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. 2394c2d0ebSMatthew Ahrens * Copyright (c) 2011, 2017 by Delphix. All rights reserved. 24e495b6e6SSaso Kiselkov * Copyright 2015 Nexenta Systems, Inc. All rights reserved. 25bc9014e6SJustin Gibbs * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 2645818ee1SMatthew Ahrens * Copyright 2013 Saso Kiselkov. All rights reserved. 27c3d26abcSMatthew Ahrens * Copyright (c) 2014 Integros [integros.com] 281702cce7SAlek Pinchuk * Copyright (c) 2017 Datto Inc. 29fa9e4066Sahrens */ 30fa9e4066Sahrens 31fa9e4066Sahrens #include <sys/zfs_context.h> 32fa9e4066Sahrens #include <sys/spa_impl.h> 33283b8460SGeorge.Wilson #include <sys/spa_boot.h> 34fa9e4066Sahrens #include <sys/zio.h> 35fa9e4066Sahrens #include <sys/zio_checksum.h> 36fa9e4066Sahrens #include <sys/zio_compress.h> 37fa9e4066Sahrens #include <sys/dmu.h> 38fa9e4066Sahrens #include <sys/dmu_tx.h> 39fa9e4066Sahrens #include <sys/zap.h> 40fa9e4066Sahrens #include <sys/zil.h> 41fa9e4066Sahrens #include <sys/vdev_impl.h> 42fa9e4066Sahrens #include <sys/metaslab.h> 43fa9e4066Sahrens #include <sys/uberblock_impl.h> 44fa9e4066Sahrens #include <sys/txg.h> 45fa9e4066Sahrens #include <sys/avl.h> 46fa9e4066Sahrens #include <sys/unique.h> 47fa9e4066Sahrens #include <sys/dsl_pool.h> 48fa9e4066Sahrens #include <sys/dsl_dir.h> 49fa9e4066Sahrens #include <sys/dsl_prop.h> 503f9d6ad7SLin Ling #include <sys/dsl_scan.h> 51fa9e4066Sahrens #include <sys/fs/zfs.h> 526ce0521aSperrin #include <sys/metaslab_impl.h> 53e14bb325SJeff Bonwick #include <sys/arc.h> 54485bbbf5SGeorge Wilson #include <sys/ddt.h> 5591ebeef5Sahrens #include "zfs_prop.h" 5645818ee1SMatthew Ahrens #include <sys/zfeature.h> 57fa9e4066Sahrens 58fa9e4066Sahrens /* 59fa9e4066Sahrens * SPA locking 60fa9e4066Sahrens * 61fa9e4066Sahrens * There are four basic locks for managing spa_t structures: 62fa9e4066Sahrens * 63fa9e4066Sahrens * spa_namespace_lock (global mutex) 64fa9e4066Sahrens * 6544cd46caSbillm * This lock must be acquired to do any of the following: 66fa9e4066Sahrens * 6744cd46caSbillm * - Lookup a spa_t by name 6844cd46caSbillm * - Add or remove a spa_t from the namespace 6944cd46caSbillm * - Increase spa_refcount from non-zero 7044cd46caSbillm * - Check if spa_refcount is zero 7144cd46caSbillm * - Rename a spa_t 72ea8dc4b6Seschrock * - add/remove/attach/detach devices 7344cd46caSbillm * - Held for the duration of create/destroy/import/export 74fa9e4066Sahrens * 7544cd46caSbillm * It does not need to handle recursion. A create or destroy may 7644cd46caSbillm * reference objects (files or zvols) in other pools, but by 7744cd46caSbillm * definition they must have an existing reference, and will never need 7844cd46caSbillm * to lookup a spa_t by name. 79fa9e4066Sahrens * 80fa9e4066Sahrens * spa_refcount (per-spa refcount_t protected by mutex) 81fa9e4066Sahrens * 8244cd46caSbillm * This reference count keep track of any active users of the spa_t. The 8344cd46caSbillm * spa_t cannot be destroyed or freed while this is non-zero. Internally, 8444cd46caSbillm * the refcount is never really 'zero' - opening a pool implicitly keeps 85088f3894Sahrens * some references in the DMU. Internally we check against spa_minref, but 8644cd46caSbillm * present the image of a zero/non-zero value to consumers. 87fa9e4066Sahrens * 88e14bb325SJeff Bonwick * spa_config_lock[] (per-spa array of rwlocks) 89fa9e4066Sahrens * 9091ebeef5Sahrens * This protects the spa_t from config changes, and must be held in 9191ebeef5Sahrens * the following circumstances: 92fa9e4066Sahrens * 9344cd46caSbillm * - RW_READER to perform I/O to the spa 9444cd46caSbillm * - RW_WRITER to change the vdev config 95fa9e4066Sahrens * 96fa9e4066Sahrens * The locking order is fairly straightforward: 97fa9e4066Sahrens * 9844cd46caSbillm * spa_namespace_lock -> spa_refcount 99fa9e4066Sahrens * 10044cd46caSbillm * The namespace lock must be acquired to increase the refcount from 0 10144cd46caSbillm * or to check if it is zero. 102fa9e4066Sahrens * 103e14bb325SJeff Bonwick * spa_refcount -> spa_config_lock[] 104fa9e4066Sahrens * 10544cd46caSbillm * There must be at least one valid reference on the spa_t to acquire 10644cd46caSbillm * the config lock. 107fa9e4066Sahrens * 108e14bb325SJeff Bonwick * spa_namespace_lock -> spa_config_lock[] 109fa9e4066Sahrens * 11044cd46caSbillm * The namespace lock must always be taken before the config lock. 111fa9e4066Sahrens * 112fa9e4066Sahrens * 113e14bb325SJeff Bonwick * The spa_namespace_lock can be acquired directly and is globally visible. 114fa9e4066Sahrens * 115e14bb325SJeff Bonwick * The namespace is manipulated using the following functions, all of which 116e14bb325SJeff Bonwick * require the spa_namespace_lock to be held. 117fa9e4066Sahrens * 11844cd46caSbillm * spa_lookup() Lookup a spa_t by name. 119fa9e4066Sahrens * 12044cd46caSbillm * spa_add() Create a new spa_t in the namespace. 121fa9e4066Sahrens * 12244cd46caSbillm * spa_remove() Remove a spa_t from the namespace. This also 12344cd46caSbillm * frees up any memory associated with the spa_t. 124fa9e4066Sahrens * 12544cd46caSbillm * spa_next() Returns the next spa_t in the system, or the 12644cd46caSbillm * first if NULL is passed. 127fa9e4066Sahrens * 12844cd46caSbillm * spa_evict_all() Shutdown and remove all spa_t structures in 12944cd46caSbillm * the system. 130fa9e4066Sahrens * 131ea8dc4b6Seschrock * spa_guid_exists() Determine whether a pool/device guid exists. 132fa9e4066Sahrens * 133fa9e4066Sahrens * The spa_refcount is manipulated using the following functions: 134fa9e4066Sahrens * 13544cd46caSbillm * spa_open_ref() Adds a reference to the given spa_t. Must be 13644cd46caSbillm * called with spa_namespace_lock held if the 13744cd46caSbillm * refcount is currently zero. 138fa9e4066Sahrens * 13944cd46caSbillm * spa_close() Remove a reference from the spa_t. This will 14044cd46caSbillm * not free the spa_t or remove it from the 14144cd46caSbillm * namespace. No locking is required. 142fa9e4066Sahrens * 14344cd46caSbillm * spa_refcount_zero() Returns true if the refcount is currently 14444cd46caSbillm * zero. Must be called with spa_namespace_lock 14544cd46caSbillm * held. 146fa9e4066Sahrens * 147e14bb325SJeff Bonwick * The spa_config_lock[] is an array of rwlocks, ordered as follows: 148e14bb325SJeff Bonwick * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV. 149e14bb325SJeff Bonwick * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}(). 150e14bb325SJeff Bonwick * 151e14bb325SJeff Bonwick * To read the configuration, it suffices to hold one of these locks as reader. 152e14bb325SJeff Bonwick * To modify the configuration, you must hold all locks as writer. To modify 153e14bb325SJeff Bonwick * vdev state without altering the vdev tree's topology (e.g. online/offline), 154e14bb325SJeff Bonwick * you must hold SCL_STATE and SCL_ZIO as writer. 155e14bb325SJeff Bonwick * 156e14bb325SJeff Bonwick * We use these distinct config locks to avoid recursive lock entry. 157e14bb325SJeff Bonwick * For example, spa_sync() (which holds SCL_CONFIG as reader) induces 158e14bb325SJeff Bonwick * block allocations (SCL_ALLOC), which may require reading space maps 159e14bb325SJeff Bonwick * from disk (dmu_read() -> zio_read() -> SCL_ZIO). 160e14bb325SJeff Bonwick * 161e14bb325SJeff Bonwick * The spa config locks cannot be normal rwlocks because we need the 162e14bb325SJeff Bonwick * ability to hand off ownership. For example, SCL_ZIO is acquired 163e14bb325SJeff Bonwick * by the issuing thread and later released by an interrupt thread. 164e14bb325SJeff Bonwick * They do, however, obey the usual write-wanted semantics to prevent 165e14bb325SJeff Bonwick * writer (i.e. system administrator) starvation. 166e14bb325SJeff Bonwick * 167e14bb325SJeff Bonwick * The lock acquisition rules are as follows: 168e14bb325SJeff Bonwick * 169e14bb325SJeff Bonwick * SCL_CONFIG 170e14bb325SJeff Bonwick * Protects changes to the vdev tree topology, such as vdev 171e14bb325SJeff Bonwick * add/remove/attach/detach. Protects the dirty config list 172e14bb325SJeff Bonwick * (spa_config_dirty_list) and the set of spares and l2arc devices. 173e14bb325SJeff Bonwick * 174e14bb325SJeff Bonwick * SCL_STATE 175e14bb325SJeff Bonwick * Protects changes to pool state and vdev state, such as vdev 176e14bb325SJeff Bonwick * online/offline/fault/degrade/clear. Protects the dirty state list 177e14bb325SJeff Bonwick * (spa_state_dirty_list) and global pool state (spa_state). 178e14bb325SJeff Bonwick * 179e14bb325SJeff Bonwick * SCL_ALLOC 180e14bb325SJeff Bonwick * Protects changes to metaslab groups and classes. 181e14bb325SJeff Bonwick * Held as reader by metaslab_alloc() and metaslab_claim(). 182e14bb325SJeff Bonwick * 183e14bb325SJeff Bonwick * SCL_ZIO 184e14bb325SJeff Bonwick * Held by bp-level zios (those which have no io_vd upon entry) 185e14bb325SJeff Bonwick * to prevent changes to the vdev tree. The bp-level zio implicitly 186e14bb325SJeff Bonwick * protects all of its vdev child zios, which do not hold SCL_ZIO. 187e14bb325SJeff Bonwick * 188e14bb325SJeff Bonwick * SCL_FREE 189e14bb325SJeff Bonwick * Protects changes to metaslab groups and classes. 190e14bb325SJeff Bonwick * Held as reader by metaslab_free(). SCL_FREE is distinct from 191e14bb325SJeff Bonwick * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free 192e14bb325SJeff Bonwick * blocks in zio_done() while another i/o that holds either 193e14bb325SJeff Bonwick * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete. 194e14bb325SJeff Bonwick * 195e14bb325SJeff Bonwick * SCL_VDEV 196e14bb325SJeff Bonwick * Held as reader to prevent changes to the vdev tree during trivial 197b24ab676SJeff Bonwick * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the 198e14bb325SJeff Bonwick * other locks, and lower than all of them, to ensure that it's safe 199e14bb325SJeff Bonwick * to acquire regardless of caller context. 200e14bb325SJeff Bonwick * 201e14bb325SJeff Bonwick * In addition, the following rules apply: 202e14bb325SJeff Bonwick * 203e14bb325SJeff Bonwick * (a) spa_props_lock protects pool properties, spa_config and spa_config_list. 204e14bb325SJeff Bonwick * The lock ordering is SCL_CONFIG > spa_props_lock. 205e14bb325SJeff Bonwick * 206e14bb325SJeff Bonwick * (b) I/O operations on leaf vdevs. For any zio operation that takes 207e14bb325SJeff Bonwick * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(), 208e14bb325SJeff Bonwick * or zio_write_phys() -- the caller must ensure that the config cannot 209e14bb325SJeff Bonwick * cannot change in the interim, and that the vdev cannot be reopened. 210e14bb325SJeff Bonwick * SCL_STATE as reader suffices for both. 211fa9e4066Sahrens * 212ea8dc4b6Seschrock * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit(). 213fa9e4066Sahrens * 21444cd46caSbillm * spa_vdev_enter() Acquire the namespace lock and the config lock 215ea8dc4b6Seschrock * for writing. 216fa9e4066Sahrens * 21744cd46caSbillm * spa_vdev_exit() Release the config lock, wait for all I/O 21844cd46caSbillm * to complete, sync the updated configs to the 219ea8dc4b6Seschrock * cache, and release the namespace lock. 220fa9e4066Sahrens * 221e14bb325SJeff Bonwick * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit(). 222e14bb325SJeff Bonwick * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual 223e14bb325SJeff Bonwick * locking is, always, based on spa_namespace_lock and spa_config_lock[]. 224e14bb325SJeff Bonwick * 225ad135b5dSChristopher Siden * spa_rename() is also implemented within this file since it requires 226e14bb325SJeff Bonwick * manipulation of the namespace. 227fa9e4066Sahrens */ 228fa9e4066Sahrens 229fa9e4066Sahrens static avl_tree_t spa_namespace_avl; 230fa9e4066Sahrens kmutex_t spa_namespace_lock; 231fa9e4066Sahrens static kcondvar_t spa_namespace_cv; 2320373e76bSbonwick static int spa_active_count; 233416e0cd8Sek int spa_max_replication_override = SPA_DVAS_PER_BP; 234fa9e4066Sahrens 23599653d4eSeschrock static kmutex_t spa_spare_lock; 23639c23413Seschrock static avl_tree_t spa_spare_avl; 237fa94a07fSbrendan static kmutex_t spa_l2cache_lock; 238fa94a07fSbrendan static avl_tree_t spa_l2cache_avl; 23999653d4eSeschrock 240fa9e4066Sahrens kmem_cache_t *spa_buffer_pool; 2418ad4d6ddSJeff Bonwick int spa_mode_global; 242fa9e4066Sahrens 243fa9e4066Sahrens #ifdef ZFS_DEBUG 2445cabbc6bSPrashanth Sreenivasa /* 2455cabbc6bSPrashanth Sreenivasa * Everything except dprintf, spa, and indirect_remap is on by default 2465cabbc6bSPrashanth Sreenivasa * in debug builds. 2475cabbc6bSPrashanth Sreenivasa */ 2485cabbc6bSPrashanth Sreenivasa int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA | ZFS_DEBUG_INDIRECT_REMAP); 249fa9e4066Sahrens #else 250fa9e4066Sahrens int zfs_flags = 0; 251fa9e4066Sahrens #endif 252fa9e4066Sahrens 2530125049cSahrens /* 2540125049cSahrens * zfs_recover can be set to nonzero to attempt to recover from 2550125049cSahrens * otherwise-fatal errors, typically caused by on-disk corruption. When 2560125049cSahrens * set, calls to zfs_panic_recover() will turn into warning messages. 2578b36997aSMatthew Ahrens * This should only be used as a last resort, as it typically results 2588b36997aSMatthew Ahrens * in leaked space, or worse. 2590125049cSahrens */ 2607fd05ac4SMatthew Ahrens boolean_t zfs_recover = B_FALSE; 2617fd05ac4SMatthew Ahrens 2627fd05ac4SMatthew Ahrens /* 2637fd05ac4SMatthew Ahrens * If destroy encounters an EIO while reading metadata (e.g. indirect 2647fd05ac4SMatthew Ahrens * blocks), space referenced by the missing metadata can not be freed. 