Index: vendor-sys/illumos/dist/uts/common/fs/zfs/arc.c =================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/arc.c (revision 316916) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/arc.c (revision 316917) @@ -1,7157 +1,7157 @@ /* * 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 http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2012, Joyent, Inc. All rights reserved. * Copyright (c) 2011, 2017 by Delphix. All rights reserved. * Copyright (c) 2014 by Saso Kiselkov. All rights reserved. * Copyright 2015 Nexenta Systems, Inc. All rights reserved. */ /* * DVA-based Adjustable Replacement Cache * * While much of the theory of operation used here is * based on the self-tuning, low overhead replacement cache * presented by Megiddo and Modha at FAST 2003, there are some * significant differences: * * 1. The Megiddo and Modha model assumes any page is evictable. * Pages in its cache cannot be "locked" into memory. This makes * the eviction algorithm simple: evict the last page in the list. * This also make the performance characteristics easy to reason * about. Our cache is not so simple. At any given moment, some * subset of the blocks in the cache are un-evictable because we * have handed out a reference to them. Blocks are only evictable * when there are no external references active. This makes * eviction far more problematic: we choose to evict the evictable * blocks that are the "lowest" in the list. * * There are times when it is not possible to evict the requested * space. In these circumstances we are unable to adjust the cache * size. To prevent the cache growing unbounded at these times we * implement a "cache throttle" that slows the flow of new data * into the cache until we can make space available. * * 2. The Megiddo and Modha model assumes a fixed cache size. * Pages are evicted when the cache is full and there is a cache * miss. Our model has a variable sized cache. It grows with * high use, but also tries to react to memory pressure from the * operating system: decreasing its size when system memory is * tight. * * 3. The Megiddo and Modha model assumes a fixed page size. All * elements of the cache are therefore exactly the same size. So * when adjusting the cache size following a cache miss, its simply * a matter of choosing a single page to evict. In our model, we * have variable sized cache blocks (rangeing from 512 bytes to * 128K bytes). We therefore choose a set of blocks to evict to make * space for a cache miss that approximates as closely as possible * the space used by the new block. * * See also: "ARC: A Self-Tuning, Low Overhead Replacement Cache" * by N. Megiddo & D. Modha, FAST 2003 */ /* * The locking model: * * A new reference to a cache buffer can be obtained in two * ways: 1) via a hash table lookup using the DVA as a key, * or 2) via one of the ARC lists. The arc_read() interface * uses method 1, while the internal ARC algorithms for * adjusting the cache use method 2. We therefore provide two * types of locks: 1) the hash table lock array, and 2) the * ARC list locks. * * Buffers do not have their own mutexes, rather they rely on the * hash table mutexes for the bulk of their protection (i.e. most * fields in the arc_buf_hdr_t are protected by these mutexes). * * buf_hash_find() returns the appropriate mutex (held) when it * locates the requested buffer in the hash table. It returns * NULL for the mutex if the buffer was not in the table. * * buf_hash_remove() expects the appropriate hash mutex to be * already held before it is invoked. * * Each ARC state also has a mutex which is used to protect the * buffer list associated with the state. When attempting to * obtain a hash table lock while holding an ARC list lock you * must use: mutex_tryenter() to avoid deadlock. Also note that * the active state mutex must be held before the ghost state mutex. * * Note that the majority of the performance stats are manipulated * with atomic operations. * * The L2ARC uses the l2ad_mtx on each vdev for the following: * * - L2ARC buflist creation * - L2ARC buflist eviction * - L2ARC write completion, which walks L2ARC buflists * - ARC header destruction, as it removes from L2ARC buflists * - ARC header release, as it removes from L2ARC buflists */ /* * ARC operation: * * Every block that is in the ARC is tracked by an arc_buf_hdr_t structure. * This structure can point either to a block that is still in the cache or to * one that is only accessible in an L2 ARC device, or it can provide * information about a block that was recently evicted. If a block is * only accessible in the L2ARC, then the arc_buf_hdr_t only has enough * information to retrieve it from the L2ARC device. This information is * stored in the l2arc_buf_hdr_t sub-structure of the arc_buf_hdr_t. A block * that is in this state cannot access the data directly. * * Blocks that are actively being referenced or have not been evicted * are cached in the L1ARC. The L1ARC (l1arc_buf_hdr_t) is a structure within * the arc_buf_hdr_t that will point to the data block in memory. A block can * only be read by a consumer if it has an l1arc_buf_hdr_t. The L1ARC * caches data in two ways -- in a list of ARC buffers (arc_buf_t) and * also in the arc_buf_hdr_t's private physical data block pointer (b_pdata). * * The L1ARC's data pointer may or may not be uncompressed. The ARC has the * ability to store the physical data (b_pdata) associated with the DVA of the * arc_buf_hdr_t. Since the b_pdata is a copy of the on-disk physical block, * it will match its on-disk compression characteristics. This behavior can be * disabled by setting 'zfs_compressed_arc_enabled' to B_FALSE. When the * compressed ARC functionality is disabled, the b_pdata will point to an * uncompressed version of the on-disk data. * * Data in the L1ARC is not accessed by consumers of the ARC directly. Each * arc_buf_hdr_t can have multiple ARC buffers (arc_buf_t) which reference it. * Each ARC buffer (arc_buf_t) is being actively accessed by a specific ARC * consumer. The ARC will provide references to this data and will keep it * cached until it is no longer in use. The ARC caches only the L1ARC's physical * data block and will evict any arc_buf_t that is no longer referenced. The * amount of memory consumed by the arc_buf_ts' data buffers can be seen via the * "overhead_size" kstat. * * Depending on the consumer, an arc_buf_t can be requested in uncompressed or * compressed form. The typical case is that consumers will want uncompressed * data, and when that happens a new data buffer is allocated where the data is * decompressed for them to use. Currently the only consumer who wants * compressed arc_buf_t's is "zfs send", when it streams data exactly as it * exists on disk. When this happens, the arc_buf_t's data buffer is shared * with the arc_buf_hdr_t. * * Here is a diagram showing an arc_buf_hdr_t referenced by two arc_buf_t's. The * first one is owned by a compressed send consumer (and therefore references * the same compressed data buffer as the arc_buf_hdr_t) and the second could be * used by any other consumer (and has its own uncompressed copy of the data * buffer). * * arc_buf_hdr_t * +-----------+ * | fields | * | common to | * | L1- and | * | L2ARC | * +-----------+ * | l2arc_buf_hdr_t * | | * +-----------+ * | l1arc_buf_hdr_t * | | arc_buf_t * | b_buf +------------>+-----------+ arc_buf_t * | b_pdata +-+ |b_next +---->+-----------+ * +-----------+ | |-----------| |b_next +-->NULL * | |b_comp = T | +-----------+ * | |b_data +-+ |b_comp = F | * | +-----------+ | |b_data +-+ * +->+------+ | +-----------+ | * compressed | | | | * data | |<--------------+ | uncompressed * +------+ compressed, | data * shared +-->+------+ * data | | * | | * +------+ * * When a consumer reads a block, the ARC must first look to see if the * arc_buf_hdr_t is cached. If the hdr is cached then the ARC allocates a new * arc_buf_t and either copies uncompressed data into a new data buffer from an * existing uncompressed arc_buf_t, decompresses the hdr's b_pdata buffer into a * new data buffer, or shares the hdr's b_pdata buffer, depending on whether the * hdr is compressed and the desired compression characteristics of the * arc_buf_t consumer. If the arc_buf_t ends up sharing data with the * arc_buf_hdr_t and both of them are uncompressed then the arc_buf_t must be * the last buffer in the hdr's b_buf list, however a shared compressed buf can * be anywhere in the hdr's list. * * The diagram below shows an example of an uncompressed ARC hdr that is * sharing its data with an arc_buf_t (note that the shared uncompressed buf is * the last element in the buf list): * * arc_buf_hdr_t * +-----------+ * | | * | | * | | * +-----------+ * l2arc_buf_hdr_t| | * | | * +-----------+ * l1arc_buf_hdr_t| | * | | arc_buf_t (shared) * | b_buf +------------>+---------+ arc_buf_t * | | |b_next +---->+---------+ * | b_pdata +-+ |---------| |b_next +-->NULL * +-----------+ | | | +---------+ * | |b_data +-+ | | * | +---------+ | |b_data +-+ * +->+------+ | +---------+ | * | | | | * uncompressed | | | | * data +------+ | | * ^ +->+------+ | * | uncompressed | | | * | data | | | * | +------+ | * +---------------------------------+ * * Writing to the ARC requires that the ARC first discard the hdr's b_pdata * since the physical block is about to be rewritten. The new data contents * will be contained in the arc_buf_t. As the I/O pipeline performs the write, * it may compress the data before writing it to disk. The ARC will be called * with the transformed data and will bcopy the transformed on-disk block into * a newly allocated b_pdata. Writes are always done into buffers which have * either been loaned (and hence are new and don't have other readers) or * buffers which have been released (and hence have their own hdr, if there * were originally other readers of the buf's original hdr). This ensures that * the ARC only needs to update a single buf and its hdr after a write occurs. * * When the L2ARC is in use, it will also take advantage of the b_pdata. The * L2ARC will always write the contents of b_pdata to the L2ARC. This means * that when compressed ARC is enabled that the L2ARC blocks are identical * to the on-disk block in the main data pool. This provides a significant * advantage since the ARC can leverage the bp's checksum when reading from the * L2ARC to determine if the contents are valid. However, if the compressed * ARC is disabled, then the L2ARC's block must be transformed to look * like the physical block in the main data pool before comparing the * checksum and determining its validity. */ #include #include #include #include #include #include #include #include #include #include #include #include #ifdef _KERNEL #include #include #include #include #endif #include #include #include #ifndef _KERNEL /* set with ZFS_DEBUG=watch, to enable watchpoints on frozen buffers */ boolean_t arc_watch = B_FALSE; int arc_procfd; #endif static kmutex_t arc_reclaim_lock; static kcondvar_t arc_reclaim_thread_cv; static boolean_t arc_reclaim_thread_exit; static kcondvar_t arc_reclaim_waiters_cv; uint_t arc_reduce_dnlc_percent = 3; /* * The number of headers to evict in arc_evict_state_impl() before * dropping the sublist lock and evicting from another sublist. A lower * value means we're more likely to evict the "correct" header (i.e. the * oldest header in the arc state), but comes with higher overhead * (i.e. more invocations of arc_evict_state_impl()). */ int zfs_arc_evict_batch_limit = 10; /* number of seconds before growing cache again */ static int arc_grow_retry = 60; /* shift of arc_c for calculating overflow limit in arc_get_data_buf */ int zfs_arc_overflow_shift = 8; /* shift of arc_c for calculating both min and max arc_p */ static int arc_p_min_shift = 4; /* log2(fraction of arc to reclaim) */ static int arc_shrink_shift = 7; /* * log2(fraction of ARC which must be free to allow growing). * I.e. If there is less than arc_c >> arc_no_grow_shift free memory, * when reading a new block into the ARC, we will evict an equal-sized block * from the ARC. * * This must be less than arc_shrink_shift, so that when we shrink the ARC, * we will still not allow it to grow. */ int arc_no_grow_shift = 5; /* * minimum lifespan of a prefetch block in clock ticks * (initialized in arc_init()) */ static int arc_min_prefetch_lifespan; /* * If this percent of memory is free, don't throttle. */ int arc_lotsfree_percent = 10; static int arc_dead; /* * The arc has filled available memory and has now warmed up. */ static boolean_t arc_warm; /* * log2 fraction of the zio arena to keep free. */ int arc_zio_arena_free_shift = 2; /* * These tunables are for performance analysis. */ uint64_t zfs_arc_max; uint64_t zfs_arc_min; uint64_t zfs_arc_meta_limit = 0; uint64_t zfs_arc_meta_min = 0; int zfs_arc_grow_retry = 0; int zfs_arc_shrink_shift = 0; int zfs_arc_p_min_shift = 0; int zfs_arc_average_blocksize = 8 * 1024; /* 8KB */ boolean_t zfs_compressed_arc_enabled = B_TRUE; /* * Note that buffers can be in one of 6 states: * ARC_anon - anonymous (discussed below) * ARC_mru - recently used, currently cached * ARC_mru_ghost - recentely used, no longer in cache * ARC_mfu - frequently used, currently cached * ARC_mfu_ghost - frequently used, no longer in cache * ARC_l2c_only - exists in L2ARC but not other states * When there are no active references to the buffer, they are * are linked onto a list in one of these arc states. These are * the only buffers that can be evicted or deleted. Within each * state there are multiple lists, one for meta-data and one for * non-meta-data. Meta-data (indirect blocks, blocks of dnodes, * etc.) is tracked separately so that it can be managed more * explicitly: favored over data, limited explicitly. * * Anonymous buffers are buffers that are not associated with * a DVA. These are buffers that hold dirty block copies * before they are written to stable storage. By definition, * they are "ref'd" and are considered part of arc_mru * that cannot be freed. Generally, they will aquire a DVA * as they are written and migrate onto the arc_mru list. * * The ARC_l2c_only state is for buffers that are in the second * level ARC but no longer in any of the ARC_m* lists. The second * level ARC itself may also contain buffers that are in any of * the ARC_m* states - meaning that a buffer can exist in two * places. The reason for the ARC_l2c_only state is to keep the * buffer header in the hash table, so that reads that hit the * second level ARC benefit from these fast lookups. */ typedef struct arc_state { /* * list of evictable buffers */ - multilist_t arcs_list[ARC_BUFC_NUMTYPES]; + multilist_t *arcs_list[ARC_BUFC_NUMTYPES]; /* * total amount of evictable data in this state */ refcount_t arcs_esize[ARC_BUFC_NUMTYPES]; /* * total amount of data in this state; this includes: evictable, * non-evictable, ARC_BUFC_DATA, and ARC_BUFC_METADATA. */ refcount_t arcs_size; } arc_state_t; /* The 6 states: */ static arc_state_t ARC_anon; static arc_state_t ARC_mru; static arc_state_t ARC_mru_ghost; static arc_state_t ARC_mfu; static arc_state_t ARC_mfu_ghost; static arc_state_t ARC_l2c_only; typedef struct arc_stats { kstat_named_t arcstat_hits; kstat_named_t arcstat_misses; kstat_named_t arcstat_demand_data_hits; kstat_named_t arcstat_demand_data_misses; kstat_named_t arcstat_demand_metadata_hits; kstat_named_t arcstat_demand_metadata_misses; kstat_named_t arcstat_prefetch_data_hits; kstat_named_t arcstat_prefetch_data_misses; kstat_named_t arcstat_prefetch_metadata_hits; kstat_named_t arcstat_prefetch_metadata_misses; kstat_named_t arcstat_mru_hits; kstat_named_t arcstat_mru_ghost_hits; kstat_named_t arcstat_mfu_hits; kstat_named_t arcstat_mfu_ghost_hits; kstat_named_t arcstat_deleted; /* * Number of buffers that could not be evicted because the hash lock * was held by another thread. The lock may not necessarily be held * by something using the same buffer, since hash locks are shared * by multiple buffers. */ kstat_named_t arcstat_mutex_miss; /* * Number of buffers skipped because they have I/O in progress, are * indrect prefetch buffers that have not lived long enough, or are * not from the spa we're trying to evict from. */ kstat_named_t arcstat_evict_skip; /* * Number of times arc_evict_state() was unable to evict enough * buffers to reach it's target amount. */ kstat_named_t arcstat_evict_not_enough; kstat_named_t arcstat_evict_l2_cached; kstat_named_t arcstat_evict_l2_eligible; kstat_named_t arcstat_evict_l2_ineligible; kstat_named_t arcstat_evict_l2_skip; kstat_named_t arcstat_hash_elements; kstat_named_t arcstat_hash_elements_max; kstat_named_t arcstat_hash_collisions; kstat_named_t arcstat_hash_chains; kstat_named_t arcstat_hash_chain_max; kstat_named_t arcstat_p; kstat_named_t arcstat_c; kstat_named_t arcstat_c_min; kstat_named_t arcstat_c_max; kstat_named_t arcstat_size; /* * Number of compressed bytes stored in the arc_buf_hdr_t's b_pdata. * Note that the compressed bytes may match the uncompressed bytes * if the block is either not compressed or compressed arc is disabled. */ kstat_named_t arcstat_compressed_size; /* * Uncompressed size of the data stored in b_pdata. If compressed * arc is disabled then this value will be identical to the stat * above. */ kstat_named_t arcstat_uncompressed_size; /* * Number of bytes stored in all the arc_buf_t's. This is classified * as "overhead" since this data is typically short-lived and will * be evicted from the arc when it becomes unreferenced unless the * zfs_keep_uncompressed_metadata or zfs_keep_uncompressed_level * values have been set (see comment in dbuf.c for more information). */ kstat_named_t arcstat_overhead_size; /* * Number of bytes consumed by internal ARC structures necessary * for tracking purposes; these structures are not actually * backed by ARC buffers. This includes arc_buf_hdr_t structures * (allocated via arc_buf_hdr_t_full and arc_buf_hdr_t_l2only * caches), and arc_buf_t structures (allocated via arc_buf_t * cache). */ kstat_named_t arcstat_hdr_size; /* * Number of bytes consumed by ARC buffers of type equal to * ARC_BUFC_DATA. This is generally consumed by buffers backing * on disk user data (e.g. plain file contents). */ kstat_named_t arcstat_data_size; /* * Number of bytes consumed by ARC buffers of type equal to * ARC_BUFC_METADATA. This is generally consumed by buffers * backing on disk data that is used for internal ZFS * structures (e.g. ZAP, dnode, indirect blocks, etc). */ kstat_named_t arcstat_metadata_size; /* * Number of bytes consumed by various buffers and structures * not actually backed with ARC buffers. This includes bonus * buffers (allocated directly via zio_buf_* functions), * dmu_buf_impl_t structures (allocated via dmu_buf_impl_t * cache), and dnode_t structures (allocated via dnode_t cache). */ kstat_named_t arcstat_other_size; /* * Total number of bytes consumed by ARC buffers residing in the * arc_anon state. This includes *all* buffers in the arc_anon * state; e.g. data, metadata, evictable, and unevictable buffers * are all included in this value. */ kstat_named_t arcstat_anon_size; /* * Number of bytes consumed by ARC buffers that meet the * following criteria: backing buffers of type ARC_BUFC_DATA, * residing in the arc_anon state, and are eligible for eviction * (e.g. have no outstanding holds on the buffer). */ kstat_named_t arcstat_anon_evictable_data; /* * Number of bytes consumed by ARC buffers that meet the * following criteria: backing buffers of type ARC_BUFC_METADATA, * residing in the arc_anon state, and are eligible for eviction * (e.g. have no outstanding holds on the buffer). */ kstat_named_t arcstat_anon_evictable_metadata; /* * Total number of bytes consumed by ARC buffers residing in the * arc_mru state. This includes *all* buffers in the arc_mru * state; e.g. data, metadata, evictable, and unevictable buffers * are all included in this value. */ kstat_named_t arcstat_mru_size; /* * Number of bytes consumed by ARC buffers that meet the * following criteria: backing buffers of type ARC_BUFC_DATA, * residing in the arc_mru state, and are eligible for eviction * (e.g. have no outstanding holds on the buffer). */ kstat_named_t arcstat_mru_evictable_data; /* * Number of bytes consumed by ARC buffers that meet the * following criteria: backing buffers of type ARC_BUFC_METADATA, * residing in the arc_mru state, and are eligible for eviction * (e.g. have no outstanding holds on the buffer). */ kstat_named_t arcstat_mru_evictable_metadata; /* * Total number of bytes that *would have been* consumed by ARC * buffers in the arc_mru_ghost state. The key thing to note * here, is the fact that this size doesn't actually indicate * RAM consumption. The ghost lists only consist of headers and * don't actually have ARC buffers linked off of these headers. * Thus, *if* the headers had associated ARC buffers, these * buffers *would have* consumed this number of bytes. */ kstat_named_t arcstat_mru_ghost_size; /* * Number of bytes that *would have been* consumed by ARC * buffers that are eligible for eviction, of type * ARC_BUFC_DATA, and linked off the arc_mru_ghost state. */ kstat_named_t arcstat_mru_ghost_evictable_data; /* * Number of bytes that *would have been* consumed by ARC * buffers that are eligible for eviction, of type * ARC_BUFC_METADATA, and linked off the arc_mru_ghost state. */ kstat_named_t arcstat_mru_ghost_evictable_metadata; /* * Total number of bytes consumed by ARC buffers residing in the * arc_mfu state. This includes *all* buffers in the arc_mfu * state; e.g. data, metadata, evictable, and unevictable buffers * are all included in this value. */ kstat_named_t arcstat_mfu_size; /* * Number of bytes consumed by ARC buffers that are eligible for * eviction, of type ARC_BUFC_DATA, and reside in the arc_mfu * state. */ kstat_named_t arcstat_mfu_evictable_data; /* * Number of bytes consumed by ARC buffers that are eligible for * eviction, of type ARC_BUFC_METADATA, and reside in the * arc_mfu state. */ kstat_named_t arcstat_mfu_evictable_metadata; /* * Total number of bytes that *would have been* consumed by ARC * buffers in the arc_mfu_ghost state. See the comment above * arcstat_mru_ghost_size for more details. */ kstat_named_t arcstat_mfu_ghost_size; /* * Number of bytes that *would have been* consumed by ARC * buffers that are eligible for eviction, of type * ARC_BUFC_DATA, and linked off the arc_mfu_ghost state. */ kstat_named_t arcstat_mfu_ghost_evictable_data; /* * Number of bytes that *would have been* consumed by ARC * buffers that are eligible for eviction, of type * ARC_BUFC_METADATA, and linked off the arc_mru_ghost state. */ kstat_named_t arcstat_mfu_ghost_evictable_metadata; kstat_named_t arcstat_l2_hits; kstat_named_t arcstat_l2_misses; kstat_named_t arcstat_l2_feeds; kstat_named_t arcstat_l2_rw_clash; kstat_named_t arcstat_l2_read_bytes; kstat_named_t arcstat_l2_write_bytes; kstat_named_t arcstat_l2_writes_sent; kstat_named_t arcstat_l2_writes_done; kstat_named_t arcstat_l2_writes_error; kstat_named_t arcstat_l2_writes_lock_retry; kstat_named_t arcstat_l2_evict_lock_retry; kstat_named_t arcstat_l2_evict_reading; kstat_named_t arcstat_l2_evict_l1cached; kstat_named_t arcstat_l2_free_on_write; kstat_named_t arcstat_l2_abort_lowmem; kstat_named_t arcstat_l2_cksum_bad; kstat_named_t arcstat_l2_io_error; kstat_named_t arcstat_l2_size; kstat_named_t arcstat_l2_asize; kstat_named_t arcstat_l2_hdr_size; kstat_named_t arcstat_memory_throttle_count; kstat_named_t arcstat_meta_used; kstat_named_t arcstat_meta_limit; kstat_named_t arcstat_meta_max; kstat_named_t arcstat_meta_min; kstat_named_t arcstat_sync_wait_for_async; kstat_named_t arcstat_demand_hit_predictive_prefetch; } arc_stats_t; static arc_stats_t arc_stats = { { "hits", KSTAT_DATA_UINT64 }, { "misses", KSTAT_DATA_UINT64 }, { "demand_data_hits", KSTAT_DATA_UINT64 }, { "demand_data_misses", KSTAT_DATA_UINT64 }, { "demand_metadata_hits", KSTAT_DATA_UINT64 }, { "demand_metadata_misses", KSTAT_DATA_UINT64 }, { "prefetch_data_hits", KSTAT_DATA_UINT64 }, { "prefetch_data_misses", KSTAT_DATA_UINT64 }, { "prefetch_metadata_hits", KSTAT_DATA_UINT64 }, { "prefetch_metadata_misses", KSTAT_DATA_UINT64 }, { "mru_hits", KSTAT_DATA_UINT64 }, { "mru_ghost_hits", KSTAT_DATA_UINT64 }, { "mfu_hits", KSTAT_DATA_UINT64 }, { "mfu_ghost_hits", KSTAT_DATA_UINT64 }, { "deleted", KSTAT_DATA_UINT64 }, { "mutex_miss", KSTAT_DATA_UINT64 }, { "evict_skip", KSTAT_DATA_UINT64 }, { "evict_not_enough", KSTAT_DATA_UINT64 }, { "evict_l2_cached", KSTAT_DATA_UINT64 }, { "evict_l2_eligible", KSTAT_DATA_UINT64 }, { "evict_l2_ineligible", KSTAT_DATA_UINT64 }, { "evict_l2_skip", KSTAT_DATA_UINT64 }, { "hash_elements", KSTAT_DATA_UINT64 }, { "hash_elements_max", KSTAT_DATA_UINT64 }, { "hash_collisions", KSTAT_DATA_UINT64 }, { "hash_chains", KSTAT_DATA_UINT64 }, { "hash_chain_max", KSTAT_DATA_UINT64 }, { "p", KSTAT_DATA_UINT64 }, { "c", KSTAT_DATA_UINT64 }, { "c_min", KSTAT_DATA_UINT64 }, { "c_max", KSTAT_DATA_UINT64 }, { "size", KSTAT_DATA_UINT64 }, { "compressed_size", KSTAT_DATA_UINT64 }, { "uncompressed_size", KSTAT_DATA_UINT64 }, { "overhead_size", KSTAT_DATA_UINT64 }, { "hdr_size", KSTAT_DATA_UINT64 }, { "data_size", KSTAT_DATA_UINT64 }, { "metadata_size", KSTAT_DATA_UINT64 }, { "other_size", KSTAT_DATA_UINT64 }, { "anon_size", KSTAT_DATA_UINT64 }, { "anon_evictable_data", KSTAT_DATA_UINT64 }, { "anon_evictable_metadata", KSTAT_DATA_UINT64 }, { "mru_size", KSTAT_DATA_UINT64 }, { "mru_evictable_data", KSTAT_DATA_UINT64 }, { "mru_evictable_metadata", KSTAT_DATA_UINT64 }, { "mru_ghost_size", KSTAT_DATA_UINT64 }, { "mru_ghost_evictable_data", KSTAT_DATA_UINT64 }, { "mru_ghost_evictable_metadata", KSTAT_DATA_UINT64 }, { "mfu_size", KSTAT_DATA_UINT64 }, { "mfu_evictable_data", KSTAT_DATA_UINT64 }, { "mfu_evictable_metadata", KSTAT_DATA_UINT64 }, { "mfu_ghost_size", KSTAT_DATA_UINT64 }, { "mfu_ghost_evictable_data", KSTAT_DATA_UINT64 }, { "mfu_ghost_evictable_metadata", KSTAT_DATA_UINT64 }, { "l2_hits", KSTAT_DATA_UINT64 }, { "l2_misses", KSTAT_DATA_UINT64 }, { "l2_feeds", KSTAT_DATA_UINT64 }, { "l2_rw_clash", KSTAT_DATA_UINT64 }, { "l2_read_bytes", KSTAT_DATA_UINT64 }, { "l2_write_bytes", KSTAT_DATA_UINT64 }, { "l2_writes_sent", KSTAT_DATA_UINT64 }, { "l2_writes_done", KSTAT_DATA_UINT64 }, { "l2_writes_error", KSTAT_DATA_UINT64 }, { "l2_writes_lock_retry", KSTAT_DATA_UINT64 }, { "l2_evict_lock_retry", KSTAT_DATA_UINT64 }, { "l2_evict_reading", KSTAT_DATA_UINT64 }, { "l2_evict_l1cached", KSTAT_DATA_UINT64 }, { "l2_free_on_write", KSTAT_DATA_UINT64 }, { "l2_abort_lowmem", KSTAT_DATA_UINT64 }, { "l2_cksum_bad", KSTAT_DATA_UINT64 }, { "l2_io_error", KSTAT_DATA_UINT64 }, { "l2_size", KSTAT_DATA_UINT64 }, { "l2_asize", KSTAT_DATA_UINT64 }, { "l2_hdr_size", KSTAT_DATA_UINT64 }, { "memory_throttle_count", KSTAT_DATA_UINT64 }, { "arc_meta_used", KSTAT_DATA_UINT64 }, { "arc_meta_limit", KSTAT_DATA_UINT64 }, { "arc_meta_max", KSTAT_DATA_UINT64 }, { "arc_meta_min", KSTAT_DATA_UINT64 }, { "sync_wait_for_async", KSTAT_DATA_UINT64 }, { "demand_hit_predictive_prefetch", KSTAT_DATA_UINT64 }, }; #define ARCSTAT(stat) (arc_stats.stat.value.ui64) #define ARCSTAT_INCR(stat, val) \ atomic_add_64(&arc_stats.stat.value.ui64, (val)) #define ARCSTAT_BUMP(stat) ARCSTAT_INCR(stat, 1) #define ARCSTAT_BUMPDOWN(stat) ARCSTAT_INCR(stat, -1) #define ARCSTAT_MAX(stat, val) { \ uint64_t m; \ while ((val) > (m = arc_stats.stat.value.ui64) && \ (m != atomic_cas_64(&arc_stats.stat.value.ui64, m, (val)))) \ continue; \ } #define ARCSTAT_MAXSTAT(stat) \ ARCSTAT_MAX(stat##_max, arc_stats.stat.value.ui64) /* * We define a macro to allow ARC hits/misses to be easily broken down by * two separate conditions, giving a total of four different subtypes for * each of hits and misses (so eight statistics total). */ #define ARCSTAT_CONDSTAT(cond1, stat1, notstat1, cond2, stat2, notstat2, stat) \ if (cond1) { \ if (cond2) { \ ARCSTAT_BUMP(arcstat_##stat1##_##stat2##_##stat); \ } else { \ ARCSTAT_BUMP(arcstat_##stat1##_##notstat2##_##stat); \ } \ } else { \ if (cond2) { \ ARCSTAT_BUMP(arcstat_##notstat1##_##stat2##_##stat); \ } else { \ ARCSTAT_BUMP(arcstat_##notstat1##_##notstat2##_##stat);\ } \ } kstat_t *arc_ksp; static arc_state_t *arc_anon; static arc_state_t *arc_mru; static arc_state_t *arc_mru_ghost; static arc_state_t *arc_mfu; static arc_state_t *arc_mfu_ghost; static arc_state_t *arc_l2c_only; /* * There are several ARC variables that are critical to export as kstats -- * but we don't want to have to grovel around in the kstat whenever we wish to * manipulate them. For these variables, we therefore define them to be in * terms of the statistic variable. This assures that we are not introducing * the possibility of inconsistency by having shadow copies of the variables, * while still allowing the code to be readable. */ #define arc_size ARCSTAT(arcstat_size) /* actual total arc size */ #define arc_p ARCSTAT(arcstat_p) /* target size of MRU */ #define arc_c ARCSTAT(arcstat_c) /* target size of cache */ #define arc_c_min ARCSTAT(arcstat_c_min) /* min target cache size */ #define arc_c_max ARCSTAT(arcstat_c_max) /* max target cache size */ #define arc_meta_limit ARCSTAT(arcstat_meta_limit) /* max size for metadata */ #define arc_meta_min ARCSTAT(arcstat_meta_min) /* min size for metadata */ #define arc_meta_used ARCSTAT(arcstat_meta_used) /* size of metadata */ #define arc_meta_max ARCSTAT(arcstat_meta_max) /* max size of metadata */ /* compressed size of entire arc */ #define arc_compressed_size ARCSTAT(arcstat_compressed_size) /* uncompressed size of entire arc */ #define arc_uncompressed_size ARCSTAT(arcstat_uncompressed_size) /* number of bytes in the arc from arc_buf_t's */ #define arc_overhead_size ARCSTAT(arcstat_overhead_size) static int arc_no_grow; /* Don't try to grow cache size */ static uint64_t arc_tempreserve; static uint64_t arc_loaned_bytes; typedef struct arc_callback arc_callback_t; struct arc_callback { void *acb_private; arc_done_func_t *acb_done; arc_buf_t *acb_buf; boolean_t acb_compressed; zio_t *acb_zio_dummy; arc_callback_t *acb_next; }; typedef struct arc_write_callback arc_write_callback_t; struct arc_write_callback { void *awcb_private; arc_done_func_t *awcb_ready; arc_done_func_t *awcb_children_ready; arc_done_func_t *awcb_physdone; arc_done_func_t *awcb_done; arc_buf_t *awcb_buf; }; /* * ARC buffers are separated into multiple structs as a memory saving measure: * - Common fields struct, always defined, and embedded within it: * - L2-only fields, always allocated but undefined when not in L2ARC * - L1-only fields, only allocated when in L1ARC * * Buffer in L1 Buffer only in L2 * +------------------------+ +------------------------+ * | arc_buf_hdr_t | | arc_buf_hdr_t | * | | | | * | | | | * | | | | * +------------------------+ +------------------------+ * | l2arc_buf_hdr_t | | l2arc_buf_hdr_t | * | (undefined if L1-only) | | | * +------------------------+ +------------------------+ * | l1arc_buf_hdr_t | * | | * | | * | | * | | * +------------------------+ * * Because it's possible for the L2ARC to become extremely large, we can wind * up eating a lot of memory in L2ARC buffer headers, so the size of a header * is minimized by only allocating the fields necessary for an L1-cached buffer * when a header is actually in the L1 cache. The sub-headers (l1arc_buf_hdr and * l2arc_buf_hdr) are embedded rather than allocated separately to save a couple * words in pointers. arc_hdr_realloc() is used to switch a header between * these two allocation states. */ typedef struct l1arc_buf_hdr { kmutex_t b_freeze_lock; zio_cksum_t *b_freeze_cksum; #ifdef ZFS_DEBUG /* * Used for debugging with kmem_flags - by allocating and freeing * b_thawed when the buffer is thawed, we get a record of the stack * trace that thawed it. */ void *b_thawed; #endif arc_buf_t *b_buf; uint32_t b_bufcnt; /* for waiting on writes to complete */ kcondvar_t b_cv; uint8_t b_byteswap; /* protected by arc state mutex */ arc_state_t *b_state; multilist_node_t b_arc_node; /* updated atomically */ clock_t b_arc_access; /* self protecting */ refcount_t b_refcnt; arc_callback_t *b_acb; void *b_pdata; } l1arc_buf_hdr_t; typedef struct l2arc_dev l2arc_dev_t; typedef struct l2arc_buf_hdr { /* protected by arc_buf_hdr mutex */ l2arc_dev_t *b_dev; /* L2ARC device */ uint64_t b_daddr; /* disk address, offset byte */ list_node_t b_l2node; } l2arc_buf_hdr_t; struct arc_buf_hdr { /* protected by hash lock */ dva_t b_dva; uint64_t b_birth; arc_buf_contents_t b_type; arc_buf_hdr_t *b_hash_next; arc_flags_t b_flags; /* * This field stores the size of the data buffer after * compression, and is set in the arc's zio completion handlers. * It is in units of SPA_MINBLOCKSIZE (e.g. 1 == 512 bytes). * * While the block pointers can store up to 32MB in their psize * field, we can only store up to 32MB minus 512B. This is due * to the bp using a bias of 1, whereas we use a bias of 0 (i.e. * a field of zeros represents 512B in the bp). We can't use a * bias of 1 since we need to reserve a psize of zero, here, to * represent holes and embedded blocks. * * This isn't a problem in practice, since the maximum size of a * buffer is limited to 16MB, so we never need to store 32MB in * this field. Even in the upstream illumos code base, the * maximum size of a buffer is limited to 16MB. */ uint16_t b_psize; /* * This field stores the size of the data buffer before * compression, and cannot change once set. It is in units * of SPA_MINBLOCKSIZE (e.g. 2 == 1024 bytes) */ uint16_t b_lsize; /* immutable */ uint64_t b_spa; /* immutable */ /* L2ARC fields. Undefined when not in L2ARC. */ l2arc_buf_hdr_t b_l2hdr; /* L1ARC fields. Undefined when in l2arc_only state */ l1arc_buf_hdr_t b_l1hdr; }; #define GHOST_STATE(state) \ ((state) == arc_mru_ghost || (state) == arc_mfu_ghost || \ (state) == arc_l2c_only) #define HDR_IN_HASH_TABLE(hdr) ((hdr)->b_flags & ARC_FLAG_IN_HASH_TABLE) #define HDR_IO_IN_PROGRESS(hdr) ((hdr)->b_flags & ARC_FLAG_IO_IN_PROGRESS) #define HDR_IO_ERROR(hdr) ((hdr)->b_flags & ARC_FLAG_IO_ERROR) #define HDR_PREFETCH(hdr) ((hdr)->b_flags & ARC_FLAG_PREFETCH) #define HDR_COMPRESSION_ENABLED(hdr) \ ((hdr)->b_flags & ARC_FLAG_COMPRESSED_ARC) #define HDR_L2CACHE(hdr) ((hdr)->b_flags & ARC_FLAG_L2CACHE) #define HDR_L2_READING(hdr) \ (((hdr)->b_flags & ARC_FLAG_IO_IN_PROGRESS) && \ ((hdr)->b_flags & ARC_FLAG_HAS_L2HDR)) #define HDR_L2_WRITING(hdr) ((hdr)->b_flags & ARC_FLAG_L2_WRITING) #define HDR_L2_EVICTED(hdr) ((hdr)->b_flags & ARC_FLAG_L2_EVICTED) #define HDR_L2_WRITE_HEAD(hdr) ((hdr)->b_flags & ARC_FLAG_L2_WRITE_HEAD) #define HDR_SHARED_DATA(hdr) ((hdr)->b_flags & ARC_FLAG_SHARED_DATA) #define HDR_ISTYPE_METADATA(hdr) \ ((hdr)->b_flags & ARC_FLAG_BUFC_METADATA) #define HDR_ISTYPE_DATA(hdr) (!HDR_ISTYPE_METADATA(hdr)) #define HDR_HAS_L1HDR(hdr) ((hdr)->b_flags & ARC_FLAG_HAS_L1HDR) #define HDR_HAS_L2HDR(hdr) ((hdr)->b_flags & ARC_FLAG_HAS_L2HDR) /* For storing compression mode in b_flags */ #define HDR_COMPRESS_OFFSET (highbit64(ARC_FLAG_COMPRESS_0) - 1) #define HDR_GET_COMPRESS(hdr) ((enum zio_compress)BF32_GET((hdr)->b_flags, \ HDR_COMPRESS_OFFSET, SPA_COMPRESSBITS)) #define HDR_SET_COMPRESS(hdr, cmp) BF32_SET((hdr)->b_flags, \ HDR_COMPRESS_OFFSET, SPA_COMPRESSBITS, (cmp)); #define ARC_BUF_LAST(buf) ((buf)->b_next == NULL) #define ARC_BUF_SHARED(buf) ((buf)->b_flags & ARC_BUF_FLAG_SHARED) #define ARC_BUF_COMPRESSED(buf) ((buf)->b_flags & ARC_BUF_FLAG_COMPRESSED) /* * Other sizes */ #define HDR_FULL_SIZE ((int64_t)sizeof (arc_buf_hdr_t)) #define HDR_L2ONLY_SIZE ((int64_t)offsetof(arc_buf_hdr_t, b_l1hdr)) /* * Hash table routines */ #define HT_LOCK_PAD 64 struct ht_lock { kmutex_t ht_lock; #ifdef _KERNEL unsigned char pad[(HT_LOCK_PAD - sizeof (kmutex_t))]; #endif }; #define BUF_LOCKS 256 typedef struct buf_hash_table { uint64_t ht_mask; arc_buf_hdr_t **ht_table; struct ht_lock ht_locks[BUF_LOCKS]; } buf_hash_table_t; static buf_hash_table_t buf_hash_table; #define BUF_HASH_INDEX(spa, dva, birth) \ (buf_hash(spa, dva, birth) & buf_hash_table.ht_mask) #define BUF_HASH_LOCK_NTRY(idx) (buf_hash_table.ht_locks[idx & (BUF_LOCKS-1)]) #define BUF_HASH_LOCK(idx) (&(BUF_HASH_LOCK_NTRY(idx).ht_lock)) #define HDR_LOCK(hdr) \ (BUF_HASH_LOCK(BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth))) uint64_t zfs_crc64_table[256]; /* * Level 2 ARC */ #define L2ARC_WRITE_SIZE (8 * 1024 * 1024) /* initial write max */ #define L2ARC_HEADROOM 2 /* num of writes */ /* * If we discover during ARC scan any buffers to be compressed, we boost * our headroom for the next scanning cycle by this percentage multiple. */ #define L2ARC_HEADROOM_BOOST 200 #define L2ARC_FEED_SECS 1 /* caching interval secs */ #define L2ARC_FEED_MIN_MS 200 /* min caching interval ms */ #define l2arc_writes_sent ARCSTAT(arcstat_l2_writes_sent) #define l2arc_writes_done ARCSTAT(arcstat_l2_writes_done) /* L2ARC Performance Tunables */ uint64_t l2arc_write_max = L2ARC_WRITE_SIZE; /* default max write size */ uint64_t l2arc_write_boost = L2ARC_WRITE_SIZE; /* extra write during warmup */ uint64_t l2arc_headroom = L2ARC_HEADROOM; /* number of dev writes */ uint64_t l2arc_headroom_boost = L2ARC_HEADROOM_BOOST; uint64_t l2arc_feed_secs = L2ARC_FEED_SECS; /* interval seconds */ uint64_t l2arc_feed_min_ms = L2ARC_FEED_MIN_MS; /* min interval milliseconds */ boolean_t l2arc_noprefetch = B_TRUE; /* don't cache prefetch bufs */ boolean_t l2arc_feed_again = B_TRUE; /* turbo warmup */ boolean_t l2arc_norw = B_TRUE; /* no reads during writes */ /* * L2ARC Internals */ struct l2arc_dev { vdev_t *l2ad_vdev; /* vdev */ spa_t *l2ad_spa; /* spa */ uint64_t l2ad_hand; /* next write location */ uint64_t l2ad_start; /* first addr on device */ uint64_t l2ad_end; /* last addr on device */ boolean_t l2ad_first; /* first sweep through */ boolean_t l2ad_writing; /* currently writing */ kmutex_t l2ad_mtx; /* lock for buffer list */ list_t l2ad_buflist; /* buffer list */ list_node_t l2ad_node; /* device list node */ refcount_t l2ad_alloc; /* allocated bytes */ }; static list_t L2ARC_dev_list; /* device list */ static list_t *l2arc_dev_list; /* device list pointer */ static kmutex_t l2arc_dev_mtx; /* device list mutex */ static l2arc_dev_t *l2arc_dev_last; /* last device used */ static list_t L2ARC_free_on_write; /* free after write buf list */ static list_t *l2arc_free_on_write; /* free after write list ptr */ static kmutex_t l2arc_free_on_write_mtx; /* mutex for list */ static uint64_t l2arc_ndev; /* number of devices */ typedef struct l2arc_read_callback { arc_buf_hdr_t *l2rcb_hdr; /* read header */ blkptr_t l2rcb_bp; /* original blkptr */ zbookmark_phys_t l2rcb_zb; /* original bookmark */ int l2rcb_flags; /* original flags */ } l2arc_read_callback_t; typedef struct l2arc_write_callback { l2arc_dev_t *l2wcb_dev; /* device info */ arc_buf_hdr_t *l2wcb_head; /* head of write buflist */ } l2arc_write_callback_t; typedef struct l2arc_data_free { /* protected by l2arc_free_on_write_mtx */ void *l2df_data; size_t l2df_size; arc_buf_contents_t l2df_type; list_node_t l2df_list_node; } l2arc_data_free_t; static kmutex_t l2arc_feed_thr_lock; static kcondvar_t l2arc_feed_thr_cv; static uint8_t l2arc_thread_exit; static void *arc_get_data_buf(arc_buf_hdr_t *, uint64_t, void *); static void arc_free_data_buf(arc_buf_hdr_t *, void *, uint64_t, void *); static void arc_hdr_free_pdata(arc_buf_hdr_t *hdr); static void arc_hdr_alloc_pdata(arc_buf_hdr_t *); static void arc_access(arc_buf_hdr_t *, kmutex_t *); static boolean_t arc_is_overflowing(); static void arc_buf_watch(arc_buf_t *); static arc_buf_contents_t arc_buf_type(arc_buf_hdr_t *); static uint32_t arc_bufc_to_flags(arc_buf_contents_t); static inline void arc_hdr_set_flags(arc_buf_hdr_t *hdr, arc_flags_t flags); static inline void arc_hdr_clear_flags(arc_buf_hdr_t *hdr, arc_flags_t flags); static boolean_t l2arc_write_eligible(uint64_t, arc_buf_hdr_t *); static void l2arc_read_done(zio_t *); static uint64_t buf_hash(uint64_t spa, const dva_t *dva, uint64_t birth) { uint8_t *vdva = (uint8_t *)dva; uint64_t crc = -1ULL; int i; ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); for (i = 0; i < sizeof (dva_t); i++) crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ vdva[i]) & 0xFF]; crc ^= (spa>>8) ^ birth; return (crc); } #define HDR_EMPTY(hdr) \ ((hdr)->b_dva.dva_word[0] == 0 && \ (hdr)->b_dva.dva_word[1] == 0) #define HDR_EQUAL(spa, dva, birth, hdr) \ ((hdr)->b_dva.dva_word[0] == (dva)->dva_word[0]) && \ ((hdr)->b_dva.dva_word[1] == (dva)->dva_word[1]) && \ ((hdr)->b_birth == birth) && ((hdr)->b_spa == spa) static void buf_discard_identity(arc_buf_hdr_t *hdr) { hdr->b_dva.dva_word[0] = 0; hdr->b_dva.dva_word[1] = 0; hdr->b_birth = 0; } static arc_buf_hdr_t * buf_hash_find(uint64_t spa, const blkptr_t *bp, kmutex_t **lockp) { const dva_t *dva = BP_IDENTITY(bp); uint64_t birth = BP_PHYSICAL_BIRTH(bp); uint64_t idx = BUF_HASH_INDEX(spa, dva, birth); kmutex_t *hash_lock = BUF_HASH_LOCK(idx); arc_buf_hdr_t *hdr; mutex_enter(hash_lock); for (hdr = buf_hash_table.ht_table[idx]; hdr != NULL; hdr = hdr->b_hash_next) { if (HDR_EQUAL(spa, dva, birth, hdr)) { *lockp = hash_lock; return (hdr); } } mutex_exit(hash_lock); *lockp = NULL; return (NULL); } /* * Insert an entry into the hash table. If there is already an element * equal to elem in the hash table, then the already existing element * will be returned and the new element will not be inserted. * Otherwise returns NULL. * If lockp == NULL, the caller is assumed to already hold the hash lock. */ static arc_buf_hdr_t * buf_hash_insert(arc_buf_hdr_t *hdr, kmutex_t **lockp) { uint64_t idx = BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth); kmutex_t *hash_lock = BUF_HASH_LOCK(idx); arc_buf_hdr_t *fhdr; uint32_t i; ASSERT(!DVA_IS_EMPTY(&hdr->b_dva)); ASSERT(hdr->b_birth != 0); ASSERT(!HDR_IN_HASH_TABLE(hdr)); if (lockp != NULL) { *lockp = hash_lock; mutex_enter(hash_lock); } else { ASSERT(MUTEX_HELD(hash_lock)); } for (fhdr = buf_hash_table.ht_table[idx], i = 0; fhdr != NULL; fhdr = fhdr->b_hash_next, i++) { if (HDR_EQUAL(hdr->b_spa, &hdr->b_dva, hdr->b_birth, fhdr)) return (fhdr); } hdr->b_hash_next = buf_hash_table.ht_table[idx]; buf_hash_table.ht_table[idx] = hdr; arc_hdr_set_flags(hdr, ARC_FLAG_IN_HASH_TABLE); /* collect some hash table performance data */ if (i > 0) { ARCSTAT_BUMP(arcstat_hash_collisions); if (i == 1) ARCSTAT_BUMP(arcstat_hash_chains); ARCSTAT_MAX(arcstat_hash_chain_max, i); } ARCSTAT_BUMP(arcstat_hash_elements); ARCSTAT_MAXSTAT(arcstat_hash_elements); return (NULL); } static void buf_hash_remove(arc_buf_hdr_t *hdr) { arc_buf_hdr_t *fhdr, **hdrp; uint64_t idx = BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth); ASSERT(MUTEX_HELD(BUF_HASH_LOCK(idx))); ASSERT(HDR_IN_HASH_TABLE(hdr)); hdrp = &buf_hash_table.ht_table[idx]; while ((fhdr = *hdrp) != hdr) { ASSERT3P(fhdr, !=, NULL); hdrp = &fhdr->b_hash_next; } *hdrp = hdr->b_hash_next; hdr->b_hash_next = NULL; arc_hdr_clear_flags(hdr, ARC_FLAG_IN_HASH_TABLE); /* collect some hash table performance data */ ARCSTAT_BUMPDOWN(arcstat_hash_elements); if (buf_hash_table.ht_table[idx] && buf_hash_table.ht_table[idx]->b_hash_next == NULL) ARCSTAT_BUMPDOWN(arcstat_hash_chains); } /* * Global data structures and functions for the buf kmem cache. */ static kmem_cache_t *hdr_full_cache; static kmem_cache_t *hdr_l2only_cache; static kmem_cache_t *buf_cache; static void buf_fini(void) { int i; kmem_free(buf_hash_table.ht_table, (buf_hash_table.ht_mask + 1) * sizeof (void *)); for (i = 0; i < BUF_LOCKS; i++) mutex_destroy(&buf_hash_table.ht_locks[i].ht_lock); kmem_cache_destroy(hdr_full_cache); kmem_cache_destroy(hdr_l2only_cache); kmem_cache_destroy(buf_cache); } /* * Constructor callback - called when the cache is empty * and a new buf is requested. */ /* ARGSUSED */ static int hdr_full_cons(void *vbuf, void *unused, int kmflag) { arc_buf_hdr_t *hdr = vbuf; bzero(hdr, HDR_FULL_SIZE); cv_init(&hdr->b_l1hdr.b_cv, NULL, CV_DEFAULT, NULL); refcount_create(&hdr->b_l1hdr.b_refcnt); mutex_init(&hdr->b_l1hdr.b_freeze_lock, NULL, MUTEX_DEFAULT, NULL); multilist_link_init(&hdr->b_l1hdr.b_arc_node); arc_space_consume(HDR_FULL_SIZE, ARC_SPACE_HDRS); return (0); } /* ARGSUSED */ static int hdr_l2only_cons(void *vbuf, void *unused, int kmflag) { arc_buf_hdr_t *hdr = vbuf; bzero(hdr, HDR_L2ONLY_SIZE); arc_space_consume(HDR_L2ONLY_SIZE, ARC_SPACE_L2HDRS); return (0); } /* ARGSUSED */ static int buf_cons(void *vbuf, void *unused, int kmflag) { arc_buf_t *buf = vbuf; bzero(buf, sizeof (arc_buf_t)); mutex_init(&buf->b_evict_lock, NULL, MUTEX_DEFAULT, NULL); arc_space_consume(sizeof (arc_buf_t), ARC_SPACE_HDRS); return (0); } /* * Destructor callback - called when a cached buf is * no longer required. */ /* ARGSUSED */ static void hdr_full_dest(void *vbuf, void *unused) { arc_buf_hdr_t *hdr = vbuf; ASSERT(HDR_EMPTY(hdr)); cv_destroy(&hdr->b_l1hdr.b_cv); refcount_destroy(&hdr->b_l1hdr.b_refcnt); mutex_destroy(&hdr->b_l1hdr.b_freeze_lock); ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); arc_space_return(HDR_FULL_SIZE, ARC_SPACE_HDRS); } /* ARGSUSED */ static void hdr_l2only_dest(void *vbuf, void *unused) { arc_buf_hdr_t *hdr = vbuf; ASSERT(HDR_EMPTY(hdr)); arc_space_return(HDR_L2ONLY_SIZE, ARC_SPACE_L2HDRS); } /* ARGSUSED */ static void buf_dest(void *vbuf, void *unused) { arc_buf_t *buf = vbuf; mutex_destroy(&buf->b_evict_lock); arc_space_return(sizeof (arc_buf_t), ARC_SPACE_HDRS); } /* * Reclaim callback -- invoked when memory is low. */ /* ARGSUSED */ static void hdr_recl(void *unused) { dprintf("hdr_recl called\n"); /* * umem calls the reclaim func when we destroy the buf cache, * which is after we do arc_fini(). */ if (!arc_dead) cv_signal(&arc_reclaim_thread_cv); } static void buf_init(void) { uint64_t *ct; uint64_t hsize = 1ULL << 12; int i, j; /* * The hash table is big enough to fill all of physical memory * with an average block size of zfs_arc_average_blocksize (default 8K). * By default, the table will take up * totalmem * sizeof(void*) / 8K (1MB per GB with 8-byte pointers). */ while (hsize * zfs_arc_average_blocksize < physmem * PAGESIZE) hsize <<= 1; retry: buf_hash_table.ht_mask = hsize - 1; buf_hash_table.ht_table = kmem_zalloc(hsize * sizeof (void*), KM_NOSLEEP); if (buf_hash_table.ht_table == NULL) { ASSERT(hsize > (1ULL << 8)); hsize >>= 1; goto retry; } hdr_full_cache = kmem_cache_create("arc_buf_hdr_t_full", HDR_FULL_SIZE, 0, hdr_full_cons, hdr_full_dest, hdr_recl, NULL, NULL, 0); hdr_l2only_cache = kmem_cache_create("arc_buf_hdr_t_l2only", HDR_L2ONLY_SIZE, 0, hdr_l2only_cons, hdr_l2only_dest, hdr_recl, NULL, NULL, 0); buf_cache = kmem_cache_create("arc_buf_t", sizeof (arc_buf_t), 0, buf_cons, buf_dest, NULL, NULL, NULL, 0); for (i = 0; i < 256; i++) for (ct = zfs_crc64_table + i, *ct = i, j = 8; j > 0; j--) *ct = (*ct >> 1) ^ (-(*ct & 1) & ZFS_CRC64_POLY); for (i = 0; i < BUF_LOCKS; i++) { mutex_init(&buf_hash_table.ht_locks[i].ht_lock, NULL, MUTEX_DEFAULT, NULL); } } /* * This is the size that the buf occupies in memory. If the buf is compressed, * it will correspond to the compressed size. You should use this method of * getting the buf size unless you explicitly need the logical size. */ int32_t arc_buf_size(arc_buf_t *buf) { return (ARC_BUF_COMPRESSED(buf) ? HDR_GET_PSIZE(buf->b_hdr) : HDR_GET_LSIZE(buf->b_hdr)); } int32_t arc_buf_lsize(arc_buf_t *buf) { return (HDR_GET_LSIZE(buf->b_hdr)); } enum zio_compress arc_get_compression(arc_buf_t *buf) { return (ARC_BUF_COMPRESSED(buf) ? HDR_GET_COMPRESS(buf->b_hdr) : ZIO_COMPRESS_OFF); } #define ARC_MINTIME (hz>>4) /* 62 ms */ static inline boolean_t arc_buf_is_shared(arc_buf_t *buf) { boolean_t shared = (buf->b_data != NULL && buf->b_data == buf->b_hdr->b_l1hdr.b_pdata); IMPLY(shared, HDR_SHARED_DATA(buf->b_hdr)); IMPLY(shared, ARC_BUF_SHARED(buf)); IMPLY(shared, ARC_BUF_COMPRESSED(buf) || ARC_BUF_LAST(buf)); /* * It would be nice to assert arc_can_share() too, but the "hdr isn't * already being shared" requirement prevents us from doing that. */ return (shared); } /* * Free the checksum associated with this header. If there is no checksum, this * is a no-op. */ static inline void arc_cksum_free(arc_buf_hdr_t *hdr) { ASSERT(HDR_HAS_L1HDR(hdr)); mutex_enter(&hdr->b_l1hdr.b_freeze_lock); if (hdr->b_l1hdr.b_freeze_cksum != NULL) { kmem_free(hdr->b_l1hdr.b_freeze_cksum, sizeof (zio_cksum_t)); hdr->b_l1hdr.b_freeze_cksum = NULL; } mutex_exit(&hdr->b_l1hdr.b_freeze_lock); } /* * Return true iff at least one of the bufs on hdr is not compressed. */ static boolean_t arc_hdr_has_uncompressed_buf(arc_buf_hdr_t *hdr) { for (arc_buf_t *b = hdr->b_l1hdr.b_buf; b != NULL; b = b->b_next) { if (!ARC_BUF_COMPRESSED(b)) { return (B_TRUE); } } return (B_FALSE); } /* * If we've turned on the ZFS_DEBUG_MODIFY flag, verify that the buf's data * matches the checksum that is stored in the hdr. If there is no checksum, * or if the buf is compressed, this is a no-op. */ static void arc_cksum_verify(arc_buf_t *buf) { arc_buf_hdr_t *hdr = buf->b_hdr; zio_cksum_t zc; if (!(zfs_flags & ZFS_DEBUG_MODIFY)) return; if (ARC_BUF_COMPRESSED(buf)) { ASSERT(hdr->b_l1hdr.b_freeze_cksum == NULL || arc_hdr_has_uncompressed_buf(hdr)); return; } ASSERT(HDR_HAS_L1HDR(hdr)); mutex_enter(&hdr->b_l1hdr.b_freeze_lock); if (hdr->b_l1hdr.b_freeze_cksum == NULL || HDR_IO_ERROR(hdr)) { mutex_exit(&hdr->b_l1hdr.b_freeze_lock); return; } fletcher_2_native(buf->b_data, arc_buf_size(buf), NULL, &zc); if (!ZIO_CHECKSUM_EQUAL(*hdr->b_l1hdr.b_freeze_cksum, zc)) panic("buffer modified while frozen!"); mutex_exit(&hdr->b_l1hdr.b_freeze_lock); } static boolean_t arc_cksum_is_equal(arc_buf_hdr_t *hdr, zio_t *zio) { enum zio_compress compress = BP_GET_COMPRESS(zio->io_bp); boolean_t valid_cksum; ASSERT(!BP_IS_EMBEDDED(zio->io_bp)); VERIFY3U(BP_GET_PSIZE(zio->io_bp), ==, HDR_GET_PSIZE(hdr)); /* * We rely on the blkptr's checksum to determine if the block * is valid or not. When compressed arc is enabled, the l2arc * writes the block to the l2arc just as it appears in the pool. * This allows us to use the blkptr's checksum to validate the * data that we just read off of the l2arc without having to store * a separate checksum in the arc_buf_hdr_t. However, if compressed * arc is disabled, then the data written to the l2arc is always * uncompressed and won't match the block as it exists in the main * pool. When this is the case, we must first compress it if it is * compressed on the main pool before we can validate the checksum. */ if (!HDR_COMPRESSION_ENABLED(hdr) && compress != ZIO_COMPRESS_OFF) { ASSERT3U(HDR_GET_COMPRESS(hdr), ==, ZIO_COMPRESS_OFF); uint64_t lsize = HDR_GET_LSIZE(hdr); uint64_t csize; void *cbuf = zio_buf_alloc(HDR_GET_PSIZE(hdr)); csize = zio_compress_data(compress, zio->io_data, cbuf, lsize); ASSERT3U(csize, <=, HDR_GET_PSIZE(hdr)); if (csize < HDR_GET_PSIZE(hdr)) { /* * Compressed blocks are always a multiple of the * smallest ashift in the pool. Ideally, we would * like to round up the csize to the next * spa_min_ashift but that value may have changed * since the block was last written. Instead, * we rely on the fact that the hdr's psize * was set to the psize of the block when it was * last written. We set the csize to that value * and zero out any part that should not contain * data. */ bzero((char *)cbuf + csize, HDR_GET_PSIZE(hdr) - csize); csize = HDR_GET_PSIZE(hdr); } zio_push_transform(zio, cbuf, csize, HDR_GET_PSIZE(hdr), NULL); } /* * Block pointers always store the checksum for the logical data. * If the block pointer has the gang bit set, then the checksum * it represents is for the reconstituted data and not for an * individual gang member. The zio pipeline, however, must be able to * determine the checksum of each of the gang constituents so it * treats the checksum comparison differently than what we need * for l2arc blocks. This prevents us from using the * zio_checksum_error() interface directly. Instead we must call the * zio_checksum_error_impl() so that we can ensure the checksum is * generated using the correct checksum algorithm and accounts for the * logical I/O size and not just a gang fragment. */ valid_cksum = (zio_checksum_error_impl(zio->io_spa, zio->io_bp, BP_GET_CHECKSUM(zio->io_bp), zio->io_data, zio->io_size, zio->io_offset, NULL) == 0); zio_pop_transforms(zio); return (valid_cksum); } /* * Given a buf full of data, if ZFS_DEBUG_MODIFY is enabled this computes a * checksum and attaches it to the buf's hdr so that we can ensure that the buf * isn't modified later on. If buf is compressed or there is already a checksum * on the hdr, this is a no-op (we only checksum uncompressed bufs). */ static void arc_cksum_compute(arc_buf_t *buf) { arc_buf_hdr_t *hdr = buf->b_hdr; if (!(zfs_flags & ZFS_DEBUG_MODIFY)) return; ASSERT(HDR_HAS_L1HDR(hdr)); mutex_enter(&buf->b_hdr->b_l1hdr.b_freeze_lock); if (hdr->b_l1hdr.b_freeze_cksum != NULL) { ASSERT(arc_hdr_has_uncompressed_buf(hdr)); mutex_exit(&hdr->b_l1hdr.b_freeze_lock); return; } else if (ARC_BUF_COMPRESSED(buf)) { mutex_exit(&hdr->b_l1hdr.b_freeze_lock); return; } ASSERT(!ARC_BUF_COMPRESSED(buf)); hdr->b_l1hdr.b_freeze_cksum = kmem_alloc(sizeof (zio_cksum_t), KM_SLEEP); fletcher_2_native(buf->b_data, arc_buf_size(buf), NULL, hdr->b_l1hdr.b_freeze_cksum); mutex_exit(&hdr->b_l1hdr.b_freeze_lock); arc_buf_watch(buf); } #ifndef _KERNEL typedef struct procctl { long cmd; prwatch_t prwatch; } procctl_t; #endif /* ARGSUSED */ static void arc_buf_unwatch(arc_buf_t *buf) { #ifndef _KERNEL if (arc_watch) { int result; procctl_t ctl; ctl.cmd = PCWATCH; ctl.prwatch.pr_vaddr = (uintptr_t)buf->b_data; ctl.prwatch.pr_size = 0; ctl.prwatch.pr_wflags = 0; result = write(arc_procfd, &ctl, sizeof (ctl)); ASSERT3U(result, ==, sizeof (ctl)); } #endif } /* ARGSUSED */ static void arc_buf_watch(arc_buf_t *buf) { #ifndef _KERNEL if (arc_watch) { int result; procctl_t ctl; ctl.cmd = PCWATCH; ctl.prwatch.pr_vaddr = (uintptr_t)buf->b_data; ctl.prwatch.pr_size = arc_buf_size(buf); ctl.prwatch.pr_wflags = WA_WRITE; result = write(arc_procfd, &ctl, sizeof (ctl)); ASSERT3U(result, ==, sizeof (ctl)); } #endif } static arc_buf_contents_t arc_buf_type(arc_buf_hdr_t *hdr) { arc_buf_contents_t type; if (HDR_ISTYPE_METADATA(hdr)) { type = ARC_BUFC_METADATA; } else { type = ARC_BUFC_DATA; } VERIFY3U(hdr->b_type, ==, type); return (type); } boolean_t arc_is_metadata(arc_buf_t *buf) { return (HDR_ISTYPE_METADATA(buf->b_hdr) != 0); } static uint32_t arc_bufc_to_flags(arc_buf_contents_t type) { switch (type) { case ARC_BUFC_DATA: /* metadata field is 0 if buffer contains normal data */ return (0); case ARC_BUFC_METADATA: return (ARC_FLAG_BUFC_METADATA); default: break; } panic("undefined ARC buffer type!"); return ((uint32_t)-1); } void arc_buf_thaw(arc_buf_t *buf) { arc_buf_hdr_t *hdr = buf->b_hdr; ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon); ASSERT(!HDR_IO_IN_PROGRESS(hdr)); arc_cksum_verify(buf); /* * Compressed buffers do not manipulate the b_freeze_cksum or * allocate b_thawed. */ if (ARC_BUF_COMPRESSED(buf)) { ASSERT(hdr->b_l1hdr.b_freeze_cksum == NULL || arc_hdr_has_uncompressed_buf(hdr)); return; } ASSERT(HDR_HAS_L1HDR(hdr)); arc_cksum_free(hdr); mutex_enter(&hdr->b_l1hdr.b_freeze_lock); #ifdef ZFS_DEBUG if (zfs_flags & ZFS_DEBUG_MODIFY) { if (hdr->b_l1hdr.b_thawed != NULL) kmem_free(hdr->b_l1hdr.b_thawed, 1); hdr->b_l1hdr.b_thawed = kmem_alloc(1, KM_SLEEP); } #endif mutex_exit(&hdr->b_l1hdr.b_freeze_lock); arc_buf_unwatch(buf); } void arc_buf_freeze(arc_buf_t *buf) { arc_buf_hdr_t *hdr = buf->b_hdr; kmutex_t *hash_lock; if (!(zfs_flags & ZFS_DEBUG_MODIFY)) return; if (ARC_BUF_COMPRESSED(buf)) { ASSERT(hdr->b_l1hdr.b_freeze_cksum == NULL || arc_hdr_has_uncompressed_buf(hdr)); return; } hash_lock = HDR_LOCK(hdr); mutex_enter(hash_lock); ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT(hdr->b_l1hdr.b_freeze_cksum != NULL || hdr->b_l1hdr.b_state == arc_anon); arc_cksum_compute(buf); mutex_exit(hash_lock); } /* * The arc_buf_hdr_t's b_flags should never be modified directly. Instead, * the following functions should be used to ensure that the flags are * updated in a thread-safe way. When manipulating the flags either * the hash_lock must be held or the hdr must be undiscoverable. This * ensures that we're not racing with any other threads when updating * the flags. */ static inline void arc_hdr_set_flags(arc_buf_hdr_t *hdr, arc_flags_t flags) { ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); hdr->b_flags |= flags; } static inline void arc_hdr_clear_flags(arc_buf_hdr_t *hdr, arc_flags_t flags) { ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); hdr->b_flags &= ~flags; } /* * Setting the compression bits in the arc_buf_hdr_t's b_flags is * done in a special way since we have to clear and set bits * at the same time. Consumers that wish to set the compression bits * must use this function to ensure that the flags are updated in * thread-safe manner. */ static void arc_hdr_set_compress(arc_buf_hdr_t *hdr, enum zio_compress cmp) { ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); /* * Holes and embedded blocks will always have a psize = 0 so * we ignore the compression of the blkptr and set the * arc_buf_hdr_t's compression to ZIO_COMPRESS_OFF. * Holes and embedded blocks remain anonymous so we don't * want to uncompress them. Mark them as uncompressed. */ if (!zfs_compressed_arc_enabled || HDR_GET_PSIZE(hdr) == 0) { arc_hdr_clear_flags(hdr, ARC_FLAG_COMPRESSED_ARC); HDR_SET_COMPRESS(hdr, ZIO_COMPRESS_OFF); ASSERT(!HDR_COMPRESSION_ENABLED(hdr)); ASSERT3U(HDR_GET_COMPRESS(hdr), ==, ZIO_COMPRESS_OFF); } else { arc_hdr_set_flags(hdr, ARC_FLAG_COMPRESSED_ARC); HDR_SET_COMPRESS(hdr, cmp); ASSERT3U(HDR_GET_COMPRESS(hdr), ==, cmp); ASSERT(HDR_COMPRESSION_ENABLED(hdr)); } } /* * Looks for another buf on the same hdr which has the data decompressed, copies * from it, and returns true. If no such buf exists, returns false. */ static boolean_t arc_buf_try_copy_decompressed_data(arc_buf_t *buf) { arc_buf_hdr_t *hdr = buf->b_hdr; boolean_t copied = B_FALSE; ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT3P(buf->b_data, !=, NULL); ASSERT(!ARC_BUF_COMPRESSED(buf)); for (arc_buf_t *from = hdr->b_l1hdr.b_buf; from != NULL; from = from->b_next) { /* can't use our own data buffer */ if (from == buf) { continue; } if (!ARC_BUF_COMPRESSED(from)) { bcopy(from->b_data, buf->b_data, arc_buf_size(buf)); copied = B_TRUE; break; } } /* * There were no decompressed bufs, so there should not be a * checksum on the hdr either. */ EQUIV(!copied, hdr->b_l1hdr.b_freeze_cksum == NULL); return (copied); } /* * Given a buf that has a data buffer attached to it, this function will * efficiently fill the buf with data of the specified compression setting from * the hdr and update the hdr's b_freeze_cksum if necessary. If the buf and hdr * are already sharing a data buf, no copy is performed. * * If the buf is marked as compressed but uncompressed data was requested, this * will allocate a new data buffer for the buf, remove that flag, and fill the * buf with uncompressed data. You can't request a compressed buf on a hdr with * uncompressed data, and (since we haven't added support for it yet) if you * want compressed data your buf must already be marked as compressed and have * the correct-sized data buffer. */ static int arc_buf_fill(arc_buf_t *buf, boolean_t compressed) { arc_buf_hdr_t *hdr = buf->b_hdr; boolean_t hdr_compressed = (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF); dmu_object_byteswap_t bswap = hdr->b_l1hdr.b_byteswap; ASSERT3P(buf->b_data, !=, NULL); IMPLY(compressed, hdr_compressed); IMPLY(compressed, ARC_BUF_COMPRESSED(buf)); if (hdr_compressed == compressed) { if (!arc_buf_is_shared(buf)) { bcopy(hdr->b_l1hdr.b_pdata, buf->b_data, arc_buf_size(buf)); } } else { ASSERT(hdr_compressed); ASSERT(!compressed); ASSERT3U(HDR_GET_LSIZE(hdr), !=, HDR_GET_PSIZE(hdr)); /* * If the buf is sharing its data with the hdr, unlink it and * allocate a new data buffer for the buf. */ if (arc_buf_is_shared(buf)) { ASSERT(ARC_BUF_COMPRESSED(buf)); /* We need to give the buf it's own b_data */ buf->b_flags &= ~ARC_BUF_FLAG_SHARED; buf->b_data = arc_get_data_buf(hdr, HDR_GET_LSIZE(hdr), buf); arc_hdr_clear_flags(hdr, ARC_FLAG_SHARED_DATA); /* Previously overhead was 0; just add new overhead */ ARCSTAT_INCR(arcstat_overhead_size, HDR_GET_LSIZE(hdr)); } else if (ARC_BUF_COMPRESSED(buf)) { /* We need to reallocate the buf's b_data */ arc_free_data_buf(hdr, buf->b_data, HDR_GET_PSIZE(hdr), buf); buf->b_data = arc_get_data_buf(hdr, HDR_GET_LSIZE(hdr), buf); /* We increased the size of b_data; update overhead */ ARCSTAT_INCR(arcstat_overhead_size, HDR_GET_LSIZE(hdr) - HDR_GET_PSIZE(hdr)); } /* * Regardless of the buf's previous compression settings, it * should not be compressed at the end of this function. */ buf->b_flags &= ~ARC_BUF_FLAG_COMPRESSED; /* * Try copying the data from another buf which already has a * decompressed version. If that's not possible, it's time to * bite the bullet and decompress the data from the hdr. */ if (arc_buf_try_copy_decompressed_data(buf)) { /* Skip byteswapping and checksumming (already done) */ ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, !=, NULL); return (0); } else { int error = zio_decompress_data(HDR_GET_COMPRESS(hdr), hdr->b_l1hdr.b_pdata, buf->b_data, HDR_GET_PSIZE(hdr), HDR_GET_LSIZE(hdr)); /* * Absent hardware errors or software bugs, this should * be impossible, but log it anyway so we can debug it. */ if (error != 0) { zfs_dbgmsg( "hdr %p, compress %d, psize %d, lsize %d", hdr, HDR_GET_COMPRESS(hdr), HDR_GET_PSIZE(hdr), HDR_GET_LSIZE(hdr)); return (SET_ERROR(EIO)); } } } /* Byteswap the buf's data if necessary */ if (bswap != DMU_BSWAP_NUMFUNCS) { ASSERT(!HDR_SHARED_DATA(hdr)); ASSERT3U(bswap, <, DMU_BSWAP_NUMFUNCS); dmu_ot_byteswap[bswap].ob_func(buf->b_data, HDR_GET_LSIZE(hdr)); } /* Compute the hdr's checksum if necessary */ arc_cksum_compute(buf); return (0); } int arc_decompress(arc_buf_t *buf) { return (arc_buf_fill(buf, B_FALSE)); } /* * Return the size of the block, b_pdata, that is stored in the arc_buf_hdr_t. */ static uint64_t arc_hdr_size(arc_buf_hdr_t *hdr) { uint64_t size; if (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF && HDR_GET_PSIZE(hdr) > 0) { size = HDR_GET_PSIZE(hdr); } else { ASSERT3U(HDR_GET_LSIZE(hdr), !=, 0); size = HDR_GET_LSIZE(hdr); } return (size); } /* * Increment the amount of evictable space in the arc_state_t's refcount. * We account for the space used by the hdr and the arc buf individually * so that we can add and remove them from the refcount individually. */ static void arc_evictable_space_increment(arc_buf_hdr_t *hdr, arc_state_t *state) { arc_buf_contents_t type = arc_buf_type(hdr); ASSERT(HDR_HAS_L1HDR(hdr)); if (GHOST_STATE(state)) { ASSERT0(hdr->b_l1hdr.b_bufcnt); ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); ASSERT3P(hdr->b_l1hdr.b_pdata, ==, NULL); (void) refcount_add_many(&state->arcs_esize[type], HDR_GET_LSIZE(hdr), hdr); return; } ASSERT(!GHOST_STATE(state)); if (hdr->b_l1hdr.b_pdata != NULL) { (void) refcount_add_many(&state->arcs_esize[type], arc_hdr_size(hdr), hdr); } for (arc_buf_t *buf = hdr->b_l1hdr.b_buf; buf != NULL; buf = buf->b_next) { if (arc_buf_is_shared(buf)) continue; (void) refcount_add_many(&state->arcs_esize[type], arc_buf_size(buf), buf); } } /* * Decrement the amount of evictable space in the arc_state_t's refcount. * We account for the space used by the hdr and the arc buf individually * so that we can add and remove them from the refcount individually. */ static void arc_evictable_space_decrement(arc_buf_hdr_t *hdr, arc_state_t *state) { arc_buf_contents_t type = arc_buf_type(hdr); ASSERT(HDR_HAS_L1HDR(hdr)); if (GHOST_STATE(state)) { ASSERT0(hdr->b_l1hdr.b_bufcnt); ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); ASSERT3P(hdr->b_l1hdr.b_pdata, ==, NULL); (void) refcount_remove_many(&state->arcs_esize[type], HDR_GET_LSIZE(hdr), hdr); return; } ASSERT(!GHOST_STATE(state)); if (hdr->b_l1hdr.b_pdata != NULL) { (void) refcount_remove_many(&state->arcs_esize[type], arc_hdr_size(hdr), hdr); } for (arc_buf_t *buf = hdr->b_l1hdr.b_buf; buf != NULL; buf = buf->b_next) { if (arc_buf_is_shared(buf)) continue; (void) refcount_remove_many(&state->arcs_esize[type], arc_buf_size(buf), buf); } } /* * Add a reference to this hdr indicating that someone is actively * referencing that memory. When the refcount transitions from 0 to 1, * we remove it from the respective arc_state_t list to indicate that * it is not evictable. */ static void add_reference(arc_buf_hdr_t *hdr, void *tag) { ASSERT(HDR_HAS_L1HDR(hdr)); if (!MUTEX_HELD(HDR_LOCK(hdr))) { ASSERT(hdr->b_l1hdr.b_state == arc_anon); ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); } arc_state_t *state = hdr->b_l1hdr.b_state; if ((refcount_add(&hdr->b_l1hdr.b_refcnt, tag) == 1) && (state != arc_anon)) { /* We don't use the L2-only state list. */ if (state != arc_l2c_only) { - multilist_remove(&state->arcs_list[arc_buf_type(hdr)], + multilist_remove(state->arcs_list[arc_buf_type(hdr)], hdr); arc_evictable_space_decrement(hdr, state); } /* remove the prefetch flag if we get a reference */ arc_hdr_clear_flags(hdr, ARC_FLAG_PREFETCH); } } /* * Remove a reference from this hdr. When the reference transitions from * 1 to 0 and we're not anonymous, then we add this hdr to the arc_state_t's * list making it eligible for eviction. */ static int remove_reference(arc_buf_hdr_t *hdr, kmutex_t *hash_lock, void *tag) { int cnt; arc_state_t *state = hdr->b_l1hdr.b_state; ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT(state == arc_anon || MUTEX_HELD(hash_lock)); ASSERT(!GHOST_STATE(state)); /* * arc_l2c_only counts as a ghost state so we don't need to explicitly * check to prevent usage of the arc_l2c_only list. */ if (((cnt = refcount_remove(&hdr->b_l1hdr.b_refcnt, tag)) == 0) && (state != arc_anon)) { - multilist_insert(&state->arcs_list[arc_buf_type(hdr)], hdr); + multilist_insert(state->arcs_list[arc_buf_type(hdr)], hdr); ASSERT3U(hdr->b_l1hdr.b_bufcnt, >, 0); arc_evictable_space_increment(hdr, state); } return (cnt); } /* * Move the supplied buffer to the indicated state. The hash lock * for the buffer must be held by the caller. */ static void arc_change_state(arc_state_t *new_state, arc_buf_hdr_t *hdr, kmutex_t *hash_lock) { arc_state_t *old_state; int64_t refcnt; uint32_t bufcnt; boolean_t update_old, update_new; arc_buf_contents_t buftype = arc_buf_type(hdr); /* * We almost always have an L1 hdr here, since we call arc_hdr_realloc() * in arc_read() when bringing a buffer out of the L2ARC. However, the * L1 hdr doesn't always exist when we change state to arc_anon before * destroying a header, in which case reallocating to add the L1 hdr is * pointless. */ if (HDR_HAS_L1HDR(hdr)) { old_state = hdr->b_l1hdr.b_state; refcnt = refcount_count(&hdr->b_l1hdr.b_refcnt); bufcnt = hdr->b_l1hdr.b_bufcnt; update_old = (bufcnt > 0 || hdr->b_l1hdr.b_pdata != NULL); } else { old_state = arc_l2c_only; refcnt = 0; bufcnt = 0; update_old = B_FALSE; } update_new = update_old; ASSERT(MUTEX_HELD(hash_lock)); ASSERT3P(new_state, !=, old_state); ASSERT(!GHOST_STATE(new_state) || bufcnt == 0); ASSERT(old_state != arc_anon || bufcnt <= 1); /* * If this buffer is evictable, transfer it from the * old state list to the new state list. */ if (refcnt == 0) { if (old_state != arc_anon && old_state != arc_l2c_only) { ASSERT(HDR_HAS_L1HDR(hdr)); - multilist_remove(&old_state->arcs_list[buftype], hdr); + multilist_remove(old_state->arcs_list[buftype], hdr); if (GHOST_STATE(old_state)) { ASSERT0(bufcnt); ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); update_old = B_TRUE; } arc_evictable_space_decrement(hdr, old_state); } if (new_state != arc_anon && new_state != arc_l2c_only) { /* * An L1 header always exists here, since if we're * moving to some L1-cached state (i.e. not l2c_only or * anonymous), we realloc the header to add an L1hdr * beforehand. */ ASSERT(HDR_HAS_L1HDR(hdr)); - multilist_insert(&new_state->arcs_list[buftype], hdr); + multilist_insert(new_state->arcs_list[buftype], hdr); if (GHOST_STATE(new_state)) { ASSERT0(bufcnt); ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); update_new = B_TRUE; } arc_evictable_space_increment(hdr, new_state); } } ASSERT(!HDR_EMPTY(hdr)); if (new_state == arc_anon && HDR_IN_HASH_TABLE(hdr)) buf_hash_remove(hdr); /* adjust state sizes (ignore arc_l2c_only) */ if (update_new && new_state != arc_l2c_only) { ASSERT(HDR_HAS_L1HDR(hdr)); if (GHOST_STATE(new_state)) { ASSERT0(bufcnt); /* * When moving a header to a ghost state, we first * remove all arc buffers. Thus, we'll have a * bufcnt of zero, and no arc buffer to use for * the reference. As a result, we use the arc * header pointer for the reference. */ (void) refcount_add_many(&new_state->arcs_size, HDR_GET_LSIZE(hdr), hdr); ASSERT3P(hdr->b_l1hdr.b_pdata, ==, NULL); } else { uint32_t buffers = 0; /* * Each individual buffer holds a unique reference, * thus we must remove each of these references one * at a time. */ for (arc_buf_t *buf = hdr->b_l1hdr.b_buf; buf != NULL; buf = buf->b_next) { ASSERT3U(bufcnt, !=, 0); buffers++; /* * When the arc_buf_t is sharing the data * block with the hdr, the owner of the * reference belongs to the hdr. Only * add to the refcount if the arc_buf_t is * not shared. */ if (arc_buf_is_shared(buf)) continue; (void) refcount_add_many(&new_state->arcs_size, arc_buf_size(buf), buf); } ASSERT3U(bufcnt, ==, buffers); if (hdr->b_l1hdr.b_pdata != NULL) { (void) refcount_add_many(&new_state->arcs_size, arc_hdr_size(hdr), hdr); } else { ASSERT(GHOST_STATE(old_state)); } } } if (update_old && old_state != arc_l2c_only) { ASSERT(HDR_HAS_L1HDR(hdr)); if (GHOST_STATE(old_state)) { ASSERT0(bufcnt); ASSERT3P(hdr->b_l1hdr.b_pdata, ==, NULL); /* * When moving a header off of a ghost state, * the header will not contain any arc buffers. * We use the arc header pointer for the reference * which is exactly what we did when we put the * header on the ghost state. */ (void) refcount_remove_many(&old_state->arcs_size, HDR_GET_LSIZE(hdr), hdr); } else { uint32_t buffers = 0; /* * Each individual buffer holds a unique reference, * thus we must remove each of these references one * at a time. */ for (arc_buf_t *buf = hdr->b_l1hdr.b_buf; buf != NULL; buf = buf->b_next) { ASSERT3U(bufcnt, !=, 0); buffers++; /* * When the arc_buf_t is sharing the data * block with the hdr, the owner of the * reference belongs to the hdr. Only * add to the refcount if the arc_buf_t is * not shared. */ if (arc_buf_is_shared(buf)) continue; (void) refcount_remove_many( &old_state->arcs_size, arc_buf_size(buf), buf); } ASSERT3U(bufcnt, ==, buffers); ASSERT3P(hdr->b_l1hdr.b_pdata, !=, NULL); (void) refcount_remove_many( &old_state->arcs_size, arc_hdr_size(hdr), hdr); } } if (HDR_HAS_L1HDR(hdr)) hdr->b_l1hdr.b_state = new_state; /* * L2 headers should never be on the L2 state list since they don't * have L1 headers allocated. */ - ASSERT(multilist_is_empty(&arc_l2c_only->arcs_list[ARC_BUFC_DATA]) && - multilist_is_empty(&arc_l2c_only->arcs_list[ARC_BUFC_METADATA])); + ASSERT(multilist_is_empty(arc_l2c_only->arcs_list[ARC_BUFC_DATA]) && + multilist_is_empty(arc_l2c_only->arcs_list[ARC_BUFC_METADATA])); } void arc_space_consume(uint64_t space, arc_space_type_t type) { ASSERT(type >= 0 && type < ARC_SPACE_NUMTYPES); switch (type) { case ARC_SPACE_DATA: ARCSTAT_INCR(arcstat_data_size, space); break; case ARC_SPACE_META: ARCSTAT_INCR(arcstat_metadata_size, space); break; case ARC_SPACE_OTHER: ARCSTAT_INCR(arcstat_other_size, space); break; case ARC_SPACE_HDRS: ARCSTAT_INCR(arcstat_hdr_size, space); break; case ARC_SPACE_L2HDRS: ARCSTAT_INCR(arcstat_l2_hdr_size, space); break; } if (type != ARC_SPACE_DATA) ARCSTAT_INCR(arcstat_meta_used, space); atomic_add_64(&arc_size, space); } void arc_space_return(uint64_t space, arc_space_type_t type) { ASSERT(type >= 0 && type < ARC_SPACE_NUMTYPES); switch (type) { case ARC_SPACE_DATA: ARCSTAT_INCR(arcstat_data_size, -space); break; case ARC_SPACE_META: ARCSTAT_INCR(arcstat_metadata_size, -space); break; case ARC_SPACE_OTHER: ARCSTAT_INCR(arcstat_other_size, -space); break; case ARC_SPACE_HDRS: ARCSTAT_INCR(arcstat_hdr_size, -space); break; case ARC_SPACE_L2HDRS: ARCSTAT_INCR(arcstat_l2_hdr_size, -space); break; } if (type != ARC_SPACE_DATA) { ASSERT(arc_meta_used >= space); if (arc_meta_max < arc_meta_used) arc_meta_max = arc_meta_used; ARCSTAT_INCR(arcstat_meta_used, -space); } ASSERT(arc_size >= space); atomic_add_64(&arc_size, -space); } /* * Given a hdr and a buf, returns whether that buf can share its b_data buffer * with the hdr's b_pdata. */ static boolean_t arc_can_share(arc_buf_hdr_t *hdr, arc_buf_t *buf) { /* * The criteria for sharing a hdr's data are: * 1. the hdr's compression matches the buf's compression * 2. the hdr doesn't need to be byteswapped * 3. the hdr isn't already being shared * 4. the buf is either compressed or it is the last buf in the hdr list * * Criterion #4 maintains the invariant that shared uncompressed * bufs must be the final buf in the hdr's b_buf list. Reading this, you * might ask, "if a compressed buf is allocated first, won't that be the * last thing in the list?", but in that case it's impossible to create * a shared uncompressed buf anyway (because the hdr must be compressed * to have the compressed buf). You might also think that #3 is * sufficient to make this guarantee, however it's possible * (specifically in the rare L2ARC write race mentioned in * arc_buf_alloc_impl()) there will be an existing uncompressed buf that * is sharable, but wasn't at the time of its allocation. Rather than * allow a new shared uncompressed buf to be created and then shuffle * the list around to make it the last element, this simply disallows * sharing if the new buf isn't the first to be added. */ ASSERT3P(buf->b_hdr, ==, hdr); boolean_t hdr_compressed = HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF; boolean_t buf_compressed = ARC_BUF_COMPRESSED(buf) != 0; return (buf_compressed == hdr_compressed && hdr->b_l1hdr.b_byteswap == DMU_BSWAP_NUMFUNCS && !HDR_SHARED_DATA(hdr) && (ARC_BUF_LAST(buf) || ARC_BUF_COMPRESSED(buf))); } /* * Allocate a buf for this hdr. If you care about the data that's in the hdr, * or if you want a compressed buffer, pass those flags in. Returns 0 if the * copy was made successfully, or an error code otherwise. */ static int arc_buf_alloc_impl(arc_buf_hdr_t *hdr, void *tag, boolean_t compressed, boolean_t fill, arc_buf_t **ret) { arc_buf_t *buf; ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT3U(HDR_GET_LSIZE(hdr), >, 0); VERIFY(hdr->b_type == ARC_BUFC_DATA || hdr->b_type == ARC_BUFC_METADATA); ASSERT3P(ret, !=, NULL); ASSERT3P(*ret, ==, NULL); buf = *ret = kmem_cache_alloc(buf_cache, KM_PUSHPAGE); buf->b_hdr = hdr; buf->b_data = NULL; buf->b_next = hdr->b_l1hdr.b_buf; buf->b_flags = 0; add_reference(hdr, tag); /* * We're about to change the hdr's b_flags. We must either * hold the hash_lock or be undiscoverable. */ ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); /* * Only honor requests for compressed bufs if the hdr is actually * compressed. */ if (compressed && HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF) buf->b_flags |= ARC_BUF_FLAG_COMPRESSED; /* * If the hdr's data can be shared then we share the data buffer and * set the appropriate bit in the hdr's b_flags to indicate the hdr is * sharing it's b_pdata with the arc_buf_t. Otherwise, we allocate a new * buffer to store the buf's data. * * There is one additional restriction here because we're sharing * hdr -> buf instead of the usual buf -> hdr: the hdr can't be actively * involved in an L2ARC write, because if this buf is used by an * arc_write() then the hdr's data buffer will be released when the * write completes, even though the L2ARC write might still be using it. */ boolean_t can_share = arc_can_share(hdr, buf) && !HDR_L2_WRITING(hdr); /* Set up b_data and sharing */ if (can_share) { buf->b_data = hdr->b_l1hdr.b_pdata; buf->b_flags |= ARC_BUF_FLAG_SHARED; arc_hdr_set_flags(hdr, ARC_FLAG_SHARED_DATA); } else { buf->b_data = arc_get_data_buf(hdr, arc_buf_size(buf), buf); ARCSTAT_INCR(arcstat_overhead_size, arc_buf_size(buf)); } VERIFY3P(buf->b_data, !=, NULL); hdr->b_l1hdr.b_buf = buf; hdr->b_l1hdr.b_bufcnt += 1; /* * If the user wants the data from the hdr, we need to either copy or * decompress the data. */ if (fill) { return (arc_buf_fill(buf, ARC_BUF_COMPRESSED(buf) != 0)); } return (0); } static char *arc_onloan_tag = "onloan"; static inline void arc_loaned_bytes_update(int64_t delta) { atomic_add_64(&arc_loaned_bytes, delta); /* assert that it did not wrap around */ ASSERT3S(atomic_add_64_nv(&arc_loaned_bytes, 0), >=, 0); } /* * Loan out an anonymous arc buffer. Loaned buffers are not counted as in * flight data by arc_tempreserve_space() until they are "returned". Loaned * buffers must be returned to the arc before they can be used by the DMU or * freed. */ arc_buf_t * arc_loan_buf(spa_t *spa, boolean_t is_metadata, int size) { arc_buf_t *buf = arc_alloc_buf(spa, arc_onloan_tag, is_metadata ? ARC_BUFC_METADATA : ARC_BUFC_DATA, size); arc_loaned_bytes_update(size); return (buf); } arc_buf_t * arc_loan_compressed_buf(spa_t *spa, uint64_t psize, uint64_t lsize, enum zio_compress compression_type) { arc_buf_t *buf = arc_alloc_compressed_buf(spa, arc_onloan_tag, psize, lsize, compression_type); arc_loaned_bytes_update(psize); return (buf); } /* * Return a loaned arc buffer to the arc. */ void arc_return_buf(arc_buf_t *buf, void *tag) { arc_buf_hdr_t *hdr = buf->b_hdr; ASSERT3P(buf->b_data, !=, NULL); ASSERT(HDR_HAS_L1HDR(hdr)); (void) refcount_add(&hdr->b_l1hdr.b_refcnt, tag); (void) refcount_remove(&hdr->b_l1hdr.b_refcnt, arc_onloan_tag); arc_loaned_bytes_update(-arc_buf_size(buf)); } /* Detach an arc_buf from a dbuf (tag) */ void arc_loan_inuse_buf(arc_buf_t *buf, void *tag) { arc_buf_hdr_t *hdr = buf->b_hdr; ASSERT3P(buf->b_data, !=, NULL); ASSERT(HDR_HAS_L1HDR(hdr)); (void) refcount_add(&hdr->b_l1hdr.b_refcnt, arc_onloan_tag); (void) refcount_remove(&hdr->b_l1hdr.b_refcnt, tag); arc_loaned_bytes_update(arc_buf_size(buf)); } static void l2arc_free_data_on_write(void *data, size_t size, arc_buf_contents_t type) { l2arc_data_free_t *df = kmem_alloc(sizeof (*df), KM_SLEEP); df->l2df_data = data; df->l2df_size = size; df->l2df_type = type; mutex_enter(&l2arc_free_on_write_mtx); list_insert_head(l2arc_free_on_write, df); mutex_exit(&l2arc_free_on_write_mtx); } static void arc_hdr_free_on_write(arc_buf_hdr_t *hdr) { arc_state_t *state = hdr->b_l1hdr.b_state; arc_buf_contents_t type = arc_buf_type(hdr); uint64_t size = arc_hdr_size(hdr); /* protected by hash lock, if in the hash table */ if (multilist_link_active(&hdr->b_l1hdr.b_arc_node)) { ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); ASSERT(state != arc_anon && state != arc_l2c_only); (void) refcount_remove_many(&state->arcs_esize[type], size, hdr); } (void) refcount_remove_many(&state->arcs_size, size, hdr); if (type == ARC_BUFC_METADATA) { arc_space_return(size, ARC_SPACE_META); } else { ASSERT(type == ARC_BUFC_DATA); arc_space_return(size, ARC_SPACE_DATA); } l2arc_free_data_on_write(hdr->b_l1hdr.b_pdata, size, type); } /* * Share the arc_buf_t's data with the hdr. Whenever we are sharing the * data buffer, we transfer the refcount ownership to the hdr and update * the appropriate kstats. */ static void arc_share_buf(arc_buf_hdr_t *hdr, arc_buf_t *buf) { arc_state_t *state = hdr->b_l1hdr.b_state; ASSERT(arc_can_share(hdr, buf)); ASSERT3P(hdr->b_l1hdr.b_pdata, ==, NULL); ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); /* * Start sharing the data buffer. We transfer the * refcount ownership to the hdr since it always owns * the refcount whenever an arc_buf_t is shared. */ refcount_transfer_ownership(&state->arcs_size, buf, hdr); hdr->b_l1hdr.b_pdata = buf->b_data; arc_hdr_set_flags(hdr, ARC_FLAG_SHARED_DATA); buf->b_flags |= ARC_BUF_FLAG_SHARED; /* * Since we've transferred ownership to the hdr we need * to increment its compressed and uncompressed kstats and * decrement the overhead size. */ ARCSTAT_INCR(arcstat_compressed_size, arc_hdr_size(hdr)); ARCSTAT_INCR(arcstat_uncompressed_size, HDR_GET_LSIZE(hdr)); ARCSTAT_INCR(arcstat_overhead_size, -arc_buf_size(buf)); } static void arc_unshare_buf(arc_buf_hdr_t *hdr, arc_buf_t *buf) { arc_state_t *state = hdr->b_l1hdr.b_state; ASSERT(arc_buf_is_shared(buf)); ASSERT3P(hdr->b_l1hdr.b_pdata, !=, NULL); ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); /* * We are no longer sharing this buffer so we need * to transfer its ownership to the rightful owner. */ refcount_transfer_ownership(&state->arcs_size, hdr, buf); arc_hdr_clear_flags(hdr, ARC_FLAG_SHARED_DATA); hdr->b_l1hdr.b_pdata = NULL; buf->b_flags &= ~ARC_BUF_FLAG_SHARED; /* * Since the buffer is no longer shared between * the arc buf and the hdr, count it as overhead. */ ARCSTAT_INCR(arcstat_compressed_size, -arc_hdr_size(hdr)); ARCSTAT_INCR(arcstat_uncompressed_size, -HDR_GET_LSIZE(hdr)); ARCSTAT_INCR(arcstat_overhead_size, arc_buf_size(buf)); } /* * Remove an arc_buf_t from the hdr's buf list and return the last * arc_buf_t on the list. If no buffers remain on the list then return * NULL. */ static arc_buf_t * arc_buf_remove(arc_buf_hdr_t *hdr, arc_buf_t *buf) { ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); arc_buf_t **bufp = &hdr->b_l1hdr.b_buf; arc_buf_t *lastbuf = NULL; /* * Remove the buf from the hdr list and locate the last * remaining buffer on the list. */ while (*bufp != NULL) { if (*bufp == buf) *bufp = buf->b_next; /* * If we've removed a buffer in the middle of * the list then update the lastbuf and update * bufp. */ if (*bufp != NULL) { lastbuf = *bufp; bufp = &(*bufp)->b_next; } } buf->b_next = NULL; ASSERT3P(lastbuf, !=, buf); IMPLY(hdr->b_l1hdr.b_bufcnt > 0, lastbuf != NULL); IMPLY(hdr->b_l1hdr.b_bufcnt > 0, hdr->b_l1hdr.b_buf != NULL); IMPLY(lastbuf != NULL, ARC_BUF_LAST(lastbuf)); return (lastbuf); } /* * Free up buf->b_data and pull the arc_buf_t off of the the arc_buf_hdr_t's * list and free it. */ static void arc_buf_destroy_impl(arc_buf_t *buf) { arc_buf_hdr_t *hdr = buf->b_hdr; /* * Free up the data associated with the buf but only if we're not * sharing this with the hdr. If we are sharing it with the hdr, the * hdr is responsible for doing the free. */ if (buf->b_data != NULL) { /* * We're about to change the hdr's b_flags. We must either * hold the hash_lock or be undiscoverable. */ ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); arc_cksum_verify(buf); arc_buf_unwatch(buf); if (arc_buf_is_shared(buf)) { arc_hdr_clear_flags(hdr, ARC_FLAG_SHARED_DATA); } else { uint64_t size = arc_buf_size(buf); arc_free_data_buf(hdr, buf->b_data, size, buf); ARCSTAT_INCR(arcstat_overhead_size, -size); } buf->b_data = NULL; ASSERT(hdr->b_l1hdr.b_bufcnt > 0); hdr->b_l1hdr.b_bufcnt -= 1; } arc_buf_t *lastbuf = arc_buf_remove(hdr, buf); if (ARC_BUF_SHARED(buf) && !ARC_BUF_COMPRESSED(buf)) { /* * If the current arc_buf_t is sharing its data buffer with the * hdr, then reassign the hdr's b_pdata to share it with the new * buffer at the end of the list. The shared buffer is always * the last one on the hdr's buffer list. * * There is an equivalent case for compressed bufs, but since * they aren't guaranteed to be the last buf in the list and * that is an exceedingly rare case, we just allow that space be * wasted temporarily. */ if (lastbuf != NULL) { /* Only one buf can be shared at once */ VERIFY(!arc_buf_is_shared(lastbuf)); /* hdr is uncompressed so can't have compressed buf */ VERIFY(!ARC_BUF_COMPRESSED(lastbuf)); ASSERT3P(hdr->b_l1hdr.b_pdata, !=, NULL); arc_hdr_free_pdata(hdr); /* * We must setup a new shared block between the * last buffer and the hdr. The data would have * been allocated by the arc buf so we need to transfer * ownership to the hdr since it's now being shared. */ arc_share_buf(hdr, lastbuf); } } else if (HDR_SHARED_DATA(hdr)) { /* * Uncompressed shared buffers are always at the end * of the list. Compressed buffers don't have the * same requirements. This makes it hard to * simply assert that the lastbuf is shared so * we rely on the hdr's compression flags to determine * if we have a compressed, shared buffer. */ ASSERT3P(lastbuf, !=, NULL); ASSERT(arc_buf_is_shared(lastbuf) || HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF); } /* * Free the checksum if we're removing the last uncompressed buf from * this hdr. */ if (!arc_hdr_has_uncompressed_buf(hdr)) { arc_cksum_free(hdr); } /* clean up the buf */ buf->b_hdr = NULL; kmem_cache_free(buf_cache, buf); } static void arc_hdr_alloc_pdata(arc_buf_hdr_t *hdr) { ASSERT3U(HDR_GET_LSIZE(hdr), >, 0); ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT(!HDR_SHARED_DATA(hdr)); ASSERT3P(hdr->b_l1hdr.b_pdata, ==, NULL); hdr->b_l1hdr.b_pdata = arc_get_data_buf(hdr, arc_hdr_size(hdr), hdr); hdr->b_l1hdr.b_byteswap = DMU_BSWAP_NUMFUNCS; ASSERT3P(hdr->b_l1hdr.b_pdata, !=, NULL); ARCSTAT_INCR(arcstat_compressed_size, arc_hdr_size(hdr)); ARCSTAT_INCR(arcstat_uncompressed_size, HDR_GET_LSIZE(hdr)); } static void arc_hdr_free_pdata(arc_buf_hdr_t *hdr) { ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT3P(hdr->b_l1hdr.b_pdata, !=, NULL); /* * If the hdr is currently being written to the l2arc then * we defer freeing the data by adding it to the l2arc_free_on_write * list. The l2arc will free the data once it's finished * writing it to the l2arc device. */ if (HDR_L2_WRITING(hdr)) { arc_hdr_free_on_write(hdr); ARCSTAT_BUMP(arcstat_l2_free_on_write); } else { arc_free_data_buf(hdr, hdr->b_l1hdr.b_pdata, arc_hdr_size(hdr), hdr); } hdr->b_l1hdr.b_pdata = NULL; hdr->b_l1hdr.b_byteswap = DMU_BSWAP_NUMFUNCS; ARCSTAT_INCR(arcstat_compressed_size, -arc_hdr_size(hdr)); ARCSTAT_INCR(arcstat_uncompressed_size, -HDR_GET_LSIZE(hdr)); } static arc_buf_hdr_t * arc_hdr_alloc(uint64_t spa, int32_t psize, int32_t lsize, enum zio_compress compression_type, arc_buf_contents_t type) { arc_buf_hdr_t *hdr; VERIFY(type == ARC_BUFC_DATA || type == ARC_BUFC_METADATA); hdr = kmem_cache_alloc(hdr_full_cache, KM_PUSHPAGE); ASSERT(HDR_EMPTY(hdr)); ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL); ASSERT3P(hdr->b_l1hdr.b_thawed, ==, NULL); HDR_SET_PSIZE(hdr, psize); HDR_SET_LSIZE(hdr, lsize); hdr->b_spa = spa; hdr->b_type = type; hdr->b_flags = 0; arc_hdr_set_flags(hdr, arc_bufc_to_flags(type) | ARC_FLAG_HAS_L1HDR); arc_hdr_set_compress(hdr, compression_type); hdr->b_l1hdr.b_state = arc_anon; hdr->b_l1hdr.b_arc_access = 0; hdr->b_l1hdr.b_bufcnt = 0; hdr->b_l1hdr.b_buf = NULL; /* * Allocate the hdr's buffer. This will contain either * the compressed or uncompressed data depending on the block * it references and compressed arc enablement. */ arc_hdr_alloc_pdata(hdr); ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); return (hdr); } /* * Transition between the two allocation states for the arc_buf_hdr struct. * The arc_buf_hdr struct can be allocated with (hdr_full_cache) or without * (hdr_l2only_cache) the fields necessary for the L1 cache - the smaller * version is used when a cache buffer is only in the L2ARC in order to reduce * memory usage. */ static arc_buf_hdr_t * arc_hdr_realloc(arc_buf_hdr_t *hdr, kmem_cache_t *old, kmem_cache_t *new) { ASSERT(HDR_HAS_L2HDR(hdr)); arc_buf_hdr_t *nhdr; l2arc_dev_t *dev = hdr->b_l2hdr.b_dev; ASSERT((old == hdr_full_cache && new == hdr_l2only_cache) || (old == hdr_l2only_cache && new == hdr_full_cache)); nhdr = kmem_cache_alloc(new, KM_PUSHPAGE); ASSERT(MUTEX_HELD(HDR_LOCK(hdr))); buf_hash_remove(hdr); bcopy(hdr, nhdr, HDR_L2ONLY_SIZE); if (new == hdr_full_cache) { arc_hdr_set_flags(nhdr, ARC_FLAG_HAS_L1HDR); /* * arc_access and arc_change_state need to be aware that a * header has just come out of L2ARC, so we set its state to * l2c_only even though it's about to change. */ nhdr->b_l1hdr.b_state = arc_l2c_only; /* Verify previous threads set to NULL before freeing */ ASSERT3P(nhdr->b_l1hdr.b_pdata, ==, NULL); } else { ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); ASSERT0(hdr->b_l1hdr.b_bufcnt); ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL); /* * If we've reached here, We must have been called from * arc_evict_hdr(), as such we should have already been * removed from any ghost list we were previously on * (which protects us from racing with arc_evict_state), * thus no locking is needed during this check. */ ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); /* * A buffer must not be moved into the arc_l2c_only * state if it's not finished being written out to the * l2arc device. Otherwise, the b_l1hdr.b_pdata field * might try to be accessed, even though it was removed. */ VERIFY(!HDR_L2_WRITING(hdr)); VERIFY3P(hdr->b_l1hdr.b_pdata, ==, NULL); #ifdef ZFS_DEBUG if (hdr->b_l1hdr.b_thawed != NULL) { kmem_free(hdr->b_l1hdr.b_thawed, 1); hdr->b_l1hdr.b_thawed = NULL; } #endif arc_hdr_clear_flags(nhdr, ARC_FLAG_HAS_L1HDR); } /* * The header has been reallocated so we need to re-insert it into any * lists it was on. */ (void) buf_hash_insert(nhdr, NULL); ASSERT(list_link_active(&hdr->b_l2hdr.b_l2node)); mutex_enter(&dev->l2ad_mtx); /* * We must place the realloc'ed header back into the list at * the same spot. Otherwise, if it's placed earlier in the list, * l2arc_write_buffers() could find it during the function's * write phase, and try to write it out to the l2arc. */ list_insert_after(&dev->l2ad_buflist, hdr, nhdr); list_remove(&dev->l2ad_buflist, hdr); mutex_exit(&dev->l2ad_mtx); /* * Since we're using the pointer address as the tag when * incrementing and decrementing the l2ad_alloc refcount, we * must remove the old pointer (that we're about to destroy) and * add the new pointer to the refcount. Otherwise we'd remove * the wrong pointer address when calling arc_hdr_destroy() later. */ (void) refcount_remove_many(&dev->l2ad_alloc, arc_hdr_size(hdr), hdr); (void) refcount_add_many(&dev->l2ad_alloc, arc_hdr_size(nhdr), nhdr); buf_discard_identity(hdr); kmem_cache_free(old, hdr); return (nhdr); } /* * Allocate a new arc_buf_hdr_t and arc_buf_t and return the buf to the caller. * The buf is returned thawed since we expect the consumer to modify it. */ arc_buf_t * arc_alloc_buf(spa_t *spa, void *tag, arc_buf_contents_t type, int32_t size) { arc_buf_hdr_t *hdr = arc_hdr_alloc(spa_load_guid(spa), size, size, ZIO_COMPRESS_OFF, type); ASSERT(!MUTEX_HELD(HDR_LOCK(hdr))); arc_buf_t *buf = NULL; VERIFY0(arc_buf_alloc_impl(hdr, tag, B_FALSE, B_FALSE, &buf)); arc_buf_thaw(buf); return (buf); } /* * Allocate a compressed buf in the same manner as arc_alloc_buf. Don't use this * for bufs containing metadata. */ arc_buf_t * arc_alloc_compressed_buf(spa_t *spa, void *tag, uint64_t psize, uint64_t lsize, enum zio_compress compression_type) { ASSERT3U(lsize, >, 0); ASSERT3U(lsize, >=, psize); ASSERT(compression_type > ZIO_COMPRESS_OFF); ASSERT(compression_type < ZIO_COMPRESS_FUNCTIONS); arc_buf_hdr_t *hdr = arc_hdr_alloc(spa_load_guid(spa), psize, lsize, compression_type, ARC_BUFC_DATA); ASSERT(!MUTEX_HELD(HDR_LOCK(hdr))); arc_buf_t *buf = NULL; VERIFY0(arc_buf_alloc_impl(hdr, tag, B_TRUE, B_FALSE, &buf)); arc_buf_thaw(buf); ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL); return (buf); } static void arc_hdr_l2hdr_destroy(arc_buf_hdr_t *hdr) { l2arc_buf_hdr_t *l2hdr = &hdr->b_l2hdr; l2arc_dev_t *dev = l2hdr->b_dev; uint64_t asize = arc_hdr_size(hdr); ASSERT(MUTEX_HELD(&dev->l2ad_mtx)); ASSERT(HDR_HAS_L2HDR(hdr)); list_remove(&dev->l2ad_buflist, hdr); ARCSTAT_INCR(arcstat_l2_asize, -asize); ARCSTAT_INCR(arcstat_l2_size, -HDR_GET_LSIZE(hdr)); vdev_space_update(dev->l2ad_vdev, -asize, 0, 0); (void) refcount_remove_many(&dev->l2ad_alloc, asize, hdr); arc_hdr_clear_flags(hdr, ARC_FLAG_HAS_L2HDR); } static void arc_hdr_destroy(arc_buf_hdr_t *hdr) { if (HDR_HAS_L1HDR(hdr)) { ASSERT(hdr->b_l1hdr.b_buf == NULL || hdr->b_l1hdr.b_bufcnt > 0); ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon); } ASSERT(!HDR_IO_IN_PROGRESS(hdr)); ASSERT(!HDR_IN_HASH_TABLE(hdr)); if (!HDR_EMPTY(hdr)) buf_discard_identity(hdr); if (HDR_HAS_L2HDR(hdr)) { l2arc_dev_t *dev = hdr->b_l2hdr.b_dev; boolean_t buflist_held = MUTEX_HELD(&dev->l2ad_mtx); if (!buflist_held) mutex_enter(&dev->l2ad_mtx); /* * Even though we checked this conditional above, we * need to check this again now that we have the * l2ad_mtx. This is because we could be racing with * another thread calling l2arc_evict() which might have * destroyed this header's L2 portion as we were waiting * to acquire the l2ad_mtx. If that happens, we don't * want to re-destroy the header's L2 portion. */ if (HDR_HAS_L2HDR(hdr)) arc_hdr_l2hdr_destroy(hdr); if (!buflist_held) mutex_exit(&dev->l2ad_mtx); } if (HDR_HAS_L1HDR(hdr)) { arc_cksum_free(hdr); while (hdr->b_l1hdr.b_buf != NULL) arc_buf_destroy_impl(hdr->b_l1hdr.b_buf); #ifdef ZFS_DEBUG if (hdr->b_l1hdr.b_thawed != NULL) { kmem_free(hdr->b_l1hdr.b_thawed, 1); hdr->b_l1hdr.b_thawed = NULL; } #endif if (hdr->b_l1hdr.b_pdata != NULL) { arc_hdr_free_pdata(hdr); } } ASSERT3P(hdr->b_hash_next, ==, NULL); if (HDR_HAS_L1HDR(hdr)) { ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL); kmem_cache_free(hdr_full_cache, hdr); } else { kmem_cache_free(hdr_l2only_cache, hdr); } } void arc_buf_destroy(arc_buf_t *buf, void* tag) { arc_buf_hdr_t *hdr = buf->b_hdr; kmutex_t *hash_lock = HDR_LOCK(hdr); if (hdr->b_l1hdr.b_state == arc_anon) { ASSERT3U(hdr->b_l1hdr.b_bufcnt, ==, 1); ASSERT(!HDR_IO_IN_PROGRESS(hdr)); VERIFY0(remove_reference(hdr, NULL, tag)); arc_hdr_destroy(hdr); return; } mutex_enter(hash_lock); ASSERT3P(hdr, ==, buf->b_hdr); ASSERT(hdr->b_l1hdr.b_bufcnt > 0); ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); ASSERT3P(hdr->b_l1hdr.b_state, !=, arc_anon); ASSERT3P(buf->b_data, !