diff --git a/include/sys/dsl_scan.h b/include/sys/dsl_scan.h index 44cca14794e0..bcb98af40067 100644 --- a/include/sys/dsl_scan.h +++ b/include/sys/dsl_scan.h @@ -1,229 +1,229 @@ // SPDX-License-Identifier: CDDL-1.0 /* * 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) 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2012, 2017 by Delphix. All rights reserved. * Copyright (c) 2017, 2019, Datto Inc. All rights reserved. */ #ifndef _SYS_DSL_SCAN_H #define _SYS_DSL_SCAN_H #include #include #include #include #include #ifdef __cplusplus extern "C" { #endif struct objset; struct dsl_dir; struct dsl_dataset; struct dsl_pool; struct dmu_tx; extern int zfs_scan_suspend_progress; /* * All members of this structure must be uint64_t, for byteswap * purposes. */ typedef struct dsl_scan_phys { uint64_t scn_func; /* pool_scan_func_t */ uint64_t scn_state; /* dsl_scan_state_t */ uint64_t scn_queue_obj; uint64_t scn_min_txg; uint64_t scn_max_txg; uint64_t scn_cur_min_txg; uint64_t scn_cur_max_txg; uint64_t scn_start_time; uint64_t scn_end_time; uint64_t scn_to_examine; /* total bytes to be scanned */ uint64_t scn_examined; /* bytes scanned so far */ uint64_t scn_skipped; /* bytes skipped by scanner */ uint64_t scn_processed; uint64_t scn_errors; /* scan I/O error count */ uint64_t scn_ddt_class_max; ddt_bookmark_t scn_ddt_bookmark; zbookmark_phys_t scn_bookmark; uint64_t scn_flags; /* dsl_scan_flags_t */ } dsl_scan_phys_t; #define SCAN_PHYS_NUMINTS (sizeof (dsl_scan_phys_t) / sizeof (uint64_t)) typedef enum dsl_scan_flags { DSF_VISIT_DS_AGAIN = 1<<0, DSF_SCRUB_PAUSED = 1<<1, } dsl_scan_flags_t; #define DSL_SCAN_FLAGS_MASK (DSF_VISIT_DS_AGAIN) typedef struct dsl_errorscrub_phys { uint64_t dep_func; /* pool_scan_func_t */ uint64_t dep_state; /* dsl_scan_state_t */ uint64_t dep_cursor; /* serialized zap cursor for tracing progress */ uint64_t dep_start_time; /* error scrub start time, unix timestamp */ uint64_t dep_end_time; /* error scrub end time, unix timestamp */ uint64_t dep_to_examine; /* total error blocks to be scrubbed */ uint64_t dep_examined; /* blocks scrubbed so far */ uint64_t dep_errors; /* error scrub I/O error count */ uint64_t dep_paused_flags; /* flag for paused */ } dsl_errorscrub_phys_t; #define ERRORSCRUB_PHYS_NUMINTS (sizeof (dsl_errorscrub_phys_t) \ / sizeof (uint64_t)) /* * Every pool will have one dsl_scan_t and this structure will contain * in-memory information about the scan and a pointer to the on-disk * representation (i.e. dsl_scan_phys_t). Most of the state of the scan * is contained on-disk to allow the scan to resume in the event of a reboot * or panic. This structure maintains information about the behavior of a * running scan, some caching information, and how it should traverse the pool. * * The following members of this structure direct the behavior of the scan: * * scn_suspending - a scan that cannot be completed in a single txg or * has exceeded its allotted time will need to suspend. * When this flag is set the scanner will stop traversing * the pool and write out the current state to disk. * * scn_restart_txg - directs the scanner to either restart or start a * a scan at the specified txg value. * * scn_done_txg - when a scan completes its traversal it will set * the completion txg to the next txg. This is necessary * to ensure that any blocks that were freed during * the scan but have not yet been processed (i.e deferred * frees) are accounted for. * * This structure also maintains information about deferred frees which are * a special kind of traversal. Deferred free can exist in either a bptree or * a bpobj structure. The scn_is_bptree flag will indicate the type of * deferred free that is in progress. If the deferred free is part of an * asynchronous destroy then the scn_async_destroying flag will be set. */ typedef struct dsl_scan { struct dsl_pool *scn_dp; uint64_t scn_restart_txg; uint64_t scn_done_txg; uint64_t scn_sync_start_time; uint64_t scn_issued_before_pass; /* for freeing blocks */ boolean_t scn_is_bptree; boolean_t scn_async_destroying; boolean_t scn_async_stalled; uint64_t scn_async_block_min_time_ms; /* flags and stats for controlling scan state */ boolean_t scn_is_sorted; /* doing sequential scan */ boolean_t scn_clearing; /* scan is issuing sequential extents */ boolean_t scn_checkpointing; /* scan is issuing all queued extents */ boolean_t scn_suspending; /* scan is suspending until next txg */ uint64_t scn_last_checkpoint; /* time of last checkpoint */ /* members for thread synchronization */ zio_t *scn_zio_root; /* root zio for waiting on IO */ taskq_t *scn_taskq; /* task queue for issuing extents */ /* for controlling scan prefetch, protected by spa_scrub_lock */ boolean_t scn_prefetch_stop; /* prefetch should stop */ zbookmark_phys_t scn_prefetch_bookmark; /* prefetch start bookmark */ avl_tree_t scn_prefetch_queue; /* priority queue of prefetch IOs */ uint64_t scn_maxinflight_bytes; /* max bytes in flight for pool */ /* per txg statistics */ uint64_t scn_visited_this_txg; /* total bps visited this txg */ - uint64_t scn_dedup_frees_this_txg; /* dedup bps freed this txg */ + uint64_t scn_async_frees_this_txg; /* async frees (dedup/clone/gang) */ uint64_t scn_holes_this_txg; uint64_t scn_lt_min_this_txg; uint64_t scn_gt_max_this_txg; uint64_t scn_ddt_contained_this_txg; uint64_t scn_objsets_visited_this_txg; uint64_t scn_avg_seg_size_this_txg; uint64_t scn_segs_this_txg; uint64_t scn_avg_zio_size_this_txg; uint64_t scn_zios_this_txg; /* zap cursor for tracing error scrub progress */ zap_cursor_t errorscrub_cursor; /* members needed for syncing scan status to disk */ dsl_scan_phys_t scn_phys; /* on disk representation of scan */ dsl_scan_phys_t scn_phys_cached; avl_tree_t scn_queue; /* queue of datasets to scan */ kmutex_t scn_queue_lock; /* serializes scn_queue inserts */ uint64_t scn_queues_pending; /* outstanding data to issue */ /* members needed for syncing error scrub status to disk */ dsl_errorscrub_phys_t errorscrub_phys; } dsl_scan_t; typedef struct { pool_scan_func_t func; uint64_t txgstart; uint64_t txgend; } setup_sync_arg_t; typedef struct dsl_scan_io_queue dsl_scan_io_queue_t; void scan_init(void); void scan_fini(void); int dsl_scan_init(struct dsl_pool *dp, uint64_t txg); int dsl_scan_setup_check(void *, dmu_tx_t *); void dsl_scan_setup_sync(void *, dmu_tx_t *); void dsl_scan_fini(struct dsl_pool *dp); void dsl_scan_sync(struct dsl_pool *, dmu_tx_t *); int dsl_scan_cancel(struct dsl_pool *); int dsl_scan(struct dsl_pool *, pool_scan_func_t, uint64_t starttxg, uint64_t txgend); void dsl_scan_assess_vdev(struct dsl_pool *dp, vdev_t *vd); boolean_t dsl_scan_scrubbing(const struct dsl_pool *dp); boolean_t dsl_errorscrubbing(const struct dsl_pool *dp); boolean_t dsl_errorscrub_active(dsl_scan_t *scn); void dsl_scan_restart_resilver(struct dsl_pool *, uint64_t txg); int dsl_scrub_set_pause_resume(const struct dsl_pool *dp, pool_scrub_cmd_t cmd); void dsl_errorscrub_sync(struct dsl_pool *, dmu_tx_t *); boolean_t dsl_scan_resilvering(struct dsl_pool *dp); boolean_t dsl_scan_resilver_scheduled(struct dsl_pool *dp); boolean_t dsl_dataset_unstable(struct dsl_dataset *ds); void dsl_scan_ddt_entry(dsl_scan_t *scn, enum zio_checksum checksum, ddt_t *ddt, ddt_lightweight_entry_t *ddlwe, dmu_tx_t *tx); void dsl_scan_ds_destroyed(struct dsl_dataset *ds, struct dmu_tx *tx); void dsl_scan_ds_snapshotted(struct dsl_dataset *ds, struct dmu_tx *tx); void dsl_scan_ds_clone_swapped(struct dsl_dataset *ds1, struct dsl_dataset *ds2, struct dmu_tx *tx); boolean_t dsl_scan_active(dsl_scan_t *scn); boolean_t dsl_scan_is_paused_scrub(const dsl_scan_t *scn); boolean_t dsl_errorscrub_is_paused(const dsl_scan_t *scn); void dsl_scan_freed(spa_t *spa, const blkptr_t *bp); void dsl_scan_io_queue_destroy(dsl_scan_io_queue_t *queue); void dsl_scan_io_queue_vdev_xfer(vdev_t *svd, vdev_t *tvd); #ifdef __cplusplus } #endif #endif /* _SYS_DSL_SCAN_H */ diff --git a/man/man4/zfs.4 b/man/man4/zfs.4 index 6f2a23a45058..50803101f1af 100644 --- a/man/man4/zfs.4 +++ b/man/man4/zfs.4 @@ -1,3052 +1,3057 @@ .\" SPDX-License-Identifier: CDDL-1.0 .\" .\" Copyright (c) 2013 by Turbo Fredriksson . All rights reserved. .\" Copyright (c) 2019, 2021 by Delphix. All rights reserved. .\" Copyright (c) 2019 Datto Inc. .\" Copyright (c) 2023, 2024, 2025, Klara, 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 September 15, 2025 .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 UINT64_MAX Ns B Pq u64 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 UINT64_MAX Ns B Pq u64 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 uint 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 uint 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 uint 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_ddt_copies Ns = Ns Sy 3 Pq uint Controls the number of copies stored for DeDup Table .Pq DDT objects. Reducing the number of copies to 1 from the previous default of 3 can reduce the write inflation caused by deduplication. This assumes redundancy for this data is provided by the vdev layer. If the DDT is damaged, space may be leaked .Pq not freed when the DDT can not report the correct reference count. . .It Sy dmu_prefetch_max Ns = Ns Sy 134217728 Ns B Po 128 MiB Pc Pq uint 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 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 u64 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 u64 Seconds between L2ARC writing. . .It Sy l2arc_headroom Ns = Ns Sy 8 Pq u64 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 u64 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 Ns | Ns 2 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 0, meaning both MRU and MFU data and metadata are cached. When turning off this feature (setting it to 0), 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 Setting it to 1 means to L2 cache only MFU data and metadata. .Pp Setting it to 2 means to L2 cache all metadata (MRU+MFU) but only MFU data (i.e. MRU data are not cached). This can be the right setting to cache as much metadata as possible even when having high data turnover. .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 uint 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 u64 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 33554432 Ns B Po 32 MiB Pc Pq u64 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 33554432 Ns B Po 32 MiB Pc Pq u64 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 u64 Minimum 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 2097152 Ns B Po 2 MiB Pc Pq u64 Metaslab group's per child vdev allocation 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 child of a top-level vdev 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 groups biasing based on their over- or under-utilization relative to the metaslab class average. If disabled, each metaslab group will receive allocations proportional to its capacity. . .It Sy metaslab_perf_bias Ns = Ns Sy 1 Ns | Ns 0 Ns | Ns 2 Pq int Controls metaslab groups biasing based on their write performance. Setting to 0 makes all metaslab groups receive fixed amounts of allocations. Setting to 2 allows faster metaslab groups to allocate more. Setting to 1 equals to 2 if the pool is write-bound or 0 otherwise. That is, if the pool is limited by write throughput, then allocate more from faster metaslab groups, but if not, try to evenly distribute the allocations. . .It Sy metaslab_force_ganging Ns = Ns Sy 16777217 Ns B Po 16 MiB + 1 B Pc Pq u64 Make some blocks above a certain size be gang blocks. This option is used by the test suite to facilitate testing. . .It Sy metaslab_force_ganging_pct Ns = Ns Sy 3 Ns % Pq uint For blocks that could be forced to be a gang block (due to .Sy metaslab_force_ganging ) , force this many of them to be gang blocks. . .It Sy brt_zap_prefetch Ns = Ns Sy 1 Ns | Ns 0 Pq int Controls prefetching BRT records for blocks which are going to be cloned. . .It Sy brt_zap_default_bs Ns = Ns Sy 13 Po 8 KiB Pc Pq int Default BRT ZAP data block size as a power of 2. Note that changing this after creating a BRT on the pool will not affect existing BRTs, only newly created ones. . .It Sy brt_zap_default_ibs Ns = Ns Sy 13 Po 8 KiB Pc Pq int Default BRT ZAP indirect block size as a power of 2. Note that changing this after creating a BRT on the pool will not affect existing BRTs, only newly created ones. . .It Sy ddt_zap_default_bs Ns = Ns Sy 15 Po 32 KiB Pc Pq int Default DDT ZAP data block size as a power of 2. Note that changing this after creating a DDT on the pool will not affect existing DDTs, only newly created ones. . .It Sy ddt_zap_default_ibs Ns = Ns Sy 15 Po 32 KiB Pc Pq int Default DDT ZAP indirect block size as a power of 2. Note that changing this after creating a DDT on the pool will not affect existing DDTs, only newly created ones. . .It Sy zfs_default_bs Ns = Ns Sy 9 Po 512 B Pc Pq int Default dnode block size as a power of 2. . .It Sy zfs_default_ibs Ns = Ns Sy 17 Po 128 KiB Pc Pq int Default dnode indirect block size as a power of 2. . .It Sy zfs_dio_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int Enable Direct I/O. If this setting is 0, then all I/O requests will be directed through the ARC acting as though the dataset property .Sy direct was set to .Sy disabled . . .It Sy zfs_dio_strict Ns = Ns Sy 0 Ns | Ns 1 Pq int Strictly enforce alignment for Direct I/O requests, returning .Sy EINVAL if not page-aligned instead of silently falling back to uncached I/O. . .It Sy zfs_history_output_max Ns = Ns Sy 1048576 Ns B Po 1 MiB Pc Pq u64 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 uint 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 u64 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 uint 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 uint 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 uint 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 uint Default lower limit for metaslab size. . .It Sy zfs_vdev_max_ms_shift Ns = Ns Sy 34 Po 16 GiB Pc Pq uint Default upper limit for metaslab size. . .It Sy zfs_vdev_max_auto_ashift Ns = Ns Sy 14 Pq uint 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_direct_write_verify Ns = Ns Sy Linux 1 | FreeBSD 0 Pq uint If non-zero, then a Direct I/O write's checksum will be verified every time the write is issued and before it is committed to the block pointer. In the event the checksum is not valid then the I/O operation will return EIO. This module parameter can be used to detect if the contents of the users buffer have changed in the process of doing a Direct I/O write. It can also help to identify if reported checksum errors are tied to Direct I/O writes. Each verify error causes a .Sy dio_verify_wr zevent. Direct Write I/O checksum verify errors can be seen with .Nm zpool Cm status Fl d . The default value for this is 1 on Linux, but is 0 for .Fx because user pages can be placed under write protection in .Fx before the Direct I/O write is issued. . .It Sy zfs_vdev_min_auto_ashift Ns = Ns Sy ASHIFT_MIN Po 9 Pc Pq uint Minimum ashift used when creating new top-level vdevs. . .It Sy zfs_vdev_min_ms_count Ns = Ns Sy 16 Pq uint 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 uint 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_preload_limit Ns = Ns Sy 10 Pq uint Maximum number of metaslabs per group to preload . .It Sy metaslab_preload_pct Ns = Ns Sy 50 Pq uint Percentage of CPUs to run a metaslab preload taskq . .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 uint 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 uint 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 uint Maximum reference holders being tracked when reference_tracking_enable is active. .It Sy raidz_expand_max_copy_bytes Ns = Ns Sy 160MB Pq ulong Max amount of memory to use for RAID-Z expansion I/O. This limits how much I/O can be outstanding at once. . .It Sy raidz_expand_max_reflow_bytes Ns = Ns Sy 0 Pq ulong For testing, pause RAID-Z expansion when reflow amount reaches this value. . .It Sy raidz_io_aggregate_rows Ns = Ns Sy 4 Pq ulong For expanded RAID-Z, aggregate reads that have more rows than this. . .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 uint 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 uint 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_num_allocators Ns = Ns Sy 4 Pq int Determines the number of block allocators to use per spa instance. Capped by the number of actual CPUs in the system via .Sy spa_cpus_per_allocator . .Pp Note that setting this value too high could result in performance degradation and/or excess fragmentation. Set value only applies to pools imported/created after that. . .It Sy spa_cpus_per_allocator Ns = Ns Sy 4 Pq int Determines the minimum number of CPUs in a system for block allocator per spa instance. Set value only applies to pools imported/created after that. . .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_read_sit_out_secs Ns = Ns Sy 600 Ns s Po 10 min Pc Pq ulong When a slow disk outlier is detected it is placed in a sit out state. While sitting out the disk will not participate in normal reads, instead its data will be reconstructed as needed from parity. Scrub operations will always read from a disk, even if it's sitting out. A number of disks in a RAID-Z or dRAID vdev may sit out at the same time, up to the number of parity devices. Writes will still be issued to a disk which is sitting out to maintain full redundancy. Defaults to 600 seconds and a value of zero disables disk sit-outs in general, including slow disk outlier detection. . .It Sy vdev_raidz_outlier_check_interval_ms Ns = Ns Sy 1000 Ns ms Po 1 sec Pc Pq ulong How often each RAID-Z and dRAID vdev will check for slow disk outliers. Increasing this interval will reduce the sensitivity of detection (since all I/Os since the last check are included in the statistics), but will slow the response to a disk developing a problem. Defaults to once per second; setting extremely small values may cause negative performance effects. . .It Sy vdev_raidz_outlier_insensitivity Ns = Ns Sy 50 Pq uint When performing slow outlier checks for RAID-Z and dRAID vdevs, this value is used to determine how far out an outlier must be before it counts as an event worth consdering. This is phrased as "insensitivity" because larger values result in fewer detections. Smaller values will result in more aggressive sitting out of disks that may have problems, but may significantly increase the rate of spurious sit-outs. .Pp To provide a more technical definition of this parameter, this is the multiple of the inter-quartile range (IQR) that is being used in a Tukey's Fence detection algorithm. This is much higher than a normal Tukey's Fence k-value, because the distribution under consideration is probably an extreme-value distribution, rather than a more typical Gaussian distribution. . .It Sy vdev_removal_max_span Ns = Ns Sy 32768 Ns B Po 32 KiB Pc Pq uint 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 u64 Logical ashift for file-based devices. . .It Sy vdev_file_physical_ashift Ns = Ns Sy 9 Po 512 B Pc Pq u64 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 zap_micro_max_size Ns = Ns Sy 131072 Ns B Po 128 KiB Pc Pq int Maximum micro ZAP size. A "micro" ZAP is upgraded to a "fat" ZAP once it grows beyond the specified size. Sizes higher than 128KiB will be clamped to 128KiB unless the .Sy large_microzap feature is enabled. . .It Sy zap_shrink_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int If set, adjacent empty ZAP blocks will be collapsed, reducing disk space. . .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_reorder Ns = Ns Sy 16777216 Ns B Po 16 MiB Pc Pq uint Requests within this byte distance from the current prefetch stream position are considered parts of the stream, reordered due to parallel processing. Such requests do not advance the stream position immediately unless .Sy zfetch_hole_shift fill threshold is reached, but saved to fill holes in the stream later. . .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 u64 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. .It Sy zfs_arc_dnode_limit_percent Ns = Ns Sy 10 Ns % Pq u64 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 u64 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 uint 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 uint 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 uint 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_evict_batches_limit Ns = Ns Sy 5 Pq uint Number of .Sy zfs_arc_evict_batch_limit batches to process per parallel eviction task under heavy load to reduce number of context switches. . .It Sy zfs_arc_evict_threads Ns = Ns Sy 0 Pq int Sets the number of ARC eviction threads to be used. .Pp If set greater than 0, ZFS will dedicate up to that many threads to ARC eviction. Each thread will process one sub-list at a time, until the eviction target is reached or all sub-lists have been processed. When set to 0, ZFS will compute a reasonable number of eviction threads based on the number of CPUs. .TS box; lb l l . CPUs Threads _ 1-4 1 5-8 2 9-15 3 16-31 4 32-63 6 64-95 8 96-127 9 128-160 11 160-191 12 192-223 13 224-255 14 256+ 16 .TE .Pp More threads may improve the responsiveness of ZFS to memory pressure. This can be important for performance when eviction from the ARC becomes a bottleneck for reads and writes. .Pp This parameter can only be set at module load time. . .It Sy zfs_arc_grow_retry Ns = Ns Sy 0 Ns s Pq uint 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 u64 Max size of ARC in bytes. If .Sy 0 , then the max size of ARC is determined by the amount of system memory installed. 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_balance Ns = Ns Sy 500 Pq uint Balance between metadata and data on ghost hits. Values above 100 increase metadata caching by proportionally reducing effect of ghost data hits on target data/metadata rate. . .It Sy zfs_arc_min Ns = Ns Sy 0 Ns B Pq u64 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 uint 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 uint 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 u64 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 uint 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_shrink_shift Ns = Ns Sy 0 Pq uint 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_ACTIVE_FILE + .Sy NR_INACTIVE_FILE ) , where that percent may exceed .Sy 100 . This only operates during memory pressure/reclaim. . .It Sy zfs_arc_shrinker_limit Ns = Ns Sy 0 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. To reduce OOM risk, this limit is applied for kswapd reclaims only. .Pp For example a value 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_shrinker_seeks Ns = Ns Sy 2 Pq int Relative cost of ARC eviction on Linux, AKA number of seeks needed to restore evicted page. Bigger values make ARC more precious and evictions smaller, comparing to other kernel subsystems. Value of 4 means parity with page cache. . .It Sy zfs_arc_sys_free Ns = Ns Sy 0 Ns B Pq u64 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_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 10 Ns % Pq uint 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 uint 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 u64 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 u64 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 uint 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_checktime_ms Ns = Ns Sy 60000 Ns ms Po 1 min Pc Pq u64 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_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_events_per_second Ns = Ns Sy 1 Ns /s Pq int Rate limit deadman zevents (which report hung I/O operations) to this many per second. . .