diff --git a/include/sys/dbuf.h b/include/sys/dbuf.h index 7886f728bbb0..06489ea84bf8 100644 --- a/include/sys/dbuf.h +++ b/include/sys/dbuf.h @@ -1,497 +1,498 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or https://opensource.org/licenses/CDDL-1.0. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2012, 2020 by Delphix. All rights reserved. * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. */ #ifndef _SYS_DBUF_H #define _SYS_DBUF_H #include #include #include #include #include #include #include #include #include #ifdef __cplusplus extern "C" { #endif #define IN_DMU_SYNC 2 /* * define flags for dbuf_read */ #define DB_RF_MUST_SUCCEED (1 << 0) #define DB_RF_CANFAIL (1 << 1) #define DB_RF_HAVESTRUCT (1 << 2) #define DB_RF_NOPREFETCH (1 << 3) #define DB_RF_NEVERWAIT (1 << 4) #define DB_RF_CACHED (1 << 5) #define DB_RF_NO_DECRYPT (1 << 6) /* * The simplified state transition diagram for dbufs looks like: * * +----> READ ----+ * | | * | V * (alloc)-->UNCACHED CACHED-->EVICTING-->(free) * | ^ ^ * | | | * +----> FILL ----+ | * | | * | | * +--------> NOFILL -------+ * * DB_SEARCH is an invalid state for a dbuf. It is used by dbuf_free_range * to find all dbufs in a range of a dnode and must be less than any other * dbuf_states_t (see comment on dn_dbufs in dnode.h). */ typedef enum dbuf_states { DB_SEARCH = -1, DB_UNCACHED, DB_FILL, DB_NOFILL, DB_READ, DB_CACHED, DB_EVICTING } dbuf_states_t; typedef enum dbuf_cached_state { DB_NO_CACHE = -1, DB_DBUF_CACHE, DB_DBUF_METADATA_CACHE, DB_CACHE_MAX } dbuf_cached_state_t; struct dnode; struct dmu_tx; /* * level = 0 means the user data * level = 1 means the single indirect block * etc. */ struct dmu_buf_impl; typedef enum override_states { DR_NOT_OVERRIDDEN, DR_IN_DMU_SYNC, DR_OVERRIDDEN } override_states_t; typedef enum db_lock_type { DLT_NONE, DLT_PARENT, DLT_OBJSET } db_lock_type_t; typedef struct dbuf_dirty_record { /* link on our parents dirty list */ list_node_t dr_dirty_node; /* transaction group this data will sync in */ uint64_t dr_txg; /* zio of outstanding write IO */ zio_t *dr_zio; /* pointer back to our dbuf */ struct dmu_buf_impl *dr_dbuf; /* list link for dbuf dirty records */ list_node_t dr_dbuf_node; /* * The dnode we are part of. Note that the dnode can not be moved or * evicted due to the hold that's added by dnode_setdirty() or * dmu_objset_sync_dnodes(), and released by dnode_rele_task() or * userquota_updates_task(). This hold is necessary for * dirty_lightweight_leaf-type dirty records, which don't have a hold * on a dbuf. */ dnode_t *dr_dnode; /* pointer to parent dirty record */ struct dbuf_dirty_record *dr_parent; /* How much space was changed to dsl_pool_dirty_space() for this? */ unsigned int dr_accounted; /* A copy of the bp that points to us */ blkptr_t dr_bp_copy; union dirty_types { struct dirty_indirect { /* protect access to list */ kmutex_t dr_mtx; /* Our list of dirty children */ list_t dr_children; } di; struct dirty_leaf { /* * dr_data is set when we dirty the buffer * so that we can retain the pointer even if it * gets COW'd in a subsequent transaction group. */ arc_buf_t *dr_data; blkptr_t dr_overridden_by; override_states_t dr_override_state; uint8_t dr_copies; boolean_t dr_nopwrite; boolean_t dr_has_raw_params; /* * If dr_has_raw_params is set, the following crypt * params will be set on the BP that's written. */ boolean_t dr_byteorder; uint8_t dr_salt[ZIO_DATA_SALT_LEN]; uint8_t dr_iv[ZIO_DATA_IV_LEN]; uint8_t dr_mac[ZIO_DATA_MAC_LEN]; } dl; struct dirty_lightweight_leaf { /* * This dirty record refers to a leaf (level=0) * block, whose dbuf has not been instantiated for * performance reasons. */ uint64_t dr_blkid; abd_t *dr_abd; zio_prop_t dr_props; enum zio_flag dr_flags; } dll; } dt; } dbuf_dirty_record_t; typedef struct dmu_buf_impl { /* * The following members are immutable, with the exception of * db.db_data, which is protected by db_mtx. */ /* the publicly visible structure */ dmu_buf_t db; /* the objset we belong to */ struct objset *db_objset; /* * handle to safely access the dnode we belong to (NULL when evicted) */ struct dnode_handle *db_dnode_handle; /* * our parent buffer; if the dnode points to us directly, * db_parent == db_dnode_handle->dnh_dnode->dn_dbuf * only accessed by sync thread ??? * (NULL when evicted) * May change from NULL to non-NULL under the protection of db_mtx * (see dbuf_check_blkptr()) */ struct dmu_buf_impl *db_parent; /* * link for hash table of all dmu_buf_impl_t's */ struct dmu_buf_impl *db_hash_next; /* * Our link on the owner dnodes's dn_dbufs list. * Protected by its dn_dbufs_mtx. Should be on the same cache line * as db_level and db_blkid for the best avl_add() performance. */ avl_node_t db_link; /* our block number */ uint64_t db_blkid; /* * Pointer to the blkptr_t which points to us. May be NULL if we * don't have one yet. (NULL when evicted) */ blkptr_t *db_blkptr; /* * Our indirection level. Data buffers have db_level==0. * Indirect buffers which point to data buffers have * db_level==1. etc. Buffers which contain dnodes have * db_level==0, since the dnodes are stored in a file. */ uint8_t db_level; /* * Protects db_buf's contents if they contain an indirect block or data * block of the meta-dnode. We use this lock to protect the structure of * the block tree. This means that when modifying this dbuf's data, we * grab its rwlock. When modifying its parent's data (including the * blkptr to this dbuf), we grab the parent's rwlock. The lock ordering * for this lock is: * 1) dn_struct_rwlock * 2) db_rwlock * We don't currently grab multiple dbufs' db_rwlocks at once. */ krwlock_t db_rwlock; /* buffer holding our data */ arc_buf_t *db_buf; /* db_mtx protects the members below */ kmutex_t db_mtx; /* * Current state of the buffer */ dbuf_states_t db_state; /* * Refcount accessed by dmu_buf_{hold,rele}. * If nonzero, the buffer can't be destroyed. * Protected by db_mtx. */ zfs_refcount_t db_holds; kcondvar_t db_changed; dbuf_dirty_record_t *db_data_pending; /* List of dirty records for the buffer sorted newest to oldest. */ list_t db_dirty_records; /* Link in dbuf_cache or dbuf_metadata_cache */ multilist_node_t db_cache_link; /* Tells us which dbuf cache this dbuf is in, if any */ dbuf_cached_state_t db_caching_status; /* Data which is unique to data (leaf) blocks: */ /* User callback information. */ dmu_buf_user_t *db_user; /* * Evict user data as soon as the dirty and reference * counts are equal. */ uint8_t db_user_immediate_evict; /* * This block was freed while a read or write was * active. */ uint8_t db_freed_in_flight; /* * dnode_evict_dbufs() or dnode_evict_bonus() tried to * evict this dbuf, but couldn't due to outstanding * references. Evict once the refcount drops to 0. */ uint8_t db_pending_evict; uint8_t db_dirtycnt; } dmu_buf_impl_t; -/* Note: the dbuf hash table is exposed only for the mdb module */ -#define DBUF_MUTEXES 2048 -#define DBUF_HASH_MUTEX(h, idx) (&(h)->hash_mutexes[(idx) & (DBUF_MUTEXES-1)]) +#define DBUF_HASH_MUTEX(h, idx) \ + (&(h)->hash_mutexes[(idx) & ((h)->hash_mutex_mask)]) + typedef struct dbuf_hash_table { uint64_t hash_table_mask; + uint64_t hash_mutex_mask; dmu_buf_impl_t **hash_table; - kmutex_t hash_mutexes[DBUF_MUTEXES] ____cacheline_aligned; + kmutex_t *hash_mutexes; } dbuf_hash_table_t; typedef void (*dbuf_prefetch_fn)(void *, uint64_t, uint64_t, boolean_t); uint64_t dbuf_whichblock(const struct dnode *di, const int64_t level, const uint64_t offset); void dbuf_create_bonus(struct dnode *dn); int dbuf_spill_set_blksz(dmu_buf_t *db, uint64_t blksz, dmu_tx_t *tx); void dbuf_rm_spill(struct dnode *dn, dmu_tx_t *tx); dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, const void *tag); dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid, const void *tag); int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid, boolean_t fail_sparse, boolean_t fail_uncached, const void *tag, dmu_buf_impl_t **dbp); int dbuf_prefetch_impl(struct dnode *dn, int64_t level, uint64_t blkid, zio_priority_t prio, arc_flags_t aflags, dbuf_prefetch_fn cb, void *arg); int dbuf_prefetch(struct dnode *dn, int64_t level, uint64_t blkid, zio_priority_t prio, arc_flags_t aflags); void dbuf_add_ref(dmu_buf_impl_t *db, const void *tag); boolean_t dbuf_try_add_ref(dmu_buf_t *db, objset_t *os, uint64_t obj, uint64_t blkid, const void *tag); uint64_t dbuf_refcount(dmu_buf_impl_t *db); void dbuf_rele(dmu_buf_impl_t *db, const void *tag); void dbuf_rele_and_unlock(dmu_buf_impl_t *db, const void *tag, boolean_t evicting); dmu_buf_impl_t *dbuf_find(struct objset *os, uint64_t object, uint8_t level, uint64_t blkid); int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags); void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx); void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx); void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx); void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx); dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx); dbuf_dirty_record_t *dbuf_dirty_lightweight(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx); arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db); void dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data, bp_embedded_type_t etype, enum zio_compress comp, int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx); int dmu_lightweight_write_by_dnode(dnode_t *dn, uint64_t offset, abd_t *abd, const struct zio_prop *zp, enum zio_flag flags, dmu_tx_t *tx); void dmu_buf_redact(dmu_buf_t *dbuf, dmu_tx_t *tx); void dbuf_destroy(dmu_buf_impl_t *db); void dbuf_unoverride(dbuf_dirty_record_t *dr); void dbuf_sync_list(list_t *list, int level, dmu_tx_t *tx); void dbuf_release_bp(dmu_buf_impl_t *db); db_lock_type_t dmu_buf_lock_parent(dmu_buf_impl_t *db, krw_t rw, const void *tag); void dmu_buf_unlock_parent(dmu_buf_impl_t *db, db_lock_type_t type, const void *tag); void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end, struct dmu_tx *); void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx); void dbuf_stats_init(dbuf_hash_table_t *hash); void dbuf_stats_destroy(void); int dbuf_dnode_findbp(dnode_t *dn, uint64_t level, uint64_t blkid, blkptr_t *bp, uint16_t *datablkszsec, uint8_t *indblkshift); #define DB_DNODE(_db) ((_db)->db_dnode_handle->dnh_dnode) #define DB_DNODE_LOCK(_db) ((_db)->db_dnode_handle->dnh_zrlock) #define DB_DNODE_ENTER(_db) (zrl_add(&DB_DNODE_LOCK(_db))) #define DB_DNODE_EXIT(_db) (zrl_remove(&DB_DNODE_LOCK(_db))) #define DB_DNODE_HELD(_db) (!zrl_is_zero(&DB_DNODE_LOCK(_db))) void dbuf_init(void); void dbuf_fini(void); boolean_t dbuf_is_metadata(dmu_buf_impl_t *db); static inline dbuf_dirty_record_t * dbuf_find_dirty_lte(dmu_buf_impl_t *db, uint64_t txg) { dbuf_dirty_record_t *dr; for (dr = list_head(&db->db_dirty_records); dr != NULL && dr->dr_txg > txg; dr = list_next(&db->db_dirty_records, dr)) continue; return (dr); } static inline dbuf_dirty_record_t * dbuf_find_dirty_eq(dmu_buf_impl_t *db, uint64_t txg) { dbuf_dirty_record_t *dr; dr = dbuf_find_dirty_lte(db, txg); if (dr && dr->dr_txg == txg) return (dr); return (NULL); } #define DBUF_GET_BUFC_TYPE(_db) \ (dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA) #define DBUF_IS_CACHEABLE(_db) \ ((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL || \ (dbuf_is_metadata(_db) && \ ((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA))) boolean_t dbuf_is_l2cacheable(dmu_buf_impl_t *db); #ifdef ZFS_DEBUG /* * There should be a ## between the string literal and fmt, to make it * clear that we're joining two strings together, but gcc does not * support that preprocessor token. */ #define dprintf_dbuf(dbuf, fmt, ...) do { \ if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ char __db_buf[32]; \ uint64_t __db_obj = (dbuf)->db.db_object; \ if (__db_obj == DMU_META_DNODE_OBJECT) \ (void) strlcpy(__db_buf, "mdn", sizeof (__db_buf)); \ else \ (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \ (u_longlong_t)__db_obj); \ dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \ "obj=%s lvl=%u blkid=%lld " fmt, \ __db_buf, (dbuf)->db_level, \ (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \ } \ } while (0) #define dprintf_dbuf_bp(db, bp, fmt, ...) do { \ if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP); \ snprintf_blkptr(__blkbuf, BP_SPRINTF_LEN, bp); \ dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf); \ kmem_free(__blkbuf, BP_SPRINTF_LEN); \ } \ } while (0) #define DBUF_VERIFY(db) dbuf_verify(db) #else #define dprintf_dbuf(db, fmt, ...) #define dprintf_dbuf_bp(db, bp, fmt, ...) #define DBUF_VERIFY(db) #endif #ifdef __cplusplus } #endif #endif /* _SYS_DBUF_H */ diff --git a/man/man4/zfs.4 b/man/man4/zfs.4 index 90a8ca788c78..805c037e3d3a 100644 --- a/man/man4/zfs.4 +++ b/man/man4/zfs.4 @@ -1,2579 +1,2585 @@ .\" .\" Copyright (c) 2013 by Turbo Fredriksson . All rights reserved. .\" Copyright (c) 2019, 2021 by Delphix. All rights reserved. .\" Copyright (c) 2019 Datto Inc. .\" The contents of this file are subject to the terms of the Common Development .\" and Distribution License (the "License"). You may not use this file except .\" in compliance with the License. You can obtain a copy of the license at .\" usr/src/OPENSOLARIS.LICENSE or https://opensource.org/licenses/CDDL-1.0. .\" .\" See the License for the specific language governing permissions and .\" limitations under the License. When distributing Covered Code, include this .\" CDDL HEADER in each file and include the License file at .\" usr/src/OPENSOLARIS.LICENSE. If applicable, add the following below this .\" CDDL HEADER, with the fields enclosed by brackets "[]" replaced with your .\" own identifying information: .\" Portions Copyright [yyyy] [name of copyright owner] .\" .Dd June 1, 2021 .Dt ZFS 4 .Os . .Sh NAME .Nm zfs .Nd tuning of the ZFS kernel module . .Sh DESCRIPTION The ZFS module supports these parameters: .Bl -tag -width Ds .It Sy dbuf_cache_max_bytes Ns = Ns Sy ULONG_MAX Ns B Pq ulong Maximum size in bytes of the dbuf cache. The target size is determined by the MIN versus .No 1/2^ Ns Sy dbuf_cache_shift Pq 1/32nd of the target ARC size. The behavior of the dbuf cache and its associated settings can be observed via the .Pa /proc/spl/kstat/zfs/dbufstats kstat. . .It Sy dbuf_metadata_cache_max_bytes Ns = Ns Sy ULONG_MAX Ns B Pq ulong Maximum size in bytes of the metadata dbuf cache. The target size is determined by the MIN versus .No 1/2^ Ns Sy dbuf_metadata_cache_shift Pq 1/64th of the target ARC size. The behavior of the metadata dbuf cache and its associated settings can be observed via the .Pa /proc/spl/kstat/zfs/dbufstats kstat. . .It Sy dbuf_cache_hiwater_pct Ns = Ns Sy 10 Ns % Pq uint The percentage over .Sy dbuf_cache_max_bytes when dbufs must be evicted directly. . .It Sy dbuf_cache_lowater_pct Ns = Ns Sy 10 Ns % Pq uint The percentage below .Sy dbuf_cache_max_bytes when the evict thread stops evicting dbufs. . .It Sy dbuf_cache_shift Ns = Ns Sy 5 Pq int Set the size of the dbuf cache .Pq Sy dbuf_cache_max_bytes to a log2 fraction of the target ARC size. . .It Sy dbuf_metadata_cache_shift Ns = Ns Sy 6 Pq int Set the size of the dbuf metadata cache .Pq Sy dbuf_metadata_cache_max_bytes to a log2 fraction of the target ARC size. . +.It Sy dbuf_mutex_cache_shift Ns = Ns Sy 0 Pq uint +Set the size of the mutex array for the dbuf cache. +When set to +.Sy 0 +the array is dynamically sized based on total system memory. +. .It Sy dmu_object_alloc_chunk_shift Ns = Ns Sy 7 Po 128 Pc Pq int dnode slots allocated in a single operation as a power of 2. The default value minimizes lock contention for the bulk operation performed. . .It Sy dmu_prefetch_max Ns = Ns Sy 134217728 Ns B Po 128 MiB Pc Pq int Limit the amount we can prefetch with one call to this amount in bytes. This helps to limit the amount of memory that can be used by prefetching. . .It Sy ignore_hole_birth Pq int Alias for .Sy send_holes_without_birth_time . . .It Sy l2arc_feed_again Ns = Ns Sy 1 Ns | Ns 0 Pq int Turbo L2ARC warm-up. When the L2ARC is cold the fill interval will be set as fast as possible. . .It Sy l2arc_feed_min_ms Ns = Ns Sy 200 Pq ulong Min feed interval in milliseconds. Requires .Sy l2arc_feed_again Ns = Ns Ar 1 and only applicable in related situations. . .It Sy l2arc_feed_secs Ns = Ns Sy 1 Pq ulong Seconds between L2ARC writing. . .It Sy l2arc_headroom Ns = Ns Sy 2 Pq ulong How far through the ARC lists to search for L2ARC cacheable content, expressed as a multiplier of .Sy l2arc_write_max . ARC persistence across reboots can be achieved with persistent L2ARC by setting this parameter to .Sy 0 , allowing the full length of ARC lists to be searched for cacheable content. . .It Sy l2arc_headroom_boost Ns = Ns Sy 200 Ns % Pq ulong Scales .Sy l2arc_headroom by this percentage when L2ARC contents are being successfully compressed before writing. A value of .Sy 100 disables this feature. . .It Sy l2arc_exclude_special Ns = Ns Sy 0 Ns | Ns 1 Pq int Controls whether buffers present on special vdevs are eligible for caching into L2ARC. If set to 1, exclude dbufs on special vdevs from being cached to L2ARC. . .It Sy l2arc_mfuonly Ns = Ns Sy 0 Ns | Ns 1 Pq int Controls whether only MFU metadata and data are cached from ARC into L2ARC. This may be desired to avoid wasting space on L2ARC when reading/writing large amounts of data that are not expected to be accessed more than once. .Pp The default is off, meaning both MRU and MFU data and metadata are cached. When turning off this feature, some MRU buffers will still be present in ARC and eventually cached on L2ARC. .No If Sy l2arc_noprefetch Ns = Ns Sy 0 , some prefetched buffers will be cached to L2ARC, and those might later transition to MRU, in which case the .Sy l2arc_mru_asize No arcstat will not be Sy 0 . .Pp Regardless of .Sy l2arc_noprefetch , some MFU buffers might be evicted from ARC, accessed later on as prefetches and transition to MRU as prefetches. If accessed again they are counted as MRU and the .Sy l2arc_mru_asize No arcstat will not be Sy 0 . .Pp The ARC status of L2ARC buffers when they were first cached in L2ARC can be seen in the .Sy l2arc_mru_asize , Sy l2arc_mfu_asize , No and Sy l2arc_prefetch_asize arcstats when importing the pool or onlining a cache device if persistent L2ARC is enabled. .Pp The .Sy evict_l2_eligible_mru arcstat does not take into account if this option is enabled as the information provided by the .Sy evict_l2_eligible_m[rf]u arcstats can be used to decide if toggling this option is appropriate for the current workload. . .It Sy l2arc_meta_percent Ns = Ns Sy 33 Ns % Pq int Percent of ARC size allowed for L2ARC-only headers. Since L2ARC buffers are not evicted on memory pressure, too many headers on a system with an irrationally large L2ARC can render it slow or unusable. This parameter limits L2ARC writes and rebuilds to achieve the target. . .It Sy l2arc_trim_ahead Ns = Ns Sy 0 Ns % Pq ulong Trims ahead of the current write size .Pq Sy l2arc_write_max on L2ARC devices by this percentage of write size if we have filled the device. If set to .Sy 100 we TRIM twice the space required to accommodate upcoming writes. A minimum of .Sy 64 MiB will be trimmed. It also enables TRIM of the whole L2ARC device upon creation or addition to an existing pool or if the header of the device is invalid upon importing a pool or onlining a cache device. A value of .Sy 0 disables TRIM on L2ARC altogether and is the default as it can put significant stress on the underlying storage devices. This will vary depending of how well the specific device handles these commands. . .It Sy l2arc_noprefetch Ns = Ns Sy 1 Ns | Ns 0 Pq int Do not write buffers to L2ARC if they were prefetched but not used by applications. In case there are prefetched buffers in L2ARC and this option is later set, we do not read the prefetched buffers from L2ARC. Unsetting this option is useful for caching sequential reads from the disks to L2ARC and serve those reads from L2ARC later on. This may be beneficial in case the L2ARC device is significantly faster in sequential reads than the disks of the pool. .Pp Use .Sy 1 to disable and .Sy 0 to enable caching/reading prefetches to/from L2ARC. . .It Sy l2arc_norw Ns = Ns Sy 0 Ns | Ns 1 Pq int No reads during writes. . .It Sy l2arc_write_boost Ns = Ns Sy 8388608 Ns B Po 8 MiB Pc Pq ulong Cold L2ARC devices will have .Sy l2arc_write_max increased by this amount while they remain cold. . .It Sy l2arc_write_max Ns = Ns Sy 8388608 Ns B Po 8 MiB Pc Pq ulong Max write bytes per interval. . .It Sy l2arc_rebuild_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int Rebuild the L2ARC when importing a pool (persistent L2ARC). This can be disabled if there are problems importing a pool or attaching an L2ARC device (e.g. the L2ARC device is slow in reading stored log metadata, or the metadata has become somehow fragmented/unusable). . .It Sy l2arc_rebuild_blocks_min_l2size Ns = Ns Sy 1073741824 Ns B Po 1 GiB Pc Pq ulong Mininum size of an L2ARC device required in order to write log blocks in it. The log blocks are used upon importing the pool to rebuild the persistent L2ARC. .Pp For L2ARC devices less than 1 GiB, the amount of data .Fn l2arc_evict evicts is significant compared to the amount of restored L2ARC data. In this case, do not write log blocks in L2ARC in order not to waste space. . .It Sy metaslab_aliquot Ns = Ns Sy 1048576 Ns B Po 1 MiB Pc Pq ulong Metaslab granularity, in bytes. This is roughly similar to what would be referred to as the "stripe size" in traditional RAID arrays. In normal operation, ZFS will try to write this amount of data to each disk before moving on to the next top-level vdev. . .It Sy metaslab_bias_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int Enable metaslab group biasing based on their vdevs' over- or under-utilization relative to the pool. . .It Sy metaslab_force_ganging Ns = Ns Sy 16777217 Ns B Po 16 MiB + 1 B Pc Pq ulong Make some blocks above a certain size be gang blocks. This option is used by the test suite to facilitate testing. . .It Sy zfs_history_output_max Ns = Ns Sy 1048576 Ns B Po 1 MiB Pc Pq int When attempting to log an output nvlist of an ioctl in the on-disk history, the output will not be stored if it is larger than this size (in bytes). This must be less than .Sy DMU_MAX_ACCESS Pq 64 MiB . This applies primarily to .Fn zfs_ioc_channel_program Pq cf. Xr zfs-program 8 . . .It Sy zfs_keep_log_spacemaps_at_export Ns = Ns Sy 0 Ns | Ns 1 Pq int Prevent log spacemaps from being destroyed during pool exports and destroys. . .It Sy zfs_metaslab_segment_weight_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int Enable/disable segment-based metaslab selection. . .It Sy zfs_metaslab_switch_threshold Ns = Ns Sy 2 Pq int When using segment-based metaslab selection, continue allocating from the active metaslab until this option's worth of buckets have been exhausted. . .It Sy metaslab_debug_load Ns = Ns Sy 0 Ns | Ns 1 Pq int Load all metaslabs during pool import. . .It Sy metaslab_debug_unload Ns = Ns Sy 0 Ns | Ns 1 Pq int Prevent metaslabs from being unloaded. . .It Sy metaslab_fragmentation_factor_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int Enable use of the fragmentation metric in computing metaslab weights. . .It Sy metaslab_df_max_search Ns = Ns Sy 16777216 Ns B Po 16 MiB Pc Pq int Maximum distance to search forward from the last offset. Without this limit, fragmented pools can see .Em >100`000 iterations and .Fn metaslab_block_picker becomes the performance limiting factor on high-performance storage. .Pp With the default setting of .Sy 16 MiB , we typically see less than .Em 500 iterations, even with very fragmented .Sy ashift Ns = Ns Sy 9 pools. The maximum number of iterations possible is .Sy metaslab_df_max_search / 2^(ashift+1) . With the default setting of .Sy 16 MiB this is .Em 16*1024 Pq with Sy ashift Ns = Ns Sy 9 or .Em 2*1024 Pq with Sy ashift Ns = Ns Sy 12 . . .It Sy metaslab_df_use_largest_segment Ns = Ns Sy 0 Ns | Ns 1 Pq int If not searching forward (due to .Sy metaslab_df_max_search , metaslab_df_free_pct , .No or Sy metaslab_df_alloc_threshold ) , this tunable controls which segment is used. If set, we will use the largest free segment. If unset, we will use a segment of at least the requested size. . .It Sy zfs_metaslab_max_size_cache_sec Ns = Ns Sy 3600 Ns s Po 1 hour Pc Pq ulong When we unload a metaslab, we cache the size of the largest free chunk. We use that cached size to determine whether or not to load a metaslab for a given allocation. As more frees accumulate in that metaslab while it's unloaded, the cached max size becomes less and less accurate. After a number of seconds controlled by this tunable, we stop considering the cached max size and start considering only the histogram instead. . .It Sy zfs_metaslab_mem_limit Ns = Ns Sy 25 Ns % Pq int When we are loading a new metaslab, we check the amount of memory being used to store metaslab range trees. If it is over a threshold, we attempt to unload the least recently used metaslab to prevent the system from clogging all of its memory with range trees. This tunable sets the percentage of total system memory that is the threshold. . .It Sy zfs_metaslab_try_hard_before_gang Ns = Ns Sy 0 Ns | Ns 1 Pq int .Bl -item -compact .It If unset, we will first try normal allocation. .It If that fails then we will do a gang allocation. .It If that fails then we will do a "try hard" gang allocation. .It If that fails then we will have a multi-layer gang block. .El .Pp .Bl -item -compact .It If set, we will first try normal allocation. .It If that fails then we will do a "try hard" allocation. .It If that fails we will do a gang allocation. .It If that fails we will do a "try hard" gang allocation. .It If that fails then we will have a multi-layer gang block. .El . .It Sy zfs_metaslab_find_max_tries Ns = Ns Sy 100 Pq int When not trying hard, we only consider this number of the best metaslabs. This improves performance, especially when there are many metaslabs per vdev and the allocation can't actually be satisfied (so we would otherwise iterate all metaslabs). . .It Sy zfs_vdev_default_ms_count Ns = Ns Sy 200 Pq int When a vdev is added, target this number of metaslabs per top-level vdev. . .It Sy zfs_vdev_default_ms_shift Ns = Ns Sy 29 Po 512 MiB Pc Pq int Default limit for metaslab size. . .It Sy zfs_vdev_max_auto_ashift Ns = Ns Sy 14 Pq ulong Maximum ashift used when optimizing for logical \[->] physical sector size on new top-level vdevs. May be increased up to .Sy ASHIFT_MAX Po 16 Pc , but this may negatively impact pool space efficiency. . .It Sy zfs_vdev_min_auto_ashift Ns = Ns Sy ASHIFT_MIN Po 9 Pc Pq ulong Minimum ashift used when creating new top-level vdevs. . .It Sy zfs_vdev_min_ms_count Ns = Ns Sy 16 Pq int Minimum number of metaslabs to create in a top-level vdev. . .It Sy vdev_validate_skip Ns = Ns Sy 0 Ns | Ns 1 Pq int Skip label validation steps during pool import. Changing is not recommended unless you know what you're doing and are recovering a damaged label. . .It Sy zfs_vdev_ms_count_limit Ns = Ns Sy 131072 Po 128k Pc Pq int Practical upper limit of total metaslabs per top-level vdev. . .It Sy metaslab_preload_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int Enable metaslab group preloading. . .It Sy metaslab_lba_weighting_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int Give more weight to metaslabs with lower LBAs, assuming they have greater bandwidth, as is typically the case on a modern constant angular velocity disk drive. . .It Sy metaslab_unload_delay Ns = Ns Sy 32 Pq int After a metaslab is used, we keep it loaded for this many TXGs, to attempt to reduce unnecessary reloading. Note that both this many TXGs and .Sy metaslab_unload_delay_ms milliseconds must pass before unloading will occur. . .It Sy metaslab_unload_delay_ms Ns = Ns Sy 600000 Ns ms Po 10 min Pc Pq int After a metaslab is used, we keep it loaded for this many milliseconds, to attempt to reduce unnecessary reloading. Note, that both this many milliseconds and .Sy metaslab_unload_delay TXGs must pass before unloading will occur. . .It Sy reference_history Ns = Ns Sy 3 Pq int Maximum reference holders being tracked when reference_tracking_enable is active. . .It Sy reference_tracking_enable Ns = Ns Sy 0 Ns | Ns 1 Pq int Track reference holders to .Sy refcount_t objects (debug builds only). . .It Sy send_holes_without_birth_time Ns = Ns Sy 1 Ns | Ns 0 Pq int When set, the .Sy hole_birth optimization will not be used, and all holes will always be sent during a .Nm zfs Cm send . This is useful if you suspect your datasets are affected by a bug in .Sy hole_birth . . .It Sy spa_config_path Ns = Ns Pa /etc/zfs/zpool.cache Pq charp SPA config file. . .It Sy spa_asize_inflation Ns = Ns Sy 24 Pq int Multiplication factor used to estimate actual disk consumption from the size of data being written. The default value is a worst case estimate, but lower values may be valid for a given pool depending on its configuration. Pool administrators who understand the factors involved may wish to specify a more realistic inflation factor, particularly if they operate close to quota or capacity limits. . .It Sy spa_load_print_vdev_tree Ns = Ns Sy 0 Ns | Ns 1 Pq int Whether to print the vdev tree in the debugging message buffer during pool import. . .It Sy spa_load_verify_data Ns = Ns Sy 1 Ns | Ns 0 Pq int Whether to traverse data blocks during an "extreme rewind" .Pq Fl X import. .Pp An extreme rewind import normally performs a full traversal of all blocks in the pool for verification. If this parameter is unset, the traversal skips non-metadata blocks. It can be toggled once the import has started to stop or start the traversal of non-metadata blocks. . .It Sy spa_load_verify_metadata Ns = Ns Sy 1 Ns | Ns 0 Pq int Whether to traverse blocks during an "extreme rewind" .Pq Fl X pool import. .Pp An extreme rewind import normally performs a full traversal of all blocks in the pool for verification. If this parameter is unset, the traversal is not performed. It can be toggled once the import has started to stop or start the traversal. . .It Sy spa_load_verify_shift Ns = Ns Sy 4 Po 1/16th Pc Pq int Sets the maximum number of bytes to consume during pool import to the log2 fraction of the target ARC size. . .It Sy spa_slop_shift Ns = Ns Sy 5 Po 1/32nd Pc Pq int Normally, we don't allow the last .Sy 3.2% Pq Sy 1/2^spa_slop_shift of space in the pool to be consumed. This ensures that we don't run the pool completely out of space, due to unaccounted changes (e.g. to the MOS). It also limits the worst-case time to allocate space. If we have less than this amount of free space, most ZPL operations (e.g. write, create) will return .Sy ENOSPC . . .It Sy spa_upgrade_errlog_limit Ns = Ns Sy 0 Pq uint Limits the number of on-disk error log entries that will be converted to the new format when enabling the .Sy head_errlog feature. The default is to convert all log entries. . .It Sy vdev_removal_max_span Ns = Ns Sy 32768 Ns B Po 32 KiB Pc Pq int During top-level vdev removal, chunks of data are copied from the vdev which may include free space in order to trade bandwidth for IOPS. This parameter determines the maximum span of free space, in bytes, which will be included as "unnecessary" data in a chunk of copied data. .Pp The default value here was chosen to align with .Sy zfs_vdev_read_gap_limit , which is a similar concept when doing regular reads (but there's no reason it has to be the same). . .It Sy vdev_file_logical_ashift Ns = Ns Sy 9 Po 512 B Pc Pq ulong Logical ashift for file-based devices. . .It Sy vdev_file_physical_ashift Ns = Ns Sy 9 Po 512 B Pc Pq ulong Physical ashift for file-based devices. . .It Sy zap_iterate_prefetch Ns = Ns Sy 1 Ns | Ns 0 Pq int If set, when we start iterating over a ZAP object, prefetch the entire object (all leaf blocks). However, this is limited by .Sy dmu_prefetch_max . . .It Sy zfetch_array_rd_sz Ns = Ns Sy 1048576 Ns B Po 1 MiB Pc Pq ulong If prefetching is enabled, disable prefetching for reads larger than this size. . .It Sy zfetch_min_distance