0) { 795fa94a07fSbrendan avl_remove(avl, aux); 796fa94a07fSbrendan kmem_free(aux, sizeof (spa_aux_t)); 797fa94a07fSbrendan } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) { 798fa94a07fSbrendan aux->aux_pool = 0ULL; 799fa94a07fSbrendan } 800fa94a07fSbrendan } 801fa94a07fSbrendan 802fa94a07fSbrendan boolean_t 80389a89ebfSlling spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl) 804fa94a07fSbrendan { 805fa94a07fSbrendan spa_aux_t search, *found; 806fa94a07fSbrendan 807fa94a07fSbrendan search.aux_guid = guid; 80889a89ebfSlling found = avl_find(avl, &search, NULL); 809fa94a07fSbrendan 810fa94a07fSbrendan if (pool) { 811fa94a07fSbrendan if (found) 812fa94a07fSbrendan *pool = found->aux_pool; 813fa94a07fSbrendan else 814fa94a07fSbrendan *pool = 0ULL; 815fa94a07fSbrendan } 816fa94a07fSbrendan 81789a89ebfSlling if (refcnt) { 81889a89ebfSlling if (found) 81989a89ebfSlling *refcnt = found->aux_count; 82089a89ebfSlling else 82189a89ebfSlling *refcnt = 0; 82289a89ebfSlling } 82389a89ebfSlling 824fa94a07fSbrendan return (found != NULL); 825fa94a07fSbrendan } 826fa94a07fSbrendan 827fa94a07fSbrendan void 828fa94a07fSbrendan spa_aux_activate(vdev_t *vd, avl_tree_t *avl) 829fa94a07fSbrendan { 830fa94a07fSbrendan spa_aux_t search, *found; 831fa94a07fSbrendan avl_index_t where; 832fa94a07fSbrendan 833fa94a07fSbrendan search.aux_guid = vd->vdev_guid; 834fa94a07fSbrendan found = avl_find(avl, &search, &where); 835fa94a07fSbrendan ASSERT(found != NULL); 836fa94a07fSbrendan ASSERT(found->aux_pool == 0ULL); 837fa94a07fSbrendan 838fa94a07fSbrendan found->aux_pool = spa_guid(vd->vdev_spa); 839fa94a07fSbrendan } 840fa94a07fSbrendan 84199653d4eSeschrock /* 84239c23413Seschrock * Spares are tracked globally due to the following constraints: 84339c23413Seschrock * 84439c23413Seschrock * - A spare may be part of multiple pools. 84539c23413Seschrock * - A spare may be added to a pool even if it's actively in use within 84639c23413Seschrock * another pool. 84739c23413Seschrock * - A spare in use in any pool can only be the source of a replacement if 84839c23413Seschrock * the target is a spare in the same pool. 84939c23413Seschrock * 85039c23413Seschrock * We keep track of all spares on the system through the use of a reference 85139c23413Seschrock * counted AVL tree. When a vdev is added as a spare, or used as a replacement 85239c23413Seschrock * spare, then we bump the reference count in the AVL tree. In addition, we set 85339c23413Seschrock * the 'vdev_isspare' member to indicate that the device is a spare (active or 85439c23413Seschrock * inactive). When a spare is made active (used to replace a device in the 85539c23413Seschrock * pool), we also keep track of which pool its been made a part of. 85639c23413Seschrock * 85739c23413Seschrock * The 'spa_spare_lock' protects the AVL tree. These functions are normally 85839c23413Seschrock * called under the spa_namespace lock as part of vdev reconfiguration. The 85939c23413Seschrock * separate spare lock exists for the status query path, which does not need to 86039c23413Seschrock * be completely consistent with respect to other vdev configuration changes. 86199653d4eSeschrock */ 86239c23413Seschrock 86399653d4eSeschrock static int 86499653d4eSeschrock spa_spare_compare(const void *a, const void *b) 86599653d4eSeschrock { 866fa94a07fSbrendan return (spa_aux_compare(a, b)); 86799653d4eSeschrock } 86899653d4eSeschrock 86999653d4eSeschrock void 87039c23413Seschrock spa_spare_add(vdev_t *vd) 87199653d4eSeschrock { 87299653d4eSeschrock mutex_enter(&spa_spare_lock); 87339c23413Seschrock ASSERT(!vd->vdev_isspare); 874fa94a07fSbrendan spa_aux_add(vd, &spa_spare_avl); 87539c23413Seschrock vd->vdev_isspare = B_TRUE; 87699653d4eSeschrock mutex_exit(&spa_spare_lock); 87799653d4eSeschrock } 87899653d4eSeschrock 87999653d4eSeschrock void 88039c23413Seschrock spa_spare_remove(vdev_t *vd) 88199653d4eSeschrock { 88299653d4eSeschrock mutex_enter(&spa_spare_lock); 88339c23413Seschrock ASSERT(vd->vdev_isspare); 884fa94a07fSbrendan spa_aux_remove(vd, &spa_spare_avl); 88539c23413Seschrock vd->vdev_isspare = B_FALSE; 88699653d4eSeschrock mutex_exit(&spa_spare_lock); 88799653d4eSeschrock } 88899653d4eSeschrock 88999653d4eSeschrock boolean_t 89089a89ebfSlling spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt) 89199653d4eSeschrock { 892fa94a07fSbrendan boolean_t found; 89399653d4eSeschrock 89499653d4eSeschrock mutex_enter(&spa_spare_lock); 89589a89ebfSlling found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl); 89699653d4eSeschrock mutex_exit(&spa_spare_lock); 89799653d4eSeschrock 898fa94a07fSbrendan return (found); 89939c23413Seschrock } 90039c23413Seschrock 90139c23413Seschrock void 90239c23413Seschrock spa_spare_activate(vdev_t *vd) 90339c23413Seschrock { 90439c23413Seschrock mutex_enter(&spa_spare_lock); 90539c23413Seschrock ASSERT(vd->vdev_isspare); 906fa94a07fSbrendan spa_aux_activate(vd, &spa_spare_avl); 907fa94a07fSbrendan mutex_exit(&spa_spare_lock); 908fa94a07fSbrendan } 90939c23413Seschrock 910fa94a07fSbrendan /* 911fa94a07fSbrendan * Level 2 ARC devices are tracked globally for the same reasons as spares. 912fa94a07fSbrendan * Cache devices currently only support one pool per cache device, and so 913fa94a07fSbrendan * for these devices the aux reference count is currently unused beyond 1. 