=, NULL); (void) remove_reference(hdr, hash_lock, tag); arc_buf_destroy_impl(buf); mutex_exit(hash_lock); } /* * Evict the arc_buf_hdr that is provided as a parameter. The resultant * state of the header is dependent on it's state prior to entering this * function. The following transitions are possible: * * - arc_mru -> arc_mru_ghost * - arc_mfu -> arc_mfu_ghost * - arc_mru_ghost -> arc_l2c_only * - arc_mru_ghost -> deleted * - arc_mfu_ghost -> arc_l2c_only * - arc_mfu_ghost -> deleted */ static int64_t arc_evict_hdr(arc_buf_hdr_t *hdr, kmutex_t *hash_lock) { arc_state_t *evicted_state, *state; int64_t bytes_evicted = 0; ASSERT(MUTEX_HELD(hash_lock)); ASSERT(HDR_HAS_L1HDR(hdr)); state = hdr->b_l1hdr.b_state; if (GHOST_STATE(state)) { ASSERT(!HDR_IO_IN_PROGRESS(hdr)); ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); /* * l2arc_write_buffers() relies on a header's L1 portion * (i.e. its b_pdata field) during its write phase. * Thus, we cannot push a header onto the arc_l2c_only * state (removing it's L1 piece) until the header is * done being written to the l2arc. */ if (HDR_HAS_L2HDR(hdr) && HDR_L2_WRITING(hdr)) { ARCSTAT_BUMP(arcstat_evict_l2_skip); return (bytes_evicted); } ARCSTAT_BUMP(arcstat_deleted); bytes_evicted += HDR_GET_LSIZE(hdr); DTRACE_PROBE1(arc__delete, arc_buf_hdr_t *, hdr); ASSERT3P(hdr->b_l1hdr.b_pdata, ==, NULL); if (HDR_HAS_L2HDR(hdr)) { /* * This buffer is cached on the 2nd Level ARC; * don't destroy the header. */ arc_change_state(arc_l2c_only, hdr, hash_lock); /* * dropping from L1+L2 cached to L2-only, * realloc to remove the L1 header. */ hdr = arc_hdr_realloc(hdr, hdr_full_cache, hdr_l2only_cache); } else { arc_change_state(arc_anon, hdr, hash_lock); arc_hdr_destroy(hdr); } return (bytes_evicted); } ASSERT(state == arc_mru || state == arc_mfu); evicted_state = (state == arc_mru) ? arc_mru_ghost : arc_mfu_ghost; /* prefetch buffers have a minimum lifespan */ if (HDR_IO_IN_PROGRESS(hdr) || ((hdr->b_flags & (ARC_FLAG_PREFETCH | ARC_FLAG_INDIRECT)) && ddi_get_lbolt() - hdr->b_l1hdr.b_arc_access < arc_min_prefetch_lifespan)) { ARCSTAT_BUMP(arcstat_evict_skip); return (bytes_evicted); } ASSERT0(refcount_count(&hdr->b_l1hdr.b_refcnt)); while (hdr->b_l1hdr.b_buf) { arc_buf_t *buf = hdr->b_l1hdr.b_buf; if (!mutex_tryenter(&buf->b_evict_lock)) { ARCSTAT_BUMP(arcstat_mutex_miss); break; } if (buf->b_data != NULL) bytes_evicted += HDR_GET_LSIZE(hdr); mutex_exit(&buf->b_evict_lock); arc_buf_destroy_impl(buf); } if (HDR_HAS_L2HDR(hdr)) { ARCSTAT_INCR(arcstat_evict_l2_cached, HDR_GET_LSIZE(hdr)); } else { if (l2arc_write_eligible(hdr->b_spa, hdr)) { ARCSTAT_INCR(arcstat_evict_l2_eligible, HDR_GET_LSIZE(hdr)); } else { ARCSTAT_INCR(arcstat_evict_l2_ineligible, HDR_GET_LSIZE(hdr)); } } if (hdr->b_l1hdr.b_bufcnt == 0) { arc_cksum_free(hdr); bytes_evicted += arc_hdr_size(hdr); /* * If this hdr is being evicted and has a compressed * buffer then we discard it here before we change states. * This ensures that the accounting is updated correctly * in arc_free_data_buf(). */ arc_hdr_free_pdata(hdr); arc_change_state(evicted_state, hdr, hash_lock); ASSERT(HDR_IN_HASH_TABLE(hdr)); arc_hdr_set_flags(hdr, ARC_FLAG_IN_HASH_TABLE); DTRACE_PROBE1(arc__evict, arc_buf_hdr_t *, hdr); } return (bytes_evicted); } static uint64_t arc_evict_state_impl(multilist_t *ml, int idx, arc_buf_hdr_t *marker, uint64_t spa, int64_t bytes) { multilist_sublist_t *mls; uint64_t bytes_evicted = 0; arc_buf_hdr_t *hdr; kmutex_t *hash_lock; int evict_count = 0; ASSERT3P(marker, !=, NULL); IMPLY(bytes < 0, bytes == ARC_EVICT_ALL); mls = multilist_sublist_lock(ml, idx); for (hdr = multilist_sublist_prev(mls, marker); hdr != NULL; hdr = multilist_sublist_prev(mls, marker)) { if ((bytes != ARC_EVICT_ALL && bytes_evicted >= bytes) || (evict_count >= zfs_arc_evict_batch_limit)) break; /* * To keep our iteration location, move the marker * forward. Since we're not holding hdr's hash lock, we * must be very careful and not remove 'hdr' from the * sublist. Otherwise, other consumers might mistake the * 'hdr' as not being on a sublist when they call the * multilist_link_active() function (they all rely on * the hash lock protecting concurrent insertions and * removals). multilist_sublist_move_forward() was * specifically implemented to ensure this is the case * (only 'marker' will be removed and re-inserted). */ multilist_sublist_move_forward(mls, marker); /* * The only case where the b_spa field should ever be * zero, is the marker headers inserted by * arc_evict_state(). It's possible for multiple threads * to be calling arc_evict_state() concurrently (e.g. * dsl_pool_close() and zio_inject_fault()), so we must * skip any markers we see from these other threads. */ if (hdr->b_spa == 0) continue; /* we're only interested in evicting buffers of a certain spa */ if (spa != 0 && hdr->b_spa != spa) { ARCSTAT_BUMP(arcstat_evict_skip); continue; } hash_lock = HDR_LOCK(hdr); /* * We aren't calling this function from any code path * that would already be holding a hash lock, so we're * asserting on this assumption to be defensive in case * this ever changes. Without this check, it would be * possible to incorrectly increment arcstat_mutex_miss * below (e.g. if the code changed such that we called * this function with a hash lock held). */ ASSERT(!MUTEX_HELD(hash_lock)); if (mutex_tryenter(hash_lock)) { uint64_t evicted = arc_evict_hdr(hdr, hash_lock); mutex_exit(hash_lock); bytes_evicted += evicted; /* * If evicted is zero, arc_evict_hdr() must have * decided to skip this header, don't increment * evict_count in this case. */ if (evicted != 0) evict_count++; /* * If arc_size isn't overflowing, signal any * threads that might happen to be waiting. * * For each header evicted, we wake up a single * thread. If we used cv_broadcast, we could * wake up "too many" threads causing arc_size * to significantly overflow arc_c; since * arc_get_data_buf() doesn't check for overflow * when it's woken up (it doesn't because it's * possible for the ARC to be overflowing while * full of un-evictable buffers, and the * function should proceed in this case). * * If threads are left sleeping, due to not * using cv_broadcast, they will be woken up * just before arc_reclaim_thread() sleeps. */ mutex_enter(&arc_reclaim_lock); if (!arc_is_overflowing()) cv_signal(&arc_reclaim_waiters_cv); mutex_exit(&arc_reclaim_lock); } else { ARCSTAT_BUMP(arcstat_mutex_miss); } } multilist_sublist_unlock(mls); return (bytes_evicted); } /* * Evict buffers from the given arc state, until we've removed the * specified number of bytes. Move the removed buffers to the * appropriate evict state. * * This function makes a "best effort". It skips over any buffers * it can't get a hash_lock on, and so, may not catch all candidates. * It may also return without evicting as much space as requested. * * If bytes is specified using the special value ARC_EVICT_ALL, this * will evict all available (i.e. unlocked and evictable) buffers from * the given arc state; which is used by arc_flush(). */ static uint64_t arc_evict_state(arc_state_t *state, uint64_t spa, int64_t bytes, arc_buf_contents_t type) { uint64_t total_evicted = 0; - multilist_t *ml = &state->arcs_list[type]; + multilist_t *ml = state->arcs_list[type]; int num_sublists; arc_buf_hdr_t **markers; IMPLY(bytes < 0, bytes == ARC_EVICT_ALL); num_sublists = multilist_get_num_sublists(ml); /* * If we've tried to evict from each sublist, made some * progress, but still have not hit the target number of bytes * to evict, we want to keep trying. The markers allow us to * pick up where we left off for each individual sublist, rather * than starting from the tail each time. */ markers = kmem_zalloc(sizeof (*markers) * num_sublists, KM_SLEEP); for (int i = 0; i < num_sublists; i++) { markers[i] = kmem_cache_alloc(hdr_full_cache, KM_SLEEP); /* * A b_spa of 0 is used to indicate that this header is * a marker. This fact is used in arc_adjust_type() and * arc_evict_state_impl(). */ markers[i]->b_spa = 0; multilist_sublist_t *mls = multilist_sublist_lock(ml, i); multilist_sublist_insert_tail(mls, markers[i]); multilist_sublist_unlock(mls); } /* * While we haven't hit our target number of bytes to evict, or * we're evicting all available buffers. */ while (total_evicted < bytes || bytes == ARC_EVICT_ALL) { /* * Start eviction using a randomly selected sublist, * this is to try and evenly balance eviction across all * sublists. Always starting at the same sublist * (e.g. index 0) would cause evictions to favor certain * sublists over others. */ int sublist_idx = multilist_get_random_index(ml); uint64_t scan_evicted = 0; for (int i = 0; i < num_sublists; i++) { uint64_t bytes_remaining; uint64_t bytes_evicted; if (bytes == ARC_EVICT_ALL) bytes_remaining = ARC_EVICT_ALL; else if (total_evicted < bytes) bytes_remaining = bytes - total_evicted; else break; bytes_evicted = arc_evict_state_impl(ml, sublist_idx, markers[sublist_idx], spa, bytes_remaining); scan_evicted += bytes_evicted; total_evicted += bytes_evicted; /* we've reached the end, wrap to the beginning */ if (++sublist_idx >= num_sublists) sublist_idx = 0; } /* * If we didn't evict anything during this scan, we have * no reason to believe we'll evict more during another * scan, so break the loop. */ if (scan_evicted == 0) { /* This isn't possible, let's make that obvious */ ASSERT3S(bytes, !=, 0); /* * When bytes is ARC_EVICT_ALL, the only way to * break the loop is when scan_evicted is zero. * In that case, we actually have evicted enough, * so we don't want to increment the kstat. */ if (bytes != ARC_EVICT_ALL) { ASSERT3S(total_evicted, <, bytes); ARCSTAT_BUMP(arcstat_evict_not_enough); } break; } } for (int i = 0; i < num_sublists; i++) { multilist_sublist_t *mls = multilist_sublist_lock(ml, i); multilist_sublist_remove(mls, markers[i]); multilist_sublist_unlock(mls); kmem_cache_free(hdr_full_cache, markers[i]); } kmem_free(markers, sizeof (*markers) * num_sublists); return (total_evicted); } /* * Flush all "evictable" data of the given type from the arc state * specified. This will not evict any "active" buffers (i.e. referenced). * * When 'retry' is set to B_FALSE, the function will make a single pass * over the state and evict any buffers that it can. Since it doesn't * continually retry the eviction, it might end up leaving some buffers * in the ARC due to lock misses. * * When 'retry' is set to B_TRUE, the function will continually retry the * eviction until *all* evictable buffers have been removed from the * state. As a result, if concurrent insertions into the state are * allowed (e.g. if the ARC isn't shutting down), this function might * wind up in an infinite loop, continually trying to evict buffers. */ static uint64_t arc_flush_state(arc_state_t *state, uint64_t spa, arc_buf_contents_t type, boolean_t retry) { uint64_t evicted = 0; while (refcount_count(&state->arcs_esize[type]) != 0) { evicted += arc_evict_state(state, spa, ARC_EVICT_ALL, type); if (!retry) break; } return (evicted); } /* * Evict the specified number of bytes from the state specified, * restricting eviction to the spa and type given. This function * prevents us from trying to evict more from a state's list than * is "evictable", and to skip evicting altogether when passed a * negative value for "bytes". In contrast, arc_evict_state() will * evict everything it can, when passed a negative value for "bytes". */ static uint64_t arc_adjust_impl(arc_state_t *state, uint64_t spa, int64_t bytes, arc_buf_contents_t type) { int64_t delta; if (bytes > 0 && refcount_count(&state->arcs_esize[type]) > 0) { delta = MIN(refcount_count(&state->arcs_esize[type]), bytes); return (arc_evict_state(state, spa, delta, type)); } return (0); } /* * Evict metadata buffers from the cache, such that arc_meta_used is * capped by the arc_meta_limit tunable. */ static uint64_t arc_adjust_meta(void) { uint64_t total_evicted = 0; int64_t target; /* * If we're over the meta limit, we want to evict enough * metadata to get back under the meta limit. We don't want to * evict so much that we drop the MRU below arc_p, though. If * we're over the meta limit more than we're over arc_p, we * evict some from the MRU here, and some from the MFU below. */ target = MIN((int64_t)(arc_meta_used - arc_meta_limit), (int64_t)(refcount_count(&arc_anon->arcs_size) + refcount_count(&arc_mru->arcs_size) - arc_p)); total_evicted += arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_METADATA); /* * Similar to the above, we want to evict enough bytes to get us * below the meta limit, but not so much as to drop us below the * space allotted to the MFU (which is defined as arc_c - arc_p). */ target = MIN((int64_t)(arc_meta_used - arc_meta_limit), (int64_t)(refcount_count(&arc_mfu->arcs_size) - (arc_c - arc_p))); total_evicted += arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_METADATA); return (total_evicted); } /* * Return the type of the oldest buffer in the given arc state * * This function will select a random sublist of type ARC_BUFC_DATA and * a random sublist of type ARC_BUFC_METADATA. The tail of each sublist * is compared, and the type which contains the "older" buffer will be * returned. */ static arc_buf_contents_t arc_adjust_type(arc_state_t *state) { - multilist_t *data_ml = &state->arcs_list[ARC_BUFC_DATA]; - multilist_t *meta_ml = &state->arcs_list[ARC_BUFC_METADATA]; + multilist_t *data_ml = state->arcs_list[ARC_BUFC_DATA]; + multilist_t *meta_ml = state->arcs_list[ARC_BUFC_METADATA]; int data_idx = multilist_get_random_index(data_ml); int meta_idx = multilist_get_random_index(meta_ml); multilist_sublist_t *data_mls; multilist_sublist_t *meta_mls; arc_buf_contents_t type; arc_buf_hdr_t *data_hdr; arc_buf_hdr_t *meta_hdr; /* * We keep the sublist lock until we're finished, to prevent * the headers from being destroyed via arc_evict_state(). */ data_mls = multilist_sublist_lock(data_ml, data_idx); meta_mls = multilist_sublist_lock(meta_ml, meta_idx); /* * These two loops are to ensure we skip any markers that * might be at the tail of the lists due to arc_evict_state(). */ for (data_hdr = multilist_sublist_tail(data_mls); data_hdr != NULL; data_hdr = multilist_sublist_prev(data_mls, data_hdr)) { if (data_hdr->b_spa != 0) break; } for (meta_hdr = multilist_sublist_tail(meta_mls); meta_hdr != NULL; meta_hdr = multilist_sublist_prev(meta_mls, meta_hdr)) { if (meta_hdr->b_spa != 0) break; } if (data_hdr == NULL && meta_hdr == NULL) { type = ARC_BUFC_DATA; } else if (data_hdr == NULL) { ASSERT3P(meta_hdr, !=, NULL); type = ARC_BUFC_METADATA; } else if (meta_hdr == NULL) { ASSERT3P(data_hdr, !=, NULL); type = ARC_BUFC_DATA; } else { ASSERT3P(data_hdr, !=, NULL); ASSERT3P(meta_hdr, !=, NULL); /* The headers can't be on the sublist without an L1 header */ ASSERT(HDR_HAS_L1HDR(data_hdr)); ASSERT(HDR_HAS_L1HDR(meta_hdr)); if (data_hdr->b_l1hdr.b_arc_access < meta_hdr->b_l1hdr.b_arc_access) { type = ARC_BUFC_DATA; } else { type = ARC_BUFC_METADATA; } } multilist_sublist_unlock(meta_mls); multilist_sublist_unlock(data_mls); return (type); } /* * Evict buffers from the cache, such that arc_size is capped by arc_c. */ static uint64_t arc_adjust(void) { uint64_t total_evicted = 0; uint64_t bytes; int64_t target; /* * If we're over arc_meta_limit, we want to correct that before * potentially evicting data buffers below. */ total_evicted += arc_adjust_meta(); /* * Adjust MRU size * * If we're over the target cache size, we want to evict enough * from the list to get back to our target size. We don't want * to evict too much from the MRU, such that it drops below * arc_p. So, if we're over our target cache size more than * the MRU is over arc_p, we'll evict enough to get back to * arc_p here, and then evict more from the MFU below. */ target = MIN((int64_t)(arc_size - arc_c), (int64_t)(refcount_count(&arc_anon->arcs_size) + refcount_count(&arc_mru->arcs_size) + arc_meta_used - arc_p)); /* * If we're below arc_meta_min, always prefer to evict data. * Otherwise, try to satisfy the requested number of bytes to * evict from the type which contains older buffers; in an * effort to keep newer buffers in the cache regardless of their * type. If we cannot satisfy the number of bytes from this * type, spill over into the next type. */ if (arc_adjust_type(arc_mru) == ARC_BUFC_METADATA && arc_meta_used > arc_meta_min) { bytes = arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_METADATA); total_evicted += bytes; /* * If we couldn't evict our target number of bytes from * metadata, we try to get the rest from data. */ target -= bytes; total_evicted += arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_DATA); } else { bytes = arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_DATA); total_evicted += bytes; /* * If we couldn't evict our target number of bytes from * data, we try to get the rest from metadata. */ target -= bytes; total_evicted += arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_METADATA); } /* * Adjust MFU size * * Now that we've tried to evict enough from the MRU to get its * size back to arc_p, if we're still above the target cache * size, we evict the rest from the MFU. */ target = arc_size - arc_c; if (arc_adjust_type(arc_mfu) == ARC_BUFC_METADATA && arc_meta_used > arc_meta_min) { bytes = arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_METADATA); total_evicted += bytes; /* * If we couldn't evict our target number of bytes from * metadata, we try to get the rest from data. */ target -= bytes; total_evicted += arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_DATA); } else { bytes = arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_DATA); total_evicted += bytes; /* * If we couldn't evict our target number of bytes from * data, we try to get the rest from data. */ target -= bytes; total_evicted += arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_METADATA); } /* * Adjust ghost lists * * In addition to the above, the ARC also defines target values * for the ghost lists. The sum of the mru list and mru ghost * list should never exceed the target size of the cache, and * the sum of the mru list, mfu list, mru ghost list, and mfu * ghost list should never exceed twice the target size of the * cache. The following logic enforces these limits on the ghost * caches, and evicts from them as needed. */ target = refcount_count(&arc_mru->arcs_size) + refcount_count(&arc_mru_ghost->arcs_size) - arc_c; bytes = arc_adjust_impl(arc_mru_ghost, 0, target, ARC_BUFC_DATA); total_evicted += bytes; target -= bytes; total_evicted += arc_adjust_impl(arc_mru_ghost, 0, target, ARC_BUFC_METADATA); /* * We assume the sum of the mru list and mfu list is less than * or equal to arc_c (we enforced this above), which means we * can use the simpler of the two equations below: * * mru + mfu + mru ghost + mfu ghost <= 2 * arc_c * mru ghost + mfu ghost <= arc_c */ target = refcount_count(&arc_mru_ghost->arcs_size) + refcount_count(&arc_mfu_ghost->arcs_size) - arc_c; bytes = arc_adjust_impl(arc_mfu_ghost, 0, target, ARC_BUFC_DATA); total_evicted += bytes; target -= bytes; total_evicted += arc_adjust_impl(arc_mfu_ghost, 0, target, ARC_BUFC_METADATA); return (total_evicted); } void arc_flush(spa_t *spa, boolean_t retry) { uint64_t guid = 0; /* * If retry is B_TRUE, a spa must not be specified since we have * no good way to determine if all of a spa's buffers have been * evicted from an arc state. */ ASSERT(!retry || spa == 0); if (spa != NULL) guid = spa_load_guid(spa); (void) arc_flush_state(arc_mru, guid, ARC_BUFC_DATA, retry); (void) arc_flush_state(arc_mru, guid, ARC_BUFC_METADATA, retry); (void) arc_flush_state(arc_mfu, guid, ARC_BUFC_DATA, retry); (void) arc_flush_state(arc_mfu, guid, ARC_BUFC_METADATA, retry); (void) arc_flush_state(arc_mru_ghost, guid, ARC_BUFC_DATA, retry); (void) arc_flush_state(arc_mru_ghost, guid, ARC_BUFC_METADATA, retry); (void) arc_flush_state(arc_mfu_ghost, guid, ARC_BUFC_DATA, retry); (void) arc_flush_state(arc_mfu_ghost, guid, ARC_BUFC_METADATA, retry); } void arc_shrink(int64_t to_free) { if (arc_c > arc_c_min) { if (arc_c > arc_c_min + to_free) atomic_add_64(&arc_c, -to_free); else arc_c = arc_c_min; atomic_add_64(&arc_p, -(arc_p >> arc_shrink_shift)); if (arc_c > arc_size) arc_c = MAX(arc_size, arc_c_min); if (arc_p > arc_c) arc_p = (arc_c >> 1); ASSERT(arc_c >= arc_c_min); ASSERT((int64_t)arc_p >= 0); } if (arc_size > arc_c) (void) arc_adjust(); } typedef enum free_memory_reason_t { FMR_UNKNOWN, FMR_NEEDFREE, FMR_LOTSFREE, FMR_SWAPFS_MINFREE, FMR_PAGES_PP_MAXIMUM, FMR_HEAP_ARENA, FMR_ZIO_ARENA, } free_memory_reason_t; int64_t last_free_memory; free_memory_reason_t last_free_reason; /* * Additional reserve of pages for pp_reserve. */ int64_t arc_pages_pp_reserve = 64; /* * Additional reserve of pages for swapfs. */ int64_t arc_swapfs_reserve = 64; /* * Return the amount of memory that can be consumed before reclaim will be * needed. Positive if there is sufficient free memory, negative indicates * the amount of memory that needs to be freed up. */ static int64_t arc_available_memory(void) { int64_t lowest = INT64_MAX; int64_t n; free_memory_reason_t r = FMR_UNKNOWN; #ifdef _KERNEL if (needfree > 0) { n = PAGESIZE * (-needfree); if (n < lowest) { lowest = n; r = FMR_NEEDFREE; } } /* * check that we're out of range of the pageout scanner. It starts to * schedule paging if freemem is less than lotsfree and needfree. * lotsfree is the high-water mark for pageout, and needfree is the * number of needed free pages. We add extra pages here to make sure * the scanner doesn't start up while we're freeing memory. */ n = PAGESIZE * (freemem - lotsfree - needfree - desfree); if (n < lowest) { lowest = n; r = FMR_LOTSFREE; } /* * check to make sure that swapfs has enough space so that anon * reservations can still succeed. anon_resvmem() checks that the * availrmem is greater than swapfs_minfree, and the number of reserved * swap pages. We also add a bit of extra here just to prevent * circumstances from getting really dire. */ n = PAGESIZE * (availrmem - swapfs_minfree - swapfs_reserve - desfree - arc_swapfs_reserve); if (n < lowest) { lowest = n; r = FMR_SWAPFS_MINFREE; } /* * Check that we have enough availrmem that memory locking (e.g., via * mlock(3C) or memcntl(2)) can still succeed. (pages_pp_maximum * stores the number of pages that cannot be locked; when availrmem * drops below pages_pp_maximum, page locking mechanisms such as * page_pp_lock() will fail.) */ n = PAGESIZE * (availrmem - pages_pp_maximum - arc_pages_pp_reserve); if (n < lowest) { lowest = n; r = FMR_PAGES_PP_MAXIMUM; } #if defined(__i386) /* * If we're on an i386 platform, it's possible that we'll exhaust the * kernel heap space before we ever run out of available physical * memory. Most checks of the size of the heap_area compare against * tune.t_minarmem, which is the minimum available real memory that we * can have in the system. However, this is generally fixed at 25 pages * which is so low that it's useless. In this comparison, we seek to * calculate the total heap-size, and reclaim if more than 3/4ths of the * heap is allocated. (Or, in the calculation, if less than 1/4th is * free) */ n = (int64_t)vmem_size(heap_arena, VMEM_FREE) - (vmem_size(heap_arena, VMEM_FREE | VMEM_ALLOC) >> 2); if (n < lowest) { lowest = n; r = FMR_HEAP_ARENA; } #endif /* * If zio data pages are being allocated out of a separate heap segment, * then enforce that the size of available vmem for this arena remains * above about 1/4th (1/(2^arc_zio_arena_free_shift)) free. * * Note that reducing the arc_zio_arena_free_shift keeps more virtual * memory (in the zio_arena) free, which can avoid memory * fragmentation issues. */ if (zio_arena != NULL) { n = (int64_t)vmem_size(zio_arena, VMEM_FREE) - (vmem_size(zio_arena, VMEM_ALLOC) >> arc_zio_arena_free_shift); if (n < lowest) { lowest = n; r = FMR_ZIO_ARENA; } } #else /* Every 100 calls, free a small amount */ if (spa_get_random(100) == 0) lowest = -1024; #endif last_free_memory = lowest; last_free_reason = r; return (lowest); } /* * Determine if the system is under memory pressure and is asking * to reclaim memory. A return value of B_TRUE indicates that the system * is under memory pressure and that the arc should adjust accordingly. */ static boolean_t arc_reclaim_needed(void) { return (arc_available_memory() < 0); } static void arc_kmem_reap_now(void) { size_t i; kmem_cache_t *prev_cache = NULL; kmem_cache_t *prev_data_cache = NULL; extern kmem_cache_t *zio_buf_cache[]; extern kmem_cache_t *zio_data_buf_cache[]; extern kmem_cache_t *range_seg_cache; #ifdef _KERNEL if (arc_meta_used >= arc_meta_limit) { /* * We are exceeding our meta-data cache limit. * Purge some DNLC entries to release holds on meta-data. */ dnlc_reduce_cache((void *)(uintptr_t)arc_reduce_dnlc_percent); } #if defined(__i386) /* * Reclaim unused memory from all kmem caches. */ kmem_reap(); #endif #endif for (i = 0; i < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; i++) { if (zio_buf_cache[i] != prev_cache) { prev_cache = zio_buf_cache[i]; kmem_cache_reap_now(zio_buf_cache[i]); } if (zio_data_buf_cache[i] != prev_data_cache) { prev_data_cache = zio_data_buf_cache[i]; kmem_cache_reap_now(zio_data_buf_cache[i]); } } kmem_cache_reap_now(buf_cache); kmem_cache_reap_now(hdr_full_cache); kmem_cache_reap_now(hdr_l2only_cache); kmem_cache_reap_now(range_seg_cache); if (zio_arena != NULL) { /* * Ask the vmem arena to reclaim unused memory from its * quantum caches. */ vmem_qcache_reap(zio_arena); } } /* * Threads can block in arc_get_data_buf() waiting for this thread to evict * enough data and signal them to proceed. When this happens, the threads in * arc_get_data_buf() are sleeping while holding the hash lock for their * particular arc header. Thus, we must be careful to never sleep on a * hash lock in this thread. This is to prevent the following deadlock: * * - Thread A sleeps on CV in arc_get_data_buf() holding hash lock "L", * waiting for the reclaim thread to signal it. * * - arc_reclaim_thread() tries to acquire hash lock "L" using mutex_enter, * fails, and goes to sleep forever. * * This possible deadlock is avoided by always acquiring a hash lock * using mutex_tryenter() from arc_reclaim_thread(). */ static void arc_reclaim_thread(void) { hrtime_t growtime = 0; callb_cpr_t cpr; CALLB_CPR_INIT(&cpr, &arc_reclaim_lock, callb_generic_cpr, FTAG); mutex_enter(&arc_reclaim_lock); while (!arc_reclaim_thread_exit) { uint64_t evicted = 0; /* * This is necessary in order for the mdb ::arc dcmd to * show up to date information. Since the ::arc command * does not call the kstat's update function, without * this call, the command may show stale stats for the * anon, mru, mru_ghost, mfu, and mfu_ghost lists. Even * with this change, the data might be up to 1 second * out of date; but that should suffice. The arc_state_t * structures can be queried directly if more accurate * information is needed. */ if (arc_ksp != NULL) arc_ksp->ks_update(arc_ksp, KSTAT_READ); mutex_exit(&arc_reclaim_lock); /* * We call arc_adjust() before (possibly) calling * arc_kmem_reap_now(), so that we can wake up * arc_get_data_buf() sooner. */ evicted = arc_adjust(); int64_t free_memory = arc_available_memory(); if (free_memory < 0) { arc_no_grow = B_TRUE; arc_warm = B_TRUE; /* * Wait at least zfs_grow_retry (default 60) seconds * before considering growing. */ growtime = gethrtime() + SEC2NSEC(arc_grow_retry); arc_kmem_reap_now(); /* * If we are still low on memory, shrink the ARC * so that we have arc_shrink_min free space. */ free_memory = arc_available_memory(); int64_t to_free = (arc_c >> arc_shrink_shift) - free_memory; if (to_free > 0) { #ifdef _KERNEL to_free = MAX(to_free, ptob(needfree)); #endif arc_shrink(to_free); } } else if (free_memory < arc_c >> arc_no_grow_shift) { arc_no_grow = B_TRUE; } else if (gethrtime() >= growtime) { arc_no_grow = B_FALSE; } mutex_enter(&arc_reclaim_lock); /* * If evicted is zero, we couldn't evict anything via * arc_adjust(). This could be due to hash lock * collisions, but more likely due to the majority of * arc buffers being unevictable. Therefore, even if * arc_size is above arc_c, another pass is unlikely to * be helpful and could potentially cause us to enter an * infinite loop. */ if (arc_size <= arc_c || evicted == 0) { /* * We're either no longer overflowing, or we * can't evict anything more, so we should wake * up any threads before we go to sleep. */ cv_broadcast(&arc_reclaim_waiters_cv); /* * Block until signaled, or after one second (we * might need to perform arc_kmem_reap_now() * even if we aren't being signalled) */ CALLB_CPR_SAFE_BEGIN(&cpr); (void) cv_timedwait_hires(&arc_reclaim_thread_cv, &arc_reclaim_lock, SEC2NSEC(1), MSEC2NSEC(1), 0); CALLB_CPR_SAFE_END(&cpr, &arc_reclaim_lock); } } arc_reclaim_thread_exit = B_FALSE; cv_broadcast(&arc_reclaim_thread_cv); CALLB_CPR_EXIT(&cpr); /* drops arc_reclaim_lock */ thread_exit(); } /* * Adapt arc info given the number of bytes we are trying to add and * the state that we are comming from. This function is only called * when we are adding new content to the cache. */ static void arc_adapt(int bytes, arc_state_t *state) { int mult; uint64_t arc_p_min = (arc_c >> arc_p_min_shift); int64_t mrug_size = refcount_count(&arc_mru_ghost->arcs_size); int64_t mfug_size = refcount_count(&arc_mfu_ghost->arcs_size); if (state == arc_l2c_only) return; ASSERT(bytes > 0); /* * Adapt the target size of the MRU list: * - if we just hit in the MRU ghost list, then increase * the target size of the MRU list. * - if we just hit in the MFU ghost list, then increase * the target size of the MFU list by decreasing the * target size of the MRU list. */ if (state == arc_mru_ghost) { mult = (mrug_size >= mfug_size) ? 1 : (mfug_size / mrug_size); mult = MIN(mult, 10); /* avoid wild arc_p adjustment */ arc_p = MIN(arc_c - arc_p_min, arc_p + bytes * mult); } else if (state == arc_mfu_ghost) { uint64_t delta; mult = (mfug_size >= mrug_size) ? 1 : (mrug_size / mfug_size); mult = MIN(mult, 10); delta = MIN(bytes * mult, arc_p); arc_p = MAX(arc_p_min, arc_p - delta); } ASSERT((int64_t)arc_p >= 0); if (arc_reclaim_needed()) { cv_signal(&arc_reclaim_thread_cv); return; } if (arc_no_grow) return; if (arc_c >= arc_c_max) return; /* * If we're within (2 * maxblocksize) bytes of the target * cache size, increment the target cache size */ if (arc_size > arc_c - (2ULL << SPA_MAXBLOCKSHIFT)) { atomic_add_64(&arc_c, (int64_t)bytes); if (arc_c > arc_c_max) arc_c = arc_c_max; else if (state == arc_anon) atomic_add_64(&arc_p, (int64_t)bytes); if (arc_p > arc_c) arc_p = arc_c; } ASSERT((int64_t)arc_p >= 0); } /* * Check if arc_size has grown past our upper threshold, determined by * zfs_arc_overflow_shift. */ static boolean_t arc_is_overflowing(void) { /* Always allow at least one block of overflow */ uint64_t overflow = MAX(SPA_MAXBLOCKSIZE, arc_c >> zfs_arc_overflow_shift); return (arc_size >= arc_c + overflow); } /* * Allocate a block and return it to the caller. If we are hitting the * hard limit for the cache size, we must sleep, waiting for the eviction * thread to catch up. If we're past the target size but below the hard * limit, we'll only signal the reclaim thread and continue on. */ static void * arc_get_data_buf(arc_buf_hdr_t *hdr, uint64_t size, void *tag) { void *datap = NULL; arc_state_t *state = hdr->b_l1hdr.b_state; arc_buf_contents_t type = arc_buf_type(hdr); arc_adapt(size, state); /* * If arc_size is currently overflowing, and has grown past our * upper limit, we must be adding data faster than the evict * thread can evict. Thus, to ensure we don't compound the * problem by adding more data and forcing arc_size to grow even * further past it's target size, we halt and wait for the * eviction thread to catch up. * * It's also possible that the reclaim thread is unable to evict * enough buffers to get arc_size below the overflow limit (e.g. * due to buffers being un-evictable, or hash lock collisions). * In this case, we want to proceed regardless if we're * overflowing; thus we don't use a while loop here. */ if (arc_is_overflowing()) { mutex_enter(&arc_reclaim_lock); /* * Now that we've acquired the lock, we may no longer be * over the overflow limit, lets check. * * We're ignoring the case of spurious wake ups. If that * were to happen, it'd let this thread consume an ARC * buffer before it should have (i.e. before we're under * the overflow limit and were signalled by the reclaim * thread). As long as that is a rare occurrence, it * shouldn't cause any harm. */ if (arc_is_overflowing()) { cv_signal(&arc_reclaim_thread_cv); cv_wait(&arc_reclaim_waiters_cv, &arc_reclaim_lock); } mutex_exit(&arc_reclaim_lock); } VERIFY3U(hdr->b_type, ==, type); if (type == ARC_BUFC_METADATA) { datap = zio_buf_alloc(size); arc_space_consume(size, ARC_SPACE_META); } else { ASSERT(type == ARC_BUFC_DATA); datap = zio_data_buf_alloc(size); arc_space_consume(size, ARC_SPACE_DATA); } /* * Update the state size. Note that ghost states have a * "ghost size" and so don't need to be updated. */ if (!GHOST_STATE(state)) { (void) refcount_add_many(&state->arcs_size, size, tag); /* * If this is reached via arc_read, the link is * protected by the hash lock. If reached via * arc_buf_alloc, the header should not be accessed by * any other thread. And, if reached via arc_read_done, * the hash lock will protect it if it's found in the * hash table; otherwise no other thread should be * trying to [add|remove]_reference it. */ if (multilist_link_active(&hdr->b_l1hdr.b_arc_node)) { ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); (void) refcount_add_many(&state->arcs_esize[type], size, tag); } /* * If we are growing the cache, and we are adding anonymous * data, and we have outgrown arc_p, update arc_p */ if (arc_size < arc_c && hdr->b_l1hdr.b_state == arc_anon && (refcount_count(&arc_anon->arcs_size) + refcount_count(&arc_mru->arcs_size) > arc_p)) arc_p = MIN(arc_c, arc_p + size); } return (datap); } /* * Free the arc data buffer. */ static void arc_free_data_buf(arc_buf_hdr_t *hdr, void *data, uint64_t size, void *tag) { arc_state_t *state = hdr->b_l1hdr.b_state; arc_buf_contents_t type = arc_buf_type(hdr); /* protected by hash lock, if in the hash table */ if (multilist_link_active(&hdr->b_l1hdr.b_arc_node)) { ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); ASSERT(state != arc_anon && state != arc_l2c_only); (void) refcount_remove_many(&state->arcs_esize[type], size, tag); } (void) refcount_remove_many(&state->arcs_size, size, tag); VERIFY3U(hdr->b_type, ==, type); if (type == ARC_BUFC_METADATA) { zio_buf_free(data, size); arc_space_return(size, ARC_SPACE_META); } else { ASSERT(type == ARC_BUFC_DATA); zio_data_buf_free(data, size); arc_space_return(size, ARC_SPACE_DATA); } } /* * This routine is called whenever a buffer is accessed. * NOTE: the hash lock is dropped in this function. */ static void arc_access(arc_buf_hdr_t *hdr, kmutex_t *hash_lock) { clock_t now; ASSERT(MUTEX_HELD(hash_lock)); ASSERT(HDR_HAS_L1HDR(hdr)); if (hdr->b_l1hdr.b_state == arc_anon) { /* * This buffer is not in the cache, and does not * appear in our "ghost" list. Add the new buffer * to the MRU state. */ ASSERT0(hdr->b_l1hdr.b_arc_access); hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); DTRACE_PROBE1(new_state__mru, arc_buf_hdr_t *, hdr); arc_change_state(arc_mru, hdr, hash_lock); } else if (hdr->b_l1hdr.b_state == arc_mru) { now = ddi_get_lbolt(); /* * If this buffer is here because of a prefetch, then either: * - clear the flag if this is a "referencing" read * (any subsequent access will bump this into the MFU state). * or * - move the buffer to the head of the list if this is * another prefetch (to make it less likely to be evicted). */ if (HDR_PREFETCH(hdr)) { if (refcount_count(&hdr->b_l1hdr.b_refcnt) == 0) { /* link protected by hash lock */ ASSERT(multilist_link_active( &hdr->b_l1hdr.b_arc_node)); } else { arc_hdr_clear_flags(hdr, ARC_FLAG_PREFETCH); ARCSTAT_BUMP(arcstat_mru_hits); } hdr->b_l1hdr.b_arc_access = now; return; } /* * This buffer has been "accessed" only once so far, * but it is still in the cache. Move it to the MFU * state. */ if (now > hdr->b_l1hdr.b_arc_access + ARC_MINTIME) { /* * More than 125ms have passed since we * instantiated this buffer. Move it to the * most frequently used state. */ hdr->b_l1hdr.b_arc_access = now; DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); arc_change_state(arc_mfu, hdr, hash_lock); } ARCSTAT_BUMP(arcstat_mru_hits); } else if (hdr->b_l1hdr.b_state == arc_mru_ghost) { arc_state_t *new_state; /* * This buffer has been "accessed" recently, but * was evicted from the cache. Move it to the * MFU state. */ if (HDR_PREFETCH(hdr)) { new_state = arc_mru; if (refcount_count(&hdr->b_l1hdr.b_refcnt) > 0) arc_hdr_clear_flags(hdr, ARC_FLAG_PREFETCH); DTRACE_PROBE1(new_state__mru, arc_buf_hdr_t *, hdr); } else { new_state = arc_mfu; DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); } hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); arc_change_state(new_state, hdr, hash_lock); ARCSTAT_BUMP(arcstat_mru_ghost_hits); } else if (hdr->b_l1hdr.b_state == arc_mfu) { /* * This buffer has been accessed more than once and is * still in the cache. Keep it in the MFU state. * * NOTE: an add_reference() that occurred when we did * the arc_read() will have kicked this off the list. * If it was a prefetch, we will explicitly move it to * the head of the list now. */ if ((HDR_PREFETCH(hdr)) != 0) { ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); /* link protected by hash_lock */ ASSERT(multilist_link_active(&hdr->b_l1hdr.b_arc_node)); } ARCSTAT_BUMP(arcstat_mfu_hits); hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); } else if (hdr->b_l1hdr.b_state == arc_mfu_ghost) { arc_state_t *new_state = arc_mfu; /* * This buffer has been accessed more than once but has * been evicted from the cache. Move it back to the * MFU state. */ if (HDR_PREFETCH(hdr)) { /* * This is a prefetch access... * move this block back to the MRU state. */ ASSERT0(refcount_count(&hdr->b_l1hdr.b_refcnt)); new_state = arc_mru; } hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); arc_change_state(new_state, hdr, hash_lock); ARCSTAT_BUMP(arcstat_mfu_ghost_hits); } else if (hdr->b_l1hdr.b_state == arc_l2c_only) { /* * This buffer is on the 2nd Level ARC. */ hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); arc_change_state(arc_mfu, hdr, hash_lock); } else { ASSERT(!"invalid arc state"); } } /* a generic arc_done_func_t which you can use */ /* ARGSUSED */ void arc_bcopy_func(zio_t *zio, arc_buf_t *buf, void *arg) { if (zio == NULL || zio->io_error == 0) bcopy(buf->b_data, arg, arc_buf_size(buf)); arc_buf_destroy(buf, arg); } /* a generic arc_done_func_t */ void arc_getbuf_func(zio_t *zio, arc_buf_t *buf, void *arg) { arc_buf_t **bufp = arg; if (zio && zio->io_error) { arc_buf_destroy(buf, arg); *bufp = NULL; } else { *bufp = buf; ASSERT(buf->b_data); } } static void arc_hdr_verify(arc_buf_hdr_t *hdr, blkptr_t *bp) { if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) { ASSERT3U(HDR_GET_PSIZE(hdr), ==, 0); ASSERT3U(HDR_GET_COMPRESS(hdr), ==, ZIO_COMPRESS_OFF); } else { if (HDR_COMPRESSION_ENABLED(hdr)) { ASSERT3U(HDR_GET_COMPRESS(hdr), ==, BP_GET_COMPRESS(bp)); } ASSERT3U(HDR_GET_LSIZE(hdr), ==, BP_GET_LSIZE(bp)); ASSERT3U(HDR_GET_PSIZE(hdr), ==, BP_GET_PSIZE(bp)); } } static void arc_read_done(zio_t *zio) { arc_buf_hdr_t *hdr = zio->io_private; kmutex_t *hash_lock = NULL; arc_callback_t *callback_list; arc_callback_t *acb; boolean_t freeable = B_FALSE; boolean_t no_zio_error = (zio->io_error == 0); /* * The hdr was inserted into hash-table and removed from lists * prior to starting I/O. We should find this header, since * it's in the hash table, and it should be legit since it's * not possible to evict it during the I/O. The only possible * reason for it not to be found is if we were freed during the * read. */ if (HDR_IN_HASH_TABLE(hdr)) { ASSERT3U(hdr->b_birth, ==, BP_PHYSICAL_BIRTH(zio->io_bp)); ASSERT3U(hdr->b_dva.dva_word[0], ==, BP_IDENTITY(zio->io_bp)->dva_word[0]); ASSERT3U(hdr->b_dva.dva_word[1], ==, BP_IDENTITY(zio->io_bp)->dva_word[1]); arc_buf_hdr_t *found = buf_hash_find(hdr->b_spa, zio->io_bp, &hash_lock); ASSERT((found == hdr && DVA_EQUAL(&hdr->b_dva, BP_IDENTITY(zio->io_bp))) || (found == hdr && HDR_L2_READING(hdr))); ASSERT3P(hash_lock, !=, NULL); } if (no_zio_error) { /* byteswap if necessary */ if (BP_SHOULD_BYTESWAP(zio->io_bp)) { if (BP_GET_LEVEL(zio->io_bp) > 0) { hdr->b_l1hdr.b_byteswap = DMU_BSWAP_UINT64; } else { hdr->b_l1hdr.b_byteswap = DMU_OT_BYTESWAP(BP_GET_TYPE(zio->io_bp)); } } else { hdr->b_l1hdr.b_byteswap = DMU_BSWAP_NUMFUNCS; } } arc_hdr_clear_flags(hdr, ARC_FLAG_L2_EVICTED); if (l2arc_noprefetch && HDR_PREFETCH(hdr)) arc_hdr_clear_flags(hdr, ARC_FLAG_L2CACHE); callback_list = hdr->b_l1hdr.b_acb; ASSERT3P(callback_list, !=, NULL); if (hash_lock && no_zio_error && hdr->b_l1hdr.b_state == arc_anon) { /* * Only call arc_access on anonymous buffers. This is because * if we've issued an I/O for an evicted buffer, we've already * called arc_access (to prevent any simultaneous readers from * getting confused). */ arc_access(hdr, hash_lock); } /* * If a read request has a callback (i.e. acb_done is not NULL), then we * make a buf containing the data according to the parameters which were * passed in. The implementation of arc_buf_alloc_impl() ensures that we * aren't needlessly decompressing the data multiple times. */ int callback_cnt = 0; for (acb = callback_list; acb != NULL; acb = acb->acb_next) { if (!acb->acb_done) continue; /* This is a demand read since prefetches don't use callbacks */ callback_cnt++; int error = arc_buf_alloc_impl(hdr, acb->acb_private, acb->acb_compressed, no_zio_error, &acb->acb_buf); if (no_zio_error) { zio->io_error = error; } } hdr->b_l1hdr.b_acb = NULL; arc_hdr_clear_flags(hdr, ARC_FLAG_IO_IN_PROGRESS); if (callback_cnt == 0) { ASSERT(HDR_PREFETCH(hdr)); ASSERT0(hdr->b_l1hdr.b_bufcnt); ASSERT3P(hdr->b_l1hdr.b_pdata, !=, NULL); } ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt) || callback_list != NULL); if (no_zio_error) { arc_hdr_verify(hdr, zio->io_bp); } else { arc_hdr_set_flags(hdr, ARC_FLAG_IO_ERROR); if (hdr->b_l1hdr.b_state != arc_anon) arc_change_state(arc_anon, hdr, hash_lock); if (HDR_IN_HASH_TABLE(hdr)) buf_hash_remove(hdr); freeable = refcount_is_zero(&hdr->b_l1hdr.b_refcnt); } /* * Broadcast before we drop the hash_lock to avoid the possibility * that the hdr (and hence the cv) might be freed before we get to * the cv_broadcast(). */ cv_broadcast(&hdr->b_l1hdr.b_cv); if (hash_lock != NULL) { mutex_exit(hash_lock); } else { /* * This block was freed while we waited for the read to * complete. It has been removed from the hash table and * moved to the anonymous state (so that it won't show up * in the cache). */ ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon); freeable = refcount_is_zero(&hdr->b_l1hdr.b_refcnt); } /* execute each callback and free its structure */ while ((acb = callback_list) != NULL) { if (acb->acb_done) acb->acb_done(zio, acb->acb_buf, acb->acb_private); if (acb->acb_zio_dummy != NULL) { acb->acb_zio_dummy->io_error = zio->io_error; zio_nowait(acb->acb_zio_dummy); } callback_list = acb->acb_next; kmem_free(acb, sizeof (arc_callback_t)); } if (freeable) arc_hdr_destroy(hdr); } /* * "Read" the block at the specified DVA (in bp) via the * cache. If the block is found in the cache, invoke the provided * callback immediately and return. Note that the `zio' parameter * in the callback will be NULL in this case, since no IO was * required. If the block is not in the cache pass the read request * on to the spa with a substitute callback function, so that the * requested block will be added to the cache. * * If a read request arrives for a block that has a read in-progress, * either wait for the in-progress read to complete (and return the * results); or, if this is a read with a "done" func, add a record * to the read to invoke the "done" func when the read completes, * and return; or just return. * * arc_read_done() will invoke all the requested "done" functions * for readers of this block. */ int arc_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, arc_done_func_t *done, void *private, zio_priority_t priority, int zio_flags, arc_flags_t *arc_flags, const zbookmark_phys_t *zb) { arc_buf_hdr_t *hdr = NULL; kmutex_t *hash_lock = NULL; zio_t *rzio; uint64_t guid = spa_load_guid(spa); boolean_t compressed_read = (zio_flags & ZIO_FLAG_RAW) != 0; ASSERT(!BP_IS_EMBEDDED(bp) || BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA); top: if (!BP_IS_EMBEDDED(bp)) { /* * Embedded BP's have no DVA and require no I/O to "read". * Create an anonymous arc buf to back it. */ hdr = buf_hash_find(guid, bp, &hash_lock); } if (hdr != NULL && HDR_HAS_L1HDR(hdr) && hdr->b_l1hdr.b_pdata != NULL) { arc_buf_t *buf = NULL; *arc_flags |= ARC_FLAG_CACHED; if (HDR_IO_IN_PROGRESS(hdr)) { if ((hdr->b_flags & ARC_FLAG_PRIO_ASYNC_READ) && priority == ZIO_PRIORITY_SYNC_READ) { /* * This sync read must wait for an * in-progress async read (e.g. a predictive * prefetch). Async reads are queued * separately at the vdev_queue layer, so * this is a form of priority inversion. * Ideally, we would "inherit" the demand * i/o's priority by moving the i/o from * the async queue to the synchronous queue, * but there is currently no mechanism to do * so. Track this so that we can evaluate * the magnitude of this potential performance * problem. * * Note that if the prefetch i/o is already * active (has been issued to the device), * the prefetch improved performance, because * we issued it sooner than we would have * without the prefetch. */ DTRACE_PROBE1(arc__sync__wait__for__async, arc_buf_hdr_t *, hdr); ARCSTAT_BUMP(arcstat_sync_wait_for_async); } if (hdr->b_flags & ARC_FLAG_PREDICTIVE_PREFETCH) { arc_hdr_clear_flags(hdr, ARC_FLAG_PREDICTIVE_PREFETCH); } if (*arc_flags & ARC_FLAG_WAIT) { cv_wait(&hdr->b_l1hdr.b_cv, hash_lock); mutex_exit(hash_lock); goto top; } ASSERT(*arc_flags & ARC_FLAG_NOWAIT); if (done) { arc_callback_t *acb = NULL; acb = kmem_zalloc(sizeof (arc_callback_t), KM_SLEEP); acb->acb_done = done; acb->acb_private = private; acb->acb_compressed = compressed_read; if (pio != NULL) acb->acb_zio_dummy = zio_null(pio, spa, NULL, NULL, NULL, zio_flags); ASSERT3P(acb->acb_done, !=, NULL); acb->acb_next = hdr->b_l1hdr.b_acb; hdr->b_l1hdr.b_acb = acb; mutex_exit(hash_lock); return (0); } mutex_exit(hash_lock); return (0); } ASSERT(hdr->b_l1hdr.b_state == arc_mru || hdr->b_l1hdr.b_state == arc_mfu); if (done) { if (hdr->b_flags & ARC_FLAG_PREDICTIVE_PREFETCH) { /* * This is a demand read which does not have to * wait for i/o because we did a predictive * prefetch i/o for it, which has completed. */ DTRACE_PROBE1( arc__demand__hit__predictive__prefetch, arc_buf_hdr_t *, hdr); ARCSTAT_BUMP( arcstat_demand_hit_predictive_prefetch); arc_hdr_clear_flags(hdr, ARC_FLAG_PREDICTIVE_PREFETCH); } ASSERT(!BP_IS_EMBEDDED(bp) || !BP_IS_HOLE(bp)); /* Get a buf with the desired data in it. */ VERIFY0(arc_buf_alloc_impl(hdr, private, compressed_read, B_TRUE, &buf)); } else if (*arc_flags & ARC_FLAG_PREFETCH && refcount_count(&hdr->b_l1hdr.b_refcnt) == 0) { arc_hdr_set_flags(hdr, ARC_FLAG_PREFETCH); } DTRACE_PROBE1(arc__hit, arc_buf_hdr_t *, hdr); arc_access(hdr, hash_lock); if (*arc_flags & ARC_FLAG_L2CACHE) arc_hdr_set_flags(hdr, ARC_FLAG_L2CACHE); mutex_exit(hash_lock); ARCSTAT_BUMP(arcstat_hits); ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr), demand, prefetch, !HDR_ISTYPE_METADATA(hdr), data, metadata, hits); if (done) done(NULL, buf, private); } else { uint64_t lsize = BP_GET_LSIZE(bp); uint64_t psize = BP_GET_PSIZE(bp); arc_callback_t *acb; vdev_t *vd = NULL; uint64_t addr = 0; boolean_t devw = B_FALSE; uint64_t size; if (hdr == NULL) { /* this block is not in the cache */ arc_buf_hdr_t *exists = NULL; arc_buf_contents_t type = BP_GET_BUFC_TYPE(bp); hdr = arc_hdr_alloc(spa_load_guid(spa), psize, lsize, BP_GET_COMPRESS(bp), type); if (!BP_IS_EMBEDDED(bp)) { hdr->b_dva = *BP_IDENTITY(bp); hdr->b_birth = BP_PHYSICAL_BIRTH(bp); exists = buf_hash_insert(hdr, &hash_lock); } if (exists != NULL) { /* somebody beat us to the hash insert */ mutex_exit(hash_lock); buf_discard_identity(hdr); arc_hdr_destroy(hdr); goto top; /* restart the IO request */ } } else { /* * This block is in the ghost cache. If it was L2-only * (and thus didn't have an L1 hdr), we realloc the * header to add an L1 hdr. */ if (!HDR_HAS_L1HDR(hdr)) { hdr = arc_hdr_realloc(hdr, hdr_l2only_cache, hdr_full_cache); } ASSERT3P(hdr->b_l1hdr.b_pdata, ==, NULL); ASSERT(GHOST_STATE(hdr->b_l1hdr.b_state)); ASSERT(!HDR_IO_IN_PROGRESS(hdr)); ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL); /* * This is a delicate dance that we play here. * This hdr is in the ghost list so we access it * to move it out of the ghost list before we * initiate the read. If it's a prefetch then * it won't have a callback so we'll remove the * reference that arc_buf_alloc_impl() created. We * do this after we've called arc_access() to * avoid hitting an assert in remove_reference(). */ arc_access(hdr, hash_lock); arc_hdr_alloc_pdata(hdr); } ASSERT3P(hdr->b_l1hdr.b_pdata, !=, NULL); size = arc_hdr_size(hdr); /* * If compression is enabled on the hdr, then will do * RAW I/O and will store the compressed data in the hdr's * data block. Otherwise, the hdr's data block will contain * the uncompressed data. */ if (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF) { zio_flags |= ZIO_FLAG_RAW; } if (*arc_flags & ARC_FLAG_PREFETCH) arc_hdr_set_flags(hdr, ARC_FLAG_PREFETCH); if (*arc_flags & ARC_FLAG_L2CACHE) arc_hdr_set_flags(hdr, ARC_FLAG_L2CACHE); if (BP_GET_LEVEL(bp) > 0) arc_hdr_set_flags(hdr, ARC_FLAG_INDIRECT); if (*arc_flags & ARC_FLAG_PREDICTIVE_PREFETCH) arc_hdr_set_flags(hdr, ARC_FLAG_PREDICTIVE_PREFETCH); ASSERT(!GHOST_STATE(hdr->b_l1hdr.b_state)); acb = kmem_zalloc(sizeof (arc_callback_t), KM_SLEEP); acb->acb_done = done; acb->acb_private = private; acb->acb_compressed = compressed_read; ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL); hdr->b_l1hdr.b_acb = acb; arc_hdr_set_flags(hdr, ARC_FLAG_IO_IN_PROGRESS); if (HDR_HAS_L2HDR(hdr) && (vd = hdr->b_l2hdr.b_dev->l2ad_vdev) != NULL) { devw = hdr->b_l2hdr.b_dev->l2ad_writing; addr = hdr->b_l2hdr.b_daddr; /* * Lock out device removal. */ if (vdev_is_dead(vd) || !spa_config_tryenter(spa, SCL_L2ARC, vd, RW_READER)) vd = NULL; } if (priority == ZIO_PRIORITY_ASYNC_READ) arc_hdr_set_flags(hdr, ARC_FLAG_PRIO_ASYNC_READ); else arc_hdr_clear_flags(hdr, ARC_FLAG_PRIO_ASYNC_READ); if (hash_lock != NULL) mutex_exit(hash_lock); /* * At this point, we have a level 1 cache miss. Try again in * L2ARC if possible. */ ASSERT3U(HDR_GET_LSIZE(hdr), ==, lsize); DTRACE_PROBE4(arc__miss, arc_buf_hdr_t *, hdr, blkptr_t *, bp, uint64_t, lsize, zbookmark_phys_t *, zb); ARCSTAT_BUMP(arcstat_misses); ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr), demand, prefetch, !HDR_ISTYPE_METADATA(hdr), data, metadata, misses); if (vd != NULL && l2arc_ndev != 0 && !(l2arc_norw && devw)) { /* * Read from the L2ARC if the following are true: * 1. The L2ARC vdev was previously cached. * 2. This buffer still has L2ARC metadata. * 3. This buffer isn't currently writing to the L2ARC. * 4. The L2ARC entry wasn't evicted, which may * also have invalidated the vdev. * 5. This isn't prefetch and l2arc_noprefetch is set. */ if (HDR_HAS_L2HDR(hdr) && !HDR_L2_WRITING(hdr) && !HDR_L2_EVICTED(hdr) && !(l2arc_noprefetch && HDR_PREFETCH(hdr))) { l2arc_read_callback_t *cb; DTRACE_PROBE1(l2arc__hit, arc_buf_hdr_t *, hdr); ARCSTAT_BUMP(arcstat_l2_hits); cb = kmem_zalloc(sizeof (l2arc_read_callback_t), KM_SLEEP); cb->l2rcb_hdr = hdr; cb->l2rcb_bp = *bp; cb->l2rcb_zb = *zb; cb->l2rcb_flags = zio_flags; ASSERT(addr >= VDEV_LABEL_START_SIZE && addr + lsize < vd->vdev_psize - VDEV_LABEL_END_SIZE); /* * l2arc read. The SCL_L2ARC lock will be * released by l2arc_read_done(). * Issue a null zio if the underlying buffer * was squashed to zero size by compression. */ ASSERT3U(HDR_GET_COMPRESS(hdr), !=, ZIO_COMPRESS_EMPTY); rzio = zio_read_phys(pio, vd, addr, size, hdr->b_l1hdr.b_pdata, ZIO_CHECKSUM_OFF, l2arc_read_done, cb, priority, zio_flags | ZIO_FLAG_DONT_CACHE | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY, B_FALSE); DTRACE_PROBE2(l2arc__read, vdev_t *, vd, zio_t *, rzio); ARCSTAT_INCR(arcstat_l2_read_bytes, size); if (*arc_flags & ARC_FLAG_NOWAIT) { zio_nowait(rzio); return (0); } ASSERT(*arc_flags & ARC_FLAG_WAIT); if (zio_wait(rzio) == 0) return (0); /* l2arc read error; goto zio_read() */ } else { DTRACE_PROBE1(l2arc__miss, arc_buf_hdr_t *, hdr); ARCSTAT_BUMP(arcstat_l2_misses); if (HDR_L2_WRITING(hdr)) ARCSTAT_BUMP(arcstat_l2_rw_clash); spa_config_exit(spa, SCL_L2ARC, vd); } } else { if (vd != NULL) spa_config_exit(spa, SCL_L2ARC, vd); if (l2arc_ndev != 0) { DTRACE_PROBE1(l2arc__miss, arc_buf_hdr_t *, hdr); ARCSTAT_BUMP(arcstat_l2_misses); } } rzio = zio_read(pio, spa, bp, hdr->b_l1hdr.b_pdata, size, arc_read_done, hdr, priority, zio_flags, zb); if (*arc_flags & ARC_FLAG_WAIT) return (zio_wait(rzio)); ASSERT(*arc_flags & ARC_FLAG_NOWAIT); zio_nowait(rzio); } return (0); } /* * Notify the arc that a block was freed, and thus will never be used again. */ void arc_freed(spa_t *spa, const blkptr_t *bp) { arc_buf_hdr_t *hdr; kmutex_t *hash_lock; uint64_t guid = spa_load_guid(spa); ASSERT(!BP_IS_EMBEDDED(bp)); hdr = buf_hash_find(guid, bp, &hash_lock); if (hdr == NULL) return; /* * We might be trying to free a block that is still doing I/O * (i.e. prefetch) or has a reference (i.e. a dedup-ed, * dmu_sync-ed block). If this block is being prefetched, then it * would still have the ARC_FLAG_IO_IN_PROGRESS flag set on the hdr * until the I/O completes. A block may also have a reference if it is * part of a dedup-ed, dmu_synced write. The dmu_sync() function would * have written the new block to its final resting place on disk but * without the dedup flag set. This would have left the hdr in the MRU * state and discoverable. When the txg finally syncs it detects that * the block was overridden in open context and issues an override I/O. * Since this is a dedup block, the override I/O will determine if the * block is already in the DDT. If so, then it will replace the io_bp * with the bp from the DDT and allow the I/O to finish. When the I/O * reaches the done callback, dbuf_write_override_done, it will * check to see if the io_bp and io_bp_override are identical. * If they are not, then it indicates that the bp was replaced with * the bp in the DDT and the override bp is freed. This allows * us to arrive here with a reference on a block that is being * freed. So if we have an I/O in progress, or a reference to * this hdr, then we don't destroy the hdr. */ if (!HDR_HAS_L1HDR(hdr) || (!HDR_IO_IN_PROGRESS(hdr) && refcount_is_zero(&hdr->b_l1hdr.b_refcnt))) { arc_change_state(arc_anon, hdr, hash_lock); arc_hdr_destroy(hdr); mutex_exit(hash_lock); } else { mutex_exit(hash_lock); } } /* * Release this buffer from the cache, making it an anonymous buffer. This * must be done after a read and prior to modifying the buffer contents. * If the buffer has more than one reference, we must make * a new hdr for the buffer. */ void arc_release(arc_buf_t *buf, void *tag) { arc_buf_hdr_t *hdr = buf->b_hdr; /* * It would be nice to assert that if it's DMU metadata (level > * 0 || it's the dnode file), then it must be syncing context. * But we don't know that information at this level. */ mutex_enter(&buf->b_evict_lock); ASSERT(HDR_HAS_L1HDR(hdr)); /* * We don't grab the hash lock prior to this check, because if * the buffer's header is in the arc_anon state, it won't be * linked into the hash table. */ if (hdr->b_l1hdr.b_state == arc_anon) { mutex_exit(&buf->b_evict_lock); ASSERT(!HDR_IO_IN_PROGRESS(hdr)); ASSERT(!HDR_IN_HASH_TABLE(hdr)); ASSERT(!HDR_HAS_L2HDR(hdr)); ASSERT(HDR_EMPTY(hdr)); ASSERT3U(hdr->b_l1hdr.b_bufcnt, ==, 1); ASSERT3S(refcount_count(&hdr->b_l1hdr.b_refcnt), ==, 1); ASSERT(!list_link_active(&hdr->b_l1hdr.b_arc_node)); hdr->b_l1hdr.b_arc_access = 0; /* * If the buf is being overridden then it may already * have a hdr that is not empty. */ buf_discard_identity(hdr); arc_buf_thaw(buf); return; } kmutex_t *hash_lock = HDR_LOCK(hdr); mutex_enter(hash_lock); /* * This assignment is only valid as long as the hash_lock is * held, we must be careful not to reference state or the * b_state field after dropping the lock. */ arc_state_t *state = hdr->b_l1hdr.b_state; ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); ASSERT3P(state, !=, arc_anon); /* this buffer is not on any list */ ASSERT3S(refcount_count(&hdr->b_l1hdr.b_refcnt), >, 0); if (HDR_HAS_L2HDR(hdr)) { mutex_enter(&hdr->b_l2hdr.b_dev->l2ad_mtx); /* * We have to recheck this conditional again now that * we're holding the l2ad_mtx to prevent a race with * another thread which might be concurrently calling * l2arc_evict(). In that case, l2arc_evict() might have * destroyed the header's L2 portion as we were waiting * to acquire the l2ad_mtx. */ if (HDR_HAS_L2HDR(hdr)) arc_hdr_l2hdr_destroy(hdr); mutex_exit(&hdr->b_l2hdr.b_dev->l2ad_mtx); } /* * Do we have more than one buf? */ if (hdr->b_l1hdr.b_bufcnt > 1) { arc_buf_hdr_t *nhdr; uint64_t spa = hdr->b_spa; uint64_t psize = HDR_GET_PSIZE(hdr); uint64_t lsize = HDR_GET_LSIZE(hdr); enum zio_compress compress = HDR_GET_COMPRESS(hdr); arc_buf_contents_t type = arc_buf_type(hdr); VERIFY3U(hdr->b_type, ==, type); ASSERT(hdr->b_l1hdr.b_buf != buf || buf->b_next != NULL); (void) remove_reference(hdr, hash_lock, tag); if (arc_buf_is_shared(buf) && !ARC_BUF_COMPRESSED(buf)) { ASSERT3P(hdr->b_l1hdr.b_buf, !=, buf); ASSERT(ARC_BUF_LAST(buf)); } /* * Pull the data off of this hdr and attach it to * a new anonymous hdr. Also find the last buffer * in the hdr's buffer list. */ arc_buf_t *lastbuf = arc_buf_remove(hdr, buf); ASSERT3P(lastbuf, !=, NULL); /* * If the current arc_buf_t and the hdr are sharing their data * buffer, then we must stop sharing that block. */ if (arc_buf_is_shared(buf)) { VERIFY(!arc_buf_is_shared(lastbuf)); /* * First, sever the block sharing relationship between * buf and the arc_buf_hdr_t. */ arc_unshare_buf(hdr, buf); /* * Now we need to recreate the hdr's b_pdata. Since we * have lastbuf handy, we try to share with it, but if * we can't then we allocate a new b_pdata and copy the * data from buf into it. */ if (arc_can_share(hdr, lastbuf)) { arc_share_buf(hdr, lastbuf); } else { arc_hdr_alloc_pdata(hdr); bcopy(buf->b_data, hdr->b_l1hdr.b_pdata, psize); } VERIFY3P(lastbuf->b_data, !=, NULL); } else if (HDR_SHARED_DATA(hdr)) { /* * Uncompressed shared buffers are always at the end * of the list. Compressed buffers don't have the * same requirements. This makes it hard to * simply assert that the lastbuf is shared so * we rely on the hdr's compression flags to determine * if we have a compressed, shared buffer. */ ASSERT(arc_buf_is_shared(lastbuf) || HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF); ASSERT(!ARC_BUF_SHARED(buf)); } ASSERT3P(hdr->b_l1hdr.b_pdata, !=, NULL); ASSERT3P(state, !=, arc_l2c_only); (void) refcount_remove_many(&state->arcs_size, arc_buf_size(buf), buf); if (refcount_is_zero(&hdr->b_l1hdr.b_refcnt)) { ASSERT3P(state, !=, arc_l2c_only); (void) refcount_remove_many(&state->arcs_esize[type], arc_buf_size(buf), buf); } hdr->b_l1hdr.b_bufcnt -= 1; arc_cksum_verify(buf); arc_buf_unwatch(buf); mutex_exit(hash_lock); /* * Allocate a new hdr. The new hdr will contain a b_pdata * buffer which will be freed in arc_write(). */ nhdr = arc_hdr_alloc(spa, psize, lsize, compress, type); ASSERT3P(nhdr->b_l1hdr.b_buf, ==, NULL); ASSERT0(nhdr->b_l1hdr.b_bufcnt); ASSERT0(refcount_count(&nhdr->b_l1hdr.b_refcnt)); VERIFY3U(nhdr->b_type, ==, type); ASSERT(!HDR_SHARED_DATA(nhdr)); nhdr->b_l1hdr.b_buf = buf; nhdr->b_l1hdr.b_bufcnt = 1; (void) refcount_add(&nhdr->b_l1hdr.b_refcnt, tag); buf->b_hdr = nhdr; mutex_exit(&buf->b_evict_lock); (void) refcount_add_many(&arc_anon->arcs_size, arc_buf_size(buf), buf); } else { mutex_exit(&buf->b_evict_lock); ASSERT(refcount_count(&hdr->b_l1hdr.b_refcnt) == 1); /* protected by hash lock, or hdr is on arc_anon */ ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); ASSERT(!HDR_IO_IN_PROGRESS(hdr)); arc_change_state(arc_anon, hdr, hash_lock); hdr->b_l1hdr.b_arc_access = 0; mutex_exit(hash_lock); buf_discard_identity(hdr); arc_buf_thaw(buf); } } int arc_released(arc_buf_t *buf) { int released; mutex_enter(&buf->b_evict_lock); released = (buf->b_data != NULL && buf->b_hdr->b_l1hdr.b_state == arc_anon); mutex_exit(&buf->b_evict_lock); return (released); } #ifdef ZFS_DEBUG int arc_referenced(arc_buf_t *buf) { int referenced; mutex_enter(&buf->b_evict_lock); referenced = (refcount_count(&buf->b_hdr->b_l1hdr.b_refcnt)); mutex_exit(&buf->b_evict_lock); return (referenced); } #endif static void arc_write_ready(zio_t *zio) { arc_write_callback_t *callback = zio->io_private; arc_buf_t *buf = callback->awcb_buf; arc_buf_hdr_t *hdr = buf->b_hdr; uint64_t psize = BP_IS_HOLE(zio->io_bp) ? 0 : BP_GET_PSIZE(zio->io_bp); ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT(!refcount_is_zero(&buf->b_hdr->b_l1hdr.b_refcnt)); ASSERT(hdr->b_l1hdr.b_bufcnt > 0); /* * If we're reexecuting this zio because the pool suspended, then * cleanup any state that was previously set the first time the * callback was invoked. */ if (zio->io_flags & ZIO_FLAG_REEXECUTED) { arc_cksum_free(hdr); arc_buf_unwatch(buf); if (hdr->b_l1hdr.b_pdata != NULL) { if (arc_buf_is_shared(buf)) { arc_unshare_buf(hdr, buf); } else { arc_hdr_free_pdata(hdr); } } } ASSERT3P(hdr->b_l1hdr.b_pdata, ==, NULL); ASSERT(!HDR_SHARED_DATA(hdr)); ASSERT(!arc_buf_is_shared(buf)); callback->awcb_ready(zio, buf, callback->awcb_private); if (HDR_IO_IN_PROGRESS(hdr)) ASSERT(zio->io_flags & ZIO_FLAG_REEXECUTED); arc_cksum_compute(buf); arc_hdr_set_flags(hdr, ARC_FLAG_IO_IN_PROGRESS); enum zio_compress compress; if (BP_IS_HOLE(zio->io_bp) || BP_IS_EMBEDDED(zio->io_bp)) { compress = ZIO_COMPRESS_OFF; } else { ASSERT3U(HDR_GET_LSIZE(hdr), ==, BP_GET_LSIZE(zio->io_bp)); compress = BP_GET_COMPRESS(zio->io_bp); } HDR_SET_PSIZE(hdr, psize); arc_hdr_set_compress(hdr, compress); /* * If the hdr is compressed, then copy the compressed * zio contents into arc_buf_hdr_t. Otherwise, copy the original * data buf into the hdr. Ideally, we would like to always copy the * io_data into b_pdata but the user may have disabled compressed * arc thus the on-disk block may or may not match what we maintain * in the hdr's b_pdata field. */ if (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF && !ARC_BUF_COMPRESSED(buf)) { ASSERT3U(BP_GET_COMPRESS(zio->io_bp), !=, ZIO_COMPRESS_OFF); ASSERT3U(psize, >, 0); arc_hdr_alloc_pdata(hdr); bcopy(zio->io_data, hdr->b_l1hdr.b_pdata, psize); } else { ASSERT3P(buf->b_data, ==, zio->io_orig_data); ASSERT3U(zio->io_orig_size, ==, arc_buf_size(buf)); ASSERT3U(hdr->b_l1hdr.b_bufcnt, ==, 1); /* * This hdr is not compressed so we're able to share * the arc_buf_t data buffer with the hdr. */ arc_share_buf(hdr, buf); ASSERT0(bcmp(zio->io_orig_data, hdr->b_l1hdr.b_pdata, HDR_GET_LSIZE(hdr))); } arc_hdr_verify(hdr, zio->io_bp); } static void arc_write_children_ready(zio_t *zio) { arc_write_callback_t *callback = zio->io_private; arc_buf_t *buf = callback->awcb_buf; callback->awcb_children_ready(zio, buf, callback->awcb_private); } /* * The SPA calls this callback for each physical write that happens on behalf * of a logical write. See the comment in dbuf_write_physdone() for details. */ static void arc_write_physdone(zio_t *zio) { arc_write_callback_t *cb = zio->io_private; if (cb->awcb_physdone != NULL) cb->awcb_physdone(zio, cb->awcb_buf, cb->awcb_private); } static void arc_write_done(zio_t *zio) { arc_write_callback_t *callback = zio->io_private; arc_buf_t *buf = callback->awcb_buf; arc_buf_hdr_t *hdr = buf->b_hdr; ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL); if (zio->io_error == 0) { arc_hdr_verify(hdr, zio->io_bp); if (BP_IS_HOLE(zio->io_bp) || BP_IS_EMBEDDED(zio->io_bp)) { buf_discard_identity(hdr); } else { hdr->b_dva = *BP_IDENTITY(zio->io_bp); hdr->b_birth = BP_PHYSICAL_BIRTH(zio->io_bp); } } else { ASSERT(HDR_EMPTY(hdr)); } /* * If the block to be written was all-zero or compressed enough to be * embedded in the BP, no write was performed so there will be no * dva/birth/checksum. The buffer must therefore remain anonymous * (and uncached). */ if (!HDR_EMPTY(hdr)) { arc_buf_hdr_t *exists; kmutex_t *hash_lock; ASSERT3U(zio->io_error, ==, 0); arc_cksum_verify(buf); exists = buf_hash_insert(hdr, &hash_lock); if (exists != NULL) { /* * This can only happen if we overwrite for * sync-to-convergence, because we remove * buffers from the hash table when we arc_free(). */ if (zio->io_flags & ZIO_FLAG_IO_REWRITE) { if (!BP_EQUAL(&zio->io_bp_orig, zio->io_bp)) panic("bad overwrite, hdr=%p exists=%p", (void *)hdr, (void *)exists); ASSERT(refcount_is_zero( &exists->b_l1hdr.b_refcnt)); arc_change_state(arc_anon, exists, hash_lock); mutex_exit(hash_lock); arc_hdr_destroy(exists); exists = buf_hash_insert(hdr, &hash_lock); ASSERT3P(exists, ==, NULL); } else if (zio->io_flags & ZIO_FLAG_NOPWRITE) { /* nopwrite */ ASSERT(zio->io_prop.zp_nopwrite); if (!BP_EQUAL(&zio->io_bp_orig, zio->io_bp)) panic("bad nopwrite, hdr=%p exists=%p", (void *)hdr, (void *)exists); } else { /* Dedup */ ASSERT(hdr->b_l1hdr.b_bufcnt == 1); ASSERT(hdr->b_l1hdr.b_state == arc_anon); ASSERT(BP_GET_DEDUP(zio->io_bp)); ASSERT(BP_GET_LEVEL(zio->io_bp) == 0); } } arc_hdr_clear_flags(hdr, ARC_FLAG_IO_IN_PROGRESS); /* if it's not anon, we are doing a scrub */ if (exists == NULL && hdr->b_l1hdr.b_state == arc_anon) arc_access(hdr, hash_lock); mutex_exit(hash_lock); } else { arc_hdr_clear_flags(hdr, ARC_FLAG_IO_IN_PROGRESS); } ASSERT(!refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); callback->awcb_done(zio, buf, callback->awcb_private); kmem_free(callback, sizeof (arc_write_callback_t)); } zio_t * arc_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, arc_buf_t *buf, boolean_t l2arc, const zio_prop_t *zp, arc_done_func_t *ready, arc_done_func_t *children_ready, arc_done_func_t *physdone, arc_done_func_t *done, void *private, zio_priority_t priority, int zio_flags, const zbookmark_phys_t *zb) { arc_buf_hdr_t *hdr = buf->b_hdr; arc_write_callback_t *callback; zio_t *zio; ASSERT3P(ready, !=, NULL); ASSERT3P(done, !=, NULL); ASSERT(!HDR_IO_ERROR(hdr)); ASSERT(!HDR_IO_IN_PROGRESS(hdr)); ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL); ASSERT3U(hdr->b_l1hdr.b_bufcnt, >, 0); if (l2arc) arc_hdr_set_flags(hdr, ARC_FLAG_L2CACHE); if (ARC_BUF_COMPRESSED(buf)) { ASSERT3U(zp->zp_compress, !=, ZIO_COMPRESS_OFF); ASSERT3U(HDR_GET_LSIZE(hdr), !=, arc_buf_size(buf)); zio_flags |= ZIO_FLAG_RAW; } callback = kmem_zalloc(sizeof (arc_write_callback_t), KM_SLEEP); callback->awcb_ready = ready; callback->awcb_children_ready = children_ready; callback->awcb_physdone = physdone; callback->awcb_done = done; callback->awcb_private = private; callback->awcb_buf = buf; /* * The hdr's b_pdata is now stale, free it now. A new data block * will be allocated when the zio pipeline calls arc_write_ready(). */ if (hdr->b_l1hdr.b_pdata != NULL) { /* * If the buf is currently sharing the data block with * the hdr then we need to break that relationship here. * The hdr will remain with a NULL data pointer and the * buf will take sole ownership of the block. */ if (arc_buf_is_shared(buf)) { arc_unshare_buf(hdr, buf); } else { arc_hdr_free_pdata(hdr); } VERIFY3P(buf->b_data, !=, NULL); arc_hdr_set_compress(hdr, ZIO_COMPRESS_OFF); } ASSERT(!arc_buf_is_shared(buf)); ASSERT3P(hdr->b_l1hdr.b_pdata, ==, NULL); zio = zio_write(pio, spa, txg, bp, buf->b_data, HDR_GET_LSIZE(hdr), arc_buf_size(buf), zp, arc_write_ready, (children_ready != NULL) ? arc_write_children_ready : NULL, arc_write_physdone, arc_write_done, callback, priority, zio_flags, zb); return (zio); } static int arc_memory_throttle(uint64_t reserve, uint64_t txg) { #ifdef _KERNEL uint64_t available_memory = ptob(freemem); static uint64_t page_load = 0; static uint64_t last_txg = 0; #if defined(__i386) available_memory = MIN(available_memory, vmem_size(heap_arena, VMEM_FREE)); #endif if (freemem > physmem * arc_lotsfree_percent / 100) return (0); if (txg > last_txg) { last_txg = txg; page_load = 0; } /* * If we are in pageout, we know that memory is already tight, * the arc is already going to be evicting, so we just want to * continue to let page writes occur as quickly as possible. */ if (curproc == proc_pageout) { if (page_load > MAX(ptob(minfree), available_memory) / 4) return (SET_ERROR(ERESTART)); /* Note: reserve is inflated, so we deflate */ page_load += reserve / 8; return (0); } else if (page_load > 0 && arc_reclaim_needed()) { /* memory is low, delay before restarting */ ARCSTAT_INCR(arcstat_memory_throttle_count, 1); return (SET_ERROR(EAGAIN)); } page_load = 0; #endif return (0); } void arc_tempreserve_clear(uint64_t reserve) { atomic_add_64(&arc_tempreserve, -reserve); ASSERT((int64_t)arc_tempreserve >= 0); } int arc_tempreserve_space(uint64_t reserve, uint64_t txg) { int error; uint64_t anon_size; if (reserve > arc_c/4 && !arc_no_grow) arc_c = MIN(arc_c_max, reserve * 4); if (reserve > arc_c) return (SET_ERROR(ENOMEM)); /* * Don't count loaned bufs as in flight dirty data to prevent long * network delays from blocking transactions that are ready to be * assigned to a txg. */ /* assert that it has not wrapped around */ ASSERT3S(atomic_add_64_nv(&arc_loaned_bytes, 0), >=, 0); anon_size = MAX((int64_t)(refcount_count(&arc_anon->arcs_size) - arc_loaned_bytes), 0); /* * Writes will, almost always, require additional memory allocations * in order to compress/encrypt/etc the data. We therefore need to * make sure that there is sufficient available memory for this. */ error = arc_memory_throttle(reserve, txg); if (error != 0) return (error); /* * Throttle writes when the amount of dirty data in the cache * gets too large. We try to keep the cache less than half full * of dirty blocks so that our sync times don't grow too large. * Note: if two requests come in concurrently, we might let them * both succeed, when one of them should fail. Not a huge deal. */ if (reserve + arc_tempreserve + anon_size > arc_c / 2 && anon_size > arc_c / 4) { uint64_t meta_esize = refcount_count(&arc_anon->arcs_esize[ARC_BUFC_METADATA]); uint64_t data_esize = refcount_count(&arc_anon->arcs_esize[ARC_BUFC_DATA]); dprintf("failing, arc_tempreserve=%lluK anon_meta=%lluK " "anon_data=%lluK tempreserve=%lluK arc_c=%lluK\n", arc_tempreserve >> 10, meta_esize >> 10, data_esize >> 10, reserve >> 10, arc_c >> 10); return (SET_ERROR(ERESTART)); } atomic_add_64(&arc_tempreserve, reserve); return (0); } static void arc_kstat_update_state(arc_state_t *state, kstat_named_t *size, kstat_named_t *evict_data, kstat_named_t *evict_metadata) { size->value.ui64 = refcount_count(&state->arcs_size); evict_data->value.ui64 = refcount_count(&state->arcs_esize[ARC_BUFC_DATA]); evict_metadata->value.ui64 = refcount_count(&state->arcs_esize[ARC_BUFC_METADATA]); } static int arc_kstat_update(kstat_t *ksp, int rw) { arc_stats_t *as = ksp->ks_data; if (rw == KSTAT_WRITE) { return (EACCES); } else { arc_kstat_update_state(arc_anon, &as->arcstat_anon_size, &as->arcstat_anon_evictable_data, &as->arcstat_anon_evictable_metadata); arc_kstat_update_state(arc_mru, &as->arcstat_mru_size, &as->arcstat_mru_evictable_data, &as->arcstat_mru_evictable_metadata); arc_kstat_update_state(arc_mru_ghost, &as->arcstat_mru_ghost_size, &as->arcstat_mru_ghost_evictable_data, &as->arcstat_mru_ghost_evictable_metadata); arc_kstat_update_state(arc_mfu, &as->arcstat_mfu_size, &as->arcstat_mfu_evictable_data, &as->arcstat_mfu_evictable_metadata); arc_kstat_update_state(arc_mfu_ghost, &as->arcstat_mfu_ghost_size, &as->arcstat_mfu_ghost_evictable_data, &as->arcstat_mfu_ghost_evictable_metadata); } return (0); } /* * This function *must* return indices evenly distributed between all * sublists of the multilist. This is needed due to how the ARC eviction * code is laid out; arc_evict_state() assumes ARC buffers are evenly * distributed between all sublists and uses this assumption when * deciding which sublist to evict from and how much to evict from it. */ unsigned int arc_state_multilist_index_func(multilist_t *ml, void *obj) { arc_buf_hdr_t *hdr = obj; /* * We rely on b_dva to generate evenly distributed index * numbers using buf_hash below. So, as an added precaution, * let's make sure we never add empty buffers to the arc lists. */ ASSERT(!HDR_EMPTY(hdr)); /* * The assumption here, is the hash value for a given * arc_buf_hdr_t will remain constant throughout it's lifetime * (i.e. it's b_spa, b_dva, and b_birth fields don't change). * Thus, we don't need to store the header's sublist index * on insertion, as this index can be recalculated on removal. * * Also, the low order bits of the hash value are thought to be * distributed evenly. Otherwise, in the case that the multilist * has a power of two number of sublists, each sublists' usage * would not be evenly distributed. */ return (buf_hash(hdr->b_spa, &hdr->b_dva, hdr->b_birth) % multilist_get_num_sublists(ml)); } static void arc_state_init(void) { arc_anon = &ARC_anon; arc_mru = &ARC_mru; arc_mru_ghost = &ARC_mru_ghost; arc_mfu = &ARC_mfu; arc_mfu_ghost = &ARC_mfu_ghost; arc_l2c_only = &ARC_l2c_only; - multilist_create(&arc_mru->arcs_list[ARC_BUFC_METADATA], - sizeof (arc_buf_hdr_t), + arc_mru->arcs_list[ARC_BUFC_METADATA] = + multilist_create(sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); - multilist_create(&arc_mru->arcs_list[ARC_BUFC_DATA], - sizeof (arc_buf_hdr_t), + arc_mru->arcs_list[ARC_BUFC_DATA] = + multilist_create(sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); - multilist_create(&arc_mru_ghost->arcs_list[ARC_BUFC_METADATA], - sizeof (arc_buf_hdr_t), + arc_mru_ghost->arcs_list[ARC_BUFC_METADATA] = + multilist_create(sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); - multilist_create(&arc_mru_ghost->arcs_list[ARC_BUFC_DATA], - sizeof (arc_buf_hdr_t), + arc_mru_ghost->arcs_list[ARC_BUFC_DATA] = + multilist_create(sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); - multilist_create(&arc_mfu->arcs_list[ARC_BUFC_METADATA], - sizeof (arc_buf_hdr_t), + arc_mfu->arcs_list[ARC_BUFC_METADATA] = + multilist_create(sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); - multilist_create(&arc_mfu->arcs_list[ARC_BUFC_DATA], - sizeof (arc_buf_hdr_t), + arc_mfu->arcs_list[ARC_BUFC_DATA] = + multilist_create(sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); - multilist_create(&arc_mfu_ghost->arcs_list[ARC_BUFC_METADATA], - sizeof (arc_buf_hdr_t), + arc_mfu_ghost->arcs_list[ARC_BUFC_METADATA] = + multilist_create(sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); - multilist_create(&arc_mfu_ghost->arcs_list[ARC_BUFC_DATA], - sizeof (arc_buf_hdr_t), + arc_mfu_ghost->arcs_list[ARC_BUFC_DATA] = + multilist_create(sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); - multilist_create(&arc_l2c_only->arcs_list[ARC_BUFC_METADATA], - sizeof (arc_buf_hdr_t), + arc_l2c_only->arcs_list[ARC_BUFC_METADATA] = + multilist_create(sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); - multilist_create(&arc_l2c_only->arcs_list[ARC_BUFC_DATA], - sizeof (arc_buf_hdr_t), + arc_l2c_only->arcs_list[ARC_BUFC_DATA] = + multilist_create(sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); refcount_create(&arc_anon->arcs_esize[ARC_BUFC_METADATA]); refcount_create(&arc_anon->arcs_esize[ARC_BUFC_DATA]); refcount_create(&arc_mru->arcs_esize[ARC_BUFC_METADATA]); refcount_create(&arc_mru->arcs_esize[ARC_BUFC_DATA]); refcount_create(&arc_mru_ghost->arcs_esize[ARC_BUFC_METADATA]); refcount_create(&arc_mru_ghost->arcs_esize[ARC_BUFC_DATA]); refcount_create(&arc_mfu->arcs_esize[ARC_BUFC_METADATA]); refcount_create(&arc_mfu->arcs_esize[ARC_BUFC_DATA]); refcount_create(&arc_mfu_ghost->arcs_esize[ARC_BUFC_METADATA]); refcount_create(&arc_mfu_ghost->arcs_esize[ARC_BUFC_DATA]); refcount_create(&arc_l2c_only->arcs_esize[ARC_BUFC_METADATA]); refcount_create(&arc_l2c_only->arcs_esize[ARC_BUFC_DATA]); refcount_create(&arc_anon->arcs_size); refcount_create(&arc_mru->arcs_size); refcount_create(&arc_mru_ghost->arcs_size); refcount_create(&arc_mfu->arcs_size); refcount_create(&arc_mfu_ghost->arcs_size); refcount_create(&arc_l2c_only->arcs_size); } static void arc_state_fini(void) { refcount_destroy(&arc_anon->arcs_esize[ARC_BUFC_METADATA]); refcount_destroy(&arc_anon->arcs_esize[ARC_BUFC_DATA]); refcount_destroy(&arc_mru->arcs_esize[ARC_BUFC_METADATA]); refcount_destroy(&arc_mru->arcs_esize[ARC_BUFC_DATA]); refcount_destroy(&arc_mru_ghost->arcs_esize[ARC_BUFC_METADATA]); refcount_destroy(&arc_mru_ghost->arcs_esize[ARC_BUFC_DATA]); refcount_destroy(&arc_mfu->arcs_esize[ARC_BUFC_METADATA]); refcount_destroy(&arc_mfu->arcs_esize[ARC_BUFC_DATA]); refcount_destroy(&arc_mfu_ghost->arcs_esize[ARC_BUFC_METADATA]); refcount_destroy(&arc_mfu_ghost->arcs_esize[ARC_BUFC_DATA]); refcount_destroy(&arc_l2c_only->arcs_esize[ARC_BUFC_METADATA]); refcount_destroy(&arc_l2c_only->arcs_esize[ARC_BUFC_DATA]); refcount_destroy(&arc_anon->arcs_size); refcount_destroy(&arc_mru->arcs_size); refcount_destroy(&arc_mru_ghost->arcs_size); refcount_destroy(&arc_mfu->arcs_size); refcount_destroy(&arc_mfu_ghost->arcs_size); refcount_destroy(&arc_l2c_only->arcs_size); - multilist_destroy(&arc_mru->arcs_list[ARC_BUFC_METADATA]); - multilist_destroy(&arc_mru_ghost->arcs_list[ARC_BUFC_METADATA]); - multilist_destroy(&arc_mfu->arcs_list[ARC_BUFC_METADATA]); - multilist_destroy(&arc_mfu_ghost->arcs_list[ARC_BUFC_METADATA]); - multilist_destroy(&arc_mru->arcs_list[ARC_BUFC_DATA]); - multilist_destroy(&arc_mru_ghost->arcs_list[ARC_BUFC_DATA]); - multilist_destroy(&arc_mfu->arcs_list[ARC_BUFC_DATA]); - multilist_destroy(&arc_mfu_ghost->arcs_list[ARC_BUFC_DATA]); + multilist_destroy(arc_mru->arcs_list[ARC_BUFC_METADATA]); + multilist_destroy(arc_mru_ghost->arcs_list[ARC_BUFC_METADATA]); + multilist_destroy(arc_mfu->arcs_list[ARC_BUFC_METADATA]); + multilist_destroy(arc_mfu_ghost->arcs_list[ARC_BUFC_METADATA]); + multilist_destroy(arc_mru->arcs_list[ARC_BUFC_DATA]); + multilist_destroy(arc_mru_ghost->arcs_list[ARC_BUFC_DATA]); + multilist_destroy(arc_mfu->arcs_list[ARC_BUFC_DATA]); + multilist_destroy(arc_mfu_ghost->arcs_list[ARC_BUFC_DATA]); } uint64_t arc_max_bytes(void) { return (arc_c_max); } void arc_init(void) { /* * allmem is "all memory that we could possibly use". */ #ifdef _KERNEL uint64_t allmem = ptob(physmem - swapfs_minfree); #else uint64_t allmem = (physmem * PAGESIZE) / 2; #endif mutex_init(&arc_reclaim_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&arc_reclaim_thread_cv, NULL, CV_DEFAULT, NULL); cv_init(&arc_reclaim_waiters_cv, NULL, CV_DEFAULT, NULL); /* Convert seconds to clock ticks */ arc_min_prefetch_lifespan = 1 * hz; /* Start out with 1/8 of all memory */ arc_c = allmem / 8; #ifdef _KERNEL /* * On architectures where the physical memory can be larger * than the addressable space (intel in 32-bit mode), we may * need to limit the cache to 1/8 of VM size. */ arc_c = MIN(arc_c, vmem_size(heap_arena, VMEM_ALLOC | VMEM_FREE) / 8); #endif /* set min cache to 1/32 of all memory, or 64MB, whichever is more */ arc_c_min = MAX(allmem / 32, 64 << 20); /* set max to 3/4 of all memory, or all but 1GB, whichever is more */ if (allmem >= 1 << 30) arc_c_max = allmem - (1 << 30); else arc_c_max = arc_c_min; arc_c_max = MAX(allmem * 3 / 4, arc_c_max); /* * In userland, there's only the memory pressure that we artificially * create (see arc_available_memory()). Don't let arc_c get too * small, because it can cause transactions to be larger than * arc_c, causing arc_tempreserve_space() to fail. */ #ifndef _KERNEL arc_c_min = arc_c_max / 2; #endif /* * Allow the tunables to override our calculations if they are * reasonable (ie. over 64MB) */ if (zfs_arc_max > 64 << 20 && zfs_arc_max < allmem) { arc_c_max = zfs_arc_max; arc_c_min = MIN(arc_c_min, arc_c_max); } if (zfs_arc_min > 64 << 20 && zfs_arc_min <= arc_c_max) arc_c_min = zfs_arc_min; arc_c = arc_c_max; arc_p = (arc_c >> 1); arc_size = 0; /* limit meta-data to 1/4 of the arc capacity */ arc_meta_limit = arc_c_max / 4; /* Allow the tunable to override if it is reasonable */ if (zfs_arc_meta_limit > 0 && zfs_arc_meta_limit <= arc_c_max) arc_meta_limit = zfs_arc_meta_limit; if (arc_c_min < arc_meta_limit / 2 && zfs_arc_min == 0) arc_c_min = arc_meta_limit / 2; if (zfs_arc_meta_min > 0) { arc_meta_min = zfs_arc_meta_min; } else { arc_meta_min = arc_c_min / 2; } if (zfs_arc_grow_retry > 0) arc_grow_retry = zfs_arc_grow_retry; if (zfs_arc_shrink_shift > 0) arc_shrink_shift = zfs_arc_shrink_shift; /* * Ensure that arc_no_grow_shift is less than arc_shrink_shift. */ if (arc_no_grow_shift >= arc_shrink_shift) arc_no_grow_shift = arc_shrink_shift - 1; if (zfs_arc_p_min_shift > 0) arc_p_min_shift = zfs_arc_p_min_shift; /* if kmem_flags are set, lets try to use less memory */ if (kmem_debugging()) arc_c = arc_c / 2; if (arc_c < arc_c_min) arc_c = arc_c_min; arc_state_init(); buf_init(); arc_reclaim_thread_exit = B_FALSE; arc_ksp = kstat_create("zfs", 0, "arcstats", "misc", KSTAT_TYPE_NAMED, sizeof (arc_stats) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL); if (arc_ksp != NULL) { arc_ksp->ks_data = &arc_stats; arc_ksp->ks_update = arc_kstat_update; kstat_install(arc_ksp); } (void) thread_create(NULL, 0, arc_reclaim_thread, NULL, 0, &p0, TS_RUN, minclsyspri); arc_dead = B_FALSE; arc_warm = B_FALSE; /* * Calculate maximum amount of dirty data per pool. * * If it has been set by /etc/system, take that. * Otherwise, use a percentage of physical memory defined by * zfs_dirty_data_max_percent (default 10%) with a cap at * zfs_dirty_data_max_max (default 4GB). */ if (zfs_dirty_data_max == 0) { zfs_dirty_data_max = physmem * PAGESIZE * zfs_dirty_data_max_percent / 100; zfs_dirty_data_max = MIN(zfs_dirty_data_max, zfs_dirty_data_max_max); } } void arc_fini(void) { mutex_enter(&arc_reclaim_lock); arc_reclaim_thread_exit = B_TRUE; /* * The reclaim thread will set arc_reclaim_thread_exit back to * B_FALSE when it is finished exiting; we're waiting for that. */ while (arc_reclaim_thread_exit) { cv_signal(&arc_reclaim_thread_cv); cv_wait(&arc_reclaim_thread_cv, &arc_reclaim_lock); } mutex_exit(&arc_reclaim_lock); /* Use B_TRUE to ensure *all* buffers are evicted */ arc_flush(NULL, B_TRUE); arc_dead = B_TRUE; if (arc_ksp != NULL) { kstat_delete(arc_ksp); arc_ksp = NULL; } mutex_destroy(&arc_reclaim_lock); cv_destroy(&arc_reclaim_thread_cv); cv_destroy(&arc_reclaim_waiters_cv); arc_state_fini(); buf_fini(); ASSERT0(arc_loaned_bytes); } /* * Level 2 ARC * * The level 2 ARC (L2ARC) is a cache layer in-between main memory and disk. * It uses dedicated storage devices to hold cached data, which are populated * using large infrequent writes. The main role of this cache is to boost * the performance of random read workloads. The intended L2ARC devices * include short-stroked disks, solid state disks, and other media with * substantially faster read latency than disk. * * +-----------------------+ * | ARC | * +-----------------------+ * | ^ ^ * | | | * l2arc_feed_thread() arc_read() * | | | * | l2arc read | * V | | * +---------------+ | * | L2ARC | | * +---------------+ | * | ^ | * l2arc_write() | | * | | | * V | | * +-------+ +-------+ * | vdev | | vdev | * | cache | | cache | * +-------+ +-------+ * +=========+ .-----. * : L2ARC : |-_____-| * : devices : | Disks | * +=========+ `-_____-' * * Read requests are satisfied from the following sources, in order: * * 1) ARC * 2) vdev cache of L2ARC devices * 3) L2ARC devices * 4) vdev cache of disks * 5) disks * * Some L2ARC device types exhibit extremely slow write performance. * To accommodate for this there are some significant differences between * the L2ARC and traditional cache design: * * 1. There is no eviction path from the ARC to the L2ARC. Evictions from * the ARC behave as usual, freeing buffers and placing headers on ghost * lists. The ARC does not send buffers to the L2ARC during eviction as * this would add inflated write latencies for all ARC memory pressure. * * 2. The L2ARC attempts to cache data from the ARC before it is evicted. * It does this by periodically scanning buffers from the eviction-end of * the MFU and MRU ARC lists, copying them to the L2ARC devices if they are * not already there. It scans until a headroom of buffers is satisfied, * which itself is a buffer for ARC eviction. If a compressible buffer is * found during scanning and selected for writing to an L2ARC device, we * temporarily boost scanning headroom during the next scan cycle to make * sure we adapt to compression effects (which might significantly reduce * the data volume we write to L2ARC). The thread that does this is * l2arc_feed_thread(), illustrated below; example sizes are included to * provide a better sense of ratio than this diagram: * * head --> tail * +---------------------+----------+ * ARC_mfu |:::::#:::::::::::::::|o#o###o###|-->. # already on L2ARC * +---------------------+----------+ | o L2ARC eligible * ARC_mru |:#:::::::::::::::::::|#o#ooo####|-->| : ARC buffer * +---------------------+----------+ | * 15.9 Gbytes ^ 32 Mbytes | * headroom | * l2arc_feed_thread() * | * l2arc write hand <--[oooo]--' * | 8 Mbyte * | write max * V * +==============================+ * L2ARC dev |####|#|###|###| |####| ... | * +==============================+ * 32 Gbytes * * 3. If an ARC buffer is copied to the L2ARC but then hit instead of * evicted, then the L2ARC has cached a buffer much sooner than it probably * needed to, potentially wasting L2ARC device bandwidth and storage. It is * safe to say that this is an uncommon case, since buffers at the end of * the ARC lists have moved there due to inactivity. * * 4. If the ARC evicts faster than the L2ARC can maintain a headroom, * then the L2ARC simply misses copying some buffers. This serves as a * pressure valve to prevent heavy read workloads from both stalling the ARC * with waits and clogging the L2ARC with writes. This also helps prevent * the potential for the L2ARC to churn if it attempts to cache content too * quickly, such as during backups of the entire pool. * * 5. After system boot and before the ARC has filled main memory, there are * no evictions from the ARC and so the tails of the ARC_mfu and ARC_mru * lists can remain mostly static. Instead of searching from tail of these * lists as pictured, the l2arc_feed_thread() will search from the list heads * for eligible buffers, greatly increasing its chance of finding them. * * The L2ARC device write speed is also boosted during this time so that * the L2ARC warms up faster. Since there have been no ARC evictions yet, * there are no L2ARC reads, and no fear of degrading read performance * through increased writes. * * 6. Writes to the L2ARC devices are grouped and sent in-sequence, so that * the vdev queue can aggregate them into larger and fewer writes. Each * device is written to in a rotor fashion, sweeping writes through * available space then repeating. * * 7. The L2ARC does not store dirty content. It never needs to flush * write buffers back to disk based storage. * * 8. If an ARC buffer is written (and dirtied) which also exists in the * L2ARC, the now stale L2ARC buffer is immediately dropped. * * The performance of the L2ARC can be tweaked by a number of tunables, which * may be necessary for different workloads: * * l2arc_write_max max write bytes per interval * l2arc_write_boost extra write bytes during device warmup * l2arc_noprefetch skip caching prefetched buffers * l2arc_headroom number of max device writes to precache * l2arc_headroom_boost when we find compressed buffers during ARC * scanning, we multiply headroom by this * percentage factor for the next scan cycle, * since more compressed buffers are likely to * be present * l2arc_feed_secs seconds between L2ARC writing * * Tunables may be removed or added as future performance improvements are * integrated, and also may become zpool properties. * * There are three key functions that control how the L2ARC warms up: * * l2arc_write_eligible() check if a buffer is eligible to cache * l2arc_write_size() calculate how much to write * l2arc_write_interval() calculate sleep delay between writes * * These three functions determine what to write, how much, and how quickly * to send writes. */ static boolean_t l2arc_write_eligible(uint64_t spa_guid, arc_buf_hdr_t *hdr) { /* * A buffer is *not* eligible for the L2ARC if it: * 1. belongs to a different spa. * 2. is already cached on the L2ARC. * 3. has an I/O in progress (it may be an incomplete read). * 4. is flagged not eligible (zfs property). */ if (hdr->b_spa != spa_guid || HDR_HAS_L2HDR(hdr) || HDR_IO_IN_PROGRESS(hdr) || !HDR_L2CACHE(hdr)) return (B_FALSE); return (B_TRUE); } static uint64_t l2arc_write_size(void) { uint64_t size; /* * Make sure our globals have meaningful values in case the user * altered them. */ size = l2arc_write_max; if (size == 0) { cmn_err(CE_NOTE, "Bad value for l2arc_write_max, value must " "be greater than zero, resetting it to the default (%d)", L2ARC_WRITE_SIZE); size = l2arc_write_max = L2ARC_WRITE_SIZE; } if (arc_warm == B_FALSE) size += l2arc_write_boost; return (size); } static clock_t l2arc_write_interval(clock_t began, uint64_t wanted, uint64_t wrote) { clock_t interval, next, now; /* * If the ARC lists are busy, increase our write rate; if the * lists are stale, idle back. This is achieved by checking * how much we previously wrote - if it was more than half of * what we wanted, schedule the next write much sooner. */ if (l2arc_feed_again && wrote > (wanted / 2)) interval = (hz * l2arc_feed_min_ms) / 1000; else interval = hz * l2arc_feed_secs; now = ddi_get_lbolt(); next = MAX(now, MIN(now + interval, began + interval)); return (next); } /* * Cycle through L2ARC devices. This is how L2ARC load balances. * If a device is returned, this also returns holding the spa config lock. */ static l2arc_dev_t * l2arc_dev_get_next(void) { l2arc_dev_t *first, *next = NULL; /* * Lock out the removal of spas (spa_namespace_lock), then removal * of cache devices (l2arc_dev_mtx). Once a device has been selected, * both locks will be dropped and a spa config lock held instead. */ mutex_enter(&spa_namespace_lock); mutex_enter(&l2arc_dev_mtx); /* if there are no vdevs, there is nothing to do */ if (l2arc_ndev == 0) goto out; first = NULL; next = l2arc_dev_last; do { /* loop around the list looking for a non-faulted vdev */ if (next == NULL) { next = list_head(l2arc_dev_list); } else { next = list_next(l2arc_dev_list, next); if (next == NULL) next = list_head(l2arc_dev_list); } /* if we have come back to the start, bail out */ if (first == NULL) first = next; else if (next == first) break; } while (vdev_is_dead(next->l2ad_vdev)); /* if we were unable to find any usable vdevs, return NULL */ if (vdev_is_dead(next->l2ad_vdev)) next = NULL; l2arc_dev_last = next; out: mutex_exit(&l2arc_dev_mtx); /* * Grab the config lock to prevent the 'next' device from being * removed while we are writing to it. */ if (next != NULL) spa_config_enter(next->l2ad_spa, SCL_L2ARC, next, RW_READER); mutex_exit(&spa_namespace_lock); return (next); } /* * Free buffers that were tagged for destruction. */ static void l2arc_do_free_on_write() { list_t *buflist; l2arc_data_free_t *df, *df_prev; mutex_enter(&l2arc_free_on_write_mtx); buflist = l2arc_free_on_write; for (df = list_tail(buflist); df; df = df_prev) { df_prev = list_prev(buflist, df); ASSERT3P(df->l2df_data, !=, NULL); if (df->l2df_type == ARC_BUFC_METADATA) { zio_buf_free(df->l2df_data, df->l2df_size); } else { ASSERT(df->l2df_type == ARC_BUFC_DATA); zio_data_buf_free(df->l2df_data, df->l2df_size); } list_remove(buflist, df); kmem_free(df, sizeof (l2arc_data_free_t)); } mutex_exit(&l2arc_free_on_write_mtx); } /* * A write to a cache device has completed. Update all headers to allow * reads from these buffers to begin. */ static void l2arc_write_done(zio_t *zio) { l2arc_write_callback_t *cb; l2arc_dev_t *dev; list_t *buflist; arc_buf_hdr_t *head, *hdr, *hdr_prev; kmutex_t *hash_lock; int64_t bytes_dropped = 0; cb = zio->io_private; ASSERT3P(cb, !=, NULL); dev = cb->l2wcb_dev; ASSERT3P(dev, !=, NULL); head = cb->l2wcb_head; ASSERT3P(head, !=, NULL); buflist = &dev->l2ad_buflist; ASSERT3P(buflist, !=, NULL); DTRACE_PROBE2(l2arc__iodone, zio_t *, zio, l2arc_write_callback_t *, cb); if (zio->io_error != 0) ARCSTAT_BUMP(arcstat_l2_writes_error); /* * All writes completed, or an error was hit. */ top: mutex_enter(&dev->l2ad_mtx); for (hdr = list_prev(buflist, head); hdr; hdr = hdr_prev) { hdr_prev = list_prev(buflist, hdr); hash_lock = HDR_LOCK(hdr); /* * We cannot use mutex_enter or else we can deadlock * with l2arc_write_buffers (due to swapping the order * the hash lock and l2ad_mtx are taken). */ if (!mutex_tryenter(hash_lock)) { /* * Missed the hash lock. We must retry so we * don't leave the ARC_FLAG_L2_WRITING bit set. */ ARCSTAT_BUMP(arcstat_l2_writes_lock_retry); /* * We don't want to rescan the headers we've * already marked as having been written out, so * we reinsert the head node so we can pick up * where we left off. */ list_remove(buflist, head); list_insert_after(buflist, hdr, head); mutex_exit(&dev->l2ad_mtx); /* * We wait for the hash lock to become available * to try and prevent busy waiting, and increase * the chance we'll be able to acquire the lock * the next time around. */ mutex_enter(hash_lock); mutex_exit(hash_lock); goto top; } /* * We could not have been moved into the arc_l2c_only * state while in-flight due to our ARC_FLAG_L2_WRITING * bit being set. Let's just ensure that's being enforced. */ ASSERT(HDR_HAS_L1HDR(hdr)); if (zio->io_error != 0) { /* * Error - drop L2ARC entry. */ list_remove(buflist, hdr); arc_hdr_clear_flags(hdr, ARC_FLAG_HAS_L2HDR); ARCSTAT_INCR(arcstat_l2_asize, -arc_hdr_size(hdr)); ARCSTAT_INCR(arcstat_l2_size, -HDR_GET_LSIZE(hdr)); bytes_dropped += arc_hdr_size(hdr); (void) refcount_remove_many(&dev->l2ad_alloc, arc_hdr_size(hdr), hdr); } /* * Allow ARC to begin reads and ghost list evictions to * this L2ARC entry. */ arc_hdr_clear_flags(hdr, ARC_FLAG_L2_WRITING); mutex_exit(hash_lock); } atomic_inc_64(&l2arc_writes_done); list_remove(buflist, head); ASSERT(!HDR_HAS_L1HDR(head)); kmem_cache_free(hdr_l2only_cache, head); mutex_exit(&dev->l2ad_mtx); vdev_space_update(dev->l2ad_vdev, -bytes_dropped, 0, 0); l2arc_do_free_on_write(); kmem_free(cb, sizeof (l2arc_write_callback_t)); } /* * A read to a cache device completed. Validate buffer contents before * handing over to the regular ARC routines. */ static void l2arc_read_done(zio_t *zio) { l2arc_read_callback_t *cb; arc_buf_hdr_t *hdr; kmutex_t *hash_lock; boolean_t valid_cksum; ASSERT3P(zio->io_vd, !=, NULL); ASSERT(zio->io_flags & ZIO_FLAG_DONT_PROPAGATE); spa_config_exit(zio->io_spa, SCL_L2ARC, zio->io_vd); cb = zio->io_private; ASSERT3P(cb, !=, NULL); hdr = cb->l2rcb_hdr; ASSERT3P(hdr, !=, NULL); hash_lock = HDR_LOCK(hdr); mutex_enter(hash_lock); ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); ASSERT3P(zio->io_data, !