2657fd05ac4SMatthew Ahrens * Normally this causes the background destroy to become "stalled", as 2667fd05ac4SMatthew Ahrens * it is unable to make forward progress. While in this stalled state, 2677fd05ac4SMatthew Ahrens * all remaining space to free from the error-encountering filesystem is 2687fd05ac4SMatthew Ahrens * "temporarily leaked". Set this flag to cause it to ignore the EIO, 2697fd05ac4SMatthew Ahrens * permanently leak the space from indirect blocks that can not be read, 2707fd05ac4SMatthew Ahrens * and continue to free everything else that it can. 2717fd05ac4SMatthew Ahrens * 2727fd05ac4SMatthew Ahrens * The default, "stalling" behavior is useful if the storage partially 2737fd05ac4SMatthew Ahrens * fails (i.e. some but not all i/os fail), and then later recovers. In 2747fd05ac4SMatthew Ahrens * this case, we will be able to continue pool operations while it is 2757fd05ac4SMatthew Ahrens * partially failed, and when it recovers, we can continue to free the 2767fd05ac4SMatthew Ahrens * space, with no leaks. However, note that this case is actually 2777fd05ac4SMatthew Ahrens * fairly rare. 2787fd05ac4SMatthew Ahrens * 2797fd05ac4SMatthew Ahrens * Typically pools either (a) fail completely (but perhaps temporarily, 2807fd05ac4SMatthew Ahrens * e.g. a top-level vdev going offline), or (b) have localized, 2817fd05ac4SMatthew Ahrens * permanent errors (e.g. disk returns the wrong data due to bit flip or 2827fd05ac4SMatthew Ahrens * firmware bug). In case (a), this setting does not matter because the 2837fd05ac4SMatthew Ahrens * pool will be suspended and the sync thread will not be able to make 2847fd05ac4SMatthew Ahrens * forward progress regardless. In case (b), because the error is 2857fd05ac4SMatthew Ahrens * permanent, the best we can do is leak the minimum amount of space, 2867fd05ac4SMatthew Ahrens * which is what setting this flag will do. Therefore, it is reasonable 2877fd05ac4SMatthew Ahrens * for this flag to normally be set, but we chose the more conservative 2887fd05ac4SMatthew Ahrens * approach of not setting it, so that there is no possibility of 2897fd05ac4SMatthew Ahrens * leaking space in the "partial temporary" failure case. 2907fd05ac4SMatthew Ahrens */ 2917fd05ac4SMatthew Ahrens boolean_t zfs_free_leak_on_eio = B_FALSE; 2920125049cSahrens 29369962b56SMatthew Ahrens /* 29469962b56SMatthew Ahrens * Expiration time in milliseconds. This value has two meanings. First it is 29569962b56SMatthew Ahrens * used to determine when the spa_deadman() logic should fire. By default the 29669962b56SMatthew Ahrens * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds. 29769962b56SMatthew Ahrens * Secondly, the value determines if an I/O is considered "hung". Any I/O that 29869962b56SMatthew Ahrens * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting 29969962b56SMatthew Ahrens * in a system panic. 30069962b56SMatthew Ahrens */ 30169962b56SMatthew Ahrens uint64_t zfs_deadman_synctime_ms = 1000000ULL; 302283b8460SGeorge.Wilson 303283b8460SGeorge.Wilson /* 30469962b56SMatthew Ahrens * Check time in milliseconds. This defines the frequency at which we check 30569962b56SMatthew Ahrens * for hung I/O. 306283b8460SGeorge.Wilson */ 30769962b56SMatthew Ahrens uint64_t zfs_deadman_checktime_ms = 5000ULL; 308283b8460SGeorge.Wilson 309283b8460SGeorge.Wilson /* 310283b8460SGeorge.Wilson * Override the zfs deadman behavior via /etc/system. By default the 311283b8460SGeorge.Wilson * deadman is enabled except on VMware and sparc deployments. 312283b8460SGeorge.Wilson */ 313283b8460SGeorge.Wilson int zfs_deadman_enabled = -1; 314283b8460SGeorge.Wilson 31569962b56SMatthew Ahrens /* 31669962b56SMatthew Ahrens * The worst case is single-sector max-parity RAID-Z blocks, in which 31769962b56SMatthew Ahrens * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1) 31869962b56SMatthew Ahrens * times the size; so just assume that. Add to this the fact that 31969962b56SMatthew Ahrens * we can have up to 3 DVAs per bp, and one more factor of 2 because 32069962b56SMatthew Ahrens * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together, 32169962b56SMatthew Ahrens * the worst case is: 32269962b56SMatthew Ahrens * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24 32369962b56SMatthew Ahrens */ 32469962b56SMatthew Ahrens int spa_asize_inflation = 24; 325fa9e4066Sahrens 3267d46dc6cSMatthew Ahrens /* 3277d46dc6cSMatthew Ahrens * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in 3287d46dc6cSMatthew Ahrens * the pool to be consumed. This ensures that we don't run the pool 3297d46dc6cSMatthew Ahrens * completely out of space, due to unaccounted changes (e.g. to the MOS). 3307d46dc6cSMatthew Ahrens * It also limits the worst-case time to allocate space. If we have 3317d46dc6cSMatthew Ahrens * less than this amount of free space, most ZPL operations (e.g. write, 3327d46dc6cSMatthew Ahrens * create) will return ENOSPC. 3337d46dc6cSMatthew Ahrens * 3347d46dc6cSMatthew Ahrens * Certain operations (e.g. file removal, most administrative actions) can 3357d46dc6cSMatthew Ahrens * use half the slop space. They will only return ENOSPC if less than half 3367d46dc6cSMatthew Ahrens * the slop space is free. Typically, once the pool has less than the slop 3377d46dc6cSMatthew Ahrens * space free, the user will use these operations to free up space in the pool. 3387d46dc6cSMatthew Ahrens * These are the operations that call dsl_pool_adjustedsize() with the netfree 3397d46dc6cSMatthew Ahrens * argument set to TRUE. 3407d46dc6cSMatthew Ahrens * 341*86714001SSerapheim Dimitropoulos * Operations that are almost guaranteed to free up space in the absence of 342*86714001SSerapheim Dimitropoulos * a pool checkpoint can use up to three quarters of the slop space 343*86714001SSerapheim Dimitropoulos * (e.g zfs destroy). 344*86714001SSerapheim Dimitropoulos * 3457d46dc6cSMatthew Ahrens * A very restricted set of operations are always permitted, regardless of 3467d46dc6cSMatthew Ahrens * the amount of free space. These are the operations that call 347*86714001SSerapheim Dimitropoulos * dsl_sync_task(ZFS_SPACE_CHECK_NONE). If these operations result in a net 348*86714001SSerapheim Dimitropoulos * increase in the amount of space used, it is possible to run the pool 349*86714001SSerapheim Dimitropoulos * completely out of space, causing it to be permanently read-only. 3507d46dc6cSMatthew Ahrens * 3514b5c8e93SMatthew Ahrens * Note that on very small pools, the slop space will be larger than 3524b5c8e93SMatthew Ahrens * 3.2%, in an effort to have it be at least spa_min_slop (128MB), 3534b5c8e93SMatthew Ahrens * but we never allow it to be more than half the pool size. 3544b5c8e93SMatthew Ahrens * 3557d46dc6cSMatthew Ahrens * See also the comments in zfs_space_check_t. 3567d46dc6cSMatthew Ahrens */ 3577d46dc6cSMatthew Ahrens int spa_slop_shift = 5; 3584b5c8e93SMatthew Ahrens uint64_t spa_min_slop = 128 * 1024 * 1024; 3597d46dc6cSMatthew Ahrens 3603ee8c80cSPavel Zakharov /*PRINTFLIKE2*/ 3613ee8c80cSPavel Zakharov void 3623ee8c80cSPavel Zakharov spa_load_failed(spa_t *spa, const char *fmt, ...) 3633ee8c80cSPavel Zakharov { 3643ee8c80cSPavel Zakharov va_list adx; 3653ee8c80cSPavel Zakharov char buf[256]; 3663ee8c80cSPavel Zakharov 3673ee8c80cSPavel Zakharov va_start(adx, fmt); 3683ee8c80cSPavel Zakharov (void) vsnprintf(buf, sizeof (buf), fmt, adx); 3693ee8c80cSPavel Zakharov va_end(adx); 3703ee8c80cSPavel Zakharov 3716f793812SPavel Zakharov zfs_dbgmsg("spa_load(%s, config %s): FAILED: %s", spa->spa_name, 3726f793812SPavel Zakharov spa->spa_trust_config ? "trusted" : "untrusted", buf); 3733ee8c80cSPavel Zakharov } 3743ee8c80cSPavel Zakharov 3753ee8c80cSPavel Zakharov /*PRINTFLIKE2*/ 3763ee8c80cSPavel Zakharov void 3773ee8c80cSPavel Zakharov spa_load_note(spa_t *spa, const char *fmt, ...) 3783ee8c80cSPavel Zakharov { 3793ee8c80cSPavel Zakharov va_list adx; 3803ee8c80cSPavel Zakharov char buf[256]; 3813ee8c80cSPavel Zakharov 3823ee8c80cSPavel Zakharov va_start(adx, fmt); 3833ee8c80cSPavel Zakharov (void) vsnprintf(buf, sizeof (buf), fmt, adx); 3843ee8c80cSPavel Zakharov va_end(adx); 3853ee8c80cSPavel Zakharov 3866f793812SPavel Zakharov zfs_dbgmsg("spa_load(%s, config %s): %s", spa->spa_name, 3876f793812SPavel Zakharov spa->spa_trust_config ? "trusted" : "untrusted", buf); 3883ee8c80cSPavel Zakharov } 3893ee8c80cSPavel Zakharov 390e05725b1Sbonwick /* 391e05725b1Sbonwick * ========================================================================== 392e05725b1Sbonwick * SPA config locking 393e05725b1Sbonwick * ========================================================================== 394e05725b1Sbonwick */ 395e05725b1Sbonwick static void 396e14bb325SJeff Bonwick spa_config_lock_init(spa_t *spa) 397e14bb325SJeff Bonwick { 398e14bb325SJeff Bonwick for (int i = 0; i < SCL_LOCKS; i++) { 399e14bb325SJeff Bonwick spa_config_lock_t *scl = &spa->spa_config_lock[i]; 400e14bb325SJeff Bonwick mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL); 401e14bb325SJeff Bonwick cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL); 4023b2aab18SMatthew Ahrens refcount_create_untracked(&scl->scl_count); 403e14bb325SJeff Bonwick scl->scl_writer = NULL; 404e14bb325SJeff Bonwick scl->scl_write_wanted = 0; 405e14bb325SJeff Bonwick } 406e05725b1Sbonwick } 407e05725b1Sbonwick 408e05725b1Sbonwick static void 409e14bb325SJeff Bonwick spa_config_lock_destroy(spa_t *spa) 410e14bb325SJeff Bonwick { 411e14bb325SJeff Bonwick for (int i = 0; i < SCL_LOCKS; i++) { 412e14bb325SJeff Bonwick spa_config_lock_t *scl = &spa->spa_config_lock[i]; 413e14bb325SJeff Bonwick mutex_destroy(&scl->scl_lock); 414e14bb325SJeff Bonwick cv_destroy(&scl->scl_cv); 415e14bb325SJeff Bonwick refcount_destroy(&scl->scl_count); 416e14bb325SJeff Bonwick ASSERT(scl->scl_writer == NULL); 417e14bb325SJeff Bonwick ASSERT(scl->scl_write_wanted == 0); 418e14bb325SJeff Bonwick } 419e14bb325SJeff Bonwick } 420e14bb325SJeff Bonwick 421e14bb325SJeff Bonwick int 422e14bb325SJeff Bonwick spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw) 423e05725b1Sbonwick { 424e14bb325SJeff Bonwick for (int i = 0; i < SCL_LOCKS; i++) { 425e14bb325SJeff Bonwick spa_config_lock_t *scl = &spa->spa_config_lock[i]; 426e14bb325SJeff Bonwick if (!(locks & (1 << i))) 427e14bb325SJeff Bonwick continue; 428e14bb325SJeff Bonwick mutex_enter(&scl->scl_lock); 429e14bb325SJeff Bonwick if (rw == RW_READER) { 430e14bb325SJeff Bonwick if (scl->scl_writer || scl->scl_write_wanted) { 431e14bb325SJeff Bonwick mutex_exit(&scl->scl_lock); 432e495b6e6SSaso Kiselkov spa_config_exit(spa, locks & ((1 << i) - 1), 433e495b6e6SSaso Kiselkov tag); 434e14bb325SJeff Bonwick return (0); 435e14bb325SJeff Bonwick } 436e14bb325SJeff Bonwick } else { 437e14bb325SJeff Bonwick ASSERT(scl->scl_writer != curthread); 438e14bb325SJeff Bonwick if (!refcount_is_zero(&scl->scl_count)) { 439e14bb325SJeff Bonwick mutex_exit(&scl->scl_lock); 440e495b6e6SSaso Kiselkov spa_config_exit(spa, locks & ((1 << i) - 1), 441e495b6e6SSaso Kiselkov tag); 442e14bb325SJeff Bonwick return (0); 443e14bb325SJeff Bonwick } 444e14bb325SJeff Bonwick scl->scl_writer = curthread; 445e14bb325SJeff Bonwick } 446e14bb325SJeff Bonwick (void) refcount_add(&scl->scl_count, tag); 447e14bb325SJeff Bonwick mutex_exit(&scl->scl_lock); 448e14bb325SJeff Bonwick } 449e14bb325SJeff Bonwick return (1); 450e05725b1Sbonwick } 451e05725b1Sbonwick 452e05725b1Sbonwick void 453e14bb325SJeff Bonwick spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw) 454e05725b1Sbonwick { 455f64c0e34SEric Taylor int wlocks_held = 0; 456f64c0e34SEric Taylor 4573b2aab18SMatthew Ahrens ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY); 4583b2aab18SMatthew Ahrens 459e14bb325SJeff Bonwick for (int i = 0; i < SCL_LOCKS; i++) { 460e14bb325SJeff Bonwick spa_config_lock_t *scl = &spa->spa_config_lock[i]; 461f64c0e34SEric Taylor if (scl->scl_writer == curthread) 462f64c0e34SEric Taylor wlocks_held |= (1 << i); 463e14bb325SJeff Bonwick if (!(locks & (1 << i))) 464e14bb325SJeff Bonwick continue; 465e14bb325SJeff Bonwick mutex_enter(&scl->scl_lock); 466e14bb325SJeff Bonwick if (rw == RW_READER) { 467e14bb325SJeff Bonwick while (scl->scl_writer || scl->scl_write_wanted) { 468e14bb325SJeff Bonwick cv_wait(&scl->scl_cv, &scl->scl_lock); 469e14bb325SJeff Bonwick } 470e14bb325SJeff Bonwick } else { 471e14bb325SJeff Bonwick ASSERT(scl->scl_writer != curthread); 472e14bb325SJeff Bonwick while (!refcount_is_zero(&scl->scl_count)) { 473e14bb325SJeff Bonwick scl->scl_write_wanted++; 474e14bb325SJeff Bonwick cv_wait(&scl->scl_cv, &scl->scl_lock); 475e14bb325SJeff Bonwick scl->scl_write_wanted--; 476e14bb325SJeff Bonwick } 477e14bb325SJeff Bonwick scl->scl_writer = curthread; 478e14bb325SJeff Bonwick } 479e14bb325SJeff Bonwick (void) refcount_add(&scl->scl_count, tag); 480e14bb325SJeff Bonwick mutex_exit(&scl->scl_lock); 481e05725b1Sbonwick } 4825cabbc6bSPrashanth Sreenivasa ASSERT3U(wlocks_held, <=, locks); 483e05725b1Sbonwick } 484e05725b1Sbonwick 485e05725b1Sbonwick void 486e14bb325SJeff Bonwick spa_config_exit(spa_t *spa, int locks, void *tag) 487e05725b1Sbonwick { 488e14bb325SJeff Bonwick for (int i = SCL_LOCKS - 1; i >= 0; i--) { 489e14bb325SJeff Bonwick spa_config_lock_t *scl = &spa->spa_config_lock[i]; 490e14bb325SJeff Bonwick if (!