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. 232a104a52SAlex Reece * Copyright (c) 2011, 2014 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. 2508b36997aSMatthew Ahrens * This should only be used as a last resort, as it typically results 2518b36997aSMatthew Ahrens * in leaked space, or worse. 2520125049cSahrens */ 2537fd05ac4SMatthew Ahrens boolean_t zfs_recover = B_FALSE; 2547fd05ac4SMatthew Ahrens 2557fd05ac4SMatthew Ahrens /* 2567fd05ac4SMatthew Ahrens * If destroy encounters an EIO while reading metadata (e.g. indirect 2577fd05ac4SMatthew Ahrens * blocks), space referenced by the missing metadata can not be freed. 2587fd05ac4SMatthew Ahrens * Normally this causes the background destroy to become "stalled", as 2597fd05ac4SMatthew Ahrens * it is unable to make forward progress. While in this stalled state, 2607fd05ac4SMatthew Ahrens * all remaining space to free from the error-encountering filesystem is 2617fd05ac4SMatthew Ahrens * "temporarily leaked". Set this flag to cause it to ignore the EIO, 2627fd05ac4SMatthew Ahrens * permanently leak the space from indirect blocks that can not be read, 2637fd05ac4SMatthew Ahrens * and continue to free everything else that it can. 2647fd05ac4SMatthew Ahrens * 2657fd05ac4SMatthew Ahrens * The default, "stalling" behavior is useful if the storage partially 2667fd05ac4SMatthew Ahrens * fails (i.e. some but not all i/os fail), and then later recovers. In 2677fd05ac4SMatthew Ahrens * this case, we will be able to continue pool operations while it is 2687fd05ac4SMatthew Ahrens * partially failed, and when it recovers, we can continue to free the 2697fd05ac4SMatthew Ahrens * space, with no leaks. However, note that this case is actually 2707fd05ac4SMatthew Ahrens * fairly rare. 2717fd05ac4SMatthew Ahrens * 2727fd05ac4SMatthew Ahrens * Typically pools either (a) fail completely (but perhaps temporarily, 2737fd05ac4SMatthew Ahrens * e.g. a top-level vdev going offline), or (b) have localized, 2747fd05ac4SMatthew Ahrens * permanent errors (e.g. disk returns the wrong data due to bit flip or 2757fd05ac4SMatthew Ahrens * firmware bug). In case (a), this setting does not matter because the 2767fd05ac4SMatthew Ahrens * pool will be suspended and the sync thread will not be able to make 2777fd05ac4SMatthew Ahrens * forward progress regardless. In case (b), because the error is 2787fd05ac4SMatthew Ahrens * permanent, the best we can do is leak the minimum amount of space, 2797fd05ac4SMatthew Ahrens * which is what setting this flag will do. Therefore, it is reasonable 2807fd05ac4SMatthew Ahrens * for this flag to normally be set, but we chose the more conservative 2817fd05ac4SMatthew Ahrens * approach of not setting it, so that there is no possibility of 2827fd05ac4SMatthew Ahrens * leaking space in the "partial temporary" failure case. 2837fd05ac4SMatthew Ahrens */ 2847fd05ac4SMatthew Ahrens boolean_t zfs_free_leak_on_eio = B_FALSE; 2850125049cSahrens 28669962b56SMatthew Ahrens /* 28769962b56SMatthew Ahrens * Expiration time in milliseconds. This value has two meanings. First it is 28869962b56SMatthew Ahrens * used to determine when the spa_deadman() logic should fire. By default the 28969962b56SMatthew Ahrens * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds. 29069962b56SMatthew Ahrens * Secondly, the value determines if an I/O is considered "hung". Any I/O that 29169962b56SMatthew Ahrens * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting 29269962b56SMatthew Ahrens * in a system panic. 29369962b56SMatthew Ahrens */ 29469962b56SMatthew Ahrens uint64_t zfs_deadman_synctime_ms = 1000000ULL; 295283b8460SGeorge.Wilson 296283b8460SGeorge.Wilson /* 29769962b56SMatthew Ahrens * Check time in milliseconds. This defines the frequency at which we check 29869962b56SMatthew Ahrens * for hung I/O. 299283b8460SGeorge.Wilson */ 30069962b56SMatthew Ahrens uint64_t zfs_deadman_checktime_ms = 5000ULL; 301283b8460SGeorge.Wilson 302283b8460SGeorge.Wilson /* 303283b8460SGeorge.Wilson * Override the zfs deadman behavior via /etc/system. By default the 304283b8460SGeorge.Wilson * deadman is enabled except on VMware and sparc deployments. 305283b8460SGeorge.Wilson */ 306283b8460SGeorge.Wilson int zfs_deadman_enabled = -1; 307283b8460SGeorge.Wilson 30869962b56SMatthew Ahrens /* 30969962b56SMatthew Ahrens * The worst case is single-sector max-parity RAID-Z blocks, in which 31069962b56SMatthew Ahrens * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1) 31169962b56SMatthew Ahrens * times the size; so just assume that. Add to this the fact that 31269962b56SMatthew Ahrens * we can have up to 3 DVAs per bp, and one more factor of 2 because 31369962b56SMatthew Ahrens * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together, 31469962b56SMatthew Ahrens * the worst case is: 31569962b56SMatthew Ahrens * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24 31669962b56SMatthew Ahrens */ 31769962b56SMatthew Ahrens int spa_asize_inflation = 24; 318fa9e4066Sahrens 3197d46dc6cSMatthew Ahrens /* 3207d46dc6cSMatthew Ahrens * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in 3217d46dc6cSMatthew Ahrens * the pool to be consumed. This ensures that we don't run the pool 3227d46dc6cSMatthew Ahrens * completely out of space, due to unaccounted changes (e.g. to the MOS). 3237d46dc6cSMatthew Ahrens * It also limits the worst-case time to allocate space. If we have 3247d46dc6cSMatthew Ahrens * less than this amount of free space, most ZPL operations (e.g. write, 3257d46dc6cSMatthew Ahrens * create) will return ENOSPC. 3267d46dc6cSMatthew Ahrens * 3277d46dc6cSMatthew Ahrens * Certain operations (e.g. file removal, most administrative actions) can 3287d46dc6cSMatthew Ahrens * use half the slop space. They will only return ENOSPC if less than half 3297d46dc6cSMatthew Ahrens * the slop space is free. Typically, once the pool has less than the slop 3307d46dc6cSMatthew Ahrens * space free, the user will use these operations to free up space in the pool. 3317d46dc6cSMatthew Ahrens * These are the operations that call dsl_pool_adjustedsize() with the netfree 3327d46dc6cSMatthew Ahrens * argument set to TRUE. 3337d46dc6cSMatthew Ahrens * 3347d46dc6cSMatthew Ahrens * A very restricted set of operations are always permitted, regardless of 3357d46dc6cSMatthew Ahrens * the amount of free space. These are the operations that call 3367d46dc6cSMatthew Ahrens * dsl_sync_task(ZFS_SPACE_CHECK_NONE), e.g. "zfs destroy". If these 3377d46dc6cSMatthew Ahrens * operations result in a net increase in the amount of space used, 3387d46dc6cSMatthew Ahrens * it is possible to run the pool completely out of space, causing it to 3397d46dc6cSMatthew Ahrens * be permanently read-only. 3407d46dc6cSMatthew Ahrens * 3417d46dc6cSMatthew Ahrens * See also the comments in zfs_space_check_t. 3427d46dc6cSMatthew Ahrens */ 3437d46dc6cSMatthew Ahrens int spa_slop_shift = 5; 3447d46dc6cSMatthew Ahrens 345e05725b1Sbonwick /* 346e05725b1Sbonwick * ========================================================================== 347e05725b1Sbonwick * SPA config locking 348e05725b1Sbonwick * ========================================================================== 349e05725b1Sbonwick */ 350e05725b1Sbonwick static void 351e14bb325SJeff Bonwick spa_config_lock_init(spa_t *spa) 352e14bb325SJeff Bonwick { 353e14bb325SJeff Bonwick for (int i = 0; i < SCL_LOCKS; i++) { 354e14bb325SJeff Bonwick spa_config_lock_t *scl = &spa->spa_config_lock[i]; 355e14bb325SJeff Bonwick mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL); 356e14bb325SJeff Bonwick cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL); 3573b2aab18SMatthew Ahrens refcount_create_untracked(&scl->scl_count); 358e14bb325SJeff Bonwick scl->scl_writer = NULL; 359e14bb325SJeff Bonwick scl->scl_write_wanted = 0; 360e14bb325SJeff Bonwick } 361e05725b1Sbonwick } 362e05725b1Sbonwick 363e05725b1Sbonwick static void 364e14bb325SJeff Bonwick spa_config_lock_destroy(spa_t *spa) 365e14bb325SJeff Bonwick { 366e14bb325SJeff Bonwick for (int i = 0; i < SCL_LOCKS; i++) { 367e14bb325SJeff Bonwick spa_config_lock_t *scl = &spa->spa_config_lock[i]; 368e14bb325SJeff Bonwick mutex_destroy(&scl->scl_lock); 369e14bb325SJeff Bonwick cv_destroy(&scl->scl_cv); 370e14bb325SJeff Bonwick refcount_destroy(&scl->scl_count); 371e14bb325SJeff Bonwick ASSERT(scl->scl_writer == NULL); 372e14bb325SJeff Bonwick ASSERT(scl->scl_write_wanted == 0); 373e14bb325SJeff Bonwick } 374e14bb325SJeff Bonwick } 375e14bb325SJeff Bonwick 376e14bb325SJeff Bonwick int 377e14bb325SJeff Bonwick spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw) 378e05725b1Sbonwick { 379e14bb325SJeff Bonwick for (int i = 0; i < SCL_LOCKS; i++) { 380e14bb325SJeff Bonwick spa_config_lock_t *scl = &spa->spa_config_lock[i]; 381e14bb325SJeff Bonwick if (!(locks & (1 << i))) 382e14bb325SJeff Bonwick continue; 383e14bb325SJeff Bonwick mutex_enter(&scl->scl_lock); 384e14bb325SJeff Bonwick if (rw == RW_READER) { 385e14bb325SJeff Bonwick if (scl->scl_writer || scl->scl_write_wanted) { 386e14bb325SJeff Bonwick mutex_exit(&scl->scl_lock); 387e14bb325SJeff Bonwick spa_config_exit(spa, locks ^ (1 << i), tag); 388e14bb325SJeff Bonwick return (0); 389e14bb325SJeff Bonwick } 390e14bb325SJeff Bonwick } else { 391e14bb325SJeff Bonwick ASSERT(scl->scl_writer != curthread); 392e14bb325SJeff Bonwick if (!refcount_is_zero(&scl->scl_count)) { 393e14bb325SJeff Bonwick mutex_exit(&scl->scl_lock); 394e14bb325SJeff Bonwick spa_config_exit(spa, locks ^ (1 << i), tag); 395e14bb325SJeff Bonwick return (0); 396e14bb325SJeff Bonwick } 397e14bb325SJeff Bonwick scl->scl_writer = curthread; 398e14bb325SJeff Bonwick } 399e14bb325SJeff Bonwick (void) refcount_add(&scl->scl_count, tag); 400e14bb325SJeff Bonwick mutex_exit(&scl->scl_lock); 401e14bb325SJeff Bonwick } 402e14bb325SJeff Bonwick return (1); 403e05725b1Sbonwick } 404e05725b1Sbonwick 405e05725b1Sbonwick void 406e14bb325SJeff Bonwick spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw) 407e05725b1Sbonwick { 408f64c0e34SEric Taylor int wlocks_held = 0; 409f64c0e34SEric Taylor 4103b2aab18SMatthew Ahrens ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY); 4113b2aab18SMatthew Ahrens 412e14bb325SJeff Bonwick for (int i = 0; i < SCL_LOCKS; i++) { 413e14bb325SJeff Bonwick spa_config_lock_t *scl = &spa->spa_config_lock[i]; 414f64c0e34SEric Taylor if (scl->scl_writer == curthread) 415f64c0e34SEric Taylor wlocks_held |= (1 << i); 416e14bb325SJeff Bonwick if (!