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_synctime_ms Ns = Ns Sy 600000 Ns ms Po 10 min Pc Pq u64 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 u64 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_dedup_log_flush_min_time_ms Ns = Ns Sy 1000 Ns Pq uint Minimum time to spend on dedup log flush each transaction. .Pp At least this long will be spent flushing dedup log entries each transaction, up to .Sy zfs_txg_timeout . This occurs even if doing so would delay the transaction, that is, other IO completes under this time. . .It Sy zfs_dedup_log_flush_entries_min Ns = Ns Sy 100 Ns Pq uint Flush at least this many entries each transaction. .Pp OpenZFS will flush a fraction of the log every TXG, to keep the size proportional to the ingest rate (see .Sy zfs_dedup_log_flush_txgs ) . This sets the minimum for that estimate, which prevents the backlog from completely draining if the ingest rate falls. Raising it can force OpenZFS to flush more aggressively, reducing the backlog to zero more quickly, but can make it less able to back off if log flushing would compete with other IO too much. . .It Sy zfs_dedup_log_flush_entries_max Ns = Ns Sy UINT_MAX Ns Pq uint Flush at most this many entries each transaction. .Pp Mostly used for debugging purposes. .It Sy zfs_dedup_log_flush_txgs Ns = Ns Sy 100 Ns Pq uint Target number of TXGs to process the whole dedup log. .Pp Every TXG, OpenZFS will process the inverse of this number times the size of the DDT backlog. This will keep the backlog at a size roughly equal to the ingest rate times this value. This offers a balance between a more efficient DDT log, with better aggregation, and shorter import times, which increase as the size of the DDT log increases. Increasing this value will result in a more efficient DDT log, but longer import times. .It Sy zfs_dedup_log_cap Ns = Ns Sy UINT_MAX Ns Pq uint Soft cap for the size of the current dedup log. .Pp If the log is larger than this size, we increase the aggressiveness of the flushing to try to bring it back down to the soft cap. Setting it will reduce import times, but will reduce the efficiency of the DDT log, increasing the expected number of IOs required to flush the same amount of data. .It Sy zfs_dedup_log_hard_cap Ns = Ns Sy 0 Ns | Ns 1 Pq uint Whether to treat the log cap as a firm cap or not. .Pp When set to 0 (the default), the .Sy zfs_dedup_log_cap will increase the maximum number of log entries we flush in a given txg. This will bring the backlog size down towards the cap, but not at the expense of making TXG syncs take longer. If this is set to 1, the cap acts more like a hard cap than a soft cap; it will also increase the minimum number of log entries we flush per TXG. Enabling it will reduce worst-case import times, at the cost of increased TXG sync times. .It Sy zfs_dedup_log_flush_flow_rate_txgs Ns = Ns Sy 10 Ns Pq uint Number of transactions to use to compute the flow rate. .Pp OpenZFS will estimate number of entries changed (ingest rate), number of entries flushed (flush rate) and time spent flushing (flush time rate) and combining these into an overall "flow rate". It will use an exponential weighted moving average over some number of recent transactions to compute these rates. This sets the number of transactions to compute these averages over. Setting it higher can help to smooth out the flow rate in the face of spiky workloads, but will take longer for the flow rate to adjust to a sustained change in the ingress rate. . .It Sy zfs_dedup_log_txg_max Ns = Ns Sy 8 Ns Pq uint Max transactions to before starting to flush dedup logs. .Pp OpenZFS maintains two dedup logs, one receiving new changes, one flushing. If there is nothing to flush, it will accumulate changes for no more than this many transactions before switching the logs and starting to flush entries out. . .It Sy zfs_dedup_log_mem_max Ns = Ns Sy 0 Ns Pq u64 Max memory to use for dedup logs. .Pp OpenZFS will spend no more than this much memory on maintaining the in-memory dedup log. Flushing will begin when around half this amount is being spent on logs. The default value of .Sy 0 will cause it to be set by .Sy zfs_dedup_log_mem_max_percent instead. . .It Sy zfs_dedup_log_mem_max_percent Ns = Ns Sy 1 Ns % Pq uint Max memory to use for dedup logs, as a percentage of total memory. .Pp If .Sy zfs_dedup_log_mem_max is not set, it will be initialized as a percentage of the total memory in the system. . .It Sy zfs_delay_min_dirty_percent Ns = Ns Sy 60 Ns % Pq uint 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_dio_write_verify_events_per_second Ns = Ns Sy 20 Ns /s Pq uint Rate limit Direct I/O write verify events to this many per second. . .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 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 u64 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 u64 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 u64 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 u64 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 u64 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 u64 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. This only applies on Linux. . .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 min(physical_ram/4, 4GiB) , or .Sy min(physical_ram/4, 1GiB) for 32-bit systems. . .It Sy zfs_dirty_data_max_max_percent Ns = Ns Sy 25 Ns % Pq uint 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 uint 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 uint 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_bclone_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int Enables access to the block cloning feature. If this setting is 0, then even if feature@block_cloning is enabled, using functions and system calls that attempt to clone blocks will act as though the feature is disabled. . .It Sy zfs_bclone_wait_dirty Ns = Ns Sy 1 Ns | Ns 0 Pq int When set to 1 the FICLONE and FICLONERANGE ioctls will wait for any dirty data to be written to disk before proceeding. This ensures that the clone operation reliably succeeds, even if a file is modified and then immediately cloned. Note that for small files this may be slower than simply copying the file. When set to 0 the clone operation will immediately fail if it encounters any dirty blocks. By default waiting is enabled. . .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 UINT64_MAX Po unlimited Pc Pq u64 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 u64 -Maximum number of dedup blocks freed in a single TXG. +.It Sy zfs_max_async_dedup_frees Ns = Ns Sy 250000 Pq u64 +Maximum number of dedup, clone or gang blocks freed in a single TXG. +These frees may require additional I/O, making them more expensive. +. +.It Sy zfs_async_free_zio_wait_interval Ns = Ns Sy 2000 Pq u64 +After freeing this many dedup, clone or gang blocks wait for all pending +I/Os to complete before continuing. . .It Sy zfs_vdev_async_read_max_active Ns = Ns Sy 3 Pq uint 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 uint 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 uint 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 uint 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 10 Pq uint 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 uint 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 uint 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 uint 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 uint 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_open_timeout_ms Ns = Ns Sy 1000 Pq uint Timeout value to wait before determining a device is missing during import. This is helpful for transient missing paths due to links being briefly removed and recreated in response to udev events. . .It Sy zfs_vdev_rebuild_max_active Ns = Ns Sy 3 Pq uint 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 uint 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 uint 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 uint 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 uint 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 uint 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 uint 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 uint 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 uint 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 uint 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 uint 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 uint 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 uint 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 uint 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_failfast_mask Ns = Ns Sy 1 Pq uint Defines if the driver should retire on a given error type. The following options may be bitwise-ored together: .TS box; lbz r l l . Value Name Description _ 1 Device No driver retries on device errors 2 Transport No driver retries on transport errors. 4 Driver No driver retries on driver errors. .TE . .It Sy zfs_vdev_disk_max_segs Ns = Ns Sy 0 Pq uint Maximum number of segments to add to a BIO (min 4). If this is higher than the maximum allowed by the device queue or the kernel itself, it will be clamped. Setting it to zero will cause the kernel's ideal size to be used. This parameter only applies on Linux. . .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_snapshot_no_setuid Ns = Ns Sy 0 Ns | Ns 1 Pq int Whether to disable .Em setuid/setgid support for snapshot mounts triggered by access to the .Sy .zfs/snapshot directory by setting the .Em nosuid mount option. . .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 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. 8192 ZFS_DEBUG_METASLAB_ALLOC Enable debugging messages when allocations fail. 16384 ZFS_DEBUG_BRT Enable BRT-related debugging messages. 32768 ZFS_DEBUG_RAIDZ_RECONSTRUCT Enabled debugging messages for raidz reconstruction. 65536 ZFS_DEBUG_DDT Enable DDT-related debugging messages. .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 cumulative: .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 uint +.It Sy zfs_free_min_time_ms Ns = Ns Sy 500 Ns ms Po 1s Pc Pq uint 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 uint 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 s64 Largest write size to store the data directly into the ZIL if .Sy logbias Ns = Ns Sy latency . Larger writes may be written indirectly similar to .Sy logbias Ns = Ns Sy throughput . In presence of SLOG this parameter is ignored, as if it was set to infinity, storing all written data into ZIL to not depend on regular vdev latency. . .It Sy zil_special_is_slog Ns = Ns Sy 1 Ns | Ns 0 Pq int When enabled, and written blocks go to normal vdevs, treat present special vdevs as SLOGs. Blocks that go to the special vdevs are still written indirectly, as with .Sy logbias Ns = Ns Sy throughput . This parameter is ignored if an SLOG is present. . .It Sy zfs_import_defer_txgs Ns = Ns Sy 5 Pq uint Number of transaction groups to wait after pool import before starting background work such as asynchronous block freeing .Pq from snapshots, clones, and deduplication and scrub or resilver operations. This allows the pool import and filesystem mounting to complete more quickly without interference from background activities. The default value of 5 transaction groups typically provides sufficient time for import and mount operations to complete on most systems. . .It Sy zfs_initialize_value Ns = Ns Sy 16045690984833335022 Po 0xDEADBEEFDEADBEEE Pc Pq u64 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 u64 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 u64 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 u64 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 u64 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 u64 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 u64 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 uint 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. .Pp Note that it is still limited by default to .Ar 1 MiB on x86_32, because Linux's 3/1 memory split doesn't leave much room for 16M chunks. . .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 u64 Minimum number of metaslabs to flush per dirty TXG. . .It Sy zfs_metaslab_fragmentation_threshold Ns = Ns Sy 77 Ns % Pq uint 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 uint 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 uint 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 uint 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 u64 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 uint 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 30 Ns % Pq u64 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 33554432 Ns B Po 32 MiB Pc Pq u64 Bytes to read per chunk. . .It Sy zfs_read_history Ns = Ns Sy 0 Pq uint 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 u64 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 67108864 Ns B Po 64 MiB Pc Pq u64 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 canceled. 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 uint 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 uint 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_defer_percent Ns = Ns Sy 10 Ns % Pq uint If the ongoing resilver progress is below this threshold, a new resilver will restart from scratch instead of being deferred after the current one finishes, even if the .Sy resilver_defer feature is enabled. . .It Sy zfs_resilver_min_time_ms Ns = Ns Sy 3000 Ns ms Po 3 s Pc Pq uint 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_after_expand Ns = Ns Sy 1 Ns | Ns 0 Pq int Automatically start a pool scrub after a RAIDZ expansion completes in order to verify the checksums of all blocks which have been copied during the expansion. This is enabled by default and strongly recommended. . .It Sy zfs_scrub_min_time_ms Ns = Ns Sy 1000 Ns ms Po 1 s Pc Pq uint 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_scrub_error_blocks_per_txg Ns = Ns Sy 4096 Pq uint Error blocks to be scrubbed in one txg. . .It Sy zfs_scan_checkpoint_intval Ns = Ns Sy 7200 Ns s Po 2 hour Pc Pq uint 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 uint 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 uint 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 uint 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 uint 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_report_txgs Ns = Ns Sy 0 Ns | Ns 1 Pq uint When reporting resilver throughput and estimated completion time use the performance observed over roughly the last .Sy zfs_scan_report_txgs TXGs. When set to zero performance is calculated over the time between checkpoints. . .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 16777216 Ns B Po 16 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 uint 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 uint 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 uint 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 uint 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 uint 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 uint 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 uint 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 uint 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 uint 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_spa_note_txg_time Ns = Ns Sy 600 Pq uint This parameter defines, in seconds, how often the TXG time database will record a new TXG if it has changed. After the specified time interval has passed, and if the TXG number has changed, the new value is recorded in the database. These timestamps can later be used for more granular operations, such as scrubbing. . .It Sy zfs_spa_flush_txg_time Ns = Ns Sy 600 Pq uint This parameter defines, in seconds, how often the ZFS will flush the TXG time database to disk. It ensures that the data is actually written to persistent storage, which helps preserve the database in case of unexpected shutdown. The database is also automatically flushed during the export sequence. . .It Sy zfs_special_class_metadata_reserve_pct Ns = Ns Sy 25 Ns % Pq uint 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 uint 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 uint Rewrite new block pointers starting in this pass. . .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 100 Pq uint 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 uint Flush dirty data to disk at least every this many seconds (maximum TXG duration). . .It Sy zfs_vdev_aggregation_limit Ns = Ns Sy 1048576 Ns B Po 1 MiB Pc Pq uint 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 uint Max vdev I/O aggregation size for non-rotating media. . .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 uint 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 uint 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_zevent_len_max Ns = Ns Sy 512 Pq uint 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 uint 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_maxcopied Ns = Ns Sy 7680 Ns B Po 7.5 KiB Pc Pq uint This sets the maximum number of write bytes logged via WR_COPIED. It tunes a tradeoff between additional memory copy and possibly worse log space efficiency vs additional range lock/unlock. . .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 67108864 Ns B Po 64 MiB Pc Pq u64 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 uint Usually, one metaslab from each normal and special 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 uint 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 uint 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 based on device performance. . .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 zfs_delete_inode Ns = Ns Sy 0 Ns | Ns 1 Pq int Sets whether the kernel should free an inode structure when the last reference is released, or cache it in memory. Intended for testing/debugging. .Pp A live inode structure "pins" versious internal OpenZFS structures in memory, which can result in large amounts of "unusable" memory on systems with lots of infrequently-accessed files, until the kernel's memory pressure mechanism asks OpenZFS to release them. .Pp The default value of .Sy 0 always caches inodes that appear to still exist on disk. Setting it to .Sy 1 will immediately release unused inodes and their associated memory back to the dbuf cache or the ARC for reuse, but may reduce performance if inodes are frequently evicted and reloaded. .Pp This parameter is only available on Linux. . .It Sy zfs_delete_dentry Ns = Ns Sy 0 Ns | Ns 1 Pq int Sets whether the kernel should free a dentry structure when it is no longer required, or hold it in the dentry cache. Intended for testing/debugging. . Since a dentry structure holds an inode reference, a cached dentry can "pin" an inode in memory indefinitely, along with associated OpenZFS structures (See .Sy zfs_delete_inode ) . .Pp The default value of .Sy 0 instructs the kernel to cache entries and their associated inodes when they are no longer directly referenced. They will be reclaimed as part of the kernel's normal cache management processes. Setting it to .Sy 1 will instruct the kernel to release directory entries and their inodes as soon as they are no longer referenced by the filesystem. .Pp This parameter is only available on Linux. . .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, encryption, checksum and parity 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. Set value only applies to pools imported/created after that. . .It Sy zio_taskq_batch_tpq Ns = Ns Sy 0 Pq uint Number of worker threads per taskq. Higher values improve I/O ordering and CPU utilization, while lower reduce lock contention. Set value only applies to pools imported/created after that. .Pp If .Sy 0 , generate a system-dependent value close to 6 threads per taskq. Set value only applies to pools imported/created after that. . .It Sy zio_taskq_write_tpq Ns = Ns Sy 16 Pq uint Determines the minimum number of threads per write issue taskq. Higher values improve CPU utilization on high throughput, while lower reduce taskq locks contention on high IOPS. Set value only applies to pools imported/created after that. . .It Sy zio_taskq_read Ns = Ns Sy fixed,1,8 null scale null Pq charp Set the queue and thread configuration for the IO read queues. This is an advanced debugging parameter. Don't change this unless you understand what it does. Each of the four values corresponds to the issue, issue high-priority, interrupt, and interrupt high-priority queues. Valid values are .Sy fixed,N,M (M queues with N threads each), .Sy scale[,MIN] (scale with CPUs, minimum MIN total threads), .Sy sync , and .Sy null . Set values only apply to pools imported/created after that. . .It Sy zio_taskq_write Ns = Ns Sy sync null scale null Pq charp Set the queue and thread configuration for the IO write queues. This is an advanced debugging parameter. Don't change this unless you understand what it does. Each of the four values corresponds to the issue, issue high-priority, interrupt, and interrupt high-priority queues. Valid values are .Sy fixed,N,M (M queues with N threads each), .Sy scale[,MIN] (scale with CPUs, minimum MIN total threads), .Sy sync , and .Sy null . Set values only apply to pools imported/created after that. . .It Sy zio_taskq_free Ns = Ns Sy scale,32 null null null Pq charp Set the queue and thread configuration for the IO free queues. This is an advanced debugging parameter. Don't change this unless you understand what it does. Each of the four values corresponds to the issue, issue high-priority, interrupt, and interrupt high-priority queues. Valid values are .Sy fixed,N,M (M queues with N threads each), .Sy scale[,MIN] (scale with CPUs, minimum MIN total threads), .Sy sync , and .Sy null . The default uses a minimum of 32 threads to improve parallelism for DDT and BRT metadata operations during frees. Set values only apply to pools imported/created after that. . .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 long 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_num_taskqs Ns = Ns Sy 0 Pq uint Number of zvol taskqs. If .Sy 0 (the default) then scaling is done internally to prefer 6 threads per taskq. This only applies on Linux. . .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 behavior 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 . .It Sy zvol_enforce_quotas Ns = Ns Sy 0 Ns | Ns 1 Pq uint Enable strict ZVOL quota enforcement. The strict quota enforcement may have a performance impact. .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 simultaneous operations also stabilizes the response time of operations from other queues, in particular synchronous ones. In broad strokes, the I/O scheduler will issue more concurrent operations from the async write queue as there is 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/dsl_scan.c b/module/zfs/dsl_scan.c index eb3f72bb0871..4a9e831dc434 100644 --- a/module/zfs/dsl_scan.c +++ b/module/zfs/dsl_scan.c @@ -1,5366 +1,5393 @@ // SPDX-License-Identifier: CDDL-1.0 /* * 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) 2008, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2011, 2021 by Delphix. All rights reserved. * Copyright 2016 Gary Mills * Copyright (c) 2017, 2019, Datto Inc. All rights reserved. * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved. * Copyright 2019 Joyent, Inc. */ #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 #include #include #ifdef _KERNEL #include #endif /* * Grand theory statement on scan queue sorting * * Scanning is implemented by recursively traversing all indirection levels * in an object and reading all blocks referenced from said objects. This * results in us approximately traversing the object from lowest logical * offset to the highest. For best performance, we would want the logical * blocks to be physically contiguous. However, this is frequently not the * case with pools given the allocation patterns of copy-on-write filesystems. * So instead, we put the I/Os into a reordering queue and issue them in a * way that will most benefit physical disks (LBA-order). * * Queue management: * * Ideally, we would want to scan all metadata and queue up all block I/O * prior to starting to issue it, because that allows us to do an optimal * sorting job. This can however consume large amounts of memory. Therefore * we continuously monitor the size of the queues and constrain them to 5% * (zfs_scan_mem_lim_fact) of physmem. If the queues grow larger than this * limit, we clear out a few of the largest extents at the head of the queues * to make room for more scanning. Hopefully, these extents will be fairly * large and contiguous, allowing us to approach sequential I/O throughput * even without a fully sorted tree. * * Metadata scanning takes place in dsl_scan_visit(), which is called from * dsl_scan_sync() every spa_sync(). If we have either fully scanned all * metadata on the pool, or we need to make room in memory because our * queues are too large, dsl_scan_visit() is postponed and * scan_io_queues_run() is called from dsl_scan_sync() instead. This implies * that metadata scanning and queued I/O issuing are mutually exclusive. This * allows us to provide maximum sequential I/O throughput for the majority of * I/O's issued since sequential I/O performance is significantly negatively * impacted if it is interleaved with random I/O. * * Implementation Notes * * One side effect of the queued scanning algorithm is that the scanning code * needs to be notified whenever a block is freed. This is needed to allow * the scanning code to remove these I/Os from the issuing queue. Additionally, * we do not attempt to queue gang blocks to be issued sequentially since this * is very hard to do and would have an extremely limited performance benefit. * Instead, we simply issue gang I/Os as soon as we find them using the legacy * algorithm. * * Backwards compatibility * * This new algorithm is backwards compatible with the legacy on-disk data * structures (and therefore does not require a new feature flag). * Periodically during scanning (see zfs_scan_checkpoint_intval), the scan * will stop scanning metadata (in logical order) and wait for all outstanding * sorted I/O to complete. Once this is done, we write out a checkpoint * bookmark, indicating that we have scanned everything logically before it. * If the pool is imported on a machine without the new sorting algorithm, * the scan simply resumes from the last checkpoint using the legacy algorithm. */ typedef int (scan_cb_t)(dsl_pool_t *, const blkptr_t *, const zbookmark_phys_t *); static scan_cb_t dsl_scan_scrub_cb; static int scan_ds_queue_compare(const void *a, const void *b); static int