=, NULL); /* * Check this survived the L2ARC journey. */ ASSERT3P(zio->io_data, ==, hdr->b_l1hdr.b_pdata); zio->io_bp_copy = cb->l2rcb_bp; /* XXX fix in L2ARC 2.0 */ zio->io_bp = &zio->io_bp_copy; /* XXX fix in L2ARC 2.0 */ valid_cksum = arc_cksum_is_equal(hdr, zio); if (valid_cksum && zio->io_error == 0 && !HDR_L2_EVICTED(hdr)) { mutex_exit(hash_lock); zio->io_private = hdr; arc_read_done(zio); } else { mutex_exit(hash_lock); /* * Buffer didn't survive caching. Increment stats and * reissue to the original storage device. */ if (zio->io_error != 0) { ARCSTAT_BUMP(arcstat_l2_io_error); } else { zio->io_error = SET_ERROR(EIO); } if (!valid_cksum) ARCSTAT_BUMP(arcstat_l2_cksum_bad); /* * If there's no waiter, issue an async i/o to the primary * storage now. If there *is* a waiter, the caller must * issue the i/o in a context where it's OK to block. */ if (zio->io_waiter == NULL) { zio_t *pio = zio_unique_parent(zio); ASSERT(!pio || pio->io_child_type == ZIO_CHILD_LOGICAL); zio_nowait(zio_read(pio, zio->io_spa, zio->io_bp, hdr->b_l1hdr.b_pdata, zio->io_size, arc_read_done, hdr, zio->io_priority, cb->l2rcb_flags, &cb->l2rcb_zb)); } } kmem_free(cb, sizeof (l2arc_read_callback_t)); } /* * This is the list priority from which the L2ARC will search for pages to * cache. This is used within loops (0..3) to cycle through lists in the * desired order. This order can have a significant effect on cache * performance. * * Currently the metadata lists are hit first, MFU then MRU, followed by * the data lists. This function returns a locked list, and also returns * the lock pointer. */ static multilist_sublist_t * l2arc_sublist_lock(int list_num) { multilist_t *ml = NULL; unsigned int idx; ASSERT(list_num >= 0 && list_num <= 3); switch (list_num) { case 0: - ml = &arc_mfu->arcs_list[ARC_BUFC_METADATA]; + ml = arc_mfu->arcs_list[ARC_BUFC_METADATA]; break; case 1: - ml = &arc_mru->arcs_list[ARC_BUFC_METADATA]; + ml = arc_mru->arcs_list[ARC_BUFC_METADATA]; break; case 2: - ml = &arc_mfu->arcs_list[ARC_BUFC_DATA]; + ml = arc_mfu->arcs_list[ARC_BUFC_DATA]; break; case 3: - ml = &arc_mru->arcs_list[ARC_BUFC_DATA]; + ml = arc_mru->arcs_list[ARC_BUFC_DATA]; break; } /* * Return a randomly-selected sublist. This is acceptable * because the caller feeds only a little bit of data for each * call (8MB). Subsequent calls will result in different * sublists being selected. */ idx = multilist_get_random_index(ml); return (multilist_sublist_lock(ml, idx)); } /* * Evict buffers from the device write hand to the distance specified in * bytes. This distance may span populated buffers, it may span nothing. * This is clearing a region on the L2ARC device ready for writing. * If the 'all' boolean is set, every buffer is evicted. */ static void l2arc_evict(l2arc_dev_t *dev, uint64_t distance, boolean_t all) { list_t *buflist; arc_buf_hdr_t *hdr, *hdr_prev; kmutex_t *hash_lock; uint64_t taddr; buflist = &dev->l2ad_buflist; if (!all && dev->l2ad_first) { /* * This is the first sweep through the device. There is * nothing to evict. */ return; } if (dev->l2ad_hand >= (dev->l2ad_end - (2 * distance))) { /* * When nearing the end of the device, evict to the end * before the device write hand jumps to the start. */ taddr = dev->l2ad_end; } else { taddr = dev->l2ad_hand + distance; } DTRACE_PROBE4(l2arc__evict, l2arc_dev_t *, dev, list_t *, buflist, uint64_t, taddr, boolean_t, all); top: mutex_enter(&dev->l2ad_mtx); for (hdr = list_tail(buflist); hdr; hdr = hdr_prev) { hdr_prev = list_prev(buflist, hdr); hash_lock = HDR_LOCK(hdr); /* * We cannot use mutex_enter or else we can deadlock * with l2arc_write_buffers (due to swapping the order * the hash lock and l2ad_mtx are taken). */ if (!mutex_tryenter(hash_lock)) { /* * Missed the hash lock. Retry. */ ARCSTAT_BUMP(arcstat_l2_evict_lock_retry); mutex_exit(&dev->l2ad_mtx); mutex_enter(hash_lock); mutex_exit(hash_lock); goto top; } if (HDR_L2_WRITE_HEAD(hdr)) { /* * We hit a write head node. Leave it for * l2arc_write_done(). */ list_remove(buflist, hdr); mutex_exit(hash_lock); continue; } if (!all && HDR_HAS_L2HDR(hdr) && (hdr->b_l2hdr.b_daddr > taddr || hdr->b_l2hdr.b_daddr < dev->l2ad_hand)) { /* * We've evicted to the target address, * or the end of the device. */ mutex_exit(hash_lock); break; } ASSERT(HDR_HAS_L2HDR(hdr)); if (!HDR_HAS_L1HDR(hdr)) { ASSERT(!HDR_L2_READING(hdr)); /* * This doesn't exist in the ARC. Destroy. * arc_hdr_destroy() will call list_remove() * and decrement arcstat_l2_size. */ arc_change_state(arc_anon, hdr, hash_lock); arc_hdr_destroy(hdr); } else { ASSERT(hdr->b_l1hdr.b_state != arc_l2c_only); ARCSTAT_BUMP(arcstat_l2_evict_l1cached); /* * Invalidate issued or about to be issued * reads, since we may be about to write * over this location. */ if (HDR_L2_READING(hdr)) { ARCSTAT_BUMP(arcstat_l2_evict_reading); arc_hdr_set_flags(hdr, ARC_FLAG_L2_EVICTED); } /* Ensure this header has finished being written */ ASSERT(!HDR_L2_WRITING(hdr)); arc_hdr_l2hdr_destroy(hdr); } mutex_exit(hash_lock); } mutex_exit(&dev->l2ad_mtx); } /* * Find and write ARC buffers to the L2ARC device. * * An ARC_FLAG_L2_WRITING flag is set so that the L2ARC buffers are not valid * for reading until they have completed writing. * The headroom_boost is an in-out parameter used to maintain headroom boost * state between calls to this function. * * Returns the number of bytes actually written (which may be smaller than * the delta by which the device hand has changed due to alignment). */ static uint64_t l2arc_write_buffers(spa_t *spa, l2arc_dev_t *dev, uint64_t target_sz) { arc_buf_hdr_t *hdr, *hdr_prev, *head; uint64_t write_asize, write_psize, write_sz, headroom; boolean_t full; l2arc_write_callback_t *cb; zio_t *pio, *wzio; uint64_t guid = spa_load_guid(spa); ASSERT3P(dev->l2ad_vdev, !=, NULL); pio = NULL; write_sz = write_asize = write_psize = 0; full = B_FALSE; head = kmem_cache_alloc(hdr_l2only_cache, KM_PUSHPAGE); arc_hdr_set_flags(head, ARC_FLAG_L2_WRITE_HEAD | ARC_FLAG_HAS_L2HDR); /* * Copy buffers for L2ARC writing. */ for (int try = 0; try <= 3; try++) { multilist_sublist_t *mls = l2arc_sublist_lock(try); uint64_t passed_sz = 0; /* * L2ARC fast warmup. * * Until the ARC is warm and starts to evict, read from the * head of the ARC lists rather than the tail. */ if (arc_warm == B_FALSE) hdr = multilist_sublist_head(mls); else hdr = multilist_sublist_tail(mls); headroom = target_sz * l2arc_headroom; if (zfs_compressed_arc_enabled) headroom = (headroom * l2arc_headroom_boost) / 100; for (; hdr; hdr = hdr_prev) { kmutex_t *hash_lock; if (arc_warm == B_FALSE) hdr_prev = multilist_sublist_next(mls, hdr); else hdr_prev = multilist_sublist_prev(mls, hdr); hash_lock = HDR_LOCK(hdr); if (!mutex_tryenter(hash_lock)) { /* * Skip this buffer rather than waiting. */ continue; } passed_sz += HDR_GET_LSIZE(hdr); if (passed_sz > headroom) { /* * Searched too far. */ mutex_exit(hash_lock); break; } if (!l2arc_write_eligible(guid, hdr)) { mutex_exit(hash_lock); continue; } if ((write_asize + HDR_GET_LSIZE(hdr)) > target_sz) { full = B_TRUE; mutex_exit(hash_lock); break; } if (pio == NULL) { /* * Insert a dummy header on the buflist so * l2arc_write_done() can find where the * write buffers begin without searching. */ mutex_enter(&dev->l2ad_mtx); list_insert_head(&dev->l2ad_buflist, head); mutex_exit(&dev->l2ad_mtx); cb = kmem_alloc( sizeof (l2arc_write_callback_t), KM_SLEEP); cb->l2wcb_dev = dev; cb->l2wcb_head = head; pio = zio_root(spa, l2arc_write_done, cb, ZIO_FLAG_CANFAIL); } hdr->b_l2hdr.b_dev = dev; hdr->b_l2hdr.b_daddr = dev->l2ad_hand; arc_hdr_set_flags(hdr, ARC_FLAG_L2_WRITING | ARC_FLAG_HAS_L2HDR); mutex_enter(&dev->l2ad_mtx); list_insert_head(&dev->l2ad_buflist, hdr); mutex_exit(&dev->l2ad_mtx); /* * We rely on the L1 portion of the header below, so * it's invalid for this header to have been evicted out * of the ghost cache, prior to being written out. The * ARC_FLAG_L2_WRITING bit ensures this won't happen. */ ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT3U(HDR_GET_PSIZE(hdr), >, 0); ASSERT3P(hdr->b_l1hdr.b_pdata, !=, NULL); ASSERT3U(arc_hdr_size(hdr), >, 0); uint64_t size = arc_hdr_size(hdr); (void) refcount_add_many(&dev->l2ad_alloc, size, hdr); /* * Normally the L2ARC can use the hdr's data, but if * we're sharing data between the hdr and one of its * bufs, L2ARC needs its own copy of the data so that * the ZIO below can't race with the buf consumer. To * ensure that this copy will be available for the * lifetime of the ZIO and be cleaned up afterwards, we * add it to the l2arc_free_on_write queue. */ void *to_write; if (!HDR_SHARED_DATA(hdr)) { to_write = hdr->b_l1hdr.b_pdata; } else { arc_buf_contents_t type = arc_buf_type(hdr); if (type == ARC_BUFC_METADATA) { to_write = zio_buf_alloc(size); } else { ASSERT3U(type, ==, ARC_BUFC_DATA); to_write = zio_data_buf_alloc(size); } bcopy(hdr->b_l1hdr.b_pdata, to_write, size); l2arc_free_data_on_write(to_write, size, type); } wzio = zio_write_phys(pio, dev->l2ad_vdev, hdr->b_l2hdr.b_daddr, size, to_write, ZIO_CHECKSUM_OFF, NULL, hdr, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_CANFAIL, B_FALSE); write_sz += HDR_GET_LSIZE(hdr); DTRACE_PROBE2(l2arc__write, vdev_t *, dev->l2ad_vdev, zio_t *, wzio); write_asize += size; /* * Keep the clock hand suitably device-aligned. */ uint64_t asize = vdev_psize_to_asize(dev->l2ad_vdev, size); write_psize += asize; dev->l2ad_hand += asize; mutex_exit(hash_lock); (void) zio_nowait(wzio); } multilist_sublist_unlock(mls); if (full == B_TRUE) break; } /* No buffers selected for writing? */ if (pio == NULL) { ASSERT0(write_sz); ASSERT(!HDR_HAS_L1HDR(head)); kmem_cache_free(hdr_l2only_cache, head); return (0); } ASSERT3U(write_asize, <=, target_sz); ARCSTAT_BUMP(arcstat_l2_writes_sent); ARCSTAT_INCR(arcstat_l2_write_bytes, write_asize); ARCSTAT_INCR(arcstat_l2_size, write_sz); ARCSTAT_INCR(arcstat_l2_asize, write_asize); vdev_space_update(dev->l2ad_vdev, write_asize, 0, 0); /* * Bump device hand to the device start if it is approaching the end. * l2arc_evict() will already have evicted ahead for this case. */ if (dev->l2ad_hand >= (dev->l2ad_end - target_sz)) { dev->l2ad_hand = dev->l2ad_start; dev->l2ad_first = B_FALSE; } dev->l2ad_writing = B_TRUE; (void) zio_wait(pio); dev->l2ad_writing = B_FALSE; return (write_asize); } /* * This thread feeds the L2ARC at regular intervals. This is the beating * heart of the L2ARC. */ static void l2arc_feed_thread(void) { callb_cpr_t cpr; l2arc_dev_t *dev; spa_t *spa; uint64_t size, wrote; clock_t begin, next = ddi_get_lbolt(); CALLB_CPR_INIT(&cpr, &l2arc_feed_thr_lock, callb_generic_cpr, FTAG); mutex_enter(&l2arc_feed_thr_lock); while (l2arc_thread_exit == 0) { CALLB_CPR_SAFE_BEGIN(&cpr); (void) cv_timedwait(&l2arc_feed_thr_cv, &l2arc_feed_thr_lock, next); CALLB_CPR_SAFE_END(&cpr, &l2arc_feed_thr_lock); next = ddi_get_lbolt() + hz; /* * Quick check for L2ARC devices. */ mutex_enter(&l2arc_dev_mtx); if (l2arc_ndev == 0) { mutex_exit(&l2arc_dev_mtx); continue; } mutex_exit(&l2arc_dev_mtx); begin = ddi_get_lbolt(); /* * This selects the next l2arc device to write to, and in * doing so the next spa to feed from: dev->l2ad_spa. This * will return NULL if there are now no l2arc devices or if * they are all faulted. * * If a device is returned, its spa's config lock is also * held to prevent device removal. l2arc_dev_get_next() * will grab and release l2arc_dev_mtx. */ if ((dev = l2arc_dev_get_next()) == NULL) continue; spa = dev->l2ad_spa; ASSERT3P(spa, !=, NULL); /* * If the pool is read-only then force the feed thread to * sleep a little longer. */ if (!spa_writeable(spa)) { next = ddi_get_lbolt() + 5 * l2arc_feed_secs * hz; spa_config_exit(spa, SCL_L2ARC, dev); continue; } /* * Avoid contributing to memory pressure. */ if (arc_reclaim_needed()) { ARCSTAT_BUMP(arcstat_l2_abort_lowmem); spa_config_exit(spa, SCL_L2ARC, dev); continue; } ARCSTAT_BUMP(arcstat_l2_feeds); size = l2arc_write_size(); /* * Evict L2ARC buffers that will be overwritten. */ l2arc_evict(dev, size, B_FALSE); /* * Write ARC buffers. */ wrote = l2arc_write_buffers(spa, dev, size); /* * Calculate interval between writes. */ next = l2arc_write_interval(begin, size, wrote); spa_config_exit(spa, SCL_L2ARC, dev); } l2arc_thread_exit = 0; cv_broadcast(&l2arc_feed_thr_cv); CALLB_CPR_EXIT(&cpr); /* drops l2arc_feed_thr_lock */ thread_exit(); } boolean_t l2arc_vdev_present(vdev_t *vd) { l2arc_dev_t *dev; mutex_enter(&l2arc_dev_mtx); for (dev = list_head(l2arc_dev_list); dev != NULL; dev = list_next(l2arc_dev_list, dev)) { if (dev->l2ad_vdev == vd) break; } mutex_exit(&l2arc_dev_mtx); return (dev != NULL); } /* * Add a vdev for use by the L2ARC. By this point the spa has already * validated the vdev and opened it. */ void l2arc_add_vdev(spa_t *spa, vdev_t *vd) { l2arc_dev_t *adddev; ASSERT(!l2arc_vdev_present(vd)); /* * Create a new l2arc device entry. */ adddev = kmem_zalloc(sizeof (l2arc_dev_t), KM_SLEEP); adddev->l2ad_spa = spa; adddev->l2ad_vdev = vd; adddev->l2ad_start = VDEV_LABEL_START_SIZE; adddev->l2ad_end = VDEV_LABEL_START_SIZE + vdev_get_min_asize(vd); adddev->l2ad_hand = adddev->l2ad_start; adddev->l2ad_first = B_TRUE; adddev->l2ad_writing = B_FALSE; mutex_init(&adddev->l2ad_mtx, NULL, MUTEX_DEFAULT, NULL); /* * This is a list of all ARC buffers that are still valid on the * device. */ list_create(&adddev->l2ad_buflist, sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l2hdr.b_l2node)); vdev_space_update(vd, 0, 0, adddev->l2ad_end - adddev->l2ad_hand); refcount_create(&adddev->l2ad_alloc); /* * Add device to global list */ mutex_enter(&l2arc_dev_mtx); list_insert_head(l2arc_dev_list, adddev); atomic_inc_64(&l2arc_ndev); mutex_exit(&l2arc_dev_mtx); } /* * Remove a vdev from the L2ARC. */ void l2arc_remove_vdev(vdev_t *vd) { l2arc_dev_t *dev, *nextdev, *remdev = NULL; /* * Find the device by vdev */ mutex_enter(&l2arc_dev_mtx); for (dev = list_head(l2arc_dev_list); dev; dev = nextdev) { nextdev = list_next(l2arc_dev_list, dev); if (vd == dev->l2ad_vdev) { remdev = dev; break; } } ASSERT3P(remdev, !=, NULL); /* * Remove device from global list */ list_remove(l2arc_dev_list, remdev); l2arc_dev_last = NULL; /* may have been invalidated */ atomic_dec_64(&l2arc_ndev); mutex_exit(&l2arc_dev_mtx); /* * Clear all buflists and ARC references. L2ARC device flush. */ l2arc_evict(remdev, 0, B_TRUE); list_destroy(&remdev->l2ad_buflist); mutex_destroy(&remdev->l2ad_mtx); refcount_destroy(&remdev->l2ad_alloc); kmem_free(remdev, sizeof (l2arc_dev_t)); } void l2arc_init(void) { l2arc_thread_exit = 0; l2arc_ndev = 0; l2arc_writes_sent = 0; l2arc_writes_done = 0; mutex_init(&l2arc_feed_thr_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&l2arc_feed_thr_cv, NULL, CV_DEFAULT, NULL); mutex_init(&l2arc_dev_mtx, NULL, MUTEX_DEFAULT, NULL); mutex_init(&l2arc_free_on_write_mtx, NULL, MUTEX_DEFAULT, NULL); l2arc_dev_list = &L2ARC_dev_list; l2arc_free_on_write = &L2ARC_free_on_write; list_create(l2arc_dev_list, sizeof (l2arc_dev_t), offsetof(l2arc_dev_t, l2ad_node)); list_create(l2arc_free_on_write, sizeof (l2arc_data_free_t), offsetof(l2arc_data_free_t, l2df_list_node)); } void l2arc_fini(void) { /* * This is called from dmu_fini(), which is called from spa_fini(); * Because of this, we can assume that all l2arc devices have * already been removed when the pools themselves were removed. */ l2arc_do_free_on_write(); mutex_destroy(&l2arc_feed_thr_lock); cv_destroy(&l2arc_feed_thr_cv); mutex_destroy(&l2arc_dev_mtx); mutex_destroy(&l2arc_free_on_write_mtx); list_destroy(l2arc_dev_list); list_destroy(l2arc_free_on_write); } void l2arc_start(void) { if (!(spa_mode_global & FWRITE)) return; (void) thread_create(NULL, 0, l2arc_feed_thread, NULL, 0, &p0, TS_RUN, minclsyspri); } void l2arc_stop(void) { if (!(spa_mode_global & FWRITE)) return; mutex_enter(&l2arc_feed_thr_lock); cv_signal(&l2arc_feed_thr_cv); /* kick thread out of startup */ l2arc_thread_exit = 1; while (l2arc_thread_exit != 0) cv_wait(&l2arc_feed_thr_cv, &l2arc_feed_thr_lock); mutex_exit(&l2arc_feed_thr_lock); } Index: vendor-sys/illumos/dist/uts/common/fs/zfs/dbuf.c =================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/dbuf.c (revision 316916) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/dbuf.c (revision 316917) @@ -1,3585 +1,3585 @@ /* * 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 http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright 2011 Nexenta Systems, Inc. All rights reserved. - * Copyright (c) 2012, 2016 by Delphix. All rights reserved. + * Copyright (c) 2012, 2017 by Delphix. All rights reserved. * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. * Copyright (c) 2013, Joyent, Inc. All rights reserved. * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. * Copyright (c) 2014 Integros [integros.com] */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include uint_t zfs_dbuf_evict_key; static boolean_t dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx); static void dbuf_write(dbuf_dirty_record_t *dr, arc_buf_t *data, dmu_tx_t *tx); #ifndef __lint extern inline void dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func_sync, dmu_buf_evict_func_t *evict_func_async, dmu_buf_t **clear_on_evict_dbufp); #endif /* ! __lint */ /* * Global data structures and functions for the dbuf cache. */ static kmem_cache_t *dbuf_kmem_cache; static taskq_t *dbu_evict_taskq; static kthread_t *dbuf_cache_evict_thread; static kmutex_t dbuf_evict_lock; static kcondvar_t dbuf_evict_cv; static boolean_t dbuf_evict_thread_exit; /* * LRU cache of dbufs. The dbuf cache maintains a list of dbufs that * are not currently held but have been recently released. These dbufs * are not eligible for arc eviction until they are aged out of the cache. * Dbufs are added to the dbuf cache once the last hold is released. If a * dbuf is later accessed and still exists in the dbuf cache, then it will * be removed from the cache and later re-added to the head of the cache. * Dbufs that are aged out of the cache will be immediately destroyed and * become eligible for arc eviction. */ -static multilist_t dbuf_cache; +static multilist_t *dbuf_cache; static refcount_t dbuf_cache_size; uint64_t dbuf_cache_max_bytes = 100 * 1024 * 1024; /* Cap the size of the dbuf cache to log2 fraction of arc size. */ int dbuf_cache_max_shift = 5; /* * The dbuf cache uses a three-stage eviction policy: * - A low water marker designates when the dbuf eviction thread * should stop evicting from the dbuf cache. * - When we reach the maximum size (aka mid water mark), we * signal the eviction thread to run. * - The high water mark indicates when the eviction thread * is unable to keep up with the incoming load and eviction must * happen in the context of the calling thread. * * The dbuf cache: * (max size) * low water mid water hi water * +----------------------------------------+----------+----------+ * | | | | * | | | | * | | | | * | | | | * +----------------------------------------+----------+----------+ * stop signal evict * evicting eviction directly * thread * * The high and low water marks indicate the operating range for the eviction * thread. The low water mark is, by default, 90% of the total size of the * cache and the high water mark is at 110% (both of these percentages can be * changed by setting dbuf_cache_lowater_pct and dbuf_cache_hiwater_pct, * respectively). The eviction thread will try to ensure that the cache remains * within this range by waking up every second and checking if the cache is * above the low water mark. The thread can also be woken up by callers adding * elements into the cache if the cache is larger than the mid water (i.e max * cache size). Once the eviction thread is woken up and eviction is required, * it will continue evicting buffers until it's able to reduce the cache size * to the low water mark. If the cache size continues to grow and hits the high * water mark, then callers adding elments to the cache will begin to evict * directly from the cache until the cache is no longer above the high water * mark. */ /* * The percentage above and below the maximum cache size. */ uint_t dbuf_cache_hiwater_pct = 10; uint_t dbuf_cache_lowater_pct = 10; /* ARGSUSED */ static int dbuf_cons(void *vdb, void *unused, int kmflag) { dmu_buf_impl_t *db = vdb; bzero(db, sizeof (dmu_buf_impl_t)); mutex_init(&db->db_mtx, NULL, MUTEX_DEFAULT, NULL); cv_init(&db->db_changed, NULL, CV_DEFAULT, NULL); multilist_link_init(&db->db_cache_link); refcount_create(&db->db_holds); return (0); } /* ARGSUSED */ static void dbuf_dest(void *vdb, void *unused) { dmu_buf_impl_t *db = vdb; mutex_destroy(&db->db_mtx); cv_destroy(&db->db_changed); ASSERT(!multilist_link_active(&db->db_cache_link)); refcount_destroy(&db->db_holds); } /* * dbuf hash table routines */ static dbuf_hash_table_t dbuf_hash_table; static uint64_t dbuf_hash_count; static uint64_t dbuf_hash(void *os, uint64_t obj, uint8_t lvl, uint64_t blkid) { uintptr_t osv = (uintptr_t)os; uint64_t crc = -1ULL; ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (lvl)) & 0xFF]; crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (osv >> 6)) & 0xFF]; crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 0)) & 0xFF]; crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 8)) & 0xFF]; crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (blkid >> 0)) & 0xFF]; crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (blkid >> 8)) & 0xFF]; crc ^= (osv>>14) ^ (obj>>16) ^ (blkid>>16); return (crc); } #define DBUF_EQUAL(dbuf, os, obj, level, blkid) \ ((dbuf)->db.db_object == (obj) && \ (dbuf)->db_objset == (os) && \ (dbuf)->db_level == (level) && \ (dbuf)->db_blkid == (blkid)) dmu_buf_impl_t * dbuf_find(objset_t *os, uint64_t obj, uint8_t level, uint64_t blkid) { dbuf_hash_table_t *h = &dbuf_hash_table; uint64_t hv = dbuf_hash(os, obj, level, blkid); uint64_t idx = hv & h->hash_table_mask; dmu_buf_impl_t *db; mutex_enter(DBUF_HASH_MUTEX(h, idx)); for (db = h->hash_table[idx]; db != NULL; db = db->db_hash_next) { if (DBUF_EQUAL(db, os, obj, level, blkid)) { mutex_enter(&db->db_mtx); if (db->db_state != DB_EVICTING) { mutex_exit(DBUF_HASH_MUTEX(h, idx)); return (db); } mutex_exit(&db->db_mtx); } } mutex_exit(DBUF_HASH_MUTEX(h, idx)); return (NULL); } static dmu_buf_impl_t * dbuf_find_bonus(objset_t *os, uint64_t object) { dnode_t *dn; dmu_buf_impl_t *db = NULL; if (dnode_hold(os, object, FTAG, &dn) == 0) { rw_enter(&dn->dn_struct_rwlock, RW_READER); if (dn->dn_bonus != NULL) { db = dn->dn_bonus; mutex_enter(&db->db_mtx); } rw_exit(&dn->dn_struct_rwlock); dnode_rele(dn, FTAG); } return (db); } /* * Insert an entry into the hash table. If there is already an element * equal to elem in the hash table, then the already existing element * will be returned and the new element will not be inserted. * Otherwise returns NULL. */ static dmu_buf_impl_t * dbuf_hash_insert(dmu_buf_impl_t *db) { dbuf_hash_table_t *h = &dbuf_hash_table; objset_t *os = db->db_objset; uint64_t obj = db->db.db_object; int level = db->db_level; uint64_t blkid = db->db_blkid; uint64_t hv = dbuf_hash(os, obj, level, blkid); uint64_t idx = hv & h->hash_table_mask; dmu_buf_impl_t *dbf; mutex_enter(DBUF_HASH_MUTEX(h, idx)); for (dbf = h->hash_table[idx]; dbf != NULL; dbf = dbf->db_hash_next) { if (DBUF_EQUAL(dbf, os, obj, level, blkid)) { mutex_enter(&dbf->db_mtx); if (dbf->db_state != DB_EVICTING) { mutex_exit(DBUF_HASH_MUTEX(h, idx)); return (dbf); } mutex_exit(&dbf->db_mtx); } } mutex_enter(&db->db_mtx); db->db_hash_next = h->hash_table[idx]; h->hash_table[idx] = db; mutex_exit(DBUF_HASH_MUTEX(h, idx)); atomic_inc_64(&dbuf_hash_count); return (NULL); } /* * Remove an entry from the hash table. It must be in the EVICTING state. */ static void dbuf_hash_remove(dmu_buf_impl_t *db) { dbuf_hash_table_t *h = &dbuf_hash_table; uint64_t hv = dbuf_hash(db->db_objset, db->db.db_object, db->db_level, db->db_blkid); uint64_t idx = hv & h->hash_table_mask; dmu_buf_impl_t *dbf, **dbp; /* * We musn't hold db_mtx to maintain lock ordering: * DBUF_HASH_MUTEX > db_mtx. */ ASSERT(refcount_is_zero(&db->db_holds)); ASSERT(db->db_state == DB_EVICTING); ASSERT(!MUTEX_HELD(&db->db_mtx)); mutex_enter(DBUF_HASH_MUTEX(h, idx)); dbp = &h->hash_table[idx]; while ((dbf = *dbp) != db) { dbp = &dbf->db_hash_next; ASSERT(dbf != NULL); } *dbp = db->db_hash_next; db->db_hash_next = NULL; mutex_exit(DBUF_HASH_MUTEX(h, idx)); atomic_dec_64(&dbuf_hash_count); } typedef enum { DBVU_EVICTING, DBVU_NOT_EVICTING } dbvu_verify_type_t; static void dbuf_verify_user(dmu_buf_impl_t *db, dbvu_verify_type_t verify_type) { #ifdef ZFS_DEBUG int64_t holds; if (db->db_user == NULL) return; /* Only data blocks support the attachment of user data. */ ASSERT(db->db_level == 0); /* Clients must resolve a dbuf before attaching user data. */ ASSERT(db->db.db_data != NULL); ASSERT3U(db->db_state, ==, DB_CACHED); holds = refcount_count(&db->db_holds); if (verify_type == DBVU_EVICTING) { /* * Immediate eviction occurs when holds == dirtycnt. * For normal eviction buffers, holds is zero on * eviction, except when dbuf_fix_old_data() calls * dbuf_clear_data(). However, the hold count can grow * during eviction even though db_mtx is held (see * dmu_bonus_hold() for an example), so we can only * test the generic invariant that holds >= dirtycnt. */ ASSERT3U(holds, >=, db->db_dirtycnt); } else { if (db->db_user_immediate_evict == TRUE) ASSERT3U(holds, >=, db->db_dirtycnt); else ASSERT3U(holds, >, 0); } #endif } static void dbuf_evict_user(dmu_buf_impl_t *db) { dmu_buf_user_t *dbu = db->db_user; ASSERT(MUTEX_HELD(&db->db_mtx)); if (dbu == NULL) return; dbuf_verify_user(db, DBVU_EVICTING); db->db_user = NULL; #ifdef ZFS_DEBUG if (dbu->dbu_clear_on_evict_dbufp != NULL) *dbu->dbu_clear_on_evict_dbufp = NULL; #endif /* * There are two eviction callbacks - one that we call synchronously * and one that we invoke via a taskq. The async one is useful for * avoiding lock order reversals and limiting stack depth. * * Note that if we have a sync callback but no async callback, * it's likely that the sync callback will free the structure * containing the dbu. In that case we need to take care to not * dereference dbu after calling the sync evict func. */ boolean_t has_async = (dbu->dbu_evict_func_async != NULL); if (dbu->dbu_evict_func_sync != NULL) dbu->dbu_evict_func_sync(dbu); if (has_async) { taskq_dispatch_ent(dbu_evict_taskq, dbu->dbu_evict_func_async, dbu, 0, &dbu->dbu_tqent); } } boolean_t dbuf_is_metadata(dmu_buf_impl_t *db) { if (db->db_level > 0) { return (B_TRUE); } else { boolean_t is_metadata; DB_DNODE_ENTER(db); is_metadata = DMU_OT_IS_METADATA(DB_DNODE(db)->dn_type); DB_DNODE_EXIT(db); return (is_metadata); } } /* * This function *must* return indices evenly distributed between all * sublists of the multilist. This is needed due to how the dbuf eviction * code is laid out; dbuf_evict_thread() assumes dbufs are evenly * distributed between all sublists and uses this assumption when * deciding which sublist to evict from and how much to evict from it. */ unsigned int dbuf_cache_multilist_index_func(multilist_t *ml, void *obj) { dmu_buf_impl_t *db = obj; /* * The assumption here, is the hash value for a given * dmu_buf_impl_t will remain constant throughout it's lifetime * (i.e. it's objset, object, level and blkid fields don't change). * Thus, we don't need to store the dbuf's sublist index * on insertion, as this index can be recalculated on removal. * * Also, the low order bits of the hash value are thought to be * distributed evenly. Otherwise, in the case that the multilist * has a power of two number of sublists, each sublists' usage * would not be evenly distributed. */ return (dbuf_hash(db->db_objset, db->db.db_object, db->db_level, db->db_blkid) % multilist_get_num_sublists(ml)); } static inline boolean_t dbuf_cache_above_hiwater(void) { uint64_t dbuf_cache_hiwater_bytes = (dbuf_cache_max_bytes * dbuf_cache_hiwater_pct) / 100; return (refcount_count(&dbuf_cache_size) > dbuf_cache_max_bytes + dbuf_cache_hiwater_bytes); } static inline boolean_t dbuf_cache_above_lowater(void) { uint64_t dbuf_cache_lowater_bytes = (dbuf_cache_max_bytes * dbuf_cache_lowater_pct) / 100; return (refcount_count(&dbuf_cache_size) > dbuf_cache_max_bytes - dbuf_cache_lowater_bytes); } /* * Evict the oldest eligible dbuf from the dbuf cache. */ static void dbuf_evict_one(void) { - int idx = multilist_get_random_index(&dbuf_cache); - multilist_sublist_t *mls = multilist_sublist_lock(&dbuf_cache, idx); + int idx = multilist_get_random_index(dbuf_cache); + multilist_sublist_t *mls = multilist_sublist_lock(dbuf_cache, idx); ASSERT(!MUTEX_HELD(&dbuf_evict_lock)); /* * Set the thread's tsd to indicate that it's processing evictions. * Once a thread stops evicting from the dbuf cache it will * reset its tsd to NULL. */ ASSERT3P(tsd_get(zfs_dbuf_evict_key), ==, NULL); (void) tsd_set(zfs_dbuf_evict_key, (void *)B_TRUE); dmu_buf_impl_t *db = multilist_sublist_tail(mls); while (db != NULL && mutex_tryenter(&db->db_mtx) == 0) { db = multilist_sublist_prev(mls, db); } DTRACE_PROBE2(dbuf__evict__one, dmu_buf_impl_t *, db, multilist_sublist_t *, mls); if (db != NULL) { multilist_sublist_remove(mls, db); multilist_sublist_unlock(mls); (void) refcount_remove_many(&dbuf_cache_size, db->db.db_size, db); dbuf_destroy(db); } else { multilist_sublist_unlock(mls); } (void) tsd_set(zfs_dbuf_evict_key, NULL); } /* * The dbuf evict thread is responsible for aging out dbufs from the * cache. Once the cache has reached it's maximum size, dbufs are removed * and destroyed. The eviction thread will continue running until the size * of the dbuf cache is at or below the maximum size. Once the dbuf is aged * out of the cache it is destroyed and becomes eligible for arc eviction. */ static void dbuf_evict_thread(void) { callb_cpr_t cpr; CALLB_CPR_INIT(&cpr, &dbuf_evict_lock, callb_generic_cpr, FTAG); mutex_enter(&dbuf_evict_lock); while (!dbuf_evict_thread_exit) { while (!dbuf_cache_above_lowater() && !dbuf_evict_thread_exit) { CALLB_CPR_SAFE_BEGIN(&cpr); (void) cv_timedwait_hires(&dbuf_evict_cv, &dbuf_evict_lock, SEC2NSEC(1), MSEC2NSEC(1), 0); CALLB_CPR_SAFE_END(&cpr, &dbuf_evict_lock); } mutex_exit(&dbuf_evict_lock); /* * Keep evicting as long as we're above the low water mark * for the cache. We do this without holding the locks to * minimize lock contention. */ while (dbuf_cache_above_lowater() && !dbuf_evict_thread_exit) { dbuf_evict_one(); } mutex_enter(&dbuf_evict_lock); } dbuf_evict_thread_exit = B_FALSE; cv_broadcast(&dbuf_evict_cv); CALLB_CPR_EXIT(&cpr); /* drops dbuf_evict_lock */ thread_exit(); } /* * Wake up the dbuf eviction thread if the dbuf cache is at its max size. * If the dbuf cache is at its high water mark, then evict a dbuf from the * dbuf cache using the callers context. */ static void dbuf_evict_notify(void) { /* * We use thread specific data to track when a thread has * started processing evictions. This allows us to avoid deeply * nested stacks that would have a call flow similar to this: * * dbuf_rele()-->dbuf_rele_and_unlock()-->dbuf_evict_notify() * ^ | * | | * +-----dbuf_destroy()<--dbuf_evict_one()<--------+ * * The dbuf_eviction_thread will always have its tsd set until * that thread exits. All other threads will only set their tsd * if they are participating in the eviction process. This only * happens if the eviction thread is unable to process evictions * fast enough. To keep the dbuf cache size in check, other threads * can evict from the dbuf cache directly. Those threads will set * their tsd values so that we ensure that they only evict one dbuf * from the dbuf cache. */ if (tsd_get(zfs_dbuf_evict_key) != NULL) return; if (refcount_count(&dbuf_cache_size) > dbuf_cache_max_bytes) { boolean_t evict_now = B_FALSE; mutex_enter(&dbuf_evict_lock); if (refcount_count(&dbuf_cache_size) > dbuf_cache_max_bytes) { evict_now = dbuf_cache_above_hiwater(); cv_signal(&dbuf_evict_cv); } mutex_exit(&dbuf_evict_lock); if (evict_now) { dbuf_evict_one(); } } } void dbuf_init(void) { uint64_t hsize = 1ULL << 16; dbuf_hash_table_t *h = &dbuf_hash_table; int i; /* * The hash table is big enough to fill all of physical memory * with an average 4K block size. The table will take up * totalmem*sizeof(void*)/4K (i.e. 2MB/GB with 8-byte pointers). */ while (hsize * 4096 < physmem * PAGESIZE) hsize <<= 1; retry: h->hash_table_mask = hsize - 1; h->hash_table = kmem_zalloc(hsize * sizeof (void *), KM_NOSLEEP); if (h->hash_table == NULL) { /* XXX - we should really return an error instead of assert */ ASSERT(hsize > (1ULL << 10)); hsize >>= 1; goto retry; } dbuf_kmem_cache = kmem_cache_create("dmu_buf_impl_t", sizeof (dmu_buf_impl_t), 0, dbuf_cons, dbuf_dest, NULL, NULL, NULL, 0); for (i = 0; i < DBUF_MUTEXES; i++) mutex_init(&h->hash_mutexes[i], NULL, MUTEX_DEFAULT, NULL); /* * Setup the parameters for the dbuf cache. We cap the size of the * dbuf cache to 1/32nd (default) of the size of the ARC. */ dbuf_cache_max_bytes = MIN(dbuf_cache_max_bytes, arc_max_bytes() >> dbuf_cache_max_shift); /* * All entries are queued via taskq_dispatch_ent(), so min/maxalloc * configuration is not required. */ dbu_evict_taskq = taskq_create("dbu_evict", 1, minclsyspri, 0, 0, 0); - multilist_create(&dbuf_cache, sizeof (dmu_buf_impl_t), + dbuf_cache = multilist_create(sizeof (dmu_buf_impl_t), offsetof(dmu_buf_impl_t, db_cache_link), dbuf_cache_multilist_index_func); refcount_create(&dbuf_cache_size); tsd_create(&zfs_dbuf_evict_key, NULL); dbuf_evict_thread_exit = B_FALSE; mutex_init(&dbuf_evict_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&dbuf_evict_cv, NULL, CV_DEFAULT, NULL); dbuf_cache_evict_thread = thread_create(NULL, 0, dbuf_evict_thread, NULL, 0, &p0, TS_RUN, minclsyspri); } void dbuf_fini(void) { dbuf_hash_table_t *h = &dbuf_hash_table; int i; for (i = 0; i < DBUF_MUTEXES; i++) mutex_destroy(&h->hash_mutexes[i]); kmem_free(h->hash_table, (h->hash_table_mask + 1) * sizeof (void *)); kmem_cache_destroy(dbuf_kmem_cache); taskq_destroy(dbu_evict_taskq); mutex_enter(&dbuf_evict_lock); dbuf_evict_thread_exit = B_TRUE; while (dbuf_evict_thread_exit) { cv_signal(&dbuf_evict_cv); cv_wait(&dbuf_evict_cv, &dbuf_evict_lock); } mutex_exit(&dbuf_evict_lock); tsd_destroy(&zfs_dbuf_evict_key); mutex_destroy(&dbuf_evict_lock); cv_destroy(&dbuf_evict_cv); refcount_destroy(&dbuf_cache_size); - multilist_destroy(&dbuf_cache); + multilist_destroy(dbuf_cache); } /* * Other stuff. */ #ifdef ZFS_DEBUG static void dbuf_verify(dmu_buf_impl_t *db) { dnode_t *dn; dbuf_dirty_record_t *dr; ASSERT(MUTEX_HELD(&db->db_mtx)); if (!(zfs_flags & ZFS_DEBUG_DBUF_VERIFY)) return; ASSERT(db->db_objset != NULL); DB_DNODE_ENTER(db); dn = DB_DNODE(db); if (dn == NULL) { ASSERT(db->db_parent == NULL); ASSERT(db->db_blkptr == NULL); } else { ASSERT3U(db->db.db_object, ==, dn->dn_object); ASSERT3P(db->db_objset, ==, dn->dn_objset); ASSERT3U(db->db_level, <, dn->dn_nlevels); ASSERT(db->db_blkid == DMU_BONUS_BLKID || db->db_blkid == DMU_SPILL_BLKID || !avl_is_empty(&dn->dn_dbufs)); } if (db->db_blkid == DMU_BONUS_BLKID) { ASSERT(dn != NULL); ASSERT3U(db->db.db_size, >=, dn->dn_bonuslen); ASSERT3U(db->db.db_offset, ==, DMU_BONUS_BLKID); } else if (db->db_blkid == DMU_SPILL_BLKID) { ASSERT(dn != NULL); ASSERT3U(db->db.db_size, >=, dn->dn_bonuslen); ASSERT0(db->db.db_offset); } else { ASSERT3U(db->db.db_offset, ==, db->db_blkid * db->db.db_size); } for (dr = db->db_data_pending; dr != NULL; dr = dr->dr_next) ASSERT(dr->dr_dbuf == db); for (dr = db->db_last_dirty; dr != NULL; dr = dr->dr_next) ASSERT(dr->dr_dbuf == db); /* * We can't assert that db_size matches dn_datablksz because it * can be momentarily different when another thread is doing * dnode_set_blksz(). */ if (db->db_level == 0 && db->db.db_object == DMU_META_DNODE_OBJECT) { dr = db->db_data_pending; /* * It should only be modified in syncing context, so * make sure we only have one copy of the data. */ ASSERT(dr == NULL || dr->dt.dl.dr_data == db->db_buf); } /* verify db->db_blkptr */ if (db->db_blkptr) { if (db->db_parent == dn->dn_dbuf) { /* db is pointed to by the dnode */ /* ASSERT3U(db->db_blkid, <, dn->dn_nblkptr); */ if (DMU_OBJECT_IS_SPECIAL(db->db.db_object)) ASSERT(db->db_parent == NULL); else ASSERT(db->db_parent != NULL); if (db->db_blkid != DMU_SPILL_BLKID) ASSERT3P(db->db_blkptr, ==, &dn->dn_phys->dn_blkptr[db->db_blkid]); } else { /* db is pointed to by an indirect block */ int epb = db->db_parent->db.db_size >> SPA_BLKPTRSHIFT; ASSERT3U(db->db_parent->db_level, ==, db->db_level+1); ASSERT3U(db->db_parent->db.db_object, ==, db->db.db_object); /* * dnode_grow_indblksz() can make this fail if we don't * have the struct_rwlock. XXX indblksz no longer * grows. safe to do this now? */ if (RW_WRITE_HELD(&dn->dn_struct_rwlock)) { ASSERT3P(db->db_blkptr, ==, ((blkptr_t *)db->db_parent->db.db_data + db->db_blkid % epb)); } } } if ((db->db_blkptr == NULL || BP_IS_HOLE(db->db_blkptr)) && (db->db_buf == NULL || db->db_buf->b_data) && db->db.db_data && db->db_blkid != DMU_BONUS_BLKID && db->db_state != DB_FILL && !dn->dn_free_txg) { /* * If the blkptr isn't set but they have nonzero data, * it had better be dirty, otherwise we'll lose that * data when we evict this buffer. * * There is an exception to this rule for indirect blocks; in * this case, if the indirect block is a hole, we fill in a few * fields on each of the child blocks (importantly, birth time) * to prevent hole birth times from being lost when you * partially fill in a hole. */ if (db->db_dirtycnt == 0) { if (db->db_level == 0) { uint64_t *buf = db->db.db_data; int i; for (i = 0; i < db->db.db_size >> 3; i++) { ASSERT(buf[i] == 0); } } else { blkptr_t *bps = db->db.db_data; ASSERT3U(1 << DB_DNODE(db)->dn_indblkshift, ==, db->db.db_size); /* * We want to verify that all the blkptrs in the * indirect block are holes, but we may have * automatically set up a few fields for them. * We iterate through each blkptr and verify * they only have those fields set. */ for (int i = 0; i < db->db.db_size / sizeof (blkptr_t); i++) { blkptr_t *bp = &bps[i]; ASSERT(ZIO_CHECKSUM_IS_ZERO( &bp->blk_cksum)); ASSERT( DVA_IS_EMPTY(&bp->blk_dva[0]) && DVA_IS_EMPTY(&bp->blk_dva[1]) && DVA_IS_EMPTY(&bp->blk_dva[2])); ASSERT0(bp->blk_fill); ASSERT0(bp->blk_pad[0]); ASSERT0(bp->blk_pad[1]); ASSERT(!BP_IS_EMBEDDED(bp)); ASSERT(BP_IS_HOLE(bp)); ASSERT0(bp->blk_phys_birth); } } } } DB_DNODE_EXIT(db); } #endif static void dbuf_clear_data(dmu_buf_impl_t *db) { ASSERT(MUTEX_HELD(&db->db_mtx)); dbuf_evict_user(db); ASSERT3P(db->db_buf, ==, NULL); db->db.db_data = NULL; if (db->db_state != DB_NOFILL) db->db_state = DB_UNCACHED; } static void dbuf_set_data(dmu_buf_impl_t *db, arc_buf_t *buf) { ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(buf != NULL); db->db_buf = buf; ASSERT(buf->b_data != NULL); db->db.db_data = buf->b_data; } /* * Loan out an arc_buf for read. Return the loaned arc_buf. */ arc_buf_t * dbuf_loan_arcbuf(dmu_buf_impl_t *db) { arc_buf_t *abuf; ASSERT(db->db_blkid != DMU_BONUS_BLKID); mutex_enter(&db->db_mtx); if (arc_released(db->db_buf) || refcount_count(&db->db_holds) > 1) { int blksz = db->db.db_size; spa_t *spa = db->db_objset->os_spa; mutex_exit(&db->db_mtx); abuf = arc_loan_buf(spa, B_FALSE, blksz); bcopy(db->db.db_data, abuf->b_data, blksz); } else { abuf = db->db_buf; arc_loan_inuse_buf(abuf, db); db->db_buf = NULL; dbuf_clear_data(db); mutex_exit(&db->db_mtx); } return (abuf); } /* * Calculate which level n block references the data at the level 0 offset * provided. */ uint64_t dbuf_whichblock(dnode_t *dn, int64_t level, uint64_t offset) { if (dn->dn_datablkshift != 0 && dn->dn_indblkshift != 0) { /* * The level n blkid is equal to the level 0 blkid divided by * the number of level 0s in a level n block. * * The level 0 blkid is offset >> datablkshift = * offset / 2^datablkshift. * * The number of level 0s in a level n is the number of block * pointers in an indirect block, raised to the power of level. * This is 2^(indblkshift - SPA_BLKPTRSHIFT)^level = * 2^(level*(indblkshift - SPA_BLKPTRSHIFT)). * * Thus, the level n blkid is: offset / * ((2^datablkshift)*(2^(level*(indblkshift - SPA_BLKPTRSHIFT))) * = offset / 2^(datablkshift + level * * (indblkshift - SPA_BLKPTRSHIFT)) * = offset >> (datablkshift + level * * (indblkshift - SPA_BLKPTRSHIFT)) */ return (offset >> (dn->dn_datablkshift + level * (dn->dn_indblkshift - SPA_BLKPTRSHIFT))); } else { ASSERT3U(offset, <, dn->dn_datablksz); return (0); } } static void dbuf_read_done(zio_t *zio, arc_buf_t *buf, void *vdb) { dmu_buf_impl_t *db = vdb; mutex_enter(&db->db_mtx); ASSERT3U(db->db_state, ==, DB_READ); /* * All reads are synchronous, so we must have a hold on the dbuf */ ASSERT(refcount_count(&db->db_holds) > 0); ASSERT(db->db_buf == NULL); ASSERT(db->db.db_data == NULL); if (db->db_level == 0 && db->db_freed_in_flight) { /* we were freed in flight; disregard any error */ arc_release(buf, db); bzero(buf->b_data, db->db.db_size); arc_buf_freeze(buf); db->db_freed_in_flight = FALSE; dbuf_set_data(db, buf); db->db_state = DB_CACHED; } else if (zio == NULL || zio->io_error == 0) { dbuf_set_data(db, buf); db->db_state = DB_CACHED; } else { ASSERT(db->db_blkid != DMU_BONUS_BLKID); ASSERT3P(db->db_buf, ==, NULL); arc_buf_destroy(buf, db); db->db_state = DB_UNCACHED; } cv_broadcast(&db->db_changed); dbuf_rele_and_unlock(db, NULL); } static void dbuf_read_impl(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags) { dnode_t *dn; zbookmark_phys_t zb; arc_flags_t aflags = ARC_FLAG_NOWAIT; DB_DNODE_ENTER(db); dn = DB_DNODE(db); ASSERT(!refcount_is_zero(&db->db_holds)); /* We need the struct_rwlock to prevent db_blkptr from changing. */ ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock)); ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(db->db_state == DB_UNCACHED); ASSERT(db->db_buf == NULL); if (db->db_blkid == DMU_BONUS_BLKID) { int bonuslen = MIN(dn->dn_bonuslen, dn->dn_phys->dn_bonuslen); ASSERT3U(bonuslen, <=, db->db.db_size); db->db.db_data = zio_buf_alloc(DN_MAX_BONUSLEN); arc_space_consume(DN_MAX_BONUSLEN, ARC_SPACE_OTHER); if (bonuslen < DN_MAX_BONUSLEN) bzero(db->db.db_data, DN_MAX_BONUSLEN); if (bonuslen) bcopy(DN_BONUS(dn->dn_phys), db->db.db_data, bonuslen); DB_DNODE_EXIT(db); db->db_state = DB_CACHED; mutex_exit(&db->db_mtx); return; } /* * Recheck BP_IS_HOLE() after dnode_block_freed() in case dnode_sync() * processes the delete record and clears the bp while we are waiting * for the dn_mtx (resulting in a "no" from block_freed). */ if (db->db_blkptr == NULL || BP_IS_HOLE(db->db_blkptr) || (db->db_level == 0 && (dnode_block_freed(dn, db->db_blkid) || BP_IS_HOLE(db->db_blkptr)))) { arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db); dbuf_set_data(db, arc_alloc_buf(db->db_objset->os_spa, db, type, db->db.db_size)); bzero(db->db.db_data, db->db.db_size); if (db->db_blkptr != NULL && db->db_level > 0 && BP_IS_HOLE(db->db_blkptr) && db->db_blkptr->blk_birth != 0) { blkptr_t *bps = db->db.db_data; for (int i = 0; i < ((1 << DB_DNODE(db)->dn_indblkshift) / sizeof (blkptr_t)); i++) { blkptr_t *bp = &bps[i]; ASSERT3U(BP_GET_LSIZE(db->db_blkptr), ==, 1 << dn->dn_indblkshift); BP_SET_LSIZE(bp, BP_GET_LEVEL(db->db_blkptr) == 1 ? dn->dn_datablksz : BP_GET_LSIZE(db->db_blkptr)); BP_SET_TYPE(bp, BP_GET_TYPE(db->db_blkptr)); BP_SET_LEVEL(bp, BP_GET_LEVEL(db->db_blkptr) - 1); BP_SET_BIRTH(bp, db->db_blkptr->blk_birth, 0); } } DB_DNODE_EXIT(db); db->db_state = DB_CACHED; mutex_exit(&db->db_mtx); return; } DB_DNODE_EXIT(db); db->db_state = DB_READ; mutex_exit(&db->db_mtx); if (DBUF_IS_L2CACHEABLE(db)) aflags |= ARC_FLAG_L2CACHE; SET_BOOKMARK(&zb, db->db_objset->os_dsl_dataset ? db->db_objset->os_dsl_dataset->ds_object : DMU_META_OBJSET, db->db.db_object, db->db_level, db->db_blkid); dbuf_add_ref(db, NULL); (void) arc_read(zio, db->db_objset->os_spa, db->db_blkptr, dbuf_read_done, db, ZIO_PRIORITY_SYNC_READ, (flags & DB_RF_CANFAIL) ? ZIO_FLAG_CANFAIL : ZIO_FLAG_MUSTSUCCEED, &aflags, &zb); } /* * This is our just-in-time copy function. It makes a copy of buffers that * have been modified in a previous transaction group before we access them in * the current active group. * * This function is used in three places: when we are dirtying a buffer for the * first time in a txg, when we are freeing a range in a dnode that includes * this buffer, and when we are accessing a buffer which was received compressed * and later referenced in a WRITE_BYREF record. * * Note that when we are called from dbuf_free_range() we do not put a hold on * the buffer, we just traverse the active dbuf list for the dnode. */ static void dbuf_fix_old_data(dmu_buf_impl_t *db, uint64_t txg) { dbuf_dirty_record_t *dr = db->db_last_dirty; ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(db->db.db_data != NULL); ASSERT(db->db_level == 0); ASSERT(db->db.db_object != DMU_META_DNODE_OBJECT); if (dr == NULL || (dr->dt.dl.dr_data != ((db->db_blkid == DMU_BONUS_BLKID) ? db->db.db_data : db->db_buf))) return; /* * If the last dirty record for this dbuf has not yet synced * and its referencing the dbuf data, either: * reset the reference to point to a new copy, * or (if there a no active holders) * just null out the current db_data pointer. */ ASSERT(dr->dr_txg >= txg - 2); if (db->db_blkid == DMU_BONUS_BLKID) { /* Note that the data bufs here are zio_bufs */ dr->dt.dl.dr_data = zio_buf_alloc(DN_MAX_BONUSLEN); arc_space_consume(DN_MAX_BONUSLEN, ARC_SPACE_OTHER); bcopy(db->db.db_data, dr->dt.dl.dr_data, DN_MAX_BONUSLEN); } else if (refcount_count(&db->db_holds) > db->db_dirtycnt) { int size = arc_buf_size(db->db_buf); arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db); spa_t *spa = db->db_objset->os_spa; enum zio_compress compress_type = arc_get_compression(db->db_buf); if (compress_type == ZIO_COMPRESS_OFF) { dr->dt.dl.dr_data = arc_alloc_buf(spa, db, type, size); } else { ASSERT3U(type, ==, ARC_BUFC_DATA); dr->dt.dl.dr_data = arc_alloc_compressed_buf(spa, db, size, arc_buf_lsize(db->db_buf), compress_type); } bcopy(db->db.db_data, dr->dt.dl.dr_data->b_data, size); } else { db->db_buf = NULL; dbuf_clear_data(db); } } int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags) { int err = 0; boolean_t havepzio = (zio != NULL); boolean_t prefetch; dnode_t *dn; /* * We don't have to hold the mutex to check db_state because it * can't be freed while we have a hold on the buffer. */ ASSERT(!refcount_is_zero(&db->db_holds)); if (db->db_state == DB_NOFILL) return (SET_ERROR(EIO)); DB_DNODE_ENTER(db); dn = DB_DNODE(db); if ((flags & DB_RF_HAVESTRUCT) == 0) rw_enter(&dn->dn_struct_rwlock, RW_READER); prefetch = db->db_level == 0 && db->db_blkid != DMU_BONUS_BLKID && (flags & DB_RF_NOPREFETCH) == 0 && dn != NULL && DBUF_IS_CACHEABLE(db); mutex_enter(&db->db_mtx); if (db->db_state == DB_CACHED) { /* * If the arc buf is compressed, we need to decompress it to * read the data. This could happen during the "zfs receive" of * a stream which is compressed and deduplicated. */ if (db->db_buf != NULL && arc_get_compression(db->db_buf) != ZIO_COMPRESS_OFF) { dbuf_fix_old_data(db, spa_syncing_txg(dmu_objset_spa(db->db_objset))); err = arc_decompress(db->db_buf); dbuf_set_data(db, db->db_buf); } mutex_exit(&db->db_mtx); if (prefetch) dmu_zfetch(&dn->dn_zfetch, db->db_blkid, 1, B_TRUE); if ((flags & DB_RF_HAVESTRUCT) == 0) rw_exit(&dn->dn_struct_rwlock); DB_DNODE_EXIT(db); } else if (db->db_state == DB_UNCACHED) { spa_t *spa = dn->dn_objset->os_spa; if (zio == NULL) zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL); dbuf_read_impl(db, zio, flags); /* dbuf_read_impl has dropped db_mtx for us */ if (prefetch) dmu_zfetch(&dn->dn_zfetch, db->db_blkid, 1, B_TRUE); if ((flags & DB_RF_HAVESTRUCT) == 0) rw_exit(&dn->dn_struct_rwlock); DB_DNODE_EXIT(db); if (!havepzio) err = zio_wait(zio); } else { /* * Another reader came in while the dbuf was in flight * between UNCACHED and CACHED. Either a writer will finish * writing the buffer (sending the dbuf to CACHED) or the * first reader's request will reach the read_done callback * and send the dbuf to CACHED. Otherwise, a failure * occurred and the dbuf went to UNCACHED. */ mutex_exit(&db->db_mtx); if (prefetch) dmu_zfetch(&dn->dn_zfetch, db->db_blkid, 1, B_TRUE); if ((flags & DB_RF_HAVESTRUCT) == 0) rw_exit(&dn->dn_struct_rwlock); DB_DNODE_EXIT(db); /* Skip the wait per the caller's request. */ mutex_enter(&db->db_mtx); if ((flags & DB_RF_NEVERWAIT) == 0) { while (db->db_state == DB_READ || db->db_state == DB_FILL) { ASSERT(db->db_state == DB_READ || (flags & DB_RF_HAVESTRUCT) == 0); DTRACE_PROBE2(blocked__read, dmu_buf_impl_t *, db, zio_t *, zio); cv_wait(&db->db_changed, &db->db_mtx); } if (db->db_state == DB_UNCACHED) err = SET_ERROR(EIO); } mutex_exit(&db->db_mtx); } ASSERT(err || havepzio || db->db_state == DB_CACHED); return (err); } static void dbuf_noread(dmu_buf_impl_t *db) { ASSERT(!refcount_is_zero(&db->db_holds)); ASSERT(db->db_blkid != DMU_BONUS_BLKID); mutex_enter(&db->db_mtx); while (db->db_state == DB_READ || db->db_state == DB_FILL) cv_wait(&db->db_changed, &db->db_mtx); if (db->db_state == DB_UNCACHED) { arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db); spa_t *spa = db->db_objset->os_spa; ASSERT(db->db_buf == NULL); ASSERT(db->db.db_data == NULL); dbuf_set_data(db, arc_alloc_buf(spa, db, type, db->db.db_size)); db->db_state = DB_FILL; } else if (db->db_state == DB_NOFILL) { dbuf_clear_data(db); } else { ASSERT3U(db->db_state, ==, DB_CACHED); } mutex_exit(&db->db_mtx); } void dbuf_unoverride(dbuf_dirty_record_t *dr) { dmu_buf_impl_t *db = dr->dr_dbuf; blkptr_t *bp = &dr->dt.dl.dr_overridden_by; uint64_t txg = dr->dr_txg; ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(dr->dt.dl.dr_override_state != DR_IN_DMU_SYNC); ASSERT(db->db_level == 0); if (db->db_blkid == DMU_BONUS_BLKID || dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN) return; ASSERT(db->db_data_pending != dr); /* free this block */ if (!BP_IS_HOLE(bp) && !dr->dt.dl.dr_nopwrite) zio_free(db->db_objset->os_spa, txg, bp); dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; dr->dt.dl.dr_nopwrite = B_FALSE; /* * Release the already-written buffer, so we leave it in * a consistent dirty state. Note that all callers are * modifying the buffer, so they will immediately do * another (redundant) arc_release(). Therefore, leave * the buf thawed to save the effort of freezing & * immediately re-thawing it. */ arc_release(dr->dt.dl.dr_data, db); } /* * Evict (if its unreferenced) or clear (if its referenced) any level-0 * data blocks in the free range, so that any future readers will find * empty blocks. */ void dbuf_free_range(dnode_t *dn, uint64_t start_blkid, uint64_t end_blkid, dmu_tx_t *tx) { dmu_buf_impl_t db_search; dmu_buf_impl_t *db, *db_next; uint64_t txg = tx->tx_txg; avl_index_t where; if (end_blkid > dn->dn_maxblkid && !(start_blkid == DMU_SPILL_BLKID || end_blkid == DMU_SPILL_BLKID)) end_blkid = dn->dn_maxblkid; dprintf_dnode(dn, "start=%llu end=%llu\n", start_blkid, end_blkid); db_search.db_level = 0; db_search.db_blkid = start_blkid; db_search.db_state = DB_SEARCH; mutex_enter(&dn->dn_dbufs_mtx); db = avl_find(&dn->dn_dbufs, &db_search, &where); ASSERT3P(db, ==, NULL); db = avl_nearest(&dn->dn_dbufs, where, AVL_AFTER); for (; db != NULL; db = db_next) { db_next = AVL_NEXT(&dn->dn_dbufs, db); ASSERT(db->db_blkid != DMU_BONUS_BLKID); if (db->db_level != 0 || db->db_blkid > end_blkid) { break; } ASSERT3U(db->db_blkid, >=, start_blkid); /* found a level 0 buffer in the range */ mutex_enter(&db->db_mtx); if (dbuf_undirty(db, tx)) { /* mutex has been dropped and dbuf destroyed */ continue; } if (db->db_state == DB_UNCACHED || db->db_state == DB_NOFILL || db->db_state == DB_EVICTING) { ASSERT(db->db.db_data == NULL); mutex_exit(&db->db_mtx); continue; } if (db->db_state == DB_READ || db->db_state == DB_FILL) { /* will be handled in dbuf_read_done or dbuf_rele */ db->db_freed_in_flight = TRUE; mutex_exit(&db->db_mtx); continue; } if (refcount_count(&db->db_holds) == 0) { ASSERT(db->db_buf); dbuf_destroy(db); continue; } /* The dbuf is referenced */ if (db->db_last_dirty != NULL) { dbuf_dirty_record_t *dr = db->db_last_dirty; if (dr->dr_txg == txg) { /* * This buffer is "in-use", re-adjust the file * size to reflect that this buffer may * contain new data when we sync. */ if (db->db_blkid != DMU_SPILL_BLKID && db->db_blkid > dn->dn_maxblkid) dn->dn_maxblkid = db->db_blkid; dbuf_unoverride(dr); } else { /* * This dbuf is not dirty in the open context. * Either uncache it (if its not referenced in * the open context) or reset its contents to * empty. */ dbuf_fix_old_data(db, txg); } } /* clear the contents if its cached */ if (db->db_state == DB_CACHED) { ASSERT(db->db.db_data != NULL); arc_release(db->db_buf, db); bzero(db->db.db_data, db->db.db_size); arc_buf_freeze(db->db_buf); } mutex_exit(&db->db_mtx); } mutex_exit(&dn->dn_dbufs_mtx); } void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx) { arc_buf_t *buf, *obuf; int osize = db->db.db_size; arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db); dnode_t *dn; ASSERT(db->db_blkid != DMU_BONUS_BLKID); DB_DNODE_ENTER(db); dn = DB_DNODE(db); /* XXX does *this* func really need the lock? */ ASSERT(RW_WRITE_HELD(&dn->dn_struct_rwlock)); /* * This call to dmu_buf_will_dirty() with the dn_struct_rwlock held * is OK, because there can be no other references to the db * when we are changing its size, so no concurrent DB_FILL can * be happening. */ /* * XXX we should be doing a dbuf_read, checking the return * value and returning that up to our callers */ dmu_buf_will_dirty(&db->db, tx); /* create the data buffer for the new block */ buf = arc_alloc_buf(dn->dn_objset->os_spa, db, type, size); /* copy old block data to the new block */ obuf = db->db_buf; bcopy(obuf->b_data, buf->b_data, MIN(osize, size)); /* zero the remainder */ if (size > osize) bzero((uint8_t *)buf->b_data + osize, size - osize); mutex_enter(&db->db_mtx); dbuf_set_data(db, buf); arc_buf_destroy(obuf, db); db->db.db_size = size; if (db->db_level == 0) { ASSERT3U(db->db_last_dirty->dr_txg, ==, tx->tx_txg); db->db_last_dirty->dt.dl.dr_data = buf; } mutex_exit(&db->db_mtx); dmu_objset_willuse_space(dn->dn_objset, size - osize, tx); DB_DNODE_EXIT(db); } void dbuf_release_bp(dmu_buf_impl_t *db) { objset_t *os = db->db_objset; ASSERT(dsl_pool_sync_context(dmu_objset_pool(os))); ASSERT(arc_released(os->os_phys_buf) || list_link_active(&os->os_dsl_dataset->ds_synced_link)); ASSERT(db->db_parent == NULL || arc_released(db->db_parent->db_buf)); (void) arc_release(db->db_buf, db); } /* * We already have a dirty record for this TXG, and we are being * dirtied again. */ static void dbuf_redirty(dbuf_dirty_record_t *dr) { dmu_buf_impl_t *db = dr->dr_dbuf; ASSERT(MUTEX_HELD(&db->db_mtx)); if (db->db_level == 0 && db->db_blkid != DMU_BONUS_BLKID) { /* * If this buffer has already been written out, * we now need to reset its state. */ dbuf_unoverride(dr); if (db->db.db_object != DMU_META_DNODE_OBJECT && db->db_state != DB_NOFILL) { /* Already released on initial dirty, so just thaw. */ ASSERT(arc_released(db->db_buf)); arc_buf_thaw(db->db_buf); } } } dbuf_dirty_record_t * dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx) { dnode_t *dn; objset_t *os; dbuf_dirty_record_t **drp, *dr; int drop_struct_lock = FALSE; int txgoff = tx->tx_txg & TXG_MASK; ASSERT(tx->tx_txg != 0); ASSERT(!refcount_is_zero(&db->db_holds)); DMU_TX_DIRTY_BUF(tx, db); DB_DNODE_ENTER(db); dn = DB_DNODE(db); /* * Shouldn't dirty a regular buffer in syncing context. Private * objects may be dirtied in syncing context, but only if they * were already pre-dirtied in open context. */ #ifdef DEBUG if (dn->dn_objset->os_dsl_dataset != NULL) { rrw_enter(&dn->dn_objset->os_dsl_dataset->ds_bp_rwlock, RW_READER, FTAG); } ASSERT(!dmu_tx_is_syncing(tx) || BP_IS_HOLE(dn->dn_objset->os_rootbp) || DMU_OBJECT_IS_SPECIAL(dn->dn_object) || dn->dn_objset->os_dsl_dataset == NULL); if (dn->dn_objset->os_dsl_dataset != NULL) rrw_exit(&dn->dn_objset->os_dsl_dataset->ds_bp_rwlock, FTAG); #endif /* * We make this assert for private objects as well, but after we * check if we're already dirty. They are allowed to re-dirty * in syncing context. */ ASSERT(dn->dn_object == DMU_META_DNODE_OBJECT || dn->dn_dirtyctx == DN_UNDIRTIED || dn->dn_dirtyctx == (dmu_tx_is_syncing(tx) ? DN_DIRTY_SYNC : DN_DIRTY_OPEN)); mutex_enter(&db->db_mtx); /* * XXX make this true for indirects too? The problem is that * transactions created with dmu_tx_create_assigned() from * syncing context don't bother holding ahead. */ ASSERT(db->db_level != 0 || db->db_state == DB_CACHED || db->db_state == DB_FILL || db->db_state == DB_NOFILL); mutex_enter(&dn->dn_mtx); /* * Don't set dirtyctx to SYNC if we're just modifying this as we * initialize the objset. */ if (dn->dn_dirtyctx == DN_UNDIRTIED) { if (dn->dn_objset->os_dsl_dataset != NULL) { rrw_enter(&dn->dn_objset->os_dsl_dataset->ds_bp_rwlock, RW_READER, FTAG); } if (!BP_IS_HOLE(dn->dn_objset->os_rootbp)) { dn->dn_dirtyctx = (dmu_tx_is_syncing(tx) ? DN_DIRTY_SYNC : DN_DIRTY_OPEN); ASSERT(dn->dn_dirtyctx_firstset == NULL); dn->dn_dirtyctx_firstset = kmem_alloc(1, KM_SLEEP); } if (dn->dn_objset->os_dsl_dataset != NULL) { rrw_exit(&dn->dn_objset->os_dsl_dataset->ds_bp_rwlock, FTAG); } } mutex_exit(&dn->dn_mtx); if (db->db_blkid == DMU_SPILL_BLKID) dn->dn_have_spill = B_TRUE; /* * If this buffer is already dirty, we're done. */ drp = &db->db_last_dirty; ASSERT(*drp == NULL || (*drp)->dr_txg <= tx->tx_txg || db->db.db_object == DMU_META_DNODE_OBJECT); while ((dr = *drp) != NULL && dr->dr_txg > tx->tx_txg) drp = &dr->dr_next; if (dr && dr->dr_txg == tx->tx_txg) { DB_DNODE_EXIT(db); dbuf_redirty(dr); mutex_exit(&db->db_mtx); return (dr); } /* * Only valid if not already dirty. */ ASSERT(dn->dn_object == 0 || dn->dn_dirtyctx == DN_UNDIRTIED || dn->dn_dirtyctx == (dmu_tx_is_syncing(tx) ? DN_DIRTY_SYNC : DN_DIRTY_OPEN)); ASSERT3U(dn->dn_nlevels, >, db->db_level); ASSERT((dn->dn_phys->dn_nlevels == 0 && db->db_level == 0) || dn->dn_phys->dn_nlevels > db->db_level || dn->dn_next_nlevels[txgoff] > db->db_level || dn->dn_next_nlevels[(tx->tx_txg-1) & TXG_MASK] > db->db_level || dn->dn_next_nlevels[(tx->tx_txg-2) & TXG_MASK] > db->db_level); /* * We should only be dirtying in syncing context if it's the * mos or we're initializing the os or it's a special object. * However, we are allowed to dirty in syncing context provided * we already dirtied it in open context. Hence we must make * this assertion only if we're not already dirty. */ os = dn->dn_objset; #ifdef DEBUG if (dn->dn_objset->os_dsl_dataset != NULL) rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_READER, FTAG); ASSERT(!dmu_tx_is_syncing(tx) || DMU_OBJECT_IS_SPECIAL(dn->dn_object) || os->os_dsl_dataset == NULL || BP_IS_HOLE(os->os_rootbp)); if (dn->dn_objset->os_dsl_dataset != NULL) rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG); #endif ASSERT(db->db.db_size != 0); dprintf_dbuf(db, "size=%llx\n", (u_longlong_t)db->db.db_size); if (db->db_blkid != DMU_BONUS_BLKID) { dmu_objset_willuse_space(os, db->db.db_size, tx); } /* * If this buffer is dirty in an old transaction group we need * to make a copy of it so that the changes we make in this * transaction group won't leak out when we sync the older txg. */ dr = kmem_zalloc(sizeof (dbuf_dirty_record_t), KM_SLEEP); if (db->db_level == 0) { void *data_old = db->db_buf; if (db->db_state != DB_NOFILL) { if (db->db_blkid == DMU_BONUS_BLKID) { dbuf_fix_old_data(db, tx->tx_txg); data_old = db->db.db_data; } else if (db->db.db_object != DMU_META_DNODE_OBJECT) { /* * Release the data buffer from the cache so * that we can modify it without impacting * possible other users of this cached data * block. Note that indirect blocks and * private objects are not released until the * syncing state (since they are only modified * then). */ arc_release(db->db_buf, db); dbuf_fix_old_data(db, tx->tx_txg); data_old = db->db_buf; } ASSERT(data_old != NULL); } dr->dt.dl.dr_data = data_old; } else { mutex_init(&dr->dt.di.dr_mtx, NULL, MUTEX_DEFAULT, NULL); list_create(&dr->dt.di.dr_children, sizeof (dbuf_dirty_record_t), offsetof(dbuf_dirty_record_t, dr_dirty_node)); } if (db->db_blkid != DMU_BONUS_BLKID && os->os_dsl_dataset != NULL) dr->dr_accounted = db->db.db_size; dr->dr_dbuf = db; dr->dr_txg = tx->tx_txg; dr->dr_next = *drp; *drp = dr; /* * We could have been freed_in_flight between the dbuf_noread * and dbuf_dirty. We win, as though the dbuf_noread() had * happened after the free. */ if (db->db_level == 0 && db->db_blkid != DMU_BONUS_BLKID && db->db_blkid != DMU_SPILL_BLKID) { mutex_enter(&dn->dn_mtx); if (dn->dn_free_ranges[txgoff] != NULL) { range_tree_clear(dn->dn_free_ranges[txgoff], db->db_blkid, 1); } mutex_exit(&dn->dn_mtx); db->db_freed_in_flight = FALSE; } /* * This buffer is now part of this txg */ dbuf_add_ref(db, (void *)(uintptr_t)tx->tx_txg); db->db_dirtycnt += 1; ASSERT3U(db->db_dirtycnt, <=, 3); mutex_exit(&db->db_mtx); if (db->db_blkid == DMU_BONUS_BLKID || db->db_blkid == DMU_SPILL_BLKID) { mutex_enter(&dn->dn_mtx); ASSERT(!list_link_active(&dr->dr_dirty_node)); list_insert_tail(&dn->dn_dirty_records[txgoff], dr); mutex_exit(&dn->dn_mtx); dnode_setdirty(dn, tx); DB_DNODE_EXIT(db); return (dr); } /* * The dn_struct_rwlock prevents db_blkptr from changing * due to a write from syncing context completing * while we are running, so we want to acquire it before * looking at db_blkptr. */ if (!RW_WRITE_HELD(&dn->dn_struct_rwlock)) { rw_enter(&dn->dn_struct_rwlock, RW_READER); drop_struct_lock = TRUE; } /* * If we are overwriting a dedup BP, then unless it is snapshotted, * when we get to syncing context we will need to decrement its * refcount in the DDT. Prefetch the relevant DDT block so that * syncing context won't have to wait for the i/o. */ ddt_prefetch(os->os_spa, db->db_blkptr); if (db->db_level == 0) { dnode_new_blkid(dn, db->db_blkid, tx, drop_struct_lock); ASSERT(dn->dn_maxblkid >= db->db_blkid); } if (db->db_level+1 < dn->dn_nlevels) { dmu_buf_impl_t *parent = db->db_parent; dbuf_dirty_record_t *di; int parent_held = FALSE; if (db->db_parent == NULL || db->db_parent == dn->dn_dbuf) { int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; parent = dbuf_hold_level(dn, db->db_level+1, db->db_blkid >> epbs, FTAG); ASSERT(parent != NULL); parent_held = TRUE; } if (drop_struct_lock) rw_exit(&dn->dn_struct_rwlock); ASSERT3U(db->db_level+1, ==, parent->db_level); di = dbuf_dirty(parent, tx); if (parent_held) dbuf_rele(parent, FTAG); mutex_enter(&db->db_mtx); /* * Since we've dropped the mutex, it's possible that * dbuf_undirty() might have changed this out from under us. */ if (db->db_last_dirty == dr || dn->dn_object == DMU_META_DNODE_OBJECT) { mutex_enter(&di->dt.di.dr_mtx); ASSERT3U(di->dr_txg, ==, tx->tx_txg); ASSERT(!list_link_active(&dr->dr_dirty_node)); list_insert_tail(&di->dt.di.dr_children, dr); mutex_exit(&di->dt.di.dr_mtx); dr->dr_parent = di; } mutex_exit(&db->db_mtx); } else { ASSERT(db->db_level+1 == dn->dn_nlevels); ASSERT(db->db_blkid < dn->dn_nblkptr); ASSERT(db->db_parent == NULL || db->db_parent == dn->dn_dbuf); mutex_enter(&dn->dn_mtx); ASSERT(!list_link_active(&dr->dr_dirty_node)); list_insert_tail(&dn->dn_dirty_records[txgoff], dr); mutex_exit(&dn->dn_mtx); if (drop_struct_lock) rw_exit(&dn->dn_struct_rwlock); } dnode_setdirty(dn, tx); DB_DNODE_EXIT(db); return (dr); } /* * Undirty a buffer in the transaction group referenced by the given * transaction. Return whether this evicted the dbuf. */ static boolean_t dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx) { dnode_t *dn; uint64_t txg = tx->tx_txg; dbuf_dirty_record_t *dr, **drp; ASSERT(txg != 0); /* * Due to our use of dn_nlevels below, this can only be called * in open context, unless we are operating on the MOS. * From syncing context, dn_nlevels may be different from the * dn_nlevels used when dbuf was dirtied. */ ASSERT(db->db_objset == dmu_objset_pool(db->db_objset)->dp_meta_objset || txg != spa_syncing_txg(dmu_objset_spa(db->db_objset))); ASSERT(db->db_blkid != DMU_BONUS_BLKID); ASSERT0(db->db_level); ASSERT(MUTEX_HELD(&db->db_mtx)); /* * If this buffer is not dirty, we're done. */ for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next) if (dr->dr_txg <= txg) break; if (dr == NULL || dr->dr_txg < txg) return (B_FALSE); ASSERT(dr->dr_txg == txg); ASSERT(dr->dr_dbuf == db); DB_DNODE_ENTER(db); dn = DB_DNODE(db); dprintf_dbuf(db, "size=%llx\n", (u_longlong_t)db->db.db_size); ASSERT(db->db.db_size != 0); dsl_pool_undirty_space(dmu_objset_pool(dn->dn_objset), dr->dr_accounted, txg); *drp = dr->dr_next; /* * Note that there are three places in dbuf_dirty() * where this dirty record may be put on a list. * Make sure to do a list_remove corresponding to * every one of those list_insert calls. */ if (dr->dr_parent) { mutex_enter(&dr->dr_parent->dt.di.dr_mtx); list_remove(&dr->dr_parent->dt.di.dr_children, dr); mutex_exit(&dr->dr_parent->dt.di.dr_mtx); } else if (db->db_blkid == DMU_SPILL_BLKID || db->db_level + 1 == dn->dn_nlevels) { ASSERT(db->db_blkptr == NULL || db->db_parent == dn->dn_dbuf); mutex_enter(&dn->dn_mtx); list_remove(&dn->dn_dirty_records[txg & TXG_MASK], dr); mutex_exit(&dn->dn_mtx); } DB_DNODE_EXIT(db); if (db->db_state != DB_NOFILL) { dbuf_unoverride(dr); ASSERT(db->db_buf != NULL); ASSERT(dr->dt.dl.dr_data != NULL); if (dr->dt.dl.dr_data != db->db_buf) arc_buf_destroy(dr->dt.dl.dr_data, db); } kmem_free(dr, sizeof (dbuf_dirty_record_t)); ASSERT(db->db_dirtycnt > 0); db->db_dirtycnt -= 1; if (refcount_remove(&db->db_holds, (void *)(uintptr_t)txg) == 0) { ASSERT(db->db_state == DB_NOFILL || arc_released(db->db_buf)); dbuf_destroy(db); return (B_TRUE); } return (B_FALSE); } void dmu_buf_will_dirty(dmu_buf_t *db_fake, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; int rf = DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH; ASSERT(tx->tx_txg != 0); ASSERT(!refcount_is_zero(&db->db_holds)); /* * Quick check for dirtyness. For already dirty blocks, this * reduces runtime of this function by >90%, and overall performance * by 50% for some workloads (e.g. file deletion with indirect blocks * cached). */ mutex_enter(&db->db_mtx); dbuf_dirty_record_t *dr; for (dr = db->db_last_dirty; dr != NULL && dr->dr_txg >= tx->tx_txg; dr = dr->dr_next) { /* * It's possible that it is already dirty but not cached, * because there are some calls to dbuf_dirty() that don't * go through dmu_buf_will_dirty(). */ if (dr->dr_txg == tx->tx_txg && db->db_state == DB_CACHED) { /* This dbuf is already dirty and cached. */ dbuf_redirty(dr); mutex_exit(&db->db_mtx); return; } } mutex_exit(&db->db_mtx); DB_DNODE_ENTER(db); if (RW_WRITE_HELD(&DB_DNODE(db)->dn_struct_rwlock)) rf |= DB_RF_HAVESTRUCT; DB_DNODE_EXIT(db); (void) dbuf_read(db, NULL, rf); (void) dbuf_dirty(db, tx); } void dmu_buf_will_not_fill(dmu_buf_t *db_fake, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; db->db_state = DB_NOFILL; dmu_buf_will_fill(db_fake, tx); } void dmu_buf_will_fill(dmu_buf_t *db_fake, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; ASSERT(db->db_blkid != DMU_BONUS_BLKID); ASSERT(tx->tx_txg != 0); ASSERT(db->db_level == 0); ASSERT(!refcount_is_zero(&db->db_holds)); ASSERT(db->db.db_object != DMU_META_DNODE_OBJECT || dmu_tx_private_ok(tx)); dbuf_noread(db); (void) dbuf_dirty(db, tx); } #pragma weak dmu_buf_fill_done = dbuf_fill_done /* ARGSUSED */ void dbuf_fill_done(dmu_buf_impl_t *db, dmu_tx_t *tx) { mutex_enter(&db->db_mtx); DBUF_VERIFY(db); if (db->db_state == DB_FILL) { if (db->db_level == 0 && db->db_freed_in_flight) { ASSERT(db->db_blkid != DMU_BONUS_BLKID); /* we were freed while filling */ /* XXX dbuf_undirty? */ bzero(db->db.db_data, db->db.db_size); db->db_freed_in_flight = FALSE; } db->db_state = DB_CACHED; cv_broadcast(&db->db_changed); } mutex_exit(&db->db_mtx); } void dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data, bp_embedded_type_t etype, enum zio_compress comp, int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbuf; struct dirty_leaf *dl; dmu_object_type_t type; if (etype == BP_EMBEDDED_TYPE_DATA) { ASSERT(spa_feature_is_active(dmu_objset_spa(db->db_objset), SPA_FEATURE_EMBEDDED_DATA)); } DB_DNODE_ENTER(db); type = DB_DNODE(db)->dn_type; DB_DNODE_EXIT(db); ASSERT0(db->db_level); ASSERT(db->db_blkid != DMU_BONUS_BLKID); dmu_buf_will_not_fill(dbuf, tx); ASSERT3U(db->db_last_dirty->dr_txg, ==, tx->tx_txg); dl = &db->db_last_dirty->dt.dl; encode_embedded_bp_compressed(&dl->dr_overridden_by, data, comp, uncompressed_size, compressed_size); BPE_SET_ETYPE(&dl->dr_overridden_by, etype); BP_SET_TYPE(&dl->dr_overridden_by, type); BP_SET_LEVEL(&dl->dr_overridden_by, 0); BP_SET_BYTEORDER(&dl->dr_overridden_by, byteorder); dl->dr_override_state = DR_OVERRIDDEN; dl->dr_overridden_by.blk_birth = db->db_last_dirty->dr_txg; } /* * Directly assign a provided arc buf to a given dbuf if it's not referenced * by anybody except our caller. Otherwise copy arcbuf's contents to dbuf. */ void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx) { ASSERT(!refcount_is_zero(&db->db_holds)); ASSERT(db->db_blkid != DMU_BONUS_BLKID); ASSERT(db->db_level == 0); ASSERT3U(dbuf_is_metadata(db), ==, arc_is_metadata(buf)); ASSERT(buf != NULL); ASSERT(arc_buf_lsize(buf) == db->db.db_size); ASSERT(tx->tx_txg != 0); arc_return_buf(buf, db); ASSERT(arc_released(buf)); mutex_enter(&db->db_mtx); while (db->db_state == DB_READ || db->db_state == DB_FILL) cv_wait(&db->db_changed, &db->db_mtx); ASSERT(db->db_state == DB_CACHED || db->db_state == DB_UNCACHED); if (db->db_state == DB_CACHED && refcount_count(&db->db_holds) - 1 > db->db_dirtycnt) { mutex_exit(&db->db_mtx); (void) dbuf_dirty(db, tx); bcopy(buf->b_data, db->db.db_data, db->db.db_size); arc_buf_destroy(buf, db); xuio_stat_wbuf_copied(); return; } xuio_stat_wbuf_nocopy(); if (db->db_state == DB_CACHED) { dbuf_dirty_record_t *dr = db->db_last_dirty; ASSERT(db->db_buf != NULL); if (dr != NULL && dr->dr_txg == tx->tx_txg) { ASSERT(dr->dt.dl.dr_data == db->db_buf); if (!arc_released(db->db_buf)) { ASSERT(dr->dt.dl.dr_override_state == DR_OVERRIDDEN); arc_release(db->db_buf, db); } dr->dt.dl.dr_data = buf; arc_buf_destroy(db->db_buf, db); } else if (dr == NULL || dr->dt.dl.dr_data != db->db_buf) { arc_release(db->db_buf, db); arc_buf_destroy(db->db_buf, db); } db->db_buf = NULL; } ASSERT(db->db_buf == NULL); dbuf_set_data(db, buf); db->db_state = DB_FILL; mutex_exit(&db->db_mtx); (void) dbuf_dirty(db, tx); dmu_buf_fill_done(&db->db, tx); } void dbuf_destroy(dmu_buf_impl_t *db) { dnode_t *dn; dmu_buf_impl_t *parent = db->db_parent; dmu_buf_impl_t *dndb; ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(refcount_is_zero(&db->db_holds)); if (db->db_buf != NULL) { arc_buf_destroy(db->db_buf, db); db->db_buf = NULL; } if (db->db_blkid == DMU_BONUS_BLKID) { ASSERT(db->db.db_data != NULL); zio_buf_free(db->db.db_data, DN_MAX_BONUSLEN); arc_space_return(DN_MAX_BONUSLEN, ARC_SPACE_OTHER); db->db_state = DB_UNCACHED; } dbuf_clear_data(db); if (multilist_link_active(&db->db_cache_link)) { - multilist_remove(&dbuf_cache, db); + multilist_remove(dbuf_cache, db); (void) refcount_remove_many(&dbuf_cache_size, db->db.db_size, db); } ASSERT(db->db_state == DB_UNCACHED || db->db_state == DB_NOFILL); ASSERT(db->db_data_pending == NULL); db->db_state = DB_EVICTING; db->db_blkptr = NULL; /* * Now that db_state is DB_EVICTING, nobody else can find this via * the hash table. We can now drop db_mtx, which allows us to * acquire the dn_dbufs_mtx. */ mutex_exit(&db->db_mtx); DB_DNODE_ENTER(db); dn = DB_DNODE(db); dndb = dn->dn_dbuf; if (db->db_blkid != DMU_BONUS_BLKID) { boolean_t needlock = !MUTEX_HELD(&dn->dn_dbufs_mtx); if (needlock) mutex_enter(&dn->dn_dbufs_mtx); avl_remove(&dn->dn_dbufs, db); atomic_dec_32(&dn->dn_dbufs_count); membar_producer(); DB_DNODE_EXIT(db); if (needlock) mutex_exit(&dn->dn_dbufs_mtx); /* * Decrementing the dbuf count means that the hold corresponding * to the removed dbuf is no longer discounted in dnode_move(), * so the dnode cannot be moved until after we release the hold. * The membar_producer() ensures visibility of the decremented * value in dnode_move(), since DB_DNODE_EXIT doesn't actually * release any lock. */ dnode_rele(dn, db); db->db_dnode_handle = NULL; dbuf_hash_remove(db); } else { DB_DNODE_EXIT(db); } ASSERT(refcount_is_zero(&db->db_holds)); db->db_parent = NULL; ASSERT(db->db_buf == NULL); ASSERT(db->db.db_data == NULL); ASSERT(db->db_hash_next == NULL); ASSERT(db->db_blkptr == NULL); ASSERT(db->db_data_pending == NULL); ASSERT(!multilist_link_active(&db->db_cache_link)); kmem_cache_free(dbuf_kmem_cache, db); arc_space_return(sizeof (dmu_buf_impl_t), ARC_SPACE_OTHER); /* * If this dbuf is referenced from an indirect dbuf, * decrement the ref count on the indirect dbuf. */ if (parent && parent != dndb) dbuf_rele(parent, db); } /* * Note: While bpp will always be updated if the function returns success, * parentp will not be updated if the dnode does not have dn_dbuf filled in; * this happens when the dnode is the meta-dnode, or a userused or groupused * object. */ static int dbuf_findbp(dnode_t *dn, int level, uint64_t blkid, int fail_sparse, dmu_buf_impl_t **parentp, blkptr_t **bpp) { *parentp = NULL; *bpp = NULL; ASSERT(blkid != DMU_BONUS_BLKID); if (blkid == DMU_SPILL_BLKID) { mutex_enter(&dn->dn_mtx); if (dn->dn_have_spill && (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR)) *bpp = &dn->dn_phys->dn_spill; else *bpp = NULL; dbuf_add_ref(dn->dn_dbuf, NULL); *parentp = dn->dn_dbuf; mutex_exit(&dn->dn_mtx); return (0); } int nlevels = (dn->dn_phys->dn_nlevels == 0) ? 1 : dn->dn_phys->dn_nlevels; int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; ASSERT3U(level * epbs, <, 64); ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock)); /* * This assertion shouldn't trip as long as the max indirect block size * is less than 1M. The reason for this is that up to that point, * the number of levels required to address an entire object with blocks * of size SPA_MINBLOCKSIZE satisfies nlevels * epbs + 1 <= 64. In * other words, if N * epbs + 1 > 64, then if (N-1) * epbs + 1 > 55 * (i.e. we can address the entire object), objects will all use at most * N-1 levels and the assertion won't overflow. However, once epbs is * 13, 4 * 13 + 1 = 53, but 5 * 13 + 1 = 66. Then, 4 levels will not be * enough to address an entire object, so objects will have 5 levels, * but then this assertion will overflow. * * All this is to say that if we ever increase DN_MAX_INDBLKSHIFT, we * need to redo this logic to handle overflows. */ ASSERT(level >= nlevels || ((nlevels - level - 1) * epbs) + highbit64(dn->dn_phys->dn_nblkptr) <= 64); if (level >= nlevels || blkid >= ((uint64_t)dn->dn_phys->dn_nblkptr << ((nlevels - level - 1) * epbs)) || (fail_sparse && blkid > (dn->dn_phys->dn_maxblkid >> (level * epbs)))) { /* the buffer has no parent yet */ return (SET_ERROR(ENOENT)); } else if (level < nlevels-1) { /* this block is referenced from an indirect block */ int err = dbuf_hold_impl(dn, level+1, blkid >> epbs, fail_sparse, FALSE, NULL, parentp); if (err) return (err); err = dbuf_read(*parentp, NULL, (DB_RF_HAVESTRUCT | DB_RF_NOPREFETCH | DB_RF_CANFAIL)); if (err) { dbuf_rele(*parentp, NULL); *parentp = NULL; return (err); } *bpp = ((blkptr_t *)(*parentp)->db.db_data) + (blkid & ((1ULL << epbs) - 1)); if (blkid > (dn->dn_phys->dn_maxblkid >> (level * epbs))) ASSERT(BP_IS_HOLE(*bpp)); return (0); } else { /* the block is referenced from the dnode */ ASSERT3U(level, ==, nlevels-1); ASSERT(dn->dn_phys->dn_nblkptr == 0 || blkid < dn->dn_phys->dn_nblkptr); if (dn->dn_dbuf) { dbuf_add_ref(dn->dn_dbuf, NULL); *parentp = dn->dn_dbuf; } *bpp = &dn->dn_phys->dn_blkptr[blkid]; return (0); } } static dmu_buf_impl_t * dbuf_create(dnode_t *dn, uint8_t level, uint64_t blkid, dmu_buf_impl_t *parent, blkptr_t *blkptr) { objset_t *os = dn->dn_objset; dmu_buf_impl_t *db, *odb; ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock)); ASSERT(dn->dn_type != DMU_OT_NONE); db = kmem_cache_alloc(dbuf_kmem_cache, KM_SLEEP); db->db_objset = os; db->db.db_object = dn->dn_object; db->db_level = level; db->db_blkid = blkid; db->db_last_dirty = NULL; db->db_dirtycnt = 0; db->db_dnode_handle = dn->dn_handle; db->db_parent = parent; db->db_blkptr = blkptr; db->db_user = NULL; db->db_user_immediate_evict = FALSE; db->db_freed_in_flight = FALSE; db->db_pending_evict = FALSE; if (blkid == DMU_BONUS_BLKID) { ASSERT3P(parent, ==, dn->dn_dbuf); db->db.db_size = DN_MAX_BONUSLEN - (dn->dn_nblkptr-1) * sizeof (blkptr_t); ASSERT3U(db->db.db_size, >=, dn->dn_bonuslen); db->db.db_offset = DMU_BONUS_BLKID; db->db_state = DB_UNCACHED; /* the bonus dbuf is not placed in the hash table */ arc_space_consume(sizeof (dmu_buf_impl_t), ARC_SPACE_OTHER); return (db); } else if (blkid == DMU_SPILL_BLKID) { db->db.db_size = (blkptr != NULL) ? BP_GET_LSIZE(blkptr) : SPA_MINBLOCKSIZE; db->db.db_offset = 0; } else { int blocksize = db->db_level ? 1 << dn->dn_indblkshift : dn->dn_datablksz; db->db.db_size = blocksize; db->db.db_offset = db->db_blkid * blocksize; } /* * Hold the dn_dbufs_mtx while we get the new dbuf * in the hash table *and* added to the dbufs list. * This prevents a possible deadlock with someone * trying to look up this dbuf before its added to the * dn_dbufs list. */ mutex_enter(&dn->dn_dbufs_mtx); db->db_state = DB_EVICTING; if ((odb = dbuf_hash_insert(db)) != NULL) { /* someone else inserted it first */ kmem_cache_free(dbuf_kmem_cache, db); mutex_exit(&dn->dn_dbufs_mtx); return (odb); } avl_add(&dn->dn_dbufs, db); db->db_state = DB_UNCACHED; mutex_exit(&dn->dn_dbufs_mtx); arc_space_consume(sizeof (dmu_buf_impl_t), ARC_SPACE_OTHER); if (parent && parent != dn->dn_dbuf) dbuf_add_ref(parent, db); ASSERT(dn->dn_object == DMU_META_DNODE_OBJECT || refcount_count(&dn->dn_holds) > 0); (void) refcount_add(&dn->dn_holds, db); atomic_inc_32(&dn->dn_dbufs_count); dprintf_dbuf(db, "db=%p\n", db); return (db); } typedef struct dbuf_prefetch_arg { spa_t *dpa_spa; /* The spa to issue the prefetch in. */ zbookmark_phys_t dpa_zb; /* The target block to prefetch. */ int dpa_epbs; /* Entries (blkptr_t's) Per Block Shift. */ int dpa_curlevel; /* The current level that we're reading */ dnode_t *dpa_dnode; /* The dnode associated with the prefetch */ zio_priority_t dpa_prio; /* The priority I/Os should be issued at. */ zio_t *dpa_zio; /* The parent zio_t for all prefetches. */ arc_flags_t dpa_aflags; /* Flags to pass to the final prefetch. */ } dbuf_prefetch_arg_t; /* * Actually issue the prefetch read for the block given. */ static void dbuf_issue_final_prefetch(dbuf_prefetch_arg_t *dpa, blkptr_t *bp) { if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) return; arc_flags_t aflags = dpa->dpa_aflags | ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH; ASSERT3U(dpa->dpa_curlevel, ==, BP_GET_LEVEL(bp)); ASSERT3U(dpa->dpa_curlevel, ==, dpa->dpa_zb.zb_level); ASSERT(dpa->dpa_zio != NULL); (void) arc_read(dpa->dpa_zio, dpa->dpa_spa, bp, NULL, NULL, dpa->dpa_prio, ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, &aflags, &dpa->dpa_zb); } /* * Called when an indirect block above our prefetch target is read in. This * will either read in the next indirect block down the tree or issue the actual * prefetch if the next block down is our target. */ static void dbuf_prefetch_indirect_done(zio_t *zio, arc_buf_t *abuf, void *private) { dbuf_prefetch_arg_t *dpa = private; ASSERT3S(dpa->dpa_zb.zb_level, <, dpa->dpa_curlevel); ASSERT3S(dpa->dpa_curlevel, >, 0); /* * The dpa_dnode is only valid if we are called with a NULL * zio. This indicates that the arc_read() returned without * first calling zio_read() to issue a physical read. Once * a physical read is made the dpa_dnode must be invalidated * as the locks guarding it may have been dropped. If the * dpa_dnode is still valid, then we want to add it to the dbuf * cache. To do so, we must hold the dbuf associated with the block * we just prefetched, read its contents so that we associate it * with an arc_buf_t, and then release it. */ if (zio != NULL) { ASSERT3S(BP_GET_LEVEL(zio->io_bp), ==, dpa->dpa_curlevel); if (zio->io_flags & ZIO_FLAG_RAW) { ASSERT3U(BP_GET_PSIZE(zio->io_bp), ==, zio->io_size); } else { ASSERT3U(BP_GET_LSIZE(zio->io_bp), ==, zio->io_size); } ASSERT3P(zio->io_spa, ==, dpa->dpa_spa); dpa->dpa_dnode = NULL; } else if (dpa->dpa_dnode != NULL) { uint64_t curblkid = dpa->dpa_zb.zb_blkid >> (dpa->dpa_epbs * (dpa->dpa_curlevel - dpa->dpa_zb.zb_level)); dmu_buf_impl_t *db = dbuf_hold_level(dpa->dpa_dnode, dpa->dpa_curlevel, curblkid, FTAG); (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH | DB_RF_HAVESTRUCT); dbuf_rele(db, FTAG); } dpa->dpa_curlevel--; uint64_t nextblkid = dpa->dpa_zb.zb_blkid >> (dpa->dpa_epbs * (dpa->dpa_curlevel - dpa->dpa_zb.zb_level)); blkptr_t *bp = ((blkptr_t *)abuf->b_data) + P2PHASE(nextblkid, 1ULL << dpa->dpa_epbs); if (BP_IS_HOLE(bp) || (zio != NULL && zio->io_error != 0)) { kmem_free(dpa, sizeof (*dpa)); } else if (dpa->dpa_curlevel == dpa->dpa_zb.zb_level) { ASSERT3U(nextblkid, ==, dpa->dpa_zb.zb_blkid); dbuf_issue_final_prefetch(dpa, bp); kmem_free(dpa, sizeof (*dpa)); } else { arc_flags_t iter_aflags = ARC_FLAG_NOWAIT; zbookmark_phys_t zb; ASSERT3U(dpa->dpa_curlevel, ==, BP_GET_LEVEL(bp)); SET_BOOKMARK(&zb, dpa->dpa_zb.zb_objset, dpa->dpa_zb.zb_object, dpa->dpa_curlevel, nextblkid); (void) arc_read(dpa->dpa_zio, dpa->dpa_spa, bp, dbuf_prefetch_indirect_done, dpa, dpa->dpa_prio, ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, &iter_aflags, &zb); } arc_buf_destroy(abuf, private); } /* * Issue prefetch reads for the given block on the given level. If the indirect * blocks above that block are not in memory, we will read them in * asynchronously. As a result, this call never blocks waiting for a read to * complete. */ void dbuf_prefetch(dnode_t *dn, int64_t level, uint64_t blkid, zio_priority_t prio, arc_flags_t aflags) { blkptr_t bp; int epbs, nlevels, curlevel; uint64_t curblkid; ASSERT(blkid != DMU_BONUS_BLKID); ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock)); if (blkid > dn->dn_maxblkid) return; if (dnode_block_freed(dn, blkid)) return; /* * This dnode hasn't been written to disk yet, so there's nothing to * prefetch. */ nlevels = dn->dn_phys->dn_nlevels; if (level >= nlevels || dn->dn_phys->dn_nblkptr == 0) return; epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; if (dn->dn_phys->dn_maxblkid < blkid << (epbs * level)) return; dmu_buf_impl_t *db = dbuf_find(dn->dn_objset, dn->dn_object, level, blkid); if (db != NULL) { mutex_exit(&db->db_mtx); /* * This dbuf already exists. It is either CACHED, or * (we assume) about to be read or filled. */ return; } /* * Find the closest ancestor (indirect block) of the target block * that is present in the cache. In this indirect block, we will * find the bp that is at curlevel, curblkid. */ curlevel = level; curblkid = blkid; while (curlevel < nlevels - 1) { int parent_level = curlevel + 1; uint64_t parent_blkid = curblkid >> epbs; dmu_buf_impl_t *db; if (dbuf_hold_impl(dn, parent_level, parent_blkid, FALSE, TRUE, FTAG, &db) == 0) { blkptr_t *bpp = db->db_buf->b_data; bp = bpp[P2PHASE(curblkid, 1 << epbs)]; dbuf_rele(db, FTAG); break; } curlevel = parent_level; curblkid = parent_blkid; } if (curlevel == nlevels - 1) { /* No cached indirect blocks found. */ ASSERT3U(curblkid, <, dn->dn_phys->dn_nblkptr); bp = dn->dn_phys->dn_blkptr[curblkid]; } if (BP_IS_HOLE(&bp)) return; ASSERT3U(curlevel, ==, BP_GET_LEVEL(&bp)); zio_t *pio = zio_root(dmu_objset_spa(dn->dn_objset), NULL, NULL, ZIO_FLAG_CANFAIL); dbuf_prefetch_arg_t *dpa = kmem_zalloc(sizeof (*dpa), KM_SLEEP); dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset; SET_BOOKMARK(&dpa->dpa_zb, ds != NULL ? ds->ds_object : DMU_META_OBJSET, dn->dn_object, level, blkid); dpa->dpa_curlevel = curlevel; dpa->dpa_prio = prio; dpa->dpa_aflags = aflags; dpa->dpa_spa = dn->dn_objset->os_spa; dpa->dpa_dnode = dn; dpa->dpa_epbs = epbs; dpa->dpa_zio = pio; /* * If we have the indirect just above us, no need to do the asynchronous * prefetch chain; we'll just run the last step ourselves. If we're at * a higher level, though, we want to issue the prefetches for all the * indirect blocks asynchronously, so we can go on with whatever we were * doing. */ if (curlevel == level) { ASSERT3U(curblkid, ==, blkid); dbuf_issue_final_prefetch(dpa, &bp); kmem_free(dpa, sizeof (*dpa)); } else { arc_flags_t iter_aflags = ARC_FLAG_NOWAIT; zbookmark_phys_t zb; SET_BOOKMARK(&zb, ds != NULL ? ds->ds_object : DMU_META_OBJSET, dn->dn_object, curlevel, curblkid); (void) arc_read(dpa->dpa_zio, dpa->dpa_spa, &bp, dbuf_prefetch_indirect_done, dpa, prio, ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, &iter_aflags, &zb); } /* * We use pio here instead of dpa_zio since it's possible that * dpa may have already been freed. */ zio_nowait(pio); } /* * Returns with db_holds incremented, and db_mtx not held. * Note: dn_struct_rwlock must be held. */ int dbuf_hold_impl(dnode_t *dn, uint8_t level, uint64_t blkid, boolean_t fail_sparse, boolean_t fail_uncached, void *tag, dmu_buf_impl_t **dbp) { dmu_buf_impl_t *db, *parent = NULL; ASSERT(blkid != DMU_BONUS_BLKID); ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock)); ASSERT3U(dn->dn_nlevels, >, level); *dbp = NULL; top: /* dbuf_find() returns with db_mtx held */ db = dbuf_find(dn->dn_objset, dn->dn_object, level, blkid); if (db == NULL) { blkptr_t *bp = NULL; int err; if (fail_uncached) return (SET_ERROR(ENOENT)); ASSERT3P(parent, ==, NULL); err = dbuf_findbp(dn, level, blkid, fail_sparse, &parent, &bp); if (fail_sparse) { if (err == 0 && bp && BP_IS_HOLE(bp)) err = SET_ERROR(ENOENT); if (err) { if (parent) dbuf_rele(parent, NULL); return (err); } } if (err && err != ENOENT) return (err); db = dbuf_create(dn, level, blkid, parent, bp); } if (fail_uncached && db->db_state != DB_CACHED) { mutex_exit(&db->db_mtx); return (SET_ERROR(ENOENT)); } if (db->db_buf != NULL) ASSERT3P(db->db.db_data, ==, db->db_buf->b_data); ASSERT(db->db_buf == NULL || arc_referenced(db->db_buf)); /* * If this buffer is currently syncing out, and we are are * still referencing it from db_data, we need to make a copy * of it in case we decide we want to dirty it again in this txg. */ if (db->db_level == 0 && db->db_blkid != DMU_BONUS_BLKID && dn->dn_object != DMU_META_DNODE_OBJECT && db->db_state == DB_CACHED && db->db_data_pending) { dbuf_dirty_record_t *dr = db->db_data_pending; if (dr->dt.dl.dr_data == db->db_buf) { arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db); dbuf_set_data(db, arc_alloc_buf(dn->dn_objset->os_spa, db, type, db->db.db_size)); bcopy(dr->dt.dl.dr_data->b_data, db->db.db_data, db->db.db_size); } } if (multilist_link_active(&db->db_cache_link)) { ASSERT(refcount_is_zero(&db->db_holds)); - multilist_remove(&dbuf_cache, db); + multilist_remove(dbuf_cache, db); (void) refcount_remove_many(&dbuf_cache_size, db->db.db_size, db); } (void) refcount_add(&db->db_holds, tag); DBUF_VERIFY(db); mutex_exit(&db->db_mtx); /* NOTE: we can't rele the parent until after we drop the db_mtx */ if (parent) dbuf_rele(parent, NULL); ASSERT3P(DB_DNODE(db), ==, dn); ASSERT3U(db->db_blkid, ==, blkid); ASSERT3U(db->db_level, ==, level); *dbp = db; return (0); } dmu_buf_impl_t * dbuf_hold(dnode_t *dn, uint64_t blkid, void *tag) { return (dbuf_hold_level(dn, 0, blkid, tag)); } dmu_buf_impl_t * dbuf_hold_level(dnode_t *dn, int level, uint64_t blkid, void *tag) { dmu_buf_impl_t *db; int err = dbuf_hold_impl(dn, level, blkid, FALSE, FALSE, tag, &db); return (err ? NULL : db); } void dbuf_create_bonus(dnode_t *dn) { ASSERT(RW_WRITE_HELD(&dn->dn_struct_rwlock)); ASSERT(dn->dn_bonus == NULL); dn->dn_bonus = dbuf_create(dn, 0, DMU_BONUS_BLKID, dn->dn_dbuf, NULL); } int dbuf_spill_set_blksz(dmu_buf_t *db_fake, uint64_t blksz, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; dnode_t *dn; if (db->db_blkid != DMU_SPILL_BLKID) return (SET_ERROR(ENOTSUP)); if (blksz == 0) blksz = SPA_MINBLOCKSIZE; ASSERT3U(blksz, <=, spa_maxblocksize(dmu_objset_spa(db->db_objset))); blksz = P2ROUNDUP(blksz, SPA_MINBLOCKSIZE); DB_DNODE_ENTER(db); dn = DB_DNODE(db); rw_enter(&dn->dn_struct_rwlock, RW_WRITER); dbuf_new_size(db, blksz, tx); rw_exit(&dn->dn_struct_rwlock); DB_DNODE_EXIT(db); return (0); } void dbuf_rm_spill(dnode_t *dn, dmu_tx_t *tx) { dbuf_free_range(dn, DMU_SPILL_BLKID, DMU_SPILL_BLKID, tx); } #pragma weak dmu_buf_add_ref = dbuf_add_ref void dbuf_add_ref(dmu_buf_impl_t *db, void *tag) { int64_t holds = refcount_add(&db->db_holds, tag); ASSERT3S(holds, >, 1); } #pragma weak dmu_buf_try_add_ref = dbuf_try_add_ref boolean_t dbuf_try_add_ref(dmu_buf_t *db_fake, objset_t *os, uint64_t obj, uint64_t blkid, void *tag) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; dmu_buf_impl_t *found_db; boolean_t result = B_FALSE; if (db->db_blkid == DMU_BONUS_BLKID) found_db = dbuf_find_bonus(os, obj); else found_db = dbuf_find(os, obj, 0, blkid); if (found_db != NULL) { if (db == found_db && dbuf_refcount(db) > db->db_dirtycnt) { (void) refcount_add(&db->db_holds, tag); result = B_TRUE; } mutex_exit(&db->db_mtx); } return (result); } /* * If you call dbuf_rele() you had better not be referencing the dnode handle * unless you have some other direct or indirect hold on the dnode. (An indirect * hold is a hold on one of the dnode's dbufs, including the bonus buffer.) * Without that, the dbuf_rele() could lead to a dnode_rele() followed by the * dnode's parent dbuf evicting its dnode handles. */ void dbuf_rele(dmu_buf_impl_t *db, void *tag) { mutex_enter(&db->db_mtx); dbuf_rele_and_unlock(db, tag); } void dmu_buf_rele(dmu_buf_t *db, void *tag) { dbuf_rele((dmu_buf_impl_t *)db, tag); } /* * dbuf_rele() for an already-locked dbuf. This is necessary to allow * db_dirtycnt and db_holds to be updated atomically. */ void dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag) { int64_t holds; ASSERT(MUTEX_HELD(&db->db_mtx)); DBUF_VERIFY(db); /* * Remove the reference to the dbuf before removing its hold on the * dnode so we can guarantee in dnode_move() that a referenced bonus * buffer has a corresponding dnode hold. */ holds = refcount_remove(&db->db_holds, tag); ASSERT(holds >= 0); /* * We can't freeze indirects if there is a possibility that they * may be modified in the current syncing context. */ if (db->db_buf != NULL && holds == (db->db_level == 0 ? db->db_dirtycnt : 0)) { arc_buf_freeze(db->db_buf); } if (holds == db->db_dirtycnt && db->db_level == 0 && db->db_user_immediate_evict) dbuf_evict_user(db); if (holds == 0) { if (db->db_blkid == DMU_BONUS_BLKID) { dnode_t *dn; boolean_t evict_dbuf = db->db_pending_evict; /* * If the dnode moves here, we cannot cross this * barrier until the move completes. */ DB_DNODE_ENTER(db); dn = DB_DNODE(db); atomic_dec_32(&dn->dn_dbufs_count); /* * Decrementing the dbuf count means that the bonus * buffer's dnode hold is no longer discounted in * dnode_move(). The dnode cannot move until after * the dnode_rele() below. */ DB_DNODE_EXIT(db); /* * Do not reference db after its lock is dropped. * Another thread may evict it. */ mutex_exit(&db->db_mtx); if (evict_dbuf) dnode_evict_bonus(dn); dnode_rele(dn, db); } else if (db->db_buf == NULL) { /* * This is a special case: we never associated this * dbuf with any data allocated from the ARC. */ ASSERT(db->db_state == DB_UNCACHED || db->db_state == DB_NOFILL); dbuf_destroy(db); } else if (arc_released(db->db_buf)) { /* * This dbuf has anonymous data associated with it. */ dbuf_destroy(db); } else { boolean_t do_arc_evict = B_FALSE; blkptr_t bp; spa_t *spa = dmu_objset_spa(db->db_objset); if (!DBUF_IS_CACHEABLE(db) && db->db_blkptr != NULL && !BP_IS_HOLE(db->db_blkptr) && !BP_IS_EMBEDDED(db->db_blkptr)) { do_arc_evict = B_TRUE; bp = *db->db_blkptr; } if (!DBUF_IS_CACHEABLE(db) || db->db_pending_evict) { dbuf_destroy(db); } else if (!multilist_link_active(&db->db_cache_link)) { - multilist_insert(&dbuf_cache, db); + multilist_insert(dbuf_cache, db); (void) refcount_add_many(&dbuf_cache_size, db->db.db_size, db); mutex_exit(&db->db_mtx); dbuf_evict_notify(); } if (do_arc_evict) arc_freed(spa, &bp); } } else { mutex_exit(&db->db_mtx); } } #pragma weak dmu_buf_refcount = dbuf_refcount uint64_t dbuf_refcount(dmu_buf_impl_t *db) { return (refcount_count(&db->db_holds)); } void * dmu_buf_replace_user(dmu_buf_t *db_fake, dmu_buf_user_t *old_user, dmu_buf_user_t *new_user) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; mutex_enter(&db->db_mtx); dbuf_verify_user(db, DBVU_NOT_EVICTING); if (db->db_user == old_user) db->db_user = new_user; else old_user = db->db_user; dbuf_verify_user(db, DBVU_NOT_EVICTING); mutex_exit(&db->db_mtx); return (old_user); } void * dmu_buf_set_user(dmu_buf_t *db_fake, dmu_buf_user_t *user) { return (dmu_buf_replace_user(db_fake, NULL, user)); } void * dmu_buf_set_user_ie(dmu_buf_t *db_fake, dmu_buf_user_t *user) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; db->db_user_immediate_evict = TRUE; return (dmu_buf_set_user(db_fake, user)); } void * dmu_buf_remove_user(dmu_buf_t *db_fake, dmu_buf_user_t *user) { return (dmu_buf_replace_user(db_fake, user, NULL)); } void * dmu_buf_get_user(dmu_buf_t *db_fake) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; dbuf_verify_user(db, DBVU_NOT_EVICTING); return (db->db_user); } void dmu_buf_user_evict_wait() { taskq_wait(dbu_evict_taskq); } blkptr_t * dmu_buf_get_blkptr(dmu_buf_t *db) { dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; return (dbi->db_blkptr); } objset_t * dmu_buf_get_objset(dmu_buf_t *db) { dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; return (dbi->db_objset); } dnode_t * dmu_buf_dnode_enter(dmu_buf_t *db) { dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; DB_DNODE_ENTER(dbi); return (DB_DNODE(dbi)); } void dmu_buf_dnode_exit(dmu_buf_t *db) { dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; DB_DNODE_EXIT(dbi); } static void dbuf_check_blkptr(dnode_t *dn, dmu_buf_impl_t *db) { /* ASSERT(dmu_tx_is_syncing(tx) */ ASSERT(MUTEX_HELD(&db->db_mtx)); if (db->db_blkptr != NULL) return; if (db->db_blkid == DMU_SPILL_BLKID) { db->db_blkptr = &dn->dn_phys->dn_spill; BP_ZERO(db->db_blkptr); return; } if (db->db_level == dn->dn_phys->dn_nlevels-1) { /* * This buffer was allocated at a time when there was * no available blkptrs from the dnode, or it was * inappropriate to hook it in (i.e., nlevels mis-match). */ ASSERT(db->db_blkid < dn->dn_phys->dn_nblkptr); ASSERT(db->db_parent == NULL); db->db_parent = dn->dn_dbuf; db->db_blkptr = &dn->dn_phys->dn_blkptr[db->db_blkid]; DBUF_VERIFY(db); } else { dmu_buf_impl_t *parent = db->db_parent; int epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; ASSERT(dn->dn_phys->dn_nlevels > 1); if (parent == NULL) { mutex_exit(&db->db_mtx); rw_enter(&dn->dn_struct_rwlock, RW_READER); parent = dbuf_hold_level(dn, db->db_level + 1, db->db_blkid >> epbs, db); rw_exit(&dn->dn_struct_rwlock); mutex_enter(&db->db_mtx); db->db_parent = parent; } db->db_blkptr = (blkptr_t *)parent->db.db_data + (db->db_blkid & ((1ULL << epbs) - 1)); DBUF_VERIFY(db); } } static void dbuf_sync_indirect(dbuf_dirty_record_t *dr, dmu_tx_t *tx) { dmu_buf_impl_t *db = dr->dr_dbuf; dnode_t *dn; zio_t *zio; ASSERT(dmu_tx_is_syncing(tx)); dprintf_dbuf_bp(db, db->db_blkptr, "blkptr=%p", db->db_blkptr); mutex_enter(&db->db_mtx); ASSERT(db->db_level > 0); DBUF_VERIFY(db); /* Read the block if it hasn't been read yet. */ if (db->db_buf == NULL) { mutex_exit(&db->db_mtx); (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED); mutex_enter(&db->db_mtx); } ASSERT3U(db->db_state, ==, DB_CACHED); ASSERT(db->db_buf != NULL); DB_DNODE_ENTER(db); dn = DB_DNODE(db); /* Indirect block size must match what the dnode thinks it is. */ ASSERT3U(db->db.db_size, ==, 1<dn_phys->dn_indblkshift); dbuf_check_blkptr(dn, db); DB_DNODE_EXIT(db); /* Provide the pending dirty record to child dbufs */ db->db_data_pending = dr; mutex_exit(&db->db_mtx); dbuf_write(dr, db->db_buf, tx); zio = dr->dr_zio; mutex_enter(&dr->dt.di.dr_mtx); dbuf_sync_list(&dr->dt.di.dr_children, db->db_level - 1, tx); ASSERT(list_head(&dr->dt.di.dr_children) == NULL); mutex_exit(&dr->dt.di.dr_mtx); zio_nowait(zio); } static void dbuf_sync_leaf(dbuf_dirty_record_t *dr, dmu_tx_t *tx) { arc_buf_t **datap = &dr->dt.dl.dr_data; dmu_buf_impl_t *db = dr->dr_dbuf; dnode_t *dn; objset_t *os; uint64_t txg = tx->tx_txg; ASSERT(dmu_tx_is_syncing(tx)); dprintf_dbuf_bp(db, db->db_blkptr, "blkptr=%p", db->db_blkptr); mutex_enter(&db->db_mtx); /* * To be synced, we must be dirtied. But we * might have been freed after the dirty. */ if (db->db_state == DB_UNCACHED) { /* This buffer has been freed since it was dirtied */ ASSERT(db->db.db_data == NULL); } else if (db->db_state == DB_FILL) { /* This buffer was freed and is now being re-filled */ ASSERT(db->db.db_data != dr->dt.dl.dr_data); } else { ASSERT(db->db_state == DB_CACHED || db->db_state == DB_NOFILL); } DBUF_VERIFY(db); DB_DNODE_ENTER(db); dn = DB_DNODE(db); if (db->db_blkid == DMU_SPILL_BLKID) { mutex_enter(&dn->dn_mtx); dn->dn_phys->dn_flags |= DNODE_FLAG_SPILL_BLKPTR; mutex_exit(&dn->dn_mtx); } /* * If this is a bonus buffer, simply copy the bonus data into the * dnode. It will be written out when the dnode is synced (and it * will be synced, since it must have been dirty for dbuf_sync to * be called). */ if (db->db_blkid == DMU_BONUS_BLKID) { dbuf_dirty_record_t **drp; ASSERT(*datap != NULL); ASSERT0(db->db_level); ASSERT3U(dn->dn_phys->dn_bonuslen, <=, DN_MAX_BONUSLEN); bcopy(*datap, DN_BONUS(dn->dn_phys), dn->dn_phys->dn_bonuslen); DB_DNODE_EXIT(db); if (*datap != db->db.db_data) { zio_buf_free(*datap, DN_MAX_BONUSLEN); arc_space_return(DN_MAX_BONUSLEN, ARC_SPACE_OTHER); } db->db_data_pending = NULL; drp = &db->db_last_dirty; while (*drp != dr) drp = &(*drp)->dr_next; ASSERT(dr->dr_next == NULL); ASSERT(dr->dr_dbuf == db); *drp = dr->dr_next; kmem_free(dr, sizeof (dbuf_dirty_record_t)); ASSERT(db->db_dirtycnt > 0); db->db_dirtycnt -= 1; dbuf_rele_and_unlock(db, (void *)(uintptr_t)txg); return; } os = dn->dn_objset; /* * This function may have dropped the db_mtx lock allowing a dmu_sync * operation to sneak in. As a result, we need to ensure that we * don't check the dr_override_state until we have returned from * dbuf_check_blkptr. */ dbuf_check_blkptr(dn, db); /* * If this buffer is in the middle of an immediate write, * wait for the synchronous IO to complete. */ while (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC) { ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); cv_wait(&db->db_changed, &db->db_mtx); ASSERT(dr->dt.dl.dr_override_state != DR_NOT_OVERRIDDEN); } if (db->db_state != DB_NOFILL && dn->dn_object != DMU_META_DNODE_OBJECT && refcount_count(&db->db_holds) > 1 && dr->dt.dl.dr_override_state != DR_OVERRIDDEN && *datap == db->db_buf) { /* * If this buffer is currently "in use" (i.e., there * are active holds and db_data still references it), * then make a copy before we start the write so that * any modifications from the open txg will not leak * into this write. * * NOTE: this copy does not need to be made for * objects only modified in the syncing context (e.g. * DNONE_DNODE blocks). */ int psize = arc_buf_size(*datap); arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db); enum zio_compress compress_type = arc_get_compression(*datap); if (compress_type == ZIO_COMPRESS_OFF) { *datap = arc_alloc_buf(os->os_spa, db, type, psize); } else { ASSERT3U(type, ==, ARC_BUFC_DATA); int lsize = arc_buf_lsize(*datap); *datap = arc_alloc_compressed_buf(os->os_spa, db, psize, lsize, compress_type); } bcopy(db->db.db_data, (*datap)->b_data, psize); } db->db_data_pending = dr; mutex_exit(&db->db_mtx); dbuf_write(dr, *datap, tx); ASSERT(!list_link_active(&dr->dr_dirty_node)); if (dn->dn_object == DMU_META_DNODE_OBJECT) { list_insert_tail(&dn->dn_dirty_records[txg&TXG_MASK], dr); DB_DNODE_EXIT(db); } else { /* * Although zio_nowait() does not "wait for an IO", it does * initiate the IO. If this is an empty write it seems plausible * that the IO could actually be completed before the nowait * returns. We need to DB_DNODE_EXIT() first in case * zio_nowait() invalidates the dbuf. */ DB_DNODE_EXIT(db); zio_nowait(dr->dr_zio); } } void dbuf_sync_list(list_t *list, int level, dmu_tx_t *tx) { dbuf_dirty_record_t *dr; while (dr = list_head(list)) { if (dr->dr_zio != NULL) { /* * If we find an already initialized zio then we * are processing the meta-dnode, and we have finished. * The dbufs for all dnodes are put back on the list * during processing, so that we can zio_wait() * these IOs after initiating all child IOs. */ ASSERT3U(dr->dr_dbuf->db.db_object, ==, DMU_META_DNODE_OBJECT); break; } if (dr->dr_dbuf->db_blkid != DMU_BONUS_BLKID && dr->dr_dbuf->db_blkid != DMU_SPILL_BLKID) { VERIFY3U(dr->dr_dbuf->db_level, ==, level); } list_remove(list, dr); if (dr->dr_dbuf->db_level > 0) dbuf_sync_indirect(dr, tx); else dbuf_sync_leaf(dr, tx); } } /* ARGSUSED */ static void dbuf_write_ready(zio_t *zio, arc_buf_t *buf, void *vdb) { dmu_buf_impl_t *db = vdb; dnode_t *dn; blkptr_t *bp = zio->io_bp; blkptr_t *bp_orig = &zio->io_bp_orig; spa_t *spa = zio->io_spa; int64_t delta; uint64_t fill = 0; int i; ASSERT3P(db->db_blkptr, !