914fa94a07fSbrendan */ 91539c23413Seschrock 916fa94a07fSbrendan static int 917fa94a07fSbrendan spa_l2cache_compare(const void *a, const void *b) 918fa94a07fSbrendan { 919fa94a07fSbrendan return (spa_aux_compare(a, b)); 920fa94a07fSbrendan } 921fa94a07fSbrendan 922fa94a07fSbrendan void 923fa94a07fSbrendan spa_l2cache_add(vdev_t *vd) 924fa94a07fSbrendan { 925fa94a07fSbrendan mutex_enter(&spa_l2cache_lock); 926fa94a07fSbrendan ASSERT(!vd->vdev_isl2cache); 927fa94a07fSbrendan spa_aux_add(vd, &spa_l2cache_avl); 928fa94a07fSbrendan vd->vdev_isl2cache = B_TRUE; 929fa94a07fSbrendan mutex_exit(&spa_l2cache_lock); 930fa94a07fSbrendan } 931fa94a07fSbrendan 932fa94a07fSbrendan void 933fa94a07fSbrendan spa_l2cache_remove(vdev_t *vd) 934fa94a07fSbrendan { 935fa94a07fSbrendan mutex_enter(&spa_l2cache_lock); 936fa94a07fSbrendan ASSERT(vd->vdev_isl2cache); 937fa94a07fSbrendan spa_aux_remove(vd, &spa_l2cache_avl); 938fa94a07fSbrendan vd->vdev_isl2cache = B_FALSE; 939fa94a07fSbrendan mutex_exit(&spa_l2cache_lock); 940fa94a07fSbrendan } 941fa94a07fSbrendan 942fa94a07fSbrendan boolean_t 943fa94a07fSbrendan spa_l2cache_exists(uint64_t guid, uint64_t *pool) 944fa94a07fSbrendan { 945fa94a07fSbrendan boolean_t found; 946fa94a07fSbrendan 947fa94a07fSbrendan mutex_enter(&spa_l2cache_lock); 94889a89ebfSlling found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl); 949fa94a07fSbrendan mutex_exit(&spa_l2cache_lock); 950fa94a07fSbrendan 951fa94a07fSbrendan return (found); 952fa94a07fSbrendan } 953fa94a07fSbrendan 954fa94a07fSbrendan void 955fa94a07fSbrendan spa_l2cache_activate(vdev_t *vd) 956fa94a07fSbrendan { 957fa94a07fSbrendan mutex_enter(&spa_l2cache_lock); 958fa94a07fSbrendan ASSERT(vd->vdev_isl2cache); 959fa94a07fSbrendan spa_aux_activate(vd, &spa_l2cache_avl); 960fa94a07fSbrendan mutex_exit(&spa_l2cache_lock); 961fa94a07fSbrendan } 962fa94a07fSbrendan 963fa9e4066Sahrens /* 964fa9e4066Sahrens * ========================================================================== 965fa9e4066Sahrens * SPA vdev locking 966fa9e4066Sahrens * ========================================================================== 967fa9e4066Sahrens */ 968fa9e4066Sahrens 969fa9e4066Sahrens /* 970ea8dc4b6Seschrock * Lock the given spa_t for the purpose of adding or removing a vdev. 971ea8dc4b6Seschrock * Grabs the global spa_namespace_lock plus the spa config lock for writing. 972fa9e4066Sahrens * It returns the next transaction group for the spa_t. 973fa9e4066Sahrens */ 974fa9e4066Sahrens uint64_t 975fa9e4066Sahrens spa_vdev_enter(spa_t *spa) 976fa9e4066Sahrens { 977a1521560SJeff Bonwick mutex_enter(&spa->spa_vdev_top_lock); 978bbfd46c4SJeff Bonwick mutex_enter(&spa_namespace_lock); 97988ecc943SGeorge Wilson return (spa_vdev_config_enter(spa)); 98088ecc943SGeorge Wilson } 98188ecc943SGeorge Wilson 98288ecc943SGeorge Wilson /* 98388ecc943SGeorge Wilson * Internal implementation for spa_vdev_enter(). Used when a vdev 98488ecc943SGeorge Wilson * operation requires multiple syncs (i.e. removing a device) while 98588ecc943SGeorge Wilson * keeping the spa_namespace_lock held. 98688ecc943SGeorge Wilson */ 98788ecc943SGeorge Wilson uint64_t 98888ecc943SGeorge Wilson spa_vdev_config_enter(spa_t *spa) 98988ecc943SGeorge Wilson { 99088ecc943SGeorge Wilson ASSERT(MUTEX_HELD(&spa_namespace_lock)); 9913d7072f8Seschrock 992e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); 993fa9e4066Sahrens 994fa9e4066Sahrens return (spa_last_synced_txg(spa) + 1); 995fa9e4066Sahrens } 996fa9e4066Sahrens 997fa9e4066Sahrens /* 99888ecc943SGeorge Wilson * Used in combination with spa_vdev_config_enter() to allow the syncing 99988ecc943SGeorge Wilson * of multiple transactions without releasing the spa_namespace_lock. 1000fa9e4066Sahrens */ 100188ecc943SGeorge Wilson void 100288ecc943SGeorge Wilson spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag) 1003fa9e4066Sahrens { 100488ecc943SGeorge Wilson ASSERT(MUTEX_HELD(&spa_namespace_lock)); 100588ecc943SGeorge Wilson 10060e34b6a7Sbonwick int config_changed = B_FALSE; 1007ea8dc4b6Seschrock 10080373e76bSbonwick ASSERT(txg > spa_last_synced_txg(spa)); 10090e34b6a7Sbonwick 1010e14bb325SJeff Bonwick spa->spa_pending_vdev = NULL; 1011e14bb325SJeff Bonwick 10120e34b6a7Sbonwick /* 10130e34b6a7Sbonwick * Reassess the DTLs. 10140e34b6a7Sbonwick */ 10150373e76bSbonwick vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE); 10160e34b6a7Sbonwick 1017e14bb325SJeff Bonwick if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) { 10180e34b6a7Sbonwick config_changed = B_TRUE; 10198f18d1faSGeorge Wilson spa->spa_config_generation++; 10200e34b6a7Sbonwick } 1021ea8dc4b6Seschrock 102288ecc943SGeorge Wilson /* 102388ecc943SGeorge Wilson * Verify the metaslab classes. 102488ecc943SGeorge Wilson */ 1025b24ab676SJeff Bonwick ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0); 1026b24ab676SJeff Bonwick ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0); 102788ecc943SGeorge Wilson 1028e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, spa); 1029fa9e4066Sahrens 103088ecc943SGeorge Wilson /* 103188ecc943SGeorge Wilson * Panic the system if the specified tag requires it. This 103288ecc943SGeorge Wilson * is useful for ensuring that configurations are updated 103388ecc943SGeorge Wilson * transactionally. 103488ecc943SGeorge Wilson */ 103588ecc943SGeorge Wilson if (zio_injection_enabled) 10361195e687SMark J Musante zio_handle_panic_injection(spa, tag, 0); 103788ecc943SGeorge Wilson 1038fa9e4066Sahrens /* 1039fa9e4066Sahrens * Note: this txg_wait_synced() is important because it ensures 1040fa9e4066Sahrens * that there won't be more than one config change per txg. 1041fa9e4066Sahrens * This allows us to use the txg as the generation number. 1042fa9e4066Sahrens */ 1043fa9e4066Sahrens if (error == 0) 1044fa9e4066Sahrens txg_wait_synced(spa->spa_dsl_pool, txg); 1045fa9e4066Sahrens 1046fa9e4066Sahrens if (vd != NULL) { 10470713e232SGeorge Wilson ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL); 10488ad4d6ddSJeff Bonwick spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); 1049fa9e4066Sahrens vdev_free(vd); 10508ad4d6ddSJeff Bonwick spa_config_exit(spa, SCL_ALL, spa); 1051fa9e4066Sahrens } 1052fa9e4066Sahrens 1053fa9e4066Sahrens /* 10540e34b6a7Sbonwick * If the config changed, update the config cache. 