(locks & (1 << i))) 417e14bb325SJeff Bonwick continue; 418e14bb325SJeff Bonwick mutex_enter(&scl->scl_lock); 419e14bb325SJeff Bonwick if (rw == RW_READER) { 420e14bb325SJeff Bonwick while (scl->scl_writer || scl->scl_write_wanted) { 421e14bb325SJeff Bonwick cv_wait(&scl->scl_cv, &scl->scl_lock); 422e14bb325SJeff Bonwick } 423e14bb325SJeff Bonwick } else { 424e14bb325SJeff Bonwick ASSERT(scl->scl_writer != curthread); 425e14bb325SJeff Bonwick while (!refcount_is_zero(&scl->scl_count)) { 426e14bb325SJeff Bonwick scl->scl_write_wanted++; 427e14bb325SJeff Bonwick cv_wait(&scl->scl_cv, &scl->scl_lock); 428e14bb325SJeff Bonwick scl->scl_write_wanted--; 429e14bb325SJeff Bonwick } 430e14bb325SJeff Bonwick scl->scl_writer = curthread; 431e14bb325SJeff Bonwick } 432e14bb325SJeff Bonwick (void) refcount_add(&scl->scl_count, tag); 433e14bb325SJeff Bonwick mutex_exit(&scl->scl_lock); 434e05725b1Sbonwick } 435f64c0e34SEric Taylor ASSERT(wlocks_held <= locks); 436e05725b1Sbonwick } 437e05725b1Sbonwick 438e05725b1Sbonwick void 439e14bb325SJeff Bonwick spa_config_exit(spa_t *spa, int locks, void *tag) 440e05725b1Sbonwick { 441e14bb325SJeff Bonwick for (int i = SCL_LOCKS - 1; i >= 0; i--) { 442e14bb325SJeff Bonwick spa_config_lock_t *scl = &spa->spa_config_lock[i]; 443e14bb325SJeff Bonwick if (!(locks & (1 << i))) 444e14bb325SJeff Bonwick continue; 445e14bb325SJeff Bonwick mutex_enter(&scl->scl_lock); 446e14bb325SJeff Bonwick ASSERT(!refcount_is_zero(&scl->scl_count)); 447e14bb325SJeff Bonwick if (refcount_remove(&scl->scl_count, tag) == 0) { 448e14bb325SJeff Bonwick ASSERT(scl->scl_writer == NULL || 449e14bb325SJeff Bonwick scl->scl_writer == curthread); 450e14bb325SJeff Bonwick scl->scl_writer = NULL; /* OK in either case */ 451e14bb325SJeff Bonwick cv_broadcast(&scl->scl_cv); 452e14bb325SJeff Bonwick } 453e14bb325SJeff Bonwick mutex_exit(&scl->scl_lock); 454e05725b1Sbonwick } 455e05725b1Sbonwick } 456e05725b1Sbonwick 457e14bb325SJeff Bonwick int 458e14bb325SJeff Bonwick spa_config_held(spa_t *spa, int locks, krw_t rw) 459e05725b1Sbonwick { 460e14bb325SJeff Bonwick int locks_held = 0; 461e05725b1Sbonwick 462e14bb325SJeff Bonwick for (int i = 0; i < SCL_LOCKS; i++) { 463e14bb325SJeff Bonwick spa_config_lock_t *scl = &spa->spa_config_lock[i]; 464e14bb325SJeff Bonwick if (!(locks & (1 << i))) 465e14bb325SJeff Bonwick continue; 466e14bb325SJeff Bonwick if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) || 467e14bb325SJeff Bonwick (rw == RW_WRITER && scl->scl_writer == curthread)) 468e14bb325SJeff Bonwick locks_held |= 1 << i; 469e14bb325SJeff Bonwick } 470e14bb325SJeff Bonwick 471e14bb325SJeff Bonwick return (locks_held); 472e05725b1Sbonwick } 473e05725b1Sbonwick 474fa9e4066Sahrens /* 475fa9e4066Sahrens * ========================================================================== 476fa9e4066Sahrens * SPA namespace functions 477fa9e4066Sahrens * ========================================================================== 478fa9e4066Sahrens */ 479fa9e4066Sahrens 480fa9e4066Sahrens /* 481fa9e4066Sahrens * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held. 482fa9e4066Sahrens * Returns NULL if no matching spa_t is found. 483fa9e4066Sahrens */ 484fa9e4066Sahrens spa_t * 485fa9e4066Sahrens spa_lookup(const char *name) 486fa9e4066Sahrens { 487e14bb325SJeff Bonwick static spa_t search; /* spa_t is large; don't allocate on stack */ 488e14bb325SJeff Bonwick spa_t *spa; 489fa9e4066Sahrens avl_index_t where; 49040feaa91Sahrens char *cp; 491fa9e4066Sahrens 492fa9e4066Sahrens ASSERT(MUTEX_HELD(&spa_namespace_lock)); 493fa9e4066Sahrens 4943b2aab18SMatthew Ahrens (void) strlcpy(search.spa_name, name, sizeof (search.spa_name)); 4953b2aab18SMatthew Ahrens 49640feaa91Sahrens /* 49740feaa91Sahrens * If it's a full dataset name, figure out the pool name and 49840feaa91Sahrens * just use that. 49940feaa91Sahrens */ 50078f17100SMatthew Ahrens cp = strpbrk(search.spa_name, "/@#"); 5013b2aab18SMatthew Ahrens if (cp != NULL) 50240feaa91Sahrens *cp = '\0'; 50340feaa91Sahrens 504fa9e4066Sahrens spa = avl_find(&spa_namespace_avl, &search, &where); 505fa9e4066Sahrens 506fa9e4066Sahrens return (spa); 507fa9e4066Sahrens } 508fa9e4066Sahrens 509283b8460SGeorge.Wilson /* 510283b8460SGeorge.Wilson * Fires when spa_sync has not completed within zfs_deadman_synctime_ms. 511283b8460SGeorge.Wilson * If the zfs_deadman_enabled flag is set then it inspects all vdev queues 512283b8460SGeorge.Wilson * looking for potentially hung I/Os. 513283b8460SGeorge.Wilson */ 514283b8460SGeorge.Wilson void 515283b8460SGeorge.Wilson spa_deadman(void *arg) 516283b8460SGeorge.Wilson { 517283b8460SGeorge.Wilson spa_t *spa = arg; 518283b8460SGeorge.Wilson 5190713e232SGeorge Wilson /* 5200713e232SGeorge Wilson * Disable the deadman timer if the pool is suspended. 5210713e232SGeorge Wilson */ 5220713e232SGeorge Wilson if (spa_suspended(spa)) { 5230713e232SGeorge Wilson VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 5240713e232SGeorge Wilson return; 5250713e232SGeorge Wilson } 5260713e232SGeorge Wilson 527283b8460SGeorge.Wilson zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu", 528283b8460SGeorge.Wilson (gethrtime() - spa->spa_sync_starttime) / NANOSEC, 529283b8460SGeorge.Wilson ++spa->spa_deadman_calls); 530283b8460SGeorge.Wilson if (zfs_deadman_enabled) 531283b8460SGeorge.Wilson vdev_deadman(spa->spa_root_vdev); 532283b8460SGeorge.Wilson } 533283b8460SGeorge.Wilson 534fa9e4066Sahrens /* 535fa9e4066Sahrens * Create an uninitialized spa_t with the given name. Requires 536fa9e4066Sahrens * spa_namespace_lock. The caller must ensure that the spa_t doesn't already 537fa9e4066Sahrens * exist by calling spa_lookup() first. 538fa9e4066Sahrens */ 539fa9e4066Sahrens spa_t * 540468c413aSTim Haley spa_add(const char *name, nvlist_t *config, const char *altroot) 541fa9e4066Sahrens { 542fa9e4066Sahrens spa_t *spa; 543c5904d13Seschrock spa_config_dirent_t *dp; 544283b8460SGeorge.Wilson cyc_handler_t hdlr; 545283b8460SGeorge.Wilson cyc_time_t when; 546fa9e4066Sahrens 547fa9e4066Sahrens ASSERT(MUTEX_HELD(&spa_namespace_lock)); 548fa9e4066Sahrens 549fa9e4066Sahrens spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP); 550fa9e4066Sahrens 551c25056deSgw mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL); 552c25056deSgw mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL); 55335a5a358SJonathan Adams mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL); 554c25056deSgw mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL); 55535a5a358SJonathan Adams mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL); 556c25056deSgw mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL); 55735a5a358SJonathan Adams mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL); 558a1521560SJeff Bonwick mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL); 559a1521560SJeff Bonwick mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL); 560c3a66015SMatthew Ahrens mutex_init(&spa->spa_iokstat_lock, NULL, MUTEX_DEFAULT, NULL); 561c25056deSgw 562c25056deSgw cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL); 56335a5a358SJonathan Adams cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL); 564c25056deSgw cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL); 565e14bb325SJeff Bonwick cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL); 566c25056deSgw 567b24ab676SJeff Bonwick for (int t = 0; t < TXG_SIZE; t++) 568cde58dbcSMatthew Ahrens bplist_create(&spa->spa_free_bplist[t]); 569b24ab676SJeff Bonwick 570e14bb325SJeff Bonwick (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name)); 571fa9e4066Sahrens spa->spa_state = POOL_STATE_UNINITIALIZED; 572fa9e4066Sahrens spa->spa_freeze_txg = UINT64_MAX; 5730373e76bSbonwick spa->spa_final_txg = UINT64_MAX; 574468c413aSTim Haley spa->spa_load_max_txg = UINT64_MAX; 57535a5a358SJonathan Adams spa->spa_proc = &p0; 57635a5a358SJonathan Adams spa->spa_proc_state = SPA_PROC_NONE; 577fa9e4066Sahrens 578283b8460SGeorge.Wilson hdlr.cyh_func = spa_deadman; 579283b8460SGeorge.Wilson hdlr.cyh_arg = spa; 580283b8460SGeorge.Wilson hdlr.cyh_level = CY_LOW_LEVEL; 581283b8460SGeorge.Wilson 58269962b56SMatthew Ahrens spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms); 583283b8460SGeorge.Wilson 584283b8460SGeorge.Wilson /* 585283b8460SGeorge.Wilson * This determines how often we need to check for hung I/Os after 586283b8460SGeorge.Wilson * the cyclic has already fired. Since checking for hung I/Os is 587283b8460SGeorge.Wilson * an expensive operation we don't want to check too frequently. 58869962b56SMatthew Ahrens * Instead wait for 5 seconds before checking again. 