=, NULL); ASSERT3P(&db->db_data_pending->dr_bp_copy, ==, bp); DB_DNODE_ENTER(db); dn = DB_DNODE(db); delta = bp_get_dsize_sync(spa, bp) - bp_get_dsize_sync(spa, bp_orig); dnode_diduse_space(dn, delta - zio->io_prev_space_delta); zio->io_prev_space_delta = delta; if (bp->blk_birth != 0) { ASSERT((db->db_blkid != DMU_SPILL_BLKID && BP_GET_TYPE(bp) == dn->dn_type) || (db->db_blkid == DMU_SPILL_BLKID && BP_GET_TYPE(bp) == dn->dn_bonustype) || BP_IS_EMBEDDED(bp)); ASSERT(BP_GET_LEVEL(bp) == db->db_level); } mutex_enter(&db->db_mtx); #ifdef ZFS_DEBUG if (db->db_blkid == DMU_SPILL_BLKID) { ASSERT(dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR); ASSERT(!(BP_IS_HOLE(bp)) && db->db_blkptr == &dn->dn_phys->dn_spill); } #endif if (db->db_level == 0) { mutex_enter(&dn->dn_mtx); if (db->db_blkid > dn->dn_phys->dn_maxblkid && db->db_blkid != DMU_SPILL_BLKID) dn->dn_phys->dn_maxblkid = db->db_blkid; mutex_exit(&dn->dn_mtx); if (dn->dn_type == DMU_OT_DNODE) { dnode_phys_t *dnp = db->db.db_data; for (i = db->db.db_size >> DNODE_SHIFT; i > 0; i--, dnp++) { if (dnp->dn_type != DMU_OT_NONE) fill++; } } else { if (BP_IS_HOLE(bp)) { fill = 0; } else { fill = 1; } } } else { blkptr_t *ibp = db->db.db_data; ASSERT3U(db->db.db_size, ==, 1<dn_phys->dn_indblkshift); for (i = db->db.db_size >> SPA_BLKPTRSHIFT; i > 0; i--, ibp++) { if (BP_IS_HOLE(ibp)) continue; fill += BP_GET_FILL(ibp); } } DB_DNODE_EXIT(db); if (!BP_IS_EMBEDDED(bp)) bp->blk_fill = fill; mutex_exit(&db->db_mtx); rw_enter(&dn->dn_struct_rwlock, RW_WRITER); *db->db_blkptr = *bp; rw_exit(&dn->dn_struct_rwlock); } /* ARGSUSED */ /* * This function gets called just prior to running through the compression * stage of the zio pipeline. If we're an indirect block comprised of only * holes, then we want this indirect to be compressed away to a hole. In * order to do that we must zero out any information about the holes that * this indirect points to prior to before we try to compress it. */ static void dbuf_write_children_ready(zio_t *zio, arc_buf_t *buf, void *vdb) { dmu_buf_impl_t *db = vdb; dnode_t *dn; blkptr_t *bp; unsigned int epbs, i; ASSERT3U(db->db_level, >, 0); DB_DNODE_ENTER(db); dn = DB_DNODE(db); epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; ASSERT3U(epbs, <, 31); /* Determine if all our children are holes */ for (i = 0, bp = db->db.db_data; i < 1 << epbs; i++, bp++) { if (!BP_IS_HOLE(bp)) break; } /* * If all the children are holes, then zero them all out so that * we may get compressed away. */ if (i == 1 << epbs) { /* * We only found holes. Grab the rwlock to prevent * anybody from reading the blocks we're about to * zero out. */ rw_enter(&dn->dn_struct_rwlock, RW_WRITER); bzero(db->db.db_data, db->db.db_size); rw_exit(&dn->dn_struct_rwlock); } DB_DNODE_EXIT(db); } /* * The SPA will call this callback several times for each zio - once * for every physical child i/o (zio->io_phys_children times). This * allows the DMU to monitor the progress of each logical i/o. For example, * there may be 2 copies of an indirect block, or many fragments of a RAID-Z * block. There may be a long delay before all copies/fragments are completed, * so this callback allows us to retire dirty space gradually, as the physical * i/os complete. */ /* ARGSUSED */ static void dbuf_write_physdone(zio_t *zio, arc_buf_t *buf, void *arg) { dmu_buf_impl_t *db = arg; objset_t *os = db->db_objset; dsl_pool_t *dp = dmu_objset_pool(os); dbuf_dirty_record_t *dr; int delta = 0; dr = db->db_data_pending; ASSERT3U(dr->dr_txg, ==, zio->io_txg); /* * The callback will be called io_phys_children times. Retire one * portion of our dirty space each time we are called. Any rounding * error will be cleaned up by dsl_pool_sync()'s call to * dsl_pool_undirty_space(). */ delta = dr->dr_accounted / zio->io_phys_children; dsl_pool_undirty_space(dp, delta, zio->io_txg); } /* ARGSUSED */ static void dbuf_write_done(zio_t *zio, arc_buf_t *buf, void *vdb) { dmu_buf_impl_t *db = vdb; blkptr_t *bp_orig = &zio->io_bp_orig; blkptr_t *bp = db->db_blkptr; objset_t *os = db->db_objset; dmu_tx_t *tx = os->os_synctx; dbuf_dirty_record_t **drp, *dr; ASSERT0(zio->io_error); ASSERT(db->db_blkptr == bp); /* * For nopwrites and rewrites we ensure that the bp matches our * original and bypass all the accounting. */ if (zio->io_flags & (ZIO_FLAG_IO_REWRITE | ZIO_FLAG_NOPWRITE)) { ASSERT(BP_EQUAL(bp, bp_orig)); } else { dsl_dataset_t *ds = os->os_dsl_dataset; (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE); dsl_dataset_block_born(ds, bp, tx); } mutex_enter(&db->db_mtx); DBUF_VERIFY(db); drp = &db->db_last_dirty; while ((dr = *drp) != db->db_data_pending) drp = &dr->dr_next; ASSERT(!list_link_active(&dr->dr_dirty_node)); ASSERT(dr->dr_dbuf == db); ASSERT(dr->dr_next == NULL); *drp = dr->dr_next; #ifdef ZFS_DEBUG if (db->db_blkid == DMU_SPILL_BLKID) { dnode_t *dn; DB_DNODE_ENTER(db); dn = DB_DNODE(db); ASSERT(dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR); ASSERT(!(BP_IS_HOLE(db->db_blkptr)) && db->db_blkptr == &dn->dn_phys->dn_spill); DB_DNODE_EXIT(db); } #endif if (db->db_level == 0) { ASSERT(db->db_blkid != DMU_BONUS_BLKID); ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); if (db->db_state != DB_NOFILL) { if (dr->dt.dl.dr_data != db->db_buf) arc_buf_destroy(dr->dt.dl.dr_data, db); } } else { dnode_t *dn; DB_DNODE_ENTER(db); dn = DB_DNODE(db); ASSERT(list_head(&dr->dt.di.dr_children) == NULL); ASSERT3U(db->db.db_size, ==, 1 << dn->dn_phys->dn_indblkshift); if (!BP_IS_HOLE(db->db_blkptr)) { int epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; ASSERT3U(db->db_blkid, <=, dn->dn_phys->dn_maxblkid >> (db->db_level * epbs)); ASSERT3U(BP_GET_LSIZE(db->db_blkptr), ==, db->db.db_size); } DB_DNODE_EXIT(db); mutex_destroy(&dr->dt.di.dr_mtx); list_destroy(&dr->dt.di.dr_children); } kmem_free(dr, sizeof (dbuf_dirty_record_t)); cv_broadcast(&db->db_changed); ASSERT(db->db_dirtycnt > 0); db->db_dirtycnt -= 1; db->db_data_pending = NULL; dbuf_rele_and_unlock(db, (void *)(uintptr_t)tx->tx_txg); } static void dbuf_write_nofill_ready(zio_t *zio) { dbuf_write_ready(zio, NULL, zio->io_private); } static void dbuf_write_nofill_done(zio_t *zio) { dbuf_write_done(zio, NULL, zio->io_private); } static void dbuf_write_override_ready(zio_t *zio) { dbuf_dirty_record_t *dr = zio->io_private; dmu_buf_impl_t *db = dr->dr_dbuf; dbuf_write_ready(zio, NULL, db); } static void dbuf_write_override_done(zio_t *zio) { dbuf_dirty_record_t *dr = zio->io_private; dmu_buf_impl_t *db = dr->dr_dbuf; blkptr_t *obp = &dr->dt.dl.dr_overridden_by; mutex_enter(&db->db_mtx); if (!BP_EQUAL(zio->io_bp, obp)) { if (!BP_IS_HOLE(obp)) dsl_free(spa_get_dsl(zio->io_spa), zio->io_txg, obp); arc_release(dr->dt.dl.dr_data, db); } mutex_exit(&db->db_mtx); dbuf_write_done(zio, NULL, db); } /* Issue I/O to commit a dirty buffer to disk. */ static void dbuf_write(dbuf_dirty_record_t *dr, arc_buf_t *data, dmu_tx_t *tx) { dmu_buf_impl_t *db = dr->dr_dbuf; dnode_t *dn; objset_t *os; dmu_buf_impl_t *parent = db->db_parent; uint64_t txg = tx->tx_txg; zbookmark_phys_t zb; zio_prop_t zp; zio_t *zio; int wp_flag = 0; ASSERT(dmu_tx_is_syncing(tx)); DB_DNODE_ENTER(db); dn = DB_DNODE(db); os = dn->dn_objset; if (db->db_state != DB_NOFILL) { if (db->db_level > 0 || dn->dn_type == DMU_OT_DNODE) { /* * Private object buffers are released here rather * than in dbuf_dirty() since they are only modified * in the syncing context and we don't want the * overhead of making multiple copies of the data. */ if (BP_IS_HOLE(db->db_blkptr)) { arc_buf_thaw(data); } else { dbuf_release_bp(db); } } } if (parent != dn->dn_dbuf) { /* Our parent is an indirect block. */ /* We have a dirty parent that has been scheduled for write. */ ASSERT(parent && parent->db_data_pending); /* Our parent's buffer is one level closer to the dnode. */ ASSERT(db->db_level == parent->db_level-1); /* * We're about to modify our parent's db_data by modifying * our block pointer, so the parent must be released. */ ASSERT(arc_released(parent->db_buf)); zio = parent->db_data_pending->dr_zio; } else { /* Our parent is the dnode itself. */ ASSERT((db->db_level == dn->dn_phys->dn_nlevels-1 && db->db_blkid != DMU_SPILL_BLKID) || (db->db_blkid == DMU_SPILL_BLKID && db->db_level == 0)); if (db->db_blkid != DMU_SPILL_BLKID) ASSERT3P(db->db_blkptr, ==, &dn->dn_phys->dn_blkptr[db->db_blkid]); zio = dn->dn_zio; } ASSERT(db->db_level == 0 || data == db->db_buf); ASSERT3U(db->db_blkptr->blk_birth, <=, txg); ASSERT(zio); SET_BOOKMARK(&zb, os->os_dsl_dataset ? os->os_dsl_dataset->ds_object : DMU_META_OBJSET, db->db.db_object, db->db_level, db->db_blkid); if (db->db_blkid == DMU_SPILL_BLKID) wp_flag = WP_SPILL; wp_flag |= (db->db_state == DB_NOFILL) ? WP_NOFILL : 0; dmu_write_policy(os, dn, db->db_level, wp_flag, (data != NULL && arc_get_compression(data) != ZIO_COMPRESS_OFF) ? arc_get_compression(data) : ZIO_COMPRESS_INHERIT, &zp); DB_DNODE_EXIT(db); /* * We copy the blkptr now (rather than when we instantiate the dirty * record), because its value can change between open context and * syncing context. We do not need to hold dn_struct_rwlock to read * db_blkptr because we are in syncing context. */ dr->dr_bp_copy = *db->db_blkptr; if (db->db_level == 0 && dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { /* * The BP for this block has been provided by open context * (by dmu_sync() or dmu_buf_write_embedded()). */ void *contents = (data != NULL) ? data->b_data : NULL; dr->dr_zio = zio_write(zio, os->os_spa, txg, &dr->dr_bp_copy, contents, db->db.db_size, db->db.db_size, &zp, dbuf_write_override_ready, NULL, NULL, dbuf_write_override_done, dr, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); mutex_enter(&db->db_mtx); dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; zio_write_override(dr->dr_zio, &dr->dt.dl.dr_overridden_by, dr->dt.dl.dr_copies, dr->dt.dl.dr_nopwrite); mutex_exit(&db->db_mtx); } else if (db->db_state == DB_NOFILL) { ASSERT(zp.zp_checksum == ZIO_CHECKSUM_OFF || zp.zp_checksum == ZIO_CHECKSUM_NOPARITY); dr->dr_zio = zio_write(zio, os->os_spa, txg, &dr->dr_bp_copy, NULL, db->db.db_size, db->db.db_size, &zp, dbuf_write_nofill_ready, NULL, NULL, dbuf_write_nofill_done, db, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED | ZIO_FLAG_NODATA, &zb); } else { ASSERT(arc_released(data)); /* * For indirect blocks, we want to setup the children * ready callback so that we can properly handle an indirect * block that only contains holes. */ arc_done_func_t *children_ready_cb = NULL; if (db->db_level != 0) children_ready_cb = dbuf_write_children_ready; dr->dr_zio = arc_write(zio, os->os_spa, txg, &dr->dr_bp_copy, data, DBUF_IS_L2CACHEABLE(db), &zp, dbuf_write_ready, children_ready_cb, dbuf_write_physdone, dbuf_write_done, db, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); } } Index: vendor-sys/illumos/dist/uts/common/fs/zfs/dmu_objset.c =================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/dmu_objset.c (revision 316916) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/dmu_objset.c (revision 316917) @@ -1,2122 +1,2242 @@ /* * 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 http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. - * Copyright (c) 2012, 2016 by Delphix. All rights reserved. + * Copyright (c) 2012, 2017 by Delphix. All rights reserved. * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. * Copyright (c) 2013, Joyent, Inc. All rights reserved. * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. * Copyright (c) 2015, STRATO AG, Inc. All rights reserved. * Copyright (c) 2014 Integros [integros.com] * Copyright 2017 Nexenta Systems, Inc. */ /* Portions Copyright 2010 Robert Milkowski */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * Needed to close a window in dnode_move() that allows the objset to be freed * before it can be safely accessed. */ krwlock_t os_lock; /* * Tunable to overwrite the maximum number of threads for the parallization * of dmu_objset_find_dp, needed to speed up the import of pools with many * datasets. * Default is 4 times the number of leaf vdevs. */ int dmu_find_threads = 0; /* * Backfill lower metadnode objects after this many have been freed. * Backfilling negatively impacts object creation rates, so only do it * if there are enough holes to fill. */ int dmu_rescan_dnode_threshold = 131072; static void dmu_objset_find_dp_cb(void *arg); void dmu_objset_init(void) { rw_init(&os_lock, NULL, RW_DEFAULT, NULL); } void dmu_objset_fini(void) { rw_destroy(&os_lock); } spa_t * dmu_objset_spa(objset_t *os) { return (os->os_spa); } zilog_t * dmu_objset_zil(objset_t *os) { return (os->os_zil); } dsl_pool_t * dmu_objset_pool(objset_t *os) { dsl_dataset_t *ds; if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir) return (ds->ds_dir->dd_pool); else return (spa_get_dsl(os->os_spa)); } dsl_dataset_t * dmu_objset_ds(objset_t *os) { return (os->os_dsl_dataset); } dmu_objset_type_t dmu_objset_type(objset_t *os) { return (os->os_phys->os_type); } void dmu_objset_name(objset_t *os, char *buf) { dsl_dataset_name(os->os_dsl_dataset, buf); } uint64_t dmu_objset_id(objset_t *os) { dsl_dataset_t *ds = os->os_dsl_dataset; return (ds ? ds->ds_object : 0); } zfs_sync_type_t dmu_objset_syncprop(objset_t *os) { return (os->os_sync); } zfs_logbias_op_t dmu_objset_logbias(objset_t *os) { return (os->os_logbias); } static void checksum_changed_cb(void *arg, uint64_t newval) { objset_t *os = arg; /* * Inheritance should have been done by now. */ ASSERT(newval != ZIO_CHECKSUM_INHERIT); os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE); } static void compression_changed_cb(void *arg, uint64_t newval) { objset_t *os = arg; /* * Inheritance and range checking should have been done by now. */ ASSERT(newval != ZIO_COMPRESS_INHERIT); os->os_compress = zio_compress_select(os->os_spa, newval, ZIO_COMPRESS_ON); } static void copies_changed_cb(void *arg, uint64_t newval) { objset_t *os = arg; /* * Inheritance and range checking should have been done by now. */ ASSERT(newval > 0); ASSERT(newval <= spa_max_replication(os->os_spa)); os->os_copies = newval; } static void dedup_changed_cb(void *arg, uint64_t newval) { objset_t *os = arg; spa_t *spa = os->os_spa; enum zio_checksum checksum; /* * Inheritance should have been done by now. */ ASSERT(newval != ZIO_CHECKSUM_INHERIT); checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF); os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK; os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY); } static void primary_cache_changed_cb(void *arg, uint64_t newval) { objset_t *os = arg; /* * Inheritance and range checking should have been done by now. */ ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || newval == ZFS_CACHE_METADATA); os->os_primary_cache = newval; } static void secondary_cache_changed_cb(void *arg, uint64_t newval) { objset_t *os = arg; /* * Inheritance and range checking should have been done by now. */ ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || newval == ZFS_CACHE_METADATA); os->os_secondary_cache = newval; } static void sync_changed_cb(void *arg, uint64_t newval) { objset_t *os = arg; /* * Inheritance and range checking should have been done by now. */ ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS || newval == ZFS_SYNC_DISABLED); os->os_sync = newval; if (os->os_zil) zil_set_sync(os->os_zil, newval); } static void redundant_metadata_changed_cb(void *arg, uint64_t newval) { objset_t *os = arg; /* * Inheritance and range checking should have been done by now. */ ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL || newval == ZFS_REDUNDANT_METADATA_MOST); os->os_redundant_metadata = newval; } static void logbias_changed_cb(void *arg, uint64_t newval) { objset_t *os = arg; ASSERT(newval == ZFS_LOGBIAS_LATENCY || newval == ZFS_LOGBIAS_THROUGHPUT); os->os_logbias = newval; if (os->os_zil) zil_set_logbias(os->os_zil, newval); } static void recordsize_changed_cb(void *arg, uint64_t newval) { objset_t *os = arg; os->os_recordsize = newval; } void dmu_objset_byteswap(void *buf, size_t size) { objset_phys_t *osp = buf; ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t)); dnode_byteswap(&osp->os_meta_dnode); byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t)); osp->os_type = BSWAP_64(osp->os_type); osp->os_flags = BSWAP_64(osp->os_flags); if (size == sizeof (objset_phys_t)) { dnode_byteswap(&osp->os_userused_dnode); dnode_byteswap(&osp->os_groupused_dnode); } } +/* + * The hash is a CRC-based hash of the objset_t pointer and the object number. + */ +static uint64_t +dnode_hash(const objset_t *os, uint64_t obj) +{ + uintptr_t osv = (uintptr_t)os; + uint64_t crc = -1ULL; + + ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); + /* + * The low 6 bits of the pointer don't have much entropy, because + * the objset_t is larger than 2^6 bytes long. + */ + crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (osv >> 6)) & 0xFF]; + crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 0)) & 0xFF]; + crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 8)) & 0xFF]; + crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 16)) & 0xFF]; + + crc ^= (osv>>14) ^ (obj>>24); + + return (crc); +} + +unsigned int +dnode_multilist_index_func(multilist_t *ml, void *obj) +{ + dnode_t *dn = obj; + return (dnode_hash(dn->dn_objset, dn->dn_object) % + multilist_get_num_sublists(ml)); +} + +/* + * Instantiates the objset_t in-memory structure corresponding to the + * objset_phys_t that's pointed to by the specified blkptr_t. + */ int dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, objset_t **osp) { objset_t *os; int i, err; ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock)); os = kmem_zalloc(sizeof (objset_t), KM_SLEEP); os->os_dsl_dataset = ds; os->os_spa = spa; os->os_rootbp = bp; if (!BP_IS_HOLE(os->os_rootbp)) { arc_flags_t aflags = ARC_FLAG_WAIT; zbookmark_phys_t zb; SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET, ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); if (DMU_OS_IS_L2CACHEABLE(os)) aflags |= ARC_FLAG_L2CACHE; dprintf_bp(os->os_rootbp, "reading %s", ""); err = arc_read(NULL, spa, os->os_rootbp, arc_getbuf_func, &os->os_phys_buf, ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb); if (err != 0) { kmem_free(os, sizeof (objset_t)); /* convert checksum errors into IO errors */ if (err == ECKSUM) err = SET_ERROR(EIO); return (err); } /* Increase the blocksize if we are permitted. */ if (spa_version(spa) >= SPA_VERSION_USERSPACE && arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) { arc_buf_t *buf = arc_alloc_buf(spa, &os->os_phys_buf, ARC_BUFC_METADATA, sizeof (objset_phys_t)); bzero(buf->b_data, sizeof (objset_phys_t)); bcopy(os->os_phys_buf->b_data, buf->b_data, arc_buf_size(os->os_phys_buf)); arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); os->os_phys_buf = buf; } os->os_phys = os->os_phys_buf->b_data; os->os_flags = os->os_phys->os_flags; } else { int size = spa_version(spa) >= SPA_VERSION_USERSPACE ? sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE; os->os_phys_buf = arc_alloc_buf(spa, &os->os_phys_buf, ARC_BUFC_METADATA, size); os->os_phys = os->os_phys_buf->b_data; bzero(os->os_phys, size); } /* * Note: the changed_cb will be called once before the register * func returns, thus changing the checksum/compression from the * default (fletcher2/off). Snapshots don't need to know about * checksum/compression/copies. */ if (ds != NULL) { boolean_t needlock = B_FALSE; /* * Note: it's valid to open the objset if the dataset is * long-held, in which case the pool_config lock will not * be held. */ if (!dsl_pool_config_held(dmu_objset_pool(os))) { needlock = B_TRUE; dsl_pool_config_enter(dmu_objset_pool(os), FTAG); } err = dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE), primary_cache_changed_cb, os); if (err == 0) { err = dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE), secondary_cache_changed_cb, os); } if (!ds->ds_is_snapshot) { if (err == 0) { err = dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum_changed_cb, os); } if (err == 0) { err = dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_COMPRESSION), compression_changed_cb, os); } if (err == 0) { err = dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_COPIES), copies_changed_cb, os); } if (err == 0) { err = dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_DEDUP), dedup_changed_cb, os); } if (err == 0) { err = dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_LOGBIAS), logbias_changed_cb, os); } if (err == 0) { err = dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_SYNC), sync_changed_cb, os); } if (err == 0) { err = dsl_prop_register(ds, zfs_prop_to_name( ZFS_PROP_REDUNDANT_METADATA), redundant_metadata_changed_cb, os); } if (err == 0) { err = dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_RECORDSIZE), recordsize_changed_cb, os); } } if (needlock) dsl_pool_config_exit(dmu_objset_pool(os), FTAG); if (err != 0) { arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); kmem_free(os, sizeof (objset_t)); return (err); } } else { /* It's the meta-objset. */ os->os_checksum = ZIO_CHECKSUM_FLETCHER_4; os->os_compress = ZIO_COMPRESS_ON; os->os_copies = spa_max_replication(spa); os->os_dedup_checksum = ZIO_CHECKSUM_OFF; os->os_dedup_verify = B_FALSE; os->os_logbias = ZFS_LOGBIAS_LATENCY; os->os_sync = ZFS_SYNC_STANDARD; os->os_primary_cache = ZFS_CACHE_ALL; os->os_secondary_cache = ZFS_CACHE_ALL; } if (ds == NULL || !ds->ds_is_snapshot) os->os_zil_header = os->os_phys->os_zil_header; os->os_zil = zil_alloc(os, &os->os_zil_header); for (i = 0; i < TXG_SIZE; i++) { - list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t), - offsetof(dnode_t, dn_dirty_link[i])); - list_create(&os->os_free_dnodes[i], sizeof (dnode_t), - offsetof(dnode_t, dn_dirty_link[i])); + os->os_dirty_dnodes[i] = multilist_create(sizeof (dnode_t), + offsetof(dnode_t, dn_dirty_link[i]), + dnode_multilist_index_func); } list_create(&os->os_dnodes, sizeof (dnode_t), offsetof(dnode_t, dn_link)); list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t), offsetof(dmu_buf_impl_t, db_link)); mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL); + mutex_init(&os->os_userused_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL); dnode_special_open(os, &os->os_phys->os_meta_dnode, DMU_META_DNODE_OBJECT, &os->os_meta_dnode); if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) { dnode_special_open(os, &os->os_phys->os_userused_dnode, DMU_USERUSED_OBJECT, &os->os_userused_dnode); dnode_special_open(os, &os->os_phys->os_groupused_dnode, DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode); } *osp = os; return (0); } int dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp) { int err = 0; /* * We shouldn't be doing anything with dsl_dataset_t's unless the * pool_config lock is held, or the dataset is long-held. */ ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool) || dsl_dataset_long_held(ds)); mutex_enter(&ds->ds_opening_lock); if (ds->ds_objset == NULL) { objset_t *os; rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); err = dmu_objset_open_impl(dsl_dataset_get_spa(ds), ds, dsl_dataset_get_blkptr(ds), &os); rrw_exit(&ds->ds_bp_rwlock, FTAG); if (err == 0) { mutex_enter(&ds->ds_lock); ASSERT(ds->ds_objset == NULL); ds->ds_objset = os; mutex_exit(&ds->ds_lock); } } *osp = ds->ds_objset; mutex_exit(&ds->ds_opening_lock); return (err); } /* * Holds the pool while the objset is held. Therefore only one objset * can be held at a time. */ int dmu_objset_hold(const char *name, void *tag, objset_t **osp) { dsl_pool_t *dp; dsl_dataset_t *ds; int err; err = dsl_pool_hold(name, tag, &dp); if (err != 0) return (err); err = dsl_dataset_hold(dp, name, tag, &ds); if (err != 0) { dsl_pool_rele(dp, tag); return (err); } err = dmu_objset_from_ds(ds, osp); if (err != 0) { dsl_dataset_rele(ds, tag); dsl_pool_rele(dp, tag); } return (err); } static int dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type, boolean_t readonly, void *tag, objset_t **osp) { int err; err = dmu_objset_from_ds(ds, osp); if (err != 0) { dsl_dataset_disown(ds, tag); } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) { dsl_dataset_disown(ds, tag); return (SET_ERROR(EINVAL)); } else if (!readonly && dsl_dataset_is_snapshot(ds)) { dsl_dataset_disown(ds, tag); return (SET_ERROR(EROFS)); } return (err); } /* * dsl_pool must not be held when this is called. * Upon successful return, there will be a longhold on the dataset, * and the dsl_pool will not be held. */ int dmu_objset_own(const char *name, dmu_objset_type_t type, boolean_t readonly, void *tag, objset_t **osp) { dsl_pool_t *dp; dsl_dataset_t *ds; int err; err = dsl_pool_hold(name, FTAG, &dp); if (err != 0) return (err); err = dsl_dataset_own(dp, name, tag, &ds); if (err != 0) { dsl_pool_rele(dp, FTAG); return (err); } err = dmu_objset_own_impl(ds, type, readonly, tag, osp); dsl_pool_rele(dp, FTAG); return (err); } int dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type, boolean_t readonly, void *tag, objset_t **osp) { dsl_dataset_t *ds; int err; err = dsl_dataset_own_obj(dp, obj, tag, &ds); if (err != 0) return (err); return (dmu_objset_own_impl(ds, type, readonly, tag, osp)); } void dmu_objset_rele(objset_t *os, void *tag) { dsl_pool_t *dp = dmu_objset_pool(os); dsl_dataset_rele(os->os_dsl_dataset, tag); dsl_pool_rele(dp, tag); } /* * When we are called, os MUST refer to an objset associated with a dataset * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner * == tag. We will then release and reacquire ownership of the dataset while * holding the pool config_rwlock to avoid intervening namespace or ownership * changes may occur. * * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to * release the hold on its dataset and acquire a new one on the dataset of the * same name so that it can be partially torn down and reconstructed. */ void dmu_objset_refresh_ownership(objset_t *os, void *tag) { dsl_pool_t *dp; dsl_dataset_t *ds, *newds; char name[ZFS_MAX_DATASET_NAME_LEN]; ds = os->os_dsl_dataset; VERIFY3P(ds, !=, NULL); VERIFY3P(ds->ds_owner, ==, tag); VERIFY(dsl_dataset_long_held(ds)); dsl_dataset_name(ds, name); dp = dmu_objset_pool(os); dsl_pool_config_enter(dp, FTAG); dmu_objset_disown(os, tag); VERIFY0(dsl_dataset_own(dp, name, tag, &newds)); VERIFY3P(newds, ==, os->os_dsl_dataset); dsl_pool_config_exit(dp, FTAG); } void dmu_objset_disown(objset_t *os, void *tag) { dsl_dataset_disown(os->os_dsl_dataset, tag); } void dmu_objset_evict_dbufs(objset_t *os) { dnode_t dn_marker; dnode_t *dn; mutex_enter(&os->os_lock); dn = list_head(&os->os_dnodes); while (dn != NULL) { /* * Skip dnodes without holds. We have to do this dance * because dnode_add_ref() only works if there is already a * hold. If the dnode has no holds, then it has no dbufs. */ if (dnode_add_ref(dn, FTAG)) { list_insert_after(&os->os_dnodes, dn, &dn_marker); mutex_exit(&os->os_lock); dnode_evict_dbufs(dn); dnode_rele(dn, FTAG); mutex_enter(&os->os_lock); dn = list_next(&os->os_dnodes, &dn_marker); list_remove(&os->os_dnodes, &dn_marker); } else { dn = list_next(&os->os_dnodes, dn); } } mutex_exit(&os->os_lock); if (DMU_USERUSED_DNODE(os) != NULL) { dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os)); dnode_evict_dbufs(DMU_USERUSED_DNODE(os)); } dnode_evict_dbufs(DMU_META_DNODE(os)); } /* * Objset eviction processing is split into into two pieces. * The first marks the objset as evicting, evicts any dbufs that * have a refcount of zero, and then queues up the objset for the * second phase of eviction. Once os->os_dnodes has been cleared by * dnode_buf_pageout()->dnode_destroy(), the second phase is executed. * The second phase closes the special dnodes, dequeues the objset from * the list of those undergoing eviction, and finally frees the objset. * * NOTE: Due to asynchronous eviction processing (invocation of * dnode_buf_pageout()), it is possible for the meta dnode for the * objset to have no holds even though os->os_dnodes is not empty. */ void dmu_objset_evict(objset_t *os) { dsl_dataset_t *ds = os->os_dsl_dataset; for (int t = 0; t < TXG_SIZE; t++) ASSERT(!dmu_objset_is_dirty(os, t)); if (ds) dsl_prop_unregister_all(ds, os); if (os->os_sa) sa_tear_down(os); dmu_objset_evict_dbufs(os); mutex_enter(&os->os_lock); spa_evicting_os_register(os->os_spa, os); if (list_is_empty(&os->os_dnodes)) { mutex_exit(&os->os_lock); dmu_objset_evict_done(os); } else { mutex_exit(&os->os_lock); } } void dmu_objset_evict_done(objset_t *os) { ASSERT3P(list_head(&os->os_dnodes), ==, NULL); dnode_special_close(&os->os_meta_dnode); if (DMU_USERUSED_DNODE(os)) { dnode_special_close(&os->os_userused_dnode); dnode_special_close(&os->os_groupused_dnode); } zil_free(os->os_zil); arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); /* * This is a barrier to prevent the objset from going away in * dnode_move() until we can safely ensure that the objset is still in * use. We consider the objset valid before the barrier and invalid * after the barrier. */ rw_enter(&os_lock, RW_READER); rw_exit(&os_lock); mutex_destroy(&os->os_lock); + mutex_destroy(&os->os_userused_lock); mutex_destroy(&os->os_obj_lock); mutex_destroy(&os->os_user_ptr_lock); + for (int i = 0; i < TXG_SIZE; i++) { + multilist_destroy(os->os_dirty_dnodes[i]); + } spa_evicting_os_deregister(os->os_spa, os); kmem_free(os, sizeof (objset_t)); } timestruc_t dmu_objset_snap_cmtime(objset_t *os) { return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir)); } /* called from dsl for meta-objset */ objset_t * dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, dmu_objset_type_t type, dmu_tx_t *tx) { objset_t *os; dnode_t *mdn; ASSERT(dmu_tx_is_syncing(tx)); if (ds != NULL) VERIFY0(dmu_objset_from_ds(ds, &os)); else VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os)); mdn = DMU_META_DNODE(os); dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT, DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx); /* * We don't want to have to increase the meta-dnode's nlevels * later, because then we could do it in quescing context while * we are also accessing it in open context. * * This precaution is not necessary for the MOS (ds == NULL), * because the MOS is only updated in syncing context. * This is most fortunate: the MOS is the only objset that * needs to be synced multiple times as spa_sync() iterates * to convergence, so minimizing its dn_nlevels matters. */ if (ds != NULL) { int levels = 1; /* * Determine the number of levels necessary for the meta-dnode * to contain DN_MAX_OBJECT dnodes. Note that in order to * ensure that we do not overflow 64 bits, there has to be * a nlevels that gives us a number of blocks > DN_MAX_OBJECT * but < 2^64. Therefore, * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT) (10) must be * less than (64 - log2(DN_MAX_OBJECT)) (16). */ while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift - DNODE_SHIFT + (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) < DN_MAX_OBJECT) levels++; mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] = mdn->dn_nlevels = levels; } ASSERT(type != DMU_OST_NONE); ASSERT(type != DMU_OST_ANY); ASSERT(type < DMU_OST_NUMTYPES); os->os_phys->os_type = type; if (dmu_objset_userused_enabled(os)) { os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; os->os_flags = os->os_phys->os_flags; } dsl_dataset_dirty(ds, tx); return (os); } typedef struct dmu_objset_create_arg { const char *doca_name; cred_t *doca_cred; void (*doca_userfunc)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx); void *doca_userarg; dmu_objset_type_t doca_type; uint64_t doca_flags; } dmu_objset_create_arg_t; /*ARGSUSED*/ static int dmu_objset_create_check(void *arg, dmu_tx_t *tx) { dmu_objset_create_arg_t *doca = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dir_t *pdd; const char *tail; int error; if (strchr(doca->doca_name, '@') != NULL) return (SET_ERROR(EINVAL)); if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN) return (SET_ERROR(ENAMETOOLONG)); error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail); if (error != 0) return (error); if (tail == NULL) { dsl_dir_rele(pdd, FTAG); return (SET_ERROR(EEXIST)); } error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, doca->doca_cred); dsl_dir_rele(pdd, FTAG); return (error); } static void dmu_objset_create_sync(void *arg, dmu_tx_t *tx) { dmu_objset_create_arg_t *doca = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dir_t *pdd; const char *tail; dsl_dataset_t *ds; uint64_t obj; blkptr_t *bp; objset_t *os; VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail)); obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags, doca->doca_cred, tx); VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds)); rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); bp = dsl_dataset_get_blkptr(ds); os = dmu_objset_create_impl(pdd->dd_pool->dp_spa, ds, bp, doca->doca_type, tx); rrw_exit(&ds->ds_bp_rwlock, FTAG); if (doca->doca_userfunc != NULL) { doca->doca_userfunc(os, doca->doca_userarg, doca->doca_cred, tx); } spa_history_log_internal_ds(ds, "create", tx, ""); dsl_dataset_rele(ds, FTAG); dsl_dir_rele(pdd, FTAG); } int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags, void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg) { dmu_objset_create_arg_t doca; doca.doca_name = name; doca.doca_cred = CRED(); doca.doca_flags = flags; doca.doca_userfunc = func; doca.doca_userarg = arg; doca.doca_type = type; return (dsl_sync_task(name, dmu_objset_create_check, dmu_objset_create_sync, &doca, 5, ZFS_SPACE_CHECK_NORMAL)); } typedef struct dmu_objset_clone_arg { const char *doca_clone; const char *doca_origin; cred_t *doca_cred; } dmu_objset_clone_arg_t; /*ARGSUSED*/ static int dmu_objset_clone_check(void *arg, dmu_tx_t *tx) { dmu_objset_clone_arg_t *doca = arg; dsl_dir_t *pdd; const char *tail; int error; dsl_dataset_t *origin; dsl_pool_t *dp = dmu_tx_pool(tx); if (strchr(doca->doca_clone, '@') != NULL) return (SET_ERROR(EINVAL)); if (strlen(doca->doca_clone) >= ZFS_MAX_DATASET_NAME_LEN) return (SET_ERROR(ENAMETOOLONG)); error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail); if (error != 0) return (error); if (tail == NULL) { dsl_dir_rele(pdd, FTAG); return (SET_ERROR(EEXIST)); } error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, doca->doca_cred); if (error != 0) { dsl_dir_rele(pdd, FTAG); return (SET_ERROR(EDQUOT)); } dsl_dir_rele(pdd, FTAG); error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin); if (error != 0) return (error); /* You can only clone snapshots, not the head datasets. */ if (!origin->ds_is_snapshot) { dsl_dataset_rele(origin, FTAG); return (SET_ERROR(EINVAL)); } dsl_dataset_rele(origin, FTAG); return (0); } static void dmu_objset_clone_sync(void *arg, dmu_tx_t *tx) { dmu_objset_clone_arg_t *doca = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dir_t *pdd; const char *tail; dsl_dataset_t *origin, *ds; uint64_t obj; char namebuf[ZFS_MAX_DATASET_NAME_LEN]; VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail)); VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin)); obj = dsl_dataset_create_sync(pdd, tail, origin, 0, doca->doca_cred, tx); VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds)); dsl_dataset_name(origin, namebuf); spa_history_log_internal_ds(ds, "clone", tx, "origin=%s (%llu)", namebuf, origin->ds_object); dsl_dataset_rele(ds, FTAG); dsl_dataset_rele(origin, FTAG); dsl_dir_rele(pdd, FTAG); } int dmu_objset_clone(const char *clone, const char *origin) { dmu_objset_clone_arg_t doca; doca.doca_clone = clone; doca.doca_origin = origin; doca.doca_cred = CRED(); return (dsl_sync_task(clone, dmu_objset_clone_check, dmu_objset_clone_sync, &doca, 5, ZFS_SPACE_CHECK_NORMAL)); } int dmu_objset_snapshot_one(const char *fsname, const char *snapname) { int err; char *longsnap = kmem_asprintf("%s@%s", fsname, snapname); nvlist_t *snaps = fnvlist_alloc(); fnvlist_add_boolean(snaps, longsnap); strfree(longsnap); err = dsl_dataset_snapshot(snaps, NULL, NULL); fnvlist_free(snaps); return (err); } static void -dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx) +dmu_objset_sync_dnodes(multilist_sublist_t *list, dmu_tx_t *tx) { dnode_t *dn; - while (dn = list_head(list)) { + while ((dn = multilist_sublist_head(list)) != NULL) { ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); ASSERT(dn->dn_dbuf->db_data_pending); /* * Initialize dn_zio outside dnode_sync() because the * meta-dnode needs to set it ouside dnode_sync(). */ dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio; ASSERT(dn->dn_zio); ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS); - list_remove(list, dn); + multilist_sublist_remove(list, dn); - if (newlist) { + multilist_t *newlist = dn->dn_objset->os_synced_dnodes; + if (newlist != NULL) { (void) dnode_add_ref(dn, newlist); - list_insert_tail(newlist, dn); + multilist_insert(newlist, dn); } dnode_sync(dn, tx); } } /* ARGSUSED */ static void dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg) { blkptr_t *bp = zio->io_bp; objset_t *os = arg; dnode_phys_t *dnp = &os->os_phys->os_meta_dnode; ASSERT(!BP_IS_EMBEDDED(bp)); ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET); ASSERT0(BP_GET_LEVEL(bp)); /* * Update rootbp fill count: it should be the number of objects * allocated in the object set (not counting the "special" * objects that are stored in the objset_phys_t -- the meta * dnode and user/group accounting objects). */ bp->blk_fill = 0; for (int i = 0; i < dnp->dn_nblkptr; i++) bp->blk_fill += BP_GET_FILL(&dnp->dn_blkptr[i]); if (os->os_dsl_dataset != NULL) rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_WRITER, FTAG); *os->os_rootbp = *bp; if (os->os_dsl_dataset != NULL) rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG); } /* ARGSUSED */ static void dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg) { blkptr_t *bp = zio->io_bp; blkptr_t *bp_orig = &zio->io_bp_orig; objset_t *os = arg; if (zio->io_flags & ZIO_FLAG_IO_REWRITE) { ASSERT(BP_EQUAL(bp, bp_orig)); } else { dsl_dataset_t *ds = os->os_dsl_dataset; dmu_tx_t *tx = os->os_synctx; (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE); dsl_dataset_block_born(ds, bp, tx); } kmem_free(bp, sizeof (*bp)); } +typedef struct sync_dnodes_arg { + multilist_t *sda_list; + int sda_sublist_idx; + multilist_t *sda_newlist; + dmu_tx_t *sda_tx; +} sync_dnodes_arg_t; + +static void +sync_dnodes_task(void *arg) +{ + sync_dnodes_arg_t *sda = arg; + + multilist_sublist_t *ms = + multilist_sublist_lock(sda->sda_list, sda->sda_sublist_idx); + + dmu_objset_sync_dnodes(ms, sda->sda_tx); + + multilist_sublist_unlock(ms); + + kmem_free(sda, sizeof (*sda)); +} + + /* called from dsl */ void dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx) { int txgoff; zbookmark_phys_t zb; zio_prop_t zp; zio_t *zio; list_t *list; - list_t *newlist = NULL; dbuf_dirty_record_t *dr; blkptr_t *blkptr_copy = kmem_alloc(sizeof (*os->os_rootbp), KM_SLEEP); *blkptr_copy = *os->os_rootbp; dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg); ASSERT(dmu_tx_is_syncing(tx)); /* XXX the write_done callback should really give us the tx... */ os->os_synctx = tx; if (os->os_dsl_dataset == NULL) { /* * This is the MOS. If we have upgraded, * spa_max_replication() could change, so reset * os_copies here. */ os->os_copies = spa_max_replication(os->os_spa); } /* * Create the root block IO */ SET_BOOKMARK(&zb, os->os_dsl_dataset ? os->os_dsl_dataset->ds_object : DMU_META_OBJSET, ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); arc_release(os->os_phys_buf, &os->os_phys_buf); dmu_write_policy(os, NULL, 0, 0, ZIO_COMPRESS_INHERIT, &zp); zio = arc_write(pio, os->os_spa, tx->tx_txg, blkptr_copy, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os), &zp, dmu_objset_write_ready, NULL, NULL, dmu_objset_write_done, os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); /* * Sync special dnodes - the parent IO for the sync is the root block */ DMU_META_DNODE(os)->dn_zio = zio; dnode_sync(DMU_META_DNODE(os), tx); os->os_phys->os_flags = os->os_flags; if (DMU_USERUSED_DNODE(os) && DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) { DMU_USERUSED_DNODE(os)->dn_zio = zio; dnode_sync(DMU_USERUSED_DNODE(os), tx); DMU_GROUPUSED_DNODE(os)->dn_zio = zio; dnode_sync(DMU_GROUPUSED_DNODE(os), tx); } txgoff = tx->tx_txg & TXG_MASK; if (dmu_objset_userused_enabled(os)) { - newlist = &os->os_synced_dnodes; /* * We must create the list here because it uses the - * dn_dirty_link[] of this txg. + * dn_dirty_link[] of this txg. But it may already + * exist because we call dsl_dataset_sync() twice per txg. */ - list_create(newlist, sizeof (dnode_t), - offsetof(dnode_t, dn_dirty_link[txgoff])); + if (os->os_synced_dnodes == NULL) { + os->os_synced_dnodes = + multilist_create(sizeof (dnode_t), + offsetof(dnode_t, dn_dirty_link[txgoff]), + dnode_multilist_index_func); + } else { + ASSERT3U(os->os_synced_dnodes->ml_offset, ==, + offsetof(dnode_t, dn_dirty_link[txgoff])); + } } - dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx); - dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx); + for (int i = 0; + i < multilist_get_num_sublists(os->os_dirty_dnodes[txgoff]); i++) { + sync_dnodes_arg_t *sda = kmem_alloc(sizeof (*sda), KM_SLEEP); + sda->sda_list = os->os_dirty_dnodes[txgoff]; + sda->sda_sublist_idx = i; + sda->sda_tx = tx; + (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, + sync_dnodes_task, sda, 0); + /* callback frees sda */ + } + taskq_wait(dmu_objset_pool(os)->dp_sync_taskq); list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff]; - while (dr = list_head(list)) { + while ((dr = list_head(list)) != NULL) { ASSERT0(dr->dr_dbuf->db_level); list_remove(list, dr); if (dr->dr_zio) zio_nowait(dr->dr_zio); } /* Enable dnode backfill if enough objects have been freed. */ if (os->os_freed_dnodes >= dmu_rescan_dnode_threshold) { os->os_rescan_dnodes = B_TRUE; os->os_freed_dnodes = 0; } /* * Free intent log blocks up to this tx. */ zil_sync(os->os_zil, tx); os->os_phys->os_zil_header = os->os_zil_header; zio_nowait(zio); } boolean_t dmu_objset_is_dirty(objset_t *os, uint64_t txg) { - return (!list_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]) || - !list_is_empty(&os->os_free_dnodes[txg & TXG_MASK])); + return (!multilist_is_empty(os->os_dirty_dnodes[txg & TXG_MASK])); } static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES]; void dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb) { used_cbs[ost] = cb; } boolean_t dmu_objset_userused_enabled(objset_t *os) { return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE && used_cbs[os->os_phys->os_type] != NULL && DMU_USERUSED_DNODE(os) != NULL); } typedef struct userquota_node { uint64_t uqn_id; int64_t uqn_delta; avl_node_t uqn_node; } userquota_node_t; typedef struct userquota_cache { avl_tree_t uqc_user_deltas; avl_tree_t uqc_group_deltas; } userquota_cache_t; static int userquota_compare(const void *l, const void *r) { const userquota_node_t *luqn = l; const userquota_node_t *ruqn = r; if (luqn->uqn_id < ruqn->uqn_id) return (-1); if (luqn->uqn_id > ruqn->uqn_id) return (1); return (0); } static void do_userquota_cacheflush(objset_t *os, userquota_cache_t *cache, dmu_tx_t *tx) { void *cookie; userquota_node_t *uqn; ASSERT(dmu_tx_is_syncing(tx)); cookie = NULL; while ((uqn = avl_destroy_nodes(&cache->uqc_user_deltas, &cookie)) != NULL) { + /* + * os_userused_lock protects against concurrent calls to + * zap_increment_int(). It's needed because zap_increment_int() + * is not thread-safe (i.e. not atomic). + */ + mutex_enter(&os->os_userused_lock); VERIFY0(zap_increment_int(os, DMU_USERUSED_OBJECT, uqn->uqn_id, uqn->uqn_delta, tx)); + mutex_exit(&os->os_userused_lock); kmem_free(uqn, sizeof (*uqn)); } avl_destroy(&cache->uqc_user_deltas); cookie = NULL; while ((uqn = avl_destroy_nodes(&cache->uqc_group_deltas, &cookie)) != NULL) { + mutex_enter(&os->os_userused_lock); VERIFY0(zap_increment_int(os, DMU_GROUPUSED_OBJECT, uqn->uqn_id, uqn->uqn_delta, tx)); + mutex_exit(&os->os_userused_lock); kmem_free(uqn, sizeof (*uqn)); } avl_destroy(&cache->uqc_group_deltas); } static void userquota_update_cache(avl_tree_t *avl, uint64_t id, int64_t delta) { userquota_node_t search = { .uqn_id = id }; avl_index_t idx; userquota_node_t *uqn = avl_find(avl, &search, &idx); if (uqn == NULL) { uqn = kmem_zalloc(sizeof (*uqn), KM_SLEEP); uqn->uqn_id = id; avl_insert(avl, uqn, idx); } uqn->uqn_delta += delta; } static void do_userquota_update(userquota_cache_t *cache, uint64_t used, uint64_t flags, uint64_t user, uint64_t group, boolean_t subtract) { if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) { int64_t delta = DNODE_SIZE + used; if (subtract) delta = -delta; userquota_update_cache(&cache->uqc_user_deltas, user, delta); userquota_update_cache(&cache->uqc_group_deltas, group, delta); } } -void -dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx) +typedef struct userquota_updates_arg { + objset_t *uua_os; + int uua_sublist_idx; + dmu_tx_t *uua_tx; +} userquota_updates_arg_t; + +static void +userquota_updates_task(void *arg) { + userquota_updates_arg_t *uua = arg; + objset_t *os = uua->uua_os; + dmu_tx_t *tx = uua->uua_tx; dnode_t *dn; - list_t *list = &os->os_synced_dnodes; userquota_cache_t cache = { 0 }; - ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os)); + multilist_sublist_t *list = + multilist_sublist_lock(os->os_synced_dnodes, uua->uua_sublist_idx); + ASSERT(multilist_sublist_head(list) == NULL || + dmu_objset_userused_enabled(os)); avl_create(&cache.uqc_user_deltas, userquota_compare, sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); avl_create(&cache.uqc_group_deltas, userquota_compare, sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); - while (dn = list_head(list)) { + while ((dn = multilist_sublist_head(list)) != NULL) { int flags; ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object)); ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE || dn->dn_phys->dn_flags & DNODE_FLAG_USERUSED_ACCOUNTED); - /* Allocate the user/groupused objects if necessary. */ - if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) { - VERIFY0(zap_create_claim(os, - DMU_USERUSED_OBJECT, - DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); - VERIFY0(zap_create_claim(os, - DMU_GROUPUSED_OBJECT, - DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); - } - flags = dn->dn_id_flags; ASSERT(flags); if (flags & DN_ID_OLD_EXIST) { do_userquota_update(&cache, dn->dn_oldused, dn->dn_oldflags, dn->dn_olduid, dn->dn_oldgid, B_TRUE); } if (flags & DN_ID_NEW_EXIST) { do_userquota_update(&cache, DN_USED_BYTES(dn->dn_phys), dn->dn_phys->dn_flags, dn->dn_newuid, dn->dn_newgid, B_FALSE); } mutex_enter(&dn->dn_mtx); dn->dn_oldused = 0; dn->dn_oldflags = 0; if (dn->dn_id_flags & DN_ID_NEW_EXIST) { dn->dn_olduid = dn->dn_newuid; dn->dn_oldgid = dn->dn_newgid; dn->dn_id_flags |= DN_ID_OLD_EXIST; if (dn->dn_bonuslen == 0) dn->dn_id_flags |= DN_ID_CHKED_SPILL; else dn->dn_id_flags |= DN_ID_CHKED_BONUS; } dn->dn_id_flags &= ~(DN_ID_NEW_EXIST); mutex_exit(&dn->dn_mtx); - list_remove(list, dn); - dnode_rele(dn, list); + multilist_sublist_remove(list, dn); + dnode_rele(dn, os->os_synced_dnodes); } do_userquota_cacheflush(os, &cache, tx); + multilist_sublist_unlock(list); + kmem_free(uua, sizeof (*uua)); +} + +void +dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx) +{ + if (!dmu_objset_userused_enabled(os)) + return; + + /* Allocate the user/groupused objects if necessary. */ + if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) { + VERIFY0(zap_create_claim(os, + DMU_USERUSED_OBJECT, + DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); + VERIFY0(zap_create_claim(os, + DMU_GROUPUSED_OBJECT, + DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); + } + + for (int i = 0; + i < multilist_get_num_sublists(os->os_synced_dnodes); i++) { + userquota_updates_arg_t *uua = + kmem_alloc(sizeof (*uua), KM_SLEEP); + uua->uua_os = os; + uua->uua_sublist_idx = i; + uua->uua_tx = tx; + /* note: caller does taskq_wait() */ + (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, + userquota_updates_task, uua, 0); + /* callback frees uua */ + } } /* * Returns a pointer to data to find uid/gid from * * If a dirty record for transaction group that is syncing can't * be found then NULL is returned. In the NULL case it is assumed * the uid/gid aren't changing. */ static void * dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx) { dbuf_dirty_record_t *dr, **drp; void *data; if (db->db_dirtycnt == 0) return (db->db.db_data); /* Nothing is changing */ for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next) if (dr->dr_txg == tx->tx_txg) break; if (dr == NULL) { data = NULL; } else { dnode_t *dn; DB_DNODE_ENTER(dr->dr_dbuf); dn = DB_DNODE(dr->dr_dbuf); if (dn->dn_bonuslen == 0 && dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID) data = dr->dt.dl.dr_data->b_data; else data = dr->dt.dl.dr_data; DB_DNODE_EXIT(dr->dr_dbuf); } return (data); } void dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx) { objset_t *os = dn->dn_objset; void *data = NULL; dmu_buf_impl_t *db = NULL; uint64_t *user = NULL; uint64_t *group = NULL; int flags = dn->dn_id_flags; int error; boolean_t have_spill = B_FALSE; if (!dmu_objset_userused_enabled(dn->dn_objset)) return; if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST| DN_ID_CHKED_SPILL))) return; if (before && dn->dn_bonuslen != 0) data = DN_BONUS(dn->dn_phys); else if (!before && dn->dn_bonuslen != 0) { if (dn->dn_bonus) { db = dn->dn_bonus; mutex_enter(&db->db_mtx); data = dmu_objset_userquota_find_data(db, tx); } else { data = DN_BONUS(dn->dn_phys); } } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) { int rf = 0; if (RW_WRITE_HELD(&dn->dn_struct_rwlock)) rf |= DB_RF_HAVESTRUCT; error = dmu_spill_hold_by_dnode(dn, rf | DB_RF_MUST_SUCCEED, FTAG, (dmu_buf_t **)&db); ASSERT(error == 0); mutex_enter(&db->db_mtx); data = (before) ? db->db.db_data : dmu_objset_userquota_find_data(db, tx); have_spill = B_TRUE; } else { mutex_enter(&dn->dn_mtx); dn->dn_id_flags |= DN_ID_CHKED_BONUS; mutex_exit(&dn->dn_mtx); return; } if (before) { ASSERT(data); user = &dn->dn_olduid; group = &dn->dn_oldgid; } else if (data) { user = &dn->dn_newuid; group = &dn->dn_newgid; } /* * Must always call the callback in case the object * type has changed and that type isn't an object type to track */ error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data, user, group); /* * Preserve existing uid/gid when the callback can't determine * what the new uid/gid are and the callback returned EEXIST. * The EEXIST error tells us to just use the existing uid/gid. * If we don't know what the old values are then just assign * them to 0, since that is a new file being created. */ if (!before && data == NULL && error == EEXIST) { if (flags & DN_ID_OLD_EXIST) { dn->dn_newuid = dn->dn_olduid; dn->dn_newgid = dn->dn_oldgid; } else { dn->dn_newuid = 0; dn->dn_newgid = 0; } error = 0; } if (db) mutex_exit(&db->db_mtx); mutex_enter(&dn->dn_mtx); if (error == 0 && before) dn->dn_id_flags |= DN_ID_OLD_EXIST; if (error == 0 && !before) dn->dn_id_flags |= DN_ID_NEW_EXIST; if (have_spill) { dn->dn_id_flags |= DN_ID_CHKED_SPILL; } else { dn->dn_id_flags |= DN_ID_CHKED_BONUS; } mutex_exit(&dn->dn_mtx); if (have_spill) dmu_buf_rele((dmu_buf_t *)db, FTAG); } boolean_t dmu_objset_userspace_present(objset_t *os) { return (os->os_phys->os_flags & OBJSET_FLAG_USERACCOUNTING_COMPLETE); } int dmu_objset_userspace_upgrade(objset_t *os) { uint64_t obj; int err = 0; if (dmu_objset_userspace_present(os)) return (0); if (!dmu_objset_userused_enabled(os)) return (SET_ERROR(ENOTSUP)); if (dmu_objset_is_snapshot(os)) return (SET_ERROR(EINVAL)); /* * We simply need to mark every object dirty, so that it will be * synced out and now accounted. If this is called * concurrently, or if we already did some work before crashing, * that's fine, since we track each object's accounted state * independently. */ for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) { dmu_tx_t *tx; dmu_buf_t *db; int objerr; if (issig(JUSTLOOKING) && issig(FORREAL)) return (SET_ERROR(EINTR)); objerr = dmu_bonus_hold(os, obj, FTAG, &db); if (objerr != 0) continue; tx = dmu_tx_create(os); dmu_tx_hold_bonus(tx, obj); objerr = dmu_tx_assign(tx, TXG_WAIT); if (objerr != 0) { dmu_tx_abort(tx); continue; } dmu_buf_will_dirty(db, tx); dmu_buf_rele(db, FTAG); dmu_tx_commit(tx); } os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; txg_wait_synced(dmu_objset_pool(os), 0); return (0); } void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp, uint64_t *usedobjsp, uint64_t *availobjsp) { dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp, usedobjsp, availobjsp); } uint64_t dmu_objset_fsid_guid(objset_t *os) { return (dsl_dataset_fsid_guid(os->os_dsl_dataset)); } void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat) { stat->dds_type = os->os_phys->os_type; if (os->os_dsl_dataset) dsl_dataset_fast_stat(os->os_dsl_dataset, stat); } void dmu_objset_stats(objset_t *os, nvlist_t *nv) { ASSERT(os->os_dsl_dataset || os->os_phys->os_type == DMU_OST_META); if (os->os_dsl_dataset != NULL) dsl_dataset_stats(os->os_dsl_dataset, nv); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE, os->os_phys->os_type); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING, dmu_objset_userspace_present(os)); } int dmu_objset_is_snapshot(objset_t *os) { if (os->os_dsl_dataset != NULL) return (os->os_dsl_dataset->ds_is_snapshot); else return (B_FALSE); } int dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen, boolean_t *conflict) { dsl_dataset_t *ds = os->os_dsl_dataset; uint64_t ignored; if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) return (SET_ERROR(ENOENT)); return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset, dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored, MT_NORMALIZE, real, maxlen, conflict)); } int dmu_snapshot_list_next(objset_t *os, int namelen, char *name, uint64_t *idp, uint64_t *offp, boolean_t *case_conflict) { dsl_dataset_t *ds = os->os_dsl_dataset; zap_cursor_t cursor; zap_attribute_t attr; ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) return (SET_ERROR(ENOENT)); zap_cursor_init_serialized(&cursor, ds->ds_dir->dd_pool->dp_meta_objset, dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp); if (zap_cursor_retrieve(&cursor, &attr) != 0) { zap_cursor_fini(&cursor); return (SET_ERROR(ENOENT)); } if (strlen(attr.za_name) + 1 > namelen) { zap_cursor_fini(&cursor); return (SET_ERROR(ENAMETOOLONG)); } (void) strcpy(name, attr.za_name); if (idp) *idp = attr.za_first_integer; if (case_conflict) *case_conflict = attr.za_normalization_conflict; zap_cursor_advance(&cursor); *offp = zap_cursor_serialize(&cursor); zap_cursor_fini(&cursor); return (0); } int dmu_dir_list_next(objset_t *os, int namelen, char *name, uint64_t *idp, uint64_t *offp) { dsl_dir_t *dd = os->os_dsl_dataset->ds_dir; zap_cursor_t cursor; zap_attribute_t attr; /* there is no next dir on a snapshot! */ if (os->os_dsl_dataset->ds_object != dsl_dir_phys(dd)->dd_head_dataset_obj) return (SET_ERROR(ENOENT)); zap_cursor_init_serialized(&cursor, dd->dd_pool->dp_meta_objset, dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp); if (zap_cursor_retrieve(&cursor, &attr) != 0) { zap_cursor_fini(&cursor); return (SET_ERROR(ENOENT)); } if (strlen(attr.za_name) + 1 > namelen) { zap_cursor_fini(&cursor); return (SET_ERROR(ENAMETOOLONG)); } (void) strcpy(name, attr.za_name); if (idp) *idp = attr.za_first_integer; zap_cursor_advance(&cursor); *offp = zap_cursor_serialize(&cursor); zap_cursor_fini(&cursor); return (0); } typedef struct dmu_objset_find_ctx { taskq_t *dc_tq; dsl_pool_t *dc_dp; uint64_t dc_ddobj; char *dc_ddname; /* last component of ddobj's name */ int (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *); void *dc_arg; int dc_flags; kmutex_t *dc_error_lock; int *dc_error; } dmu_objset_find_ctx_t; static void dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp) { dsl_pool_t *dp = dcp->dc_dp; dsl_dir_t *dd; dsl_dataset_t *ds; zap_cursor_t zc; zap_attribute_t *attr; uint64_t thisobj; int err = 0; /* don't process if there already was an error */ if (*dcp->dc_error != 0) goto out; /* * Note: passing the name (dc_ddname) here is optional, but it * improves performance because we don't need to call * zap_value_search() to determine the name. */ err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, dcp->dc_ddname, FTAG, &dd); if (err != 0) goto out; /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ if (dd->dd_myname[0] == '$') { dsl_dir_rele(dd, FTAG); goto out; } thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); /* * Iterate over all children. */ if (dcp->dc_flags & DS_FIND_CHILDREN) { for (zap_cursor_init(&zc, dp->dp_meta_objset, dsl_dir_phys(dd)->dd_child_dir_zapobj); zap_cursor_retrieve(&zc, attr) == 0; (void) zap_cursor_advance(&zc)) { ASSERT3U(attr->za_integer_length, ==, sizeof (uint64_t)); ASSERT3U(attr->za_num_integers, ==, 1); dmu_objset_find_ctx_t *child_dcp = kmem_alloc(sizeof (*child_dcp), KM_SLEEP); *child_dcp = *dcp; child_dcp->dc_ddobj = attr->za_first_integer; child_dcp->dc_ddname = spa_strdup(attr->za_name); if (dcp->dc_tq != NULL) (void) taskq_dispatch(dcp->dc_tq, dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP); else dmu_objset_find_dp_impl(child_dcp); } zap_cursor_fini(&zc); } /* * Iterate over all snapshots. */ if (dcp->dc_flags & DS_FIND_SNAPSHOTS) { dsl_dataset_t *ds; err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); if (err == 0) { uint64_t snapobj; snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; dsl_dataset_rele(ds, FTAG); for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); zap_cursor_retrieve(&zc, attr) == 0; (void) zap_cursor_advance(&zc)) { ASSERT3U(attr->za_integer_length, ==, sizeof (uint64_t)); ASSERT3U(attr->za_num_integers, ==, 1); err = dsl_dataset_hold_obj(dp, attr->za_first_integer, FTAG, &ds); if (err != 0) break; err = dcp->dc_func(dp, ds, dcp->dc_arg); dsl_dataset_rele(ds, FTAG); if (err != 0) break; } zap_cursor_fini(&zc); } } kmem_free(attr, sizeof (zap_attribute_t)); if (err != 0) { dsl_dir_rele(dd, FTAG); goto out; } /* * Apply to self. */ err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); /* * Note: we hold the dir while calling dsl_dataset_hold_obj() so * that the dir will remain cached, and we won't have to re-instantiate * it (which could be expensive due to finding its name via * zap_value_search()). */ dsl_dir_rele(dd, FTAG); if (err != 0) goto out; err = dcp->dc_func(dp, ds, dcp->dc_arg); dsl_dataset_rele(ds, FTAG); out: if (err != 0) { mutex_enter(dcp->dc_error_lock); /* only keep first error */ if (*dcp->dc_error == 0) *dcp->dc_error = err; mutex_exit(dcp->dc_error_lock); } if (dcp->dc_ddname != NULL) spa_strfree(dcp->dc_ddname); kmem_free(dcp, sizeof (*dcp)); } static void dmu_objset_find_dp_cb(void *arg) { dmu_objset_find_ctx_t *dcp = arg; dsl_pool_t *dp = dcp->dc_dp; /* * We need to get a pool_config_lock here, as there are several * asssert(pool_config_held) down the stack. Getting a lock via * dsl_pool_config_enter is risky, as it might be stalled by a * pending writer. This would deadlock, as the write lock can * only be granted when our parent thread gives up the lock. * The _prio interface gives us priority over a pending writer. */ dsl_pool_config_enter_prio(dp, FTAG); dmu_objset_find_dp_impl(dcp); dsl_pool_config_exit(dp, FTAG); } /* * Find objsets under and including ddobj, call func(ds) on each. * The order for the enumeration is completely undefined. * func is called with dsl_pool_config held. */ int dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj, int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags) { int error = 0; taskq_t *tq = NULL; int ntasks; dmu_objset_find_ctx_t *dcp; kmutex_t err_lock; mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL); dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP); dcp->dc_tq = NULL; dcp->dc_dp = dp; dcp->dc_ddobj = ddobj; dcp->dc_ddname = NULL; dcp->dc_func = func; dcp->dc_arg = arg; dcp->dc_flags = flags; dcp->dc_error_lock = &err_lock; dcp->dc_error = &error; if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) { /* * In case a write lock is held we can't make use of * parallelism, as down the stack of the worker threads * the lock is asserted via dsl_pool_config_held. * In case of a read lock this is solved by getting a read * lock in each worker thread, which isn't possible in case * of a writer lock. So we fall back to the synchronous path * here. * In the future it might be possible to get some magic into * dsl_pool_config_held in a way that it returns true for * the worker threads so that a single lock held from this * thread suffices. For now, stay single threaded. */ dmu_objset_find_dp_impl(dcp); mutex_destroy(&err_lock); return (error); } ntasks = dmu_find_threads; if (ntasks == 0) ntasks = vdev_count_leaves(dp->dp_spa) * 4; tq = taskq_create("dmu_objset_find", ntasks, minclsyspri, ntasks, INT_MAX, 0); if (tq == NULL) { kmem_free(dcp, sizeof (*dcp)); mutex_destroy(&err_lock); return (SET_ERROR(ENOMEM)); } dcp->dc_tq = tq; /* dcp will be freed by task */ (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP); /* * PORTING: this code relies on the property of taskq_wait to wait * until no more tasks are queued and no more tasks are active. As * we always queue new tasks from within other tasks, task_wait * reliably waits for the full recursion to finish, even though we * enqueue new tasks after taskq_wait has been called. * On platforms other than illumos, taskq_wait may not have this * property. */ taskq_wait(tq); taskq_destroy(tq); mutex_destroy(&err_lock); return (error); } /* * Find all objsets under name, and for each, call 'func(child_name, arg)'. * The dp_config_rwlock must not be held when this is called, and it * will not be held when the callback is called. * Therefore this function should only be used when the pool is not changing * (e.g. in syncing context), or the callback can deal with the possible races. */ static int dmu_objset_find_impl(spa_t *spa, const char *name, int func(const char *, void *), void *arg, int flags) { dsl_dir_t *dd; dsl_pool_t *dp = spa_get_dsl(spa); dsl_dataset_t *ds; zap_cursor_t zc; zap_attribute_t *attr; char *child; uint64_t thisobj; int err; dsl_pool_config_enter(dp, FTAG); err = dsl_dir_hold(dp, name, FTAG, &dd, NULL); if (err != 0) { dsl_pool_config_exit(dp, FTAG); return (err); } /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ if (dd->dd_myname[0] == '$') { dsl_dir_rele(dd, FTAG); dsl_pool_config_exit(dp, FTAG); return (0); } thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); /* * Iterate over all children. */ if (flags & DS_FIND_CHILDREN) { for (zap_cursor_init(&zc, dp->dp_meta_objset, dsl_dir_phys(dd)->dd_child_dir_zapobj); zap_cursor_retrieve(&zc, attr) == 0; (void) zap_cursor_advance(&zc)) { ASSERT3U(attr->za_integer_length, ==, sizeof (uint64_t)); ASSERT3U(attr->za_num_integers, ==, 1); child = kmem_asprintf("%s/%s", name, attr->za_name); dsl_pool_config_exit(dp, FTAG); err = dmu_objset_find_impl(spa, child, func, arg, flags); dsl_pool_config_enter(dp, FTAG); strfree(child); if (err != 0) break; } zap_cursor_fini(&zc); if (err != 0) { dsl_dir_rele(dd, FTAG); dsl_pool_config_exit(dp, FTAG); kmem_free(attr, sizeof (zap_attribute_t)); return (err); } } /* * Iterate over all snapshots. */ if (flags & DS_FIND_SNAPSHOTS) { err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); if (err == 0) { uint64_t snapobj; snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; dsl_dataset_rele(ds, FTAG); for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); zap_cursor_retrieve(&zc, attr) == 0; (void) zap_cursor_advance(&zc)) { ASSERT3U(attr->za_integer_length, ==, sizeof (uint64_t)); ASSERT3U(attr->za_num_integers, ==, 1); child = kmem_asprintf("%s@%s", name, attr->za_name); dsl_pool_config_exit(dp, FTAG); err = func(child, arg); dsl_pool_config_enter(dp, FTAG); strfree(child); if (err != 0) break; } zap_cursor_fini(&zc); } } dsl_dir_rele(dd, FTAG); kmem_free(attr, sizeof (zap_attribute_t)); dsl_pool_config_exit(dp, FTAG); if (err != 0) return (err); /* Apply to self. */ return (func(name, arg)); } /* * See comment above dmu_objset_find_impl(). */ int dmu_objset_find(char *name, int func(const char *, void *), void *arg, int flags) { spa_t *spa; int error; error = spa_open(name, &spa, FTAG); if (error != 0) return (error); error = dmu_objset_find_impl(spa, name, func, arg, flags); spa_close(spa, FTAG); return (error); } void dmu_objset_set_user(objset_t *os, void *user_ptr) { ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); os->os_user_ptr = user_ptr; } void * dmu_objset_get_user(objset_t *os) { ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); return (os->os_user_ptr); } /* * Determine name of filesystem, given name of snapshot. * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes */ int dmu_fsname(const char *snapname, char *buf) { char *atp = strchr(snapname, '@'); if (atp == NULL) return (SET_ERROR(EINVAL)); if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN) return (SET_ERROR(ENAMETOOLONG)); (void) strlcpy(buf, snapname, atp - snapname + 1); return (0); } /* * Call when we think we're going to write/free space in open context to track * the amount of dirty data in the open txg, which is also the amount * of memory that can not be evicted until this txg syncs. */ void dmu_objset_willuse_space(objset_t *os, int64_t space, dmu_tx_t *tx) { dsl_dataset_t *ds = os->os_dsl_dataset; int64_t aspace = spa_get_worst_case_asize(os->os_spa, space); if (ds != NULL) { dsl_dir_willuse_space(ds->ds_dir, aspace, tx); dsl_pool_dirty_space(dmu_tx_pool(tx), space, tx); } } Index: vendor-sys/illumos/dist/uts/common/fs/zfs/dnode.c =================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/dnode.c (revision 316916) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/dnode.c (revision 316917) @@ -1,2015 +1,1993 @@ /* * 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 http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. - * Copyright (c) 2012, 2016 by Delphix. All rights reserved. + * Copyright (c) 2012, 2017 by Delphix. All rights reserved. * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. * Copyright (c) 2014 Integros [integros.com] */ #include #include #include #include #include #include #include #include #include #include #include #include #include static kmem_cache_t *dnode_cache; /* * Define DNODE_STATS to turn on statistic gathering. By default, it is only * turned on when DEBUG is also defined. */ #ifdef DEBUG #define DNODE_STATS #endif /* DEBUG */ #ifdef DNODE_STATS #define DNODE_STAT_ADD(stat) ((stat)++) #else #define DNODE_STAT_ADD(stat) /* nothing */ #endif /* DNODE_STATS */ static dnode_phys_t dnode_phys_zero; int zfs_default_bs = SPA_MINBLOCKSHIFT; int zfs_default_ibs = DN_MAX_INDBLKSHIFT; static kmem_cbrc_t dnode_move(void *, void *, size_t, void *); static int dbuf_compare(const void *x1, const void *x2) { const dmu_buf_impl_t *d1 = x1; const dmu_buf_impl_t *d2 = x2; if (d1->db_level < d2->db_level) { return (-1); } if (d1->db_level > d2->db_level) { return (1); } if (d1->db_blkid < d2->db_blkid) { return (-1); } if (d1->db_blkid > d2->db_blkid) { return (1); } if (d1->db_state == DB_SEARCH) { ASSERT3S(d2->db_state, !=, DB_SEARCH); return (-1); } else if (d2->db_state == DB_SEARCH) { ASSERT3S(d1->db_state, !=, DB_SEARCH); return (1); } if ((uintptr_t)d1 < (uintptr_t)d2) { return (-1); } if ((uintptr_t)d1 > (uintptr_t)d2) { return (1); } return (0); } /* ARGSUSED */ static int dnode_cons(void *arg, void *unused, int kmflag) { dnode_t *dn = arg; int i; rw_init(&dn->dn_struct_rwlock, NULL, RW_DEFAULT, NULL); mutex_init(&dn->dn_mtx, NULL, MUTEX_DEFAULT, NULL); mutex_init(&dn->dn_dbufs_mtx, NULL, MUTEX_DEFAULT, NULL); cv_init(&dn->dn_notxholds, NULL, CV_DEFAULT, NULL); /* * Every dbuf has a reference, and dropping a tracked reference is * O(number of references), so don't track dn_holds. */ refcount_create_untracked(&dn->dn_holds); refcount_create(&dn->dn_tx_holds); list_link_init(&dn->dn_link); bzero(&dn->dn_next_nblkptr[0], sizeof (dn->dn_next_nblkptr)); bzero(&dn->dn_next_nlevels[0], sizeof (dn->dn_next_nlevels)); bzero(&dn->dn_next_indblkshift[0], sizeof (dn->dn_next_indblkshift)); bzero(&dn->dn_next_bonustype[0], sizeof (dn->dn_next_bonustype)); bzero(&dn->dn_rm_spillblk[0], sizeof (dn->dn_rm_spillblk)); bzero(&dn->dn_next_bonuslen[0], sizeof (dn->dn_next_bonuslen)); bzero(&dn->dn_next_blksz[0], sizeof (dn->dn_next_blksz)); for (i = 0; i < TXG_SIZE; i++) { list_link_init(&dn->dn_dirty_link[i]); dn->dn_free_ranges[i] = NULL; list_create(&dn->dn_dirty_records[i], sizeof (dbuf_dirty_record_t), offsetof(dbuf_dirty_record_t, dr_dirty_node)); } dn->dn_allocated_txg = 0; dn->dn_free_txg = 0; dn->dn_assigned_txg = 0; dn->dn_dirtyctx = 0; dn->dn_dirtyctx_firstset = NULL; dn->dn_bonus = NULL; dn->dn_have_spill = B_FALSE; dn->dn_zio = NULL; dn->dn_oldused = 0; dn->dn_oldflags = 0; dn->dn_olduid = 0; dn->dn_oldgid = 0; dn->dn_newuid = 0; dn->dn_newgid = 0; dn->dn_id_flags = 0; dn->dn_dbufs_count = 0; avl_create(&dn->dn_dbufs, dbuf_compare, sizeof (dmu_buf_impl_t), offsetof(dmu_buf_impl_t, db_link)); dn->dn_moved = 0; return (0); } /* ARGSUSED */ static void dnode_dest(void *arg, void *unused) { int i; dnode_t *dn = arg; rw_destroy(&dn->dn_struct_rwlock); mutex_destroy(&dn->dn_mtx); mutex_destroy(&dn->dn_dbufs_mtx); cv_destroy(&dn->dn_notxholds); refcount_destroy(&dn->dn_holds); refcount_destroy(&dn->dn_tx_holds); ASSERT(!list_link_active(&dn->dn_link)); for (i = 0; i < TXG_SIZE; i++) { ASSERT(!list_link_active(&dn->dn_dirty_link[i])); ASSERT3P(dn->dn_free_ranges[i], ==, NULL); list_destroy(&dn->dn_dirty_records[i]); ASSERT0(dn->dn_next_nblkptr[i]); ASSERT0(dn->dn_next_nlevels[i]); ASSERT0(dn->dn_next_indblkshift[i]); ASSERT0(dn->dn_next_bonustype[i]); ASSERT0(dn->dn_rm_spillblk[i]); ASSERT0(dn->dn_next_bonuslen[i]); ASSERT0(dn->dn_next_blksz[i]); } ASSERT0(dn->dn_allocated_txg); ASSERT0(dn->dn_free_txg); ASSERT0(dn->dn_assigned_txg); ASSERT0(dn->dn_dirtyctx); ASSERT3P(dn->dn_dirtyctx_firstset, ==, NULL); ASSERT3P(dn->dn_bonus, ==, NULL); ASSERT(!dn->dn_have_spill); ASSERT3P(dn->dn_zio, ==, NULL); ASSERT0(dn->dn_oldused); ASSERT0(dn->dn_oldflags); ASSERT0(dn->dn_olduid); ASSERT0(dn->dn_oldgid); ASSERT0(dn->dn_newuid); ASSERT0(dn->dn_newgid); ASSERT0(dn->dn_id_flags); ASSERT0(dn->dn_dbufs_count); avl_destroy(&dn->dn_dbufs); } void dnode_init(void) { ASSERT(dnode_cache == NULL); dnode_cache = kmem_cache_create("dnode_t", sizeof (dnode_t), 0, dnode_cons, dnode_dest, NULL, NULL, NULL, 0); kmem_cache_set_move(dnode_cache, dnode_move); } void dnode_fini(void) { kmem_cache_destroy(dnode_cache); dnode_cache = NULL; } #ifdef ZFS_DEBUG void dnode_verify(dnode_t *dn) { int drop_struct_lock = FALSE; ASSERT(dn->dn_phys); ASSERT(dn->dn_objset); ASSERT(dn->dn_handle->dnh_dnode == dn); ASSERT(DMU_OT_IS_VALID(dn->dn_phys->dn_type)); if (!(zfs_flags & ZFS_DEBUG_DNODE_VERIFY)) return; if (!RW_WRITE_HELD(&dn->dn_struct_rwlock)) { rw_enter(&dn->dn_struct_rwlock, RW_READER); drop_struct_lock = TRUE; } if (dn->dn_phys->dn_type != DMU_OT_NONE || dn->dn_allocated_txg != 0) { int i; ASSERT3U(dn->dn_indblkshift, >=, 0); ASSERT3U(dn->dn_indblkshift, <=, SPA_MAXBLOCKSHIFT); if (dn->dn_datablkshift) { ASSERT3U(dn->dn_datablkshift, >=, SPA_MINBLOCKSHIFT); ASSERT3U(dn->dn_datablkshift, <=, SPA_MAXBLOCKSHIFT); ASSERT3U(1<dn_datablkshift, ==, dn->dn_datablksz); } ASSERT3U(dn->dn_nlevels, <=, 30); ASSERT(DMU_OT_IS_VALID(dn->dn_type)); ASSERT3U(dn->dn_nblkptr, >=, 1); ASSERT3U(dn->dn_nblkptr, <=, DN_MAX_NBLKPTR); ASSERT3U(dn->dn_bonuslen, <=, DN_MAX_BONUSLEN); ASSERT3U(dn->dn_datablksz, ==, dn->dn_datablkszsec << SPA_MINBLOCKSHIFT); ASSERT3U(ISP2(dn->dn_datablksz), ==, dn->dn_datablkshift != 0); ASSERT3U((dn->dn_nblkptr - 1) * sizeof (blkptr_t) + dn->dn_bonuslen, <=, DN_MAX_BONUSLEN); for (i = 0; i < TXG_SIZE; i++) { ASSERT3U(dn->dn_next_nlevels[i], <=, dn->dn_nlevels); } } if (dn->dn_phys->dn_type != DMU_OT_NONE) ASSERT3U(dn->dn_phys->dn_nlevels, <=, dn->dn_nlevels); ASSERT(DMU_OBJECT_IS_SPECIAL(dn->dn_object) || dn->dn_dbuf != NULL); if (dn->dn_dbuf != NULL) { ASSERT3P(dn->dn_phys, ==, (dnode_phys_t *)dn->dn_dbuf->db.db_data + (dn->dn_object % (dn->dn_dbuf->db.db_size >> DNODE_SHIFT))); } if (drop_struct_lock) rw_exit(&dn->dn_struct_rwlock); } #endif void dnode_byteswap(dnode_phys_t *dnp) { uint64_t *buf64 = (void*)&dnp->dn_blkptr; int i; if (dnp->dn_type == DMU_OT_NONE) { bzero(dnp, sizeof (dnode_phys_t)); return; } dnp->dn_datablkszsec = BSWAP_16(dnp->dn_datablkszsec); dnp->dn_bonuslen = BSWAP_16(dnp->dn_bonuslen); dnp->dn_maxblkid = BSWAP_64(dnp->dn_maxblkid); dnp->dn_used = BSWAP_64(dnp->dn_used); /* * dn_nblkptr is only one byte, so it's OK to read it in either * byte order. We can't read dn_bouslen. */ ASSERT(dnp->dn_indblkshift <= SPA_MAXBLOCKSHIFT); ASSERT(dnp->dn_nblkptr <= DN_MAX_NBLKPTR); for (i = 0; i < dnp->dn_nblkptr * sizeof (blkptr_t)/8; i++) buf64[i] = BSWAP_64(buf64[i]); /* * OK to check dn_bonuslen for zero, because it won't matter if * we have the wrong byte order. This is necessary because the * dnode dnode is smaller than a regular dnode. */ if (dnp->dn_bonuslen != 0) { /* * Note that the bonus length calculated here may be * longer than the actual bonus buffer. This is because * we always put the bonus buffer after the last block * pointer (instead of packing it against the end of the * dnode buffer). */ int off = (dnp->dn_nblkptr-1) * sizeof (blkptr_t); size_t len = DN_MAX_BONUSLEN - off; ASSERT(DMU_OT_IS_VALID(dnp->dn_bonustype)); dmu_object_byteswap_t byteswap = DMU_OT_BYTESWAP(dnp->dn_bonustype); dmu_ot_byteswap[byteswap].ob_func(dnp->dn_bonus + off, len); } /* Swap SPILL block if we have one */ if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) byteswap_uint64_array(&dnp->dn_spill, sizeof (blkptr_t)); } void dnode_buf_byteswap(void *vbuf, size_t size) { dnode_phys_t *buf = vbuf; int i; ASSERT3U(sizeof (dnode_phys_t), ==, (1<>= DNODE_SHIFT; for (i = 0; i < size; i++) { dnode_byteswap(buf); buf++; } } void dnode_setbonuslen(dnode_t *dn, int newsize, dmu_tx_t *tx) { ASSERT3U(refcount_count(&dn->dn_holds), >=, 1); dnode_setdirty(dn, tx); rw_enter(&dn->dn_struct_rwlock, RW_WRITER); ASSERT3U(newsize, <=, DN_MAX_BONUSLEN - (dn->dn_nblkptr-1) * sizeof (blkptr_t)); dn->dn_bonuslen = newsize; if (newsize == 0) dn->dn_next_bonuslen[tx->tx_txg & TXG_MASK] = DN_ZERO_BONUSLEN; else dn->dn_next_bonuslen[tx->tx_txg & TXG_MASK] = dn->dn_bonuslen; rw_exit(&dn->dn_struct_rwlock); } void dnode_setbonus_type(dnode_t *dn, dmu_object_type_t newtype, dmu_tx_t *tx) { ASSERT3U(refcount_count(&dn->dn_holds), >=, 1); dnode_setdirty(dn, tx); rw_enter(&dn->dn_struct_rwlock, RW_WRITER); dn->dn_bonustype = newtype; dn->dn_next_bonustype[tx->tx_txg & TXG_MASK] = dn->dn_bonustype; rw_exit(&dn->dn_struct_rwlock); } void dnode_rm_spill(dnode_t *dn, dmu_tx_t *tx) { ASSERT3U(refcount_count(&dn->dn_holds), >=, 1); ASSERT(RW_WRITE_HELD(&dn->dn_struct_rwlock)); dnode_setdirty(dn, tx); dn->dn_rm_spillblk[tx->tx_txg&TXG_MASK] = DN_KILL_SPILLBLK; dn->dn_have_spill = B_FALSE; } static void dnode_setdblksz(dnode_t *dn, int size) { ASSERT0(P2PHASE(size, SPA_MINBLOCKSIZE)); ASSERT3U(size, <=, SPA_MAXBLOCKSIZE); ASSERT3U(size, >=, SPA_MINBLOCKSIZE); ASSERT3U(size >> SPA_MINBLOCKSHIFT, <, 1<<(sizeof (dn->dn_phys->dn_datablkszsec) * 8)); dn->dn_datablksz = size; dn->dn_datablkszsec = size >> SPA_MINBLOCKSHIFT; dn->dn_datablkshift = ISP2(size) ? highbit64(size - 1) : 0; } static dnode_t * dnode_create(objset_t *os, dnode_phys_t *dnp, dmu_buf_impl_t *db, uint64_t object, dnode_handle_t *dnh) { dnode_t *dn; dn = kmem_cache_alloc(dnode_cache, KM_SLEEP); ASSERT(!POINTER_IS_VALID(dn->dn_objset)); dn->dn_moved = 0; /* * Defer setting dn_objset until the dnode is ready to be a candidate * for the dnode_move() callback. */ dn->dn_object = object; dn->dn_dbuf = db; dn->dn_handle = dnh; dn->dn_phys = dnp; if (dnp->dn_datablkszsec) { dnode_setdblksz(dn, dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT); } else { dn->dn_datablksz = 0; dn->dn_datablkszsec = 0; dn->dn_datablkshift = 0; } dn->dn_indblkshift = dnp->dn_indblkshift; dn->dn_nlevels = dnp->dn_nlevels; dn->dn_type = dnp->dn_type; dn->dn_nblkptr = dnp->dn_nblkptr; dn->dn_checksum = dnp->dn_checksum; dn->dn_compress = dnp->dn_compress; dn->dn_bonustype = dnp->dn_bonustype; dn->dn_bonuslen = dnp->dn_bonuslen; dn->dn_maxblkid = dnp->dn_maxblkid; dn->dn_have_spill = ((dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0); dn->dn_id_flags = 0; dmu_zfetch_init(&dn->dn_zfetch, dn); ASSERT(DMU_OT_IS_VALID(dn->dn_phys->dn_type)); mutex_enter(&os->os_lock); if (dnh->dnh_dnode != NULL) { /* Lost the allocation race. */ mutex_exit(&os->os_lock); kmem_cache_free(dnode_cache, dn); return (dnh->dnh_dnode); } /* * Exclude special dnodes from os_dnodes so an empty os_dnodes * signifies that the special dnodes have no references from * their children (the entries in os_dnodes). This allows * dnode_destroy() to easily determine if the last child has * been removed and then complete eviction of the objset. */ if (!DMU_OBJECT_IS_SPECIAL(object)) list_insert_head(&os->os_dnodes, dn); membar_producer(); /* * Everything else must be valid before assigning dn_objset * makes the dnode eligible for dnode_move(). */ dn->dn_objset = os; dnh->dnh_dnode = dn; mutex_exit(&os->os_lock); arc_space_consume(sizeof (dnode_t), ARC_SPACE_OTHER); return (dn); } /* * Caller must be holding the dnode handle, which is released upon return. */ static void dnode_destroy(dnode_t *dn) { objset_t *os = dn->dn_objset; boolean_t complete_os_eviction = B_FALSE; ASSERT((dn->dn_id_flags & DN_ID_NEW_EXIST) == 0); mutex_enter(&os->os_lock); POINTER_INVALIDATE(&dn->dn_objset); if (!DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { list_remove(&os->os_dnodes, dn); complete_os_eviction = list_is_empty(&os->os_dnodes) && list_link_active(&os->os_evicting_node); } mutex_exit(&os->os_lock); /* the dnode can no longer move, so we can release the handle */ zrl_remove(&dn->dn_handle->dnh_zrlock); dn->dn_allocated_txg = 0; dn->dn_free_txg = 0; dn->dn_assigned_txg = 0; dn->dn_dirtyctx = 0; if (dn->dn_dirtyctx_firstset != NULL) { kmem_free(dn->dn_dirtyctx_firstset, 1); dn->dn_dirtyctx_firstset = NULL; } if (dn->dn_bonus != NULL) { mutex_enter(&dn->dn_bonus->db_mtx); dbuf_destroy(dn->dn_bonus); dn->dn_bonus = NULL; } dn->dn_zio = NULL; dn->dn_have_spill = B_FALSE; dn->dn_oldused = 0; dn->dn_oldflags = 0; dn->dn_olduid = 0; dn->dn_oldgid = 0; dn->dn_newuid = 0; dn->dn_newgid = 0; dn->dn_id_flags = 0; dmu_zfetch_fini(&dn->dn_zfetch); kmem_cache_free(dnode_cache, dn); arc_space_return(sizeof (dnode_t), ARC_SPACE_OTHER); if (complete_os_eviction) dmu_objset_evict_done(os); } void dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) { int i; ASSERT3U(blocksize, <=, spa_maxblocksize(dmu_objset_spa(dn->dn_objset))); if (blocksize == 0) blocksize = 1 << zfs_default_bs; else blocksize = P2ROUNDUP(blocksize, SPA_MINBLOCKSIZE); if (ibs == 0) ibs = zfs_default_ibs; ibs = MIN(MAX(ibs, DN_MIN_INDBLKSHIFT), DN_MAX_INDBLKSHIFT); dprintf("os=%p obj=%llu txg=%llu blocksize=%d ibs=%d\n", dn->dn_objset, dn->dn_object, tx->tx_txg, blocksize, ibs); ASSERT(dn->dn_type == DMU_OT_NONE); ASSERT(bcmp(dn->dn_phys, &dnode_phys_zero, sizeof (dnode_phys_t)) == 0); ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE); ASSERT(ot != DMU_OT_NONE); ASSERT(DMU_OT_IS_VALID(ot)); ASSERT((bonustype == DMU_OT_NONE && bonuslen == 0) || (bonustype == DMU_OT_SA && bonuslen == 0) || (bonustype != DMU_OT_NONE && bonuslen != 0)); ASSERT(DMU_OT_IS_VALID(bonustype)); ASSERT3U(bonuslen, <=, DN_MAX_BONUSLEN); ASSERT(dn->dn_type == DMU_OT_NONE); ASSERT0(dn->dn_maxblkid); ASSERT0(dn->dn_allocated_txg); ASSERT0(dn->dn_assigned_txg); ASSERT(refcount_is_zero(&dn->dn_tx_holds)); ASSERT3U(refcount_count(&dn->dn_holds), <=, 1); ASSERT(avl_is_empty(&dn->dn_dbufs)); for (i = 0; i < TXG_SIZE; i++) { ASSERT0(dn->dn_next_nblkptr[i]); ASSERT0(dn->dn_next_nlevels[i]); ASSERT0(dn->dn_next_indblkshift[i]); ASSERT0(dn->dn_next_bonuslen[i]); ASSERT0(dn->dn_next_bonustype[i]); ASSERT0(dn->dn_rm_spillblk[i]); ASSERT0(dn->dn_next_blksz[i]); ASSERT(!list_link_active(&dn->dn_dirty_link[i])); ASSERT3P(list_head(&dn->dn_dirty_records[i]), ==, NULL); ASSERT3P(dn->dn_free_ranges[i], ==, NULL); } dn->dn_type = ot; dnode_setdblksz(dn, blocksize); dn->dn_indblkshift = ibs; dn->dn_nlevels = 1; if (bonustype == DMU_OT_SA) /* Maximize bonus space for SA */ dn->dn_nblkptr = 1; else dn->dn_nblkptr = 1 + ((DN_MAX_BONUSLEN - bonuslen) >> SPA_BLKPTRSHIFT); dn->dn_bonustype = bonustype; dn->dn_bonuslen = bonuslen; dn->dn_checksum = ZIO_CHECKSUM_INHERIT; dn->dn_compress = ZIO_COMPRESS_INHERIT; dn->dn_dirtyctx = 0; dn->dn_free_txg = 0; if (dn->dn_dirtyctx_firstset) { kmem_free(dn->dn_dirtyctx_firstset, 1); dn->dn_dirtyctx_firstset = NULL; } dn->dn_allocated_txg = tx->tx_txg; dn->dn_id_flags = 0; dnode_setdirty(dn, tx); dn->dn_next_indblkshift[tx->tx_txg & TXG_MASK] = ibs; dn->dn_next_bonuslen[tx->tx_txg & TXG_MASK] = dn->dn_bonuslen; dn->dn_next_bonustype[tx->tx_txg & TXG_MASK] = dn->dn_bonustype; dn->dn_next_blksz[tx->tx_txg & TXG_MASK] = dn->dn_datablksz; } void dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx) { int nblkptr; ASSERT3U(blocksize, >=, SPA_MINBLOCKSIZE); ASSERT3U(blocksize, <=, spa_maxblocksize(dmu_objset_spa(dn->dn_objset))); ASSERT0(blocksize % SPA_MINBLOCKSIZE); ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT || dmu_tx_private_ok(tx)); ASSERT(tx->tx_txg != 0); ASSERT((bonustype == DMU_OT_NONE && bonuslen == 0) || (bonustype != DMU_OT_NONE && bonuslen != 0) || (bonustype == DMU_OT_SA && bonuslen == 0)); ASSERT(DMU_OT_IS_VALID(bonustype)); ASSERT3U(bonuslen, <=, DN_MAX_BONUSLEN); /* clean up any unreferenced dbufs */ dnode_evict_dbufs(dn); dn->dn_id_flags = 0; rw_enter(&dn->dn_struct_rwlock, RW_WRITER); dnode_setdirty(dn, tx); if (dn->dn_datablksz != blocksize) { /* change blocksize */ ASSERT(dn->dn_maxblkid == 0 && (BP_IS_HOLE(&dn->dn_phys->dn_blkptr[0]) || dnode_block_freed(dn, 0))); dnode_setdblksz(dn, blocksize); dn->dn_next_blksz[tx->tx_txg&TXG_MASK] = blocksize; } if (dn->dn_bonuslen != bonuslen) dn->dn_next_bonuslen[tx->tx_txg&TXG_MASK] = bonuslen; if (bonustype == DMU_OT_SA) /* Maximize bonus space for SA */ nblkptr = 1; else nblkptr = 1 + ((DN_MAX_BONUSLEN - bonuslen) >> SPA_BLKPTRSHIFT); if (dn->dn_bonustype != bonustype) dn->dn_next_bonustype[tx->tx_txg&TXG_MASK] = bonustype; if (dn->dn_nblkptr != nblkptr) dn->dn_next_nblkptr[tx->tx_txg&TXG_MASK] = nblkptr; if (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR) { dbuf_rm_spill(dn, tx); dnode_rm_spill(dn, tx); } rw_exit(&dn->dn_struct_rwlock); /* change type */ dn->dn_type = ot; /* change bonus size and type */ mutex_enter(&dn->dn_mtx); dn->dn_bonustype = bonustype; dn->dn_bonuslen = bonuslen; dn->dn_nblkptr = nblkptr; dn->dn_checksum = ZIO_CHECKSUM_INHERIT; dn->dn_compress = ZIO_COMPRESS_INHERIT; ASSERT3U(dn->dn_nblkptr, <=, DN_MAX_NBLKPTR); /* fix up the bonus db_size */ if (dn->dn_bonus) { dn->dn_bonus->db.db_size = DN_MAX_BONUSLEN - (dn->dn_nblkptr-1) * sizeof (blkptr_t); ASSERT(dn->dn_bonuslen <= dn->dn_bonus->db.db_size); } dn->dn_allocated_txg = tx->tx_txg; mutex_exit(&dn->dn_mtx); } #ifdef DNODE_STATS static struct { uint64_t dms_dnode_invalid; uint64_t dms_dnode_recheck1; uint64_t dms_dnode_recheck2; uint64_t dms_dnode_special; uint64_t dms_dnode_handle; uint64_t dms_dnode_rwlock; uint64_t dms_dnode_active; } dnode_move_stats; #endif /* DNODE_STATS */ static void dnode_move_impl(dnode_t *odn, dnode_t *ndn) { int i; ASSERT(!RW_LOCK_HELD(&odn->dn_struct_rwlock)); ASSERT(MUTEX_NOT_HELD(&odn->dn_mtx)); ASSERT(MUTEX_NOT_HELD(&odn->dn_dbufs_mtx)); ASSERT(!RW_LOCK_HELD(&odn->dn_zfetch.zf_rwlock)); /* Copy fields. */ ndn->dn_objset = odn->dn_objset; ndn->dn_object = odn->dn_object; ndn->dn_dbuf = odn->dn_dbuf; ndn->dn_handle = odn->dn_handle; ndn->dn_phys = odn->dn_phys; ndn->dn_type = odn->dn_type; ndn->dn_bonuslen = odn->dn_bonuslen; ndn->dn_bonustype = odn->dn_bonustype; ndn->dn_nblkptr = odn->dn_nblkptr; ndn->dn_checksum = odn->dn_checksum; ndn->dn_compress = odn->dn_compress; ndn->dn_nlevels = odn->dn_nlevels; ndn->dn_indblkshift = odn->dn_indblkshift; ndn->dn_datablkshift = odn->dn_datablkshift; ndn->dn_datablkszsec = odn->dn_datablkszsec; ndn->dn_datablksz = odn->dn_datablksz; ndn->dn_maxblkid = odn->dn_maxblkid; bcopy(&odn->dn_next_nblkptr[0], &ndn->dn_next_nblkptr[0], sizeof (odn->dn_next_nblkptr)); bcopy(&odn->dn_next_nlevels[0], &ndn->dn_next_nlevels[0], sizeof (odn->dn_next_nlevels)); bcopy(&odn->dn_next_indblkshift[0], &ndn->dn_next_indblkshift[0], sizeof (odn->dn_next_indblkshift)); bcopy(&odn->dn_next_bonustype[0], &ndn->dn_next_bonustype[0], sizeof (odn->dn_next_bonustype)); bcopy(&odn->dn_rm_spillblk[0], &ndn->dn_rm_spillblk[0], sizeof (odn->dn_rm_spillblk)); bcopy(&odn->dn_next_bonuslen[0], &ndn->dn_next_bonuslen[0], sizeof (odn->dn_next_bonuslen)); bcopy(&odn->dn_next_blksz[0], &ndn->dn_next_blksz[0], sizeof (odn->dn_next_blksz)); for (i = 0; i < TXG_SIZE; i++) { list_move_tail(&ndn->dn_dirty_records[i], &odn->dn_dirty_records[i]); } bcopy(&odn->dn_free_ranges[0], &ndn->dn_free_ranges[0], sizeof (odn->dn_free_ranges)); ndn->dn_allocated_txg = odn->dn_allocated_txg; ndn->dn_free_txg = odn->dn_free_txg; ndn->dn_assigned_txg = odn->dn_assigned_txg; ndn->dn_dirtyctx = odn->dn_dirtyctx; ndn->dn_dirtyctx_firstset = odn->dn_dirtyctx_firstset; ASSERT(refcount_count(&odn->dn_tx_holds) == 0); refcount_transfer(&ndn->dn_holds, &odn->dn_holds); ASSERT(avl_is_empty(&ndn->dn_dbufs)); avl_swap(&ndn->dn_dbufs, &odn->dn_dbufs); ndn->dn_dbufs_count = odn->dn_dbufs_count; ndn->dn_bonus = odn->dn_bonus; ndn->dn_have_spill = odn->dn_have_spill; ndn->dn_zio = odn->dn_zio; ndn->dn_oldused = odn->dn_oldused; ndn->dn_oldflags = odn->dn_oldflags; ndn->dn_olduid = odn->dn_olduid; ndn->dn_oldgid = odn->dn_oldgid; ndn->dn_newuid = odn->dn_newuid; ndn->dn_newgid = odn->dn_newgid; ndn->dn_id_flags = odn->dn_id_flags; dmu_zfetch_init(&ndn->dn_zfetch, NULL); list_move_tail(&ndn->dn_zfetch.zf_stream, &odn->dn_zfetch.zf_stream); ndn->dn_zfetch.zf_dnode = odn->dn_zfetch.zf_dnode; /* * Update back pointers. Updating the handle fixes the back pointer of * every descendant dbuf as well as the bonus dbuf. */ ASSERT(ndn->dn_handle->dnh_dnode == odn); ndn->dn_handle->dnh_dnode = ndn; if (ndn->dn_zfetch.zf_dnode == odn) { ndn->dn_zfetch.zf_dnode = ndn; } /* * Invalidate the original dnode by clearing all of its back pointers. */ odn->dn_dbuf = NULL; odn->dn_handle = NULL; avl_create(&odn->dn_dbufs, dbuf_compare, sizeof (dmu_buf_impl_t), offsetof(dmu_buf_impl_t, db_link)); odn->dn_dbufs_count = 0; odn->dn_bonus = NULL; odn->dn_zfetch.zf_dnode = NULL; /* * Set the low bit of the objset pointer to ensure that dnode_move() * recognizes the dnode as invalid in any subsequent callback. */ POINTER_INVALIDATE(&odn->dn_objset); /* * Satisfy the destructor. */ for (i = 0; i < TXG_SIZE; i++) { list_create(&odn->dn_dirty_records[i], sizeof (dbuf_dirty_record_t), offsetof(dbuf_dirty_record_t, dr_dirty_node)); odn->dn_free_ranges[i] = NULL; odn->dn_next_nlevels[i] = 0; odn->dn_next_indblkshift[i] = 0; odn->dn_next_bonustype[i] = 0; odn->dn_rm_spillblk[i] = 0; odn->dn_next_bonuslen[i] = 0; odn->dn_next_blksz[i] = 0; } odn->dn_allocated_txg = 0; odn->dn_free_txg = 0; odn->dn_assigned_txg = 0; odn->dn_dirtyctx = 0; odn->dn_dirtyctx_firstset = NULL; odn->dn_have_spill = B_FALSE; odn->dn_zio = NULL; odn->dn_oldused = 0; odn->dn_oldflags = 0; odn->dn_olduid = 0; odn->dn_oldgid = 0; odn->dn_newuid = 0; odn->dn_newgid = 0; odn->dn_id_flags = 0; /* * Mark the dnode. */ ndn->dn_moved = 1; odn->dn_moved = (uint8_t)-1; } #ifdef _KERNEL /*ARGSUSED*/ static kmem_cbrc_t dnode_move(void *buf, void *newbuf, size_t size, void *arg) { dnode_t *odn = buf, *ndn = newbuf; objset_t *os; int64_t refcount; uint32_t dbufs; /* * The dnode is on the objset's list of known dnodes if the objset * pointer is valid. We set the low bit of the objset pointer when * freeing the dnode to invalidate it, and the memory patterns written * by kmem (baddcafe and deadbeef) set at least one of the two low bits. * A newly created dnode sets the objset pointer last of all to indicate * that the dnode is known and in a valid state to be moved by this * function. */ os = odn->dn_objset; if (!POINTER_IS_VALID(os)) { DNODE_STAT_ADD(dnode_move_stats.dms_dnode_invalid); return (KMEM_CBRC_DONT_KNOW); } /* * Ensure that the objset does not go away during the move. */ rw_enter(&os_lock, RW_WRITER); if (os != odn->dn_objset) { rw_exit(&os_lock); DNODE_STAT_ADD(dnode_move_stats.dms_dnode_recheck1); return (KMEM_CBRC_DONT_KNOW); } /* * If the dnode is still valid, then so is the objset. We know that no * valid objset can be freed while we hold os_lock, so we can safely * ensure that the objset remains in use. */ mutex_enter(&os->os_lock); /* * Recheck the objset pointer in case the dnode was removed just before * acquiring the lock. */ if (os != odn->dn_objset) { mutex_exit(&os->os_lock); rw_exit(&os_lock); DNODE_STAT_ADD(dnode_move_stats.dms_dnode_recheck2); return (KMEM_CBRC_DONT_KNOW); } /* * At this point we know that as long as we hold os->os_lock, the dnode * cannot be freed and fields within the dnode can be safely accessed. * The objset listing this dnode cannot go away as long as this dnode is * on its list. */ rw_exit(&os_lock); if (DMU_OBJECT_IS_SPECIAL(odn->dn_object)) { mutex_exit(&os->os_lock); DNODE_STAT_ADD(dnode_move_stats.dms_dnode_special); return (KMEM_CBRC_NO); } ASSERT(odn->dn_dbuf != NULL); /* only "special" dnodes have no parent */ /* * Lock the dnode handle to prevent the dnode from obtaining any new * holds. This also prevents the descendant dbufs and the bonus dbuf * from accessing the dnode, so that we can discount their holds. The * handle is safe to access because we know that while the dnode cannot * go away, neither can its handle. Once we hold dnh_zrlock, we can * safely move any dnode referenced only by dbufs. */ if (!zrl_tryenter(&odn->dn_handle->dnh_zrlock)) { mutex_exit(&os->os_lock); DNODE_STAT_ADD(dnode_move_stats.dms_dnode_handle); return (KMEM_CBRC_LATER); } /* * Ensure a consistent view of the dnode's holds and the dnode's dbufs. * We need to guarantee that there is a hold for every dbuf in order to * determine whether the dnode is actively referenced. Falsely matching * a dbuf to an active hold would lead to an unsafe move. It's possible * that a thread already having an active dnode hold is about to add a * dbuf, and we can't compare hold and dbuf counts while the add is in * progress. */ if (!rw_tryenter(&odn->dn_struct_rwlock, RW_WRITER)) { zrl_exit(&odn->dn_handle->dnh_zrlock); mutex_exit(&os->os_lock); DNODE_STAT_ADD(dnode_move_stats.dms_dnode_rwlock); return (KMEM_CBRC_LATER); } /* * A dbuf may be removed (evicted) without an active dnode hold. In that * case, the dbuf count is decremented under the handle lock before the * dbuf's hold is released. This order ensures that if we count the hold * after the dbuf is removed but before its hold is released, we will * treat the unmatched hold as active and exit safely. If we count the * hold before the dbuf is removed, the hold is discounted, and the * removal is blocked until the move completes. */ refcount = refcount_count(&odn->dn_holds); ASSERT(refcount >= 0); dbufs = odn->dn_dbufs_count; /* We can't have more dbufs than dnode holds. */ ASSERT3U(dbufs, <=, refcount); DTRACE_PROBE3(dnode__move, dnode_t *, odn, int64_t, refcount, uint32_t, dbufs); if (refcount > dbufs) { rw_exit(&odn->dn_struct_rwlock); zrl_exit(&odn->dn_handle->dnh_zrlock); mutex_exit(&os->os_lock); DNODE_STAT_ADD(dnode_move_stats.dms_dnode_active); return (KMEM_CBRC_LATER); } rw_exit(&odn->dn_struct_rwlock); /* * At this point we know that anyone with a hold on the dnode is not * actively referencing it. The dnode is known and in a valid state to * move. We're holding the locks needed to execute the critical section. */ dnode_move_impl(odn, ndn); list_link_replace(&odn->dn_link, &ndn->dn_link); /* If the dnode was safe to move, the refcount cannot have changed. */ ASSERT(refcount == refcount_count(&ndn->dn_holds)); ASSERT(dbufs == ndn->dn_dbufs_count); zrl_exit(&ndn->dn_handle->dnh_zrlock); /* handle has moved */ mutex_exit(&os->os_lock); return (KMEM_CBRC_YES); } #endif /* _KERNEL */ void dnode_special_close(dnode_handle_t *dnh) { dnode_t *dn = dnh->dnh_dnode; /* * Wait for final references to the dnode to clear. This can * only happen if the arc is asyncronously evicting state that * has a hold on this dnode while we are trying to evict this * dnode. */ while (refcount_count(&dn->dn_holds) > 0) delay(1); ASSERT(dn->dn_dbuf == NULL || dmu_buf_get_user(&dn->dn_dbuf->db) == NULL); zrl_add(&dnh->dnh_zrlock); dnode_destroy(dn); /* implicit zrl_remove() */ zrl_destroy(&dnh->dnh_zrlock); dnh->dnh_dnode = NULL; } void dnode_special_open(objset_t *os, dnode_phys_t *dnp, uint64_t object, dnode_handle_t *dnh) { dnode_t *dn; dn = dnode_create(os, dnp, NULL, object, dnh); zrl_init(&dnh->dnh_zrlock); DNODE_VERIFY(dn); } static void dnode_buf_evict_async(void *dbu) { dnode_children_t *children_dnodes = dbu; int i; for (i = 0; i < children_dnodes->dnc_count; i++) { dnode_handle_t *dnh = &children_dnodes->dnc_children[i]; dnode_t *dn; /* * The dnode handle lock guards against the dnode moving to * another valid address, so there is no need here to guard * against changes to or from NULL. */ if (dnh->dnh_dnode == NULL) { zrl_destroy(&dnh->dnh_zrlock); continue; } zrl_add(&dnh->dnh_zrlock); dn = dnh->dnh_dnode; /* * If there are holds on this dnode, then there should * be holds on the dnode's containing dbuf as well; thus * it wouldn't be eligible for eviction and this function * would not have been called. */ ASSERT(refcount_is_zero(&dn->dn_holds)); ASSERT(refcount_is_zero(&dn->dn_tx_holds)); dnode_destroy(dn); /* implicit zrl_remove() */ zrl_destroy(&dnh->dnh_zrlock); dnh->dnh_dnode = NULL; } kmem_free(children_dnodes, sizeof (dnode_children_t) + children_dnodes->dnc_count * sizeof (dnode_handle_t)); } /* * errors: * EINVAL - invalid object number. * EIO - i/o error. * succeeds even for free dnodes. */ int dnode_hold_impl(objset_t *os, uint64_t object, int flag, void *tag, dnode_t **dnp) { int epb, idx, err; int drop_struct_lock = FALSE; int type; uint64_t blk; dnode_t *mdn, *dn; dmu_buf_impl_t *db; dnode_children_t *children_dnodes; dnode_handle_t *dnh; /* * If you are holding the spa config lock as writer, you shouldn't * be asking the DMU to do *anything* unless it's the root pool * which may require us to read from the root filesystem while * holding some (not all) of the locks as writer. */ ASSERT(spa_config_held(os->os_spa, SCL_ALL, RW_WRITER) == 0 || (spa_is_root(os->os_spa) && spa_config_held(os->os_spa, SCL_STATE, RW_WRITER))); if (object == DMU_USERUSED_OBJECT || object == DMU_GROUPUSED_OBJECT) { dn = (object == DMU_USERUSED_OBJECT) ? DMU_USERUSED_DNODE(os) : DMU_GROUPUSED_DNODE(os); if (dn == NULL) return (SET_ERROR(ENOENT)); type = dn->dn_type; if ((flag & DNODE_MUST_BE_ALLOCATED) && type == DMU_OT_NONE) return (SET_ERROR(ENOENT)); if ((flag & DNODE_MUST_BE_FREE) && type != DMU_OT_NONE) return (SET_ERROR(EEXIST)); DNODE_VERIFY(dn); (void) refcount_add(&dn->dn_holds, tag); *dnp = dn; return (0); } if (object == 0 || object >= DN_MAX_OBJECT) return (SET_ERROR(EINVAL)); mdn = DMU_META_DNODE(os); ASSERT(mdn->dn_object == DMU_META_DNODE_OBJECT); DNODE_VERIFY(mdn); if (!RW_WRITE_HELD(&mdn->dn_struct_rwlock)) { rw_enter(&mdn->dn_struct_rwlock, RW_READER); drop_struct_lock = TRUE; } blk = dbuf_whichblock(mdn, 0, object * sizeof (dnode_phys_t)); db = dbuf_hold(mdn, blk, FTAG); if (drop_struct_lock) rw_exit(&mdn->dn_struct_rwlock); if (db == NULL) return (SET_ERROR(EIO)); err = dbuf_read(db, NULL, DB_RF_CANFAIL); if (err) { dbuf_rele(db, FTAG); return (err); } ASSERT3U(db->db.db_size, >=, 1<db.db_size >> DNODE_SHIFT; idx = object & (epb-1); ASSERT(DB_DNODE(db)->dn_type == DMU_OT_DNODE); children_dnodes = dmu_buf_get_user(&db->db); if (children_dnodes == NULL) { int i; dnode_children_t *winner; children_dnodes = kmem_zalloc(sizeof (dnode_children_t) + epb * sizeof (dnode_handle_t), KM_SLEEP); children_dnodes->dnc_count = epb; dnh = &children_dnodes->dnc_children[0]; for (i = 0; i < epb; i++) { zrl_init(&dnh[i].dnh_zrlock); } dmu_buf_init_user(&children_dnodes->dnc_dbu, NULL, dnode_buf_evict_async, NULL); winner = dmu_buf_set_user(&db->db, &children_dnodes->dnc_dbu); if (winner != NULL) { for (i = 0; i < epb; i++) { zrl_destroy(&dnh[i].dnh_zrlock); } kmem_free(children_dnodes, sizeof (dnode_children_t) + epb * sizeof (dnode_handle_t)); children_dnodes = winner; } } ASSERT(children_dnodes->dnc_count == epb); dnh = &children_dnodes->dnc_children[idx]; zrl_add(&dnh->dnh_zrlock); dn = dnh->dnh_dnode; if (dn == NULL) { dnode_phys_t *phys = (dnode_phys_t *)db->db.db_data+idx; dn = dnode_create(os, phys, db, object, dnh); } mutex_enter(&dn->dn_mtx); type = dn->dn_type; if (dn->dn_free_txg || ((flag & DNODE_MUST_BE_ALLOCATED) && type == DMU_OT_NONE) || ((flag & DNODE_MUST_BE_FREE) && (type != DMU_OT_NONE || !refcount_is_zero(&dn->dn_holds)))) { mutex_exit(&dn->dn_mtx); zrl_remove(&dnh->dnh_zrlock); dbuf_rele(db, FTAG); return (type == DMU_OT_NONE ? ENOENT : EEXIST); } if (refcount_add(&dn->dn_holds, tag) == 1) dbuf_add_ref(db, dnh); mutex_exit(&dn->dn_mtx); /* Now we can rely on the hold to prevent the dnode from moving. */ zrl_remove(&dnh->dnh_zrlock); DNODE_VERIFY(dn); ASSERT3P(dn->dn_dbuf, ==, db); ASSERT3U(dn->dn_object, ==, object); dbuf_rele(db, FTAG); *dnp = dn; return (0); } /* * Return held dnode if the object is allocated, NULL if not. */ int dnode_hold(objset_t *os, uint64_t object, void *tag, dnode_t **dnp) { return (dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, tag, dnp)); } /* * Can only add a reference if there is already at least one * reference on the dnode. Returns FALSE if unable to add a * new reference. */ boolean_t dnode_add_ref(dnode_t *dn, void *tag) { mutex_enter(&dn->dn_mtx); if (refcount_is_zero(&dn->dn_holds)) { mutex_exit(&dn->dn_mtx); return (FALSE); } VERIFY(1 < refcount_add(&dn->dn_holds, tag)); mutex_exit(&dn->dn_mtx); return (TRUE); } void dnode_rele(dnode_t *dn, void *tag) { mutex_enter(&dn->dn_mtx); dnode_rele_and_unlock(dn, tag); } void dnode_rele_and_unlock(dnode_t *dn, void *tag) { uint64_t refs; /* Get while the hold prevents the dnode from moving. */ dmu_buf_impl_t *db = dn->dn_dbuf; dnode_handle_t *dnh = dn->dn_handle; refs = refcount_remove(&dn->dn_holds, tag); mutex_exit(&dn->dn_mtx); /* * It's unsafe to release the last hold on a dnode by dnode_rele() or * indirectly by dbuf_rele() while relying on the dnode handle to * prevent the dnode from moving, since releasing the last hold could * result in the dnode's parent dbuf evicting its dnode handles. For * that reason anyone calling dnode_rele() or dbuf_rele() without some * other direct or indirect hold on the dnode must first drop the dnode * handle. */ ASSERT(refs > 0 || dnh->dnh_zrlock.zr_owner != curthread); /* NOTE: the DNODE_DNODE does not have a dn_dbuf */ if (refs == 0 && db != NULL) { /* * Another thread could add a hold to the dnode handle in * dnode_hold_impl() while holding the parent dbuf. Since the * hold on the parent dbuf prevents the handle from being * destroyed, the hold on the handle is OK. We can't yet assert * that the handle has zero references, but that will be * asserted anyway when the handle gets destroyed. */ dbuf_rele(db, dnh); } } void dnode_setdirty(dnode_t *dn, dmu_tx_t *tx) { objset_t *os = dn->dn_objset; uint64_t txg = tx->tx_txg; if (DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { dsl_dataset_dirty(os->os_dsl_dataset, tx); return; } DNODE_VERIFY(dn); #ifdef ZFS_DEBUG mutex_enter(&dn->dn_mtx); ASSERT(dn->dn_phys->dn_type || dn->dn_allocated_txg); ASSERT(dn->dn_free_txg == 0 || dn->dn_free_txg >= txg); mutex_exit(&dn->dn_mtx); #endif /* * Determine old uid/gid when necessary */ dmu_objset_userquota_get_ids(dn, B_TRUE, tx); - mutex_enter(&os->os_lock); + multilist_t *dirtylist = os->os_dirty_dnodes[txg & TXG_MASK]; + multilist_sublist_t *mls = multilist_sublist_lock_obj(dirtylist, dn); /* * If we are already marked dirty, we're done. */ if (list_link_active(&dn->dn_dirty_link[txg & TXG_MASK])) { - mutex_exit(&os->os_lock); + multilist_sublist_unlock(mls); return; } ASSERT(!refcount_is_zero(&dn->dn_holds) || !avl_is_empty(&dn->dn_dbufs)); ASSERT(dn->dn_datablksz != 0); ASSERT0(dn->dn_next_bonuslen[txg&TXG_MASK]); ASSERT0(dn->dn_next_blksz[txg&TXG_MASK]); ASSERT0(dn->dn_next_bonustype[txg&TXG_MASK]); dprintf_ds(os->os_dsl_dataset, "obj=%llu txg=%llu\n", dn->dn_object, txg); - if (dn->dn_free_txg > 0 && dn->dn_free_txg <= txg) { - list_insert_tail(&os->os_free_dnodes[txg&TXG_MASK], dn); - } else { - list_insert_tail(&os->os_dirty_dnodes[txg&TXG_MASK], dn); - } + multilist_sublist_insert_head(mls, dn); - mutex_exit(&os->os_lock); + multilist_sublist_unlock(mls); /* * The dnode maintains a hold on its containing dbuf as * long as there are holds on it. Each instantiated child * dbuf maintains a hold on the dnode. When the last child * drops its hold, the dnode will drop its hold on the * containing dbuf. We add a "dirty hold" here so that the * dnode will hang around after we finish processing its * children. */ VERIFY(dnode_add_ref(dn, (void *)(uintptr_t)tx->tx_txg)); (void) dbuf_dirty(dn->dn_dbuf, tx); dsl_dataset_dirty(os->os_dsl_dataset, tx); } void dnode_free(dnode_t *dn, dmu_tx_t *tx) { - int txgoff = tx->tx_txg & TXG_MASK; - - dprintf("dn=%p txg=%llu\n", dn, tx->tx_txg); - - /* we should be the only holder... hopefully */ - /* ASSERT3U(refcount_count(&dn->dn_holds), ==, 1); */ - mutex_enter(&dn->dn_mtx); if (dn->dn_type == DMU_OT_NONE || dn->dn_free_txg) { mutex_exit(&dn->dn_mtx); return; } dn->dn_free_txg = tx->tx_txg; mutex_exit(&dn->dn_mtx); - /* - * If the dnode is already dirty, it needs to be moved from - * the dirty list to the free list. - */ - mutex_enter(&dn->dn_objset->os_lock); - if (list_link_active(&dn->dn_dirty_link[txgoff])) { - list_remove(&dn->dn_objset->os_dirty_dnodes[txgoff], dn); - list_insert_tail(&dn->dn_objset->os_free_dnodes[txgoff], dn); - mutex_exit(&dn->dn_objset->os_lock); - } else { - mutex_exit(&dn->dn_objset->os_lock); - dnode_setdirty(dn, tx); - } + dnode_setdirty(dn, tx); } /* * Try to change the block size for the indicated dnode. This can only * succeed if there are no blocks allocated or dirty beyond first block */ int dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx) { dmu_buf_impl_t *db; int err; ASSERT3U(size, <=, spa_maxblocksize(dmu_objset_spa(dn->dn_objset))); if (size == 0) size = SPA_MINBLOCKSIZE; else size = P2ROUNDUP(size, SPA_MINBLOCKSIZE); if (ibs == dn->dn_indblkshift) ibs = 0; if (size >> SPA_MINBLOCKSHIFT == dn->dn_datablkszsec && ibs == 0) return (0); rw_enter(&dn->dn_struct_rwlock, RW_WRITER); /* Check for any allocated blocks beyond the first */ if (dn->dn_maxblkid != 0) goto fail; mutex_enter(&dn->dn_dbufs_mtx); for (db = avl_first(&dn->dn_dbufs); db != NULL; db = AVL_NEXT(&dn->dn_dbufs, db)) { if (db->db_blkid != 0 && db->db_blkid != DMU_BONUS_BLKID && db->db_blkid != DMU_SPILL_BLKID) { mutex_exit(&dn->dn_dbufs_mtx); goto fail; } } mutex_exit(&dn->dn_dbufs_mtx); if (ibs && dn->dn_nlevels != 1) goto fail; /* resize the old block */ err = dbuf_hold_impl(dn, 0, 0, TRUE, FALSE, FTAG, &db); if (err == 0) dbuf_new_size(db, size, tx); else if (err != ENOENT) goto fail; dnode_setdblksz(dn, size); dnode_setdirty(dn, tx); dn->dn_next_blksz[tx->tx_txg&TXG_MASK] = size; if (ibs) { dn->dn_indblkshift = ibs; dn->dn_next_indblkshift[tx->tx_txg&TXG_MASK] = ibs; } /* rele after we have fixed the blocksize in the dnode */ if (db) dbuf_rele(db, FTAG); rw_exit(&dn->dn_struct_rwlock); return (0); fail: rw_exit(&dn->dn_struct_rwlock); return (SET_ERROR(ENOTSUP)); } /* read-holding callers must not rely on the lock being continuously held */ void dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx, boolean_t have_read) { uint64_t txgoff = tx->tx_txg & TXG_MASK; int epbs, new_nlevels; uint64_t sz; ASSERT(blkid != DMU_BONUS_BLKID); ASSERT(have_read ? RW_READ_HELD(&dn->dn_struct_rwlock) : RW_WRITE_HELD(&dn->dn_struct_rwlock)); /* * if we have a read-lock, check to see if we need to do any work * before upgrading to a write-lock. */ if (have_read) { if (blkid <= dn->dn_maxblkid) return; if (!rw_tryupgrade(&dn->dn_struct_rwlock)) { rw_exit(&dn->dn_struct_rwlock); rw_enter(&dn->dn_struct_rwlock, RW_WRITER); } } if (blkid <= dn->dn_maxblkid) goto out; dn->dn_maxblkid = blkid; /* * Compute the number of levels necessary to support the new maxblkid. */ new_nlevels = 1; epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; for (sz = dn->dn_nblkptr; sz <= blkid && sz >= dn->dn_nblkptr; sz <<= epbs) new_nlevels++; if (new_nlevels > dn->dn_nlevels) { int old_nlevels = dn->dn_nlevels; dmu_buf_impl_t *db; list_t *list; dbuf_dirty_record_t *new, *dr, *dr_next; dn->dn_nlevels = new_nlevels; ASSERT3U(new_nlevels, >, dn->dn_next_nlevels[txgoff]); dn->dn_next_nlevels[txgoff] = new_nlevels; /* dirty the left indirects */ db = dbuf_hold_level(dn, old_nlevels, 0, FTAG); ASSERT(db != NULL); new = dbuf_dirty(db, tx); dbuf_rele(db, FTAG); /* transfer the dirty records to the new indirect */ mutex_enter(&dn->dn_mtx); mutex_enter(&new->dt.di.dr_mtx); list = &dn->dn_dirty_records[txgoff]; for (dr = list_head(list); dr; dr = dr_next) { dr_next = list_next(&dn->dn_dirty_records[txgoff], dr); if (dr->dr_dbuf->db_level != new_nlevels-1 && dr->dr_dbuf->db_blkid != DMU_BONUS_BLKID && dr->dr_dbuf->db_blkid != DMU_SPILL_BLKID) { ASSERT(dr->dr_dbuf->db_level == old_nlevels-1); list_remove(&dn->dn_dirty_records[txgoff], dr); list_insert_tail(&new->dt.di.dr_children, dr); dr->dr_parent = new; } } mutex_exit(&new->dt.di.dr_mtx); mutex_exit(&dn->dn_mtx); } out: if (have_read) rw_downgrade(&dn->dn_struct_rwlock); } static void dnode_dirty_l1(dnode_t *dn, uint64_t l1blkid, dmu_tx_t *tx) { dmu_buf_impl_t *db = dbuf_hold_level(dn, 1, l1blkid, FTAG); if (db != NULL) { dmu_buf_will_dirty(&db->db, tx); dbuf_rele(db, FTAG); } } void dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx) { dmu_buf_impl_t *db; uint64_t blkoff, blkid, nblks; int blksz, blkshift, head, tail; int trunc = FALSE; int epbs; rw_enter(&dn->dn_struct_rwlock, RW_WRITER); blksz = dn->dn_datablksz; blkshift = dn->dn_datablkshift; epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; if (len == DMU_OBJECT_END) { len = UINT64_MAX - off; trunc = TRUE; } /* * First, block align the region to free: */ if (ISP2(blksz)) { head = P2NPHASE(off, blksz); blkoff = P2PHASE(off, blksz); if ((off >> blkshift) > dn->dn_maxblkid) goto out; } else { ASSERT(dn->dn_maxblkid == 0); if (off == 0 && len >= blksz) { /* * Freeing the whole block; fast-track this request. * Note that we won't dirty any indirect blocks, * which is fine because we will be freeing the entire * file and thus all indirect blocks will be freed * by free_children(). */ blkid = 0; nblks = 1; goto done; } else if (off >= blksz) { /* Freeing past end-of-data */ goto out; } else { /* Freeing part of the block. */ head = blksz - off; ASSERT3U(head, >, 0); } blkoff = off; } /* zero out any partial block data at the start of the range */ if (head) { ASSERT3U(blkoff + head, ==, blksz); if (len < head) head = len; if (dbuf_hold_impl(dn, 0, dbuf_whichblock(dn, 0, off), TRUE, FALSE, FTAG, &db) == 0) { caddr_t data; /* don't dirty if it isn't on disk and isn't dirty */ if (db->db_last_dirty || (db->db_blkptr && !BP_IS_HOLE(db->db_blkptr))) { rw_exit(&dn->dn_struct_rwlock); dmu_buf_will_dirty(&db->db, tx); rw_enter(&dn->dn_struct_rwlock, RW_WRITER); data = db->db.db_data; bzero(data + blkoff, head); } dbuf_rele(db, FTAG); } off += head; len -= head; } /* If the range was less than one block, we're done */ if (len == 0) goto out; /* If the remaining range is past end of file, we're done */ if ((off >> blkshift) > dn->dn_maxblkid) goto out; ASSERT(ISP2(blksz)); if (trunc) tail = 0; else tail = P2PHASE(len, blksz); ASSERT0(P2PHASE(off, blksz)); /* zero out any partial block data at the end of the range */ if (tail) { if (len < tail) tail = len; if (dbuf_hold_impl(dn, 0, dbuf_whichblock(dn, 0, off+len), TRUE, FALSE, FTAG, &db) == 0) { /* don't dirty if not on disk and not dirty */ if (db->db_last_dirty || (db->db_blkptr && !BP_IS_HOLE(db->db_blkptr))) { rw_exit(&dn->dn_struct_rwlock); dmu_buf_will_dirty(&db->db, tx); rw_enter(&dn->dn_struct_rwlock, RW_WRITER); bzero(db->db.db_data, tail); } dbuf_rele(db, FTAG); } len -= tail; } /* If the range did not include a full block, we are done */ if (len == 0) goto out; ASSERT(IS_P2ALIGNED(off, blksz)); ASSERT(trunc || IS_P2ALIGNED(len, blksz)); blkid = off >> blkshift; nblks = len >> blkshift; if (trunc) nblks += 1; /* * Dirty all the indirect blocks in this range. Note that only * the first and last indirect blocks can actually be written * (if they were partially freed) -- they must be dirtied, even if * they do not exist on disk yet. The interior blocks will * be freed by free_children(), so they will not actually be written. * Even though these interior blocks will not be written, we * dirty them for two reasons: * * - It ensures that the indirect blocks remain in memory until * syncing context. (They have already been prefetched by * dmu_tx_hold_free(), so we don't have to worry about reading * them serially here.) * * - The dirty space accounting will put pressure on the txg sync * mechanism to begin syncing, and to delay transactions if there * is a large amount of freeing. Even though these indirect * blocks will not be written, we could need to write the same * amount of space if we copy the freed BPs into deadlists. */ if (dn->dn_nlevels > 1) { uint64_t first, last; first = blkid >> epbs; dnode_dirty_l1(dn, first, tx); if (trunc) last = dn->dn_maxblkid >> epbs; else last = (blkid + nblks - 1) >> epbs; if (last != first) dnode_dirty_l1(dn, last, tx); int shift = dn->dn_datablkshift + dn->dn_indblkshift - SPA_BLKPTRSHIFT; for (uint64_t i = first + 1; i < last; i++) { /* * Set i to the blockid of the next non-hole * level-1 indirect block at or after i. Note * that dnode_next_offset() operates in terms of * level-0-equivalent bytes. */ uint64_t ibyte = i << shift; int err = dnode_next_offset(dn, DNODE_FIND_HAVELOCK, &ibyte, 2, 1, 0); i = ibyte >> shift; if (i >= last) break; /* * Normally we should not see an error, either * from dnode_next_offset() or dbuf_hold_level() * (except for ESRCH from dnode_next_offset). * If there is an i/o error, then when we read * this block in syncing context, it will use * ZIO_FLAG_MUSTSUCCEED, and thus hang/panic according * to the "failmode" property. dnode_next_offset() * doesn't have a flag to indicate MUSTSUCCEED. */ if (err != 0) break; dnode_dirty_l1(dn, i, tx); } } done: /* * Add this range to the dnode range list. * We will finish up this free operation in the syncing phase. */ mutex_enter(&dn->dn_mtx); int txgoff = tx->tx_txg & TXG_MASK; if (dn->dn_free_ranges[txgoff] == NULL) { dn->dn_free_ranges[txgoff] = range_tree_create(NULL, NULL, &dn->dn_mtx); } range_tree_clear(dn->dn_free_ranges[txgoff], blkid, nblks); range_tree_add(dn->dn_free_ranges[txgoff], blkid, nblks); dprintf_dnode(dn, "blkid=%llu nblks=%llu txg=%llu\n", blkid, nblks, tx->tx_txg); mutex_exit(&dn->dn_mtx); dbuf_free_range(dn, blkid, blkid + nblks - 1, tx); dnode_setdirty(dn, tx); out: rw_exit(&dn->dn_struct_rwlock); } static boolean_t dnode_spill_freed(dnode_t *dn) { int i; mutex_enter(&dn->dn_mtx); for (i = 0; i < TXG_SIZE; i++) { if (dn->dn_rm_spillblk[i] == DN_KILL_SPILLBLK) break; } mutex_exit(&dn->dn_mtx); return (i < TXG_SIZE); } /* return TRUE if this blkid was freed in a recent txg, or FALSE if it wasn't */ uint64_t dnode_block_freed(dnode_t *dn, uint64_t blkid) { void *dp = spa_get_dsl(dn->dn_objset->os_spa); int i; if (blkid == DMU_BONUS_BLKID) return (FALSE); /* * If we're in the process of opening the pool, dp will not be * set yet, but there shouldn't be anything dirty. */ if (dp == NULL) return (FALSE); if (dn->dn_free_txg) return (TRUE); if (blkid == DMU_SPILL_BLKID) return (dnode_spill_freed(dn)); mutex_enter(&dn->dn_mtx); for (i = 0; i < TXG_SIZE; i++) { if (dn->dn_free_ranges[i] != NULL && range_tree_contains(dn->dn_free_ranges[i], blkid, 1)) break; } mutex_exit(&dn->dn_mtx); return (i < TXG_SIZE); } /* call from syncing context when we actually write/free space for this dnode */ void dnode_diduse_space(dnode_t *dn, int64_t delta) { uint64_t space; dprintf_dnode(dn, "dn=%p dnp=%p used=%llu delta=%lld\n", dn, dn->dn_phys, (u_longlong_t)dn->dn_phys->dn_used, (longlong_t)delta); mutex_enter(&dn->dn_mtx); space = DN_USED_BYTES(dn->dn_phys); if (delta > 0) { ASSERT3U(space + delta, >=, space); /* no overflow */ } else { ASSERT3U(space, >=, -delta); /* no underflow */ } space += delta; if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_DNODE_BYTES) { ASSERT((dn->dn_phys->dn_flags & DNODE_FLAG_USED_BYTES) == 0); ASSERT0(P2PHASE(space, 1<dn_phys->dn_used = space >> DEV_BSHIFT; } else { dn->dn_phys->dn_used = space; dn->dn_phys->dn_flags |= DNODE_FLAG_USED_BYTES; } mutex_exit(&dn->dn_mtx); } /* * Scans a block at the indicated "level" looking for a hole or data, * depending on 'flags'. * * If level > 0, then we are scanning an indirect block looking at its * pointers. If level == 0, then we are looking at a block of dnodes. * * If we don't find what we are looking for in the block, we return ESRCH. * Otherwise, return with *offset pointing to the beginning (if searching * forwards) or end (if searching backwards) of the range covered by the * block pointer we matched on (or dnode). * * The basic search algorithm used below by dnode_next_offset() is to * use this function to search up the block tree (widen the search) until * we find something (i.e., we don't return ESRCH) and then search back * down the tree (narrow the search) until we reach our original search * level. */ static int dnode_next_offset_level(dnode_t *dn, int flags, uint64_t *offset, int lvl, uint64_t blkfill, uint64_t txg) { dmu_buf_impl_t *db = NULL; void *data = NULL; uint64_t epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; uint64_t epb = 1ULL << epbs; uint64_t minfill, maxfill; boolean_t hole; int i, inc, error, span; dprintf("probing object %llu offset %llx level %d of %u\n", dn->dn_object, *offset, lvl, dn->dn_phys->dn_nlevels); hole = ((flags & DNODE_FIND_HOLE) != 0); inc = (flags & DNODE_FIND_BACKWARDS) ? -1 : 1; ASSERT(txg == 0 || !hole); if (lvl == dn->dn_phys->dn_nlevels) { error = 0; epb = dn->dn_phys->dn_nblkptr; data = dn->dn_phys->dn_blkptr; } else { uint64_t blkid = dbuf_whichblock(dn, lvl, *offset); error = dbuf_hold_impl(dn, lvl, blkid, TRUE, FALSE, FTAG, &db); if (error) { if (error != ENOENT) return (error); if (hole) return (0); /* * This can only happen when we are searching up * the block tree for data. We don't really need to * adjust the offset, as we will just end up looking * at the pointer to this block in its parent, and its * going to be unallocated, so we will skip over it. */ return (SET_ERROR(ESRCH)); } error = dbuf_read(db, NULL, DB_RF_CANFAIL | DB_RF_HAVESTRUCT); if (error) { dbuf_rele(db, FTAG); return (error); } data = db->db.db_data; } if (db != NULL && txg != 0 && (db->db_blkptr == NULL || db->db_blkptr->blk_birth <= txg || BP_IS_HOLE(db->db_blkptr))) { /* * This can only happen when we are searching up the tree * and these conditions mean that we need to keep climbing. */ error = SET_ERROR(ESRCH); } else if (lvl == 0) { dnode_phys_t *dnp = data; span = DNODE_SHIFT; ASSERT(dn->dn_type == DMU_OT_DNODE); for (i = (*offset >> span) & (blkfill - 1); i >= 0 && i < blkfill; i += inc) { if ((dnp[i].dn_type == DMU_OT_NONE) == hole) break; *offset += (1ULL << span) * inc; } if (i < 0 || i == blkfill) error = SET_ERROR(ESRCH); } else { blkptr_t *bp = data; uint64_t start = *offset; span = (lvl - 1) * epbs + dn->dn_datablkshift; minfill = 0; maxfill = blkfill << ((lvl - 1) * epbs); if (hole) maxfill--; else minfill++; *offset = *offset >> span; for (i = BF64_GET(*offset, 0, epbs); i >= 0 && i < epb; i += inc) { if (BP_GET_FILL(&bp[i]) >= minfill && BP_GET_FILL(&bp[i]) <= maxfill && (hole || bp[i].blk_birth > txg)) break; if (inc > 0 || *offset > 0) *offset += inc; } *offset = *offset << span; if (inc < 0) { /* traversing backwards; position offset at the end */ ASSERT3U(*offset, <=, start); *offset = MIN(*offset + (1ULL << span) - 1, start); } else if (*offset < start) { *offset = start; } if (i < 0 || i >= epb) error = SET_ERROR(ESRCH); } if (db) dbuf_rele(db, FTAG); return (error); } /* * Find the next hole, data, or sparse region at or after *offset. * The value 'blkfill' tells us how many items we expect to find * in an L0 data block; this value is 1 for normal objects, * DNODES_PER_BLOCK for the meta dnode, and some fraction of * DNODES_PER_BLOCK when searching for sparse regions thereof. * * Examples: * * dnode_next_offset(dn, flags, offset, 1, 1, 0); * Finds the next/previous hole/data in a file. * Used in dmu_offset_next(). * * dnode_next_offset(mdn, flags, offset, 0, DNODES_PER_BLOCK, txg); * Finds the next free/allocated dnode an objset's meta-dnode. * Only finds objects that have new contents since txg (ie. * bonus buffer changes and content removal are ignored). * Used in dmu_object_next(). * * dnode_next_offset(mdn, DNODE_FIND_HOLE, offset, 2, DNODES_PER_BLOCK >> 2, 0); * Finds the next L2 meta-dnode bp that's at most 1/4 full. * Used in dmu_object_alloc(). */ int dnode_next_offset(dnode_t *dn, int flags, uint64_t *offset, int minlvl, uint64_t blkfill, uint64_t txg) { uint64_t initial_offset = *offset; int lvl, maxlvl; int error = 0; if (!(flags & DNODE_FIND_HAVELOCK)) rw_enter(&dn->dn_struct_rwlock, RW_READER); if (dn->dn_phys->dn_nlevels == 0) { error = SET_ERROR(ESRCH); goto out; } if (dn->dn_datablkshift == 0) { if (*offset < dn->dn_datablksz) { if (flags & DNODE_FIND_HOLE) *offset = dn->dn_datablksz; } else { error = SET_ERROR(ESRCH); } goto out; } maxlvl = dn->dn_phys->dn_nlevels; for (lvl = minlvl; lvl <= maxlvl; lvl++) { error = dnode_next_offset_level(dn, flags, offset, lvl, blkfill, txg); if (error != ESRCH) break; } while (error == 0 && --lvl >= minlvl) { error = dnode_next_offset_level(dn, flags, offset, lvl, blkfill, txg); } /* * There's always a "virtual hole" at the end of the object, even * if all BP's which physically exist are non-holes. */ if ((flags & DNODE_FIND_HOLE) && error == ESRCH && txg == 0 && minlvl == 1 && blkfill == 1 && !(flags & DNODE_FIND_BACKWARDS)) { error = 0; } if (error == 0 && (flags & DNODE_FIND_BACKWARDS ? initial_offset < *offset : initial_offset > *offset)) error = SET_ERROR(ESRCH); out: if (!(flags & DNODE_FIND_HAVELOCK)) rw_exit(&dn->dn_struct_rwlock); return (error); } Index: vendor-sys/illumos/dist/uts/common/fs/zfs/dnode_sync.c =================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/dnode_sync.c (revision 316916) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/dnode_sync.c (revision 316917) @@ -1,728 +1,728 @@ /* * 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 http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. - * Copyright (c) 2012, 2016 by Delphix. All rights reserved. + * Copyright (c) 2012, 2017 by Delphix. All rights reserved. * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. */ #include #include #include #include #include #include #include #include #include #include static void dnode_increase_indirection(dnode_t *dn, dmu_tx_t *tx) { dmu_buf_impl_t *db; int txgoff = tx->tx_txg & TXG_MASK; int nblkptr = dn->dn_phys->dn_nblkptr; int old_toplvl = dn->dn_phys->dn_nlevels - 1; int new_level = dn->dn_next_nlevels[txgoff]; int i; rw_enter(&dn->dn_struct_rwlock, RW_WRITER); /* this dnode can't be paged out because it's dirty */ ASSERT(dn->dn_phys->dn_type != DMU_OT_NONE); ASSERT(RW_WRITE_HELD(&dn->dn_struct_rwlock)); ASSERT(new_level > 1 && dn->dn_phys->dn_nlevels > 0); db = dbuf_hold_level(dn, dn->dn_phys->dn_nlevels, 0, FTAG); ASSERT(db != NULL); dn->dn_phys->dn_nlevels = new_level; dprintf("os=%p obj=%llu, increase to %d\n", dn->dn_objset, dn->dn_object, dn->dn_phys->dn_nlevels); /* transfer dnode's block pointers to new indirect block */ (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED|DB_RF_HAVESTRUCT); ASSERT(db->db.db_data); ASSERT(arc_released(db->db_buf)); ASSERT3U(sizeof (blkptr_t) * nblkptr, <=, db->db.db_size); bcopy(dn->dn_phys->dn_blkptr, db->db.db_data, sizeof (blkptr_t) * nblkptr); arc_buf_freeze(db->db_buf); /* set dbuf's parent pointers to new indirect buf */ for (i = 0; i < nblkptr; i++) { dmu_buf_impl_t *child = dbuf_find(dn->dn_objset, dn->dn_object, old_toplvl, i); if (child == NULL) continue; #ifdef DEBUG DB_DNODE_ENTER(child); ASSERT3P(DB_DNODE(child), ==, dn); DB_DNODE_EXIT(child); #endif /* DEBUG */ if (child->db_parent && child->db_parent != dn->dn_dbuf) { ASSERT(child->db_parent->db_level == db->db_level); ASSERT(child->db_blkptr != &dn->dn_phys->dn_blkptr[child->db_blkid]); mutex_exit(&child->db_mtx); continue; } ASSERT(child->db_parent == NULL || child->db_parent == dn->dn_dbuf); child->db_parent = db; dbuf_add_ref(db, child); if (db->db.db_data) child->db_blkptr = (blkptr_t *)db->db.db_data + i; else child->db_blkptr = NULL; dprintf_dbuf_bp(child, child->db_blkptr, "changed db_blkptr to new indirect %s", ""); mutex_exit(&child->db_mtx); } bzero(dn->dn_phys->dn_blkptr, sizeof (blkptr_t) * nblkptr); dbuf_rele(db, FTAG); rw_exit(&dn->dn_struct_rwlock); } static void free_blocks(dnode_t *dn, blkptr_t *bp, int num, dmu_tx_t *tx) { dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset; uint64_t bytesfreed = 0; dprintf("ds=%p obj=%llx num=%d\n", ds, dn->dn_object, num); for (int i = 0; i < num; i++, bp++) { if (BP_IS_HOLE(bp)) continue; bytesfreed += dsl_dataset_block_kill(ds, bp, tx, B_FALSE); ASSERT3U(bytesfreed, <=, DN_USED_BYTES(dn->dn_phys)); /* * Save some useful information on the holes being * punched, including logical size, type, and indirection * level. Retaining birth time enables detection of when * holes are punched for reducing the number of free * records transmitted during a zfs send. */ uint64_t lsize = BP_GET_LSIZE(bp); dmu_object_type_t type = BP_GET_TYPE(bp); uint64_t lvl = BP_GET_LEVEL(bp); bzero(bp, sizeof (blkptr_t)); if (spa_feature_is_active(dn->dn_objset->os_spa, SPA_FEATURE_HOLE_BIRTH)) { BP_SET_LSIZE(bp, lsize); BP_SET_TYPE(bp, type); BP_SET_LEVEL(bp, lvl); BP_SET_BIRTH(bp, dmu_tx_get_txg(tx), 0); } } dnode_diduse_space(dn, -bytesfreed); } #ifdef ZFS_DEBUG static void free_verify(dmu_buf_impl_t *db, uint64_t start, uint64_t end, dmu_tx_t *tx) { int off, num; int i, err, epbs; uint64_t txg = tx->tx_txg; dnode_t *dn; DB_DNODE_ENTER(db); dn = DB_DNODE(db); epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; off = start - (db->db_blkid * 1<=, 0); ASSERT3U(num, >=, 0); ASSERT3U(db->db_level, >, 0); ASSERT3U(db->db.db_size, ==, 1 << dn->dn_phys->dn_indblkshift); ASSERT3U(off+num, <=, db->db.db_size >> SPA_BLKPTRSHIFT); ASSERT(db->db_blkptr != NULL); for (i = off; i < off+num; i++) { uint64_t *buf; dmu_buf_impl_t *child; dbuf_dirty_record_t *dr; int j; ASSERT(db->db_level == 1); rw_enter(&dn->dn_struct_rwlock, RW_READER); err = dbuf_hold_impl(dn, db->db_level-1, (db->db_blkid << epbs) + i, TRUE, FALSE, FTAG, &child); rw_exit(&dn->dn_struct_rwlock); if (err == ENOENT) continue; ASSERT(err == 0); ASSERT(child->db_level == 0); dr = child->db_last_dirty; while (dr && dr->dr_txg > txg) dr = dr->dr_next; ASSERT(dr == NULL || dr->dr_txg == txg); /* data_old better be zeroed */ if (dr) { buf = dr->dt.dl.dr_data->b_data; for (j = 0; j < child->db.db_size >> 3; j++) { if (buf[j] != 0) { panic("freed data not zero: " "child=%p i=%d off=%d num=%d\n", (void *)child, i, off, num); } } } /* * db_data better be zeroed unless it's dirty in a * future txg. */ mutex_enter(&child->db_mtx); buf = child->db.db_data; if (buf != NULL && child->db_state != DB_FILL && child->db_last_dirty == NULL) { for (j = 0; j < child->db.db_size >> 3; j++) { if (buf[j] != 0) { panic("freed data not zero: " "child=%p i=%d off=%d num=%d\n", (void *)child, i, off, num); } } } mutex_exit(&child->db_mtx); dbuf_rele(child, FTAG); } DB_DNODE_EXIT(db); } #endif static void free_children(dmu_buf_impl_t *db, uint64_t blkid, uint64_t nblks, dmu_tx_t *tx) { dnode_t *dn; blkptr_t *bp; dmu_buf_impl_t *subdb; uint64_t start, end, dbstart, dbend; unsigned int epbs, shift, i; /* * There is a small possibility that this block will not be cached: * 1 - if level > 1 and there are no children with level <= 1 * 2 - if this block was evicted since we read it from * dmu_tx_hold_free(). */ if (db->db_state != DB_CACHED) (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED); dbuf_release_bp(db); bp = db->db.db_data; DB_DNODE_ENTER(db); dn = DB_DNODE(db); epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; ASSERT3U(epbs, <, 31); shift = (db->db_level - 1) * epbs; dbstart = db->db_blkid << epbs; start = blkid >> shift; if (dbstart < start) { bp += start - dbstart; } else { start = dbstart; } dbend = ((db->db_blkid + 1) << epbs) - 1; end = (blkid + nblks - 1) >> shift; if (dbend <= end) end = dbend; ASSERT3U(start, <=, end); if (db->db_level == 1) { FREE_VERIFY(db, start, end, tx); free_blocks(dn, bp, end-start+1, tx); } else { for (uint64_t id = start; id <= end; id++, bp++) { if (BP_IS_HOLE(bp)) continue; rw_enter(&dn->dn_struct_rwlock, RW_READER); VERIFY0(dbuf_hold_impl(dn, db->db_level - 1, id, TRUE, FALSE, FTAG, &subdb)); rw_exit(&dn->dn_struct_rwlock); ASSERT3P(bp, ==, subdb->db_blkptr); free_children(subdb, blkid, nblks, tx); dbuf_rele(subdb, FTAG); } } /* If this whole block is free, free ourself too. */ for (i = 0, bp = db->db.db_data; i < 1 << epbs; i++, bp++) { if (!BP_IS_HOLE(bp)) break; } if (i == 1 << epbs) { /* * We only found holes. Grab the rwlock to prevent * anybody from reading the blocks we're about to * zero out. */ rw_enter(&dn->dn_struct_rwlock, RW_WRITER); bzero(db->db.db_data, db->db.db_size); rw_exit(&dn->dn_struct_rwlock); free_blocks(dn, db->db_blkptr, 1, tx); } else { /* * Partial block free; must be marked dirty so that it * will be written out. */ ASSERT(db->db_dirtycnt > 0); } DB_DNODE_EXIT(db); arc_buf_freeze(db->db_buf); } /* * Traverse the indicated range of the provided file * and "free" all the blocks contained there. */ static void dnode_sync_free_range_impl(dnode_t *dn, uint64_t blkid, uint64_t nblks, dmu_tx_t *tx) { blkptr_t *bp = dn->dn_phys->dn_blkptr; int dnlevel = dn->dn_phys->dn_nlevels; boolean_t trunc = B_FALSE; if (blkid > dn->dn_phys->dn_maxblkid) return; ASSERT(dn->dn_phys->dn_maxblkid < UINT64_MAX); if (blkid + nblks > dn->dn_phys->dn_maxblkid) { nblks = dn->dn_phys->dn_maxblkid - blkid + 1; trunc = B_TRUE; } /* There are no indirect blocks in the object */ if (dnlevel == 1) { if (blkid >= dn->dn_phys->dn_nblkptr) { /* this range was never made persistent */ return; } ASSERT3U(blkid + nblks, <=, dn->dn_phys->dn_nblkptr); free_blocks(dn, bp + blkid, nblks, tx); } else { int shift = (dnlevel - 1) * (dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT); int start = blkid >> shift; int end = (blkid + nblks - 1) >> shift; dmu_buf_impl_t *db; ASSERT(start < dn->dn_phys->dn_nblkptr); bp += start; for (int i = start; i <= end; i++, bp++) { if (BP_IS_HOLE(bp)) continue; rw_enter(&dn->dn_struct_rwlock, RW_READER); VERIFY0(dbuf_hold_impl(dn, dnlevel - 1, i, TRUE, FALSE, FTAG, &db)); rw_exit(&dn->dn_struct_rwlock); free_children(db, blkid, nblks, tx); dbuf_rele(db, FTAG); } } if (trunc) { dn->dn_phys->dn_maxblkid = blkid == 0 ? 0 : blkid - 1; uint64_t off = (dn->dn_phys->dn_maxblkid + 1) * (dn->dn_phys->dn_datablkszsec << SPA_MINBLOCKSHIFT); ASSERT(off < dn->dn_phys->dn_maxblkid || dn->dn_phys->dn_maxblkid == 0 || dnode_next_offset(dn, 0, &off, 1, 1, 0) != 0); } } typedef struct dnode_sync_free_range_arg { dnode_t *dsfra_dnode; dmu_tx_t *dsfra_tx; } dnode_sync_free_range_arg_t; static void dnode_sync_free_range(void *arg, uint64_t blkid, uint64_t nblks) { dnode_sync_free_range_arg_t *dsfra = arg; dnode_t *dn = dsfra->dsfra_dnode; mutex_exit(&dn->dn_mtx); dnode_sync_free_range_impl(dn, blkid, nblks, dsfra->dsfra_tx); mutex_enter(&dn->dn_mtx); } /* * Try to kick all the dnode's dbufs out of the cache... */ void dnode_evict_dbufs(dnode_t *dn) { dmu_buf_impl_t db_marker; dmu_buf_impl_t *db, *db_next; mutex_enter(&dn->dn_dbufs_mtx); for (db = avl_first(&dn->dn_dbufs); db != NULL; db = db_next) { #ifdef DEBUG DB_DNODE_ENTER(db); ASSERT3P(DB_DNODE(db), ==, dn); DB_DNODE_EXIT(db); #endif /* DEBUG */ mutex_enter(&db->db_mtx); if (db->db_state != DB_EVICTING && refcount_is_zero(&db->db_holds)) { db_marker.db_level = db->db_level; db_marker.db_blkid = db->db_blkid; db_marker.db_state = DB_SEARCH; avl_insert_here(&dn->dn_dbufs, &db_marker, db, AVL_BEFORE); dbuf_destroy(db); db_next = AVL_NEXT(&dn->dn_dbufs, &db_marker); avl_remove(&dn->dn_dbufs, &db_marker); } else { db->db_pending_evict = TRUE; mutex_exit(&db->db_mtx); db_next = AVL_NEXT(&dn->dn_dbufs, db); } } mutex_exit(&dn->dn_dbufs_mtx); dnode_evict_bonus(dn); } void dnode_evict_bonus(dnode_t *dn) { rw_enter(&dn->dn_struct_rwlock, RW_WRITER); if (dn->dn_bonus != NULL) { if (refcount_is_zero(&dn->dn_bonus->db_holds)) { mutex_enter(&dn->dn_bonus->db_mtx); dbuf_destroy(dn->dn_bonus); dn->dn_bonus = NULL; } else { dn->dn_bonus->db_pending_evict = TRUE; } } rw_exit(&dn->dn_struct_rwlock); } static void dnode_undirty_dbufs(list_t *list) { dbuf_dirty_record_t *dr; while (dr = list_head(list)) { dmu_buf_impl_t *db = dr->dr_dbuf; uint64_t txg = dr->dr_txg; if (db->db_level != 0) dnode_undirty_dbufs(&dr->dt.di.dr_children); mutex_enter(&db->db_mtx); /* XXX - use dbuf_undirty()? */ list_remove(list, dr); ASSERT(db->db_last_dirty == dr); db->db_last_dirty = NULL; db->db_dirtycnt -= 1; if (db->db_level == 0) { ASSERT(db->db_blkid == DMU_BONUS_BLKID || dr->dt.dl.dr_data == db->db_buf); dbuf_unoverride(dr); } else { mutex_destroy(&dr->dt.di.dr_mtx); list_destroy(&dr->dt.di.dr_children); } kmem_free(dr, sizeof (dbuf_dirty_record_t)); dbuf_rele_and_unlock(db, (void *)(uintptr_t)txg); } } static void dnode_sync_free(dnode_t *dn, dmu_tx_t *tx) { int txgoff = tx->tx_txg & TXG_MASK; ASSERT(dmu_tx_is_syncing(tx)); /* * Our contents should have been freed in dnode_sync() by the * free range record inserted by the caller of dnode_free(). */ ASSERT0(DN_USED_BYTES(dn->dn_phys)); ASSERT(BP_IS_HOLE(dn->dn_phys->dn_blkptr)); dnode_undirty_dbufs(&dn->dn_dirty_records[txgoff]); dnode_evict_dbufs(dn); /* * XXX - It would be nice to assert this, but we may still * have residual holds from async evictions from the arc... * * zfs_obj_to_path() also depends on this being * commented out. * * ASSERT3U(refcount_count(&dn->dn_holds), ==, 1); */ /* Undirty next bits */ dn->dn_next_nlevels[txgoff] = 0; dn->dn_next_indblkshift[txgoff] = 0; dn->dn_next_blksz[txgoff] = 0; /* ASSERT(blkptrs are zero); */ ASSERT(dn->dn_phys->dn_type != DMU_OT_NONE); ASSERT(dn->dn_type != DMU_OT_NONE); ASSERT(dn->dn_free_txg > 0); if (dn->dn_allocated_txg != dn->dn_free_txg) dmu_buf_will_dirty(&dn->dn_dbuf->db, tx); bzero(dn->dn_phys, sizeof (dnode_phys_t)); mutex_enter(&dn->dn_mtx); dn->dn_type = DMU_OT_NONE; dn->dn_maxblkid = 0; dn->dn_allocated_txg = 0; dn->dn_free_txg = 0; dn->dn_have_spill = B_FALSE; mutex_exit(&dn->dn_mtx); ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); dnode_rele(dn, (void *)(uintptr_t)tx->tx_txg); /* * Now that we've released our hold, the dnode may * be evicted, so we musn't access it. */ } /* * Write out the dnode's dirty buffers. */ void dnode_sync(dnode_t *dn, dmu_tx_t *tx) { dnode_phys_t *dnp = dn->dn_phys; int txgoff = tx->tx_txg & TXG_MASK; list_t *list = &dn->dn_dirty_records[txgoff]; static const dnode_phys_t zerodn = { 0 }; boolean_t kill_spill = B_FALSE; ASSERT(dmu_tx_is_syncing(tx)); ASSERT(dnp->dn_type != DMU_OT_NONE || dn->dn_allocated_txg); ASSERT(dnp->dn_type != DMU_OT_NONE || bcmp(dnp, &zerodn, DNODE_SIZE) == 0); DNODE_VERIFY(dn); ASSERT(dn->dn_dbuf == NULL || arc_released(dn->dn_dbuf->db_buf)); if (dmu_objset_userused_enabled(dn->dn_objset) && !DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { mutex_enter(&dn->dn_mtx); dn->dn_oldused = DN_USED_BYTES(dn->dn_phys); dn->dn_oldflags = dn->dn_phys->dn_flags; dn->dn_phys->dn_flags |= DNODE_FLAG_USERUSED_ACCOUNTED; mutex_exit(&dn->dn_mtx); dmu_objset_userquota_get_ids(dn, B_FALSE, tx); } else { /* Once we account for it, we should always account for it. */ ASSERT(!(dn->dn_phys->dn_flags & DNODE_FLAG_USERUSED_ACCOUNTED)); } mutex_enter(&dn->dn_mtx); if (dn->dn_allocated_txg == tx->tx_txg) { /* The dnode is newly allocated or reallocated */ if (dnp->dn_type == DMU_OT_NONE) { /* this is a first alloc, not a realloc */ dnp->dn_nlevels = 1; dnp->dn_nblkptr = dn->dn_nblkptr; } dnp->dn_type = dn->dn_type; dnp->dn_bonustype = dn->dn_bonustype; dnp->dn_bonuslen = dn->dn_bonuslen; } ASSERT(dnp->dn_nlevels > 1 || BP_IS_HOLE(&dnp->dn_blkptr[0]) || BP_IS_EMBEDDED(&dnp->dn_blkptr[0]) || BP_GET_LSIZE(&dnp->dn_blkptr[0]) == dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT); ASSERT(dnp->dn_nlevels < 2 || BP_IS_HOLE(&dnp->dn_blkptr[0]) || BP_GET_LSIZE(&dnp->dn_blkptr[0]) == 1 << dnp->dn_indblkshift); if (dn->dn_next_type[txgoff] != 0) { dnp->dn_type = dn->dn_type; dn->dn_next_type[txgoff] = 0; } if (dn->dn_next_blksz[txgoff] != 0) { ASSERT(P2PHASE(dn->dn_next_blksz[txgoff], SPA_MINBLOCKSIZE) == 0); ASSERT(BP_IS_HOLE(&dnp->dn_blkptr[0]) || dn->dn_maxblkid == 0 || list_head(list) != NULL || dn->dn_next_blksz[txgoff] >> SPA_MINBLOCKSHIFT == dnp->dn_datablkszsec || range_tree_space(dn->dn_free_ranges[txgoff]) != 0); dnp->dn_datablkszsec = dn->dn_next_blksz[txgoff] >> SPA_MINBLOCKSHIFT; dn->dn_next_blksz[txgoff] = 0; } if (dn->dn_next_bonuslen[txgoff] != 0) { if (dn->dn_next_bonuslen[txgoff] == DN_ZERO_BONUSLEN) dnp->dn_bonuslen = 0; else dnp->dn_bonuslen = dn->dn_next_bonuslen[txgoff]; ASSERT(dnp->dn_bonuslen <= DN_MAX_BONUSLEN); dn->dn_next_bonuslen[txgoff] = 0; } if (dn->dn_next_bonustype[txgoff] != 0) { ASSERT(DMU_OT_IS_VALID(dn->dn_next_bonustype[txgoff])); dnp->dn_bonustype = dn->dn_next_bonustype[txgoff]; dn->dn_next_bonustype[txgoff] = 0; } boolean_t freeing_dnode = dn->dn_free_txg > 0 && dn->dn_free_txg <= tx->tx_txg; /* * Remove the spill block if we have been explicitly asked to * remove it, or if the object is being removed. */ if (dn->dn_rm_spillblk[txgoff] || freeing_dnode) { if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) kill_spill = B_TRUE; dn->dn_rm_spillblk[txgoff] = 0; } if (dn->dn_next_indblkshift[txgoff] != 0) { ASSERT(dnp->dn_nlevels == 1); dnp->dn_indblkshift = dn->dn_next_indblkshift[txgoff]; dn->dn_next_indblkshift[txgoff] = 0; } /* * Just take the live (open-context) values for checksum and compress. * Strictly speaking it's a future leak, but nothing bad happens if we * start using the new checksum or compress algorithm a little early. */ dnp->dn_checksum = dn->dn_checksum; dnp->dn_compress = dn->dn_compress; mutex_exit(&dn->dn_mtx); if (kill_spill) { free_blocks(dn, &dn->dn_phys->dn_spill, 1, tx); mutex_enter(&dn->dn_mtx); dnp->dn_flags &= ~DNODE_FLAG_SPILL_BLKPTR; mutex_exit(&dn->dn_mtx); } /* process all the "freed" ranges in the file */ if (dn->dn_free_ranges[txgoff] != NULL) { dnode_sync_free_range_arg_t dsfra; dsfra.dsfra_dnode = dn; dsfra.dsfra_tx = tx; mutex_enter(&dn->dn_mtx); range_tree_vacate(dn->dn_free_ranges[txgoff], dnode_sync_free_range, &dsfra); range_tree_destroy(dn->dn_free_ranges[txgoff]); dn->dn_free_ranges[txgoff] = NULL; mutex_exit(&dn->dn_mtx); } if (freeing_dnode) { dn->dn_objset->os_freed_dnodes++; dnode_sync_free(dn, tx); return; } if (dn->dn_next_nlevels[txgoff]) { dnode_increase_indirection(dn, tx); dn->dn_next_nlevels[txgoff] = 0; } if (dn->dn_next_nblkptr[txgoff]) { /* this should only happen on a realloc */ ASSERT(dn->dn_allocated_txg == tx->tx_txg); if (dn->dn_next_nblkptr[txgoff] > dnp->dn_nblkptr) { /* zero the new blkptrs we are gaining */ bzero(dnp->dn_blkptr + dnp->dn_nblkptr, sizeof (blkptr_t) * (dn->dn_next_nblkptr[txgoff] - dnp->dn_nblkptr)); #ifdef ZFS_DEBUG } else { int i; ASSERT(dn->dn_next_nblkptr[txgoff] < dnp->dn_nblkptr); /* the blkptrs we are losing better be unallocated */ for (i = dn->dn_next_nblkptr[txgoff]; i < dnp->dn_nblkptr; i++) ASSERT(BP_IS_HOLE(&dnp->dn_blkptr[i])); #endif } mutex_enter(&dn->dn_mtx); dnp->dn_nblkptr = dn->dn_next_nblkptr[txgoff]; dn->dn_next_nblkptr[txgoff] = 0; mutex_exit(&dn->dn_mtx); } dbuf_sync_list(list, dn->dn_phys->dn_nlevels - 1, tx); if (!DMU_OBJECT_IS_SPECIAL(dn->dn_object)) { ASSERT3P(list_head(list), ==, NULL); dnode_rele(dn, (void *)(uintptr_t)tx->tx_txg); } /* * Although we have dropped our reference to the dnode, it * can't be evicted until its written, and we haven't yet * initiated the IO for the dnode's dbuf. */ } Index: vendor-sys/illumos/dist/uts/common/fs/zfs/dsl_dataset.c =================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/dsl_dataset.c (revision 316916) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/dsl_dataset.c (revision 316917) @@ -1,3611 +1,3616 @@ /* * 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 http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2011, 2016 by Delphix. All rights reserved. * Copyright (c) 2014, Joyent, Inc. All rights reserved. * Copyright (c) 2014 RackTop Systems. * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. * Copyright (c) 2014 Integros [integros.com] * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved. * Copyright 2017 Nexenta Systems, 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 /* * The SPA supports block sizes up to 16MB. However, very large blocks * can have an impact on i/o latency (e.g. tying up a spinning disk for * ~300ms), and also potentially on the memory allocator. Therefore, * we do not allow the recordsize to be set larger than zfs_max_recordsize * (default 1MB). Larger blocks can be created by changing this tunable, * and pools with larger blocks can always be imported and used, regardless * of this setting. */ int zfs_max_recordsize = 1 * 1024 * 1024; #define SWITCH64(x, y) \ { \ uint64_t __tmp = (x); \ (x) = (y); \ (y) = __tmp; \ } #define DS_REF_MAX (1ULL << 62) extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds); extern int spa_asize_inflation; static zil_header_t zero_zil; /* * Figure out how much of this delta should be propogated to the dsl_dir * layer. If there's a refreservation, that space has already been * partially accounted for in our ancestors. */ static int64_t parent_delta(dsl_dataset_t *ds, int64_t delta) { dsl_dataset_phys_t *ds_phys; uint64_t old_bytes, new_bytes; if (ds->ds_reserved == 0) return (delta); ds_phys = dsl_dataset_phys(ds); old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved); new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved); ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta)); return (new_bytes - old_bytes); } void dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx) { int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); int compressed = BP_GET_PSIZE(bp); int uncompressed = BP_GET_UCSIZE(bp); int64_t delta; dprintf_bp(bp, "ds=%p", ds); ASSERT(dmu_tx_is_syncing(tx)); /* It could have been compressed away to nothing */ if (BP_IS_HOLE(bp)) return; ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE); ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp))); if (ds == NULL) { dsl_pool_mos_diduse_space(tx->tx_pool, used, compressed, uncompressed); return; } ASSERT3U(bp->blk_birth, >, dsl_dataset_phys(ds)->ds_prev_snap_txg); dmu_buf_will_dirty(ds->ds_dbuf, tx); mutex_enter(&ds->ds_lock); delta = parent_delta(ds, used); dsl_dataset_phys(ds)->ds_referenced_bytes += used; dsl_dataset_phys(ds)->ds_compressed_bytes += compressed; dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed; dsl_dataset_phys(ds)->ds_unique_bytes += used; if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) { ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] = B_TRUE; } spa_feature_t f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp)); if (f != SPA_FEATURE_NONE) ds->ds_feature_activation_needed[f] = B_TRUE; mutex_exit(&ds->ds_lock); dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta, compressed, uncompressed, tx); dsl_dir_transfer_space(ds->ds_dir, used - delta, DD_USED_REFRSRV, DD_USED_HEAD, tx); } int dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx, boolean_t async) { int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); int compressed = BP_GET_PSIZE(bp); int uncompressed = BP_GET_UCSIZE(bp); if (BP_IS_HOLE(bp)) return (0); ASSERT(dmu_tx_is_syncing(tx)); ASSERT(bp->blk_birth <= tx->tx_txg); if (ds == NULL) { dsl_free(tx->tx_pool, tx->tx_txg, bp); dsl_pool_mos_diduse_space(tx->tx_pool, -used, -compressed, -uncompressed); return (used); } ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); ASSERT(!ds->ds_is_snapshot); dmu_buf_will_dirty(ds->ds_dbuf, tx); if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) { int64_t delta; dprintf_bp(bp, "freeing ds=%llu", ds->ds_object); dsl_free(tx->tx_pool, tx->tx_txg, bp); mutex_enter(&ds->ds_lock); ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used || !DS_UNIQUE_IS_ACCURATE(ds)); delta = parent_delta(ds, -used); dsl_dataset_phys(ds)->ds_unique_bytes -= used; mutex_exit(&ds->ds_lock); dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta, -compressed, -uncompressed, tx); dsl_dir_transfer_space(ds->ds_dir, -used - delta, DD_USED_REFRSRV, DD_USED_HEAD, tx); } else { dprintf_bp(bp, "putting on dead list: %s", ""); if (async) { /* * We are here as part of zio's write done callback, * which means we're a zio interrupt thread. We can't * call dsl_deadlist_insert() now because it may block * waiting for I/O. Instead, put bp on the deferred * queue and let dsl_pool_sync() finish the job. */ bplist_append(&ds->ds_pending_deadlist, bp); } else { dsl_deadlist_insert(&ds->ds_deadlist, bp, tx); } ASSERT3U(ds->ds_prev->ds_object, ==, dsl_dataset_phys(ds)->ds_prev_snap_obj); ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0); /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */ if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == ds->ds_object && bp->blk_birth > dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) { dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); mutex_enter(&ds->ds_prev->ds_lock); dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used; mutex_exit(&ds->ds_prev->ds_lock); } if (bp->blk_birth > ds->ds_dir->dd_origin_txg) { dsl_dir_transfer_space(ds->ds_dir, used, DD_USED_HEAD, DD_USED_SNAP, tx); } } mutex_enter(&ds->ds_lock); ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used); dsl_dataset_phys(ds)->ds_referenced_bytes -= used; ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed); dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed; ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed); dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed; mutex_exit(&ds->ds_lock); return (used); } /* * We have to release the fsid syncronously or we risk that a subsequent * mount of the same dataset will fail to unique_insert the fsid. This * failure would manifest itself as the fsid of this dataset changing * between mounts which makes NFS clients quite unhappy. */ static void dsl_dataset_evict_sync(void *dbu) { dsl_dataset_t *ds = dbu; ASSERT(ds->ds_owner == NULL); unique_remove(ds->ds_fsid_guid); } static void dsl_dataset_evict_async(void *dbu) { dsl_dataset_t *ds = dbu; ASSERT(ds->ds_owner == NULL); ds->ds_dbuf = NULL; if (ds->ds_objset != NULL) dmu_objset_evict(ds->ds_objset); if (ds->ds_prev) { dsl_dataset_rele(ds->ds_prev, ds); ds->ds_prev = NULL; } bplist_destroy(&ds->ds_pending_deadlist); if (ds->ds_deadlist.dl_os != NULL) dsl_deadlist_close(&ds->ds_deadlist); if (ds->ds_dir) dsl_dir_async_rele(ds->ds_dir, ds); ASSERT(!list_link_active(&ds->ds_synced_link)); list_destroy(&ds->ds_prop_cbs); mutex_destroy(&ds->ds_lock); mutex_destroy(&ds->ds_opening_lock); mutex_destroy(&ds->ds_sendstream_lock); refcount_destroy(&ds->ds_longholds); rrw_destroy(&ds->ds_bp_rwlock); kmem_free(ds, sizeof (dsl_dataset_t)); } int dsl_dataset_get_snapname(dsl_dataset_t *ds) { dsl_dataset_phys_t *headphys; int err; dmu_buf_t *headdbuf; dsl_pool_t *dp = ds->ds_dir->dd_pool; objset_t *mos = dp->dp_meta_objset; if (ds->ds_snapname[0]) return (0); if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0) return (0); err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, FTAG, &headdbuf); if (err != 0) return (err); headphys = headdbuf->db_data; err = zap_value_search(dp->dp_meta_objset, headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname); dmu_buf_rele(headdbuf, FTAG); return (err); } int dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value) { objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; matchtype_t mt = 0; int err; if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) mt = MT_NORMALIZE; err = zap_lookup_norm(mos, snapobj, name, 8, 1, value, mt, NULL, 0, NULL); if (err == ENOTSUP && (mt & MT_NORMALIZE)) err = zap_lookup(mos, snapobj, name, 8, 1, value); return (err); } int dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx, boolean_t adj_cnt) { objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; matchtype_t mt = 0; int err; dsl_dir_snap_cmtime_update(ds->ds_dir); if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) mt = MT_NORMALIZE; err = zap_remove_norm(mos, snapobj, name, mt, tx); if (err == ENOTSUP && (mt & MT_NORMALIZE)) err = zap_remove(mos, snapobj, name, tx); if (err == 0 && adj_cnt) dsl_fs_ss_count_adjust(ds->ds_dir, -1, DD_FIELD_SNAPSHOT_COUNT, tx); return (err); } boolean_t dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag) { dmu_buf_t *dbuf = ds->ds_dbuf; boolean_t result = B_FALSE; if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset, ds->ds_object, DMU_BONUS_BLKID, tag)) { if (ds == dmu_buf_get_user(dbuf)) result = B_TRUE; else dmu_buf_rele(dbuf, tag); } return (result); } int dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, dsl_dataset_t **dsp) { objset_t *mos = dp->dp_meta_objset; dmu_buf_t *dbuf; dsl_dataset_t *ds; int err; dmu_object_info_t doi; ASSERT(dsl_pool_config_held(dp)); err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); if (err != 0) return (err); /* Make sure dsobj has the correct object type. */ dmu_object_info_from_db(dbuf, &doi); if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) { dmu_buf_rele(dbuf, tag); return (SET_ERROR(EINVAL)); } ds = dmu_buf_get_user(dbuf); if (ds == NULL) { dsl_dataset_t *winner = NULL; ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); ds->ds_dbuf = dbuf; ds->ds_object = dsobj; ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0; mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL); rrw_init(&ds->ds_bp_rwlock, B_FALSE); refcount_create(&ds->ds_longholds); bplist_create(&ds->ds_pending_deadlist); dsl_deadlist_open(&ds->ds_deadlist, mos, dsl_dataset_phys(ds)->ds_deadlist_obj); list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t), offsetof(dmu_sendarg_t, dsa_link)); list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t), offsetof(dsl_prop_cb_record_t, cbr_ds_node)); if (doi.doi_type == DMU_OTN_ZAP_METADATA) { for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { if (!(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET)) continue; err = zap_contains(mos, dsobj, spa_feature_table[f].fi_guid); if (err == 0) { ds->ds_feature_inuse[f] = B_TRUE; } else { ASSERT3U(err, ==, ENOENT); err = 0; } } } err = dsl_dir_hold_obj(dp, dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, &ds->ds_dir); if (err != 0) { mutex_destroy(&ds->ds_lock); mutex_destroy(&ds->ds_opening_lock); mutex_destroy(&ds->ds_sendstream_lock); refcount_destroy(&ds->ds_longholds); bplist_destroy(&ds->ds_pending_deadlist); dsl_deadlist_close(&ds->ds_deadlist); kmem_free(ds, sizeof (dsl_dataset_t)); dmu_buf_rele(dbuf, tag); return (err); } if (!ds->ds_is_snapshot) { ds->ds_snapname[0] = '\0'; if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { err = dsl_dataset_hold_obj(dp, dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev); } if (doi.doi_type == DMU_OTN_ZAP_METADATA) { int zaperr = zap_lookup(mos, ds->ds_object, DS_FIELD_BOOKMARK_NAMES, sizeof (ds->ds_bookmarks), 1, &ds->ds_bookmarks); if (zaperr != ENOENT) VERIFY0(zaperr); } } else { if (zfs_flags & ZFS_DEBUG_SNAPNAMES) err = dsl_dataset_get_snapname(ds); if (err == 0 && dsl_dataset_phys(ds)->ds_userrefs_obj != 0) { err = zap_count( ds->ds_dir->dd_pool->dp_meta_objset, dsl_dataset_phys(ds)->ds_userrefs_obj, &ds->ds_userrefs); } } if (err == 0 && !ds->ds_is_snapshot) { err = dsl_prop_get_int_ds(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &ds->ds_reserved); if (err == 0) { err = dsl_prop_get_int_ds(ds, zfs_prop_to_name(ZFS_PROP_REFQUOTA), &ds->ds_quota); } } else { ds->ds_reserved = ds->ds_quota = 0; } dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict_sync, dsl_dataset_evict_async, &ds->ds_dbuf); if (err == 0) winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu); if (err != 0 || winner != NULL) { bplist_destroy(&ds->ds_pending_deadlist); dsl_deadlist_close(&ds->ds_deadlist); if (ds->ds_prev) dsl_dataset_rele(ds->ds_prev, ds); dsl_dir_rele(ds->ds_dir, ds); mutex_destroy(&ds->ds_lock); mutex_destroy(&ds->ds_opening_lock); mutex_destroy(&ds->ds_sendstream_lock); refcount_destroy(&ds->ds_longholds); kmem_free(ds, sizeof (dsl_dataset_t)); if (err != 0) { dmu_buf_rele(dbuf, tag); return (err); } ds = winner; } else { ds->ds_fsid_guid = unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid); if (ds->ds_fsid_guid != dsl_dataset_phys(ds)->ds_fsid_guid) { zfs_dbgmsg("ds_fsid_guid changed from " "%llx to %llx for pool %s dataset id %llu", (long long) dsl_dataset_phys(ds)->ds_fsid_guid, (long long)ds->ds_fsid_guid, spa_name(dp->dp_spa), dsobj); } } } ASSERT3P(ds->ds_dbuf, ==, dbuf); ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data); ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 || spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN || dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap); *dsp = ds; return (0); } int dsl_dataset_hold(dsl_pool_t *dp, const char *name, void *tag, dsl_dataset_t **dsp) { dsl_dir_t *dd; const char *snapname; uint64_t obj; int err = 0; dsl_dataset_t *ds; err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname); if (err != 0) return (err); ASSERT(dsl_pool_config_held(dp)); obj = dsl_dir_phys(dd)->dd_head_dataset_obj; if (obj != 0) err = dsl_dataset_hold_obj(dp, obj, tag, &ds); else err = SET_ERROR(ENOENT); /* we may be looking for a snapshot */ if (err == 0 && snapname != NULL) { dsl_dataset_t *snap_ds; if (*snapname++ != '@') { dsl_dataset_rele(ds, tag); dsl_dir_rele(dd, FTAG); return (SET_ERROR(ENOENT)); } dprintf("looking for snapshot '%s'\n", snapname); err = dsl_dataset_snap_lookup(ds, snapname, &obj); if (err == 0) err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds); dsl_dataset_rele(ds, tag); if (err == 0) { mutex_enter(&snap_ds->ds_lock); if (snap_ds->ds_snapname[0] == 0) (void) strlcpy(snap_ds->ds_snapname, snapname, sizeof (snap_ds->ds_snapname)); mutex_exit(&snap_ds->ds_lock); ds = snap_ds; } } if (err == 0) *dsp = ds; dsl_dir_rele(dd, FTAG); return (err); } int dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, dsl_dataset_t **dsp) { int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp); if (err != 0) return (err); if (!dsl_dataset_tryown(*dsp, tag)) { dsl_dataset_rele(*dsp, tag); *dsp = NULL; return (SET_ERROR(EBUSY)); } return (0); } int dsl_dataset_own(dsl_pool_t *dp, const char *name, void *tag, dsl_dataset_t **dsp) { int err = dsl_dataset_hold(dp, name, tag, dsp); if (err != 0) return (err); if (!dsl_dataset_tryown(*dsp, tag)) { dsl_dataset_rele(*dsp, tag); return (SET_ERROR(EBUSY)); } return (0); } /* * See the comment above dsl_pool_hold() for details. In summary, a long * hold is used to prevent destruction of a dataset while the pool hold * is dropped, allowing other concurrent operations (e.g. spa_sync()). * * The dataset and pool must be held when this function is called. After it * is called, the pool hold may be released while the dataset is still held * and accessed. */ void dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag) { ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); (void) refcount_add(&ds->ds_longholds, tag); } void dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag) { (void) refcount_remove(&ds->ds_longholds, tag); } /* Return B_TRUE if there are any long holds on this dataset. */ boolean_t dsl_dataset_long_held(dsl_dataset_t *ds) { return (!refcount_is_zero(&ds->ds_longholds)); } void dsl_dataset_name(dsl_dataset_t *ds, char *name) { if (ds == NULL) { (void) strcpy(name, "mos"); } else { dsl_dir_name(ds->ds_dir, name); VERIFY0(dsl_dataset_get_snapname(ds)); if (ds->ds_snapname[0]) { VERIFY3U(strlcat(name, "@", ZFS_MAX_DATASET_NAME_LEN), <, ZFS_MAX_DATASET_NAME_LEN); /* * We use a "recursive" mutex so that we * can call dprintf_ds() with ds_lock held. */ if (!MUTEX_HELD(&ds->ds_lock)) { mutex_enter(&ds->ds_lock); VERIFY3U(strlcat(name, ds->ds_snapname, ZFS_MAX_DATASET_NAME_LEN), <, ZFS_MAX_DATASET_NAME_LEN); mutex_exit(&ds->ds_lock); } else { VERIFY3U(strlcat(name, ds->ds_snapname, ZFS_MAX_DATASET_NAME_LEN), <, ZFS_MAX_DATASET_NAME_LEN); } } } } int dsl_dataset_namelen(dsl_dataset_t *ds) { VERIFY0(dsl_dataset_get_snapname(ds)); mutex_enter(&ds->ds_lock); int len = dsl_dir_namelen(ds->ds_dir) + 1 + strlen(ds->ds_snapname); mutex_exit(&ds->ds_lock); return (len); } void dsl_dataset_rele(dsl_dataset_t *ds, void *tag) { dmu_buf_rele(ds->ds_dbuf, tag); } void dsl_dataset_disown(dsl_dataset_t *ds, void *tag) { ASSERT3P(ds->ds_owner, ==, tag); ASSERT(ds->ds_dbuf != NULL); mutex_enter(&ds->ds_lock); ds->ds_owner = NULL; mutex_exit(&ds->ds_lock); dsl_dataset_long_rele(ds, tag); dsl_dataset_rele(ds, tag); } boolean_t dsl_dataset_tryown(dsl_dataset_t *ds, void *tag) { boolean_t gotit = FALSE; ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); mutex_enter(&ds->ds_lock); if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) { ds->ds_owner = tag; dsl_dataset_long_hold(ds, tag); gotit = TRUE; } mutex_exit(&ds->ds_lock); return (gotit); } boolean_t dsl_dataset_has_owner(dsl_dataset_t *ds) { boolean_t rv; mutex_enter(&ds->ds_lock); rv = (ds->ds_owner != NULL); mutex_exit(&ds->ds_lock); return (rv); } static void dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx) { spa_t *spa = dmu_tx_pool(tx)->dp_spa; objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset; uint64_t zero = 0; VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET); spa_feature_incr(spa, f, tx); dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx); VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid, sizeof (zero), 1, &zero, tx)); } void dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx) { spa_t *spa = dmu_tx_pool(tx)->dp_spa; objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset; VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET); VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx)); spa_feature_decr(spa, f, tx); } uint64_t dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, uint64_t flags, dmu_tx_t *tx) { dsl_pool_t *dp = dd->dd_pool; dmu_buf_t *dbuf; dsl_dataset_phys_t *dsphys; uint64_t dsobj; objset_t *mos = dp->dp_meta_objset; if (origin == NULL) origin = dp->dp_origin_snap; ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp); ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0); ASSERT(dmu_tx_is_syncing(tx)); ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0); dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); dmu_buf_will_dirty(dbuf, tx); dsphys = dbuf->db_data; bzero(dsphys, sizeof (dsl_dataset_phys_t)); dsphys->ds_dir_obj = dd->dd_object; dsphys->ds_flags = flags; dsphys->ds_fsid_guid = unique_create(); (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, sizeof (dsphys->ds_guid)); dsphys->ds_snapnames_zapobj = zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP, DMU_OT_NONE, 0, tx); dsphys->ds_creation_time = gethrestime_sec(); dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg; if (origin == NULL) { dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx); } else { dsl_dataset_t *ohds; /* head of the origin snapshot */ dsphys->ds_prev_snap_obj = origin->ds_object; dsphys->ds_prev_snap_txg = dsl_dataset_phys(origin)->ds_creation_txg; dsphys->ds_referenced_bytes = dsl_dataset_phys(origin)->ds_referenced_bytes; dsphys->ds_compressed_bytes = dsl_dataset_phys(origin)->ds_compressed_bytes; dsphys->ds_uncompressed_bytes = dsl_dataset_phys(origin)->ds_uncompressed_bytes; rrw_enter(&origin->ds_bp_rwlock, RW_READER, FTAG); dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp; rrw_exit(&origin->ds_bp_rwlock, FTAG); /* * Inherit flags that describe the dataset's contents * (INCONSISTENT) or properties (Case Insensitive). */ dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags & (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET); for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { if (origin->ds_feature_inuse[f]) dsl_dataset_activate_feature(dsobj, f, tx); } dmu_buf_will_dirty(origin->ds_dbuf, tx); dsl_dataset_phys(origin)->ds_num_children++; VERIFY0(dsl_dataset_hold_obj(dp, dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj, FTAG, &ohds)); dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist, dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx); dsl_dataset_rele(ohds, FTAG); if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) { if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) { dsl_dataset_phys(origin)->ds_next_clones_obj = zap_create(mos, DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); } VERIFY0(zap_add_int(mos, dsl_dataset_phys(origin)->ds_next_clones_obj, dsobj, tx)); } dmu_buf_will_dirty(dd->dd_dbuf, tx); dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object; if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) { dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); dsl_dir_phys(origin->ds_dir)->dd_clones = zap_create(mos, DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); } VERIFY0(zap_add_int(mos, dsl_dir_phys(origin->ds_dir)->dd_clones, dsobj, tx)); } } if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; dmu_buf_rele(dbuf, FTAG); dmu_buf_will_dirty(dd->dd_dbuf, tx); dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj; return (dsobj); } static void dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx) { objset_t *os; VERIFY0(dmu_objset_from_ds(ds, &os)); if (bcmp(&os->os_zil_header, &zero_zil, sizeof (zero_zil)) != 0) { dsl_pool_t *dp = ds->ds_dir->dd_pool; zio_t *zio; bzero(&os->os_zil_header, sizeof (os->os_zil_header)); zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); dsl_dataset_sync(ds, zio, tx); VERIFY0(zio_wait(zio)); /* dsl_dataset_sync_done will drop this reference. */ dmu_buf_add_ref(ds->ds_dbuf, ds); dsl_dataset_sync_done(ds, tx); } } uint64_t dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx) { dsl_pool_t *dp = pdd->dd_pool; uint64_t dsobj, ddobj; dsl_dir_t *dd; ASSERT(dmu_tx_is_syncing(tx)); ASSERT(lastname[0] != '@'); ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx); VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd)); dsobj = dsl_dataset_create_sync_dd(dd, origin, flags & ~DS_CREATE_FLAG_NODIRTY, tx); dsl_deleg_set_create_perms(dd, tx, cr); /* * Since we're creating a new node we know it's a leaf, so we can * initialize the counts if the limit feature is active. */ if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) { uint64_t cnt = 0; objset_t *os = dd->dd_pool->dp_meta_objset; dsl_dir_zapify(dd, tx); VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, sizeof (cnt), 1, &cnt, tx)); VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, sizeof (cnt), 1, &cnt, tx)); } dsl_dir_rele(dd, FTAG); /* * If we are creating a clone, make sure we zero out any stale * data from the origin snapshots zil header. */ if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) { dsl_dataset_t *ds; VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); dsl_dataset_zero_zil(ds, tx); dsl_dataset_rele(ds, FTAG); } return (dsobj); } /* * The unique space in the head dataset can be calculated by subtracting * the space used in the most recent snapshot, that is still being used * in this file system, from the space currently in use. To figure out * the space in the most recent snapshot still in use, we need to take * the total space used in the snapshot and subtract out the space that * has been freed up since the snapshot was taken. */ void dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) { uint64_t mrs_used; uint64_t dlused, dlcomp, dluncomp; ASSERT(!ds->ds_is_snapshot); if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes; else mrs_used = 0; dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp); ASSERT3U(dlused, <=, mrs_used); dsl_dataset_phys(ds)->ds_unique_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused); if (spa_version(ds->ds_dir->dd_pool->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; } void dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx) { objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; uint64_t count; int err; ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2); err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, obj, tx); /* * The err should not be ENOENT, but 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. * If we knew that the pool was created after * SPA_VERSION_NEXT_CLONES, we could assert that it isn't * ENOENT. However, at least we can check that we don't have * too many entries in the next_clones_obj even after failing to * remove this one. */ if (err != ENOENT) VERIFY0(err); ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, &count)); ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2); } blkptr_t * dsl_dataset_get_blkptr(dsl_dataset_t *ds) { return (&dsl_dataset_phys(ds)->ds_bp); } spa_t * dsl_dataset_get_spa(dsl_dataset_t *ds) { return (ds->ds_dir->dd_pool->dp_spa); } void dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) { dsl_pool_t *dp; if (ds == NULL) /* this is the meta-objset */ return; ASSERT(ds->ds_objset != NULL); if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) panic("dirtying snapshot!"); /* Must not dirty a dataset in the same txg where it got snapshotted. */ ASSERT3U(tx->tx_txg, >, dsl_dataset_phys(ds)->ds_prev_snap_txg); dp = ds->ds_dir->dd_pool; if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { /* up the hold count until we can be written out */ dmu_buf_add_ref(ds->ds_dbuf, ds); } } boolean_t dsl_dataset_is_dirty(dsl_dataset_t *ds) { for (int t = 0; t < TXG_SIZE; t++) { if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets, ds, t)) return (B_TRUE); } return (B_FALSE); } static int dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) { uint64_t asize; if (!dmu_tx_is_syncing(tx)) return (0); /* * If there's an fs-only reservation, any blocks that might become * owned by the snapshot dataset must be accommodated by space * outside of the reservation. */ ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds)); asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved); if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) return (SET_ERROR(ENOSPC)); /* * Propagate any reserved space for this snapshot to other * snapshot checks in this sync group. */ if (asize > 0) dsl_dir_willuse_space(ds->ds_dir, asize, tx); return (0); } typedef struct dsl_dataset_snapshot_arg { nvlist_t *ddsa_snaps; nvlist_t *ddsa_props; nvlist_t *ddsa_errors; cred_t *ddsa_cr; } dsl_dataset_snapshot_arg_t; int dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname, dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr) { int error; uint64_t value; ds->ds_trysnap_txg = tx->tx_txg; if (!dmu_tx_is_syncing(tx)) return (0); /* * We don't allow multiple snapshots of the same txg. If there * is already one, try again. */ if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) return (SET_ERROR(EAGAIN)); /* * Check for conflicting snapshot name. */ error = dsl_dataset_snap_lookup(ds, snapname, &value); if (error == 0) return (SET_ERROR(EEXIST)); if (error != ENOENT) return (error); /* * We don't allow taking snapshots of inconsistent datasets, such as * those into which we are currently receiving. However, if we are * creating this snapshot as part of a receive, this check will be * executed atomically with respect to the completion of the receive * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this * case we ignore this, knowing it will be fixed up for us shortly in * dmu_recv_end_sync(). */ if (!recv && DS_IS_INCONSISTENT(ds)) return (SET_ERROR(EBUSY)); /* * Skip the check for temporary snapshots or if we have already checked * the counts in dsl_dataset_snapshot_check. This means we really only * check the count here when we're receiving a stream. */ if (cnt != 0 && cr != NULL) { error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr); if (error != 0) return (error); } error = dsl_dataset_snapshot_reserve_space(ds, tx); if (error != 0) return (error); return (0); } static int dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx) { dsl_dataset_snapshot_arg_t *ddsa = arg; dsl_pool_t *dp = dmu_tx_pool(tx); nvpair_t *pair; int rv = 0; /* * Pre-compute how many total new snapshots will be created for each * level in the tree and below. This is needed for validating the * snapshot limit when either taking a recursive snapshot or when * taking multiple snapshots. * * The problem is that the counts are not actually adjusted when * we are checking, only when we finally sync. For a single snapshot, * this is easy, the count will increase by 1 at each node up the tree, * but its more complicated for the recursive/multiple snapshot case. * * The dsl_fs_ss_limit_check function does recursively check the count * at each level up the tree but since it is validating each snapshot * independently we need to be sure that we are validating the complete * count for the entire set of snapshots. We do this by rolling up the * counts for each component of the name into an nvlist and then * checking each of those cases with the aggregated count. * * This approach properly handles not only the recursive snapshot * case (where we get all of those on the ddsa_snaps list) but also * the sibling case (e.g. snapshot a/b and a/c so that we will also * validate the limit on 'a' using a count of 2). * * We validate the snapshot names in the third loop and only report * name errors once. */ if (dmu_tx_is_syncing(tx)) { nvlist_t *cnt_track = NULL; cnt_track = fnvlist_alloc(); /* Rollup aggregated counts into the cnt_track list */ for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { char *pdelim; uint64_t val; char nm[MAXPATHLEN]; (void) strlcpy(nm, nvpair_name(pair), sizeof (nm)); pdelim = strchr(nm, '@'); if (pdelim == NULL) continue; *pdelim = '\0'; do { if (nvlist_lookup_uint64(cnt_track, nm, &val) == 0) { /* update existing entry */ fnvlist_add_uint64(cnt_track, nm, val + 1); } else { /* add to list */ fnvlist_add_uint64(cnt_track, nm, 1); } pdelim = strrchr(nm, '/'); if (pdelim != NULL) *pdelim = '\0'; } while (pdelim != NULL); } /* Check aggregated counts at each level */ for (pair = nvlist_next_nvpair(cnt_track, NULL); pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) { int error = 0; char *name; uint64_t cnt = 0; dsl_dataset_t *ds; name = nvpair_name(pair); cnt = fnvpair_value_uint64(pair); ASSERT(cnt > 0); error = dsl_dataset_hold(dp, name, FTAG, &ds); if (error == 0) { error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, ZFS_PROP_SNAPSHOT_LIMIT, NULL, ddsa->ddsa_cr); dsl_dataset_rele(ds, FTAG); } if (error != 0) { if (ddsa->ddsa_errors != NULL) fnvlist_add_int32(ddsa->ddsa_errors, name, error); rv = error; /* only report one error for this check */ break; } } nvlist_free(cnt_track); } for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { int error = 0; dsl_dataset_t *ds; char *name, *atp; char dsname[ZFS_MAX_DATASET_NAME_LEN]; name = nvpair_name(pair); if (strlen(name) >= ZFS_MAX_DATASET_NAME_LEN) error = SET_ERROR(ENAMETOOLONG); if (error == 0) { atp = strchr(name, '@'); if (atp == NULL) error = SET_ERROR(EINVAL); if (error == 0) (void) strlcpy(dsname, name, atp - name + 1); } if (error == 0) error = dsl_dataset_hold(dp, dsname, FTAG, &ds); if (error == 0) { /* passing 0/NULL skips dsl_fs_ss_limit_check */ error = dsl_dataset_snapshot_check_impl(ds, atp + 1, tx, B_FALSE, 0, NULL); dsl_dataset_rele(ds, FTAG); } if (error != 0) { if (ddsa->ddsa_errors != NULL) { fnvlist_add_int32(ddsa->ddsa_errors, name, error); } rv = error; } } return (rv); } void dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname, dmu_tx_t *tx) { dsl_pool_t *dp = ds->ds_dir->dd_pool; dmu_buf_t *dbuf; dsl_dataset_phys_t *dsphys; uint64_t dsobj, crtxg; objset_t *mos = dp->dp_meta_objset; objset_t *os; ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock)); /* * If we are on an old pool, the zil must not be active, in which * case it will be zeroed. Usually zil_suspend() accomplishes this. */ ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP || dmu_objset_from_ds(ds, &os) != 0 || bcmp(&os->os_phys->os_zil_header, &zero_zil, sizeof (zero_zil)) == 0); /* Should not snapshot a dirty dataset. */ ASSERT(!txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets, ds, tx->tx_txg)); dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx); /* * The origin's ds_creation_txg has to be < TXG_INITIAL */ if (strcmp(snapname, ORIGIN_DIR_NAME) == 0) crtxg = 1; else crtxg = tx->tx_txg; dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); dmu_buf_will_dirty(dbuf, tx); dsphys = dbuf->db_data; bzero(dsphys, sizeof (dsl_dataset_phys_t)); dsphys->ds_dir_obj = ds->ds_dir->dd_object; dsphys->ds_fsid_guid = unique_create(); (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, sizeof (dsphys->ds_guid)); dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg; dsphys->ds_next_snap_obj = ds->ds_object; dsphys->ds_num_children = 1; dsphys->ds_creation_time = gethrestime_sec(); dsphys->ds_creation_txg = crtxg; dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj; dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes; dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes; dsphys->ds_uncompressed_bytes = dsl_dataset_phys(ds)->ds_uncompressed_bytes; dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags; rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp; rrw_exit(&ds->ds_bp_rwlock, FTAG); dmu_buf_rele(dbuf, FTAG); for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { if (ds->ds_feature_inuse[f]) dsl_dataset_activate_feature(dsobj, f, tx); } ASSERT3U(ds->ds_prev != 0, ==, dsl_dataset_phys(ds)->ds_prev_snap_obj != 0); if (ds->ds_prev) { uint64_t next_clones_obj = dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj; ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == ds->ds_object || dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1); if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == ds->ds_object) { dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==, dsl_dataset_phys(ds->ds_prev)->ds_creation_txg); dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj; } else if (next_clones_obj != 0) { dsl_dataset_remove_from_next_clones(ds->ds_prev, dsphys->ds_next_snap_obj, tx); VERIFY0(zap_add_int(mos, next_clones_obj, dsobj, tx)); } } /* * If we have a reference-reservation on this dataset, we will * need to increase the amount of refreservation being charged * since our unique space is going to zero. */ if (ds->ds_reserved) { int64_t delta; ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved); dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); } dmu_buf_will_dirty(ds->ds_dbuf, tx); dsl_dataset_phys(ds)->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX, dsl_dataset_phys(ds)->ds_prev_snap_obj, tx); dsl_deadlist_close(&ds->ds_deadlist); dsl_deadlist_open(&ds->ds_deadlist, mos, dsl_dataset_phys(ds)->ds_deadlist_obj); dsl_deadlist_add_key(&ds->ds_deadlist, dsl_dataset_phys(ds)->ds_prev_snap_txg, tx); ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg); dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj; dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg; dsl_dataset_phys(ds)->ds_unique_bytes = 0; if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj, snapname, 8, 1, &dsobj, tx)); if (ds->ds_prev) dsl_dataset_rele(ds->ds_prev, ds); VERIFY0(dsl_dataset_hold_obj(dp, dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev)); dsl_scan_ds_snapshotted(ds, tx); dsl_dir_snap_cmtime_update(ds->ds_dir); spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, ""); } static void dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx) { dsl_dataset_snapshot_arg_t *ddsa = arg; dsl_pool_t *dp = dmu_tx_pool(tx); nvpair_t *pair; for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { dsl_dataset_t *ds; char *name, *atp; char dsname[ZFS_MAX_DATASET_NAME_LEN]; name = nvpair_name(pair); atp = strchr(name, '@'); (void) strlcpy(dsname, name, atp - name + 1); VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds)); dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx); if (ddsa->ddsa_props != NULL) { dsl_props_set_sync_impl(ds->ds_prev, ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx); } dsl_dataset_rele(ds, FTAG); } } /* * The snapshots must all be in the same pool. * All-or-nothing: if there are any failures, nothing will be modified. */ int dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors) { dsl_dataset_snapshot_arg_t ddsa; nvpair_t *pair; boolean_t needsuspend; int error; spa_t *spa; char *firstname; nvlist_t *suspended = NULL; pair = nvlist_next_nvpair(snaps, NULL); if (pair == NULL) return (0); firstname = nvpair_name(pair); error = spa_open(firstname, &spa, FTAG); if (error != 0) return (error); needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); spa_close(spa, FTAG); if (needsuspend) { suspended = fnvlist_alloc(); for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; pair = nvlist_next_nvpair(snaps, pair)) { char fsname[ZFS_MAX_DATASET_NAME_LEN]; char *snapname = nvpair_name(pair); char *atp; void *cookie; atp = strchr(snapname, '@'); if (atp == NULL) { error = SET_ERROR(EINVAL); break; } (void) strlcpy(fsname, snapname, atp - snapname + 1); error = zil_suspend(fsname, &cookie); if (error != 0) break; fnvlist_add_uint64(suspended, fsname, (uintptr_t)cookie); } } ddsa.ddsa_snaps = snaps; ddsa.ddsa_props = props; ddsa.ddsa_errors = errors; ddsa.ddsa_cr = CRED(); if (error == 0) { error = dsl_sync_task(firstname, dsl_dataset_snapshot_check, dsl_dataset_snapshot_sync, &ddsa, fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL); } if (suspended != NULL) { for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL; pair = nvlist_next_nvpair(suspended, pair)) { zil_resume((void *)(uintptr_t) fnvpair_value_uint64(pair)); } fnvlist_free(suspended); } return (error); } typedef struct dsl_dataset_snapshot_tmp_arg { const char *ddsta_fsname; const char *ddsta_snapname; minor_t ddsta_cleanup_minor; const char *ddsta_htag; } dsl_dataset_snapshot_tmp_arg_t; static int dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx) { dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *ds; int error; error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds); if (error != 0) return (error); /* NULL cred means no limit check for tmp snapshot */ error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname, tx, B_FALSE, 0, NULL); if (error != 0) { dsl_dataset_rele(ds, FTAG); return (error); } if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(ENOTSUP)); } error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag, B_TRUE, tx); if (error != 0) { dsl_dataset_rele(ds, FTAG); return (error); } dsl_dataset_rele(ds, FTAG); return (0); } static void dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx) { dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *ds; VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds)); dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx); dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag, ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx); dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx); dsl_dataset_rele(ds, FTAG); } int dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname, minor_t cleanup_minor, const char *htag) { dsl_dataset_snapshot_tmp_arg_t ddsta; int error; spa_t *spa; boolean_t needsuspend; void *cookie; ddsta.ddsta_fsname = fsname; ddsta.ddsta_snapname = snapname; ddsta.ddsta_cleanup_minor = cleanup_minor; ddsta.ddsta_htag = htag; error = spa_open(fsname, &spa, FTAG); if (error != 0) return (error); needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); spa_close(spa, FTAG); if (needsuspend) { error = zil_suspend(fsname, &cookie); if (error != 0) return (error); } error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check, dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED); if (needsuspend) zil_resume(cookie); return (error); } void dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) { ASSERT(dmu_tx_is_syncing(tx)); ASSERT(ds->ds_objset != NULL); ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0); /* * in case we had to change ds_fsid_guid when we opened it, * sync it out now. */ dmu_buf_will_dirty(ds->ds_dbuf, tx); dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid; if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) { VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1, &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx)); VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1, &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx)); VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1, &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx)); ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0; ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0; ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0; } dmu_objset_sync(ds->ds_objset, zio, tx); for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { if (ds->ds_feature_activation_needed[f]) { if (ds->ds_feature_inuse[f]) continue; dsl_dataset_activate_feature(ds->ds_object, f, tx); ds->ds_feature_inuse[f] = B_TRUE; } } } static int deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) { dsl_deadlist_t *dl = arg; dsl_deadlist_insert(dl, bp, tx); return (0); } void dsl_dataset_sync_done(dsl_dataset_t *ds, dmu_tx_t *tx) { objset_t *os = ds->ds_objset; bplist_iterate(&ds->ds_pending_deadlist, deadlist_enqueue_cb, &ds->ds_deadlist, tx); + if (os->os_synced_dnodes != NULL) { + multilist_destroy(os->os_synced_dnodes); + os->os_synced_dnodes = NULL; + } + ASSERT(!dmu_objset_is_dirty(os, dmu_tx_get_txg(tx))); dmu_buf_rele(ds->ds_dbuf, ds); } static void get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv) { uint64_t count = 0; objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; zap_cursor_t zc; zap_attribute_t za; nvlist_t *propval = fnvlist_alloc(); nvlist_t *val; ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); /* * We use nvlist_alloc() instead of fnvlist_alloc() because the * latter would allocate the list with NV_UNIQUE_NAME flag. * As a result, every time a clone name is appended to the list * it would be (linearly) searched for for a duplicate name. * We already know that all clone names must be unique and we * want avoid the quadratic complexity of double-checking that * because we can have a large number of clones. */ VERIFY0(nvlist_alloc(&val, 0, KM_SLEEP)); /* * There may be missing entries in ds_next_clones_obj * due to a bug in a previous version of the code. * Only trust it if it has the right number of entries. */ if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) { VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, &count)); } if (count != dsl_dataset_phys(ds)->ds_num_children - 1) goto fail; for (zap_cursor_init(&zc, mos, dsl_dataset_phys(ds)->ds_next_clones_obj); zap_cursor_retrieve(&zc, &za) == 0; zap_cursor_advance(&zc)) { dsl_dataset_t *clone; char buf[ZFS_MAX_DATASET_NAME_LEN]; VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool, za.za_first_integer, FTAG, &clone)); dsl_dir_name(clone->ds_dir, buf); fnvlist_add_boolean(val, buf); dsl_dataset_rele(clone, FTAG); } zap_cursor_fini(&zc); fnvlist_add_nvlist(propval, ZPROP_VALUE, val); fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval); fail: nvlist_free(val); nvlist_free(propval); } static void get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv) { dsl_pool_t *dp = ds->ds_dir->dd_pool; if (dsl_dataset_has_resume_receive_state(ds)) { char *str; void *packed; uint8_t *compressed; uint64_t val; nvlist_t *token_nv = fnvlist_alloc(); size_t packed_size, compressed_size; if (zap_lookup(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) { fnvlist_add_uint64(token_nv, "fromguid", val); } if (zap_lookup(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) { fnvlist_add_uint64(token_nv, "object", val); } if (zap_lookup(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) { fnvlist_add_uint64(token_nv, "offset", val); } if (zap_lookup(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) { fnvlist_add_uint64(token_nv, "bytes", val); } if (zap_lookup(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) { fnvlist_add_uint64(token_nv, "toguid", val); } char buf[256]; if (zap_lookup(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) { fnvlist_add_string(token_nv, "toname", buf); } if (zap_contains(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_LARGEBLOCK) == 0) { fnvlist_add_boolean(token_nv, "largeblockok"); } if (zap_contains(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_EMBEDOK) == 0) { fnvlist_add_boolean(token_nv, "embedok"); } if (zap_contains(dp->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_COMPRESSOK) == 0) { fnvlist_add_boolean(token_nv, "compressok"); } packed = fnvlist_pack(token_nv, &packed_size); fnvlist_free(token_nv); compressed = kmem_alloc(packed_size, KM_SLEEP); compressed_size = gzip_compress(packed, compressed, packed_size, packed_size, 6); zio_cksum_t cksum; fletcher_4_native(compressed, compressed_size, NULL, &cksum); str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP); for (int i = 0; i < compressed_size; i++) { (void) sprintf(str + i * 2, "%02x", compressed[i]); } str[compressed_size * 2] = '\0'; char *propval = kmem_asprintf("%u-%llx-%llx-%s", ZFS_SEND_RESUME_TOKEN_VERSION, (longlong_t)cksum.zc_word[0], (longlong_t)packed_size, str); dsl_prop_nvlist_add_string(nv, ZFS_PROP_RECEIVE_RESUME_TOKEN, propval); kmem_free(packed, packed_size); kmem_free(str, compressed_size * 2 + 1); kmem_free(compressed, packed_size); strfree(propval); } } void dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) { dsl_pool_t *dp = ds->ds_dir->dd_pool; uint64_t refd, avail, uobjs, aobjs, ratio; ASSERT(dsl_pool_config_held(dp)); ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 : (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 / dsl_dataset_phys(ds)->ds_compressed_bytes); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED, dsl_dataset_phys(ds)->ds_uncompressed_bytes); if (ds->ds_is_snapshot) { dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, dsl_dataset_phys(ds)->ds_unique_bytes); get_clones_stat(ds, nv); } else { if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) { char buf[ZFS_MAX_DATASET_NAME_LEN]; dsl_dataset_name(ds->ds_prev, buf); dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf); } dsl_dir_stats(ds->ds_dir, nv); } dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, dsl_dataset_phys(ds)->ds_creation_time); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, dsl_dataset_phys(ds)->ds_creation_txg); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA, ds->ds_quota); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION, ds->ds_reserved); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID, dsl_dataset_phys(ds)->ds_guid); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE, dsl_dataset_phys(ds)->ds_unique_bytes); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID, ds->ds_object); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, ds->ds_userrefs); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY, DS_IS_DEFER_DESTROY(ds) ? 