1055fa9e4066Sahrens */ 10560e34b6a7Sbonwick if (config_changed) 1057c5904d13Seschrock spa_config_sync(spa, B_FALSE, B_TRUE); 105888ecc943SGeorge Wilson } 1059ea8dc4b6Seschrock 106088ecc943SGeorge Wilson /* 106188ecc943SGeorge Wilson * Unlock the spa_t after adding or removing a vdev. Besides undoing the 106288ecc943SGeorge Wilson * locking of spa_vdev_enter(), we also want make sure the transactions have 106388ecc943SGeorge Wilson * synced to disk, and then update the global configuration cache with the new 106488ecc943SGeorge Wilson * information. 106588ecc943SGeorge Wilson */ 106688ecc943SGeorge Wilson int 106788ecc943SGeorge Wilson spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error) 106888ecc943SGeorge Wilson { 106988ecc943SGeorge Wilson spa_vdev_config_exit(spa, vd, txg, error, FTAG); 1070ea8dc4b6Seschrock mutex_exit(&spa_namespace_lock); 1071bbfd46c4SJeff Bonwick mutex_exit(&spa->spa_vdev_top_lock); 1072fa9e4066Sahrens 1073fa9e4066Sahrens return (error); 1074fa9e4066Sahrens } 1075fa9e4066Sahrens 1076e14bb325SJeff Bonwick /* 1077e14bb325SJeff Bonwick * Lock the given spa_t for the purpose of changing vdev state. 1078e14bb325SJeff Bonwick */ 1079e14bb325SJeff Bonwick void 10808f18d1faSGeorge Wilson spa_vdev_state_enter(spa_t *spa, int oplocks) 1081e14bb325SJeff Bonwick { 10828f18d1faSGeorge Wilson int locks = SCL_STATE_ALL | oplocks; 10838f18d1faSGeorge Wilson 1084dcba9f3fSGeorge Wilson /* 1085dcba9f3fSGeorge Wilson * Root pools may need to read of the underlying devfs filesystem 1086dcba9f3fSGeorge Wilson * when opening up a vdev. Unfortunately if we're holding the 1087dcba9f3fSGeorge Wilson * SCL_ZIO lock it will result in a deadlock when we try to issue 1088dcba9f3fSGeorge Wilson * the read from the root filesystem. Instead we "prefetch" 1089dcba9f3fSGeorge Wilson * the associated vnodes that we need prior to opening the 1090dcba9f3fSGeorge Wilson * underlying devices and cache them so that we can prevent 1091dcba9f3fSGeorge Wilson * any I/O when we are doing the actual open. 1092dcba9f3fSGeorge Wilson */ 1093dcba9f3fSGeorge Wilson if (spa_is_root(spa)) { 10949842588bSGeorge Wilson int low = locks & ~(SCL_ZIO - 1); 10959842588bSGeorge Wilson int high = locks & ~low; 10969842588bSGeorge Wilson 10979842588bSGeorge Wilson spa_config_enter(spa, high, spa, RW_WRITER); 1098dcba9f3fSGeorge Wilson vdev_hold(spa->spa_root_vdev); 10999842588bSGeorge Wilson spa_config_enter(spa, low, spa, RW_WRITER); 1100dcba9f3fSGeorge Wilson } else { 1101dcba9f3fSGeorge Wilson spa_config_enter(spa, locks, spa, RW_WRITER); 1102dcba9f3fSGeorge Wilson } 11038f18d1faSGeorge Wilson spa->spa_vdev_locks = locks; 1104e14bb325SJeff Bonwick } 1105e14bb325SJeff Bonwick 1106e14bb325SJeff Bonwick int 1107e14bb325SJeff Bonwick spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error) 1108e14bb325SJeff Bonwick { 1109c6065d0fSGeorge Wilson boolean_t config_changed = B_FALSE; 1110c6065d0fSGeorge Wilson 1111b24ab676SJeff Bonwick if (vd != NULL || error == 0) 1112b24ab676SJeff Bonwick vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev, 1113b24ab676SJeff Bonwick 0, 0, B_FALSE); 1114b24ab676SJeff Bonwick 11158f18d1faSGeorge Wilson if (vd != NULL) { 1116e14bb325SJeff Bonwick vdev_state_dirty(vd->vdev_top); 1117c6065d0fSGeorge Wilson config_changed = B_TRUE; 11188f18d1faSGeorge Wilson spa->spa_config_generation++; 11198f18d1faSGeorge Wilson } 1120e14bb325SJeff Bonwick 1121dcba9f3fSGeorge Wilson if (spa_is_root(spa)) 1122dcba9f3fSGeorge Wilson vdev_rele(spa->spa_root_vdev); 1123dcba9f3fSGeorge Wilson 11248f18d1faSGeorge Wilson ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL); 11258f18d1faSGeorge Wilson spa_config_exit(spa, spa->spa_vdev_locks, spa); 1126e14bb325SJeff Bonwick 11278ad4d6ddSJeff Bonwick /* 11288ad4d6ddSJeff Bonwick * If anything changed, wait for it to sync. This ensures that, 11298ad4d6ddSJeff Bonwick * from the system administrator's perspective, zpool(1M) commands 11308ad4d6ddSJeff Bonwick * are synchronous. This is important for things like zpool offline: 11318ad4d6ddSJeff Bonwick * when the command completes, you expect no further I/O from ZFS. 11328ad4d6ddSJeff Bonwick */ 11338ad4d6ddSJeff Bonwick if (vd != NULL) 11348ad4d6ddSJeff Bonwick txg_wait_synced(spa->spa_dsl_pool, 0); 11358ad4d6ddSJeff Bonwick 1136c6065d0fSGeorge Wilson /* 1137c6065d0fSGeorge Wilson * If the config changed, update the config cache. 1138c6065d0fSGeorge Wilson */ 1139c6065d0fSGeorge Wilson if (config_changed) { 1140c6065d0fSGeorge Wilson mutex_enter(&spa_namespace_lock); 1141c6065d0fSGeorge Wilson spa_config_sync(spa, B_FALSE, B_TRUE); 1142c6065d0fSGeorge Wilson mutex_exit(&spa_namespace_lock); 1143c6065d0fSGeorge Wilson } 1144c6065d0fSGeorge Wilson 1145e14bb325SJeff Bonwick return (error); 1146e14bb325SJeff Bonwick } 1147e14bb325SJeff Bonwick 1148fa9e4066Sahrens /* 1149fa9e4066Sahrens * ========================================================================== 1150fa9e4066Sahrens * Miscellaneous functions 1151fa9e4066Sahrens * ========================================================================== 1152fa9e4066Sahrens */ 1153fa9e4066Sahrens 1154ad135b5dSChristopher Siden void 115543466aaeSMax Grossman spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx) 1156ad135b5dSChristopher Siden { 11572acef22dSMatthew Ahrens if (!nvlist_exists(spa->spa_label_features, feature)) { 11582acef22dSMatthew Ahrens fnvlist_add_boolean(spa->spa_label_features, feature); 115943466aaeSMax Grossman /* 116043466aaeSMax Grossman * When we are creating the pool (tx_txg==TXG_INITIAL), we can't 116143466aaeSMax Grossman * dirty the vdev config because lock SCL_CONFIG is not held. 116243466aaeSMax Grossman * Thankfully, in this case we don't need to dirty the config 116343466aaeSMax Grossman * because it will be written out anyway when we finish 116443466aaeSMax Grossman * creating the pool. 116543466aaeSMax Grossman */ 116643466aaeSMax Grossman if (tx->tx_txg != TXG_INITIAL) 116743466aaeSMax Grossman vdev_config_dirty(spa->spa_root_vdev); 11682acef22dSMatthew Ahrens } 1169ad135b5dSChristopher