scan_prefetch_queue_compare(const void *a, const void *b); static void scan_ds_queue_clear(dsl_scan_t *scn); static void scan_ds_prefetch_queue_clear(dsl_scan_t *scn); static boolean_t scan_ds_queue_contains(dsl_scan_t *scn, uint64_t dsobj, uint64_t *txg); static void scan_ds_queue_insert(dsl_scan_t *scn, uint64_t dsobj, uint64_t txg); static void scan_ds_queue_remove(dsl_scan_t *scn, uint64_t dsobj); static void scan_ds_queue_sync(dsl_scan_t *scn, dmu_tx_t *tx); static uint64_t dsl_scan_count_data_disks(spa_t *spa); static void read_by_block_level(dsl_scan_t *scn, zbookmark_phys_t zb); extern uint_t zfs_vdev_async_write_active_min_dirty_percent; static int zfs_scan_blkstats = 0; /* * 'zpool status' uses bytes processed per pass to report throughput and * estimate time remaining. We define a pass to start when the scanning * phase completes for a sequential resilver. Optionally, this value * may be used to reset the pass statistics every N txgs to provide an * estimated completion time based on currently observed performance. */ static uint_t zfs_scan_report_txgs = 0; /* * By default zfs will check to ensure it is not over the hard memory * limit before each txg. If finer-grained control of this is needed * this value can be set to 1 to enable checking before scanning each * block. */ static int zfs_scan_strict_mem_lim = B_FALSE; /* * Maximum number of parallelly executed bytes per leaf vdev. We attempt * to strike a balance here between keeping the vdev queues full of I/Os * at all times and not overflowing the queues to cause long latency, * which would cause long txg sync times. No matter what, we will not * overload the drives with I/O, since that is protected by * zfs_vdev_scrub_max_active. */ static uint64_t zfs_scan_vdev_limit = 16 << 20; static uint_t zfs_scan_issue_strategy = 0; /* don't queue & sort zios, go direct */ static int zfs_scan_legacy = B_FALSE; static uint64_t zfs_scan_max_ext_gap = 2 << 20; /* in bytes */ /* * fill_weight is non-tunable at runtime, so we copy it at module init from * zfs_scan_fill_weight. Runtime adjustments to zfs_scan_fill_weight would * break queue sorting. */ static uint_t zfs_scan_fill_weight = 3; static uint64_t fill_weight; /* See dsl_scan_should_clear() for details on the memory limit tunables */ static const uint64_t zfs_scan_mem_lim_min = 16 << 20; /* bytes */ static const uint64_t zfs_scan_mem_lim_soft_max = 128 << 20; /* bytes */ /* fraction of physmem */ static uint_t zfs_scan_mem_lim_fact = 20; /* fraction of mem lim above */ static uint_t zfs_scan_mem_lim_soft_fact = 20; /* minimum milliseconds to scrub per txg */ static uint_t zfs_scrub_min_time_ms = 1000; /* minimum milliseconds to obsolete per txg */ static uint_t zfs_obsolete_min_time_ms = 500; /* minimum milliseconds to free per txg */ -static uint_t zfs_free_min_time_ms = 1000; +static uint_t zfs_free_min_time_ms = 500; /* minimum milliseconds to resilver per txg */ static uint_t zfs_resilver_min_time_ms = 3000; static uint_t zfs_scan_checkpoint_intval = 7200; /* in seconds */ int zfs_scan_suspend_progress = 0; /* set to prevent scans from progressing */ static int zfs_no_scrub_io = B_FALSE; /* set to disable scrub i/o */ static int zfs_no_scrub_prefetch = B_FALSE; /* set to disable scrub prefetch */ static const ddt_class_t zfs_scrub_ddt_class_max = DDT_CLASS_DUPLICATE; /* max number of blocks to free in a single TXG */ static uint64_t zfs_async_block_max_blocks = UINT64_MAX; /* max number of dedup blocks to free in a single TXG */ -static uint64_t zfs_max_async_dedup_frees = 100000; +static uint64_t zfs_max_async_dedup_frees = 250000; + +/* + * After freeing this many async ZIOs (dedup, clone, gang blocks), wait for + * them to complete before continuing. This prevents unbounded I/O queueing. + */ +static uint64_t zfs_async_free_zio_wait_interval = 2000; /* set to disable resilver deferring */ static int zfs_resilver_disable_defer = B_FALSE; /* Don't defer a resilver if the one in progress only got this far: */ static uint_t zfs_resilver_defer_percent = 10; /* * Number of TXGs to wait after importing before starting background * work (async destroys, scan/scrub/resilver operations). This allows * the import command and filesystem mounts to complete quickly without * being delayed by background activities. The value is somewhat arbitrary * since userspace triggers filesystem mounts asynchronously, but 5 TXGs * provides a reasonable window for import completion in most cases. */ static uint_t zfs_import_defer_txgs = 5; #define DSL_SCAN_IS_SCRUB_RESILVER(scn) \ ((scn)->scn_phys.scn_func == POOL_SCAN_SCRUB || \ (scn)->scn_phys.scn_func == POOL_SCAN_RESILVER) #define DSL_SCAN_IS_SCRUB(scn) \ ((scn)->scn_phys.scn_func == POOL_SCAN_SCRUB) #define DSL_SCAN_IS_RESILVER(scn) \ ((scn)->scn_phys.scn_func == POOL_SCAN_RESILVER) /* * Enable/disable the processing of the free_bpobj object. */ static int zfs_free_bpobj_enabled = 1; /* Error blocks to be scrubbed in one txg. */ static uint_t zfs_scrub_error_blocks_per_txg = 1 << 12; /* the order has to match pool_scan_type */ static scan_cb_t *scan_funcs[POOL_SCAN_FUNCS] = { NULL, dsl_scan_scrub_cb, /* POOL_SCAN_SCRUB */ dsl_scan_scrub_cb, /* POOL_SCAN_RESILVER */ }; /* In core node for the scn->scn_queue. Represents a dataset to be scanned */ typedef struct { uint64_t sds_dsobj; uint64_t sds_txg; avl_node_t sds_node; } scan_ds_t; /* * This controls what conditions are placed on dsl_scan_sync_state(): * SYNC_OPTIONAL) write out scn_phys iff scn_queues_pending == 0 * SYNC_MANDATORY) write out scn_phys always. scn_queues_pending must be 0. * SYNC_CACHED) if scn_queues_pending == 0, write out scn_phys. Otherwise * write out the scn_phys_cached version. * See dsl_scan_sync_state for details. */ typedef enum { SYNC_OPTIONAL, SYNC_MANDATORY, SYNC_CACHED } state_sync_type_t; /* * This struct represents the minimum information needed to reconstruct a * zio for sequential scanning. This is useful because many of these will * accumulate in the sequential IO queues before being issued, so saving * memory matters here. */ typedef struct scan_io { /* fields from blkptr_t */ uint64_t sio_blk_prop; uint64_t sio_phys_birth; uint64_t sio_birth; zio_cksum_t sio_cksum; uint32_t sio_nr_dvas; /* fields from zio_t */ uint32_t sio_flags; zbookmark_phys_t sio_zb; /* members for queue sorting */ union { avl_node_t sio_addr_node; /* link into issuing queue */ list_node_t sio_list_node; /* link for issuing to disk */ } sio_nodes; /* * There may be up to SPA_DVAS_PER_BP DVAs here from the bp, * depending on how many were in the original bp. Only the * first DVA is really used for sorting and issuing purposes. * The other DVAs (if provided) simply exist so that the zio * layer can find additional copies to repair from in the * event of an error. This array must go at the end of the * struct to allow this for the variable number of elements. */ dva_t sio_dva[]; } scan_io_t; #define SIO_SET_OFFSET(sio, x) DVA_SET_OFFSET(&(sio)->sio_dva[0], x) #define SIO_SET_ASIZE(sio, x) DVA_SET_ASIZE(&(sio)->sio_dva[0], x) #define SIO_GET_OFFSET(sio) DVA_GET_OFFSET(&(sio)->sio_dva[0]) #define SIO_GET_ASIZE(sio) DVA_GET_ASIZE(&(sio)->sio_dva[0]) #define SIO_GET_END_OFFSET(sio) \ (SIO_GET_OFFSET(sio) + SIO_GET_ASIZE(sio)) #define SIO_GET_MUSED(sio) \ (sizeof (scan_io_t) + ((sio)->sio_nr_dvas * sizeof (dva_t))) struct dsl_scan_io_queue { dsl_scan_t *q_scn; /* associated dsl_scan_t */ vdev_t *q_vd; /* top-level vdev that this queue represents */ zio_t *q_zio; /* scn_zio_root child for waiting on IO */ /* trees used for sorting I/Os and extents of I/Os */ zfs_range_tree_t *q_exts_by_addr; zfs_btree_t q_exts_by_size; avl_tree_t q_sios_by_addr; uint64_t q_sio_memused; uint64_t q_last_ext_addr; /* members for zio rate limiting */ uint64_t q_maxinflight_bytes; uint64_t q_inflight_bytes; kcondvar_t q_zio_cv; /* used under vd->vdev_scan_io_queue_lock */ /* per txg statistics */ uint64_t q_total_seg_size_this_txg; uint64_t q_segs_this_txg; uint64_t q_total_zio_size_this_txg; uint64_t q_zios_this_txg; }; /* private data for dsl_scan_prefetch_cb() */ typedef struct scan_prefetch_ctx { zfs_refcount_t spc_refcnt; /* refcount for memory management */ dsl_scan_t *spc_scn; /* dsl_scan_t for the pool */ boolean_t spc_root; /* is this prefetch for an objset? */ uint8_t spc_indblkshift; /* dn_indblkshift of current dnode */ uint16_t spc_datablkszsec; /* dn_idatablkszsec of current dnode */ } scan_prefetch_ctx_t; /* private data for dsl_scan_prefetch() */ typedef struct scan_prefetch_issue_ctx { avl_node_t spic_avl_node; /* link into scn->scn_prefetch_queue */ scan_prefetch_ctx_t *spic_spc; /* spc for the callback */ blkptr_t spic_bp; /* bp to prefetch */ zbookmark_phys_t spic_zb; /* bookmark to prefetch */ } scan_prefetch_issue_ctx_t; static void scan_exec_io(dsl_pool_t *dp, const blkptr_t *bp, int zio_flags, const zbookmark_phys_t *zb, dsl_scan_io_queue_t *queue); static void scan_io_queue_insert_impl(dsl_scan_io_queue_t *queue, scan_io_t *sio); static dsl_scan_io_queue_t *scan_io_queue_create(vdev_t *vd); static void scan_io_queues_destroy(dsl_scan_t *scn); static kmem_cache_t *sio_cache[SPA_DVAS_PER_BP]; /* sio->sio_nr_dvas must be set so we know which cache to free from */ static void sio_free(scan_io_t *sio) { ASSERT3U(sio->sio_nr_dvas, >, 0); ASSERT3U(sio->sio_nr_dvas, <=, SPA_DVAS_PER_BP); kmem_cache_free(sio_cache[sio->sio_nr_dvas - 1], sio); } /* It is up to the caller to set sio->sio_nr_dvas for freeing */ static scan_io_t * sio_alloc(unsigned short nr_dvas) { ASSERT3U(nr_dvas, >, 0); ASSERT3U(nr_dvas, <=, SPA_DVAS_PER_BP); return (kmem_cache_alloc(sio_cache[nr_dvas - 1], KM_SLEEP)); } void scan_init(void) { /* * This is used in ext_size_compare() to weight segments * based on how sparse they are. This cannot be changed * mid-scan and the tree comparison functions don't currently * have a mechanism for passing additional context to the * compare functions. Thus we store this value globally and * we only allow it to be set at module initialization time */ fill_weight = zfs_scan_fill_weight; for (int i = 0; i < SPA_DVAS_PER_BP; i++) { char name[36]; (void) snprintf(name, sizeof (name), "sio_cache_%d", i); sio_cache[i] = kmem_cache_create(name, (sizeof (scan_io_t) + ((i + 1) * sizeof (dva_t))), 0, NULL, NULL, NULL, NULL, NULL, 0); } } void scan_fini(void) { for (int i = 0; i < SPA_DVAS_PER_BP; i++) { kmem_cache_destroy(sio_cache[i]); } } static inline boolean_t dsl_scan_is_running(const dsl_scan_t *scn) { return (scn->scn_phys.scn_state == DSS_SCANNING); } boolean_t dsl_scan_resilvering(dsl_pool_t *dp) { return (dsl_scan_is_running(dp->dp_scan) && dp->dp_scan->scn_phys.scn_func == POOL_SCAN_RESILVER); } static inline void sio2bp(const scan_io_t *sio, blkptr_t *bp) { memset(bp, 0, sizeof (*bp)); bp->blk_prop = sio->sio_blk_prop; BP_SET_PHYSICAL_BIRTH(bp, sio->sio_phys_birth); BP_SET_LOGICAL_BIRTH(bp, sio->sio_birth); bp->blk_fill = 1; /* we always only work with data pointers */ bp->blk_cksum = sio->sio_cksum; ASSERT3U(sio->sio_nr_dvas, >, 0); ASSERT3U(sio->sio_nr_dvas, <=, SPA_DVAS_PER_BP); memcpy(bp->blk_dva, sio->sio_dva, sio->sio_nr_dvas * sizeof (dva_t)); } static inline void bp2sio(const blkptr_t *bp, scan_io_t *sio, int dva_i) { sio->sio_blk_prop = bp->blk_prop; sio->sio_phys_birth = BP_GET_RAW_PHYSICAL_BIRTH(bp); sio->sio_birth = BP_GET_LOGICAL_BIRTH(bp); sio->sio_cksum = bp->blk_cksum; sio->sio_nr_dvas = BP_GET_NDVAS(bp); /* * Copy the DVAs to the sio. We need all copies of the block so * that the self healing code can use the alternate copies if the * first is corrupted. We want the DVA at index dva_i to be first * in the sio since this is the primary one that we want to issue. */ for (int i = 0, j = dva_i; i < sio->sio_nr_dvas; i++, j++) { sio->sio_dva[i] = bp->blk_dva[j % sio->sio_nr_dvas]; } } int dsl_scan_init(dsl_pool_t *dp, uint64_t txg) { int err; dsl_scan_t *scn; spa_t *spa = dp->dp_spa; uint64_t f; scn = dp->dp_scan = kmem_zalloc(sizeof (dsl_scan_t), KM_SLEEP); scn->scn_dp = dp; /* * It's possible that we're resuming a scan after a reboot so * make sure that the scan_async_destroying flag is initialized * appropriately. */ ASSERT(!scn->scn_async_destroying); scn->scn_async_destroying = spa_feature_is_active(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY); /* * Calculate the max number of in-flight bytes for pool-wide * scanning operations (minimum 1MB, maximum 1/4 of arc_c_max). * Limits for the issuing phase are done per top-level vdev and * are handled separately. */ scn->scn_maxinflight_bytes = MIN(arc_c_max / 4, MAX(1ULL << 20, zfs_scan_vdev_limit * dsl_scan_count_data_disks(spa))); avl_create(&scn->scn_queue, scan_ds_queue_compare, sizeof (scan_ds_t), offsetof(scan_ds_t, sds_node)); mutex_init(&scn->scn_queue_lock, NULL, MUTEX_DEFAULT, NULL); avl_create(&scn->scn_prefetch_queue, scan_prefetch_queue_compare, sizeof (scan_prefetch_issue_ctx_t), offsetof(scan_prefetch_issue_ctx_t, spic_avl_node)); err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, "scrub_func", sizeof (uint64_t), 1, &f); if (err == 0) { /* * There was an old-style scrub in progress. Restart a * new-style scrub from the beginning. */ scn->scn_restart_txg = txg; zfs_dbgmsg("old-style scrub was in progress for %s; " "restarting new-style scrub in txg %llu", spa->spa_name, (longlong_t)scn->scn_restart_txg); /* * Load the queue obj from the old location so that it * can be freed by dsl_scan_done(). */ (void) zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, "scrub_queue", sizeof (uint64_t), 1, &scn->scn_phys.scn_queue_obj); } else { err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRORSCRUB, sizeof (uint64_t), ERRORSCRUB_PHYS_NUMINTS, &scn->errorscrub_phys); if (err != 0 && err != ENOENT) return (err); err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS, &scn->scn_phys); /* * Detect if the pool contains the signature of #2094. If it * does properly update the scn->scn_phys structure and notify * the administrator by setting an errata for the pool. */ if (err == EOVERFLOW) { uint64_t zaptmp[SCAN_PHYS_NUMINTS + 1]; VERIFY3S(SCAN_PHYS_NUMINTS, ==, 24); VERIFY3S(offsetof(dsl_scan_phys_t, scn_flags), ==, (23 * sizeof (uint64_t))); err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS + 1, &zaptmp); if (err == 0) { uint64_t overflow = zaptmp[SCAN_PHYS_NUMINTS]; if (overflow & ~DSL_SCAN_FLAGS_MASK || scn->scn_async_destroying) { spa->spa_errata = ZPOOL_ERRATA_ZOL_2094_ASYNC_DESTROY; return (EOVERFLOW); } memcpy(&scn->scn_phys, zaptmp, SCAN_PHYS_NUMINTS * sizeof (uint64_t)); scn->scn_phys.scn_flags = overflow; /* Required scrub already in progress. */ if (scn->scn_phys.scn_state == DSS_FINISHED || scn->scn_phys.scn_state == DSS_CANCELED) spa->spa_errata = ZPOOL_ERRATA_ZOL_2094_SCRUB; } } if (err == ENOENT) return (0); else if (err) return (err); /* * We might be restarting after a reboot, so jump the issued * counter to how far we've scanned. We know we're consistent * up to here. */ scn->scn_issued_before_pass = scn->scn_phys.scn_examined - scn->scn_phys.scn_skipped; if (dsl_scan_is_running(scn) && spa_prev_software_version(dp->dp_spa) < SPA_VERSION_SCAN) { /* * A new-type scrub was in progress on an old * pool, and the pool was accessed by old * software. Restart from the beginning, since * the old software may have changed the pool in * the meantime. */ scn->scn_restart_txg = txg; zfs_dbgmsg("new-style scrub for %s was modified " "by old software; restarting in txg %llu", spa->spa_name, (longlong_t)scn->scn_restart_txg); } else if (dsl_scan_resilvering(dp)) { /* * If a resilver is in progress and there are already * errors, restart it instead of finishing this scan and * then restarting it. If there haven't been any errors * then remember that the incore DTL is valid. */ if (scn->scn_phys.scn_errors > 0) { scn->scn_restart_txg = txg; zfs_dbgmsg("resilver can't excise DTL_MISSING " "when finished; restarting on %s in txg " "%llu", spa->spa_name, (u_longlong_t)scn->scn_restart_txg); } else { /* it's safe to excise DTL when finished */ spa->spa_scrub_started = B_TRUE; } } } memcpy(&scn->scn_phys_cached, &scn->scn_phys, sizeof (scn->scn_phys)); /* reload the queue into the in-core state */ if (scn->scn_phys.scn_queue_obj != 0) { zap_cursor_t zc; zap_attribute_t *za = zap_attribute_alloc(); for (zap_cursor_init(&zc, dp->dp_meta_objset, scn->scn_phys.scn_queue_obj); zap_cursor_retrieve(&zc, za) == 0; (void) zap_cursor_advance(&zc)) { scan_ds_queue_insert(scn, zfs_strtonum(za->za_name, NULL), za->za_first_integer); } zap_cursor_fini(&zc); zap_attribute_free(za); } ddt_walk_init(spa, scn->scn_phys.scn_max_txg); spa_scan_stat_init(spa); vdev_scan_stat_init(spa->spa_root_vdev); return (0); } void dsl_scan_fini(dsl_pool_t *dp) { if (dp->dp_scan != NULL) { dsl_scan_t *scn = dp->dp_scan; if (scn->scn_taskq != NULL) taskq_destroy(scn->scn_taskq); scan_ds_queue_clear(scn); avl_destroy(&scn->scn_queue); mutex_destroy(&scn->scn_queue_lock); scan_ds_prefetch_queue_clear(scn); avl_destroy(&scn->scn_prefetch_queue); kmem_free(dp->dp_scan, sizeof (dsl_scan_t)); dp->dp_scan = NULL; } } static boolean_t dsl_scan_restarting(dsl_scan_t *scn, dmu_tx_t *tx) { return (scn->scn_restart_txg != 0 && scn->scn_restart_txg <= tx->tx_txg); } boolean_t dsl_scan_resilver_scheduled(dsl_pool_t *dp) { return ((dp->dp_scan && dp->dp_scan->scn_restart_txg != 0) || (spa_async_tasks(dp->dp_spa) & SPA_ASYNC_RESILVER)); } boolean_t dsl_scan_scrubbing(const dsl_pool_t *dp) { dsl_scan_phys_t *scn_phys = &dp->dp_scan->scn_phys; return (scn_phys->scn_state == DSS_SCANNING && scn_phys->scn_func == POOL_SCAN_SCRUB); } boolean_t dsl_errorscrubbing(const dsl_pool_t *dp) { dsl_errorscrub_phys_t *errorscrub_phys = &dp->dp_scan->errorscrub_phys; return (errorscrub_phys->dep_state == DSS_ERRORSCRUBBING && errorscrub_phys->dep_func == POOL_SCAN_ERRORSCRUB); } boolean_t dsl_errorscrub_is_paused(const dsl_scan_t *scn) { return (dsl_errorscrubbing(scn->scn_dp) && scn->errorscrub_phys.dep_paused_flags); } boolean_t dsl_scan_is_paused_scrub(const dsl_scan_t *scn) { return (dsl_scan_scrubbing(scn->scn_dp) && scn->scn_phys.scn_flags & DSF_SCRUB_PAUSED); } static void dsl_errorscrub_sync_state(dsl_scan_t *scn, dmu_tx_t *tx) { scn->errorscrub_phys.dep_cursor = zap_cursor_serialize(&scn->errorscrub_cursor); VERIFY0(zap_update(scn->scn_dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRORSCRUB, sizeof (uint64_t), ERRORSCRUB_PHYS_NUMINTS, &scn->errorscrub_phys, tx)); } static void dsl_errorscrub_setup_sync(void *arg, dmu_tx_t *tx) { dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan; pool_scan_func_t *funcp = arg; dsl_pool_t *dp = scn->scn_dp; spa_t *spa = dp->dp_spa; ASSERT(!dsl_scan_is_running(scn)); ASSERT(!dsl_errorscrubbing(scn->scn_dp)); ASSERT(*funcp > POOL_SCAN_NONE && *funcp < POOL_SCAN_FUNCS); memset(&scn->errorscrub_phys, 0, sizeof (scn->errorscrub_phys)); scn->errorscrub_phys.dep_func = *funcp; scn->errorscrub_phys.dep_state = DSS_ERRORSCRUBBING; scn->errorscrub_phys.dep_start_time = gethrestime_sec(); scn->errorscrub_phys.dep_to_examine = spa_get_last_errlog_size(spa); scn->errorscrub_phys.dep_examined = 0; scn->errorscrub_phys.dep_errors = 0; scn->errorscrub_phys.dep_cursor = 0; zap_cursor_init_serialized(&scn->errorscrub_cursor, spa->spa_meta_objset, spa->spa_errlog_last, scn->errorscrub_phys.dep_cursor); vdev_config_dirty(spa->spa_root_vdev); spa_event_notify(spa, NULL, NULL, ESC_ZFS_ERRORSCRUB_START); dsl_errorscrub_sync_state(scn, tx); spa_history_log_internal(spa, "error scrub setup", tx, "func=%u mintxg=%u maxtxg=%llu", *funcp, 0, (u_longlong_t)tx->tx_txg); } static int dsl_errorscrub_setup_check(void *arg, dmu_tx_t *tx) { (void) arg; dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan; if (dsl_scan_is_running(scn) || (dsl_errorscrubbing(scn->scn_dp))) { return (SET_ERROR(EBUSY)); } if (spa_get_last_errlog_size(scn->scn_dp->dp_spa) == 0) { return (ECANCELED); } return (0); } /* * Writes out a persistent dsl_scan_phys_t record to the pool directory. * Because we can be running in the block sorting algorithm, we do not always * want to write out the record, only when it is "safe" to do so. This safety * condition is achieved by making sure that the sorting queues are empty * (scn_queues_pending == 0). When this condition is not true, the sync'd state * is inconsistent with how much actual scanning progress has been made. The * kind of sync to be performed is specified by the sync_type argument. If the * sync is optional, we only sync if the queues are empty. If the sync is * mandatory, we do a hard ASSERT to make sure that the queues are empty. The * third possible state is a "cached" sync. This is done in response to: * 1) The dataset that was in the last sync'd dsl_scan_phys_t having been * destroyed, so we wouldn't be able to restart scanning from it. * 2) The snapshot that was in the last sync'd dsl_scan_phys_t having been * superseded by a newer snapshot. * 3) The dataset that was in the last sync'd dsl_scan_phys_t having been * swapped with its clone. * In all cases, a cached sync simply rewrites the last record we've written, * just slightly modified. For the modifications that are performed to the * last written dsl_scan_phys_t, see dsl_scan_ds_destroyed, * dsl_scan_ds_snapshotted and dsl_scan_ds_clone_swapped. */ static void dsl_scan_sync_state(dsl_scan_t *scn, dmu_tx_t *tx, state_sync_type_t sync_type) { int i; spa_t *spa = scn->scn_dp->dp_spa; ASSERT(sync_type != SYNC_MANDATORY || scn->scn_queues_pending == 0); if (scn->scn_queues_pending == 0) { for (i = 0; i < spa->spa_root_vdev->vdev_children; i++) { vdev_t *vd = spa->spa_root_vdev->vdev_child[i]; dsl_scan_io_queue_t *q = vd->vdev_scan_io_queue; if (q == NULL) continue; mutex_enter(&vd->vdev_scan_io_queue_lock); ASSERT3P(avl_first(&q->q_sios_by_addr), ==, NULL); ASSERT3P(zfs_btree_first(&q->q_exts_by_size, NULL), ==, NULL); ASSERT3P(zfs_range_tree_first(q->q_exts_by_addr), ==, NULL); mutex_exit(&vd->vdev_scan_io_queue_lock); } if (scn->scn_phys.scn_queue_obj != 0) scan_ds_queue_sync(scn, tx); VERIFY0(zap_update(scn->scn_dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS, &scn->scn_phys, tx)); memcpy(&scn->scn_phys_cached, &scn->scn_phys, sizeof (scn->scn_phys)); if (scn->scn_checkpointing) zfs_dbgmsg("finish scan checkpoint for %s", spa->spa_name); scn->scn_checkpointing = B_FALSE; scn->scn_last_checkpoint = ddi_get_lbolt(); } else if (sync_type == SYNC_CACHED) { VERIFY0(zap_update(scn->scn_dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS, &scn->scn_phys_cached, tx)); } } int dsl_scan_setup_check(void *arg, dmu_tx_t *tx) { (void) arg; dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan; vdev_t *rvd = scn->scn_dp->dp_spa->spa_root_vdev; if (dsl_scan_is_running(scn) || vdev_rebuild_active(rvd) || dsl_errorscrubbing(scn->scn_dp)) return (SET_ERROR(EBUSY)); return (0); } void dsl_scan_setup_sync(void *arg, dmu_tx_t *tx) { setup_sync_arg_t *setup_sync_arg = (setup_sync_arg_t *)arg; dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan; dmu_object_type_t ot = 0; dsl_pool_t *dp = scn->scn_dp; spa_t *spa = dp->dp_spa; ASSERT(!dsl_scan_is_running(scn)); ASSERT3U(setup_sync_arg->func, >, POOL_SCAN_NONE); ASSERT3U(setup_sync_arg->func, <, POOL_SCAN_FUNCS); memset(&scn->scn_phys, 0, sizeof (scn->scn_phys)); /* * If we are starting a fresh scrub, we erase the error scrub * information from disk. */ memset(&scn->errorscrub_phys, 0, sizeof (scn->errorscrub_phys)); dsl_errorscrub_sync_state(scn, tx); scn->scn_phys.scn_func = setup_sync_arg->func; scn->scn_phys.scn_state = DSS_SCANNING; scn->scn_phys.scn_min_txg = setup_sync_arg->txgstart; if (setup_sync_arg->txgend == 0) { scn->scn_phys.scn_max_txg = tx->tx_txg; } else { scn->scn_phys.scn_max_txg = setup_sync_arg->txgend; } scn->scn_phys.scn_ddt_class_max = DDT_CLASSES - 1; /* the entire DDT */ scn->scn_phys.scn_start_time = gethrestime_sec(); scn->scn_phys.scn_errors = 0; scn->scn_phys.scn_to_examine = spa->spa_root_vdev->vdev_stat.vs_alloc; scn->scn_issued_before_pass = 0; scn->scn_restart_txg = 0; scn->scn_done_txg = 0; scn->scn_last_checkpoint = 0; scn->scn_checkpointing = B_FALSE; spa_scan_stat_init(spa); vdev_scan_stat_init(spa->spa_root_vdev); if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) { scn->scn_phys.scn_ddt_class_max = zfs_scrub_ddt_class_max; /* rewrite all disk labels */ vdev_config_dirty(spa->spa_root_vdev); if (vdev_resilver_needed(spa->spa_root_vdev, &scn->scn_phys.scn_min_txg, &scn->scn_phys.scn_max_txg)) { nvlist_t *aux = fnvlist_alloc(); fnvlist_add_string(aux, ZFS_EV_RESILVER_TYPE, "healing"); spa_event_notify(spa, NULL, aux, ESC_ZFS_RESILVER_START); nvlist_free(aux); } else { spa_event_notify(spa, NULL, NULL, ESC_ZFS_SCRUB_START); } spa->spa_scrub_started = B_TRUE; /* * If this is an incremental scrub, limit the DDT scrub phase * to just the auto-ditto class (for correctness); the rest * of the scrub should go faster using top-down pruning. */ if (scn->scn_phys.scn_min_txg > TXG_INITIAL) scn->scn_phys.scn_ddt_class_max = DDT_CLASS_DITTO; /* * When starting a resilver clear any existing rebuild state. * This is required to prevent stale rebuild status from * being reported when a rebuild is run, then a resilver and * finally a scrub. In which case only the scrub status * should be reported by 'zpool