1 : 0); if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { uint64_t written, comp, uncomp; dsl_pool_t *dp = ds->ds_dir->dd_pool; dsl_dataset_t *prev; int err = dsl_dataset_hold_obj(dp, dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); if (err == 0) { err = dsl_dataset_space_written(prev, ds, &written, &comp, &uncomp); dsl_dataset_rele(prev, FTAG); if (err == 0) { dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN, written); } } } if (!dsl_dataset_is_snapshot(ds)) { /* * A failed "newfs" (e.g. full) resumable receive leaves * the stats set on this dataset. Check here for the prop. */ get_receive_resume_stats(ds, nv); /* * A failed incremental resumable receive leaves the * stats set on our child named "%recv". Check the child * for the prop. */ /* 6 extra bytes for /%recv */ char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; dsl_dataset_t *recv_ds; dsl_dataset_name(ds, recvname); if (strlcat(recvname, "/", sizeof (recvname)) < sizeof (recvname) && strlcat(recvname, recv_clone_name, sizeof (recvname)) < sizeof (recvname) && dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) { get_receive_resume_stats(recv_ds, nv); dsl_dataset_rele(recv_ds, FTAG); } } } void dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) { dsl_pool_t *dp = ds->ds_dir->dd_pool; ASSERT(dsl_pool_config_held(dp)); stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg; stat->dds_inconsistent = dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT; stat->dds_guid = dsl_dataset_phys(ds)->ds_guid; stat->dds_origin[0] = '\0'; if (ds->ds_is_snapshot) { stat->dds_is_snapshot = B_TRUE; stat->dds_num_clones = dsl_dataset_phys(ds)->ds_num_children - 1; } else { stat->dds_is_snapshot = B_FALSE; stat->dds_num_clones = 0; if (dsl_dir_is_clone(ds->ds_dir)) { dsl_dataset_t *ods; VERIFY0(dsl_dataset_hold_obj(dp, dsl_dir_phys(ds->ds_dir)->dd_origin_obj, FTAG, &ods)); dsl_dataset_name(ods, stat->dds_origin); dsl_dataset_rele(ods, FTAG); } } } uint64_t dsl_dataset_fsid_guid(dsl_dataset_t *ds) { return (ds->ds_fsid_guid); } void dsl_dataset_space(dsl_dataset_t *ds, uint64_t *refdbytesp, uint64_t *availbytesp, uint64_t *usedobjsp, uint64_t *availobjsp) { *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes; *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) *availbytesp += ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes; if (ds->ds_quota != 0) { /* * Adjust available bytes according to refquota */ if (*refdbytesp < ds->ds_quota) *availbytesp = MIN(*availbytesp, ds->ds_quota - *refdbytesp); else *availbytesp = 0; } rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp); rrw_exit(&ds->ds_bp_rwlock, FTAG); *availobjsp = DN_MAX_OBJECT - *usedobjsp; } boolean_t dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap) { dsl_pool_t *dp = ds->ds_dir->dd_pool; uint64_t birth; ASSERT(dsl_pool_config_held(dp)); if (snap == NULL) return (B_FALSE); rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); birth = dsl_dataset_get_blkptr(ds)->blk_birth; rrw_exit(&ds->ds_bp_rwlock, FTAG); if (birth > dsl_dataset_phys(snap)->ds_creation_txg) { objset_t *os, *os_snap; /* * It may be that only the ZIL differs, because it was * reset in the head. Don't count that as being * modified. */ if (dmu_objset_from_ds(ds, &os) != 0) return (B_TRUE); if (dmu_objset_from_ds(snap, &os_snap) != 0) return (B_TRUE); return (bcmp(&os->os_phys->os_meta_dnode, &os_snap->os_phys->os_meta_dnode, sizeof (os->os_phys->os_meta_dnode)) != 0); } return (B_FALSE); } typedef struct dsl_dataset_rename_snapshot_arg { const char *ddrsa_fsname; const char *ddrsa_oldsnapname; const char *ddrsa_newsnapname; boolean_t ddrsa_recursive; dmu_tx_t *ddrsa_tx; } dsl_dataset_rename_snapshot_arg_t; /* ARGSUSED */ static int dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) { dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; int error; uint64_t val; error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); if (error != 0) { /* ignore nonexistent snapshots */ return (error == ENOENT ? 0 : error); } /* new name should not exist */ error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val); if (error == 0) error = SET_ERROR(EEXIST); else if (error == ENOENT) error = 0; /* dataset name + 1 for the "@" + the new snapshot name must fit */ if (dsl_dir_namelen(hds->ds_dir) + 1 + strlen(ddrsa->ddrsa_newsnapname) >= ZFS_MAX_DATASET_NAME_LEN) error = SET_ERROR(ENAMETOOLONG); return (error); } static int dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx) { dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *hds; int error; error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds); if (error != 0) return (error); if (ddrsa->ddrsa_recursive) { error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object, dsl_dataset_rename_snapshot_check_impl, ddrsa, DS_FIND_CHILDREN); } else { error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa); } dsl_dataset_rele(hds, FTAG); return (error); } static int dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) { dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; dsl_dataset_t *ds; uint64_t val; dmu_tx_t *tx = ddrsa->ddrsa_tx; int error; error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); ASSERT(error == 0 || error == ENOENT); if (error == ENOENT) { /* ignore nonexistent snapshots */ return (0); } VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds)); /* log before we change the name */ spa_history_log_internal_ds(ds, "rename", tx, "-> @%s", ddrsa->ddrsa_newsnapname); VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx, B_FALSE)); mutex_enter(&ds->ds_lock); (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname); mutex_exit(&ds->ds_lock); VERIFY0(zap_add(dp->dp_meta_objset, dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname, 8, 1, &ds->ds_object, tx)); dsl_dataset_rele(ds, FTAG); return (0); } static void dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx) { dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *hds; VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds)); ddrsa->ddrsa_tx = tx; if (ddrsa->ddrsa_recursive) { VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object, dsl_dataset_rename_snapshot_sync_impl, ddrsa, DS_FIND_CHILDREN)); } else { VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa)); } dsl_dataset_rele(hds, FTAG); } int dsl_dataset_rename_snapshot(const char *fsname, const char *oldsnapname, const char *newsnapname, boolean_t recursive) { dsl_dataset_rename_snapshot_arg_t ddrsa; ddrsa.ddrsa_fsname = fsname; ddrsa.ddrsa_oldsnapname = oldsnapname; ddrsa.ddrsa_newsnapname = newsnapname; ddrsa.ddrsa_recursive = recursive; return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check, dsl_dataset_rename_snapshot_sync, &ddrsa, 1, ZFS_SPACE_CHECK_RESERVED)); } /* * If we're doing an ownership handoff, we need to make sure that there is * only one long hold on the dataset. We're not allowed to change anything here * so we don't permanently release the long hold or regular hold here. We want * to do this only when syncing to avoid the dataset unexpectedly going away * when we release the long hold. */ static int dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx) { boolean_t held; if (!dmu_tx_is_syncing(tx)) return (0); if (owner != NULL) { VERIFY3P(ds->ds_owner, ==, owner); dsl_dataset_long_rele(ds, owner); } held = dsl_dataset_long_held(ds); if (owner != NULL) dsl_dataset_long_hold(ds, owner); if (held) return (SET_ERROR(EBUSY)); return (0); } typedef struct dsl_dataset_rollback_arg { const char *ddra_fsname; void *ddra_owner; nvlist_t *ddra_result; } dsl_dataset_rollback_arg_t; static int dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx) { dsl_dataset_rollback_arg_t *ddra = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *ds; int64_t unused_refres_delta; int error; error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds); if (error != 0) return (error); /* must not be a snapshot */ if (ds->ds_is_snapshot) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(EINVAL)); } /* must have a most recent snapshot */ if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(EINVAL)); } /* * No rollback to a snapshot created in the current txg, because * the rollback may dirty the dataset and create blocks that are * not reachable from the rootbp while having a birth txg that * falls into the snapshot's range. */ if (dmu_tx_is_syncing(tx) && dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(EAGAIN)); } /* must not have any bookmarks after the most recent snapshot */ nvlist_t *proprequest = fnvlist_alloc(); fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG)); nvlist_t *bookmarks = fnvlist_alloc(); error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks); fnvlist_free(proprequest); if (error != 0) return (error); for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL); pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) { nvlist_t *valuenv = fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair), zfs_prop_to_name(ZFS_PROP_CREATETXG)); uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value"); if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) { fnvlist_free(bookmarks); dsl_dataset_rele(ds, FTAG); return (SET_ERROR(EEXIST)); } } fnvlist_free(bookmarks); error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx); if (error != 0) { dsl_dataset_rele(ds, FTAG); return (error); } /* * Check if the snap we are rolling back to uses more than * the refquota. */ if (ds->ds_quota != 0 && dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(EDQUOT)); } /* * When we do the clone swap, we will temporarily use more space * due to the refreservation (the head will no longer have any * unique space, so the entire amount of the refreservation will need * to be free). We will immediately destroy the clone, freeing * this space, but the freeing happens over many txg's. */ unused_refres_delta = (int64_t)MIN(ds->ds_reserved, dsl_dataset_phys(ds)->ds_unique_bytes); if (unused_refres_delta > 0 && unused_refres_delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(ENOSPC)); } dsl_dataset_rele(ds, FTAG); return (0); } static void dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx) { dsl_dataset_rollback_arg_t *ddra = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *ds, *clone; uint64_t cloneobj; char namebuf[ZFS_MAX_DATASET_NAME_LEN]; VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds)); dsl_dataset_name(ds->ds_prev, namebuf); fnvlist_add_string(ddra->ddra_result, "target", namebuf); cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback", ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx); VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone)); dsl_dataset_clone_swap_sync_impl(clone, ds, tx); dsl_dataset_zero_zil(ds, tx); dsl_destroy_head_sync_impl(clone, tx); dsl_dataset_rele(clone, FTAG); dsl_dataset_rele(ds, FTAG); } /* * Rolls back the given filesystem or volume to the most recent snapshot. * The name of the most recent snapshot will be returned under key "target" * in the result nvlist. * * If owner != NULL: * - The existing dataset MUST be owned by the specified owner at entry * - Upon return, dataset will still be held by the same owner, whether we * succeed or not. * * This mode is required any time the existing filesystem is mounted. See * notes above zfs_suspend_fs() for further details. */ int dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result) { dsl_dataset_rollback_arg_t ddra; ddra.ddra_fsname = fsname; ddra.ddra_owner = owner; ddra.ddra_result = result; return (dsl_sync_task(fsname, dsl_dataset_rollback_check, dsl_dataset_rollback_sync, &ddra, 1, ZFS_SPACE_CHECK_RESERVED)); } struct promotenode { list_node_t link; dsl_dataset_t *ds; }; typedef struct dsl_dataset_promote_arg { const char *ddpa_clonename; dsl_dataset_t *ddpa_clone; list_t shared_snaps, origin_snaps, clone_snaps; dsl_dataset_t *origin_origin; /* origin of the origin */ uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap; char *err_ds; cred_t *cr; } dsl_dataset_promote_arg_t; static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep); static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag); static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag); static int dsl_dataset_promote_check(void *arg, dmu_tx_t *tx) { dsl_dataset_promote_arg_t *ddpa = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *hds; struct promotenode *snap; dsl_dataset_t *origin_ds; int err; uint64_t unused; uint64_t ss_mv_cnt; size_t max_snap_len; err = promote_hold(ddpa, dp, FTAG); if (err != 0) return (err); hds = ddpa->ddpa_clone; max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1; if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) { promote_rele(ddpa, FTAG); return (SET_ERROR(EXDEV)); } /* * Compute and check the amount of space to transfer. Since this is * so expensive, don't do the preliminary check. */ if (!dmu_tx_is_syncing(tx)) { promote_rele(ddpa, FTAG); return (0); } snap = list_head(&ddpa->shared_snaps); origin_ds = snap->ds; /* compute origin's new unique space */ snap = list_tail(&ddpa->clone_snaps); ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, origin_ds->ds_object); dsl_deadlist_space_range(&snap->ds->ds_deadlist, dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX, &ddpa->unique, &unused, &unused); /* * Walk the snapshots that we are moving * * Compute space to transfer. Consider the incremental changes * to used by each snapshot: * (my used) = (prev's used) + (blocks born) - (blocks killed) * So each snapshot gave birth to: * (blocks born) = (my used) - (prev's used) + (blocks killed) * So a sequence would look like: * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0) * Which simplifies to: * uN + kN + kN-1 + ... + k1 + k0 * Note however, if we stop before we reach the ORIGIN we get: * uN + kN + kN-1 + ... + kM - uM-1 */ ss_mv_cnt = 0; ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes; ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes; ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes; for (snap = list_head(&ddpa->shared_snaps); snap; snap = list_next(&ddpa->shared_snaps, snap)) { uint64_t val, dlused, dlcomp, dluncomp; dsl_dataset_t *ds = snap->ds; ss_mv_cnt++; /* * If there are long holds, we won't be able to evict * the objset. */ if (dsl_dataset_long_held(ds)) { err = SET_ERROR(EBUSY); goto out; } /* Check that the snapshot name does not conflict */ VERIFY0(dsl_dataset_get_snapname(ds)); if (strlen(ds->ds_snapname) >= max_snap_len) { err = SET_ERROR(ENAMETOOLONG); goto out; } err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val); if (err == 0) { (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname); err = SET_ERROR(EEXIST); goto out; } if (err != ENOENT) goto out; /* The very first snapshot does not have a deadlist */ if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0) continue; dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp); ddpa->used += dlused; ddpa->comp += dlcomp; ddpa->uncomp += dluncomp; } /* * If we are a clone of a clone then we never reached ORIGIN, * so we need to subtract out the clone origin's used space. */ if (ddpa->origin_origin) { ddpa->used -= dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes; ddpa->comp -= dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes; ddpa->uncomp -= dsl_dataset_phys(ddpa->origin_origin)-> ds_uncompressed_bytes; } /* Check that there is enough space and limit headroom here */ err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir, 0, ss_mv_cnt, ddpa->used, ddpa->cr); if (err != 0) goto out; /* * Compute the amounts of space that will be used by snapshots * after the promotion (for both origin and clone). For each, * it is the amount of space that will be on all of their * deadlists (that was not born before their new origin). */ if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) { uint64_t space; /* * Note, typically this will not be a clone of a clone, * so dd_origin_txg will be < TXG_INITIAL, so * these snaplist_space() -> dsl_deadlist_space_range() * calls will be fast because they do not have to * iterate over all bps. */ snap = list_head(&ddpa->origin_snaps); err = snaplist_space(&ddpa->shared_snaps, snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap); if (err != 0) goto out; err = snaplist_space(&ddpa->clone_snaps, snap->ds->ds_dir->dd_origin_txg, &space); if (err != 0) goto out; ddpa->cloneusedsnap += space; } if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) { err = snaplist_space(&ddpa->origin_snaps, dsl_dataset_phys(origin_ds)->ds_creation_txg, &ddpa->originusedsnap); if (err != 0) goto out; } out: promote_rele(ddpa, FTAG); return (err); } static void dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx) { dsl_dataset_promote_arg_t *ddpa = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *hds; struct promotenode *snap; dsl_dataset_t *origin_ds; dsl_dataset_t *origin_head; dsl_dir_t *dd; dsl_dir_t *odd = NULL; uint64_t oldnext_obj; int64_t delta; VERIFY0(promote_hold(ddpa, dp, FTAG)); hds = ddpa->ddpa_clone; ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE); snap = list_head(&ddpa->shared_snaps); origin_ds = snap->ds; dd = hds->ds_dir; snap = list_head(&ddpa->origin_snaps); origin_head = snap->ds; /* * We need to explicitly open odd, since origin_ds's dd will be * changing. */ VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object, NULL, FTAG, &odd)); /* change origin's next snap */ dmu_buf_will_dirty(origin_ds->ds_dbuf, tx); oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj; snap = list_tail(&ddpa->clone_snaps); ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, origin_ds->ds_object); dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object; /* change the origin's next clone */ if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) { dsl_dataset_remove_from_next_clones(origin_ds, snap->ds->ds_object, tx); VERIFY0(zap_add_int(dp->dp_meta_objset, dsl_dataset_phys(origin_ds)->ds_next_clones_obj, oldnext_obj, tx)); } /* change origin */ dmu_buf_will_dirty(dd->dd_dbuf, tx); ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object); dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj; dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg; dmu_buf_will_dirty(odd->dd_dbuf, tx); dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object; origin_head->ds_dir->dd_origin_txg = dsl_dataset_phys(origin_ds)->ds_creation_txg; /* change dd_clone entries */ if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { VERIFY0(zap_remove_int(dp->dp_meta_objset, dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx)); VERIFY0(zap_add_int(dp->dp_meta_objset, dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, hds->ds_object, tx)); VERIFY0(zap_remove_int(dp->dp_meta_objset, dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, origin_head->ds_object, tx)); if (dsl_dir_phys(dd)->dd_clones == 0) { dsl_dir_phys(dd)->dd_clones = zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); } VERIFY0(zap_add_int(dp->dp_meta_objset, dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx)); } /* move snapshots to this dir */ for (snap = list_head(&ddpa->shared_snaps); snap; snap = list_next(&ddpa->shared_snaps, snap)) { dsl_dataset_t *ds = snap->ds; /* * Property callbacks are registered to a particular * dsl_dir. Since ours is changing, evict the objset * so that they will be unregistered from the old dsl_dir. */ if (ds->ds_objset) { dmu_objset_evict(ds->ds_objset); ds->ds_objset = NULL; } /* move snap name entry */ VERIFY0(dsl_dataset_get_snapname(ds)); VERIFY0(dsl_dataset_snap_remove(origin_head, ds->ds_snapname, tx, B_TRUE)); VERIFY0(zap_add(dp->dp_meta_objset, dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname, 8, 1, &ds->ds_object, tx)); dsl_fs_ss_count_adjust(hds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx); /* change containing dsl_dir */ dmu_buf_will_dirty(ds->ds_dbuf, tx); ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object); dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object; ASSERT3P(ds->ds_dir, ==, odd); dsl_dir_rele(ds->ds_dir, ds); VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object, NULL, ds, &ds->ds_dir)); /* move any clone references */ if (dsl_dataset_phys(ds)->ds_next_clones_obj && spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { zap_cursor_t zc; zap_attribute_t za; for (zap_cursor_init(&zc, dp->dp_meta_objset, dsl_dataset_phys(ds)->ds_next_clones_obj); zap_cursor_retrieve(&zc, &za) == 0; zap_cursor_advance(&zc)) { dsl_dataset_t *cnds; uint64_t o; if (za.za_first_integer == oldnext_obj) { /* * We've already moved the * origin's reference. */ continue; } VERIFY0(dsl_dataset_hold_obj(dp, za.za_first_integer, FTAG, &cnds)); o = dsl_dir_phys(cnds->ds_dir)-> dd_head_dataset_obj; VERIFY0(zap_remove_int(dp->dp_meta_objset, dsl_dir_phys(odd)->dd_clones, o, tx)); VERIFY0(zap_add_int(dp->dp_meta_objset, dsl_dir_phys(dd)->dd_clones, o, tx)); dsl_dataset_rele(cnds, FTAG); } zap_cursor_fini(&zc); } ASSERT(!dsl_prop_hascb(ds)); } /* * Change space accounting. * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either * both be valid, or both be 0 (resulting in delta == 0). This * is true for each of {clone,origin} independently. */ delta = ddpa->cloneusedsnap - dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]; ASSERT3S(delta, >=, 0); ASSERT3U(ddpa->used, >=, delta); dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx); dsl_dir_diduse_space(dd, DD_USED_HEAD, ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx); delta = ddpa->originusedsnap - dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP]; ASSERT3S(delta, <=, 0); ASSERT3U(ddpa->used, >=, -delta); dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx); dsl_dir_diduse_space(odd, DD_USED_HEAD, -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx); dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique; /* log history record */ spa_history_log_internal_ds(hds, "promote", tx, ""); dsl_dir_rele(odd, FTAG); promote_rele(ddpa, FTAG); } /* * Make a list of dsl_dataset_t's for the snapshots between first_obj * (exclusive) and last_obj (inclusive). The list will be in reverse * order (last_obj will be the list_head()). If first_obj == 0, do all * snapshots back to this dataset's origin. */ static int snaplist_make(dsl_pool_t *dp, uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag) { uint64_t obj = last_obj; list_create(l, sizeof (struct promotenode), offsetof(struct promotenode, link)); while (obj != first_obj) { dsl_dataset_t *ds; struct promotenode *snap; int err; err = dsl_dataset_hold_obj(dp, obj, tag, &ds); ASSERT(err != ENOENT); if (err != 0) return (err); if (first_obj == 0) first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj; snap = kmem_alloc(sizeof (*snap), KM_SLEEP); snap->ds = ds; list_insert_tail(l, snap); obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; } return (0); } static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep) { struct promotenode *snap; *spacep = 0; for (snap = list_head(l); snap; snap = list_next(l, snap)) { uint64_t used, comp, uncomp; dsl_deadlist_space_range(&snap->ds->ds_deadlist, mintxg, UINT64_MAX, &used, &comp, &uncomp); *spacep += used; } return (0); } static void snaplist_destroy(list_t *l, void *tag) { struct promotenode *snap; if (l == NULL || !list_link_active(&l->list_head)) return; while ((snap = list_tail(l)) != NULL) { list_remove(l, snap); dsl_dataset_rele(snap->ds, tag); kmem_free(snap, sizeof (*snap)); } list_destroy(l); } static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag) { int error; dsl_dir_t *dd; struct promotenode *snap; error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag, &ddpa->ddpa_clone); if (error != 0) return (error); dd = ddpa->ddpa_clone->ds_dir; if (ddpa->ddpa_clone->ds_is_snapshot || !dsl_dir_is_clone(dd)) { dsl_dataset_rele(ddpa->ddpa_clone, tag); return (SET_ERROR(EINVAL)); } error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj, &ddpa->shared_snaps, tag); if (error != 0) goto out; error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object, &ddpa->clone_snaps, tag); if (error != 0) goto out; snap = list_head(&ddpa->shared_snaps); ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj); error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj, dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj, &ddpa->origin_snaps, tag); if (error != 0) goto out; if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) { error = dsl_dataset_hold_obj(dp, dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj, tag, &ddpa->origin_origin); if (error != 0) goto out; } out: if (error != 0) promote_rele(ddpa, tag); return (error); } static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag) { snaplist_destroy(&ddpa->shared_snaps, tag); snaplist_destroy(&ddpa->clone_snaps, tag); snaplist_destroy(&ddpa->origin_snaps, tag); if (ddpa->origin_origin != NULL) dsl_dataset_rele(ddpa->origin_origin, tag); dsl_dataset_rele(ddpa->ddpa_clone, tag); } /* * Promote a clone. * * If it fails due to a conflicting snapshot name, "conflsnap" will be filled * in with the name. (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.) */ int dsl_dataset_promote(const char *name, char *conflsnap) { dsl_dataset_promote_arg_t ddpa = { 0 }; uint64_t numsnaps; int error; objset_t *os; /* * We will modify space proportional to the number of * snapshots. Compute numsnaps. */ error = dmu_objset_hold(name, FTAG, &os); if (error != 0) return (error); error = zap_count(dmu_objset_pool(os)->dp_meta_objset, dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj, &numsnaps); dmu_objset_rele(os, FTAG); if (error != 0) return (error); ddpa.ddpa_clonename = name; ddpa.err_ds = conflsnap; ddpa.cr = CRED(); return (dsl_sync_task(name, dsl_dataset_promote_check, dsl_dataset_promote_sync, &ddpa, 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED)); } int dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone, dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx) { /* * "slack" factor for received datasets with refquota set on them. * See the bottom of this function for details on its use. */ uint64_t refquota_slack = DMU_MAX_ACCESS * spa_asize_inflation; int64_t unused_refres_delta; /* they should both be heads */ if (clone->ds_is_snapshot || origin_head->ds_is_snapshot) return (SET_ERROR(EINVAL)); /* if we are not forcing, the branch point should be just before them */ if (!force && clone->ds_prev != origin_head->ds_prev) return (SET_ERROR(EINVAL)); /* clone should be the clone (unless they are unrelated) */ if (clone->ds_prev != NULL && clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap && origin_head->ds_dir != clone->ds_prev->ds_dir) return (SET_ERROR(EINVAL)); /* the clone should be a child of the origin */ if (clone->ds_dir->dd_parent != origin_head->ds_dir) return (SET_ERROR(EINVAL)); /* origin_head shouldn't be modified unless 'force' */ if (!force && dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev)) return (SET_ERROR(ETXTBSY)); /* origin_head should have no long holds (e.g. is not mounted) */ if (dsl_dataset_handoff_check(origin_head, owner, tx)) return (SET_ERROR(EBUSY)); /* check amount of any unconsumed refreservation */ unused_refres_delta = (int64_t)MIN(origin_head->ds_reserved, dsl_dataset_phys(origin_head)->ds_unique_bytes) - (int64_t)MIN(origin_head->ds_reserved, dsl_dataset_phys(clone)->ds_unique_bytes); if (unused_refres_delta > 0 && unused_refres_delta > dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE)) return (SET_ERROR(ENOSPC)); /* * The clone can't be too much over the head's refquota. * * To ensure that the entire refquota can be used, we allow one * transaction to exceed the the refquota. Therefore, this check * needs to also allow for the space referenced to be more than the * refquota. The maximum amount of space that one transaction can use * on disk is DMU_MAX_ACCESS * spa_asize_inflation. Allowing this * overage ensures that we are able to receive a filesystem that * exceeds the refquota on the source system. * * So that overage is the refquota_slack we use below. */ if (origin_head->ds_quota != 0 && dsl_dataset_phys(clone)->ds_referenced_bytes > origin_head->ds_quota + refquota_slack) return (SET_ERROR(EDQUOT)); return (0); } void dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone, dsl_dataset_t *origin_head, dmu_tx_t *tx) { dsl_pool_t *dp = dmu_tx_pool(tx); int64_t unused_refres_delta; ASSERT(clone->ds_reserved == 0); /* * NOTE: On DEBUG kernels there could be a race between this and * the check function if spa_asize_inflation is adjusted... */ ASSERT(origin_head->ds_quota == 0 || dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota + DMU_MAX_ACCESS * spa_asize_inflation); ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev); /* * Swap per-dataset feature flags. */ for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { if (!(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET)) { ASSERT(!clone->ds_feature_inuse[f]); ASSERT(!origin_head->ds_feature_inuse[f]); continue; } boolean_t clone_inuse = clone->ds_feature_inuse[f]; boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f]; if (clone_inuse) { dsl_dataset_deactivate_feature(clone->ds_object, f, tx); clone->ds_feature_inuse[f] = B_FALSE; } if (origin_head_inuse) { dsl_dataset_deactivate_feature(origin_head->ds_object, f, tx); origin_head->ds_feature_inuse[f] = B_FALSE; } if (clone_inuse) { dsl_dataset_activate_feature(origin_head->ds_object, f, tx); origin_head->ds_feature_inuse[f] = B_TRUE; } if (origin_head_inuse) { dsl_dataset_activate_feature(clone->ds_object, f, tx); clone->ds_feature_inuse[f] = B_TRUE; } } dmu_buf_will_dirty(clone->ds_dbuf, tx); dmu_buf_will_dirty(origin_head->ds_dbuf, tx); if (clone->ds_objset != NULL) { dmu_objset_evict(clone->ds_objset); clone->ds_objset = NULL; } if (origin_head->ds_objset != NULL) { dmu_objset_evict(origin_head->ds_objset); origin_head->ds_objset = NULL; } unused_refres_delta = (int64_t)MIN(origin_head->ds_reserved, dsl_dataset_phys(origin_head)->ds_unique_bytes) - (int64_t)MIN(origin_head->ds_reserved, dsl_dataset_phys(clone)->ds_unique_bytes); /* * Reset origin's unique bytes, if it exists. */ if (clone->ds_prev) { dsl_dataset_t *origin = clone->ds_prev; uint64_t comp, uncomp; dmu_buf_will_dirty(origin->ds_dbuf, tx); dsl_deadlist_space_range(&clone->ds_deadlist, dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX, &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp); } /* swap blkptrs */ { rrw_enter(&clone->ds_bp_rwlock, RW_WRITER, FTAG); rrw_enter(&origin_head->ds_bp_rwlock, RW_WRITER, FTAG); blkptr_t tmp; tmp = dsl_dataset_phys(origin_head)->ds_bp; dsl_dataset_phys(origin_head)->ds_bp = dsl_dataset_phys(clone)->ds_bp; dsl_dataset_phys(clone)->ds_bp = tmp; rrw_exit(&origin_head->ds_bp_rwlock, FTAG); rrw_exit(&clone->ds_bp_rwlock, FTAG); } /* set dd_*_bytes */ { int64_t dused, dcomp, duncomp; uint64_t cdl_used, cdl_comp, cdl_uncomp; uint64_t odl_used, odl_comp, odl_uncomp; ASSERT3U(dsl_dir_phys(clone->ds_dir)-> dd_used_breakdown[DD_USED_SNAP], ==, 0); dsl_deadlist_space(&clone->ds_deadlist, &cdl_used, &cdl_comp, &cdl_uncomp); dsl_deadlist_space(&origin_head->ds_deadlist, &odl_used, &odl_comp, &odl_uncomp); dused = dsl_dataset_phys(clone)->ds_referenced_bytes + cdl_used - (dsl_dataset_phys(origin_head)->ds_referenced_bytes + odl_used); dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes + cdl_comp - (dsl_dataset_phys(origin_head)->ds_compressed_bytes + odl_comp); duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes + cdl_uncomp - (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes + odl_uncomp); dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD, dused, dcomp, duncomp, tx); dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD, -dused, -dcomp, -duncomp, tx); /* * The difference in the space used by snapshots is the * difference in snapshot space due to the head's * deadlist (since that's the only thing that's * changing that affects the snapused). */ dsl_deadlist_space_range(&clone->ds_deadlist, origin_head->ds_dir->dd_origin_txg, UINT64_MAX, &cdl_used, &cdl_comp, &cdl_uncomp); dsl_deadlist_space_range(&origin_head->ds_deadlist, origin_head->ds_dir->dd_origin_txg, UINT64_MAX, &odl_used, &odl_comp, &odl_uncomp); dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used, DD_USED_HEAD, DD_USED_SNAP, tx); } /* swap ds_*_bytes */ SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes, dsl_dataset_phys(clone)->ds_referenced_bytes); SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes, dsl_dataset_phys(clone)->ds_compressed_bytes); SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes, dsl_dataset_phys(clone)->ds_uncompressed_bytes); SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes, dsl_dataset_phys(clone)->ds_unique_bytes); /* apply any parent delta for change in unconsumed refreservation */ dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV, unused_refres_delta, 0, 0, tx); /* * Swap deadlists. */ dsl_deadlist_close(&clone->ds_deadlist); dsl_deadlist_close(&origin_head->ds_deadlist); SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj, dsl_dataset_phys(clone)->ds_deadlist_obj); dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset, dsl_dataset_phys(clone)->ds_deadlist_obj); dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset, dsl_dataset_phys(origin_head)->ds_deadlist_obj); dsl_scan_ds_clone_swapped(origin_head, clone, tx); spa_history_log_internal_ds(clone, "clone swap", tx, "parent=%s", origin_head->ds_dir->dd_myname); } /* * Given a pool name and a dataset object number in that pool, * return the name of that dataset. */ int dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) { dsl_pool_t *dp; dsl_dataset_t *ds; int error; error = dsl_pool_hold(pname, FTAG, &dp); if (error != 0) return (error); error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds); if (error == 0) { dsl_dataset_name(ds, buf); dsl_dataset_rele(ds, FTAG); } dsl_pool_rele(dp, FTAG); return (error); } int dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv) { int error = 0; ASSERT3S(asize, >, 0); /* * *ref_rsrv is the portion of asize that will come from any * unconsumed refreservation space. */ *ref_rsrv = 0; mutex_enter(&ds->ds_lock); /* * Make a space adjustment for reserved bytes. */ if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) { ASSERT3U(*used, >=, ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); *used -= (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); *ref_rsrv = asize - MIN(asize, parent_delta(ds, asize + inflight)); } if (!check_quota || ds->ds_quota == 0) { mutex_exit(&ds->ds_lock); return (0); } /* * If they are requesting more space, and our current estimate * is over quota, they get to try again unless the actual * on-disk is over quota and there are no pending changes (which * may free up space for us). */ if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >= ds->ds_quota) { if (inflight > 0 || dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota) error = SET_ERROR(ERESTART); else error = SET_ERROR(EDQUOT); } mutex_exit(&ds->ds_lock); return (error); } typedef struct dsl_dataset_set_qr_arg { const char *ddsqra_name; zprop_source_t ddsqra_source; uint64_t ddsqra_value; } dsl_dataset_set_qr_arg_t; /* ARGSUSED */ static int dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx) { dsl_dataset_set_qr_arg_t *ddsqra = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *ds; int error; uint64_t newval; if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA) return (SET_ERROR(ENOTSUP)); error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); if (error != 0) return (error); if (ds->ds_is_snapshot) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(EINVAL)); } error = dsl_prop_predict(ds->ds_dir, zfs_prop_to_name(ZFS_PROP_REFQUOTA), ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); if (error != 0) { dsl_dataset_rele(ds, FTAG); return (error); } if (newval == 0) { dsl_dataset_rele(ds, FTAG); return (0); } if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes || newval < ds->ds_reserved) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(ENOSPC)); } dsl_dataset_rele(ds, FTAG); return (0); } static void dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx) { dsl_dataset_set_qr_arg_t *ddsqra = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *ds; uint64_t newval; VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFQUOTA), ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, &ddsqra->ddsqra_value, tx); VERIFY0(dsl_prop_get_int_ds(ds, zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval)); if (ds->ds_quota != newval) { dmu_buf_will_dirty(ds->ds_dbuf, tx); ds->ds_quota = newval; } dsl_dataset_rele(ds, FTAG); } int dsl_dataset_set_refquota(const char *dsname, zprop_source_t source, uint64_t refquota) { dsl_dataset_set_qr_arg_t ddsqra; ddsqra.ddsqra_name = dsname; ddsqra.ddsqra_source = source; ddsqra.ddsqra_value = refquota; return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check, dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE)); } static int dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx) { dsl_dataset_set_qr_arg_t *ddsqra = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *ds; int error; uint64_t newval, unique; if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION) return (SET_ERROR(ENOTSUP)); error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); if (error != 0) return (error); if (ds->ds_is_snapshot) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(EINVAL)); } error = dsl_prop_predict(ds->ds_dir, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); if (error != 0) { dsl_dataset_rele(ds, FTAG); return (error); } /* * If we are doing the preliminary check in open context, the * space estimates may be inaccurate. */ if (!dmu_tx_is_syncing(tx)) { dsl_dataset_rele(ds, FTAG); return (0); } mutex_enter(&ds->ds_lock); if (!DS_UNIQUE_IS_ACCURATE(ds)) dsl_dataset_recalc_head_uniq(ds); unique = dsl_dataset_phys(ds)->ds_unique_bytes; mutex_exit(&ds->ds_lock); if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) { uint64_t delta = MAX(unique, newval) - MAX(unique, ds->ds_reserved); if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) || (ds->ds_quota > 0 && newval > ds->ds_quota)) { dsl_dataset_rele(ds, FTAG); return (SET_ERROR(ENOSPC)); } } dsl_dataset_rele(ds, FTAG); return (0); } void dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds, zprop_source_t source, uint64_t value, dmu_tx_t *tx) { uint64_t newval; uint64_t unique; int64_t delta; dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), source, sizeof (value), 1, &value, tx); VERIFY0(dsl_prop_get_int_ds(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval)); dmu_buf_will_dirty(ds->ds_dbuf, tx); mutex_enter(&ds->ds_dir->dd_lock); mutex_enter(&ds->ds_lock); ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); unique = dsl_dataset_phys(ds)->ds_unique_bytes; delta = MAX(0, (int64_t)(newval - unique)) - MAX(0, (int64_t)(ds->ds_reserved - unique)); ds->ds_reserved = newval; mutex_exit(&ds->ds_lock); dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); mutex_exit(&ds->ds_dir->dd_lock); } static void dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx) { dsl_dataset_set_qr_arg_t *ddsqra = arg; dsl_pool_t *dp = dmu_tx_pool(tx); dsl_dataset_t *ds; VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); dsl_dataset_set_refreservation_sync_impl(ds, ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx); dsl_dataset_rele(ds, FTAG); } int dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source, uint64_t refreservation) { dsl_dataset_set_qr_arg_t ddsqra; ddsqra.ddsqra_name = dsname; ddsqra.ddsqra_source = source; ddsqra.ddsqra_value = refreservation; return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check, dsl_dataset_set_refreservation_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE)); } /* * Return (in *usedp) the amount of space written in new that is not * present in oldsnap. New may be a snapshot or the head. Old must be * a snapshot before new, in new's filesystem (or its origin). If not then * fail and return EINVAL. * * The written space is calculated by considering two components: First, we * ignore any freed space, and calculate the written as new's used space * minus old's used space. Next, we add in the amount of space that was freed * between the two snapshots, thus reducing new's used space relative to old's. * Specifically, this is the space that was born before old->ds_creation_txg, * and freed before new (ie. on new's deadlist or a previous deadlist). * * space freed [---------------------] * snapshots ---O-------O--------O-------O------ * oldsnap new */ int dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) { int err = 0; uint64_t snapobj; dsl_pool_t *dp = new->ds_dir->dd_pool; ASSERT(dsl_pool_config_held(dp)); *usedp = 0; *usedp += dsl_dataset_phys(new)->ds_referenced_bytes; *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes; *compp = 0; *compp += dsl_dataset_phys(new)->ds_compressed_bytes; *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes; *uncompp = 0; *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes; *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes; snapobj = new->ds_object; while (snapobj != oldsnap->ds_object) { dsl_dataset_t *snap; uint64_t used, comp, uncomp; if (snapobj == new->ds_object) { snap = new; } else { err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap); if (err != 0) break; } if (dsl_dataset_phys(snap)->ds_prev_snap_txg == dsl_dataset_phys(oldsnap)->ds_creation_txg) { /* * The blocks in the deadlist can not be born after * ds_prev_snap_txg, so get the whole deadlist space, * which is more efficient (especially for old-format * deadlists). Unfortunately the deadlist code * doesn't have enough information to make this * optimization itself. */ dsl_deadlist_space(&snap->ds_deadlist, &used, &comp, &uncomp); } else { dsl_deadlist_space_range(&snap->ds_deadlist, 0, dsl_dataset_phys(oldsnap)->ds_creation_txg, &used, &comp, &uncomp); } *usedp += used; *compp += comp; *uncompp += uncomp; /* * If we get to the beginning of the chain of snapshots * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap * was not a snapshot of/before new. */ snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj; if (snap != new) dsl_dataset_rele(snap, FTAG); if (snapobj == 0) { err = SET_ERROR(EINVAL); break; } } return (err); } /* * Return (in *usedp) the amount of space that will be reclaimed if firstsnap, * lastsnap, and all snapshots in between are deleted. * * blocks that would be freed [---------------------------] * snapshots ---O-------O--------O-------O--------O * firstsnap lastsnap * * This is the set of blocks that were born after the snap before firstsnap, * (birth > firstsnap->prev_snap_txg) and died before the snap after the * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist). * We calculate this by iterating over the relevant deadlists (from the snap * after lastsnap, backward to the snap after firstsnap), summing up the * space on the deadlist that was born after the snap before firstsnap. */ int dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, dsl_dataset_t *lastsnap, uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) { int err = 0; uint64_t snapobj; dsl_pool_t *dp = firstsnap->ds_dir->dd_pool; ASSERT(firstsnap->ds_is_snapshot); ASSERT(lastsnap->ds_is_snapshot); /* * Check that the snapshots are in the same dsl_dir, and firstsnap * is before lastsnap. */ if (firstsnap->ds_dir != lastsnap->ds_dir || dsl_dataset_phys(firstsnap)->ds_creation_txg > dsl_dataset_phys(lastsnap)->ds_creation_txg) return (SET_ERROR(EINVAL)); *usedp = *compp = *uncompp = 0; snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj; while (snapobj != firstsnap->ds_object) { dsl_dataset_t *ds; uint64_t used, comp, uncomp; err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds); if (err != 0) break; dsl_deadlist_space_range(&ds->ds_deadlist, dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX, &used, &comp, &uncomp); *usedp += used; *compp += comp; *uncompp += uncomp; snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj; ASSERT3U(snapobj, !=, 0); dsl_dataset_rele(ds, FTAG); } return (err); } /* * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline. * For example, they could both be snapshots of the same filesystem, and * 'earlier' is before 'later'. Or 'earlier' could be the origin of * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's * filesystem. Or 'earlier' could be the origin's origin. * * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg. */ boolean_t dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier, uint64_t earlier_txg) { dsl_pool_t *dp = later->ds_dir->dd_pool; int error; boolean_t ret; ASSERT(dsl_pool_config_held(dp)); ASSERT(earlier->ds_is_snapshot || earlier_txg != 0); if (earlier_txg == 0) earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg; if (later->ds_is_snapshot && earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg) return (B_FALSE); if (later->ds_dir == earlier->ds_dir) return (B_TRUE); if (!dsl_dir_is_clone(later->ds_dir)) return (B_FALSE); if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object) return (B_TRUE); dsl_dataset_t *origin; error = dsl_dataset_hold_obj(dp, dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin); if (error != 0) return (B_FALSE); ret = dsl_dataset_is_before(origin, earlier, earlier_txg); dsl_dataset_rele(origin, FTAG); return (ret); } void dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx) { objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx); } boolean_t dsl_dataset_is_zapified(dsl_dataset_t *ds) { dmu_object_info_t doi; dmu_object_info_from_db(ds->ds_dbuf, &doi); return (doi.doi_type == DMU_OTN_ZAP_METADATA); } boolean_t dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds) { return (dsl_dataset_is_zapified(ds) && zap_contains(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0); } Index: vendor-sys/illumos/dist/uts/common/fs/zfs/dsl_pool.c =================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/dsl_pool.c (revision 316916) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/dsl_pool.c (revision 316917) @@ -1,1082 +1,1095 @@ /* * 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 http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. - * Copyright (c) 2011, 2016 by Delphix. All rights reserved. + * Copyright (c) 2011, 2017 by Delphix. All rights reserved. * Copyright (c) 2013 Steven Hartland. All rights reserved. * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. * Copyright (c) 2014 Integros [integros.com] * Copyright 2016 Nexenta Systems, Inc. All rights reserved. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * ZFS Write Throttle * ------------------ * * ZFS must limit the rate of incoming writes to the rate at which it is able * to sync data modifications to the backend storage. Throttling by too much * creates an artificial limit; throttling by too little can only be sustained * for short periods and would lead to highly lumpy performance. On a per-pool * basis, ZFS tracks the amount of modified (dirty) data. As operations change * data, the amount of dirty data increases; as ZFS syncs out data, the amount * of dirty data decreases. When the amount of dirty data exceeds a * predetermined threshold further modifications are blocked until the amount * of dirty data decreases (as data is synced out). * * The limit on dirty data is tunable, and should be adjusted according to * both the IO capacity and available memory of the system. The larger the * window, the more ZFS is able to aggregate and amortize metadata (and data) * changes. However, memory is a limited resource, and allowing for more dirty * data comes at the cost of keeping other useful data in memory (for example * ZFS data cached by the ARC). * * Implementation * * As buffers are modified dsl_pool_willuse_space() increments both the per- * txg (dp_dirty_pertxg[]) and poolwide (dp_dirty_total) accounting of * dirty space used; dsl_pool_dirty_space() decrements those values as data * is synced out from dsl_pool_sync(). While only the poolwide value is * relevant, the per-txg value is useful for debugging. The tunable * zfs_dirty_data_max determines the dirty space limit. Once that value is * exceeded, new writes are halted until space frees up. * * The zfs_dirty_data_sync tunable dictates the threshold at which we * ensure that there is a txg syncing (see the comment in txg.c for a full * description of transaction group stages). * * The IO scheduler uses both the dirty space limit and current amount of * dirty data as inputs. Those values affect the number of concurrent IOs ZFS * issues. See the comment in vdev_queue.c for details of the IO scheduler. * * The delay is also calculated based on the amount of dirty data. See the * comment above dmu_tx_delay() for details. */ /* * zfs_dirty_data_max will be set to zfs_dirty_data_max_percent% of all memory, * capped at zfs_dirty_data_max_max. It can also be overridden in /etc/system. */ uint64_t zfs_dirty_data_max; uint64_t zfs_dirty_data_max_max = 4ULL * 1024 * 1024 * 1024; int zfs_dirty_data_max_percent = 10; /* * If there is at least this much dirty data, push out a txg. */ uint64_t zfs_dirty_data_sync = 64 * 1024 * 1024; /* * Once there is this amount of dirty data, the dmu_tx_delay() will kick in * and delay each transaction. * This value should be >= zfs_vdev_async_write_active_max_dirty_percent. */ int zfs_delay_min_dirty_percent = 60; /* * This controls how quickly the delay approaches infinity. * Larger values cause it to delay more for a given amount of dirty data. * Therefore larger values will cause there to be less dirty data for a * given throughput. * * 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 10x and 1/10th this number. * * Note: zfs_delay_scale * zfs_dirty_data_max must be < 2^64, due to the * multiply in dmu_tx_delay(). */ uint64_t zfs_delay_scale = 1000 * 1000 * 1000 / 2000; +/* + * This determines the number of threads used by the dp_sync_taskq. + */ +int zfs_sync_taskq_batch_pct = 75; hrtime_t zfs_throttle_delay = MSEC2NSEC(10); hrtime_t zfs_throttle_resolution = MSEC2NSEC(10); int dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp) { uint64_t obj; int err; err = zap_lookup(dp->dp_meta_objset, dsl_dir_phys(dp->dp_root_dir)->dd_child_dir_zapobj, name, sizeof (obj), 1, &obj); if (err) return (err); return (dsl_dir_hold_obj(dp, obj, name, dp, ddp)); } static dsl_pool_t * dsl_pool_open_impl(spa_t *spa, uint64_t txg) { dsl_pool_t *dp; blkptr_t *bp = spa_get_rootblkptr(spa); dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP); dp->dp_spa = spa; dp->dp_meta_rootbp = *bp; rrw_init(&dp->dp_config_rwlock, B_TRUE); txg_init(dp, txg); txg_list_create(&dp->dp_dirty_datasets, offsetof(dsl_dataset_t, ds_dirty_link)); txg_list_create(&dp->dp_dirty_zilogs, offsetof(zilog_t, zl_dirty_link)); txg_list_create(&dp->dp_dirty_dirs, offsetof(dsl_dir_t, dd_dirty_link)); txg_list_create(&dp->dp_sync_tasks, offsetof(dsl_sync_task_t, dst_node)); + dp->dp_sync_taskq = taskq_create("dp_sync_taskq", + zfs_sync_taskq_batch_pct, minclsyspri, 1, INT_MAX, + TASKQ_THREADS_CPU_PCT); + mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&dp->dp_spaceavail_cv, NULL, CV_DEFAULT, NULL); dp->dp_vnrele_taskq = taskq_create("zfs_vn_rele_taskq", 1, minclsyspri, 1, 4, 0); return (dp); } int dsl_pool_init(spa_t *spa, uint64_t txg, dsl_pool_t **dpp) { int err; dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp, &dp->dp_meta_objset); if (err != 0) dsl_pool_close(dp); else *dpp = dp; return (err); } int dsl_pool_open(dsl_pool_t *dp) { int err; dsl_dir_t *dd; dsl_dataset_t *ds; uint64_t obj; rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &dp->dp_root_dir_obj); if (err) goto out; err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, dp, &dp->dp_root_dir); if (err) goto out; err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir); if (err) goto out; if (spa_version(dp->dp_spa) >= SPA_VERSION_ORIGIN) { err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd); if (err) goto out; err = dsl_dataset_hold_obj(dp, dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds); if (err == 0) { err = dsl_dataset_hold_obj(dp, dsl_dataset_phys(ds)->ds_prev_snap_obj, dp, &dp->dp_origin_snap); dsl_dataset_rele(ds, FTAG); } dsl_dir_rele(dd, dp); if (err) goto out; } if (spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) { err = dsl_pool_open_special_dir(dp, FREE_DIR_NAME, &dp->dp_free_dir); if (err) goto out; err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj); if (err) goto out; VERIFY0(bpobj_open(&dp->dp_free_bpobj, dp->dp_meta_objset, obj)); } /* * Note: errors ignored, because the leak dir will not exist if we * have not encountered a leak yet. */ (void) dsl_pool_open_special_dir(dp, LEAK_DIR_NAME, &dp->dp_leak_dir); if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY)) { err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1, &dp->dp_bptree_obj); if (err != 0) goto out; } if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMPTY_BPOBJ)) { err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1, &dp->dp_empty_bpobj); if (err != 0) goto out; } err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_TMP_USERREFS, sizeof (uint64_t), 1, &dp->dp_tmp_userrefs_obj); if (err == ENOENT) err = 0; if (err) goto out; err = dsl_scan_init(dp, dp->dp_tx.tx_open_txg); out: rrw_exit(&dp->dp_config_rwlock, FTAG); return (err); } void dsl_pool_close(dsl_pool_t *dp) { /* * Drop our references from dsl_pool_open(). * * Since we held the origin_snap from "syncing" context (which * includes pool-opening context), it actually only got a "ref" * and not a hold, so just drop that here. */ if (dp->dp_origin_snap) dsl_dataset_rele(dp->dp_origin_snap, dp); if (dp->dp_mos_dir) dsl_dir_rele(dp->dp_mos_dir, dp); if (dp->dp_free_dir) dsl_dir_rele(dp->dp_free_dir, dp); if (dp->dp_leak_dir) dsl_dir_rele(dp->dp_leak_dir, dp); if (dp->dp_root_dir) dsl_dir_rele(dp->dp_root_dir, dp); bpobj_close(&dp->dp_free_bpobj); /* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */ if (dp->dp_meta_objset) dmu_objset_evict(dp->dp_meta_objset); txg_list_destroy(&dp->dp_dirty_datasets); txg_list_destroy(&dp->dp_dirty_zilogs); txg_list_destroy(&dp->dp_sync_tasks); txg_list_destroy(&dp->dp_dirty_dirs); + taskq_destroy(dp->dp_sync_taskq); + /* * We can't set retry to TRUE since we're explicitly specifying * a spa to flush. This is good enough; any missed buffers for * this spa won't cause trouble, and they'll eventually fall * out of the ARC just like any other unused buffer. */ arc_flush(dp->dp_spa, FALSE); txg_fini(dp); dsl_scan_fini(dp); dmu_buf_user_evict_wait(); rrw_destroy(&dp->dp_config_rwlock); mutex_destroy(&dp->dp_lock); taskq_destroy(dp->dp_vnrele_taskq); if (dp->dp_blkstats) kmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t)); kmem_free(dp, sizeof (dsl_pool_t)); } dsl_pool_t * dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg) { int err; dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg); objset_t *os; dsl_dataset_t *ds; uint64_t obj; rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); /* create and open the MOS (meta-objset) */ dp->dp_meta_objset = dmu_objset_create_impl(spa, NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx); /* create the pool directory */ err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx); ASSERT0(err); /* Initialize scan structures */ VERIFY0(dsl_scan_init(dp, txg)); /* create and open the root dir */ dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx); VERIFY0(dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, dp, &dp->dp_root_dir)); /* create and open the meta-objset dir */ (void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx); VERIFY0(dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir)); if (spa_version(spa) >= SPA_VERSION_DEADLISTS) { /* create and open the free dir */ (void) dsl_dir_create_sync(dp, dp->dp_root_dir, FREE_DIR_NAME, tx); VERIFY0(dsl_pool_open_special_dir(dp, FREE_DIR_NAME, &dp->dp_free_dir)); /* create and open the free_bplist */ obj = bpobj_alloc(dp->dp_meta_objset, SPA_OLD_MAXBLOCKSIZE, tx); VERIFY(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx) == 0); VERIFY0(bpobj_open(&dp->dp_free_bpobj, dp->dp_meta_objset, obj)); } if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB) dsl_pool_create_origin(dp, tx); /* create the root dataset */ obj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx); /* create the root objset */ VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, &ds)); rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); os = dmu_objset_create_impl(dp->dp_spa, ds, dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx); rrw_exit(&ds->ds_bp_rwlock, FTAG); #ifdef _KERNEL zfs_create_fs(os, kcred, zplprops, tx); #endif dsl_dataset_rele(ds, FTAG); dmu_tx_commit(tx); rrw_exit(&dp->dp_config_rwlock, FTAG); return (dp); } /* * Account for the meta-objset space in its placeholder dsl_dir. */ void dsl_pool_mos_diduse_space(dsl_pool_t *dp, int64_t used, int64_t comp, int64_t uncomp) { ASSERT3U(comp, ==, uncomp); /* it's all metadata */ mutex_enter(&dp->dp_lock); dp->dp_mos_used_delta += used; dp->dp_mos_compressed_delta += comp; dp->dp_mos_uncompressed_delta += uncomp; mutex_exit(&dp->dp_lock); } static void dsl_pool_sync_mos(dsl_pool_t *dp, dmu_tx_t *tx) { zio_t *zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); dmu_objset_sync(dp->dp_meta_objset, zio, tx); VERIFY0(zio_wait(zio)); dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", ""); spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp); } static void dsl_pool_dirty_delta(dsl_pool_t *dp, int64_t delta) { ASSERT(MUTEX_HELD(&dp->dp_lock)); if (delta < 0) ASSERT3U(-delta, <=, dp->dp_dirty_total); dp->dp_dirty_total += delta; /* * Note: we signal even when increasing dp_dirty_total. * This ensures forward progress -- each thread wakes the next waiter. */ if (dp->dp_dirty_total <= zfs_dirty_data_max) cv_signal(&dp->dp_spaceavail_cv); } void dsl_pool_sync(dsl_pool_t *dp, uint64_t txg) { zio_t *zio; dmu_tx_t *tx; dsl_dir_t *dd; dsl_dataset_t *ds; objset_t *mos = dp->dp_meta_objset; list_t synced_datasets; list_create(&synced_datasets, sizeof (dsl_dataset_t), offsetof(dsl_dataset_t, ds_synced_link)); tx = dmu_tx_create_assigned(dp, txg); /* * Write out all dirty blocks of dirty datasets. */ zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) != NULL) { /* * We must not sync any non-MOS datasets twice, because * we may have taken a snapshot of them. However, we * may sync newly-created datasets on pass 2. */ ASSERT(!list_link_active(&ds->ds_synced_link)); list_insert_tail(&synced_datasets, ds); dsl_dataset_sync(ds, zio, tx); } VERIFY0(zio_wait(zio)); /* * We have written all of the accounted dirty data, so our * dp_space_towrite should now be zero. However, some seldom-used * code paths do not adhere to this (e.g. dbuf_undirty(), also * rounding error in dbuf_write_physdone). * Shore up the accounting of any dirtied space now. */ dsl_pool_undirty_space(dp, dp->dp_dirty_pertxg[txg & TXG_MASK], txg); /* * Update the long range free counter after * we're done syncing user data */ mutex_enter(&dp->dp_lock); ASSERT(spa_sync_pass(dp->dp_spa) == 1 || dp->dp_long_free_dirty_pertxg[txg & TXG_MASK] == 0); dp->dp_long_free_dirty_pertxg[txg & TXG_MASK] = 0; mutex_exit(&dp->dp_lock); /* * After the data blocks have been written (ensured by the zio_wait() - * above), update the user/group space accounting. + * above), update the user/group space accounting. This happens + * in tasks dispatched to dp_sync_taskq, so wait for them before + * continuing. */ for (ds = list_head(&synced_datasets); ds != NULL; ds = list_next(&synced_datasets, ds)) { dmu_objset_do_userquota_updates(ds->ds_objset, tx); } + taskq_wait(dp->dp_sync_taskq); /* * Sync the datasets again to push out the changes due to * userspace updates. This must be done before we process the * sync tasks, so that any snapshots will have the correct * user accounting information (and we won't get confused * about which blocks are part of the snapshot). */ zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) != NULL) { ASSERT(list_link_active(&ds->ds_synced_link)); dmu_buf_rele(ds->ds_dbuf, ds); dsl_dataset_sync(ds, zio, tx); } VERIFY0(zio_wait(zio)); /* * Now that the datasets have been completely synced, we can * clean up our in-memory structures accumulated while syncing: * * - move dead blocks from the pending deadlist to the on-disk deadlist * - release hold from dsl_dataset_dirty() */ while ((ds = list_remove_head(&synced_datasets)) != NULL) { dsl_dataset_sync_done(ds, tx); } while ((dd = txg_list_remove(&dp->dp_dirty_dirs, txg)) != NULL) { dsl_dir_sync(dd, tx); } /* * The MOS's space is accounted for in the pool/$MOS * (dp_mos_dir). We can't modify the mos while we're syncing * it, so we remember the deltas and apply them here. */ if (dp->dp_mos_used_delta != 0 || dp->dp_mos_compressed_delta != 0 || dp->dp_mos_uncompressed_delta != 0) { dsl_dir_diduse_space(dp->dp_mos_dir, DD_USED_HEAD, dp->dp_mos_used_delta, dp->dp_mos_compressed_delta, dp->dp_mos_uncompressed_delta, tx); dp->dp_mos_used_delta = 0; dp->dp_mos_compressed_delta = 0; dp->dp_mos_uncompressed_delta = 0; } - if (list_head(&mos->os_dirty_dnodes[txg & TXG_MASK]) != NULL || - list_head(&mos->os_free_dnodes[txg & TXG_MASK]) != NULL) { + if (!multilist_is_empty(mos->os_dirty_dnodes[txg & TXG_MASK])) { dsl_pool_sync_mos(dp, tx); } /* * If we modify a dataset in the same txg that we want to destroy it, * its dsl_dir's dd_dbuf will be dirty, and thus have a hold on it. * dsl_dir_destroy_check() will fail if there are unexpected holds. * Therefore, we want to sync the MOS (thus syncing the dd_dbuf * and clearing the hold on it) before we process the sync_tasks. * The MOS data dirtied by the sync_tasks will be synced on the next * pass. */ if (!txg_list_empty(&dp->dp_sync_tasks, txg)) { dsl_sync_task_t *dst; /* * No more sync tasks should have been added while we * were syncing. */ ASSERT3U(spa_sync_pass(dp->dp_spa), ==, 1); while ((dst = txg_list_remove(&dp->dp_sync_tasks, txg)) != NULL) dsl_sync_task_sync(dst, tx); } dmu_tx_commit(tx); DTRACE_PROBE2(dsl_pool_sync__done, dsl_pool_t *dp, dp, uint64_t, txg); } void dsl_pool_sync_done(dsl_pool_t *dp, uint64_t txg) { zilog_t *zilog; while (zilog = txg_list_head(&dp->dp_dirty_zilogs, txg)) { dsl_dataset_t *ds = dmu_objset_ds(zilog->zl_os); /* * We don't remove the zilog from the dp_dirty_zilogs * list until after we've cleaned it. This ensures that * callers of zilog_is_dirty() receive an accurate * answer when they are racing with the spa sync thread. */ zil_clean(zilog, txg); (void) txg_list_remove_this(&dp->dp_dirty_zilogs, zilog, txg); ASSERT(!dmu_objset_is_dirty(zilog->zl_os, txg)); dmu_buf_rele(ds->ds_dbuf, zilog); } ASSERT(!dmu_objset_is_dirty(dp->dp_meta_objset, txg)); } /* * TRUE if the current thread is the tx_sync_thread or if we * are being called from SPA context during pool initialization. */ int dsl_pool_sync_context(dsl_pool_t *dp) { return (curthread == dp->dp_tx.tx_sync_thread || - spa_is_initializing(dp->dp_spa)); + spa_is_initializing(dp->dp_spa) || + taskq_member(dp->dp_sync_taskq, curthread)); } uint64_t dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree) { uint64_t space, resv; /* * If we're trying to assess whether it's OK to do a free, * cut the reservation in half to allow forward progress * (e.g. make it possible to rm(1) files from a full pool). */ space = spa_get_dspace(dp->dp_spa); resv = spa_get_slop_space(dp->dp_spa); if (netfree) resv >>= 1; return (space - resv); } boolean_t dsl_pool_need_dirty_delay(dsl_pool_t *dp) { uint64_t delay_min_bytes = zfs_dirty_data_max * zfs_delay_min_dirty_percent / 100; boolean_t rv; mutex_enter(&dp->dp_lock); if (dp->dp_dirty_total > zfs_dirty_data_sync) txg_kick(dp); rv = (dp->dp_dirty_total > delay_min_bytes); mutex_exit(&dp->dp_lock); return (rv); } void dsl_pool_dirty_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx) { if (space > 0) { mutex_enter(&dp->dp_lock); dp->dp_dirty_pertxg[tx->tx_txg & TXG_MASK] += space; dsl_pool_dirty_delta(dp, space); mutex_exit(&dp->dp_lock); } } void dsl_pool_undirty_space(dsl_pool_t *dp, int64_t space, uint64_t txg) { ASSERT3S(space, >=, 0); if (space == 0) return; mutex_enter(&dp->dp_lock); if (dp->dp_dirty_pertxg[txg & TXG_MASK] < space) { /* XXX writing something we didn't dirty? */ space = dp->dp_dirty_pertxg[txg & TXG_MASK]; } ASSERT3U(dp->dp_dirty_pertxg[txg & TXG_MASK], >=, space); dp->dp_dirty_pertxg[txg & TXG_MASK] -= space; ASSERT3U(dp->dp_dirty_total, >=, space); dsl_pool_dirty_delta(dp, -space); mutex_exit(&dp->dp_lock); } /* ARGSUSED */ static int upgrade_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) { dmu_tx_t *tx = arg; dsl_dataset_t *ds, *prev = NULL; int err; 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) { 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 (dsl_dataset_phys(prev)->ds_next_snap_obj != ds->ds_object) break; dsl_dataset_rele(ds, FTAG); ds = prev; prev = NULL; } if (prev == NULL) { prev = dp->dp_origin_snap; /* * The $ORIGIN can't have any data, or the accounting * will be wrong. */ rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); ASSERT0(dsl_dataset_phys(prev)->ds_bp.blk_birth); rrw_exit(&ds->ds_bp_rwlock, FTAG); /* The origin doesn't get attached to itself */ if (ds->ds_object == prev->ds_object) { dsl_dataset_rele(ds, FTAG); return (0); } dmu_buf_will_dirty(ds->ds_dbuf, tx); dsl_dataset_phys(ds)->ds_prev_snap_obj = prev->ds_object; dsl_dataset_phys(ds)->ds_prev_snap_txg = dsl_dataset_phys(prev)->ds_creation_txg; dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); dsl_dir_phys(ds->ds_dir)->dd_origin_obj = prev->ds_object; dmu_buf_will_dirty(prev->ds_dbuf, tx); dsl_dataset_phys(prev)->ds_num_children++; if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0) { ASSERT(ds->ds_prev == NULL); VERIFY0(dsl_dataset_hold_obj(dp, dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev)); } } ASSERT3U(dsl_dir_phys(ds->ds_dir)->dd_origin_obj, ==, prev->ds_object); ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_obj, ==, prev->ds_object); if (dsl_dataset_phys(prev)->ds_next_clones_obj == 0) { dmu_buf_will_dirty(prev->ds_dbuf, tx); dsl_dataset_phys(prev)->ds_next_clones_obj = zap_create(dp->dp_meta_objset, DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); } VERIFY0(zap_add_int(dp->dp_meta_objset, dsl_dataset_phys(prev)->ds_next_clones_obj, ds->ds_object, tx)); dsl_dataset_rele(ds, FTAG); if (prev != dp->dp_origin_snap) dsl_dataset_rele(prev, FTAG); return (0); } void dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx) { ASSERT(dmu_tx_is_syncing(tx)); ASSERT(dp->dp_origin_snap != NULL); VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, upgrade_clones_cb, tx, DS_FIND_CHILDREN | DS_FIND_SERIALIZE)); } /* ARGSUSED */ static int upgrade_dir_clones_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) { dmu_tx_t *tx = arg; objset_t *mos = dp->dp_meta_objset; if (dsl_dir_phys(ds->ds_dir)->dd_origin_obj != 0) { dsl_dataset_t *origin; VERIFY0(dsl_dataset_hold_obj(dp, dsl_dir_phys(ds->ds_dir)->dd_origin_obj, FTAG, &origin)); if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) { dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); dsl_dir_phys(origin->ds_dir)->dd_clones = zap_create(mos, DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); } VERIFY0(zap_add_int(dp->dp_meta_objset, dsl_dir_phys(origin->ds_dir)->dd_clones, ds->ds_object, tx)); dsl_dataset_rele(origin, FTAG); } return (0); } void dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx) { ASSERT(dmu_tx_is_syncing(tx)); uint64_t obj; (void) dsl_dir_create_sync(dp, dp->dp_root_dir, FREE_DIR_NAME, tx); VERIFY0(dsl_pool_open_special_dir(dp, FREE_DIR_NAME, &dp->dp_free_dir)); /* * We can't use bpobj_alloc(), because spa_version() still * returns the old version, and we need a new-version bpobj with * subobj support. So call dmu_object_alloc() directly. */ obj = dmu_object_alloc(dp->dp_meta_objset, DMU_OT_BPOBJ, SPA_OLD_MAXBLOCKSIZE, DMU_OT_BPOBJ_HDR, sizeof (bpobj_phys_t), tx); VERIFY0(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx)); VERIFY0(bpobj_open(&dp->dp_free_bpobj, dp->dp_meta_objset, obj)); VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, upgrade_dir_clones_cb, tx, DS_FIND_CHILDREN | DS_FIND_SERIALIZE)); } void dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx) { uint64_t dsobj; dsl_dataset_t *ds; ASSERT(dmu_tx_is_syncing(tx)); ASSERT(dp->dp_origin_snap == NULL); ASSERT(rrw_held(&dp->dp_config_rwlock, RW_WRITER)); /* create the origin dir, ds, & snap-ds */ dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME, NULL, 0, kcred, tx); VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); dsl_dataset_snapshot_sync_impl(ds, ORIGIN_DIR_NAME, tx); VERIFY0(dsl_dataset_hold_obj(dp, dsl_dataset_phys(ds)->ds_prev_snap_obj, dp, &dp->dp_origin_snap)); dsl_dataset_rele(ds, FTAG); } taskq_t * dsl_pool_vnrele_taskq(dsl_pool_t *dp) { return (dp->dp_vnrele_taskq); } /* * Walk through the pool-wide zap object of temporary snapshot user holds * and release them. */ void dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp) { zap_attribute_t za; zap_cursor_t zc; objset_t *mos = dp->dp_meta_objset; uint64_t zapobj = dp->dp_tmp_userrefs_obj; nvlist_t *holds; if (zapobj == 0) return; ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); holds = fnvlist_alloc(); for (zap_cursor_init(&zc, mos, zapobj); zap_cursor_retrieve(&zc, &za) == 0; zap_cursor_advance(&zc)) { char *htag; nvlist_t *tags; htag = strchr(za.za_name, '-'); *htag = '\0'; ++htag; if (nvlist_lookup_nvlist(holds, za.za_name, &tags) != 0) { tags = fnvlist_alloc(); fnvlist_add_boolean(tags, htag); fnvlist_add_nvlist(holds, za.za_name, tags); fnvlist_free(tags); } else { fnvlist_add_boolean(tags, htag); } } dsl_dataset_user_release_tmp(dp, holds); fnvlist_free(holds); zap_cursor_fini(&zc); } /* * Create the pool-wide zap object for storing temporary snapshot holds. */ void dsl_pool_user_hold_create_obj(dsl_pool_t *dp, dmu_tx_t *tx) { objset_t *mos = dp->dp_meta_objset; ASSERT(dp->dp_tmp_userrefs_obj == 0); ASSERT(dmu_tx_is_syncing(tx)); dp->dp_tmp_userrefs_obj = zap_create_link(mos, DMU_OT_USERREFS, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_TMP_USERREFS, tx); } static int dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj, const char *tag, uint64_t now, dmu_tx_t *tx, boolean_t holding) { objset_t *mos = dp->dp_meta_objset; uint64_t zapobj = dp->dp_tmp_userrefs_obj; char *name; int error; ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); ASSERT(dmu_tx_is_syncing(tx)); /* * If the pool was created prior to SPA_VERSION_USERREFS, the * zap object for temporary holds might not exist yet. */ if (zapobj == 0) { if (holding) { dsl_pool_user_hold_create_obj(dp, tx); zapobj = dp->dp_tmp_userrefs_obj; } else { return (SET_ERROR(ENOENT)); } } name = kmem_asprintf("%llx-%s", (u_longlong_t)dsobj, tag); if (holding) error = zap_add(mos, zapobj, name, 8, 1, &now, tx); else error = zap_remove(mos, zapobj, name, tx); strfree(name); return (error); } /* * Add a temporary hold for the given dataset object and tag. */ int dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, const char *tag, uint64_t now, dmu_tx_t *tx) { return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, now, tx, B_TRUE)); } /* * Release a temporary hold for the given dataset object and tag. */ int dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, const char *tag, dmu_tx_t *tx) { return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, NULL, tx, B_FALSE)); } /* * DSL Pool Configuration Lock * * The dp_config_rwlock protects against changes to DSL state (e.g. dataset * creation / destruction / rename / property setting). It must be held for * read to hold a dataset or dsl_dir. I.e. you must call * dsl_pool_config_enter() or dsl_pool_hold() before calling * dsl_{dataset,dir}_hold{_obj}. In most circumstances, the dp_config_rwlock * must be held continuously until all datasets and dsl_dirs are released. * * The only exception to this rule is that if a "long hold" is placed on * a dataset, then the dp_config_rwlock may be dropped while the dataset * is still held. The long hold will prevent the dataset from being * destroyed -- the destroy will fail with EBUSY. A long hold can be * obtained by calling dsl_dataset_long_hold(), or by "owning" a dataset * (by calling dsl_{dataset,objset}_{try}own{_obj}). * * Legitimate long-holders (including owners) should be long-running, cancelable * tasks that should cause "zfs destroy" to fail. This includes DMU * consumers (i.e. a ZPL filesystem being mounted or ZVOL being open), * "zfs send", and "zfs diff". There are several other long-holders whose * uses are suboptimal (e.g. "zfs promote", and zil_suspend()). * * The usual formula for long-holding would be: * dsl_pool_hold() * dsl_dataset_hold() * ... perform checks ... * dsl_dataset_long_hold() * dsl_pool_rele() * ... perform long-running task ... * dsl_dataset_long_rele() * dsl_dataset_rele() * * Note that when the long hold is released, the dataset is still held but * the pool is not held. The dataset may change arbitrarily during this time * (e.g. it could be destroyed). Therefore you shouldn't do anything to the * dataset except release it. * * User-initiated operations (e.g. ioctls, zfs_ioc_*()) are either read-only * or modifying operations. * * Modifying operations should generally use dsl_sync_task(). The synctask * infrastructure enforces proper locking strategy with respect to the * dp_config_rwlock. See the comment above dsl_sync_task() for details. * * Read-only operations will manually hold the pool, then the dataset, obtain * information from the dataset, then release the pool and dataset. * dmu_objset_{hold,rele}() are convenience routines that also do the pool * hold/rele. */ int dsl_pool_hold(const char *name, void *tag, dsl_pool_t **dp) { spa_t *spa; int error; error = spa_open(name, &spa, tag); if (error == 0) { *dp = spa_get_dsl(spa); dsl_pool_config_enter(*dp, tag); } return (error); } void dsl_pool_rele(dsl_pool_t *dp, void *tag) { dsl_pool_config_exit(dp, tag); spa_close(dp->dp_spa, tag); } void dsl_pool_config_enter(dsl_pool_t *dp, void *tag) { /* * We use a "reentrant" reader-writer lock, but not reentrantly. * * The rrwlock can (with the track_all flag) track all reading threads, * which is very useful for debugging which code path failed to release * the lock, and for verifying that the *current* thread does hold * the lock. * * (Unlike a rwlock, which knows that N threads hold it for * read, but not *which* threads, so rw_held(RW_READER) returns TRUE * if any thread holds it for read, even if this thread doesn't). */ ASSERT(!rrw_held(&dp->dp_config_rwlock, RW_READER)); rrw_enter(&dp->dp_config_rwlock, RW_READER, tag); } void dsl_pool_config_enter_prio(dsl_pool_t *dp, void *tag) { ASSERT(!rrw_held(&dp->dp_config_rwlock, RW_READER)); rrw_enter_read_prio(&dp->dp_config_rwlock, tag); } void dsl_pool_config_exit(dsl_pool_t *dp, void *tag) { rrw_exit(&dp->dp_config_rwlock, tag); } boolean_t dsl_pool_config_held(dsl_pool_t *dp) { return (RRW_LOCK_HELD(&dp->dp_config_rwlock)); } boolean_t dsl_pool_config_held_writer(dsl_pool_t *dp) { return (RRW_WRITE_HELD(&dp->dp_config_rwlock)); } Index: vendor-sys/illumos/dist/uts/common/fs/zfs/multilist.c =================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/multilist.c (revision 316916) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/multilist.c (revision 316917) @@ -1,392 +1,401 @@ /* * CDDL HEADER START * * This file and its contents are supplied under the terms of the * Common Development and Distribution License ("CDDL"), version 1.0. * You may only use this file in accordance with the terms of version * 1.0 of the CDDL. * * A full copy of the text of the CDDL should have accompanied this * source. A copy of the CDDL is also available via the Internet at * http://www.illumos.org/license/CDDL. * * CDDL HEADER END */ /* * Copyright (c) 2013, 2017 by Delphix. All rights reserved. */ #include #include /* needed for spa_get_random() */ #include /* * This overrides the number of sublists in each multilist_t, which defaults * to the number of CPUs in the system (see multilist_create()). */ int zfs_multilist_num_sublists = 0; /* * Given the object contained on the list, return a pointer to the * object's multilist_node_t structure it contains. */ static multilist_node_t * multilist_d2l(multilist_t *ml, void *obj) { return ((multilist_node_t *)((char *)obj + ml->ml_offset)); } /* * Initialize a new mutlilist using the parameters specified. * * - 'size' denotes the size of the structure containing the * multilist_node_t. * - 'offset' denotes the byte offset of the mutlilist_node_t within * the structure that contains it. * - 'num' specifies the number of internal sublists to create. * - 'index_func' is used to determine which sublist to insert into * when the multilist_insert() function is called; as well as which * sublist to remove from when multilist_remove() is called. The * requirements this function must meet, are the following: * * - It must always return the same value when called on the same * object (to ensure the object is removed from the list it was * inserted into). * * - It must return a value in the range [0, number of sublists). * The multilist_get_num_sublists() function may be used to * determine the number of sublists in the multilist. * * Also, in order to reduce internal contention between the sublists * during insertion and removal, this function should choose evenly * between all available sublists when inserting. This isn't a hard * requirement, but a general rule of thumb in order to garner the * best multi-threaded performance out of the data structure. */ -static void -multilist_create_impl(multilist_t *ml, size_t size, size_t offset, +static multilist_t * +multilist_create_impl(size_t size, size_t offset, unsigned int num, multilist_sublist_index_func_t *index_func) { - ASSERT3P(ml, !=, NULL); ASSERT3U(size, >, 0); ASSERT3U(size, >=, offset + sizeof (multilist_node_t)); ASSERT3U(num, >, 0); ASSERT3P(index_func, !=, NULL); + multilist_t *ml = kmem_alloc(sizeof (*ml), KM_SLEEP); ml->ml_offset = offset; ml->ml_num_sublists = num; ml->ml_index_func = index_func; ml->ml_sublists = kmem_zalloc(sizeof (multilist_sublist_t) * ml->ml_num_sublists, KM_SLEEP); ASSERT3P(ml->ml_sublists, !=, NULL); for (int i = 0; i < ml->ml_num_sublists; i++) { multilist_sublist_t *mls = &ml->ml_sublists[i]; mutex_init(&mls->mls_lock, NULL, MUTEX_DEFAULT, NULL); list_create(&mls->mls_list, size, offset); } + return (ml); } /* - * Initialize a new sublist, using the default number of sublists + * Allocate a new multilist, using the default number of sublists * (the number of CPUs, or at least 4, or the tunable * zfs_multilist_num_sublists). */ -void -multilist_create(multilist_t *ml, size_t size, size_t offset, +multilist_t * +multilist_create(size_t size, size_t offset, multilist_sublist_index_func_t *index_func) { int num_sublists; if (zfs_multilist_num_sublists > 0) { num_sublists = zfs_multilist_num_sublists; } else { num_sublists = MAX(boot_ncpus, 4); } - multilist_create_impl(ml, size, offset, num_sublists, index_func); + return (multilist_create_impl(size, offset, num_sublists, index_func)); } /* * Destroy the given multilist object, and free up any memory it holds. */ void multilist_destroy(multilist_t *ml) { ASSERT(multilist_is_empty(ml)); for (int i = 0; i < ml->ml_num_sublists; i++) { multilist_sublist_t *mls = &ml->ml_sublists[i]; ASSERT(list_is_empty(&mls->mls_list)); list_destroy(&mls->mls_list); mutex_destroy(&mls->mls_lock); } ASSERT3P(ml->ml_sublists, !=, NULL); kmem_free(ml->ml_sublists, sizeof (multilist_sublist_t) * ml->ml_num_sublists); ml->ml_num_sublists = 0; ml->ml_offset = 0; + kmem_free(ml, sizeof (multilist_t)); } /* * Insert the given object into the multilist. * * This function will insert the object specified into the sublist * determined using the function given at multilist creation time. * * The sublist locks are automatically acquired if not already held, to * ensure consistency when inserting and removing from multiple threads. */ void multilist_insert(multilist_t *ml, void *obj) { unsigned int sublist_idx = ml->ml_index_func(ml, obj); multilist_sublist_t *mls; boolean_t need_lock; DTRACE_PROBE3(multilist__insert, multilist_t *, ml, unsigned int, sublist_idx, void *, obj); ASSERT3U(sublist_idx, <, ml->ml_num_sublists); mls = &ml->ml_sublists[sublist_idx]; /* * Note: Callers may already hold the sublist lock by calling * multilist_sublist_lock(). Here we rely on MUTEX_HELD() * returning TRUE if and only if the current thread holds the * lock. While it's a little ugly to make the lock recursive in * this way, it works and allows the calling code to be much * simpler -- otherwise it would have to pass around a flag * indicating that it already has the lock. */ need_lock = !MUTEX_HELD(&mls->mls_lock); if (need_lock) mutex_enter(&mls->mls_lock); ASSERT(!multilist_link_active(multilist_d2l(ml, obj))); multilist_sublist_insert_head(mls, obj); if (need_lock) mutex_exit(&mls->mls_lock); } /* * Remove the given object from the multilist. * * This function will remove the object specified from the sublist * determined using the function given at multilist creation time. * * The necessary sublist locks are automatically acquired, to ensure * consistency when inserting and removing from multiple threads. */ void multilist_remove(multilist_t *ml, void *obj) { unsigned int sublist_idx = ml->ml_index_func(ml, obj); multilist_sublist_t *mls; boolean_t need_lock; DTRACE_PROBE3(multilist__remove, multilist_t *, ml, unsigned int, sublist_idx, void *, obj); ASSERT3U(sublist_idx, <, ml->ml_num_sublists); mls = &ml->ml_sublists[sublist_idx]; /* See comment in multilist_insert(). */ need_lock = !MUTEX_HELD(&mls->mls_lock); if (need_lock) mutex_enter(&mls->mls_lock); ASSERT(multilist_link_active(multilist_d2l(ml, obj))); multilist_sublist_remove(mls, obj); if (need_lock) mutex_exit(&mls->mls_lock); } /* * Check to see if this multilist object is empty. * * This will return TRUE if it finds all of the sublists of this * multilist to be empty, and FALSE otherwise. Each sublist lock will be * automatically acquired as necessary. * * If concurrent insertions and removals are occurring, the semantics * of this function become a little fuzzy. Instead of locking all * sublists for the entire call time of the function, each sublist is * only locked as it is individually checked for emptiness. Thus, it's * possible for this function to return TRUE with non-empty sublists at * the time the function returns. This would be due to another thread * inserting into a given sublist, after that specific sublist was check * and deemed empty, but before all sublists have been checked. */ int multilist_is_empty(multilist_t *ml) { for (int i = 0; i < ml->ml_num_sublists; i++) { multilist_sublist_t *mls = &ml->ml_sublists[i]; /* See comment in multilist_insert(). */ boolean_t need_lock = !MUTEX_HELD(&mls->mls_lock); if (need_lock) mutex_enter(&mls->mls_lock); if (!list_is_empty(&mls->mls_list)) { if (need_lock) mutex_exit(&mls->mls_lock); return (FALSE); } if (need_lock) mutex_exit(&mls->mls_lock); } return (TRUE); } /* Return the number of sublists composing this multilist */ unsigned int multilist_get_num_sublists(multilist_t *ml) { return (ml->ml_num_sublists); } /* Return a randomly selected, valid sublist index for this multilist */ unsigned int multilist_get_random_index(multilist_t *ml) { return (spa_get_random(ml->ml_num_sublists)); } /* Lock and return the sublist specified at the given index */ multilist_sublist_t * multilist_sublist_lock(multilist_t *ml, unsigned int sublist_idx) { multilist_sublist_t *mls; ASSERT3U(sublist_idx, <, ml->ml_num_sublists); mls = &ml->ml_sublists[sublist_idx]; mutex_enter(&mls->mls_lock); return (mls); +} + +/* Lock and return the sublist that would be used to store the specified obj */ +multilist_sublist_t * +multilist_sublist_lock_obj(multilist_t *ml, void *obj) +{ + return (multilist_sublist_lock(ml, ml->ml_index_func(ml, obj))); } void multilist_sublist_unlock(multilist_sublist_t *mls) { mutex_exit(&mls->mls_lock); } /* * We're allowing any object to be inserted into this specific sublist, * but this can lead to trouble if multilist_remove() is called to * remove this object. Specifically, if calling ml_index_func on this * object returns an index for sublist different than what is passed as * a parameter here, any call to multilist_remove() with this newly * inserted object is undefined! (the call to multilist_remove() will * remove the object from a list that it isn't contained in) */ void multilist_sublist_insert_head(multilist_sublist_t *mls, void *obj) { ASSERT(MUTEX_HELD(&mls->mls_lock)); list_insert_head(&mls->mls_list, obj); } /* please see comment above multilist_sublist_insert_head */ void multilist_sublist_insert_tail(multilist_sublist_t *mls, void *obj) { ASSERT(MUTEX_HELD(&mls->mls_lock)); list_insert_tail(&mls->mls_list, obj); } /* * Move the object one element forward in the list. * * This function will move the given object forward in the list (towards * the head) by one object. So, in essence, it will swap its position in * the list with its "prev" pointer. If the given object is already at the * head of the list, it cannot be moved forward any more than it already * is, so no action is taken. * * NOTE: This function **must not** remove any object from the list other * than the object given as the parameter. This is relied upon in * arc_evict_state_impl(). */ void multilist_sublist_move_forward(multilist_sublist_t *mls, void *obj) { void *prev = list_prev(&mls->mls_list, obj); ASSERT(MUTEX_HELD(&mls->mls_lock)); ASSERT(!list_is_empty(&mls->mls_list)); /* 'obj' must be at the head of the list, nothing to do */ if (prev == NULL) return; list_remove(&mls->mls_list, obj); list_insert_before(&mls->mls_list, prev, obj); } void multilist_sublist_remove(multilist_sublist_t *mls, void *obj) { ASSERT(MUTEX_HELD(&mls->mls_lock)); list_remove(&mls->mls_list, obj); } void * multilist_sublist_head(multilist_sublist_t *mls) { ASSERT(MUTEX_HELD(&mls->mls_lock)); return (list_head(&mls->mls_list)); } void * multilist_sublist_tail(multilist_sublist_t *mls) { ASSERT(MUTEX_HELD(&mls->mls_lock)); return (list_tail(&mls->mls_list)); } void * multilist_sublist_next(multilist_sublist_t *mls, void *obj) { ASSERT(MUTEX_HELD(&mls->mls_lock)); return (list_next(&mls->mls_list, obj)); } void * multilist_sublist_prev(multilist_sublist_t *mls, void *obj) { ASSERT(MUTEX_HELD(&mls->mls_lock)); return (list_prev(&mls->mls_list, obj)); } void multilist_link_init(multilist_node_t *link) { list_link_init(link); } int multilist_link_active(multilist_node_t *link) { return (list_link_active(link)); } Index: vendor-sys/illumos/dist/uts/common/fs/zfs/spa_misc.c =================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/spa_misc.c (revision 316916) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/spa_misc.c (revision 316917) @@ -1,2048 +1,2048 @@ /* * 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 http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. - * Copyright (c) 2011, 2015 by Delphix. All rights reserved. + * Copyright (c) 2011, 2017 by Delphix. All rights reserved. * Copyright 2015 Nexenta Systems, Inc. All rights reserved. * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. * Copyright 2013 Saso Kiselkov. All rights reserved. * Copyright (c) 2014 Integros [integros.com] */ #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 "zfs_prop.h" #include /* * SPA locking * * There are four basic locks for managing spa_t structures: * * spa_namespace_lock (global mutex) * * This lock must be acquired to do any of the following: * * - Lookup a spa_t by name * - Add or remove a spa_t from the namespace * - Increase spa_refcount from non-zero * - Check if spa_refcount is zero * - Rename a spa_t * - add/remove/attach/detach devices * - Held for the duration of create/destroy/import/export * * It does not need to handle recursion. A create or destroy may * reference objects (files or zvols) in other pools, but by * definition they must have an existing reference, and will never need * to lookup a spa_t by name. * * spa_refcount (per-spa refcount_t protected by mutex) * * This reference count keep track of any active users of the spa_t. The * spa_t cannot be destroyed or freed while this is non-zero. Internally, * the refcount is never really 'zero' - opening a pool implicitly keeps * some references in the DMU. Internally we check against spa_minref, but * present the image of a zero/non-zero value to consumers. * * spa_config_lock[] (per-spa array of rwlocks) * * This protects the spa_t from config changes, and must be held in * the following circumstances: * * - RW_READER to perform I/O to the spa * - RW_WRITER to change the vdev config * * The locking order is fairly straightforward: * * spa_namespace_lock -> spa_refcount * * The namespace lock must be acquired to increase the refcount from 0 * or to check if it is zero. * * spa_refcount -> spa_config_lock[] * * There must be at least one valid reference on the spa_t to acquire * the config lock. * * spa_namespace_lock -> spa_config_lock[] * * The namespace lock must always be taken before the config lock. * * * The spa_namespace_lock can be acquired directly and is globally visible. * * The namespace is manipulated using the following functions, all of which * require the spa_namespace_lock to be held. * * spa_lookup() Lookup a spa_t by name. * * spa_add() Create a new spa_t in the namespace. * * spa_remove() Remove a spa_t from the namespace. This also * frees up any memory associated with the spa_t. * * spa_next() Returns the next spa_t in the system, or the * first if NULL is passed. * * spa_evict_all() Shutdown and remove all spa_t structures in * the system. * * spa_guid_exists() Determine whether a pool/device guid exists. * * The spa_refcount is manipulated using the following functions: * * spa_open_ref() Adds a reference to the given spa_t. Must be * called with spa_namespace_lock held if the * refcount is currently zero. * * spa_close() Remove a reference from the spa_t. This will * not free the spa_t or remove it from the * namespace. No locking is required. * * spa_refcount_zero() Returns true if the refcount is currently * zero. Must be called with spa_namespace_lock * held. * * The spa_config_lock[] is an array of rwlocks, ordered as follows: * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV. * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}(). * * To read the configuration, it suffices to hold one of these locks as reader. * To modify the configuration, you must hold all locks as writer. To modify * vdev state without altering the vdev tree's topology (e.g. online/offline), * you must hold SCL_STATE and SCL_ZIO as writer. * * We use these distinct config locks to avoid recursive lock entry. * For example, spa_sync() (which holds SCL_CONFIG as reader) induces * block allocations (SCL_ALLOC), which may require reading space maps * from disk (dmu_read() -> zio_read() -> SCL_ZIO). * * The spa config locks cannot be normal rwlocks because we need the * ability to hand off ownership. For example, SCL_ZIO is acquired * by the issuing thread and later released by an interrupt thread. * They do, however, obey the usual write-wanted semantics to prevent * writer (i.e. system administrator) starvation. * * The lock acquisition rules are as follows: * * SCL_CONFIG * Protects changes to the vdev tree topology, such as vdev * add/remove/attach/detach. Protects the dirty config list * (spa_config_dirty_list) and the set of spares and l2arc devices. * * SCL_STATE * Protects changes to pool state and vdev state, such as vdev * online/offline/fault/degrade/clear. Protects the dirty state list * (spa_state_dirty_list) and global pool state (spa_state). * * SCL_ALLOC * Protects changes to metaslab groups and classes. * Held as reader by metaslab_alloc() and metaslab_claim(). * * SCL_ZIO * Held by bp-level zios (those which have no io_vd upon entry) * to prevent changes to the vdev tree. The bp-level zio implicitly * protects all of its vdev child zios, which do not hold SCL_ZIO. * * SCL_FREE * Protects changes to metaslab groups and classes. * Held as reader by metaslab_free(). SCL_FREE is distinct from * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free * blocks in zio_done() while another i/o that holds either * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete. * * SCL_VDEV * Held as reader to prevent changes to the vdev tree during trivial * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the * other locks, and lower than all of them, to ensure that it's safe * to acquire regardless of caller context. * * In addition, the following rules apply: * * (a) spa_props_lock protects pool properties, spa_config and spa_config_list. * The lock ordering is SCL_CONFIG > spa_props_lock. * * (b) I/O operations on leaf vdevs. For any zio operation that takes * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(), * or zio_write_phys() -- the caller must ensure that the config cannot * cannot change in the interim, and that the vdev cannot be reopened. * SCL_STATE as reader suffices for both. * * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit(). * * spa_vdev_enter() Acquire the namespace lock and the config lock * for writing. * * spa_vdev_exit() Release the config lock, wait for all I/O * to complete, sync the updated configs to the * cache, and release the namespace lock. * * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit(). * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual * locking is, always, based on spa_namespace_lock and spa_config_lock[]. * * spa_rename() is also implemented within this file since it requires * manipulation of the namespace. */ static avl_tree_t spa_namespace_avl; kmutex_t spa_namespace_lock; static kcondvar_t spa_namespace_cv; static int spa_active_count; int spa_max_replication_override = SPA_DVAS_PER_BP; static kmutex_t spa_spare_lock; static avl_tree_t spa_spare_avl; static kmutex_t spa_l2cache_lock; static avl_tree_t spa_l2cache_avl; kmem_cache_t *spa_buffer_pool; int spa_mode_global; #ifdef ZFS_DEBUG /* Everything except dprintf and spa is on by default in debug builds */ int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA); #else int zfs_flags = 0; #endif /* * zfs_recover can be set to nonzero to attempt to recover from * otherwise-fatal errors, typically caused by on-disk corruption. When * set, calls to zfs_panic_recover() will turn into warning messages. * This should only be used as a last resort, as it typically results * in leaked space, or worse. */ boolean_t zfs_recover = B_FALSE; /* * If destroy encounters an 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 EIO, * permanently leak the space from indirect blocks that can not be read, * and continue to free everything else that it can. * * The default, "stalling" behavior is useful if the storage partially * fails (i.e. some but not all i/os 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. However, note that this case is actually * fairly rare. * * Typically pools either (a) fail completely (but perhaps temporarily, * e.g. a top-level vdev going offline), or (b) have localized, * permanent errors (e.g. disk returns the wrong data due to bit flip or * firmware bug). In case (a), 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 case (b), 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. Therefore, it is 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. */ boolean_t zfs_free_leak_on_eio = B_FALSE; /* * Expiration time in milliseconds. This value has two meanings. First it is * used to determine when the spa_deadman() logic should fire. By default the * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds. * Secondly, the value determines if an I/O is considered "hung". Any I/O that * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting * in a system panic. */ uint64_t zfs_deadman_synctime_ms = 1000000ULL; /* * Check time in milliseconds. This defines the frequency at which we check * for hung I/O. */ uint64_t zfs_deadman_checktime_ms = 5000ULL; /* * Override the zfs deadman behavior via /etc/system. By default the * deadman is enabled except on VMware and sparc deployments. */ int zfs_deadman_enabled = -1; /* * The worst case is single-sector max-parity RAID-Z blocks, in which * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1) * times the size; so just assume that. Add to this the fact that * we can have up to 3 DVAs per bp, and one more factor of 2 because * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together, * the worst case is: * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24 */ int spa_asize_inflation = 24; /* * Normally, we don't allow the last 3.2% (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 ENOSPC. * * Certain operations (e.g. file removal, most administrative actions) can * use half the slop space. They will only return ENOSPC if less than half * the slop space is free. Typically, once the pool has less than the slop * space free, the user will use these operations to free up space in the pool. * These are the operations that call dsl_pool_adjustedsize() with the netfree * argument set to TRUE. * * A very restricted set of operations are always permitted, regardless of * the amount of free space. These are the operations that call * dsl_sync_task(ZFS_SPACE_CHECK_NONE), e.g. "zfs destroy". If these * operations result in a net increase in the amount of space used, * it is possible to run the pool completely out of space, causing it to * be permanently read-only. * * Note that on very small pools, the slop space will be larger than * 3.2%, in an effort to have it be at least spa_min_slop (128MB), * but we never allow it to be more than half the pool size. * * See also the comments in zfs_space_check_t. */ int spa_slop_shift = 5; uint64_t spa_min_slop = 128 * 1024 * 1024; /* * ========================================================================== * SPA config locking * ========================================================================== */ static void spa_config_lock_init(spa_t *spa) { for (int i = 0; i < SCL_LOCKS; i++) { spa_config_lock_t *scl = &spa->spa_config_lock[i]; mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL); refcount_create_untracked(&scl->scl_count); scl->scl_writer = NULL; scl->scl_write_wanted = 0; } } static void spa_config_lock_destroy(spa_t *spa) { for (int i = 0; i < SCL_LOCKS; i++) { spa_config_lock_t *scl = &spa->spa_config_lock[i]; mutex_destroy(&scl->scl_lock); cv_destroy(&scl->scl_cv); refcount_destroy(&scl->scl_count); ASSERT(scl->scl_writer == NULL); ASSERT(scl->scl_write_wanted == 0); } } int spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw) { for (int i = 0; i < SCL_LOCKS; i++) { spa_config_lock_t *scl = &spa->spa_config_lock[i]; if (!