Siden } 1170ad135b5dSChristopher Siden 1171ad135b5dSChristopher Siden void 1172ad135b5dSChristopher Siden spa_deactivate_mos_feature(spa_t *spa, const char *feature) 1173ad135b5dSChristopher Siden { 11742acef22dSMatthew Ahrens if (nvlist_remove_all(spa->spa_label_features, feature) == 0) 11752acef22dSMatthew Ahrens vdev_config_dirty(spa->spa_root_vdev); 1176ad135b5dSChristopher Siden } 1177ad135b5dSChristopher Siden 1178fa9e4066Sahrens /* 1179fa9e4066Sahrens * Rename a spa_t. 1180fa9e4066Sahrens */ 1181fa9e4066Sahrens int 1182fa9e4066Sahrens spa_rename(const char *name, const char *newname) 1183fa9e4066Sahrens { 1184fa9e4066Sahrens spa_t *spa; 1185fa9e4066Sahrens int err; 1186fa9e4066Sahrens 1187fa9e4066Sahrens /* 1188fa9e4066Sahrens * Lookup the spa_t and grab the config lock for writing. We need to 1189fa9e4066Sahrens * actually open the pool so that we can sync out the necessary labels. 1190fa9e4066Sahrens * It's OK to call spa_open() with the namespace lock held because we 1191ea8dc4b6Seschrock * allow recursive calls for other reasons. 1192fa9e4066Sahrens */ 1193fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 1194fa9e4066Sahrens if ((err = spa_open(name, &spa, FTAG)) != 0) { 1195fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 1196fa9e4066Sahrens return (err); 1197fa9e4066Sahrens } 1198fa9e4066Sahrens 1199e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1200fa9e4066Sahrens 1201fa9e4066Sahrens avl_remove(&spa_namespace_avl, spa); 1202e14bb325SJeff Bonwick (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name)); 1203fa9e4066Sahrens avl_add(&spa_namespace_avl, spa); 1204fa9e4066Sahrens 1205fa9e4066Sahrens /* 1206fa9e4066Sahrens * Sync all labels to disk with the new names by marking the root vdev 1207fa9e4066Sahrens * dirty and waiting for it to sync. It will pick up the new pool name 1208fa9e4066Sahrens * during the sync. 1209fa9e4066Sahrens */ 1210fa9e4066Sahrens vdev_config_dirty(spa->spa_root_vdev); 1211fa9e4066Sahrens 1212e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 1213fa9e4066Sahrens 12140373e76bSbonwick txg_wait_synced(spa->spa_dsl_pool, 0); 1215fa9e4066Sahrens 1216fa9e4066Sahrens /* 1217fa9e4066Sahrens * Sync the updated config cache. 1218fa9e4066Sahrens */ 1219c5904d13Seschrock spa_config_sync(spa, B_FALSE, B_TRUE); 1220fa9e4066Sahrens 1221fa9e4066Sahrens spa_close(spa, FTAG); 1222fa9e4066Sahrens 1223fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 1224fa9e4066Sahrens 1225fa9e4066Sahrens return (0); 1226fa9e4066Sahrens } 1227fa9e4066Sahrens 1228fa9e4066Sahrens /* 1229f9af39baSGeorge Wilson * Return the spa_t associated with given pool_guid, if it exists. If 1230f9af39baSGeorge Wilson * device_guid is non-zero, determine whether the pool exists *and* contains 1231f9af39baSGeorge Wilson * a device with the specified device_guid. 1232fa9e4066Sahrens */ 1233f9af39baSGeorge Wilson spa_t * 1234f9af39baSGeorge Wilson spa_by_guid(uint64_t pool_guid, uint64_t device_guid) 1235fa9e4066Sahrens { 1236fa9e4066Sahrens spa_t *spa; 1237fa9e4066Sahrens avl_tree_t *t = &spa_namespace_avl; 1238fa9e4066Sahrens 1239ea8dc4b6Seschrock ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1240fa9e4066Sahrens 1241fa9e4066Sahrens for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) { 1242fa9e4066Sahrens if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1243fa9e4066Sahrens continue; 1244fa9e4066Sahrens if (spa->spa_root_vdev == NULL) 1245fa9e4066Sahrens continue; 124639c23413Seschrock if (spa_guid(spa) == pool_guid) { 124739c23413Seschrock if (device_guid == 0) 124839c23413Seschrock break; 124939c23413Seschrock 125039c23413Seschrock if (vdev_lookup_by_guid(spa->spa_root_vdev, 125139c23413Seschrock device_guid) != NULL) 125239c23413Seschrock break; 125339c23413Seschrock 125439c23413Seschrock /* 12558654d025Sperrin * Check any devices we may be in the process of adding. 125639c23413Seschrock */ 125739c23413Seschrock if (spa->spa_pending_vdev) { 125839c23413Seschrock if (vdev_lookup_by_guid(spa->spa_pending_vdev, 125939c23413Seschrock device_guid) != NULL) 126039c23413Seschrock break; 126139c23413Seschrock } 126239c23413Seschrock } 1263fa9e4066Sahrens } 1264fa9e4066Sahrens 1265f9af39baSGeorge Wilson return (spa); 1266f9af39baSGeorge Wilson } 1267f9af39baSGeorge Wilson 1268f9af39baSGeorge Wilson /* 1269f9af39baSGeorge Wilson * Determine whether a pool with the given pool_guid exists. 1270f9af39baSGeorge Wilson */ 1271f9af39baSGeorge Wilson boolean_t 1272f9af39baSGeorge Wilson spa_guid_exists(uint64_t pool_guid, uint64_t device_guid) 1273f9af39baSGeorge Wilson { 1274f9af39baSGeorge Wilson return (spa_by_guid(pool_guid, device_guid) != NULL); 1275fa9e4066Sahrens } 1276fa9e4066Sahrens 1277fa9e4066Sahrens char * 1278fa9e4066Sahrens spa_strdup(const char *s) 1279fa9e4066Sahrens { 1280fa9e4066Sahrens size_t len; 1281fa9e4066Sahrens char *new; 1282fa9e4066Sahrens 1283fa9e4066Sahrens len = strlen(s); 1284fa9e4066Sahrens new = kmem_alloc(len + 1, KM_SLEEP); 1285fa9e4066Sahrens bcopy(s, new, len); 1286fa9e4066Sahrens new[len] = '\0'; 1287fa9e4066Sahrens 1288fa9e4066Sahrens return (new); 1289fa9e4066Sahrens } 1290fa9e4066Sahrens 1291fa9e4066Sahrens void 1292fa9e4066Sahrens spa_strfree(char *s) 1293fa9e4066Sahrens { 1294fa9e4066Sahrens kmem_free(s, strlen(s) + 1); 1295fa9e4066Sahrens } 1296fa9e4066Sahrens 1297fa9e4066Sahrens uint64_t 1298fa9e4066Sahrens spa_get_random(uint64_t range) 1299fa9e4066Sahrens { 1300fa9e4066Sahrens uint64_t r; 1301fa9e4066Sahrens 1302fa9e4066Sahrens ASSERT(range != 0); 1303fa9e4066Sahrens 1304fa9e4066Sahrens (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t)); 1305fa9e4066Sahrens 1306fa9e4066Sahrens return (r % range); 1307fa9e4066Sahrens } 1308fa9e4066Sahrens 13091195e687SMark J Musante uint64_t 13101195e687SMark J Musante spa_generate_guid(spa_t *spa) 13111195e687SMark J Musante { 13121195e687SMark J Musante