Ns = Ns Sy 4194304 Ns B Po 4 MiB Pc Pq uint Min bytes to prefetch per stream. Prefetch distance starts from the demand access size and quickly grows to this value, doubling on each hit. After that it may grow further by 1/8 per hit, but only if some prefetch since last time haven't completed in time to satisfy demand request, i.e. prefetch depth didn't cover the read latency or the pool got saturated. . .It Sy zfetch_max_distance Ns = Ns Sy 67108864 Ns B Po 64 MiB Pc Pq uint Max bytes to prefetch per stream. . .It Sy zfetch_max_idistance Ns = Ns Sy 67108864 Ns B Po 64 MiB Pc Pq uint Max bytes to prefetch indirects for per stream. . .It Sy zfetch_max_streams Ns = Ns Sy 8 Pq uint Max number of streams per zfetch (prefetch streams per file). . .It Sy zfetch_min_sec_reap Ns = Ns Sy 1 Pq uint Min time before inactive prefetch stream can be reclaimed . .It Sy zfetch_max_sec_reap Ns = Ns Sy 2 Pq uint Max time before inactive prefetch stream can be deleted . .It Sy zfs_abd_scatter_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int Enables ARC from using scatter/gather lists and forces all allocations to be linear in kernel memory. Disabling can improve performance in some code paths at the expense of fragmented kernel memory. . .It Sy zfs_abd_scatter_max_order Ns = Ns Sy MAX_ORDER\-1 Pq uint Maximum number of consecutive memory pages allocated in a single block for scatter/gather lists. .Pp The value of .Sy MAX_ORDER depends on kernel configuration. . .It Sy zfs_abd_scatter_min_size Ns = Ns Sy 1536 Ns B Po 1.5 KiB Pc Pq uint This is the minimum allocation size that will use scatter (page-based) ABDs. Smaller allocations will use linear ABDs. . .It Sy zfs_arc_dnode_limit Ns = Ns Sy 0 Ns B Pq ulong When the number of bytes consumed by dnodes in the ARC exceeds this number of bytes, try to unpin some of it in response to demand for non-metadata. This value acts as a ceiling to the amount of dnode metadata, and defaults to .Sy 0 , which indicates that a percent which is based on .Sy zfs_arc_dnode_limit_percent of the ARC meta buffers that may be used for dnodes. .Pp Also see .Sy zfs_arc_meta_prune which serves a similar purpose but is used when the amount of metadata in the ARC exceeds .Sy zfs_arc_meta_limit rather than in response to overall demand for non-metadata. . .It Sy zfs_arc_dnode_limit_percent Ns = Ns Sy 10 Ns % Pq ulong Percentage that can be consumed by dnodes of ARC meta buffers. .Pp See also .Sy zfs_arc_dnode_limit , which serves a similar purpose but has a higher priority if nonzero. . .It Sy zfs_arc_dnode_reduce_percent Ns = Ns Sy 10 Ns % Pq ulong Percentage of ARC dnodes to try to scan in response to demand for non-metadata when the number of bytes consumed by dnodes exceeds .Sy zfs_arc_dnode_limit . . .It Sy zfs_arc_average_blocksize Ns = Ns Sy 8192 Ns B Po 8 KiB Pc Pq int The ARC's buffer hash table is sized based on the assumption of an average block size of this value. This works out to roughly 1 MiB of hash table per 1 GiB of physical memory with 8-byte pointers. For configurations with a known larger average block size, this value can be increased to reduce the memory footprint. . .It Sy zfs_arc_eviction_pct Ns = Ns Sy 200 Ns % Pq int When .Fn arc_is_overflowing , .Fn arc_get_data_impl waits for this percent of the requested amount of data to be evicted. For example, by default, for every .Em 2 KiB that's evicted, .Em 1 KiB of it may be "reused" by a new allocation. Since this is above .Sy 100 Ns % , it ensures that progress is made towards getting .Sy arc_size No under Sy arc_c . Since this is finite, it ensures that allocations can still happen, even during the potentially long time that .Sy arc_size No is more than Sy arc_c . . .It Sy zfs_arc_evict_batch_limit Ns = Ns Sy 10 Pq int Number ARC headers to evict per sub-list before proceeding to another sub-list. This batch-style operation prevents entire sub-lists from being evicted at once but comes at a cost of additional unlocking and locking. . .It Sy zfs_arc_grow_retry Ns = Ns Sy 0 Ns s Pq int If set to a non zero value, it will replace the .Sy arc_grow_retry value with this value. The .Sy arc_grow_retry .No value Pq default Sy 5 Ns s is the number of seconds the ARC will wait before trying to resume growth after a memory pressure event. . .It Sy zfs_arc_lotsfree_percent Ns = Ns Sy 10 Ns % Pq int Throttle I/O when free system memory drops below this percentage of total system memory. Setting this value to .Sy 0 will disable the throttle. . .It Sy zfs_arc_max Ns = Ns Sy 0 Ns B Pq ulong Max size of ARC in bytes. If .Sy 0 , then the max size of ARC is determined by the amount of system memory installed. Under Linux, half of system memory will be used as the limit. Under .Fx , the larger of .Sy all_system_memory No \- Sy 1 GiB and .Sy 5/8 No \(mu Sy all_system_memory will be used as the limit. This value must be at least .Sy 67108864 Ns B Pq 64 MiB . .Pp This value can be changed dynamically, with some caveats. It cannot be set back to .Sy 0 while running, and reducing it below the current ARC size will not cause the ARC to shrink without memory pressure to induce shrinking. . .It Sy zfs_arc_meta_adjust_restarts Ns = Ns Sy 4096 Pq ulong The number of restart passes to make while scanning the ARC attempting the free buffers in order to stay below the .Sy fs_arc_meta_limit . This value should not need to be tuned but is available to facilitate performance analysis. . .It Sy zfs_arc_meta_limit Ns = Ns Sy 0 Ns B Pq ulong The maximum allowed size in bytes that metadata buffers are allowed to consume in the ARC. When this limit is reached, metadata buffers will be reclaimed, even if the overall .Sy arc_c_max has not been reached. It defaults to .Sy 0 , which indicates that a percentage based on .Sy zfs_arc_meta_limit_percent of the ARC may be used for metadata. .Pp This value my be changed dynamically, except that must be set to an explicit value .Pq cannot be set back to Sy 0 . . .It Sy zfs_arc_meta_limit_percent Ns = Ns Sy 75 Ns % Pq ulong Percentage of ARC buffers that can be used for metadata. .Pp See also .Sy zfs_arc_meta_limit , which serves a similar purpose but has a higher priority if nonzero. . .It Sy zfs_arc_meta_min Ns = Ns Sy 0 Ns B Pq ulong The minimum allowed size in bytes that metadata buffers may consume in the ARC. . .It Sy zfs_arc_meta_prune Ns = Ns Sy 10000 Pq int The number of dentries and inodes to be scanned looking for entries which can be dropped. This may be required when the ARC reaches the .Sy zfs_arc_meta_limit because dentries and inodes can pin buffers in the ARC. Increasing this value will cause to dentry and inode caches to be pruned more aggressively. Setting this value to .Sy 0 will disable pruning the inode and dentry caches. . .It Sy zfs_arc_meta_strategy Ns = Ns Sy 1 Ns | Ns 0 Pq int Define the strategy for ARC metadata buffer eviction (meta reclaim strategy): .Bl -tag -compact -offset 4n -width "0 (META_ONLY)" .It Sy 0 Pq META_ONLY evict only the ARC metadata buffers .It Sy 1 Pq BALANCED additional data buffers may be evicted if required to evict the required number of metadata buffers. .El . .It Sy zfs_arc_min Ns = Ns Sy 0 Ns B Pq ulong Min size of ARC in bytes. .No If set to Sy 0 , arc_c_min will default to consuming the larger of .Sy 32 MiB and .Sy all_system_memory No / Sy 32 . . .It Sy zfs_arc_min_prefetch_ms Ns = Ns Sy 0 Ns ms Ns Po Ns ≡ Ns 1s Pc Pq int Minimum time prefetched blocks are locked in the ARC. . .It Sy zfs_arc_min_prescient_prefetch_ms Ns = Ns Sy 0 Ns ms Ns Po Ns ≡ Ns 6s Pc Pq int Minimum time "prescient prefetched" blocks are locked in the ARC. These blocks are meant to be prefetched fairly aggressively ahead of the code that may use them. . .It Sy zfs_arc_prune_task_threads Ns = Ns Sy 1 Pq int Number of arc_prune threads. .Fx does not need more than one. Linux may theoretically use one per mount point up to number of CPUs, but that was not proven to be useful. . .It Sy zfs_max_missing_tvds Ns = Ns Sy 0 Pq int Number of missing top-level vdevs which will be allowed during pool import (only in read-only mode). . .It Sy zfs_max_nvlist_src_size Ns = Sy 0 Pq ulong Maximum size in bytes allowed to be passed as .Sy zc_nvlist_src_size for ioctls on .Pa /dev/zfs . This prevents a user from causing the kernel to allocate an excessive amount of memory. When the limit is exceeded, the ioctl fails with .Sy EINVAL and a description of the error is sent to the .Pa zfs-dbgmsg log. This parameter should not need to be touched under normal circumstances. If .Sy 0 , equivalent to a quarter of the user-wired memory limit under .Fx and to .Sy 134217728 Ns B Pq 128 MiB under Linux. . .It Sy zfs_multilist_num_sublists Ns = Ns Sy 0 Pq int To allow more fine-grained locking, each ARC state contains a series of lists for both data and metadata objects. Locking is performed at the level of these "sub-lists". This parameters controls the number of sub-lists per ARC state, and also applies to other uses of the multilist data structure. .Pp If .Sy 0 , equivalent to the greater of the number of online CPUs and .Sy 4 . . .It Sy zfs_arc_overflow_shift Ns = Ns Sy 8 Pq int The ARC size is considered to be overflowing if it exceeds the current ARC target size .Pq Sy arc_c by thresholds determined by this parameter. Exceeding by .Sy ( arc_c No >> Sy zfs_arc_overflow_shift ) No / Sy 2 starts ARC reclamation process. If that appears insufficient, exceeding by .Sy ( arc_c No >> Sy zfs_arc_overflow_shift ) No \(mu Sy 1.5 blocks new buffer allocation until the reclaim thread catches up. Started reclamation process continues till ARC size returns below the target size. .Pp The default value of .Sy 8 causes the ARC to start reclamation if it exceeds the target size by .Em 0.2% of the target size, and block allocations by .Em 0.6% . . .It Sy zfs_arc_p_min_shift Ns = Ns Sy 0 Pq int If nonzero, this will update .Sy arc_p_min_shift Pq default Sy 4 with the new value. .Sy arc_p_min_shift No is used as a shift of Sy arc_c when calculating the minumum .Sy arc_p No size. . .It Sy zfs_arc_p_dampener_disable Ns = Ns Sy 1 Ns | Ns 0 Pq int Disable .Sy arc_p adapt dampener, which reduces the maximum single adjustment to .Sy arc_p . . .It Sy zfs_arc_shrink_shift Ns = Ns Sy 0 Pq int If nonzero, this will update .Sy arc_shrink_shift Pq default Sy 7 with the new value. . .It Sy zfs_arc_pc_percent Ns = Ns Sy 0 Ns % Po off Pc Pq uint Percent of pagecache to reclaim ARC to. .Pp This tunable allows the ZFS ARC to play more nicely with the kernel's LRU pagecache. It can guarantee that the ARC size won't collapse under scanning pressure on the pagecache, yet still allows the ARC to be reclaimed down to .Sy zfs_arc_min if necessary. This value is specified as percent of pagecache size (as measured by .Sy NR_FILE_PAGES ) , where that percent may exceed .Sy 100 . This only operates during memory pressure/reclaim. . .It Sy zfs_arc_shrinker_limit Ns = Ns Sy 10000 Pq int This is a limit on how many pages the ARC shrinker makes available for eviction in response to one page allocation attempt. Note that in practice, the kernel's shrinker can ask us to evict up to about four times this for one allocation attempt. .Pp The default limit of .Sy 10000 Pq in practice, Em 160 MiB No per allocation attempt with 4 KiB pages limits the amount of time spent attempting to reclaim ARC memory to less than 100 ms per allocation attempt, even with a small average compressed block size of ~8 KiB. .Pp The parameter can be set to 0 (zero) to disable the limit, and only applies on Linux. . .It Sy zfs_arc_sys_free Ns = Ns Sy 0 Ns B Pq ulong The target number of bytes the ARC should leave as free memory on the system. If zero, equivalent to the bigger of .Sy 512 KiB No and Sy all_system_memory/64 . . .It Sy zfs_autoimport_disable Ns = Ns Sy 1 Ns | Ns 0 Pq int Disable pool import at module load by ignoring the cache file .Pq Sy spa_config_path . . .It Sy zfs_checksum_events_per_second Ns = Ns Sy 20 Ns /s Pq uint Rate limit checksum events to this many per second. Note that this should not be set below the ZED thresholds (currently 10 checksums over 10 seconds) or else the daemon may not trigger any action. . .It Sy zfs_commit_timeout_pct Ns = Ns Sy 5 Ns % Pq int This controls the amount of time that a ZIL block (lwb) will remain "open" when it isn't "full", and it has a thread waiting for it to be committed to stable storage. The timeout is scaled based on a percentage of the last lwb latency to avoid significantly impacting the latency of each individual transaction record (itx). . .It Sy zfs_condense_indirect_commit_entry_delay_ms Ns = Ns Sy 0 Ns ms Pq int Vdev indirection layer (used for device removal) sleeps for this many milliseconds during mapping generation. Intended for use with the test suite to throttle vdev removal speed. . .It Sy zfs_condense_indirect_obsolete_pct Ns = Ns Sy 25 Ns % Pq int Minimum percent of obsolete bytes in vdev mapping required to attempt to condense .Pq see Sy zfs_condense_indirect_vdevs_enable . Intended for use with the test suite to facilitate triggering condensing as needed. . .It Sy zfs_condense_indirect_vdevs_enable Ns = Ns Sy 1 Ns | Ns 0 Pq int Enable condensing indirect vdev mappings. When set, attempt to condense indirect vdev mappings if the mapping uses more than .Sy zfs_condense_min_mapping_bytes bytes of memory and if the obsolete space map object uses more than .Sy zfs_condense_max_obsolete_bytes bytes on-disk. The condensing process is an attempt to save memory by removing obsolete mappings. . .It Sy zfs_condense_max_obsolete_bytes Ns = Ns Sy 1073741824 Ns B Po 1 GiB Pc Pq ulong Only attempt to condense indirect vdev mappings if the on-disk size of the obsolete space map object is greater than this number of bytes .Pq see Sy zfs_condense_indirect_vdevs_enable . . .It Sy zfs_condense_min_mapping_bytes Ns = Ns Sy 131072 Ns B Po 128 KiB Pc Pq ulong Minimum size vdev mapping to attempt to condense .Pq see Sy zfs_condense_indirect_vdevs_enable . . .It Sy zfs_dbgmsg_enable Ns = Ns Sy 1 Ns | Ns 0 Pq int Internally ZFS keeps a small log to facilitate debugging. The log is enabled by default, and can be disabled by unsetting this option. The contents of the log can be accessed by reading .Pa /proc/spl/kstat/zfs/dbgmsg . Writing .Sy 0 to the file clears the log. .Pp This setting does not influence debug prints due to .Sy zfs_flags . . .It Sy zfs_dbgmsg_maxsize Ns = Ns Sy 4194304 Ns B Po 4 MiB Pc Pq int Maximum size of the internal ZFS debug log. . .It Sy zfs_dbuf_state_index Ns = Ns Sy 0 Pq int Historically used for controlling what reporting was available under .Pa /proc/spl/kstat/zfs . No effect. . .It Sy zfs_deadman_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int When a pool sync operation takes longer than .Sy zfs_deadman_synctime_ms , or when an individual I/O operation takes longer than .Sy zfs_deadman_ziotime_ms , then the operation is considered to be "hung". If .Sy zfs_deadman_enabled is set, then the deadman behavior is invoked as described by .Sy zfs_deadman_failmode . By default, the deadman is enabled and set to .Sy wait which results in "hung" I/O operations only being logged. The deadman is automatically disabled when a pool gets suspended. . .It Sy zfs_deadman_failmode Ns = Ns Sy wait Pq charp Controls the failure behavior when the deadman detects a "hung" I/O operation. Valid values are: .Bl -tag -compact -offset 4n -width "continue" .It Sy wait Wait for a "hung" operation to complete. For each "hung" operation a "deadman" event will be posted describing that operation. .It Sy continue Attempt to recover from a "hung" operation by re-dispatching it to the I/O pipeline if possible. .It Sy panic Panic the system. This can be used to facilitate automatic fail-over to a properly configured fail-over partner. .El . .It Sy zfs_deadman_checktime_ms Ns = Ns Sy 60000 Ns ms Po 1 min Pc Pq int Check time in milliseconds. This defines the frequency at which we check for hung I/O requests and potentially invoke the .Sy zfs_deadman_failmode behavior. . .It Sy zfs_deadman_synctime_ms Ns = Ns Sy 600000 Ns ms Po 10 min Pc Pq ulong Interval in milliseconds after which the deadman is triggered and also the interval after which a pool sync operation is considered to be "hung". Once this limit is exceeded the deadman will be invoked every .Sy zfs_deadman_checktime_ms milliseconds until the pool sync completes. . .It Sy zfs_deadman_ziotime_ms Ns = Ns Sy 300000 Ns ms Po 5 min Pc Pq ulong Interval in milliseconds after which the deadman is triggered and an individual I/O operation is considered to be "hung". As long as the operation remains "hung", the deadman will be invoked every .Sy zfs_deadman_checktime_ms milliseconds until the operation completes. . .It Sy zfs_dedup_prefetch Ns = Ns Sy 0 Ns | Ns 1 Pq int Enable prefetching dedup-ed blocks which are going to be freed. . .It Sy zfs_delay_min_dirty_percent Ns = Ns Sy 60 Ns % Pq int Start to delay each transaction once there is this amount of dirty data, expressed as a percentage of .Sy zfs_dirty_data_max . This value should be at least .Sy zfs_vdev_async_write_active_max_dirty_percent . .No See Sx ZFS TRANSACTION DELAY . . .It Sy zfs_delay_scale Ns = Ns Sy 500000 Pq int This controls how quickly the transaction delay approaches infinity. Larger values cause longer delays for a given amount of dirty data. .Pp For the smoothest delay, this value should be about 1 billion divided by the maximum number of operations per second. This will smoothly handle between ten times and a tenth of this number. .No See Sx ZFS TRANSACTION DELAY . .Pp .Sy zfs_delay_scale No \(mu Sy zfs_dirty_data_max Em must No be smaller than Sy 2^64 . . .It Sy zfs_disable_ivset_guid_check Ns = Ns Sy 0 Ns | Ns 1 Pq int Disables requirement for IVset GUIDs to be present and match when doing a raw receive of encrypted datasets. Intended for users whose pools were created with OpenZFS pre-release versions and now have compatibility issues. . .It Sy zfs_key_max_salt_uses Ns = Ns Sy 400000000 Po 4*10^8 Pc Pq ulong Maximum number of uses of a single salt value before generating a new one for encrypted datasets. The default value is also the maximum. . .It Sy zfs_object_mutex_size Ns = Ns Sy 64 Pq uint Size of the znode hashtable used for holds. .Pp Due to the need to hold locks on objects that may not exist yet, kernel mutexes are not created per-object and instead a hashtable is used where collisions will result in objects waiting when there is not actually contention on the same object. . .It Sy zfs_slow_io_events_per_second Ns = Ns Sy 20 Ns /s Pq int Rate limit delay and deadman zevents (which report slow I/O operations) to this many per second. . .It Sy zfs_unflushed_max_mem_amt Ns = Ns Sy 1073741824 Ns B Po 1 GiB Pc Pq ulong Upper-bound limit for unflushed metadata changes to be held by the log spacemap in memory, in bytes. . .It Sy zfs_unflushed_max_mem_ppm Ns = Ns Sy 1000 Ns ppm Po 0.1% Pc Pq ulong Part of overall system memory that ZFS allows to be used for unflushed metadata changes by the log spacemap, in millionths. . .It Sy zfs_unflushed_log_block_max Ns = Ns Sy 131072 Po 128k Pc Pq ulong Describes the maximum number of log spacemap blocks allowed for each pool. The default value means that the space in all the log spacemaps can add up to no more than .Sy 131072 blocks (which means .Em 16 GiB of logical space before compression and ditto blocks, assuming that blocksize is .Em 128 KiB ) . .Pp This tunable is important because it involves a trade-off between import time after an unclean export and the frequency of flushing metaslabs. The higher this number is, the more log blocks we allow when the pool is active which means that we flush metaslabs less often and thus decrease the number of I/O operations for spacemap updates per TXG. At the same time though, that means that in the event of an unclean export, there will be more log spacemap blocks for us to read, inducing overhead in the import time of the pool. The lower the number, the amount of flushing increases, destroying log blocks quicker as they become obsolete faster, which leaves less blocks to be read during import time after a crash. .Pp Each log spacemap block existing during pool import leads to approximately one extra logical I/O issued. This is the reason why this tunable is exposed in terms of blocks rather than space used. . .It Sy zfs_unflushed_log_block_min Ns = Ns Sy 1000 Pq ulong If the number of metaslabs is small and our incoming rate is high, we could get into a situation that we are flushing all our metaslabs every TXG. Thus we always allow at least this many log blocks. . .It Sy zfs_unflushed_log_block_pct Ns = Ns Sy 400 Ns % Pq ulong Tunable used to determine the number of blocks that can be used for the spacemap log, expressed as a percentage of the total number of unflushed metaslabs in the pool. . .It Sy zfs_unflushed_log_txg_max Ns = Ns Sy 1000 Pq ulong Tunable limiting maximum time in TXGs any metaslab may remain unflushed. It effectively limits maximum number of unflushed per-TXG spacemap logs that need to be read after unclean pool export. . .It Sy zfs_unlink_suspend_progress Ns = Ns Sy 0 Ns | Ns 1 Pq uint When enabled, files will not be asynchronously removed from the list of pending unlinks and the space they consume will be leaked. Once this option has been disabled and the dataset is remounted, the pending unlinks will be processed and the freed space returned to the pool. This option is used by the test suite. . .It Sy zfs_delete_blocks Ns = Ns Sy 20480 Pq ulong This is the used to define a large file for the purposes of deletion. Files containing more than .Sy zfs_delete_blocks will be deleted asynchronously, while smaller files are deleted synchronously. Decreasing this value will reduce the time spent in an .Xr unlink 2 system call, at the expense of a longer delay before the freed space is available. . .It Sy zfs_dirty_data_max Ns = Pq int Determines the dirty space limit in bytes. Once this limit is exceeded, new writes are halted until space frees up. This parameter takes precedence over .Sy zfs_dirty_data_max_percent . .No See Sx ZFS TRANSACTION DELAY . .Pp Defaults to .Sy physical_ram/10 , capped at .Sy zfs_dirty_data_max_max . . .It Sy zfs_dirty_data_max_max Ns = Pq int Maximum allowable value of .Sy zfs_dirty_data_max , expressed in bytes. This limit is only enforced at module load time, and will be ignored if .Sy zfs_dirty_data_max is later changed. This parameter takes precedence over .Sy zfs_dirty_data_max_max_percent . .No See Sx ZFS TRANSACTION DELAY . .Pp Defaults to .Sy physical_ram/4 , . .It Sy zfs_dirty_data_max_max_percent Ns = Ns Sy 25 Ns % Pq int Maximum allowable value of .Sy zfs_dirty_data_max , expressed as a percentage of physical RAM. This limit is only enforced at module load time, and will be ignored if .Sy zfs_dirty_data_max is later changed. The parameter .Sy zfs_dirty_data_max_max takes precedence over this one. .No See Sx ZFS TRANSACTION DELAY . . .It Sy zfs_dirty_data_max_percent Ns = Ns Sy 10 Ns % Pq int Determines the dirty space limit, expressed as a percentage of all memory. Once this limit is exceeded, new writes are halted until space frees up. The parameter .Sy zfs_dirty_data_max takes precedence over this one. .No See Sx ZFS TRANSACTION DELAY . .Pp Subject to .Sy zfs_dirty_data_max_max . . .It Sy zfs_dirty_data_sync_percent Ns = Ns Sy 20 Ns % Pq int Start syncing out a transaction group if there's at least this much dirty data .Pq as a percentage of Sy zfs_dirty_data_max . This should be less than .Sy zfs_vdev_async_write_active_min_dirty_percent . . .It Sy zfs_wrlog_data_max Ns = Pq int The upper limit of write-transaction zil log data size in bytes. Write operations are throttled when approaching the limit until log data is cleared out after transaction group sync. Because of some overhead, it should be set at least 2 times the size of .Sy zfs_dirty_data_max .No to prevent harming normal write throughput. It also should be smaller than the size of the slog device if slog is present. .Pp Defaults to .Sy zfs_dirty_data_max*2 . .It Sy zfs_fallocate_reserve_percent Ns = Ns Sy 110 Ns % Pq uint Since ZFS is a copy-on-write filesystem with snapshots, blocks cannot be preallocated for a file in order to guarantee that later writes will not run out of space. Instead, .Xr fallocate 2 space preallocation only checks that sufficient space is currently available in the pool or the user's project quota allocation, and then creates a sparse file of the requested size. The requested space is multiplied by .Sy zfs_fallocate_reserve_percent to allow additional space for indirect blocks and other internal metadata. Setting this to .Sy 0 disables support for .Xr fallocate 2 and causes it to return .Sy EOPNOTSUPP . . .It Sy zfs_fletcher_4_impl Ns = Ns Sy fastest Pq string Select a fletcher 4 implementation. .Pp Supported selectors are: .Sy fastest , scalar , sse2 , ssse3 , avx2 , avx512f , avx512bw , .No and Sy aarch64_neon . All except .Sy fastest No and Sy scalar require instruction set extensions to be available, and will only appear if ZFS detects that they are present at runtime. If multiple implementations of fletcher 4 are available, the .Sy fastest will be chosen using a micro benchmark. Selecting .Sy scalar results in the original CPU-based calculation being used. Selecting any option other than .Sy fastest No or Sy scalar results in vector instructions from the respective CPU instruction set being used. . .It Sy zfs_blake3_impl Ns = Ns Sy fastest Pq string Select a BLAKE3 implementation. .Pp Supported selectors are: .Sy cycle , fastest , generic , sse2 , sse41 , avx2 , avx512 . All except .Sy cycle , fastest No and Sy generic require instruction set extensions to be available, and will only appear if ZFS detects that they are present at runtime. If multiple implementations of BLAKE3 are available, the .Sy fastest will be chosen using a micro benchmark. You can see the benchmark results by reading this kstat file: .Pa /proc/spl/kstat/zfs/chksum_bench . . .It Sy zfs_free_bpobj_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int Enable/disable the processing of the free_bpobj object. . .It Sy zfs_async_block_max_blocks Ns = Ns Sy ULONG_MAX Po unlimited Pc Pq ulong Maximum number of blocks freed in a single TXG. . .It Sy zfs_max_async_dedup_frees Ns = Ns Sy 100000 Po 10^5 Pc Pq ulong Maximum number of dedup blocks freed in a single TXG. . .It Sy zfs_vdev_async_read_max_active Ns = Ns Sy 3 Pq int Maximum asynchronous read I/O operations active to each device. .No See Sx ZFS I/O SCHEDULER . . .It Sy zfs_vdev_async_read_min_active Ns = Ns Sy 1 Pq int Minimum asynchronous read I/O operation active to each device. .No See Sx ZFS I/O SCHEDULER . . .It Sy zfs_vdev_async_write_active_max_dirty_percent Ns = Ns Sy 60 Ns % Pq int When the pool has more than this much dirty data, use .Sy zfs_vdev_async_write_max_active to limit active async writes. If the dirty data is between the minimum and maximum, the active I/O limit is linearly interpolated. .No See Sx ZFS I/O SCHEDULER . . .It Sy zfs_vdev_async_write_active_min_dirty_percent Ns = Ns Sy 30 Ns % Pq int When the pool has less than this much dirty data, use .Sy zfs_vdev_async_write_min_active to limit active async writes. If the dirty data is between the minimum and maximum, the active I/O limit is linearly interpolated. .No See Sx ZFS I/O SCHEDULER . . .It Sy zfs_vdev_async_write_max_active Ns = Ns Sy 30 Pq int Maximum asynchronous write I/O operations active to each device. .No See Sx ZFS I/O SCHEDULER . . .It Sy zfs_vdev_async_write_min_active Ns = Ns Sy 2 Pq int Minimum asynchronous write I/O operations active to each device. .No See Sx ZFS I/O SCHEDULER . .Pp Lower values are associated with better latency on rotational media but poorer resilver performance. The default value of .Sy 2 was chosen as a compromise. A value of .Sy 3 has been shown to improve resilver performance further at a cost of further increasing latency. . .It Sy zfs_vdev_initializing_max_active Ns = Ns Sy 1 Pq int Maximum initializing I/O operations active to each device. .No See Sx ZFS I/O SCHEDULER . . .It Sy zfs_vdev_initializing_min_active Ns = Ns Sy 1 Pq int Minimum initializing I/O operations active to each device. .No See Sx ZFS I/O SCHEDULER . . .It Sy zfs_vdev_max_active Ns = Ns Sy 1000 Pq int The maximum number of I/O operations active to each device. Ideally, this will be at least the sum of each queue's .Sy max_active . .No See Sx ZFS I/O SCHEDULER . . .It Sy zfs_vdev_rebuild_max_active Ns = Ns Sy 3 Pq int Maximum sequential resilver I/O operations active to each device. .No See Sx ZFS I/O SCHEDULER . . .It Sy zfs_vdev_rebuild_min_active Ns = Ns Sy 1 Pq int Minimum sequential resilver I/O operations active to each device. .No See Sx ZFS I/O SCHEDULER . . .It Sy zfs_vdev_removal_max_active Ns = Ns Sy 2 Pq int Maximum removal I/O operations active to each device. .No See Sx ZFS I/O SCHEDULER . . .It Sy zfs_vdev_removal_min_active Ns = Ns Sy 1 Pq int Minimum removal I/O operations active to each device. .No See Sx ZFS I/O SCHEDULER . . .It Sy zfs_vdev_scrub_max_active Ns = Ns Sy 2 Pq int Maximum scrub I/O operations active to each device. .No See Sx ZFS I/O SCHEDULER . . .It Sy zfs_vdev_scrub_min_active Ns = Ns Sy 1 Pq int Minimum scrub I/O operations active to each device. .No See Sx ZFS I/O SCHEDULER . . .It Sy zfs_vdev_sync_read_max_active Ns = Ns Sy 10 Pq int Maximum synchronous read I/O operations active to each device. .No See Sx ZFS I/O SCHEDULER . . .It Sy zfs_vdev_sync_read_min_active Ns = Ns Sy 10 Pq int Minimum synchronous read I/O operations active to each device. .No See Sx ZFS I/O SCHEDULER . . .It Sy zfs_vdev_sync_write_max_active Ns = Ns Sy 10 Pq int Maximum synchronous write I/O operations active to each device. .No See Sx ZFS I/O SCHEDULER . . .It Sy zfs_vdev_sync_write_min_active Ns = Ns Sy 10 Pq int Minimum synchronous write I/O operations active to each device. .No See Sx ZFS I/O SCHEDULER . . .It Sy zfs_vdev_trim_max_active Ns = Ns Sy 2 Pq int Maximum trim/discard I/O operations active to each device. .No See Sx ZFS I/O SCHEDULER . . .It Sy zfs_vdev_trim_min_active Ns = Ns Sy 1 Pq int Minimum trim/discard I/O operations active to each device. .No See Sx ZFS I/O SCHEDULER . . .It Sy zfs_vdev_nia_delay Ns = Ns Sy 5 Pq int For non-interactive I/O (scrub, resilver, removal, initialize and rebuild), the number of concurrently-active I/O operations is limited to .Sy zfs_*_min_active , unless the vdev is "idle". When there are no interactive I/O operations active (synchronous or otherwise), and .Sy zfs_vdev_nia_delay operations have completed since the last interactive operation, then the vdev is considered to be "idle", and the number of concurrently-active non-interactive operations is increased to .Sy zfs_*_max_active . .No See Sx ZFS I/O SCHEDULER . . .It Sy zfs_vdev_nia_credit Ns = Ns Sy 5 Pq int Some HDDs tend to prioritize sequential I/O so strongly, that concurrent random I/O latency reaches several seconds. On some HDDs this happens even if sequential I/O operations are submitted one at a time, and so setting .Sy zfs_*_max_active Ns = Sy 1 does not help. To prevent non-interactive I/O, like scrub, from monopolizing the device, no more than .Sy zfs_vdev_nia_credit operations can be sent while there are outstanding incomplete interactive operations. This enforced wait ensures the HDD services the interactive I/O within a reasonable amount of time. .No See Sx ZFS I/O SCHEDULER . . .It Sy zfs_vdev_queue_depth_pct Ns = Ns Sy 1000 Ns % Pq int Maximum number of queued allocations per top-level vdev expressed as a percentage of .Sy zfs_vdev_async_write_max_active , which allows the system to detect devices that are more capable of handling allocations and to allocate more blocks to those devices. This allows for dynamic allocation distribution when devices are imbalanced, as fuller devices will tend to be slower than empty devices. .Pp Also see .Sy zio_dva_throttle_enabled . . .It Sy zfs_expire_snapshot Ns = Ns Sy 300 Ns s Pq int Time before expiring .Pa .zfs/snapshot . . .It Sy zfs_admin_snapshot Ns = Ns Sy 0 Ns | Ns 1 Pq int Allow the creation, removal, or renaming of entries in the .Sy .zfs/snapshot directory to cause the creation, destruction, or renaming of snapshots. When enabled, this functionality works both locally and over NFS exports which have the .Em no_root_squash option set. . .It Sy zfs_flags Ns = Ns Sy 0 Pq int Set additional debugging flags. The following flags may be bitwise-ored together: .TS box; lbz r l l . Value Symbolic Name Description _ 1 ZFS_DEBUG_DPRINTF Enable dprintf entries in the debug log. * 2 ZFS_DEBUG_DBUF_VERIFY Enable extra dbuf verifications. * 4 ZFS_DEBUG_DNODE_VERIFY Enable extra dnode verifications. 