(locks & (1 << i))) continue; mutex_enter(&scl->scl_lock); if (rw == RW_READER) { if (scl->scl_writer || scl->scl_write_wanted) { mutex_exit(&scl->scl_lock); spa_config_exit(spa, locks & ((1 << i) - 1), tag); return (0); } } else { ASSERT(scl->scl_writer != curthread); if (!refcount_is_zero(&scl->scl_count)) { mutex_exit(&scl->scl_lock); spa_config_exit(spa, locks & ((1 << i) - 1), tag); return (0); } scl->scl_writer = curthread; } (void) refcount_add(&scl->scl_count, tag); mutex_exit(&scl->scl_lock); } return (1); } void spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw) { int wlocks_held = 0; ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY); for (int i = 0; i < SCL_LOCKS; i++) { spa_config_lock_t *scl = &spa->spa_config_lock[i]; if (scl->scl_writer == curthread) wlocks_held |= (1 << i); if (!(locks & (1 << i))) continue; mutex_enter(&scl->scl_lock); if (rw == RW_READER) { while (scl->scl_writer || scl->scl_write_wanted) { cv_wait(&scl->scl_cv, &scl->scl_lock); } } else { ASSERT(scl->scl_writer != curthread); while (!refcount_is_zero(&scl->scl_count)) { scl->scl_write_wanted++; cv_wait(&scl->scl_cv, &scl->scl_lock); scl->scl_write_wanted--; } scl->scl_writer = curthread; } (void) refcount_add(&scl->scl_count, tag); mutex_exit(&scl->scl_lock); } ASSERT(wlocks_held <= locks); } void spa_config_exit(spa_t *spa, int locks, void *tag) { for (int i = SCL_LOCKS - 1; i >= 0; i--) { spa_config_lock_t *scl = &spa->spa_config_lock[i]; if (!(locks & (1 << i))) continue; mutex_enter(&scl->scl_lock); ASSERT(!refcount_is_zero(&scl->scl_count)); if (refcount_remove(&scl->scl_count, tag) == 0) { ASSERT(scl->scl_writer == NULL || scl->scl_writer == curthread); scl->scl_writer = NULL; /* OK in either case */ cv_broadcast(&scl->scl_cv); } mutex_exit(&scl->scl_lock); } } int spa_config_held(spa_t *spa, int locks, krw_t rw) { int locks_held = 0; for (int i = 0; i < SCL_LOCKS; i++) { spa_config_lock_t *scl = &spa->spa_config_lock[i]; if (!(locks & (1 << i))) continue; if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) || (rw == RW_WRITER && scl->scl_writer == curthread)) locks_held |= 1 << i; } return (locks_held); } /* * ========================================================================== * SPA namespace functions * ========================================================================== */ /* * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held. * Returns NULL if no matching spa_t is found. */ spa_t * spa_lookup(const char *name) { static spa_t search; /* spa_t is large; don't allocate on stack */ spa_t *spa; avl_index_t where; char *cp; ASSERT(MUTEX_HELD(&spa_namespace_lock)); (void) strlcpy(search.spa_name, name, sizeof (search.spa_name)); /* * If it's a full dataset name, figure out the pool name and * just use that. */ cp = strpbrk(search.spa_name, "/@#"); if (cp != NULL) *cp = '\0'; spa = avl_find(&spa_namespace_avl, &search, &where); return (spa); } /* * Fires when spa_sync has not completed within zfs_deadman_synctime_ms. * If the zfs_deadman_enabled flag is set then it inspects all vdev queues * looking for potentially hung I/Os. */ void spa_deadman(void *arg) { spa_t *spa = arg; /* * Disable the deadman timer if the pool is suspended. */ if (spa_suspended(spa)) { VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); return; } zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu", (gethrtime() - spa->spa_sync_starttime) / NANOSEC, ++spa->spa_deadman_calls); if (zfs_deadman_enabled) vdev_deadman(spa->spa_root_vdev); } /* * Create an uninitialized spa_t with the given name. Requires * spa_namespace_lock. The caller must ensure that the spa_t doesn't already * exist by calling spa_lookup() first. */ spa_t * spa_add(const char *name, nvlist_t *config, const char *altroot) { spa_t *spa; spa_config_dirent_t *dp; cyc_handler_t hdlr; cyc_time_t when; ASSERT(MUTEX_HELD(&spa_namespace_lock)); spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP); mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_evicting_os_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_cksum_tmpls_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_iokstat_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_alloc_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL); cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL); cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL); cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL); cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL); for (int t = 0; t < TXG_SIZE; t++) bplist_create(&spa->spa_free_bplist[t]); (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name)); spa->spa_state = POOL_STATE_UNINITIALIZED; spa->spa_freeze_txg = UINT64_MAX; spa->spa_final_txg = UINT64_MAX; spa->spa_load_max_txg = UINT64_MAX; spa->spa_proc = &p0; spa->spa_proc_state = SPA_PROC_NONE; hdlr.cyh_func = spa_deadman; hdlr.cyh_arg = spa; hdlr.cyh_level = CY_LOW_LEVEL; spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms); /* * This determines how often we need to check for hung I/Os after * the cyclic has already fired. Since checking for hung I/Os is * an expensive operation we don't want to check too frequently. * Instead wait for 5 seconds before checking again. */ when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms); when.cyt_when = CY_INFINITY; mutex_enter(&cpu_lock); spa->spa_deadman_cycid = cyclic_add(&hdlr, &when); mutex_exit(&cpu_lock); refcount_create(&spa->spa_refcount); spa_config_lock_init(spa); avl_add(&spa_namespace_avl, spa); /* * Set the alternate root, if there is one. */ if (altroot) { spa->spa_root = spa_strdup(altroot); spa_active_count++; } - avl_create(&spa->spa_alloc_tree, zio_timestamp_compare, + avl_create(&spa->spa_alloc_tree, zio_bookmark_compare, sizeof (zio_t), offsetof(zio_t, io_alloc_node)); /* * Every pool starts with the default cachefile */ list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t), offsetof(spa_config_dirent_t, scd_link)); dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP); dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path); list_insert_head(&spa->spa_config_list, dp); VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME, KM_SLEEP) == 0); if (config != NULL) { nvlist_t *features; if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ, &features) == 0) { VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); } VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0); } if (spa->spa_label_features == NULL) { VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME, KM_SLEEP) == 0); } spa->spa_iokstat = kstat_create("zfs", 0, name, "disk", KSTAT_TYPE_IO, 1, 0); if (spa->spa_iokstat) { spa->spa_iokstat->ks_lock = &spa->spa_iokstat_lock; kstat_install(spa->spa_iokstat); } spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0); spa->spa_min_ashift = INT_MAX; spa->spa_max_ashift = 0; /* * As a pool is being created, treat all features as disabled by * setting SPA_FEATURE_DISABLED for all entries in the feature * refcount cache. */ for (int i = 0; i < SPA_FEATURES; i++) { spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED; } return (spa); } /* * Removes a spa_t from the namespace, freeing up any memory used. Requires * spa_namespace_lock. This is called only after the spa_t has been closed and * deactivated. */ void spa_remove(spa_t *spa) { spa_config_dirent_t *dp; ASSERT(MUTEX_HELD(&spa_namespace_lock)); ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); ASSERT3U(refcount_count(&spa->spa_refcount), ==, 0); nvlist_free(spa->spa_config_splitting); avl_remove(&spa_namespace_avl, spa); cv_broadcast(&spa_namespace_cv); if (spa->spa_root) { spa_strfree(spa->spa_root); spa_active_count--; } while ((dp = list_head(&spa->spa_config_list)) != NULL) { list_remove(&spa->spa_config_list, dp); if (dp->scd_path != NULL) spa_strfree(dp->scd_path); kmem_free(dp, sizeof (spa_config_dirent_t)); } avl_destroy(&spa->spa_alloc_tree); list_destroy(&spa->spa_config_list); nvlist_free(spa->spa_label_features); nvlist_free(spa->spa_load_info); spa_config_set(spa, NULL); mutex_enter(&cpu_lock); if (spa->spa_deadman_cycid != CYCLIC_NONE) cyclic_remove(spa->spa_deadman_cycid); mutex_exit(&cpu_lock); spa->spa_deadman_cycid = CYCLIC_NONE; refcount_destroy(&spa->spa_refcount); spa_config_lock_destroy(spa); kstat_delete(spa->spa_iokstat); spa->spa_iokstat = NULL; for (int t = 0; t < TXG_SIZE; t++) bplist_destroy(&spa->spa_free_bplist[t]); zio_checksum_templates_free(spa); cv_destroy(&spa->spa_async_cv); cv_destroy(&spa->spa_evicting_os_cv); cv_destroy(&spa->spa_proc_cv); cv_destroy(&spa->spa_scrub_io_cv); cv_destroy(&spa->spa_suspend_cv); mutex_destroy(&spa->spa_alloc_lock); mutex_destroy(&spa->spa_async_lock); mutex_destroy(&spa->spa_errlist_lock); mutex_destroy(&spa->spa_errlog_lock); mutex_destroy(&spa->spa_evicting_os_lock); mutex_destroy(&spa->spa_history_lock); mutex_destroy(&spa->spa_proc_lock); mutex_destroy(&spa->spa_props_lock); mutex_destroy(&spa->spa_cksum_tmpls_lock); mutex_destroy(&spa->spa_scrub_lock); mutex_destroy(&spa->spa_suspend_lock); mutex_destroy(&spa->spa_vdev_top_lock); mutex_destroy(&spa->spa_iokstat_lock); kmem_free(spa, sizeof (spa_t)); } /* * Given a pool, return the next pool in the namespace, or NULL if there is * none. If 'prev' is NULL, return the first pool. */ spa_t * spa_next(spa_t *prev) { ASSERT(MUTEX_HELD(&spa_namespace_lock)); if (prev) return (AVL_NEXT(&spa_namespace_avl, prev)); else return (avl_first(&spa_namespace_avl)); } /* * ========================================================================== * SPA refcount functions * ========================================================================== */ /* * Add a reference to the given spa_t. Must have at least one reference, or * have the namespace lock held. */ void spa_open_ref(spa_t *spa, void *tag) { ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref || MUTEX_HELD(&spa_namespace_lock)); (void) refcount_add(&spa->spa_refcount, tag); } /* * Remove a reference to the given spa_t. Must have at least one reference, or * have the namespace lock held. */ void spa_close(spa_t *spa, void *tag) { ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref || MUTEX_HELD(&spa_namespace_lock)); (void) refcount_remove(&spa->spa_refcount, tag); } /* * Remove a reference to the given spa_t held by a dsl dir that is * being asynchronously released. Async releases occur from a taskq * performing eviction of dsl datasets and dirs. The namespace lock * isn't held and the hold by the object being evicted may contribute to * spa_minref (e.g. dataset or directory released during pool export), * so the asserts in spa_close() do not apply. */ void spa_async_close(spa_t *spa, void *tag) { (void) refcount_remove(&spa->spa_refcount, tag); } /* * Check to see if the spa refcount is zero. Must be called with * spa_namespace_lock held. We really compare against spa_minref, which is the * number of references acquired when opening a pool */ boolean_t spa_refcount_zero(spa_t *spa) { ASSERT(MUTEX_HELD(&spa_namespace_lock)); return (refcount_count(&spa->spa_refcount) == spa->spa_minref); } /* * ========================================================================== * SPA spare and l2cache tracking * ========================================================================== */ /* * Hot spares and cache devices are tracked using the same code below, * for 'auxiliary' devices. */ typedef struct spa_aux { uint64_t aux_guid; uint64_t aux_pool; avl_node_t aux_avl; int aux_count; } spa_aux_t; static int spa_aux_compare(const void *a, const void *b) { const spa_aux_t *sa = a; const spa_aux_t *sb = b; if (sa->aux_guid < sb->aux_guid) return (-1); else if (sa->aux_guid > sb->aux_guid) return (1); else return (0); } void spa_aux_add(vdev_t *vd, avl_tree_t *avl) { avl_index_t where; spa_aux_t search; spa_aux_t *aux; search.aux_guid = vd->vdev_guid; if ((aux = avl_find(avl, &search, &where)) != NULL) { aux->aux_count++; } else { aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP); aux->aux_guid = vd->vdev_guid; aux->aux_count = 1; avl_insert(avl, aux, where); } } void spa_aux_remove(vdev_t *vd, avl_tree_t *avl) { spa_aux_t search; spa_aux_t *aux; avl_index_t where; search.aux_guid = vd->vdev_guid; aux = avl_find(avl, &search, &where); ASSERT(aux != NULL); if (--aux->aux_count == 0) { avl_remove(avl, aux); kmem_free(aux, sizeof (spa_aux_t)); } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) { aux->aux_pool = 0ULL; } } boolean_t spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl) { spa_aux_t search, *found; search.aux_guid = guid; found = avl_find(avl, &search, NULL); if (pool) { if (found) *pool = found->aux_pool; else *pool = 0ULL; } if (refcnt) { if (found) *refcnt = found->aux_count; else *refcnt = 0; } return (found != NULL); } void spa_aux_activate(vdev_t *vd, avl_tree_t *avl) { spa_aux_t search, *found; avl_index_t where; search.aux_guid = vd->vdev_guid; found = avl_find(avl, &search, &where); ASSERT(found != NULL); ASSERT(found->aux_pool == 0ULL); found->aux_pool = spa_guid(vd->vdev_spa); } /* * Spares are tracked globally due to the following constraints: * * - A spare may be part of multiple pools. * - A spare may be added to a pool even if it's actively in use within * another pool. * - A spare in use in any pool can only be the source of a replacement if * the target is a spare in the same pool. * * We keep track of all spares on the system through the use of a reference * counted AVL tree. When a vdev is added as a spare, or used as a replacement * spare, then we bump the reference count in the AVL tree. In addition, we set * the 'vdev_isspare' member to indicate that the device is a spare (active or * inactive). When a spare is made active (used to replace a device in the * pool), we also keep track of which pool its been made a part of. * * The 'spa_spare_lock' protects the AVL tree. These functions are normally * called under the spa_namespace lock as part of vdev reconfiguration. The * separate spare lock exists for the status query path, which does not need to * be completely consistent with respect to other vdev configuration changes. */ static int spa_spare_compare(const void *a, const void *b) { return (spa_aux_compare(a, b)); } void spa_spare_add(vdev_t *vd) { mutex_enter(&spa_spare_lock); ASSERT(!vd->vdev_isspare); spa_aux_add(vd, &spa_spare_avl); vd->vdev_isspare = B_TRUE; mutex_exit(&spa_spare_lock); } void spa_spare_remove(vdev_t *vd) { mutex_enter(&spa_spare_lock); ASSERT(vd->vdev_isspare); spa_aux_remove(vd, &spa_spare_avl); vd->vdev_isspare = B_FALSE; mutex_exit(&spa_spare_lock); } boolean_t spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt) { boolean_t found; mutex_enter(&spa_spare_lock); found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl); mutex_exit(&spa_spare_lock); return (found); } void spa_spare_activate(vdev_t *vd) { mutex_enter(&spa_spare_lock); ASSERT(vd->vdev_isspare); spa_aux_activate(vd, &spa_spare_avl); mutex_exit(&spa_spare_lock); } /* * Level 2 ARC devices are tracked globally for the same reasons as spares. * Cache devices currently only support one pool per cache device, and so * for these devices the aux reference count is currently unused beyond 1. */ static int spa_l2cache_compare(const void *a, const void *b) { return (spa_aux_compare(a, b)); } void spa_l2cache_add(vdev_t *vd) { mutex_enter(&spa_l2cache_lock); ASSERT(!vd->vdev_isl2cache); spa_aux_add(vd, &spa_l2cache_avl); vd->vdev_isl2cache = B_TRUE; mutex_exit(&spa_l2cache_lock); } void spa_l2cache_remove(vdev_t *vd) { mutex_enter(&spa_l2cache_lock); ASSERT(vd->vdev_isl2cache); spa_aux_remove(vd, &spa_l2cache_avl); vd->vdev_isl2cache = B_FALSE; mutex_exit(&spa_l2cache_lock); } boolean_t spa_l2cache_exists(uint64_t guid, uint64_t *pool) { boolean_t found; mutex_enter(&spa_l2cache_lock); found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl); mutex_exit(&spa_l2cache_lock); return (found); } void spa_l2cache_activate(vdev_t *vd) { mutex_enter(&spa_l2cache_lock); ASSERT(vd->vdev_isl2cache); spa_aux_activate(vd, &spa_l2cache_avl); mutex_exit(&spa_l2cache_lock); } /* * ========================================================================== * SPA vdev locking * ========================================================================== */ /* * Lock the given spa_t for the purpose of adding or removing a vdev. * Grabs the global spa_namespace_lock plus the spa config lock for writing. * It returns the next transaction group for the spa_t. */ uint64_t spa_vdev_enter(spa_t *spa) { mutex_enter(&spa->spa_vdev_top_lock); mutex_enter(&spa_namespace_lock); return (spa_vdev_config_enter(spa)); } /* * Internal implementation for spa_vdev_enter(). Used when a vdev * operation requires multiple syncs (i.e. removing a device) while * keeping the spa_namespace_lock held. */ uint64_t spa_vdev_config_enter(spa_t *spa) { ASSERT(MUTEX_HELD(&spa_namespace_lock)); spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); return (spa_last_synced_txg(spa) + 1); } /* * Used in combination with spa_vdev_config_enter() to allow the syncing * of multiple transactions without releasing the spa_namespace_lock. */ void spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag) { ASSERT(MUTEX_HELD(&spa_namespace_lock)); int config_changed = B_FALSE; ASSERT(txg > spa_last_synced_txg(spa)); spa->spa_pending_vdev = NULL; /* * Reassess the DTLs. */ vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE); if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) { config_changed = B_TRUE; spa->spa_config_generation++; } /* * Verify the metaslab classes. */ ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0); ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0); spa_config_exit(spa, SCL_ALL, spa); /* * Panic the system if the specified tag requires it. This * is useful for ensuring that configurations are updated * transactionally. */ if (zio_injection_enabled) zio_handle_panic_injection(spa, tag, 0); /* * Note: this txg_wait_synced() is important because it ensures * that there won't be more than one config change per txg. * This allows us to use the txg as the generation number. */ if (error == 0) txg_wait_synced(spa->spa_dsl_pool, txg); if (vd != NULL) { ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL); spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); vdev_free(vd); spa_config_exit(spa, SCL_ALL, spa); } /* * If the config changed, update the config cache. */ if (config_changed) spa_config_sync(spa, B_FALSE, B_TRUE); } /* * Unlock the spa_t after adding or removing a vdev. Besides undoing the * locking of spa_vdev_enter(), we also want make sure the transactions have * synced to disk, and then update the global configuration cache with the new * information. */ int spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error) { spa_vdev_config_exit(spa, vd, txg, error, FTAG); mutex_exit(&spa_namespace_lock); mutex_exit(&spa->spa_vdev_top_lock); return (error); } /* * Lock the given spa_t for the purpose of changing vdev state. */ void spa_vdev_state_enter(spa_t *spa, int oplocks) { int locks = SCL_STATE_ALL | oplocks; /* * Root pools may need to read of the underlying devfs filesystem * when opening up a vdev. Unfortunately if we're holding the * SCL_ZIO lock it will result in a deadlock when we try to issue * the read from the root filesystem. Instead we "prefetch" * the associated vnodes that we need prior to opening the * underlying devices and cache them so that we can prevent * any I/O when we are doing the actual open. */ if (spa_is_root(spa)) { int low = locks & ~(SCL_ZIO - 1); int high = locks & ~low; spa_config_enter(spa, high, spa, RW_WRITER); vdev_hold(spa->spa_root_vdev); spa_config_enter(spa, low, spa, RW_WRITER); } else { spa_config_enter(spa, locks, spa, RW_WRITER); } spa->spa_vdev_locks = locks; } int spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error) { boolean_t config_changed = B_FALSE; if (vd != NULL || error == 0) vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev, 0, 0, B_FALSE); if (vd != NULL) { vdev_state_dirty(vd->vdev_top); config_changed = B_TRUE; spa->spa_config_generation++; } if (spa_is_root(spa)) vdev_rele(spa->spa_root_vdev); ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL); spa_config_exit(spa, spa->spa_vdev_locks, spa); /* * If anything changed, wait for it to sync. This ensures that, * from the system administrator's perspective, zpool(1M) commands * are synchronous. This is important for things like zpool offline: * when the command completes, you expect no further I/O from ZFS. */ if (vd != NULL) txg_wait_synced(spa->spa_dsl_pool, 0); /* * If the config changed, update the config cache. */ if (config_changed) { mutex_enter(&spa_namespace_lock); spa_config_sync(spa, B_FALSE, B_TRUE); mutex_exit(&spa_namespace_lock); } return (error); } /* * ========================================================================== * Miscellaneous functions * ========================================================================== */ void spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx) { if (!nvlist_exists(spa->spa_label_features, feature)) { fnvlist_add_boolean(spa->spa_label_features, feature); /* * When we are creating the pool (tx_txg==TXG_INITIAL), we can't * dirty the vdev config because lock SCL_CONFIG is not held. * Thankfully, in this case we don't need to dirty the config * because it will be written out anyway when we finish * creating the pool. */ if (tx->tx_txg != TXG_INITIAL) vdev_config_dirty(spa->spa_root_vdev); } } void spa_deactivate_mos_feature(spa_t *spa, const char *feature) { if (nvlist_remove_all(spa->spa_label_features, feature) == 0) vdev_config_dirty(spa->spa_root_vdev); } /* * Rename a spa_t. */ int spa_rename(const char *name, const char *newname) { spa_t *spa; int err; /* * Lookup the spa_t and grab the config lock for writing. We need to * actually open the pool so that we can sync out the necessary labels. * It's OK to call spa_open() with the namespace lock held because we * allow recursive calls for other reasons. */ mutex_enter(&spa_namespace_lock); if ((err = spa_open(name, &spa, FTAG)) != 0) { mutex_exit(&spa_namespace_lock); return (err); } spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); avl_remove(&spa_namespace_avl, spa); (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name)); avl_add(&spa_namespace_avl, spa); /* * Sync all labels to disk with the new names by marking the root vdev * dirty and waiting for it to sync. It will pick up the new pool name * during the sync. */ vdev_config_dirty(spa->spa_root_vdev); spa_config_exit(spa, SCL_ALL, FTAG); txg_wait_synced(spa->spa_dsl_pool, 0); /* * Sync the updated config cache. */ spa_config_sync(spa, B_FALSE, B_TRUE); spa_close(spa, FTAG); mutex_exit(&spa_namespace_lock); return (0); } /* * Return the spa_t associated with given pool_guid, if it exists. If * device_guid is non-zero, determine whether the pool exists *and* contains * a device with the specified device_guid. */ spa_t * spa_by_guid(uint64_t pool_guid, uint64_t device_guid) { spa_t *spa; avl_tree_t *t = &spa_namespace_avl; ASSERT(MUTEX_HELD(&spa_namespace_lock)); for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) { if (spa->spa_state == POOL_STATE_UNINITIALIZED) continue; if (spa->spa_root_vdev == NULL) continue; if (spa_guid(spa) == pool_guid) { if (device_guid == 0) break; if (vdev_lookup_by_guid(spa->spa_root_vdev, device_guid) != NULL) break; /* * Check any devices we may be in the process of adding. */ if (spa->spa_pending_vdev) { if (vdev_lookup_by_guid(spa->spa_pending_vdev, device_guid) != NULL) break; } } } return (spa); } /* * Determine whether a pool with the given pool_guid exists. */ boolean_t spa_guid_exists(uint64_t pool_guid, uint64_t device_guid) { return (spa_by_guid(pool_guid, device_guid) != NULL); } char * spa_strdup(const char *s) { size_t len; char *new; len = strlen(s); new = kmem_alloc(len + 1, KM_SLEEP); bcopy(s, new, len); new[len] = '\0'; return (new); } void spa_strfree(char *s) { kmem_free(s, strlen(s) + 1); } uint64_t spa_get_random(uint64_t range) { uint64_t r; ASSERT(range != 0); (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t)); return (r % range); } uint64_t spa_generate_guid(spa_t *spa) { uint64_t guid = spa_get_random(-1ULL); if (spa != NULL) { while (guid == 0 || spa_guid_exists(spa_guid(spa), guid)) guid = spa_get_random(-1ULL); } else { while (guid == 0 || spa_guid_exists(guid, 0)) guid = spa_get_random(-1ULL); } return (guid); } void snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp) { char type[256]; char *checksum = NULL; char *compress = NULL; if (bp != NULL) { if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) { dmu_object_byteswap_t bswap = DMU_OT_BYTESWAP(BP_GET_TYPE(bp)); (void) snprintf(type, sizeof (type), "bswap %s %s", DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ? "metadata" : "data", dmu_ot_byteswap[bswap].ob_name); } else { (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name, sizeof (type)); } if (!BP_IS_EMBEDDED(bp)) { checksum = zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name; } compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name; } SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum, compress); } void spa_freeze(spa_t *spa) { uint64_t freeze_txg = 0; spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); if (spa->spa_freeze_txg == UINT64_MAX) { freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE; spa->spa_freeze_txg = freeze_txg; } spa_config_exit(spa, SCL_ALL, FTAG); if (freeze_txg != 0) txg_wait_synced(spa_get_dsl(spa), freeze_txg); } void zfs_panic_recover(const char *fmt, ...) { va_list adx; va_start(adx, fmt); vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx); va_end(adx); } /* * This is a stripped-down version of strtoull, suitable only for converting * lowercase hexadecimal numbers that don't overflow. */ uint64_t strtonum(const char *str, char **nptr) { uint64_t val = 0; char c; int digit; while ((c = *str) != '\0') { if (c >= '0' && c <= '9') digit = c - '0'; else if (c >= 'a' && c <= 'f') digit = 10 + c - 'a'; else break; val *= 16; val += digit; str++; } if (nptr) *nptr = (char *)str; return (val); } /* * ========================================================================== * Accessor functions * ========================================================================== */ boolean_t spa_shutting_down(spa_t *spa) { return (spa->spa_async_suspended); } dsl_pool_t * spa_get_dsl(spa_t *spa) { return (spa->spa_dsl_pool); } boolean_t spa_is_initializing(spa_t *spa) { return (spa->spa_is_initializing); } blkptr_t * spa_get_rootblkptr(spa_t *spa) { return (&spa->spa_ubsync.ub_rootbp); } void spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp) { spa->spa_uberblock.ub_rootbp = *bp; } void spa_altroot(spa_t *spa, char *buf, size_t buflen) { if (spa->spa_root == NULL) buf[0] = '\0'; else (void) strncpy(buf, spa->spa_root, buflen); } int spa_sync_pass(spa_t *spa) { return (spa->spa_sync_pass); } char * spa_name(spa_t *spa) { return (spa->spa_name); } uint64_t spa_guid(spa_t *spa) { dsl_pool_t *dp = spa_get_dsl(spa); uint64_t guid; /* * If we fail to parse the config during spa_load(), we can go through * the error path (which posts an ereport) and end up here with no root * vdev. We stash the original pool guid in 'spa_config_guid' to handle * this case. */ if (spa->spa_root_vdev == NULL) return (spa->spa_config_guid); guid = spa->spa_last_synced_guid != 0 ? spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid; /* * Return the most recently synced out guid unless we're * in syncing context. */ if (dp && dsl_pool_sync_context(dp)) return (spa->spa_root_vdev->vdev_guid); else return (guid); } uint64_t spa_load_guid(spa_t *spa) { /* * This is a GUID that exists solely as a reference for the * purposes of the arc. It is generated at load time, and * is never written to persistent storage. */ return (spa->spa_load_guid); } uint64_t spa_last_synced_txg(spa_t *spa) { return (spa->spa_ubsync.ub_txg); } uint64_t spa_first_txg(spa_t *spa) { return (spa->spa_first_txg); } uint64_t spa_syncing_txg(spa_t *spa) { return (spa->spa_syncing_txg); } pool_state_t spa_state(spa_t *spa) { return (spa->spa_state); } spa_load_state_t spa_load_state(spa_t *spa) { return (spa->spa_load_state); } uint64_t spa_freeze_txg(spa_t *spa) { return (spa->spa_freeze_txg); } /* ARGSUSED */ uint64_t spa_get_worst_case_asize(spa_t *spa, uint64_t lsize) { return (lsize * spa_asize_inflation); } /* * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%), * or at least 128MB, unless that would cause it to be more than half the * pool size. * * See the comment above spa_slop_shift for details. */ uint64_t spa_get_slop_space(spa_t *spa) { uint64_t space = spa_get_dspace(spa); return (MAX(space >> spa_slop_shift, MIN(space >> 1, spa_min_slop))); } uint64_t spa_get_dspace(spa_t *spa) { return (spa->spa_dspace); } void spa_update_dspace(spa_t *spa) { spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) + ddt_get_dedup_dspace(spa); } /* * Return the failure mode that has been set to this pool. The default * behavior will be to block all I/Os when a complete failure occurs. */ uint8_t spa_get_failmode(spa_t *spa) { return (spa->spa_failmode); } boolean_t spa_suspended(spa_t *spa) { return (spa->spa_suspended); } uint64_t spa_version(spa_t *spa) { return (spa->spa_ubsync.ub_version); } boolean_t spa_deflate(spa_t *spa) { return (spa->spa_deflate); } metaslab_class_t * spa_normal_class(spa_t *spa) { return (spa->spa_normal_class); } metaslab_class_t * spa_log_class(spa_t *spa) { return (spa->spa_log_class); } void spa_evicting_os_register(spa_t *spa, objset_t *os) { mutex_enter(&spa->spa_evicting_os_lock); list_insert_head(&spa->spa_evicting_os_list, os); mutex_exit(&spa->spa_evicting_os_lock); } void spa_evicting_os_deregister(spa_t *spa, objset_t *os) { mutex_enter(&spa->spa_evicting_os_lock); list_remove(&spa->spa_evicting_os_list, os); cv_broadcast(&spa->spa_evicting_os_cv); mutex_exit(&spa->spa_evicting_os_lock); } void spa_evicting_os_wait(spa_t *spa) { mutex_enter(&spa->spa_evicting_os_lock); while (!list_is_empty(&spa->spa_evicting_os_list)) cv_wait(&spa->spa_evicting_os_cv, &spa->spa_evicting_os_lock); mutex_exit(&spa->spa_evicting_os_lock); dmu_buf_user_evict_wait(); } int spa_max_replication(spa_t *spa) { /* * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to * handle BPs with more than one DVA allocated. Set our max * replication level accordingly. */ if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS) return (1); return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override)); } int spa_prev_software_version(spa_t *spa) { return (spa->spa_prev_software_version); } uint64_t spa_deadman_synctime(spa_t *spa) { return (spa->spa_deadman_synctime); } uint64_t dva_get_dsize_sync(spa_t *spa, const dva_t *dva) { uint64_t asize = DVA_GET_ASIZE(dva); uint64_t dsize = asize; ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); if (asize != 0 && spa->spa_deflate) { vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva)); dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio; } return (dsize); } uint64_t bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp) { uint64_t dsize = 0; for (int d = 0; d < BP_GET_NDVAS(bp); d++) dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); return (dsize); } uint64_t bp_get_dsize(spa_t *spa, const blkptr_t *bp) { uint64_t dsize = 0; spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); for (int d = 0; d < BP_GET_NDVAS(bp); d++) dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); spa_config_exit(spa, SCL_VDEV, FTAG); return (dsize); } /* * ========================================================================== * Initialization and Termination * ========================================================================== */ static int spa_name_compare(const void *a1, const void *a2) { const spa_t *s1 = a1; const spa_t *s2 = a2; int s; s = strcmp(s1->spa_name, s2->spa_name); if (s > 0) return (1); if (s < 0) return (-1); return (0); } int spa_busy(void) { return (spa_active_count); } void spa_boot_init() { spa_config_load(); } void spa_init(int mode) { mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL); avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t), offsetof(spa_t, spa_avl)); avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t), offsetof(spa_aux_t, aux_avl)); avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t), offsetof(spa_aux_t, aux_avl)); spa_mode_global = mode; #ifdef _KERNEL spa_arch_init(); #else if (spa_mode_global != FREAD && dprintf_find_string("watch")) { arc_procfd = open("/proc/self/ctl", O_WRONLY); if (arc_procfd == -1) { perror("could not enable watchpoints: " "opening /proc/self/ctl failed: "); } else { arc_watch = B_TRUE; } } #endif refcount_init(); unique_init(); range_tree_init(); metaslab_alloc_trace_init(); zio_init(); dmu_init(); zil_init(); vdev_cache_stat_init(); zfs_prop_init(); zpool_prop_init(); zpool_feature_init(); spa_config_load(); l2arc_start(); } void spa_fini(void) { l2arc_stop(); spa_evict_all(); vdev_cache_stat_fini(); zil_fini(); dmu_fini(); zio_fini(); metaslab_alloc_trace_fini(); range_tree_fini(); unique_fini(); refcount_fini(); avl_destroy(&spa_namespace_avl); avl_destroy(&spa_spare_avl); avl_destroy(&spa_l2cache_avl); cv_destroy(&spa_namespace_cv); mutex_destroy(&spa_namespace_lock); mutex_destroy(&spa_spare_lock); mutex_destroy(&spa_l2cache_lock); } /* * Return whether this pool has slogs. No locking needed. * It's not a problem if the wrong answer is returned as it's only for * performance and not correctness */ boolean_t spa_has_slogs(spa_t *spa) { return (spa->spa_log_class->mc_rotor != NULL); } spa_log_state_t spa_get_log_state(spa_t *spa) { return (spa->spa_log_state); } void spa_set_log_state(spa_t *spa, spa_log_state_t state) { spa->spa_log_state = state; } boolean_t spa_is_root(spa_t *spa) { return (spa->spa_is_root); } boolean_t spa_writeable(spa_t *spa) { return (!!(spa->spa_mode & FWRITE)); } /* * Returns true if there is a pending sync task in any of the current * syncing txg, the current quiescing txg, or the current open txg. */ boolean_t spa_has_pending_synctask(spa_t *spa) { return (!txg_all_lists_empty(&spa->spa_dsl_pool->dp_sync_tasks)); } int spa_mode(spa_t *spa) { return (spa->spa_mode); } uint64_t spa_bootfs(spa_t *spa) { return (spa->spa_bootfs); } uint64_t spa_delegation(spa_t *spa) { return (spa->spa_delegation); } objset_t * spa_meta_objset(spa_t *spa) { return (spa->spa_meta_objset); } enum zio_checksum spa_dedup_checksum(spa_t *spa) { return (spa->spa_dedup_checksum); } /* * Reset pool scan stat per scan pass (or reboot). */ void spa_scan_stat_init(spa_t *spa) { /* data not stored on disk */ spa->spa_scan_pass_start = gethrestime_sec(); spa->spa_scan_pass_exam = 0; vdev_scan_stat_init(spa->spa_root_vdev); } /* * Get scan stats for zpool status reports */ int spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps) { dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL; if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE) return (SET_ERROR(ENOENT)); bzero(ps, sizeof (pool_scan_stat_t)); /* data stored on disk */ ps->pss_func = scn->scn_phys.scn_func; ps->pss_start_time = scn->scn_phys.scn_start_time; ps->pss_end_time = scn->scn_phys.scn_end_time; ps->pss_to_examine = scn->scn_phys.scn_to_examine; ps->pss_examined = scn->scn_phys.scn_examined; ps->pss_to_process = scn->scn_phys.scn_to_process; ps->pss_processed = scn->scn_phys.scn_processed; ps->pss_errors = scn->scn_phys.scn_errors; ps->pss_state = scn->scn_phys.scn_state; /* data not stored on disk */ ps->pss_pass_start = spa->spa_scan_pass_start; ps->pss_pass_exam = spa->spa_scan_pass_exam; return (0); } boolean_t spa_debug_enabled(spa_t *spa) { return (spa->spa_debug); } int spa_maxblocksize(spa_t *spa) { if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) return (SPA_MAXBLOCKSIZE); else return (SPA_OLD_MAXBLOCKSIZE); } Index: vendor-sys/illumos/dist/uts/common/fs/zfs/sys/dmu_objset.h =================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/sys/dmu_objset.h (revision 316916) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/sys/dmu_objset.h (revision 316917) @@ -1,196 +1,198 @@ /* * 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 http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. - * Copyright (c) 2012, 2016 by Delphix. All rights reserved. + * Copyright (c) 2012, 2017 by Delphix. All rights reserved. * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. * Copyright (c) 2014 Integros [integros.com] */ /* Portions Copyright 2010 Robert Milkowski */ #ifndef _SYS_DMU_OBJSET_H #define _SYS_DMU_OBJSET_H #include #include #include #include #include #include #include #include #ifdef __cplusplus extern "C" { #endif extern krwlock_t os_lock; struct dsl_pool; struct dsl_dataset; struct dmu_tx; #define OBJSET_PHYS_SIZE 2048 #define OBJSET_OLD_PHYS_SIZE 1024 #define OBJSET_BUF_HAS_USERUSED(buf) \ (arc_buf_size(buf) > OBJSET_OLD_PHYS_SIZE) #define OBJSET_FLAG_USERACCOUNTING_COMPLETE (1ULL<<0) typedef struct objset_phys { dnode_phys_t os_meta_dnode; zil_header_t os_zil_header; uint64_t os_type; uint64_t os_flags; char os_pad[OBJSET_PHYS_SIZE - sizeof (dnode_phys_t)*3 - sizeof (zil_header_t) - sizeof (uint64_t)*2]; dnode_phys_t os_userused_dnode; dnode_phys_t os_groupused_dnode; } objset_phys_t; struct objset { /* Immutable: */ struct dsl_dataset *os_dsl_dataset; spa_t *os_spa; arc_buf_t *os_phys_buf; objset_phys_t *os_phys; /* * The following "special" dnodes have no parent, are exempt * from dnode_move(), and are not recorded in os_dnodes, but they * root their descendents in this objset using handles anyway, so * that all access to dnodes from dbufs consistently uses handles. */ dnode_handle_t os_meta_dnode; dnode_handle_t os_userused_dnode; dnode_handle_t os_groupused_dnode; zilog_t *os_zil; list_node_t os_evicting_node; /* can change, under dsl_dir's locks: */ enum zio_checksum os_checksum; enum zio_compress os_compress; uint8_t os_copies; enum zio_checksum os_dedup_checksum; boolean_t os_dedup_verify; zfs_logbias_op_t os_logbias; zfs_cache_type_t os_primary_cache; zfs_cache_type_t os_secondary_cache; zfs_sync_type_t os_sync; zfs_redundant_metadata_type_t os_redundant_metadata; int os_recordsize; /* * Pointer is constant; the blkptr it points to is protected by * os_dsl_dataset->ds_bp_rwlock */ blkptr_t *os_rootbp; /* no lock needed: */ struct dmu_tx *os_synctx; /* XXX sketchy */ zil_header_t os_zil_header; - list_t os_synced_dnodes; + multilist_t *os_synced_dnodes; uint64_t os_flags; uint64_t os_freed_dnodes; boolean_t os_rescan_dnodes; /* Protected by os_obj_lock */ kmutex_t os_obj_lock; uint64_t os_obj_next; /* Protected by os_lock */ kmutex_t os_lock; - list_t os_dirty_dnodes[TXG_SIZE]; - list_t os_free_dnodes[TXG_SIZE]; + multilist_t *os_dirty_dnodes[TXG_SIZE]; list_t os_dnodes; list_t os_downgraded_dbufs; + + /* Protects changes to DMU_{USER,GROUP}USED_OBJECT */ + kmutex_t os_userused_lock; /* stuff we store for the user */ kmutex_t os_user_ptr_lock; void *os_user_ptr; sa_os_t *os_sa; }; #define DMU_META_OBJSET 0 #define DMU_META_DNODE_OBJECT 0 #define DMU_OBJECT_IS_SPECIAL(obj) ((int64_t)(obj) <= 0) #define DMU_META_DNODE(os) ((os)->os_meta_dnode.dnh_dnode) #define DMU_USERUSED_DNODE(os) ((os)->os_userused_dnode.dnh_dnode) #define DMU_GROUPUSED_DNODE(os) ((os)->os_groupused_dnode.dnh_dnode) #define DMU_OS_IS_L2CACHEABLE(os) \ ((os)->os_secondary_cache == ZFS_CACHE_ALL || \ (os)->os_secondary_cache == ZFS_CACHE_METADATA) #define DMU_OS_IS_L2COMPRESSIBLE(os) (zfs_mdcomp_disable == B_FALSE) /* called from zpl */ int dmu_objset_hold(const char *name, void *tag, objset_t **osp); int dmu_objset_own(const char *name, dmu_objset_type_t type, boolean_t readonly, void *tag, objset_t **osp); int dmu_objset_own_obj(struct dsl_pool *dp, uint64_t obj, dmu_objset_type_t type, boolean_t readonly, void *tag, objset_t **osp); void dmu_objset_refresh_ownership(objset_t *os, void *tag); void dmu_objset_rele(objset_t *os, void *tag); void dmu_objset_disown(objset_t *os, void *tag); int dmu_objset_from_ds(struct dsl_dataset *ds, objset_t **osp); void dmu_objset_stats(objset_t *os, nvlist_t *nv); void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat); void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp, uint64_t *usedobjsp, uint64_t *availobjsp); uint64_t dmu_objset_fsid_guid(objset_t *os); int dmu_objset_find_dp(struct dsl_pool *dp, uint64_t ddobj, int func(struct dsl_pool *, struct dsl_dataset *, void *), void *arg, int flags); int dmu_objset_prefetch(const char *name, void *arg); void dmu_objset_evict_dbufs(objset_t *os); timestruc_t dmu_objset_snap_cmtime(objset_t *os); /* called from dsl */ void dmu_objset_sync(objset_t *os, zio_t *zio, dmu_tx_t *tx); boolean_t dmu_objset_is_dirty(objset_t *os, uint64_t txg); objset_t *dmu_objset_create_impl(spa_t *spa, struct dsl_dataset *ds, blkptr_t *bp, dmu_objset_type_t type, dmu_tx_t *tx); int dmu_objset_open_impl(spa_t *spa, struct dsl_dataset *ds, blkptr_t *bp, objset_t **osp); void dmu_objset_evict(objset_t *os); void dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx); void dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx); boolean_t dmu_objset_userused_enabled(objset_t *os); int dmu_objset_userspace_upgrade(objset_t *os); boolean_t dmu_objset_userspace_present(objset_t *os); int dmu_fsname(const char *snapname, char *buf); void dmu_objset_evict_done(objset_t *os); void dmu_objset_willuse_space(objset_t *os, int64_t space, dmu_tx_t *tx); void dmu_objset_init(void); void dmu_objset_fini(void); #ifdef __cplusplus } #endif #endif /* _SYS_DMU_OBJSET_H */ Index: vendor-sys/illumos/dist/uts/common/fs/zfs/sys/dnode.h =================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/sys/dnode.h (revision 316916) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/sys/dnode.h (revision 316917) @@ -1,361 +1,362 @@ /* * 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 http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. - * Copyright (c) 2012, 2016 by Delphix. All rights reserved. + * Copyright (c) 2012, 2017 by Delphix. All rights reserved. * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. */ #ifndef _SYS_DNODE_H #define _SYS_DNODE_H #include #include #include #include #include #include #include #include +#include #ifdef __cplusplus extern "C" { #endif /* * dnode_hold() flags. */ #define DNODE_MUST_BE_ALLOCATED 1 #define DNODE_MUST_BE_FREE 2 /* * dnode_next_offset() flags. */ #define DNODE_FIND_HOLE 1 #define DNODE_FIND_BACKWARDS 2 #define DNODE_FIND_HAVELOCK 4 /* * Fixed constants. */ #define DNODE_SHIFT 9 /* 512 bytes */ #define DN_MIN_INDBLKSHIFT 12 /* 4k */ /* * If we ever increase this value beyond 20, we need to revisit all logic that * does x << level * ebps to handle overflow. With a 1M indirect block size, * 4 levels of indirect blocks would not be able to guarantee addressing an * entire object, so 5 levels will be used, but 5 * (20 - 7) = 65. */ #define DN_MAX_INDBLKSHIFT 17 /* 128k */ #define DNODE_BLOCK_SHIFT 14 /* 16k */ #define DNODE_CORE_SIZE 64 /* 64 bytes for dnode sans blkptrs */ #define DN_MAX_OBJECT_SHIFT 48 /* 256 trillion (zfs_fid_t limit) */ #define DN_MAX_OFFSET_SHIFT 64 /* 2^64 bytes in a dnode */ /* * dnode id flags * * Note: a file will never ever have its * ids moved from bonus->spill * and only in a crypto environment would it be on spill */ #define DN_ID_CHKED_BONUS 0x1 #define DN_ID_CHKED_SPILL 0x2 #define DN_ID_OLD_EXIST 0x4 #define DN_ID_NEW_EXIST 0x8 /* * Derived constants. */ #define DNODE_SIZE (1 << DNODE_SHIFT) #define DN_MAX_NBLKPTR ((DNODE_SIZE - DNODE_CORE_SIZE) >> SPA_BLKPTRSHIFT) #define DN_MAX_BONUSLEN (DNODE_SIZE - DNODE_CORE_SIZE - (1 << SPA_BLKPTRSHIFT)) #define DN_MAX_OBJECT (1ULL << DN_MAX_OBJECT_SHIFT) #define DN_ZERO_BONUSLEN (DN_MAX_BONUSLEN + 1) #define DN_KILL_SPILLBLK (1) #define DNODES_PER_BLOCK_SHIFT (DNODE_BLOCK_SHIFT - DNODE_SHIFT) #define DNODES_PER_BLOCK (1ULL << DNODES_PER_BLOCK_SHIFT) /* * This is inaccurate if the indblkshift of the particular object is not the * max. But it's only used by userland to calculate the zvol reservation. */ #define DNODES_PER_LEVEL_SHIFT (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT) #define DNODES_PER_LEVEL (1ULL << DNODES_PER_LEVEL_SHIFT) /* The +2 here is a cheesy way to round up */ #define DN_MAX_LEVELS (2 + ((DN_MAX_OFFSET_SHIFT - SPA_MINBLOCKSHIFT) / \ (DN_MIN_INDBLKSHIFT - SPA_BLKPTRSHIFT))) #define DN_BONUS(dnp) ((void*)((dnp)->dn_bonus + \ (((dnp)->dn_nblkptr - 1) * sizeof (blkptr_t)))) #define DN_USED_BYTES(dnp) (((dnp)->dn_flags & DNODE_FLAG_USED_BYTES) ? \ (dnp)->dn_used : (dnp)->dn_used << SPA_MINBLOCKSHIFT) #define EPB(blkshift, typeshift) (1 << (blkshift - typeshift)) struct dmu_buf_impl; struct objset; struct zio; enum dnode_dirtycontext { DN_UNDIRTIED, DN_DIRTY_OPEN, DN_DIRTY_SYNC }; /* Is dn_used in bytes? if not, it's in multiples of SPA_MINBLOCKSIZE */ #define DNODE_FLAG_USED_BYTES (1<<0) #define DNODE_FLAG_USERUSED_ACCOUNTED (1<<1) /* Does dnode have a SA spill blkptr in bonus? */ #define DNODE_FLAG_SPILL_BLKPTR (1<<2) typedef struct dnode_phys { uint8_t dn_type; /* dmu_object_type_t */ uint8_t dn_indblkshift; /* ln2(indirect block size) */ uint8_t dn_nlevels; /* 1=dn_blkptr->data blocks */ uint8_t dn_nblkptr; /* length of dn_blkptr */ uint8_t dn_bonustype; /* type of data in bonus buffer */ uint8_t dn_checksum; /* ZIO_CHECKSUM type */ uint8_t dn_compress; /* ZIO_COMPRESS type */ uint8_t dn_flags; /* DNODE_FLAG_* */ uint16_t dn_datablkszsec; /* data block size in 512b sectors */ uint16_t dn_bonuslen; /* length of dn_bonus */ uint8_t dn_pad2[4]; /* accounting is protected by dn_dirty_mtx */ uint64_t dn_maxblkid; /* largest allocated block ID */ uint64_t dn_used; /* bytes (or sectors) of disk space */ uint64_t dn_pad3[4]; blkptr_t dn_blkptr[1]; uint8_t dn_bonus[DN_MAX_BONUSLEN - sizeof (blkptr_t)]; blkptr_t dn_spill; } dnode_phys_t; struct dnode { /* * Protects the structure of the dnode, including the number of levels * of indirection (dn_nlevels), dn_maxblkid, and dn_next_* */ krwlock_t dn_struct_rwlock; /* Our link on dn_objset->os_dnodes list; protected by os_lock. */ list_node_t dn_link; /* immutable: */ struct objset *dn_objset; uint64_t dn_object; struct dmu_buf_impl *dn_dbuf; struct dnode_handle *dn_handle; dnode_phys_t *dn_phys; /* pointer into dn->dn_dbuf->db.db_data */ /* * Copies of stuff in dn_phys. They're valid in the open * context (eg. even before the dnode is first synced). * Where necessary, these are protected by dn_struct_rwlock. */ dmu_object_type_t dn_type; /* object type */ uint16_t dn_bonuslen; /* bonus length */ uint8_t dn_bonustype; /* bonus type */ uint8_t dn_nblkptr; /* number of blkptrs (immutable) */ uint8_t dn_checksum; /* ZIO_CHECKSUM type */ uint8_t dn_compress; /* ZIO_COMPRESS type */ uint8_t dn_nlevels; uint8_t dn_indblkshift; uint8_t dn_datablkshift; /* zero if blksz not power of 2! */ uint8_t dn_moved; /* Has this dnode been moved? */ uint16_t dn_datablkszsec; /* in 512b sectors */ uint32_t dn_datablksz; /* in bytes */ uint64_t dn_maxblkid; uint8_t dn_next_type[TXG_SIZE]; uint8_t dn_next_nblkptr[TXG_SIZE]; uint8_t dn_next_nlevels[TXG_SIZE]; uint8_t dn_next_indblkshift[TXG_SIZE]; uint8_t dn_next_bonustype[TXG_SIZE]; uint8_t dn_rm_spillblk[TXG_SIZE]; /* for removing spill blk */ uint16_t dn_next_bonuslen[TXG_SIZE]; uint32_t dn_next_blksz[TXG_SIZE]; /* next block size in bytes */ /* protected by dn_dbufs_mtx; declared here to fill 32-bit hole */ uint32_t dn_dbufs_count; /* count of dn_dbufs */ /* protected by os_lock: */ - list_node_t dn_dirty_link[TXG_SIZE]; /* next on dataset's dirty */ + multilist_node_t dn_dirty_link[TXG_SIZE]; /* next on dataset's dirty */ /* protected by dn_mtx: */ kmutex_t dn_mtx; list_t dn_dirty_records[TXG_SIZE]; struct range_tree *dn_free_ranges[TXG_SIZE]; uint64_t dn_allocated_txg; uint64_t dn_free_txg; uint64_t dn_assigned_txg; kcondvar_t dn_notxholds; enum dnode_dirtycontext dn_dirtyctx; uint8_t *dn_dirtyctx_firstset; /* dbg: contents meaningless */ /* protected by own devices */ refcount_t dn_tx_holds; refcount_t dn_holds; kmutex_t dn_dbufs_mtx; /* * Descendent dbufs, ordered by dbuf_compare. Note that dn_dbufs * can contain multiple dbufs of the same (level, blkid) when a * dbuf is marked DB_EVICTING without being removed from * dn_dbufs. To maintain the avl invariant that there cannot be * duplicate entries, we order the dbufs by an arbitrary value - * their address in memory. This means that dn_dbufs cannot be used to * directly look up a dbuf. Instead, callers must use avl_walk, have * a reference to the dbuf, or look up a non-existant node with * db_state = DB_SEARCH (see dbuf_free_range for an example). */ avl_tree_t dn_dbufs; /* protected by dn_struct_rwlock */ struct dmu_buf_impl *dn_bonus; /* bonus buffer dbuf */ boolean_t dn_have_spill; /* have spill or are spilling */ /* parent IO for current sync write */ zio_t *dn_zio; /* used in syncing context */ uint64_t dn_oldused; /* old phys used bytes */ uint64_t dn_oldflags; /* old phys dn_flags */ uint64_t dn_olduid, dn_oldgid; uint64_t dn_newuid, dn_newgid; int dn_id_flags; /* holds prefetch structure */ struct zfetch dn_zfetch; }; /* * Adds a level of indirection between the dbuf and the dnode to avoid * iterating descendent dbufs in dnode_move(). Handles are not allocated * individually, but as an array of child dnodes in dnode_hold_impl(). */ typedef struct dnode_handle { /* Protects dnh_dnode from modification by dnode_move(). */ zrlock_t dnh_zrlock; dnode_t *dnh_dnode; } dnode_handle_t; typedef struct dnode_children { dmu_buf_user_t dnc_dbu; /* User evict data */ size_t dnc_count; /* number of children */ dnode_handle_t dnc_children[]; /* sized dynamically */ } dnode_children_t; typedef struct free_range { avl_node_t fr_node; uint64_t fr_blkid; uint64_t fr_nblks; } free_range_t; void dnode_special_open(struct objset *dd, dnode_phys_t *dnp, uint64_t object, dnode_handle_t *dnh); void dnode_special_close(dnode_handle_t *dnh); void dnode_setbonuslen(dnode_t *dn, int newsize, dmu_tx_t *tx); void dnode_setbonus_type(dnode_t *dn, dmu_object_type_t, dmu_tx_t *tx); void dnode_rm_spill(dnode_t *dn, dmu_tx_t *tx); int dnode_hold(struct objset *dd, uint64_t object, void *ref, dnode_t **dnp); int dnode_hold_impl(struct objset *dd, uint64_t object, int flag, void *ref, dnode_t **dnp); boolean_t dnode_add_ref(dnode_t *dn, void *ref); void dnode_rele(dnode_t *dn, void *ref); void dnode_rele_and_unlock(dnode_t *dn, void *tag); void dnode_setdirty(dnode_t *dn, dmu_tx_t *tx); void dnode_sync(dnode_t *dn, dmu_tx_t *tx); void dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx); void dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx); void dnode_free(dnode_t *dn, dmu_tx_t *tx); void dnode_byteswap(dnode_phys_t *dnp); void dnode_buf_byteswap(void *buf, size_t size); void dnode_verify(dnode_t *dn); int dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx); void dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx); void dnode_diduse_space(dnode_t *dn, int64_t space); void dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx, boolean_t); uint64_t dnode_block_freed(dnode_t *dn, uint64_t blkid); void dnode_init(void); void dnode_fini(void); int dnode_next_offset(dnode_t *dn, int flags, uint64_t *off, int minlvl, uint64_t blkfill, uint64_t txg); void dnode_evict_dbufs(dnode_t *dn); void dnode_evict_bonus(dnode_t *dn); #define DNODE_IS_CACHEABLE(_dn) \ ((_dn)->dn_objset->os_primary_cache == ZFS_CACHE_ALL || \ (DMU_OT_IS_METADATA((_dn)->dn_type) && \ (_dn)->dn_objset->os_primary_cache == ZFS_CACHE_METADATA)) #define DNODE_META_IS_CACHEABLE(_dn) \ ((_dn)->dn_objset->os_primary_cache == ZFS_CACHE_ALL || \ (_dn)->dn_objset->os_primary_cache == ZFS_CACHE_METADATA) #ifdef ZFS_DEBUG /* * There should be a ## between the string literal and fmt, to make it * clear that we're joining two strings together, but that piece of shit * gcc doesn't support that preprocessor token. */ #define dprintf_dnode(dn, fmt, ...) do { \ if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ char __db_buf[32]; \ uint64_t __db_obj = (dn)->dn_object; \ if (__db_obj == DMU_META_DNODE_OBJECT) \ (void) strcpy(__db_buf, "mdn"); \ else \ (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \ (u_longlong_t)__db_obj);\ dprintf_ds((dn)->dn_objset->os_dsl_dataset, "obj=%s " fmt, \ __db_buf, __VA_ARGS__); \ } \ _NOTE(CONSTCOND) } while (0) #define DNODE_VERIFY(dn) dnode_verify(dn) #define FREE_VERIFY(db, start, end, tx) free_verify(db, start, end, tx) #else #define dprintf_dnode(db, fmt, ...) #define DNODE_VERIFY(dn) #define FREE_VERIFY(db, start, end, tx) #endif #ifdef __cplusplus } #endif #endif /* _SYS_DNODE_H */ Index: vendor-sys/illumos/dist/uts/common/fs/zfs/sys/dsl_pool.h =================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/sys/dsl_pool.h (revision 316916) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/sys/dsl_pool.h (revision 316917) @@ -1,178 +1,179 @@ /* * 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 http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. - * Copyright (c) 2013 by Delphix. All rights reserved. + * Copyright (c) 2013, 2017 by Delphix. All rights reserved. * Copyright 2016 Nexenta Systems, Inc. All rights reserved. */ #ifndef _SYS_DSL_POOL_H #define _SYS_DSL_POOL_H #include #include #include #include #include #include #include #include #include #include #include #ifdef __cplusplus extern "C" { #endif struct objset; struct dsl_dir; struct dsl_dataset; struct dsl_pool; struct dmu_tx; struct dsl_scan; extern uint64_t zfs_dirty_data_max; extern uint64_t zfs_dirty_data_max_max; extern uint64_t zfs_dirty_data_sync; extern int zfs_dirty_data_max_percent; extern int zfs_delay_min_dirty_percent; extern uint64_t zfs_delay_scale; /* These macros are for indexing into the zfs_all_blkstats_t. */ #define DMU_OT_DEFERRED DMU_OT_NONE #define DMU_OT_OTHER DMU_OT_NUMTYPES /* place holder for DMU_OT() types */ #define DMU_OT_TOTAL (DMU_OT_NUMTYPES + 1) typedef struct zfs_blkstat { uint64_t zb_count; uint64_t zb_asize; uint64_t zb_lsize; uint64_t zb_psize; uint64_t zb_gangs; uint64_t zb_ditto_2_of_2_samevdev; uint64_t zb_ditto_2_of_3_samevdev; uint64_t zb_ditto_3_of_3_samevdev; } zfs_blkstat_t; typedef struct zfs_all_blkstats { zfs_blkstat_t zab_type[DN_MAX_LEVELS + 1][DMU_OT_TOTAL + 1]; } zfs_all_blkstats_t; typedef struct dsl_pool { /* Immutable */ spa_t *dp_spa; struct objset *dp_meta_objset; struct dsl_dir *dp_root_dir; struct dsl_dir *dp_mos_dir; struct dsl_dir *dp_free_dir; struct dsl_dir *dp_leak_dir; struct dsl_dataset *dp_origin_snap; uint64_t dp_root_dir_obj; struct taskq *dp_vnrele_taskq; /* No lock needed - sync context only */ blkptr_t dp_meta_rootbp; uint64_t dp_tmp_userrefs_obj; bpobj_t dp_free_bpobj; uint64_t dp_bptree_obj; uint64_t dp_empty_bpobj; struct dsl_scan *dp_scan; /* Uses dp_lock */ kmutex_t dp_lock; kcondvar_t dp_spaceavail_cv; uint64_t dp_dirty_pertxg[TXG_SIZE]; uint64_t dp_dirty_total; uint64_t dp_long_free_dirty_pertxg[TXG_SIZE]; uint64_t dp_mos_used_delta; uint64_t dp_mos_compressed_delta; uint64_t dp_mos_uncompressed_delta; /* * Time of most recently scheduled (furthest in the future) * wakeup for delayed transactions. */ hrtime_t dp_last_wakeup; /* Has its own locking */ tx_state_t dp_tx; txg_list_t dp_dirty_datasets; txg_list_t dp_dirty_zilogs; txg_list_t dp_dirty_dirs; txg_list_t dp_sync_tasks; + taskq_t *dp_sync_taskq; /* * Protects administrative changes (properties, namespace) * * It is only held for write in syncing context. Therefore * syncing context does not need to ever have it for read, since * nobody else could possibly have it for write. */ rrwlock_t dp_config_rwlock; zfs_all_blkstats_t *dp_blkstats; } dsl_pool_t; int dsl_pool_init(spa_t *spa, uint64_t txg, dsl_pool_t **dpp); int dsl_pool_open(dsl_pool_t *dp); void dsl_pool_close(dsl_pool_t *dp); dsl_pool_t *dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg); void dsl_pool_sync(dsl_pool_t *dp, uint64_t txg); void dsl_pool_sync_done(dsl_pool_t *dp, uint64_t txg); int dsl_pool_sync_context(dsl_pool_t *dp); uint64_t dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree); uint64_t dsl_pool_adjustedfree(dsl_pool_t *dp, boolean_t netfree); void dsl_pool_dirty_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx); void dsl_pool_undirty_space(dsl_pool_t *dp, int64_t space, uint64_t txg); void dsl_free(dsl_pool_t *dp, uint64_t txg, const blkptr_t *bpp); void dsl_free_sync(zio_t *pio, dsl_pool_t *dp, uint64_t txg, const blkptr_t *bpp); void dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx); void dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx); void dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx); void dsl_pool_mos_diduse_space(dsl_pool_t *dp, int64_t used, int64_t comp, int64_t uncomp); void dsl_pool_config_enter(dsl_pool_t *dp, void *tag); void dsl_pool_config_enter_prio(dsl_pool_t *dp, void *tag); void dsl_pool_config_exit(dsl_pool_t *dp, void *tag); boolean_t dsl_pool_config_held(dsl_pool_t *dp); boolean_t dsl_pool_config_held_writer(dsl_pool_t *dp); boolean_t dsl_pool_need_dirty_delay(dsl_pool_t *dp); taskq_t *dsl_pool_vnrele_taskq(dsl_pool_t *dp); int dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, const char *tag, uint64_t now, dmu_tx_t *tx); int dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, const char *tag, dmu_tx_t *tx); void dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp); int dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **); int dsl_pool_hold(const char *name, void *tag, dsl_pool_t **dp); void dsl_pool_rele(dsl_pool_t *dp, void *tag); #ifdef __cplusplus } #endif #endif /* _SYS_DSL_POOL_H */ Index: vendor-sys/illumos/dist/uts/common/fs/zfs/sys/multilist.h =================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/sys/multilist.h (revision 316916) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/sys/multilist.h (revision 316917) @@ -1,106 +1,106 @@ /* * CDDL HEADER START * * This file and its contents are supplied under the terms of the * Common Development and Distribution License ("CDDL"), version 1.0. * You may only use this file in accordance with the terms of version * 1.0 of the CDDL. * * A full copy of the text of the CDDL should have accompanied this * source. A copy of the CDDL is also available via the Internet at * http://www.illumos.org/license/CDDL. * * CDDL HEADER END */ /* * Copyright (c) 2013, 2017 by Delphix. All rights reserved. */ #ifndef _SYS_MULTILIST_H #define _SYS_MULTILIST_H #include #ifdef __cplusplus extern "C" { #endif typedef list_node_t multilist_node_t; typedef struct multilist multilist_t; typedef struct multilist_sublist multilist_sublist_t; typedef unsigned int multilist_sublist_index_func_t(multilist_t *, void *); struct multilist_sublist { /* * The mutex used internally to implement thread safe insertions * and removals to this individual sublist. It can also be locked * by a consumer using multilist_sublist_{lock,unlock}, which is * useful if a consumer needs to traverse the list in a thread * safe manner. */ kmutex_t mls_lock; /* * The actual list object containing all objects in this sublist. */ list_t mls_list; /* * Pad to cache line (64 bytes), in an effort to try and prevent * cache line contention. */ uint8_t mls_pad[24]; }; struct multilist { /* * This is used to get to the multilist_node_t structure given * the void *object contained on the list. */ size_t ml_offset; /* * The number of sublists used internally by this multilist. */ uint64_t ml_num_sublists; /* * The array of pointers to the actual sublists. */ multilist_sublist_t *ml_sublists; /* * Pointer to function which determines the sublist to use * when inserting and removing objects from this multilist. * Please see the comment above multilist_create for details. */ multilist_sublist_index_func_t *ml_index_func; }; void multilist_destroy(multilist_t *); -void multilist_create(multilist_t *, size_t, size_t, - multilist_sublist_index_func_t *); +multilist_t *multilist_create(size_t, size_t, multilist_sublist_index_func_t *); void multilist_insert(multilist_t *, void *); void multilist_remove(multilist_t *, void *); int multilist_is_empty(multilist_t *); unsigned int multilist_get_num_sublists(multilist_t *); unsigned int multilist_get_random_index(multilist_t *); multilist_sublist_t *multilist_sublist_lock(multilist_t *, unsigned int); +multilist_sublist_t *multilist_sublist_lock_obj(multilist_t *, void *); void multilist_sublist_unlock(multilist_sublist_t *); void multilist_sublist_insert_head(multilist_sublist_t *, void *); void multilist_sublist_insert_tail(multilist_sublist_t *, void *); void multilist_sublist_move_forward(multilist_sublist_t *mls, void *obj); void multilist_sublist_remove(multilist_sublist_t *, void *); void *multilist_sublist_head(multilist_sublist_t *); void *multilist_sublist_tail(multilist_sublist_t *); void *multilist_sublist_next(multilist_sublist_t *, void *); void *multilist_sublist_prev(multilist_sublist_t *, void *); void multilist_link_init(multilist_node_t *); int multilist_link_active(multilist_node_t *); #ifdef __cplusplus } #endif #endif /* _SYS_MULTILIST_H */ Index: vendor-sys/illumos/dist/uts/common/fs/zfs/sys/zio.h =================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/sys/zio.h (revision 316916) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/sys/zio.h (revision 316917) @@ -1,614 +1,614 @@ /* * 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 http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright 2011 Nexenta Systems, Inc. All rights reserved. - * Copyright (c) 2012, 2016 by Delphix. All rights reserved. + * Copyright (c) 2012, 2017 by Delphix. All rights reserved. * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. * Copyright (c) 2013, Joyent, Inc. All rights reserved. * Copyright 2016 Toomas Soome */ #ifndef _ZIO_H #define _ZIO_H #include #include #include #include #include #include #include #ifdef __cplusplus extern "C" { #endif /* * Embedded checksum */ #define ZEC_MAGIC 0x210da7ab10c7a11ULL typedef struct zio_eck { uint64_t zec_magic; /* for validation, endianness */ zio_cksum_t zec_cksum; /* 256-bit checksum */ } zio_eck_t; /* * Gang block headers are self-checksumming and contain an array * of block pointers. */ #define SPA_GANGBLOCKSIZE SPA_MINBLOCKSIZE #define SPA_GBH_NBLKPTRS ((SPA_GANGBLOCKSIZE - \ sizeof (zio_eck_t)) / sizeof (blkptr_t)) #define SPA_GBH_FILLER ((SPA_GANGBLOCKSIZE - \ sizeof (zio_eck_t) - \ (SPA_GBH_NBLKPTRS * sizeof (blkptr_t))) /\ sizeof (uint64_t)) typedef struct zio_gbh { blkptr_t zg_blkptr[SPA_GBH_NBLKPTRS]; uint64_t zg_filler[SPA_GBH_FILLER]; zio_eck_t zg_tail; } zio_gbh_phys_t; enum zio_checksum { ZIO_CHECKSUM_INHERIT = 0, ZIO_CHECKSUM_ON, ZIO_CHECKSUM_OFF, ZIO_CHECKSUM_LABEL, ZIO_CHECKSUM_GANG_HEADER, ZIO_CHECKSUM_ZILOG, ZIO_CHECKSUM_FLETCHER_2, ZIO_CHECKSUM_FLETCHER_4, ZIO_CHECKSUM_SHA256, ZIO_CHECKSUM_ZILOG2, ZIO_CHECKSUM_NOPARITY, ZIO_CHECKSUM_SHA512, ZIO_CHECKSUM_SKEIN, ZIO_CHECKSUM_EDONR, ZIO_CHECKSUM_FUNCTIONS }; /* * The number of "legacy" compression functions which can be set on individual * objects. */ #define ZIO_CHECKSUM_LEGACY_FUNCTIONS ZIO_CHECKSUM_ZILOG2 #define ZIO_CHECKSUM_ON_VALUE ZIO_CHECKSUM_FLETCHER_4 #define ZIO_CHECKSUM_DEFAULT ZIO_CHECKSUM_ON #define ZIO_CHECKSUM_MASK 0xffULL #define ZIO_CHECKSUM_VERIFY (1 << 8) #define ZIO_DEDUPCHECKSUM ZIO_CHECKSUM_SHA256 #define ZIO_DEDUPDITTO_MIN 100 /* * The number of "legacy" compression functions which can be set on individual * objects. */ #define ZIO_COMPRESS_LEGACY_FUNCTIONS ZIO_COMPRESS_LZ4 /* * The meaning of "compress = on" selected by the compression features enabled * on a given pool. */ #define ZIO_COMPRESS_LEGACY_ON_VALUE ZIO_COMPRESS_LZJB #define ZIO_COMPRESS_LZ4_ON_VALUE ZIO_COMPRESS_LZ4 #define ZIO_COMPRESS_DEFAULT ZIO_COMPRESS_OFF #define BOOTFS_COMPRESS_VALID(compress) \ ((compress) == ZIO_COMPRESS_LZJB || \ (compress) == ZIO_COMPRESS_LZ4 || \ (compress) == ZIO_COMPRESS_GZIP_1 || \ (compress) == ZIO_COMPRESS_GZIP_2 || \ (compress) == ZIO_COMPRESS_GZIP_3 || \ (compress) == ZIO_COMPRESS_GZIP_4 || \ (compress) == ZIO_COMPRESS_GZIP_5 || \ (compress) == ZIO_COMPRESS_GZIP_6 || \ (compress) == ZIO_COMPRESS_GZIP_7 || \ (compress) == ZIO_COMPRESS_GZIP_8 || \ (compress) == ZIO_COMPRESS_GZIP_9 || \ (compress) == ZIO_COMPRESS_ON || \ (compress) == ZIO_COMPRESS_OFF) #define ZIO_FAILURE_MODE_WAIT 0 #define ZIO_FAILURE_MODE_CONTINUE 1 #define ZIO_FAILURE_MODE_PANIC 2 enum zio_flag { /* * Flags inherited by gang, ddt, and vdev children, * and that must be equal for two zios to aggregate */ ZIO_FLAG_DONT_AGGREGATE = 1 << 0, ZIO_FLAG_IO_REPAIR = 1 << 1, ZIO_FLAG_SELF_HEAL = 1 << 2, ZIO_FLAG_RESILVER = 1 << 3, ZIO_FLAG_SCRUB = 1 << 4, ZIO_FLAG_SCAN_THREAD = 1 << 5, ZIO_FLAG_PHYSICAL = 1 << 6, #define ZIO_FLAG_AGG_INHERIT (ZIO_FLAG_CANFAIL - 1) /* * Flags inherited by ddt, gang, and vdev children. */ ZIO_FLAG_CANFAIL = 1 << 7, /* must be first for INHERIT */ ZIO_FLAG_SPECULATIVE = 1 << 8, ZIO_FLAG_CONFIG_WRITER = 1 << 9, ZIO_FLAG_DONT_RETRY = 1 << 10, ZIO_FLAG_DONT_CACHE = 1 << 11, ZIO_FLAG_NODATA = 1 << 12, ZIO_FLAG_INDUCE_DAMAGE = 1 << 13, ZIO_FLAG_IO_ALLOCATING = 1 << 14, #define ZIO_FLAG_DDT_INHERIT (ZIO_FLAG_IO_RETRY - 1) #define ZIO_FLAG_GANG_INHERIT (ZIO_FLAG_IO_RETRY - 1) /* * Flags inherited by vdev children. */ ZIO_FLAG_IO_RETRY = 1 << 15, /* must be first for INHERIT */ ZIO_FLAG_PROBE = 1 << 16, ZIO_FLAG_TRYHARD = 1 << 17, ZIO_FLAG_OPTIONAL = 1 << 18, #define ZIO_FLAG_VDEV_INHERIT (ZIO_FLAG_DONT_QUEUE - 1) /* * Flags not inherited by any children. */ ZIO_FLAG_DONT_QUEUE = 1 << 19, /* must be first for INHERIT */ ZIO_FLAG_DONT_PROPAGATE = 1 << 20, ZIO_FLAG_IO_BYPASS = 1 << 21, ZIO_FLAG_IO_REWRITE = 1 << 22, ZIO_FLAG_RAW = 1 << 23, ZIO_FLAG_GANG_CHILD = 1 << 24, ZIO_FLAG_DDT_CHILD = 1 << 25, ZIO_FLAG_GODFATHER = 1 << 26, ZIO_FLAG_NOPWRITE = 1 << 27, ZIO_FLAG_REEXECUTED = 1 << 28, ZIO_FLAG_DELEGATED = 1 << 29, }; #define ZIO_FLAG_MUSTSUCCEED 0 #define ZIO_DDT_CHILD_FLAGS(zio) \ (((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) | \ ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL) #define ZIO_GANG_CHILD_FLAGS(zio) \ (((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) | \ ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL) #define ZIO_VDEV_CHILD_FLAGS(zio) \ (((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) | \ ZIO_FLAG_CANFAIL) enum zio_child { ZIO_CHILD_VDEV = 0, ZIO_CHILD_GANG, ZIO_CHILD_DDT, ZIO_CHILD_LOGICAL, ZIO_CHILD_TYPES }; enum zio_wait_type { ZIO_WAIT_READY = 0, ZIO_WAIT_DONE, ZIO_WAIT_TYPES }; /* * We'll take the unused errnos, 'EBADE' and 'EBADR' (from the Convergent * graveyard) to indicate checksum errors and fragmentation. */ #define ECKSUM EBADE #define EFRAGS EBADR typedef void zio_done_func_t(zio_t *zio); extern boolean_t zio_dva_throttle_enabled; extern const char *zio_type_name[ZIO_TYPES]; /* * A bookmark is a four-tuple that uniquely * identifies any block in the pool. By convention, the meta-objset (MOS) * is objset 0, and the meta-dnode is object 0. This covers all blocks * except root blocks and ZIL blocks, which are defined as follows: * * Root blocks (objset_phys_t) are object 0, level -1: . * ZIL blocks are bookmarked . * dmu_sync()ed ZIL data blocks are bookmarked . * dnode visit bookmarks are . * * Note: this structure is called a bookmark because its original purpose * was to remember where to resume a pool-wide traverse. * * Note: this structure is passed between userland and the kernel, and is * stored on disk (by virtue of being incorporated into other on-disk * structures, e.g. dsl_scan_phys_t). */ typedef struct zbookmark_phys { uint64_t zb_objset; uint64_t zb_object; int64_t zb_level; uint64_t zb_blkid; } zbookmark_phys_t; #define SET_BOOKMARK(zb, objset, object, level, blkid) \ { \ (zb)->zb_objset = objset; \ (zb)->zb_object = object; \ (zb)->zb_level = level; \ (zb)->zb_blkid = blkid; \ } #define ZB_DESTROYED_OBJSET (-1ULL) #define ZB_ROOT_OBJECT (0ULL) #define ZB_ROOT_LEVEL (-1LL) #define ZB_ROOT_BLKID (0ULL) #define ZB_ZIL_OBJECT (0ULL) #define ZB_ZIL_LEVEL (-2LL) #define ZB_DNODE_LEVEL (-3LL) #define ZB_DNODE_BLKID (0ULL) #define ZB_IS_ZERO(zb) \ ((zb)->zb_objset == 0 && (zb)->zb_object == 0 && \ (zb)->zb_level == 0 && (zb)->zb_blkid == 0) #define ZB_IS_ROOT(zb) \ ((zb)->zb_object == ZB_ROOT_OBJECT && \ (zb)->zb_level == ZB_ROOT_LEVEL && \ (zb)->zb_blkid == ZB_ROOT_BLKID) typedef struct zio_prop { enum zio_checksum zp_checksum; enum zio_compress zp_compress; dmu_object_type_t zp_type; uint8_t zp_level; uint8_t zp_copies; boolean_t zp_dedup; boolean_t zp_dedup_verify; boolean_t zp_nopwrite; } zio_prop_t; typedef struct zio_cksum_report zio_cksum_report_t; typedef void zio_cksum_finish_f(zio_cksum_report_t *rep, const void *good_data); typedef void zio_cksum_free_f(void *cbdata, size_t size); struct zio_bad_cksum; /* defined in zio_checksum.h */ struct dnode_phys; struct zio_cksum_report { struct zio_cksum_report *zcr_next; nvlist_t *zcr_ereport; nvlist_t *zcr_detector; void *zcr_cbdata; size_t zcr_cbinfo; /* passed to zcr_free() */ uint64_t zcr_align; uint64_t zcr_length; zio_cksum_finish_f *zcr_finish; zio_cksum_free_f *zcr_free; /* internal use only */ struct zio_bad_cksum *zcr_ckinfo; /* information from failure */ }; typedef void zio_vsd_cksum_report_f(zio_t *zio, zio_cksum_report_t *zcr, void *arg); zio_vsd_cksum_report_f zio_vsd_default_cksum_report; typedef struct zio_vsd_ops { zio_done_func_t *vsd_free; zio_vsd_cksum_report_f *vsd_cksum_report; } zio_vsd_ops_t; typedef struct zio_gang_node { zio_gbh_phys_t *gn_gbh; struct zio_gang_node *gn_child[SPA_GBH_NBLKPTRS]; } zio_gang_node_t; typedef zio_t *zio_gang_issue_func_t(zio_t *zio, blkptr_t *bp, zio_gang_node_t *gn, void *data); typedef void zio_transform_func_t(zio_t *zio, void *data, uint64_t size); typedef struct zio_transform { void *zt_orig_data; uint64_t zt_orig_size; uint64_t zt_bufsize; zio_transform_func_t *zt_transform; struct zio_transform *zt_next; } zio_transform_t; typedef int zio_pipe_stage_t(zio_t *zio); /* * The io_reexecute flags are distinct from io_flags because the child must * be able to propagate them to the parent. The normal io_flags are local * to the zio, not protected by any lock, and not modifiable by children; * the reexecute flags are protected by io_lock, modifiable by children, * and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set. */ #define ZIO_REEXECUTE_NOW 0x01 #define ZIO_REEXECUTE_SUSPEND 0x02 typedef struct zio_alloc_list { list_t zal_list; uint64_t zal_size; } zio_alloc_list_t; typedef struct zio_link { zio_t *zl_parent; zio_t *zl_child; list_node_t zl_parent_node; list_node_t zl_child_node; } zio_link_t; struct zio { /* Core information about this I/O */ zbookmark_phys_t io_bookmark; zio_prop_t io_prop; zio_type_t io_type; enum zio_child io_child_type; int io_cmd; zio_priority_t io_priority; uint8_t io_reexecute; uint8_t io_state[ZIO_WAIT_TYPES]; uint64_t io_txg; spa_t *io_spa; blkptr_t *io_bp; blkptr_t *io_bp_override; blkptr_t io_bp_copy; list_t io_parent_list; list_t io_child_list; zio_t *io_logical; zio_transform_t *io_transform_stack; /* Callback info */ zio_done_func_t *io_ready; zio_done_func_t *io_children_ready; zio_done_func_t *io_physdone; zio_done_func_t *io_done; void *io_private; int64_t io_prev_space_delta; /* DMU private */ blkptr_t io_bp_orig; /* Data represented by this I/O */ void *io_data; void *io_orig_data; uint64_t io_size; uint64_t io_orig_size; /* io_lsize != io_orig_size iff this is a raw write */ uint64_t io_lsize; /* Stuff for the vdev stack */ vdev_t *io_vd; void *io_vsd; const zio_vsd_ops_t *io_vsd_ops; uint64_t io_offset; hrtime_t io_timestamp; hrtime_t io_queued_timestamp; hrtime_t io_target_timestamp; avl_node_t io_queue_node; avl_node_t io_offset_node; avl_node_t io_alloc_node; zio_alloc_list_t io_alloc_list; /* Internal pipeline state */ enum zio_flag io_flags; enum zio_stage io_stage; enum zio_stage io_pipeline; enum zio_flag io_orig_flags; enum zio_stage io_orig_stage; enum zio_stage io_orig_pipeline; enum zio_stage io_pipeline_trace; int io_error; int io_child_error[ZIO_CHILD_TYPES]; uint64_t io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES]; uint64_t io_child_count; uint64_t io_phys_children; uint64_t io_parent_count; uint64_t *io_stall; zio_t *io_gang_leader; zio_gang_node_t *io_gang_tree; void *io_executor; void *io_waiter; kmutex_t io_lock; kcondvar_t io_cv; /* FMA state */ zio_cksum_report_t *io_cksum_report; uint64_t io_ena; /* Taskq dispatching state */ taskq_ent_t io_tqent; }; -extern int zio_timestamp_compare(const void *, const void *); +extern int zio_bookmark_compare(const void *, const void *); extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd, zio_done_func_t *done, void *private, enum zio_flag flags); extern zio_t *zio_root(spa_t *spa, zio_done_func_t *done, void *private, enum zio_flag flags); extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, void *data, uint64_t lsize, zio_done_func_t *done, void *private, zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb); extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, void *data, uint64_t size, uint64_t psize, const zio_prop_t *zp, zio_done_func_t *ready, zio_done_func_t *children_ready, zio_done_func_t *physdone, zio_done_func_t *done, void *private, zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb); extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, void *data, uint64_t size, zio_done_func_t *done, void *private, zio_priority_t priority, enum zio_flag flags, zbookmark_phys_t *zb); extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies, boolean_t nopwrite); extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp); extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, zio_done_func_t *done, void *private, enum zio_flag flags); extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd, zio_done_func_t *done, void *private, enum zio_flag flags); extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size, void *data, int checksum, zio_done_func_t *done, void *private, zio_priority_t priority, enum zio_flag flags, boolean_t labels); extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size, void *data, int checksum, zio_done_func_t *done, void *private, zio_priority_t priority, enum zio_flag flags, boolean_t labels); extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, enum zio_flag flags); extern int zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp, blkptr_t *old_bp, uint64_t size, boolean_t use_slog); extern void zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp); extern void zio_flush(zio_t *zio, vdev_t *vd); extern void zio_shrink(zio_t *zio, uint64_t size); extern int zio_wait(zio_t *zio); extern void zio_nowait(zio_t *zio); extern void zio_execute(zio_t *zio); extern void zio_interrupt(zio_t *zio); extern void zio_delay_init(zio_t *zio); extern void zio_delay_interrupt(zio_t *zio); extern zio_t *zio_walk_parents(zio_t *cio, zio_link_t **); extern zio_t *zio_walk_children(zio_t *pio, zio_link_t **); extern zio_t *zio_unique_parent(zio_t *cio); extern void zio_add_child(zio_t *pio, zio_t *cio); extern void *zio_buf_alloc(size_t size); extern void zio_buf_free(void *buf, size_t size); extern void *zio_data_buf_alloc(size_t size); extern void zio_data_buf_free(void *buf, size_t size); extern void zio_push_transform(zio_t *zio, void *data, uint64_t size, uint64_t bufsize, zio_transform_func_t *transform); extern void zio_pop_transforms(zio_t *zio); extern void zio_resubmit_stage_async(void *); extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd, uint64_t offset, void *data, uint64_t size, int type, zio_priority_t priority, enum zio_flag flags, zio_done_func_t *done, void *private); extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset, void *data, uint64_t size, int type, zio_priority_t priority, enum zio_flag flags, zio_done_func_t *done, void *private); extern void zio_vdev_io_bypass(zio_t *zio); extern void zio_vdev_io_reissue(zio_t *zio); extern void zio_vdev_io_redone(zio_t *zio); extern void zio_checksum_verified(zio_t *zio); extern int zio_worst_error(int e1, int e2); extern enum zio_checksum zio_checksum_select(enum zio_checksum child, enum zio_checksum parent); extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa, enum zio_checksum child, enum zio_checksum parent); extern enum zio_compress zio_compress_select(spa_t *spa, enum zio_compress child, enum zio_compress parent); extern void zio_suspend(spa_t *spa, zio_t *zio); extern int zio_resume(spa_t *spa); extern void zio_resume_wait(spa_t *spa); /* * Initial setup and teardown. */ extern void zio_init(void); extern void zio_fini(void); /* * Fault injection */ struct zinject_record; extern uint32_t zio_injection_enabled; extern int zio_inject_fault(char *name, int flags, int *id, struct zinject_record *record); extern int zio_inject_list_next(int *id, char *name, size_t buflen, struct zinject_record *record); extern int zio_clear_fault(int id); extern void zio_handle_panic_injection(spa_t *spa, char *tag, uint64_t type); extern int zio_handle_fault_injection(zio_t *zio, int error); extern int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error); extern int zio_handle_label_injection(zio_t *zio, int error); extern void zio_handle_ignored_writes(zio_t *zio); extern hrtime_t zio_handle_io_delay(zio_t *zio); /* * Checksum ereport functions */ extern void zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd, struct zio *zio, uint64_t offset, uint64_t length, void *arg, struct zio_bad_cksum *info); extern void zfs_ereport_finish_checksum(zio_cksum_report_t *report, const void *good_data, const void *bad_data, boolean_t drop_if_identical); extern void zfs_ereport_send_interim_checksum(zio_cksum_report_t *report); extern void zfs_ereport_free_checksum(zio_cksum_report_t *report); /* If we have the good data in hand, this function can be used */ extern void zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd, struct zio *zio, uint64_t offset, uint64_t length, const void *good_data, const void *bad_data, struct zio_bad_cksum *info); /* Called from spa_sync(), but primarily an injection handler */ extern void spa_handle_ignored_writes(spa_t *spa); /* zbookmark_phys functions */ boolean_t zbookmark_subtree_completed(const struct dnode_phys *dnp, const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block); int zbookmark_compare(uint16_t dbss1, uint8_t ibs1, uint16_t dbss2, uint8_t ibs2, const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2); #ifdef __cplusplus } #endif #endif /* _ZIO_H */ Index: vendor-sys/illumos/dist/uts/common/fs/zfs/zio.c =================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/zio.c (revision 316916) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/zio.c (revision 316917) @@ -1,3914 +1,3922 @@ /* * 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 http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. - * Copyright (c) 2011, 2016 by Delphix. All rights reserved. + * Copyright (c) 2011, 2017 by Delphix. All rights reserved. * Copyright (c) 2011 Nexenta Systems, Inc. All rights reserved. * Copyright (c) 2014 Integros [integros.com] */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * ========================================================================== * I/O type descriptions * ========================================================================== */ const char *zio_type_name[ZIO_TYPES] = { "zio_null", "zio_read", "zio_write", "zio_free", "zio_claim", "zio_ioctl" }; boolean_t zio_dva_throttle_enabled = B_TRUE; /* * ========================================================================== * I/O kmem caches * ========================================================================== */ kmem_cache_t *zio_cache; kmem_cache_t *zio_link_cache; kmem_cache_t *zio_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT]; kmem_cache_t *zio_data_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT]; #ifdef _KERNEL extern vmem_t *zio_alloc_arena; #endif #define ZIO_PIPELINE_CONTINUE 0x100 #define ZIO_PIPELINE_STOP 0x101 #define BP_SPANB(indblkshift, level) \ (((uint64_t)1) << ((level) * ((indblkshift) - SPA_BLKPTRSHIFT))) #define COMPARE_META_LEVEL 0x80000000ul /* * The following actions directly effect the spa's sync-to-convergence logic. * The values below define the sync pass when we start performing the action. * Care should be taken when changing these values as they directly impact * spa_sync() performance. Tuning these values may introduce subtle performance * pathologies and should only be done in the context of performance analysis. * These tunables will eventually be removed and replaced with #defines once * enough analysis has been done to determine optimal values. * * The 'zfs_sync_pass_deferred_free' pass must be greater than 1 to ensure that * regular blocks are not deferred. */ int zfs_sync_pass_deferred_free = 2; /* defer frees starting in this pass */ int zfs_sync_pass_dont_compress = 5; /* don't compress starting in this pass */ int zfs_sync_pass_rewrite = 2; /* rewrite new bps starting in this pass */ /* * An allocating zio is one that either currently has the DVA allocate * stage set or will have it later in its lifetime. */ #define IO_IS_ALLOCATING(zio) ((zio)->io_orig_pipeline & ZIO_STAGE_DVA_ALLOCATE) boolean_t zio_requeue_io_start_cut_in_line = B_TRUE; #ifdef ZFS_DEBUG int zio_buf_debug_limit = 16384; #else int zio_buf_debug_limit = 0; #endif static void zio_taskq_dispatch(zio_t *, zio_taskq_type_t, boolean_t); void zio_init(void) { size_t c; vmem_t *data_alloc_arena = NULL; #ifdef _KERNEL data_alloc_arena = zio_alloc_arena; #endif zio_cache = kmem_cache_create("zio_cache", sizeof (zio_t), 0, NULL, NULL, NULL, NULL, NULL, 0); zio_link_cache = kmem_cache_create("zio_link_cache", sizeof (zio_link_t), 0, NULL, NULL, NULL, NULL, NULL, 0); /* * For small buffers, we want a cache for each multiple of * SPA_MINBLOCKSIZE. For larger buffers, we want a cache * for each quarter-power of 2. */ for (c = 0; c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; c++) { size_t size = (c + 1) << SPA_MINBLOCKSHIFT; size_t p2 = size; size_t align = 0; size_t cflags = (size > zio_buf_debug_limit) ? KMC_NODEBUG : 0; while (!ISP2(p2)) p2 &= p2 - 1; #ifndef _KERNEL /* * If we are using watchpoints, put each buffer on its own page, * to eliminate the performance overhead of trapping to the * kernel when modifying a non-watched buffer that shares the * page with a watched buffer. */ if (arc_watch && !IS_P2ALIGNED(size, PAGESIZE)) continue; #endif if (size <= 4 * SPA_MINBLOCKSIZE) { align = SPA_MINBLOCKSIZE; } else if (IS_P2ALIGNED(size, p2 >> 2)) { align = MIN(p2 >> 2, PAGESIZE); } if (align != 0) { char name[36]; (void) sprintf(name, "zio_buf_%lu", (ulong_t)size); zio_buf_cache[c] = kmem_cache_create(name, size, align, NULL, NULL, NULL, NULL, NULL, cflags); /* * Since zio_data bufs do not appear in crash dumps, we * pass KMC_NOTOUCH so that no allocator metadata is * stored with the buffers. */ (void) sprintf(name, "zio_data_buf_%lu", (ulong_t)size); zio_data_buf_cache[c] = kmem_cache_create(name, size, align, NULL, NULL, NULL, NULL, data_alloc_arena, cflags | KMC_NOTOUCH); } } while (--c != 0) { ASSERT(zio_buf_cache[c] != NULL); if (zio_buf_cache[c - 1] == NULL) zio_buf_cache[c - 1] = zio_buf_cache[c]; ASSERT(zio_data_buf_cache[c] != NULL); if (zio_data_buf_cache[c - 1] == NULL) zio_data_buf_cache[c - 1] = zio_data_buf_cache[c]; } zio_inject_init(); } void zio_fini(void) { size_t c; kmem_cache_t *last_cache = NULL; kmem_cache_t *last_data_cache = NULL; for (c = 0; c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; c++) { if (zio_buf_cache[c] != last_cache) { last_cache = zio_buf_cache[c]; kmem_cache_destroy(zio_buf_cache[c]); } zio_buf_cache[c] = NULL; if (zio_data_buf_cache[c] != last_data_cache) { last_data_cache = zio_data_buf_cache[c]; kmem_cache_destroy(zio_data_buf_cache[c]); } zio_data_buf_cache[c] = NULL; } kmem_cache_destroy(zio_link_cache); kmem_cache_destroy(zio_cache); zio_inject_fini(); } /* * ========================================================================== * Allocate and free I/O buffers * ========================================================================== */ /* * Use zio_buf_alloc to allocate ZFS metadata. This data will appear in a * crashdump if the kernel panics, so use it judiciously. Obviously, it's * useful to inspect ZFS metadata, but if possible, we should avoid keeping * excess / transient data in-core during a crashdump. */ void * zio_buf_alloc(size_t size) { size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; VERIFY3U(c, <, SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); return (kmem_cache_alloc(zio_buf_cache[c], KM_PUSHPAGE)); } /* * Use zio_data_buf_alloc to allocate data. The data will not appear in a * crashdump if the kernel panics. This exists so that we will limit the amount * of ZFS data that shows up in a kernel crashdump. (Thus reducing the amount * of kernel heap dumped to disk when the kernel panics) */ void * zio_data_buf_alloc(size_t size) { size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; VERIFY3U(c, <, SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); return (kmem_cache_alloc(zio_data_buf_cache[c], KM_PUSHPAGE)); } void zio_buf_free(void *buf, size_t size) { size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; VERIFY3U(c, <, SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); kmem_cache_free(zio_buf_cache[c], buf); } void zio_data_buf_free(void *buf, size_t size) { size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; VERIFY3U(c, <, SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); kmem_cache_free(zio_data_buf_cache[c], buf); } /* * ========================================================================== * Push and pop I/O transform buffers * ========================================================================== */ void zio_push_transform(zio_t *zio, void *data, uint64_t size, uint64_t bufsize, zio_transform_func_t *transform) { zio_transform_t *zt = kmem_alloc(sizeof (zio_transform_t), KM_SLEEP); zt->zt_orig_data = zio->io_data; zt->zt_orig_size = zio->io_size; zt->zt_bufsize = bufsize; zt->zt_transform = transform; zt->zt_next = zio->io_transform_stack; zio->io_transform_stack = zt; zio->io_data = data; zio->io_size = size; } void zio_pop_transforms(zio_t *zio) { zio_transform_t *zt; while ((zt = zio->io_transform_stack) != NULL) { if (zt->zt_transform != NULL) zt->zt_transform(zio, zt->zt_orig_data, zt->zt_orig_size); if (zt->zt_bufsize != 0) zio_buf_free(zio->io_data, zt->zt_bufsize); zio->io_data = zt->zt_orig_data; zio->io_size = zt->zt_orig_size; zio->io_transform_stack = zt->zt_next; kmem_free(zt, sizeof (zio_transform_t)); } } /* * ========================================================================== * I/O transform callbacks for subblocks and decompression * ========================================================================== */ static void zio_subblock(zio_t *zio, void *data, uint64_t size) { ASSERT(zio->io_size > size); if (zio->io_type == ZIO_TYPE_READ) bcopy(zio->io_data, data, size); } static void zio_decompress(zio_t *zio, void *data, uint64_t size) { if (zio->io_error == 0 && zio_decompress_data(BP_GET_COMPRESS(zio->io_bp), zio->io_data, data, zio->io_size, size) != 0) zio->io_error = SET_ERROR(EIO); } /* * ========================================================================== * I/O parent/child relationships and pipeline interlocks * ========================================================================== */ zio_t * zio_walk_parents(zio_t *cio, zio_link_t **zl) { list_t *pl = &cio->io_parent_list; *zl = (*zl == NULL) ? list_head(pl) : list_next(pl, *zl); if (*zl == NULL) return (NULL); ASSERT((*zl)->zl_child == cio); return ((*zl)->zl_parent); } zio_t * zio_walk_children(zio_t *pio, zio_link_t **zl) { list_t *cl = &pio->io_child_list; *zl = (*zl == NULL) ? list_head(cl) : list_next(cl, *zl); if (*zl == NULL) return (NULL); ASSERT((*zl)->zl_parent == pio); return ((*zl)->zl_child); } zio_t * zio_unique_parent(zio_t *cio) { zio_link_t *zl = NULL; zio_t *pio = zio_walk_parents(cio, &zl); VERIFY3P(zio_walk_parents(cio, &zl), ==, NULL); return (pio); } void zio_add_child(zio_t *pio, zio_t *cio) { zio_link_t *zl = kmem_cache_alloc(zio_link_cache, KM_SLEEP); /* * Logical I/Os can have logical, gang, or vdev children. * Gang I/Os can have gang or vdev children. * Vdev I/Os can only have vdev children. * The following ASSERT captures all of these constraints. */ ASSERT(cio->io_child_type <= pio->io_child_type); zl->zl_parent = pio; zl->zl_child = cio; mutex_enter(&cio->io_lock); mutex_enter(&pio->io_lock); ASSERT(pio->io_state[ZIO_WAIT_DONE] == 0); for (int w = 0; w < ZIO_WAIT_TYPES; w++) pio->io_children[cio->io_child_type][w] += !cio->io_state[w]; list_insert_head(&pio->io_child_list, zl); list_insert_head(&cio->io_parent_list, zl); pio->io_child_count++; cio->io_parent_count++; mutex_exit(&pio->io_lock); mutex_exit(&cio->io_lock); } static void zio_remove_child(zio_t *pio, zio_t *cio, zio_link_t *zl) { ASSERT(zl->zl_parent == pio); ASSERT(zl->zl_child == cio); mutex_enter(&cio->io_lock); mutex_enter(&pio->io_lock); list_remove(&pio->io_child_list, zl); list_remove(&cio->io_parent_list, zl); pio->io_child_count--; cio->io_parent_count--; mutex_exit(&pio->io_lock); mutex_exit(&cio->io_lock); kmem_cache_free(zio_link_cache, zl); } static boolean_t zio_wait_for_children(zio_t *zio, enum zio_child child, enum zio_wait_type wait) { uint64_t *countp = &zio->io_children[child][wait]; boolean_t waiting = B_FALSE; mutex_enter(&zio->io_lock); ASSERT(zio->io_stall == NULL); if (*countp != 0) { zio->io_stage >>= 1; ASSERT3U(zio->io_stage, !=, ZIO_STAGE_OPEN); zio->io_stall = countp; waiting = B_TRUE; } mutex_exit(&zio->io_lock); return (waiting); } static void zio_notify_parent(zio_t *pio, zio_t *zio, enum zio_wait_type wait) { uint64_t *countp = &pio->io_children[zio->io_child_type][wait]; int *errorp = &pio->io_child_error[zio->io_child_type]; mutex_enter(&pio->io_lock); if (zio->io_error && !(zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) *errorp = zio_worst_error(*errorp, zio->io_error); pio->io_reexecute |= zio->io_reexecute; ASSERT3U(*countp, >, 0); (*countp)--; if (*countp == 0 && pio->io_stall == countp) { zio_taskq_type_t type = pio->io_stage < ZIO_STAGE_VDEV_IO_START ? ZIO_TASKQ_ISSUE : ZIO_TASKQ_INTERRUPT; pio->io_stall = NULL; mutex_exit(&pio->io_lock); /* * Dispatch the parent zio in its own taskq so that * the child can continue to make progress. This also * prevents overflowing the stack when we have deeply nested * parent-child relationships. */ zio_taskq_dispatch(pio, type, B_FALSE); } else { mutex_exit(&pio->io_lock); } } static void zio_inherit_child_errors(zio_t *zio, enum zio_child c) { if (zio->io_child_error[c] != 0 && zio->io_error == 0) zio->io_error = zio->io_child_error[c]; } int -zio_timestamp_compare(const void *x1, const void *x2) +zio_bookmark_compare(const void *x1, const void *x2) { const zio_t *z1 = x1; const zio_t *z2 = x2; - if (z1->io_queued_timestamp < z2->io_queued_timestamp) + if (z1->io_bookmark.zb_objset < z2->io_bookmark.zb_objset) return (-1); - if (z1->io_queued_timestamp > z2->io_queued_timestamp) + if (z1->io_bookmark.zb_objset > z2->io_bookmark.zb_objset) return (1); - if (z1->io_offset < z2->io_offset) + if (z1->io_bookmark.zb_object < z2->io_bookmark.zb_object) return (-1); - if (z1->io_offset > z2->io_offset) + if (z1->io_bookmark.zb_object > z2->io_bookmark.zb_object) return (1); + if (z1->io_bookmark.zb_level < z2->io_bookmark.zb_level) + return (-1); + if (z1->io_bookmark.zb_level > z2->io_bookmark.zb_level) + return (1); + + if (z1->io_bookmark.zb_blkid < z2->io_bookmark.zb_blkid) + return (-1); + if (z1->io_bookmark.zb_blkid > z2->io_bookmark.zb_blkid) + return (1); + if (z1 < z2) return (-1); if (z1 > z2) return (1); return (0); } /* * ========================================================================== * Create the various types of I/O (read, write, free, etc) * ========================================================================== */ static zio_t * zio_create(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, void *data, uint64_t lsize, uint64_t psize, zio_done_func_t *done, void *private, zio_type_t type, zio_priority_t priority, enum zio_flag flags, vdev_t *vd, uint64_t offset, const zbookmark_phys_t *zb, enum zio_stage stage, enum zio_stage pipeline) { zio_t *zio; ASSERT3U(psize, <=, SPA_MAXBLOCKSIZE); ASSERT(P2PHASE(psize, SPA_MINBLOCKSIZE) == 0); ASSERT(P2PHASE(offset, SPA_MINBLOCKSIZE) == 0); ASSERT(!vd || spa_config_held(spa, SCL_STATE_ALL, RW_READER)); ASSERT(!bp || !