uint64_t guid = spa_get_random(-1ULL); 13131195e687SMark J Musante 13141195e687SMark J Musante if (spa != NULL) { 13151195e687SMark J Musante while (guid == 0 || spa_guid_exists(spa_guid(spa), guid)) 13161195e687SMark J Musante guid = spa_get_random(-1ULL); 13171195e687SMark J Musante } else { 13181195e687SMark J Musante while (guid == 0 || spa_guid_exists(guid, 0)) 13191195e687SMark J Musante guid = spa_get_random(-1ULL); 13201195e687SMark J Musante } 13211195e687SMark J Musante 13221195e687SMark J Musante return (guid); 13231195e687SMark J Musante } 13241195e687SMark J Musante 1325fa9e4066Sahrens void 132643466aaeSMax Grossman snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp) 1327fa9e4066Sahrens { 1328ad135b5dSChristopher Siden char type[256]; 1329f0ba89beSJeff Bonwick char *checksum = NULL; 1330f0ba89beSJeff Bonwick char *compress = NULL; 1331f0ba89beSJeff Bonwick 1332f0ba89beSJeff Bonwick if (bp != NULL) { 1333ad135b5dSChristopher Siden if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) { 1334ad135b5dSChristopher Siden dmu_object_byteswap_t bswap = 1335ad135b5dSChristopher Siden DMU_OT_BYTESWAP(BP_GET_TYPE(bp)); 1336ad135b5dSChristopher Siden (void) snprintf(type, sizeof (type), "bswap %s %s", 1337ad135b5dSChristopher Siden DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ? 1338ad135b5dSChristopher Siden "metadata" : "data", 1339ad135b5dSChristopher Siden dmu_ot_byteswap[bswap].ob_name); 1340ad135b5dSChristopher Siden } else { 1341ad135b5dSChristopher Siden (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name, 1342ad135b5dSChristopher Siden sizeof (type)); 1343ad135b5dSChristopher Siden } 1344f0ba89beSJeff Bonwick checksum = zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name; 1345f0ba89beSJeff Bonwick compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name; 1346f0ba89beSJeff Bonwick } 1347fa9e4066Sahrens 134843466aaeSMax Grossman SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum, 134943466aaeSMax Grossman compress); 1350fa9e4066Sahrens } 1351fa9e4066Sahrens 1352fa9e4066Sahrens void 1353fa9e4066Sahrens spa_freeze(spa_t *spa) 1354fa9e4066Sahrens { 1355fa9e4066Sahrens uint64_t freeze_txg = 0; 1356fa9e4066Sahrens 1357e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1358fa9e4066Sahrens if (spa->spa_freeze_txg == UINT64_MAX) { 1359fa9e4066Sahrens freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE; 1360fa9e4066Sahrens spa->spa_freeze_txg = freeze_txg; 1361fa9e4066Sahrens } 1362e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 1363fa9e4066Sahrens if (freeze_txg != 0) 1364fa9e4066Sahrens txg_wait_synced(spa_get_dsl(spa), freeze_txg); 1365fa9e4066Sahrens } 1366fa9e4066Sahrens 13670125049cSahrens void 13680125049cSahrens zfs_panic_recover(const char *fmt, ...) 13690125049cSahrens { 13700125049cSahrens va_list adx; 13710125049cSahrens 13720125049cSahrens va_start(adx, fmt); 13730125049cSahrens vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx); 13740125049cSahrens va_end(adx); 13750125049cSahrens } 13760125049cSahrens 13773f9d6ad7SLin Ling /* 13783f9d6ad7SLin Ling * This is a stripped-down version of strtoull, suitable only for converting 1379f7170741SWill Andrews * lowercase hexadecimal numbers that don't overflow. 13803f9d6ad7SLin Ling */ 13813f9d6ad7SLin Ling uint64_t 13823f9d6ad7SLin Ling strtonum(const char *str, char **nptr) 13833f9d6ad7SLin Ling { 13843f9d6ad7SLin Ling uint64_t val = 0; 13853f9d6ad7SLin Ling char c; 13863f9d6ad7SLin Ling int digit; 13873f9d6ad7SLin Ling 13883f9d6ad7SLin Ling while ((c = *str) != '\0') { 13893f9d6ad7SLin Ling if (c >= '0' && c <= '9') 13903f9d6ad7SLin Ling digit = c - '0'; 13913f9d6ad7SLin Ling else if (c >= 'a' && c <= 'f') 13923f9d6ad7SLin Ling digit = 10 + c - 'a'; 13933f9d6ad7SLin Ling else 13943f9d6ad7SLin Ling break; 13953f9d6ad7SLin Ling 13963f9d6ad7SLin Ling val *= 16; 13973f9d6ad7SLin Ling val += digit; 13983f9d6ad7SLin Ling 13993f9d6ad7SLin Ling str++; 14003f9d6ad7SLin Ling } 14013f9d6ad7SLin Ling 14023f9d6ad7SLin Ling if (nptr) 14033f9d6ad7SLin Ling *nptr = (char *)str; 14043f9d6ad7SLin Ling 14053f9d6ad7SLin Ling return (val); 14063f9d6ad7SLin Ling } 14073f9d6ad7SLin Ling 1408fa9e4066Sahrens /* 1409fa9e4066Sahrens * ========================================================================== 1410fa9e4066Sahrens * Accessor functions 1411fa9e4066Sahrens * ========================================================================== 1412fa9e4066Sahrens */ 1413fa9e4066Sahrens 1414088f3894Sahrens boolean_t 141588b7b0f2SMatthew Ahrens spa_shutting_down(spa_t *spa) 1416fa9e4066Sahrens { 141788b7b0f2SMatthew Ahrens return (spa->spa_async_suspended); 1418fa9e4066Sahrens } 1419fa9e4066Sahrens 1420fa9e4066Sahrens dsl_pool_t * 1421fa9e4066Sahrens spa_get_dsl(spa_t *spa) 1422fa9e4066Sahrens { 1423fa9e4066Sahrens return (spa->spa_dsl_pool); 1424fa9e4066Sahrens } 1425fa9e4066Sahrens 1426ad135b5dSChristopher Siden boolean_t 1427ad135b5dSChristopher Siden spa_is_initializing(spa_t *spa) 1428ad135b5dSChristopher Siden { 1429ad135b5dSChristopher Siden return (spa->spa_is_initializing); 1430ad135b5dSChristopher Siden } 1431ad135b5dSChristopher Siden 1432fa9e4066Sahrens blkptr_t * 1433fa9e4066Sahrens spa_get_rootblkptr(spa_t *spa) 1434fa9e4066Sahrens { 1435fa9e4066Sahrens return (&spa->spa_ubsync.ub_rootbp); 1436fa9e4066Sahrens } 1437fa9e4066Sahrens 1438fa9e4066Sahrens void 1439fa9e4066Sahrens spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp) 1440fa9e4066Sahrens { 1441fa9e4066Sahrens spa->spa_uberblock.ub_rootbp = *bp; 1442fa9e4066Sahrens } 1443fa9e4066Sahrens 1444fa9e4066Sahrens void 1445fa9e4066Sahrens spa_altroot(spa_t *spa, char *buf, size_t buflen) 1446fa9e4066Sahrens { 1447fa9e4066Sahrens if (spa->spa_root == NULL) 1448fa9e4066Sahrens buf[0] = '\0'; 1449fa9e4066Sahrens else 1450fa9e4066Sahrens (void) strncpy(buf, spa->spa_root, buflen); 1451fa9e4066Sahrens } 1452fa9e4066Sahrens 