status'. */ if (scn->scn_phys.scn_func == POOL_SCAN_RESILVER) { vdev_t *rvd = spa->spa_root_vdev; for (uint64_t i = 0; i < rvd->vdev_children; i++) { vdev_t *vd = rvd->vdev_child[i]; vdev_rebuild_clear_sync( (void *)(uintptr_t)vd->vdev_id, tx); } } } /* back to the generic stuff */ if (zfs_scan_blkstats) { if (dp->dp_blkstats == NULL) { dp->dp_blkstats = vmem_alloc(sizeof (zfs_all_blkstats_t), KM_SLEEP); } memset(&dp->dp_blkstats->zab_type, 0, sizeof (dp->dp_blkstats->zab_type)); } else { if (dp->dp_blkstats) { vmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t)); dp->dp_blkstats = NULL; } } if (spa_version(spa) < SPA_VERSION_DSL_SCRUB) ot = DMU_OT_ZAP_OTHER; scn->scn_phys.scn_queue_obj = zap_create(dp->dp_meta_objset, ot ? ot : DMU_OT_SCAN_QUEUE, DMU_OT_NONE, 0, tx); memcpy(&scn->scn_phys_cached, &scn->scn_phys, sizeof (scn->scn_phys)); ddt_walk_init(spa, scn->scn_phys.scn_max_txg); dsl_scan_sync_state(scn, tx, SYNC_MANDATORY); spa_history_log_internal(spa, "scan setup", tx, "func=%u mintxg=%llu maxtxg=%llu", setup_sync_arg->func, (u_longlong_t)scn->scn_phys.scn_min_txg, (u_longlong_t)scn->scn_phys.scn_max_txg); } /* * Called by ZFS_IOC_POOL_SCRUB and ZFS_IOC_POOL_SCAN ioctl to start a scrub, * error scrub or resilver. Can also be called to resume a paused scrub or * error scrub. */ int dsl_scan(dsl_pool_t *dp, pool_scan_func_t func, uint64_t txgstart, uint64_t txgend) { spa_t *spa = dp->dp_spa; dsl_scan_t *scn = dp->dp_scan; setup_sync_arg_t setup_sync_arg; if (func != POOL_SCAN_SCRUB && (txgstart != 0 || txgend != 0)) { return (EINVAL); } /* * Purge all vdev caches and probe all devices. We do this here * rather than in sync context because this requires a writer lock * on the spa_config lock, which we can't do from sync context. The * spa_scrub_reopen flag indicates that vdev_open() should not * attempt to start another scrub. */ spa_vdev_state_enter(spa, SCL_NONE); spa->spa_scrub_reopen = B_TRUE; vdev_reopen(spa->spa_root_vdev); spa->spa_scrub_reopen = B_FALSE; (void) spa_vdev_state_exit(spa, NULL, 0); if (func == POOL_SCAN_RESILVER) { dsl_scan_restart_resilver(spa->spa_dsl_pool, 0); return (0); } if (func == POOL_SCAN_ERRORSCRUB) { if (dsl_errorscrub_is_paused(dp->dp_scan)) { /* * got error scrub start cmd, resume paused error scrub. */ int err = dsl_scrub_set_pause_resume(scn->scn_dp, POOL_SCRUB_NORMAL); if (err == 0) { spa_event_notify(spa, NULL, NULL, ESC_ZFS_ERRORSCRUB_RESUME); return (0); } return (SET_ERROR(err)); } return (dsl_sync_task(spa_name(dp->dp_spa), dsl_errorscrub_setup_check, dsl_errorscrub_setup_sync, &func, 0, ZFS_SPACE_CHECK_RESERVED)); } if (func == POOL_SCAN_SCRUB && dsl_scan_is_paused_scrub(scn)) { /* got scrub start cmd, resume paused scrub */ int err = dsl_scrub_set_pause_resume(scn->scn_dp, POOL_SCRUB_NORMAL); if (err == 0) { spa_event_notify(spa, NULL, NULL, ESC_ZFS_SCRUB_RESUME); return (0); } return (SET_ERROR(err)); } setup_sync_arg.func = func; setup_sync_arg.txgstart = txgstart; setup_sync_arg.txgend = txgend; return (dsl_sync_task(spa_name(spa), dsl_scan_setup_check, dsl_scan_setup_sync, &setup_sync_arg, 0, ZFS_SPACE_CHECK_EXTRA_RESERVED)); } static void dsl_errorscrub_done(dsl_scan_t *scn, boolean_t complete, dmu_tx_t *tx) { dsl_pool_t *dp = scn->scn_dp; spa_t *spa = dp->dp_spa; if (complete) { spa_event_notify(spa, NULL, NULL, ESC_ZFS_ERRORSCRUB_FINISH); spa_history_log_internal(spa, "error scrub done", tx, "errors=%llu", (u_longlong_t)spa_approx_errlog_size(spa)); } else { spa_history_log_internal(spa, "error scrub canceled", tx, "errors=%llu", (u_longlong_t)spa_approx_errlog_size(spa)); } scn->errorscrub_phys.dep_state = complete ? DSS_FINISHED : DSS_CANCELED; spa->spa_scrub_active = B_FALSE; spa_errlog_rotate(spa); scn->errorscrub_phys.dep_end_time = gethrestime_sec(); zap_cursor_fini(&scn->errorscrub_cursor); if (spa->spa_errata == ZPOOL_ERRATA_ZOL_2094_SCRUB) spa->spa_errata = 0; ASSERT(!dsl_errorscrubbing(scn->scn_dp)); } static void dsl_scan_done(dsl_scan_t *scn, boolean_t complete, dmu_tx_t *tx) { static const char *old_names[] = { "scrub_bookmark", "scrub_ddt_bookmark", "scrub_ddt_class_max", "scrub_queue", "scrub_min_txg", "scrub_max_txg", "scrub_func", "scrub_errors", NULL }; dsl_pool_t *dp = scn->scn_dp; spa_t *spa = dp->dp_spa; int i; /* Remove any remnants of an old-style scrub. */ for (i = 0; old_names[i]; i++) { (void) zap_remove(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, old_names[i], tx); } if (scn->scn_phys.scn_queue_obj != 0) { VERIFY0(dmu_object_free(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, tx)); scn->scn_phys.scn_queue_obj = 0; } scan_ds_queue_clear(scn); scan_ds_prefetch_queue_clear(scn); scn->scn_phys.scn_flags &= ~DSF_SCRUB_PAUSED; /* * If we were "restarted" from a stopped state, don't bother * with anything else. */ if (!dsl_scan_is_running(scn)) { ASSERT(!scn->scn_is_sorted); return; } if (scn->scn_is_sorted) { scan_io_queues_destroy(scn); scn->scn_is_sorted = B_FALSE; if (scn->scn_taskq != NULL) { taskq_destroy(scn->scn_taskq); scn->scn_taskq = NULL; } } if (dsl_scan_restarting(scn, tx)) { spa_history_log_internal(spa, "scan aborted, restarting", tx, "errors=%llu", (u_longlong_t)spa_approx_errlog_size(spa)); } else if (!complete) { spa_history_log_internal(spa, "scan cancelled", tx, "errors=%llu", (u_longlong_t)spa_approx_errlog_size(spa)); } else { spa_history_log_internal(spa, "scan done", tx, "errors=%llu", (u_longlong_t)spa_approx_errlog_size(spa)); if (DSL_SCAN_IS_SCRUB(scn)) { VERIFY0(zap_update(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_LAST_SCRUBBED_TXG, sizeof (uint64_t), 1, &scn->scn_phys.scn_max_txg, tx)); spa->spa_scrubbed_last_txg = scn->scn_phys.scn_max_txg; } } if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) { spa->spa_scrub_active = B_FALSE; /* * If the scrub/resilver completed, update all DTLs to * reflect this. Whether it succeeded or not, vacate * all temporary scrub DTLs. * * As the scrub does not currently support traversing * data that have been freed but are part of a checkpoint, * we don't mark the scrub as done in the DTLs as faults * may still exist in those vdevs. */ if (complete && !spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { vdev_dtl_reassess(spa->spa_root_vdev, tx->tx_txg, scn->scn_phys.scn_max_txg, B_TRUE, B_FALSE); if (DSL_SCAN_IS_RESILVER(scn)) { nvlist_t *aux = fnvlist_alloc(); fnvlist_add_string(aux, ZFS_EV_RESILVER_TYPE, "healing"); spa_event_notify(spa, NULL, aux, ESC_ZFS_RESILVER_FINISH); nvlist_free(aux); } else { spa_event_notify(spa, NULL, NULL, ESC_ZFS_SCRUB_FINISH); } } else { vdev_dtl_reassess(spa->spa_root_vdev, tx->tx_txg, 0, B_TRUE, B_FALSE); } spa_errlog_rotate(spa); /* * Don't clear flag until after vdev_dtl_reassess to ensure that * DTL_MISSING will get updated when possible. */ scn->scn_phys.scn_state = complete ? DSS_FINISHED : DSS_CANCELED; scn->scn_phys.scn_end_time = gethrestime_sec(); spa->spa_scrub_started = B_FALSE; /* * We may have finished replacing a device. * Let the async thread assess this and handle the detach. */ spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); /* * Clear any resilver_deferred flags in the config. * If there are drives that need resilvering, kick * off an asynchronous request to start resilver. * vdev_clear_resilver_deferred() may update the config * before the resilver can restart. In the event of * a crash during this period, the spa loading code * will find the drives that need to be resilvered * and start the resilver then. */ if (spa_feature_is_enabled(spa, SPA_FEATURE_RESILVER_DEFER) && vdev_clear_resilver_deferred(spa->spa_root_vdev, tx)) { spa_history_log_internal(spa, "starting deferred resilver", tx, "errors=%llu", (u_longlong_t)spa_approx_errlog_size(spa)); spa_async_request(spa, SPA_ASYNC_RESILVER); } /* Clear recent error events (i.e. duplicate events tracking) */ if (complete) zfs_ereport_clear(spa, NULL); } else { scn->scn_phys.scn_state = complete ? DSS_FINISHED : DSS_CANCELED; scn->scn_phys.scn_end_time = gethrestime_sec(); } spa_notify_waiters(spa); if (spa->spa_errata == ZPOOL_ERRATA_ZOL_2094_SCRUB) spa->spa_errata = 0; ASSERT(!dsl_scan_is_running(scn)); } static int dsl_errorscrub_pause_resume_check(void *arg, dmu_tx_t *tx) { pool_scrub_cmd_t *cmd = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_scan_t *scn = dp->dp_scan; if (*cmd == POOL_SCRUB_PAUSE) { /* * can't pause a error scrub when there is no in-progress * error scrub. */ if (!dsl_errorscrubbing(dp)) return (SET_ERROR(ENOENT)); /* can't pause a paused error scrub */ if (dsl_errorscrub_is_paused(scn)) return (SET_ERROR(EBUSY)); } else if (*cmd != POOL_SCRUB_NORMAL) { return (SET_ERROR(ENOTSUP)); } return (0); } static void dsl_errorscrub_pause_resume_sync(void *arg, dmu_tx_t *tx) { pool_scrub_cmd_t *cmd = arg; dsl_pool_t *dp = dmu_tx_pool(tx); spa_t *spa = dp->dp_spa; dsl_scan_t *scn = dp->dp_scan; if (*cmd == POOL_SCRUB_PAUSE) { spa->spa_scan_pass_errorscrub_pause = gethrestime_sec(); scn->errorscrub_phys.dep_paused_flags = B_TRUE; dsl_errorscrub_sync_state(scn, tx); spa_event_notify(spa, NULL, NULL, ESC_ZFS_ERRORSCRUB_PAUSED); } else { ASSERT3U(*cmd, ==, POOL_SCRUB_NORMAL); if (dsl_errorscrub_is_paused(scn)) { /* * We need to keep track of how much time we spend * paused per pass so that we can adjust the error scrub * rate shown in the output of 'zpool status'. */ spa->spa_scan_pass_errorscrub_spent_paused += gethrestime_sec() - spa->spa_scan_pass_errorscrub_pause; spa->spa_scan_pass_errorscrub_pause = 0; scn->errorscrub_phys.dep_paused_flags = B_FALSE; zap_cursor_init_serialized( &scn->errorscrub_cursor, spa->spa_meta_objset, spa->spa_errlog_last, scn->errorscrub_phys.dep_cursor); dsl_errorscrub_sync_state(scn, tx); } } } static int dsl_errorscrub_cancel_check(void *arg, dmu_tx_t *tx) { (void) arg; dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan; /* can't cancel a error scrub when there is no one in-progress */ if (!dsl_errorscrubbing(scn->scn_dp)) return (SET_ERROR(ENOENT)); return (0); } static void dsl_errorscrub_cancel_sync(void *arg, dmu_tx_t *tx) { (void) arg; dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan; dsl_errorscrub_done(scn, B_FALSE, tx); dsl_errorscrub_sync_state(scn, tx); spa_event_notify(scn->scn_dp->dp_spa, NULL, NULL, ESC_ZFS_ERRORSCRUB_ABORT); } static int dsl_scan_cancel_check(void *arg, dmu_tx_t *tx) { (void) arg; dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan; if (!dsl_scan_is_running(scn)) return (SET_ERROR(ENOENT)); return (0); } static void dsl_scan_cancel_sync(void *arg, dmu_tx_t *tx) { (void) arg; dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan; dsl_scan_done(scn, B_FALSE, tx); dsl_scan_sync_state(scn, tx, SYNC_MANDATORY); spa_event_notify(scn->scn_dp->dp_spa, NULL, NULL, ESC_ZFS_SCRUB_ABORT); } int dsl_scan_cancel(dsl_pool_t *dp) { if (dsl_errorscrubbing(dp)) { return (dsl_sync_task(spa_name(dp->dp_spa), dsl_errorscrub_cancel_check, dsl_errorscrub_cancel_sync, NULL, 3, ZFS_SPACE_CHECK_RESERVED)); } return (dsl_sync_task(spa_name(dp->dp_spa), dsl_scan_cancel_check, dsl_scan_cancel_sync, NULL, 3, ZFS_SPACE_CHECK_RESERVED)); } static int dsl_scrub_pause_resume_check(void *arg, dmu_tx_t *tx) { pool_scrub_cmd_t *cmd = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_scan_t *scn = dp->dp_scan; if (*cmd == POOL_SCRUB_PAUSE) { /* can't pause a scrub when there is no in-progress scrub */ if (!dsl_scan_scrubbing(dp)) return (SET_ERROR(ENOENT)); /* can't pause a paused scrub */ if (dsl_scan_is_paused_scrub(scn)) return (SET_ERROR(EBUSY)); } else if (*cmd != POOL_SCRUB_NORMAL) { return (SET_ERROR(ENOTSUP)); } return (0); } static void dsl_scrub_pause_resume_sync(void *arg, dmu_tx_t *tx) { pool_scrub_cmd_t *cmd = arg; dsl_pool_t *dp = dmu_tx_pool(tx); spa_t *spa = dp->dp_spa; dsl_scan_t *scn = dp->dp_scan; if (*cmd == POOL_SCRUB_PAUSE) { /* can't pause a scrub when there is no in-progress scrub */ spa->spa_scan_pass_scrub_pause = gethrestime_sec(); scn->scn_phys.scn_flags |= DSF_SCRUB_PAUSED; scn->scn_phys_cached.scn_flags |= DSF_SCRUB_PAUSED; dsl_scan_sync_state(scn, tx, SYNC_CACHED); spa_event_notify(spa, NULL, NULL, ESC_ZFS_SCRUB_PAUSED); spa_notify_waiters(spa); } else { ASSERT3U(*cmd, ==, POOL_SCRUB_NORMAL); if (dsl_scan_is_paused_scrub(scn)) { /* * We need to keep track of how much time we spend * paused per pass so that we can adjust the scrub rate * shown in the output of 'zpool status' */ spa->spa_scan_pass_scrub_spent_paused += gethrestime_sec() - spa->spa_scan_pass_scrub_pause; spa->spa_scan_pass_scrub_pause = 0; scn->scn_phys.scn_flags &= ~DSF_SCRUB_PAUSED; scn->scn_phys_cached.scn_flags &= ~DSF_SCRUB_PAUSED; dsl_scan_sync_state(scn, tx, SYNC_CACHED); } } } /* * Set scrub pause/resume state if it makes sense to do so */ int dsl_scrub_set_pause_resume(const dsl_pool_t *dp, pool_scrub_cmd_t cmd) { if (dsl_errorscrubbing(dp)) { return (dsl_sync_task(spa_name(dp->dp_spa), dsl_errorscrub_pause_resume_check, dsl_errorscrub_pause_resume_sync, &cmd, 3, ZFS_SPACE_CHECK_RESERVED)); } return (dsl_sync_task(spa_name(dp->dp_spa), dsl_scrub_pause_resume_check, dsl_scrub_pause_resume_sync, &cmd, 3, ZFS_SPACE_CHECK_RESERVED)); } /* start a new scan, or restart an existing one. */ void dsl_scan_restart_resilver(dsl_pool_t *dp, uint64_t txg) { if (txg == 0) { dmu_tx_t *tx; tx = dmu_tx_create_dd(dp->dp_mos_dir); VERIFY0(dmu_tx_assign(tx, DMU_TX_WAIT | DMU_TX_SUSPEND)); txg = dmu_tx_get_txg(tx); dp->dp_scan->scn_restart_txg = txg; dmu_tx_commit(tx); } else { dp->dp_scan->scn_restart_txg = txg; } zfs_dbgmsg("restarting resilver for %s at txg=%llu", dp->dp_spa->spa_name, (longlong_t)txg); } void dsl_free(dsl_pool_t *dp, uint64_t txg, const blkptr_t *bp) { zio_free(dp->dp_spa, txg, bp); } void dsl_free_sync(zio_t *pio, dsl_pool_t *dp, uint64_t txg, const blkptr_t *bpp) { ASSERT(dsl_pool_sync_context(dp)); zio_nowait(zio_free_sync(pio, dp->dp_spa, txg, bpp, pio->io_flags)); } static int scan_ds_queue_compare(const void *a, const void *b) { const scan_ds_t *sds_a = a, *sds_b = b; if (sds_a->sds_dsobj < sds_b->sds_dsobj) return (-1); if (sds_a->sds_dsobj == sds_b->sds_dsobj) return (0); return (1); } static void scan_ds_queue_clear(dsl_scan_t *scn) { void *cookie = NULL; scan_ds_t *sds; while ((sds = avl_destroy_nodes(&scn->scn_queue, &cookie)) != NULL) { kmem_free(sds, sizeof (*sds)); } } static boolean_t scan_ds_queue_contains(dsl_scan_t *scn, uint64_t dsobj, uint64_t *txg) { scan_ds_t srch, *sds; srch.sds_dsobj = dsobj; sds = avl_find(&scn->scn_queue, &srch, NULL); if (sds != NULL && txg != NULL) *txg = sds->sds_txg; return (sds != NULL); } static void scan_ds_queue_insert(dsl_scan_t *scn, uint64_t dsobj, uint64_t txg) { scan_ds_t *sds; avl_index_t where; sds = kmem_zalloc(sizeof (*sds), KM_SLEEP); sds->sds_dsobj = dsobj; sds->sds_txg = txg; VERIFY3P(avl_find(&scn->scn_queue, sds, &where), ==, NULL); avl_insert(&scn->scn_queue, sds, where); } static void scan_ds_queue_remove(dsl_scan_t *scn, uint64_t dsobj) { scan_ds_t srch, *sds; srch.sds_dsobj = dsobj; sds = avl_find(&scn->scn_queue, &srch, NULL); VERIFY(sds != NULL); avl_remove(&scn->scn_queue, sds); kmem_free(sds, sizeof (*sds)); } static void scan_ds_queue_sync(dsl_scan_t *scn, dmu_tx_t *tx) { dsl_pool_t *dp = scn->scn_dp; spa_t *spa = dp->dp_spa; dmu_object_type_t ot = (spa_version(spa) >= SPA_VERSION_DSL_SCRUB) ? DMU_OT_SCAN_QUEUE : DMU_OT_ZAP_OTHER; ASSERT0(scn->scn_queues_pending); ASSERT(scn->scn_phys.scn_queue_obj != 0); VERIFY0(dmu_object_free(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, tx)); scn->scn_phys.scn_queue_obj = zap_create(dp->dp_meta_objset, ot, DMU_OT_NONE, 0, tx); for (scan_ds_t *sds = avl_first(&scn->scn_queue); sds != NULL; sds = AVL_NEXT(&scn->scn_queue, sds)) { VERIFY0(zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, sds->sds_dsobj, sds->sds_txg, tx)); } } /* * Computes the memory limit state that we're currently in. A sorted scan * needs quite a bit of memory to hold the sorting queue, so we need to * reasonably constrain the size so it doesn't impact overall system * performance. We compute two limits: * 1) Hard memory limit: if the amount of memory used by the sorting * queues on a pool gets above this value, we stop the metadata * scanning portion and start issuing the queued up and sorted * I/Os to reduce memory usage. * This limit is calculated as a fraction of physmem (by default 5%). * We constrain the lower bound of the hard limit to an absolute * minimum of zfs_scan_mem_lim_min (default: 16 MiB). We also constrain * the upper bound to 5% of the total pool size - no chance we'll * ever need that much memory, but just to keep the value in check. * 2) Soft memory limit: once we hit the hard memory limit, we start * issuing I/O to reduce queue memory usage, but we don't want to * completely empty out the queues, since we might be able to find I/Os * that will fill in the gaps of our non-sequential IOs at some point * in the future. So we stop the issuing of I/Os once the amount of * memory used drops below the soft limit (at which point we stop issuing * I/O and start scanning metadata again). * * This limit is calculated by subtracting a fraction of the hard * limit from the hard limit. By default this fraction is 5%, so * the soft limit is 95% of the hard limit. We cap the size of the * difference between the hard and soft limits at an absolute * maximum of zfs_scan_mem_lim_soft_max (default: 128 MiB) - this is * sufficient to not cause too frequent switching between the * metadata scan and I/O issue (even at 2k recordsize, 128 MiB's * worth of queues is about 1.2 GiB of on-pool data, so scanning * that should take at least a decent fraction of a second). */ static boolean_t dsl_scan_should_clear(dsl_scan_t *scn) { spa_t *spa = scn->scn_dp->dp_spa; vdev_t *rvd = scn->scn_dp->dp_spa->spa_root_vdev; uint64_t alloc, mlim_hard, mlim_soft, mused; alloc = metaslab_class_get_alloc(spa_normal_class(spa)); alloc += metaslab_class_get_alloc(spa_special_class(spa)); alloc += metaslab_class_get_alloc(spa_dedup_class(spa)); mlim_hard = MAX((physmem / zfs_scan_mem_lim_fact) * PAGESIZE, zfs_scan_mem_lim_min); mlim_hard = MIN(mlim_hard, alloc / 20); mlim_soft = mlim_hard - MIN(mlim_hard / zfs_scan_mem_lim_soft_fact, zfs_scan_mem_lim_soft_max); mused = 0; for (uint64_t i = 0; i < rvd->vdev_children; i++) { vdev_t *tvd = rvd->vdev_child[i]; dsl_scan_io_queue_t *queue; mutex_enter(&tvd->vdev_scan_io_queue_lock); queue = tvd->vdev_scan_io_queue; if (queue != NULL) { /* * # of extents in exts_by_addr = # in exts_by_size. * B-tree efficiency is ~75%, but can be as low as 50%. */ mused += zfs_btree_numnodes(&queue->q_exts_by_size) * (( sizeof (zfs_range_seg_gap_t) + sizeof (uint64_t)) * 3 / 2) + queue->q_sio_memused; } mutex_exit(&tvd->vdev_scan_io_queue_lock); } dprintf("current scan memory usage: %llu bytes\n", (longlong_t)mused); if (mused == 0) ASSERT0(scn->scn_queues_pending); /* * If we are above our hard limit, we need to clear out memory. * If we are below our soft limit, we need to accumulate sequential IOs. * Otherwise, we should keep doing whatever we are currently doing. */ if (mused >= mlim_hard) return (B_TRUE); else if (mused < mlim_soft) return (B_FALSE); else return (scn->scn_clearing); } static boolean_t dsl_scan_check_suspend(dsl_scan_t *scn, const zbookmark_phys_t *zb) { /* we never skip user/group accounting objects */ if (zb && (int64_t)zb->zb_object < 0) return (B_FALSE); if (scn->scn_suspending) return (B_TRUE); /* we're already suspending */ if (!ZB_IS_ZERO(&scn->scn_phys.scn_bookmark)) return (B_FALSE); /* we're resuming */ /* We only know how to resume from level-0 and objset blocks. */ if (zb && (zb->zb_level != 0 && zb->zb_level != ZB_ROOT_LEVEL)) return (B_FALSE); /* * We suspend if: * - we have scanned for at least the minimum time (default 1 sec * for scrub, 3 sec for resilver), and either we have sufficient * dirty data that we are starting to write more quickly * (default 30%), someone is explicitly waiting for this txg * to complete, or we have used up all of the time in the txg * timeout (default 5 sec). * or * - the spa is shutting down because this pool is being exported * or the machine is rebooting. * or * - the scan queue has reached its memory use limit */ uint64_t curr_time_ns = gethrtime(); uint64_t scan_time_ns = curr_time_ns - scn->scn_sync_start_time; uint64_t sync_time_ns = curr_time_ns - scn->scn_dp->dp_spa->spa_sync_starttime; uint64_t dirty_min_bytes = zfs_dirty_data_max * zfs_vdev_async_write_active_min_dirty_percent / 100; uint_t mintime = (scn->scn_phys.scn_func == POOL_SCAN_RESILVER) ? zfs_resilver_min_time_ms : zfs_scrub_min_time_ms; if ((NSEC2MSEC(scan_time_ns) > mintime && (scn->scn_dp->dp_dirty_total >= dirty_min_bytes || txg_sync_waiting(scn->scn_dp) || NSEC2SEC(sync_time_ns) >= zfs_txg_timeout)) || spa_shutting_down(scn->scn_dp->dp_spa) || (zfs_scan_strict_mem_lim && dsl_scan_should_clear(scn)) || !ddt_walk_ready(scn->scn_dp->dp_spa)) { if (zb && zb->zb_level == ZB_ROOT_LEVEL) { dprintf("suspending at first available bookmark " "%llx/%llx/%llx/%llx\n", (longlong_t)zb->zb_objset, (longlong_t)zb->zb_object, (longlong_t)zb->zb_level, (longlong_t)zb->zb_blkid); SET_BOOKMARK(&scn->scn_phys.scn_bookmark, zb->zb_objset, 0, 0, 0); } else if (zb != NULL) { dprintf("suspending at bookmark %llx/%llx/%llx/%llx\n", (longlong_t)zb->zb_objset, (longlong_t)zb->zb_object, (longlong_t)zb->zb_level, (longlong_t)zb->zb_blkid); scn->scn_phys.scn_bookmark = *zb; } else { #ifdef ZFS_DEBUG dsl_scan_phys_t *scnp = &scn->scn_phys; dprintf("suspending at at DDT bookmark " "%llx/%llx/%llx/%llx\n", (longlong_t)scnp->scn_ddt_bookmark.ddb_class, (longlong_t)scnp->scn_ddt_bookmark.ddb_type, (longlong_t)scnp->scn_ddt_bookmark.ddb_checksum, (longlong_t)scnp->scn_ddt_bookmark.ddb_cursor); #endif } scn->scn_suspending = B_TRUE; return (B_TRUE); } return (B_FALSE); } static boolean_t dsl_error_scrub_check_suspend(dsl_scan_t *scn, const zbookmark_phys_t *zb) { /* * We suspend if: * - we have scrubbed for at least the minimum time (default 1 sec * for error scrub), someone is explicitly waiting for this txg * to complete, or we have used up all of the time in the txg * timeout (default 5 sec). * or * - the spa is shutting down because this pool is being exported * or the machine is rebooting. */ uint64_t curr_time_ns = gethrtime(); uint64_t error_scrub_time_ns = curr_time_ns - scn->scn_sync_start_time; uint64_t sync_time_ns = curr_time_ns - scn->scn_dp->dp_spa->spa_sync_starttime; int mintime = zfs_scrub_min_time_ms; if ((NSEC2MSEC(error_scrub_time_ns) > mintime && (txg_sync_waiting(scn->scn_dp) || NSEC2SEC(sync_time_ns) >= zfs_txg_timeout)) || spa_shutting_down(scn->scn_dp->dp_spa)) { if (zb) { dprintf("error scrub suspending at bookmark " "%llx/%llx/%llx/%llx\n", (longlong_t)zb->zb_objset, (longlong_t)zb->zb_object, (longlong_t)zb->zb_level, (longlong_t)zb->zb_blkid); } return (B_TRUE); } return (B_FALSE); } typedef struct zil_scan_arg { dsl_pool_t *zsa_dp; zil_header_t *zsa_zh; } zil_scan_arg_t; static int dsl_scan_zil_block(zilog_t *zilog, const blkptr_t *bp, void *arg, uint64_t claim_txg) { (void) zilog; zil_scan_arg_t *zsa = arg; dsl_pool_t *dp = zsa->zsa_dp; dsl_scan_t *scn = dp->dp_scan; zil_header_t *zh = zsa->zsa_zh; zbookmark_phys_t zb; ASSERT(!BP_IS_REDACTED(bp)); if (BP_IS_HOLE(bp) || BP_GET_BIRTH(bp) <= scn->scn_phys.scn_cur_min_txg) return (0); /* * One block ("stubby") can be allocated a long time ago; we * want to visit that one because it has been allocated * (on-disk) even if it hasn't been claimed (even though for * scrub there's nothing to do to it). */ if (claim_txg == 0 && BP_GET_BIRTH(bp) >= spa_min_claim_txg(dp->dp_spa)) return (0); SET_BOOKMARK(&zb, zh->zh_log.blk_cksum.zc_word[ZIL_ZC_OBJSET], ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, bp->blk_cksum.zc_word[ZIL_ZC_SEQ]); VERIFY0(scan_funcs[scn->scn_phys.scn_func](dp, bp, &zb)); return (0); } static int dsl_scan_zil_record(zilog_t *zilog, const lr_t *lrc, void *arg, uint64_t claim_txg) { (void) zilog; if (lrc->lrc_txtype == TX_WRITE) { zil_scan_arg_t *zsa = arg; dsl_pool_t *dp = zsa->zsa_dp; dsl_scan_t *scn = dp->dp_scan; zil_header_t *zh = zsa->zsa_zh; const lr_write_t *lr = (const lr_write_t *)lrc; const blkptr_t *bp = &lr->lr_blkptr; zbookmark_phys_t zb; ASSERT(!BP_IS_REDACTED(bp)); if (BP_IS_HOLE(bp) || BP_GET_BIRTH(bp) <= scn->scn_phys.scn_cur_min_txg) return (0); /* * birth can be < claim_txg if this record's txg is * already txg sync'ed (but this log block contains * other records that are not synced) */ if (claim_txg == 0 || BP_GET_BIRTH(bp) < claim_txg) return (0); ASSERT3U(BP_GET_LSIZE(bp), !