(locks & (1 << i))) 491e14bb325SJeff Bonwick continue; 492e14bb325SJeff Bonwick mutex_enter(&scl->scl_lock); 493e14bb325SJeff Bonwick ASSERT(!refcount_is_zero(&scl->scl_count)); 494e14bb325SJeff Bonwick if (refcount_remove(&scl->scl_count, tag) == 0) { 495e14bb325SJeff Bonwick ASSERT(scl->scl_writer == NULL || 496e14bb325SJeff Bonwick scl->scl_writer == curthread); 497e14bb325SJeff Bonwick scl->scl_writer = NULL; /* OK in either case */ 498e14bb325SJeff Bonwick cv_broadcast(&scl->scl_cv); 499e14bb325SJeff Bonwick } 500e14bb325SJeff Bonwick mutex_exit(&scl->scl_lock); 501e05725b1Sbonwick } 502e05725b1Sbonwick } 503e05725b1Sbonwick 504e14bb325SJeff Bonwick int 505e14bb325SJeff Bonwick spa_config_held(spa_t *spa, int locks, krw_t rw) 506e05725b1Sbonwick { 507e14bb325SJeff Bonwick int locks_held = 0; 508e05725b1Sbonwick 509e14bb325SJeff Bonwick for (int i = 0; i < SCL_LOCKS; i++) { 510e14bb325SJeff Bonwick spa_config_lock_t *scl = &spa->spa_config_lock[i]; 511e14bb325SJeff Bonwick if (!(locks & (1 << i))) 512e14bb325SJeff Bonwick continue; 513e14bb325SJeff Bonwick if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) || 514e14bb325SJeff Bonwick (rw == RW_WRITER && scl->scl_writer == curthread)) 515e14bb325SJeff Bonwick locks_held |= 1 << i; 516e14bb325SJeff Bonwick } 517e14bb325SJeff Bonwick 518e14bb325SJeff Bonwick return (locks_held); 519e05725b1Sbonwick } 520e05725b1Sbonwick 521fa9e4066Sahrens /* 522fa9e4066Sahrens * ========================================================================== 523fa9e4066Sahrens * SPA namespace functions 524fa9e4066Sahrens * ========================================================================== 525fa9e4066Sahrens */ 526fa9e4066Sahrens 527fa9e4066Sahrens /* 528fa9e4066Sahrens * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held. 529fa9e4066Sahrens * Returns NULL if no matching spa_t is found. 530fa9e4066Sahrens */ 531fa9e4066Sahrens spa_t * 532fa9e4066Sahrens spa_lookup(const char *name) 533fa9e4066Sahrens { 534e14bb325SJeff Bonwick static spa_t search; /* spa_t is large; don't allocate on stack */ 535e14bb325SJeff Bonwick spa_t *spa; 536fa9e4066Sahrens avl_index_t where; 53740feaa91Sahrens char *cp; 538fa9e4066Sahrens 539fa9e4066Sahrens ASSERT(MUTEX_HELD(&spa_namespace_lock)); 540fa9e4066Sahrens 5413b2aab18SMatthew Ahrens (void) strlcpy(search.spa_name, name, sizeof (search.spa_name)); 5423b2aab18SMatthew Ahrens 54340feaa91Sahrens /* 54440feaa91Sahrens * If it's a full dataset name, figure out the pool name and 54540feaa91Sahrens * just use that. 54640feaa91Sahrens */ 54778f17100SMatthew Ahrens cp = strpbrk(search.spa_name, "/@#"); 5483b2aab18SMatthew Ahrens if (cp != NULL) 54940feaa91Sahrens *cp = '\0'; 55040feaa91Sahrens 551fa9e4066Sahrens spa = avl_find(&spa_namespace_avl, &search, &where); 552fa9e4066Sahrens 553fa9e4066Sahrens return (spa); 554fa9e4066Sahrens } 555fa9e4066Sahrens 556283b8460SGeorge.Wilson /* 557283b8460SGeorge.Wilson * Fires when spa_sync has not completed within zfs_deadman_synctime_ms. 558283b8460SGeorge.Wilson * If the zfs_deadman_enabled flag is set then it inspects all vdev queues 559283b8460SGeorge.Wilson * looking for potentially hung I/Os. 560283b8460SGeorge.Wilson */ 561283b8460SGeorge.Wilson void 562283b8460SGeorge.Wilson spa_deadman(void *arg) 563283b8460SGeorge.Wilson { 564283b8460SGeorge.Wilson spa_t *spa = arg; 565283b8460SGeorge.Wilson 5660713e232SGeorge Wilson /* 5670713e232SGeorge Wilson * Disable the deadman timer if the pool is suspended. 5680713e232SGeorge Wilson */ 5690713e232SGeorge Wilson if (spa_suspended(spa)) { 5700713e232SGeorge Wilson VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 5710713e232SGeorge Wilson return; 5720713e232SGeorge Wilson } 5730713e232SGeorge Wilson 574283b8460SGeorge.Wilson zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu", 575283b8460SGeorge.Wilson (gethrtime() - spa->spa_sync_starttime) / NANOSEC, 576283b8460SGeorge.Wilson ++spa->spa_deadman_calls); 577283b8460SGeorge.Wilson if (zfs_deadman_enabled) 578283b8460SGeorge.Wilson vdev_deadman(spa->spa_root_vdev); 579283b8460SGeorge.Wilson } 580283b8460SGeorge.Wilson 581fa9e4066Sahrens /* 582fa9e4066Sahrens * Create an uninitialized spa_t with the given name. Requires 583fa9e4066Sahrens * spa_namespace_lock. The caller must ensure that the spa_t doesn't already 584fa9e4066Sahrens * exist by calling spa_lookup() first. 585fa9e4066Sahrens */ 586fa9e4066Sahrens spa_t * 587468c413aSTim Haley spa_add(const char *name, nvlist_t *config, const char *altroot) 588fa9e4066Sahrens { 589fa9e4066Sahrens spa_t *spa; 590c5904d13Seschrock spa_config_dirent_t *dp; 591283b8460SGeorge.Wilson cyc_handler_t hdlr; 592283b8460SGeorge.Wilson cyc_time_t when; 593fa9e4066Sahrens 594fa9e4066Sahrens ASSERT(MUTEX_HELD(&spa_namespace_lock)); 595fa9e4066Sahrens 596fa9e4066Sahrens spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP); 597fa9e4066Sahrens 598c25056deSgw mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL); 599c25056deSgw mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL); 60035a5a358SJonathan Adams mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL); 601bc9014e6SJustin Gibbs mutex_init(&spa->spa_evicting_os_lock, NULL, MUTEX_DEFAULT, NULL); 602c25056deSgw mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL); 60335a5a358SJonathan Adams mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL); 604c25056deSgw mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL); 60545818ee1SMatthew Ahrens mutex_init(&spa->spa_cksum_tmpls_lock, NULL, MUTEX_DEFAULT, NULL); 60635a5a358SJonathan Adams mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL); 607a1521560SJeff Bonwick mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL); 608a1521560SJeff Bonwick mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL); 609c3a66015SMatthew Ahrens mutex_init(&spa->spa_iokstat_lock, NULL, MUTEX_DEFAULT, NULL); 6100f7643c7SGeorge Wilson mutex_init(&spa->spa_alloc_lock, NULL, MUTEX_DEFAULT, NULL); 611c25056deSgw 612c25056deSgw cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL); 613bc9014e6SJustin Gibbs cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL); 61435a5a358SJonathan Adams cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL); 615c25056deSgw cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL); 616e14bb325SJeff Bonwick cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL); 617c25056deSgw 618b24ab676SJeff Bonwick for (int t = 0; t < TXG_SIZE; t++) 619cde58dbcSMatthew Ahrens bplist_create(&spa->spa_free_bplist[t]); 620b24ab676SJeff Bonwick 621e14bb325SJeff Bonwick (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name)); 622fa9e4066Sahrens spa->spa_state = POOL_STATE_UNINITIALIZED; 623fa9e4066Sahrens spa->spa_freeze_txg = UINT64_MAX; 6240373e76bSbonwick spa->spa_final_txg = UINT64_MAX; 625468c413aSTim Haley spa->spa_load_max_txg = UINT64_MAX; 62635a5a358SJonathan Adams spa->spa_proc = &p0; 62735a5a358SJonathan Adams spa->spa_proc_state = SPA_PROC_NONE; 6286f793812SPavel Zakharov spa->spa_trust_config = B_TRUE; 629fa9e4066Sahrens 630283b8460SGeorge.Wilson hdlr.cyh_func = spa_deadman; 631283b8460SGeorge.Wilson hdlr.cyh_arg = spa; 632283b8460SGeorge.Wilson hdlr.cyh_level = CY_LOW_LEVEL; 633283b8460SGeorge.Wilson 63469962b56SMatthew Ahrens spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms); 635283b8460SGeorge.Wilson 636283b8460SGeorge.Wilson /* 637283b8460SGeorge.Wilson * This determines how often we need to check for hung I/Os after 638283b8460SGeorge.Wilson * the cyclic has already fired. Since checking for hung I/Os is 639283b8460SGeorge.Wilson * an expensive operation we don't want to check too frequently. 64069962b56SMatthew Ahrens * Instead wait for 5 seconds before checking again. 641283b8460SGeorge.Wilson */ 64269962b56SMatthew Ahrens when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms); 643283b8460SGeorge.Wilson when.cyt_when = CY_INFINITY; 644283b8460SGeorge.Wilson mutex_enter(&cpu_lock); 645283b8460SGeorge.Wilson spa->spa_deadman_cycid = cyclic_add(&hdlr, &when); 646283b8460SGeorge.Wilson mutex_exit(&cpu_lock); 647283b8460SGeorge.Wilson 648fa9e4066Sahrens refcount_create(&spa->spa_refcount); 649e14bb325SJeff Bonwick spa_config_lock_init(spa); 650fa9e4066Sahrens 651fa9e4066Sahrens avl_add(&spa_namespace_avl, spa); 652fa9e4066Sahrens 6530373e76bSbonwick /* 6540373e76bSbonwick * Set the alternate root, if there is one. 6550373e76bSbonwick */ 6560373e76bSbonwick if (altroot) { 6570373e76bSbonwick spa->spa_root = spa_strdup(altroot); 6580373e76bSbonwick spa_active_count++; 6590373e76bSbonwick } 6600373e76bSbonwick 66194c2d0ebSMatthew Ahrens avl_create(&spa->spa_alloc_tree, zio_bookmark_compare, 6620f7643c7SGeorge Wilson sizeof (zio_t), offsetof(zio_t, io_alloc_node)); 6630f7643c7SGeorge Wilson 664c5904d13Seschrock /* 665c5904d13Seschrock * Every pool starts with the default cachefile 666c5904d13Seschrock */ 667c5904d13Seschrock list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t), 668c5904d13Seschrock offsetof(spa_config_dirent_t, scd_link)); 669c5904d13Seschrock 670c5904d13Seschrock dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP); 671ef912c80STim Haley dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path); 672c5904d13Seschrock list_insert_head(&spa->spa_config_list, dp); 673c5904d13Seschrock 6744b964adaSGeorge Wilson VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME, 6754b964adaSGeorge Wilson KM_SLEEP) == 0); 6764b964adaSGeorge Wilson 677ad135b5dSChristopher Siden if (config != NULL) { 678ad135b5dSChristopher Siden nvlist_t *features; 679ad135b5dSChristopher Siden 680ad135b5dSChristopher Siden if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ, 681ad135b5dSChristopher Siden &features) == 0) { 682ad135b5dSChristopher Siden VERIFY(nvlist_dup(features, &spa->spa_label_features, 683ad135b5dSChristopher Siden 0) == 0); 684ad135b5dSChristopher Siden } 685ad135b5dSChristopher Siden 686468c413aSTim Haley VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0); 687ad135b5dSChristopher Siden } 688ad135b5dSChristopher Siden 689ad135b5dSChristopher Siden if (spa->spa_label_features == NULL) { 690ad135b5dSChristopher Siden VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME, 691ad135b5dSChristopher Siden KM_SLEEP) == 0); 692ad135b5dSChristopher Siden } 693468c413aSTim Haley 694c3a66015SMatthew Ahrens spa->spa_iokstat = kstat_create("zfs", 0, name, 695c3a66015SMatthew Ahrens "disk", KSTAT_TYPE_IO, 1, 0); 696c3a66015SMatthew Ahrens if (spa->spa_iokstat) { 697c3a66015SMatthew Ahrens spa->spa_iokstat->ks_lock = &spa->spa_iokstat_lock; 698c3a66015SMatthew Ahrens kstat_install(spa->spa_iokstat); 699c3a66015SMatthew Ahrens } 700c3a66015SMatthew Ahrens 7013b2aab18SMatthew Ahrens spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0); 7023b2aab18SMatthew Ahrens 70381cd5c55SMatthew Ahrens spa->spa_min_ashift = INT_MAX; 70481cd5c55SMatthew Ahrens spa->spa_max_ashift = 0; 70581cd5c55SMatthew Ahrens 70643466aaeSMax Grossman /* 70743466aaeSMax Grossman * As a pool is being created, treat all features as disabled by 70843466aaeSMax Grossman * setting SPA_FEATURE_DISABLED for all entries in the feature 70943466aaeSMax Grossman * refcount cache. 71043466aaeSMax Grossman */ 71143466aaeSMax Grossman for (int i = 0; i < SPA_FEATURES; i++) { 71243466aaeSMax Grossman spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED; 71343466aaeSMax Grossman } 71443466aaeSMax Grossman 715fa9e4066Sahrens return (spa); 716fa9e4066Sahrens } 717fa9e4066Sahrens 718fa9e4066Sahrens /* 719fa9e4066Sahrens * Removes a spa_t from the namespace, freeing up any memory used. Requires 720fa9e4066Sahrens * spa_namespace_lock. This is called only after the spa_t has been closed and 721fa9e4066Sahrens * deactivated. 722fa9e4066Sahrens */ 723fa9e4066Sahrens void 724fa9e4066Sahrens spa_remove(spa_t *spa) 725fa9e4066Sahrens { 726c5904d13Seschrock spa_config_dirent_t *dp; 727c5904d13Seschrock 728fa9e4066Sahrens ASSERT(MUTEX_HELD(&spa_namespace_lock)); 729fa9e4066Sahrens ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 730bc9014e6SJustin Gibbs ASSERT3U(refcount_count(&spa->spa_refcount), ==, 0); 731fa9e4066Sahrens 7321195e687SMark J Musante nvlist_free(spa->spa_config_splitting); 7331195e687SMark J Musante 734fa9e4066Sahrens avl_remove(&spa_namespace_avl, spa); 735fa9e4066Sahrens cv_broadcast(&spa_namespace_cv); 736fa9e4066Sahrens 7370373e76bSbonwick if (spa->spa_root) { 738fa9e4066Sahrens spa_strfree(spa->spa_root); 7390373e76bSbonwick spa_active_count--; 7400373e76bSbonwick } 741fa9e4066Sahrens 742c5904d13Seschrock while ((dp = list_head(&spa->spa_config_list)) != NULL) { 743c5904d13Seschrock list_remove(&spa->spa_config_list, dp); 744c5904d13Seschrock if (dp->scd_path != NULL) 745c5904d13Seschrock spa_strfree(dp->scd_path); 746c5904d13Seschrock kmem_free(dp, sizeof (spa_config_dirent_t)); 747c5904d13Seschrock } 748c5904d13Seschrock 7490f7643c7SGeorge Wilson avl_destroy(&spa->spa_alloc_tree); 750c5904d13Seschrock list_destroy(&spa->spa_config_list); 7512f8aaab3Seschrock 752ad135b5dSChristopher Siden nvlist_free(spa->spa_label_features); 7534b964adaSGeorge Wilson nvlist_free(spa->spa_load_info); 754fa9e4066Sahrens spa_config_set(spa, NULL); 755fa9e4066Sahrens 756283b8460SGeorge.Wilson mutex_enter(&cpu_lock); 757283b8460SGeorge.Wilson if (spa->spa_deadman_cycid != CYCLIC_NONE) 758283b8460SGeorge.Wilson cyclic_remove(spa->spa_deadman_cycid); 759283b8460SGeorge.Wilson mutex_exit(&cpu_lock); 760283b8460SGeorge.Wilson spa->spa_deadman_cycid = CYCLIC_NONE; 761283b8460SGeorge.Wilson 762fa9e4066Sahrens refcount_destroy(&spa->spa_refcount); 76391ebeef5Sahrens 764e14bb325SJeff Bonwick spa_config_lock_destroy(spa); 765fa9e4066Sahrens 766c3a66015SMatthew Ahrens kstat_delete(spa->spa_iokstat); 767c3a66015SMatthew Ahrens