(flags & ZIO_FLAG_CONFIG_WRITER)); ASSERT(vd || stage == ZIO_STAGE_OPEN); IMPLY(lsize != psize, (flags & ZIO_FLAG_RAW) != 0); zio = kmem_cache_alloc(zio_cache, KM_SLEEP); bzero(zio, sizeof (zio_t)); mutex_init(&zio->io_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&zio->io_cv, NULL, CV_DEFAULT, NULL); list_create(&zio->io_parent_list, sizeof (zio_link_t), offsetof(zio_link_t, zl_parent_node)); list_create(&zio->io_child_list, sizeof (zio_link_t), offsetof(zio_link_t, zl_child_node)); metaslab_trace_init(&zio->io_alloc_list); if (vd != NULL) zio->io_child_type = ZIO_CHILD_VDEV; else if (flags & ZIO_FLAG_GANG_CHILD) zio->io_child_type = ZIO_CHILD_GANG; else if (flags & ZIO_FLAG_DDT_CHILD) zio->io_child_type = ZIO_CHILD_DDT; else zio->io_child_type = ZIO_CHILD_LOGICAL; if (bp != NULL) { zio->io_bp = (blkptr_t *)bp; zio->io_bp_copy = *bp; zio->io_bp_orig = *bp; if (type != ZIO_TYPE_WRITE || zio->io_child_type == ZIO_CHILD_DDT) zio->io_bp = &zio->io_bp_copy; /* so caller can free */ if (zio->io_child_type == ZIO_CHILD_LOGICAL) zio->io_logical = zio; if (zio->io_child_type > ZIO_CHILD_GANG && BP_IS_GANG(bp)) pipeline |= ZIO_GANG_STAGES; } zio->io_spa = spa; zio->io_txg = txg; zio->io_done = done; zio->io_private = private; zio->io_type = type; zio->io_priority = priority; zio->io_vd = vd; zio->io_offset = offset; zio->io_orig_data = zio->io_data = data; zio->io_orig_size = zio->io_size = psize; zio->io_lsize = lsize; zio->io_orig_flags = zio->io_flags = flags; zio->io_orig_stage = zio->io_stage = stage; zio->io_orig_pipeline = zio->io_pipeline = pipeline; zio->io_pipeline_trace = ZIO_STAGE_OPEN; zio->io_state[ZIO_WAIT_READY] = (stage >= ZIO_STAGE_READY); zio->io_state[ZIO_WAIT_DONE] = (stage >= ZIO_STAGE_DONE); if (zb != NULL) zio->io_bookmark = *zb; if (pio != NULL) { if (zio->io_logical == NULL) zio->io_logical = pio->io_logical; if (zio->io_child_type == ZIO_CHILD_GANG) zio->io_gang_leader = pio->io_gang_leader; zio_add_child(pio, zio); } return (zio); } static void zio_destroy(zio_t *zio) { metaslab_trace_fini(&zio->io_alloc_list); list_destroy(&zio->io_parent_list); list_destroy(&zio->io_child_list); mutex_destroy(&zio->io_lock); cv_destroy(&zio->io_cv); kmem_cache_free(zio_cache, zio); } zio_t * zio_null(zio_t *pio, spa_t *spa, vdev_t *vd, zio_done_func_t *done, void *private, enum zio_flag flags) { zio_t *zio; zio = zio_create(pio, spa, 0, NULL, NULL, 0, 0, done, private, ZIO_TYPE_NULL, ZIO_PRIORITY_NOW, flags, vd, 0, NULL, ZIO_STAGE_OPEN, ZIO_INTERLOCK_PIPELINE); return (zio); } zio_t * zio_root(spa_t *spa, zio_done_func_t *done, void *private, enum zio_flag flags) { return (zio_null(NULL, spa, NULL, done, private, flags)); } void zfs_blkptr_verify(spa_t *spa, const blkptr_t *bp) { if (!DMU_OT_IS_VALID(BP_GET_TYPE(bp))) { zfs_panic_recover("blkptr at %p has invalid TYPE %llu", bp, (longlong_t)BP_GET_TYPE(bp)); } if (BP_GET_CHECKSUM(bp) >= ZIO_CHECKSUM_FUNCTIONS || BP_GET_CHECKSUM(bp) <= ZIO_CHECKSUM_ON) { zfs_panic_recover("blkptr at %p has invalid CHECKSUM %llu", bp, (longlong_t)BP_GET_CHECKSUM(bp)); } if (BP_GET_COMPRESS(bp) >= ZIO_COMPRESS_FUNCTIONS || BP_GET_COMPRESS(bp) <= ZIO_COMPRESS_ON) { zfs_panic_recover("blkptr at %p has invalid COMPRESS %llu", bp, (longlong_t)BP_GET_COMPRESS(bp)); } if (BP_GET_LSIZE(bp) > SPA_MAXBLOCKSIZE) { zfs_panic_recover("blkptr at %p has invalid LSIZE %llu", bp, (longlong_t)BP_GET_LSIZE(bp)); } if (BP_GET_PSIZE(bp) > SPA_MAXBLOCKSIZE) { zfs_panic_recover("blkptr at %p has invalid PSIZE %llu", bp, (longlong_t)BP_GET_PSIZE(bp)); } if (BP_IS_EMBEDDED(bp)) { if (BPE_GET_ETYPE(bp) > NUM_BP_EMBEDDED_TYPES) { zfs_panic_recover("blkptr at %p has invalid ETYPE %llu", bp, (longlong_t)BPE_GET_ETYPE(bp)); } } /* * Pool-specific checks. * * Note: it would be nice to verify that the blk_birth and * BP_PHYSICAL_BIRTH() are not too large. However, spa_freeze() * allows the birth time of log blocks (and dmu_sync()-ed blocks * that are in the log) to be arbitrarily large. */ for (int i = 0; i < BP_GET_NDVAS(bp); i++) { uint64_t vdevid = DVA_GET_VDEV(&bp->blk_dva[i]); if (vdevid >= spa->spa_root_vdev->vdev_children) { zfs_panic_recover("blkptr at %p DVA %u has invalid " "VDEV %llu", bp, i, (longlong_t)vdevid); continue; } vdev_t *vd = spa->spa_root_vdev->vdev_child[vdevid]; if (vd == NULL) { zfs_panic_recover("blkptr at %p DVA %u has invalid " "VDEV %llu", bp, i, (longlong_t)vdevid); continue; } if (vd->vdev_ops == &vdev_hole_ops) { zfs_panic_recover("blkptr at %p DVA %u has hole " "VDEV %llu", bp, i, (longlong_t)vdevid); continue; } if (vd->vdev_ops == &vdev_missing_ops) { /* * "missing" vdevs are valid during import, but we * don't have their detailed info (e.g. asize), so * we can't perform any more checks on them. */ continue; } uint64_t offset = DVA_GET_OFFSET(&bp->blk_dva[i]); uint64_t asize = DVA_GET_ASIZE(&bp->blk_dva[i]); if (BP_IS_GANG(bp)) asize = vdev_psize_to_asize(vd, SPA_GANGBLOCKSIZE); if (offset + asize > vd->vdev_asize) { zfs_panic_recover("blkptr at %p DVA %u has invalid " "OFFSET %llu", bp, i, (longlong_t)offset); } } } zio_t * zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, void *data, uint64_t size, zio_done_func_t *done, void *private, zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb) { zio_t *zio; zfs_blkptr_verify(spa, bp); zio = zio_create(pio, spa, BP_PHYSICAL_BIRTH(bp), bp, data, size, size, done, private, ZIO_TYPE_READ, priority, flags, NULL, 0, zb, ZIO_STAGE_OPEN, (flags & ZIO_FLAG_DDT_CHILD) ? ZIO_DDT_CHILD_READ_PIPELINE : ZIO_READ_PIPELINE); return (zio); } zio_t * zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, void *data, uint64_t lsize, uint64_t psize, const zio_prop_t *zp, zio_done_func_t *ready, zio_done_func_t *children_ready, zio_done_func_t *physdone, zio_done_func_t *done, void *private, zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb) { zio_t *zio; ASSERT(zp->zp_checksum >= ZIO_CHECKSUM_OFF && zp->zp_checksum < ZIO_CHECKSUM_FUNCTIONS && zp->zp_compress >= ZIO_COMPRESS_OFF && zp->zp_compress < ZIO_COMPRESS_FUNCTIONS && DMU_OT_IS_VALID(zp->zp_type) && zp->zp_level < 32 && zp->zp_copies > 0 && zp->zp_copies <= spa_max_replication(spa)); zio = zio_create(pio, spa, txg, bp, data, lsize, psize, done, private, ZIO_TYPE_WRITE, priority, flags, NULL, 0, zb, ZIO_STAGE_OPEN, (flags & ZIO_FLAG_DDT_CHILD) ? ZIO_DDT_CHILD_WRITE_PIPELINE : ZIO_WRITE_PIPELINE); zio->io_ready = ready; zio->io_children_ready = children_ready; zio->io_physdone = physdone; zio->io_prop = *zp; /* * Data can be NULL if we are going to call zio_write_override() to * provide the already-allocated BP. But we may need the data to * verify a dedup hit (if requested). In this case, don't try to * dedup (just take the already-allocated BP verbatim). */ if (data == NULL && zio->io_prop.zp_dedup_verify) { zio->io_prop.zp_dedup = zio->io_prop.zp_dedup_verify = B_FALSE; } return (zio); } zio_t * zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, void *data, uint64_t size, zio_done_func_t *done, void *private, zio_priority_t priority, enum zio_flag flags, zbookmark_phys_t *zb) { zio_t *zio; zio = zio_create(pio, spa, txg, bp, data, size, size, done, private, ZIO_TYPE_WRITE, priority, flags | ZIO_FLAG_IO_REWRITE, NULL, 0, zb, ZIO_STAGE_OPEN, ZIO_REWRITE_PIPELINE); return (zio); } void zio_write_override(zio_t *zio, blkptr_t *bp, int copies, boolean_t nopwrite) { ASSERT(zio->io_type == ZIO_TYPE_WRITE); ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); ASSERT(zio->io_stage == ZIO_STAGE_OPEN); ASSERT(zio->io_txg == spa_syncing_txg(zio->io_spa)); /* * We must reset the io_prop to match the values that existed * when the bp was first written by dmu_sync() keeping in mind * that nopwrite and dedup are mutually exclusive. */ zio->io_prop.zp_dedup = nopwrite ? B_FALSE : zio->io_prop.zp_dedup; zio->io_prop.zp_nopwrite = nopwrite; zio->io_prop.zp_copies = copies; zio->io_bp_override = bp; } void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp) { /* * The check for EMBEDDED is a performance optimization. We * process the free here (by ignoring it) rather than * putting it on the list and then processing it in zio_free_sync(). */ if (BP_IS_EMBEDDED(bp)) return; metaslab_check_free(spa, bp); /* * Frees that are for the currently-syncing txg, are not going to be * deferred, and which will not need to do a read (i.e. not GANG or * DEDUP), can be processed immediately. Otherwise, put them on the * in-memory list for later processing. */ if (BP_IS_GANG(bp) || BP_GET_DEDUP(bp) || txg != spa->spa_syncing_txg || spa_sync_pass(spa) >= zfs_sync_pass_deferred_free) { bplist_append(&spa->spa_free_bplist[txg & TXG_MASK], bp); } else { VERIFY0(zio_wait(zio_free_sync(NULL, spa, txg, bp, 0))); } } zio_t * zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, enum zio_flag flags) { zio_t *zio; enum zio_stage stage = ZIO_FREE_PIPELINE; ASSERT(!BP_IS_HOLE(bp)); ASSERT(spa_syncing_txg(spa) == txg); ASSERT(spa_sync_pass(spa) < zfs_sync_pass_deferred_free); if (BP_IS_EMBEDDED(bp)) return (zio_null(pio, spa, NULL, NULL, NULL, 0)); metaslab_check_free(spa, bp); arc_freed(spa, bp); /* * GANG and DEDUP blocks can induce a read (for the gang block header, * or the DDT), so issue them asynchronously so that this thread is * not tied up. */ if (BP_IS_GANG(bp) || BP_GET_DEDUP(bp)) stage |= ZIO_STAGE_ISSUE_ASYNC; zio = zio_create(pio, spa, txg, bp, NULL, BP_GET_PSIZE(bp), BP_GET_PSIZE(bp), NULL, NULL, ZIO_TYPE_FREE, ZIO_PRIORITY_NOW, flags, NULL, 0, NULL, ZIO_STAGE_OPEN, stage); return (zio); } zio_t * zio_claim(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, zio_done_func_t *done, void *private, enum zio_flag flags) { zio_t *zio; dprintf_bp(bp, "claiming in txg %llu", txg); if (BP_IS_EMBEDDED(bp)) return (zio_null(pio, spa, NULL, NULL, NULL, 0)); /* * A claim is an allocation of a specific block. Claims are needed * to support immediate writes in the intent log. The issue is that * immediate writes contain committed data, but in a txg that was * *not* committed. Upon opening the pool after an unclean shutdown, * the intent log claims all blocks that contain immediate write data * so that the SPA knows they're in use. * * All claims *must* be resolved in the first txg -- before the SPA * starts allocating blocks -- so that nothing is allocated twice. * If txg == 0 we just verify that the block is claimable. */ ASSERT3U(spa->spa_uberblock.ub_rootbp.blk_birth, <, spa_first_txg(spa)); ASSERT(txg == spa_first_txg(spa) || txg == 0); ASSERT(!BP_GET_DEDUP(bp) || !spa_writeable(spa)); /* zdb(1M) */ zio = zio_create(pio, spa, txg, bp, NULL, BP_GET_PSIZE(bp), BP_GET_PSIZE(bp), done, private, ZIO_TYPE_CLAIM, ZIO_PRIORITY_NOW, flags, NULL, 0, NULL, ZIO_STAGE_OPEN, ZIO_CLAIM_PIPELINE); ASSERT0(zio->io_queued_timestamp); return (zio); } zio_t * zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd, zio_done_func_t *done, void *private, enum zio_flag flags) { zio_t *zio; int c; if (vd->vdev_children == 0) { zio = zio_create(pio, spa, 0, NULL, NULL, 0, 0, done, private, ZIO_TYPE_IOCTL, ZIO_PRIORITY_NOW, flags, vd, 0, NULL, ZIO_STAGE_OPEN, ZIO_IOCTL_PIPELINE); zio->io_cmd = cmd; } else { zio = zio_null(pio, spa, NULL, NULL, NULL, flags); for (c = 0; c < vd->vdev_children; c++) zio_nowait(zio_ioctl(zio, spa, vd->vdev_child[c], cmd, done, private, flags)); } return (zio); } zio_t * zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size, void *data, int checksum, zio_done_func_t *done, void *private, zio_priority_t priority, enum zio_flag flags, boolean_t labels) { zio_t *zio; ASSERT(vd->vdev_children == 0); ASSERT(!labels || offset + size <= VDEV_LABEL_START_SIZE || offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE); ASSERT3U(offset + size, <=, vd->vdev_psize); zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, size, done, private, ZIO_TYPE_READ, priority, flags | ZIO_FLAG_PHYSICAL, vd, offset, NULL, ZIO_STAGE_OPEN, ZIO_READ_PHYS_PIPELINE); zio->io_prop.zp_checksum = checksum; return (zio); } zio_t * zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size, void *data, int checksum, zio_done_func_t *done, void *private, zio_priority_t priority, enum zio_flag flags, boolean_t labels) { zio_t *zio; ASSERT(vd->vdev_children == 0); ASSERT(!labels || offset + size <= VDEV_LABEL_START_SIZE || offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE); ASSERT3U(offset + size, <=, vd->vdev_psize); zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, size, done, private, ZIO_TYPE_WRITE, priority, flags | ZIO_FLAG_PHYSICAL, vd, offset, NULL, ZIO_STAGE_OPEN, ZIO_WRITE_PHYS_PIPELINE); zio->io_prop.zp_checksum = checksum; if (zio_checksum_table[checksum].ci_flags & ZCHECKSUM_FLAG_EMBEDDED) { /* * zec checksums are necessarily destructive -- they modify * the end of the write buffer to hold the verifier/checksum. * Therefore, we must make a local copy in case the data is * being written to multiple places in parallel. */ void *wbuf = zio_buf_alloc(size); bcopy(data, wbuf, size); zio_push_transform(zio, wbuf, size, size, NULL); } return (zio); } /* * Create a child I/O to do some work for us. */ zio_t * zio_vdev_child_io(zio_t *pio, blkptr_t *bp, vdev_t *vd, uint64_t offset, void *data, uint64_t size, int type, zio_priority_t priority, enum zio_flag flags, zio_done_func_t *done, void *private) { enum zio_stage pipeline = ZIO_VDEV_CHILD_PIPELINE; zio_t *zio; ASSERT(vd->vdev_parent == (pio->io_vd ? pio->io_vd : pio->io_spa->spa_root_vdev)); if (type == ZIO_TYPE_READ && bp != NULL) { /* * If we have the bp, then the child should perform the * checksum and the parent need not. This pushes error * detection as close to the leaves as possible and * eliminates redundant checksums in the interior nodes. */ pipeline |= ZIO_STAGE_CHECKSUM_VERIFY; pio->io_pipeline &= ~ZIO_STAGE_CHECKSUM_VERIFY; } if (vd->vdev_children == 0) offset += VDEV_LABEL_START_SIZE; flags |= ZIO_VDEV_CHILD_FLAGS(pio) | ZIO_FLAG_DONT_PROPAGATE; /* * If we've decided to do a repair, the write is not speculative -- * even if the original read was. */ if (flags & ZIO_FLAG_IO_REPAIR) flags &= ~ZIO_FLAG_SPECULATIVE; /* * If we're creating a child I/O that is not associated with a * top-level vdev, then the child zio is not an allocating I/O. * If this is a retried I/O then we ignore it since we will * have already processed the original allocating I/O. */ if (flags & ZIO_FLAG_IO_ALLOCATING && (vd != vd->vdev_top || (flags & ZIO_FLAG_IO_RETRY))) { metaslab_class_t *mc = spa_normal_class(pio->io_spa); ASSERT(mc->mc_alloc_throttle_enabled); ASSERT(type == ZIO_TYPE_WRITE); ASSERT(priority == ZIO_PRIORITY_ASYNC_WRITE); ASSERT(!(flags & ZIO_FLAG_IO_REPAIR)); ASSERT(!(pio->io_flags & ZIO_FLAG_IO_REWRITE) || pio->io_child_type == ZIO_CHILD_GANG); flags &= ~ZIO_FLAG_IO_ALLOCATING; } zio = zio_create(pio, pio->io_spa, pio->io_txg, bp, data, size, size, done, private, type, priority, flags, vd, offset, &pio->io_bookmark, ZIO_STAGE_VDEV_IO_START >> 1, pipeline); ASSERT3U(zio->io_child_type, ==, ZIO_CHILD_VDEV); zio->io_physdone = pio->io_physdone; if (vd->vdev_ops->vdev_op_leaf && zio->io_logical != NULL) zio->io_logical->io_phys_children++; return (zio); } zio_t * zio_vdev_delegated_io(vdev_t *vd, uint64_t offset, void *data, uint64_t size, int type, zio_priority_t priority, enum zio_flag flags, zio_done_func_t *done, void *private) { zio_t *zio; ASSERT(vd->vdev_ops->vdev_op_leaf); zio = zio_create(NULL, vd->vdev_spa, 0, NULL, data, size, size, done, private, type, priority, flags | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY | ZIO_FLAG_DELEGATED, vd, offset, NULL, ZIO_STAGE_VDEV_IO_START >> 1, ZIO_VDEV_CHILD_PIPELINE); return (zio); } void zio_flush(zio_t *zio, vdev_t *vd) { zio_nowait(zio_ioctl(zio, zio->io_spa, vd, DKIOCFLUSHWRITECACHE, NULL, NULL, ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY)); } void zio_shrink(zio_t *zio, uint64_t size) { ASSERT(zio->io_executor == NULL); ASSERT(zio->io_orig_size == zio->io_size); ASSERT(size <= zio->io_size); /* * We don't shrink for raidz because of problems with the * reconstruction when reading back less than the block size. * Note, BP_IS_RAIDZ() assumes no compression. */ ASSERT(BP_GET_COMPRESS(zio->io_bp) == ZIO_COMPRESS_OFF); if (!BP_IS_RAIDZ(zio->io_bp)) { /* we are not doing a raw write */ ASSERT3U(zio->io_size, ==, zio->io_lsize); zio->io_orig_size = zio->io_size = zio->io_lsize = size; } } /* * ========================================================================== * Prepare to read and write logical blocks * ========================================================================== */ static int zio_read_bp_init(zio_t *zio) { blkptr_t *bp = zio->io_bp; if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_OFF && zio->io_child_type == ZIO_CHILD_LOGICAL && !(zio->io_flags & ZIO_FLAG_RAW)) { uint64_t psize = BP_IS_EMBEDDED(bp) ? BPE_GET_PSIZE(bp) : BP_GET_PSIZE(bp); void *cbuf = zio_buf_alloc(psize); zio_push_transform(zio, cbuf, psize, psize, zio_decompress); } if (BP_IS_EMBEDDED(bp) && BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA) { zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; decode_embedded_bp_compressed(bp, zio->io_data); } else { ASSERT(!BP_IS_EMBEDDED(bp)); } if (!DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) && BP_GET_LEVEL(bp) == 0) zio->io_flags |= ZIO_FLAG_DONT_CACHE; if (BP_GET_TYPE(bp) == DMU_OT_DDT_ZAP) zio->io_flags |= ZIO_FLAG_DONT_CACHE; if (BP_GET_DEDUP(bp) && zio->io_child_type == ZIO_CHILD_LOGICAL) zio->io_pipeline = ZIO_DDT_READ_PIPELINE; return (ZIO_PIPELINE_CONTINUE); } static int zio_write_bp_init(zio_t *zio) { if (!IO_IS_ALLOCATING(zio)) return (ZIO_PIPELINE_CONTINUE); ASSERT(zio->io_child_type != ZIO_CHILD_DDT); if (zio->io_bp_override) { blkptr_t *bp = zio->io_bp; zio_prop_t *zp = &zio->io_prop; ASSERT(bp->blk_birth != zio->io_txg); ASSERT(BP_GET_DEDUP(zio->io_bp_override) == 0); *bp = *zio->io_bp_override; zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; if (BP_IS_EMBEDDED(bp)) return (ZIO_PIPELINE_CONTINUE); /* * If we've been overridden and nopwrite is set then * set the flag accordingly to indicate that a nopwrite * has already occurred. */ if (!BP_IS_HOLE(bp) && zp->zp_nopwrite) { ASSERT(!zp->zp_dedup); ASSERT3U(BP_GET_CHECKSUM(bp), ==, zp->zp_checksum); zio->io_flags |= ZIO_FLAG_NOPWRITE; return (ZIO_PIPELINE_CONTINUE); } ASSERT(!zp->zp_nopwrite); if (BP_IS_HOLE(bp) || !zp->zp_dedup) return (ZIO_PIPELINE_CONTINUE); ASSERT((zio_checksum_table[zp->zp_checksum].ci_flags & ZCHECKSUM_FLAG_DEDUP) || zp->zp_dedup_verify); if (BP_GET_CHECKSUM(bp) == zp->zp_checksum) { BP_SET_DEDUP(bp, 1); zio->io_pipeline |= ZIO_STAGE_DDT_WRITE; return (ZIO_PIPELINE_CONTINUE); } /* * We were unable to handle this as an override bp, treat * it as a regular write I/O. */ zio->io_bp_override = NULL; *bp = zio->io_bp_orig; zio->io_pipeline = zio->io_orig_pipeline; } return (ZIO_PIPELINE_CONTINUE); } static int zio_write_compress(zio_t *zio) { spa_t *spa = zio->io_spa; zio_prop_t *zp = &zio->io_prop; enum zio_compress compress = zp->zp_compress; blkptr_t *bp = zio->io_bp; uint64_t lsize = zio->io_lsize; uint64_t psize = zio->io_size; int pass = 1; EQUIV(lsize != psize, (zio->io_flags & ZIO_FLAG_RAW) != 0); /* * If our children haven't all reached the ready stage, * wait for them and then repeat this pipeline stage. */ if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_READY) || zio_wait_for_children(zio, ZIO_CHILD_LOGICAL, ZIO_WAIT_READY)) return (ZIO_PIPELINE_STOP); if (!IO_IS_ALLOCATING(zio)) return (ZIO_PIPELINE_CONTINUE); if (zio->io_children_ready != NULL) { /* * Now that all our children are ready, run the callback * associated with this zio in case it wants to modify the * data to be written. */ ASSERT3U(zp->zp_level, >, 0); zio->io_children_ready(zio); } ASSERT(zio->io_child_type != ZIO_CHILD_DDT); ASSERT(zio->io_bp_override == NULL); if (!BP_IS_HOLE(bp) && bp->blk_birth == zio->io_txg) { /* * We're rewriting an existing block, which means we're * working on behalf of spa_sync(). For spa_sync() to * converge, it must eventually be the case that we don't * have to allocate new blocks. But compression changes * the blocksize, which forces a reallocate, and makes * convergence take longer. Therefore, after the first * few passes, stop compressing to ensure convergence. */ pass = spa_sync_pass(spa); ASSERT(zio->io_txg == spa_syncing_txg(spa)); ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); ASSERT(!BP_GET_DEDUP(bp)); if (pass >= zfs_sync_pass_dont_compress) compress = ZIO_COMPRESS_OFF; /* Make sure someone doesn't change their mind on overwrites */ ASSERT(BP_IS_EMBEDDED(bp) || MIN(zp->zp_copies + BP_IS_GANG(bp), spa_max_replication(spa)) == BP_GET_NDVAS(bp)); } /* If it's a compressed write that is not raw, compress the buffer. */ if (compress != ZIO_COMPRESS_OFF && psize == lsize) { void *cbuf = zio_buf_alloc(lsize); psize = zio_compress_data(compress, zio->io_data, cbuf, lsize); if (psize == 0 || psize == lsize) { compress = ZIO_COMPRESS_OFF; zio_buf_free(cbuf, lsize); } else if (!zp->zp_dedup && psize <= BPE_PAYLOAD_SIZE && zp->zp_level == 0 && !DMU_OT_HAS_FILL(zp->zp_type) && spa_feature_is_enabled(spa, SPA_FEATURE_EMBEDDED_DATA)) { encode_embedded_bp_compressed(bp, cbuf, compress, lsize, psize); BPE_SET_ETYPE(bp, BP_EMBEDDED_TYPE_DATA); BP_SET_TYPE(bp, zio->io_prop.zp_type); BP_SET_LEVEL(bp, zio->io_prop.zp_level); zio_buf_free(cbuf, lsize); bp->blk_birth = zio->io_txg; zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; ASSERT(spa_feature_is_active(spa, SPA_FEATURE_EMBEDDED_DATA)); return (ZIO_PIPELINE_CONTINUE); } else { /* * Round up compressed size up to the ashift * of the smallest-ashift device, and zero the tail. * This ensures that the compressed size of the BP * (and thus compressratio property) are correct, * in that we charge for the padding used to fill out * the last sector. */ ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT); size_t rounded = (size_t)P2ROUNDUP(psize, 1ULL << spa->spa_min_ashift); if (rounded >= lsize) { compress = ZIO_COMPRESS_OFF; zio_buf_free(cbuf, lsize); psize = lsize; } else { bzero((char *)cbuf + psize, rounded - psize); psize = rounded; zio_push_transform(zio, cbuf, psize, lsize, NULL); } } /* * We were unable to handle this as an override bp, treat * it as a regular write I/O. */ zio->io_bp_override = NULL; *bp = zio->io_bp_orig; zio->io_pipeline = zio->io_orig_pipeline; } else { ASSERT3U(psize, !=, 0); } /* * The final pass of spa_sync() must be all rewrites, but the first * few passes offer a trade-off: allocating blocks defers convergence, * but newly allocated blocks are sequential, so they can be written * to disk faster. Therefore, we allow the first few passes of * spa_sync() to allocate new blocks, but force rewrites after that. * There should only be a handful of blocks after pass 1 in any case. */ if (!BP_IS_HOLE(bp) && bp->blk_birth == zio->io_txg && BP_GET_PSIZE(bp) == psize && pass >= zfs_sync_pass_rewrite) { ASSERT(psize != 0); enum zio_stage gang_stages = zio->io_pipeline & ZIO_GANG_STAGES; zio->io_pipeline = ZIO_REWRITE_PIPELINE | gang_stages; zio->io_flags |= ZIO_FLAG_IO_REWRITE; } else { BP_ZERO(bp); zio->io_pipeline = ZIO_WRITE_PIPELINE; } if (psize == 0) { if (zio->io_bp_orig.blk_birth != 0 && spa_feature_is_active(spa, SPA_FEATURE_HOLE_BIRTH)) { BP_SET_LSIZE(bp, lsize); BP_SET_TYPE(bp, zp->zp_type); BP_SET_LEVEL(bp, zp->zp_level); BP_SET_BIRTH(bp, zio->io_txg, 0); } zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; } else { ASSERT(zp->zp_checksum != ZIO_CHECKSUM_GANG_HEADER); BP_SET_LSIZE(bp, lsize); BP_SET_TYPE(bp, zp->zp_type); BP_SET_LEVEL(bp, zp->zp_level); BP_SET_PSIZE(bp, psize); BP_SET_COMPRESS(bp, compress); BP_SET_CHECKSUM(bp, zp->zp_checksum); BP_SET_DEDUP(bp, zp->zp_dedup); BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER); if (zp->zp_dedup) { ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE)); zio->io_pipeline = ZIO_DDT_WRITE_PIPELINE; } if (zp->zp_nopwrite) { ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE)); zio->io_pipeline |= ZIO_STAGE_NOP_WRITE; } } return (ZIO_PIPELINE_CONTINUE); } static int zio_free_bp_init(zio_t *zio) { blkptr_t *bp = zio->io_bp; if (zio->io_child_type == ZIO_CHILD_LOGICAL) { if (BP_GET_DEDUP(bp)) zio->io_pipeline = ZIO_DDT_FREE_PIPELINE; } return (ZIO_PIPELINE_CONTINUE); } /* * ========================================================================== * Execute the I/O pipeline * ========================================================================== */ static void zio_taskq_dispatch(zio_t *zio, zio_taskq_type_t q, boolean_t cutinline) { spa_t *spa = zio->io_spa; zio_type_t t = zio->io_type; int flags = (cutinline ? TQ_FRONT : 0); /* * If we're a config writer or a probe, the normal issue and * interrupt threads may all be blocked waiting for the config lock. * In this case, select the otherwise-unused taskq for ZIO_TYPE_NULL. */ if (zio->io_flags & (ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_PROBE)) t = ZIO_TYPE_NULL; /* * A similar issue exists for the L2ARC write thread until L2ARC 2.0. */ if (t == ZIO_TYPE_WRITE && zio->io_vd && zio->io_vd->vdev_aux) t = ZIO_TYPE_NULL; /* * If this is a high priority I/O, then use the high priority taskq if * available. */ if (zio->io_priority == ZIO_PRIORITY_NOW && spa->spa_zio_taskq[t][q + 1].stqs_count != 0) q++; ASSERT3U(q, <, ZIO_TASKQ_TYPES); /* * NB: We are assuming that the zio can only be dispatched * to a single taskq at a time. It would be a grievous error * to dispatch the zio to another taskq at the same time. */ ASSERT(zio->io_tqent.tqent_next == NULL); spa_taskq_dispatch_ent(spa, t, q, (task_func_t *)zio_execute, zio, flags, &zio->io_tqent); } static boolean_t zio_taskq_member(zio_t *zio, zio_taskq_type_t q) { kthread_t *executor = zio->io_executor; spa_t *spa = zio->io_spa; for (zio_type_t t = 0; t < ZIO_TYPES; t++) { spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; uint_t i; for (i = 0; i < tqs->stqs_count; i++) { if (taskq_member(tqs->stqs_taskq[i], executor)) return (B_TRUE); } } return (B_FALSE); } static int zio_issue_async(zio_t *zio) { zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_FALSE); return (ZIO_PIPELINE_STOP); } void zio_interrupt(zio_t *zio) { zio_taskq_dispatch(zio, ZIO_TASKQ_INTERRUPT, B_FALSE); } void zio_delay_interrupt(zio_t *zio) { /* * The timeout_generic() function isn't defined in userspace, so * rather than trying to implement the function, the zio delay * functionality has been disabled for userspace builds. */ #ifdef _KERNEL /* * If io_target_timestamp is zero, then no delay has been registered * for this IO, thus jump to the end of this function and "skip" the * delay; issuing it directly to the zio layer. */ if (zio->io_target_timestamp != 0) { hrtime_t now = gethrtime(); if (now >= zio->io_target_timestamp) { /* * This IO has already taken longer than the target * delay to complete, so we don't want to delay it * any longer; we "miss" the delay and issue it * directly to the zio layer. This is likely due to * the target latency being set to a value less than * the underlying hardware can satisfy (e.g. delay * set to 1ms, but the disks take 10ms to complete an * IO request). */ DTRACE_PROBE2(zio__delay__miss, zio_t *, zio, hrtime_t, now); zio_interrupt(zio); } else { hrtime_t diff = zio->io_target_timestamp - now; DTRACE_PROBE3(zio__delay__hit, zio_t *, zio, hrtime_t, now, hrtime_t, diff); (void) timeout_generic(CALLOUT_NORMAL, (void (*)(void *))zio_interrupt, zio, diff, 1, 0); } return; } #endif DTRACE_PROBE1(zio__delay__skip, zio_t *, zio); zio_interrupt(zio); } /* * Execute the I/O pipeline until one of the following occurs: * * (1) the I/O completes * (2) the pipeline stalls waiting for dependent child I/Os * (3) the I/O issues, so we're waiting for an I/O completion interrupt * (4) the I/O is delegated by vdev-level caching or aggregation * (5) the I/O is deferred due to vdev-level queueing * (6) the I/O is handed off to another thread. * * In all cases, the pipeline stops whenever there's no CPU work; it never * burns a thread in cv_wait(). * * There's no locking on io_stage because there's no legitimate way * for multiple threads to be attempting to process the same I/O. */ static zio_pipe_stage_t *zio_pipeline[]; void zio_execute(zio_t *zio) { zio->io_executor = curthread; ASSERT3U(zio->io_queued_timestamp, >, 0); while (zio->io_stage < ZIO_STAGE_DONE) { enum zio_stage pipeline = zio->io_pipeline; enum zio_stage stage = zio->io_stage; int rv; ASSERT(!MUTEX_HELD(&zio->io_lock)); ASSERT(ISP2(stage)); ASSERT(zio->io_stall == NULL); do { stage <<= 1; } while ((stage & pipeline) == 0); ASSERT(stage <= ZIO_STAGE_DONE); /* * If we are in interrupt context and this pipeline stage * will grab a config lock that is held across I/O, * or may wait for an I/O that needs an interrupt thread * to complete, issue async to avoid deadlock. * * For VDEV_IO_START, we cut in line so that the io will * be sent to disk promptly. */ if ((stage & ZIO_BLOCKING_STAGES) && zio->io_vd == NULL && zio_taskq_member(zio, ZIO_TASKQ_INTERRUPT)) { boolean_t cut = (stage == ZIO_STAGE_VDEV_IO_START) ? zio_requeue_io_start_cut_in_line : B_FALSE; zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, cut); return; } zio->io_stage = stage; zio->io_pipeline_trace |= zio->io_stage; rv = zio_pipeline[highbit64(stage) - 1](zio); if (rv == ZIO_PIPELINE_STOP) return; ASSERT(rv == ZIO_PIPELINE_CONTINUE); } } /* * ========================================================================== * Initiate I/O, either sync or async * ========================================================================== */ int zio_wait(zio_t *zio) { int error; ASSERT(zio->io_stage == ZIO_STAGE_OPEN); ASSERT(zio->io_executor == NULL); zio->io_waiter = curthread; ASSERT0(zio->io_queued_timestamp); zio->io_queued_timestamp = gethrtime(); zio_execute(zio); mutex_enter(&zio->io_lock); while (zio->io_executor != NULL) cv_wait(&zio->io_cv, &zio->io_lock); mutex_exit(&zio->io_lock); error = zio->io_error; zio_destroy(zio); return (error); } void zio_nowait(zio_t *zio) { ASSERT(zio->io_executor == NULL); if (zio->io_child_type == ZIO_CHILD_LOGICAL && zio_unique_parent(zio) == NULL) { /* * This is a logical async I/O with no parent to wait for it. * We add it to the spa_async_root_zio "Godfather" I/O which * will ensure they complete prior to unloading the pool. */ spa_t *spa = zio->io_spa; zio_add_child(spa->spa_async_zio_root[CPU_SEQID], zio); } ASSERT0(zio->io_queued_timestamp); zio->io_queued_timestamp = gethrtime(); zio_execute(zio); } /* * ========================================================================== * Reexecute or suspend/resume failed I/O * ========================================================================== */ static void zio_reexecute(zio_t *pio) { zio_t *cio, *cio_next; ASSERT(pio->io_child_type == ZIO_CHILD_LOGICAL); ASSERT(pio->io_orig_stage == ZIO_STAGE_OPEN); ASSERT(pio->io_gang_leader == NULL); ASSERT(pio->io_gang_tree == NULL); pio->io_flags = pio->io_orig_flags; pio->io_stage = pio->io_orig_stage; pio->io_pipeline = pio->io_orig_pipeline; pio->io_reexecute = 0; pio->io_flags |= ZIO_FLAG_REEXECUTED; pio->io_pipeline_trace = 0; pio->io_error = 0; for (int w = 0; w < ZIO_WAIT_TYPES; w++) pio->io_state[w] = 0; for (int c = 0; c < ZIO_CHILD_TYPES; c++) pio->io_child_error[c] = 0; if (IO_IS_ALLOCATING(pio)) BP_ZERO(pio->io_bp); /* * As we reexecute pio's children, new children could be created. * New children go to the head of pio's io_child_list, however, * so we will (correctly) not reexecute them. The key is that * the remainder of pio's io_child_list, from 'cio_next' onward, * cannot be affected by any side effects of reexecuting 'cio'. */ zio_link_t *zl = NULL; for (cio = zio_walk_children(pio, &zl); cio != NULL; cio = cio_next) { cio_next = zio_walk_children(pio, &zl); mutex_enter(&pio->io_lock); for (int w = 0; w < ZIO_WAIT_TYPES; w++) pio->io_children[cio->io_child_type][w]++; mutex_exit(&pio->io_lock); zio_reexecute(cio); } /* * Now that all children have been reexecuted, execute the parent. * We don't reexecute "The Godfather" I/O here as it's the * responsibility of the caller to wait on it. */ if (!(pio->io_flags & ZIO_FLAG_GODFATHER)) { pio->io_queued_timestamp = gethrtime(); zio_execute(pio); } } void zio_suspend(spa_t *spa, zio_t *zio) { if (spa_get_failmode(spa) == ZIO_FAILURE_MODE_PANIC) fm_panic("Pool '%s' has encountered an uncorrectable I/O " "failure and the failure mode property for this pool " "is set to panic.", spa_name(spa)); zfs_ereport_post(FM_EREPORT_ZFS_IO_FAILURE, spa, NULL, NULL, 0, 0); mutex_enter(&spa->spa_suspend_lock); if (spa->spa_suspend_zio_root == NULL) spa->spa_suspend_zio_root = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER); spa->spa_suspended = B_TRUE; if (zio != NULL) { ASSERT(!(zio->io_flags & ZIO_FLAG_GODFATHER)); ASSERT(zio != spa->spa_suspend_zio_root); ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); ASSERT(zio_unique_parent(zio) == NULL); ASSERT(zio->io_stage == ZIO_STAGE_DONE); zio_add_child(spa->spa_suspend_zio_root, zio); } mutex_exit(&spa->spa_suspend_lock); } int zio_resume(spa_t *spa) { zio_t *pio; /* * Reexecute all previously suspended i/o. */ mutex_enter(&spa->spa_suspend_lock); spa->spa_suspended = B_FALSE; cv_broadcast(&spa->spa_suspend_cv); pio = spa->spa_suspend_zio_root; spa->spa_suspend_zio_root = NULL; mutex_exit(&spa->spa_suspend_lock); if (pio == NULL) return (0); zio_reexecute(pio); return (zio_wait(pio)); } void zio_resume_wait(spa_t *spa) { mutex_enter(&spa->spa_suspend_lock); while (spa_suspended(spa)) cv_wait(&spa->spa_suspend_cv, &spa->spa_suspend_lock); mutex_exit(&spa->spa_suspend_lock); } /* * ========================================================================== * Gang blocks. * * A gang block is a collection of small blocks that looks to the DMU * like one large block. When zio_dva_allocate() cannot find a block * of the requested size, due to either severe fragmentation or the pool * being nearly full, it calls zio_write_gang_block() to construct the * block from smaller fragments. * * A gang block consists of a gang header (zio_gbh_phys_t) and up to * three (SPA_GBH_NBLKPTRS) gang members. The gang header is just like * an indirect block: it's an array of block pointers. It consumes * only one sector and hence is allocatable regardless of fragmentation. * The gang header's bps point to its gang members, which hold the data. * * Gang blocks are self-checksumming, using the bp's * as the verifier to ensure uniqueness of the SHA256 checksum. * Critically, the gang block bp's blk_cksum is the checksum of the data, * not the gang header. This ensures that data block signatures (needed for * deduplication) are independent of how the block is physically stored. * * Gang blocks can be nested: a gang member may itself be a gang block. * Thus every gang block is a tree in which root and all interior nodes are * gang headers, and the leaves are normal blocks that contain user data. * The root of the gang tree is called the gang leader. * * To perform any operation (read, rewrite, free, claim) on a gang block, * zio_gang_assemble() first assembles the gang tree (minus data leaves) * in the io_gang_tree field of the original logical i/o by recursively * reading the gang leader and all gang headers below it. This yields * an in-core tree containing the contents of every gang header and the * bps for every constituent of the gang block. * * With the gang tree now assembled, zio_gang_issue() just walks the gang tree * and invokes a callback on each bp. To free a gang block, zio_gang_issue() * calls zio_free_gang() -- a trivial wrapper around zio_free() -- for each bp. * zio_claim_gang() provides a similarly trivial wrapper for zio_claim(). * zio_read_gang() is a wrapper around zio_read() that omits reading gang * headers, since we already have those in io_gang_tree. zio_rewrite_gang() * performs a zio_rewrite() of the data or, for gang headers, a zio_rewrite() * of the gang header plus zio_checksum_compute() of the data to update the * gang header's blk_cksum as described above. * * The two-phase assemble/issue model solves the problem of partial failure -- * what if you'd freed part of a gang block but then couldn't read the * gang header for another part? Assembling the entire gang tree first * ensures that all the necessary gang header I/O has succeeded before * starting the actual work of free, claim, or write. Once the gang tree * is assembled, free and claim are in-memory operations that cannot fail. * * In the event that a gang write fails, zio_dva_unallocate() walks the * gang tree to immediately free (i.e. insert back into the space map) * everything we've allocated. This ensures that we don't get ENOSPC * errors during repeated suspend/resume cycles due to a flaky device. * * Gang rewrites only happen during sync-to-convergence. If we can't assemble * the gang tree, we won't modify the block, so we can safely defer the free * (knowing that the block is still intact). If we *can* assemble the gang * tree, then even if some of the rewrites fail, zio_dva_unallocate() will free * each constituent bp and we can allocate a new block on the next sync pass. * * In all cases, the gang tree allows complete recovery from partial failure. * ========================================================================== */ static zio_t * zio_read_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) { if (gn != NULL) return (pio); return (zio_read(pio, pio->io_spa, bp, data, BP_GET_PSIZE(bp), NULL, NULL, pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark)); } zio_t * zio_rewrite_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) { zio_t *zio; if (gn != NULL) { zio = zio_rewrite(pio, pio->io_spa, pio->io_txg, bp, gn->gn_gbh, SPA_GANGBLOCKSIZE, NULL, NULL, pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); /* * As we rewrite each gang header, the pipeline will compute * a new gang block header checksum for it; but no one will * compute a new data checksum, so we do that here. The one * exception is the gang leader: the pipeline already computed * its data checksum because that stage precedes gang assembly. * (Presently, nothing actually uses interior data checksums; * this is just good hygiene.) */ if (gn != pio->io_gang_leader->io_gang_tree) { zio_checksum_compute(zio, BP_GET_CHECKSUM(bp), data, BP_GET_PSIZE(bp)); } /* * If we are here to damage data for testing purposes, * leave the GBH alone so that we can detect the damage. */ if (pio->io_gang_leader->io_flags & ZIO_FLAG_INDUCE_DAMAGE) zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES; } else { zio = zio_rewrite(pio, pio->io_spa, pio->io_txg, bp, data, BP_GET_PSIZE(bp), NULL, NULL, pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); } return (zio); } /* ARGSUSED */ zio_t * zio_free_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) { return (zio_free_sync(pio, pio->io_spa, pio->io_txg, bp, ZIO_GANG_CHILD_FLAGS(pio))); } /* ARGSUSED */ zio_t * zio_claim_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) { return (zio_claim(pio, pio->io_spa, pio->io_txg, bp, NULL, NULL, ZIO_GANG_CHILD_FLAGS(pio))); } static zio_gang_issue_func_t *zio_gang_issue_func[ZIO_TYPES] = { NULL, zio_read_gang, zio_rewrite_gang, zio_free_gang, zio_claim_gang, NULL }; static void zio_gang_tree_assemble_done(zio_t *zio); static zio_gang_node_t * zio_gang_node_alloc(zio_gang_node_t **gnpp) { zio_gang_node_t *gn; ASSERT(*gnpp == NULL); gn = kmem_zalloc(sizeof (*gn), KM_SLEEP); gn->gn_gbh = zio_buf_alloc(SPA_GANGBLOCKSIZE); *gnpp = gn; return (gn); } static void zio_gang_node_free(zio_gang_node_t **gnpp) { zio_gang_node_t *gn = *gnpp; for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) ASSERT(gn->gn_child[g] == NULL); zio_buf_free(gn->gn_gbh, SPA_GANGBLOCKSIZE); kmem_free(gn, sizeof (*gn)); *gnpp = NULL; } static void zio_gang_tree_free(zio_gang_node_t **gnpp) { zio_gang_node_t *gn = *gnpp; if (gn == NULL) return; for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) zio_gang_tree_free(&gn->gn_child[g]); zio_gang_node_free(gnpp); } static void zio_gang_tree_assemble(zio_t *gio, blkptr_t *bp, zio_gang_node_t **gnpp) { zio_gang_node_t *gn = zio_gang_node_alloc(gnpp); ASSERT(gio->io_gang_leader == gio); ASSERT(BP_IS_GANG(bp)); zio_nowait(zio_read(gio, gio->io_spa, bp, gn->gn_gbh, SPA_GANGBLOCKSIZE, zio_gang_tree_assemble_done, gn, gio->io_priority, ZIO_GANG_CHILD_FLAGS(gio), &gio->io_bookmark)); } static void zio_gang_tree_assemble_done(zio_t *zio) { zio_t *gio = zio->io_gang_leader; zio_gang_node_t *gn = zio->io_private; blkptr_t *bp = zio->io_bp; ASSERT(gio == zio_unique_parent(zio)); ASSERT(zio->io_child_count == 0); if (zio->io_error) return; if (BP_SHOULD_BYTESWAP(bp)) byteswap_uint64_array(zio->io_data, zio->io_size); ASSERT(zio->io_data == gn->gn_gbh); ASSERT(zio->io_size == SPA_GANGBLOCKSIZE); ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC); for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) { blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g]; if (!BP_IS_GANG(gbp)) continue; zio_gang_tree_assemble(gio, gbp, &gn->gn_child[g]); } } static void zio_gang_tree_issue(zio_t *pio, zio_gang_node_t *gn, blkptr_t *bp, void *data) { zio_t *gio = pio->io_gang_leader; zio_t *zio; ASSERT(BP_IS_GANG(bp) == !!gn); ASSERT(BP_GET_CHECKSUM(bp) == BP_GET_CHECKSUM(gio->io_bp)); ASSERT(BP_GET_LSIZE(bp) == BP_GET_PSIZE(bp) || gn == gio->io_gang_tree); /* * If you're a gang header, your data is in gn->gn_gbh. * If you're a gang member, your data is in 'data' and gn == NULL. */ zio = zio_gang_issue_func[gio->io_type](pio, bp, gn, data); if (gn != NULL) { ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC); for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) { blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g]; if (BP_IS_HOLE(gbp)) continue; zio_gang_tree_issue(zio, gn->gn_child[g], gbp, data); data = (char *)data + BP_GET_PSIZE(gbp); } } if (gn == gio->io_gang_tree) ASSERT3P((char *)gio->io_data + gio->io_size, ==, data); if (zio != pio) zio_nowait(zio); } static int zio_gang_assemble(zio_t *zio) { blkptr_t *bp = zio->io_bp; ASSERT(BP_IS_GANG(bp) && zio->io_gang_leader == NULL); ASSERT(zio->io_child_type > ZIO_CHILD_GANG); zio->io_gang_leader = zio; zio_gang_tree_assemble(zio, bp, &zio->io_gang_tree); return (ZIO_PIPELINE_CONTINUE); } static int zio_gang_issue(zio_t *zio) { blkptr_t *bp = zio->io_bp; if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_DONE)) return (ZIO_PIPELINE_STOP); ASSERT(BP_IS_GANG(bp) && zio->io_gang_leader == zio); ASSERT(zio->io_child_type > ZIO_CHILD_GANG); if (zio->io_child_error[ZIO_CHILD_GANG] == 0) zio_gang_tree_issue(zio, zio->io_gang_tree, bp, zio->io_data); else zio_gang_tree_free(&zio->io_gang_tree); zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; return (ZIO_PIPELINE_CONTINUE); } static void zio_write_gang_member_ready(zio_t *zio) { zio_t *pio = zio_unique_parent(zio); zio_t *gio = zio->io_gang_leader; dva_t *cdva = zio->io_bp->blk_dva; dva_t *pdva = pio->io_bp->blk_dva; uint64_t asize; if (BP_IS_HOLE(zio->io_bp)) return; ASSERT(BP_IS_HOLE(&zio->io_bp_orig)); ASSERT(zio->io_child_type == ZIO_CHILD_GANG); ASSERT3U(zio->io_prop.zp_copies, ==, gio->io_prop.zp_copies); ASSERT3U(zio->io_prop.zp_copies, <=, BP_GET_NDVAS(zio->io_bp)); ASSERT3U(pio->io_prop.zp_copies, <=, BP_GET_NDVAS(pio->io_bp)); ASSERT3U(BP_GET_NDVAS(zio->io_bp), <=, BP_GET_NDVAS(pio->io_bp)); mutex_enter(&pio->io_lock); for (int d = 0; d < BP_GET_NDVAS(zio->io_bp); d++) { ASSERT(DVA_GET_GANG(&pdva[d])); asize = DVA_GET_ASIZE(&pdva[d]); asize += DVA_GET_ASIZE(&cdva[d]); DVA_SET_ASIZE(&pdva[d], asize); } mutex_exit(&pio->io_lock); } static int zio_write_gang_block(zio_t *pio) { spa_t *spa = pio->io_spa; metaslab_class_t *mc = spa_normal_class(spa); blkptr_t *bp = pio->io_bp; zio_t *gio = pio->io_gang_leader; zio_t *zio; zio_gang_node_t *gn, **gnpp; zio_gbh_phys_t *gbh; uint64_t txg = pio->io_txg; uint64_t resid = pio->io_size; uint64_t lsize; int copies = gio->io_prop.zp_copies; int gbh_copies = MIN(copies + 1, spa_max_replication(spa)); zio_prop_t zp; int error; int flags = METASLAB_HINTBP_FAVOR | METASLAB_GANG_HEADER; if (pio->io_flags & ZIO_FLAG_IO_ALLOCATING) { ASSERT(pio->io_priority == ZIO_PRIORITY_ASYNC_WRITE); ASSERT(!(pio->io_flags & ZIO_FLAG_NODATA)); flags |= METASLAB_ASYNC_ALLOC; VERIFY(refcount_held(&mc->mc_alloc_slots, pio)); /* * The logical zio has already placed a reservation for * 'copies' allocation slots but gang blocks may require * additional copies. These additional copies * (i.e. gbh_copies - copies) are guaranteed to succeed * since metaslab_class_throttle_reserve() always allows * additional reservations for gang blocks. */ VERIFY(metaslab_class_throttle_reserve(mc, gbh_copies - copies, pio, flags)); } error = metaslab_alloc(spa, mc, SPA_GANGBLOCKSIZE, bp, gbh_copies, txg, pio == gio ? NULL : gio->io_bp, flags, &pio->io_alloc_list, pio); if (error) { if (pio->io_flags & ZIO_FLAG_IO_ALLOCATING) { ASSERT(pio->io_priority == ZIO_PRIORITY_ASYNC_WRITE); ASSERT(!(pio->io_flags & ZIO_FLAG_NODATA)); /* * If we failed to allocate the gang block header then * we remove any additional allocation reservations that * we placed here. The original reservation will * be removed when the logical I/O goes to the ready * stage. */ metaslab_class_throttle_unreserve(mc, gbh_copies - copies, pio); } pio->io_error = error; return (ZIO_PIPELINE_CONTINUE); } if (pio == gio) { gnpp = &gio->io_gang_tree; } else { gnpp = pio->io_private; ASSERT(pio->io_ready == zio_write_gang_member_ready); } gn = zio_gang_node_alloc(gnpp); gbh = gn->gn_gbh; bzero(gbh, SPA_GANGBLOCKSIZE); /* * Create the gang header. */ zio = zio_rewrite(pio, spa, txg, bp, gbh, SPA_GANGBLOCKSIZE, NULL, NULL, pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); /* * Create and nowait the gang children. */ for (int g = 0; resid != 0; resid -= lsize, g++) { lsize = P2ROUNDUP(resid / (SPA_GBH_NBLKPTRS - g), SPA_MINBLOCKSIZE); ASSERT(lsize >= SPA_MINBLOCKSIZE && lsize <= resid); zp.zp_checksum = gio->io_prop.zp_checksum; zp.zp_compress = ZIO_COMPRESS_OFF; zp.zp_type = DMU_OT_NONE; zp.zp_level = 0; zp.zp_copies = gio->io_prop.zp_copies; zp.zp_dedup = B_FALSE; zp.zp_dedup_verify = B_FALSE; zp.zp_nopwrite = B_FALSE; zio_t *cio = zio_write(zio, spa, txg, &gbh->zg_blkptr[g], (char *)pio->io_data + (pio->io_size - resid), lsize, lsize, &zp, zio_write_gang_member_ready, NULL, NULL, NULL, &gn->gn_child[g], pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); if (pio->io_flags & ZIO_FLAG_IO_ALLOCATING) { ASSERT(pio->io_priority == ZIO_PRIORITY_ASYNC_WRITE); ASSERT(!(pio->io_flags & ZIO_FLAG_NODATA)); /* * Gang children won't throttle but we should * account for their work, so reserve an allocation * slot for them here. */ VERIFY(metaslab_class_throttle_reserve(mc, zp.zp_copies, cio, flags)); } zio_nowait(cio); } /* * Set pio's pipeline to just wait for zio to finish. */ pio->io_pipeline = ZIO_INTERLOCK_PIPELINE; zio_nowait(zio); return (ZIO_PIPELINE_CONTINUE); } /* * The zio_nop_write stage in the pipeline determines if allocating a * new bp is necessary. The nopwrite feature can handle writes in * either syncing or open context (i.e. zil writes) and as a result is * mutually exclusive with dedup. * * By leveraging a cryptographically secure checksum, such as SHA256, we * can compare the checksums of the new data and the old to determine if * allocating a new block is required. Note that our requirements for * cryptographic strength are fairly weak: there can't be any accidental * hash collisions, but we don't need to be secure against intentional * (malicious) collisions. To trigger a nopwrite, you have to be able * to write the file to begin with, and triggering an incorrect (hash * collision) nopwrite is no worse than simply writing to the file. * That said, there are no known attacks against the checksum algorithms * used for nopwrite, assuming that the salt and the checksums * themselves remain secret. */ static int zio_nop_write(zio_t *zio) { blkptr_t *bp = zio->io_bp; blkptr_t *bp_orig = &zio->io_bp_orig; zio_prop_t *zp = &zio->io_prop; ASSERT(BP_GET_LEVEL(bp) == 0); ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE)); ASSERT(zp->zp_nopwrite); ASSERT(!zp->zp_dedup); ASSERT(zio->io_bp_override == NULL); ASSERT(IO_IS_ALLOCATING(zio)); /* * Check to see if the original bp and the new bp have matching * characteristics (i.e. same checksum, compression algorithms, etc). * If they don't then just continue with the pipeline which will * allocate a new bp. */ if (BP_IS_HOLE(bp_orig) || !(zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_flags & ZCHECKSUM_FLAG_NOPWRITE) || BP_GET_CHECKSUM(bp) != BP_GET_CHECKSUM(bp_orig) || BP_GET_COMPRESS(bp) != BP_GET_COMPRESS(bp_orig) || BP_GET_DEDUP(bp) != BP_GET_DEDUP(bp_orig) || zp->zp_copies != BP_GET_NDVAS(bp_orig)) return (ZIO_PIPELINE_CONTINUE); /* * If the checksums match then reset the pipeline so that we * avoid allocating a new bp and issuing any I/O. */ if (ZIO_CHECKSUM_EQUAL(bp->blk_cksum, bp_orig->blk_cksum)) { ASSERT(zio_checksum_table[zp->zp_checksum].ci_flags & ZCHECKSUM_FLAG_NOPWRITE); ASSERT3U(BP_GET_PSIZE(bp), ==, BP_GET_PSIZE(bp_orig)); ASSERT3U(BP_GET_LSIZE(bp), ==, BP_GET_LSIZE(bp_orig)); ASSERT(zp->zp_compress != ZIO_COMPRESS_OFF); ASSERT(bcmp(&bp->blk_prop, &bp_orig->blk_prop, sizeof (uint64_t)) == 0); *bp = *bp_orig; zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; zio->io_flags |= ZIO_FLAG_NOPWRITE; } return (ZIO_PIPELINE_CONTINUE); } /* * ========================================================================== * Dedup * ========================================================================== */ static void zio_ddt_child_read_done(zio_t *zio) { blkptr_t *bp = zio->io_bp; ddt_entry_t *dde = zio->io_private; ddt_phys_t *ddp; zio_t *pio = zio_unique_parent(zio); mutex_enter(&pio->io_lock); ddp = ddt_phys_select(dde, bp); if (zio->io_error == 0) ddt_phys_clear(ddp); /* this ddp doesn't need repair */ if (zio->io_error == 0 && dde->dde_repair_data == NULL) dde->dde_repair_data = zio->io_data; else zio_buf_free(zio->io_data, zio->io_size); mutex_exit(&pio->io_lock); } static int zio_ddt_read_start(zio_t *zio) { blkptr_t *bp = zio->io_bp; ASSERT(BP_GET_DEDUP(bp)); ASSERT(BP_GET_PSIZE(bp) == zio->io_size); ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); if (zio->io_child_error[ZIO_CHILD_DDT]) { ddt_t *ddt = ddt_select(zio->io_spa, bp); ddt_entry_t *dde = ddt_repair_start(ddt, bp); ddt_phys_t *ddp = dde->dde_phys; ddt_phys_t *ddp_self = ddt_phys_select(dde, bp); blkptr_t blk; ASSERT(zio->io_vsd == NULL); zio->io_vsd = dde; if (ddp_self == NULL) return (ZIO_PIPELINE_CONTINUE); for (int p = 0; p < DDT_PHYS_TYPES; p++, ddp++) { if (ddp->ddp_phys_birth == 0 || ddp == ddp_self) continue; ddt_bp_create(ddt->ddt_checksum, &dde->dde_key, ddp, &blk); zio_nowait(zio_read(zio, zio->io_spa, &blk, zio_buf_alloc(zio->io_size), zio->io_size, zio_ddt_child_read_done, dde, zio->io_priority, ZIO_DDT_CHILD_FLAGS(zio) | ZIO_FLAG_DONT_PROPAGATE, &zio->io_bookmark)); } return (ZIO_PIPELINE_CONTINUE); } zio_nowait(zio_read(zio, zio->io_spa, bp, zio->io_data, zio->io_size, NULL, NULL, zio->io_priority, ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark)); return (ZIO_PIPELINE_CONTINUE); } static int zio_ddt_read_done(zio_t *zio) { blkptr_t *bp = zio->io_bp; if (zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_DONE)) return (ZIO_PIPELINE_STOP); ASSERT(BP_GET_DEDUP(bp)); ASSERT(BP_GET_PSIZE(bp) == zio->io_size); ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); if (zio->io_child_error[ZIO_CHILD_DDT]) { ddt_t *ddt = ddt_select(zio->io_spa, bp); ddt_entry_t *dde = zio->io_vsd; if (ddt == NULL) { ASSERT(spa_load_state(zio->io_spa) != SPA_LOAD_NONE); return (ZIO_PIPELINE_CONTINUE); } if (dde == NULL) { zio->io_stage = ZIO_STAGE_DDT_READ_START >> 1; zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_FALSE); return (ZIO_PIPELINE_STOP); } if (dde->dde_repair_data != NULL) { bcopy(dde->dde_repair_data, zio->io_data, zio->io_size); zio->io_child_error[ZIO_CHILD_DDT] = 0; } ddt_repair_done(ddt, dde); zio->io_vsd = NULL; } ASSERT(zio->io_vsd == NULL); return (ZIO_PIPELINE_CONTINUE); } static boolean_t zio_ddt_collision(zio_t *zio, ddt_t *ddt, ddt_entry_t *dde) { spa_t *spa = zio->io_spa; boolean_t do_raw = (zio->io_flags & ZIO_FLAG_RAW); /* We should never get a raw, override zio */ ASSERT(!(zio->io_bp_override && do_raw)); /* * Note: we compare the original data, not the transformed data, * because when zio->io_bp is an override bp, we will not have * pushed the I/O transforms. That's an important optimization * because otherwise we'd compress/encrypt all dmu_sync() data twice. */ for (int p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) { zio_t *lio = dde->dde_lead_zio[p]; if (lio != NULL) { return (lio->io_orig_size != zio->io_orig_size || bcmp(zio->io_orig_data, lio->io_orig_data, zio->io_orig_size) != 0); } } for (int p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) { ddt_phys_t *ddp = &dde->dde_phys[p]; if (ddp->ddp_phys_birth != 0) { arc_buf_t *abuf = NULL; arc_flags_t aflags = ARC_FLAG_WAIT; int zio_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE; blkptr_t blk = *zio->io_bp; int error; ddt_bp_fill(ddp, &blk, ddp->ddp_phys_birth); ddt_exit(ddt); /* * Intuitively, it would make more sense to compare * io_data than io_orig_data in the raw case since you * don't want to look at any transformations that have * happened to the data. However, for raw I/Os the * data will actually be the same in io_data and * io_orig_data, so all we have to do is issue this as * a raw ARC read. */ if (do_raw) { zio_flags |= ZIO_FLAG_RAW; ASSERT3U(zio->io_size, ==, zio->io_orig_size); ASSERT0(bcmp(zio->io_data, zio->io_orig_data, zio->io_size)); ASSERT3P(zio->io_transform_stack, ==, NULL); } error = arc_read(NULL, spa, &blk, arc_getbuf_func, &abuf, ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zio->io_bookmark); if (error == 0) { if (arc_buf_size(abuf) != zio->io_orig_size || bcmp(abuf->b_data, zio->io_orig_data, zio->io_orig_size) != 0) error = SET_ERROR(EEXIST); arc_buf_destroy(abuf, &abuf); } ddt_enter(ddt); return (error != 0); } } return (B_FALSE); } static void zio_ddt_child_write_ready(zio_t *zio) { int p = zio->io_prop.zp_copies; ddt_t *ddt = ddt_select(zio->io_spa, zio->io_bp); ddt_entry_t *dde = zio->io_private; ddt_phys_t *ddp = &dde->dde_phys[p]; zio_t *pio; if (zio->io_error) return; ddt_enter(ddt); ASSERT(dde->dde_lead_zio[p] == zio); ddt_phys_fill(ddp, zio->io_bp); zio_link_t *zl = NULL; while ((pio = zio_walk_parents(zio, &zl)) != NULL) ddt_bp_fill(ddp, pio->io_bp, zio->io_txg); ddt_exit(ddt); } static void zio_ddt_child_write_done(zio_t *zio) { int p = zio->io_prop.zp_copies; ddt_t *ddt = ddt_select(zio->io_spa, zio->io_bp); ddt_entry_t *dde = zio->io_private; ddt_phys_t *ddp = &dde->dde_phys[p]; ddt_enter(ddt); ASSERT(ddp->ddp_refcnt == 0); ASSERT(dde->dde_lead_zio[p] == zio); dde->dde_lead_zio[p] = NULL; if (zio->io_error == 0) { zio_link_t *zl = NULL; while (zio_walk_parents(zio, &zl) != NULL) ddt_phys_addref(ddp); } else { ddt_phys_clear(ddp); } ddt_exit(ddt); } static void zio_ddt_ditto_write_done(zio_t *zio) { int p = DDT_PHYS_DITTO; zio_prop_t *zp = &zio->io_prop; blkptr_t *bp = zio->io_bp; ddt_t *ddt = ddt_select(zio->io_spa, bp); ddt_entry_t *dde = zio->io_private; ddt_phys_t *ddp = &dde->dde_phys[p]; ddt_key_t *ddk = &dde->dde_key; ddt_enter(ddt); ASSERT(ddp->ddp_refcnt == 0); ASSERT(dde->dde_lead_zio[p] == zio); dde->dde_lead_zio[p] = NULL; if (zio->io_error == 0) { ASSERT(ZIO_CHECKSUM_EQUAL(bp->blk_cksum, ddk->ddk_cksum)); ASSERT(zp->zp_copies < SPA_DVAS_PER_BP); ASSERT(zp->zp_copies == BP_GET_NDVAS(bp) - BP_IS_GANG(bp)); if (ddp->ddp_phys_birth != 0) ddt_phys_free(ddt, ddk, ddp, zio->io_txg); ddt_phys_fill(ddp, bp); } ddt_exit(ddt); } static int zio_ddt_write(zio_t *zio) { spa_t *spa = zio->io_spa; blkptr_t *bp = zio->io_bp; uint64_t txg = zio->io_txg; zio_prop_t *zp = &zio->io_prop; int p = zp->zp_copies; int ditto_copies; zio_t *cio = NULL; zio_t *dio = NULL; ddt_t *ddt = ddt_select(spa, bp); ddt_entry_t *dde; ddt_phys_t *ddp; ASSERT(BP_GET_DEDUP(bp)); ASSERT(BP_GET_CHECKSUM(bp) == zp->zp_checksum); ASSERT(BP_IS_HOLE(bp) || zio->io_bp_override); ASSERT(!(zio->io_bp_override && (zio->io_flags & ZIO_FLAG_RAW))); ddt_enter(ddt); dde = ddt_lookup(ddt, bp, B_TRUE); ddp = &dde->dde_phys[p]; if (zp->zp_dedup_verify && zio_ddt_collision(zio, ddt, dde)) { /* * If we're using a weak checksum, upgrade to a strong checksum * and try again. If we're already using a strong checksum, * we can't resolve it, so just convert to an ordinary write. * (And automatically e-mail a paper to Nature?) */ if (!(zio_checksum_table[zp->zp_checksum].ci_flags & ZCHECKSUM_FLAG_DEDUP)) { zp->zp_checksum = spa_dedup_checksum(spa); zio_pop_transforms(zio); zio->io_stage = ZIO_STAGE_OPEN; BP_ZERO(bp); } else { zp->zp_dedup = B_FALSE; BP_SET_DEDUP(bp, B_FALSE); } ASSERT(!BP_GET_DEDUP(bp)); zio->io_pipeline = ZIO_WRITE_PIPELINE; ddt_exit(ddt); return (ZIO_PIPELINE_CONTINUE); } ditto_copies = ddt_ditto_copies_needed(ddt, dde, ddp); ASSERT(ditto_copies < SPA_DVAS_PER_BP); if (ditto_copies > ddt_ditto_copies_present(dde) && dde->dde_lead_zio[DDT_PHYS_DITTO] == NULL) { zio_prop_t czp = *zp; czp.zp_copies = ditto_copies; /* * If we arrived here with an override bp, we won't have run * the transform stack, so we won't have the data we need to * generate a child i/o. So, toss the override bp and restart. * This is safe, because using the override bp is just an * optimization; and it's rare, so the cost doesn't matter. */ if (zio->io_bp_override) { zio_pop_transforms(zio); zio->io_stage = ZIO_STAGE_OPEN; zio->io_pipeline = ZIO_WRITE_PIPELINE; zio->io_bp_override = NULL; BP_ZERO(bp); ddt_exit(ddt); return (ZIO_PIPELINE_CONTINUE); } dio = zio_write(zio, spa, txg, bp, zio->io_orig_data, zio->io_orig_size, zio->io_orig_size, &czp, NULL, NULL, NULL, zio_ddt_ditto_write_done, dde, zio->io_priority, ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark); zio_push_transform(dio, zio->io_data, zio->io_size, 0, NULL); dde->dde_lead_zio[DDT_PHYS_DITTO] = dio; } if (ddp->ddp_phys_birth != 0 || dde->dde_lead_zio[p] != NULL) { if (ddp->ddp_phys_birth != 0) ddt_bp_fill(ddp, bp, txg); if (dde->dde_lead_zio[p] != NULL) zio_add_child(zio, dde->dde_lead_zio[p]); else ddt_phys_addref(ddp); } else if (zio->io_bp_override) { ASSERT(bp->blk_birth == txg); ASSERT(BP_EQUAL(bp, zio->io_bp_override)); ddt_phys_fill(ddp, bp); ddt_phys_addref(ddp); } else { cio = zio_write(zio, spa, txg, bp, zio->io_orig_data, zio->io_orig_size, zio->io_orig_size, zp, zio_ddt_child_write_ready, NULL, NULL, zio_ddt_child_write_done, dde, zio->io_priority, ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark); zio_push_transform(cio, zio->io_data, zio->io_size, 0, NULL); dde->dde_lead_zio[p] = cio; } ddt_exit(ddt); if (cio) zio_nowait(cio); if (dio) zio_nowait(dio); return (ZIO_PIPELINE_CONTINUE); } ddt_entry_t *freedde; /* for debugging */ static int zio_ddt_free(zio_t *zio) { spa_t *spa = zio->io_spa; blkptr_t *bp = zio->io_bp; ddt_t *ddt = ddt_select(spa, bp); ddt_entry_t *dde; ddt_phys_t *ddp; ASSERT(BP_GET_DEDUP(bp)); ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); ddt_enter(ddt); freedde = dde = ddt_lookup(ddt, bp, B_TRUE); ddp = ddt_phys_select(dde, bp); ddt_phys_decref(ddp); ddt_exit(ddt); return (ZIO_PIPELINE_CONTINUE); } /* * ========================================================================== * Allocate and free blocks * ========================================================================== */ static zio_t * zio_io_to_allocate(spa_t *spa) { zio_t *zio; ASSERT(MUTEX_HELD(&spa->spa_alloc_lock)); zio = avl_first(&spa->spa_alloc_tree); if (zio == NULL) return (NULL); ASSERT(IO_IS_ALLOCATING(zio)); /* * Try to place a reservation for this zio. If we're unable to * reserve then we throttle. */ if (!metaslab_class_throttle_reserve(spa_normal_class(spa), zio->io_prop.zp_copies, zio, 0)) { return (NULL); } avl_remove(&spa->spa_alloc_tree, zio); ASSERT3U(zio->io_stage, <, ZIO_STAGE_DVA_ALLOCATE); return (zio); } static int zio_dva_throttle(zio_t *zio) { spa_t *spa = zio->io_spa; zio_t *nio; if (zio->io_priority == ZIO_PRIORITY_SYNC_WRITE || !spa_normal_class(zio->io_spa)->mc_alloc_throttle_enabled || zio->io_child_type == ZIO_CHILD_GANG || zio->io_flags & ZIO_FLAG_NODATA) { return (ZIO_PIPELINE_CONTINUE); } ASSERT(zio->io_child_type > ZIO_CHILD_GANG); ASSERT3U(zio->io_queued_timestamp, >, 0); ASSERT(zio->io_stage == ZIO_STAGE_DVA_THROTTLE); mutex_enter(&spa->spa_alloc_lock); ASSERT(zio->io_type == ZIO_TYPE_WRITE); avl_add(&spa->spa_alloc_tree, zio); nio = zio_io_to_allocate(zio->io_spa); mutex_exit(&spa->spa_alloc_lock); if (nio == zio) return (ZIO_PIPELINE_CONTINUE); if (nio != NULL) { - ASSERT3U(nio->io_queued_timestamp, <=, - zio->io_queued_timestamp); ASSERT(nio->io_stage == ZIO_STAGE_DVA_THROTTLE); /* * We are passing control to a new zio so make sure that * it is processed by a different thread. We do this to * avoid stack overflows that can occur when parents are * throttled and children are making progress. We allow * it to go to the head of the taskq since it's already * been waiting. */ zio_taskq_dispatch(nio, ZIO_TASKQ_ISSUE, B_TRUE); } return (ZIO_PIPELINE_STOP); } void zio_allocate_dispatch(spa_t *spa) { zio_t *zio; mutex_enter(&spa->spa_alloc_lock); zio = zio_io_to_allocate(spa); mutex_exit(&spa->spa_alloc_lock); if (zio == NULL) return; ASSERT3U(zio->io_stage, ==, ZIO_STAGE_DVA_THROTTLE); ASSERT0(zio->io_error); zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_TRUE); } static int zio_dva_allocate(zio_t *zio) { spa_t *spa = zio->io_spa; metaslab_class_t *mc = spa_normal_class(spa); blkptr_t *bp = zio->io_bp; int error; int flags = 0; if (zio->io_gang_leader == NULL) { ASSERT(zio->io_child_type > ZIO_CHILD_GANG); zio->io_gang_leader = zio; } ASSERT(BP_IS_HOLE(bp)); ASSERT0(BP_GET_NDVAS(bp)); ASSERT3U(zio->io_prop.zp_copies, >, 0); ASSERT3U(zio->io_prop.zp_copies, <=, spa_max_replication(spa)); ASSERT3U(zio->io_size, ==, BP_GET_PSIZE(bp)); if (zio->io_flags & ZIO_FLAG_NODATA) { flags |= METASLAB_DONT_THROTTLE; } if (zio->io_flags & ZIO_FLAG_GANG_CHILD) { flags |= METASLAB_GANG_CHILD; } if (zio->io_priority == ZIO_PRIORITY_ASYNC_WRITE) { flags |= METASLAB_ASYNC_ALLOC; } error = metaslab_alloc(spa, mc, zio->io_size, bp, zio->io_prop.zp_copies, zio->io_txg, NULL, flags, &zio->io_alloc_list, zio); if (error != 0) { spa_dbgmsg(spa, "%s: metaslab allocation failure: zio %p, " "size %llu, error %d", spa_name(spa), zio, zio->io_size, error); if (error == ENOSPC && zio->io_size > SPA_MINBLOCKSIZE) return (zio_write_gang_block(zio)); zio->io_error = error; } return (ZIO_PIPELINE_CONTINUE); } static int zio_dva_free(zio_t *zio) { metaslab_free(zio->io_spa, zio->io_bp, zio->io_txg, B_FALSE); return (ZIO_PIPELINE_CONTINUE); } static int zio_dva_claim(zio_t *zio) { int error; error = metaslab_claim(zio->io_spa, zio->io_bp, zio->io_txg); if (error) zio->io_error = error; return (ZIO_PIPELINE_CONTINUE); } /* * Undo an allocation. This is used by zio_done() when an I/O fails * and we want to give back the block we just allocated. * This handles both normal blocks and gang blocks. */ static void zio_dva_unallocate(zio_t *zio, zio_gang_node_t *gn, blkptr_t *bp) { ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp)); ASSERT(zio->io_bp_override == NULL); if (!BP_IS_HOLE(bp)) metaslab_free(zio->io_spa, bp, bp->blk_birth, B_TRUE); if (gn != NULL) { for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) { zio_dva_unallocate(zio, gn->gn_child[g], &gn->gn_gbh->zg_blkptr[g]); } } } /* * Try to allocate an intent log block. Return 0 on success, errno on failure. */ int zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp, blkptr_t *old_bp, uint64_t size, boolean_t use_slog) { int error = 1; zio_alloc_list_t io_alloc_list; ASSERT(txg > spa_syncing_txg(spa)); metaslab_trace_init(&io_alloc_list); if (use_slog) { error = metaslab_alloc(spa, spa_log_class(spa), size, new_bp, 1, txg, old_bp, METASLAB_HINTBP_AVOID, &io_alloc_list, NULL); } if (error) { error = metaslab_alloc(spa, spa_normal_class(spa), size, new_bp, 1, txg, old_bp, METASLAB_HINTBP_AVOID, &io_alloc_list, NULL); } metaslab_trace_fini(&io_alloc_list); if (error == 0) { BP_SET_LSIZE(new_bp, size); BP_SET_PSIZE(new_bp, size); BP_SET_COMPRESS(new_bp, ZIO_COMPRESS_OFF); BP_SET_CHECKSUM(new_bp, spa_version(spa) >= SPA_VERSION_SLIM_ZIL ? ZIO_CHECKSUM_ZILOG2 : ZIO_CHECKSUM_ZILOG); BP_SET_TYPE(new_bp, DMU_OT_INTENT_LOG); BP_SET_LEVEL(new_bp, 0); BP_SET_DEDUP(new_bp, 0); BP_SET_BYTEORDER(new_bp, ZFS_HOST_BYTEORDER); } return (error); } /* * Free an intent log block. */ void zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp) { ASSERT(BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG); ASSERT(!BP_IS_GANG(bp)); zio_free(spa, txg, bp); } /* * ========================================================================== * Read and write to physical devices * ========================================================================== */ /* * Issue an I/O to the underlying vdev. Typically the issue pipeline * stops after this stage and will resume upon I/O completion. * However, there are instances where the vdev layer may need to * continue the pipeline when an I/O was not issued. Since the I/O * that was sent to the vdev layer might be different than the one * currently active in the pipeline (see vdev_queue_io()), we explicitly * force the underlying vdev layers to call either zio_execute() or * zio_interrupt() to ensure that the pipeline continues with the correct I/O. */ static int zio_vdev_io_start(zio_t *zio) { vdev_t *vd = zio->io_vd; uint64_t align; spa_t *spa = zio->io_spa; ASSERT(zio->io_error == 0); ASSERT(zio->io_child_error[ZIO_CHILD_VDEV] == 0); if (vd == NULL) { if (!(zio->io_flags & ZIO_FLAG_CONFIG_WRITER)) spa_config_enter(spa, SCL_ZIO, zio, RW_READER); /* * The mirror_ops handle multiple DVAs in a single BP. */ vdev_mirror_ops.vdev_op_io_start(zio); return (ZIO_PIPELINE_STOP); } ASSERT3P(zio->io_logical, !=, zio); /* * We keep track of time-sensitive I/Os so that the scan thread * can quickly react to certain workloads. In particular, we care * about non-scrubbing, top-level reads and writes with the following * characteristics: * - synchronous writes of user data to non-slog devices * - any reads of user data * When these conditions are met, adjust the timestamp of spa_last_io * which allows the scan thread to adjust its workload accordingly. */ if (!(zio->io_flags & ZIO_FLAG_SCAN_THREAD) && zio->io_bp != NULL && vd == vd->vdev_top && !vd->vdev_islog && zio->io_bookmark.zb_objset != DMU_META_OBJSET && zio->io_txg != spa_syncing_txg(spa)) { uint64_t old = spa->spa_last_io; uint64_t new = ddi_get_lbolt64(); if (old != new) (void) atomic_cas_64(&spa->spa_last_io, old, new); } align = 1ULL << vd->vdev_top->vdev_ashift; if (!(zio->io_flags & ZIO_FLAG_PHYSICAL) && P2PHASE(zio->io_size, align) != 0) { /* Transform logical writes to be a full physical block size. */ uint64_t asize = P2ROUNDUP(zio->io_size, align); char *abuf = zio_buf_alloc(asize); ASSERT(vd == vd->vdev_top); if (zio->io_type == ZIO_TYPE_WRITE) { bcopy(zio->io_data, abuf, zio->io_size); bzero(abuf + zio->io_size, asize - zio->io_size); } zio_push_transform(zio, abuf, asize, asize, zio_subblock); } /* * If this is not a physical io, make sure that it is properly aligned * before proceeding. */ if (!(zio->io_flags & ZIO_FLAG_PHYSICAL)) { ASSERT0(P2PHASE(zio->io_offset, align)); ASSERT0(P2PHASE(zio->io_size, align)); } else { /* * For physical writes, we allow 512b aligned writes and assume * the device will perform a read-modify-write as necessary. */ ASSERT0(P2PHASE(zio->io_offset, SPA_MINBLOCKSIZE)); ASSERT0(P2PHASE(zio->io_size, SPA_MINBLOCKSIZE)); } VERIFY(zio->io_type != ZIO_TYPE_WRITE || spa_writeable(spa)); /* * If this is a repair I/O, and there's no self-healing involved -- * that is, we're just resilvering what we expect to resilver -- * then don't do the I/O unless zio's txg is actually in vd's DTL. * This prevents spurious resilvering with nested replication. * For example, given a mirror of mirrors, (A+B)+(C+D), if only * A is out of date, we'll read from C+D, then use the data to * resilver A+B -- but we don't actually want to resilver B, just A. * The top-level mirror has no way to know this, so instead we just * discard unnecessary repairs as we work our way down the vdev tree. * The same logic applies to any form of nested replication: * ditto + mirror, RAID-Z + replacing, etc. This covers them all. */ if ((zio->io_flags & ZIO_FLAG_IO_REPAIR) && !(zio->io_flags & ZIO_FLAG_SELF_HEAL) && zio->io_txg != 0 && /* not a delegated i/o */ !vdev_dtl_contains(vd, DTL_PARTIAL, zio->io_txg, 1)) { ASSERT(zio->io_type == ZIO_TYPE_WRITE); zio_vdev_io_bypass(zio); return (ZIO_PIPELINE_CONTINUE); } if (vd->vdev_ops->vdev_op_leaf && (zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE)) { if (zio->io_type == ZIO_TYPE_READ && vdev_cache_read(zio)) return (ZIO_PIPELINE_CONTINUE); if ((zio = vdev_queue_io(zio)) == NULL) return (ZIO_PIPELINE_STOP); if (!vdev_accessible(vd, zio)) { zio->io_error = SET_ERROR(ENXIO); zio_interrupt(zio); return (ZIO_PIPELINE_STOP); } } vd->vdev_ops->vdev_op_io_start(zio); return (ZIO_PIPELINE_STOP); } static int zio_vdev_io_done(zio_t *zio) { vdev_t *vd = zio->io_vd; vdev_ops_t *ops = vd ? vd->vdev_ops : &vdev_mirror_ops; boolean_t unexpected_error = B_FALSE; if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE)) return (ZIO_PIPELINE_STOP); ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE); if (vd != NULL && vd->vdev_ops->vdev_op_leaf) { vdev_queue_io_done(zio); if (zio->io_type == ZIO_TYPE_WRITE) vdev_cache_write(zio); if (zio_injection_enabled && zio->io_error == 0) zio->io_error = zio_handle_device_injection(vd, zio, EIO); if (zio_injection_enabled && zio->io_error == 0) zio->io_error = zio_handle_label_injection(zio, EIO); if (zio->io_error) { if (!vdev_accessible(vd, zio)) { zio->io_error = SET_ERROR(ENXIO); } else { unexpected_error = B_TRUE; } } } ops->vdev_op_io_done(zio); if (unexpected_error) VERIFY(vdev_probe(vd, zio) == NULL); return (ZIO_PIPELINE_CONTINUE); } /* * For non-raidz ZIOs, we can just copy aside the bad data read from the * disk, and use that to finish the checksum ereport later. */ static void zio_vsd_default_cksum_finish(zio_cksum_report_t *zcr, const void *good_buf) { /* no processing needed */ zfs_ereport_finish_checksum(zcr, good_buf, zcr->zcr_cbdata, B_FALSE); } /*ARGSUSED*/ void zio_vsd_default_cksum_report(zio_t *zio, zio_cksum_report_t *zcr, void *ignored) { void *buf = zio_buf_alloc(zio->io_size); bcopy(zio->io_data, buf, zio->io_size); zcr->zcr_cbinfo = zio->io_size; zcr->zcr_cbdata = buf; zcr->zcr_finish = zio_vsd_default_cksum_finish; zcr->zcr_free = zio_buf_free; } static int zio_vdev_io_assess(zio_t *zio) { vdev_t *vd = zio->io_vd; if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE)) return (ZIO_PIPELINE_STOP); if (vd == NULL && !(zio->io_flags & ZIO_FLAG_CONFIG_WRITER)) spa_config_exit(zio->io_spa, SCL_ZIO, zio); if (zio->io_vsd != NULL) { zio->io_vsd_ops->vsd_free(zio); zio->io_vsd = NULL; } if (zio_injection_enabled && zio->io_error == 0) zio->io_error = zio_handle_fault_injection(zio, EIO); /* * If the I/O failed, determine whether we should attempt to retry it. * * On retry, we cut in line in the issue queue, since we don't want * compression/checksumming/etc. work to prevent our (cheap) IO reissue. */ if (zio->io_error && vd == NULL && !(zio->io_flags & (ZIO_FLAG_DONT_RETRY | ZIO_FLAG_IO_RETRY))) { ASSERT(!(zio->io_flags & ZIO_FLAG_DONT_QUEUE)); /* not a leaf */ ASSERT(!(zio->io_flags & ZIO_FLAG_IO_BYPASS)); /* not a leaf */ zio->io_error = 0; zio->io_flags |= ZIO_FLAG_IO_RETRY | ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE; zio->io_stage = ZIO_STAGE_VDEV_IO_START >> 1; zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, zio_requeue_io_start_cut_in_line); return (ZIO_PIPELINE_STOP); } /* * If we got an error on a leaf device, convert it to ENXIO * if the device is not accessible at all. */ if (zio->io_error && vd != NULL && vd->vdev_ops->vdev_op_leaf && !vdev_accessible(vd, zio)) zio->io_error = SET_ERROR(ENXIO); /* * If we can't write to an interior vdev (mirror or RAID-Z), * set vdev_cant_write so that we stop trying to allocate from it. */ if (zio->io_error == ENXIO && zio->io_type == ZIO_TYPE_WRITE && vd != NULL && !vd->vdev_ops->vdev_op_leaf) { vd->vdev_cant_write = B_TRUE; } /* * If a cache flush returns ENOTSUP or ENOTTY, we know that no future * attempts will ever succeed. In this case we set a persistent bit so * that we don't bother with it in the future. */ if ((zio->io_error == ENOTSUP || zio->io_error == ENOTTY) && zio->io_type == ZIO_TYPE_IOCTL && zio->io_cmd == DKIOCFLUSHWRITECACHE && vd != NULL) vd->vdev_nowritecache = B_TRUE; if (zio->io_error) zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; if (vd != NULL && vd->vdev_ops->vdev_op_leaf && zio->io_physdone != NULL) { ASSERT(!(zio->io_flags & ZIO_FLAG_DELEGATED)); ASSERT(zio->io_child_type == ZIO_CHILD_VDEV); zio->io_physdone(zio->io_logical); } return (ZIO_PIPELINE_CONTINUE); } void zio_vdev_io_reissue(zio_t *zio) { ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_START); ASSERT(zio->io_error == 0); zio->io_stage >>= 1; } void zio_vdev_io_redone(zio_t *zio) { ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_DONE); zio->io_stage >>= 1; } void zio_vdev_io_bypass(zio_t *zio) { ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_START); ASSERT(zio->io_error == 0); zio->io_flags |= ZIO_FLAG_IO_BYPASS; zio->io_stage = ZIO_STAGE_VDEV_IO_ASSESS >> 1; } /* * ========================================================================== * Generate and verify checksums * ========================================================================== */ static int zio_checksum_generate(zio_t *zio) { blkptr_t *bp = zio->io_bp; enum zio_checksum checksum; if (bp == NULL) { /* * This is zio_write_phys(). * We're either generating a label checksum, or none at all. */ checksum = zio->io_prop.zp_checksum; if (checksum == ZIO_CHECKSUM_OFF) return (ZIO_PIPELINE_CONTINUE); ASSERT(checksum == ZIO_CHECKSUM_LABEL); } else { if (BP_IS_GANG(bp) && zio->io_child_type == ZIO_CHILD_GANG) { ASSERT(!IO_IS_ALLOCATING(zio)); checksum = ZIO_CHECKSUM_GANG_HEADER; } else { checksum = BP_GET_CHECKSUM(bp); } } zio_checksum_compute(zio, checksum, zio->io_data, zio->io_size); return (ZIO_PIPELINE_CONTINUE); } static int zio_checksum_verify(zio_t *zio) { zio_bad_cksum_t info; blkptr_t *bp = zio->io_bp; int error; ASSERT(zio->io_vd != NULL); if (bp == NULL) { /* * This is zio_read_phys(). * We're either verifying a label checksum, or nothing at all. */ if (zio->io_prop.zp_checksum == ZIO_CHECKSUM_OFF) return (ZIO_PIPELINE_CONTINUE); ASSERT(zio->io_prop.zp_checksum == ZIO_CHECKSUM_LABEL); } if ((error = zio_checksum_error(zio, &info)) != 0) { zio->io_error = error; if (error == ECKSUM && !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) { zfs_ereport_start_checksum(zio->io_spa, zio->io_vd, zio, zio->io_offset, zio->io_size, NULL, &info); } } return (ZIO_PIPELINE_CONTINUE); } /* * Called by RAID-Z to ensure we don't compute the checksum twice. */ void zio_checksum_verified(zio_t *zio) { zio->io_pipeline &= ~ZIO_STAGE_CHECKSUM_VERIFY; } /* * ========================================================================== * Error rank. Error are ranked in the order 0, ENXIO, ECKSUM, EIO, other. * An error of 0 indicates success. ENXIO indicates whole-device failure, * which may be transient (e.g. unplugged) or permament. ECKSUM and EIO * indicate errors that are specific to one I/O, and most likely permanent. * Any other error is presumed to be worse because we weren't expecting it. * ========================================================================== */ int zio_worst_error(int e1, int e2) { static int zio_error_rank[] = { 0, ENXIO, ECKSUM, EIO }; int r1, r2; for (r1 = 0; r1 < sizeof (zio_error_rank) / sizeof (int); r1++) if (e1 == zio_error_rank[r1]) break; for (r2 = 0; r2 < sizeof (zio_error_rank) / sizeof (int); r2++) if (e2 == zio_error_rank[r2]) break; return (r1 > r2 ? e1 : e2); } /* * ========================================================================== * I/O completion * ========================================================================== */ static int zio_ready(zio_t *zio) { blkptr_t *bp = zio->io_bp; zio_t *pio, *pio_next; zio_link_t *zl = NULL; if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_READY) || zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_READY)) return (ZIO_PIPELINE_STOP); if (zio->io_ready) { ASSERT(IO_IS_ALLOCATING(zio)); ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp) || (zio->io_flags & ZIO_FLAG_NOPWRITE)); ASSERT(zio->io_children[ZIO_CHILD_GANG][ZIO_WAIT_READY] == 0); zio->io_ready(zio); } if (bp != NULL && bp != &zio->io_bp_copy) zio->io_bp_copy = *bp; if (zio->io_error != 0) { zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; if (zio->io_flags & ZIO_FLAG_IO_ALLOCATING) { ASSERT(IO_IS_ALLOCATING(zio)); ASSERT(zio->io_priority == ZIO_PRIORITY_ASYNC_WRITE); /* * We were unable to allocate anything, unreserve and * issue the next I/O to allocate. */ metaslab_class_throttle_unreserve( spa_normal_class(zio->io_spa), zio->io_prop.zp_copies, zio); zio_allocate_dispatch(zio->io_spa); } } mutex_enter(&zio->io_lock); zio->io_state[ZIO_WAIT_READY] = 1; pio = zio_walk_parents(zio, &zl); mutex_exit(&zio->io_lock); /* * As we notify zio's parents, new parents could be added. * New parents go to the head of zio's io_parent_list, however, * so we will (correctly) not notify them. The remainder of zio's * io_parent_list, from 'pio_next' onward, cannot change because * all parents must wait for us to be done before they can be done. */ for (; pio != NULL; pio = pio_next) { pio_next = zio_walk_parents(zio, &zl); zio_notify_parent(pio, zio, ZIO_WAIT_READY); } if (zio->io_flags & ZIO_FLAG_NODATA) { if (BP_IS_GANG(bp)) { zio->io_flags &= ~ZIO_FLAG_NODATA; } else { ASSERT((uintptr_t)zio->io_data < SPA_MAXBLOCKSIZE); zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES; } } if (zio_injection_enabled && zio->io_spa->spa_syncing_txg == zio->io_txg) zio_handle_ignored_writes(zio); return (ZIO_PIPELINE_CONTINUE); } /* * Update the allocation throttle accounting. */ static void zio_dva_throttle_done(zio_t *zio) { zio_t *lio = zio->io_logical; zio_t *pio = zio_unique_parent(zio); vdev_t *vd = zio->io_vd; int flags = METASLAB_ASYNC_ALLOC; ASSERT3P(zio->io_bp, !=, NULL); ASSERT3U(zio->io_type, ==, ZIO_TYPE_WRITE); ASSERT3U(zio->io_priority, ==, ZIO_PRIORITY_ASYNC_WRITE); ASSERT3U(zio->io_child_type, ==, ZIO_CHILD_VDEV); ASSERT(vd != NULL); ASSERT3P(vd, ==, vd->vdev_top); ASSERT(!(zio->io_flags & (ZIO_FLAG_IO_REPAIR | ZIO_FLAG_IO_RETRY))); ASSERT(zio->io_flags & ZIO_FLAG_IO_ALLOCATING); ASSERT(!(lio->io_flags & ZIO_FLAG_IO_REWRITE)); ASSERT(!(lio->io_orig_flags & ZIO_FLAG_NODATA)); /* * Parents of gang children can have two flavors -- ones that * allocated the gang header (will have ZIO_FLAG_IO_REWRITE set) * and ones that allocated the constituent blocks. The allocation * throttle needs to know the allocating parent zio so we must find * it here. */ if (pio->io_child_type == ZIO_CHILD_GANG) { /* * If our parent is a rewrite gang child then our grandparent * would have been the one that performed the allocation. */ if (pio->io_flags & ZIO_FLAG_IO_REWRITE) pio = zio_unique_parent(pio); flags |= METASLAB_GANG_CHILD; } ASSERT(IO_IS_ALLOCATING(pio)); ASSERT3P(zio, !=, zio->io_logical); ASSERT(zio->io_logical != NULL); ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REPAIR)); ASSERT0(zio->io_flags & ZIO_FLAG_NOPWRITE); mutex_enter(&pio->io_lock); metaslab_group_alloc_decrement(zio->io_spa, vd->vdev_id, pio, flags); mutex_exit(&pio->io_lock); metaslab_class_throttle_unreserve(spa_normal_class(zio->io_spa), 1, pio); /* * Call into the pipeline to see if there is more work that * needs to be done. If there is work to be done it will be * dispatched to another taskq thread. */ zio_allocate_dispatch(zio->io_spa); } static int zio_done(zio_t *zio) { spa_t *spa = zio->io_spa; zio_t *lio = zio->io_logical; blkptr_t *bp = zio->io_bp; vdev_t *vd = zio->io_vd; uint64_t psize = zio->io_size; zio_t *pio, *pio_next; metaslab_class_t *mc = spa_normal_class(spa); zio_link_t *zl = NULL; /* * If our children haven't all completed, * wait for them and then repeat this pipeline stage. */ if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE) || zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_DONE) || zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_DONE) || zio_wait_for_children(zio, ZIO_CHILD_LOGICAL, ZIO_WAIT_DONE)) return (ZIO_PIPELINE_STOP); /* * If the allocation throttle is enabled, then update the accounting. * We only track child I/Os that are part of an allocating async * write. We must do this since the allocation is performed * by the logical I/O but the actual write is done by child I/Os. */ if (zio->io_flags & ZIO_FLAG_IO_ALLOCATING && zio->io_child_type == ZIO_CHILD_VDEV) { ASSERT(mc->mc_alloc_throttle_enabled); zio_dva_throttle_done(zio); } /* * If the allocation throttle is enabled, verify that * we have decremented the refcounts for every I/O that was throttled. */ if (zio->io_flags & ZIO_FLAG_IO_ALLOCATING) { ASSERT(zio->io_type == ZIO_TYPE_WRITE); ASSERT(zio->io_priority == ZIO_PRIORITY_ASYNC_WRITE); ASSERT(bp != NULL); metaslab_group_alloc_verify(spa, zio->io_bp, zio); VERIFY(refcount_not_held(&mc->mc_alloc_slots, zio)); } for (int c = 0; c < ZIO_CHILD_TYPES; c++) for (int w = 0; w < ZIO_WAIT_TYPES; w++) ASSERT(zio->io_children[c][w] == 0); if (bp != NULL && !BP_IS_EMBEDDED(bp)) { ASSERT(bp->blk_pad[0] == 0); ASSERT(bp->blk_pad[1] == 0); ASSERT(bcmp(bp, &zio->io_bp_copy, sizeof (blkptr_t)) == 0 || (bp == zio_unique_parent(zio)->io_bp)); if (zio->io_type == ZIO_TYPE_WRITE && !BP_IS_HOLE(bp) && zio->io_bp_override == NULL && !(zio->io_flags & ZIO_FLAG_IO_REPAIR)) { ASSERT(!BP_SHOULD_BYTESWAP(bp)); ASSERT3U(zio->io_prop.zp_copies, <=, BP_GET_NDVAS(bp)); ASSERT(BP_COUNT_GANG(bp) == 0 || (BP_COUNT_GANG(bp) == BP_GET_NDVAS(bp))); } if (zio->io_flags & ZIO_FLAG_NOPWRITE) VERIFY(BP_EQUAL(bp, &zio->io_bp_orig)); } /* * If there were child vdev/gang/ddt errors, they apply to us now. */ zio_inherit_child_errors(zio, ZIO_CHILD_VDEV); zio_inherit_child_errors(zio, ZIO_CHILD_GANG); zio_inherit_child_errors(zio, ZIO_CHILD_DDT); /* * If the I/O on the transformed data was successful, generate any * checksum reports now while we still have the transformed data. */ if (zio->io_error == 0) { while (zio->io_cksum_report != NULL) { zio_cksum_report_t *zcr = zio->io_cksum_report; uint64_t align = zcr->zcr_align; uint64_t asize = P2ROUNDUP(psize, align); char *abuf = zio->io_data; if (asize != psize) { abuf = zio_buf_alloc(asize); bcopy(zio->io_data, abuf, psize); bzero(abuf + psize, asize - psize); } zio->io_cksum_report = zcr->zcr_next; zcr->zcr_next = NULL; zcr->zcr_finish(zcr, abuf); zfs_ereport_free_checksum(zcr); if (asize != psize) zio_buf_free(abuf, asize); } } zio_pop_transforms(zio); /* note: may set zio->io_error */ vdev_stat_update(zio, psize); if (zio->io_error) { /* * If this I/O is attached to a particular vdev, * generate an error message describing the I/O failure * at the block level. We ignore these errors if the * device is currently unavailable. */ if (zio->io_error != ECKSUM && vd != NULL && !vdev_is_dead(vd)) zfs_ereport_post(FM_EREPORT_ZFS_IO, spa, vd, zio, 0, 0); if ((zio->io_error == EIO || !(zio->io_flags & (ZIO_FLAG_SPECULATIVE | ZIO_FLAG_DONT_PROPAGATE))) && zio == lio) { /* * For logical I/O requests, tell the SPA to log the * error and generate a logical data ereport. */ spa_log_error(spa, zio); zfs_ereport_post(FM_EREPORT_ZFS_DATA, spa, NULL, zio, 0, 0); } } if (zio->io_error && zio == lio) { /* * Determine whether zio should be reexecuted. This will * propagate all the way to the root via zio_notify_parent(). */ ASSERT(vd == NULL && bp != NULL); ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); if (IO_IS_ALLOCATING(zio) && !(zio->io_flags & ZIO_FLAG_CANFAIL)) { if (zio->io_error != ENOSPC) zio->io_reexecute |= ZIO_REEXECUTE_NOW; else zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; } if ((zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_FREE) && !(zio->io_flags & ZIO_FLAG_SCAN_THREAD) && zio->io_error == ENXIO && spa_load_state(spa) == SPA_LOAD_NONE && spa_get_failmode(spa) != ZIO_FAILURE_MODE_CONTINUE) zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; if (!(zio->io_flags & ZIO_FLAG_CANFAIL) && !zio->io_reexecute) zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; /* * Here is a possibly good place to attempt to do * either combinatorial reconstruction or error correction * based on checksums. It also might be a good place * to send out preliminary ereports before we suspend * processing. */ } /* * If there were logical child errors, they apply to us now. * We defer this until now to avoid conflating logical child * errors with errors that happened to the zio itself when * updating vdev stats and reporting FMA events above. */ zio_inherit_child_errors(zio, ZIO_CHILD_LOGICAL); if ((zio->io_error || zio->io_reexecute) && IO_IS_ALLOCATING(zio) && zio->io_gang_leader == zio && !(zio->io_flags & (ZIO_FLAG_IO_REWRITE | ZIO_FLAG_NOPWRITE))) zio_dva_unallocate(zio, zio->io_gang_tree, bp); zio_gang_tree_free(&zio->io_gang_tree); /* * Godfather I/Os should never suspend. */ if ((zio->io_flags & ZIO_FLAG_GODFATHER) && (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND)) zio->io_reexecute = 0; if (zio->io_reexecute) { /* * This is a logical I/O that wants to reexecute. * * Reexecute is top-down. When an i/o fails, if it's not * the root, it simply notifies its parent and sticks around. * The parent, seeing that it still has children in zio_done(), * does the same. This percolates all the way up to the root. * The root i/o will reexecute or suspend the entire tree. * * This approach ensures that zio_reexecute() honors * all the original i/o dependency relationships, e.g. * parents not executing until children are ready. */ ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); zio->io_gang_leader = NULL; mutex_enter(&zio->io_lock); zio->io_state[ZIO_WAIT_DONE] = 1; mutex_exit(&zio->io_lock); /* * "The Godfather" I/O monitors its children but is * not a true parent to them. It will track them through * the pipeline but severs its ties whenever they get into * trouble (e.g. suspended). This allows "The Godfather" * I/O to return status without blocking. */ zl = NULL; for (pio = zio_walk_parents(zio, &zl); pio != NULL; pio = pio_next) { zio_link_t *remove_zl = zl; pio_next = zio_walk_parents(zio, &zl); if ((pio->io_flags & ZIO_FLAG_GODFATHER) && (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND)) { zio_remove_child(pio, zio, remove_zl); zio_notify_parent(pio, zio, ZIO_WAIT_DONE); } } if ((pio = zio_unique_parent(zio)) != NULL) { /* * We're not a root i/o, so there's nothing to do * but notify our parent. Don't propagate errors * upward since we haven't permanently failed yet. */ ASSERT(!(zio->io_flags & ZIO_FLAG_GODFATHER)); zio->io_flags |= ZIO_FLAG_DONT_PROPAGATE; zio_notify_parent(pio, zio, ZIO_WAIT_DONE); } else if (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND) { /* * We'd fail again if we reexecuted now, so suspend * until conditions improve (e.g. device comes online). */ zio_suspend(spa, zio); } else { /* * Reexecution is potentially a huge amount of work. * Hand it off to the otherwise-unused claim taskq. */ ASSERT(zio->io_tqent.tqent_next == NULL); spa_taskq_dispatch_ent(spa, ZIO_TYPE_CLAIM, ZIO_TASKQ_ISSUE, (task_func_t *)zio_reexecute, zio, 0, &zio->io_tqent); } return (ZIO_PIPELINE_STOP); } ASSERT(zio->io_child_count == 0); ASSERT(zio->io_reexecute == 0); ASSERT(zio->io_error == 0 || (zio->io_flags & ZIO_FLAG_CANFAIL)); /* * Report any checksum errors, since the I/O is complete. */ while (zio->io_cksum_report != NULL) { zio_cksum_report_t *zcr = zio->io_cksum_report; zio->io_cksum_report = zcr->zcr_next; zcr->zcr_next = NULL; zcr->zcr_finish(zcr, NULL); zfs_ereport_free_checksum(zcr); } /* * It is the responsibility of the done callback to ensure that this * particular zio is no longer discoverable for adoption, and as * such, cannot acquire any new parents. */ if (zio->io_done) zio->io_done(zio); mutex_enter(&zio->io_lock); zio->io_state[ZIO_WAIT_DONE] = 1; mutex_exit(&zio->io_lock); zl = NULL; for (pio = zio_walk_parents(zio, &zl); pio != NULL; pio = pio_next) { zio_link_t *remove_zl = zl; pio_next = zio_walk_parents(zio, &zl); zio_remove_child(pio, zio, remove_zl); zio_notify_parent(pio, zio, ZIO_WAIT_DONE); } if (zio->io_waiter != NULL) { mutex_enter(&zio->io_lock); zio->io_executor = NULL; cv_broadcast(&zio->io_cv); mutex_exit(&zio->io_lock); } else { zio_destroy(zio); } return (ZIO_PIPELINE_STOP); } /* * ========================================================================== * I/O pipeline definition * ========================================================================== */ static zio_pipe_stage_t *zio_pipeline[] = { NULL, zio_read_bp_init, zio_write_bp_init, zio_free_bp_init, zio_issue_async, zio_write_compress, zio_checksum_generate, zio_nop_write, zio_ddt_read_start, zio_ddt_read_done, zio_ddt_write, zio_ddt_free, zio_gang_assemble, zio_gang_issue, zio_dva_throttle, zio_dva_allocate, zio_dva_free, zio_dva_claim, zio_ready, zio_vdev_io_start, zio_vdev_io_done, zio_vdev_io_assess, zio_checksum_verify, zio_done }; /* * Compare two zbookmark_phys_t's to see which we would reach first in a * pre-order traversal of the object tree. * * This is simple in every case aside from the meta-dnode object. For all other * objects, we traverse them in order (object 1 before object 2, and so on). * However, all of these objects are traversed while traversing object 0, since * the data it points to is the list of objects. Thus, we need to convert to a * canonical representation so we can compare meta-dnode bookmarks to * non-meta-dnode bookmarks. * * We do this by calculating "equivalents" for each field of the zbookmark. * zbookmarks outside of the meta-dnode use their own object and level, and * calculate the level 0 equivalent (the first L0 blkid that is contained in the * blocks this bookmark refers to) by multiplying their blkid by their span * (the number of L0 blocks contained within one block at their level). * zbookmarks inside the meta-dnode calculate their object equivalent * (which is L0equiv * dnodes per data block), use 0 for their L0equiv, and use * level + 1<<31 (any value larger than a level could ever be) for their level. * This causes them to always compare before a bookmark in their object * equivalent, compare appropriately to bookmarks in other objects, and to * compare appropriately to other bookmarks in the meta-dnode. */ int zbookmark_compare(uint16_t dbss1, uint8_t ibs1, uint16_t dbss2, uint8_t ibs2, const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2) { /* * These variables represent the "equivalent" values for the zbookmark, * after converting zbookmarks inside the meta dnode to their * normal-object equivalents. */ uint64_t zb1obj, zb2obj; uint64_t zb1L0, zb2L0; uint64_t zb1level, zb2level; if (zb1->zb_object == zb2->zb_object && zb1->zb_level == zb2->zb_level && zb1->zb_blkid == zb2->zb_blkid) return (0); /* * BP_SPANB calculates the span in blocks. */ zb1L0 = (zb1->zb_blkid) * BP_SPANB(ibs1, zb1->zb_level); zb2L0 = (zb2->zb_blkid) * BP_SPANB(ibs2, zb2->zb_level); if (zb1->zb_object == DMU_META_DNODE_OBJECT) { zb1obj = zb1L0 * (dbss1 << (SPA_MINBLOCKSHIFT - DNODE_SHIFT)); zb1L0 = 0; zb1level = zb1->zb_level + COMPARE_META_LEVEL; } else { zb1obj = zb1->zb_object; zb1level = zb1->zb_level; } if (zb2->zb_object == DMU_META_DNODE_OBJECT) { zb2obj = zb2L0 * (dbss2 << (SPA_MINBLOCKSHIFT - DNODE_SHIFT)); zb2L0 = 0; zb2level = zb2->zb_level + COMPARE_META_LEVEL; } else { zb2obj = zb2->zb_object; zb2level = zb2->zb_level; } /* Now that we have a canonical representation, do the comparison. */ if (zb1obj != zb2obj) return (zb1obj < zb2obj ? -1 : 1); else if (zb1L0 != zb2L0) return (zb1L0 < zb2L0 ? -1 : 1); else if (zb1level != zb2level) return (zb1level > zb2level ? -1 : 1); /* * This can (theoretically) happen if the bookmarks have the same object * and level, but different blkids, if the block sizes are not the same. * There is presently no way to change the indirect block sizes */ return (0); } /* * This function checks the following: given that last_block is the place that * our traversal stopped last time, does that guarantee that we've visited * every node under subtree_root? Therefore, we can't just use the raw output * of zbookmark_compare. We have to pass in a modified version of * subtree_root; by incrementing the block id, and then checking whether * last_block is before or equal to that, we can tell whether or not having * visited last_block implies that all of subtree_root's children have been * visited. */ boolean_t zbookmark_subtree_completed(const dnode_phys_t *dnp, const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block) { zbookmark_phys_t mod_zb = *subtree_root; mod_zb.zb_blkid++; ASSERT(last_block->zb_level == 0); /* The objset_phys_t isn't before anything. */ if (dnp == NULL) return (B_FALSE); /* * We pass in 1ULL << (DNODE_BLOCK_SHIFT - SPA_MINBLOCKSHIFT) for the * data block size in sectors, because that variable is only used if * the bookmark refers to a block in the meta-dnode. Since we don't * know without examining it what object it refers to, and there's no * harm in passing in this value in other cases, we always pass it in. * * We pass in 0 for the indirect block size shift because zb2 must be * level 0. The indirect block size is only used to calculate the span * of the bookmark, but since the bookmark must be level 0, the span is * always 1, so the math works out. * * If you make changes to how the zbookmark_compare code works, be sure * to make sure that this code still works afterwards. */ return (zbookmark_compare(dnp->dn_datablkszsec, dnp->dn_indblkshift, 1ULL << (DNODE_BLOCK_SHIFT - SPA_MINBLOCKSHIFT), 0, &mod_zb, last_block) <= 0); }