=, 0); SET_BOOKMARK(&zb, zh->zh_log.blk_cksum.zc_word[ZIL_ZC_OBJSET], lr->lr_foid, ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp)); VERIFY0(scan_funcs[scn->scn_phys.scn_func](dp, bp, &zb)); } return (0); } static void dsl_scan_zil(dsl_pool_t *dp, zil_header_t *zh) { uint64_t claim_txg = zh->zh_claim_txg; zil_scan_arg_t zsa = { dp, zh }; zilog_t *zilog; ASSERT(spa_writeable(dp->dp_spa)); /* * We only want to visit blocks that have been claimed but not yet * replayed (or, in read-only mode, blocks that *would* be claimed). */ if (claim_txg == 0) return; zilog = zil_alloc(dp->dp_meta_objset, zh); (void) zil_parse(zilog, dsl_scan_zil_block, dsl_scan_zil_record, &zsa, claim_txg, B_FALSE); zil_free(zilog); } /* * We compare scan_prefetch_issue_ctx_t's based on their bookmarks. The idea * here is to sort the AVL tree by the order each block will be needed. */ static int scan_prefetch_queue_compare(const void *a, const void *b) { const scan_prefetch_issue_ctx_t *spic_a = a, *spic_b = b; const scan_prefetch_ctx_t *spc_a = spic_a->spic_spc; const scan_prefetch_ctx_t *spc_b = spic_b->spic_spc; return (zbookmark_compare(spc_a->spc_datablkszsec, spc_a->spc_indblkshift, spc_b->spc_datablkszsec, spc_b->spc_indblkshift, &spic_a->spic_zb, &spic_b->spic_zb)); } static void scan_prefetch_ctx_rele(scan_prefetch_ctx_t *spc, const void *tag) { if (zfs_refcount_remove(&spc->spc_refcnt, tag) == 0) { zfs_refcount_destroy(&spc->spc_refcnt); kmem_free(spc, sizeof (scan_prefetch_ctx_t)); } } static scan_prefetch_ctx_t * scan_prefetch_ctx_create(dsl_scan_t *scn, dnode_phys_t *dnp, const void *tag) { scan_prefetch_ctx_t *spc; spc = kmem_alloc(sizeof (scan_prefetch_ctx_t), KM_SLEEP); zfs_refcount_create(&spc->spc_refcnt); zfs_refcount_add(&spc->spc_refcnt, tag); spc->spc_scn = scn; if (dnp != NULL) { spc->spc_datablkszsec = dnp->dn_datablkszsec; spc->spc_indblkshift = dnp->dn_indblkshift; spc->spc_root = B_FALSE; } else { spc->spc_datablkszsec = 0; spc->spc_indblkshift = 0; spc->spc_root = B_TRUE; } return (spc); } static void scan_prefetch_ctx_add_ref(scan_prefetch_ctx_t *spc, const void *tag) { zfs_refcount_add(&spc->spc_refcnt, tag); } static void scan_ds_prefetch_queue_clear(dsl_scan_t *scn) { spa_t *spa = scn->scn_dp->dp_spa; void *cookie = NULL; scan_prefetch_issue_ctx_t *spic = NULL; mutex_enter(&spa->spa_scrub_lock); while ((spic = avl_destroy_nodes(&scn->scn_prefetch_queue, &cookie)) != NULL) { scan_prefetch_ctx_rele(spic->spic_spc, scn); kmem_free(spic, sizeof (scan_prefetch_issue_ctx_t)); } mutex_exit(&spa->spa_scrub_lock); } static boolean_t dsl_scan_check_prefetch_resume(scan_prefetch_ctx_t *spc, const zbookmark_phys_t *zb) { zbookmark_phys_t *last_zb = &spc->spc_scn->scn_prefetch_bookmark; dnode_phys_t tmp_dnp; dnode_phys_t *dnp = (spc->spc_root) ? NULL : &tmp_dnp; if (zb->zb_objset != last_zb->zb_objset) return (B_TRUE); if ((int64_t)zb->zb_object < 0) return (B_FALSE); tmp_dnp.dn_datablkszsec = spc->spc_datablkszsec; tmp_dnp.dn_indblkshift = spc->spc_indblkshift; if (zbookmark_subtree_completed(dnp, zb, last_zb)) return (B_TRUE); return (B_FALSE); } static void dsl_scan_prefetch(scan_prefetch_ctx_t *spc, blkptr_t *bp, zbookmark_phys_t *zb) { avl_index_t idx; dsl_scan_t *scn = spc->spc_scn; spa_t *spa = scn->scn_dp->dp_spa; scan_prefetch_issue_ctx_t *spic; if (zfs_no_scrub_prefetch || BP_IS_REDACTED(bp)) return; if (BP_IS_HOLE(bp) || BP_GET_BIRTH(bp) <= scn->scn_phys.scn_cur_min_txg || (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_DNODE && BP_GET_TYPE(bp) != DMU_OT_OBJSET)) return; if (dsl_scan_check_prefetch_resume(spc, zb)) return; scan_prefetch_ctx_add_ref(spc, scn); spic = kmem_alloc(sizeof (scan_prefetch_issue_ctx_t), KM_SLEEP); spic->spic_spc = spc; spic->spic_bp = *bp; spic->spic_zb = *zb; /* * Add the IO to the queue of blocks to prefetch. This allows us to * prioritize blocks that we will need first for the main traversal * thread. */ mutex_enter(&spa->spa_scrub_lock); if (avl_find(&scn->scn_prefetch_queue, spic, &idx) != NULL) { /* this block is already queued for prefetch */ kmem_free(spic, sizeof (scan_prefetch_issue_ctx_t)); scan_prefetch_ctx_rele(spc, scn); mutex_exit(&spa->spa_scrub_lock); return; } avl_insert(&scn->scn_prefetch_queue, spic, idx); cv_broadcast(&spa->spa_scrub_io_cv); mutex_exit(&spa->spa_scrub_lock); } static void dsl_scan_prefetch_dnode(dsl_scan_t *scn, dnode_phys_t *dnp, uint64_t objset, uint64_t object) { int i; zbookmark_phys_t zb; scan_prefetch_ctx_t *spc; if (dnp->dn_nblkptr == 0 && !(dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR)) return; SET_BOOKMARK(&zb, objset, object, 0, 0); spc = scan_prefetch_ctx_create(scn, dnp, FTAG); for (i = 0; i < dnp->dn_nblkptr; i++) { zb.zb_level = BP_GET_LEVEL(&dnp->dn_blkptr[i]); zb.zb_blkid = i; dsl_scan_prefetch(spc, &dnp->dn_blkptr[i], &zb); } if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) { zb.zb_level = 0; zb.zb_blkid = DMU_SPILL_BLKID; dsl_scan_prefetch(spc, DN_SPILL_BLKPTR(dnp), &zb); } scan_prefetch_ctx_rele(spc, FTAG); } static void dsl_scan_prefetch_cb(zio_t *zio, const zbookmark_phys_t *zb, const blkptr_t *bp, arc_buf_t *buf, void *private) { (void) zio; scan_prefetch_ctx_t *spc = private; dsl_scan_t *scn = spc->spc_scn; spa_t *spa = scn->scn_dp->dp_spa; /* broadcast that the IO has completed for rate limiting purposes */ mutex_enter(&spa->spa_scrub_lock); ASSERT3U(spa->spa_scrub_inflight, >=, BP_GET_PSIZE(bp)); spa->spa_scrub_inflight -= BP_GET_PSIZE(bp); cv_broadcast(&spa->spa_scrub_io_cv); mutex_exit(&spa->spa_scrub_lock); /* if there was an error or we are done prefetching, just cleanup */ if (buf == NULL || scn->scn_prefetch_stop) goto out; if (BP_GET_LEVEL(bp) > 0) { int i; blkptr_t *cbp; int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT; zbookmark_phys_t czb; for (i = 0, cbp = buf->b_data; i < epb; i++, cbp++) { SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object, zb->zb_level - 1, zb->zb_blkid * epb + i); dsl_scan_prefetch(spc, cbp, &czb); } } else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) { dnode_phys_t *cdnp; int i; int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT; for (i = 0, cdnp = buf->b_data; i < epb; i += cdnp->dn_extra_slots + 1, cdnp += cdnp->dn_extra_slots + 1) { dsl_scan_prefetch_dnode(scn, cdnp, zb->zb_objset, zb->zb_blkid * epb + i); } } else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) { objset_phys_t *osp = buf->b_data; dsl_scan_prefetch_dnode(scn, &osp->os_meta_dnode, zb->zb_objset, DMU_META_DNODE_OBJECT); if (OBJSET_BUF_HAS_USERUSED(buf)) { if (OBJSET_BUF_HAS_PROJECTUSED(buf)) { dsl_scan_prefetch_dnode(scn, &osp->os_projectused_dnode, zb->zb_objset, DMU_PROJECTUSED_OBJECT); } dsl_scan_prefetch_dnode(scn, &osp->os_groupused_dnode, zb->zb_objset, DMU_GROUPUSED_OBJECT); dsl_scan_prefetch_dnode(scn, &osp->os_userused_dnode, zb->zb_objset, DMU_USERUSED_OBJECT); } } out: if (buf != NULL) arc_buf_destroy(buf, private); scan_prefetch_ctx_rele(spc, scn); } static void dsl_scan_prefetch_thread(void *arg) { dsl_scan_t *scn = arg; spa_t *spa = scn->scn_dp->dp_spa; scan_prefetch_issue_ctx_t *spic; /* loop until we are told to stop */ while (!scn->scn_prefetch_stop) { arc_flags_t flags = ARC_FLAG_NOWAIT | ARC_FLAG_PRESCIENT_PREFETCH | ARC_FLAG_PREFETCH; int zio_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCAN_THREAD; mutex_enter(&spa->spa_scrub_lock); /* * Wait until we have an IO to issue and are not above our * maximum in flight limit. */ while (!scn->scn_prefetch_stop && (avl_numnodes(&scn->scn_prefetch_queue) == 0 || spa->spa_scrub_inflight >= scn->scn_maxinflight_bytes)) { cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); } /* recheck if we should stop since we waited for the cv */ if (scn->scn_prefetch_stop) { mutex_exit(&spa->spa_scrub_lock); break; } /* remove the prefetch IO from the tree */ spic = avl_first(&scn->scn_prefetch_queue); spa->spa_scrub_inflight += BP_GET_PSIZE(&spic->spic_bp); avl_remove(&scn->scn_prefetch_queue, spic); mutex_exit(&spa->spa_scrub_lock); if (BP_IS_PROTECTED(&spic->spic_bp)) { ASSERT(BP_GET_TYPE(&spic->spic_bp) == DMU_OT_DNODE || BP_GET_TYPE(&spic->spic_bp) == DMU_OT_OBJSET); ASSERT3U(BP_GET_LEVEL(&spic->spic_bp), ==, 0); zio_flags |= ZIO_FLAG_RAW; } /* We don't need data L1 buffer since we do not prefetch L0. */ blkptr_t *bp = &spic->spic_bp; if (BP_GET_LEVEL(bp) == 1 && BP_GET_TYPE(bp) != DMU_OT_DNODE && BP_GET_TYPE(bp) != DMU_OT_OBJSET) flags |= ARC_FLAG_NO_BUF; /* issue the prefetch asynchronously */ (void) arc_read(scn->scn_zio_root, spa, bp, dsl_scan_prefetch_cb, spic->spic_spc, ZIO_PRIORITY_SCRUB, zio_flags, &flags, &spic->spic_zb); kmem_free(spic, sizeof (scan_prefetch_issue_ctx_t)); } ASSERT(scn->scn_prefetch_stop); /* free any prefetches we didn't get to complete */ mutex_enter(&spa->spa_scrub_lock); while ((spic = avl_first(&scn->scn_prefetch_queue)) != NULL) { avl_remove(&scn->scn_prefetch_queue, spic); scan_prefetch_ctx_rele(spic->spic_spc, scn); kmem_free(spic, sizeof (scan_prefetch_issue_ctx_t)); } ASSERT0(avl_numnodes(&scn->scn_prefetch_queue)); mutex_exit(&spa->spa_scrub_lock); } static boolean_t dsl_scan_check_resume(dsl_scan_t *scn, const dnode_phys_t *dnp, const zbookmark_phys_t *zb) { /* * We never skip over user/group accounting objects (obj<0) */ if (!ZB_IS_ZERO(&scn->scn_phys.scn_bookmark) && (int64_t)zb->zb_object >= 0) { /* * If we already visited this bp & everything below (in * a prior txg sync), don't bother doing it again. */ if (zbookmark_subtree_completed(dnp, zb, &scn->scn_phys.scn_bookmark)) return (B_TRUE); /* * If we found the block we're trying to resume from, or * we went past it, zero it out to indicate that it's OK * to start checking for suspending again. */ if (zbookmark_subtree_tbd(dnp, zb, &scn->scn_phys.scn_bookmark)) { dprintf("resuming at %llx/%llx/%llx/%llx\n", (longlong_t)zb->zb_objset, (longlong_t)zb->zb_object, (longlong_t)zb->zb_level, (longlong_t)zb->zb_blkid); memset(&scn->scn_phys.scn_bookmark, 0, sizeof (*zb)); } } return (B_FALSE); } static void dsl_scan_visitbp(const blkptr_t *bp, const zbookmark_phys_t *zb, dnode_phys_t *dnp, dsl_dataset_t *ds, dsl_scan_t *scn, dmu_objset_type_t ostype, dmu_tx_t *tx); inline __attribute__((always_inline)) static void dsl_scan_visitdnode( dsl_scan_t *, dsl_dataset_t *ds, dmu_objset_type_t ostype, dnode_phys_t *dnp, uint64_t object, dmu_tx_t *tx); /* * Return nonzero on i/o error. * Return new buf to write out in *bufp. */ inline __attribute__((always_inline)) static int dsl_scan_recurse(dsl_scan_t *scn, dsl_dataset_t *ds, dmu_objset_type_t ostype, dnode_phys_t *dnp, const blkptr_t *bp, const zbookmark_phys_t *zb, dmu_tx_t *tx) { dsl_pool_t *dp = scn->scn_dp; spa_t *spa = dp->dp_spa; int zio_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCAN_THREAD; int err; ASSERT(!BP_IS_REDACTED(bp)); /* * There is an unlikely case of encountering dnodes with contradicting * dn_bonuslen and DNODE_FLAG_SPILL_BLKPTR flag before in files created * or modified before commit 4254acb was merged. As it is not possible * to know which of the two is correct, report an error. */ if (dnp != NULL && dnp->dn_bonuslen > DN_MAX_BONUS_LEN(dnp)) { scn->scn_phys.scn_errors++; spa_log_error(spa, zb, BP_GET_PHYSICAL_BIRTH(bp)); return (SET_ERROR(EINVAL)); } if (BP_GET_LEVEL(bp) > 0) { arc_flags_t flags = ARC_FLAG_WAIT; int i; blkptr_t *cbp; int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT; arc_buf_t *buf; err = arc_read(NULL, spa, bp, arc_getbuf_func, &buf, ZIO_PRIORITY_SCRUB, zio_flags, &flags, zb); if (err) { scn->scn_phys.scn_errors++; return (err); } for (i = 0, cbp = buf->b_data; i < epb; i++, cbp++) { zbookmark_phys_t czb; SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object, zb->zb_level - 1, zb->zb_blkid * epb + i); dsl_scan_visitbp(cbp, &czb, dnp, ds, scn, ostype, tx); } arc_buf_destroy(buf, &buf); } else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) { arc_flags_t flags = ARC_FLAG_WAIT; dnode_phys_t *cdnp; int i; int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT; arc_buf_t *buf; if (BP_IS_PROTECTED(bp)) { ASSERT3U(BP_GET_COMPRESS(bp), ==, ZIO_COMPRESS_OFF); zio_flags |= ZIO_FLAG_RAW; } err = arc_read(NULL, spa, bp, arc_getbuf_func, &buf, ZIO_PRIORITY_SCRUB, zio_flags, &flags, zb); if (err) { scn->scn_phys.scn_errors++; return (err); } for (i = 0, cdnp = buf->b_data; i < epb; i += cdnp->dn_extra_slots + 1, cdnp += cdnp->dn_extra_slots + 1) { dsl_scan_visitdnode(scn, ds, ostype, cdnp, zb->zb_blkid * epb + i, tx); } arc_buf_destroy(buf, &buf); } else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) { arc_flags_t flags = ARC_FLAG_WAIT; objset_phys_t *osp; arc_buf_t *buf; err = arc_read(NULL, spa, bp, arc_getbuf_func, &buf, ZIO_PRIORITY_SCRUB, zio_flags, &flags, zb); if (err) { scn->scn_phys.scn_errors++; return (err); } osp = buf->b_data; dsl_scan_visitdnode(scn, ds, osp->os_type, &osp->os_meta_dnode, DMU_META_DNODE_OBJECT, tx); if (OBJSET_BUF_HAS_USERUSED(buf)) { /* * We also always visit user/group/project accounting * objects, and never skip them, even if we are * suspending. This is necessary so that the * space deltas from this txg get integrated. */ if (OBJSET_BUF_HAS_PROJECTUSED(buf)) dsl_scan_visitdnode(scn, ds, osp->os_type, &osp->os_projectused_dnode, DMU_PROJECTUSED_OBJECT, tx); dsl_scan_visitdnode(scn, ds, osp->os_type, &osp->os_groupused_dnode, DMU_GROUPUSED_OBJECT, tx); dsl_scan_visitdnode(scn, ds, osp->os_type, &osp->os_userused_dnode, DMU_USERUSED_OBJECT, tx); } arc_buf_destroy(buf, &buf); } else if (zfs_blkptr_verify(spa, bp, BLK_CONFIG_NEEDED, BLK_VERIFY_LOG)) { /* * Sanity check the block pointer contents, this is handled * by arc_read() for the cases above. */ scn->scn_phys.scn_errors++; spa_log_error(spa, zb, BP_GET_PHYSICAL_BIRTH(bp)); return (SET_ERROR(EINVAL)); } return (0); } inline __attribute__((always_inline)) static void dsl_scan_visitdnode(dsl_scan_t *scn, dsl_dataset_t *ds, dmu_objset_type_t ostype, dnode_phys_t *dnp, uint64_t object, dmu_tx_t *tx) { int j; for (j = 0; j < dnp->dn_nblkptr; j++) { zbookmark_phys_t czb; SET_BOOKMARK(&czb, ds ? ds->ds_object : 0, object, dnp->dn_nlevels - 1, j); dsl_scan_visitbp(&dnp->dn_blkptr[j], &czb, dnp, ds, scn, ostype, tx); } if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) { zbookmark_phys_t czb; SET_BOOKMARK(&czb, ds ? ds->ds_object : 0, object, 0, DMU_SPILL_BLKID); dsl_scan_visitbp(DN_SPILL_BLKPTR(dnp), &czb, dnp, ds, scn, ostype, tx); } } /* * The arguments are in this order because mdb can only print the * first 5; we want them to be useful. */ static void dsl_scan_visitbp(const blkptr_t *bp, const zbookmark_phys_t *zb, dnode_phys_t *dnp, dsl_dataset_t *ds, dsl_scan_t *scn, dmu_objset_type_t ostype, dmu_tx_t *tx) { dsl_pool_t *dp = scn->scn_dp; if (dsl_scan_check_suspend(scn, zb)) return; if (dsl_scan_check_resume(scn, dnp, zb)) return; scn->scn_visited_this_txg++; if (BP_IS_HOLE(bp)) { scn->scn_holes_this_txg++; return; } if (BP_IS_REDACTED(bp)) { ASSERT(dsl_dataset_feature_is_active(ds, SPA_FEATURE_REDACTED_DATASETS)); return; } /* * Check if this block contradicts any filesystem flags. */ spa_feature_t f = SPA_FEATURE_LARGE_BLOCKS; if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) ASSERT(dsl_dataset_feature_is_active(ds, f)); f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp)); if (f != SPA_FEATURE_NONE) ASSERT(dsl_dataset_feature_is_active(ds, f)); f = zio_compress_to_feature(BP_GET_COMPRESS(bp)); if (f != SPA_FEATURE_NONE) ASSERT(dsl_dataset_feature_is_active(ds, f)); /* * Recurse any blocks that were written either logically or physically * at or after cur_min_txg. About logical birth we care for traversal, * looking for any changes, while about physical for the actual scan. */ if (BP_GET_BIRTH(bp) <= scn->scn_phys.scn_cur_min_txg) { scn->scn_lt_min_this_txg++; return; } if (dsl_scan_recurse(scn, ds, ostype, dnp, bp, zb, tx) != 0) return; /* * If dsl_scan_ddt() has already visited this block, it will have * already done any translations or scrubbing, so don't call the * callback again. */ if (ddt_class_contains(dp->dp_spa, scn->scn_phys.scn_ddt_class_max, bp)) { scn->scn_ddt_contained_this_txg++; return; } /* * If this block is from the future (after cur_max_txg), then we * are doing this on behalf of a deleted snapshot, and we will * revisit the future block on the next pass of this dataset. * Don't scan it now unless we need to because something * under it was modified. */ if (BP_GET_PHYSICAL_BIRTH(bp) > scn->scn_phys.scn_cur_max_txg) { scn->scn_gt_max_this_txg++; return; } scan_funcs[scn->scn_phys.scn_func](dp, bp, zb); } static void dsl_scan_visit_rootbp(dsl_scan_t *scn, dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) { zbookmark_phys_t zb; scan_prefetch_ctx_t *spc; SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET, ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); if (ZB_IS_ZERO(&scn->scn_phys.scn_bookmark)) { SET_BOOKMARK(&scn->scn_prefetch_bookmark, zb.zb_objset, 0, 0, 0); } else { scn->scn_prefetch_bookmark = scn->scn_phys.scn_bookmark; } scn->scn_objsets_visited_this_txg++; spc = scan_prefetch_ctx_create(scn, NULL, FTAG); dsl_scan_prefetch(spc, bp, &zb); scan_prefetch_ctx_rele(spc, FTAG); dsl_scan_visitbp(bp, &zb, NULL, ds, scn, DMU_OST_NONE, tx); dprintf_ds(ds, "finished scan%s", ""); } static void ds_destroyed_scn_phys(dsl_dataset_t *ds, dsl_scan_phys_t *scn_phys) { if (scn_phys->scn_bookmark.zb_objset == ds->ds_object) { if (ds->ds_is_snapshot) { /* * Note: * - scn_cur_{min,max}_txg stays the same. * - Setting the flag is not really necessary if * scn_cur_max_txg == scn_max_txg, because there * is nothing after this snapshot that we care * about. However, we set it anyway and then * ignore it when we retraverse it in * dsl_scan_visitds(). */ scn_phys->scn_bookmark.zb_objset = dsl_dataset_phys(ds)->ds_next_snap_obj; zfs_dbgmsg("destroying ds %llu on %s; currently " "traversing; reset zb_objset to %llu", (u_longlong_t)ds->ds_object, ds->ds_dir->dd_pool->dp_spa->spa_name, (u_longlong_t)dsl_dataset_phys(ds)-> ds_next_snap_obj); scn_phys->scn_flags |= DSF_VISIT_DS_AGAIN; } else { SET_BOOKMARK(&scn_phys->scn_bookmark, ZB_DESTROYED_OBJSET, 0, 0, 0); zfs_dbgmsg("destroying ds %llu on %s; currently " "traversing; reset bookmark to -1,0,0,0", (u_longlong_t)ds->ds_object, ds->ds_dir->dd_pool->dp_spa->spa_name); } } } /* * Invoked when a dataset is destroyed. We need to make sure that: * * 1) If it is the dataset that was currently being scanned, we write * a new dsl_scan_phys_t and marking the objset reference in it * as destroyed. * 2) Remove it from the work queue, if it was present. * * If the dataset was actually a snapshot, instead of marking the dataset * as destroyed, we instead substitute the next snapshot in line. */ void dsl_scan_ds_destroyed(dsl_dataset_t *ds, dmu_tx_t *tx) { dsl_pool_t *dp = ds->ds_dir->dd_pool; dsl_scan_t *scn = dp->dp_scan; uint64_t mintxg; if (!dsl_scan_is_running(scn)) return; ds_destroyed_scn_phys(ds, &scn->scn_phys); ds_destroyed_scn_phys(ds, &scn->scn_phys_cached); if (scan_ds_queue_contains(scn, ds->ds_object, &mintxg)) { scan_ds_queue_remove(scn, ds->ds_object); if (ds->ds_is_snapshot) scan_ds_queue_insert(scn, dsl_dataset_phys(ds)->ds_next_snap_obj, mintxg); } if (zap_lookup_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds->ds_object, &mintxg) == 0) { ASSERT3U(dsl_dataset_phys(ds)->ds_num_children, <=, 1); VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds->ds_object, tx)); if (ds->ds_is_snapshot) { /* * We keep the same mintxg; it could be > * ds_creation_txg if the previous snapshot was * deleted too. */ VERIFY(zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, dsl_dataset_phys(ds)->ds_next_snap_obj, mintxg, tx) == 0); zfs_dbgmsg("destroying ds %llu on %s; in queue; " "replacing with %llu", (u_longlong_t)ds->ds_object, dp->dp_spa->spa_name, (u_longlong_t)dsl_dataset_phys(ds)-> ds_next_snap_obj); } else { zfs_dbgmsg("destroying ds %llu on %s; in queue; " "removing", (u_longlong_t)ds->ds_object, dp->dp_spa->spa_name); } } /* * dsl_scan_sync() should be called after this, and should sync * out our changed state, but just to be safe, do it here. */ dsl_scan_sync_state(scn, tx, SYNC_CACHED); } static void ds_snapshotted_bookmark(dsl_dataset_t *ds, zbookmark_phys_t *scn_bookmark) { if (scn_bookmark->zb_objset == ds->ds_object) { scn_bookmark->zb_objset = dsl_dataset_phys(ds)->ds_prev_snap_obj; zfs_dbgmsg("snapshotting ds %llu on %s; currently traversing; " "reset zb_objset to %llu", (u_longlong_t)ds->ds_object, ds->ds_dir->dd_pool->dp_spa->spa_name, (u_longlong_t)dsl_dataset_phys(ds)->ds_prev_snap_obj); } } /* * Called when a dataset is snapshotted. If we were currently traversing * this snapshot, we reset our bookmark to point at the newly created * snapshot. We also modify our work queue to remove the old snapshot and * replace with the new one. */ void dsl_scan_ds_snapshotted(dsl_dataset_t *ds, dmu_tx_t *tx) { dsl_pool_t *dp = ds->ds_dir->dd_pool; dsl_scan_t *scn = dp->dp_scan; uint64_t mintxg; if (!dsl_scan_is_running(scn)) return; ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0); ds_snapshotted_bookmark(ds, &scn->scn_phys.scn_bookmark); ds_snapshotted_bookmark(ds, &scn->scn_phys_cached.scn_bookmark); if (scan_ds_queue_contains(scn, ds->ds_object, &mintxg)) { scan_ds_queue_remove(scn, ds->ds_object); scan_ds_queue_insert(scn, dsl_dataset_phys(ds)->ds_prev_snap_obj, mintxg); } if (zap_lookup_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds->ds_object, &mintxg) == 0) { VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds->ds_object, tx)); VERIFY(zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, dsl_dataset_phys(ds)->ds_prev_snap_obj, mintxg, tx) == 0); zfs_dbgmsg("snapshotting ds %llu on %s; in queue; " "replacing with %llu", (u_longlong_t)ds->ds_object, dp->dp_spa->spa_name, (u_longlong_t)dsl_dataset_phys(ds)->ds_prev_snap_obj); } dsl_scan_sync_state(scn, tx, SYNC_CACHED); } static void ds_clone_swapped_bookmark(dsl_dataset_t *ds1, dsl_dataset_t *ds2, zbookmark_phys_t *scn_bookmark) { if (scn_bookmark->zb_objset == ds1->ds_object) { scn_bookmark->zb_objset = ds2->ds_object; zfs_dbgmsg("clone_swap ds %llu on %s; currently traversing; " "reset zb_objset to %llu", (u_longlong_t)ds1->ds_object, ds1->ds_dir->dd_pool->dp_spa->spa_name, (u_longlong_t)ds2->ds_object); } else if (scn_bookmark->zb_objset == ds2->ds_object) { scn_bookmark->zb_objset = ds1->ds_object; zfs_dbgmsg("clone_swap ds %llu on %s; currently traversing; " "reset zb_objset to %llu", (u_longlong_t)ds2->ds_object, ds2->ds_dir->dd_pool->dp_spa->spa_name, (u_longlong_t)ds1->ds_object); } } /* * Called when an origin dataset and its clone are swapped. If we were * currently traversing the dataset, we need to switch to traversing the * newly promoted clone. */ void dsl_scan_ds_clone_swapped(dsl_dataset_t *ds1, dsl_dataset_t *ds2, dmu_tx_t *tx) { dsl_pool_t *dp = ds1->ds_dir->dd_pool; dsl_scan_t *scn = dp->dp_scan; uint64_t mintxg1, mintxg2; boolean_t ds1_queued, ds2_queued; if (!dsl_scan_is_running(scn)) return; ds_clone_swapped_bookmark(ds1, ds2, &scn->scn_phys.scn_bookmark); ds_clone_swapped_bookmark(ds1, ds2, &scn->scn_phys_cached.scn_bookmark); /* * Handle the in-memory scan queue. */ ds1_queued = scan_ds_queue_contains(scn, ds1->ds_object, &mintxg1); ds2_queued = scan_ds_queue_contains(scn, ds2->ds_object, &mintxg2); /* Sanity checking. */ if (ds1_queued) { ASSERT3U(mintxg1, ==, dsl_dataset_phys(ds1)->ds_prev_snap_txg); ASSERT3U(mintxg1, ==, dsl_dataset_phys(ds2)->ds_prev_snap_txg); } if (ds2_queued) { ASSERT3U(mintxg2, ==, dsl_dataset_phys(ds1)->ds_prev_snap_txg); ASSERT3U(mintxg2, ==, dsl_dataset_phys(ds2)->ds_prev_snap_txg); } if (ds1_queued && ds2_queued) { /* * If both are queued, we don't need to do anything. * The swapping code below would not handle this case correctly, * since we can't insert ds2 if it is already there. That's * because scan_ds_queue_insert() prohibits a duplicate insert * and panics. */ } else if (ds1_queued) { scan_ds_queue_remove(scn, ds1->ds_object); scan_ds_queue_insert(scn, ds2->ds_object, mintxg1); } else if (ds2_queued) { scan_ds_queue_remove(scn, ds2->ds_object); scan_ds_queue_insert(scn, ds1->ds_object, mintxg2); } /* * Handle the on-disk scan queue. * The on-disk state is an out-of-date version of the in-memory state, * so the in-memory and on-disk values for ds1_queued and ds2_queued may * be different. Therefore we need to apply the swap logic to the * on-disk state independently of the in-memory state. */ ds1_queued = zap_lookup_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds1->ds_object, &mintxg1) == 0; ds2_queued = zap_lookup_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds2->ds_object, &mintxg2) == 0; /* Sanity checking. */ if (ds1_queued) { ASSERT3U(mintxg1, ==, dsl_dataset_phys(ds1)->ds_prev_snap_txg); ASSERT3U(mintxg1, ==, dsl_dataset_phys(ds2)->ds_prev_snap_txg); } if (ds2_queued) { ASSERT3U(mintxg2, ==, dsl_dataset_phys(ds1)->ds_prev_snap_txg); ASSERT3U(mintxg2, ==, dsl_dataset_phys(ds2)->ds_prev_snap_txg); } if (ds1_queued && ds2_queued) { /* * If both are queued, we don't need to do anything. * Alternatively, we could check for EEXIST from * zap_add_int_key() and back out to the original state, but * that would be more work than checking for this case upfront. */ } else if (ds1_queued) { VERIFY3S(0, ==, zap_remove_int(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds1->ds_object, tx)); VERIFY3S(0, ==, zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds2->ds_object, mintxg1, tx)); zfs_dbgmsg("clone_swap ds %llu on %s; in queue; " "replacing with %llu", (u_longlong_t)ds1->ds_object, dp->dp_spa->spa_name, (u_longlong_t)ds2->ds_object); } else if (ds2_queued) { VERIFY3S(0, ==, zap_remove_int(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds2->ds_object, tx)); VERIFY3S(0, ==, zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds1->ds_object, mintxg2, tx)); zfs_dbgmsg("clone_swap ds %llu on %s; in queue; " "replacing with %llu", (u_longlong_t)ds2->ds_object, dp->dp_spa->spa_name, (u_longlong_t)ds1->ds_object); } dsl_scan_sync_state(scn, tx, SYNC_CACHED); } static int enqueue_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) { uint64_t originobj = *(uint64_t *)arg; dsl_dataset_t *ds; int err; dsl_scan_t *scn = dp->dp_scan; if (dsl_dir_phys(hds->ds_dir)->dd_origin_obj != originobj) return (0); err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds); if (err) return (err); while (dsl_dataset_phys(ds)->ds_prev_snap_obj != originobj) { dsl_dataset_t *prev; err = dsl_dataset_hold_obj(dp, dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); dsl_dataset_rele(ds, FTAG); if (err) return (err); ds = prev; } mutex_enter(&scn->scn_queue_lock); scan_ds_queue_insert(scn, ds->ds_object, dsl_dataset_phys(ds)->ds_prev_snap_txg); mutex_exit(&scn->scn_queue_lock); dsl_dataset_rele(ds, FTAG); return (0); } static void dsl_scan_visitds(dsl_scan_t *scn, uint64_t dsobj, dmu_tx_t *tx) { dsl_pool_t *dp = scn->scn_dp; dsl_dataset_t *ds; VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); if (scn->scn_phys.scn_cur_min_txg >= scn->scn_phys.scn_max_txg) { /* * This can happen if this snapshot was created after the * scan started, and we already completed a previous snapshot * that was created after the scan started. This snapshot * only references blocks with: * * birth < our ds_creation_txg * cur_min_txg is no less than ds_creation_txg. * We have already visited these blocks. * or * birth > scn_max_txg * The scan requested not to visit these blocks. * * Subsequent snapshots (and clones) can reference our * blocks, or blocks with even higher birth times. * Therefore we do not need to visit them either, * so we do not add them to the work queue. * * Note that checking for cur_min_txg >= cur_max_txg * is not sufficient, because in that case we may need to * visit subsequent snapshots. This happens when min_txg > 0, * which raises cur_min_txg. In this case we will visit * this dataset but skip all of its blocks, because the * rootbp's birth time is < cur_min_txg. Then we will * add the next snapshots/clones to the work queue. */ char *dsname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP); dsl_dataset_name(ds, dsname); zfs_dbgmsg("scanning dataset %llu (%s) is unnecessary because " "cur_min_txg (%llu) >= max_txg (%llu)", (longlong_t)dsobj, dsname, (longlong_t)scn->scn_phys.scn_cur_min_txg, (longlong_t)scn->scn_phys.scn_max_txg); kmem_free(dsname, MAXNAMELEN); goto out; } /* * Only the ZIL in the head (non-snapshot) is valid. Even though * snapshots can have ZIL block pointers (which may be the same * BP as in the head), they must be ignored. In addition, $ORIGIN * doesn't have a objset (i.e. its ds_bp is a hole) so we don't * need to look for a ZIL in it either. So we traverse the ZIL here, * rather than in scan_recurse(), because the regular snapshot * block-sharing rules don't apply to it. */ if (!dsl_dataset_is_snapshot(ds) && (dp->dp_origin_snap == NULL || ds->ds_dir != dp->dp_origin_snap->ds_dir)) { objset_t *os; if (dmu_objset_from_ds(ds, &os) != 0) { goto out; } dsl_scan_zil(dp, &os->os_zil_header); } /* * Iterate over the bps in this ds. */ dmu_buf_will_dirty(ds->ds_dbuf, tx); rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); dsl_scan_visit_rootbp(scn, ds, &dsl_dataset_phys(ds)->ds_bp, tx); rrw_exit(&ds->ds_bp_rwlock, FTAG); char *dsname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP); dsl_dataset_name(ds, dsname); zfs_dbgmsg("scanned dataset %llu (%s) with min=%llu max=%llu; " "suspending=%u", (longlong_t)dsobj, dsname, (longlong_t)scn->scn_phys.scn_cur_min_txg, (longlong_t)scn->scn_phys.scn_cur_max_txg, (int)scn->scn_suspending); kmem_free(dsname, ZFS_MAX_DATASET_NAME_LEN); if (scn->scn_suspending) goto out; /* * We've finished this pass over this dataset. */ /* * If we did not completely visit this dataset, do another pass. */ if (scn->scn_phys.scn_flags & DSF_VISIT_DS_AGAIN) { zfs_dbgmsg("incomplete pass on %s; visiting again", dp->dp_spa->spa_name); scn->scn_phys.scn_flags &= ~DSF_VISIT_DS_AGAIN; scan_ds_queue_insert(scn, ds->ds_object, scn->scn_phys.scn_cur_max_txg); goto out; } /* * Add descendant datasets to work queue. */ if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) { scan_ds_queue_insert(scn, dsl_dataset_phys(ds)->ds_next_snap_obj, dsl_dataset_phys(ds)->ds_creation_txg); } if (dsl_dataset_phys(ds)->ds_num_children > 1) { boolean_t usenext = B_FALSE; if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) { uint64_t count; /* * A bug in a previous version of the code could * cause upgrade_clones_cb() to not set * ds_next_snap_obj when it should, leading to a * missing entry. Therefore we can only use the * next_clones_obj when its count is correct. */ int err = zap_count(dp->dp_meta_objset, dsl_dataset_phys(ds)->ds_next_clones_obj, &count); if (err == 0 && count == dsl_dataset_phys(ds)->ds_num_children - 1) usenext = B_TRUE; } if (usenext) { zap_cursor_t zc; zap_attribute_t *za = zap_attribute_alloc(); for (zap_cursor_init(&zc, dp->dp_meta_objset, dsl_dataset_phys(ds)->ds_next_clones_obj); zap_cursor_retrieve(&zc, za) == 0; (void) zap_cursor_advance(&zc)) { scan_ds_queue_insert(scn, zfs_strtonum(za->za_name, NULL), dsl_dataset_phys(ds)->ds_creation_txg); } zap_cursor_fini(&zc); zap_attribute_free(za); } else { VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, enqueue_clones_cb, &ds->ds_object, DS_FIND_CHILDREN)); } } out: dsl_dataset_rele(ds, FTAG); } static int enqueue_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) { (void) arg; dsl_dataset_t *ds; int err; dsl_scan_t *scn = dp->dp_scan; err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds); if (err) return (err); while (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { dsl_dataset_t *prev; err = dsl_dataset_hold_obj(dp, dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); if (err) { dsl_dataset_rele(ds, FTAG); return (err); } /* * If this is a clone, we don't need to worry about it for now. */ if (dsl_dataset_phys(prev)->ds_next_snap_obj != ds->ds_object) { dsl_dataset_rele(ds, FTAG); dsl_dataset_rele(prev, FTAG); return (0); } dsl_dataset_rele(ds, FTAG); ds = prev; } mutex_enter(&scn->scn_queue_lock); scan_ds_queue_insert(scn, ds->ds_object, dsl_dataset_phys(ds)->ds_prev_snap_txg); mutex_exit(&scn->scn_queue_lock); dsl_dataset_rele(ds, FTAG); return (0); } void dsl_scan_ddt_entry(dsl_scan_t *scn, enum zio_checksum checksum, ddt_t *ddt, ddt_lightweight_entry_t *ddlwe, dmu_tx_t *tx) { (void) tx; const ddt_key_t *ddk = &ddlwe->ddlwe_key; blkptr_t bp; zbookmark_phys_t zb = { 0 }; if (!dsl_scan_is_running(scn)) return; /* * This function is special because it is the only thing * that can add scan_io_t's to the vdev scan queues from * outside dsl_scan_sync(). For the most part this is ok * as long as it is called from within syncing context. * However, dsl_scan_sync() expects that no new sio's will * be added between when all the work for a scan is done * and the next txg when the scan is actually marked as * completed. This check ensures we do not issue new sio's * during this period. */ if (scn->scn_done_txg != 0) return; for (int p = 0; p < DDT_NPHYS(ddt); p++) { ddt_phys_variant_t v = DDT_PHYS_VARIANT(ddt, p); uint64_t phys_birth = ddt_phys_birth(&ddlwe->ddlwe_phys, v); if (phys_birth == 0 || phys_birth > scn->scn_phys.scn_max_txg) continue; ddt_bp_create(checksum, ddk, &ddlwe->ddlwe_phys, v, &bp); scn->scn_visited_this_txg++; scan_funcs[scn->scn_phys.scn_func](scn->scn_dp, &bp, &zb); } } /* * Scrub/dedup interaction. * * If there are N references to a deduped block, we don't want to scrub it * N times -- ideally, we should scrub it exactly once. * * We leverage the fact that the dde's replication class (ddt_class_t) * is ordered from highest replication class (DDT_CLASS_DITTO) to lowest * (DDT_CLASS_UNIQUE) so that we may walk the DDT in that order. * * To prevent excess scrubbing, the scrub begins by walking the DDT * to find all blocks with refcnt > 1, and scrubs each of these once. * Since there are two replication classes which contain blocks with * refcnt > 1, we scrub the highest replication class (DDT_CLASS_DITTO) first. * Finally the top-down scrub begins, only visiting blocks with refcnt == 1. * * There would be nothing more to say if a block's refcnt couldn't change * during a scrub, but of course it can so we must account for changes * in a block's replication class. * * Here's an example of what can occur: * * If a block has refcnt > 1 during the DDT scrub phase, but has refcnt == 1 * when visited during the top-down scrub phase, it will be scrubbed twice. * This negates our scrub optimization, but is otherwise harmless. * * If a block has refcnt == 1 during the DDT scrub phase, but has refcnt > 1 * on each visit during the top-down scrub phase, it will never be scrubbed. * To catch this, ddt_sync_entry() notifies the scrub code whenever a block's * reference class transitions to a higher level (i.e DDT_CLASS_UNIQUE to * DDT_CLASS_DUPLICATE); if it transitions from refcnt == 1 to refcnt > 1 * while a scrub is in progress, it scrubs the block right then. */ static void dsl_scan_ddt(dsl_scan_t *scn, dmu_tx_t *tx) { ddt_bookmark_t *ddb = &scn->scn_phys.scn_ddt_bookmark; ddt_lightweight_entry_t ddlwe = {0}; int error; uint64_t n = 0; while ((error = ddt_walk(scn->scn_dp->dp_spa, ddb, &ddlwe)) == 0) { ddt_t *ddt; if (ddb->ddb_class > scn->scn_phys.scn_ddt_class_max) break; dprintf("visiting ddb=%llu/%llu/%llu/%llx\n", (longlong_t)ddb->ddb_class, (longlong_t)ddb->ddb_type, (longlong_t)ddb->ddb_checksum, (longlong_t)ddb->ddb_cursor); /* There should be no pending changes to the dedup table */ ddt = scn->scn_dp->dp_spa->spa_ddt[ddb->ddb_checksum]; ASSERT(avl_first(&ddt->ddt_tree) == NULL); dsl_scan_ddt_entry(scn, ddb->ddb_checksum, ddt, &ddlwe, tx); n++; if (dsl_scan_check_suspend(scn, NULL)) break; } if (error == EAGAIN) { dsl_scan_check_suspend(scn, NULL); error = 0; zfs_dbgmsg("waiting for ddt to become ready for scan " "on %s with class_max = %u; suspending=%u", scn->scn_dp->dp_spa->spa_name, (int)scn->scn_phys.scn_ddt_class_max, (int)scn->scn_suspending); } else zfs_dbgmsg("scanned %llu ddt entries on %s with " "class_max = %u; suspending=%u", (longlong_t)n, scn->scn_dp->dp_spa->spa_name, (int)scn->scn_phys.scn_ddt_class_max, (int)scn->scn_suspending); ASSERT(error == 0 || error == ENOENT); ASSERT(error != ENOENT || ddb->ddb_class > scn->scn_phys.scn_ddt_class_max); } static uint64_t dsl_scan_ds_maxtxg(dsl_dataset_t *ds) { uint64_t smt = ds->ds_dir->dd_pool->dp_scan->scn_phys.scn_max_txg; if (ds->ds_is_snapshot) return (MIN(smt, dsl_dataset_phys(ds)->ds_creation_txg)); return (smt); } static void dsl_scan_visit(dsl_scan_t *scn, dmu_tx_t *tx) { scan_ds_t *sds; dsl_pool_t *dp = scn->scn_dp; if (scn->scn_phys.scn_ddt_bookmark.ddb_class <= scn->scn_phys.scn_ddt_class_max) { scn->scn_phys.scn_cur_min_txg = scn->scn_phys.scn_min_txg; scn->scn_phys.scn_cur_max_txg = scn->scn_phys.scn_max_txg; dsl_scan_ddt(scn, tx); if (scn->scn_suspending) return; } if (scn->scn_phys.scn_bookmark.zb_objset == DMU_META_OBJSET) { /* First do the MOS & ORIGIN */ scn->scn_phys.scn_cur_min_txg = scn->scn_phys.scn_min_txg; scn->scn_phys.scn_cur_max_txg = scn->scn_phys.scn_max_txg; dsl_scan_visit_rootbp(scn, NULL, &dp->dp_meta_rootbp, tx); if (scn->scn_suspending) return; if (spa_version(dp->dp_spa) < SPA_VERSION_DSL_SCRUB) { VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, enqueue_cb, NULL, DS_FIND_CHILDREN)); } else { dsl_scan_visitds(scn, dp->dp_origin_snap->ds_object, tx); } ASSERT(!scn->scn_suspending); } else if (scn->scn_phys.scn_bookmark.zb_objset != ZB_DESTROYED_OBJSET) { uint64_t dsobj = scn->scn_phys.scn_bookmark.zb_objset; /* * If we were suspended, continue from here. Note if the * ds we were suspended on was deleted, the zb_objset may * be -1, so we will skip this and find a new objset * below. */ dsl_scan_visitds(scn, dsobj, tx); if (scn->scn_suspending) return; } /* * In case we suspended right at the end of the ds, zero the * bookmark so we don't think that we're still trying to resume. */ memset(&scn->scn_phys.scn_bookmark, 0, sizeof (zbookmark_phys_t)); /* * Keep pulling things out of the dataset avl queue. Updates to the * persistent zap-object-as-queue happen only at checkpoints. */ while ((sds = avl_first(&scn->scn_queue)) != NULL) { dsl_dataset_t *ds; uint64_t dsobj = sds->sds_dsobj; uint64_t txg = sds->sds_txg; /* dequeue and free the ds from the queue */ scan_ds_queue_remove(scn, dsobj); sds = NULL; /* set up min / max txg */ VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); if (txg != 0) { scn->scn_phys.scn_cur_min_txg = MAX(scn->scn_phys.scn_min_txg, txg); } else { scn->scn_phys.scn_cur_min_txg = MAX(scn->scn_phys.scn_min_txg, dsl_dataset_phys(ds)->ds_prev_snap_txg); } scn->scn_phys.scn_cur_max_txg = dsl_scan_ds_maxtxg(ds); dsl_dataset_rele(ds, FTAG); dsl_scan_visitds(scn, dsobj, tx); if (scn->scn_suspending) return; } /* No more objsets to fetch, we're done */ scn->scn_phys.scn_bookmark.zb_objset = ZB_DESTROYED_OBJSET; ASSERT0(scn->scn_suspending); } static uint64_t dsl_scan_count_data_disks(spa_t *spa) { vdev_t *rvd = spa->spa_root_vdev; uint64_t i, leaves = 0; for (i = 0; i < rvd->vdev_children; i++) { vdev_t *vd = rvd->vdev_child[i]; if (vd->vdev_islog || vd->vdev_isspare || vd->vdev_isl2cache) continue; leaves += vdev_get_ndisks(vd) - vdev_get_nparity(vd); } return (leaves); } static void scan_io_queues_update_zio_stats(dsl_scan_io_queue_t *q, const blkptr_t *bp) { int i; uint64_t cur_size = 0; for (i = 0; i < BP_GET_NDVAS(bp); i++) { cur_size += DVA_GET_ASIZE(&bp->blk_dva[i]); } q->q_total_zio_size_this_txg += cur_size; q->q_zios_this_txg++; } static void scan_io_queues_update_seg_stats(dsl_scan_io_queue_t *q, uint64_t start, uint64_t end) { q->q_total_seg_size_this_txg += end - start; q->q_segs_this_txg++; } static boolean_t scan_io_queue_check_suspend(dsl_scan_t *scn) { /* See comment in dsl_scan_check_suspend() */ uint64_t curr_time_ns = gethrtime(); uint64_t scan_time_ns = curr_time_ns - scn->scn_sync_start_time; uint64_t sync_time_ns = curr_time_ns - scn->scn_dp->dp_spa->spa_sync_starttime; uint64_t dirty_min_bytes = zfs_dirty_data_max * zfs_vdev_async_write_active_min_dirty_percent / 100; uint_t mintime = (scn->scn_phys.scn_func == POOL_SCAN_RESILVER) ? zfs_resilver_min_time_ms : zfs_scrub_min_time_ms; return ((NSEC2MSEC(scan_time_ns) > mintime && (scn->scn_dp->dp_dirty_total >= dirty_min_bytes || txg_sync_waiting(scn->scn_dp) || NSEC2SEC(sync_time_ns) >= zfs_txg_timeout)) || spa_shutting_down(scn->scn_dp->dp_spa)); } /* * Given a list of scan_io_t's in io_list, this issues the I/Os out to * disk. This consumes the io_list and frees the scan_io_t's. This is * called when emptying queues, either when we're up against the memory * limit or when we have finished scanning. Returns B_TRUE if we stopped * processing the list before we finished. Any sios that were not issued * will remain in the io_list. */ static boolean_t scan_io_queue_issue(dsl_scan_io_queue_t *queue, list_t *io_list) { dsl_scan_t *scn = queue->q_scn; scan_io_t *sio; boolean_t suspended = B_FALSE; while ((sio = list_head(io_list)) != NULL) { blkptr_t bp; if (scan_io_queue_check_suspend(scn)) { suspended = B_TRUE; break; } sio2bp(sio, &bp); scan_exec_io(scn->scn_dp, &bp, sio->sio_flags, &sio->sio_zb, queue); (void) list_remove_head(io_list); scan_io_queues_update_zio_stats(queue, &bp); sio_free(sio); } return (suspended); } /* * This function removes sios from an IO queue which reside within a given * zfs_range_seg_t and inserts them (in offset order) into a list. Note that * we only ever return a maximum of 32 sios at once. If there are more sios * to process within this segment that did not make it onto the list we * return B_TRUE and otherwise B_FALSE. */ static boolean_t scan_io_queue_gather(dsl_scan_io_queue_t *queue, zfs_range_seg_t *rs, list_t *list) { scan_io_t *srch_sio, *sio, *next_sio; avl_index_t idx; uint_t num_sios = 0; int64_t bytes_issued = 0; ASSERT(rs != NULL); ASSERT(MUTEX_HELD(&queue->q_vd->vdev_scan_io_queue_lock)); srch_sio = sio_alloc(1); srch_sio->sio_nr_dvas = 1; SIO_SET_OFFSET(srch_sio, zfs_rs_get_start(rs, queue->q_exts_by_addr)); /* * The exact start of the extent might not contain any matching zios, * so if that's the case, examine the next one in the tree. */ sio = avl_find(&queue->q_sios_by_addr, srch_sio, &idx); sio_free(srch_sio); if (sio == NULL) sio = avl_nearest(&queue->q_sios_by_addr, idx, AVL_AFTER); while (sio != NULL && SIO_GET_OFFSET(sio) < zfs_rs_get_end(rs, queue->q_exts_by_addr) && num_sios <= 32) { ASSERT3U(SIO_GET_OFFSET(sio), >=, zfs_rs_get_start(rs, queue->q_exts_by_addr)); ASSERT3U(SIO_GET_END_OFFSET(sio), <=, zfs_rs_get_end(rs, queue->q_exts_by_addr)); next_sio = AVL_NEXT(&queue->q_sios_by_addr, sio); avl_remove(&queue->q_sios_by_addr, sio); if (avl_is_empty(&queue->q_sios_by_addr)) atomic_add_64(&queue->q_scn->scn_queues_pending, -1); queue->q_sio_memused -= SIO_GET_MUSED(sio); bytes_issued += SIO_GET_ASIZE(sio); num_sios++; list_insert_tail(list, sio); sio = next_sio; } /* * We limit the number of sios we process at once to 32 to avoid * biting off more than we can chew. If we didn't take everything * in the segment we update it to reflect the work we were able to * complete. Otherwise, we remove it from the range tree entirely. */ if (sio != NULL && SIO_GET_OFFSET(sio) < zfs_rs_get_end(rs, queue->q_exts_by_addr)) { zfs_range_tree_adjust_fill(queue->q_exts_by_addr, rs, -bytes_issued); zfs_range_tree_resize_segment(queue->q_exts_by_addr, rs, SIO_GET_OFFSET(sio), zfs_rs_get_end(rs, queue->q_exts_by_addr) - SIO_GET_OFFSET(sio)); queue->q_last_ext_addr = SIO_GET_OFFSET(sio); return (B_TRUE); } else { uint64_t rstart = zfs_rs_get_start(rs, queue->q_exts_by_addr); uint64_t rend = zfs_rs_get_end(rs, queue->q_exts_by_addr); zfs_range_tree_remove(queue->q_exts_by_addr, rstart, rend - rstart); queue->q_last_ext_addr = -1; return (B_FALSE); } } /* * This is called from the queue emptying thread and selects the next * extent from which we are to issue I/Os. The behavior of this function * depends on the state of the scan, the current memory consumption and * whether or not we are performing a scan shutdown. * 1) We select extents in an elevator algorithm (LBA-order) if the scan * needs to perform a checkpoint * 2) We select the largest available extent if we are up against the * memory limit. * 3) Otherwise we don't select any extents. */ static zfs_range_seg_t * scan_io_queue_fetch_ext(dsl_scan_io_queue_t *queue) { dsl_scan_t *scn = queue->q_scn; zfs_range_tree_t *rt = queue->q_exts_by_addr; ASSERT(MUTEX_HELD(&queue->q_vd->vdev_scan_io_queue_lock)); ASSERT(scn->scn_is_sorted); if (!scn->scn_checkpointing && !scn->scn_clearing) return (NULL); /* * During normal clearing, we want to issue our largest segments * first, keeping IO as sequential as possible, and leaving the * smaller extents for later with the hope that they might eventually * grow to larger sequential segments. However, when the scan is * checkpointing, no new extents will be added to the sorting queue, * so the way we are sorted now is as good as it will ever get. * In this case, we instead switch to issuing extents in LBA order. */ if ((zfs_scan_issue_strategy < 1 && scn->scn_checkpointing) || zfs_scan_issue_strategy == 1) return (zfs_range_tree_first(rt)); /* * Try to continue previous extent if it is not completed yet. After * shrink in scan_io_queue_gather() it may no longer be the best, but * otherwise we leave shorter remnant every txg. */ uint64_t start; uint64_t size = 1ULL << rt->rt_shift; zfs_range_seg_t *addr_rs; if (queue->q_last_ext_addr != -1) { start = queue->q_last_ext_addr; addr_rs = zfs_range_tree_find(rt, start, size); if (addr_rs != NULL) return (addr_rs); } /* * Nothing to continue, so find new best extent. */ uint64_t *v = zfs_btree_first(&queue->q_exts_by_size, NULL); if (v == NULL) return (NULL); queue->q_last_ext_addr = start = *v << rt->rt_shift; /* * We need to get the original entry in the by_addr tree so we can * modify it. */ addr_rs = zfs_range_tree_find(rt, start, size); ASSERT3P(addr_rs, !=, NULL); ASSERT3U(zfs_rs_get_start(addr_rs, rt), ==, start); ASSERT3U(zfs_rs_get_end(addr_rs, rt), >, start); return (addr_rs); } static void scan_io_queues_run_one(void *arg) { dsl_scan_io_queue_t *queue = arg; kmutex_t *q_lock = &queue->q_vd->vdev_scan_io_queue_lock; boolean_t suspended = B_FALSE; zfs_range_seg_t *rs; scan_io_t *sio; zio_t *zio; list_t sio_list; ASSERT(queue->q_scn->scn_is_sorted); list_create(&sio_list, sizeof (scan_io_t), offsetof(scan_io_t, sio_nodes.sio_list_node)); zio = zio_null(queue->q_scn->scn_zio_root, queue->q_scn->scn_dp->dp_spa, NULL, NULL, NULL, ZIO_FLAG_CANFAIL); mutex_enter(q_lock); queue->q_zio = zio; /* Calculate maximum in-flight bytes for this vdev. */ queue->q_maxinflight_bytes = MAX(1, zfs_scan_vdev_limit * (vdev_get_ndisks(queue->q_vd) - vdev_get_nparity(queue->q_vd))); /* reset per-queue scan statistics for this txg */ queue->q_total_seg_size_this_txg = 0; queue->q_segs_this_txg = 0; queue->q_total_zio_size_this_txg = 0; queue->q_zios_this_txg = 0; /* loop until we run out of time or sios */ while ((rs = scan_io_queue_fetch_ext(queue)) != NULL) { uint64_t seg_start = 0, seg_end = 0; boolean_t more_left; ASSERT(list_is_empty(&sio_list)); /* loop while we still have sios left to process in this rs */ do { scan_io_t *first_sio, *last_sio; /* * We have selected which extent needs to be * processed next. Gather up the corresponding sios. */ more_left = scan_io_queue_gather(queue, rs, &sio_list); ASSERT(!list_is_empty(&sio_list)); first_sio = list_head(&sio_list); last_sio = list_tail(&sio_list); seg_end = SIO_GET_END_OFFSET(last_sio); if (seg_start == 0) seg_start = SIO_GET_OFFSET(first_sio); /* * Issuing sios can take a long time so drop the * queue lock. The sio queue won't be updated by * other threads since we're in syncing context so * we can be sure that our trees will remain exactly * as we left them. */ mutex_exit(q_lock); suspended = scan_io_queue_issue(queue, &sio_list); mutex_enter(q_lock); if (suspended) break; } while (more_left); /* update statistics for debugging purposes */ scan_io_queues_update_seg_stats(queue, seg_start, seg_end); if (suspended) break; } /* * If we were suspended in the middle of processing, * requeue any unfinished sios and exit. */ while ((sio = list_remove_head(&sio_list)) != NULL) scan_io_queue_insert_impl(queue, sio); queue->q_zio = NULL; mutex_exit(q_lock); zio_nowait(zio); list_destroy(&sio_list); } /* * Performs an emptying run on all scan queues in the pool. This just * punches out one thread per top-level vdev, each of which processes * only that vdev's scan queue. We can parallelize the I/O here because * we know that each queue's I/Os only affect its own top-level vdev. * * This function waits for the queue runs to complete, and must be * called from dsl_scan_sync (or in general, syncing context). */ static void scan_io_queues_run(dsl_scan_t *scn) { spa_t *spa = scn->scn_dp->dp_spa; ASSERT(scn->scn_is_sorted); ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); if (scn->scn_queues_pending == 0) return; if (scn->scn_taskq == NULL) { int nthreads = spa->spa_root_vdev->vdev_children; /* * We need to make this taskq *always* execute as many * threads in parallel as we have top-level vdevs and no * less, otherwise strange serialization of the calls to * scan_io_queues_run_one can occur during spa_sync runs * and that significantly impacts performance. */ scn->scn_taskq = taskq_create("dsl_scan_iss", nthreads, minclsyspri, nthreads, nthreads, TASKQ_PREPOPULATE); } for (uint64_t i = 0; i < spa->spa_root_vdev->vdev_children; i++) { vdev_t *vd = spa->spa_root_vdev->vdev_child[i]; mutex_enter(&vd->vdev_scan_io_queue_lock); if (vd->vdev_scan_io_queue != NULL) { VERIFY(taskq_dispatch(scn->scn_taskq, scan_io_queues_run_one, vd->vdev_scan_io_queue, TQ_SLEEP) != TASKQID_INVALID); } mutex_exit(&vd->vdev_scan_io_queue_lock); } /* * Wait for the queues to finish issuing their IOs for this run * before we return. There may still be IOs in flight at this * point. */ taskq_wait(scn->scn_taskq); } static boolean_t dsl_scan_async_block_should_pause(dsl_scan_t *scn) { uint64_t elapsed_nanosecs; if (zfs_recover) return (B_FALSE); if (zfs_async_block_max_blocks != 0 && scn->scn_visited_this_txg >= zfs_async_block_max_blocks) { return (B_TRUE); } if (zfs_max_async_dedup_frees != 0 && - scn->scn_dedup_frees_this_txg >= zfs_max_async_dedup_frees) { + scn->scn_async_frees_this_txg >= zfs_max_async_dedup_frees) { return (B_TRUE); } elapsed_nanosecs = gethrtime() - scn->scn_sync_start_time; - return (elapsed_nanosecs / NANOSEC > zfs_txg_timeout || + return (elapsed_nanosecs / (NANOSEC / 2) > zfs_txg_timeout || (NSEC2MSEC(elapsed_nanosecs) > scn->scn_async_block_min_time_ms && txg_sync_waiting(scn->scn_dp)) || spa_shutting_down(scn->scn_dp->dp_spa)); } static int dsl_scan_free_block_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) { dsl_scan_t *scn = arg; if (!scn->scn_is_bptree || (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_OBJSET)) { if (dsl_scan_async_block_should_pause(scn)) return (SET_ERROR(ERESTART)); } - zio_nowait(zio_free_sync(scn->scn_zio_root, scn->scn_dp->dp_spa, - dmu_tx_get_txg(tx), bp, 0)); + zio_t *zio = zio_free_sync(scn->scn_zio_root, scn->scn_dp->dp_spa, + dmu_tx_get_txg(tx), bp, 0); dsl_dir_diduse_space(tx->tx_pool->dp_free_dir, DD_USED_HEAD, -bp_get_dsize_sync(scn->scn_dp->dp_spa, bp), -BP_GET_PSIZE(bp), -BP_GET_UCSIZE(bp), tx); scn->scn_visited_this_txg++; - if (BP_GET_DEDUP(bp)) - scn->scn_dedup_frees_this_txg++; + if (zio != NULL) { + /* + * zio_free_sync() returned a ZIO, meaning this is an + * async I/O (dedup, clone or gang block). + */ + scn->scn_async_frees_this_txg++; + zio_nowait(zio); + + /* + * After issuing N async ZIOs, wait for them to complete. + * This makes time limits work with actual I/O completion + * times, not just queuing times. + */ + uint64_t i = zfs_async_free_zio_wait_interval; + if (i != 0 && (scn->scn_async_frees_this_txg % i) == 0) { + VERIFY0(zio_wait(scn->scn_zio_root)); + scn->scn_zio_root = zio_root(scn->scn_dp->dp_spa, NULL, + NULL, ZIO_FLAG_MUSTSUCCEED); + } + } return (0); } static void dsl_scan_update_stats(dsl_scan_t *scn) { spa_t *spa = scn->scn_dp->dp_spa; uint64_t i; uint64_t seg_size_total = 0, zio_size_total = 0; uint64_t seg_count_total = 0, zio_count_total = 0; for (i = 0; i < spa->spa_root_vdev->vdev_children; i++) { vdev_t *vd = spa->spa_root_vdev->vdev_child[i]; dsl_scan_io_queue_t *queue = vd->vdev_scan_io_queue; if (queue == NULL) continue; seg_size_total += queue->q_total_seg_size_this_txg; zio_size_total += queue->q_total_zio_size_this_txg; seg_count_total += queue->q_segs_this_txg; zio_count_total += queue->q_zios_this_txg; } if (seg_count_total == 0 || zio_count_total == 0) { scn->scn_avg_seg_size_this_txg = 0; scn->scn_avg_zio_size_this_txg = 0; scn->scn_segs_this_txg = 0; scn->scn_zios_this_txg = 0; return; } scn->scn_avg_seg_size_this_txg = seg_size_total / seg_count_total; scn->scn_avg_zio_size_this_txg = zio_size_total / zio_count_total; scn->scn_segs_this_txg = seg_count_total; scn->scn_zios_this_txg = zio_count_total; } static int bpobj_dsl_scan_free_block_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, dmu_tx_t *tx) { ASSERT(!bp_freed); return (dsl_scan_free_block_cb(arg, bp, tx)); } static int dsl_scan_obsolete_block_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, dmu_tx_t *tx) { ASSERT(!bp_freed); dsl_scan_t *scn = arg; const dva_t *dva = &bp->blk_dva[0]; if (dsl_scan_async_block_should_pause(scn)) return (SET_ERROR(ERESTART)); spa_vdev_indirect_mark_obsolete(scn->scn_dp->dp_spa, DVA_GET_VDEV(dva), DVA_GET_OFFSET(dva), DVA_GET_ASIZE(dva), tx); scn->scn_visited_this_txg++; return (0); } boolean_t dsl_scan_active(dsl_scan_t *scn) { spa_t *spa = scn->scn_dp->dp_spa; uint64_t used = 0, comp, uncomp; boolean_t clones_left; if (spa->spa_load_state != SPA_LOAD_NONE) return (B_FALSE); if (spa_shutting_down(spa)) return (B_FALSE); if ((dsl_scan_is_running(scn) && !dsl_scan_is_paused_scrub(scn)) || (scn->scn_async_destroying && !scn->scn_async_stalled)) return (B_TRUE); if (spa_version(scn->scn_dp->dp_spa) >= SPA_VERSION_DEADLISTS) { (void) bpobj_space(&scn->scn_dp->dp_free_bpobj, &used, &comp, &uncomp); } clones_left = spa_livelist_delete_check(spa); return ((used != 0) || (clones_left)); } boolean_t dsl_errorscrub_active(dsl_scan_t *scn) { spa_t *spa = scn->scn_dp->dp_spa; if (spa->spa_load_state != SPA_LOAD_NONE) return (B_FALSE); if (spa_shutting_down(spa)) return (B_FALSE); if (dsl_errorscrubbing(scn->scn_dp)) return (B_TRUE); return (B_FALSE); } static boolean_t dsl_scan_check_deferred(vdev_t *vd) { boolean_t need_resilver = B_FALSE; for (int c = 0; c < vd->vdev_children; c++) { need_resilver |= dsl_scan_check_deferred(vd->vdev_child[c]); } if (!vdev_is_concrete(vd) || vd->vdev_aux || !vd->vdev_ops->vdev_op_leaf) return (need_resilver); if (!vd->vdev_resilver_deferred) need_resilver = B_TRUE; return (need_resilver); } static boolean_t dsl_scan_need_resilver(spa_t *spa, const dva_t *dva, size_t psize, uint64_t phys_birth) { vdev_t *vd; vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva)); if (vd->vdev_ops == &vdev_indirect_ops) { /* * The indirect vdev can point to multiple * vdevs. For simplicity, always create * the resilver zio_t. zio_vdev_io_start() * will bypass the child resilver i/o's if * they are on vdevs that don't have DTL's. */ return (B_TRUE); } if (DVA_GET_GANG(dva)) { /* * Gang members may be spread across multiple * vdevs, so the best estimate we have is the * scrub range, which has already been checked. * XXX -- it would be better to change our * allocation policy to ensure that all * gang members reside on the same vdev. */ return (B_TRUE); } /* * Check if the top-level vdev must resilver this offset. * When the offset does not intersect with a dirty leaf DTL * then it may be possible to skip the resilver IO. The psize * is provided instead of asize to simplify the check for RAIDZ. */ if (!vdev_dtl_need_resilver(vd, dva, psize, phys_birth)) return (B_FALSE); /* * Check that this top-level vdev has a device under it which * is resilvering and is not deferred. */ if (!dsl_scan_check_deferred(vd)) return (B_FALSE); return (B_TRUE); } static int dsl_process_async_destroys(dsl_pool_t *dp, dmu_tx_t *tx) { dsl_scan_t *scn = dp->dp_scan; spa_t *spa = dp->dp_spa; int err = 0; if (spa_suspend_async_destroy(spa)) return (0); if (zfs_free_bpobj_enabled && spa_version(spa) >= SPA_VERSION_DEADLISTS) { scn->scn_is_bptree = B_FALSE; scn->scn_async_block_min_time_ms = zfs_free_min_time_ms; scn->scn_zio_root = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); err = bpobj_iterate(&dp->dp_free_bpobj, bpobj_dsl_scan_free_block_cb, scn, tx); VERIFY0(zio_wait(scn->scn_zio_root)); scn->scn_zio_root = NULL; if (err != 0 && err != ERESTART) zfs_panic_recover("error %u from bpobj_iterate()", err); } if (err == 0 && spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY)) { ASSERT(scn->scn_async_destroying); scn->scn_is_bptree = B_TRUE; scn->scn_zio_root = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); err = bptree_iterate(dp->dp_meta_objset, dp->dp_bptree_obj, B_TRUE, dsl_scan_free_block_cb, scn, tx); VERIFY0(zio_wait(scn->scn_zio_root)); scn->scn_zio_root = NULL; if (err == EIO || err == ECKSUM) { err = 0; } else if (err != 0 && err != ERESTART) { zfs_panic_recover("error %u from " "traverse_dataset_destroyed()", err); } if (bptree_is_empty(dp->dp_meta_objset, dp->dp_bptree_obj)) { /* finished; deactivate async destroy feature */ spa_feature_decr(spa, SPA_FEATURE_ASYNC_DESTROY, tx); ASSERT(!spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY)); VERIFY0(zap_remove(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_BPTREE_OBJ, tx)); VERIFY0(bptree_free(dp->dp_meta_objset, dp->dp_bptree_obj, tx)); dp->dp_bptree_obj = 0; scn->scn_async_destroying = B_FALSE; scn->scn_async_stalled = B_FALSE; } else { /* * If we didn't make progress, mark the async * destroy as stalled, so that we will not initiate * a spa_sync() on its behalf. Note that we only * check this if we are not finished, because if the * bptree had no blocks for us to visit, we can * finish without "making progress". */ scn->scn_async_stalled = (scn->scn_visited_this_txg == 0); } } if (scn->scn_visited_this_txg) { zfs_dbgmsg("freed %llu blocks in %llums from " "free_bpobj/bptree on %s in txg %llu; err=%u", (longlong_t)scn->scn_visited_this_txg, (longlong_t) NSEC2MSEC(gethrtime() - scn->scn_sync_start_time), spa->spa_name, (longlong_t)tx->tx_txg, err); scn->scn_visited_this_txg = 0; - scn->scn_dedup_frees_this_txg = 0; + scn->scn_async_frees_this_txg = 0; /* * Write out changes to the DDT and the BRT that may be required * as a result of the blocks freed. This ensures that the DDT * and the BRT are clean when a scrub/resilver runs. */ ddt_sync(spa, tx->tx_txg); brt_sync(spa, tx->tx_txg); } if (err != 0) return (err); if (dp->dp_free_dir != NULL && !scn->scn_async_destroying && zfs_free_leak_on_eio && (dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes != 0 || dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes != 0 || dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes != 0)) { /* * We have finished background destroying, but there is still * some space left in the dp_free_dir. Transfer this leaked * space to the dp_leak_dir. */ if (dp->dp_leak_dir == NULL) { rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); (void) dsl_dir_create_sync(dp, dp->dp_root_dir, LEAK_DIR_NAME, tx); VERIFY0(dsl_pool_open_special_dir(dp, LEAK_DIR_NAME, &dp->dp_leak_dir)); rrw_exit(&dp->dp_config_rwlock, FTAG); } dsl_dir_diduse_space(dp->dp_leak_dir, DD_USED_HEAD, dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes, dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes, dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes, tx); dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD, -dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes, -dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes, -dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes, tx); } if (dp->dp_free_dir != NULL && !scn->scn_async_destroying && !spa_livelist_delete_check(spa)) { /* finished; verify that space accounting went to zero */ ASSERT0(dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes); ASSERT0(dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes); ASSERT0(dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes); } spa_notify_waiters(spa); EQUIV(bpobj_is_open(&dp->dp_obsolete_bpobj), 0 == zap_contains(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_OBSOLETE_BPOBJ)); if (err == 0 && bpobj_is_open(&dp->dp_obsolete_bpobj)) { ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_OBSOLETE_COUNTS)); scn->scn_is_bptree = B_FALSE; scn->scn_async_block_min_time_ms = zfs_obsolete_min_time_ms; err = bpobj_iterate(&dp->dp_obsolete_bpobj, dsl_scan_obsolete_block_cb, scn, tx); if (err != 0 && err != ERESTART) zfs_panic_recover("error %u from bpobj_iterate()", err); if (bpobj_is_empty(&dp->dp_obsolete_bpobj)) dsl_pool_destroy_obsolete_bpobj(dp, tx); } return (0); } static void name_to_bookmark(char *buf, zbookmark_phys_t *zb) { zb->zb_objset = zfs_strtonum(buf, &buf); ASSERT(*buf == ':'); zb->zb_object = zfs_strtonum(buf + 1, &buf); ASSERT(*buf == ':'); zb->zb_level = (int)zfs_strtonum(buf + 1, &buf); ASSERT(*buf == ':'); zb->zb_blkid = zfs_strtonum(buf + 1, &buf); ASSERT(*buf == '\0'); } static void name_to_object(char *buf, uint64_t *obj) { *obj = zfs_strtonum(buf, &buf); ASSERT(*buf == '\0'); } static void read_by_block_level(dsl_scan_t *scn, zbookmark_phys_t zb) { dsl_pool_t *dp = scn->scn_dp; dsl_dataset_t *ds; objset_t *os; if (dsl_dataset_hold_obj(dp, zb.zb_objset, FTAG, &ds) != 0) return; if (dmu_objset_from_ds(ds, &os) != 0) { dsl_dataset_rele(ds, FTAG); return; } /* * If the key is not loaded dbuf_dnode_findbp() will error out with * EACCES. However in that case dnode_hold() will eventually call * dbuf_read()->zio_wait() which may call spa_log_error(). This will * lead to a deadlock due to us holding the mutex spa_errlist_lock. * Avoid this by checking here if the keys are loaded, if not return. * If the keys are not loaded the head_errlog feature is meaningless * as we cannot figure out the birth txg of the block pointer. */ if (dsl_dataset_get_keystatus(ds->ds_dir) == ZFS_KEYSTATUS_UNAVAILABLE) { dsl_dataset_rele(ds, FTAG); return; } dnode_t *dn; blkptr_t bp; if (dnode_hold(os, zb.zb_object, FTAG, &dn) != 0) { dsl_dataset_rele(ds, FTAG); return; } rw_enter(&dn->dn_struct_rwlock, RW_READER); int error = dbuf_dnode_findbp(dn, zb.zb_level, zb.zb_blkid, &bp, NULL, NULL); if (error) { rw_exit(&dn->dn_struct_rwlock); dnode_rele(dn, FTAG); dsl_dataset_rele(ds, FTAG); return; } if (!error && BP_IS_HOLE(&bp)) { rw_exit(&dn->dn_struct_rwlock); dnode_rele(dn, FTAG); dsl_dataset_rele(ds, FTAG); return; } int zio_flags = ZIO_FLAG_SCAN_THREAD | ZIO_FLAG_RAW | ZIO_FLAG_CANFAIL | ZIO_FLAG_SCRUB; /* If it's an intent log block, failure is expected. */ if (zb.zb_level == ZB_ZIL_LEVEL) zio_flags |= ZIO_FLAG_SPECULATIVE; ASSERT(!BP_IS_EMBEDDED(&bp)); scan_exec_io(dp, &bp, zio_flags, &zb, NULL); rw_exit(&dn->dn_struct_rwlock); dnode_rele(dn, FTAG); dsl_dataset_rele(ds, FTAG); } /* * We keep track of the scrubbed error blocks in "count". This will be used * when deciding whether we exceeded zfs_scrub_error_blocks_per_txg. This * function is modelled after check_filesystem(). */ static int scrub_filesystem(spa_t *spa, uint64_t fs, zbookmark_err_phys_t *zep, int *count) { dsl_dataset_t *ds; dsl_pool_t *dp = spa->spa_dsl_pool; dsl_scan_t *scn = dp->dp_scan; int error = dsl_dataset_hold_obj(dp, fs, FTAG, &ds); if (error != 0) return (error); uint64_t latest_txg; uint64_t txg_to_consider = spa->spa_syncing_txg; boolean_t check_snapshot = B_TRUE; error = find_birth_txg(ds, zep, &latest_txg); /* * If find_birth_txg() errors out, then err on the side of caution and * proceed. In worst case scenario scrub all objects. If zep->zb_birth * is 0 (e.g. in case of encryption with unloaded keys) also proceed to * scrub all objects. */ if (error == 0 && zep->zb_birth == latest_txg) { /* Block neither free nor re written. */ zbookmark_phys_t zb; zep_to_zb(fs, zep, &zb); scn->scn_zio_root = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL); /* We have already acquired the config lock for spa */ read_by_block_level(scn, zb); (void) zio_wait(scn->scn_zio_root); scn->scn_zio_root = NULL; scn->errorscrub_phys.dep_examined++; scn->errorscrub_phys.dep_to_examine--; (*count)++; if ((*count) == zfs_scrub_error_blocks_per_txg || dsl_error_scrub_check_suspend(scn, &zb)) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(EFAULT)); } check_snapshot = B_FALSE; } else if (error == 0) { txg_to_consider = latest_txg; } /* * Retrieve the number of snapshots if the dataset is not a snapshot. */ uint64_t snap_count = 0; if (dsl_dataset_phys(ds)->ds_snapnames_zapobj != 0) { error = zap_count(spa->spa_meta_objset, dsl_dataset_phys(ds)->ds_snapnames_zapobj, &snap_count); if (error != 0) { dsl_dataset_rele(ds, FTAG); return (error); } } if (snap_count == 0) { /* Filesystem without snapshots. */ dsl_dataset_rele(ds, FTAG); return (0); } uint64_t snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; uint64_t snap_obj_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg; dsl_dataset_rele(ds, FTAG); /* Check only snapshots created from this file system. */ while (snap_obj != 0 && zep->zb_birth < snap_obj_txg && snap_obj_txg <= txg_to_consider) { error = dsl_dataset_hold_obj(dp, snap_obj, FTAG, &ds); if (error != 0) return (error); if (dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj != fs) { snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; snap_obj_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg; dsl_dataset_rele(ds, FTAG); continue; } boolean_t affected = B_TRUE; if (check_snapshot) { uint64_t blk_txg; error = find_birth_txg(ds, zep, &blk_txg); /* * Scrub the snapshot also when zb_birth == 0 or when * find_birth_txg() returns an error. */ affected = (error == 0 && zep->zb_birth == blk_txg) || (error != 0) || (zep->zb_birth == 0); } /* Scrub snapshots. */ if (affected) { zbookmark_phys_t zb; zep_to_zb(snap_obj, zep, &zb); scn->scn_zio_root = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL); /* We have already acquired the config lock for spa */ read_by_block_level(scn, zb); (void) zio_wait(scn->scn_zio_root); scn->scn_zio_root = NULL; scn->errorscrub_phys.dep_examined++; scn->errorscrub_phys.dep_to_examine--; (*count)++; if ((*count) == zfs_scrub_error_blocks_per_txg || dsl_error_scrub_check_suspend(scn, &zb)) { dsl_dataset_rele(ds, FTAG); return (EFAULT); } } snap_obj_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg; snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; dsl_dataset_rele(ds, FTAG); } return (0); } void dsl_errorscrub_sync(dsl_pool_t *dp, dmu_tx_t *tx) { spa_t *spa = dp->dp_spa; dsl_scan_t *scn = dp->dp_scan; /* * Only process scans in sync pass 1. */ if (spa_sync_pass(spa) > 1) return; /* * If the spa is shutting down, then stop scanning. This will * ensure that the scan does not dirty any new data during the * shutdown phase. */ if (spa_shutting_down(spa)) return; if (!dsl_errorscrub_active(scn) || dsl_errorscrub_is_paused(scn)) { return; } if (dsl_scan_resilvering(scn->scn_dp)) { /* cancel the error scrub if resilver started */ dsl_scan_cancel(scn->scn_dp); return; } spa->spa_scrub_active = B_TRUE; scn->scn_sync_start_time = gethrtime(); /* * zfs_scan_suspend_progress can be set to disable scrub progress. * See more detailed comment in dsl_scan_sync(). */ if (zfs_scan_suspend_progress) { uint64_t scan_time_ns = gethrtime() - scn->scn_sync_start_time; int mintime = zfs_scrub_min_time_ms; while (zfs_scan_suspend_progress && !txg_sync_waiting(scn->scn_dp) && !spa_shutting_down(scn->scn_dp->dp_spa) && NSEC2MSEC(scan_time_ns) < mintime) { delay(hz); scan_time_ns = gethrtime() - scn->scn_sync_start_time; } return; } int i = 0; zap_attribute_t *za; zbookmark_phys_t *zb; boolean_t limit_exceeded = B_FALSE; za = zap_attribute_alloc(); zb = kmem_zalloc(sizeof (zbookmark_phys_t), KM_SLEEP); if (!spa_feature_is_enabled(spa, SPA_FEATURE_HEAD_ERRLOG)) { for (; zap_cursor_retrieve(&scn->errorscrub_cursor, za) == 0; zap_cursor_advance(&scn->errorscrub_cursor)) { name_to_bookmark(za->za_name, zb); scn->scn_zio_root = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_CANFAIL); dsl_pool_config_enter(dp, FTAG); read_by_block_level(scn, *zb); dsl_pool_config_exit(dp, FTAG); (void) zio_wait(scn->scn_zio_root); scn->scn_zio_root = NULL; scn->errorscrub_phys.dep_examined += 1; scn->errorscrub_phys.dep_to_examine -= 1; i++; if (i == zfs_scrub_error_blocks_per_txg || dsl_error_scrub_check_suspend(scn, zb)) { limit_exceeded = B_TRUE; break; } } if (!limit_exceeded) dsl_errorscrub_done(scn, B_TRUE, tx); dsl_errorscrub_sync_state(scn, tx); zap_attribute_free(za); kmem_free(zb, sizeof (*zb)); return; } int error = 0; for (; zap_cursor_retrieve(&scn->errorscrub_cursor, za) == 0; zap_cursor_advance(&scn->errorscrub_cursor)) { zap_cursor_t *head_ds_cursor; zap_attribute_t *head_ds_attr; zbookmark_err_phys_t head_ds_block; head_ds_cursor = kmem_zalloc(sizeof (zap_cursor_t), KM_SLEEP); head_ds_attr = zap_attribute_alloc(); uint64_t head_ds_err_obj = za->za_first_integer; uint64_t head_ds; name_to_object(za->za_name, &head_ds); boolean_t config_held = B_FALSE; uint64_t top_affected_fs; for (zap_cursor_init(head_ds_cursor, spa->spa_meta_objset, head_ds_err_obj); zap_cursor_retrieve(head_ds_cursor, head_ds_attr) == 0; zap_cursor_advance(head_ds_cursor)) { name_to_errphys(head_ds_attr->za_name, &head_ds_block); /* * In case we are called from spa_sync the pool * config is already held. */ if (!dsl_pool_config_held(dp)) { dsl_pool_config_enter(dp, FTAG); config_held = B_TRUE; } error = find_top_affected_fs(spa, head_ds, &head_ds_block, &top_affected_fs); if (error) break; error = scrub_filesystem(spa, top_affected_fs, &head_ds_block, &i); if (error == SET_ERROR(EFAULT)) { limit_exceeded = B_TRUE; break; } } zap_cursor_fini(head_ds_cursor); kmem_free(head_ds_cursor, sizeof (*head_ds_cursor)); zap_attribute_free(head_ds_attr); if (config_held) dsl_pool_config_exit(dp, FTAG); } zap_attribute_free(za); kmem_free(zb, sizeof (*zb)); if (!limit_exceeded) dsl_errorscrub_done(scn, B_TRUE, tx); dsl_errorscrub_sync_state(scn, tx); } /* * This is the primary entry point for scans that is called from syncing * context. Scans must happen entirely during syncing context so that we * can guarantee that blocks we are currently scanning will not change out * from under us. While a scan is active, this function controls how quickly * transaction groups proceed, instead of the normal handling provided by * txg_sync_thread(). */ void dsl_scan_sync(dsl_pool_t *dp, dmu_tx_t *tx) { int err = 0; dsl_scan_t *scn = dp->dp_scan; spa_t *spa = dp->dp_spa; state_sync_type_t sync_type = SYNC_OPTIONAL; int restart_early = 0; if (spa->spa_resilver_deferred) { uint64_t to_issue, issued; if (!spa_feature_is_active(dp->dp_spa, SPA_FEATURE_RESILVER_DEFER)) spa_feature_incr(spa, SPA_FEATURE_RESILVER_DEFER, tx); /* * See print_scan_scrub_resilver_status() issued/total_i * @ cmd/zpool/zpool_main.c */ to_issue = scn->scn_phys.scn_to_examine - scn->scn_phys.scn_skipped; issued = scn->scn_issued_before_pass + spa->spa_scan_pass_issued; restart_early = zfs_resilver_disable_defer || (issued < (to_issue * zfs_resilver_defer_percent / 100)); } /* * Only process scans in sync pass 1. */ if (spa_sync_pass(spa) > 1) return; /* * Check for scn_restart_txg before checking spa_load_state, so * that we can restart an old-style scan while the pool is being * imported (see dsl_scan_init). We also restart scans if there * is a deferred resilver and the user has manually disabled * deferred resilvers via zfs_resilver_disable_defer, or if the * current scan progress is below zfs_resilver_defer_percent. */ if (dsl_scan_restarting(scn, tx) || restart_early) { setup_sync_arg_t setup_sync_arg = { .func = POOL_SCAN_SCRUB, .txgstart = 0, .txgend = 0, }; dsl_scan_done(scn, B_FALSE, tx); if (vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) setup_sync_arg.func = POOL_SCAN_RESILVER; zfs_dbgmsg("restarting scan func=%u on %s txg=%llu early=%d", setup_sync_arg.func, dp->dp_spa->spa_name, (longlong_t)tx->tx_txg, restart_early); dsl_scan_setup_sync(&setup_sync_arg, tx); } /* * If the spa is shutting down, then stop scanning. This will * ensure that the scan does not dirty any new data during the * shutdown phase. */ if (spa_shutting_down(spa)) return; /* * Wait a few txgs after importing before doing background work * (async destroys and scanning). This should help the import * command to complete quickly. */ if (spa->spa_syncing_txg < spa->spa_first_txg + zfs_import_defer_txgs) return; /* * If the scan is inactive due to a stalled async destroy, try again. */ if (!scn->scn_async_stalled && !dsl_scan_active(scn)) return; /* reset scan statistics */ scn->scn_visited_this_txg = 0; - scn->scn_dedup_frees_this_txg = 0; + scn->scn_async_frees_this_txg = 0; scn->scn_holes_this_txg = 0; scn->scn_lt_min_this_txg = 0; scn->scn_gt_max_this_txg = 0; scn->scn_ddt_contained_this_txg = 0; scn->scn_objsets_visited_this_txg = 0; scn->scn_avg_seg_size_this_txg = 0; scn->scn_segs_this_txg = 0; scn->scn_avg_zio_size_this_txg = 0; scn->scn_zios_this_txg = 0; scn->scn_suspending = B_FALSE; scn->scn_sync_start_time = gethrtime(); spa->spa_scrub_active = B_TRUE; /* * First process the async destroys. If we suspend, don't do * any scrubbing or resilvering. This ensures that there are no * async destroys while we are scanning, so the scan code doesn't * have to worry about traversing it. It is also faster to free the * blocks than to scrub them. */ err = dsl_process_async_destroys(dp, tx); if (err != 0) return; if (!dsl_scan_is_running(scn) || dsl_scan_is_paused_scrub(scn)) return; /* * zfs_scan_suspend_progress can be set to disable scan progress. * We don't want to spin the txg_sync thread, so we add a delay * here to simulate the time spent doing a scan. This is mostly * useful for testing and debugging. */ if (zfs_scan_suspend_progress) { uint64_t scan_time_ns = gethrtime() - scn->scn_sync_start_time; uint_t mintime = (scn->scn_phys.scn_func == POOL_SCAN_RESILVER) ? zfs_resilver_min_time_ms : zfs_scrub_min_time_ms; while (zfs_scan_suspend_progress && !txg_sync_waiting(scn->scn_dp) && !spa_shutting_down(scn->scn_dp->dp_spa) && NSEC2MSEC(scan_time_ns) < mintime) { delay(hz); scan_time_ns = gethrtime() - scn->scn_sync_start_time; } return; } /* * Disabled by default, set zfs_scan_report_txgs to report * average performance over the last zfs_scan_report_txgs TXGs. */ if (zfs_scan_report_txgs != 0 && tx->tx_txg % zfs_scan_report_txgs == 0) { scn->scn_issued_before_pass += spa->spa_scan_pass_issued; spa_scan_stat_init(spa); } /* * It is possible to switch from unsorted to sorted at any time, * but afterwards the scan will remain sorted unless reloaded from * a checkpoint after a reboot. */ if (!zfs_scan_legacy) { scn->scn_is_sorted = B_TRUE; if (scn->scn_last_checkpoint == 0) scn->scn_last_checkpoint = ddi_get_lbolt(); } /* * For sorted scans, determine what kind of work we will be doing * this txg based on our memory limitations and whether or not we * need to perform a checkpoint. */ if (scn->scn_is_sorted) { /* * If we are over our checkpoint interval, set scn_clearing * so that we can begin checkpointing immediately. The * checkpoint allows us to save a consistent bookmark * representing how much data we have scrubbed so far. * Otherwise, use the memory limit to determine if we should * scan for metadata or start issue scrub IOs. We accumulate * metadata until we hit our hard memory limit at which point * we issue scrub IOs until we are at our soft memory limit. */ if (scn->scn_checkpointing || ddi_get_lbolt() - scn->scn_last_checkpoint > SEC_TO_TICK(zfs_scan_checkpoint_intval)) { if (!scn->scn_checkpointing) zfs_dbgmsg("begin scan checkpoint for %s", spa->spa_name); scn->scn_checkpointing = B_TRUE; scn->scn_clearing = B_TRUE; } else { boolean_t should_clear = dsl_scan_should_clear(scn); if (should_clear && !scn->scn_clearing) { zfs_dbgmsg("begin scan clearing for %s", spa->spa_name); scn->scn_clearing = B_TRUE; } else if (!should_clear && scn->scn_clearing) { zfs_dbgmsg("finish