589283b8460SGeorge.Wilson */ 59069962b56SMatthew Ahrens when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms); 591283b8460SGeorge.Wilson when.cyt_when = CY_INFINITY; 592283b8460SGeorge.Wilson mutex_enter(&cpu_lock); 593283b8460SGeorge.Wilson spa->spa_deadman_cycid = cyclic_add(&hdlr, &when); 594283b8460SGeorge.Wilson mutex_exit(&cpu_lock); 595283b8460SGeorge.Wilson 596fa9e4066Sahrens refcount_create(&spa->spa_refcount); 597e14bb325SJeff Bonwick spa_config_lock_init(spa); 598fa9e4066Sahrens 599fa9e4066Sahrens avl_add(&spa_namespace_avl, spa); 600fa9e4066Sahrens 6010373e76bSbonwick /* 6020373e76bSbonwick * Set the alternate root, if there is one. 6030373e76bSbonwick */ 6040373e76bSbonwick if (altroot) { 6050373e76bSbonwick spa->spa_root = spa_strdup(altroot); 6060373e76bSbonwick spa_active_count++; 6070373e76bSbonwick } 6080373e76bSbonwick 609c5904d13Seschrock /* 610c5904d13Seschrock * Every pool starts with the default cachefile 611c5904d13Seschrock */ 612c5904d13Seschrock list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t), 613c5904d13Seschrock offsetof(spa_config_dirent_t, scd_link)); 614c5904d13Seschrock 615c5904d13Seschrock dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP); 616ef912c80STim Haley dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path); 617c5904d13Seschrock list_insert_head(&spa->spa_config_list, dp); 618c5904d13Seschrock 6194b964adaSGeorge Wilson VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME, 6204b964adaSGeorge Wilson KM_SLEEP) == 0); 6214b964adaSGeorge Wilson 622ad135b5dSChristopher Siden if (config != NULL) { 623ad135b5dSChristopher Siden nvlist_t *features; 624ad135b5dSChristopher Siden 625ad135b5dSChristopher Siden if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ, 626ad135b5dSChristopher Siden &features) == 0) { 627ad135b5dSChristopher Siden VERIFY(nvlist_dup(features, &spa->spa_label_features, 628ad135b5dSChristopher Siden 0) == 0); 629ad135b5dSChristopher Siden } 630ad135b5dSChristopher Siden 631468c413aSTim Haley VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0); 632ad135b5dSChristopher Siden } 633ad135b5dSChristopher Siden 634ad135b5dSChristopher Siden if (spa->spa_label_features == NULL) { 635ad135b5dSChristopher Siden VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME, 636ad135b5dSChristopher Siden KM_SLEEP) == 0); 637ad135b5dSChristopher Siden } 638468c413aSTim Haley 639c3a66015SMatthew Ahrens spa->spa_iokstat = kstat_create("zfs", 0, name, 640c3a66015SMatthew Ahrens "disk", KSTAT_TYPE_IO, 1, 0); 641c3a66015SMatthew Ahrens if (spa->spa_iokstat) { 642c3a66015SMatthew Ahrens spa->spa_iokstat->ks_lock = &spa->spa_iokstat_lock; 643c3a66015SMatthew Ahrens kstat_install(spa->spa_iokstat); 644c3a66015SMatthew Ahrens } 645c3a66015SMatthew Ahrens 6463b2aab18SMatthew Ahrens spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0); 6473b2aab18SMatthew Ahrens 64843466aaeSMax Grossman /* 64943466aaeSMax Grossman * As a pool is being created, treat all features as disabled by 65043466aaeSMax Grossman * setting SPA_FEATURE_DISABLED for all entries in the feature 65143466aaeSMax Grossman * refcount cache. 65243466aaeSMax Grossman */ 65343466aaeSMax Grossman for (int i = 0; i < SPA_FEATURES; i++) { 65443466aaeSMax Grossman spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED; 65543466aaeSMax Grossman } 65643466aaeSMax Grossman 657fa9e4066Sahrens return (spa); 658fa9e4066Sahrens } 659fa9e4066Sahrens 660fa9e4066Sahrens /* 661fa9e4066Sahrens * Removes a spa_t from the namespace, freeing up any memory used. Requires 662fa9e4066Sahrens * spa_namespace_lock. This is called only after the spa_t has been closed and 663fa9e4066Sahrens * deactivated. 664fa9e4066Sahrens */ 665fa9e4066Sahrens void 666fa9e4066Sahrens spa_remove(spa_t *spa) 667fa9e4066Sahrens { 668c5904d13Seschrock spa_config_dirent_t *dp; 669c5904d13Seschrock 670fa9e4066Sahrens ASSERT(MUTEX_HELD(&spa_namespace_lock)); 671fa9e4066Sahrens ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 672fa9e4066Sahrens 6731195e687SMark J Musante nvlist_free(spa->spa_config_splitting); 6741195e687SMark J Musante 675fa9e4066Sahrens avl_remove(&spa_namespace_avl, spa); 676fa9e4066Sahrens cv_broadcast(&spa_namespace_cv); 677fa9e4066Sahrens 6780373e76bSbonwick if (spa->spa_root) { 679fa9e4066Sahrens spa_strfree(spa->spa_root); 6800373e76bSbonwick spa_active_count--; 6810373e76bSbonwick } 682fa9e4066Sahrens 683c5904d13Seschrock while ((dp = list_head(&spa->spa_config_list)) != NULL) { 684c5904d13Seschrock list_remove(&spa->spa_config_list, dp); 685c5904d13Seschrock if (dp->scd_path != NULL) 686c5904d13Seschrock spa_strfree(dp->scd_path); 687c5904d13Seschrock kmem_free(dp, sizeof (spa_config_dirent_t)); 688c5904d13Seschrock } 689c5904d13Seschrock 690c5904d13Seschrock list_destroy(&spa->spa_config_list); 6912f8aaab3Seschrock 692ad135b5dSChristopher Siden nvlist_free(spa->spa_label_features); 6934b964adaSGeorge Wilson nvlist_free(spa->spa_load_info); 694fa9e4066Sahrens spa_config_set(spa, NULL); 695fa9e4066Sahrens 696283b8460SGeorge.Wilson mutex_enter(&cpu_lock); 697283b8460SGeorge.Wilson if (spa->spa_deadman_cycid != CYCLIC_NONE) 698283b8460SGeorge.Wilson cyclic_remove(spa->spa_deadman_cycid); 699283b8460SGeorge.Wilson mutex_exit(&cpu_lock); 700283b8460SGeorge.Wilson spa->spa_deadman_cycid = CYCLIC_NONE; 701283b8460SGeorge.Wilson 702fa9e4066Sahrens refcount_destroy(&spa->spa_refcount); 70391ebeef5Sahrens 704e14bb325SJeff Bonwick spa_config_lock_destroy(spa); 705fa9e4066Sahrens 706c3a66015SMatthew Ahrens kstat_delete(spa->spa_iokstat); 707c3a66015SMatthew Ahrens spa->spa_iokstat = NULL; 708c3a66015SMatthew Ahrens 709b24ab676SJeff Bonwick for (int t = 0; t < TXG_SIZE; t++) 710cde58dbcSMatthew Ahrens bplist_destroy(&spa->spa_free_bplist[t]); 711b24ab676SJeff Bonwick 712c25056deSgw cv_destroy(&spa->spa_async_cv); 71335a5a358SJonathan Adams cv_destroy(&spa->spa_proc_cv); 714c25056deSgw cv_destroy(&spa->spa_scrub_io_cv); 715e14bb325SJeff Bonwick cv_destroy(&spa->spa_suspend_cv); 716c25056deSgw 7175ad82045Snd mutex_destroy(&spa->spa_async_lock); 718c25056deSgw mutex_destroy(&spa->spa_errlist_lock); 71935a5a358SJonathan Adams mutex_destroy(&spa->spa_errlog_lock); 72006eeb2adSek mutex_destroy(&spa->spa_history_lock); 72135a5a358SJonathan Adams mutex_destroy(&spa->spa_proc_lock); 722b1b8ab34Slling mutex_destroy(&spa->spa_props_lock); 72335a5a358SJonathan Adams mutex_destroy(&spa->spa_scrub_lock); 724e14bb325SJeff Bonwick mutex_destroy(&spa->spa_suspend_lock); 725a1521560SJeff Bonwick mutex_destroy(&spa->spa_vdev_top_lock); 726c3a66015SMatthew Ahrens mutex_destroy(&spa->spa_iokstat_lock); 7275ad82045Snd 728fa9e4066Sahrens kmem_free(spa, sizeof (spa_t)); 729fa9e4066Sahrens } 730fa9e4066Sahrens 731fa9e4066Sahrens /* 732fa9e4066Sahrens * Given a pool, return the next pool in the namespace, or NULL if there is 733fa9e4066Sahrens * none. If 'prev' is NULL, return the first pool. 734fa9e4066Sahrens */ 735fa9e4066Sahrens spa_t * 736fa9e4066Sahrens spa_next(spa_t *prev) 737fa9e4066Sahrens { 738fa9e4066Sahrens ASSERT(MUTEX_HELD(&spa_namespace_lock)); 739fa9e4066Sahrens 740fa9e4066Sahrens if (prev) 741fa9e4066Sahrens return (AVL_NEXT(&spa_namespace_avl, prev)); 742fa9e4066Sahrens else 743fa9e4066Sahrens return (avl_first(&spa_namespace_avl)); 744fa9e4066Sahrens } 745fa9e4066Sahrens 746fa9e4066Sahrens /* 747fa9e4066Sahrens * ========================================================================== 748fa9e4066Sahrens * SPA refcount functions 749fa9e4066Sahrens * ========================================================================== 750fa9e4066Sahrens */ 751fa9e4066Sahrens 752fa9e4066Sahrens /* 753fa9e4066Sahrens * Add a reference to the given spa_t. Must have at least one reference, or 754fa9e4066Sahrens * have the namespace lock held. 755fa9e4066Sahrens */ 756fa9e4066Sahrens void 757fa9e4066Sahrens spa_open_ref(spa_t *spa, void *tag) 758fa9e4066Sahrens { 759088f3894Sahrens ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref || 760fa9e4066Sahrens MUTEX_HELD(&spa_namespace_lock)); 761fa9e4066Sahrens (void) refcount_add(&spa->spa_refcount, tag); 762fa9e4066Sahrens } 763fa9e4066Sahrens 764fa9e4066Sahrens /* 765fa9e4066Sahrens * Remove a reference to the given spa_t. Must have at least one reference, or 766fa9e4066Sahrens * have the namespace lock held. 767fa9e4066Sahrens */ 768fa9e4066Sahrens void 769fa9e4066Sahrens spa_close(spa_t *spa, void *tag) 770fa9e4066Sahrens { 771088f3894Sahrens ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref || 772fa9e4066Sahrens MUTEX_HELD(&spa_namespace_lock)); 773fa9e4066Sahrens (void) refcount_remove(&spa->spa_refcount, tag); 774fa9e4066Sahrens } 775fa9e4066Sahrens 776fa9e4066Sahrens /* 777fa9e4066Sahrens * Check to see if the spa refcount is zero. Must be called with 778088f3894Sahrens * spa_namespace_lock held. We really compare against spa_minref, which is the 779fa9e4066Sahrens * number of references acquired when opening a pool 780fa9e4066Sahrens */ 781fa9e4066Sahrens boolean_t 782fa9e4066Sahrens spa_refcount_zero(spa_t *spa) 783fa9e4066Sahrens { 784fa9e4066Sahrens ASSERT(MUTEX_HELD(&spa_namespace_lock)); 785fa9e4066Sahrens 786088f3894Sahrens return (refcount_count(&spa->spa_refcount) == spa->spa_minref); 787fa9e4066Sahrens } 788fa9e4066Sahrens 78999653d4eSeschrock /* 79099653d4eSeschrock * ========================================================================== 791fa94a07fSbrendan * SPA spare and l2cache tracking 79299653d4eSeschrock * ========================================================================== 79399653d4eSeschrock */ 79499653d4eSeschrock 795fa94a07fSbrendan /* 796fa94a07fSbrendan * Hot spares and cache devices are tracked using the same code below, 797fa94a07fSbrendan * for 'auxiliary' devices. 