1453fa9e4066Sahrens int 1454fa9e4066Sahrens spa_sync_pass(spa_t *spa) 1455fa9e4066Sahrens { 1456fa9e4066Sahrens return (spa->spa_sync_pass); 1457fa9e4066Sahrens } 1458fa9e4066Sahrens 1459fa9e4066Sahrens char * 1460fa9e4066Sahrens spa_name(spa_t *spa) 1461fa9e4066Sahrens { 1462fa9e4066Sahrens return (spa->spa_name); 1463fa9e4066Sahrens } 1464fa9e4066Sahrens 1465fa9e4066Sahrens uint64_t 1466fa9e4066Sahrens spa_guid(spa_t *spa) 1467fa9e4066Sahrens { 1468dfbb9432SGeorge Wilson dsl_pool_t *dp = spa_get_dsl(spa); 1469dfbb9432SGeorge Wilson uint64_t guid; 1470dfbb9432SGeorge Wilson 1471b5989ec7Seschrock /* 1472b5989ec7Seschrock * If we fail to parse the config during spa_load(), we can go through 1473b5989ec7Seschrock * the error path (which posts an ereport) and end up here with no root 1474e9103aaeSGarrett D'Amore * vdev. We stash the original pool guid in 'spa_config_guid' to handle 1475b5989ec7Seschrock * this case. 1476b5989ec7Seschrock */ 1477dfbb9432SGeorge Wilson if (spa->spa_root_vdev == NULL) 1478dfbb9432SGeorge Wilson return (spa->spa_config_guid); 1479dfbb9432SGeorge Wilson 1480dfbb9432SGeorge Wilson guid = spa->spa_last_synced_guid != 0 ? 1481dfbb9432SGeorge Wilson spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid; 1482dfbb9432SGeorge Wilson 1483dfbb9432SGeorge Wilson /* 1484dfbb9432SGeorge Wilson * Return the most recently synced out guid unless we're 1485dfbb9432SGeorge Wilson * in syncing context. 1486dfbb9432SGeorge Wilson */ 1487dfbb9432SGeorge Wilson if (dp && dsl_pool_sync_context(dp)) 1488b5989ec7Seschrock return (spa->spa_root_vdev->vdev_guid); 1489b5989ec7Seschrock else 1490dfbb9432SGeorge Wilson return (guid); 1491e9103aaeSGarrett D'Amore } 1492e9103aaeSGarrett D'Amore 1493e9103aaeSGarrett D'Amore uint64_t 1494e9103aaeSGarrett D'Amore spa_load_guid(spa_t *spa) 1495e9103aaeSGarrett D'Amore { 1496e9103aaeSGarrett D'Amore /* 1497e9103aaeSGarrett D'Amore * This is a GUID that exists solely as a reference for the 1498e9103aaeSGarrett D'Amore * purposes of the arc. It is generated at load time, and 1499e9103aaeSGarrett D'Amore * is never written to persistent storage. 1500e9103aaeSGarrett D'Amore */ 1501e9103aaeSGarrett D'Amore return (spa->spa_load_guid); 1502fa9e4066Sahrens } 1503fa9e4066Sahrens 1504fa9e4066Sahrens uint64_t 1505fa9e4066Sahrens spa_last_synced_txg(spa_t *spa) 1506fa9e4066Sahrens { 1507fa9e4066Sahrens return (spa->spa_ubsync.ub_txg); 1508fa9e4066Sahrens } 1509fa9e4066Sahrens 1510fa9e4066Sahrens uint64_t 1511fa9e4066Sahrens spa_first_txg(spa_t *spa) 1512fa9e4066Sahrens { 1513fa9e4066Sahrens return (spa->spa_first_txg); 1514fa9e4066Sahrens } 1515fa9e4066Sahrens 1516b24ab676SJeff Bonwick uint64_t 1517b24ab676SJeff Bonwick spa_syncing_txg(spa_t *spa) 1518b24ab676SJeff Bonwick { 1519b24ab676SJeff Bonwick return (spa->spa_syncing_txg); 1520b24ab676SJeff Bonwick } 1521b24ab676SJeff Bonwick 152288b7b0f2SMatthew Ahrens pool_state_t 1523fa9e4066Sahrens spa_state(spa_t *spa) 1524fa9e4066Sahrens { 1525fa9e4066Sahrens return (spa->spa_state); 1526fa9e4066Sahrens } 1527fa9e4066Sahrens 1528b16da2e2SGeorge Wilson spa_load_state_t 1529b16da2e2SGeorge Wilson spa_load_state(spa_t *spa) 1530b16da2e2SGeorge Wilson { 1531b16da2e2SGeorge Wilson return (spa->spa_load_state); 1532b16da2e2SGeorge Wilson } 1533b16da2e2SGeorge Wilson 1534fa9e4066Sahrens uint64_t 1535fa9e4066Sahrens spa_freeze_txg(spa_t *spa) 1536fa9e4066Sahrens { 1537fa9e4066Sahrens return (spa->spa_freeze_txg); 1538fa9e4066Sahrens } 1539fa9e4066Sahrens 1540fa9e4066Sahrens /* ARGSUSED */ 1541fa9e4066Sahrens uint64_t 1542fa9e4066Sahrens spa_get_asize(spa_t *spa, uint64_t lsize) 1543fa9e4066Sahrens { 154469962b56SMatthew Ahrens return (lsize * spa_asize_inflation); 154544cd46caSbillm } 154644cd46caSbillm 1547485bbbf5SGeorge Wilson uint64_t 1548485bbbf5SGeorge Wilson spa_get_dspace(spa_t *spa) 1549485bbbf5SGeorge Wilson { 1550485bbbf5SGeorge Wilson return (spa->spa_dspace); 1551485bbbf5SGeorge Wilson } 1552485bbbf5SGeorge Wilson 1553485bbbf5SGeorge Wilson void 1554485bbbf5SGeorge Wilson spa_update_dspace(spa_t *spa) 1555485bbbf5SGeorge Wilson { 1556485bbbf5SGeorge Wilson spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) + 1557485bbbf5SGeorge Wilson ddt_get_dedup_dspace(spa); 1558485bbbf5SGeorge Wilson } 1559485bbbf5SGeorge Wilson 15600a4e9518Sgw /* 15610a4e9518Sgw * Return the failure mode that has been set to this pool. The default 15620a4e9518Sgw * behavior will be to block all I/Os when a complete failure occurs. 15630a4e9518Sgw */ 15640a4e9518Sgw uint8_t 15650a4e9518Sgw spa_get_failmode(spa_t *spa) 15660a4e9518Sgw { 15670a4e9518Sgw return (spa->spa_failmode); 15680a4e9518Sgw } 15690a4e9518Sgw 1570e14bb325SJeff Bonwick boolean_t 1571e14bb325SJeff Bonwick spa_suspended(spa_t *spa) 1572e14bb325SJeff Bonwick { 1573e14bb325SJeff Bonwick return (spa->spa_suspended); 1574e14bb325SJeff Bonwick } 1575e14bb325SJeff Bonwick 157644cd46caSbillm uint64_t 157744cd46caSbillm spa_version(spa_t *spa) 157844cd46caSbillm { 157944cd46caSbillm return (spa->spa_ubsync.ub_version); 158044cd46caSbillm } 158144cd46caSbillm 1582b24ab676SJeff Bonwick boolean_t 1583b24ab676SJeff Bonwick spa_deflate(spa_t *spa) 1584b24ab676SJeff Bonwick { 1585b24ab676SJeff Bonwick return (spa->spa_deflate); 1586b24ab676SJeff Bonwick } 1587b24ab676SJeff Bonwick 1588b24ab676SJeff Bonwick metaslab_class_t * 1589b24ab676SJeff Bonwick spa_normal_class(spa_t *spa) 1590b24ab676SJeff Bonwick { 1591b24ab676SJeff Bonwick return (spa->spa_normal_class); 1592b24ab676SJeff Bonwick } 1593b24ab676SJeff Bonwick 1594b24ab676SJeff Bonwick metaslab_class_t * 1595b24ab676SJeff Bonwick spa_log_class(spa_t *spa) 1596b24ab676SJeff Bonwick { 1597b24ab676SJeff Bonwick return (spa->spa_log_class); 1598b24ab676SJeff Bonwick } 1599b24ab676SJeff Bonwick 160044cd46caSbillm int 160144cd46caSbillm spa_max_replication(spa_t *spa) 160244cd46caSbillm { 160344cd46caSbillm /* 1604e7437265Sahrens * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to 160544cd46caSbillm * handle BPs with more than one DVA allocated. Set our max 160644cd46caSbillm * replication level accordingly. 