spa->spa_iokstat = NULL; 768c3a66015SMatthew Ahrens 769b24ab676SJeff Bonwick for (int t = 0; t < TXG_SIZE; t++) 770cde58dbcSMatthew Ahrens bplist_destroy(&spa->spa_free_bplist[t]); 771b24ab676SJeff Bonwick 77245818ee1SMatthew Ahrens zio_checksum_templates_free(spa); 77345818ee1SMatthew Ahrens 774c25056deSgw cv_destroy(&spa->spa_async_cv); 775bc9014e6SJustin Gibbs cv_destroy(&spa->spa_evicting_os_cv); 77635a5a358SJonathan Adams cv_destroy(&spa->spa_proc_cv); 777c25056deSgw cv_destroy(&spa->spa_scrub_io_cv); 778e14bb325SJeff Bonwick cv_destroy(&spa->spa_suspend_cv); 779c25056deSgw 7800f7643c7SGeorge Wilson mutex_destroy(&spa->spa_alloc_lock); 7815ad82045Snd mutex_destroy(&spa->spa_async_lock); 782c25056deSgw mutex_destroy(&spa->spa_errlist_lock); 78335a5a358SJonathan Adams mutex_destroy(&spa->spa_errlog_lock); 784bc9014e6SJustin Gibbs mutex_destroy(&spa->spa_evicting_os_lock); 78506eeb2adSek mutex_destroy(&spa->spa_history_lock); 78635a5a358SJonathan Adams mutex_destroy(&spa->spa_proc_lock); 787b1b8ab34Slling mutex_destroy(&spa->spa_props_lock); 78845818ee1SMatthew Ahrens mutex_destroy(&spa->spa_cksum_tmpls_lock); 78935a5a358SJonathan Adams mutex_destroy(&spa->spa_scrub_lock); 790e14bb325SJeff Bonwick mutex_destroy(&spa->spa_suspend_lock); 791a1521560SJeff Bonwick mutex_destroy(&spa->spa_vdev_top_lock); 792c3a66015SMatthew Ahrens mutex_destroy(&spa->spa_iokstat_lock); 7935ad82045Snd 794fa9e4066Sahrens kmem_free(spa, sizeof (spa_t)); 795fa9e4066Sahrens } 796fa9e4066Sahrens 797fa9e4066Sahrens /* 798fa9e4066Sahrens * Given a pool, return the next pool in the namespace, or NULL if there is 799fa9e4066Sahrens * none. If 'prev' is NULL, return the first pool. 800fa9e4066Sahrens */ 801fa9e4066Sahrens spa_t * 802fa9e4066Sahrens spa_next(spa_t *prev) 803fa9e4066Sahrens { 804fa9e4066Sahrens ASSERT(MUTEX_HELD(&spa_namespace_lock)); 805fa9e4066Sahrens 806fa9e4066Sahrens if (prev) 807fa9e4066Sahrens return (AVL_NEXT(&spa_namespace_avl, prev)); 808fa9e4066Sahrens else 809fa9e4066Sahrens return (avl_first(&spa_namespace_avl)); 810fa9e4066Sahrens } 811fa9e4066Sahrens 812fa9e4066Sahrens /* 813fa9e4066Sahrens * ========================================================================== 814fa9e4066Sahrens * SPA refcount functions 815fa9e4066Sahrens * ========================================================================== 816fa9e4066Sahrens */ 817fa9e4066Sahrens 818fa9e4066Sahrens /* 819fa9e4066Sahrens * Add a reference to the given spa_t. Must have at least one reference, or 820fa9e4066Sahrens * have the namespace lock held. 821fa9e4066Sahrens */ 822fa9e4066Sahrens void 823fa9e4066Sahrens spa_open_ref(spa_t *spa, void *tag) 824fa9e4066Sahrens { 825088f3894Sahrens ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref || 826fa9e4066Sahrens MUTEX_HELD(&spa_namespace_lock)); 827fa9e4066Sahrens (void) refcount_add(&spa->spa_refcount, tag); 828fa9e4066Sahrens } 829fa9e4066Sahrens 830fa9e4066Sahrens /* 831fa9e4066Sahrens * Remove a reference to the given spa_t. Must have at least one reference, or 832fa9e4066Sahrens * have the namespace lock held. 833fa9e4066Sahrens */ 834fa9e4066Sahrens void 835fa9e4066Sahrens spa_close(spa_t *spa, void *tag) 836fa9e4066Sahrens { 837088f3894Sahrens ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref || 838fa9e4066Sahrens MUTEX_HELD(&spa_namespace_lock)); 839fa9e4066Sahrens (void) refcount_remove(&spa->spa_refcount, tag); 840fa9e4066Sahrens } 841fa9e4066Sahrens 842bc9014e6SJustin Gibbs /* 843bc9014e6SJustin Gibbs * Remove a reference to the given spa_t held by a dsl dir that is 844bc9014e6SJustin Gibbs * being asynchronously released. Async releases occur from a taskq 845bc9014e6SJustin Gibbs * performing eviction of dsl datasets and dirs. The namespace lock 846bc9014e6SJustin Gibbs * isn't held and the hold by the object being evicted may contribute to 847bc9014e6SJustin Gibbs * spa_minref (e.g. dataset or directory released during pool export), 848bc9014e6SJustin Gibbs * so the asserts in spa_close() do not apply. 849bc9014e6SJustin Gibbs */ 850bc9014e6SJustin Gibbs void 851bc9014e6SJustin Gibbs spa_async_close(spa_t *spa, void *tag) 852bc9014e6SJustin Gibbs { 853bc9014e6SJustin Gibbs (void) refcount_remove(&spa->spa_refcount, tag); 854bc9014e6SJustin Gibbs } 855bc9014e6SJustin Gibbs 856fa9e4066Sahrens /* 857fa9e4066Sahrens * Check to see if the spa refcount is zero. Must be called with 858088f3894Sahrens * spa_namespace_lock held. We really compare against spa_minref, which is the 859fa9e4066Sahrens * number of references acquired when opening a pool 860fa9e4066Sahrens */ 861fa9e4066Sahrens boolean_t 862fa9e4066Sahrens spa_refcount_zero(spa_t *spa) 863fa9e4066Sahrens { 864fa9e4066Sahrens ASSERT(MUTEX_HELD(&spa_namespace_lock)); 865fa9e4066Sahrens 866088f3894Sahrens return (refcount_count(&spa->spa_refcount) == spa->spa_minref); 867fa9e4066Sahrens } 868fa9e4066Sahrens 86999653d4eSeschrock /* 87099653d4eSeschrock * ========================================================================== 871fa94a07fSbrendan * SPA spare and l2cache tracking 87299653d4eSeschrock * ========================================================================== 87399653d4eSeschrock */ 87499653d4eSeschrock 875fa94a07fSbrendan /* 876fa94a07fSbrendan * Hot spares and cache devices are tracked using the same code below, 877fa94a07fSbrendan * for 'auxiliary' devices. 878fa94a07fSbrendan */ 879fa94a07fSbrendan 880fa94a07fSbrendan typedef struct spa_aux { 881fa94a07fSbrendan uint64_t aux_guid; 882fa94a07fSbrendan uint64_t aux_pool; 883fa94a07fSbrendan avl_node_t aux_avl; 884fa94a07fSbrendan int aux_count; 885fa94a07fSbrendan } spa_aux_t; 886fa94a07fSbrendan 887fa94a07fSbrendan static int 888fa94a07fSbrendan spa_aux_compare(const void *a, const void *b) 889fa94a07fSbrendan { 890fa94a07fSbrendan const spa_aux_t *sa = a; 891fa94a07fSbrendan const spa_aux_t *sb = b; 892fa94a07fSbrendan 893fa94a07fSbrendan if (sa->aux_guid < sb->aux_guid) 894fa94a07fSbrendan return (-1); 895fa94a07fSbrendan else if (sa->aux_guid > sb->aux_guid) 896fa94a07fSbrendan return (1); 897fa94a07fSbrendan else 898fa94a07fSbrendan return (0); 899fa94a07fSbrendan } 900fa94a07fSbrendan 901fa94a07fSbrendan void 902fa94a07fSbrendan spa_aux_add(vdev_t *vd, avl_tree_t *avl) 903fa94a07fSbrendan { 904fa94a07fSbrendan avl_index_t where; 905fa94a07fSbrendan spa_aux_t search; 906fa94a07fSbrendan spa_aux_t *aux; 907fa94a07fSbrendan 908fa94a07fSbrendan search.aux_guid = vd->vdev_guid; 909fa94a07fSbrendan if ((aux = avl_find(avl, &search, &where)) != NULL) { 910fa94a07fSbrendan aux->aux_count++; 911fa94a07fSbrendan } else { 912fa94a07fSbrendan aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP); 913fa94a07fSbrendan aux->aux_guid = vd->vdev_guid; 914fa94a07fSbrendan aux->aux_count = 1; 915fa94a07fSbrendan avl_insert(avl, aux, where); 916fa94a07fSbrendan } 917fa94a07fSbrendan } 918fa94a07fSbrendan 919fa94a07fSbrendan void 920fa94a07fSbrendan spa_aux_remove(vdev_t *vd, avl_tree_t *avl) 921fa94a07fSbrendan { 922fa94a07fSbrendan spa_aux_t search; 923fa94a07fSbrendan spa_aux_t *aux; 924fa94a07fSbrendan avl_index_t where; 925fa94a07fSbrendan 926fa94a07fSbrendan search.aux_guid = vd->vdev_guid; 927fa94a07fSbrendan aux = avl_find(avl, &search, &where); 928fa94a07fSbrendan 929fa94a07fSbrendan ASSERT(aux != NULL); 930fa94a07fSbrendan 931fa94a07fSbrendan if (--aux->aux_count == 0) { 932fa94a07fSbrendan avl_remove(avl, aux); 933fa94a07fSbrendan kmem_free(aux, sizeof (spa_aux_t)); 934fa94a07fSbrendan } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) { 935fa94a07fSbrendan aux->aux_pool = 0ULL; 936fa94a07fSbrendan } 937fa94a07fSbrendan } 938fa94a07fSbrendan 939fa94a07fSbrendan boolean_t 94089a89ebfSlling spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl) 941fa94a07fSbrendan { 942fa94a07fSbrendan spa_aux_t search, *found; 943fa94a07fSbrendan 944fa94a07fSbrendan search.aux_guid = guid; 94589a89ebfSlling found = avl_find(avl, &search, NULL); 946fa94a07fSbrendan 947fa94a07fSbrendan if (pool) { 948fa94a07fSbrendan if (found) 949fa94a07fSbrendan *pool = found->aux_pool; 950fa94a07fSbrendan else 951fa94a07fSbrendan *pool = 0ULL; 952fa94a07fSbrendan } 953fa94a07fSbrendan 95489a89ebfSlling if (refcnt) { 95589a89ebfSlling if (found) 95689a89ebfSlling *refcnt = found->aux_count; 95789a89ebfSlling else 95889a89ebfSlling *refcnt = 0; 95989a89ebfSlling } 96089a89ebfSlling 961fa94a07fSbrendan return (found != NULL); 962fa94a07fSbrendan } 963fa94a07fSbrendan 964fa94a07fSbrendan void 965fa94a07fSbrendan spa_aux_activate(vdev_t *vd, avl_tree_t *avl) 966fa94a07fSbrendan { 967fa94a07fSbrendan spa_aux_t search, *found; 968fa94a07fSbrendan avl_index_t where; 969fa94a07fSbrendan 970fa94a07fSbrendan search.aux_guid = vd->vdev_guid; 971fa94a07fSbrendan found = avl_find(avl, &search, &where); 972fa94a07fSbrendan ASSERT(found != NULL); 973fa94a07fSbrendan ASSERT(found->aux_pool == 0ULL); 974fa94a07fSbrendan 975fa94a07fSbrendan found->aux_pool = spa_guid(vd->vdev_spa); 976fa94a07fSbrendan } 977fa94a07fSbrendan 97899653d4eSeschrock /* 97939c23413Seschrock * Spares are tracked globally due to the following constraints: 98039c23413Seschrock * 98139c23413Seschrock * - A spare may be part of multiple pools. 98239c23413Seschrock * - A spare may be added to a pool even if it's actively in use within 98339c23413Seschrock * another pool. 98439c23413Seschrock * - A spare in use in any pool can only be the source of a replacement if 98539c23413Seschrock * the target is a spare in the same pool. 98639c23413Seschrock * 98739c23413Seschrock * We keep track of all spares on the system through the use of a reference 98839c23413Seschrock * counted AVL tree. When a vdev is added as a spare, or used as a replacement 98939c23413Seschrock * spare, then we bump the reference count in the AVL tree. In addition, we set 99039c23413Seschrock * the 'vdev_isspare' member to indicate that the device is a spare (active or 99139c23413Seschrock * inactive). When a spare is made active (used to replace a device in the 99239c23413Seschrock * pool), we also keep track of which pool its been made a part of. 99339c23413Seschrock * 99439c23413Seschrock * The 'spa_spare_lock' protects the AVL tree. These functions are normally 99539c23413Seschrock * called under the spa_namespace lock as part of vdev reconfiguration. The 99639c23413Seschrock * separate spare lock exists for the status query path, which does not need to 99739c23413Seschrock * be completely consistent with respect to other vdev configuration changes. 99899653d4eSeschrock */ 99939c23413Seschrock 100099653d4eSeschrock static int 100199653d4eSeschrock spa_spare_compare(const void *a, const void *b) 100299653d4eSeschrock { 1003fa94a07fSbrendan return (spa_aux_compare(a, b)); 100499653d4eSeschrock } 100599653d4eSeschrock 100699653d4eSeschrock void 100739c23413Seschrock spa_spare_add(vdev_t *vd) 100899653d4eSeschrock { 100999653d4eSeschrock mutex_enter(&spa_spare_lock); 101039c23413Seschrock ASSERT(!vd->vdev_isspare); 1011fa94a07fSbrendan spa_aux_add(vd, &spa_spare_avl); 101239c23413Seschrock vd->vdev_isspare = B_TRUE; 101399653d4eSeschrock mutex_exit(&spa_spare_lock); 101499653d4eSeschrock } 101599653d4eSeschrock 101699653d4eSeschrock void 101739c23413Seschrock spa_spare_remove(vdev_t *vd) 101899653d4eSeschrock { 101999653d4eSeschrock mutex_enter(&spa_spare_lock); 102039c23413Seschrock ASSERT(vd->vdev_isspare); 1021fa94a07fSbrendan spa_aux_remove(vd, &spa_spare_avl); 102239c23413Seschrock vd->vdev_isspare = B_FALSE; 102399653d4eSeschrock mutex_exit(&spa_spare_lock); 102499653d4eSeschrock } 102599653d4eSeschrock 102699653d4eSeschrock boolean_t 102789a89ebfSlling spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt) 102899653d4eSeschrock { 1029fa94a07fSbrendan boolean_t found; 103099653d4eSeschrock 103199653d4eSeschrock mutex_enter(&spa_spare_lock); 103289a89ebfSlling found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl); 103399653d4eSeschrock mutex_exit(&spa_spare_lock); 103499653d4eSeschrock 1035fa94a07fSbrendan return (found); 103639c23413Seschrock } 103739c23413Seschrock 103839c23413Seschrock void 103939c23413Seschrock spa_spare_activate(vdev_t *vd) 104039c23413Seschrock { 104139c23413Seschrock mutex_enter(&spa_spare_lock); 104239c23413Seschrock ASSERT(vd->vdev_isspare); 1043fa94a07fSbrendan spa_aux_activate(vd, &spa_spare_avl); 1044fa94a07fSbrendan mutex_exit(&spa_spare_lock); 1045fa94a07fSbrendan } 104639c23413Seschrock 1047fa94a07fSbrendan /* 1048fa94a07fSbrendan * Level 2 ARC devices are tracked globally for the same reasons as spares. 1049fa94a07fSbrendan * Cache devices currently only support one pool per cache device, and so 1050fa94a07fSbrendan * for these devices the aux reference count is currently unused beyond 1. 