798fa94a07fSbrendan */ 799fa94a07fSbrendan 800fa94a07fSbrendan typedef struct spa_aux { 801fa94a07fSbrendan uint64_t aux_guid; 802fa94a07fSbrendan uint64_t aux_pool; 803fa94a07fSbrendan avl_node_t aux_avl; 804fa94a07fSbrendan int aux_count; 805fa94a07fSbrendan } spa_aux_t; 806fa94a07fSbrendan 807fa94a07fSbrendan static int 808fa94a07fSbrendan spa_aux_compare(const void *a, const void *b) 809fa94a07fSbrendan { 810fa94a07fSbrendan const spa_aux_t *sa = a; 811fa94a07fSbrendan const spa_aux_t *sb = b; 812fa94a07fSbrendan 813fa94a07fSbrendan if (sa->aux_guid < sb->aux_guid) 814fa94a07fSbrendan return (-1); 815fa94a07fSbrendan else if (sa->aux_guid > sb->aux_guid) 816fa94a07fSbrendan return (1); 817fa94a07fSbrendan else 818fa94a07fSbrendan return (0); 819fa94a07fSbrendan } 820fa94a07fSbrendan 821fa94a07fSbrendan void 822fa94a07fSbrendan spa_aux_add(vdev_t *vd, avl_tree_t *avl) 823fa94a07fSbrendan { 824fa94a07fSbrendan avl_index_t where; 825fa94a07fSbrendan spa_aux_t search; 826fa94a07fSbrendan spa_aux_t *aux; 827fa94a07fSbrendan 828fa94a07fSbrendan search.aux_guid = vd->vdev_guid; 829fa94a07fSbrendan if ((aux = avl_find(avl, &search, &where)) != NULL) { 830fa94a07fSbrendan aux->aux_count++; 831fa94a07fSbrendan } else { 832fa94a07fSbrendan aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP); 833fa94a07fSbrendan aux->aux_guid = vd->vdev_guid; 834fa94a07fSbrendan aux->aux_count = 1; 835fa94a07fSbrendan avl_insert(avl, aux, where); 836fa94a07fSbrendan } 837fa94a07fSbrendan } 838fa94a07fSbrendan 839fa94a07fSbrendan void 840fa94a07fSbrendan spa_aux_remove(vdev_t *vd, avl_tree_t *avl) 841fa94a07fSbrendan { 842fa94a07fSbrendan spa_aux_t search; 843fa94a07fSbrendan spa_aux_t *aux; 844fa94a07fSbrendan avl_index_t where; 845fa94a07fSbrendan 846fa94a07fSbrendan search.aux_guid = vd->vdev_guid; 847fa94a07fSbrendan aux = avl_find(avl, &search, &where); 848fa94a07fSbrendan 849fa94a07fSbrendan ASSERT(aux != NULL); 850fa94a07fSbrendan 851fa94a07fSbrendan if (--aux->aux_count == 0) { 852fa94a07fSbrendan avl_remove(avl, aux); 853fa94a07fSbrendan kmem_free(aux, sizeof (spa_aux_t)); 854fa94a07fSbrendan } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) { 855fa94a07fSbrendan aux->aux_pool = 0ULL; 856fa94a07fSbrendan } 857fa94a07fSbrendan } 858fa94a07fSbrendan 859fa94a07fSbrendan boolean_t 86089a89ebfSlling spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl) 861fa94a07fSbrendan { 862fa94a07fSbrendan spa_aux_t search, *found; 863fa94a07fSbrendan 864fa94a07fSbrendan search.aux_guid = guid; 86589a89ebfSlling found = avl_find(avl, &search, NULL); 866fa94a07fSbrendan 867fa94a07fSbrendan if (pool) { 868fa94a07fSbrendan if (found) 869fa94a07fSbrendan *pool = found->aux_pool; 870fa94a07fSbrendan else 871fa94a07fSbrendan *pool = 0ULL; 872fa94a07fSbrendan } 873fa94a07fSbrendan 87489a89ebfSlling if (refcnt) { 87589a89ebfSlling if (found) 87689a89ebfSlling *refcnt = found->aux_count; 87789a89ebfSlling else 87889a89ebfSlling *refcnt = 0; 87989a89ebfSlling } 88089a89ebfSlling 881fa94a07fSbrendan return (found != NULL); 882fa94a07fSbrendan } 883fa94a07fSbrendan 884fa94a07fSbrendan void 885fa94a07fSbrendan spa_aux_activate(vdev_t *vd, avl_tree_t *avl) 886fa94a07fSbrendan { 887fa94a07fSbrendan spa_aux_t search, *found; 888fa94a07fSbrendan avl_index_t where; 889fa94a07fSbrendan 890fa94a07fSbrendan search.aux_guid = vd->vdev_guid; 891fa94a07fSbrendan found = avl_find(avl, &search, &where); 892fa94a07fSbrendan ASSERT(found != NULL); 893fa94a07fSbrendan ASSERT(found->aux_pool == 0ULL); 894fa94a07fSbrendan 895fa94a07fSbrendan found->aux_pool = spa_guid(vd->vdev_spa); 896fa94a07fSbrendan } 897fa94a07fSbrendan 89899653d4eSeschrock /* 89939c23413Seschrock * Spares are tracked globally due to the following constraints: 90039c23413Seschrock * 90139c23413Seschrock * - A spare may be part of multiple pools. 90239c23413Seschrock * - A spare may be added to a pool even if it's actively in use within 90339c23413Seschrock * another pool. 90439c23413Seschrock * - A spare in use in any pool can only be the source of a replacement if 90539c23413Seschrock * the target is a spare in the same pool. 90639c23413Seschrock * 90739c23413Seschrock * We keep track of all spares on the system through the use of a reference 90839c23413Seschrock * counted AVL tree. When a vdev is added as a spare, or used as a replacement 90939c23413Seschrock * spare, then we bump the reference count in the AVL tree. In addition, we set 91039c23413Seschrock * the 'vdev_isspare' member to indicate that the device is a spare (active or 91139c23413Seschrock * inactive). When a spare is made active (used to replace a device in the 91239c23413Seschrock * pool), we also keep track of which pool its been made a part of. 91339c23413Seschrock * 91439c23413Seschrock * The 'spa_spare_lock' protects the AVL tree. These functions are normally 91539c23413Seschrock * called under the spa_namespace lock as part of vdev reconfiguration. The 91639c23413Seschrock * separate spare lock exists for the status query path, which does not need to 91739c23413Seschrock * be completely consistent with respect to other vdev configuration changes. 91899653d4eSeschrock */ 91939c23413Seschrock 92099653d4eSeschrock static int 92199653d4eSeschrock spa_spare_compare(const void *a, const void *b) 92299653d4eSeschrock { 923fa94a07fSbrendan return (spa_aux_compare(a, b)); 92499653d4eSeschrock } 92599653d4eSeschrock 92699653d4eSeschrock void 92739c23413Seschrock spa_spare_add(vdev_t *vd) 92899653d4eSeschrock { 92999653d4eSeschrock mutex_enter(&spa_spare_lock); 93039c23413Seschrock ASSERT(!vd->vdev_isspare); 931fa94a07fSbrendan spa_aux_add(vd, &spa_spare_avl); 93239c23413Seschrock vd->vdev_isspare = B_TRUE; 93399653d4eSeschrock mutex_exit(&spa_spare_lock); 93499653d4eSeschrock } 93599653d4eSeschrock 93699653d4eSeschrock void 93739c23413Seschrock spa_spare_remove(vdev_t *vd) 93899653d4eSeschrock { 93999653d4eSeschrock mutex_enter(&spa_spare_lock); 94039c23413Seschrock ASSERT(vd->vdev_isspare); 941fa94a07fSbrendan spa_aux_remove(vd, &spa_spare_avl); 94239c23413Seschrock vd->vdev_isspare = B_FALSE; 94399653d4eSeschrock mutex_exit(&spa_spare_lock); 94499653d4eSeschrock } 94599653d4eSeschrock 94699653d4eSeschrock boolean_t 94789a89ebfSlling spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt) 94899653d4eSeschrock { 949fa94a07fSbrendan boolean_t found; 95099653d4eSeschrock 95199653d4eSeschrock mutex_enter(&spa_spare_lock); 95289a89ebfSlling found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl); 95399653d4eSeschrock mutex_exit(&spa_spare_lock); 95499653d4eSeschrock 955fa94a07fSbrendan return (found); 95639c23413Seschrock } 95739c23413Seschrock 95839c23413Seschrock void 95939c23413Seschrock spa_spare_activate(vdev_t *vd) 96039c23413Seschrock { 96139c23413Seschrock mutex_enter(&spa_spare_lock); 96239c23413Seschrock ASSERT(vd->vdev_isspare); 963fa94a07fSbrendan spa_aux_activate(vd, &spa_spare_avl); 964fa94a07fSbrendan mutex_exit(&spa_spare_lock); 965fa94a07fSbrendan } 96639c23413Seschrock 967fa94a07fSbrendan /* 968fa94a07fSbrendan * Level 2 ARC devices are tracked globally for the same reasons as spares. 969fa94a07fSbrendan * Cache devices currently only support one pool per cache device, and so 970fa94a07fSbrendan * for these devices the aux reference count is currently unused beyond 1. 971fa94a07fSbrendan */ 97239c23413Seschrock 973fa94a07fSbrendan static int 974fa94a07fSbrendan spa_l2cache_compare(const void *a, const void *b) 975fa94a07fSbrendan { 976fa94a07fSbrendan return (spa_aux_compare(a, b)); 977fa94a07fSbrendan } 978fa94a07fSbrendan 979fa94a07fSbrendan void 980fa94a07fSbrendan spa_l2cache_add(vdev_t *vd) 981fa94a07fSbrendan { 982fa94a07fSbrendan mutex_enter(&spa_l2cache_lock); 983fa94a07fSbrendan ASSERT(!vd->vdev_isl2cache); 984fa94a07fSbrendan spa_aux_add(vd, &spa_l2cache_avl); 985fa94a07fSbrendan vd->vdev_isl2cache = B_TRUE; 986fa94a07fSbrendan mutex_exit(&spa_l2cache_lock); 987fa94a07fSbrendan } 988fa94a07fSbrendan 989fa94a07fSbrendan void 990fa94a07fSbrendan spa_l2cache_remove(vdev_t *vd) 991fa94a07fSbrendan { 992fa94a07fSbrendan mutex_enter(&spa_l2cache_lock); 993fa94a07fSbrendan ASSERT(vd->vdev_isl2cache); 994fa94a07fSbrendan spa_aux_remove(vd, &spa_l2cache_avl); 995fa94a07fSbrendan vd->vdev_isl2cache = B_FALSE; 996fa94a07fSbrendan mutex_exit(&spa_l2cache_lock); 997fa94a07fSbrendan } 998fa94a07fSbrendan 999fa94a07fSbrendan boolean_t 1000fa94a07fSbrendan spa_l2cache_exists(uint64_t guid, uint64_t *pool) 1001fa94a07fSbrendan { 1002fa94a07fSbrendan boolean_t found; 1003fa94a07fSbrendan 1004fa94a07fSbrendan mutex_enter(&spa_l2cache_lock); 100589a89ebfSlling found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl); 1006fa94a07fSbrendan mutex_exit(&spa_l2cache_lock); 1007fa94a07fSbrendan 1008fa94a07fSbrendan return (found); 1009fa94a07fSbrendan } 1010fa94a07fSbrendan 1011fa94a07fSbrendan void 1012fa94a07fSbrendan spa_l2cache_activate(vdev_t *vd) 1013fa94a07fSbrendan { 1014fa94a07fSbrendan mutex_enter(&spa_l2cache_lock); 1015fa94a07fSbrendan ASSERT(vd->vdev_isl2cache); 1016fa94a07fSbrendan spa_aux_activate(vd, &spa_l2cache_avl); 1017fa94a07fSbrendan mutex_exit(&spa_l2cache_lock); 1018fa94a07fSbrendan } 1019fa94a07fSbrendan 1020fa9e4066Sahrens /* 1021fa9e4066Sahrens * ========================================================================== 1022fa9e4066Sahrens * SPA vdev locking 1023fa9e4066Sahrens * ========================================================================== 1024fa9e4066Sahrens */ 1025fa9e4066Sahrens 1026fa9e4066Sahrens /* 1027ea8dc4b6Seschrock * Lock the given spa_t for the purpose of adding or removing a vdev. 1028ea8dc4b6Seschrock * Grabs the global spa_namespace_lock plus the spa config lock for writing. 