1607fa9e4066Sahrens */ 1608e7437265Sahrens if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS) 160944cd46caSbillm return (1); 161044cd46caSbillm return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override)); 1611fa9e4066Sahrens } 1612fa9e4066Sahrens 16133f9d6ad7SLin Ling int 16143f9d6ad7SLin Ling spa_prev_software_version(spa_t *spa) 16153f9d6ad7SLin Ling { 16163f9d6ad7SLin Ling return (spa->spa_prev_software_version); 16173f9d6ad7SLin Ling } 16183f9d6ad7SLin Ling 1619283b8460SGeorge.Wilson uint64_t 1620283b8460SGeorge.Wilson spa_deadman_synctime(spa_t *spa) 1621283b8460SGeorge.Wilson { 1622283b8460SGeorge.Wilson return (spa->spa_deadman_synctime); 1623283b8460SGeorge.Wilson } 1624283b8460SGeorge.Wilson 162599653d4eSeschrock uint64_t 1626b24ab676SJeff Bonwick dva_get_dsize_sync(spa_t *spa, const dva_t *dva) 162799653d4eSeschrock { 1628b24ab676SJeff Bonwick uint64_t asize = DVA_GET_ASIZE(dva); 1629b24ab676SJeff Bonwick uint64_t dsize = asize; 163099653d4eSeschrock 1631b24ab676SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 163299653d4eSeschrock 1633b24ab676SJeff Bonwick if (asize != 0 && spa->spa_deflate) { 1634b24ab676SJeff Bonwick vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva)); 1635b24ab676SJeff Bonwick dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio; 163699653d4eSeschrock } 1637b24ab676SJeff Bonwick 1638b24ab676SJeff Bonwick return (dsize); 1639b24ab676SJeff Bonwick } 1640b24ab676SJeff Bonwick 1641b24ab676SJeff Bonwick uint64_t 1642b24ab676SJeff Bonwick bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp) 1643b24ab676SJeff Bonwick { 1644b24ab676SJeff Bonwick uint64_t dsize = 0; 1645b24ab676SJeff Bonwick 1646b24ab676SJeff Bonwick for (int d = 0; d < SPA_DVAS_PER_BP; d++) 1647b24ab676SJeff Bonwick dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); 1648b24ab676SJeff Bonwick 1649b24ab676SJeff Bonwick return (dsize); 1650b24ab676SJeff Bonwick } 1651b24ab676SJeff Bonwick 1652b24ab676SJeff Bonwick uint64_t 1653b24ab676SJeff Bonwick bp_get_dsize(spa_t *spa, const blkptr_t *bp) 1654b24ab676SJeff Bonwick { 1655b24ab676SJeff Bonwick uint64_t dsize = 0; 1656b24ab676SJeff Bonwick 1657b24ab676SJeff Bonwick spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 1658b24ab676SJeff Bonwick 1659b24ab676SJeff Bonwick for (int d = 0; d < SPA_DVAS_PER_BP; d++) 1660b24ab676SJeff Bonwick dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); 1661b24ab676SJeff Bonwick 1662e14bb325SJeff Bonwick spa_config_exit(spa, SCL_VDEV, FTAG); 1663b24ab676SJeff Bonwick 1664b24ab676SJeff Bonwick return (dsize); 166599653d4eSeschrock } 166699653d4eSeschrock 1667fa9e4066Sahrens /* 1668fa9e4066Sahrens * ========================================================================== 1669fa9e4066Sahrens * Initialization and Termination 1670fa9e4066Sahrens * ========================================================================== 1671fa9e4066Sahrens */ 1672fa9e4066Sahrens 1673fa9e4066Sahrens static int 1674fa9e4066Sahrens spa_name_compare(const void *a1, const void *a2) 1675fa9e4066Sahrens { 1676fa9e4066Sahrens const spa_t *s1 = a1; 1677fa9e4066Sahrens const spa_t *s2 = a2; 1678fa9e4066Sahrens int s; 1679fa9e4066Sahrens 1680fa9e4066Sahrens s = strcmp(s1->spa_name, s2->spa_name); 1681fa9e4066Sahrens if (s > 0) 1682fa9e4066Sahrens return (1); 1683fa9e4066Sahrens if (s < 0) 1684fa9e4066Sahrens return (-1); 1685fa9e4066Sahrens return (0); 1686fa9e4066Sahrens } 1687fa9e4066Sahrens 16880373e76bSbonwick int 16890373e76bSbonwick spa_busy(void) 16900373e76bSbonwick { 16910373e76bSbonwick return (spa_active_count); 16920373e76bSbonwick } 16930373e76bSbonwick 1694e7cbe64fSgw void 1695e7cbe64fSgw spa_boot_init() 1696e7cbe64fSgw { 1697e7cbe64fSgw spa_config_load(); 1698e7cbe64fSgw } 1699e7cbe64fSgw 1700fa9e4066Sahrens void 1701fa9e4066Sahrens spa_init(int mode) 1702fa9e4066Sahrens { 1703fa9e4066Sahrens mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL); 1704c25056deSgw mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL); 1705fa94a07fSbrendan mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL); 1706fa9e4066Sahrens cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL); 1707fa9e4066Sahrens 1708fa9e4066Sahrens avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t), 1709fa9e4066Sahrens offsetof(spa_t, spa_avl)); 1710fa9e4066Sahrens 1711fa94a07fSbrendan avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t), 1712fa94a07fSbrendan offsetof(spa_aux_t, aux_avl)); 1713fa94a07fSbrendan 1714fa94a07fSbrendan avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t), 1715fa94a07fSbrendan offsetof(spa_aux_t, aux_avl)); 171699653d4eSeschrock 17178ad4d6ddSJeff Bonwick spa_mode_global = mode; 1718fa9e4066Sahrens 1719283b8460SGeorge.Wilson #ifdef _KERNEL 1720283b8460SGeorge.Wilson spa_arch_init(); 1721283b8460SGeorge.Wilson #else 1722cd1c8b85SMatthew Ahrens if (spa_mode_global != FREAD && dprintf_find_string("watch")) { 1723cd1c8b85SMatthew Ahrens arc_procfd = open("/proc/self/ctl", O_WRONLY); 1724cd1c8b85SMatthew Ahrens if (arc_procfd == -1) { 1725cd1c8b85SMatthew Ahrens perror("could not enable watchpoints: " 1726cd1c8b85SMatthew Ahrens "opening /proc/self/ctl failed: "); 1727cd1c8b85SMatthew Ahrens } else { 1728cd1c8b85SMatthew Ahrens arc_watch = B_TRUE; 1729cd1c8b85SMatthew Ahrens } 1730cd1c8b85SMatthew Ahrens } 1731cd1c8b85SMatthew Ahrens #endif 1732cd1c8b85SMatthew Ahrens 1733fa9e4066Sahrens refcount_init(); 1734fa9e4066Sahrens unique_init(); 17350713e232SGeorge Wilson range_tree_init(); 1736fa9e4066Sahrens zio_init(); 1737fa9e4066Sahrens dmu_init(); 1738fa9e4066Sahrens zil_init(); 173987db74c1Sek vdev_cache_stat_init(); 174091ebeef5Sahrens zfs_prop_init(); 1741990b4856Slling