1051fa94a07fSbrendan */ 105239c23413Seschrock 1053fa94a07fSbrendan static int 1054fa94a07fSbrendan spa_l2cache_compare(const void *a, const void *b) 1055fa94a07fSbrendan { 1056fa94a07fSbrendan return (spa_aux_compare(a, b)); 1057fa94a07fSbrendan } 1058fa94a07fSbrendan 1059fa94a07fSbrendan void 1060fa94a07fSbrendan spa_l2cache_add(vdev_t *vd) 1061fa94a07fSbrendan { 1062fa94a07fSbrendan mutex_enter(&spa_l2cache_lock); 1063fa94a07fSbrendan ASSERT(!vd->vdev_isl2cache); 1064fa94a07fSbrendan spa_aux_add(vd, &spa_l2cache_avl); 1065fa94a07fSbrendan vd->vdev_isl2cache = B_TRUE; 1066fa94a07fSbrendan mutex_exit(&spa_l2cache_lock); 1067fa94a07fSbrendan } 1068fa94a07fSbrendan 1069fa94a07fSbrendan void 1070fa94a07fSbrendan spa_l2cache_remove(vdev_t *vd) 1071fa94a07fSbrendan { 1072fa94a07fSbrendan mutex_enter(&spa_l2cache_lock); 1073fa94a07fSbrendan ASSERT(vd->vdev_isl2cache); 1074fa94a07fSbrendan spa_aux_remove(vd, &spa_l2cache_avl); 1075fa94a07fSbrendan vd->vdev_isl2cache = B_FALSE; 1076fa94a07fSbrendan mutex_exit(&spa_l2cache_lock); 1077fa94a07fSbrendan } 1078fa94a07fSbrendan 1079fa94a07fSbrendan boolean_t 1080fa94a07fSbrendan spa_l2cache_exists(uint64_t guid, uint64_t *pool) 1081fa94a07fSbrendan { 1082fa94a07fSbrendan boolean_t found; 1083fa94a07fSbrendan 1084fa94a07fSbrendan mutex_enter(&spa_l2cache_lock); 108589a89ebfSlling found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl); 1086fa94a07fSbrendan mutex_exit(&spa_l2cache_lock); 1087fa94a07fSbrendan 1088fa94a07fSbrendan return (found); 1089fa94a07fSbrendan } 1090fa94a07fSbrendan 1091fa94a07fSbrendan void 1092fa94a07fSbrendan spa_l2cache_activate(vdev_t *vd) 1093fa94a07fSbrendan { 1094fa94a07fSbrendan mutex_enter(&spa_l2cache_lock); 1095fa94a07fSbrendan ASSERT(vd->vdev_isl2cache); 1096fa94a07fSbrendan spa_aux_activate(vd, &spa_l2cache_avl); 1097fa94a07fSbrendan mutex_exit(&spa_l2cache_lock); 1098fa94a07fSbrendan } 1099fa94a07fSbrendan 1100fa9e4066Sahrens /* 1101fa9e4066Sahrens * ========================================================================== 1102fa9e4066Sahrens * SPA vdev locking 1103fa9e4066Sahrens * ========================================================================== 1104fa9e4066Sahrens */ 1105fa9e4066Sahrens 1106fa9e4066Sahrens /* 1107ea8dc4b6Seschrock * Lock the given spa_t for the purpose of adding or removing a vdev. 1108ea8dc4b6Seschrock * Grabs the global spa_namespace_lock plus the spa config lock for writing. 1109fa9e4066Sahrens * It returns the next transaction group for the spa_t. 1110fa9e4066Sahrens */ 1111fa9e4066Sahrens uint64_t 1112fa9e4066Sahrens spa_vdev_enter(spa_t *spa) 1113fa9e4066Sahrens { 1114a1521560SJeff Bonwick mutex_enter(&spa->spa_vdev_top_lock); 1115bbfd46c4SJeff Bonwick mutex_enter(&spa_namespace_lock); 111688ecc943SGeorge Wilson return (spa_vdev_config_enter(spa)); 111788ecc943SGeorge Wilson } 111888ecc943SGeorge Wilson 111988ecc943SGeorge Wilson /* 112088ecc943SGeorge Wilson * Internal implementation for spa_vdev_enter(). Used when a vdev 112188ecc943SGeorge Wilson * operation requires multiple syncs (i.e. removing a device) while 112288ecc943SGeorge Wilson * keeping the spa_namespace_lock held. 112388ecc943SGeorge Wilson */ 112488ecc943SGeorge Wilson uint64_t 112588ecc943SGeorge Wilson spa_vdev_config_enter(spa_t *spa) 112688ecc943SGeorge Wilson { 112788ecc943SGeorge Wilson ASSERT(MUTEX_HELD(&spa_namespace_lock)); 11283d7072f8Seschrock 1129e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); 1130fa9e4066Sahrens 1131fa9e4066Sahrens return (spa_last_synced_txg(spa) + 1); 1132fa9e4066Sahrens } 1133fa9e4066Sahrens 1134fa9e4066Sahrens /* 113588ecc943SGeorge Wilson * Used in combination with spa_vdev_config_enter() to allow the syncing 113688ecc943SGeorge Wilson * of multiple transactions without releasing the spa_namespace_lock. 1137fa9e4066Sahrens */ 113888ecc943SGeorge Wilson void 113988ecc943SGeorge Wilson spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag) 1140fa9e4066Sahrens { 114188ecc943SGeorge Wilson ASSERT(MUTEX_HELD(&spa_namespace_lock)); 114288ecc943SGeorge Wilson 11430e34b6a7Sbonwick int config_changed = B_FALSE; 1144ea8dc4b6Seschrock 11450373e76bSbonwick ASSERT(txg > spa_last_synced_txg(spa)); 11460e34b6a7Sbonwick 1147e14bb325SJeff Bonwick spa->spa_pending_vdev = NULL; 1148e14bb325SJeff Bonwick 11490e34b6a7Sbonwick /* 11500e34b6a7Sbonwick * Reassess the DTLs. 11510e34b6a7Sbonwick */ 11520373e76bSbonwick vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE); 11530e34b6a7Sbonwick 1154e14bb325SJeff Bonwick if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) { 11550e34b6a7Sbonwick config_changed = B_TRUE; 11568f18d1faSGeorge Wilson spa->spa_config_generation++; 11570e34b6a7Sbonwick } 1158ea8dc4b6Seschrock 115988ecc943SGeorge Wilson /* 116088ecc943SGeorge Wilson * Verify the metaslab classes. 116188ecc943SGeorge Wilson */ 1162b24ab676SJeff Bonwick ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0); 1163b24ab676SJeff Bonwick ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0); 116488ecc943SGeorge Wilson 1165e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, spa); 1166fa9e4066Sahrens 116788ecc943SGeorge Wilson /* 116888ecc943SGeorge Wilson * Panic the system if the specified tag requires it. This 116988ecc943SGeorge Wilson * is useful for ensuring that configurations are updated 117088ecc943SGeorge Wilson * transactionally. 117188ecc943SGeorge Wilson */ 117288ecc943SGeorge Wilson if (zio_injection_enabled) 11731195e687SMark J Musante zio_handle_panic_injection(spa, tag, 0); 117488ecc943SGeorge Wilson 1175fa9e4066Sahrens /* 1176fa9e4066Sahrens * Note: this txg_wait_synced() is important because it ensures 1177fa9e4066Sahrens * that there won't be more than one config change per txg. 1178fa9e4066Sahrens * This allows us to use the txg as the generation number. 1179fa9e4066Sahrens */ 1180fa9e4066Sahrens if (error == 0) 1181fa9e4066Sahrens txg_wait_synced(spa->spa_dsl_pool, txg); 1182fa9e4066Sahrens 1183fa9e4066Sahrens if (vd != NULL) { 11840713e232SGeorge Wilson ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL); 11858ad4d6ddSJeff Bonwick spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); 1186fa9e4066Sahrens vdev_free(vd); 11878ad4d6ddSJeff Bonwick spa_config_exit(spa, SCL_ALL, spa); 1188fa9e4066Sahrens } 1189fa9e4066Sahrens 1190fa9e4066Sahrens /* 11910e34b6a7Sbonwick * If the config changed, update the config cache. 1192fa9e4066Sahrens */ 11930e34b6a7Sbonwick if (config_changed) 11945cabbc6bSPrashanth Sreenivasa spa_write_cachefile(spa, B_FALSE, B_TRUE); 119588ecc943SGeorge Wilson } 1196ea8dc4b6Seschrock 119788ecc943SGeorge Wilson /* 119888ecc943SGeorge Wilson * Unlock the spa_t after adding or removing a vdev. Besides undoing the 119988ecc943SGeorge Wilson * locking of spa_vdev_enter(), we also want make sure the transactions have 120088ecc943SGeorge Wilson * synced to disk, and then update the global configuration cache with the new 120188ecc943SGeorge Wilson * information. 120288ecc943SGeorge Wilson */ 120388ecc943SGeorge Wilson int 120488ecc943SGeorge Wilson spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error) 120588ecc943SGeorge Wilson { 120688ecc943SGeorge Wilson spa_vdev_config_exit(spa, vd, txg, error, FTAG); 1207ea8dc4b6Seschrock mutex_exit(&spa_namespace_lock); 1208bbfd46c4SJeff Bonwick mutex_exit(&spa->spa_vdev_top_lock); 1209fa9e4066Sahrens 1210fa9e4066Sahrens return (error); 1211fa9e4066Sahrens } 1212fa9e4066Sahrens 1213e14bb325SJeff Bonwick /* 1214e14bb325SJeff Bonwick * Lock the given spa_t for the purpose of changing vdev state. 1215e14bb325SJeff Bonwick */ 1216e14bb325SJeff Bonwick void 12178f18d1faSGeorge Wilson spa_vdev_state_enter(spa_t *spa, int oplocks) 1218e14bb325SJeff Bonwick { 12198f18d1faSGeorge Wilson int locks = SCL_STATE_ALL | oplocks; 12208f18d1faSGeorge Wilson 1221dcba9f3fSGeorge Wilson /* 1222dcba9f3fSGeorge Wilson * Root pools may need to read of the underlying devfs filesystem 1223dcba9f3fSGeorge Wilson * when opening up a vdev. Unfortunately if we're holding the 1224dcba9f3fSGeorge Wilson * SCL_ZIO lock it will result in a deadlock when we try to issue 1225dcba9f3fSGeorge Wilson * the read from the root filesystem. Instead we "prefetch" 1226dcba9f3fSGeorge Wilson * the associated vnodes that we need prior to opening the 1227dcba9f3fSGeorge Wilson * underlying devices and cache them so that we can prevent 1228dcba9f3fSGeorge Wilson * any I/O when we are doing the actual open. 1229dcba9f3fSGeorge Wilson */ 1230dcba9f3fSGeorge Wilson if (spa_is_root(spa)) { 12319842588bSGeorge Wilson int low = locks & ~(SCL_ZIO - 1); 12329842588bSGeorge Wilson int high = locks & ~low; 12339842588bSGeorge Wilson 12349842588bSGeorge Wilson spa_config_enter(spa, high, spa, RW_WRITER); 1235dcba9f3fSGeorge Wilson vdev_hold(spa->spa_root_vdev); 12369842588bSGeorge Wilson spa_config_enter(spa, low, spa, RW_WRITER); 1237dcba9f3fSGeorge Wilson } else { 1238dcba9f3fSGeorge Wilson spa_config_enter(spa, locks, spa, RW_WRITER); 1239dcba9f3fSGeorge Wilson } 12408f18d1faSGeorge Wilson spa->spa_vdev_locks = locks; 1241e14bb325SJeff Bonwick } 1242e14bb325SJeff Bonwick 1243e14bb325SJeff Bonwick int 1244e14bb325SJeff Bonwick spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error) 1245e14bb325SJeff Bonwick { 1246c6065d0fSGeorge Wilson boolean_t config_changed = B_FALSE; 1247c6065d0fSGeorge Wilson 1248b24ab676SJeff Bonwick if (vd != NULL || error == 0) 1249b24ab676SJeff Bonwick vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev, 1250b24ab676SJeff Bonwick 0, 0, B_FALSE); 1251b24ab676SJeff Bonwick 12528f18d1faSGeorge Wilson if (vd != NULL) { 1253e14bb325SJeff Bonwick vdev_state_dirty(vd->vdev_top); 1254c6065d0fSGeorge Wilson config_changed = B_TRUE; 12558f18d1faSGeorge Wilson spa->spa_config_generation++; 12568f18d1faSGeorge Wilson } 1257e14bb325SJeff Bonwick 1258dcba9f3fSGeorge Wilson if (spa_is_root(spa)) 1259dcba9f3fSGeorge Wilson vdev_rele(spa->spa_root_vdev); 1260dcba9f3fSGeorge Wilson 12618f18d1faSGeorge Wilson ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL); 12628f18d1faSGeorge Wilson spa_config_exit(spa, spa->spa_vdev_locks, spa); 1263e14bb325SJeff Bonwick 12648ad4d6ddSJeff Bonwick /* 12658ad4d6ddSJeff Bonwick * If anything changed, wait for it to sync. This ensures that, 12668ad4d6ddSJeff Bonwick * from the system administrator's perspective, zpool(1M) commands 12678ad4d6ddSJeff Bonwick * are synchronous. This is important for things like zpool offline: 12688ad4d6ddSJeff Bonwick * when the command completes, you expect no further I/O from ZFS. 12698ad4d6ddSJeff Bonwick */ 12708ad4d6ddSJeff Bonwick if (vd != NULL) 12718ad4d6ddSJeff Bonwick txg_wait_synced(spa->spa_dsl_pool, 0); 12728ad4d6ddSJeff Bonwick 1273c6065d0fSGeorge Wilson /* 1274c6065d0fSGeorge Wilson * If the config changed, update the config cache. 1275c6065d0fSGeorge Wilson */ 1276c6065d0fSGeorge Wilson if (config_changed) { 1277c6065d0fSGeorge Wilson mutex_enter(&spa_namespace_lock); 12785cabbc6bSPrashanth Sreenivasa spa_write_cachefile(spa, B_FALSE, B_TRUE); 1279c6065d0fSGeorge Wilson mutex_exit(&spa_namespace_lock); 1280c6065d0fSGeorge Wilson } 1281c6065d0fSGeorge Wilson 1282e14bb325SJeff Bonwick return (error); 1283e14bb325SJeff Bonwick } 1284e14bb325SJeff Bonwick 1285fa9e4066Sahrens /* 1286fa9e4066Sahrens * ========================================================================== 1287fa9e4066Sahrens * Miscellaneous functions 1288fa9e4066Sahrens * ========================================================================== 1289fa9e4066Sahrens */ 1290fa9e4066Sahrens 1291ad135b5dSChristopher Siden void 129243466aaeSMax Grossman spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx) 1293ad135b5dSChristopher Siden { 12942acef22dSMatthew Ahrens if (!nvlist_exists(spa->spa_label_features, feature)) { 12952acef22dSMatthew Ahrens fnvlist_add_boolean(spa->spa_label_features, feature); 129643466aaeSMax Grossman /* 129743466aaeSMax Grossman * When we are creating the pool (tx_txg==TXG_INITIAL), we can't 129843466aaeSMax Grossman * dirty the vdev config because lock SCL_CONFIG is not held. 129943466aaeSMax Grossman * Thankfully, in this case we don't need to dirty the config 130043466aaeSMax Grossman * because it will be written out anyway when we finish 130143466aaeSMax Grossman * creating the pool. 130243466aaeSMax Grossman */ 130343466aaeSMax Grossman if (tx->tx_txg != TXG_INITIAL) 130443466aaeSMax Grossman vdev_config_dirty(spa->spa_root_vdev); 13052acef22dSMatthew Ahrens } 1306ad135b5dSChristopher Siden } 1307ad135b5dSChristopher Siden 1308ad135b5dSChristopher Siden void 1309ad135b5dSChristopher Siden spa_deactivate_mos_feature(spa_t *spa, const char *feature) 1310ad135b5dSChristopher Siden { 13112acef22dSMatthew Ahrens if (nvlist_remove_all(spa->spa_label_features, feature) == 0) 13122acef22dSMatthew Ahrens vdev_config_dirty(spa->spa_root_vdev); 1313ad135b5dSChristopher Siden } 1314ad135b5dSChristopher Siden 1315fa9e4066Sahrens /* 1316fa9e4066Sahrens * Rename a spa_t. 1317fa9e4066Sahrens */ 1318fa9e4066Sahrens int 1319fa9e4066Sahrens spa_rename(const char *name, const char *newname) 1320fa9e4066Sahrens { 1321fa9e4066Sahrens spa_t *spa; 1322fa9e4066Sahrens int err; 1323fa9e4066Sahrens 1324fa9e4066Sahrens /* 1325fa9e4066Sahrens * Lookup the spa_t and grab the config lock for writing. We need to 1326fa9e4066Sahrens * actually open the pool so that we can sync out the necessary labels. 1327fa9e4066Sahrens * It's OK to call spa_open() with the namespace lock held because we 1328ea8dc4b6Seschrock * allow recursive calls for other reasons. 1329fa9e4066Sahrens */ 1330fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 1331fa9e4066Sahrens if ((err = spa_open(name, &spa, FTAG)) != 0) { 1332fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 1333fa9e4066Sahrens return (err); 1334fa9e4066Sahrens } 1335fa9e4066Sahrens 1336e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1337fa9e4066Sahrens 1338fa9e4066Sahrens avl_remove(&spa_namespace_avl, spa); 1339e14bb325SJeff Bonwick (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name)); 1340fa9e4066Sahrens avl_add(&spa_namespace_avl, spa); 1341fa9e4066Sahrens 1342fa9e4066Sahrens /* 1343fa9e4066Sahrens * Sync all labels to disk with the new names by marking the root vdev 1344fa9e4066Sahrens * dirty and waiting for it to sync. It will pick up the new pool name 1345fa9e4066Sahrens * during the sync. 1346fa9e4066Sahrens */ 1347fa9e4066Sahrens vdev_config_dirty(spa->spa_root_vdev); 1348fa9e4066Sahrens 1349e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 1350fa9e4066Sahrens 13510373e76bSbonwick txg_wait_synced(spa->spa_dsl_pool, 0); 1352fa9e4066Sahrens 1353fa9e4066Sahrens /* 1354fa9e4066Sahrens * Sync the updated config cache. 