1029fa9e4066Sahrens * It returns the next transaction group for the spa_t. 1030fa9e4066Sahrens */ 1031fa9e4066Sahrens uint64_t 1032fa9e4066Sahrens spa_vdev_enter(spa_t *spa) 1033fa9e4066Sahrens { 1034a1521560SJeff Bonwick mutex_enter(&spa->spa_vdev_top_lock); 1035bbfd46c4SJeff Bonwick mutex_enter(&spa_namespace_lock); 103688ecc943SGeorge Wilson return (spa_vdev_config_enter(spa)); 103788ecc943SGeorge Wilson } 103888ecc943SGeorge Wilson 103988ecc943SGeorge Wilson /* 104088ecc943SGeorge Wilson * Internal implementation for spa_vdev_enter(). Used when a vdev 104188ecc943SGeorge Wilson * operation requires multiple syncs (i.e. removing a device) while 104288ecc943SGeorge Wilson * keeping the spa_namespace_lock held. 104388ecc943SGeorge Wilson */ 104488ecc943SGeorge Wilson uint64_t 104588ecc943SGeorge Wilson spa_vdev_config_enter(spa_t *spa) 104688ecc943SGeorge Wilson { 104788ecc943SGeorge Wilson ASSERT(MUTEX_HELD(&spa_namespace_lock)); 10483d7072f8Seschrock 1049e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); 1050fa9e4066Sahrens 1051fa9e4066Sahrens return (spa_last_synced_txg(spa) + 1); 1052fa9e4066Sahrens } 1053fa9e4066Sahrens 1054fa9e4066Sahrens /* 105588ecc943SGeorge Wilson * Used in combination with spa_vdev_config_enter() to allow the syncing 105688ecc943SGeorge Wilson * of multiple transactions without releasing the spa_namespace_lock. 1057fa9e4066Sahrens */ 105888ecc943SGeorge Wilson void 105988ecc943SGeorge Wilson spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag) 1060fa9e4066Sahrens { 106188ecc943SGeorge Wilson ASSERT(MUTEX_HELD(&spa_namespace_lock)); 106288ecc943SGeorge Wilson 10630e34b6a7Sbonwick int config_changed = B_FALSE; 1064ea8dc4b6Seschrock 10650373e76bSbonwick ASSERT(txg > spa_last_synced_txg(spa)); 10660e34b6a7Sbonwick 1067e14bb325SJeff Bonwick spa->spa_pending_vdev = NULL; 1068e14bb325SJeff Bonwick 10690e34b6a7Sbonwick /* 10700e34b6a7Sbonwick * Reassess the DTLs. 10710e34b6a7Sbonwick */ 10720373e76bSbonwick vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE); 10730e34b6a7Sbonwick 1074e14bb325SJeff Bonwick if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) { 10750e34b6a7Sbonwick config_changed = B_TRUE; 10768f18d1faSGeorge Wilson spa->spa_config_generation++; 10770e34b6a7Sbonwick } 1078ea8dc4b6Seschrock 107988ecc943SGeorge Wilson /* 108088ecc943SGeorge Wilson * Verify the metaslab classes. 108188ecc943SGeorge Wilson */ 1082b24ab676SJeff Bonwick ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0); 1083b24ab676SJeff Bonwick ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0); 108488ecc943SGeorge Wilson 1085e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, spa); 1086fa9e4066Sahrens 108788ecc943SGeorge Wilson /* 108888ecc943SGeorge Wilson * Panic the system if the specified tag requires it. This 108988ecc943SGeorge Wilson * is useful for ensuring that configurations are updated 109088ecc943SGeorge Wilson * transactionally. 109188ecc943SGeorge Wilson */ 109288ecc943SGeorge Wilson if (zio_injection_enabled) 10931195e687SMark J Musante zio_handle_panic_injection(spa, tag, 0); 109488ecc943SGeorge Wilson 1095fa9e4066Sahrens /* 1096fa9e4066Sahrens * Note: this txg_wait_synced() is important because it ensures 1097fa9e4066Sahrens * that there won't be more than one config change per txg. 1098fa9e4066Sahrens * This allows us to use the txg as the generation number. 1099fa9e4066Sahrens */ 1100fa9e4066Sahrens if (error == 0) 1101fa9e4066Sahrens txg_wait_synced(spa->spa_dsl_pool, txg); 1102fa9e4066Sahrens 1103fa9e4066Sahrens if (vd != NULL) { 11040713e232SGeorge Wilson ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL); 11058ad4d6ddSJeff Bonwick spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); 1106fa9e4066Sahrens vdev_free(vd); 11078ad4d6ddSJeff Bonwick spa_config_exit(spa, SCL_ALL, spa); 1108fa9e4066Sahrens } 1109fa9e4066Sahrens 1110fa9e4066Sahrens /* 11110e34b6a7Sbonwick * If the config changed, update the config cache. 1112fa9e4066Sahrens */ 11130e34b6a7Sbonwick if (config_changed) 1114c5904d13Seschrock spa_config_sync(spa, B_FALSE, B_TRUE); 111588ecc943SGeorge Wilson } 1116ea8dc4b6Seschrock 111788ecc943SGeorge Wilson /* 111888ecc943SGeorge Wilson * Unlock the spa_t after adding or removing a vdev. Besides undoing the 111988ecc943SGeorge Wilson * locking of spa_vdev_enter(), we also want make sure the transactions have 112088ecc943SGeorge Wilson * synced to disk, and then update the global configuration cache with the new 112188ecc943SGeorge Wilson * information. 112288ecc943SGeorge Wilson */ 112388ecc943SGeorge Wilson int 112488ecc943SGeorge Wilson spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error) 112588ecc943SGeorge Wilson { 112688ecc943SGeorge Wilson spa_vdev_config_exit(spa, vd, txg, error, FTAG); 1127ea8dc4b6Seschrock mutex_exit(&spa_namespace_lock); 1128bbfd46c4SJeff Bonwick mutex_exit(&spa->spa_vdev_top_lock); 1129fa9e4066Sahrens 1130fa9e4066Sahrens return (error); 1131fa9e4066Sahrens } 1132fa9e4066Sahrens 1133e14bb325SJeff Bonwick /* 1134e14bb325SJeff Bonwick * Lock the given spa_t for the purpose of changing vdev state. 1135e14bb325SJeff Bonwick */ 1136e14bb325SJeff Bonwick void 11378f18d1faSGeorge Wilson spa_vdev_state_enter(spa_t *spa, int oplocks) 1138e14bb325SJeff Bonwick { 11398f18d1faSGeorge Wilson int locks = SCL_STATE_ALL | oplocks; 11408f18d1faSGeorge Wilson 1141dcba9f3fSGeorge Wilson /* 1142dcba9f3fSGeorge Wilson * Root pools may need to read of the underlying devfs filesystem 1143dcba9f3fSGeorge Wilson * when opening up a vdev. Unfortunately if we're holding the 1144dcba9f3fSGeorge Wilson * SCL_ZIO lock it will result in a deadlock when we try to issue 1145dcba9f3fSGeorge Wilson * the read from the root filesystem. Instead we "prefetch" 1146dcba9f3fSGeorge Wilson * the associated vnodes that we need prior to opening the 1147dcba9f3fSGeorge Wilson * underlying devices and cache them so that we can prevent 1148dcba9f3fSGeorge Wilson * any I/O when we are doing the actual open. 1149dcba9f3fSGeorge Wilson */ 1150dcba9f3fSGeorge Wilson if (spa_is_root(spa)) { 11519842588bSGeorge Wilson int low = locks & ~(SCL_ZIO - 1); 11529842588bSGeorge Wilson int high = locks & ~low; 11539842588bSGeorge Wilson 11549842588bSGeorge Wilson spa_config_enter(spa, high, spa, RW_WRITER); 1155dcba9f3fSGeorge Wilson vdev_hold(spa->spa_root_vdev); 11569842588bSGeorge Wilson spa_config_enter(spa, low, spa, RW_WRITER); 1157dcba9f3fSGeorge Wilson } else { 1158dcba9f3fSGeorge Wilson spa_config_enter(spa, locks, spa, RW_WRITER); 1159dcba9f3fSGeorge Wilson } 11608f18d1faSGeorge Wilson spa->spa_vdev_locks = locks; 1161e14bb325SJeff Bonwick } 1162e14bb325SJeff Bonwick 1163e14bb325SJeff Bonwick int 1164e14bb325SJeff Bonwick spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error) 1165e14bb325SJeff Bonwick { 1166c6065d0fSGeorge Wilson boolean_t config_changed = B_FALSE; 1167c6065d0fSGeorge Wilson 1168b24ab676SJeff Bonwick if (vd != NULL || error == 0) 1169b24ab676SJeff Bonwick vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev, 1170b24ab676SJeff Bonwick 0, 0, B_FALSE); 1171b24ab676SJeff Bonwick 11728f18d1faSGeorge Wilson if (vd != NULL) { 1173e14bb325SJeff Bonwick vdev_state_dirty(vd->vdev_top); 1174c6065d0fSGeorge Wilson config_changed = B_TRUE; 11758f18d1faSGeorge Wilson spa->spa_config_generation++; 11768f18d1faSGeorge Wilson } 1177e14bb325SJeff Bonwick 1178dcba9f3fSGeorge Wilson if (spa_is_root(spa)) 1179dcba9f3fSGeorge Wilson vdev_rele(spa->spa_root_vdev); 1180dcba9f3fSGeorge Wilson 11818f18d1faSGeorge Wilson ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL); 11828f18d1faSGeorge Wilson spa_config_exit(spa, spa->spa_vdev_locks, spa); 1183e14bb325SJeff Bonwick 11848ad4d6ddSJeff Bonwick /* 11858ad4d6ddSJeff Bonwick * If anything changed, wait for it to sync. This ensures that, 11868ad4d6ddSJeff Bonwick * from the system administrator's perspective, zpool(1M) commands 11878ad4d6ddSJeff Bonwick * are synchronous. This is important for things like zpool offline: 11888ad4d6ddSJeff Bonwick * when the command completes, you expect no further I/O from ZFS. 11898ad4d6ddSJeff Bonwick */ 11908ad4d6ddSJeff Bonwick if (vd != NULL) 11918ad4d6ddSJeff Bonwick txg_wait_synced(spa->spa_dsl_pool, 0); 11928ad4d6ddSJeff Bonwick 1193c6065d0fSGeorge Wilson /* 1194c6065d0fSGeorge Wilson * If the config changed, update the config cache. 1195c6065d0fSGeorge Wilson */ 1196c6065d0fSGeorge Wilson if (config_changed) { 1197c6065d0fSGeorge Wilson mutex_enter(&spa_namespace_lock); 1198c6065d0fSGeorge Wilson spa_config_sync(spa, B_FALSE, B_TRUE); 1199c6065d0fSGeorge Wilson mutex_exit(&spa_namespace_lock); 1200c6065d0fSGeorge Wilson } 1201c6065d0fSGeorge Wilson 1202e14bb325SJeff Bonwick return (error); 1203e14bb325SJeff Bonwick } 1204e14bb325SJeff Bonwick 1205fa9e4066Sahrens /* 1206fa9e4066Sahrens * ========================================================================== 1207fa9e4066Sahrens * Miscellaneous functions 1208fa9e4066Sahrens * ========================================================================== 1209fa9e4066Sahrens */ 1210fa9e4066Sahrens 1211ad135b5dSChristopher Siden void 121243466aaeSMax Grossman spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx) 1213ad135b5dSChristopher Siden { 12142acef22dSMatthew Ahrens if (!nvlist_exists(spa->spa_label_features, feature)) { 12152acef22dSMatthew Ahrens fnvlist_add_boolean(spa->spa_label_features, feature); 121643466aaeSMax Grossman /* 121743466aaeSMax Grossman * When we are creating the pool (tx_txg==TXG_INITIAL), we can't 121843466aaeSMax Grossman * dirty the vdev config because lock SCL_CONFIG is not held. 121943466aaeSMax Grossman * Thankfully, in this case we don't need to dirty the config 122043466aaeSMax Grossman * because it will be written out anyway when we finish 122143466aaeSMax Grossman * creating the pool. 122243466aaeSMax Grossman */ 122343466aaeSMax Grossman if (tx->tx_txg != TXG_INITIAL) 122443466aaeSMax Grossman vdev_config_dirty(spa->spa_root_vdev); 12252acef22dSMatthew Ahrens } 1226ad135b5dSChristopher Siden } 1227ad135b5dSChristopher Siden 1228ad135b5dSChristopher Siden void 1229ad135b5dSChristopher Siden spa_deactivate_mos_feature(spa_t *spa, const char *feature) 1230ad135b5dSChristopher Siden { 12312acef22dSMatthew Ahrens if (nvlist_remove_all(spa->spa_label_features, feature) == 0) 12322acef22dSMatthew Ahrens vdev_config_dirty(spa->spa_root_vdev); 1233ad135b5dSChristopher Siden } 1234ad135b5dSChristopher Siden 1235fa9e4066Sahrens /* 1236fa9e4066Sahrens * Rename a spa_t. 1237fa9e4066Sahrens */ 1238fa9e4066Sahrens int 1239fa9e4066Sahrens spa_rename(const char *name, const char *newname) 1240fa9e4066Sahrens { 1241fa9e4066Sahrens spa_t *spa; 1242fa9e4066Sahrens int err; 1243fa9e4066Sahrens 1244fa9e4066Sahrens /* 1245fa9e4066Sahrens * Lookup the spa_t and grab the config lock for writing. We need to 1246fa9e4066Sahrens * actually open the pool so that we can sync out the necessary labels. 