scan clearing for %s", spa->spa_name); scn->scn_clearing = B_FALSE; } } } else { ASSERT0(scn->scn_checkpointing); ASSERT0(scn->scn_clearing); } if (!scn->scn_clearing && scn->scn_done_txg == 0) { /* Need to scan metadata for more blocks to scrub */ dsl_scan_phys_t *scnp = &scn->scn_phys; taskqid_t prefetch_tqid; /* * Calculate the max number of in-flight bytes for pool-wide * scanning operations (minimum 1MB, maximum 1/4 of arc_c_max). * Limits for the issuing phase are done per top-level vdev and * are handled separately. */ scn->scn_maxinflight_bytes = MIN(arc_c_max / 4, MAX(1ULL << 20, zfs_scan_vdev_limit * dsl_scan_count_data_disks(spa))); if (scnp->scn_ddt_bookmark.ddb_class <= scnp->scn_ddt_class_max) { ASSERT(ZB_IS_ZERO(&scnp->scn_bookmark)); zfs_dbgmsg("doing scan sync for %s txg %llu; " "ddt bm=%llu/%llu/%llu/%llx", spa->spa_name, (longlong_t)tx->tx_txg, (longlong_t)scnp->scn_ddt_bookmark.ddb_class, (longlong_t)scnp->scn_ddt_bookmark.ddb_type, (longlong_t)scnp->scn_ddt_bookmark.ddb_checksum, (longlong_t)scnp->scn_ddt_bookmark.ddb_cursor); } else { zfs_dbgmsg("doing scan sync for %s txg %llu; " "bm=%llu/%llu/%llu/%llu", spa->spa_name, (longlong_t)tx->tx_txg, (longlong_t)scnp->scn_bookmark.zb_objset, (longlong_t)scnp->scn_bookmark.zb_object, (longlong_t)scnp->scn_bookmark.zb_level, (longlong_t)scnp->scn_bookmark.zb_blkid); } scn->scn_zio_root = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_CANFAIL); scn->scn_prefetch_stop = B_FALSE; prefetch_tqid = taskq_dispatch(dp->dp_sync_taskq, dsl_scan_prefetch_thread, scn, TQ_SLEEP); ASSERT(prefetch_tqid != TASKQID_INVALID); dsl_pool_config_enter(dp, FTAG); dsl_scan_visit(scn, tx); dsl_pool_config_exit(dp, FTAG); mutex_enter(&dp->dp_spa->spa_scrub_lock); scn->scn_prefetch_stop = B_TRUE; cv_broadcast(&spa->spa_scrub_io_cv); mutex_exit(&dp->dp_spa->spa_scrub_lock); taskq_wait_id(dp->dp_sync_taskq, prefetch_tqid); (void) zio_wait(scn->scn_zio_root); scn->scn_zio_root = NULL; zfs_dbgmsg("scan visited %llu blocks of %s in %llums " "(%llu os's, %llu holes, %llu < mintxg, " "%llu in ddt, %llu > maxtxg)", (longlong_t)scn->scn_visited_this_txg, spa->spa_name, (longlong_t)NSEC2MSEC(gethrtime() - scn->scn_sync_start_time), (longlong_t)scn->scn_objsets_visited_this_txg, (longlong_t)scn->scn_holes_this_txg, (longlong_t)scn->scn_lt_min_this_txg, (longlong_t)scn->scn_ddt_contained_this_txg, (longlong_t)scn->scn_gt_max_this_txg); if (!scn->scn_suspending) { ASSERT0(avl_numnodes(&scn->scn_queue)); scn->scn_done_txg = tx->tx_txg + 1; if (scn->scn_is_sorted) { scn->scn_checkpointing = B_TRUE; scn->scn_clearing = B_TRUE; scn->scn_issued_before_pass += spa->spa_scan_pass_issued; spa_scan_stat_init(spa); } zfs_dbgmsg("scan complete for %s txg %llu", spa->spa_name, (longlong_t)tx->tx_txg); } } else if (scn->scn_is_sorted && scn->scn_queues_pending != 0) { ASSERT(scn->scn_clearing); /* need to issue scrubbing IOs from per-vdev queues */ scn->scn_zio_root = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_CANFAIL); scan_io_queues_run(scn); (void) zio_wait(scn->scn_zio_root); scn->scn_zio_root = NULL; /* calculate and dprintf the current memory usage */ (void) dsl_scan_should_clear(scn); dsl_scan_update_stats(scn); zfs_dbgmsg("scan issued %llu blocks for %s (%llu segs) " "in %llums (avg_block_size = %llu, avg_seg_size = %llu)", (longlong_t)scn->scn_zios_this_txg, spa->spa_name, (longlong_t)scn->scn_segs_this_txg, (longlong_t)NSEC2MSEC(gethrtime() - scn->scn_sync_start_time), (longlong_t)scn->scn_avg_zio_size_this_txg, (longlong_t)scn->scn_avg_seg_size_this_txg); } else if (scn->scn_done_txg != 0 && scn->scn_done_txg <= tx->tx_txg) { /* Finished with everything. Mark the scrub as complete */ zfs_dbgmsg("scan issuing complete txg %llu for %s", (longlong_t)tx->tx_txg, spa->spa_name); ASSERT3U(scn->scn_done_txg, !=, 0); ASSERT0(spa->spa_scrub_inflight); ASSERT0(scn->scn_queues_pending); dsl_scan_done(scn, B_TRUE, tx); sync_type = SYNC_MANDATORY; } dsl_scan_sync_state(scn, tx, sync_type); } static void count_block_issued(spa_t *spa, const blkptr_t *bp, boolean_t all) { /* * Don't count embedded bp's, since we already did the work of * scanning these when we scanned the containing block. */ if (BP_IS_EMBEDDED(bp)) return; /* * Update the spa's stats on how many bytes we have issued. * Sequential scrubs create a zio for each DVA of the bp. Each * of these will include all DVAs for repair purposes, but the * zio code will only try the first one unless there is an issue. * Therefore, we should only count the first DVA for these IOs. */ atomic_add_64(&spa->spa_scan_pass_issued, all ? BP_GET_ASIZE(bp) : DVA_GET_ASIZE(&bp->blk_dva[0])); } static void count_block_skipped(dsl_scan_t *scn, const blkptr_t *bp, boolean_t all) { if (BP_IS_EMBEDDED(bp)) return; atomic_add_64(&scn->scn_phys.scn_skipped, all ? BP_GET_ASIZE(bp) : DVA_GET_ASIZE(&bp->blk_dva[0])); } static void count_block(zfs_all_blkstats_t *zab, const blkptr_t *bp) { /* * If we resume after a reboot, zab will be NULL; don't record * incomplete stats in that case. */ if (zab == NULL) return; for (int i = 0; i < 4; i++) { int l = (i < 2) ? BP_GET_LEVEL(bp) : DN_MAX_LEVELS; int t = (i & 1) ? BP_GET_TYPE(bp) : DMU_OT_TOTAL; if (t & DMU_OT_NEWTYPE) t = DMU_OT_OTHER; zfs_blkstat_t *zb = &zab->zab_type[l][t]; int equal; zb->zb_count++; zb->zb_asize += BP_GET_ASIZE(bp); zb->zb_lsize += BP_GET_LSIZE(bp); zb->zb_psize += BP_GET_PSIZE(bp); zb->zb_gangs += BP_COUNT_GANG(bp); switch (BP_GET_NDVAS(bp)) { case 2: if (DVA_GET_VDEV(&bp->blk_dva[0]) == DVA_GET_VDEV(&bp->blk_dva[1])) zb->zb_ditto_2_of_2_samevdev++; break; case 3: equal = (DVA_GET_VDEV(&bp->blk_dva[0]) == DVA_GET_VDEV(&bp->blk_dva[1])) + (DVA_GET_VDEV(&bp->blk_dva[0]) == DVA_GET_VDEV(&bp->blk_dva[2])) + (DVA_GET_VDEV(&bp->blk_dva[1]) == DVA_GET_VDEV(&bp->blk_dva[2])); if (equal == 1) zb->zb_ditto_2_of_3_samevdev++; else if (equal == 3) zb->zb_ditto_3_of_3_samevdev++; break; } } } static void scan_io_queue_insert_impl(dsl_scan_io_queue_t *queue, scan_io_t *sio) { avl_index_t idx; dsl_scan_t *scn = queue->q_scn; ASSERT(MUTEX_HELD(&queue->q_vd->vdev_scan_io_queue_lock)); if (unlikely(avl_is_empty(&queue->q_sios_by_addr))) atomic_add_64(&scn->scn_queues_pending, 1); if (avl_find(&queue->q_sios_by_addr, sio, &idx) != NULL) { /* block is already scheduled for reading */ sio_free(sio); return; } avl_insert(&queue->q_sios_by_addr, sio, idx); queue->q_sio_memused += SIO_GET_MUSED(sio); zfs_range_tree_add(queue->q_exts_by_addr, SIO_GET_OFFSET(sio), SIO_GET_ASIZE(sio)); } /* * Given all the info we got from our metadata scanning process, we * construct a scan_io_t and insert it into the scan sorting queue. The * I/O must already be suitable for us to process. This is controlled * by dsl_scan_enqueue(). */ static void scan_io_queue_insert(dsl_scan_io_queue_t *queue, const blkptr_t *bp, int dva_i, int zio_flags, const zbookmark_phys_t *zb) { scan_io_t *sio = sio_alloc(BP_GET_NDVAS(bp)); ASSERT0(BP_IS_GANG(bp)); ASSERT(MUTEX_HELD(&queue->q_vd->vdev_scan_io_queue_lock)); bp2sio(bp, sio, dva_i); sio->sio_flags = zio_flags; sio->sio_zb = *zb; queue->q_last_ext_addr = -1; scan_io_queue_insert_impl(queue, sio); } /* * Given a set of I/O parameters as discovered by the metadata traversal * process, attempts to place the I/O into the sorted queues (if allowed), * or immediately executes the I/O. */ static void dsl_scan_enqueue(dsl_pool_t *dp, const blkptr_t *bp, int zio_flags, const zbookmark_phys_t *zb) { spa_t *spa = dp->dp_spa; ASSERT(!BP_IS_EMBEDDED(bp)); /* * Gang blocks are hard to issue sequentially, so we just issue them * here immediately instead of queuing them. */ if (!dp->dp_scan->scn_is_sorted || BP_IS_GANG(bp)) { scan_exec_io(dp, bp, zio_flags, zb, NULL); return; } for (int i = 0; i < BP_GET_NDVAS(bp); i++) { dva_t dva; vdev_t *vdev; dva = bp->blk_dva[i]; vdev = vdev_lookup_top(spa, DVA_GET_VDEV(&dva)); ASSERT(vdev != NULL); mutex_enter(&vdev->vdev_scan_io_queue_lock); if (vdev->vdev_scan_io_queue == NULL) vdev->vdev_scan_io_queue = scan_io_queue_create(vdev); ASSERT(dp->dp_scan != NULL); scan_io_queue_insert(vdev->vdev_scan_io_queue, bp, i, zio_flags, zb); mutex_exit(&vdev->vdev_scan_io_queue_lock); } } static int dsl_scan_scrub_cb(dsl_pool_t *dp, const blkptr_t *bp, const zbookmark_phys_t *zb) { dsl_scan_t *scn = dp->dp_scan; spa_t *spa = dp->dp_spa; uint64_t phys_birth = BP_GET_PHYSICAL_BIRTH(bp); size_t psize = BP_GET_PSIZE(bp); boolean_t needs_io = B_FALSE; int zio_flags = ZIO_FLAG_SCAN_THREAD | ZIO_FLAG_RAW | ZIO_FLAG_CANFAIL; count_block(dp->dp_blkstats, bp); if (phys_birth <= scn->scn_phys.scn_min_txg || phys_birth >= scn->scn_phys.scn_max_txg) { count_block_skipped(scn, bp, B_TRUE); return (0); } /* Embedded BP's have phys_birth==0, so we reject them above. */ ASSERT(!BP_IS_EMBEDDED(bp)); ASSERT(DSL_SCAN_IS_SCRUB_RESILVER(scn)); if (scn->scn_phys.scn_func == POOL_SCAN_SCRUB) { zio_flags |= ZIO_FLAG_SCRUB; needs_io = B_TRUE; } else { ASSERT3U(scn->scn_phys.scn_func, ==, POOL_SCAN_RESILVER); zio_flags |= ZIO_FLAG_RESILVER; needs_io = B_FALSE; } /* If it's an intent log block, failure is expected. */ if (zb->zb_level == ZB_ZIL_LEVEL) zio_flags |= ZIO_FLAG_SPECULATIVE; for (int d = 0; d < BP_GET_NDVAS(bp); d++) { const dva_t *dva = &bp->blk_dva[d]; /* * Keep track of how much data we've examined so that * zpool(8) status can make useful progress reports. */ uint64_t asize = DVA_GET_ASIZE(dva); scn->scn_phys.scn_examined += asize; spa->spa_scan_pass_exam += asize; /* if it's a resilver, this may not be in the target range */ if (!needs_io) needs_io = dsl_scan_need_resilver(spa, dva, psize, phys_birth); } if (needs_io && !zfs_no_scrub_io) { dsl_scan_enqueue(dp, bp, zio_flags, zb); } else { count_block_skipped(scn, bp, B_TRUE); } /* do not relocate this block */ return (0); } static void dsl_scan_scrub_done(zio_t *zio) { spa_t *spa = zio->io_spa; blkptr_t *bp = zio->io_bp; dsl_scan_io_queue_t *queue = zio->io_private; abd_free(zio->io_abd); if (queue == NULL) { mutex_enter(&spa->spa_scrub_lock); ASSERT3U(spa->spa_scrub_inflight, >=, BP_GET_PSIZE(bp)); spa->spa_scrub_inflight -= BP_GET_PSIZE(bp); cv_broadcast(&spa->spa_scrub_io_cv); mutex_exit(&spa->spa_scrub_lock); } else { mutex_enter(&queue->q_vd->vdev_scan_io_queue_lock); ASSERT3U(queue->q_inflight_bytes, >=, BP_GET_PSIZE(bp)); queue->q_inflight_bytes -= BP_GET_PSIZE(bp); cv_broadcast(&queue->q_zio_cv); mutex_exit(&queue->q_vd->vdev_scan_io_queue_lock); } if (zio->io_error && (zio->io_error != ECKSUM || !(zio->io_flags & ZIO_FLAG_SPECULATIVE))) { if (dsl_errorscrubbing(spa->spa_dsl_pool) && !dsl_errorscrub_is_paused(spa->spa_dsl_pool->dp_scan)) { atomic_inc_64(&spa->spa_dsl_pool->dp_scan ->errorscrub_phys.dep_errors); } else { atomic_inc_64(&spa->spa_dsl_pool->dp_scan->scn_phys .scn_errors); } } } /* * Given a scanning zio's information, executes the zio. The zio need * not necessarily be only sortable, this function simply executes the * zio, no matter what it is. The optional queue argument allows the * caller to specify that they want per top level vdev IO rate limiting * instead of the legacy global limiting. */ static void scan_exec_io(dsl_pool_t *dp, const blkptr_t *bp, int zio_flags, const zbookmark_phys_t *zb, dsl_scan_io_queue_t *queue) { spa_t *spa = dp->dp_spa; dsl_scan_t *scn = dp->dp_scan; size_t size = BP_GET_PSIZE(bp); abd_t *data = abd_alloc_for_io(size, B_FALSE); zio_t *pio; if (queue == NULL) { ASSERT3U(scn->scn_maxinflight_bytes, >, 0); mutex_enter(&spa->spa_scrub_lock); while (spa->spa_scrub_inflight >= scn->scn_maxinflight_bytes) cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); spa->spa_scrub_inflight += BP_GET_PSIZE(bp); mutex_exit(&spa->spa_scrub_lock); pio = scn->scn_zio_root; } else { kmutex_t *q_lock = &queue->q_vd->vdev_scan_io_queue_lock; ASSERT3U(queue->q_maxinflight_bytes, >, 0); mutex_enter(q_lock); while (queue->q_inflight_bytes >= queue->q_maxinflight_bytes) cv_wait(&queue->q_zio_cv, q_lock); queue->q_inflight_bytes += BP_GET_PSIZE(bp); pio = queue->q_zio; mutex_exit(q_lock); } ASSERT(pio != NULL); count_block_issued(spa, bp, queue == NULL); zio_nowait(zio_read(pio, spa, bp, data, size, dsl_scan_scrub_done, queue, ZIO_PRIORITY_SCRUB, zio_flags, zb)); } /* * This is the primary extent sorting algorithm. We balance two parameters: * 1) how many bytes of I/O are in an extent * 2) how well the extent is filled with I/O (as a fraction of its total size) * Since we allow extents to have gaps between their constituent I/Os, it's * possible to have a fairly large extent that contains the same amount of * I/O bytes than a much smaller extent, which just packs the I/O more tightly. * The algorithm sorts based on a score calculated from the extent's size, * the relative fill volume (in %) and a "fill weight" parameter that controls * the split between whether we prefer larger extents or more well populated * extents: * * SCORE = FILL_IN_BYTES + (FILL_IN_PERCENT * FILL_IN_BYTES * FILL_WEIGHT) * * Example: * 1) assume extsz = 64 MiB * 2) assume fill = 32 MiB (extent is half full) * 3) assume fill_weight = 3 * 4) SCORE = 32M + (((32M * 100) / 64M) * 3 * 32M) / 100 * SCORE = 32M + (50 * 3 * 32M) / 100 * SCORE = 32M + (4800M / 100) * SCORE = 32M + 48M * ^ ^ * | +--- final total relative fill-based score * +--------- final total fill-based score * SCORE = 80M * * As can be seen, at fill_ratio=3, the algorithm is slightly biased towards * extents that are more completely filled (in a 3:2 ratio) vs just larger. * Note that as an optimization, we replace multiplication and division by * 100 with bitshifting by 7 (which effectively multiplies and divides by 128). * * Since we do not care if one extent is only few percent better than another, * compress the score into 6 bits via binary logarithm AKA highbit64() and * put into otherwise unused due to ashift high bits of offset. This allows * to reduce q_exts_by_size B-tree elements to only 64 bits and compare them * with single operation. Plus it makes scrubs more sequential and reduces * chances that minor extent change move it within the B-tree. */ __attribute__((always_inline)) inline static int ext_size_compare(const void *x, const void *y) { const uint64_t *a = x, *b = y; return (TREE_CMP(*a, *b)); } ZFS_BTREE_FIND_IN_BUF_FUNC(ext_size_find_in_buf, uint64_t, ext_size_compare) static void ext_size_create(zfs_range_tree_t *rt, void *arg) { (void) rt; zfs_btree_t *size_tree = arg; zfs_btree_create(size_tree, ext_size_compare, ext_size_find_in_buf, sizeof (uint64_t)); } static void ext_size_destroy(zfs_range_tree_t *rt, void *arg) { (void) rt; zfs_btree_t *size_tree = arg; ASSERT0(zfs_btree_numnodes(size_tree)); zfs_btree_destroy(size_tree); } static uint64_t ext_size_value(zfs_range_tree_t *rt, zfs_range_seg_gap_t *rsg) { (void) rt; uint64_t size = rsg->rs_end - rsg->rs_start; uint64_t score = rsg->rs_fill + ((((rsg->rs_fill << 7) / size) * fill_weight * rsg->rs_fill) >> 7); ASSERT3U(rt->rt_shift, >=, 8); return (((uint64_t)(64 - highbit64(score)) << 56) | rsg->rs_start); } static void ext_size_add(zfs_range_tree_t *rt, zfs_range_seg_t *rs, void *arg) { zfs_btree_t *size_tree = arg; ASSERT3U(rt->rt_type, ==, ZFS_RANGE_SEG_GAP); uint64_t v = ext_size_value(rt, (zfs_range_seg_gap_t *)rs); zfs_btree_add(size_tree, &v); } static void ext_size_remove(zfs_range_tree_t *rt, zfs_range_seg_t *rs, void *arg) { zfs_btree_t *size_tree = arg; ASSERT3U(rt->rt_type, ==, ZFS_RANGE_SEG_GAP); uint64_t v = ext_size_value(rt, (zfs_range_seg_gap_t *)rs); zfs_btree_remove(size_tree, &v); } static void ext_size_vacate(zfs_range_tree_t *rt, void *arg) { zfs_btree_t *size_tree = arg; zfs_btree_clear(size_tree); zfs_btree_destroy(size_tree); ext_size_create(rt, arg); } static const zfs_range_tree_ops_t ext_size_ops = { .rtop_create = ext_size_create, .rtop_destroy = ext_size_destroy, .rtop_add = ext_size_add, .rtop_remove = ext_size_remove, .rtop_vacate = ext_size_vacate }; /* * Comparator for the q_sios_by_addr tree. Sorting is simply performed * based on LBA-order (from lowest to highest). */ static int sio_addr_compare(const void *x, const void *y) { const scan_io_t *a = x, *b = y; return (TREE_CMP(SIO_GET_OFFSET(a), SIO_GET_OFFSET(b))); } /* IO queues are created on demand when they are needed. */ static dsl_scan_io_queue_t * scan_io_queue_create(vdev_t *vd) { dsl_scan_t *scn = vd->vdev_spa->spa_dsl_pool->dp_scan; dsl_scan_io_queue_t *q = kmem_zalloc(sizeof (*q), KM_SLEEP); q->q_scn = scn; q->q_vd = vd; q->q_sio_memused = 0; q->q_last_ext_addr = -1; cv_init(&q->q_zio_cv, NULL, CV_DEFAULT, NULL); q->q_exts_by_addr = zfs_range_tree_create_gap(&ext_size_ops, ZFS_RANGE_SEG_GAP, &q->q_exts_by_size, 0, vd->vdev_ashift, zfs_scan_max_ext_gap); avl_create(&q->q_sios_by_addr, sio_addr_compare, sizeof (scan_io_t), offsetof(scan_io_t, sio_nodes.sio_addr_node)); return (q); } /* * Destroys a scan queue and all segments and scan_io_t's contained in it. * No further execution of I/O occurs, anything pending in the queue is * simply freed without being executed. */ void dsl_scan_io_queue_destroy(dsl_scan_io_queue_t *queue) { dsl_scan_t *scn = queue->q_scn; scan_io_t *sio; void *cookie = NULL; ASSERT(MUTEX_HELD(&queue->q_vd->vdev_scan_io_queue_lock)); if (!avl_is_empty(&queue->q_sios_by_addr)) atomic_add_64(&scn->scn_queues_pending, -1); while ((sio = avl_destroy_nodes(&queue->q_sios_by_addr, &cookie)) != NULL) { ASSERT(zfs_range_tree_contains(queue->q_exts_by_addr, SIO_GET_OFFSET(sio), SIO_GET_ASIZE(sio))); queue->q_sio_memused -= SIO_GET_MUSED(sio); sio_free(sio); } ASSERT0(queue->q_sio_memused); zfs_range_tree_vacate(queue->q_exts_by_addr, NULL, queue); zfs_range_tree_destroy(queue->q_exts_by_addr); avl_destroy(&queue->q_sios_by_addr); cv_destroy(&queue->q_zio_cv); kmem_free(queue, sizeof (*queue)); } /* * Properly transfers a dsl_scan_queue_t from `svd' to `tvd'. This is * called on behalf of vdev_top_transfer when creating or destroying * a mirror vdev due to zpool attach/detach. */ void dsl_scan_io_queue_vdev_xfer(vdev_t *svd, vdev_t *tvd) { mutex_enter(&svd->vdev_scan_io_queue_lock); mutex_enter(&tvd->vdev_scan_io_queue_lock); VERIFY0P(tvd->vdev_scan_io_queue); tvd->vdev_scan_io_queue = svd->vdev_scan_io_queue; svd->vdev_scan_io_queue = NULL; if (tvd->vdev_scan_io_queue != NULL) tvd->vdev_scan_io_queue->q_vd = tvd; mutex_exit(&tvd->vdev_scan_io_queue_lock); mutex_exit(&svd->vdev_scan_io_queue_lock); } static void scan_io_queues_destroy(dsl_scan_t *scn) { vdev_t *rvd = scn->scn_dp->dp_spa->spa_root_vdev; for (uint64_t i = 0; i < rvd->vdev_children; i++) { vdev_t *tvd = rvd->vdev_child[i]; mutex_enter(&tvd->vdev_scan_io_queue_lock); if (tvd->vdev_scan_io_queue != NULL) dsl_scan_io_queue_destroy(tvd->vdev_scan_io_queue); tvd->vdev_scan_io_queue = NULL; mutex_exit(&tvd->vdev_scan_io_queue_lock); } } static void dsl_scan_freed_dva(spa_t *spa, const blkptr_t *bp, int dva_i) { dsl_pool_t *dp = spa->spa_dsl_pool; dsl_scan_t *scn = dp->dp_scan; vdev_t *vdev; kmutex_t *q_lock; dsl_scan_io_queue_t *queue; scan_io_t *srch_sio, *sio; avl_index_t idx; uint64_t start, size; vdev = vdev_lookup_top(spa, DVA_GET_VDEV(&bp->blk_dva[dva_i])); ASSERT(vdev != NULL); q_lock = &vdev->vdev_scan_io_queue_lock; queue = vdev->vdev_scan_io_queue; mutex_enter(q_lock); if (queue == NULL) { mutex_exit(q_lock); return; } srch_sio = sio_alloc(BP_GET_NDVAS(bp)); bp2sio(bp, srch_sio, dva_i); start = SIO_GET_OFFSET(srch_sio); size = SIO_GET_ASIZE(srch_sio); /* * We can find the zio in two states: * 1) Cold, just sitting in the queue of zio's to be issued at * some point in the future. In this case, all we do is * remove the zio from the q_sios_by_addr tree, decrement * its data volume from the containing zfs_range_seg_t and * resort the q_exts_by_size tree to reflect that the * zfs_range_seg_t has lost some of its 'fill'. We don't shorten * the zfs_range_seg_t - this is usually rare enough not to be * worth the extra hassle of trying keep track of precise * extent boundaries. * 2) Hot, where the zio is currently in-flight in * dsl_scan_issue_ios. In this case, we can't simply * reach in and stop the in-flight zio's, so we instead * block the caller. Eventually, dsl_scan_issue_ios will * be done with issuing the zio's it gathered and will * signal us. */ sio = avl_find(&queue->q_sios_by_addr, srch_sio, &idx); sio_free(srch_sio); if (sio != NULL) { blkptr_t tmpbp; /* Got it while it was cold in the queue */ ASSERT3U(start, ==, SIO_GET_OFFSET(sio)); ASSERT3U(size, ==, SIO_GET_ASIZE(sio)); avl_remove(&queue->q_sios_by_addr, sio); if (avl_is_empty(&queue->q_sios_by_addr)) atomic_add_64(&scn->scn_queues_pending, -1); queue->q_sio_memused -= SIO_GET_MUSED(sio); ASSERT(zfs_range_tree_contains(queue->q_exts_by_addr, start, size)); zfs_range_tree_remove_fill(queue->q_exts_by_addr, start, size); /* count the block as though we skipped it */ sio2bp(sio, &tmpbp); count_block_skipped(scn, &tmpbp, B_FALSE); sio_free(sio); } mutex_exit(q_lock); } /* * Callback invoked when a zio_free() zio is executing. This needs to be * intercepted to prevent the zio from deallocating a particular portion * of disk space and it then getting reallocated and written to, while we * still have it queued up for processing. */ void dsl_scan_freed(spa_t *spa, const blkptr_t *bp) { dsl_pool_t *dp = spa->spa_dsl_pool; dsl_scan_t *scn = dp->dp_scan; ASSERT(!BP_IS_EMBEDDED(bp)); ASSERT(scn != NULL); if (!dsl_scan_is_running(scn)) return; for (int i = 0; i < BP_GET_NDVAS(bp); i++) dsl_scan_freed_dva(spa, bp, i); } /* * Check if a vdev needs resilvering (non-empty DTL), if so, and resilver has * not started, start it. Otherwise, only restart if max txg in DTL range is * greater than the max txg in the current scan. If the DTL max is less than * the scan max, then the vdev has not missed any new data since the resilver * started, so a restart is not needed. */ void dsl_scan_assess_vdev(dsl_pool_t *dp, vdev_t *vd) { uint64_t min, max; if (!vdev_resilver_needed(vd, &min, &max)) return; if (!dsl_scan_resilvering(dp)) { spa_async_request(dp->dp_spa, SPA_ASYNC_RESILVER); return; } if (max <= dp->dp_scan->scn_phys.scn_max_txg) return; /* restart is needed, check if it can be deferred */ if (spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_RESILVER_DEFER)) vdev_defer_resilver(vd); else spa_async_request(dp->dp_spa, SPA_ASYNC_RESILVER); } ZFS_MODULE_PARAM(zfs, zfs_, scan_vdev_limit, U64, ZMOD_RW, "Max bytes in flight per leaf vdev for scrubs and resilvers"); ZFS_MODULE_PARAM(zfs, zfs_, scrub_min_time_ms, UINT, ZMOD_RW, "Min millisecs to scrub per txg"); ZFS_MODULE_PARAM(zfs, zfs_, obsolete_min_time_ms, UINT, ZMOD_RW, "Min millisecs to obsolete per txg"); ZFS_MODULE_PARAM(zfs, zfs_, free_min_time_ms, UINT, ZMOD_RW, "Min millisecs to free per txg"); ZFS_MODULE_PARAM(zfs, zfs_, resilver_min_time_ms, UINT, ZMOD_RW, "Min millisecs to resilver per txg"); ZFS_MODULE_PARAM(zfs, zfs_, scan_suspend_progress, INT, ZMOD_RW, "Set to prevent scans from progressing"); ZFS_MODULE_PARAM(zfs, zfs_, no_scrub_io, INT, ZMOD_RW, "Set to disable scrub I/O"); ZFS_MODULE_PARAM(zfs, zfs_, no_scrub_prefetch, INT, ZMOD_RW, "Set to disable scrub prefetching"); ZFS_MODULE_PARAM(zfs, zfs_, async_block_max_blocks, U64, ZMOD_RW, "Max number of blocks freed in one txg"); ZFS_MODULE_PARAM(zfs, zfs_, max_async_dedup_frees, U64, ZMOD_RW, - "Max number of dedup blocks freed in one txg"); + "Max number of dedup, clone or gang blocks freed in one txg"); + +ZFS_MODULE_PARAM(zfs, zfs_, async_free_zio_wait_interval, U64, ZMOD_RW, + "Wait for pending free I/Os after issuing this many asynchronously"); ZFS_MODULE_PARAM(zfs, zfs_, free_bpobj_enabled, INT, ZMOD_RW, "Enable processing of the free_bpobj"); ZFS_MODULE_PARAM(zfs, zfs_, scan_blkstats, INT, ZMOD_RW, "Enable block statistics calculation during scrub"); ZFS_MODULE_PARAM(zfs, zfs_, scan_mem_lim_fact, UINT, ZMOD_RW, "Fraction of RAM for scan hard limit"); ZFS_MODULE_PARAM(zfs, zfs_, scan_issue_strategy, UINT, ZMOD_RW, "IO issuing strategy during scrubbing. 0 = default, 1 = LBA, 2 = size"); ZFS_MODULE_PARAM(zfs, zfs_, scan_legacy, INT, ZMOD_RW, "Scrub using legacy non-sequential method"); ZFS_MODULE_PARAM(zfs, zfs_, import_defer_txgs, UINT, ZMOD_RW, "Number of TXGs to defer background work after pool import"); ZFS_MODULE_PARAM(zfs, zfs_, scan_checkpoint_intval, UINT, ZMOD_RW, "Scan progress on-disk checkpointing interval"); ZFS_MODULE_PARAM(zfs, zfs_, scan_max_ext_gap, U64, ZMOD_RW, "Max gap in bytes between sequential scrub / resilver I/Os"); ZFS_MODULE_PARAM(zfs, zfs_, scan_mem_lim_soft_fact, UINT, ZMOD_RW, "Fraction of hard limit used as soft limit"); ZFS_MODULE_PARAM(zfs, zfs_, scan_strict_mem_lim, INT, ZMOD_RW, "Tunable to attempt to reduce lock contention"); ZFS_MODULE_PARAM(zfs, zfs_, scan_fill_weight, UINT, ZMOD_RW, "Tunable to adjust bias towards more filled segments during scans"); ZFS_MODULE_PARAM(zfs, zfs_, scan_report_txgs, UINT, ZMOD_RW, "Tunable to report resilver performance over the last N txgs"); ZFS_MODULE_PARAM(zfs, zfs_, resilver_disable_defer, INT, ZMOD_RW, "Process all resilvers immediately"); ZFS_MODULE_PARAM(zfs, zfs_, resilver_defer_percent, UINT, ZMOD_RW, "Issued IO percent complete after which resilvers are deferred"); ZFS_MODULE_PARAM(zfs, zfs_, scrub_error_blocks_per_txg, UINT, ZMOD_RW, "Error blocks to be scrubbed in one txg");