8 ZFS_DEBUG_SNAPNAMES Enable snapshot name verification. 16 ZFS_DEBUG_MODIFY Check for illegally modified ARC buffers. 64 ZFS_DEBUG_ZIO_FREE Enable verification of block frees. 128 ZFS_DEBUG_HISTOGRAM_VERIFY Enable extra spacemap histogram verifications. 256 ZFS_DEBUG_METASLAB_VERIFY Verify space accounting on disk matches in-memory \fBrange_trees\fP. 512 ZFS_DEBUG_SET_ERROR Enable \fBSET_ERROR\fP and dprintf entries in the debug log. 1024 ZFS_DEBUG_INDIRECT_REMAP Verify split blocks created by device removal. 2048 ZFS_DEBUG_TRIM Verify TRIM ranges are always within the allocatable range tree. 4096 ZFS_DEBUG_LOG_SPACEMAP Verify that the log summary is consistent with the spacemap log and enable \fBzfs_dbgmsgs\fP for metaslab loading and flushing. .TE .Sy \& * No Requires debug build. . .It Sy zfs_btree_verify_intensity Ns = Ns Sy 0 Pq uint Enables btree verification. The following settings are culminative: .TS box; lbz r l l . Value Description 1 Verify height. 2 Verify pointers from children to parent. 3 Verify element counts. 4 Verify element order. (expensive) * 5 Verify unused memory is poisoned. (expensive) .TE .Sy \& * No Requires debug build. . .It Sy zfs_free_leak_on_eio Ns = Ns Sy 0 Ns | Ns 1 Pq int If destroy encounters an .Sy EIO while reading metadata (e.g. indirect blocks), space referenced by the missing metadata can not be freed. Normally this causes the background destroy to become "stalled", as it is unable to make forward progress. While in this stalled state, all remaining space to free from the error-encountering filesystem is "temporarily leaked". Set this flag to cause it to ignore the .Sy EIO , permanently leak the space from indirect blocks that can not be read, and continue to free everything else that it can. .Pp The default "stalling" behavior is useful if the storage partially fails (i.e. some but not all I/O operations fail), and then later recovers. In this case, we will be able to continue pool operations while it is partially failed, and when it recovers, we can continue to free the space, with no leaks. Note, however, that this case is actually fairly rare. .Pp Typically pools either .Bl -enum -compact -offset 4n -width "1." .It fail completely (but perhaps temporarily, e.g. due to a top-level vdev going offline), or .It have localized, permanent errors (e.g. disk returns the wrong data due to bit flip or firmware bug). .El In the former case, this setting does not matter because the pool will be suspended and the sync thread will not be able to make forward progress regardless. In the latter, because the error is permanent, the best we can do is leak the minimum amount of space, which is what setting this flag will do. It is therefore reasonable for this flag to normally be set, but we chose the more conservative approach of not setting it, so that there is no possibility of leaking space in the "partial temporary" failure case. . .It Sy zfs_free_min_time_ms Ns = Ns Sy 1000 Ns ms Po 1s Pc Pq int During a .Nm zfs Cm destroy operation using the .Sy async_destroy feature, a minimum of this much time will be spent working on freeing blocks per TXG. . .It Sy zfs_obsolete_min_time_ms Ns = Ns Sy 500 Ns ms Pq int Similar to .Sy zfs_free_min_time_ms , but for cleanup of old indirection records for removed vdevs. . .It Sy zfs_immediate_write_sz Ns = Ns Sy 32768 Ns B Po 32 KiB Pc Pq long Largest data block to write to the ZIL. Larger blocks will be treated as if the dataset being written to had the .Sy logbias Ns = Ns Sy throughput property set. . .It Sy zfs_initialize_value Ns = Ns Sy 16045690984833335022 Po 0xDEADBEEFDEADBEEE Pc Pq ulong Pattern written to vdev free space by .Xr zpool-initialize 8 . . .It Sy zfs_initialize_chunk_size Ns = Ns Sy 1048576 Ns B Po 1 MiB Pc Pq ulong Size of writes used by .Xr zpool-initialize 8 . This option is used by the test suite. . .It Sy zfs_livelist_max_entries Ns = Ns Sy 500000 Po 5*10^5 Pc Pq ulong The threshold size (in block pointers) at which we create a new sub-livelist. Larger sublists are more costly from a memory perspective but the fewer sublists there are, the lower the cost of insertion. . .It Sy zfs_livelist_min_percent_shared Ns = Ns Sy 75 Ns % Pq int If the amount of shared space between a snapshot and its clone drops below this threshold, the clone turns off the livelist and reverts to the old deletion method. This is in place because livelists no long give us a benefit once a clone has been overwritten enough. . .It Sy zfs_livelist_condense_new_alloc Ns = Ns Sy 0 Pq int Incremented each time an extra ALLOC blkptr is added to a livelist entry while it is being condensed. This option is used by the test suite to track race conditions. . .It Sy zfs_livelist_condense_sync_cancel Ns = Ns Sy 0 Pq int Incremented each time livelist condensing is canceled while in .Fn spa_livelist_condense_sync . This option is used by the test suite to track race conditions. . .It Sy zfs_livelist_condense_sync_pause Ns = Ns Sy 0 Ns | Ns 1 Pq int When set, the livelist condense process pauses indefinitely before executing the synctask \(em .Fn spa_livelist_condense_sync . This option is used by the test suite to trigger race conditions. . .It Sy zfs_livelist_condense_zthr_cancel Ns = Ns Sy 0 Pq int Incremented each time livelist condensing is canceled while in .Fn spa_livelist_condense_cb . This option is used by the test suite to track race conditions. . .It Sy zfs_livelist_condense_zthr_pause Ns = Ns Sy 0 Ns | Ns 1 Pq int When set, the livelist condense process pauses indefinitely before executing the open context condensing work in .Fn spa_livelist_condense_cb . This option is used by the test suite to trigger race conditions. . .It Sy zfs_lua_max_instrlimit Ns = Ns Sy 100000000 Po 10^8 Pc Pq ulong The maximum execution time limit that can be set for a ZFS channel program, specified as a number of Lua instructions. . .It Sy zfs_lua_max_memlimit Ns = Ns Sy 104857600 Po 100 MiB Pc Pq ulong The maximum memory limit that can be set for a ZFS channel program, specified in bytes. . .It Sy zfs_max_dataset_nesting Ns = Ns Sy 50 Pq int The maximum depth of nested datasets. This value can be tuned temporarily to fix existing datasets that exceed the predefined limit. . .It Sy zfs_max_log_walking Ns = Ns Sy 5 Pq ulong The number of past TXGs that the flushing algorithm of the log spacemap feature uses to estimate incoming log blocks. . .It Sy zfs_max_logsm_summary_length Ns = Ns Sy 10 Pq ulong Maximum number of rows allowed in the summary of the spacemap log. . .It Sy zfs_max_recordsize Ns = Ns Sy 16777216 Po 16 MiB Pc Pq int We currently support block sizes from .Em 512 Po 512 B Pc No to Em 16777216 Po 16 MiB Pc . The benefits of larger blocks, and thus larger I/O, need to be weighed against the cost of COWing a giant block to modify one byte. Additionally, very large blocks can have an impact on I/O latency, and also potentially on the memory allocator. Therefore, we formerly forbade creating blocks larger than 1M. Larger blocks could be created by changing it, and pools with larger blocks can always be imported and used, regardless of this setting. . .It Sy zfs_allow_redacted_dataset_mount Ns = Ns Sy 0 Ns | Ns 1 Pq int Allow datasets received with redacted send/receive to be mounted. Normally disabled because these datasets may be missing key data. . .It Sy zfs_min_metaslabs_to_flush Ns = Ns Sy 1 Pq ulong Minimum number of metaslabs to flush per dirty TXG. . .It Sy zfs_metaslab_fragmentation_threshold Ns = Ns Sy 70 Ns % Pq int Allow metaslabs to keep their active state as long as their fragmentation percentage is no more than this value. An active metaslab that exceeds this threshold will no longer keep its active status allowing better metaslabs to be selected. . .It Sy zfs_mg_fragmentation_threshold Ns = Ns Sy 95 Ns % Pq int Metaslab groups are considered eligible for allocations if their fragmentation metric (measured as a percentage) is less than or equal to this value. If a metaslab group exceeds this threshold then it will be skipped unless all metaslab groups within the metaslab class have also crossed this threshold. . .It Sy zfs_mg_noalloc_threshold Ns = Ns Sy 0 Ns % Pq int Defines a threshold at which metaslab groups should be eligible for allocations. The value is expressed as a percentage of free space beyond which a metaslab group is always eligible for allocations. If a metaslab group's free space is less than or equal to the threshold, the allocator will avoid allocating to that group unless all groups in the pool have reached the threshold. Once all groups have reached the threshold, all groups are allowed to accept allocations. The default value of .Sy 0 disables the feature and causes all metaslab groups to be eligible for allocations. .Pp This parameter allows one to deal with pools having heavily imbalanced vdevs such as would be the case when a new vdev has been added. Setting the threshold to a non-zero percentage will stop allocations from being made to vdevs that aren't filled to the specified percentage and allow lesser filled vdevs to acquire more allocations than they otherwise would under the old .Sy zfs_mg_alloc_failures facility. . .It Sy zfs_ddt_data_is_special Ns = Ns Sy 1 Ns | Ns 0 Pq int If enabled, ZFS will place DDT data into the special allocation class. . .It Sy zfs_user_indirect_is_special Ns = Ns Sy 1 Ns | Ns 0 Pq int If enabled, ZFS will place user data indirect blocks into the special allocation class. . .It Sy zfs_multihost_history Ns = Ns Sy 0 Pq int Historical statistics for this many latest multihost updates will be available in .Pa /proc/spl/kstat/zfs/ Ns Ao Ar pool Ac Ns Pa /multihost . . .It Sy zfs_multihost_interval Ns = Ns Sy 1000 Ns ms Po 1 s Pc Pq ulong Used to control the frequency of multihost writes which are performed when the .Sy multihost pool property is on. This is one of the factors used to determine the length of the activity check during import. .Pp The multihost write period is .Sy zfs_multihost_interval No / Sy leaf-vdevs . On average a multihost write will be issued for each leaf vdev every .Sy zfs_multihost_interval milliseconds. In practice, the observed period can vary with the I/O load and this observed value is the delay which is stored in the uberblock. . .It Sy zfs_multihost_import_intervals Ns = Ns Sy 20 Pq uint Used to control the duration of the activity test on import. Smaller values of .Sy zfs_multihost_import_intervals will reduce the import time but increase the risk of failing to detect an active pool. The total activity check time is never allowed to drop below one second. .Pp On import the activity check waits a minimum amount of time determined by .Sy zfs_multihost_interval No \(mu Sy zfs_multihost_import_intervals , or the same product computed on the host which last had the pool imported, whichever is greater. The activity check time may be further extended if the value of MMP delay found in the best uberblock indicates actual multihost updates happened at longer intervals than .Sy zfs_multihost_interval . A minimum of .Em 100 ms is enforced. .Pp .Sy 0 No is equivalent to Sy 1 . . .It Sy zfs_multihost_fail_intervals Ns = Ns Sy 10 Pq uint Controls the behavior of the pool when multihost write failures or delays are detected. .Pp When .Sy 0 , multihost write failures or delays are ignored. The failures will still be reported to the ZED which depending on its configuration may take action such as suspending the pool or offlining a device. .Pp Otherwise, the pool will be suspended if .Sy zfs_multihost_fail_intervals No \(mu Sy zfs_multihost_interval milliseconds pass without a successful MMP write. This guarantees the activity test will see MMP writes if the pool is imported. .Sy 1 No is equivalent to Sy 2 ; this is necessary to prevent the pool from being suspended due to normal, small I/O latency variations. . .It Sy zfs_no_scrub_io Ns = Ns Sy 0 Ns | Ns 1 Pq int Set to disable scrub I/O. This results in scrubs not actually scrubbing data and simply doing a metadata crawl of the pool instead. . .It Sy zfs_no_scrub_prefetch Ns = Ns Sy 0 Ns | Ns 1 Pq int Set to disable block prefetching for scrubs. . .It Sy zfs_nocacheflush Ns = Ns Sy 0 Ns | Ns 1 Pq int Disable cache flush operations on disks when writing. Setting this will cause pool corruption on power loss if a volatile out-of-order write cache is enabled. . .It Sy zfs_nopwrite_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int Allow no-operation writes. The occurrence of nopwrites will further depend on other pool properties .Pq i.a. the checksumming and compression algorithms . . .It Sy zfs_dmu_offset_next_sync Ns = Ns Sy 1 Ns | Ns 0 Pq int Enable forcing TXG sync to find holes. When enabled forces ZFS to sync data when .Sy SEEK_HOLE No or Sy SEEK_DATA flags are used allowing holes in a file to be accurately reported. When disabled holes will not be reported in recently dirtied files. . .It Sy zfs_pd_bytes_max Ns = Ns Sy 52428800 Ns B Po 50 MiB Pc Pq int The number of bytes which should be prefetched during a pool traversal, like .Nm zfs Cm send or other data crawling operations. . .It Sy zfs_traverse_indirect_prefetch_limit Ns = Ns Sy 32 Pq int The number of blocks pointed by indirect (non-L0) block which should be prefetched during a pool traversal, like .Nm zfs Cm send or other data crawling operations. . .It Sy zfs_per_txg_dirty_frees_percent Ns = Ns Sy 5 Ns % Pq ulong Control percentage of dirtied indirect blocks from frees allowed into one TXG. After this threshold is crossed, additional frees will wait until the next TXG. .Sy 0 No disables this throttle. . .It Sy zfs_prefetch_disable Ns = Ns Sy 0 Ns | Ns 1 Pq int Disable predictive prefetch. Note that it leaves "prescient" prefetch .Pq for, e.g., Nm zfs Cm send intact. Unlike predictive prefetch, prescient prefetch never issues I/O that ends up not being needed, so it can't hurt performance. . .It Sy zfs_qat_checksum_disable Ns = Ns Sy 0 Ns | Ns 1 Pq int Disable QAT hardware acceleration for SHA256 checksums. May be unset after the ZFS modules have been loaded to initialize the QAT hardware as long as support is compiled in and the QAT driver is present. . .It Sy zfs_qat_compress_disable Ns = Ns Sy 0 Ns | Ns 1 Pq int Disable QAT hardware acceleration for gzip compression. May be unset after the ZFS modules have been loaded to initialize the QAT hardware as long as support is compiled in and the QAT driver is present. . .It Sy zfs_qat_encrypt_disable Ns = Ns Sy 0 Ns | Ns 1 Pq int Disable QAT hardware acceleration for AES-GCM encryption. May be unset after the ZFS modules have been loaded to initialize the QAT hardware as long as support is compiled in and the QAT driver is present. . .It Sy zfs_vnops_read_chunk_size Ns = Ns Sy 1048576 Ns B Po 1 MiB Pc Pq long Bytes to read per chunk. . .It Sy zfs_read_history Ns = Ns Sy 0 Pq int Historical statistics for this many latest reads will be available in .Pa /proc/spl/kstat/zfs/ Ns Ao Ar pool Ac Ns Pa /reads . . .It Sy zfs_read_history_hits Ns = Ns Sy 0 Ns | Ns 1 Pq int Include cache hits in read history . .It Sy zfs_rebuild_max_segment Ns = Ns Sy 1048576 Ns B Po 1 MiB Pc Pq ulong Maximum read segment size to issue when sequentially resilvering a top-level vdev. . .It Sy zfs_rebuild_scrub_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int Automatically start a pool scrub when the last active sequential resilver completes in order to verify the checksums of all blocks which have been resilvered. This is enabled by default and strongly recommended. . .It Sy zfs_rebuild_vdev_limit Ns = Ns Sy 33554432 Ns B Po 32 MiB Pc Pq ulong Maximum amount of I/O that can be concurrently issued for a sequential resilver per leaf device, given in bytes. . .It Sy zfs_reconstruct_indirect_combinations_max Ns = Ns Sy 4096 Pq int If an indirect split block contains more than this many possible unique combinations when being reconstructed, consider it too computationally expensive to check them all. Instead, try at most this many randomly selected combinations each time the block is accessed. This allows all segment copies to participate fairly in the reconstruction when all combinations cannot be checked and prevents repeated use of one bad copy. . .It Sy zfs_recover Ns = Ns Sy 0 Ns | Ns 1 Pq int Set to attempt to recover from fatal errors. This should only be used as a last resort, as it typically results in leaked space, or worse. . .It Sy zfs_removal_ignore_errors Ns = Ns Sy 0 Ns | Ns 1 Pq int Ignore hard I/O errors during device removal. When set, if a device encounters a hard I/O error during the removal process the removal will not be cancelled. This can result in a normally recoverable block becoming permanently damaged and is hence not recommended. This should only be used as a last resort when the pool cannot be returned to a healthy state prior to removing the device. . .It Sy zfs_removal_suspend_progress Ns = Ns Sy 0 Ns | Ns 1 Pq int This is used by the test suite so that it can ensure that certain actions happen while in the middle of a removal. . .It Sy zfs_remove_max_segment Ns = Ns Sy 16777216 Ns B Po 16 MiB Pc Pq int The largest contiguous segment that we will attempt to allocate when removing a device. If there is a performance problem with attempting to allocate large blocks, consider decreasing this. The default value is also the maximum. . .It Sy zfs_resilver_disable_defer Ns = Ns Sy 0 Ns | Ns 1 Pq int Ignore the .Sy resilver_defer feature, causing an operation that would start a resilver to immediately restart the one in progress. . .It Sy zfs_resilver_min_time_ms Ns = Ns Sy 3000 Ns ms Po 3 s Pc Pq int Resilvers are processed by the sync thread. While resilvering, it will spend at least this much time working on a resilver between TXG flushes. . .It Sy zfs_scan_ignore_errors Ns = Ns Sy 0 Ns | Ns 1 Pq int If set, remove the DTL (dirty time list) upon completion of a pool scan (scrub), even if there were unrepairable errors. Intended to be used during pool repair or recovery to stop resilvering when the pool is next imported. . .It Sy zfs_scrub_min_time_ms Ns = Ns Sy 1000 Ns ms Po 1 s Pc Pq int Scrubs are processed by the sync thread. While scrubbing, it will spend at least this much time working on a scrub between TXG flushes. . .It Sy zfs_scan_checkpoint_intval Ns = Ns Sy 7200 Ns s Po 2 hour Pc Pq int To preserve progress across reboots, the sequential scan algorithm periodically needs to stop metadata scanning and issue all the verification I/O to disk. The frequency of this flushing is determined by this tunable. . .It Sy zfs_scan_fill_weight Ns = Ns Sy 3 Pq int This tunable affects how scrub and resilver I/O segments are ordered. A higher number indicates that we care more about how filled in a segment is, while a lower number indicates we care more about the size of the extent without considering the gaps within a segment. This value is only tunable upon module insertion. Changing the value afterwards will have no effect on scrub or resilver performance. . .It Sy zfs_scan_issue_strategy Ns = Ns Sy 0 Pq int Determines the order that data will be verified while scrubbing or resilvering: .Bl -tag -compact -offset 4n -width "a" .It Sy 1 Data will be verified as sequentially as possible, given the amount of memory reserved for scrubbing .Pq see Sy zfs_scan_mem_lim_fact . This may improve scrub performance if the pool's data is very fragmented. .It Sy 2 The largest mostly-contiguous chunk of found data will be verified first. By deferring scrubbing of small segments, we may later find adjacent data to coalesce and increase the segment size. .It Sy 0 .No Use strategy Sy 1 No during normal verification .No and strategy Sy 2 No while taking a checkpoint. .El . .It Sy zfs_scan_legacy Ns = Ns Sy 0 Ns | Ns 1 Pq int If unset, indicates that scrubs and resilvers will gather metadata in memory before issuing sequential I/O. Otherwise indicates that the legacy algorithm will be used, where I/O is initiated as soon as it is discovered. Unsetting will not affect scrubs or resilvers that are already in progress. . .It Sy zfs_scan_max_ext_gap Ns = Ns Sy 2097152 Ns B Po 2 MiB Pc Pq int Sets the largest gap in bytes between scrub/resilver I/O operations that will still be considered sequential for sorting purposes. Changing this value will not affect scrubs or resilvers that are already in progress. . .It Sy zfs_scan_mem_lim_fact Ns = Ns Sy 20 Ns ^-1 Pq int Maximum fraction of RAM used for I/O sorting by sequential scan algorithm. This tunable determines the hard limit for I/O sorting memory usage. When the hard limit is reached we stop scanning metadata and start issuing data verification I/O. This is done until we get below the soft limit. . .It Sy zfs_scan_mem_lim_soft_fact Ns = Ns Sy 20 Ns ^-1 Pq int The fraction of the hard limit used to determined the soft limit for I/O sorting by the sequential scan algorithm. When we cross this limit from below no action is taken. When we cross this limit from above it is because we are issuing verification I/O. In this case (unless the metadata scan is done) we stop issuing verification I/O and start scanning metadata again until we get to the hard limit. . .It Sy zfs_scan_strict_mem_lim Ns = Ns Sy 0 Ns | Ns 1 Pq int Enforce tight memory limits on pool scans when a sequential scan is in progress. When disabled, the memory limit may be exceeded by fast disks. . .It Sy zfs_scan_suspend_progress Ns = Ns Sy 0 Ns | Ns 1 Pq int Freezes a scrub/resilver in progress without actually pausing it. Intended for testing/debugging. . .It Sy zfs_scan_vdev_limit Ns = Ns Sy 4194304 Ns B Po 4 MiB Pc Pq int Maximum amount of data that can be concurrently issued at once for scrubs and resilvers per leaf device, given in bytes. . .It Sy zfs_send_corrupt_data Ns = Ns Sy 0 Ns | Ns 1 Pq int Allow sending of corrupt data (ignore read/checksum errors when sending). . .It Sy zfs_send_unmodified_spill_blocks Ns = Ns Sy 1 Ns | Ns 0 Pq int Include unmodified spill blocks in the send stream. Under certain circumstances, previous versions of ZFS could incorrectly remove the spill block from an existing object. Including unmodified copies of the spill blocks creates a backwards-compatible stream which will recreate a spill block if it was incorrectly removed. . .It Sy zfs_send_no_prefetch_queue_ff Ns = Ns Sy 20 Ns ^\-1 Pq int The fill fraction of the .Nm zfs Cm send internal queues. The fill fraction controls the timing with which internal threads are woken up. . .It Sy zfs_send_no_prefetch_queue_length Ns = Ns Sy 1048576 Ns B Po 1 MiB Pc Pq int The maximum number of bytes allowed in .Nm zfs Cm send Ns 's internal queues. . .It Sy zfs_send_queue_ff Ns = Ns Sy 20 Ns ^\-1 Pq int The fill fraction of the .Nm zfs Cm send prefetch queue. The fill fraction controls the timing with which internal threads are woken up. . .It Sy zfs_send_queue_length Ns = Ns Sy 16777216 Ns B Po 16 MiB Pc Pq int The maximum number of bytes allowed that will be prefetched by .Nm zfs Cm send . This value must be at least twice the maximum block size in use. . .It Sy zfs_recv_queue_ff Ns = Ns Sy 20 Ns ^\-1 Pq int The fill fraction of the .Nm zfs Cm receive queue. The fill fraction controls the timing with which internal threads are woken up. . .It Sy zfs_recv_queue_length Ns = Ns Sy 16777216 Ns B Po 16 MiB Pc Pq int The maximum number of bytes allowed in the .Nm zfs Cm receive queue. This value must be at least twice the maximum block size in use. . .It Sy zfs_recv_write_batch_size Ns = Ns Sy 1048576 Ns B Po 1 MiB Pc Pq int The maximum amount of data, in bytes, that .Nm zfs Cm receive will write in one DMU transaction. This is the uncompressed size, even when receiving a compressed send stream. This setting will not reduce the write size below a single block. Capped at a maximum of .Sy 32 MiB . . .It Sy zfs_recv_best_effort_corrective Ns = Ns Sy 0 Pq int When this variable is set to non-zero a corrective receive: .Bl -enum -compact -offset 4n -width "1." .It Does not enforce the restriction of source & destination snapshot GUIDs matching. .It If there is an error during healing, the healing receive is not terminated instead it moves on to the next record. .El . .It Sy zfs_override_estimate_recordsize Ns = Ns Sy 0 Ns | Ns 1 Pq ulong Setting this variable overrides the default logic for estimating block sizes when doing a .Nm zfs Cm send . The default heuristic is that the average block size will be the current recordsize. Override this value if most data in your dataset is not of that size and you require accurate zfs send size estimates. . .It Sy zfs_sync_pass_deferred_free Ns = Ns Sy 2 Pq int Flushing of data to disk is done in passes. Defer frees starting in this pass. . .It Sy zfs_spa_discard_memory_limit Ns = Ns Sy 16777216 Ns B Po 16 MiB Pc Pq int Maximum memory used for prefetching a checkpoint's space map on each vdev while discarding the checkpoint. . .It Sy zfs_special_class_metadata_reserve_pct Ns = Ns Sy 25 Ns % Pq int Only allow small data blocks to be allocated on the special and dedup vdev types when the available free space percentage on these vdevs exceeds this value. This ensures reserved space is available for pool metadata as the special vdevs approach capacity. . .It Sy zfs_sync_pass_dont_compress Ns = Ns Sy 8 Pq int Starting in this sync pass, disable compression (including of metadata). With the default setting, in practice, we don't have this many sync passes, so this has no effect. .Pp The original intent was that disabling compression would help the sync passes to converge. However, in practice, disabling compression increases the average number of sync passes; because when we turn compression off, many blocks' size will change, and thus we have to re-allocate (not overwrite) them. It also increases the number of .Em 128 KiB allocations (e.g. for indirect blocks and spacemaps) because these will not be compressed. The .Em 128 KiB allocations are especially detrimental to performance on highly fragmented systems, which may have very few free segments of this size, and may need to load new metaslabs to satisfy these allocations. . .It Sy zfs_sync_pass_rewrite Ns = Ns Sy 2 Pq int Rewrite new block pointers starting in this pass. . .It Sy zfs_sync_taskq_batch_pct Ns = Ns Sy 75 Ns % Pq int This controls the number of threads used by .Sy dp_sync_taskq . The default value of .Sy 75% will create a maximum of one thread per CPU. . .It Sy zfs_trim_extent_bytes_max Ns = Ns Sy 134217728 Ns B Po 128 MiB Pc Pq uint Maximum size of TRIM command. Larger ranges will be split into chunks no larger than this value before issuing. . .It Sy zfs_trim_extent_bytes_min Ns = Ns Sy 32768 Ns B Po 32 KiB Pc Pq uint Minimum size of TRIM commands. TRIM ranges smaller than this will be skipped, unless they're part of a larger range which was chunked. This is done because it's common for these small TRIMs to negatively impact overall performance. . .It Sy zfs_trim_metaslab_skip Ns = Ns Sy 0 Ns | Ns 1 Pq uint Skip uninitialized metaslabs during the TRIM process. This option is useful for pools constructed from large thinly-provisioned devices where TRIM operations are slow. As a pool ages, an increasing fraction of the pool's metaslabs will be initialized, progressively degrading the usefulness of this option. This setting is stored when starting a manual TRIM and will persist for the duration of the requested TRIM. . .It Sy zfs_trim_queue_limit Ns = Ns Sy 10 Pq uint Maximum number of queued TRIMs outstanding per leaf vdev. The number of concurrent TRIM commands issued to the device is controlled by .Sy zfs_vdev_trim_min_active No and Sy zfs_vdev_trim_max_active . . .It Sy zfs_trim_txg_batch Ns = Ns Sy 32 Pq uint The number of transaction groups' worth of frees which should be aggregated before TRIM operations are issued to the device. This setting represents a trade-off between issuing larger, more efficient TRIM operations and the delay before the recently trimmed space is available for use by the device. .Pp Increasing this value will allow frees to be aggregated for a longer time. This will result is larger TRIM operations and potentially increased memory usage. Decreasing this value will have the opposite effect. The default of .Sy 32 was determined to be a reasonable compromise. . .It Sy zfs_txg_history Ns = Ns Sy 0 Pq int Historical statistics for this many latest TXGs will be available in .Pa /proc/spl/kstat/zfs/ Ns Ao Ar pool Ac Ns Pa /TXGs . . .It Sy zfs_txg_timeout Ns = Ns Sy 5 Ns s Pq int Flush dirty data to disk at least every this many seconds (maximum TXG duration). . .It Sy zfs_vdev_aggregate_trim Ns = Ns Sy 0 Ns | Ns 1 Pq int Allow TRIM I/O operations to be aggregated. This is normally not helpful because the extents to be trimmed will have been already been aggregated by the metaslab. This option is provided for debugging and performance analysis. . .It Sy zfs_vdev_aggregation_limit Ns = Ns Sy 1048576 Ns B Po 1 MiB Pc Pq int Max vdev I/O aggregation size. . .It Sy zfs_vdev_aggregation_limit_non_rotating Ns = Ns Sy 131072 Ns B Po 128 KiB Pc Pq int Max vdev I/O aggregation size for non-rotating media. . .It Sy zfs_vdev_cache_bshift Ns = Ns Sy 16 Po 64 KiB Pc Pq int Shift size to inflate reads to. . .It Sy zfs_vdev_cache_max Ns = Ns Sy 16384 Ns B Po 16 KiB Pc Pq int Inflate reads smaller than this value to meet the .Sy zfs_vdev_cache_bshift size .Pq default Sy 64 KiB . . .It Sy zfs_vdev_cache_size Ns = Ns Sy 0 Pq int Total size of the per-disk cache in bytes. .Pp Currently this feature is disabled, as it has been found to not be helpful for performance and in some cases harmful. . .It Sy zfs_vdev_mirror_rotating_inc Ns = Ns Sy 0 Pq int A number by which the balancing algorithm increments the load calculation for the purpose of selecting the least busy mirror member when an I/O operation immediately follows its predecessor on rotational vdevs for the purpose of making decisions based on load. . .It Sy zfs_vdev_mirror_rotating_seek_inc Ns = Ns Sy 5 Pq int A number by which the balancing algorithm increments the load calculation for the purpose of selecting the least busy mirror member when an I/O operation lacks locality as defined by .Sy zfs_vdev_mirror_rotating_seek_offset . Operations within this that are not immediately following the previous operation are incremented by half. . .It Sy zfs_vdev_mirror_rotating_seek_offset Ns = Ns Sy 1048576 Ns B Po 1 MiB Pc Pq int The maximum distance for the last queued I/O operation in which the balancing algorithm considers an operation to have locality. .No See Sx ZFS I/O SCHEDULER . . .It Sy zfs_vdev_mirror_non_rotating_inc Ns = Ns Sy 0 Pq int A number by which the balancing algorithm increments the load calculation for the purpose of selecting the least busy mirror member on non-rotational vdevs when I/O operations do not immediately follow one another. . .It Sy zfs_vdev_mirror_non_rotating_seek_inc Ns = Ns Sy 1 Pq int A number by which the balancing algorithm increments the load calculation for the purpose of selecting the least busy mirror member when an I/O operation lacks locality as defined by the .Sy zfs_vdev_mirror_rotating_seek_offset . Operations within this that are not immediately following the previous operation are incremented by half. . .It Sy zfs_vdev_read_gap_limit Ns = Ns Sy 32768 Ns B Po 32 KiB Pc Pq int Aggregate read I/O operations if the on-disk gap between them is within this threshold. . .It Sy zfs_vdev_write_gap_limit Ns = Ns Sy 4096 Ns B