zpool_prop_init(); 1742ad135b5dSChristopher Siden zpool_feature_init(); 1743fa9e4066Sahrens spa_config_load(); 1744e14bb325SJeff Bonwick l2arc_start(); 1745fa9e4066Sahrens } 1746fa9e4066Sahrens 1747fa9e4066Sahrens void 1748fa9e4066Sahrens spa_fini(void) 1749fa9e4066Sahrens { 1750e14bb325SJeff Bonwick l2arc_stop(); 1751e14bb325SJeff Bonwick 1752fa9e4066Sahrens spa_evict_all(); 1753fa9e4066Sahrens 175487db74c1Sek vdev_cache_stat_fini(); 1755fa9e4066Sahrens zil_fini(); 1756fa9e4066Sahrens dmu_fini(); 1757fa9e4066Sahrens zio_fini(); 17580713e232SGeorge Wilson range_tree_fini(); 175991ebeef5Sahrens unique_fini(); 1760fa9e4066Sahrens refcount_fini(); 1761fa9e4066Sahrens 1762fa9e4066Sahrens avl_destroy(&spa_namespace_avl); 176399653d4eSeschrock avl_destroy(&spa_spare_avl); 1764fa94a07fSbrendan avl_destroy(&spa_l2cache_avl); 1765fa9e4066Sahrens 1766fa9e4066Sahrens cv_destroy(&spa_namespace_cv); 1767fa9e4066Sahrens mutex_destroy(&spa_namespace_lock); 1768c25056deSgw mutex_destroy(&spa_spare_lock); 1769fa94a07fSbrendan mutex_destroy(&spa_l2cache_lock); 1770fa9e4066Sahrens } 17716ce0521aSperrin 17726ce0521aSperrin /* 17736ce0521aSperrin * Return whether this pool has slogs. No locking needed. 17746ce0521aSperrin * It's not a problem if the wrong answer is returned as it's only for 17756ce0521aSperrin * performance and not correctness 17766ce0521aSperrin */ 17776ce0521aSperrin boolean_t 17786ce0521aSperrin spa_has_slogs(spa_t *spa) 17796ce0521aSperrin { 17806ce0521aSperrin return (spa->spa_log_class->mc_rotor != NULL); 17816ce0521aSperrin } 1782bf82a41bSeschrock 1783b24ab676SJeff Bonwick spa_log_state_t 1784b24ab676SJeff Bonwick spa_get_log_state(spa_t *spa) 1785b24ab676SJeff Bonwick { 1786b24ab676SJeff Bonwick return (spa->spa_log_state); 1787b24ab676SJeff Bonwick } 1788b24ab676SJeff Bonwick 1789b24ab676SJeff Bonwick void 1790b24ab676SJeff Bonwick spa_set_log_state(spa_t *spa, spa_log_state_t state) 1791b24ab676SJeff Bonwick { 1792b24ab676SJeff Bonwick spa->spa_log_state = state; 1793b24ab676SJeff Bonwick } 1794b24ab676SJeff Bonwick 1795bf82a41bSeschrock boolean_t 1796bf82a41bSeschrock spa_is_root(spa_t *spa) 1797bf82a41bSeschrock { 1798bf82a41bSeschrock return (spa->spa_is_root); 1799bf82a41bSeschrock } 18008ad4d6ddSJeff Bonwick 18018ad4d6ddSJeff Bonwick boolean_t 18028ad4d6ddSJeff Bonwick spa_writeable(spa_t *spa) 18038ad4d6ddSJeff Bonwick { 18048ad4d6ddSJeff Bonwick return (!!(spa->spa_mode & FWRITE)); 18058ad4d6ddSJeff Bonwick } 18068ad4d6ddSJeff Bonwick 18078ad4d6ddSJeff Bonwick int 18088ad4d6ddSJeff Bonwick spa_mode(spa_t *spa) 18098ad4d6ddSJeff Bonwick { 18108ad4d6ddSJeff Bonwick return (spa->spa_mode); 18118ad4d6ddSJeff Bonwick } 1812b24ab676SJeff Bonwick 1813b24ab676SJeff Bonwick uint64_t 1814b24ab676SJeff Bonwick spa_bootfs(spa_t *spa) 1815b24ab676SJeff Bonwick { 1816b24ab676SJeff Bonwick return (spa->spa_bootfs); 1817b24ab676SJeff Bonwick } 1818b24ab676SJeff Bonwick 1819b24ab676SJeff Bonwick uint64_t 1820b24ab676SJeff Bonwick spa_delegation(spa_t *spa) 1821b24ab676SJeff Bonwick { 1822b24ab676SJeff Bonwick return (spa->spa_delegation); 1823b24ab676SJeff Bonwick } 1824b24ab676SJeff Bonwick 1825b24ab676SJeff Bonwick objset_t * 1826b24ab676SJeff Bonwick spa_meta_objset(spa_t *spa) 1827b24ab676SJeff Bonwick { 1828b24ab676SJeff Bonwick return (spa->spa_meta_objset); 1829b24ab676SJeff Bonwick } 1830b24ab676SJeff Bonwick 1831b24ab676SJeff Bonwick enum zio_checksum 1832b24ab676SJeff Bonwick spa_dedup_checksum(spa_t *spa) 1833b24ab676SJeff Bonwick { 1834b24ab676SJeff Bonwick return (spa->spa_dedup_checksum); 1835b24ab676SJeff Bonwick } 18363f9d6ad7SLin Ling 18373f9d6ad7SLin Ling /* 18383f9d6ad7SLin Ling * Reset pool scan stat per scan pass (or reboot). 18393f9d6ad7SLin Ling */ 18403f9d6ad7SLin Ling void 18413f9d6ad7SLin Ling spa_scan_stat_init(spa_t *spa) 18423f9d6ad7SLin Ling { 18433f9d6ad7SLin Ling /* data not stored on disk */ 18443f9d6ad7SLin Ling spa->spa_scan_pass_start = gethrestime_sec(); 18453f9d6ad7SLin Ling spa->spa_scan_pass_exam = 0; 18463f9d6ad7SLin Ling vdev_scan_stat_init(spa->spa_root_vdev); 18473f9d6ad7SLin Ling } 18483f9d6ad7SLin Ling 18493f9d6ad7SLin Ling /* 18503f9d6ad7SLin Ling * Get scan stats for zpool status reports 18513f9d6ad7SLin Ling */ 18523f9d6ad7SLin Ling int 18533f9d6ad7SLin Ling spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps) 18543f9d6ad7SLin Ling { 18553f9d6ad7SLin Ling dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL; 18563f9d6ad7SLin Ling 18573f9d6ad7SLin Ling if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE) 1858be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 18593f9d6ad7SLin Ling bzero(ps, sizeof (pool_scan_stat_t)); 18603f9d6ad7SLin Ling 18613f9d6ad7SLin Ling /* data stored on disk */ 18623f9d6ad7SLin Ling ps->pss_func = scn->scn_phys.scn_func; 18633f9d6ad7SLin Ling ps->pss_start_time = scn->scn_phys.scn_start_time; 18643f9d6ad7SLin Ling ps->pss_end_time = scn->scn_phys.scn_end_time; 18653f9d6ad7SLin Ling ps->pss_to_examine = scn->scn_phys.scn_to_examine; 18663f9d6ad7SLin Ling ps->pss_examined = scn->scn_phys.scn_examined; 18673f9d6ad7SLin Ling ps->pss_to_process = scn->scn_phys.scn_to_process; 18683f9d6ad7SLin Ling ps->pss_processed = scn->scn_phys.scn_processed; 18693f9d6ad7SLin Ling ps->pss_errors = scn->scn_phys.scn_errors; 18703f9d6ad7SLin Ling ps->pss_state = scn->scn_phys.scn_state; 18713f9d6ad7SLin Ling 18723f9d6ad7SLin Ling /* data not stored on disk */ 18733f9d6ad7SLin Ling ps->pss_pass_start = spa->spa_scan_pass_start; 18743f9d6ad7SLin Ling ps->pss_pass_exam = spa->spa_scan_pass_exam; 18753f9d6ad7SLin Ling 18763f9d6ad7SLin Ling return (0); 18773f9d6ad7SLin Ling } 187809c9d376SGeorge Wilson 187909c9d376SGeorge Wilson boolean_t 188009c9d376SGeorge Wilson spa_debug_enabled(spa_t *spa) 188109c9d376SGeorge Wilson { 188209c9d376SGeorge Wilson return (spa->spa_debug); 188309c9d376SGeorge Wilson } 1884