1355fa9e4066Sahrens */ 13565cabbc6bSPrashanth Sreenivasa spa_write_cachefile(spa, B_FALSE, B_TRUE); 1357fa9e4066Sahrens 1358fa9e4066Sahrens spa_close(spa, FTAG); 1359fa9e4066Sahrens 1360fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 1361fa9e4066Sahrens 1362fa9e4066Sahrens return (0); 1363fa9e4066Sahrens } 1364fa9e4066Sahrens 1365fa9e4066Sahrens /* 1366f9af39baSGeorge Wilson * Return the spa_t associated with given pool_guid, if it exists. If 1367f9af39baSGeorge Wilson * device_guid is non-zero, determine whether the pool exists *and* contains 1368f9af39baSGeorge Wilson * a device with the specified device_guid. 1369fa9e4066Sahrens */ 1370f9af39baSGeorge Wilson spa_t * 1371f9af39baSGeorge Wilson spa_by_guid(uint64_t pool_guid, uint64_t device_guid) 1372fa9e4066Sahrens { 1373fa9e4066Sahrens spa_t *spa; 1374fa9e4066Sahrens avl_tree_t *t = &spa_namespace_avl; 1375fa9e4066Sahrens 1376ea8dc4b6Seschrock ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1377fa9e4066Sahrens 1378fa9e4066Sahrens for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) { 1379fa9e4066Sahrens if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1380fa9e4066Sahrens continue; 1381fa9e4066Sahrens if (spa->spa_root_vdev == NULL) 1382fa9e4066Sahrens continue; 138339c23413Seschrock if (spa_guid(spa) == pool_guid) { 138439c23413Seschrock if (device_guid == 0) 138539c23413Seschrock break; 138639c23413Seschrock 138739c23413Seschrock if (vdev_lookup_by_guid(spa->spa_root_vdev, 138839c23413Seschrock device_guid) != NULL) 138939c23413Seschrock break; 139039c23413Seschrock 139139c23413Seschrock /* 13928654d025Sperrin * Check any devices we may be in the process of adding. 139339c23413Seschrock */ 139439c23413Seschrock if (spa->spa_pending_vdev) { 139539c23413Seschrock if (vdev_lookup_by_guid(spa->spa_pending_vdev, 139639c23413Seschrock device_guid) != NULL) 139739c23413Seschrock break; 139839c23413Seschrock } 139939c23413Seschrock } 1400fa9e4066Sahrens } 1401fa9e4066Sahrens 1402f9af39baSGeorge Wilson return (spa); 1403f9af39baSGeorge Wilson } 1404f9af39baSGeorge Wilson 1405f9af39baSGeorge Wilson /* 1406f9af39baSGeorge Wilson * Determine whether a pool with the given pool_guid exists. 1407f9af39baSGeorge Wilson */ 1408f9af39baSGeorge Wilson boolean_t 1409f9af39baSGeorge Wilson spa_guid_exists(uint64_t pool_guid, uint64_t device_guid) 1410f9af39baSGeorge Wilson { 1411f9af39baSGeorge Wilson return (spa_by_guid(pool_guid, device_guid) != NULL); 1412fa9e4066Sahrens } 1413fa9e4066Sahrens 1414fa9e4066Sahrens char * 1415fa9e4066Sahrens spa_strdup(const char *s) 1416fa9e4066Sahrens { 1417fa9e4066Sahrens size_t len; 1418fa9e4066Sahrens char *new; 1419fa9e4066Sahrens 1420fa9e4066Sahrens len = strlen(s); 1421fa9e4066Sahrens new = kmem_alloc(len + 1, KM_SLEEP); 1422fa9e4066Sahrens bcopy(s, new, len); 1423fa9e4066Sahrens new[len] = '\0'; 1424fa9e4066Sahrens 1425fa9e4066Sahrens return (new); 1426fa9e4066Sahrens } 1427fa9e4066Sahrens 1428fa9e4066Sahrens void 1429fa9e4066Sahrens spa_strfree(char *s) 1430fa9e4066Sahrens { 1431fa9e4066Sahrens kmem_free(s, strlen(s) + 1); 1432fa9e4066Sahrens } 1433fa9e4066Sahrens 1434fa9e4066Sahrens uint64_t 1435fa9e4066Sahrens spa_get_random(uint64_t range) 1436fa9e4066Sahrens { 1437fa9e4066Sahrens uint64_t r; 1438fa9e4066Sahrens 1439fa9e4066Sahrens ASSERT(range != 0); 1440fa9e4066Sahrens 1441fa9e4066Sahrens (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t)); 1442fa9e4066Sahrens 1443fa9e4066Sahrens return (r % range); 1444fa9e4066Sahrens } 1445fa9e4066Sahrens 14461195e687SMark J Musante uint64_t 14471195e687SMark J Musante spa_generate_guid(spa_t *spa) 14481195e687SMark J Musante { 14491195e687SMark J Musante uint64_t guid = spa_get_random(-1ULL); 14501195e687SMark J Musante 14511195e687SMark J Musante if (spa != NULL) { 14521195e687SMark J Musante while (guid == 0 || spa_guid_exists(spa_guid(spa), guid)) 14531195e687SMark J Musante guid = spa_get_random(-1ULL); 14541195e687SMark J Musante } else { 14551195e687SMark J Musante while (guid == 0 || spa_guid_exists(guid, 0)) 14561195e687SMark J Musante guid = spa_get_random(-1ULL); 14571195e687SMark J Musante } 14581195e687SMark J Musante 14591195e687SMark J Musante return (guid); 14601195e687SMark J Musante } 14611195e687SMark J Musante 1462fa9e4066Sahrens void 146343466aaeSMax Grossman snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp) 1464fa9e4066Sahrens { 1465ad135b5dSChristopher Siden char type[256]; 1466f0ba89beSJeff Bonwick char *checksum = NULL; 1467f0ba89beSJeff Bonwick char *compress = NULL; 1468f0ba89beSJeff Bonwick 1469f0ba89beSJeff Bonwick if (bp != NULL) { 1470ad135b5dSChristopher Siden if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) { 1471ad135b5dSChristopher Siden dmu_object_byteswap_t bswap = 1472ad135b5dSChristopher Siden DMU_OT_BYTESWAP(BP_GET_TYPE(bp)); 1473ad135b5dSChristopher Siden (void) snprintf(type, sizeof (type), "bswap %s %s", 1474ad135b5dSChristopher Siden DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ? 1475ad135b5dSChristopher Siden "metadata" : "data", 1476ad135b5dSChristopher Siden dmu_ot_byteswap[bswap].ob_name); 1477ad135b5dSChristopher Siden } else { 1478ad135b5dSChristopher Siden (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name, 1479ad135b5dSChristopher Siden sizeof (type)); 1480ad135b5dSChristopher Siden } 14815d7b4d43SMatthew Ahrens if (!BP_IS_EMBEDDED(bp)) { 14825d7b4d43SMatthew Ahrens checksum = 14835d7b4d43SMatthew Ahrens zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name; 14845d7b4d43SMatthew Ahrens } 1485f0ba89beSJeff Bonwick compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name; 1486f0ba89beSJeff Bonwick } 1487fa9e4066Sahrens 148843466aaeSMax Grossman SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum, 148943466aaeSMax Grossman compress); 1490fa9e4066Sahrens } 1491fa9e4066Sahrens 1492fa9e4066Sahrens void 1493fa9e4066Sahrens spa_freeze(spa_t *spa) 1494fa9e4066Sahrens { 1495fa9e4066Sahrens uint64_t freeze_txg = 0; 1496fa9e4066Sahrens 1497e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1498fa9e4066Sahrens if (spa->spa_freeze_txg == UINT64_MAX) { 1499fa9e4066Sahrens freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE; 1500fa9e4066Sahrens spa->spa_freeze_txg = freeze_txg; 1501fa9e4066Sahrens } 1502e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 1503fa9e4066Sahrens if (freeze_txg != 0) 1504fa9e4066Sahrens txg_wait_synced(spa_get_dsl(spa), freeze_txg); 1505fa9e4066Sahrens } 1506fa9e4066Sahrens 15070125049cSahrens void 15080125049cSahrens zfs_panic_recover(const char *fmt, ...) 15090125049cSahrens { 15100125049cSahrens va_list adx; 15110125049cSahrens 15120125049cSahrens va_start(adx, fmt); 15130125049cSahrens vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx); 15140125049cSahrens va_end(adx); 15150125049cSahrens } 15160125049cSahrens 15173f9d6ad7SLin Ling /* 15183f9d6ad7SLin Ling * This is a stripped-down version of strtoull, suitable only for converting 1519f7170741SWill Andrews * lowercase hexadecimal numbers that don't overflow. 15203f9d6ad7SLin Ling */ 15213f9d6ad7SLin Ling uint64_t 15224585130bSYuri Pankov zfs_strtonum(const char *str, char **nptr) 15233f9d6ad7SLin Ling { 15243f9d6ad7SLin Ling uint64_t val = 0; 15253f9d6ad7SLin Ling char c; 15263f9d6ad7SLin Ling int digit; 15273f9d6ad7SLin Ling 15283f9d6ad7SLin Ling while ((c = *str) != '\0') { 15293f9d6ad7SLin Ling if (c >= '0' && c <= '9') 15303f9d6ad7SLin Ling digit = c - '0'; 15313f9d6ad7SLin Ling else if (c >= 'a' && c <= 'f') 15323f9d6ad7SLin Ling digit = 10 + c - 'a'; 15333f9d6ad7SLin Ling else 15343f9d6ad7SLin Ling break; 15353f9d6ad7SLin Ling 15363f9d6ad7SLin Ling val *= 16; 15373f9d6ad7SLin Ling val += digit; 15383f9d6ad7SLin Ling 15393f9d6ad7SLin Ling str++; 15403f9d6ad7SLin Ling } 15413f9d6ad7SLin Ling 15423f9d6ad7SLin Ling if (nptr) 15433f9d6ad7SLin Ling *nptr = (char *)str; 15443f9d6ad7SLin Ling 15453f9d6ad7SLin Ling return (val); 15463f9d6ad7SLin Ling } 15473f9d6ad7SLin Ling 1548fa9e4066Sahrens /* 1549fa9e4066Sahrens * ========================================================================== 1550fa9e4066Sahrens * Accessor functions 1551fa9e4066Sahrens * ========================================================================== 1552fa9e4066Sahrens */ 1553fa9e4066Sahrens 1554088f3894Sahrens boolean_t 155588b7b0f2SMatthew Ahrens spa_shutting_down(spa_t *spa) 1556fa9e4066Sahrens { 155788b7b0f2SMatthew Ahrens return (spa->spa_async_suspended); 1558fa9e4066Sahrens } 1559fa9e4066Sahrens 1560fa9e4066Sahrens dsl_pool_t * 1561fa9e4066Sahrens spa_get_dsl(spa_t *spa) 1562fa9e4066Sahrens { 1563fa9e4066Sahrens return (spa->spa_dsl_pool); 1564fa9e4066Sahrens } 1565fa9e4066Sahrens 1566ad135b5dSChristopher Siden boolean_t 1567ad135b5dSChristopher Siden spa_is_initializing(spa_t *spa) 1568ad135b5dSChristopher Siden { 1569ad135b5dSChristopher Siden return (spa->spa_is_initializing); 1570ad135b5dSChristopher Siden } 1571ad135b5dSChristopher Siden 15725cabbc6bSPrashanth Sreenivasa boolean_t 15735cabbc6bSPrashanth Sreenivasa spa_indirect_vdevs_loaded(spa_t *spa) 15745cabbc6bSPrashanth Sreenivasa { 15755cabbc6bSPrashanth Sreenivasa return (spa->spa_indirect_vdevs_loaded); 15765cabbc6bSPrashanth Sreenivasa } 15775cabbc6bSPrashanth Sreenivasa 1578fa9e4066Sahrens blkptr_t * 1579fa9e4066Sahrens spa_get_rootblkptr(spa_t *spa) 1580fa9e4066Sahrens { 1581fa9e4066Sahrens return (&spa->spa_ubsync.ub_rootbp); 1582fa9e4066Sahrens } 1583fa9e4066Sahrens 1584fa9e4066Sahrens void 1585fa9e4066Sahrens spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp) 1586fa9e4066Sahrens { 1587fa9e4066Sahrens spa->spa_uberblock.ub_rootbp = *bp; 1588fa9e4066Sahrens } 1589fa9e4066Sahrens 1590fa9e4066Sahrens void 1591fa9e4066Sahrens spa_altroot(spa_t *spa, char *buf, size_t buflen) 1592fa9e4066Sahrens { 1593fa9e4066Sahrens if (spa->spa_root == NULL) 1594fa9e4066Sahrens buf[0] = '\0'; 1595fa9e4066Sahrens else 1596fa9e4066Sahrens (void) strncpy(buf, spa->spa_root, buflen); 1597fa9e4066Sahrens } 1598fa9e4066Sahrens 1599fa9e4066Sahrens int 1600fa9e4066Sahrens spa_sync_pass(spa_t *spa) 1601fa9e4066Sahrens { 1602fa9e4066Sahrens return (spa->spa_sync_pass); 1603fa9e4066Sahrens } 1604fa9e4066Sahrens 1605fa9e4066Sahrens char * 1606fa9e4066Sahrens spa_name(spa_t *spa) 1607fa9e4066Sahrens { 1608fa9e4066Sahrens return (spa->spa_name); 1609fa9e4066Sahrens } 1610fa9e4066Sahrens 1611fa9e4066Sahrens uint64_t 1612fa9e4066Sahrens spa_guid(spa_t *spa) 1613fa9e4066Sahrens { 1614dfbb9432SGeorge Wilson dsl_pool_t *dp = spa_get_dsl(spa); 1615dfbb9432SGeorge Wilson uint64_t guid; 1616dfbb9432SGeorge Wilson 1617b5989ec7Seschrock /* 1618b5989ec7Seschrock * If we fail to parse the config during spa_load(), we can go through 1619b5989ec7Seschrock * the error path (which posts an ereport) and end up here with no root 1620e9103aaeSGarrett D'Amore * vdev. We stash the original pool guid in 'spa_config_guid' to handle 1621b5989ec7Seschrock * this case. 1622b5989ec7Seschrock */ 1623dfbb9432SGeorge Wilson if (spa->spa_root_vdev == NULL) 1624dfbb9432SGeorge Wilson return (spa->spa_config_guid); 1625dfbb9432SGeorge Wilson 1626dfbb9432SGeorge Wilson guid = spa->spa_last_synced_guid != 0 ? 1627dfbb9432SGeorge Wilson spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid; 1628dfbb9432SGeorge Wilson 1629dfbb9432SGeorge Wilson /* 1630dfbb9432SGeorge Wilson * Return the most recently synced out guid unless we're 1631dfbb9432SGeorge Wilson * in syncing context. 1632dfbb9432SGeorge Wilson */ 1633dfbb9432SGeorge Wilson if (dp && dsl_pool_sync_context(dp)) 1634b5989ec7Seschrock return (spa->spa_root_vdev->vdev_guid); 1635b5989ec7Seschrock else 1636dfbb9432SGeorge Wilson return (guid); 1637e9103aaeSGarrett D'Amore } 1638e9103aaeSGarrett D'Amore 1639e9103aaeSGarrett D'Amore uint64_t 1640e9103aaeSGarrett D'Amore spa_load_guid(spa_t *spa) 1641e9103aaeSGarrett D'Amore { 1642e9103aaeSGarrett D'Amore /* 1643e9103aaeSGarrett D'Amore * This is a GUID that exists solely as a reference for the 1644e9103aaeSGarrett D'Amore * purposes of the arc. It is generated at load time, and 1645e9103aaeSGarrett D'Amore * is never written to persistent storage. 1646e9103aaeSGarrett D'Amore */ 1647e9103aaeSGarrett D'Amore return (spa->spa_load_guid); 1648fa9e4066Sahrens } 1649fa9e4066Sahrens 1650fa9e4066Sahrens uint64_t 1651fa9e4066Sahrens spa_last_synced_txg(spa_t *spa) 1652fa9e4066Sahrens { 1653fa9e4066Sahrens return (spa->spa_ubsync.ub_txg); 1654fa9e4066Sahrens } 1655fa9e4066Sahrens 1656fa9e4066Sahrens uint64_t 1657fa9e4066Sahrens spa_first_txg(spa_t *spa) 1658fa9e4066Sahrens { 1659fa9e4066Sahrens return (spa->spa_first_txg); 1660fa9e4066Sahrens } 1661fa9e4066Sahrens 1662b24ab676SJeff Bonwick uint64_t 1663b24ab676SJeff Bonwick spa_syncing_txg(spa_t *spa) 1664b24ab676SJeff Bonwick { 1665b24ab676SJeff Bonwick return (spa->spa_syncing_txg); 1666b24ab676SJeff Bonwick } 1667b24ab676SJeff Bonwick 16683991b535SGeorge Wilson /* 16693991b535SGeorge Wilson * Return the last txg where data can be dirtied. The final txgs 16703991b535SGeorge Wilson * will be used to just clear out any deferred frees that remain. 