1247fa9e4066Sahrens * It's OK to call spa_open() with the namespace lock held because we 1248ea8dc4b6Seschrock * allow recursive calls for other reasons. 1249fa9e4066Sahrens */ 1250fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 1251fa9e4066Sahrens if ((err = spa_open(name, &spa, FTAG)) != 0) { 1252fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 1253fa9e4066Sahrens return (err); 1254fa9e4066Sahrens } 1255fa9e4066Sahrens 1256e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1257fa9e4066Sahrens 1258fa9e4066Sahrens avl_remove(&spa_namespace_avl, spa); 1259e14bb325SJeff Bonwick (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name)); 1260fa9e4066Sahrens avl_add(&spa_namespace_avl, spa); 1261fa9e4066Sahrens 1262fa9e4066Sahrens /* 1263fa9e4066Sahrens * Sync all labels to disk with the new names by marking the root vdev 1264fa9e4066Sahrens * dirty and waiting for it to sync. It will pick up the new pool name 1265fa9e4066Sahrens * during the sync. 1266fa9e4066Sahrens */ 1267fa9e4066Sahrens vdev_config_dirty(spa->spa_root_vdev); 1268fa9e4066Sahrens 1269e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 1270fa9e4066Sahrens 12710373e76bSbonwick txg_wait_synced(spa->spa_dsl_pool, 0); 1272fa9e4066Sahrens 1273fa9e4066Sahrens /* 1274fa9e4066Sahrens * Sync the updated config cache. 1275fa9e4066Sahrens */ 1276c5904d13Seschrock spa_config_sync(spa, B_FALSE, B_TRUE); 1277fa9e4066Sahrens 1278fa9e4066Sahrens spa_close(spa, FTAG); 1279fa9e4066Sahrens 1280fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 1281fa9e4066Sahrens 1282fa9e4066Sahrens return (0); 1283fa9e4066Sahrens } 1284fa9e4066Sahrens 1285fa9e4066Sahrens /* 1286f9af39baSGeorge Wilson * Return the spa_t associated with given pool_guid, if it exists. If 1287f9af39baSGeorge Wilson * device_guid is non-zero, determine whether the pool exists *and* contains 1288f9af39baSGeorge Wilson * a device with the specified device_guid. 1289fa9e4066Sahrens */ 1290f9af39baSGeorge Wilson spa_t * 1291f9af39baSGeorge Wilson spa_by_guid(uint64_t pool_guid, uint64_t device_guid) 1292fa9e4066Sahrens { 1293fa9e4066Sahrens spa_t *spa; 1294fa9e4066Sahrens avl_tree_t *t = &spa_namespace_avl; 1295fa9e4066Sahrens 1296ea8dc4b6Seschrock ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1297fa9e4066Sahrens 1298fa9e4066Sahrens for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) { 1299fa9e4066Sahrens if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1300fa9e4066Sahrens continue; 1301fa9e4066Sahrens if (spa->spa_root_vdev == NULL) 1302fa9e4066Sahrens continue; 130339c23413Seschrock if (spa_guid(spa) == pool_guid) { 130439c23413Seschrock if (device_guid == 0) 130539c23413Seschrock break; 130639c23413Seschrock 130739c23413Seschrock if (vdev_lookup_by_guid(spa->spa_root_vdev, 130839c23413Seschrock device_guid) != NULL) 130939c23413Seschrock break; 131039c23413Seschrock 131139c23413Seschrock /* 13128654d025Sperrin * Check any devices we may be in the process of adding. 131339c23413Seschrock */ 131439c23413Seschrock if (spa->spa_pending_vdev) { 131539c23413Seschrock if (vdev_lookup_by_guid(spa->spa_pending_vdev, 131639c23413Seschrock device_guid) != NULL) 131739c23413Seschrock break; 131839c23413Seschrock } 131939c23413Seschrock } 1320fa9e4066Sahrens } 1321fa9e4066Sahrens 1322f9af39baSGeorge Wilson return (spa); 1323f9af39baSGeorge Wilson } 1324f9af39baSGeorge Wilson 1325f9af39baSGeorge Wilson /* 1326f9af39baSGeorge Wilson * Determine whether a pool with the given pool_guid exists. 1327f9af39baSGeorge Wilson */ 1328f9af39baSGeorge Wilson boolean_t 1329f9af39baSGeorge Wilson spa_guid_exists(uint64_t pool_guid, uint64_t device_guid) 1330f9af39baSGeorge Wilson { 1331f9af39baSGeorge Wilson return (spa_by_guid(pool_guid, device_guid) != NULL); 1332fa9e4066Sahrens } 1333fa9e4066Sahrens 1334fa9e4066Sahrens char * 1335fa9e4066Sahrens spa_strdup(const char *s) 1336fa9e4066Sahrens { 1337fa9e4066Sahrens size_t len; 1338fa9e4066Sahrens char *new; 1339fa9e4066Sahrens 1340fa9e4066Sahrens len = strlen(s); 1341fa9e4066Sahrens new = kmem_alloc(len + 1, KM_SLEEP); 1342fa9e4066Sahrens bcopy(s, new, len); 1343fa9e4066Sahrens new[len] = '\0'; 1344fa9e4066Sahrens 1345fa9e4066Sahrens return (new); 1346fa9e4066Sahrens } 1347fa9e4066Sahrens 1348fa9e4066Sahrens void 1349fa9e4066Sahrens spa_strfree(char *s) 1350fa9e4066Sahrens { 1351fa9e4066Sahrens kmem_free(s, strlen(s) + 1); 1352fa9e4066Sahrens } 1353fa9e4066Sahrens 1354fa9e4066Sahrens uint64_t 1355fa9e4066Sahrens spa_get_random(uint64_t range) 1356fa9e4066Sahrens { 1357fa9e4066Sahrens uint64_t r; 1358fa9e4066Sahrens 1359fa9e4066Sahrens ASSERT(range != 0); 1360fa9e4066Sahrens 1361fa9e4066Sahrens (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t)); 1362fa9e4066Sahrens 1363fa9e4066Sahrens return (r % range); 1364fa9e4066Sahrens } 1365fa9e4066Sahrens 13661195e687SMark J Musante uint64_t 13671195e687SMark J Musante spa_generate_guid(spa_t *spa) 13681195e687SMark J Musante { 13691195e687SMark J Musante uint64_t guid = spa_get_random(-1ULL); 13701195e687SMark J Musante 13711195e687SMark J Musante if (spa != NULL) { 13721195e687SMark J Musante while (guid == 0 || spa_guid_exists(spa_guid(spa), guid)) 13731195e687SMark J Musante guid = spa_get_random(-1ULL); 13741195e687SMark J Musante } else { 13751195e687SMark J Musante while (guid == 0 || spa_guid_exists(guid, 0)) 13761195e687SMark J Musante guid = spa_get_random(-1ULL); 13771195e687SMark J Musante } 13781195e687SMark J Musante 13791195e687SMark J Musante return (guid); 13801195e687SMark J Musante } 13811195e687SMark J Musante 1382fa9e4066Sahrens void 138343466aaeSMax Grossman snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp) 1384fa9e4066Sahrens { 1385ad135b5dSChristopher Siden char type[256]; 1386f0ba89beSJeff Bonwick char *checksum = NULL; 1387f0ba89beSJeff Bonwick char *compress = NULL; 1388f0ba89beSJeff Bonwick 1389f0ba89beSJeff Bonwick if (bp != NULL) { 1390ad135b5dSChristopher Siden if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) { 1391ad135b5dSChristopher Siden dmu_object_byteswap_t bswap = 1392ad135b5dSChristopher Siden DMU_OT_BYTESWAP(BP_GET_TYPE(bp)); 1393ad135b5dSChristopher Siden (void) snprintf(type, sizeof (type), "bswap %s %s", 1394ad135b5dSChristopher Siden DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ? 1395ad135b5dSChristopher Siden "metadata" : "data", 1396ad135b5dSChristopher Siden dmu_ot_byteswap[bswap].ob_name); 1397ad135b5dSChristopher Siden } else { 1398ad135b5dSChristopher Siden (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name, 1399ad135b5dSChristopher Siden sizeof (type)); 1400ad135b5dSChristopher Siden } 14015d7b4d43SMatthew Ahrens if (!BP_IS_EMBEDDED(bp)) { 14025d7b4d43SMatthew Ahrens checksum = 14035d7b4d43SMatthew Ahrens zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name; 14045d7b4d43SMatthew Ahrens } 1405f0ba89beSJeff Bonwick compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name; 1406f0ba89beSJeff Bonwick } 1407fa9e4066Sahrens 140843466aaeSMax Grossman SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum, 140943466aaeSMax Grossman compress); 1410fa9e4066Sahrens } 1411fa9e4066Sahrens 1412fa9e4066Sahrens void 1413fa9e4066Sahrens spa_freeze(spa_t *spa) 1414fa9e4066Sahrens { 1415fa9e4066Sahrens uint64_t freeze_txg = 0; 1416fa9e4066Sahrens 1417e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1418fa9e4066Sahrens if (spa->spa_freeze_txg == UINT64_MAX) { 1419fa9e4066Sahrens freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE; 1420fa9e4066Sahrens spa->spa_freeze_txg = freeze_txg; 1421fa9e4066Sahrens } 1422e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 1423fa9e4066Sahrens if (freeze_txg != 0) 1424fa9e4066Sahrens txg_wait_synced(spa_get_dsl(spa), freeze_txg); 1425fa9e4066Sahrens } 1426fa9e4066Sahrens 14270125049cSahrens void 14280125049cSahrens zfs_panic_recover(const char *fmt, ...) 14290125049cSahrens { 14300125049cSahrens va_list adx; 14310125049cSahrens 14320125049cSahrens va_start(adx, fmt); 14330125049cSahrens vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx); 14340125049cSahrens va_end(adx); 14350125049cSahrens } 14360125049cSahrens 14373f9d6ad7SLin Ling /* 14383f9d6ad7SLin Ling * This is a stripped-down version of strtoull, suitable only for converting 1439f7170741SWill Andrews * lowercase hexadecimal numbers that don't overflow. 14403f9d6ad7SLin Ling */ 14413f9d6ad7SLin Ling uint64_t 14423f9d6ad7SLin Ling strtonum(const char *str, char **nptr) 14433f9d6ad7SLin Ling { 14443f9d6ad7SLin Ling uint64_t val = 0; 14453f9d6ad7SLin Ling char c; 14463f9d6ad7SLin Ling int digit; 14473f9d6ad7SLin Ling 14483f9d6ad7SLin Ling while ((c = *str) != '\0') { 14493f9d6ad7SLin Ling if (c >= '0' && c <= '9') 14503f9d6ad7SLin Ling digit = c - '0'; 14513f9d6ad7SLin Ling else if (c >= 'a' && c <= 'f') 14523f9d6ad7SLin Ling digit = 10 + c - 'a'; 14533f9d6ad7SLin Ling else 14543f9d6ad7SLin Ling break; 14553f9d6ad7SLin Ling 14563f9d6ad7SLin Ling val *= 16; 14573f9d6ad7SLin Ling val += digit; 14583f9d6ad7SLin Ling 14593f9d6ad7SLin Ling str++; 14603f9d6ad7SLin Ling } 14613f9d6ad7SLin Ling 14623f9d6ad7SLin Ling if (nptr) 14633f9d6ad7SLin Ling *nptr = (char *)str; 14643f9d6ad7SLin Ling 14653f9d6ad7SLin Ling return (val); 14663f9d6ad7SLin Ling } 14673f9d6ad7SLin Ling 1468fa9e4066Sahrens /* 1469fa9e4066Sahrens * ========================================================================== 1470fa9e4066Sahrens * Accessor functions 1471fa9e4066Sahrens * ========================================================================== 1472fa9e4066Sahrens */ 1473fa9e4066Sahrens 1474088f3894Sahrens boolean_t 147588b7b0f2SMatthew Ahrens spa_shutting_down(spa_t *spa) 1476fa9e4066Sahrens { 147788b7b0f2SMatthew Ahrens return (spa->spa_async_suspended); 1478fa9e4066Sahrens } 1479fa9e4066Sahrens 1480fa9e4066Sahrens dsl_pool_t * 1481fa9e4066Sahrens spa_get_dsl(spa_t *spa) 1482fa9e4066Sahrens { 1483fa9e4066Sahrens return (spa->spa_dsl_pool); 1484fa9e4066Sahrens } 1485fa9e4066Sahrens 1486ad135b5dSChristopher