Po 4 KiB Pc Pq int Aggregate write I/O operations if the on-disk gap between them is within this threshold. . .It Sy zfs_vdev_raidz_impl Ns = Ns Sy fastest Pq string Select the raidz parity implementation to use. .Pp Variants that don't depend on CPU-specific features may be selected on module load, as they are supported on all systems. The remaining options may only be set after the module is loaded, as they are available only if the implementations are compiled in and supported on the running system. .Pp Once the module is loaded, .Pa /sys/module/zfs/parameters/zfs_vdev_raidz_impl will show the available options, with the currently selected one enclosed in square brackets. .Pp .TS lb l l . fastest selected by built-in benchmark original original implementation scalar scalar implementation sse2 SSE2 instruction set 64-bit x86 ssse3 SSSE3 instruction set 64-bit x86 avx2 AVX2 instruction set 64-bit x86 avx512f AVX512F instruction set 64-bit x86 avx512bw AVX512F & AVX512BW instruction sets 64-bit x86 aarch64_neon NEON Aarch64/64-bit ARMv8 aarch64_neonx2 NEON with more unrolling Aarch64/64-bit ARMv8 powerpc_altivec Altivec PowerPC .TE . .It Sy zfs_vdev_scheduler Pq charp .Sy DEPRECATED . Prints warning to kernel log for compatibility. . .It Sy zfs_zevent_len_max Ns = Ns Sy 512 Pq int Max event queue length. Events in the queue can be viewed with .Xr zpool-events 8 . . .It Sy zfs_zevent_retain_max Ns = Ns Sy 2000 Pq int Maximum recent zevent records to retain for duplicate checking. Setting this to .Sy 0 disables duplicate detection. . .It Sy zfs_zevent_retain_expire_secs Ns = Ns Sy 900 Ns s Po 15 min Pc Pq int Lifespan for a recent ereport that was retained for duplicate checking. . .It Sy zfs_zil_clean_taskq_maxalloc Ns = Ns Sy 1048576 Pq int The maximum number of taskq entries that are allowed to be cached. When this limit is exceeded transaction records (itxs) will be cleaned synchronously. . .It Sy zfs_zil_clean_taskq_minalloc Ns = Ns Sy 1024 Pq int The number of taskq entries that are pre-populated when the taskq is first created and are immediately available for use. . .It Sy zfs_zil_clean_taskq_nthr_pct Ns = Ns Sy 100 Ns % Pq int This controls the number of threads used by .Sy dp_zil_clean_taskq . The default value of .Sy 100% will create a maximum of one thread per cpu. . .It Sy zil_maxblocksize Ns = Ns Sy 131072 Ns B Po 128 KiB Pc Pq int This sets the maximum block size used by the ZIL. On very fragmented pools, lowering this .Pq typically to Sy 36 KiB can improve performance. . .It Sy zil_nocacheflush Ns = Ns Sy 0 Ns | Ns 1 Pq int Disable the cache flush commands that are normally sent to disk by the ZIL after an LWB write has completed. Setting this will cause ZIL corruption on power loss if a volatile out-of-order write cache is enabled. . .It Sy zil_replay_disable Ns = Ns Sy 0 Ns | Ns 1 Pq int Disable intent logging replay. Can be disabled for recovery from corrupted ZIL. . .It Sy zil_slog_bulk Ns = Ns Sy 786432 Ns B Po 768 KiB Pc Pq ulong Limit SLOG write size per commit executed with synchronous priority. Any writes above that will be executed with lower (asynchronous) priority to limit potential SLOG device abuse by single active ZIL writer. . .It Sy zfs_zil_saxattr Ns = Ns Sy 1 Ns | Ns 0 Pq int Setting this tunable to zero disables ZIL logging of new .Sy xattr Ns = Ns Sy sa records if the .Sy org.openzfs:zilsaxattr feature is enabled on the pool. This would only be necessary to work around bugs in the ZIL logging or replay code for this record type. The tunable has no effect if the feature is disabled. . .It Sy zfs_embedded_slog_min_ms Ns = Ns Sy 64 Pq int Usually, one metaslab from each normal-class vdev is dedicated for use by the ZIL to log synchronous writes. However, if there are fewer than .Sy zfs_embedded_slog_min_ms metaslabs in the vdev, this functionality is disabled. This ensures that we don't set aside an unreasonable amount of space for the ZIL. . .It Sy zstd_earlyabort_pass Ns = Ns Sy 1 Pq int Whether heuristic for detection of incompressible data with zstd levels >= 3 using LZ4 and zstd-1 passes is enabled. . .It Sy zstd_abort_size Ns = Ns Sy 131072 Pq int Minimal uncompressed size (inclusive) of a record before the early abort heuristic will be attempted. . .It Sy zio_deadman_log_all Ns = Ns Sy 0 Ns | Ns 1 Pq int If non-zero, the zio deadman will produce debugging messages .Pq see Sy zfs_dbgmsg_enable for all zios, rather than only for leaf zios possessing a vdev. This is meant to be used by developers to gain diagnostic information for hang conditions which don't involve a mutex or other locking primitive: typically conditions in which a thread in the zio pipeline is looping indefinitely. . .It Sy zio_slow_io_ms Ns = Ns Sy 30000 Ns ms Po 30 s Pc Pq int When an I/O operation takes more than this much time to complete, it's marked as slow. Each slow operation causes a delay zevent. Slow I/O counters can be seen with .Nm zpool Cm status Fl s . . .It Sy zio_dva_throttle_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int Throttle block allocations in the I/O pipeline. This allows for dynamic allocation distribution when devices are imbalanced. When enabled, the maximum number of pending allocations per top-level vdev is limited by .Sy zfs_vdev_queue_depth_pct . . .It Sy zfs_xattr_compat Ns = Ns 0 Ns | Ns 1 Pq int Control the naming scheme used when setting new xattrs in the user namespace. If .Sy 0 .Pq the default on Linux , user namespace xattr names are prefixed with the namespace, to be backwards compatible with previous versions of ZFS on Linux. If .Sy 1 .Pq the default on Fx , user namespace xattr names are not prefixed, to be backwards compatible with previous versions of ZFS on illumos and .Fx . .Pp Either naming scheme can be read on this and future versions of ZFS, regardless of this tunable, but legacy ZFS on illumos or .Fx are unable to read user namespace xattrs written in the Linux format, and legacy versions of ZFS on Linux are unable to read user namespace xattrs written in the legacy ZFS format. .Pp An existing xattr with the alternate naming scheme is removed when overwriting the xattr so as to not accumulate duplicates. . .It Sy zio_requeue_io_start_cut_in_line Ns = Ns Sy 0 Ns | Ns 1 Pq int Prioritize requeued I/O. . .It Sy zio_taskq_batch_pct Ns = Ns Sy 80 Ns % Pq uint Percentage of online CPUs which will run a worker thread for I/O. These workers are responsible for I/O work such as compression and checksum calculations. Fractional number of CPUs will be rounded down. .Pp The default value of .Sy 80% was chosen to avoid using all CPUs which can result in latency issues and inconsistent application performance, especially when slower compression and/or checksumming is enabled. . .It Sy zio_taskq_batch_tpq Ns = Ns Sy 0 Pq uint Number of worker threads per taskq. Lower values improve I/O ordering and CPU utilization, while higher reduces lock contention. .Pp If .Sy 0 , generate a system-dependent value close to 6 threads per taskq. . .It Sy zvol_inhibit_dev Ns = Ns Sy 0 Ns | Ns 1 Pq uint Do not create zvol device nodes. This may slightly improve startup time on systems with a very large number of zvols. . .It Sy zvol_major Ns = Ns Sy 230 Pq uint Major number for zvol block devices. . .It Sy zvol_max_discard_blocks Ns = Ns Sy 16384 Pq ulong Discard (TRIM) operations done on zvols will be done in batches of this many blocks, where block size is determined by the .Sy volblocksize property of a zvol. . .It Sy zvol_prefetch_bytes Ns = Ns Sy 131072 Ns B Po 128 KiB Pc Pq uint When adding a zvol to the system, prefetch this many bytes from the start and end of the volume. Prefetching these regions of the volume is desirable, because they are likely to be accessed immediately by .Xr blkid 8 or the kernel partitioner. . .It Sy zvol_request_sync Ns = Ns Sy 0 Ns | Ns 1 Pq uint When processing I/O requests for a zvol, submit them synchronously. This effectively limits the queue depth to .Em 1 for each I/O submitter. When unset, requests are handled asynchronously by a thread pool. The number of requests which can be handled concurrently is controlled by .Sy zvol_threads . .Sy zvol_request_sync is ignored when running on a kernel that supports block multiqueue .Pq Li blk-mq . . .It Sy zvol_threads Ns = Ns Sy 0 Pq uint The number of system wide threads to use for processing zvol block IOs. If .Sy 0 (the default) then internally set .Sy zvol_threads to the number of CPUs present or 32 (whichever is greater). . .It Sy zvol_blk_mq_threads Ns = Ns Sy 0 Pq uint The number of threads per zvol to use for queuing IO requests. This parameter will only appear if your kernel supports .Li blk-mq and is only read and assigned to a zvol at zvol load time. If .Sy 0 (the default) then internally set .Sy zvol_blk_mq_threads to the number of CPUs present. . .It Sy zvol_use_blk_mq Ns = Ns Sy 0 Ns | Ns 1 Pq uint Set to .Sy 1 to use the .Li blk-mq API for zvols. Set to .Sy 0 (the default) to use the legacy zvol APIs. This setting can give better or worse zvol performance depending on the workload. This parameter will only appear if your kernel supports .Li blk-mq and is only read and assigned to a zvol at zvol load time. . .It Sy zvol_blk_mq_blocks_per_thread Ns = Ns Sy 8 Pq uint If .Sy zvol_use_blk_mq is enabled, then process this number of .Sy volblocksize Ns -sized blocks per zvol thread. This tunable can be use to favor better performance for zvol reads (lower values) or writes (higher values). If set to .Sy 0 , then the zvol layer will process the maximum number of blocks per thread that it can. This parameter will only appear if your kernel supports .Li blk-mq and is only applied at each zvol's load time. . .It Sy zvol_blk_mq_queue_depth Ns = Ns Sy 0 Pq uint The queue_depth value for the zvol .Li blk-mq interface. This parameter will only appear if your kernel supports .Li blk-mq and is only applied at each zvol's load time. If .Sy 0 (the default) then use the kernel's default queue depth. Values are clamped to the kernel's .Dv BLKDEV_MIN_RQ and .Dv BLKDEV_MAX_RQ Ns / Ns Dv BLKDEV_DEFAULT_RQ limits. . .It Sy zvol_volmode Ns = Ns Sy 1 Pq uint Defines zvol block devices behaviour when .Sy volmode Ns = Ns Sy default : .Bl -tag -compact -offset 4n -width "a" .It Sy 1 .No equivalent to Sy full .It Sy 2 .No equivalent to Sy dev .It Sy 3 .No equivalent to Sy none .El .El . .Sh ZFS I/O SCHEDULER ZFS issues I/O operations to leaf vdevs to satisfy and complete I/O operations. The scheduler determines when and in what order those operations are issued. The scheduler divides operations into five I/O classes, prioritized in the following order: sync read, sync write, async read, async write, and scrub/resilver. Each queue defines the minimum and maximum number of concurrent operations that may be issued to the device. In addition, the device has an aggregate maximum, .Sy zfs_vdev_max_active . Note that the sum of the per-queue minima must not exceed the aggregate maximum. If the sum of the per-queue maxima exceeds the aggregate maximum, then the number of active operations may reach .Sy zfs_vdev_max_active , in which case no further operations will be issued, regardless of whether all per-queue minima have been met. .Pp For many physical devices, throughput increases with the number of concurrent operations, but latency typically suffers. Furthermore, physical devices typically have a limit at which more concurrent operations have no effect on throughput or can actually cause it to decrease. .Pp The scheduler selects the next operation to issue by first looking for an I/O class whose minimum has not been satisfied. Once all are satisfied and the aggregate maximum has not been hit, the scheduler looks for classes whose maximum has not been satisfied. Iteration through the I/O classes is done in the order specified above. No further operations are issued if the aggregate maximum number of concurrent operations has been hit, or if there are no operations queued for an I/O class that has not hit its maximum. Every time an I/O operation is queued or an operation completes, the scheduler looks for new operations to issue. .Pp In general, smaller .Sy max_active Ns s will lead to lower latency of synchronous operations. Larger .Sy max_active Ns s may lead to higher overall throughput, depending on underlying storage. .Pp The ratio of the queues' .Sy max_active Ns s determines the balance of performance between reads, writes, and scrubs. For example, increasing .Sy zfs_vdev_scrub_max_active will cause the scrub or resilver to complete more quickly, but reads and writes to have higher latency and lower throughput. .Pp All I/O classes have a fixed maximum number of outstanding operations, except for the async write class. Asynchronous writes represent the data that is committed to stable storage during the syncing stage for transaction groups. Transaction groups enter the syncing state periodically, so the number of queued async writes will quickly burst up and then bleed down to zero. Rather than servicing them as quickly as possible, the I/O scheduler changes the maximum number of active async write operations according to the amount of dirty data in the pool. Since both throughput and latency typically increase with the number of concurrent operations issued to physical devices, reducing the burstiness in the number of concurrent operations also stabilizes the response time of operations from other – and in particular synchronous – queues. In broad strokes, the I/O scheduler will issue more concurrent operations from the async write queue as there's more dirty data in the pool. . .Ss Async Writes The number of concurrent operations issued for the async write I/O class follows a piece-wise linear function defined by a few adjustable points: .Bd -literal | o---------| <-- \fBzfs_vdev_async_write_max_active\fP ^ | /^ | | | / | | active | / | | I/O | / | | count | / | | | / | | |-------o | | <-- \fBzfs_vdev_async_write_min_active\fP 0|_______^______|_________| 0% | | 100% of \fBzfs_dirty_data_max\fP | | | `-- \fBzfs_vdev_async_write_active_max_dirty_percent\fP `--------- \fBzfs_vdev_async_write_active_min_dirty_percent\fP .Ed .Pp Until the amount of dirty data exceeds a minimum percentage of the dirty data allowed in the pool, the I/O scheduler will limit the number of concurrent operations to the minimum. As that threshold is crossed, the number of concurrent operations issued increases linearly to the maximum at the specified maximum percentage of the dirty data allowed in the pool. .Pp Ideally, the amount of dirty data on a busy pool will stay in the sloped part of the function between .Sy zfs_vdev_async_write_active_min_dirty_percent and .Sy zfs_vdev_async_write_active_max_dirty_percent . If it exceeds the maximum percentage, this indicates that the rate of incoming data is greater than the rate that the backend storage can handle. In this case, we must further throttle incoming writes, as described in the next section. . .Sh ZFS TRANSACTION DELAY We delay transactions when we've determined that the backend storage isn't able to accommodate the rate of incoming writes. .Pp If there is already a transaction waiting, we delay relative to when that transaction will finish waiting. This way the calculated delay time is independent of the number of threads concurrently executing transactions. .Pp If we are the only waiter, wait relative to when the transaction started, rather than the current time. This credits the transaction for "time already served", e.g. reading indirect blocks. .Pp The minimum time for a transaction to take is calculated as .D1 min_time = min( Ns Sy zfs_delay_scale No \(mu Po Sy dirty No \- Sy min Pc / Po Sy max No \- Sy dirty Pc , 100ms) .Pp The delay has two degrees of freedom that can be adjusted via tunables. The percentage of dirty data at which we start to delay is defined by .Sy zfs_delay_min_dirty_percent . This should typically be at or above .Sy zfs_vdev_async_write_active_max_dirty_percent , so that we only start to delay after writing at full speed has failed to keep up with the incoming write rate. The scale of the curve is defined by .Sy zfs_delay_scale . Roughly speaking, this variable determines the amount of delay at the midpoint of the curve. .Bd -literal delay 10ms +-------------------------------------------------------------*+ | *| 9ms + *+ | *| 8ms + *+ | * | 7ms + * + | * | 6ms + * + | * | 5ms + * + | * | 4ms + * + | * | 3ms + * + | * | 2ms + (midpoint) * + | | ** | 1ms + v *** + | \fBzfs_delay_scale\fP ----------> ******** | 0 +-------------------------------------*********----------------+ 0% <- \fBzfs_dirty_data_max\fP -> 100% .Ed .Pp Note, that since the delay is added to the outstanding time remaining on the most recent transaction it's effectively the inverse of IOPS. Here, the midpoint of .Em 500 us translates to .Em 2000 IOPS . The shape of the curve was chosen such that small changes in the amount of accumulated dirty data in the first three quarters of the curve yield relatively small differences in the amount of delay. .Pp The effects can be easier to understand when the amount of delay is represented on a logarithmic scale: .Bd -literal delay 100ms +-------------------------------------------------------------++ + + | | + *+ 10ms + *+ + ** + | (midpoint) ** | + | ** + 1ms + v **** + + \fBzfs_delay_scale\fP ----------> ***** + | **** | + **** + 100us + ** + + * + | * | + * + 10us + * + + + | | + + +--------------------------------------------------------------+ 0% <- \fBzfs_dirty_data_max\fP -> 100% .Ed .Pp Note here that only as the amount of dirty data approaches its limit does the delay start to increase rapidly. The goal of a properly tuned system should be to keep the amount of dirty data out of that range by first ensuring that the appropriate limits are set for the I/O scheduler to reach optimal throughput on the back-end storage, and then by changing the value of .Sy zfs_delay_scale to increase the steepness of the curve. diff --git a/module/zfs/dbuf.c b/module/zfs/dbuf.c index 9e36afd31057..85ba4e3221ad 100644 --- a/module/zfs/dbuf.c +++ b/module/zfs/dbuf.c @@ -1,5119 +1,5142 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or https://opensource.org/licenses/CDDL-1.0. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright 2011 Nexenta Systems, Inc. All rights reserved. * Copyright (c) 2012, 2020 by Delphix. All rights reserved. * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. * Copyright (c) 2019, Klara Inc. * Copyright (c) 2019, Allan Jude */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static kstat_t *dbuf_ksp; typedef struct dbuf_stats { /* * Various statistics about the size of the dbuf cache. */ kstat_named_t cache_count; kstat_named_t cache_size_bytes; kstat_named_t cache_size_bytes_max; /* * Statistics regarding the bounds on the dbuf cache size. */ kstat_named_t cache_target_bytes; kstat_named_t cache_lowater_bytes; kstat_named_t cache_hiwater_bytes; /* * Total number of dbuf cache evictions that have occurred. */ kstat_named_t cache_total_evicts; /* * The distribution of dbuf levels in the dbuf cache and * the total size of all dbufs at each level. */ kstat_named_t cache_levels[DN_MAX_LEVELS]; kstat_named_t cache_levels_bytes[DN_MAX_LEVELS]; /* * Statistics about the dbuf hash table. */ kstat_named_t hash_hits; kstat_named_t hash_misses; kstat_named_t hash_collisions; kstat_named_t hash_elements; kstat_named_t hash_elements_max; /* * Number of sublists containing more than one dbuf in the dbuf * hash table. Keep track of the longest hash chain. */ kstat_named_t hash_chains; kstat_named_t hash_chain_max; /* * Number of times a dbuf_create() discovers that a dbuf was * already created and in the dbuf hash table. */ kstat_named_t hash_insert_race; + /* + * Number of entries in the hash table dbuf and mutex arrays. + */ + kstat_named_t hash_table_count; + kstat_named_t hash_mutex_count; /* * Statistics about the size of the metadata dbuf cache. */ kstat_named_t metadata_cache_count; kstat_named_t metadata_cache_size_bytes; kstat_named_t metadata_cache_size_bytes_max; /* * For diagnostic purposes, this is incremented whenever we can't add * something to the metadata cache because it's full, and instead put * the data in the regular dbuf cache. */ kstat_named_t metadata_cache_overflow; } dbuf_stats_t; dbuf_stats_t dbuf_stats = { { "cache_count", KSTAT_DATA_UINT64 }, { "cache_size_bytes", KSTAT_DATA_UINT64 }, { "cache_size_bytes_max", KSTAT_DATA_UINT64 }, { "cache_target_bytes", KSTAT_DATA_UINT64 }, { "cache_lowater_bytes", KSTAT_DATA_UINT64 }, { "cache_hiwater_bytes", KSTAT_DATA_UINT64 }, { "cache_total_evicts", KSTAT_DATA_UINT64 }, { { "cache_levels_N", KSTAT_DATA_UINT64 } }, { { "cache_levels_bytes_N", KSTAT_DATA_UINT64 } }, { "hash_hits", KSTAT_DATA_UINT64 }, { "hash_misses", KSTAT_DATA_UINT64 }, { "hash_collisions", KSTAT_DATA_UINT64 }, { "hash_elements", KSTAT_DATA_UINT64 }, { "hash_elements_max", KSTAT_DATA_UINT64 }, { "hash_chains", KSTAT_DATA_UINT64 }, { "hash_chain_max", KSTAT_DATA_UINT64 }, { "hash_insert_race", KSTAT_DATA_UINT64 }, + { "hash_table_count", KSTAT_DATA_UINT64 }, + { "hash_mutex_count", KSTAT_DATA_UINT64 }, { "metadata_cache_count", KSTAT_DATA_UINT64 }, { "metadata_cache_size_bytes", KSTAT_DATA_UINT64 }, { "metadata_cache_size_bytes_max", KSTAT_DATA_UINT64 }, { "metadata_cache_overflow", KSTAT_DATA_UINT64 } }; struct { wmsum_t cache_count; wmsum_t cache_total_evicts; wmsum_t cache_levels[DN_MAX_LEVELS]; wmsum_t cache_levels_bytes[DN_MAX_LEVELS]; wmsum_t hash_hits; wmsum_t hash_misses; wmsum_t hash_collisions; wmsum_t hash_chains; wmsum_t hash_insert_race; wmsum_t metadata_cache_count; wmsum_t metadata_cache_overflow; } dbuf_sums; #define DBUF_STAT_INCR(stat, val) \ wmsum_add(&dbuf_sums.stat, val); #define DBUF_STAT_DECR(stat, val) \ DBUF_STAT_INCR(stat, -(val)); #define DBUF_STAT_BUMP(stat) \ DBUF_STAT_INCR(stat, 1); #define DBUF_STAT_BUMPDOWN(stat) \ DBUF_STAT_INCR(stat, -1); #define DBUF_STAT_MAX(stat, v) { \ uint64_t _m; \ while ((v) > (_m = dbuf_stats.stat.value.ui64) && \ (_m != atomic_cas_64(&dbuf_stats.stat.value.ui64, _m, (v))))\ continue; \ } static boolean_t dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx); static void dbuf_write(dbuf_dirty_record_t *dr, arc_buf_t *data, dmu_tx_t *tx); static void dbuf_sync_leaf_verify_bonus_dnode(dbuf_dirty_record_t *dr); static int dbuf_read_verify_dnode_crypt(dmu_buf_impl_t *db, uint32_t flags); /* * Global data structures and functions for the dbuf cache. */ static kmem_cache_t *dbuf_kmem_cache; static taskq_t *dbu_evict_taskq; static kthread_t *dbuf_cache_evict_thread; static kmutex_t dbuf_evict_lock; static kcondvar_t dbuf_evict_cv; static boolean_t dbuf_evict_thread_exit; /* * There are two dbuf caches; each dbuf can only be in one of them at a time. * * 1. Cache of metadata dbufs, to help make read-heavy administrative commands * from /sbin/zfs run faster. The "metadata cache" specifically stores dbufs * that represent the metadata that describes filesystems/snapshots/ * bookmarks/properties/etc. We only evict from this cache when we export a * pool, to short-circuit as much I/O as possible for all administrative * commands that need the metadata. There is no eviction policy for this * cache, because we try to only include types in it which would occupy a * very small amount of space per object but create a large impact on the * performance of these commands. Instead, after it reaches a maximum size * (which should only happen on very small memory systems with a very large * number of filesystem objects), we stop taking new dbufs into the * metadata cache, instead putting them in the normal dbuf cache. * * 2. LRU cache of dbufs. The dbuf cache maintains a list of dbufs that * are not currently held but have been recently released. These dbufs * are not eligible for arc eviction until they are aged out of the cache. * Dbufs that are aged out of the cache will be immediately destroyed and * become eligible for arc eviction. * * Dbufs are added to these caches once the last hold is released. If a dbuf is * later accessed and still exists in the dbuf cache, then it will be removed * from the cache and later re-added to the head of the cache. * * If a given dbuf meets the requirements for the metadata cache, it will go * there, otherwise it will be considered for the generic LRU dbuf cache. The * caches and the refcounts tracking their sizes are stored in an array indexed * by those caches' matching enum values (from dbuf_cached_state_t). */ typedef struct dbuf_cache { multilist_t cache; zfs_refcount_t size ____cacheline_aligned; } dbuf_cache_t; dbuf_cache_t dbuf_caches[DB_CACHE_MAX]; /* Size limits for the caches */ static unsigned long dbuf_cache_max_bytes = ULONG_MAX; static unsigned long dbuf_metadata_cache_max_bytes = ULONG_MAX; /* Set the default sizes of the caches to log2 fraction of arc size */ static int dbuf_cache_shift = 5; static int dbuf_metadata_cache_shift = 6; +/* Set the dbuf hash mutex count as log2 shift (dynamic by default) */ +static uint32_t dbuf_mutex_cache_shift = 0; + static unsigned long dbuf_cache_target_bytes(void); static unsigned long dbuf_metadata_cache_target_bytes(void); /* * The LRU dbuf cache uses a three-stage eviction policy: * - A low water marker designates when the dbuf eviction thread * should stop evicting from the dbuf cache. * - When we reach the maximum size (aka mid water mark), we * signal the eviction thread to run. * - The high water mark indicates when the eviction thread * is unable to keep up with the incoming load and eviction must * happen in the context of the calling thread. * * The dbuf cache: * (max size) * low water mid water hi water * +----------------------------------------+----------+----------+ * | | | | * | | | | * | | | | * | | | | * +----------------------------------------+----------+----------+ * stop signal evict * evicting eviction directly * thread * * The high and low water marks indicate the operating range for the eviction * thread. The low water mark is, by default, 90% of the total size of the * cache and the high water mark is at 110% (both of these percentages can be * changed by setting dbuf_cache_lowater_pct and dbuf_cache_hiwater_pct, * respectively). The eviction thread will try to ensure that the cache remains * within this range by waking up every second and checking if the cache is * above the low water mark. The thread can also be woken up by callers adding * elements into the cache if the cache is larger than the mid water (i.e max * cache size). Once the eviction thread is woken up and eviction is required, * it will continue evicting buffers until it's able to reduce the cache size * to the low water mark. If the cache size continues to grow and hits the high * water mark, then callers adding elements to the cache will begin to evict * directly from the cache until the cache is no longer above the high water * mark. */ /* * The percentage above and below the maximum cache size. */ static uint_t dbuf_cache_hiwater_pct = 10; static uint_t dbuf_cache_lowater_pct = 10; static int dbuf_cons(void *vdb, void *unused, int kmflag) { (void) unused, (void) kmflag; dmu_buf_impl_t *db = vdb; memset(db, 0, sizeof (dmu_buf_impl_t)); mutex_init(&db->db_mtx, NULL, MUTEX_DEFAULT, NULL); rw_init(&db->db_rwlock, NULL, RW_DEFAULT, NULL); cv_init(&db->db_changed, NULL, CV_DEFAULT, NULL); multilist_link_init(&db->db_cache_link); zfs_refcount_create(&db->db_holds); return (0); } static void dbuf_dest(void *vdb, void *unused) { (void) unused; dmu_buf_impl_t *db = vdb; mutex_destroy(&db->db_mtx); rw_destroy(&db->db_rwlock); cv_destroy(&db->db_changed); ASSERT(!multilist_link_active(&db->db_cache_link)); zfs_refcount_destroy(&db->db_holds); } /* * dbuf hash table routines */ static dbuf_hash_table_t dbuf_hash_table; /* * We use Cityhash for this. It's fast, and has good hash properties without * requiring any large static buffers. */ static uint64_t dbuf_hash(void *os, uint64_t obj, uint8_t lvl, uint64_t blkid) { return (cityhash4((uintptr_t)os, obj, (uint64_t)lvl, blkid)); } #define DTRACE_SET_STATE(db, why) \ DTRACE_PROBE2(dbuf__state_change, dmu_buf_impl_t *, db, \ const char *, why) #define DBUF_EQUAL(dbuf, os, obj, level, blkid) \ ((dbuf)->db.db_object == (obj) && \ (dbuf)->db_objset == (os) && \ (dbuf)->db_level == (level) && \ (dbuf)->db_blkid == (blkid)) dmu_buf_impl_t * dbuf_find(objset_t *os, uint64_t obj, uint8_t level, uint64_t blkid) { dbuf_hash_table_t *h = &dbuf_hash_table; uint64_t hv; uint64_t idx; dmu_buf_impl_t *db; hv = dbuf_hash(os, obj, level, blkid); idx = hv & h->hash_table_mask; mutex_enter(DBUF_HASH_MUTEX(h, idx)); for (db = h->hash_table[idx]; db != NULL; db = db->db_hash_next) { if (DBUF_EQUAL(db, os, obj, level, blkid)) { mutex_enter(&db->db_mtx); if (db->db_state != DB_EVICTING) { mutex_exit(DBUF_HASH_MUTEX(h, idx)); return (db); } mutex_exit(&db->db_mtx); } } mutex_exit(DBUF_HASH_MUTEX(h, idx)); return (NULL); } static dmu_buf_impl_t * dbuf_find_bonus(objset_t *os, uint64_t object) { dnode_t *dn; dmu_buf_impl_t *db = NULL; if (dnode_hold(os, object, FTAG, &dn) == 0) { rw_enter(&dn->dn_struct_rwlock, RW_READER); if (dn->dn_bonus != NULL) { db = dn->dn_bonus; mutex_enter(&db->db_mtx); } rw_exit(&dn->dn_struct_rwlock); dnode_rele(dn, FTAG); } return (db); } /* * Insert an entry into the hash table. If there is already an element * equal to elem in the hash table, then the already existing element * will be returned and the new element will not be inserted. * Otherwise returns NULL. */ static dmu_buf_impl_t * dbuf_hash_insert(dmu_buf_impl_t *db) { dbuf_hash_table_t *h = &dbuf_hash_table; objset_t *os = db->db_objset; uint64_t obj = db->db.db_object; int level = db->db_level; uint64_t blkid, hv, idx; dmu_buf_impl_t *dbf; uint32_t i; blkid = db->db_blkid; hv = dbuf_hash(os, obj, level, blkid); idx = hv & h->hash_table_mask; mutex_enter(DBUF_HASH_MUTEX(h, idx)); for (dbf = h->hash_table[idx], i = 0; dbf != NULL; dbf = dbf->db_hash_next, i++) { if (DBUF_EQUAL(dbf, os, obj, level, blkid)) { mutex_enter(&dbf->db_mtx); if (dbf->db_state != DB_EVICTING) { mutex_exit(DBUF_HASH_MUTEX(h, idx)); return (dbf); } mutex_exit(&dbf->db_mtx); } } if (i > 0) { DBUF_STAT_BUMP(hash_collisions); if (i == 1) DBUF_STAT_BUMP(hash_chains); DBUF_STAT_MAX(hash_chain_max, i); } mutex_enter(&db->db_mtx); db->db_hash_next = h->hash_table[idx]; h->hash_table[idx] = db; mutex_exit(DBUF_HASH_MUTEX(h, idx)); uint64_t he = atomic_inc_64_nv(&dbuf_stats.hash_elements.value.ui64); DBUF_STAT_MAX(hash_elements_max, he); return (NULL); } /* * This returns whether this dbuf should be stored in the metadata cache, which * is based on whether it's from one of the dnode types that store data related * to traversing dataset hierarchies. */ static boolean_t dbuf_include_in_metadata_cache(dmu_buf_impl_t *db) { DB_DNODE_ENTER(db); dmu_object_type_t type = DB_DNODE(db)->dn_type; DB_DNODE_EXIT(db); /* Check if this dbuf is one of the types we care about */ if (DMU_OT_IS_METADATA_CACHED(type)) { /* If we hit this, then we set something up wrong in dmu_ot */ ASSERT(DMU_OT_IS_METADATA(type)); /* * Sanity check for small-memory systems: don't allocate too * much memory for this purpose. */ if (zfs_refcount_count( &dbuf_caches[DB_DBUF_METADATA_CACHE].size) > dbuf_metadata_cache_target_bytes()) { DBUF_STAT_BUMP(metadata_cache_overflow); return (B_FALSE); } return (B_TRUE); } return (B_FALSE); } /* * Remove an entry from the hash table. It must be in the EVICTING state. */ static void dbuf_hash_remove(dmu_buf_impl_t *db) { dbuf_hash_table_t *h = &dbuf_hash_table; uint64_t hv, idx; dmu_buf_impl_t *dbf, **dbp; hv = dbuf_hash(db->db_objset, db->db.db_object, db->db_level, db->db_blkid); idx = hv & h->hash_table_mask; /* * We mustn't hold db_mtx to maintain lock ordering: * DBUF_HASH_MUTEX > db_mtx. */ ASSERT(zfs_refcount_is_zero(&db->db_holds)); ASSERT(db->db_state == DB_EVICTING); ASSERT(!MUTEX_HELD(&db->db_mtx)); mutex_enter(DBUF_HASH_MUTEX(h, idx)); dbp = &h->hash_table[idx]; while ((dbf = *dbp) != db) { dbp = &dbf->db_hash_next; ASSERT(dbf != NULL); } *dbp = db->db_hash_next; db->db_hash_next = NULL; if (h->hash_table[idx] && h->hash_table[idx]->db_hash_next == NULL) DBUF_STAT_BUMPDOWN(hash_chains); mutex_exit(DBUF_HASH_MUTEX(h, idx)); atomic_dec_64(&dbuf_stats.hash_elements.value.ui64); } typedef enum { DBVU_EVICTING, DBVU_NOT_EVICTING } dbvu_verify_type_t; static void dbuf_verify_user(dmu_buf_impl_t *db, dbvu_verify_type_t verify_type) { #ifdef ZFS_DEBUG int64_t holds; if (db->db_user == NULL) return; /* Only data blocks support the attachment of user data. */ ASSERT(db->db_level == 0); /* Clients must resolve a dbuf before attaching user data. */ ASSERT(db->db.db_data != NULL); ASSERT3U(db->db_state, ==, DB_CACHED); holds = zfs_refcount_count(&db->db_holds); if (verify_type == DBVU_EVICTING) { /* * Immediate eviction occurs when holds == dirtycnt. * For normal eviction buffers, holds is zero on * eviction, except when dbuf_fix_old_data() calls * dbuf_clear_data(). However, the hold count can grow * during eviction even though db_mtx is held (see * dmu_bonus_hold() for an example), so we can only * test the generic invariant that holds >= dirtycnt. */ ASSERT3U(holds, >=, db->db_dirtycnt); } else { if (db->db_user_immediate_evict == TRUE) ASSERT3U(holds, >=, db->db_dirtycnt); else ASSERT3U(holds, >, 0); } #endif } static void dbuf_evict_user(dmu_buf_impl_t *db) { dmu_buf_user_t *dbu = db->db_user; ASSERT(MUTEX_HELD(&db->db_mtx)); if (dbu == NULL) return; dbuf_verify_user(db, DBVU_EVICTING); db->db_user = NULL; #ifdef ZFS_DEBUG if (dbu->dbu_clear_on_evict_dbufp != NULL) *dbu->dbu_clear_on_evict_dbufp = NULL; #endif /* * There are two eviction callbacks - one that we call synchronously * and one that we invoke via a taskq. The async one is useful for * avoiding lock order reversals and limiting stack depth. * * Note that if we have a sync callback but no async callback, * it's likely that the sync callback will free the structure * containing the dbu. In that case we need to take care to not * dereference dbu after calling the sync evict func. */ boolean_t has_async = (dbu->dbu_evict_func_async != NULL); if (dbu->dbu_evict_func_sync != NULL) dbu->dbu_evict_func_sync(dbu); if (has_async) { taskq_dispatch_ent(dbu_evict_taskq, dbu->dbu_evict_func_async, dbu, 0, &dbu->dbu_tqent); } } boolean_t dbuf_is_metadata(dmu_buf_impl_t *db) { /* * Consider indirect blocks and spill blocks to be meta data. */ if (db->db_level > 0 || db->db_blkid == DMU_SPILL_BLKID) { return (B_TRUE); } else { boolean_t is_metadata; DB_DNODE_ENTER(db); is_metadata = DMU_OT_IS_METADATA(DB_DNODE(db)->dn_type); DB_DNODE_EXIT(db); return (is_metadata); } } /* * We want to exclude buffers that are on a special allocation class from * L2ARC. */ boolean_t dbuf_is_l2cacheable(dmu_buf_impl_t *db) { vdev_t *vd = NULL; zfs_cache_type_t cache = db->db_objset->os_secondary_cache; blkptr_t *bp = db->db_blkptr; if (bp != NULL && !BP_IS_HOLE(bp)) { uint64_t vdev = DVA_GET_VDEV(bp->blk_dva); vdev_t *rvd = db->db_objset->os_spa->spa_root_vdev; if (vdev < rvd->vdev_children) vd = rvd->vdev_child[vdev]; if (cache == ZFS_CACHE_ALL || (dbuf_is_metadata(db) && cache == ZFS_CACHE_METADATA)) { if (vd == NULL) return (B_TRUE); if ((vd->vdev_alloc_bias != VDEV_BIAS_SPECIAL && vd->vdev_alloc_bias != VDEV_BIAS_DEDUP) || l2arc_exclude_special == 0) return (B_TRUE); } } return (B_FALSE); } static inline boolean_t dnode_level_is_l2cacheable(blkptr_t *bp, dnode_t *dn, int64_t level) { vdev_t *vd = NULL; zfs_cache_type_t cache = dn->dn_objset->os_secondary_cache; if (bp != NULL && !BP_IS_HOLE(bp)) { uint64_t vdev = DVA_GET_VDEV(bp->blk_dva); vdev_t *rvd = dn->dn_objset->os_spa->spa_root_vdev; if (vdev < rvd->vdev_children) vd = rvd->vdev_child[vdev]; if (cache == ZFS_CACHE_ALL || ((level > 0 || DMU_OT_IS_METADATA(dn->dn_handle->dnh_dnode->dn_type)) && cache == ZFS_CACHE_METADATA)) { if (vd == NULL) return (B_TRUE); if ((vd->vdev_alloc_bias != VDEV_BIAS_SPECIAL && vd->vdev_alloc_bias != VDEV_BIAS_DEDUP) || l2arc_exclude_special == 0) return (B_TRUE); } } return (B_FALSE); } /* * This function *must* return indices evenly distributed between all * sublists of the multilist. This is needed due to how the dbuf eviction * code is laid out; dbuf_evict_thread() assumes dbufs are evenly * distributed between all sublists and uses this assumption when * deciding which sublist to evict from and how much to evict from it. */ static unsigned int dbuf_cache_multilist_index_func(multilist_t *ml, void *obj) { dmu_buf_impl_t *db = obj; /* * The assumption here, is the hash value for a given * dmu_buf_impl_t will remain constant throughout it's lifetime * (i.e. it's objset, object, level and blkid fields don't change). * Thus, we don't need to store the dbuf's sublist index * on insertion, as this index can be recalculated on removal. * * Also, the low order bits of the hash value are thought to be * distributed evenly. Otherwise, in the case that the multilist * has a power of two number of sublists, each sublists' usage * would not be evenly distributed. In this context full 64bit * division would be a waste of time, so limit it to 32 bits. */ return ((unsigned int)dbuf_hash(db->db_objset, db->db.db_object, db->db_level, db->db_blkid) % multilist_get_num_sublists(ml)); } /* * The target size of the dbuf cache can grow with the ARC target, * unless limited by the tunable dbuf_cache_max_bytes. */ static inline unsigned long dbuf_cache_target_bytes(void) { return (MIN(dbuf_cache_max_bytes, arc_target_bytes() >> dbuf_cache_shift)); } /* * The target size of the dbuf metadata cache can grow with the ARC target, * unless limited by the tunable dbuf_metadata_cache_max_bytes. */ static inline unsigned long dbuf_metadata_cache_target_bytes(void) { return (MIN(dbuf_metadata_cache_max_bytes, arc_target_bytes() >> dbuf_metadata_cache_shift)); } static inline uint64_t dbuf_cache_hiwater_bytes(void) { uint64_t dbuf_cache_target = dbuf_cache_target_bytes(); return (dbuf_cache_target + (dbuf_cache_target * dbuf_cache_hiwater_pct) / 100); } static inline uint64_t dbuf_cache_lowater_bytes(void) { uint64_t dbuf_cache_target = dbuf_cache_target_bytes(); return (dbuf_cache_target - (dbuf_cache_target * dbuf_cache_lowater_pct) / 100); } static inline boolean_t dbuf_cache_above_lowater(void) { return (zfs_refcount_count(&dbuf_caches[DB_DBUF_CACHE].size) > dbuf_cache_lowater_bytes()); } /* * Evict the oldest eligible dbuf from the dbuf cache. */ static void dbuf_evict_one(void) { int idx = multilist_get_random_index(&dbuf_caches[DB_DBUF_CACHE].cache); multilist_sublist_t *mls = multilist_sublist_lock( &dbuf_caches[DB_DBUF_CACHE].cache, idx); ASSERT(!MUTEX_HELD(&dbuf_evict_lock)); dmu_buf_impl_t *db = multilist_sublist_tail(mls); while (db != NULL && mutex_tryenter(&db->db_mtx) == 0) { db = multilist_sublist_prev(mls, db); } DTRACE_PROBE2(dbuf__evict__one, dmu_buf_impl_t *, db, multilist_sublist_t *, mls); if (db != NULL) { multilist_sublist_remove(mls, db); multilist_sublist_unlock(mls); (void) zfs_refcount_remove_many( &dbuf_caches[DB_DBUF_CACHE].size, db->db.db_size, db); DBUF_STAT_BUMPDOWN(cache_levels[db->db_level]); DBUF_STAT_BUMPDOWN(cache_count); DBUF_STAT_DECR(cache_levels_bytes[db->db_level], db->db.db_size); ASSERT3U(db->db_caching_status, ==, DB_DBUF_CACHE); db->db_caching_status = DB_NO_CACHE; dbuf_destroy(db); DBUF_STAT_BUMP(cache_total_evicts); } else { multilist_sublist_unlock(mls); } } /* * The dbuf evict thread is responsible for aging out dbufs from the * cache. Once the cache has reached it's maximum size, dbufs are removed * and destroyed. The eviction thread will continue running until the size * of the dbuf cache is at or below the maximum size. Once the dbuf is aged * out of the cache it is destroyed and becomes eligible for arc eviction. */ static __attribute__((noreturn)) void dbuf_evict_thread(void *unused) { (void) unused; callb_cpr_t cpr; CALLB_CPR_INIT(&cpr, &dbuf_evict_lock, callb_generic_cpr, FTAG); mutex_enter(&dbuf_evict_lock); while (!dbuf_evict_thread_exit) { while (!dbuf_cache_above_lowater() && !dbuf_evict_thread_exit) { CALLB_CPR_SAFE_BEGIN(&cpr); (void) cv_timedwait_idle_hires(&dbuf_evict_cv, &dbuf_evict_lock, SEC2NSEC(1), MSEC2NSEC(1), 0); CALLB_CPR_SAFE_END(&cpr, &dbuf_evict_lock); } mutex_exit(&dbuf_evict_lock); /* * Keep evicting as long as we're above the low water mark * for the cache. We do this without holding the locks to * minimize lock contention. */ while (dbuf_cache_above_lowater() && !dbuf_evict_thread_exit) { dbuf_evict_one(); } mutex_enter(&dbuf_evict_lock); } dbuf_evict_thread_exit = B_FALSE; cv_broadcast(&dbuf_evict_cv); CALLB_CPR_EXIT(&cpr); /* drops dbuf_evict_lock */ thread_exit(); } /* * Wake up the dbuf eviction thread if the dbuf cache is at its max size. * If the dbuf cache is at its high water mark, then evict a dbuf from the * dbuf cache using the caller's context. */ static void dbuf_evict_notify(uint64_t size) { /* * We check if we should evict without holding the dbuf_evict_lock, * because it's OK to occasionally make the wrong decision here, * and grabbing the lock results in massive lock contention. */ if (size > dbuf_cache_target_bytes()) { if (size > dbuf_cache_hiwater_bytes()) dbuf_evict_one(); cv_signal(&dbuf_evict_cv); } } static int dbuf_kstat_update(kstat_t *ksp, int rw) { dbuf_stats_t *ds = ksp->ks_data; + dbuf_hash_table_t *h = &dbuf_hash_table; if (rw == KSTAT_WRITE) return (SET_ERROR(EACCES)); ds->cache_count.value.ui64 = wmsum_value(&dbuf_sums.cache_count); ds->cache_size_bytes.value.ui64 = zfs_refcount_count(&dbuf_caches[DB_DBUF_CACHE].size); ds->cache_target_bytes.value.ui64 = dbuf_cache_target_bytes(); ds->cache_hiwater_bytes.value.ui64 = dbuf_cache_hiwater_bytes(); ds->cache_lowater_bytes.value.ui64 = dbuf_cache_lowater_bytes(); ds->cache_total_evicts.value.ui64 = wmsum_value(&dbuf_sums.cache_total_evicts); for (int i = 0; i < DN_MAX_LEVELS; i++) { ds->cache_levels[i].value.ui64 = wmsum_value(&dbuf_sums.cache_levels[i]); ds->cache_levels_bytes[i].value.ui64 = wmsum_value(&dbuf_sums.cache_levels_bytes[i]); } ds->hash_hits.value.ui64 = wmsum_value(&dbuf_sums.hash_hits); ds->hash_misses.value.ui64 = wmsum_value(&dbuf_sums.hash_misses); ds->hash_collisions.value.ui64 = wmsum_value(&dbuf_sums.hash_collisions); ds->hash_chains.value.ui64 = wmsum_value(&dbuf_sums.hash_chains); ds->hash_insert_race.value.ui64 = wmsum_value(&dbuf_sums.hash_insert_race); + ds->hash_table_count.value.ui64 = h->hash_table_mask + 1; + ds->hash_mutex_count.value.ui64 = h->hash_mutex_mask + 1; ds->metadata_cache_count.value.ui64 = wmsum_value(&dbuf_sums.metadata_cache_count); ds->metadata_cache_size_bytes.value.ui64 = zfs_refcount_count( &dbuf_caches[DB_DBUF_METADATA_CACHE].size); ds->metadata_cache_overflow.value.ui64 = wmsum_value(&dbuf_sums.metadata_cache_overflow); return (0); } void dbuf_init(void) { - uint64_t hsize = 1ULL << 16; + uint64_t hmsize, hsize = 1ULL << 16; dbuf_hash_table_t *h = &dbuf_hash_table; - int i; /* * The hash table is big enough to fill one eighth of physical memory * with an average block size of zfs_arc_average_blocksize (default 8K). * By default, the table will take up * totalmem * sizeof(void*) / 8K (1MB per GB with 8-byte pointers). */ while (hsize * zfs_arc_average_blocksize < arc_all_memory() / 8) hsize <<= 1; -retry: - h->hash_table_mask = hsize - 1; -#if defined(_KERNEL) + h->hash_table = NULL; + while (h->hash_table == NULL) { + h->hash_table_mask = hsize - 1; + + h->hash_table = vmem_zalloc(hsize * sizeof (void *), KM_SLEEP); + if (h->hash_table == NULL) + hsize >>= 1; + + ASSERT3U(hsize, >=, 1ULL << 10); + } + /* - * Large allocations which do not require contiguous pages - * should be using vmem_alloc() in the linux kernel + * The hash table buckets are protected by an array of mutexes where + * each mutex is reponsible for protecting 128 buckets. A minimum + * array size of 8192 is targeted to avoid contention. */ - h->hash_table = vmem_zalloc(hsize * sizeof (void *), KM_SLEEP); -#else - h->hash_table = kmem_zalloc(hsize * sizeof (void *), KM_NOSLEEP); -#endif - if (h->hash_table == NULL) { - /* XXX - we should really return an error instead of assert */ - ASSERT(hsize > (1ULL << 10)); - hsize >>= 1; - goto retry; + if (dbuf_mutex_cache_shift == 0) + hmsize = MAX(hsize >> 7, 1ULL << 13); + else + hmsize = 1ULL << MIN(dbuf_mutex_cache_shift, 24); + + h->hash_mutexes = NULL; + while (h->hash_mutexes == NULL) { + h->hash_mutex_mask = hmsize - 1; + + h->hash_mutexes = vmem_zalloc(hmsize * sizeof (kmutex_t), + KM_SLEEP); + if (h->hash_mutexes == NULL) + hmsize >>= 1; } dbuf_kmem_cache = kmem_cache_create("dmu_buf_impl_t", sizeof (dmu_buf_impl_t), 0, dbuf_cons, dbuf_dest, NULL, NULL, NULL, 0); - for (i = 0; i < DBUF_MUTEXES; i++) + for (int i = 0; i < hmsize; i++) mutex_init(&h->hash_mutexes[i], NULL, MUTEX_DEFAULT, NULL); dbuf_stats_init(h); /* * All entries are queued via taskq_dispatch_ent(), so min/maxalloc * configuration is not required. */ dbu_evict_taskq = taskq_create("dbu_evict", 1, defclsyspri, 0, 0, 0); for (dbuf_cached_state_t dcs = 0; dcs < DB_CACHE_MAX; dcs++) { multilist_create(&dbuf_caches[dcs].cache, sizeof (dmu_buf_impl_t), offsetof(dmu_buf_impl_t, db_cache_link), dbuf_cache_multilist_index_func); zfs_refcount_create(&dbuf_caches[dcs].size); } dbuf_evict_thread_exit = B_FALSE; mutex_init(&dbuf_evict_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&dbuf_evict_cv, NULL, CV_DEFAULT, NULL); dbuf_cache_evict_thread = thread_create(NULL, 0, dbuf_evict_thread, NULL, 0, &p0, TS_RUN, minclsyspri); wmsum_init(&dbuf_sums.cache_count, 0); wmsum_init(&dbuf_sums.cache_total_evicts, 0); - for (i = 0; i < DN_MAX_LEVELS; i++) { + for (int i = 0; i < DN_MAX_LEVELS; i++) { wmsum_init(&dbuf_sums.cache_levels[i], 0); wmsum_init(&dbuf_sums.cache_levels_bytes[i], 0); } wmsum_init(&dbuf_sums.hash_hits, 0); wmsum_init(&dbuf_sums.hash_misses, 0); wmsum_init(&dbuf_sums.hash_collisions, 0); wmsum_init(&dbuf_sums.hash_chains, 0); wmsum_init(&dbuf_sums.hash_insert_race, 0); wmsum_init(&dbuf_sums.metadata_cache_count, 0); wmsum_init(&dbuf_sums.metadata_cache_overflow, 0); dbuf_ksp = kstat_create("zfs", 0, "dbufstats", "misc", KSTAT_TYPE_NAMED, sizeof (dbuf_stats) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL); if (dbuf_ksp != NULL) { - for (i = 0; i < DN_MAX_LEVELS; i++) { + for (int i = 0; i < DN_MAX_LEVELS; i++) { snprintf(dbuf_stats.cache_levels[i].name, KSTAT_STRLEN, "cache_level_%d", i); dbuf_stats.cache_levels[i].data_type = KSTAT_DATA_UINT64; snprintf(dbuf_stats.cache_levels_bytes[i].name, KSTAT_STRLEN, "cache_level_%d_bytes", i); dbuf_stats.cache_levels_bytes[i].data_type = KSTAT_DATA_UINT64; } dbuf_ksp->ks_data = &dbuf_stats; dbuf_ksp->ks_update = dbuf_kstat_update; kstat_install(dbuf_ksp); } } void dbuf_fini(void) { dbuf_hash_table_t *h = &dbuf_hash_table; - int i; dbuf_stats_destroy(); - for (i = 0; i < DBUF_MUTEXES; i++) + for (int i = 0; i < (h->hash_mutex_mask + 1); i++) mutex_destroy(&h->hash_mutexes[i]); -#if defined(_KERNEL) - /* - * Large allocations which do not require contiguous pages - * should be using vmem_free() in the linux kernel - */ + vmem_free(h->hash_table, (h->hash_table_mask + 1) * sizeof (void *)); -#else - kmem_free(h->hash_table, (h->hash_table_mask + 1) * sizeof (void *)); -#endif + vmem_free(h->hash_mutexes, (h->hash_mutex_mask + 1) * + sizeof (kmutex_t)); + kmem_cache_destroy(dbuf_kmem_cache); taskq_destroy(dbu_evict_taskq); mutex_enter(&dbuf_evict_lock); dbuf_evict_thread_exit = B_TRUE; while (dbuf_evict_thread_exit) { cv_signal(&dbuf_evict_cv); cv_wait(&dbuf_evict_cv, &dbuf_evict_lock); } mutex_exit(&dbuf_evict_lock); mutex_destroy(&dbuf_evict_lock); cv_destroy(&dbuf_evict_cv); for (dbuf_cached_state_t dcs = 0; dcs < DB_CACHE_MAX; dcs++) { zfs_refcount_destroy(&dbuf_caches[dcs].size); multilist_destroy(&dbuf_caches[dcs].cache); } if (dbuf_ksp != NULL) { kstat_delete(dbuf_ksp); dbuf_ksp = NULL; } wmsum_fini(&dbuf_sums.cache_count); wmsum_fini(&dbuf_sums.cache_total_evicts); - for (i = 0; i < DN_MAX_LEVELS; i++) { + for (int i = 0; i < DN_MAX_LEVELS; i++) { wmsum_fini(&dbuf_sums.cache_levels[i]); wmsum_fini(&dbuf_sums.cache_levels_bytes[i]); } wmsum_fini(&dbuf_sums.hash_hits); wmsum_fini(&dbuf_sums.hash_misses); wmsum_fini(&dbuf_sums.hash_collisions); wmsum_fini(&dbuf_sums.hash_chains); wmsum_fini(&dbuf_sums.hash_insert_race); wmsum_fini(&dbuf_sums.metadata_cache_count); wmsum_fini(&dbuf_sums.metadata_cache_overflow); } /* * Other stuff. */ #ifdef ZFS_DEBUG static void dbuf_verify(dmu_buf_impl_t *db) { dnode_t *dn; dbuf_dirty_record_t *dr; uint32_t txg_prev; ASSERT(MUTEX_HELD(&db->db_mtx)); if (!(zfs_flags & ZFS_DEBUG_DBUF_VERIFY)) return; ASSERT(db->db_objset != NULL); DB_DNODE_ENTER(db); dn = DB_DNODE(db); if (dn == NULL) { ASSERT(db->db_parent == NULL); ASSERT(db->db_blkptr == NULL); } else { ASSERT3U(db->db.db_object, ==, dn->dn_object); ASSERT3P(db->db_objset, ==, dn->dn_objset); ASSERT3U(db->db_level, <, dn->dn_nlevels); ASSERT(db->db_blkid == DMU_BONUS_BLKID || db->db_blkid == DMU_SPILL_BLKID || !avl_is_empty(&dn->dn_dbufs)); } if (db->db_blkid == DMU_BONUS_BLKID) { ASSERT(dn != NULL); ASSERT3U(db->db.db_size, >=, dn->dn_bonuslen); ASSERT3U(db->db.db_offset, ==, DMU_BONUS_BLKID); } else if (db->db_blkid == DMU_SPILL_BLKID) { ASSERT(dn != NULL); ASSERT0(db->db.db_offset); } else { ASSERT3U(db->db.db_offset, ==, db->db_blkid * db->db.db_size); } if ((dr = list_head(&db->db_dirty_records)) != NULL) { ASSERT(dr->dr_dbuf == db); txg_prev = dr->dr_txg; for (dr = list_next(&db->db_dirty_records, dr); dr != NULL; dr = list_next(&db->db_dirty_records, dr)) { ASSERT(dr->dr_dbuf == db); ASSERT(txg_prev > dr->dr_txg); txg_prev = dr->dr_txg; } } /* * We can't assert that db_size matches dn_datablksz because it * can be momentarily different when another thread is doing * dnode_set_blksz(). */ if (db->db_level == 0 && db->db.db_object == DMU_META_DNODE_OBJECT) { dr = db->db_data_pending; /* * It should only be modified in syncing context, so * make sure we only have one copy of the data. */ ASSERT(dr == NULL || dr->dt.dl.dr_data == db->db_buf); } /* verify db->db_blkptr */ if (db->db_blkptr) { if (db->db_parent == dn->dn_dbuf) { /* db is pointed to by the dnode */ /* ASSERT3U(db->db_blkid, <, dn->dn_nblkptr); */ if (DMU_OBJECT_IS_SPECIAL(db->db.db_object)) ASSERT(db->db_parent == NULL); else ASSERT(db->db_parent != NULL); if (db->db_blkid != DMU_SPILL_BLKID) ASSERT3P(db->db_blkptr, ==, &dn->dn_phys->dn_blkptr[db->db_blkid]); } else { /* db is pointed to by an indirect block */ int epb __maybe_unused = db->db_parent->db.db_size >> SPA_BLKPTRSHIFT; ASSERT3U(db->db_parent->db_level, ==, db->db_level+1); ASSERT3U(db->db_parent->db.db_object, ==, db->db.db_object); /* * dnode_grow_indblksz() can make this fail if we don't * have the parent's rwlock. XXX indblksz no longer * grows. safe to do this now? */ if (RW_LOCK_HELD(&db->db_parent->db_rwlock)) { ASSERT3P(db->db_blkptr, ==, ((blkptr_t *)db->db_parent->db.db_data + db->db_blkid % epb)); } } } if ((db->db_blkptr == NULL || BP_IS_HOLE(db->db_blkptr)) && (db->db_buf == NULL || db->db_buf->b_data) && db->db.db_data && db->db_blkid != DMU_BONUS_BLKID && db->db_state != DB_FILL && !dn->dn_free_txg) { /* * If the blkptr isn't set but they have nonzero data, * it had better be dirty, otherwise we'll lose that * data when we evict this buffer. * * There is an exception to this rule for indirect blocks; in * this case, if the indirect block is a hole, we fill in a few * fields on each of the child blocks (importantly, birth time) * to prevent hole birth times from being lost when you * partially fill in a hole. */ if (db->db_dirtycnt == 0) { if (db->db_level == 0) { uint64_t *buf = db->db.db_data; int i; for (i = 0; i < db->db.db_size >> 3; i++) { ASSERT(buf[i] == 0); } } else { blkptr_t *bps = db->db.db_data; ASSERT3U(1 << DB_DNODE(db)->dn_indblkshift, ==, db->db.db_size); /* * We want to verify that all the blkptrs in the * indirect block are holes, but we may have * automatically set up a few fields for them. * We iterate through each blkptr and verify * they only have those fields set. */ for (int i = 0; i < db->db.db_size / sizeof (blkptr_t); i++) { blkptr_t *bp = &bps[i]; ASSERT(ZIO_CHECKSUM_IS_ZERO( &bp->blk_cksum)); ASSERT( DVA_IS_EMPTY(&bp->blk_dva[0]) && DVA_IS_EMPTY(&bp->blk_dva[1]) && DVA_IS_EMPTY(&bp->blk_dva[2])); ASSERT0(bp->blk_fill); ASSERT0(bp->blk_pad[0]); ASSERT0(bp->blk_pad[1]); ASSERT(!BP_IS_EMBEDDED(bp)); ASSERT(BP_IS_HOLE(bp)); ASSERT0(bp->blk_phys_birth); } } } } DB_DNODE_EXIT(db); } #endif static void dbuf_clear_data(dmu_buf_impl_t *db) { ASSERT(MUTEX_HELD(&db->db_mtx)); dbuf_evict_user(db); ASSERT3P(db->db_buf, ==, NULL); db->db.db_data = NULL; if (db->db_state != DB_NOFILL) { db->db_state = DB_UNCACHED; DTRACE_SET_STATE(db, "clear data"); } } static void dbuf_set_data(dmu_buf_impl_t *db, arc_buf_t *buf) { ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(buf != NULL); db->db_buf = buf; ASSERT(buf->b_data != NULL); db->db.db_data = buf->b_data; } static arc_buf_t * dbuf_alloc_arcbuf(dmu_buf_impl_t *db) { spa_t *spa = db->db_objset->os_spa; return (arc_alloc_buf(spa, db, DBUF_GET_BUFC_TYPE(db), db->db.db_size)); } /* * Loan out an arc_buf for read. Return the loaned arc_buf. */ arc_buf_t * dbuf_loan_arcbuf(dmu_buf_impl_t *db) { arc_buf_t *abuf; ASSERT(db->db_blkid != DMU_BONUS_BLKID); mutex_enter(&db->db_mtx); if (arc_released(db->db_buf) || zfs_refcount_count(&db->db_holds) > 1) { int blksz = db->db.db_size; spa_t *spa = db->db_objset->os_spa; mutex_exit(&db->db_mtx); abuf = arc_loan_buf(spa, B_FALSE, blksz); memcpy(abuf->b_data, db->db.db_data, blksz); } else { abuf = db->db_buf; arc_loan_inuse_buf(abuf, db); db->db_buf = NULL; dbuf_clear_data(db); mutex_exit(&db->db_mtx); } return (abuf); } /* * Calculate which level n block references the data at the level 0 offset * provided. */ uint64_t dbuf_whichblock(const dnode_t *dn, const int64_t level, const uint64_t offset) { if (dn->dn_datablkshift != 0 && dn->dn_indblkshift != 0) { /* * The level n blkid is equal to the level 0 blkid divided by * the number of level 0s in a level n block. * * The level 0 blkid is offset >> datablkshift = * offset / 2^datablkshift. * * The number of level 0s in a level n is the number of block * pointers in an indirect block, raised to the power of level. * This is 2^(indblkshift - SPA_BLKPTRSHIFT)^level = * 2^(level*(indblkshift - SPA_BLKPTRSHIFT)). * * Thus, the level n blkid is: offset / * ((2^datablkshift)*(2^(level*(indblkshift-SPA_BLKPTRSHIFT)))) * = offset / 2^(datablkshift + level * * (indblkshift - SPA_BLKPTRSHIFT)) * = offset >> (datablkshift + level * * (indblkshift - SPA_BLKPTRSHIFT)) */ const unsigned exp = dn->dn_datablkshift + level * (dn->dn_indblkshift - SPA_BLKPTRSHIFT); if (exp >= 8 * sizeof (offset)) { /* This only happens on the highest indirection level */ ASSERT3U(level, ==, dn->dn_nlevels - 1); return (0); } ASSERT3U(exp, <, 8 * sizeof (offset)); return (offset >> exp); } else { ASSERT3U(offset, <, dn->dn_datablksz); return (0); } } /* * This function is used to lock the parent of the provided dbuf. This should be * used when modifying or reading db_blkptr. */ db_lock_type_t dmu_buf_lock_parent(dmu_buf_impl_t *db, krw_t rw, const void *tag) { enum db_lock_type ret = DLT_NONE; if (db->db_parent != NULL) { rw_enter(&db->db_parent->db_rwlock, rw); ret = DLT_PARENT; } else if (dmu_objset_ds(db->db_objset) != NULL) { rrw_enter(&dmu_objset_ds(db->db_objset)->ds_bp_rwlock, rw, tag); ret = DLT_OBJSET; } /* * We only return a DLT_NONE lock when it's the top-most indirect block * of the meta-dnode of the MOS. */ return (ret); } /* * We need to pass the lock type in because it's possible that the block will * move from being the topmost indirect block in a dnode (and thus, have no * parent) to not the top-most via an indirection increase. This would cause a * panic if we didn't pass the lock type in. */ void dmu_buf_unlock_parent(dmu_buf_impl_t *db, db_lock_type_t type, const void *tag) { if (type == DLT_PARENT) rw_exit(&db->db_parent->db_rwlock); else if (type == DLT_OBJSET) rrw_exit(&dmu_objset_ds(db->db_objset)->ds_bp_rwlock, tag); } static void dbuf_read_done(zio_t *zio, const zbookmark_phys_t *zb, const blkptr_t *bp, arc_buf_t *buf, void *vdb) { (void) zb, (void) bp; dmu_buf_impl_t *db = vdb; mutex_enter(&db->db_mtx); ASSERT3U(db->db_state, ==, DB_READ); /* * All reads are synchronous, so we must have a hold on the dbuf */ ASSERT(zfs_refcount_count(&db->db_holds) > 0); ASSERT(db->db_buf == NULL); ASSERT(db->db.db_data == NULL); if (buf == NULL) { /* i/o error */ ASSERT(zio == NULL || zio->io_error != 0); ASSERT(db->db_blkid != DMU_BONUS_BLKID); ASSERT3P(db->db_buf, ==, NULL); db->db_state = DB_UNCACHED; DTRACE_SET_STATE(db, "i/o error"); } else if (db->db_level == 0 && db->db_freed_in_flight) { /* freed in flight */ ASSERT(zio == NULL || zio->io_error == 0); arc_release(buf, db); memset(buf->b_data, 0, db->db.db_size); arc_buf_freeze(buf); db->db_freed_in_flight = FALSE; dbuf_set_data(db, buf); db->db_state = DB_CACHED; DTRACE_SET_STATE(db, "freed in flight"); } else { /* success */ ASSERT(zio == NULL || zio->io_error == 0); dbuf_set_data(db, buf); db->db_state = DB_CACHED; DTRACE_SET_STATE(db, "successful read"); } cv_broadcast(&db->db_changed); dbuf_rele_and_unlock(db, NULL, B_FALSE); } /* * Shortcut for performing reads on bonus dbufs. Returns * an error if we fail to verify the dnode associated with * a decrypted block. Otherwise success. */ static int dbuf_read_bonus(dmu_buf_impl_t *db, dnode_t *dn, uint32_t flags) { int bonuslen, max_bonuslen, err; err = dbuf_read_verify_dnode_crypt(db, flags); if (err) return (err); bonuslen = MIN(dn->dn_bonuslen, dn->dn_phys->dn_bonuslen); max_bonuslen = DN_SLOTS_TO_BONUSLEN(dn->dn_num_slots); ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(DB_DNODE_HELD(db)); ASSERT3U(bonuslen, <=, db->db.db_size); db->db.db_data = kmem_alloc(max_bonuslen, KM_SLEEP); arc_space_consume(max_bonuslen, ARC_SPACE_BONUS); if (bonuslen < max_bonuslen) memset(db->db.db_data, 0, max_bonuslen); if (bonuslen) memcpy(db->db.db_data, DN_BONUS(dn->dn_phys), bonuslen); db->db_state = DB_CACHED; DTRACE_SET_STATE(db, "bonus buffer filled"); return (0); } static void dbuf_handle_indirect_hole(dmu_buf_impl_t *db, dnode_t *dn) { blkptr_t *bps = db->db.db_data; uint32_t indbs = 1ULL << dn->dn_indblkshift; int n_bps = indbs >> SPA_BLKPTRSHIFT; for (int i = 0; i < n_bps; i++) { blkptr_t *bp = &bps[i]; ASSERT3U(BP_GET_LSIZE(db->db_blkptr), ==, indbs); BP_SET_LSIZE(bp, BP_GET_LEVEL(db->db_blkptr) == 1 ? dn->dn_datablksz : BP_GET_LSIZE(db->db_blkptr)); BP_SET_TYPE(bp, BP_GET_TYPE(db->db_blkptr)); BP_SET_LEVEL(bp, BP_GET_LEVEL(db->db_blkptr) - 1); BP_SET_BIRTH(bp, db->db_blkptr->blk_birth, 0); } } /* * Handle reads on dbufs that are holes, if necessary. This function * requires that the dbuf's mutex is held. Returns success (0) if action * was taken, ENOENT if no action was taken. */ static int dbuf_read_hole(dmu_buf_impl_t *db, dnode_t *dn) { ASSERT(MUTEX_HELD(&db->db_mtx)); int is_hole = db->db_blkptr == NULL || BP_IS_HOLE(db->db_blkptr); /* * For level 0 blocks only, if the above check fails: * Recheck BP_IS_HOLE() after dnode_block_freed() in case dnode_sync() * processes the delete record and clears the bp while we are waiting * for the dn_mtx (resulting in a "no" from block_freed). */ if (!is_hole && db->db_level == 0) { is_hole = dnode_block_freed(dn, db->db_blkid) || BP_IS_HOLE(db->db_blkptr); } if (is_hole) { dbuf_set_data(db, dbuf_alloc_arcbuf(db)); memset(db->db.db_data, 0, db->db.db_size); if (db->db_blkptr != NULL && db->db_level > 0 && BP_IS_HOLE(db->db_blkptr) && db->db_blkptr->blk_birth != 0) { dbuf_handle_indirect_hole(db, dn); } db->db_state = DB_CACHED; DTRACE_SET_STATE(db, "hole read satisfied"); return (0); } return (ENOENT); } /* * This function ensures that, when doing a decrypting read of a block, * we make sure we have decrypted the dnode associated with it. We must do * this so that we ensure we are fully authenticating the checksum-of-MACs * tree from the root of the objset down to this block. Indirect blocks are * always verified against their secure checksum-of-MACs assuming that the * dnode containing them is correct. Now that we are doing a decrypting read, * we can be sure that the key is loaded and verify that assumption. This is * especially important considering that we always read encrypted dnode * blocks as raw data (without verifying their MACs) to start, and * decrypt / authenticate them when we need to read an encrypted bonus buffer. */ static int dbuf_read_verify_dnode_crypt(dmu_buf_impl_t *db, uint32_t flags) { int err = 0; objset_t *os = db->db_objset; arc_buf_t *dnode_abuf; dnode_t *dn; zbookmark_phys_t zb; ASSERT(MUTEX_HELD(&db->db_mtx)); if (!os->os_encrypted || os->os_raw_receive || (flags & DB_RF_NO_DECRYPT) != 0) return (0); DB_DNODE_ENTER(db); dn = DB_DNODE(db); dnode_abuf = (dn->dn_dbuf != NULL) ? dn->dn_dbuf->db_buf : NULL; if (dnode_abuf == NULL || !arc_is_encrypted(dnode_abuf)) { DB_DNODE_EXIT(db); return (0); } SET_BOOKMARK(&zb, dmu_objset_id(os), DMU_META_DNODE_OBJECT, 0, dn->dn_dbuf->db_blkid); err = arc_untransform(dnode_abuf, os->os_spa, &zb, B_TRUE); /* * An error code of EACCES tells us that the key is still not * available. This is ok if we are only reading authenticated * (and therefore non-encrypted) blocks. */ if (err == EACCES && ((db->db_blkid != DMU_BONUS_BLKID && !DMU_OT_IS_ENCRYPTED(dn->dn_type)) || (db->db_blkid == DMU_BONUS_BLKID && !DMU_OT_IS_ENCRYPTED(dn->dn_bonustype)))) err = 0; DB_DNODE_EXIT(db); return (err); } /* * Drops db_mtx and the parent lock specified by dblt and tag before * returning. */ static int dbuf_read_impl(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags, db_lock_type_t dblt, const void *tag) { dnode_t *dn; zbookmark_phys_t zb; uint32_t aflags = ARC_FLAG_NOWAIT; int err, zio_flags; err = zio_flags = 0; DB_DNODE_ENTER(db); dn = DB_DNODE(db); ASSERT(!zfs_refcount_is_zero(&db->db_holds)); ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(db->db_state == DB_UNCACHED); ASSERT(db->db_buf == NULL); ASSERT(db->db_parent == NULL || RW_LOCK_HELD(&db->db_parent->db_rwlock)); if (db->db_blkid == DMU_BONUS_BLKID) { err = dbuf_read_bonus(db, dn, flags); goto early_unlock; } err = dbuf_read_hole(db, dn); if (err == 0) goto early_unlock; /* * Any attempt to read a redacted block should result in an error. This * will never happen under normal conditions, but can be useful for * debugging purposes. */ if (BP_IS_REDACTED(db->db_blkptr)) { ASSERT(dsl_dataset_feature_is_active( db->db_objset->os_dsl_dataset, SPA_FEATURE_REDACTED_DATASETS)); err = SET_ERROR(EIO); goto early_unlock; } SET_BOOKMARK(&zb, dmu_objset_id(db->db_objset), db->db.db_object, db->db_level, db->db_blkid); /* * All bps of an encrypted os should have the encryption bit set. * If this is not true it indicates tampering and we report an error. */ if (db->db_objset->os_encrypted && !BP_USES_CRYPT(db->db_blkptr)) { spa_log_error(db->db_objset->os_spa, &zb); zfs_panic_recover("unencrypted block in encrypted " "object set %llu", dmu_objset_id(db->db_objset)); err = SET_ERROR(EIO); goto early_unlock; } err = dbuf_read_verify_dnode_crypt(db, flags); if (err != 0) goto early_unlock; DB_DNODE_EXIT(db); db->db_state = DB_READ; DTRACE_SET_STATE(db, "read issued"); mutex_exit(&db->db_mtx); if (dbuf_is_l2cacheable(db)) aflags |= ARC_FLAG_L2CACHE; dbuf_add_ref(db, NULL); zio_flags = (flags & DB_RF_CANFAIL) ? ZIO_FLAG_CANFAIL : ZIO_FLAG_MUSTSUCCEED; if ((flags & DB_RF_NO_DECRYPT) && BP_IS_PROTECTED(db->db_blkptr)) zio_flags |= ZIO_FLAG_RAW; /* * The zio layer will copy the provided blkptr later, but we need to * do this now so that we can release the parent's rwlock. We have to * do that now so that if dbuf_read_done is called synchronously (on * an l1 cache hit) we don't acquire the db_mtx while holding the * parent's rwlock, which would be a lock ordering violation. */ blkptr_t bp = *db->db_blkptr; dmu_buf_unlock_parent(db, dblt, tag); (void) arc_read(zio, db->db_objset->os_spa, &bp, dbuf_read_done, db, ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb); return (err); early_unlock: DB_DNODE_EXIT(db); mutex_exit(&db->db_mtx); dmu_buf_unlock_parent(db, dblt, tag); return (err); } /* * This is our just-in-time copy function. It makes a copy of buffers that * have been modified in a previous transaction group before we access them in * the current active group. * * This function is used in three places: when we are dirtying a buffer for the * first time in a txg, when we are freeing a range in a dnode that includes * this buffer, and when we are accessing a buffer which was received compressed * and later referenced in a WRITE_BYREF record. * * Note that when we are called from dbuf_free_range() we do not put a hold on * the buffer, we just traverse the active dbuf list for the dnode. */ static void dbuf_fix_old_data(dmu_buf_impl_t *db, uint64_t txg) { dbuf_dirty_record_t *dr = list_head(&db->db_dirty_records); ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(db->db.db_data != NULL); ASSERT(db->db_level == 0); ASSERT(db->db.db_object != DMU_META_DNODE_OBJECT); if (dr == NULL || (dr->dt.dl.dr_data != ((db->db_blkid == DMU_BONUS_BLKID) ? db->db.db_data : db->db_buf))) return; /* * If the last dirty record for this dbuf has not yet synced * and its referencing the dbuf data, either: * reset the reference to point to a new copy, * or (if there a no active holders) * just null out the current db_data pointer. */ ASSERT3U(dr->dr_txg, >=, txg - 2); if (db->db_blkid == DMU_BONUS_BLKID) { dnode_t *dn = DB_DNODE(db); int bonuslen = DN_SLOTS_TO_BONUSLEN(dn->dn_num_slots); dr->dt.dl.dr_data = kmem_alloc(bonuslen, KM_SLEEP); arc_space_consume(bonuslen, ARC_SPACE_BONUS); memcpy(dr->dt.dl.dr_data, db->db.db_data, bonuslen); } else if (zfs_refcount_count(&db->db_holds) > db->db_dirtycnt) { dnode_t *dn = DB_DNODE(db); int size = arc_buf_size(db->db_buf); arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db); spa_t *spa = db->db_objset->os_spa; enum zio_compress compress_type = arc_get_compression(db->db_buf); uint8_t complevel = arc_get_complevel(db->db_buf); if (arc_is_encrypted(db->db_buf)) { boolean_t byteorder; uint8_t salt[ZIO_DATA_SALT_LEN]; uint8_t iv[ZIO_DATA_IV_LEN]; uint8_t mac[ZIO_DATA_MAC_LEN]; arc_get_raw_params(db->db_buf, &byteorder, salt, iv, mac); dr->dt.dl.dr_data = arc_alloc_raw_buf(spa, db, dmu_objset_id(dn->dn_objset), byteorder, salt, iv, mac, dn->dn_type, size, arc_buf_lsize(db->db_buf), compress_type, complevel); } else if (compress_type != ZIO_COMPRESS_OFF) { ASSERT3U(type, ==, ARC_BUFC_DATA); dr->dt.dl.dr_data = arc_alloc_compressed_buf(spa, db, size, arc_buf_lsize(db->db_buf), compress_type, complevel); } else { dr->dt.dl.dr_data = arc_alloc_buf(spa, db, type, size); } memcpy(dr->dt.dl.dr_data->b_data, db->db.db_data, size); } else { db->db_buf = NULL; dbuf_clear_data(db); } } int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags) { int err = 0; boolean_t prefetch; dnode_t *dn; /* * We don't have to hold the mutex to check db_state because it * can't be freed while we have a hold on the buffer. */ ASSERT(!zfs_refcount_is_zero(&db->db_holds)); if (db->db_state == DB_NOFILL) return (SET_ERROR(EIO)); DB_DNODE_ENTER(db); dn = DB_DNODE(db); prefetch = db->db_level == 0 && db->db_blkid != DMU_BONUS_BLKID && (flags & DB_RF_NOPREFETCH) == 0 && dn != NULL && DBUF_IS_CACHEABLE(db); mutex_enter(&db->db_mtx); if (db->db_state == DB_CACHED) { spa_t *spa = dn->dn_objset->os_spa; /* * Ensure that this block's dnode has been decrypted if * the caller has requested decrypted data. */ err = dbuf_read_verify_dnode_crypt(db, flags); /* * If the arc buf is compressed or encrypted and the caller * requested uncompressed data, we need to untransform it * before returning. We also call arc_untransform() on any * unauthenticated blocks, which will verify their MAC if * the key is now available. */ if (err == 0 && db->db_buf != NULL && (flags & DB_RF_NO_DECRYPT) == 0 && (arc_is_encrypted(db->db_buf) || arc_is_unauthenticated(db->db_buf) || arc_get_compression(db->db_buf) != ZIO_COMPRESS_OFF)) { zbookmark_phys_t zb; SET_BOOKMARK(&zb, dmu_objset_id(db->db_objset), db->db.db_object, db->db_level, db->db_blkid); dbuf_fix_old_data(db, spa_syncing_txg(spa)); err = arc_untransform(db->db_buf, spa, &zb, B_FALSE); dbuf_set_data(db, db->db_buf); } mutex_exit(&db->db_mtx); if (err == 0 && prefetch) { dmu_zfetch(&dn->dn_zfetch, db->db_blkid, 1, B_TRUE, B_FALSE, flags & DB_RF_HAVESTRUCT); } DB_DNODE_EXIT(db); DBUF_STAT_BUMP(hash_hits); } else if (db->db_state == DB_UNCACHED) { spa_t *spa = dn->dn_objset->os_spa; boolean_t need_wait = B_FALSE; db_lock_type_t dblt = dmu_buf_lock_parent(db, RW_READER, FTAG); if (zio == NULL && db->db_blkptr != NULL && !BP_IS_HOLE(db->db_blkptr)) { zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL); need_wait = B_TRUE; } err = dbuf_read_impl(db, zio, flags, dblt, FTAG); /* * dbuf_read_impl has dropped db_mtx and our parent's rwlock * for us */ if (!err && prefetch) { dmu_zfetch(&dn->dn_zfetch, db->db_blkid, 1, B_TRUE, db->db_state != DB_CACHED, flags & DB_RF_HAVESTRUCT); } DB_DNODE_EXIT(db); DBUF_STAT_BUMP(hash_misses); /* * If we created a zio_root we must execute it to avoid * leaking it, even if it isn't attached to any work due * to an error in dbuf_read_impl(). */ if (need_wait) { if (err == 0) err = zio_wait(zio); else VERIFY0(zio_wait(zio)); } } else { /* * Another reader came in while the dbuf was in flight * between UNCACHED and CACHED. Either a writer will finish * writing the buffer (sending the dbuf to CACHED) or the * first reader's request will reach the read_done callback * and send the dbuf to CACHED. Otherwise, a failure * occurred and the dbuf went to UNCACHED. */ mutex_exit(&db->db_mtx); if (prefetch) { dmu_zfetch(&dn->dn_zfetch, db->db_blkid, 1, B_TRUE, B_TRUE, flags & DB_RF_HAVESTRUCT); } DB_DNODE_EXIT(db); DBUF_STAT_BUMP(hash_misses); /* Skip the wait per the caller's request. */ if ((flags & DB_RF_NEVERWAIT) == 0) { mutex_enter(&db->db_mtx); while (db->db_state == DB_READ || db->db_state == DB_FILL) { ASSERT(db->db_state == DB_READ || (flags & DB_RF_HAVESTRUCT) == 0); DTRACE_PROBE2(blocked__read, dmu_buf_impl_t *, db, zio_t *, zio); cv_wait(&db->db_changed, &db->db_mtx); } if (db->db_state == DB_UNCACHED) err = SET_ERROR(EIO); mutex_exit(&db->db_mtx); } } return (err); } static void dbuf_noread(dmu_buf_impl_t *db) { ASSERT(!zfs_refcount_is_zero(&db->db_holds)); ASSERT(db->db_blkid != DMU_BONUS_BLKID); mutex_enter(&db->db_mtx); while (db->db_state == DB_READ || db->db_state == DB_FILL) cv_wait(&db->db_changed, &db->db_mtx); if (db->db_state == DB_UNCACHED) { ASSERT(db->db_buf == NULL); ASSERT(db->db.db_data == NULL); dbuf_set_data(db, dbuf_alloc_arcbuf(db)); db->db_state = DB_FILL; DTRACE_SET_STATE(db, "assigning filled buffer"); } else if (db->db_state == DB_NOFILL) { dbuf_clear_data(db); } else { ASSERT3U(db->db_state, ==, DB_CACHED); } mutex_exit(&db->db_mtx); } void dbuf_unoverride(dbuf_dirty_record_t *dr) { dmu_buf_impl_t *db = dr->dr_dbuf; blkptr_t *bp = &dr->dt.dl.dr_overridden_by; uint64_t txg = dr->dr_txg; ASSERT(MUTEX_HELD(&db->db_mtx)); /* * This assert is valid because dmu_sync() expects to be called by * a zilog's get_data while holding a range lock. This call only * comes from dbuf_dirty() callers who must also hold a range lock. */ ASSERT(dr->dt.dl.dr_override_state != DR_IN_DMU_SYNC); ASSERT(db->db_level == 0); if (db->db_blkid == DMU_BONUS_BLKID || dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN) return; ASSERT(db->db_data_pending != dr); /* free this block */ if (!BP_IS_HOLE(bp) && !dr->dt.dl.dr_nopwrite) zio_free(db->db_objset->os_spa, txg, bp); dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; dr->dt.dl.dr_nopwrite = B_FALSE; dr->dt.dl.dr_has_raw_params = B_FALSE; /* * Release the already-written buffer, so we leave it in * a consistent dirty state. Note that all callers are * modifying the buffer, so they will immediately do * another (redundant) arc_release(). Therefore, leave * the buf thawed to save the effort of freezing & * immediately re-thawing it. */ arc_release(dr->dt.dl.dr_data, db); } /* * Evict (if its unreferenced) or clear (if its referenced) any level-0 * data blocks in the free range, so that any future readers will find * empty blocks. */ void dbuf_free_range(dnode_t *dn, uint64_t start_blkid, uint64_t end_blkid, dmu_tx_t *tx) { dmu_buf_impl_t *db_search; dmu_buf_impl_t *db, *db_next; uint64_t txg = tx->tx_txg; avl_index_t where; dbuf_dirty_record_t *dr; if (end_blkid > dn->dn_maxblkid && !(start_blkid == DMU_SPILL_BLKID || end_blkid == DMU_SPILL_BLKID)) end_blkid = dn->dn_maxblkid; dprintf_dnode(dn, "start=%llu end=%llu\n", (u_longlong_t)start_blkid, (u_longlong_t)end_blkid); db_search = kmem_alloc(sizeof (dmu_buf_impl_t), KM_SLEEP); db_search->db_level = 0; db_search->db_blkid = start_blkid; db_search->db_state = DB_SEARCH; mutex_enter(&dn->dn_dbufs_mtx); db = avl_find(&dn->dn_dbufs, db_search, &where); ASSERT3P(db, ==, NULL); db = avl_nearest(&dn->dn_dbufs, where, AVL_AFTER); for (; db != NULL; db = db_next) { db_next = AVL_NEXT(&dn->dn_dbufs, db); ASSERT(db->db_blkid != DMU_BONUS_BLKID); if (db->db_level != 0 || db->db_blkid > end_blkid) { break; } ASSERT3U(db->db_blkid, >=, start_blkid); /* found a level 0 buffer in the range */ mutex_enter(&db->db_mtx); if (dbuf_undirty(db, tx)) { /* mutex has been dropped and dbuf destroyed */ continue; } if (db->db_state == DB_UNCACHED || db->db_state == DB_NOFILL || db->db_state == DB_EVICTING) { ASSERT(db->db.db_data == NULL); mutex_exit(&db->db_mtx); continue; } if (db->db_state == DB_READ || db->db_state == DB_FILL) { /* will be handled in dbuf_read_done or dbuf_rele */ db->db_freed_in_flight = TRUE; mutex_exit(&db->db_mtx); continue; } if (zfs_refcount_count(&db->db_holds) == 0) { ASSERT(db->db_buf); dbuf_destroy(db); continue; } /* The dbuf is referenced */ dr = list_head(&db->db_dirty_records); if (dr != NULL) { if (dr->dr_txg == txg) { /* * This buffer is "in-use", re-adjust the file * size to reflect that this buffer may * contain new data when we sync. */ if (db->db_blkid != DMU_SPILL_BLKID && db->db_blkid > dn->dn_maxblkid) dn->dn_maxblkid = db->db_blkid; dbuf_unoverride(dr); } else { /* * This dbuf is not dirty in the open context. * Either uncache it (if its not referenced in * the open context) or reset its contents to * empty. */ dbuf_fix_old_data(db, txg); } } /* clear the contents if its cached */ if (db->db_state == DB_CACHED) { ASSERT(db->db.db_data != NULL); arc_release(db->db_buf, db); rw_enter(&db->db_rwlock, RW_WRITER); memset(db->db.db_data, 0, db->db.db_size); rw_exit(&db->db_rwlock); arc_buf_freeze(db->db_buf); } mutex_exit(&db->db_mtx); } mutex_exit(&dn->dn_dbufs_mtx); kmem_free(db_search, sizeof (dmu_buf_impl_t)); } void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx) { arc_buf_t *buf, *old_buf; dbuf_dirty_record_t *dr; int osize = db->db.db_size; arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db); dnode_t *dn; ASSERT(db->db_blkid != DMU_BONUS_BLKID); DB_DNODE_ENTER(db); dn = DB_DNODE(db); /* * XXX we should be doing a dbuf_read, checking the return * value and returning that up to our callers */ dmu_buf_will_dirty(&db->db, tx); /* create the data buffer for the new block */ buf = arc_alloc_buf(dn->dn_objset->os_spa, db, type, size); /* copy old block data to the new block */ old_buf = db->db_buf; memcpy(buf->b_data, old_buf->b_data, MIN(osize, size)); /* zero the remainder */ if (size > osize) memset((uint8_t *)buf->b_data + osize, 0, size - osize); mutex_enter(&db->db_mtx); dbuf_set_data(db, buf); arc_buf_destroy(old_buf, db); db->db.db_size = size; dr = list_head(&db->db_dirty_records); /* dirty record added by dmu_buf_will_dirty() */ VERIFY(dr != NULL); if (db->db_level == 0) dr->dt.dl.dr_data = buf; ASSERT3U(dr->dr_txg, ==, tx->tx_txg); ASSERT3U(dr->dr_accounted, ==, osize); dr->dr_accounted = size; mutex_exit(&db->db_mtx); dmu_objset_willuse_space(dn->dn_objset, size - osize, tx); DB_DNODE_EXIT(db); } void dbuf_release_bp(dmu_buf_impl_t *db) { objset_t *os __maybe_unused = db->db_objset; ASSERT(dsl_pool_sync_context(dmu_objset_pool(os))); ASSERT(arc_released(os->os_phys_buf) || list_link_active(&os->os_dsl_dataset->ds_synced_link)); ASSERT(db->db_parent == NULL || arc_released(db->db_parent->db_buf)); (void) arc_release(db->db_buf, db); } /* * We already have a dirty record for this TXG, and we are being * dirtied again. */ static void dbuf_redirty(dbuf_dirty_record_t *dr) { dmu_buf_impl_t *db = dr->dr_dbuf; ASSERT(MUTEX_HELD(&db->db_mtx)); if (db->db_level == 0 && db->db_blkid != DMU_BONUS_BLKID) { /* * If this buffer has already been written out, * we now need to reset its state. */ dbuf_unoverride(dr); if (db->db.db_object != DMU_META_DNODE_OBJECT && db->db_state != DB_NOFILL) { /* Already released on initial dirty, so just thaw. */ ASSERT(arc_released(db->db_buf)); arc_buf_thaw(db->db_buf); } } } dbuf_dirty_record_t * dbuf_dirty_lightweight(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx) { rw_enter(&dn->dn_struct_rwlock, RW_READER); IMPLY(dn->dn_objset->os_raw_receive, dn->dn_maxblkid >= blkid); dnode_new_blkid(dn, blkid, tx, B_TRUE, B_FALSE); ASSERT(dn->dn_maxblkid >= blkid); dbuf_dirty_record_t *dr = kmem_zalloc(sizeof (*dr), KM_SLEEP); list_link_init(&dr->dr_dirty_node); list_link_init(&dr->dr_dbuf_node); dr->dr_dnode = dn; dr->dr_txg = tx->tx_txg; dr->dt.dll.dr_blkid = blkid; dr->dr_accounted = dn->dn_datablksz; /* * There should not be any dbuf for the block that we're dirtying. * Otherwise the buffer contents could be inconsistent between the * dbuf and the lightweight dirty record. */ ASSERT3P(NULL, ==, dbuf_find(dn->dn_objset, dn->dn_object, 0, blkid)); mutex_enter(&dn->dn_mtx); int txgoff = tx->tx_txg & TXG_MASK; if (dn->dn_free_ranges[txgoff] != NULL) { range_tree_clear(dn->dn_free_ranges[txgoff], blkid, 1); } if (dn->dn_nlevels == 1) { ASSERT3U(blkid, <, dn->dn_nblkptr); list_insert_tail(&dn->dn_dirty_records[txgoff], dr); mutex_exit(&dn->dn_mtx); rw_exit(&dn->dn_struct_rwlock); dnode_setdirty(dn, tx); } else { mutex_exit(&dn->dn_mtx); int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; dmu_buf_impl_t *parent_db = dbuf_hold_level(dn, 1, blkid >> epbs, FTAG); rw_exit(&dn->dn_struct_rwlock); if (parent_db == NULL) { kmem_free(dr, sizeof (*dr)); return (NULL); } int err = dbuf_read(parent_db, NULL, (DB_RF_NOPREFETCH | DB_RF_CANFAIL)); if (err != 0) { dbuf_rele(parent_db, FTAG); kmem_free(dr, sizeof (*dr)); return (NULL); } dbuf_dirty_record_t *parent_dr = dbuf_dirty(parent_db, tx); dbuf_rele(parent_db, FTAG); mutex_enter(&parent_dr->dt.di.dr_mtx); ASSERT3U(parent_dr->dr_txg, ==, tx->tx_txg); list_insert_tail(&parent_dr->dt.di.dr_children, dr); mutex_exit(&parent_dr->dt.di.dr_mtx); dr->dr_parent = parent_dr; } dmu_objset_willuse_space(dn->dn_objset, dr->dr_accounted, tx); return (dr); } dbuf_dirty_record_t * dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx) { dnode_t *dn; objset_t *os; dbuf_dirty_record_t *dr, *dr_next, *dr_head; int txgoff = tx->tx_txg & TXG_MASK; boolean_t drop_struct_rwlock = B_FALSE; ASSERT(tx->tx_txg != 0); ASSERT(!zfs_refcount_is_zero(&db->db_holds)); DMU_TX_DIRTY_BUF(tx, db); DB_DNODE_ENTER(db); dn = DB_DNODE(db); /* * Shouldn't dirty a regular buffer in syncing context. Private * objects may be dirtied in syncing context, but only if they * were already pre-dirtied in open context. */ #ifdef ZFS_DEBUG if (dn->dn_objset->os_dsl_dataset != NULL) { rrw_enter(&dn->dn_objset->os_dsl_dataset->ds_bp_rwlock, RW_READER, FTAG); } ASSERT(!dmu_tx_is_syncing(tx) || BP_IS_HOLE(dn->dn_objset->os_rootbp) || DMU_OBJECT_IS_SPECIAL(dn->dn_object) || dn->dn_objset->os_dsl_dataset == NULL); if (dn->dn_objset->os_dsl_dataset != NULL) rrw_exit(&dn->dn_objset->os_dsl_dataset->ds_bp_rwlock, FTAG); #endif /* * We make this assert for private objects as well, but after we * check if we're already dirty. They are allowed to re-dirty * in syncing context. */ ASSERT(dn->dn_object == DMU_META_DNODE_OBJECT || dn->dn_dirtyctx == DN_UNDIRTIED || dn->dn_dirtyctx == (dmu_tx_is_syncing(tx) ? DN_DIRTY_SYNC : DN_DIRTY_OPEN)); mutex_enter(&db->db_mtx); /* * XXX make this true for indirects too? The problem is that * transactions created with dmu_tx_create_assigned() from * syncing context don't bother holding ahead. */ ASSERT(db->db_level != 0 || db->db_state == DB_CACHED || db->db_state == DB_FILL || db->db_state == DB_NOFILL); mutex_enter(&dn->dn_mtx); dnode_set_dirtyctx(dn, tx, db); if (tx->tx_txg > dn->dn_dirty_txg) dn->dn_dirty_txg = tx->tx_txg; mutex_exit(&dn->dn_mtx); if (db->db_blkid == DMU_SPILL_BLKID) dn->dn_have_spill = B_TRUE; /* * If this buffer is already dirty, we're done. */ dr_head = list_head(&db->db_dirty_records); ASSERT(dr_head == NULL || dr_head->dr_txg <= tx->tx_txg || db->db.db_object == DMU_META_DNODE_OBJECT); dr_next = dbuf_find_dirty_lte(db, tx->tx_txg); if (dr_next && dr_next->dr_txg == tx->tx_txg) { DB_DNODE_EXIT(db); dbuf_redirty(dr_next); mutex_exit(&db->db_mtx); return (dr_next); } /* * Only valid if not already dirty. */ ASSERT(dn->dn_object == 0 || dn->dn_dirtyctx == DN_UNDIRTIED || dn->dn_dirtyctx == (dmu_tx_is_syncing(tx) ? DN_DIRTY_SYNC : DN_DIRTY_OPEN)); ASSERT3U(dn->dn_nlevels, >, db->db_level); /* * We should only be dirtying in syncing context if it's the * mos or we're initializing the os or it's a special object. * However, we are allowed to dirty in syncing context provided * we already dirtied it in open context. Hence we must make * this assertion only if we're not already dirty. */ os = dn->dn_objset; VERIFY3U(tx->tx_txg, <=, spa_final_dirty_txg(os->os_spa)); #ifdef ZFS_DEBUG if (dn->dn_objset->os_dsl_dataset != NULL) rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_READER, FTAG); ASSERT(!dmu_tx_is_syncing(tx) || DMU_OBJECT_IS_SPECIAL(dn->dn_object) || os->os_dsl_dataset == NULL || BP_IS_HOLE(os->os_rootbp)); if (dn->dn_objset->os_dsl_dataset != NULL) rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG); #endif ASSERT(db->db.db_size != 0); dprintf_dbuf(db, "size=%llx\n", (u_longlong_t)db->db.db_size); if (db->db_blkid != DMU_BONUS_BLKID) { dmu_objset_willuse_space(os, db->db.db_size, tx); } /* * If this buffer is dirty in an old transaction group we need * to make a copy of it so that the changes we make in this * transaction group won't leak out when we sync the older txg. */ dr = kmem_zalloc(sizeof (dbuf_dirty_record_t), KM_SLEEP); list_link_init(&dr->dr_dirty_node); list_link_init(&dr->dr_dbuf_node); dr->dr_dnode = dn; if (db->db_level == 0) { void *data_old = db->db_buf; if (db->db_state != DB_NOFILL) { if (db->db_blkid == DMU_BONUS_BLKID) { dbuf_fix_old_data(db, tx->tx_txg); data_old = db->db.db_data; } else if (db->db.db_object != DMU_META_DNODE_OBJECT) { /* * Release the data buffer from the cache so * that we can modify it without impacting * possible other users of this cached data * block. Note that indirect blocks and * private objects are not released until the * syncing state (since they are only modified * then). */ arc_release(db->db_buf, db); dbuf_fix_old_data(db, tx->tx_txg); data_old = db->db_buf; } ASSERT(data_old != NULL); } dr->dt.dl.dr_data = data_old; } else { mutex_init(&dr->dt.di.dr_mtx, NULL, MUTEX_NOLOCKDEP, NULL); list_create(&dr->dt.di.dr_children, sizeof (dbuf_dirty_record_t), offsetof(dbuf_dirty_record_t, dr_dirty_node)); } if (db->db_blkid != DMU_BONUS_BLKID) dr->dr_accounted = db->db.db_size; dr->dr_dbuf = db; dr->dr_txg = tx->tx_txg; list_insert_before(&db->db_dirty_records, dr_next, dr); /* * We could have been freed_in_flight between the dbuf_noread * and dbuf_dirty. We win, as though the dbuf_noread() had * happened after the free. */ if (db->db_level == 0 && db->db_blkid != DMU_BONUS_BLKID && db->db_blkid != DMU_SPILL_BLKID) { mutex_enter(&dn->dn_mtx); if (dn->dn_free_ranges[txgoff] != NULL) { range_tree_clear(dn->dn_free_ranges[txgoff], db->db_blkid, 1); } mutex_exit(&dn->dn_mtx); db->db_freed_in_flight = FALSE; } /* * This buffer is now part of this txg */ dbuf_add_ref(db, (void *)(uintptr_t)tx->tx_txg); db->db_dirtycnt += 1; ASSERT3U(db->db_dirtycnt, <=, 3); mutex_exit(&db->db_mtx); if (db->db_blkid == DMU_BONUS_BLKID || db->db_blkid == DMU_SPILL_BLKID) { mutex_enter(&dn->dn_mtx); ASSERT(!list_link_active(&dr->dr_dirty_node)); list_insert_tail(&dn->dn_dirty_records[txgoff], dr); mutex_exit(&dn->dn_mtx); dnode_setdirty(dn, tx); DB_DNODE_EXIT(db); return (dr); } if (!RW_WRITE_HELD(&dn->dn_struct_rwlock)) { rw_enter(&dn->dn_struct_rwlock, RW_READER); drop_struct_rwlock = B_TRUE; } /* * If we are overwriting a dedup BP, then unless it is snapshotted, * when we get to syncing context we will need to decrement its * refcount in the DDT. Prefetch the relevant DDT block so that * syncing context won't have to wait for the i/o. */ if (db->db_blkptr != NULL) { db_lock_type_t dblt = dmu_buf_lock_parent(db, RW_READER, FTAG); ddt_prefetch(os->os_spa, db->db_blkptr); dmu_buf_unlock_parent(db, dblt, FTAG); } /* * We need to hold the dn_struct_rwlock to make this assertion, * because it protects dn_phys / dn_next_nlevels from changing. */ ASSERT((dn->dn_phys->dn_nlevels == 0 && db->db_level == 0) || dn->dn_phys->dn_nlevels > db->db_level || dn->dn_next_nlevels[txgoff] > db->db_level || dn->dn_next_nlevels[(tx->tx_txg-1) & TXG_MASK] > db->db_level || dn->dn_next_nlevels[(tx->tx_txg-2) & TXG_MASK] > db->db_level); if (db->db_level == 0) { ASSERT(!db->db_objset->os_raw_receive || dn->dn_maxblkid >= db->db_blkid); dnode_new_blkid(dn, db->db_blkid, tx, drop_struct_rwlock, B_FALSE); ASSERT(dn->dn_maxblkid >= db->db_blkid); } if (db->db_level+1 < dn->dn_nlevels) { dmu_buf_impl_t *parent = db->db_parent; dbuf_dirty_record_t *di; int parent_held = FALSE; if (db->db_parent == NULL || db->db_parent == dn->dn_dbuf) { int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; parent = dbuf_hold_level(dn, db->db_level + 1, db->db_blkid >> epbs, FTAG); ASSERT(parent != NULL); parent_held = TRUE; } if (drop_struct_rwlock) rw_exit(&dn->dn_struct_rwlock); ASSERT3U(db->db_level + 1, ==, parent->db_level); di = dbuf_dirty(parent, tx); if (parent_held) dbuf_rele(parent, FTAG); mutex_enter(&db->db_mtx); /* * Since we've dropped the mutex, it's possible that * dbuf_undirty() might have changed this out from under us. */ if (list_head(&db->db_dirty_records) == dr || dn->dn_object == DMU_META_DNODE_OBJECT) { mutex_enter(&di->dt.di.dr_mtx); ASSERT3U(di->dr_txg, ==, tx->tx_txg); ASSERT(!list_link_active(&dr->dr_dirty_node)); list_insert_tail(&di->dt.di.dr_children, dr); mutex_exit(&di->dt.di.dr_mtx); dr->dr_parent = di; } mutex_exit(&db->db_mtx); } else { ASSERT(db->db_level + 1 == dn->dn_nlevels); ASSERT(db->db_blkid < dn->dn_nblkptr); ASSERT(db->db_parent == NULL || db->db_parent == dn->dn_dbuf); mutex_enter(&dn->dn_mtx); ASSERT(!list_link_active(&dr->dr_dirty_node)); list_insert_tail(&dn->dn_dirty_records[txgoff], dr); mutex_exit(&dn->dn_mtx); if (drop_struct_rwlock) rw_exit(&dn->dn_struct_rwlock); } dnode_setdirty(dn, tx); DB_DNODE_EXIT(db); return (dr); } static void dbuf_undirty_bonus(dbuf_dirty_record_t *dr) { dmu_buf_impl_t *db = dr->dr_dbuf; if (dr->dt.dl.dr_data != db->db.db_data) { struct dnode *dn = dr->dr_dnode; int max_bonuslen = DN_SLOTS_TO_BONUSLEN(dn->dn_num_slots); kmem_free(dr->dt.dl.dr_data, max_bonuslen); arc_space_return(max_bonuslen, ARC_SPACE_BONUS); } db->db_data_pending = NULL; ASSERT(list_next(&db->db_dirty_records, dr) == NULL); list_remove(&db->db_dirty_records, dr); if (dr->dr_dbuf->db_level != 0) { mutex_destroy(&dr->dt.di.dr_mtx); list_destroy(&dr->dt.di.dr_children); } kmem_free(dr, sizeof (dbuf_dirty_record_t)); ASSERT3U(db->db_dirtycnt, >, 0); db->db_dirtycnt -= 1; } /* * Undirty a buffer in the transaction group referenced by the given * transaction. Return whether this evicted the dbuf. */ static boolean_t dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx) { uint64_t txg = tx->tx_txg; ASSERT(txg != 0); /* * Due to our use of dn_nlevels below, this can only be called * in open context, unless we are operating on the MOS. * From syncing context, dn_nlevels may be different from the * dn_nlevels used when dbuf was dirtied. */ ASSERT(db->db_objset == dmu_objset_pool(db->db_objset)->dp_meta_objset || txg != spa_syncing_txg(dmu_objset_spa(db->db_objset))); ASSERT(db->db_blkid != DMU_BONUS_BLKID); ASSERT0(db->db_level); ASSERT(MUTEX_HELD(&db->db_mtx)); /* * If this buffer is not dirty, we're done. */ dbuf_dirty_record_t *dr = dbuf_find_dirty_eq(db, txg); if (dr == NULL) return (B_FALSE); ASSERT(dr->dr_dbuf == db); dnode_t *dn = dr->dr_dnode; dprintf_dbuf(db, "size=%llx\n", (u_longlong_t)db->db.db_size); ASSERT(db->db.db_size != 0); dsl_pool_undirty_space(dmu_objset_pool(dn->dn_objset), dr->dr_accounted, txg); list_remove(&db->db_dirty_records, dr); /* * Note that there are three places in dbuf_dirty() * where this dirty record may be put on a list. * Make sure to do a list_remove corresponding to * every one of those list_insert calls. */ if (dr->dr_parent) { mutex_enter(&dr->dr_parent->dt.di.dr_mtx); list_remove(&dr->dr_parent->dt.di.dr_children, dr); mutex_exit(&dr->dr_parent->dt.di.dr_mtx); } else if (db->db_blkid == DMU_SPILL_BLKID || db->db_level + 1 == dn->dn_nlevels) { ASSERT(db->db_blkptr == NULL || db->db_parent == dn->dn_dbuf); mutex_enter(&dn->dn_mtx); list_remove(&dn->dn_dirty_records[txg & TXG_MASK], dr); mutex_exit(&dn->dn_mtx); } if (db->db_state != DB_NOFILL) { dbuf_unoverride(dr); ASSERT(db->db_buf != NULL); ASSERT(dr->dt.dl.dr_data != NULL); if (dr->dt.dl.dr_data != db->db_buf) arc_buf_destroy(dr->dt.dl.dr_data, db); } kmem_free(dr, sizeof (dbuf_dirty_record_t)); ASSERT(db->db_dirtycnt > 0); db->db_dirtycnt -= 1; if (zfs_refcount_remove(&db->db_holds, (void *)(uintptr_t)txg) == 0) { ASSERT(db->db_state == DB_NOFILL || arc_released(db->db_buf)); dbuf_destroy(db); return (B_TRUE); } return (B_FALSE); } static void dmu_buf_will_dirty_impl(dmu_buf_t *db_fake, int flags, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; ASSERT(tx->tx_txg != 0); ASSERT(!zfs_refcount_is_zero(&db->db_holds)); /* * Quick check for dirtiness. For already dirty blocks, this * reduces runtime of this function by >90%, and overall performance * by 50% for some workloads (e.g. file deletion with indirect blocks * cached). */ mutex_enter(&db->db_mtx); if (db->db_state == DB_CACHED) { dbuf_dirty_record_t *dr = dbuf_find_dirty_eq(db, tx->tx_txg); /* * It's possible that it is already dirty but not cached, * because there are some calls to dbuf_dirty() that don't * go through dmu_buf_will_dirty(). */ if (dr != NULL) { /* This dbuf is already dirty and cached. */ dbuf_redirty(dr); mutex_exit(&db->db_mtx); return; } } mutex_exit(&db->db_mtx); DB_DNODE_ENTER(db); if (RW_WRITE_HELD(&DB_DNODE(db)->dn_struct_rwlock)) flags |= DB_RF_HAVESTRUCT; DB_DNODE_EXIT(db); (void) dbuf_read(db, NULL, flags); (void) dbuf_dirty(db, tx); } void dmu_buf_will_dirty(dmu_buf_t *db_fake, dmu_tx_t *tx) { dmu_buf_will_dirty_impl(db_fake, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH, tx); } boolean_t dmu_buf_is_dirty(dmu_buf_t *db_fake, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; dbuf_dirty_record_t *dr; mutex_enter(&db->db_mtx); dr = dbuf_find_dirty_eq(db, tx->tx_txg); mutex_exit(&db->db_mtx); return (dr != NULL); } void dmu_buf_will_not_fill(dmu_buf_t *db_fake, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; db->db_state = DB_NOFILL; DTRACE_SET_STATE(db, "allocating NOFILL buffer"); dmu_buf_will_fill(db_fake, tx); } void dmu_buf_will_fill(dmu_buf_t *db_fake, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; ASSERT(db->db_blkid != DMU_BONUS_BLKID); ASSERT(tx->tx_txg != 0); ASSERT(db->db_level == 0); ASSERT(!zfs_refcount_is_zero(&db->db_holds)); ASSERT(db->db.db_object != DMU_META_DNODE_OBJECT || dmu_tx_private_ok(tx)); dbuf_noread(db); (void) dbuf_dirty(db, tx); } /* * This function is effectively the same as dmu_buf_will_dirty(), but * indicates the caller expects raw encrypted data in the db, and provides * the crypt params (byteorder, salt, iv, mac) which should be stored in the * blkptr_t when this dbuf is written. This is only used for blocks of * dnodes, during raw receive. */ void dmu_buf_set_crypt_params(dmu_buf_t *db_fake, boolean_t byteorder, const uint8_t *salt, const uint8_t *iv, const uint8_t *mac, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; dbuf_dirty_record_t *dr; /* * dr_has_raw_params is only processed for blocks of dnodes * (see dbuf_sync_dnode_leaf_crypt()). */ ASSERT3U(db->db.db_object, ==, DMU_META_DNODE_OBJECT); ASSERT3U(db->db_level, ==, 0); ASSERT(db->db_objset->os_raw_receive); dmu_buf_will_dirty_impl(db_fake, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH | DB_RF_NO_DECRYPT, tx); dr = dbuf_find_dirty_eq(db, tx->tx_txg); ASSERT3P(dr, !