16713991b535SGeorge Wilson */ 16723991b535SGeorge Wilson uint64_t 16733991b535SGeorge Wilson spa_final_dirty_txg(spa_t *spa) 16743991b535SGeorge Wilson { 16753991b535SGeorge Wilson return (spa->spa_final_txg - TXG_DEFER_SIZE); 16763991b535SGeorge Wilson } 16773991b535SGeorge Wilson 167888b7b0f2SMatthew Ahrens pool_state_t 1679fa9e4066Sahrens spa_state(spa_t *spa) 1680fa9e4066Sahrens { 1681fa9e4066Sahrens return (spa->spa_state); 1682fa9e4066Sahrens } 1683fa9e4066Sahrens 1684b16da2e2SGeorge Wilson spa_load_state_t 1685b16da2e2SGeorge Wilson spa_load_state(spa_t *spa) 1686b16da2e2SGeorge Wilson { 1687b16da2e2SGeorge Wilson return (spa->spa_load_state); 1688b16da2e2SGeorge Wilson } 1689b16da2e2SGeorge Wilson 1690fa9e4066Sahrens uint64_t 1691fa9e4066Sahrens spa_freeze_txg(spa_t *spa) 1692fa9e4066Sahrens { 1693fa9e4066Sahrens return (spa->spa_freeze_txg); 1694fa9e4066Sahrens } 1695fa9e4066Sahrens 1696fa9e4066Sahrens /* ARGSUSED */ 1697fa9e4066Sahrens uint64_t 169861e255ceSMatthew Ahrens spa_get_worst_case_asize(spa_t *spa, uint64_t lsize) 1699fa9e4066Sahrens { 170069962b56SMatthew Ahrens return (lsize * spa_asize_inflation); 170144cd46caSbillm } 170244cd46caSbillm 17037d46dc6cSMatthew Ahrens /* 17047d46dc6cSMatthew Ahrens * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%), 17054b5c8e93SMatthew Ahrens * or at least 128MB, unless that would cause it to be more than half the 17064b5c8e93SMatthew Ahrens * pool size. 17077d46dc6cSMatthew Ahrens * 17087d46dc6cSMatthew Ahrens * See the comment above spa_slop_shift for details. 17097d46dc6cSMatthew Ahrens */ 17107d46dc6cSMatthew Ahrens uint64_t 17114b5c8e93SMatthew Ahrens spa_get_slop_space(spa_t *spa) 17124b5c8e93SMatthew Ahrens { 17137d46dc6cSMatthew Ahrens uint64_t space = spa_get_dspace(spa); 17144b5c8e93SMatthew Ahrens return (MAX(space >> spa_slop_shift, MIN(space >> 1, spa_min_slop))); 17157d46dc6cSMatthew Ahrens } 17167d46dc6cSMatthew Ahrens 1717485bbbf5SGeorge Wilson uint64_t 1718485bbbf5SGeorge Wilson spa_get_dspace(spa_t *spa) 1719485bbbf5SGeorge Wilson { 1720485bbbf5SGeorge Wilson return (spa->spa_dspace); 1721485bbbf5SGeorge Wilson } 1722485bbbf5SGeorge Wilson 1723*86714001SSerapheim Dimitropoulos uint64_t 1724*86714001SSerapheim Dimitropoulos spa_get_checkpoint_space(spa_t *spa) 1725*86714001SSerapheim Dimitropoulos { 1726*86714001SSerapheim Dimitropoulos return (spa->spa_checkpoint_info.sci_dspace); 1727*86714001SSerapheim Dimitropoulos } 1728*86714001SSerapheim Dimitropoulos 1729485bbbf5SGeorge Wilson void 1730485bbbf5SGeorge Wilson spa_update_dspace(spa_t *spa) 1731485bbbf5SGeorge Wilson { 1732485bbbf5SGeorge Wilson spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) + 1733485bbbf5SGeorge Wilson ddt_get_dedup_dspace(spa); 17345cabbc6bSPrashanth Sreenivasa if (spa->spa_vdev_removal != NULL) { 17355cabbc6bSPrashanth Sreenivasa /* 17365cabbc6bSPrashanth Sreenivasa * We can't allocate from the removing device, so 17375cabbc6bSPrashanth Sreenivasa * subtract its size. This prevents the DMU/DSL from 17385cabbc6bSPrashanth Sreenivasa * filling up the (now smaller) pool while we are in the 17395cabbc6bSPrashanth Sreenivasa * middle of removing the device. 17405cabbc6bSPrashanth Sreenivasa * 17415cabbc6bSPrashanth Sreenivasa * Note that the DMU/DSL doesn't actually know or care 17425cabbc6bSPrashanth Sreenivasa * how much space is allocated (it does its own tracking 17435cabbc6bSPrashanth Sreenivasa * of how much space has been logically used). So it 17445cabbc6bSPrashanth Sreenivasa * doesn't matter that the data we are moving may be 17455cabbc6bSPrashanth Sreenivasa * allocated twice (on the old device and the new 17465cabbc6bSPrashanth Sreenivasa * device). 17475cabbc6bSPrashanth Sreenivasa */ 17485cabbc6bSPrashanth Sreenivasa vdev_t *vd = spa->spa_vdev_removal->svr_vdev; 17495cabbc6bSPrashanth Sreenivasa spa->spa_dspace -= spa_deflate(spa) ? 17505cabbc6bSPrashanth Sreenivasa vd->vdev_stat.vs_dspace : vd->vdev_stat.vs_space; 17515cabbc6bSPrashanth Sreenivasa } 1752485bbbf5SGeorge Wilson } 1753485bbbf5SGeorge Wilson 17540a4e9518Sgw /* 17550a4e9518Sgw * Return the failure mode that has been set to this pool. The default 17560a4e9518Sgw * behavior will be to block all I/Os when a complete failure occurs. 17570a4e9518Sgw */ 17580a4e9518Sgw uint8_t 17590a4e9518Sgw spa_get_failmode(spa_t *spa) 17600a4e9518Sgw { 17610a4e9518Sgw return (spa->spa_failmode); 17620a4e9518Sgw } 17630a4e9518Sgw 1764e14bb325SJeff Bonwick boolean_t 1765e14bb325SJeff Bonwick spa_suspended(spa_t *spa) 1766e14bb325SJeff Bonwick { 1767e14bb325SJeff Bonwick return (spa->spa_suspended); 1768e14bb325SJeff Bonwick } 1769e14bb325SJeff Bonwick 177044cd46caSbillm uint64_t 177144cd46caSbillm spa_version(spa_t *spa) 177244cd46caSbillm { 177344cd46caSbillm return (spa->spa_ubsync.ub_version); 177444cd46caSbillm } 177544cd46caSbillm 1776b24ab676SJeff Bonwick boolean_t 1777b24ab676SJeff Bonwick spa_deflate(spa_t *spa) 1778b24ab676SJeff Bonwick { 1779b24ab676SJeff Bonwick return (spa->spa_deflate); 1780b24ab676SJeff Bonwick } 1781b24ab676SJeff Bonwick 1782b24ab676SJeff Bonwick metaslab_class_t * 1783b24ab676SJeff Bonwick spa_normal_class(spa_t *spa) 1784b24ab676SJeff Bonwick { 1785b24ab676SJeff Bonwick return (spa->spa_normal_class); 1786b24ab676SJeff Bonwick } 1787b24ab676SJeff Bonwick 1788b24ab676SJeff Bonwick metaslab_class_t * 1789b24ab676SJeff Bonwick spa_log_class(spa_t *spa) 1790b24ab676SJeff Bonwick { 1791b24ab676SJeff Bonwick return (spa->spa_log_class); 1792b24ab676SJeff Bonwick } 1793b24ab676SJeff Bonwick 1794bc9014e6SJustin Gibbs void 1795bc9014e6SJustin Gibbs spa_evicting_os_register(spa_t *spa, objset_t *os) 1796bc9014e6SJustin Gibbs { 1797bc9014e6SJustin Gibbs mutex_enter(&spa->spa_evicting_os_lock); 1798bc9014e6SJustin Gibbs list_insert_head(&spa->spa_evicting_os_list, os); 1799bc9014e6SJustin Gibbs mutex_exit(&spa->spa_evicting_os_lock); 1800bc9014e6SJustin Gibbs } 1801bc9014e6SJustin Gibbs 1802bc9014e6SJustin Gibbs void 1803bc9014e6SJustin Gibbs spa_evicting_os_deregister(spa_t *spa, objset_t *os) 1804bc9014e6SJustin Gibbs { 1805bc9014e6SJustin Gibbs mutex_enter(&spa->spa_evicting_os_lock); 1806bc9014e6SJustin Gibbs list_remove(&spa->spa_evicting_os_list, os); 1807bc9014e6SJustin Gibbs cv_broadcast(&spa->spa_evicting_os_cv); 1808bc9014e6SJustin Gibbs mutex_exit(&spa->spa_evicting_os_lock); 1809bc9014e6SJustin Gibbs } 1810bc9014e6SJustin Gibbs 1811bc9014e6SJustin Gibbs void 1812bc9014e6SJustin Gibbs spa_evicting_os_wait(spa_t *spa) 1813bc9014e6SJustin Gibbs { 1814bc9014e6SJustin Gibbs mutex_enter(&spa->spa_evicting_os_lock); 1815bc9014e6SJustin Gibbs while (!list_is_empty(&spa->spa_evicting_os_list)) 1816bc9014e6SJustin Gibbs cv_wait(&spa->spa_evicting_os_cv, &spa->spa_evicting_os_lock); 1817bc9014e6SJustin Gibbs mutex_exit(&spa->spa_evicting_os_lock); 1818bc9014e6SJustin Gibbs 1819bc9014e6SJustin Gibbs dmu_buf_user_evict_wait(); 1820bc9014e6SJustin Gibbs } 1821bc9014e6SJustin Gibbs 182244cd46caSbillm int 182344cd46caSbillm spa_max_replication(spa_t *spa) 182444cd46caSbillm { 182544cd46caSbillm /* 1826e7437265Sahrens * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to 182744cd46caSbillm * handle BPs with more than one DVA allocated. Set our max 182844cd46caSbillm * replication level accordingly. 1829fa9e4066Sahrens */ 1830e7437265Sahrens if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS) 183144cd46caSbillm return (1); 183244cd46caSbillm return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override)); 1833fa9e4066Sahrens } 1834fa9e4066Sahrens 18353f9d6ad7SLin Ling int 18363f9d6ad7SLin Ling spa_prev_software_version(spa_t *spa) 18373f9d6ad7SLin Ling { 18383f9d6ad7SLin Ling return (spa->spa_prev_software_version); 18393f9d6ad7SLin Ling } 18403f9d6ad7SLin Ling 1841283b8460SGeorge.Wilson uint64_t 1842283b8460SGeorge.Wilson spa_deadman_synctime(spa_t *spa) 1843283b8460SGeorge.Wilson { 1844283b8460SGeorge.Wilson return (spa->spa_deadman_synctime); 1845283b8460SGeorge.Wilson } 1846283b8460SGeorge.Wilson 184799653d4eSeschrock uint64_t 1848b24ab676SJeff Bonwick dva_get_dsize_sync(spa_t *spa, const dva_t *dva) 184999653d4eSeschrock { 1850b24ab676SJeff Bonwick uint64_t asize = DVA_GET_ASIZE(dva); 1851b24ab676SJeff Bonwick uint64_t dsize = asize; 185299653d4eSeschrock 1853b24ab676SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 185499653d4eSeschrock 1855b24ab676SJeff Bonwick if (asize != 0 && spa->spa_deflate) { 1856b24ab676SJeff Bonwick vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva)); 1857b24ab676SJeff Bonwick dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio; 185899653d4eSeschrock } 1859b24ab676SJeff Bonwick 1860b24ab676SJeff Bonwick return (dsize); 1861b24ab676SJeff Bonwick } 1862b24ab676SJeff Bonwick 1863b24ab676SJeff Bonwick uint64_t 1864b24ab676SJeff Bonwick bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp) 1865b24ab676SJeff Bonwick { 1866b24ab676SJeff Bonwick uint64_t dsize = 0; 1867b24ab676SJeff Bonwick 18685d7b4d43SMatthew Ahrens for (int d = 0; d < BP_GET_NDVAS(bp); d++) 1869b24ab676SJeff Bonwick dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); 1870b24ab676SJeff Bonwick 1871b24ab676SJeff Bonwick return (dsize); 1872b24ab676SJeff Bonwick } 1873b24ab676SJeff Bonwick 1874b24ab676SJeff Bonwick uint64_t 1875b24ab676SJeff Bonwick bp_get_dsize(spa_t *spa, const blkptr_t *bp) 1876b24ab676SJeff Bonwick { 1877b24ab676SJeff Bonwick uint64_t dsize = 0; 1878b24ab676SJeff Bonwick 1879b24ab676SJeff Bonwick spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 1880b24ab676SJeff Bonwick 18815d7b4d43SMatthew Ahrens for (int d = 0; d < BP_GET_NDVAS(bp); d++) 1882b24ab676SJeff Bonwick dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); 1883b24ab676SJeff Bonwick 1884e14bb325SJeff Bonwick spa_config_exit(spa, SCL_VDEV, FTAG); 1885b24ab676SJeff Bonwick 1886b24ab676SJeff Bonwick return (dsize); 188799653d4eSeschrock } 188899653d4eSeschrock 1889fa9e4066Sahrens /* 1890fa9e4066Sahrens * ========================================================================== 1891fa9e4066Sahrens * Initialization and Termination 1892fa9e4066Sahrens * ========================================================================== 1893fa9e4066Sahrens */ 1894fa9e4066Sahrens 1895fa9e4066Sahrens static int 1896fa9e4066Sahrens spa_name_compare(const void *a1, const void *a2) 1897fa9e4066Sahrens { 1898fa9e4066Sahrens const spa_t *s1 = a1; 1899fa9e4066Sahrens const spa_t *s2 = a2; 1900fa9e4066Sahrens int s; 1901fa9e4066Sahrens 1902fa9e4066Sahrens s = strcmp(s1->spa_name, s2->spa_name); 1903fa9e4066Sahrens if (s > 0) 1904fa9e4066Sahrens return (1); 1905fa9e4066Sahrens if (s < 0) 1906fa9e4066Sahrens return (-1); 1907fa9e4066Sahrens return (0); 1908fa9e4066Sahrens } 1909fa9e4066Sahrens 19100373e76bSbonwick int 19110373e76bSbonwick spa_busy(void) 19120373e76bSbonwick { 19130373e76bSbonwick return (spa_active_count); 19140373e76bSbonwick } 19150373e76bSbonwick 1916e7cbe64fSgw void 1917e7cbe64fSgw spa_boot_init() 1918e7cbe64fSgw { 1919e7cbe64fSgw spa_config_load(); 1920e7cbe64fSgw } 1921e7cbe64fSgw 1922fa9e4066Sahrens void 1923fa9e4066Sahrens spa_init(int mode) 1924fa9e4066Sahrens { 1925fa9e4066Sahrens mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL); 1926c25056deSgw mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL); 1927fa94a07fSbrendan mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL); 1928fa9e4066Sahrens cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL); 1929fa9e4066Sahrens 1930fa9e4066Sahrens avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t), 1931fa9e4066Sahrens offsetof(spa_t, spa_avl)); 1932fa9e4066Sahrens 1933fa94a07fSbrendan avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t), 1934fa94a07fSbrendan offsetof(spa_aux_t, aux_avl)); 1935fa94a07fSbrendan 1936fa94a07fSbrendan avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t), 1937fa94a07fSbrendan offsetof(spa_aux_t, aux_avl)); 193899653d4eSeschrock 19398ad4d6ddSJeff Bonwick spa_mode_global = mode; 1940fa9e4066Sahrens 1941283b8460SGeorge.Wilson #ifdef _KERNEL 1942283b8460SGeorge.Wilson spa_arch_init(); 1943283b8460SGeorge.Wilson #else 1944cd1c8b85SMatthew Ahrens if (spa_mode_global != FREAD && dprintf_find_string("watch")) { 1945cd1c8b85SMatthew Ahrens arc_procfd = open("/proc/self/ctl", O_WRONLY); 1946cd1c8b85SMatthew Ahrens if (arc_procfd == -1) { 1947cd1c8b85SMatthew Ahrens perror("could not enable watchpoints: " 1948cd1c8b85SMatthew Ahrens "opening /proc/self/ctl failed: "); 1949cd1c8b85SMatthew Ahrens } else { 1950cd1c8b85SMatthew Ahrens arc_watch = B_TRUE; 1951cd1c8b85SMatthew Ahrens } 1952cd1c8b85SMatthew Ahrens } 1953cd1c8b85SMatthew Ahrens #endif 1954cd1c8b85SMatthew Ahrens 1955fa9e4066Sahrens refcount_init(); 1956fa9e4066Sahrens unique_init(); 19570713e232SGeorge Wilson range_tree_init(); 19588363e80aSGeorge Wilson metaslab_alloc_trace_init(); 1959fa9e4066Sahrens zio_init(); 1960fa9e4066Sahrens dmu_init(); 1961fa9e4066Sahrens zil_init(); 196287db74c1Sek vdev_cache_stat_init(); 196391ebeef5Sahrens zfs_prop_init(); 1964990b4856Slling zpool_prop_init(); 1965ad135b5dSChristopher Siden zpool_feature_init(); 1966fa9e4066Sahrens spa_config_load(); 1967e14bb325SJeff Bonwick l2arc_start(); 1968fa9e4066Sahrens } 1969fa9e4066Sahrens 1970fa9e4066Sahrens void 1971fa9e4066Sahrens spa_fini(void) 1972fa9e4066Sahrens { 1973e14bb325SJeff Bonwick l2arc_stop(); 1974e14bb325SJeff Bonwick 1975fa9e4066Sahrens spa_evict_all(); 1976fa9e4066Sahrens 197787db74c1Sek vdev_cache_stat_fini(); 1978fa9e4066Sahrens zil_fini(); 1979fa9e4066Sahrens dmu_fini(); 1980fa9e4066Sahrens zio_fini(); 19818363e80aSGeorge Wilson metaslab_alloc_trace_fini(); 19820713e232SGeorge Wilson range_tree_fini(); 198391ebeef5Sahrens unique_fini(); 1984fa9e4066Sahrens refcount_fini(); 1985fa9e4066Sahrens 1986fa9e4066Sahrens avl_destroy(&spa_namespace_avl); 198799653d4eSeschrock avl_destroy(&spa_spare_avl); 1988fa94a07fSbrendan avl_destroy(&spa_l2cache_avl); 1989fa9e4066Sahrens 1990fa9e4066Sahrens cv_destroy(&spa_namespace_cv); 1991fa9e4066Sahrens mutex_destroy(&spa_namespace_lock); 1992c25056deSgw mutex_destroy(&spa_spare_lock); 1993fa94a07fSbrendan mutex_destroy(&spa_l2cache_lock); 1994fa9e4066Sahrens } 19956ce0521aSperrin 19966ce0521aSperrin /* 19976ce0521aSperrin * Return whether this pool has slogs. No locking needed. 