Siden boolean_t 1487ad135b5dSChristopher Siden spa_is_initializing(spa_t *spa) 1488ad135b5dSChristopher Siden { 1489ad135b5dSChristopher Siden return (spa->spa_is_initializing); 1490ad135b5dSChristopher Siden } 1491ad135b5dSChristopher Siden 1492fa9e4066Sahrens blkptr_t * 1493fa9e4066Sahrens spa_get_rootblkptr(spa_t *spa) 1494fa9e4066Sahrens { 1495fa9e4066Sahrens return (&spa->spa_ubsync.ub_rootbp); 1496fa9e4066Sahrens } 1497fa9e4066Sahrens 1498fa9e4066Sahrens void 1499fa9e4066Sahrens spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp) 1500fa9e4066Sahrens { 1501fa9e4066Sahrens spa->spa_uberblock.ub_rootbp = *bp; 1502fa9e4066Sahrens } 1503fa9e4066Sahrens 1504fa9e4066Sahrens void 1505fa9e4066Sahrens spa_altroot(spa_t *spa, char *buf, size_t buflen) 1506fa9e4066Sahrens { 1507fa9e4066Sahrens if (spa->spa_root == NULL) 1508fa9e4066Sahrens buf[0] = '\0'; 1509fa9e4066Sahrens else 1510fa9e4066Sahrens (void) strncpy(buf, spa->spa_root, buflen); 1511fa9e4066Sahrens } 1512fa9e4066Sahrens 1513fa9e4066Sahrens int 1514fa9e4066Sahrens spa_sync_pass(spa_t *spa) 1515fa9e4066Sahrens { 1516fa9e4066Sahrens return (spa->spa_sync_pass); 1517fa9e4066Sahrens } 1518fa9e4066Sahrens 1519fa9e4066Sahrens char * 1520fa9e4066Sahrens spa_name(spa_t *spa) 1521fa9e4066Sahrens { 1522fa9e4066Sahrens return (spa->spa_name); 1523fa9e4066Sahrens } 1524fa9e4066Sahrens 1525fa9e4066Sahrens uint64_t 1526fa9e4066Sahrens spa_guid(spa_t *spa) 1527fa9e4066Sahrens { 1528dfbb9432SGeorge Wilson dsl_pool_t *dp = spa_get_dsl(spa); 1529dfbb9432SGeorge Wilson uint64_t guid; 1530dfbb9432SGeorge Wilson 1531b5989ec7Seschrock /* 1532b5989ec7Seschrock * If we fail to parse the config during spa_load(), we can go through 1533b5989ec7Seschrock * the error path (which posts an ereport) and end up here with no root 1534e9103aaeSGarrett D'Amore * vdev. We stash the original pool guid in 'spa_config_guid' to handle 1535b5989ec7Seschrock * this case. 1536b5989ec7Seschrock */ 1537dfbb9432SGeorge Wilson if (spa->spa_root_vdev == NULL) 1538dfbb9432SGeorge Wilson return (spa->spa_config_guid); 1539dfbb9432SGeorge Wilson 1540dfbb9432SGeorge Wilson guid = spa->spa_last_synced_guid != 0 ? 1541dfbb9432SGeorge Wilson spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid; 1542dfbb9432SGeorge Wilson 1543dfbb9432SGeorge Wilson /* 1544dfbb9432SGeorge Wilson * Return the most recently synced out guid unless we're 1545dfbb9432SGeorge Wilson * in syncing context. 1546dfbb9432SGeorge Wilson */ 1547dfbb9432SGeorge Wilson if (dp && dsl_pool_sync_context(dp)) 1548b5989ec7Seschrock return (spa->spa_root_vdev->vdev_guid); 1549b5989ec7Seschrock else 1550dfbb9432SGeorge Wilson return (guid); 1551e9103aaeSGarrett D'Amore } 1552e9103aaeSGarrett D'Amore 1553e9103aaeSGarrett D'Amore uint64_t 1554e9103aaeSGarrett D'Amore spa_load_guid(spa_t *spa) 1555e9103aaeSGarrett D'Amore { 1556e9103aaeSGarrett D'Amore /* 1557e9103aaeSGarrett D'Amore * This is a GUID that exists solely as a reference for the 1558e9103aaeSGarrett D'Amore * purposes of the arc. It is generated at load time, and 1559e9103aaeSGarrett D'Amore * is never written to persistent storage. 1560e9103aaeSGarrett D'Amore */ 1561e9103aaeSGarrett D'Amore return (spa->spa_load_guid); 1562fa9e4066Sahrens } 1563fa9e4066Sahrens 1564fa9e4066Sahrens uint64_t 1565fa9e4066Sahrens spa_last_synced_txg(spa_t *spa) 1566fa9e4066Sahrens { 1567fa9e4066Sahrens return (spa->spa_ubsync.ub_txg); 1568fa9e4066Sahrens } 1569fa9e4066Sahrens 1570fa9e4066Sahrens uint64_t 1571fa9e4066Sahrens spa_first_txg(spa_t *spa) 1572fa9e4066Sahrens { 1573fa9e4066Sahrens return (spa->spa_first_txg); 1574fa9e4066Sahrens } 1575fa9e4066Sahrens 1576b24ab676SJeff Bonwick uint64_t 1577b24ab676SJeff Bonwick spa_syncing_txg(spa_t *spa) 1578b24ab676SJeff Bonwick { 1579b24ab676SJeff Bonwick return (spa->spa_syncing_txg); 1580b24ab676SJeff Bonwick } 1581b24ab676SJeff Bonwick 158288b7b0f2SMatthew Ahrens pool_state_t 1583fa9e4066Sahrens spa_state(spa_t *spa) 1584fa9e4066Sahrens { 1585fa9e4066Sahrens return (spa->spa_state); 1586fa9e4066Sahrens } 1587fa9e4066Sahrens 1588b16da2e2SGeorge Wilson spa_load_state_t 1589b16da2e2SGeorge Wilson spa_load_state(spa_t *spa) 1590b16da2e2SGeorge Wilson { 1591b16da2e2SGeorge Wilson return (spa->spa_load_state); 1592b16da2e2SGeorge Wilson } 1593b16da2e2SGeorge Wilson 1594fa9e4066Sahrens uint64_t 1595fa9e4066Sahrens spa_freeze_txg(spa_t *spa) 1596fa9e4066Sahrens { 1597fa9e4066Sahrens return (spa->spa_freeze_txg); 1598fa9e4066Sahrens } 1599fa9e4066Sahrens 1600fa9e4066Sahrens /* ARGSUSED */ 1601fa9e4066Sahrens uint64_t 1602fa9e4066Sahrens spa_get_asize(spa_t *spa, uint64_t lsize) 1603fa9e4066Sahrens { 160469962b56SMatthew Ahrens return (lsize * spa_asize_inflation); 160544cd46caSbillm } 160644cd46caSbillm 16077d46dc6cSMatthew Ahrens /* 16087d46dc6cSMatthew Ahrens * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%), 16097d46dc6cSMatthew Ahrens * or at least 32MB. 16107d46dc6cSMatthew Ahrens * 16117d46dc6cSMatthew Ahrens * See the comment above spa_slop_shift for details. 16127d46dc6cSMatthew Ahrens */ 16137d46dc6cSMatthew Ahrens uint64_t 16147d46dc6cSMatthew Ahrens spa_get_slop_space(spa_t *spa) { 16157d46dc6cSMatthew Ahrens uint64_t space = spa_get_dspace(spa); 16167d46dc6cSMatthew Ahrens return (MAX(space >> spa_slop_shift, SPA_MINDEVSIZE >> 1)); 16177d46dc6cSMatthew Ahrens } 16187d46dc6cSMatthew Ahrens 1619485bbbf5SGeorge Wilson uint64_t 1620485bbbf5SGeorge Wilson spa_get_dspace(spa_t *spa) 1621485bbbf5SGeorge Wilson { 1622485bbbf5SGeorge Wilson return (spa->spa_dspace); 1623485bbbf5SGeorge Wilson } 1624485bbbf5SGeorge Wilson 1625485bbbf5SGeorge Wilson void 1626485bbbf5SGeorge Wilson spa_update_dspace(spa_t *spa) 1627485bbbf5SGeorge Wilson { 1628485bbbf5SGeorge Wilson spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) + 1629485bbbf5SGeorge Wilson ddt_get_dedup_dspace(spa); 1630485bbbf5SGeorge Wilson } 1631485bbbf5SGeorge Wilson 16320a4e9518Sgw /* 16330a4e9518Sgw * Return the failure mode that has been set to this pool. The default 16340a4e9518Sgw * behavior will be to block all I/Os when a complete failure occurs. 16350a4e9518Sgw */ 16360a4e9518Sgw uint8_t 16370a4e9518Sgw spa_get_failmode(spa_t *spa) 16380a4e9518Sgw { 16390a4e9518Sgw return (spa->spa_failmode); 16400a4e9518Sgw } 16410a4e9518Sgw 1642e14bb325SJeff Bonwick boolean_t 1643e14bb325SJeff Bonwick spa_suspended(spa_t *spa) 1644e14bb325SJeff Bonwick { 1645e14bb325SJeff Bonwick return (spa->spa_suspended); 1646e14bb325SJeff Bonwick } 1647e14bb325SJeff Bonwick 164844cd46caSbillm uint64_t 164944cd46caSbillm spa_version(spa_t *spa) 165044cd46caSbillm { 165144cd46caSbillm return (spa->spa_ubsync.ub_version); 165244cd46caSbillm } 165344cd46caSbillm 1654b24ab676SJeff Bonwick boolean_t 1655b24ab676SJeff Bonwick spa_deflate(spa_t *spa) 1656b24ab676SJeff Bonwick { 1657b24ab676SJeff Bonwick return (spa->spa_deflate); 1658b24ab676SJeff Bonwick } 1659b24ab676SJeff Bonwick 1660b24ab676SJeff Bonwick metaslab_class_t * 1661b24ab676SJeff Bonwick spa_normal_class(spa_t *spa) 1662b24ab676SJeff Bonwick { 1663b24ab676SJeff Bonwick return (spa->spa_normal_class); 1664b24ab676SJeff Bonwick } 1665b24ab676SJeff Bonwick 1666b24ab676SJeff Bonwick metaslab_class_t * 1667b24ab676SJeff Bonwick spa_log_class(spa_t *spa) 1668b24ab676SJeff Bonwick { 1669b24ab676SJeff Bonwick return (spa->spa_log_class); 1670b24ab676SJeff Bonwick } 1671b24ab676SJeff Bonwick 167244cd46caSbillm int 167344cd46caSbillm spa_max_replication(spa_t *spa) 167444cd46caSbillm { 167544cd46caSbillm /* 1676e7437265Sahrens * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to 167744cd46caSbillm * handle BPs with more than one DVA allocated. Set our max 167844cd46caSbillm * replication level accordingly. 1679fa9e4066Sahrens */ 1680e7437265Sahrens if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS) 168144cd46caSbillm return (1); 168244cd46caSbillm return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override)); 1683fa9e4066Sahrens } 1684fa9e4066Sahrens 16853f9d6ad7SLin Ling int 16863f9d6ad7SLin Ling spa_prev_software_version(spa_t *spa) 16873f9d6ad7SLin Ling { 16883f9d6ad7SLin Ling return (spa->spa_prev_software_version); 16893f9d6ad7SLin Ling } 16903f9d6ad7SLin Ling 1691283b8460SGeorge.Wilson uint64_t 1692283b8460SGeorge.Wilson spa_deadman_synctime(spa_t *spa) 1693283b8460SGeorge.Wilson { 1694283b8460SGeorge.Wilson return (spa->spa_deadman_synctime); 1695283b8460SGeorge.Wilson } 1696283b8460SGeorge.Wilson 169799653d4eSeschrock uint64_t 1698b24ab676SJeff Bonwick dva_get_dsize_sync(spa_t *spa, const dva_t *dva) 169999653d4eSeschrock { 1700b24ab676SJeff Bonwick uint64_t asize = DVA_GET_ASIZE(dva); 1701b24ab676SJeff Bonwick uint64_t dsize = asize; 170299653d4eSeschrock 1703b24ab676SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 170499653d4eSeschrock 1705b24ab676SJeff Bonwick if (asize != 0 && spa->spa_deflate) { 1706b24ab676SJeff Bonwick vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva)); 1707b24ab676SJeff Bonwick dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio; 170899653d4eSeschrock } 1709b24ab676SJeff Bonwick 1710b24ab676SJeff Bonwick return (dsize); 1711b24ab676SJeff Bonwick } 1712b24ab676SJeff Bonwick 1713b24ab676SJeff Bonwick uint64_t 1714b24ab676SJeff Bonwick bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp) 1715b24ab676SJeff Bonwick { 1716b24ab676SJeff Bonwick uint64_t dsize = 0; 1717b24ab676SJeff Bonwick 17185d7b4d43SMatthew Ahrens for (int d = 0; d < BP_GET_NDVAS(bp); d++) 1719b24ab676SJeff Bonwick dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); 1720b24ab676SJeff Bonwick 1721b24ab676SJeff Bonwick return (dsize); 1722b24ab676SJeff Bonwick } 1723b24ab676SJeff Bonwick 1724b24ab676SJeff Bonwick uint64_t 1725b24ab676SJeff Bonwick bp_get_dsize(spa_t *spa, const blkptr_t *bp) 1726b24ab676SJeff Bonwick { 1727b24ab676SJeff Bonwick uint64_t dsize = 0; 1728b24ab676SJeff Bonwick 1729b24ab676SJeff Bonwick spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 1730b24ab676SJeff Bonwick 17315d7b4d43SMatthew Ahrens for (int d = 0; d < BP_GET_NDVAS(bp); d++) 1732b24ab676SJeff Bonwick dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); 1733b24ab676SJeff Bonwick 1734e14bb325SJeff Bonwick spa_config_exit(spa, SCL_VDEV, FTAG); 1735b24ab676SJeff Bonwick 1736b24ab676SJeff Bonwick return (dsize); 173799653d4eSeschrock } 173899653d4eSeschrock 1739fa9e4066Sahrens /* 1740fa9e4066Sahrens * ========================================================================== 1741fa9e4066Sahrens * Initialization and Termination 1742fa9e4066Sahrens * ========================================================================== 1743fa9e4066Sahrens */ 1744fa9e4066Sahrens 1745fa9e4066Sahrens static int 1746fa9e4066Sahrens spa_name_compare(const void *a1, const void *a2) 1747fa9e4066Sahrens { 1748fa9e4066Sahrens const spa_t *s1 = a1; 1749fa9e4066Sahrens const spa_t *s2 = a2; 1750fa9e4066Sahrens int s; 1751fa9e4066Sahrens 1752fa9e4066Sahrens s = strcmp(s1->spa_name, s2->spa_name); 1753fa9e4066Sahrens if (s > 0) 1754fa9e4066Sahrens return (1); 1755fa9e4066Sahrens if (s < 0) 1756fa9e4066Sahrens return (-1); 1757fa9e4066Sahrens return (0); 1758fa9e4066Sahrens } 1759fa9e4066Sahrens 17600373e76bSbonwick int 17610373e76bSbonwick spa_busy(void) 17620373e76bSbonwick { 17630373e76bSbonwick return (spa_active_count); 17640373e76bSbonwick } 17650373e76bSbonwick 1766e7cbe64fSgw void 1767e7cbe64fSgw spa_boot_init() 1768e7cbe64fSgw { 1769e7cbe64fSgw spa_config_load(); 1770e7cbe64fSgw } 1771e7cbe64fSgw 1772fa9e4066Sahrens void 1773fa9e4066Sahrens spa_init(int mode) 1774fa9e4066Sahrens { 1775fa9e4066Sahrens mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL); 1776c25056deSgw mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL); 1777fa94a07fSbrendan mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL); 1778fa9e4066Sahrens cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL); 1779fa9e4066Sahrens 1780fa9e4066Sahrens avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t), 1781fa9e4066Sahrens offsetof(spa_t, spa_avl)); 1782fa9e4066Sahrens 1783fa94a07fSbrendan avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t), 1784fa94a07fSbrendan offsetof(spa_aux_t, aux_avl)); 1785fa94a07fSbrendan 1786fa94a07fSbrendan avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t), 1787fa94a07fSbrendan offsetof(spa_aux_t, aux_avl)); 178899653d4eSeschrock 17898ad4d6ddSJeff Bonwick spa_mode_global = mode; 1790fa9e4066Sahrens 1791283b8460SGeorge.Wilson #ifdef _KERNEL 1792283b8460SGeorge.Wilson spa_arch_init(); 1793283b8460SGeorge.Wilson #else 1794cd1c8b85SMatthew Ahrens if (spa_mode_global != FREAD && dprintf_find_string("watch")) { 1795cd1c8b85SMatthew Ahrens arc_procfd = open("/proc/self/ctl", O_WRONLY); 1796cd1c8b85SMatthew Ahrens if (arc_procfd == -1) { 1797cd1c8b85SMatthew Ahrens perror("could not enable watchpoints: " 1798cd1c8b85SMatthew Ahrens "opening /proc/self/ctl failed: "); 1799cd1c8b85SMatthew Ahrens } else { 1800cd1c8b85SMatthew Ahrens arc_watch = B_TRUE; 1801cd1c8b85SMatthew Ahrens } 1802cd1c8b85SMatthew Ahrens } 1803cd1c8b85SMatthew Ahrens #endif 1804cd1c8b85SMatthew Ahrens 1805fa9e4066Sahrens refcount_init(); 1806fa9e4066Sahrens unique_init(); 18070713e232SGeorge Wilson range_tree_init(); 1808fa9e4066Sahrens zio_init(); 1809fa9e4066Sahrens dmu_init(); 1810fa9e4066Sahrens zil_init(); 181187db74c1Sek vdev_cache_stat_init(); 181291ebeef5Sahrens zfs_prop_init(); 1813990b4856Slling zpool_prop_init(); 1814ad135b5dSChristopher Siden zpool_feature_init(); 1815fa9e4066Sahrens spa_config_load(); 1816e14bb325SJeff Bonwick l2arc_start(); 1817fa9e4066Sahrens } 1818fa9e4066Sahrens 1819fa9e4066Sahrens void 1820fa9e4066Sahrens spa_fini(void) 1821fa9e4066Sahrens { 1822e14bb325SJeff Bonwick l2arc_stop(); 1823e14bb325SJeff Bonwick 1824fa9e4066Sahrens spa_evict_all(); 1825fa9e4066Sahrens 182687db74c1Sek vdev_cache_stat_fini(); 1827fa9e4066Sahrens zil_fini(); 1828fa9e4066Sahrens dmu_fini(); 1829fa9e4066Sahrens zio_fini(); 18300713e232SGeorge Wilson range_tree_fini(); 183191ebeef5Sahrens unique_fini(); 1832fa9e4066Sahrens refcount_fini(); 1833fa9e4066Sahrens 1834fa9e4066Sahrens avl_destroy(&spa_namespace_avl); 183599653d4eSeschrock avl_destroy(&spa_spare_avl); 1836fa94a07fSbrendan avl_destroy(&spa_l2cache_avl); 1837fa9e4066Sahrens 1838fa9e4066Sahrens cv_destroy(&spa_namespace_cv); 1839fa9e4066Sahrens mutex_destroy(&spa_namespace_lock); 1840c25056deSgw mutex_destroy(&spa_spare_lock); 1841fa94a07fSbrendan mutex_destroy(&spa_l2cache_lock); 1842fa9e4066Sahrens } 18436ce0521aSperrin 18446ce0521aSperrin /* 18456ce0521aSperrin * Return whether this pool has slogs. No locking needed. 18466ce0521aSperrin * It's not a problem if the wrong answer is returned as it's only for 18476ce0521aSperrin * performance and not correctness 18486ce0521aSperrin */ 18496ce0521aSperrin boolean_t 18506ce0521aSperrin spa_has_slogs(spa_t *spa) 18516ce0521aSperrin { 18526ce0521aSperrin return (spa->spa_log_class->mc_rotor != NULL); 18536ce0521aSperrin } 1854bf82a41bSeschrock 1855b24ab676SJeff Bonwick spa_log_state_t 1856b24ab676SJeff Bonwick spa_get_log_state(spa_t *spa) 1857b24ab676SJeff Bonwick { 1858b24ab676SJeff Bonwick return (spa->spa_log_state); 1859b24ab676SJeff Bonwick } 1860b24ab676SJeff Bonwick 1861b24ab676SJeff Bonwick void 1862b24ab676SJeff Bonwick spa_set_log_state(spa_t *spa, spa_log_state_t state) 1863b24ab676SJeff Bonwick { 1864b24ab676SJeff Bonwick spa->spa_log_state = state; 1865b24ab676SJeff Bonwick } 1866b24ab676SJeff Bonwick 1867bf82a41bSeschrock boolean_t 1868bf82a41bSeschrock spa_is_root(spa_t *spa) 1869bf82a41bSeschrock { 1870bf82a41bSeschrock return (spa->spa_is_root); 1871bf82a41bSeschrock } 18728ad4d6ddSJeff Bonwick 18738ad4d6ddSJeff Bonwick boolean_t 18748ad4d6ddSJeff Bonwick spa_writeable(spa_t *spa) 18758ad4d6ddSJeff Bonwick { 18768ad4d6ddSJeff Bonwick return (!!(spa->spa_mode & FWRITE)); 18778ad4d6ddSJeff Bonwick } 18788ad4d6ddSJeff Bonwick 1879*73527f44SAlex Reece /* 1880*73527f44SAlex Reece * Returns true if there is a pending sync task in any of the current 1881*73527f44SAlex Reece * syncing txg, the current quiescing txg, or the current open txg. 1882*73527f44SAlex Reece */ 1883*73527f44SAlex Reece boolean_t 1884*73527f44SAlex Reece spa_has_pending_synctask(spa_t *spa) 1885*73527f44SAlex Reece { 1886*73527f44SAlex Reece return (!txg_all_lists_empty(&spa->spa_dsl_pool->dp_sync_tasks)); 1887*73527f44SAlex Reece } 1888*73527f44SAlex Reece 18898ad4d6ddSJeff Bonwick int 18908ad4d6ddSJeff Bonwick spa_mode(spa_t *spa) 18918ad4d6ddSJeff Bonwick { 18928ad4d6ddSJeff Bonwick return (spa->spa_mode); 18938ad4d6ddSJeff Bonwick } 1894b24ab676SJeff Bonwick 1895b24ab676SJeff Bonwick uint64_t 1896b24ab676SJeff Bonwick spa_bootfs(spa_t *spa) 1897b24ab676SJeff Bonwick { 1898b24ab676SJeff Bonwick return (spa->spa_bootfs); 1899b24ab676SJeff Bonwick } 1900b24ab676SJeff Bonwick 1901b24ab676SJeff Bonwick uint64_t 1902b24ab676SJeff Bonwick spa_delegation(spa_t *spa) 1903b24ab676SJeff Bonwick { 1904b24ab676SJeff Bonwick return (spa->spa_delegation); 1905b24ab676SJeff Bonwick } 1906b24ab676SJeff Bonwick 1907b24ab676SJeff Bonwick objset_t * 1908b24ab676SJeff Bonwick spa_meta_objset(spa_t *spa) 1909b24ab676SJeff Bonwick { 1910b24ab676SJeff Bonwick return (spa->spa_meta_objset); 1911b24ab676SJeff Bonwick } 1912b24ab676SJeff Bonwick 1913b24ab676SJeff Bonwick enum zio_checksum 1914b24ab676SJeff Bonwick spa_dedup_checksum(spa_t *spa) 1915b24ab676SJeff Bonwick { 1916b24ab676SJeff Bonwick return (spa->spa_dedup_checksum); 1917b24ab676SJeff Bonwick } 19183f9d6ad7SLin Ling 19193f9d6ad7SLin Ling /* 19203f9d6ad7SLin Ling * Reset pool scan stat per scan pass (or reboot). 19213f9d6ad7SLin Ling */ 19223f9d6ad7SLin Ling void 19233f9d6ad7SLin Ling spa_scan_stat_init(spa_t *spa) 19243f9d6ad7SLin Ling { 19253f9d6ad7SLin Ling /* data not stored on disk */ 19263f9d6ad7SLin Ling spa->spa_scan_pass_start = gethrestime_sec(); 19273f9d6ad7SLin Ling spa->spa_scan_pass_exam = 0; 19283f9d6ad7SLin Ling vdev_scan_stat_init(spa->spa_root_vdev); 19293f9d6ad7SLin Ling } 19303f9d6ad7SLin Ling 19313f9d6ad7SLin Ling /* 19323f9d6ad7SLin Ling * Get scan stats for zpool status reports 19333f9d6ad7SLin Ling */ 19343f9d6ad7SLin Ling int 19353f9d6ad7SLin Ling spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps) 19363f9d6ad7SLin Ling { 19373f9d6ad7SLin Ling dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL; 19383f9d6ad7SLin Ling 19393f9d6ad7SLin Ling if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE) 1940be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 19413f9d6ad7SLin Ling bzero(ps, sizeof (pool_scan_stat_t)); 19423f9d6ad7SLin Ling 19433f9d6ad7SLin Ling /* data stored on disk */ 19443f9d6ad7SLin Ling ps->pss_func = scn->scn_phys.scn_func; 19453f9d6ad7SLin Ling ps->pss_start_time = scn->scn_phys.scn_start_time; 19463f9d6ad7SLin Ling ps->pss_end_time = scn->scn_phys.scn_end_time; 19473f9d6ad7SLin Ling ps->pss_to_examine = scn->scn_phys.scn_to_examine; 19483f9d6ad7SLin Ling ps->pss_examined = scn->scn_phys.scn_examined; 19493f9d6ad7SLin Ling ps->pss_to_process = scn->scn_phys.scn_to_process; 19503f9d6ad7SLin Ling ps->pss_processed = scn->scn_phys.scn_processed; 19513f9d6ad7SLin Ling ps->pss_errors = scn->scn_phys.scn_errors; 19523f9d6ad7SLin Ling ps->pss_state = scn->scn_phys.scn_state; 19533f9d6ad7SLin Ling 19543f9d6ad7SLin Ling /* data not stored on disk */ 19553f9d6ad7SLin Ling ps->pss_pass_start = spa->spa_scan_pass_start; 19563f9d6ad7SLin Ling ps->pss_pass_exam = spa->spa_scan_pass_exam; 19573f9d6ad7SLin Ling 19583f9d6ad7SLin Ling return (0); 19593f9d6ad7SLin Ling } 196009c9d376SGeorge Wilson 196109c9d376SGeorge Wilson boolean_t 196209c9d376SGeorge Wilson spa_debug_enabled(spa_t *spa) 196309c9d376SGeorge Wilson { 196409c9d376SGeorge Wilson return (spa->spa_debug); 196509c9d376SGeorge Wilson } 1966