=, NULL); dr->dt.dl.dr_has_raw_params = B_TRUE; dr->dt.dl.dr_byteorder = byteorder; memcpy(dr->dt.dl.dr_salt, salt, ZIO_DATA_SALT_LEN); memcpy(dr->dt.dl.dr_iv, iv, ZIO_DATA_IV_LEN); memcpy(dr->dt.dl.dr_mac, mac, ZIO_DATA_MAC_LEN); } static void dbuf_override_impl(dmu_buf_impl_t *db, const blkptr_t *bp, dmu_tx_t *tx) { struct dirty_leaf *dl; dbuf_dirty_record_t *dr; dr = list_head(&db->db_dirty_records); ASSERT3U(dr->dr_txg, ==, tx->tx_txg); dl = &dr->dt.dl; dl->dr_overridden_by = *bp; dl->dr_override_state = DR_OVERRIDDEN; dl->dr_overridden_by.blk_birth = dr->dr_txg; } void dmu_buf_fill_done(dmu_buf_t *dbuf, dmu_tx_t *tx) { (void) tx; dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbuf; dbuf_states_t old_state; mutex_enter(&db->db_mtx); DBUF_VERIFY(db); old_state = db->db_state; db->db_state = DB_CACHED; if (old_state == DB_FILL) { if (db->db_level == 0 && db->db_freed_in_flight) { ASSERT(db->db_blkid != DMU_BONUS_BLKID); /* we were freed while filling */ /* XXX dbuf_undirty? */ memset(db->db.db_data, 0, db->db.db_size); db->db_freed_in_flight = FALSE; DTRACE_SET_STATE(db, "fill done handling freed in flight"); } else { DTRACE_SET_STATE(db, "fill done"); } cv_broadcast(&db->db_changed); } mutex_exit(&db->db_mtx); } void dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data, bp_embedded_type_t etype, enum zio_compress comp, int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbuf; struct dirty_leaf *dl; dmu_object_type_t type; dbuf_dirty_record_t *dr; if (etype == BP_EMBEDDED_TYPE_DATA) { ASSERT(spa_feature_is_active(dmu_objset_spa(db->db_objset), SPA_FEATURE_EMBEDDED_DATA)); } DB_DNODE_ENTER(db); type = DB_DNODE(db)->dn_type; DB_DNODE_EXIT(db); ASSERT0(db->db_level); ASSERT(db->db_blkid != DMU_BONUS_BLKID); dmu_buf_will_not_fill(dbuf, tx); dr = list_head(&db->db_dirty_records); ASSERT3U(dr->dr_txg, ==, tx->tx_txg); dl = &dr->dt.dl; encode_embedded_bp_compressed(&dl->dr_overridden_by, data, comp, uncompressed_size, compressed_size); BPE_SET_ETYPE(&dl->dr_overridden_by, etype); BP_SET_TYPE(&dl->dr_overridden_by, type); BP_SET_LEVEL(&dl->dr_overridden_by, 0); BP_SET_BYTEORDER(&dl->dr_overridden_by, byteorder); dl->dr_override_state = DR_OVERRIDDEN; dl->dr_overridden_by.blk_birth = dr->dr_txg; } void dmu_buf_redact(dmu_buf_t *dbuf, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbuf; dmu_object_type_t type; ASSERT(dsl_dataset_feature_is_active(db->db_objset->os_dsl_dataset, SPA_FEATURE_REDACTED_DATASETS)); DB_DNODE_ENTER(db); type = DB_DNODE(db)->dn_type; DB_DNODE_EXIT(db); ASSERT0(db->db_level); dmu_buf_will_not_fill(dbuf, tx); blkptr_t bp = { { { {0} } } }; BP_SET_TYPE(&bp, type); BP_SET_LEVEL(&bp, 0); BP_SET_BIRTH(&bp, tx->tx_txg, 0); BP_SET_REDACTED(&bp); BPE_SET_LSIZE(&bp, dbuf->db_size); dbuf_override_impl(db, &bp, tx); } /* * Directly assign a provided arc buf to a given dbuf if it's not referenced * by anybody except our caller. Otherwise copy arcbuf's contents to dbuf. */ void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx) { ASSERT(!zfs_refcount_is_zero(&db->db_holds)); ASSERT(db->db_blkid != DMU_BONUS_BLKID); ASSERT(db->db_level == 0); ASSERT3U(dbuf_is_metadata(db), ==, arc_is_metadata(buf)); ASSERT(buf != NULL); ASSERT3U(arc_buf_lsize(buf), ==, db->db.db_size); ASSERT(tx->tx_txg != 0); arc_return_buf(buf, db); ASSERT(arc_released(buf)); mutex_enter(&db->db_mtx); while (db->db_state == DB_READ || db->db_state == DB_FILL) cv_wait(&db->db_changed, &db->db_mtx); ASSERT(db->db_state == DB_CACHED || db->db_state == DB_UNCACHED); if (db->db_state == DB_CACHED && zfs_refcount_count(&db->db_holds) - 1 > db->db_dirtycnt) { /* * In practice, we will never have a case where we have an * encrypted arc buffer while additional holds exist on the * dbuf. We don't handle this here so we simply assert that * fact instead. */ ASSERT(!arc_is_encrypted(buf)); mutex_exit(&db->db_mtx); (void) dbuf_dirty(db, tx); memcpy(db->db.db_data, buf->b_data, db->db.db_size); arc_buf_destroy(buf, db); return; } if (db->db_state == DB_CACHED) { dbuf_dirty_record_t *dr = list_head(&db->db_dirty_records); ASSERT(db->db_buf != NULL); if (dr != NULL && dr->dr_txg == tx->tx_txg) { ASSERT(dr->dt.dl.dr_data == db->db_buf); if (!arc_released(db->db_buf)) { ASSERT(dr->dt.dl.dr_override_state == DR_OVERRIDDEN); arc_release(db->db_buf, db); } dr->dt.dl.dr_data = buf; arc_buf_destroy(db->db_buf, db); } else if (dr == NULL || dr->dt.dl.dr_data != db->db_buf) { arc_release(db->db_buf, db); arc_buf_destroy(db->db_buf, db); } db->db_buf = NULL; } ASSERT(db->db_buf == NULL); dbuf_set_data(db, buf); db->db_state = DB_FILL; DTRACE_SET_STATE(db, "filling assigned arcbuf"); mutex_exit(&db->db_mtx); (void) dbuf_dirty(db, tx); dmu_buf_fill_done(&db->db, tx); } void dbuf_destroy(dmu_buf_impl_t *db) { dnode_t *dn; dmu_buf_impl_t *parent = db->db_parent; dmu_buf_impl_t *dndb; ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(zfs_refcount_is_zero(&db->db_holds)); if (db->db_buf != NULL) { arc_buf_destroy(db->db_buf, db); db->db_buf = NULL; } if (db->db_blkid == DMU_BONUS_BLKID) { int slots = DB_DNODE(db)->dn_num_slots; int bonuslen = DN_SLOTS_TO_BONUSLEN(slots); if (db->db.db_data != NULL) { kmem_free(db->db.db_data, bonuslen); arc_space_return(bonuslen, ARC_SPACE_BONUS); db->db_state = DB_UNCACHED; DTRACE_SET_STATE(db, "buffer cleared"); } } dbuf_clear_data(db); if (multilist_link_active(&db->db_cache_link)) { ASSERT(db->db_caching_status == DB_DBUF_CACHE || db->db_caching_status == DB_DBUF_METADATA_CACHE); multilist_remove(&dbuf_caches[db->db_caching_status].cache, db); (void) zfs_refcount_remove_many( &dbuf_caches[db->db_caching_status].size, db->db.db_size, db); if (db->db_caching_status == DB_DBUF_METADATA_CACHE) { DBUF_STAT_BUMPDOWN(metadata_cache_count); } else { DBUF_STAT_BUMPDOWN(cache_levels[db->db_level]); DBUF_STAT_BUMPDOWN(cache_count); DBUF_STAT_DECR(cache_levels_bytes[db->db_level], db->db.db_size); } db->db_caching_status = DB_NO_CACHE; } ASSERT(db->db_state == DB_UNCACHED || db->db_state == DB_NOFILL); ASSERT(db->db_data_pending == NULL); ASSERT(list_is_empty(&db->db_dirty_records)); db->db_state = DB_EVICTING; DTRACE_SET_STATE(db, "buffer eviction started"); db->db_blkptr = NULL; /* * Now that db_state is DB_EVICTING, nobody else can find this via * the hash table. We can now drop db_mtx, which allows us to * acquire the dn_dbufs_mtx. */ mutex_exit(&db->db_mtx); DB_DNODE_ENTER(db); dn = DB_DNODE(db); dndb = dn->dn_dbuf; if (db->db_blkid != DMU_BONUS_BLKID) { boolean_t needlock = !MUTEX_HELD(&dn->dn_dbufs_mtx); if (needlock) mutex_enter_nested(&dn->dn_dbufs_mtx, NESTED_SINGLE); avl_remove(&dn->dn_dbufs, db); membar_producer(); DB_DNODE_EXIT(db); if (needlock) mutex_exit(&dn->dn_dbufs_mtx); /* * Decrementing the dbuf count means that the hold corresponding * to the removed dbuf is no longer discounted in dnode_move(), * so the dnode cannot be moved until after we release the hold. * The membar_producer() ensures visibility of the decremented * value in dnode_move(), since DB_DNODE_EXIT doesn't actually * release any lock. */ mutex_enter(&dn->dn_mtx); dnode_rele_and_unlock(dn, db, B_TRUE); db->db_dnode_handle = NULL; dbuf_hash_remove(db); } else { DB_DNODE_EXIT(db); } ASSERT(zfs_refcount_is_zero(&db->db_holds)); db->db_parent = NULL; ASSERT(db->db_buf == NULL); ASSERT(db->db.db_data == NULL); ASSERT(db->db_hash_next == NULL); ASSERT(db->db_blkptr == NULL); ASSERT(db->db_data_pending == NULL); ASSERT3U(db->db_caching_status, ==, DB_NO_CACHE); ASSERT(!multilist_link_active(&db->db_cache_link)); /* * If this dbuf is referenced from an indirect dbuf, * decrement the ref count on the indirect dbuf. */ if (parent && parent != dndb) { mutex_enter(&parent->db_mtx); dbuf_rele_and_unlock(parent, db, B_TRUE); } kmem_cache_free(dbuf_kmem_cache, db); arc_space_return(sizeof (dmu_buf_impl_t), ARC_SPACE_DBUF); } /* * Note: While bpp will always be updated if the function returns success, * parentp will not be updated if the dnode does not have dn_dbuf filled in; * this happens when the dnode is the meta-dnode, or {user|group|project}used * object. */ __attribute__((always_inline)) static inline int dbuf_findbp(dnode_t *dn, int level, uint64_t blkid, int fail_sparse, dmu_buf_impl_t **parentp, blkptr_t **bpp) { *parentp = NULL; *bpp = NULL; ASSERT(blkid != DMU_BONUS_BLKID); if (blkid == DMU_SPILL_BLKID) { mutex_enter(&dn->dn_mtx); if (dn->dn_have_spill && (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR)) *bpp = DN_SPILL_BLKPTR(dn->dn_phys); else *bpp = NULL; dbuf_add_ref(dn->dn_dbuf, NULL); *parentp = dn->dn_dbuf; mutex_exit(&dn->dn_mtx); return (0); } int nlevels = (dn->dn_phys->dn_nlevels == 0) ? 1 : dn->dn_phys->dn_nlevels; int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; ASSERT3U(level * epbs, <, 64); ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock)); /* * This assertion shouldn't trip as long as the max indirect block size * is less than 1M. The reason for this is that up to that point, * the number of levels required to address an entire object with blocks * of size SPA_MINBLOCKSIZE satisfies nlevels * epbs + 1 <= 64. In * other words, if N * epbs + 1 > 64, then if (N-1) * epbs + 1 > 55 * (i.e. we can address the entire object), objects will all use at most * N-1 levels and the assertion won't overflow. However, once epbs is * 13, 4 * 13 + 1 = 53, but 5 * 13 + 1 = 66. Then, 4 levels will not be * enough to address an entire object, so objects will have 5 levels, * but then this assertion will overflow. * * All this is to say that if we ever increase DN_MAX_INDBLKSHIFT, we * need to redo this logic to handle overflows. */ ASSERT(level >= nlevels || ((nlevels - level - 1) * epbs) + highbit64(dn->dn_phys->dn_nblkptr) <= 64); if (level >= nlevels || blkid >= ((uint64_t)dn->dn_phys->dn_nblkptr << ((nlevels - level - 1) * epbs)) || (fail_sparse && blkid > (dn->dn_phys->dn_maxblkid >> (level * epbs)))) { /* the buffer has no parent yet */ return (SET_ERROR(ENOENT)); } else if (level < nlevels-1) { /* this block is referenced from an indirect block */ int err; err = dbuf_hold_impl(dn, level + 1, blkid >> epbs, fail_sparse, FALSE, NULL, parentp); if (err) return (err); err = dbuf_read(*parentp, NULL, (DB_RF_HAVESTRUCT | DB_RF_NOPREFETCH | DB_RF_CANFAIL)); if (err) { dbuf_rele(*parentp, NULL); *parentp = NULL; return (err); } rw_enter(&(*parentp)->db_rwlock, RW_READER); *bpp = ((blkptr_t *)(*parentp)->db.db_data) + (blkid & ((1ULL << epbs) - 1)); if (blkid > (dn->dn_phys->dn_maxblkid >> (level * epbs))) ASSERT(BP_IS_HOLE(*bpp)); rw_exit(&(*parentp)->db_rwlock); return (0); } else { /* the block is referenced from the dnode */ ASSERT3U(level, ==, nlevels-1); ASSERT(dn->dn_phys->dn_nblkptr == 0 || blkid < dn->dn_phys->dn_nblkptr); if (dn->dn_dbuf) { dbuf_add_ref(dn->dn_dbuf, NULL); *parentp = dn->dn_dbuf; } *bpp = &dn->dn_phys->dn_blkptr[blkid]; return (0); } } static dmu_buf_impl_t * dbuf_create(dnode_t *dn, uint8_t level, uint64_t blkid, dmu_buf_impl_t *parent, blkptr_t *blkptr) { objset_t *os = dn->dn_objset; dmu_buf_impl_t *db, *odb; ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock)); ASSERT(dn->dn_type != DMU_OT_NONE); db = kmem_cache_alloc(dbuf_kmem_cache, KM_SLEEP); list_create(&db->db_dirty_records, sizeof (dbuf_dirty_record_t), offsetof(dbuf_dirty_record_t, dr_dbuf_node)); db->db_objset = os; db->db.db_object = dn->dn_object; db->db_level = level; db->db_blkid = blkid; db->db_dirtycnt = 0; db->db_dnode_handle = dn->dn_handle; db->db_parent = parent; db->db_blkptr = blkptr; db->db_user = NULL; db->db_user_immediate_evict = FALSE; db->db_freed_in_flight = FALSE; db->db_pending_evict = FALSE; if (blkid == DMU_BONUS_BLKID) { ASSERT3P(parent, ==, dn->dn_dbuf); db->db.db_size = DN_SLOTS_TO_BONUSLEN(dn->dn_num_slots) - (dn->dn_nblkptr-1) * sizeof (blkptr_t); ASSERT3U(db->db.db_size, >=, dn->dn_bonuslen); db->db.db_offset = DMU_BONUS_BLKID; db->db_state = DB_UNCACHED; DTRACE_SET_STATE(db, "bonus buffer created"); db->db_caching_status = DB_NO_CACHE; /* the bonus dbuf is not placed in the hash table */ arc_space_consume(sizeof (dmu_buf_impl_t), ARC_SPACE_DBUF); return (db); } else if (blkid == DMU_SPILL_BLKID) { db->db.db_size = (blkptr != NULL) ? BP_GET_LSIZE(blkptr) : SPA_MINBLOCKSIZE; db->db.db_offset = 0; } else { int blocksize = db->db_level ? 1 << dn->dn_indblkshift : dn->dn_datablksz; db->db.db_size = blocksize; db->db.db_offset = db->db_blkid * blocksize; } /* * Hold the dn_dbufs_mtx while we get the new dbuf * in the hash table *and* added to the dbufs list. * This prevents a possible deadlock with someone * trying to look up this dbuf before it's added to the * dn_dbufs list. */ mutex_enter(&dn->dn_dbufs_mtx); db->db_state = DB_EVICTING; /* not worth logging this state change */ if ((odb = dbuf_hash_insert(db)) != NULL) { /* someone else inserted it first */ mutex_exit(&dn->dn_dbufs_mtx); kmem_cache_free(dbuf_kmem_cache, db); DBUF_STAT_BUMP(hash_insert_race); return (odb); } avl_add(&dn->dn_dbufs, db); db->db_state = DB_UNCACHED; DTRACE_SET_STATE(db, "regular buffer created"); db->db_caching_status = DB_NO_CACHE; mutex_exit(&dn->dn_dbufs_mtx); arc_space_consume(sizeof (dmu_buf_impl_t), ARC_SPACE_DBUF); if (parent && parent != dn->dn_dbuf) dbuf_add_ref(parent, db); ASSERT(dn->dn_object == DMU_META_DNODE_OBJECT || zfs_refcount_count(&dn->dn_holds) > 0); (void) zfs_refcount_add(&dn->dn_holds, db); dprintf_dbuf(db, "db=%p\n", db); return (db); } /* * This function returns a block pointer and information about the object, * given a dnode and a block. This is a publicly accessible version of * dbuf_findbp that only returns some information, rather than the * dbuf. Note that the dnode passed in must be held, and the dn_struct_rwlock * should be locked as (at least) a reader. */ int dbuf_dnode_findbp(dnode_t *dn, uint64_t level, uint64_t blkid, blkptr_t *bp, uint16_t *datablkszsec, uint8_t *indblkshift) { dmu_buf_impl_t *dbp = NULL; blkptr_t *bp2; int err = 0; ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock)); err = dbuf_findbp(dn, level, blkid, B_FALSE, &dbp, &bp2); if (err == 0) { *bp = *bp2; if (dbp != NULL) dbuf_rele(dbp, NULL); if (datablkszsec != NULL) *datablkszsec = dn->dn_phys->dn_datablkszsec; if (indblkshift != NULL) *indblkshift = dn->dn_phys->dn_indblkshift; } return (err); } typedef struct dbuf_prefetch_arg { spa_t *dpa_spa; /* The spa to issue the prefetch in. */ zbookmark_phys_t dpa_zb; /* The target block to prefetch. */ int dpa_epbs; /* Entries (blkptr_t's) Per Block Shift. */ int dpa_curlevel; /* The current level that we're reading */ dnode_t *dpa_dnode; /* The dnode associated with the prefetch */ zio_priority_t dpa_prio; /* The priority I/Os should be issued at. */ zio_t *dpa_zio; /* The parent zio_t for all prefetches. */ arc_flags_t dpa_aflags; /* Flags to pass to the final prefetch. */ dbuf_prefetch_fn dpa_cb; /* prefetch completion callback */ void *dpa_arg; /* prefetch completion arg */ } dbuf_prefetch_arg_t; static void dbuf_prefetch_fini(dbuf_prefetch_arg_t *dpa, boolean_t io_done) { if (dpa->dpa_cb != NULL) { dpa->dpa_cb(dpa->dpa_arg, dpa->dpa_zb.zb_level, dpa->dpa_zb.zb_blkid, io_done); } kmem_free(dpa, sizeof (*dpa)); } static void dbuf_issue_final_prefetch_done(zio_t *zio, const zbookmark_phys_t *zb, const blkptr_t *iobp, arc_buf_t *abuf, void *private) { (void) zio, (void) zb, (void) iobp; dbuf_prefetch_arg_t *dpa = private; if (abuf != NULL) arc_buf_destroy(abuf, private); dbuf_prefetch_fini(dpa, B_TRUE); } /* * Actually issue the prefetch read for the block given. */ static void dbuf_issue_final_prefetch(dbuf_prefetch_arg_t *dpa, blkptr_t *bp) { ASSERT(!BP_IS_REDACTED(bp) || dsl_dataset_feature_is_active( dpa->dpa_dnode->dn_objset->os_dsl_dataset, SPA_FEATURE_REDACTED_DATASETS)); if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp) || BP_IS_REDACTED(bp)) return (dbuf_prefetch_fini(dpa, B_FALSE)); int zio_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE; arc_flags_t aflags = dpa->dpa_aflags | ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH | ARC_FLAG_NO_BUF; /* dnodes are always read as raw and then converted later */ if (BP_GET_TYPE(bp) == DMU_OT_DNODE && BP_IS_PROTECTED(bp) && dpa->dpa_curlevel == 0) zio_flags |= ZIO_FLAG_RAW; ASSERT3U(dpa->dpa_curlevel, ==, BP_GET_LEVEL(bp)); ASSERT3U(dpa->dpa_curlevel, ==, dpa->dpa_zb.zb_level); ASSERT(dpa->dpa_zio != NULL); (void) arc_read(dpa->dpa_zio, dpa->dpa_spa, bp, dbuf_issue_final_prefetch_done, dpa, dpa->dpa_prio, zio_flags, &aflags, &dpa->dpa_zb); } /* * Called when an indirect block above our prefetch target is read in. This * will either read in the next indirect block down the tree or issue the actual * prefetch if the next block down is our target. */ static void dbuf_prefetch_indirect_done(zio_t *zio, const zbookmark_phys_t *zb, const blkptr_t *iobp, arc_buf_t *abuf, void *private) { (void) zb, (void) iobp; dbuf_prefetch_arg_t *dpa = private; ASSERT3S(dpa->dpa_zb.zb_level, <, dpa->dpa_curlevel); ASSERT3S(dpa->dpa_curlevel, >, 0); if (abuf == NULL) { ASSERT(zio == NULL || zio->io_error != 0); dbuf_prefetch_fini(dpa, B_TRUE); return; } ASSERT(zio == NULL || zio->io_error == 0); /* * The dpa_dnode is only valid if we are called with a NULL * zio. This indicates that the arc_read() returned without * first calling zio_read() to issue a physical read. Once * a physical read is made the dpa_dnode must be invalidated * as the locks guarding it may have been dropped. If the * dpa_dnode is still valid, then we want to add it to the dbuf * cache. To do so, we must hold the dbuf associated with the block * we just prefetched, read its contents so that we associate it * with an arc_buf_t, and then release it. */ if (zio != NULL) { ASSERT3S(BP_GET_LEVEL(zio->io_bp), ==, dpa->dpa_curlevel); if (zio->io_flags & ZIO_FLAG_RAW_COMPRESS) { ASSERT3U(BP_GET_PSIZE(zio->io_bp), ==, zio->io_size); } else { ASSERT3U(BP_GET_LSIZE(zio->io_bp), ==, zio->io_size); } ASSERT3P(zio->io_spa, ==, dpa->dpa_spa); dpa->dpa_dnode = NULL; } else if (dpa->dpa_dnode != NULL) { uint64_t curblkid = dpa->dpa_zb.zb_blkid >> (dpa->dpa_epbs * (dpa->dpa_curlevel - dpa->dpa_zb.zb_level)); dmu_buf_impl_t *db = dbuf_hold_level(dpa->dpa_dnode, dpa->dpa_curlevel, curblkid, FTAG); if (db == NULL) { arc_buf_destroy(abuf, private); dbuf_prefetch_fini(dpa, B_TRUE); return; } (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH | DB_RF_HAVESTRUCT); dbuf_rele(db, FTAG); } dpa->dpa_curlevel--; uint64_t nextblkid = dpa->dpa_zb.zb_blkid >> (dpa->dpa_epbs * (dpa->dpa_curlevel - dpa->dpa_zb.zb_level)); blkptr_t *bp = ((blkptr_t *)abuf->b_data) + P2PHASE(nextblkid, 1ULL << dpa->dpa_epbs); ASSERT(!BP_IS_REDACTED(bp) || dsl_dataset_feature_is_active( dpa->dpa_dnode->dn_objset->os_dsl_dataset, SPA_FEATURE_REDACTED_DATASETS)); if (BP_IS_HOLE(bp) || BP_IS_REDACTED(bp)) { arc_buf_destroy(abuf, private); dbuf_prefetch_fini(dpa, B_TRUE); return; } else if (dpa->dpa_curlevel == dpa->dpa_zb.zb_level) { ASSERT3U(nextblkid, ==, dpa->dpa_zb.zb_blkid); dbuf_issue_final_prefetch(dpa, bp); } else { arc_flags_t iter_aflags = ARC_FLAG_NOWAIT; zbookmark_phys_t zb; /* flag if L2ARC eligible, l2arc_noprefetch then decides */ if (dpa->dpa_aflags & ARC_FLAG_L2CACHE) iter_aflags |= ARC_FLAG_L2CACHE; ASSERT3U(dpa->dpa_curlevel, ==, BP_GET_LEVEL(bp)); SET_BOOKMARK(&zb, dpa->dpa_zb.zb_objset, dpa->dpa_zb.zb_object, dpa->dpa_curlevel, nextblkid); (void) arc_read(dpa->dpa_zio, dpa->dpa_spa, bp, dbuf_prefetch_indirect_done, dpa, ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, &iter_aflags, &zb); } arc_buf_destroy(abuf, private); } /* * Issue prefetch reads for the given block on the given level. If the indirect * blocks above that block are not in memory, we will read them in * asynchronously. As a result, this call never blocks waiting for a read to * complete. Note that the prefetch might fail if the dataset is encrypted and * the encryption key is unmapped before the IO completes. */ int dbuf_prefetch_impl(dnode_t *dn, int64_t level, uint64_t blkid, zio_priority_t prio, arc_flags_t aflags, dbuf_prefetch_fn cb, void *arg) { blkptr_t bp; int epbs, nlevels, curlevel; uint64_t curblkid; ASSERT(blkid != DMU_BONUS_BLKID); ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock)); if (blkid > dn->dn_maxblkid) goto no_issue; if (level == 0 && dnode_block_freed(dn, blkid)) goto no_issue; /* * This dnode hasn't been written to disk yet, so there's nothing to * prefetch. */ nlevels = dn->dn_phys->dn_nlevels; if (level >= nlevels || dn->dn_phys->dn_nblkptr == 0) goto no_issue; epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; if (dn->dn_phys->dn_maxblkid < blkid << (epbs * level)) goto no_issue; dmu_buf_impl_t *db = dbuf_find(dn->dn_objset, dn->dn_object, level, blkid); if (db != NULL) { mutex_exit(&db->db_mtx); /* * This dbuf already exists. It is either CACHED, or * (we assume) about to be read or filled. */ goto no_issue; } /* * Find the closest ancestor (indirect block) of the target block * that is present in the cache. In this indirect block, we will * find the bp that is at curlevel, curblkid. */ curlevel = level; curblkid = blkid; while (curlevel < nlevels - 1) { int parent_level = curlevel + 1; uint64_t parent_blkid = curblkid >> epbs; dmu_buf_impl_t *db; if (dbuf_hold_impl(dn, parent_level, parent_blkid, FALSE, TRUE, FTAG, &db) == 0) { blkptr_t *bpp = db->db_buf->b_data; bp = bpp[P2PHASE(curblkid, 1 << epbs)]; dbuf_rele(db, FTAG); break; } curlevel = parent_level; curblkid = parent_blkid; } if (curlevel == nlevels - 1) { /* No cached indirect blocks found. */ ASSERT3U(curblkid, <, dn->dn_phys->dn_nblkptr); bp = dn->dn_phys->dn_blkptr[curblkid]; } ASSERT(!BP_IS_REDACTED(&bp) || dsl_dataset_feature_is_active(dn->dn_objset->os_dsl_dataset, SPA_FEATURE_REDACTED_DATASETS)); if (BP_IS_HOLE(&bp) || BP_IS_REDACTED(&bp)) goto no_issue; ASSERT3U(curlevel, ==, BP_GET_LEVEL(&bp)); zio_t *pio = zio_root(dmu_objset_spa(dn->dn_objset), NULL, NULL, ZIO_FLAG_CANFAIL); dbuf_prefetch_arg_t *dpa = kmem_zalloc(sizeof (*dpa), KM_SLEEP); dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset; SET_BOOKMARK(&dpa->dpa_zb, ds != NULL ? ds->ds_object : DMU_META_OBJSET, dn->dn_object, level, blkid); dpa->dpa_curlevel = curlevel; dpa->dpa_prio = prio; dpa->dpa_aflags = aflags; dpa->dpa_spa = dn->dn_objset->os_spa; dpa->dpa_dnode = dn; dpa->dpa_epbs = epbs; dpa->dpa_zio = pio; dpa->dpa_cb = cb; dpa->dpa_arg = arg; /* flag if L2ARC eligible, l2arc_noprefetch then decides */ if (dnode_level_is_l2cacheable(&bp, dn, level)) dpa->dpa_aflags |= ARC_FLAG_L2CACHE; /* * If we have the indirect just above us, no need to do the asynchronous * prefetch chain; we'll just run the last step ourselves. If we're at * a higher level, though, we want to issue the prefetches for all the * indirect blocks asynchronously, so we can go on with whatever we were * doing. */ if (curlevel == level) { ASSERT3U(curblkid, ==, blkid); dbuf_issue_final_prefetch(dpa, &bp); } else { arc_flags_t iter_aflags = ARC_FLAG_NOWAIT; zbookmark_phys_t zb; /* flag if L2ARC eligible, l2arc_noprefetch then decides */ if (dnode_level_is_l2cacheable(&bp, dn, level)) iter_aflags |= ARC_FLAG_L2CACHE; SET_BOOKMARK(&zb, ds != NULL ? ds->ds_object : DMU_META_OBJSET, dn->dn_object, curlevel, curblkid); (void) arc_read(dpa->dpa_zio, dpa->dpa_spa, &bp, dbuf_prefetch_indirect_done, dpa, ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, &iter_aflags, &zb); } /* * We use pio here instead of dpa_zio since it's possible that * dpa may have already been freed. */ zio_nowait(pio); return (1); no_issue: if (cb != NULL) cb(arg, level, blkid, B_FALSE); return (0); } int dbuf_prefetch(dnode_t *dn, int64_t level, uint64_t blkid, zio_priority_t prio, arc_flags_t aflags) { return (dbuf_prefetch_impl(dn, level, blkid, prio, aflags, NULL, NULL)); } /* * Helper function for dbuf_hold_impl() to copy a buffer. Handles * the case of encrypted, compressed and uncompressed buffers by * allocating the new buffer, respectively, with arc_alloc_raw_buf(), * arc_alloc_compressed_buf() or arc_alloc_buf().* * * NOTE: Declared noinline to avoid stack bloat in dbuf_hold_impl(). */ noinline static void dbuf_hold_copy(dnode_t *dn, dmu_buf_impl_t *db) { dbuf_dirty_record_t *dr = db->db_data_pending; arc_buf_t *data = dr->dt.dl.dr_data; enum zio_compress compress_type = arc_get_compression(data); uint8_t complevel = arc_get_complevel(data); if (arc_is_encrypted(data)) { boolean_t byteorder; uint8_t salt[ZIO_DATA_SALT_LEN]; uint8_t iv[ZIO_DATA_IV_LEN]; uint8_t mac[ZIO_DATA_MAC_LEN]; arc_get_raw_params(data, &byteorder, salt, iv, mac); dbuf_set_data(db, arc_alloc_raw_buf(dn->dn_objset->os_spa, db, dmu_objset_id(dn->dn_objset), byteorder, salt, iv, mac, dn->dn_type, arc_buf_size(data), arc_buf_lsize(data), compress_type, complevel)); } else if (compress_type != ZIO_COMPRESS_OFF) { dbuf_set_data(db, arc_alloc_compressed_buf( dn->dn_objset->os_spa, db, arc_buf_size(data), arc_buf_lsize(data), compress_type, complevel)); } else { dbuf_set_data(db, arc_alloc_buf(dn->dn_objset->os_spa, db, DBUF_GET_BUFC_TYPE(db), db->db.db_size)); } rw_enter(&db->db_rwlock, RW_WRITER); memcpy(db->db.db_data, data->b_data, arc_buf_size(data)); rw_exit(&db->db_rwlock); } /* * Returns with db_holds incremented, and db_mtx not held. * Note: dn_struct_rwlock must be held. */ int dbuf_hold_impl(dnode_t *dn, uint8_t level, uint64_t blkid, boolean_t fail_sparse, boolean_t fail_uncached, const void *tag, dmu_buf_impl_t **dbp) { dmu_buf_impl_t *db, *parent = NULL; /* If the pool has been created, verify the tx_sync_lock is not held */ spa_t *spa = dn->dn_objset->os_spa; dsl_pool_t *dp = spa->spa_dsl_pool; if (dp != NULL) { ASSERT(!MUTEX_HELD(&dp->dp_tx.tx_sync_lock)); } ASSERT(blkid != DMU_BONUS_BLKID); ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock)); ASSERT3U(dn->dn_nlevels, >, level); *dbp = NULL; /* dbuf_find() returns with db_mtx held */ db = dbuf_find(dn->dn_objset, dn->dn_object, level, blkid); if (db == NULL) { blkptr_t *bp = NULL; int err; if (fail_uncached) return (SET_ERROR(ENOENT)); ASSERT3P(parent, ==, NULL); err = dbuf_findbp(dn, level, blkid, fail_sparse, &parent, &bp); if (fail_sparse) { if (err == 0 && bp && BP_IS_HOLE(bp)) err = SET_ERROR(ENOENT); if (err) { if (parent) dbuf_rele(parent, NULL); return (err); } } if (err && err != ENOENT) return (err); db = dbuf_create(dn, level, blkid, parent, bp); } if (fail_uncached && db->db_state != DB_CACHED) { mutex_exit(&db->db_mtx); return (SET_ERROR(ENOENT)); } if (db->db_buf != NULL) { arc_buf_access(db->db_buf); ASSERT3P(db->db.db_data, ==, db->db_buf->b_data); } ASSERT(db->db_buf == NULL || arc_referenced(db->db_buf)); /* * If this buffer is currently syncing out, and we are * still referencing it from db_data, we need to make a copy * of it in case we decide we want to dirty it again in this txg. */ if (db->db_level == 0 && db->db_blkid != DMU_BONUS_BLKID && dn->dn_object != DMU_META_DNODE_OBJECT && db->db_state == DB_CACHED && db->db_data_pending) { dbuf_dirty_record_t *dr = db->db_data_pending; if (dr->dt.dl.dr_data == db->db_buf) dbuf_hold_copy(dn, db); } if (multilist_link_active(&db->db_cache_link)) { ASSERT(zfs_refcount_is_zero(&db->db_holds)); ASSERT(db->db_caching_status == DB_DBUF_CACHE || db->db_caching_status == DB_DBUF_METADATA_CACHE); multilist_remove(&dbuf_caches[db->db_caching_status].cache, db); (void) zfs_refcount_remove_many( &dbuf_caches[db->db_caching_status].size, db->db.db_size, db); if (db->db_caching_status == DB_DBUF_METADATA_CACHE) { DBUF_STAT_BUMPDOWN(metadata_cache_count); } else { DBUF_STAT_BUMPDOWN(cache_levels[db->db_level]); DBUF_STAT_BUMPDOWN(cache_count); DBUF_STAT_DECR(cache_levels_bytes[db->db_level], db->db.db_size); } db->db_caching_status = DB_NO_CACHE; } (void) zfs_refcount_add(&db->db_holds, tag); DBUF_VERIFY(db); mutex_exit(&db->db_mtx); /* NOTE: we can't rele the parent until after we drop the db_mtx */ if (parent) dbuf_rele(parent, NULL); ASSERT3P(DB_DNODE(db), ==, dn); ASSERT3U(db->db_blkid, ==, blkid); ASSERT3U(db->db_level, ==, level); *dbp = db; return (0); } dmu_buf_impl_t * dbuf_hold(dnode_t *dn, uint64_t blkid, const void *tag) { return (dbuf_hold_level(dn, 0, blkid, tag)); } dmu_buf_impl_t * dbuf_hold_level(dnode_t *dn, int level, uint64_t blkid, const void *tag) { dmu_buf_impl_t *db; int err = dbuf_hold_impl(dn, level, blkid, FALSE, FALSE, tag, &db); return (err ? NULL : db); } void dbuf_create_bonus(dnode_t *dn) { ASSERT(RW_WRITE_HELD(&dn->dn_struct_rwlock)); ASSERT(dn->dn_bonus == NULL); dn->dn_bonus = dbuf_create(dn, 0, DMU_BONUS_BLKID, dn->dn_dbuf, NULL); } int dbuf_spill_set_blksz(dmu_buf_t *db_fake, uint64_t blksz, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; if (db->db_blkid != DMU_SPILL_BLKID) return (SET_ERROR(ENOTSUP)); if (blksz == 0) blksz = SPA_MINBLOCKSIZE; ASSERT3U(blksz, <=, spa_maxblocksize(dmu_objset_spa(db->db_objset))); blksz = P2ROUNDUP(blksz, SPA_MINBLOCKSIZE); dbuf_new_size(db, blksz, tx); return (0); } void dbuf_rm_spill(dnode_t *dn, dmu_tx_t *tx) { dbuf_free_range(dn, DMU_SPILL_BLKID, DMU_SPILL_BLKID, tx); } #pragma weak dmu_buf_add_ref = dbuf_add_ref void dbuf_add_ref(dmu_buf_impl_t *db, const void *tag) { int64_t holds = zfs_refcount_add(&db->db_holds, tag); VERIFY3S(holds, >, 1); } #pragma weak dmu_buf_try_add_ref = dbuf_try_add_ref boolean_t dbuf_try_add_ref(dmu_buf_t *db_fake, objset_t *os, uint64_t obj, uint64_t blkid, const void *tag) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; dmu_buf_impl_t *found_db; boolean_t result = B_FALSE; if (blkid == DMU_BONUS_BLKID) found_db = dbuf_find_bonus(os, obj); else found_db = dbuf_find(os, obj, 0, blkid); if (found_db != NULL) { if (db == found_db && dbuf_refcount(db) > db->db_dirtycnt) { (void) zfs_refcount_add(&db->db_holds, tag); result = B_TRUE; } mutex_exit(&found_db->db_mtx); } return (result); } /* * If you call dbuf_rele() you had better not be referencing the dnode handle * unless you have some other direct or indirect hold on the dnode. (An indirect * hold is a hold on one of the dnode's dbufs, including the bonus buffer.) * Without that, the dbuf_rele() could lead to a dnode_rele() followed by the * dnode's parent dbuf evicting its dnode handles. */ void dbuf_rele(dmu_buf_impl_t *db, const void *tag) { mutex_enter(&db->db_mtx); dbuf_rele_and_unlock(db, tag, B_FALSE); } void dmu_buf_rele(dmu_buf_t *db, const void *tag) { dbuf_rele((dmu_buf_impl_t *)db, tag); } /* * dbuf_rele() for an already-locked dbuf. This is necessary to allow * db_dirtycnt and db_holds to be updated atomically. The 'evicting' * argument should be set if we are already in the dbuf-evicting code * path, in which case we don't want to recursively evict. This allows us to * avoid deeply nested stacks that would have a call flow similar to this: * * dbuf_rele()-->dbuf_rele_and_unlock()-->dbuf_evict_notify() * ^ | * | | * +-----dbuf_destroy()<--dbuf_evict_one()<--------+ * */ void dbuf_rele_and_unlock(dmu_buf_impl_t *db, const void *tag, boolean_t evicting) { int64_t holds; uint64_t size; ASSERT(MUTEX_HELD(&db->db_mtx)); DBUF_VERIFY(db); /* * Remove the reference to the dbuf before removing its hold on the * dnode so we can guarantee in dnode_move() that a referenced bonus * buffer has a corresponding dnode hold. */ holds = zfs_refcount_remove(&db->db_holds, tag); ASSERT(holds >= 0); /* * We can't freeze indirects if there is a possibility that they * may be modified in the current syncing context. */ if (db->db_buf != NULL && holds == (db->db_level == 0 ? db->db_dirtycnt : 0)) { arc_buf_freeze(db->db_buf); } if (holds == db->db_dirtycnt && db->db_level == 0 && db->db_user_immediate_evict) dbuf_evict_user(db); if (holds == 0) { if (db->db_blkid == DMU_BONUS_BLKID) { dnode_t *dn; boolean_t evict_dbuf = db->db_pending_evict; /* * If the dnode moves here, we cannot cross this * barrier until the move completes. */ DB_DNODE_ENTER(db); dn = DB_DNODE(db); atomic_dec_32(&dn->dn_dbufs_count); /* * Decrementing the dbuf count means that the bonus * buffer's dnode hold is no longer discounted in * dnode_move(). The dnode cannot move until after * the dnode_rele() below. */ DB_DNODE_EXIT(db); /* * Do not reference db after its lock is dropped. * Another thread may evict it. */ mutex_exit(&db->db_mtx); if (evict_dbuf) dnode_evict_bonus(dn); dnode_rele(dn, db); } else if (db->db_buf == NULL) { /* * This is a special case: we never associated this * dbuf with any data allocated from the ARC. */ ASSERT(db->db_state == DB_UNCACHED || db->db_state == DB_NOFILL); dbuf_destroy(db); } else if (arc_released(db->db_buf)) { /* * This dbuf has anonymous data associated with it. */ dbuf_destroy(db); } else { boolean_t do_arc_evict = B_FALSE; blkptr_t bp; spa_t *spa = dmu_objset_spa(db->db_objset); if (!DBUF_IS_CACHEABLE(db) && db->db_blkptr != NULL && !BP_IS_HOLE(db->db_blkptr) && !BP_IS_EMBEDDED(db->db_blkptr)) { do_arc_evict = B_TRUE; bp = *db->db_blkptr; } if (!DBUF_IS_CACHEABLE(db) || db->db_pending_evict) { dbuf_destroy(db); } else if (!multilist_link_active(&db->db_cache_link)) { ASSERT3U(db->db_caching_status, ==, DB_NO_CACHE); dbuf_cached_state_t dcs = dbuf_include_in_metadata_cache(db) ? DB_DBUF_METADATA_CACHE : DB_DBUF_CACHE; db->db_caching_status = dcs; multilist_insert(&dbuf_caches[dcs].cache, db); uint64_t db_size = db->db.db_size; size = zfs_refcount_add_many( &dbuf_caches[dcs].size, db_size, db); uint8_t db_level = db->db_level; mutex_exit(&db->db_mtx); if (dcs == DB_DBUF_METADATA_CACHE) { DBUF_STAT_BUMP(metadata_cache_count); DBUF_STAT_MAX( metadata_cache_size_bytes_max, size); } else { DBUF_STAT_BUMP(cache_count); DBUF_STAT_MAX(cache_size_bytes_max, size); DBUF_STAT_BUMP(cache_levels[db_level]); DBUF_STAT_INCR( cache_levels_bytes[db_level], db_size); } if (dcs == DB_DBUF_CACHE && !evicting) dbuf_evict_notify(size); } if (do_arc_evict) arc_freed(spa, &bp); } } else { mutex_exit(&db->db_mtx); } } #pragma weak dmu_buf_refcount = dbuf_refcount uint64_t dbuf_refcount(dmu_buf_impl_t *db) { return (zfs_refcount_count(&db->db_holds)); } uint64_t dmu_buf_user_refcount(dmu_buf_t *db_fake) { uint64_t holds; dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; mutex_enter(&db->db_mtx); ASSERT3U(zfs_refcount_count(&db->db_holds), >=, db->db_dirtycnt); holds = zfs_refcount_count(&db->db_holds) - db->db_dirtycnt; mutex_exit(&db->db_mtx); return (holds); } void * dmu_buf_replace_user(dmu_buf_t *db_fake, dmu_buf_user_t *old_user, dmu_buf_user_t *new_user) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; mutex_enter(&db->db_mtx); dbuf_verify_user(db, DBVU_NOT_EVICTING); if (db->db_user == old_user) db->db_user = new_user; else old_user = db->db_user; dbuf_verify_user(db, DBVU_NOT_EVICTING); mutex_exit(&db->db_mtx); return (old_user); } void * dmu_buf_set_user(dmu_buf_t *db_fake, dmu_buf_user_t *user) { return (dmu_buf_replace_user(db_fake, NULL, user)); } void * dmu_buf_set_user_ie(dmu_buf_t *db_fake, dmu_buf_user_t *user) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; db->db_user_immediate_evict = TRUE; return (dmu_buf_set_user(db_fake, user)); } void * dmu_buf_remove_user(dmu_buf_t *db_fake, dmu_buf_user_t *user) { return (dmu_buf_replace_user(db_fake, user, NULL)); } void * dmu_buf_get_user(dmu_buf_t *db_fake) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; dbuf_verify_user(db, DBVU_NOT_EVICTING); return (db->db_user); } void dmu_buf_user_evict_wait(void) { taskq_wait(dbu_evict_taskq); } blkptr_t * dmu_buf_get_blkptr(dmu_buf_t *db) { dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; return (dbi->db_blkptr); } objset_t * dmu_buf_get_objset(dmu_buf_t *db) { dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; return (dbi->db_objset); } dnode_t * dmu_buf_dnode_enter(dmu_buf_t *db) { dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; DB_DNODE_ENTER(dbi); return (DB_DNODE(dbi)); } void dmu_buf_dnode_exit(dmu_buf_t *db) { dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; DB_DNODE_EXIT(dbi); } static void dbuf_check_blkptr(dnode_t *dn, dmu_buf_impl_t *db) { /* ASSERT(dmu_tx_is_syncing(tx) */ ASSERT(MUTEX_HELD(&db->db_mtx)); if (db->db_blkptr != NULL) return; if (db->db_blkid == DMU_SPILL_BLKID) { db->db_blkptr = DN_SPILL_BLKPTR(dn->dn_phys); BP_ZERO(db->db_blkptr); return; } if (db->db_level == dn->dn_phys->dn_nlevels-1) { /* * This buffer was allocated at a time when there was * no available blkptrs from the dnode, or it was * inappropriate to hook it in (i.e., nlevels mismatch). */ ASSERT(db->db_blkid < dn->dn_phys->dn_nblkptr); ASSERT(db->db_parent == NULL); db->db_parent = dn->dn_dbuf; db->db_blkptr = &dn->dn_phys->dn_blkptr[db->db_blkid]; DBUF_VERIFY(db); } else { dmu_buf_impl_t *parent = db->db_parent; int epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; ASSERT(dn->dn_phys->dn_nlevels > 1); if (parent == NULL) { mutex_exit(&db->db_mtx); rw_enter(&dn->dn_struct_rwlock, RW_READER); parent = dbuf_hold_level(dn, db->db_level + 1, db->db_blkid >> epbs, db); rw_exit(&dn->dn_struct_rwlock); mutex_enter(&db->db_mtx); db->db_parent = parent; } db->db_blkptr = (blkptr_t *)parent->db.db_data + (db->db_blkid & ((1ULL << epbs) - 1)); DBUF_VERIFY(db); } } static void dbuf_sync_bonus(dbuf_dirty_record_t *dr, dmu_tx_t *tx) { dmu_buf_impl_t *db = dr->dr_dbuf; void *data = dr->dt.dl.dr_data; ASSERT0(db->db_level); ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(db->db_blkid == DMU_BONUS_BLKID); ASSERT(data != NULL); dnode_t *dn = dr->dr_dnode; ASSERT3U(DN_MAX_BONUS_LEN(dn->dn_phys), <=, DN_SLOTS_TO_BONUSLEN(dn->dn_phys->dn_extra_slots + 1)); memcpy(DN_BONUS(dn->dn_phys), data, DN_MAX_BONUS_LEN(dn->dn_phys)); dbuf_sync_leaf_verify_bonus_dnode(dr); dbuf_undirty_bonus(dr); dbuf_rele_and_unlock(db, (void *)(uintptr_t)tx->tx_txg, B_FALSE); } /* * When syncing out a blocks of dnodes, adjust the block to deal with * encryption. Normally, we make sure the block is decrypted before writing * it. If we have crypt params, then we are writing a raw (encrypted) block, * from a raw receive. In this case, set the ARC buf's crypt params so * that the BP will be filled with the correct byteorder, salt, iv, and mac. */ static void dbuf_prepare_encrypted_dnode_leaf(dbuf_dirty_record_t *dr) { int err; dmu_buf_impl_t *db = dr->dr_dbuf; ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT3U(db->db.db_object, ==, DMU_META_DNODE_OBJECT); ASSERT3U(db->db_level, ==, 0); if (!db->db_objset->os_raw_receive && arc_is_encrypted(db->db_buf)) { zbookmark_phys_t zb; /* * Unfortunately, there is currently no mechanism for * syncing context to handle decryption errors. An error * here is only possible if an attacker maliciously * changed a dnode block and updated the associated * checksums going up the block tree. */ SET_BOOKMARK(&zb, dmu_objset_id(db->db_objset), db->db.db_object, db->db_level, db->db_blkid); err = arc_untransform(db->db_buf, db->db_objset->os_spa, &zb, B_TRUE); if (err) panic("Invalid dnode block MAC"); } else if (dr->dt.dl.dr_has_raw_params) { (void) arc_release(dr->dt.dl.dr_data, db); arc_convert_to_raw(dr->dt.dl.dr_data, dmu_objset_id(db->db_objset), dr->dt.dl.dr_byteorder, DMU_OT_DNODE, dr->dt.dl.dr_salt, dr->dt.dl.dr_iv, dr->dt.dl.dr_mac); } } /* * dbuf_sync_indirect() is called recursively from dbuf_sync_list() so it * is critical the we not allow the compiler to inline this function in to * dbuf_sync_list() thereby drastically bloating the stack usage. */ noinline static void dbuf_sync_indirect(dbuf_dirty_record_t *dr, dmu_tx_t *tx) { dmu_buf_impl_t *db = dr->dr_dbuf; dnode_t *dn = dr->dr_dnode; ASSERT(dmu_tx_is_syncing(tx)); dprintf_dbuf_bp(db, db->db_blkptr, "blkptr=%p", db->db_blkptr); mutex_enter(&db->db_mtx); ASSERT(db->db_level > 0); DBUF_VERIFY(db); /* Read the block if it hasn't been read yet. */ if (db->db_buf == NULL) { mutex_exit(&db->db_mtx); (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED); mutex_enter(&db->db_mtx); } ASSERT3U(db->db_state, ==, DB_CACHED); ASSERT(db->db_buf != NULL); /* Indirect block size must match what the dnode thinks it is. */ ASSERT3U(db->db.db_size, ==, 1<dn_phys->dn_indblkshift); dbuf_check_blkptr(dn, db); /* Provide the pending dirty record to child dbufs */ db->db_data_pending = dr; mutex_exit(&db->db_mtx); dbuf_write(dr, db->db_buf, tx); zio_t *zio = dr->dr_zio; mutex_enter(&dr->dt.di.dr_mtx); dbuf_sync_list(&dr->dt.di.dr_children, db->db_level - 1, tx); ASSERT(list_head(&dr->dt.di.dr_children) == NULL); mutex_exit(&dr->dt.di.dr_mtx); zio_nowait(zio); } /* * Verify that the size of the data in our bonus buffer does not exceed * its recorded size. * * The purpose of this verification is to catch any cases in development * where the size of a phys structure (i.e space_map_phys_t) grows and, * due to incorrect feature management, older pools expect to read more * data even though they didn't actually write it to begin with. * * For a example, this would catch an error in the feature logic where we * open an older pool and we expect to write the space map histogram of * a space map with size SPACE_MAP_SIZE_V0. */ static void dbuf_sync_leaf_verify_bonus_dnode(dbuf_dirty_record_t *dr) { #ifdef ZFS_DEBUG dnode_t *dn = dr->dr_dnode; /* * Encrypted bonus buffers can have data past their bonuslen. * Skip the verification of these blocks. */ if (DMU_OT_IS_ENCRYPTED(dn->dn_bonustype)) return; uint16_t bonuslen = dn->dn_phys->dn_bonuslen; uint16_t maxbonuslen = DN_SLOTS_TO_BONUSLEN(dn->dn_num_slots); ASSERT3U(bonuslen, <=, maxbonuslen); arc_buf_t *datap = dr->dt.dl.dr_data; char *datap_end = ((char *)datap) + bonuslen; char *datap_max = ((char *)datap) + maxbonuslen; /* ensure that everything is zero after our data */ for (; datap_end < datap_max; datap_end++) ASSERT(*datap_end == 0); #endif } static blkptr_t * dbuf_lightweight_bp(dbuf_dirty_record_t *dr) { /* This must be a lightweight dirty record. */ ASSERT3P(dr->dr_dbuf, ==, NULL); dnode_t *dn = dr->dr_dnode; if (dn->dn_phys->dn_nlevels == 1) { VERIFY3U(dr->dt.dll.dr_blkid, <, dn->dn_phys->dn_nblkptr); return (&dn->dn_phys->dn_blkptr[dr->dt.dll.dr_blkid]); } else { dmu_buf_impl_t *parent_db = dr->dr_parent->dr_dbuf; int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; VERIFY3U(parent_db->db_level, ==, 1); VERIFY3P(parent_db->db_dnode_handle->dnh_dnode, ==, dn); VERIFY3U(dr->dt.dll.dr_blkid >> epbs, ==, parent_db->db_blkid); blkptr_t *bp = parent_db->db.db_data; return (&bp[dr->dt.dll.dr_blkid & ((1 << epbs) - 1)]); } } static void dbuf_lightweight_ready(zio_t *zio) { dbuf_dirty_record_t *dr = zio->io_private; blkptr_t *bp = zio->io_bp; if (zio->io_error != 0) return; dnode_t *dn = dr->dr_dnode; blkptr_t *bp_orig = dbuf_lightweight_bp(dr); spa_t *spa = dmu_objset_spa(dn->dn_objset); int64_t delta = bp_get_dsize_sync(spa, bp) - bp_get_dsize_sync(spa, bp_orig); dnode_diduse_space(dn, delta); uint64_t blkid = dr->dt.dll.dr_blkid; mutex_enter(&dn->dn_mtx); if (blkid > dn->dn_phys->dn_maxblkid) { ASSERT0(dn->dn_objset->os_raw_receive); dn->dn_phys->dn_maxblkid = blkid; } mutex_exit(&dn->dn_mtx); if (!BP_IS_EMBEDDED(bp)) { uint64_t fill = BP_IS_HOLE(bp) ? 0 : 1; BP_SET_FILL(bp, fill); } dmu_buf_impl_t *parent_db; EQUIV(dr->dr_parent == NULL, dn->dn_phys->dn_nlevels == 1); if (dr->dr_parent == NULL) { parent_db = dn->dn_dbuf; } else { parent_db = dr->dr_parent->dr_dbuf; } rw_enter(&parent_db->db_rwlock, RW_WRITER); *bp_orig = *bp; rw_exit(&parent_db->db_rwlock); } static void dbuf_lightweight_physdone(zio_t *zio) { dbuf_dirty_record_t *dr = zio->io_private; dsl_pool_t *dp = spa_get_dsl(zio->io_spa); ASSERT3U(dr->dr_txg, ==, zio->io_txg); /* * The callback will be called io_phys_children times. Retire one * portion of our dirty space each time we are called. Any rounding * error will be cleaned up by dbuf_lightweight_done(). */ int delta = dr->dr_accounted / zio->io_phys_children; dsl_pool_undirty_space(dp, delta, zio->io_txg); } static void dbuf_lightweight_done(zio_t *zio) { dbuf_dirty_record_t *dr = zio->io_private; VERIFY0(zio->io_error); objset_t *os = dr->dr_dnode->dn_objset; dmu_tx_t *tx = os->os_synctx; if (zio->io_flags & (ZIO_FLAG_IO_REWRITE | ZIO_FLAG_NOPWRITE)) { ASSERT(BP_EQUAL(zio->io_bp, &zio->io_bp_orig)); } else { dsl_dataset_t *ds = os->os_dsl_dataset; (void) dsl_dataset_block_kill(ds, &zio->io_bp_orig, tx, B_TRUE); dsl_dataset_block_born(ds, zio->io_bp, tx); } /* * See comment in dbuf_write_done(). */ if (zio->io_phys_children == 0) { dsl_pool_undirty_space(dmu_objset_pool(os), dr->dr_accounted, zio->io_txg); } else { dsl_pool_undirty_space(dmu_objset_pool(os), dr->dr_accounted % zio->io_phys_children, zio->io_txg); } abd_free(dr->dt.dll.dr_abd); kmem_free(dr, sizeof (*dr)); } noinline static void dbuf_sync_lightweight(dbuf_dirty_record_t *dr, dmu_tx_t *tx) { dnode_t *dn = dr->dr_dnode; zio_t *pio; if (dn->dn_phys->dn_nlevels == 1) { pio = dn->dn_zio; } else { pio = dr->dr_parent->dr_zio; } zbookmark_phys_t zb = { .zb_objset = dmu_objset_id(dn->dn_objset), .zb_object = dn->dn_object, .zb_level = 0, .zb_blkid = dr->dt.dll.dr_blkid, }; /* * See comment in dbuf_write(). This is so that zio->io_bp_orig * will have the old BP in dbuf_lightweight_done(). */ dr->dr_bp_copy = *dbuf_lightweight_bp(dr); dr->dr_zio = zio_write(pio, dmu_objset_spa(dn->dn_objset), dmu_tx_get_txg(tx), &dr->dr_bp_copy, dr->dt.dll.dr_abd, dn->dn_datablksz, abd_get_size(dr->dt.dll.dr_abd), &dr->dt.dll.dr_props, dbuf_lightweight_ready, NULL, dbuf_lightweight_physdone, dbuf_lightweight_done, dr, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED | dr->dt.dll.dr_flags, &zb); zio_nowait(dr->dr_zio); } /* * dbuf_sync_leaf() is called recursively from dbuf_sync_list() so it is * critical the we not allow the compiler to inline this function in to * dbuf_sync_list() thereby drastically bloating the stack usage. */ noinline static void dbuf_sync_leaf(dbuf_dirty_record_t *dr, dmu_tx_t *tx) { arc_buf_t **datap = &dr->dt.dl.dr_data; dmu_buf_impl_t *db = dr->dr_dbuf; dnode_t *dn = dr->dr_dnode; objset_t *os; uint64_t txg = tx->tx_txg; ASSERT(dmu_tx_is_syncing(tx)); dprintf_dbuf_bp(db, db->db_blkptr, "blkptr=%p", db->db_blkptr); mutex_enter(&db->db_mtx); /* * To be synced, we must be dirtied. But we * might have been freed after the dirty. */ if (db->db_state == DB_UNCACHED) { /* This buffer has been freed since it was dirtied */ ASSERT(db->db.db_data == NULL); } else if (db->db_state == DB_FILL) { /* This buffer was freed and is now being re-filled */ ASSERT(db->db.db_data != dr->dt.dl.dr_data); } else { ASSERT(db->db_state == DB_CACHED || db->db_state == DB_NOFILL); } DBUF_VERIFY(db); if (db->db_blkid == DMU_SPILL_BLKID) { mutex_enter(&dn->dn_mtx); if (!(dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR)) { /* * In the previous transaction group, the bonus buffer * was entirely used to store the attributes for the * dnode which overrode the dn_spill field. However, * when adding more attributes to the file a spill * block was required to hold the extra attributes. * * Make sure to clear the garbage left in the dn_spill * field from the previous attributes in the bonus * buffer. Otherwise, after writing out the spill * block to the new allocated dva, it will free * the old block pointed to by the invalid dn_spill. */ db->db_blkptr = NULL; } dn->dn_phys->dn_flags |= DNODE_FLAG_SPILL_BLKPTR; mutex_exit(&dn->dn_mtx); } /* * If this is a bonus buffer, simply copy the bonus data into the * dnode. It will be written out when the dnode is synced (and it * will be synced, since it must have been dirty for dbuf_sync to * be called). */ if (db->db_blkid == DMU_BONUS_BLKID) { ASSERT(dr->dr_dbuf == db); dbuf_sync_bonus(dr, tx); return; } os = dn->dn_objset; /* * This function may have dropped the db_mtx lock allowing a dmu_sync * operation to sneak in. As a result, we need to ensure that we * don't check the dr_override_state until we have returned from * dbuf_check_blkptr. */ dbuf_check_blkptr(dn, db); /* * If this buffer is in the middle of an immediate write, * wait for the synchronous IO to complete. */ while (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC) { ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); cv_wait(&db->db_changed, &db->db_mtx); ASSERT(dr->dt.dl.dr_override_state != DR_NOT_OVERRIDDEN); } /* * If this is a dnode block, ensure it is appropriately encrypted * or decrypted, depending on what we are writing to it this txg. */ if (os->os_encrypted && dn->dn_object == DMU_META_DNODE_OBJECT) dbuf_prepare_encrypted_dnode_leaf(dr); if (db->db_state != DB_NOFILL && dn->dn_object != DMU_META_DNODE_OBJECT && zfs_refcount_count(&db->db_holds) > 1 && dr->dt.dl.dr_override_state != DR_OVERRIDDEN && *datap == db->db_buf) { /* * If this buffer is currently "in use" (i.e., there * are active holds and db_data still references it), * then make a copy before we start the write so that * any modifications from the open txg will not leak * into this write. * * NOTE: this copy does not need to be made for * objects only modified in the syncing context (e.g. * DNONE_DNODE blocks). */ int psize = arc_buf_size(*datap); int lsize = arc_buf_lsize(*datap); arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db); enum zio_compress compress_type = arc_get_compression(*datap); uint8_t complevel = arc_get_complevel(*datap); if (arc_is_encrypted(*datap)) { boolean_t byteorder; uint8_t salt[ZIO_DATA_SALT_LEN]; uint8_t iv[ZIO_DATA_IV_LEN]; uint8_t mac[ZIO_DATA_MAC_LEN]; arc_get_raw_params(*datap, &byteorder, salt, iv, mac); *datap = arc_alloc_raw_buf(os->os_spa, db, dmu_objset_id(os), byteorder, salt, iv, mac, dn->dn_type, psize, lsize, compress_type, complevel); } else if (compress_type != ZIO_COMPRESS_OFF) { ASSERT3U(type, ==, ARC_BUFC_DATA); *datap = arc_alloc_compressed_buf(os->os_spa, db, psize, lsize, compress_type, complevel); } else { *datap = arc_alloc_buf(os->os_spa, db, type, psize); } memcpy((*datap)->b_data, db->db.db_data, psize); } db->db_data_pending = dr; mutex_exit(&db->db_mtx); dbuf_write(dr, *datap, tx); ASSERT(!list_link_active(&dr->dr_dirty_node)); if (dn->dn_object == DMU_META_DNODE_OBJECT) { list_insert_tail(&dn->dn_dirty_records[txg & TXG_MASK], dr); } else { zio_nowait(dr->dr_zio); } } void dbuf_sync_list(list_t *list, int level, dmu_tx_t *tx) { dbuf_dirty_record_t *dr; while ((dr = list_head(list))) { if (dr->dr_zio != NULL) { /* * If we find an already initialized zio then we * are processing the meta-dnode, and we have finished. * The dbufs for all dnodes are put back on the list * during processing, so that we can zio_wait() * these IOs after initiating all child IOs. */ ASSERT3U(dr->dr_dbuf->db.db_object, ==, DMU_META_DNODE_OBJECT); break; } list_remove(list, dr); if (dr->dr_dbuf == NULL) { dbuf_sync_lightweight(dr, tx); } else { if (dr->dr_dbuf->db_blkid != DMU_BONUS_BLKID && dr->dr_dbuf->db_blkid != DMU_SPILL_BLKID) { VERIFY3U(dr->dr_dbuf->db_level, ==, level); } if (dr->dr_dbuf->db_level > 0) dbuf_sync_indirect(dr, tx); else dbuf_sync_leaf(dr, tx); } } } static void dbuf_write_ready(zio_t *zio, arc_buf_t *buf, void *vdb) { (void) buf; dmu_buf_impl_t *db = vdb; dnode_t *dn; blkptr_t *bp = zio->io_bp; blkptr_t *bp_orig = &zio->io_bp_orig; spa_t *spa = zio->io_spa; int64_t delta; uint64_t fill = 0; int i; ASSERT3P(db->db_blkptr, !=, NULL); ASSERT3P(&db->db_data_pending->dr_bp_copy, ==, bp); DB_DNODE_ENTER(db); dn = DB_DNODE(db); delta = bp_get_dsize_sync(spa, bp) - bp_get_dsize_sync(spa, bp_orig); dnode_diduse_space(dn, delta - zio->io_prev_space_delta); zio->io_prev_space_delta = delta; if (bp->blk_birth != 0) { ASSERT((db->db_blkid != DMU_SPILL_BLKID && BP_GET_TYPE(bp) == dn->dn_type) || (db->db_blkid == DMU_SPILL_BLKID && BP_GET_TYPE(bp) == dn->dn_bonustype) || BP_IS_EMBEDDED(bp)); ASSERT(BP_GET_LEVEL(bp) == db->db_level); } mutex_enter(&db->db_mtx); #ifdef ZFS_DEBUG if (db->db_blkid == DMU_SPILL_BLKID) { ASSERT(dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR); ASSERT(!(BP_IS_HOLE(bp)) && db->db_blkptr == DN_SPILL_BLKPTR(dn->dn_phys)); } #endif if (db->db_level == 0) { mutex_enter(&dn->dn_mtx); if (db->db_blkid > dn->dn_phys->dn_maxblkid && db->db_blkid != DMU_SPILL_BLKID) { ASSERT0(db->db_objset->os_raw_receive); dn->dn_phys->dn_maxblkid = db->db_blkid; } mutex_exit(&dn->dn_mtx); if (dn->dn_type == DMU_OT_DNODE) { i = 0; while (i < db->db.db_size) { dnode_phys_t *dnp = (void *)(((char *)db->db.db_data) + i); i += DNODE_MIN_SIZE; if (dnp->dn_type != DMU_OT_NONE) { fill++; i += dnp->dn_extra_slots * DNODE_MIN_SIZE; } } } else { if (BP_IS_HOLE(bp)) { fill = 0; } else { fill = 1; } } } else { blkptr_t *ibp = db->db.db_data; ASSERT3U(db->db.db_size, ==, 1<dn_phys->dn_indblkshift); for (i = db->db.db_size >> SPA_BLKPTRSHIFT; i > 0; i--, ibp++) { if (BP_IS_HOLE(ibp)) continue; fill += BP_GET_FILL(ibp); } } DB_DNODE_EXIT(db); if (!BP_IS_EMBEDDED(bp)) BP_SET_FILL(bp, fill); mutex_exit(&db->db_mtx); db_lock_type_t dblt = dmu_buf_lock_parent(db, RW_WRITER, FTAG); *db->db_blkptr = *bp; dmu_buf_unlock_parent(db, dblt, FTAG); } /* * This function gets called just prior to running through the compression * stage of the zio pipeline. If we're an indirect block comprised of only * holes, then we want this indirect to be compressed away to a hole. In * order to do that we must zero out any information about the holes that * this indirect points to prior to before we try to compress it. */ static void dbuf_write_children_ready(zio_t *zio, arc_buf_t *buf, void *vdb) { (void) zio, (void) buf; dmu_buf_impl_t *db = vdb; dnode_t *dn; blkptr_t *bp; unsigned int epbs, i; ASSERT3U(db->db_level, >, 0); DB_DNODE_ENTER(db); dn = DB_DNODE(db); epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; ASSERT3U(epbs, <, 31); /* Determine if all our children are holes */ for (i = 0, bp = db->db.db_data; i < 1ULL << epbs; i++, bp++) { if (!BP_IS_HOLE(bp)) break; } /* * If all the children are holes, then zero them all out so that * we may get compressed away. */ if (i == 1ULL << epbs) { /* * We only found holes. Grab the rwlock to prevent * anybody from reading the blocks we're about to * zero out. */ rw_enter(&db->db_rwlock, RW_WRITER); memset(db->db.db_data, 0, db->db.db_size); rw_exit(&db->db_rwlock); } DB_DNODE_EXIT(db); } /* * The SPA will call this callback several times for each zio - once * for every physical child i/o (zio->io_phys_children times). This * allows the DMU to monitor the progress of each logical i/o. For example, * there may be 2 copies of an indirect block, or many fragments of a RAID-Z * block. There may be a long delay before all copies/fragments are completed, * so this callback allows us to retire dirty space gradually, as the physical * i/os complete. */ static void dbuf_write_physdone(zio_t *zio, arc_buf_t *buf, void *arg) { (void) buf; dmu_buf_impl_t *db = arg; objset_t *os = db->db_objset; dsl_pool_t *dp = dmu_objset_pool(os); dbuf_dirty_record_t *dr; int delta = 0; dr = db->db_data_pending; ASSERT3U(dr->dr_txg, ==, zio->io_txg); /* * The callback will be called io_phys_children times. Retire one * portion of our dirty space each time we are called. Any rounding * error will be cleaned up by dbuf_write_done(). */ delta = dr->dr_accounted / zio->io_phys_children; dsl_pool_undirty_space(dp, delta, zio->io_txg); } static void dbuf_write_done(zio_t *zio, arc_buf_t *buf, void *vdb) { (void) buf; dmu_buf_impl_t *db = vdb; blkptr_t *bp_orig = &zio->io_bp_orig; blkptr_t *bp = db->db_blkptr; objset_t *os = db->db_objset; dmu_tx_t *tx = os->os_synctx; ASSERT0(zio->io_error); ASSERT(db->db_blkptr == bp); /* * For nopwrites and rewrites we ensure that the bp matches our * original and bypass all the accounting. */ if (zio->io_flags & (ZIO_FLAG_IO_REWRITE | ZIO_FLAG_NOPWRITE)) { ASSERT(BP_EQUAL(bp, bp_orig)); } else { dsl_dataset_t *ds = os->os_dsl_dataset; (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE); dsl_dataset_block_born(ds, bp, tx); } mutex_enter(&db->db_mtx); DBUF_VERIFY(db); dbuf_dirty_record_t *dr = db->db_data_pending; dnode_t *dn = dr->dr_dnode; ASSERT(!list_link_active(&dr->dr_dirty_node)); ASSERT(dr->dr_dbuf == db); ASSERT(list_next(&db->db_dirty_records, dr) == NULL); list_remove(&db->db_dirty_records, dr); #ifdef ZFS_DEBUG if (db->db_blkid == DMU_SPILL_BLKID) { ASSERT(dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR); ASSERT(!(BP_IS_HOLE(db->db_blkptr)) && db->db_blkptr == DN_SPILL_BLKPTR(dn->dn_phys)); } #endif if (db->db_level == 0) { ASSERT(db->db_blkid != DMU_BONUS_BLKID); ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); if (db->db_state != DB_NOFILL) { if (dr->dt.dl.dr_data != db->db_buf) arc_buf_destroy(dr->dt.dl.dr_data, db); } } else { ASSERT(list_head(&dr->dt.di.dr_children) == NULL); ASSERT3U(db->db.db_size, ==, 1 << dn->dn_phys->dn_indblkshift); if (!BP_IS_HOLE(db->db_blkptr)) { int epbs __maybe_unused = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; ASSERT3U(db->db_blkid, <=, dn->dn_phys->dn_maxblkid >> (db->db_level * epbs)); ASSERT3U(BP_GET_LSIZE(db->db_blkptr), ==, db->db.db_size); } mutex_destroy(&dr->dt.di.dr_mtx); list_destroy(&dr->dt.di.dr_children); } cv_broadcast(&db->db_changed); ASSERT(db->db_dirtycnt > 0); db->db_dirtycnt -= 1; db->db_data_pending = NULL; dbuf_rele_and_unlock(db, (void *)(uintptr_t)tx->tx_txg, B_FALSE); /* * If we didn't do a physical write in this ZIO and we * still ended up here, it means that the space of the * dbuf that we just released (and undirtied) above hasn't * been marked as undirtied in the pool's accounting. * * Thus, we undirty that space in the pool's view of the * world here. For physical writes this type of update * happens in dbuf_write_physdone(). * * If we did a physical write, cleanup any rounding errors * that came up due to writing multiple copies of a block * on disk [see dbuf_write_physdone()]. */ if (zio->io_phys_children == 0) { dsl_pool_undirty_space(dmu_objset_pool(os), dr->dr_accounted, zio->io_txg); } else { dsl_pool_undirty_space(dmu_objset_pool(os), dr->dr_accounted % zio->io_phys_children, zio->io_txg); } kmem_free(dr, sizeof (dbuf_dirty_record_t)); } static void dbuf_write_nofill_ready(zio_t *zio) { dbuf_write_ready(zio, NULL, zio->io_private); } static void dbuf_write_nofill_done(zio_t *zio) { dbuf_write_done(zio, NULL, zio->io_private); } static void dbuf_write_override_ready(zio_t *zio) { dbuf_dirty_record_t *dr = zio->io_private; dmu_buf_impl_t *db = dr->dr_dbuf; dbuf_write_ready(zio, NULL, db); } static void dbuf_write_override_done(zio_t *zio) { dbuf_dirty_record_t *dr = zio->io_private; dmu_buf_impl_t *db = dr->dr_dbuf; blkptr_t *obp = &dr->dt.dl.dr_overridden_by; mutex_enter(&db->db_mtx); if (!BP_EQUAL(zio->io_bp, obp)) { if (!BP_IS_HOLE(obp)) dsl_free(spa_get_dsl(zio->io_spa), zio->io_txg, obp); arc_release(dr->dt.dl.dr_data, db); } mutex_exit(&db->db_mtx); dbuf_write_done(zio, NULL, db); if (zio->io_abd != NULL) abd_free(zio->io_abd); } typedef struct dbuf_remap_impl_callback_arg { objset_t *drica_os; uint64_t drica_blk_birth; dmu_tx_t *drica_tx; } dbuf_remap_impl_callback_arg_t; static void dbuf_remap_impl_callback(uint64_t vdev, uint64_t offset, uint64_t size, void *arg) { dbuf_remap_impl_callback_arg_t *drica = arg; objset_t *os = drica->drica_os; spa_t *spa = dmu_objset_spa(os); dmu_tx_t *tx = drica->drica_tx; ASSERT(dsl_pool_sync_context(spa_get_dsl(spa))); if (os == spa_meta_objset(spa)) { spa_vdev_indirect_mark_obsolete(spa, vdev, offset, size, tx); } else { dsl_dataset_block_remapped(dmu_objset_ds(os), vdev, offset, size, drica->drica_blk_birth, tx); } } static void dbuf_remap_impl(dnode_t *dn, blkptr_t *bp, krwlock_t *rw, dmu_tx_t *tx) { blkptr_t bp_copy = *bp; spa_t *spa = dmu_objset_spa(dn->dn_objset); dbuf_remap_impl_callback_arg_t drica; ASSERT(dsl_pool_sync_context(spa_get_dsl(spa))); drica.drica_os = dn->dn_objset; drica.drica_blk_birth = bp->blk_birth; drica.drica_tx = tx; if (spa_remap_blkptr(spa, &bp_copy, dbuf_remap_impl_callback, &drica)) { /* * If the blkptr being remapped is tracked by a livelist, * then we need to make sure the livelist reflects the update. * First, cancel out the old blkptr by appending a 'FREE' * entry. Next, add an 'ALLOC' to track the new version. This * way we avoid trying to free an inaccurate blkptr at delete. * Note that embedded blkptrs are not tracked in livelists. */ if (dn->dn_objset != spa_meta_objset(spa)) { dsl_dataset_t *ds = dmu_objset_ds(dn->dn_objset); if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist) && bp->blk_birth > ds->ds_dir->dd_origin_txg) { ASSERT(!BP_IS_EMBEDDED(bp)); ASSERT(dsl_dir_is_clone(ds->ds_dir)); ASSERT(spa_feature_is_enabled(spa, SPA_FEATURE_LIVELIST)); bplist_append(&ds->ds_dir->dd_pending_frees, bp); bplist_append(&ds->ds_dir->dd_pending_allocs, &bp_copy); } } /* * The db_rwlock prevents dbuf_read_impl() from * dereferencing the BP while we are changing it. To * avoid lock contention, only grab it when we are actually * changing the BP. */ if (rw != NULL) rw_enter(rw, RW_WRITER); *bp = bp_copy; if (rw != NULL) rw_exit(rw); } } /* * Remap any existing BP's to concrete vdevs, if possible. */ static void dbuf_remap(dnode_t *dn, dmu_buf_impl_t *db, dmu_tx_t *tx) { spa_t *spa = dmu_objset_spa(db->db_objset); ASSERT(dsl_pool_sync_context(spa_get_dsl(spa))); if (!spa_feature_is_active(spa, SPA_FEATURE_DEVICE_REMOVAL)) return; if (db->db_level > 0) { blkptr_t *bp = db->db.db_data; for (int i = 0; i < db->db.db_size >> SPA_BLKPTRSHIFT; i++) { dbuf_remap_impl(dn, &bp[i], &db->db_rwlock, tx); } } else if (db->db.db_object == DMU_META_DNODE_OBJECT) { dnode_phys_t *dnp = db->db.db_data; ASSERT3U(db->db_dnode_handle->dnh_dnode->dn_type, ==, DMU_OT_DNODE); for (int i = 0; i < db->db.db_size >> DNODE_SHIFT; i += dnp[i].dn_extra_slots + 1) { for (int j = 0; j < dnp[i].dn_nblkptr; j++) { krwlock_t *lock = (dn->dn_dbuf == NULL ? NULL : &dn->dn_dbuf->db_rwlock); dbuf_remap_impl(dn, &dnp[i].dn_blkptr[j], lock, tx); } } } } /* Issue I/O to commit a dirty buffer to disk. */ static void dbuf_write(dbuf_dirty_record_t *dr, arc_buf_t *data, dmu_tx_t *tx) { dmu_buf_impl_t *db = dr->dr_dbuf; dnode_t *dn = dr->dr_dnode; objset_t *os; dmu_buf_impl_t *parent = db->db_parent; uint64_t txg = tx->tx_txg; zbookmark_phys_t zb; zio_prop_t zp; zio_t *pio; /* parent I/O */ int wp_flag = 0; ASSERT(dmu_tx_is_syncing(tx)); os = dn->dn_objset; if (db->db_state != DB_NOFILL) { if (db->db_level > 0 || dn->dn_type == DMU_OT_DNODE) { /* * Private object buffers are released here rather * than in dbuf_dirty() since they are only modified * in the syncing context and we don't want the * overhead of making multiple copies of the data. */ if (BP_IS_HOLE(db->db_blkptr)) { arc_buf_thaw(data); } else { dbuf_release_bp(db); } dbuf_remap(dn, db, tx); } } if (parent != dn->dn_dbuf) { /* Our parent is an indirect block. */ /* We have a dirty parent that has been scheduled for write. */ ASSERT(parent && parent->db_data_pending); /* Our parent's buffer is one level closer to the dnode. */ ASSERT(db->db_level == parent->db_level-1); /* * We're about to modify our parent's db_data by modifying * our block pointer, so the parent must be released. */ ASSERT(arc_released(parent->db_buf)); pio = parent->db_data_pending->dr_zio; } else { /* Our parent is the dnode itself. */ ASSERT((db->db_level == dn->dn_phys->dn_nlevels-1 && db->db_blkid != DMU_SPILL_BLKID) || (db->db_blkid == DMU_SPILL_BLKID && db->db_level == 0)); if (db->db_blkid != DMU_SPILL_BLKID) ASSERT3P(db->db_blkptr, ==, &dn->dn_phys->dn_blkptr[db->db_blkid]); pio = dn->dn_zio; } ASSERT(db->db_level == 0 || data == db->db_buf); ASSERT3U(db->db_blkptr->blk_birth, <=, txg); ASSERT(pio); SET_BOOKMARK(&zb, os->os_dsl_dataset ? os->os_dsl_dataset->ds_object : DMU_META_OBJSET, db->db.db_object, db->db_level, db->db_blkid); if (db->db_blkid == DMU_SPILL_BLKID) wp_flag = WP_SPILL; wp_flag |= (db->db_state == DB_NOFILL) ? WP_NOFILL : 0; dmu_write_policy(os, dn, db->db_level, wp_flag, &zp); /* * We copy the blkptr now (rather than when we instantiate the dirty * record), because its value can change between open context and * syncing context. We do not need to hold dn_struct_rwlock to read * db_blkptr because we are in syncing context. */ dr->dr_bp_copy = *db->db_blkptr; if (db->db_level == 0 && dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { /* * The BP for this block has been provided by open context * (by dmu_sync() or dmu_buf_write_embedded()). */ abd_t *contents = (data != NULL) ? abd_get_from_buf(data->b_data, arc_buf_size(data)) : NULL; dr->dr_zio = zio_write(pio, os->os_spa, txg, &dr->dr_bp_copy, contents, db->db.db_size, db->db.db_size, &zp, dbuf_write_override_ready, NULL, NULL, dbuf_write_override_done, dr, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); mutex_enter(&db->db_mtx); dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; zio_write_override(dr->dr_zio, &dr->dt.dl.dr_overridden_by, dr->dt.dl.dr_copies, dr->dt.dl.dr_nopwrite); mutex_exit(&db->db_mtx); } else if (db->db_state == DB_NOFILL) { ASSERT(zp.zp_checksum == ZIO_CHECKSUM_OFF || zp.zp_checksum == ZIO_CHECKSUM_NOPARITY); dr->dr_zio = zio_write(pio, os->os_spa, txg, &dr->dr_bp_copy, NULL, db->db.db_size, db->db.db_size, &zp, dbuf_write_nofill_ready, NULL, NULL, dbuf_write_nofill_done, db, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED | ZIO_FLAG_NODATA, &zb); } else { ASSERT(arc_released(data)); /* * For indirect blocks, we want to setup the children * ready callback so that we can properly handle an indirect * block that only contains holes. */ arc_write_done_func_t *children_ready_cb = NULL; if (db->db_level != 0) children_ready_cb = dbuf_write_children_ready; dr->dr_zio = arc_write(pio, os->os_spa, txg, &dr->dr_bp_copy, data, dbuf_is_l2cacheable(db), &zp, dbuf_write_ready, children_ready_cb, dbuf_write_physdone, dbuf_write_done, db, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); } } EXPORT_SYMBOL(dbuf_find); EXPORT_SYMBOL(dbuf_is_metadata); EXPORT_SYMBOL(dbuf_destroy); EXPORT_SYMBOL(dbuf_loan_arcbuf); EXPORT_SYMBOL(dbuf_whichblock); EXPORT_SYMBOL(dbuf_read); EXPORT_SYMBOL(dbuf_unoverride); EXPORT_SYMBOL(dbuf_free_range); EXPORT_SYMBOL(dbuf_new_size); EXPORT_SYMBOL(dbuf_release_bp); EXPORT_SYMBOL(dbuf_dirty); EXPORT_SYMBOL(dmu_buf_set_crypt_params); EXPORT_SYMBOL(dmu_buf_will_dirty); EXPORT_SYMBOL(dmu_buf_is_dirty); EXPORT_SYMBOL(dmu_buf_will_not_fill); EXPORT_SYMBOL(dmu_buf_will_fill); EXPORT_SYMBOL(dmu_buf_fill_done); EXPORT_SYMBOL(dmu_buf_rele); EXPORT_SYMBOL(dbuf_assign_arcbuf); EXPORT_SYMBOL(dbuf_prefetch); EXPORT_SYMBOL(dbuf_hold_impl); EXPORT_SYMBOL(dbuf_hold); EXPORT_SYMBOL(dbuf_hold_level); EXPORT_SYMBOL(dbuf_create_bonus); EXPORT_SYMBOL(dbuf_spill_set_blksz); EXPORT_SYMBOL(dbuf_rm_spill); EXPORT_SYMBOL(dbuf_add_ref); EXPORT_SYMBOL(dbuf_rele); EXPORT_SYMBOL(dbuf_rele_and_unlock); EXPORT_SYMBOL(dbuf_refcount); EXPORT_SYMBOL(dbuf_sync_list); EXPORT_SYMBOL(dmu_buf_set_user); EXPORT_SYMBOL(dmu_buf_set_user_ie); EXPORT_SYMBOL(dmu_buf_get_user); EXPORT_SYMBOL(dmu_buf_get_blkptr); ZFS_MODULE_PARAM(zfs_dbuf_cache, dbuf_cache_, max_bytes, ULONG, ZMOD_RW, "Maximum size in bytes of the dbuf cache."); ZFS_MODULE_PARAM(zfs_dbuf_cache, dbuf_cache_, hiwater_pct, UINT, ZMOD_RW, "Percentage over dbuf_cache_max_bytes for direct dbuf eviction."); ZFS_MODULE_PARAM(zfs_dbuf_cache, dbuf_cache_, lowater_pct, UINT, ZMOD_RW, "Percentage below dbuf_cache_max_bytes when dbuf eviction stops."); ZFS_MODULE_PARAM(zfs_dbuf, dbuf_, metadata_cache_max_bytes, ULONG, ZMOD_RW, "Maximum size in bytes of dbuf metadata cache."); ZFS_MODULE_PARAM(zfs_dbuf, dbuf_, cache_shift, INT, ZMOD_RW, "Set size of dbuf cache to log2 fraction of arc size."); ZFS_MODULE_PARAM(zfs_dbuf, dbuf_, metadata_cache_shift, INT, ZMOD_RW, "Set size of dbuf metadata cache to log2 fraction of arc size."); + +ZFS_MODULE_PARAM(zfs_dbuf, dbuf_, mutex_cache_shift, UINT, ZMOD_RD, + "Set size of dbuf cache mutex array as log2 shift.");