19986ce0521aSperrin * It's not a problem if the wrong answer is returned as it's only for 19996ce0521aSperrin * performance and not correctness 20006ce0521aSperrin */ 20016ce0521aSperrin boolean_t 20026ce0521aSperrin spa_has_slogs(spa_t *spa) 20036ce0521aSperrin { 20046ce0521aSperrin return (spa->spa_log_class->mc_rotor != NULL); 20056ce0521aSperrin } 2006bf82a41bSeschrock 2007b24ab676SJeff Bonwick spa_log_state_t 2008b24ab676SJeff Bonwick spa_get_log_state(spa_t *spa) 2009b24ab676SJeff Bonwick { 2010b24ab676SJeff Bonwick return (spa->spa_log_state); 2011b24ab676SJeff Bonwick } 2012b24ab676SJeff Bonwick 2013b24ab676SJeff Bonwick void 2014b24ab676SJeff Bonwick spa_set_log_state(spa_t *spa, spa_log_state_t state) 2015b24ab676SJeff Bonwick { 2016b24ab676SJeff Bonwick spa->spa_log_state = state; 2017b24ab676SJeff Bonwick } 2018b24ab676SJeff Bonwick 2019bf82a41bSeschrock boolean_t 2020bf82a41bSeschrock spa_is_root(spa_t *spa) 2021bf82a41bSeschrock { 2022bf82a41bSeschrock return (spa->spa_is_root); 2023bf82a41bSeschrock } 20248ad4d6ddSJeff Bonwick 20258ad4d6ddSJeff Bonwick boolean_t 20268ad4d6ddSJeff Bonwick spa_writeable(spa_t *spa) 20278ad4d6ddSJeff Bonwick { 20286f793812SPavel Zakharov return (!!(spa->spa_mode & FWRITE) && spa->spa_trust_config); 20298ad4d6ddSJeff Bonwick } 20308ad4d6ddSJeff Bonwick 203173527f44SAlex Reece /* 203273527f44SAlex Reece * Returns true if there is a pending sync task in any of the current 203373527f44SAlex Reece * syncing txg, the current quiescing txg, or the current open txg. 203473527f44SAlex Reece */ 203573527f44SAlex Reece boolean_t 203673527f44SAlex Reece spa_has_pending_synctask(spa_t *spa) 203773527f44SAlex Reece { 2038*86714001SSerapheim Dimitropoulos return (!txg_all_lists_empty(&spa->spa_dsl_pool->dp_sync_tasks) || 2039*86714001SSerapheim Dimitropoulos !txg_all_lists_empty(&spa->spa_dsl_pool->dp_early_sync_tasks)); 204073527f44SAlex Reece } 204173527f44SAlex Reece 20428ad4d6ddSJeff Bonwick int 20438ad4d6ddSJeff Bonwick spa_mode(spa_t *spa) 20448ad4d6ddSJeff Bonwick { 20458ad4d6ddSJeff Bonwick return (spa->spa_mode); 20468ad4d6ddSJeff Bonwick } 2047b24ab676SJeff Bonwick 2048b24ab676SJeff Bonwick uint64_t 2049b24ab676SJeff Bonwick spa_bootfs(spa_t *spa) 2050b24ab676SJeff Bonwick { 2051b24ab676SJeff Bonwick return (spa->spa_bootfs); 2052b24ab676SJeff Bonwick } 2053b24ab676SJeff Bonwick 2054b24ab676SJeff Bonwick uint64_t 2055b24ab676SJeff Bonwick spa_delegation(spa_t *spa) 2056b24ab676SJeff Bonwick { 2057b24ab676SJeff Bonwick return (spa->spa_delegation); 2058b24ab676SJeff Bonwick } 2059b24ab676SJeff Bonwick 2060b24ab676SJeff Bonwick objset_t * 2061b24ab676SJeff Bonwick spa_meta_objset(spa_t *spa) 2062b24ab676SJeff Bonwick { 2063b24ab676SJeff Bonwick return (spa->spa_meta_objset); 2064b24ab676SJeff Bonwick } 2065b24ab676SJeff Bonwick 2066b24ab676SJeff Bonwick enum zio_checksum 2067b24ab676SJeff Bonwick spa_dedup_checksum(spa_t *spa) 2068b24ab676SJeff Bonwick { 2069b24ab676SJeff Bonwick return (spa->spa_dedup_checksum); 2070b24ab676SJeff Bonwick } 20713f9d6ad7SLin Ling 20723f9d6ad7SLin Ling /* 20733f9d6ad7SLin Ling * Reset pool scan stat per scan pass (or reboot). 20743f9d6ad7SLin Ling */ 20753f9d6ad7SLin Ling void 20763f9d6ad7SLin Ling spa_scan_stat_init(spa_t *spa) 20773f9d6ad7SLin Ling { 20783f9d6ad7SLin Ling /* data not stored on disk */ 20793f9d6ad7SLin Ling spa->spa_scan_pass_start = gethrestime_sec(); 20801702cce7SAlek Pinchuk if (dsl_scan_is_paused_scrub(spa->spa_dsl_pool->dp_scan)) 20811702cce7SAlek Pinchuk spa->spa_scan_pass_scrub_pause = spa->spa_scan_pass_start; 20821702cce7SAlek Pinchuk else 20831702cce7SAlek Pinchuk spa->spa_scan_pass_scrub_pause = 0; 20841702cce7SAlek Pinchuk spa->spa_scan_pass_scrub_spent_paused = 0; 20853f9d6ad7SLin Ling spa->spa_scan_pass_exam = 0; 20863f9d6ad7SLin Ling vdev_scan_stat_init(spa->spa_root_vdev); 20873f9d6ad7SLin Ling } 20883f9d6ad7SLin Ling 20893f9d6ad7SLin Ling /* 20903f9d6ad7SLin Ling * Get scan stats for zpool status reports 20913f9d6ad7SLin Ling */ 20923f9d6ad7SLin Ling int 20933f9d6ad7SLin Ling spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps) 20943f9d6ad7SLin Ling { 20953f9d6ad7SLin Ling dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL; 20963f9d6ad7SLin Ling 20973f9d6ad7SLin Ling if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE) 2098be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 20993f9d6ad7SLin Ling bzero(ps, sizeof (pool_scan_stat_t)); 21003f9d6ad7SLin Ling 21013f9d6ad7SLin Ling /* data stored on disk */ 21023f9d6ad7SLin Ling ps->pss_func = scn->scn_phys.scn_func; 21033f9d6ad7SLin Ling ps->pss_start_time = scn->scn_phys.scn_start_time; 21043f9d6ad7SLin Ling ps->pss_end_time = scn->scn_phys.scn_end_time; 21053f9d6ad7SLin Ling ps->pss_to_examine = scn->scn_phys.scn_to_examine; 21063f9d6ad7SLin Ling ps->pss_examined = scn->scn_phys.scn_examined; 21073f9d6ad7SLin Ling ps->pss_to_process = scn->scn_phys.scn_to_process; 21083f9d6ad7SLin Ling ps->pss_processed = scn->scn_phys.scn_processed; 21093f9d6ad7SLin Ling ps->pss_errors = scn->scn_phys.scn_errors; 21103f9d6ad7SLin Ling ps->pss_state = scn->scn_phys.scn_state; 21113f9d6ad7SLin Ling 21123f9d6ad7SLin Ling /* data not stored on disk */ 21133f9d6ad7SLin Ling ps->pss_pass_start = spa->spa_scan_pass_start; 21143f9d6ad7SLin Ling ps->pss_pass_exam = spa->spa_scan_pass_exam; 21151702cce7SAlek Pinchuk ps->pss_pass_scrub_pause = spa->spa_scan_pass_scrub_pause; 21161702cce7SAlek Pinchuk ps->pss_pass_scrub_spent_paused = spa->spa_scan_pass_scrub_spent_paused; 21173f9d6ad7SLin Ling 21183f9d6ad7SLin Ling return (0); 21193f9d6ad7SLin Ling } 212009c9d376SGeorge Wilson 212109c9d376SGeorge Wilson boolean_t 212209c9d376SGeorge Wilson spa_debug_enabled(spa_t *spa) 212309c9d376SGeorge Wilson { 212409c9d376SGeorge Wilson return (spa->spa_debug); 212509c9d376SGeorge Wilson } 2126b5152584SMatthew Ahrens 2127b5152584SMatthew Ahrens int 2128b5152584SMatthew Ahrens spa_maxblocksize(spa_t *spa) 2129b5152584SMatthew Ahrens { 2130b5152584SMatthew Ahrens if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) 2131b5152584SMatthew Ahrens return (SPA_MAXBLOCKSIZE); 2132b5152584SMatthew Ahrens else 2133b5152584SMatthew Ahrens return (SPA_OLD_MAXBLOCKSIZE); 2134b5152584SMatthew Ahrens } 21355cabbc6bSPrashanth Sreenivasa 21365cabbc6bSPrashanth Sreenivasa /* 21375cabbc6bSPrashanth Sreenivasa * Returns the txg that the last device removal completed. No indirect mappings 21385cabbc6bSPrashanth Sreenivasa * have been added since this txg. 21395cabbc6bSPrashanth Sreenivasa */ 21405cabbc6bSPrashanth Sreenivasa uint64_t 21415cabbc6bSPrashanth Sreenivasa spa_get_last_removal_txg(spa_t *spa) 21425cabbc6bSPrashanth Sreenivasa { 21435cabbc6bSPrashanth Sreenivasa uint64_t vdevid; 21445cabbc6bSPrashanth Sreenivasa uint64_t ret = -1ULL; 21455cabbc6bSPrashanth Sreenivasa 21465cabbc6bSPrashanth Sreenivasa spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 21475cabbc6bSPrashanth Sreenivasa /* 21485cabbc6bSPrashanth Sreenivasa * sr_prev_indirect_vdev is only modified while holding all the 21495cabbc6bSPrashanth Sreenivasa * config locks, so it is sufficient to hold SCL_VDEV as reader when 21505cabbc6bSPrashanth Sreenivasa * examining it. 21515cabbc6bSPrashanth Sreenivasa */ 21525cabbc6bSPrashanth Sreenivasa vdevid = spa->spa_removing_phys.sr_prev_indirect_vdev; 21535cabbc6bSPrashanth Sreenivasa 21545cabbc6bSPrashanth Sreenivasa while (vdevid != -1ULL) { 21555cabbc6bSPrashanth Sreenivasa vdev_t *vd = vdev_lookup_top(spa, vdevid); 21565cabbc6bSPrashanth Sreenivasa vdev_indirect_births_t *vib = vd->vdev_indirect_births; 21575cabbc6bSPrashanth Sreenivasa 21585cabbc6bSPrashanth Sreenivasa ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops); 21595cabbc6bSPrashanth Sreenivasa 21605cabbc6bSPrashanth Sreenivasa /* 21615cabbc6bSPrashanth Sreenivasa * If the removal did not remap any data, we don't care. 21625cabbc6bSPrashanth Sreenivasa */ 21635cabbc6bSPrashanth Sreenivasa if (vdev_indirect_births_count(vib) != 0) { 21645cabbc6bSPrashanth Sreenivasa ret = vdev_indirect_births_last_entry_txg(vib); 21655cabbc6bSPrashanth Sreenivasa break; 21665cabbc6bSPrashanth Sreenivasa } 21675cabbc6bSPrashanth Sreenivasa 21685cabbc6bSPrashanth Sreenivasa vdevid = vd->vdev_indirect_config.vic_prev_indirect_vdev; 21695cabbc6bSPrashanth Sreenivasa } 21705cabbc6bSPrashanth Sreenivasa spa_config_exit(spa, SCL_VDEV, FTAG); 21715cabbc6bSPrashanth Sreenivasa 21725cabbc6bSPrashanth Sreenivasa IMPLY(ret != -1ULL, 21735cabbc6bSPrashanth Sreenivasa spa_feature_is_active(spa, SPA_FEATURE_DEVICE_REMOVAL)); 21745cabbc6bSPrashanth Sreenivasa 21755cabbc6bSPrashanth Sreenivasa return (ret); 21765cabbc6bSPrashanth Sreenivasa } 21776f793812SPavel Zakharov 21786f793812SPavel Zakharov boolean_t 21796f793812SPavel Zakharov spa_trust_config(spa_t *spa) 21806f793812SPavel Zakharov { 21816f793812SPavel Zakharov return (spa->spa_trust_config); 21826f793812SPavel Zakharov } 21836f793812SPavel Zakharov 21846f793812SPavel Zakharov uint64_t 21856f793812SPavel Zakharov spa_missing_tvds_allowed(spa_t *spa) 21866f793812SPavel Zakharov { 21876f793812SPavel Zakharov return (spa->spa_missing_tvds_allowed); 21886f793812SPavel Zakharov } 21896f793812SPavel Zakharov 21906f793812SPavel Zakharov void 21916f793812SPavel Zakharov spa_set_missing_tvds(spa_t *spa, uint64_t missing) 21926f793812SPavel Zakharov { 21936f793812SPavel Zakharov spa->spa_missing_tvds = missing; 21946f793812SPavel Zakharov } 2195*86714001SSerapheim Dimitropoulos 2196*86714001SSerapheim Dimitropoulos boolean_t 2197*86714001SSerapheim Dimitropoulos spa_top_vdevs_spacemap_addressable(spa_t *spa) 2198*86714001SSerapheim Dimitropoulos { 2199*86714001SSerapheim Dimitropoulos vdev_t *rvd = spa->spa_root_vdev; 2200*86714001SSerapheim Dimitropoulos for (uint64_t c = 0; c < rvd->vdev_children; c++) { 2201*86714001SSerapheim Dimitropoulos if (!vdev_is_spacemap_addressable(rvd->vdev_child[c])) 2202*86714001SSerapheim Dimitropoulos return (B_FALSE); 2203*86714001SSerapheim Dimitropoulos } 2204*86714001SSerapheim Dimitropoulos return (B_TRUE); 2205*86714001SSerapheim Dimitropoulos } 2206*86714001SSerapheim Dimitropoulos 2207*86714001SSerapheim Dimitropoulos boolean_t 2208*86714001SSerapheim Dimitropoulos spa_has_checkpoint(spa_t *spa) 2209*86714001SSerapheim Dimitropoulos { 2210*86714001SSerapheim Dimitropoulos return (spa->spa_checkpoint_txg != 0); 2211*86714001SSerapheim Dimitropoulos } 2212*86714001SSerapheim Dimitropoulos 2213*86714001SSerapheim Dimitropoulos boolean_t 2214*86714001SSerapheim Dimitropoulos spa_importing_readonly_checkpoint(spa_t *spa) 2215*86714001SSerapheim Dimitropoulos { 2216*86714001SSerapheim Dimitropoulos return ((spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT) && 2217*86714001SSerapheim Dimitropoulos spa->spa_mode == FREAD); 2218*86714001SSerapheim Dimitropoulos } 2219*86714001SSerapheim Dimitropoulos 2220*86714001SSerapheim Dimitropoulos uint64_t 2221*86714001SSerapheim Dimitropoulos spa_min_claim_txg(spa_t *spa) 2222*86714001SSerapheim Dimitropoulos { 2223*86714001SSerapheim Dimitropoulos uint64_t checkpoint_txg = spa->spa_uberblock.ub_checkpoint_txg; 2224*86714001SSerapheim Dimitropoulos 2225*86714001SSerapheim Dimitropoulos if (checkpoint_txg != 0) 2226*86714001SSerapheim Dimitropoulos return (checkpoint_txg + 1); 2227*86714001SSerapheim Dimitropoulos 2228*86714001SSerapheim Dimitropoulos return (spa->spa_first_txg); 2229*86714001SSerapheim Dimitropoulos } 2230*86714001SSerapheim Dimitropoulos 2231*86714001SSerapheim Dimitropoulos /* 2232*86714001SSerapheim Dimitropoulos * If there is a checkpoint, async destroys may consume more space from 2233*86714001SSerapheim Dimitropoulos * the pool instead of freeing it. In an attempt to save the pool from 2234*86714001SSerapheim Dimitropoulos * getting suspended when it is about to run out of space, we stop 2235*86714001SSerapheim Dimitropoulos * processing async destroys. 2236*86714001SSerapheim Dimitropoulos */ 2237*86714001SSerapheim Dimitropoulos boolean_t 2238*86714001SSerapheim Dimitropoulos spa_suspend_async_destroy(spa_t *spa) 2239*86714001SSerapheim Dimitropoulos { 2240*86714001SSerapheim Dimitropoulos dsl_pool_t *dp = spa_get_dsl(spa); 2241*86714001SSerapheim Dimitropoulos 2242*86714001SSerapheim Dimitropoulos uint64_t unreserved = dsl_pool_unreserved_space(dp, 2243*86714001SSerapheim Dimitropoulos ZFS_SPACE_CHECK_EXTRA_RESERVED); 2244*86714001SSerapheim Dimitropoulos uint64_t used = dsl_dir_phys(dp->dp_root_dir)->dd_used_bytes; 2245*86714001SSerapheim Dimitropoulos uint64_t avail = (unreserved > used) ? (unreserved - used) : 0; 2246*86714001SSerapheim Dimitropoulos 2247*86714001SSerapheim Dimitropoulos if (spa_has_checkpoint(spa) && avail == 0) 2248*86714001SSerapheim Dimitropoulos return (B_TRUE); 2249*86714001SSerapheim Dimitropoulos 2250*86714001SSerapheim Dimitropoulos return (B_FALSE); 2251*86714001SSerapheim Dimitropoulos } 2252