Index: stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/arc.c =================================================================== --- stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/arc.c (revision 304137) +++ stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/arc.c (revision 304138) @@ -1,7245 +1,7259 @@ /* * 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, 2015 by Delphix. All rights reserved. + * Copyright (c) 2011, 2016 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. * * Arc buffers may have an associated eviction callback function. * This function will be invoked prior to removing the buffer (e.g. * in arc_do_user_evicts()). Note however that the data associated * with the buffer may be evicted prior to the callback. The callback * must be made with *no locks held* (to prevent deadlock). Additionally, * the users of callbacks must ensure that their private data is * protected from simultaneous callbacks from arc_clear_callback() * and arc_do_user_evicts(). * * 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 */ #include #include #include #include #include #include #include #include #include #include #ifdef _KERNEL #include #include #endif #include #include #include #include #include #include #ifdef illumos #ifndef _KERNEL /* set with ZFS_DEBUG=watch, to enable watchpoints on frozen buffers */ boolean_t arc_watch = B_FALSE; int arc_procfd; #endif #endif /* illumos */ 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; static kmutex_t arc_user_evicts_lock; static kcondvar_t arc_user_evicts_cv; static boolean_t arc_user_evicts_thread_exit; static kmutex_t arc_dnlc_evicts_lock; static kcondvar_t arc_dnlc_evicts_cv; static boolean_t arc_dnlc_evicts_thread_exit; 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; /* * The number of sublists used for each of the arc state lists. If this * is not set to a suitable value by the user, it will be configured to * the number of CPUs on the system in arc_init(). */ int zfs_arc_num_sublists_per_state = 0; /* 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; extern boolean_t zfs_prefetch_disable; /* * The arc has filled available memory and has now warmed up. */ static boolean_t arc_warm; /* * 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_disable_dup_eviction = 0; uint64_t zfs_arc_average_blocksize = 8 * 1024; /* 8KB */ u_int zfs_arc_free_target = 0; /* Absolute min for arc min / max is 16MB. */ static uint64_t arc_abs_min = 16 << 20; static int sysctl_vfs_zfs_arc_free_target(SYSCTL_HANDLER_ARGS); static int sysctl_vfs_zfs_arc_meta_limit(SYSCTL_HANDLER_ARGS); static int sysctl_vfs_zfs_arc_max(SYSCTL_HANDLER_ARGS); static int sysctl_vfs_zfs_arc_min(SYSCTL_HANDLER_ARGS); #if defined(__FreeBSD__) && defined(_KERNEL) static void arc_free_target_init(void *unused __unused) { zfs_arc_free_target = vm_pageout_wakeup_thresh; } SYSINIT(arc_free_target_init, SI_SUB_KTHREAD_PAGE, SI_ORDER_ANY, arc_free_target_init, NULL); TUNABLE_QUAD("vfs.zfs.arc_meta_limit", &zfs_arc_meta_limit); TUNABLE_QUAD("vfs.zfs.arc_meta_min", &zfs_arc_meta_min); TUNABLE_INT("vfs.zfs.arc_shrink_shift", &zfs_arc_shrink_shift); SYSCTL_DECL(_vfs_zfs); SYSCTL_PROC(_vfs_zfs, OID_AUTO, arc_max, CTLTYPE_U64 | CTLFLAG_RWTUN, 0, sizeof(uint64_t), sysctl_vfs_zfs_arc_max, "QU", "Maximum ARC size"); SYSCTL_PROC(_vfs_zfs, OID_AUTO, arc_min, CTLTYPE_U64 | CTLFLAG_RWTUN, 0, sizeof(uint64_t), sysctl_vfs_zfs_arc_min, "QU", "Minimum ARC size"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, arc_average_blocksize, CTLFLAG_RDTUN, &zfs_arc_average_blocksize, 0, "ARC average blocksize"); SYSCTL_INT(_vfs_zfs, OID_AUTO, arc_shrink_shift, CTLFLAG_RW, &arc_shrink_shift, 0, "log2(fraction of arc to reclaim)"); /* * We don't have a tunable for arc_free_target due to the dependency on * pagedaemon initialisation. */ SYSCTL_PROC(_vfs_zfs, OID_AUTO, arc_free_target, CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(u_int), sysctl_vfs_zfs_arc_free_target, "IU", "Desired number of free pages below which ARC triggers reclaim"); static int sysctl_vfs_zfs_arc_free_target(SYSCTL_HANDLER_ARGS) { u_int val; int err; val = zfs_arc_free_target; err = sysctl_handle_int(oidp, &val, 0, req); if (err != 0 || req->newptr == NULL) return (err); if (val < minfree) return (EINVAL); if (val > vm_cnt.v_page_count) return (EINVAL); zfs_arc_free_target = val; return (0); } /* * Must be declared here, before the definition of corresponding kstat * macro which uses the same names will confuse the compiler. */ SYSCTL_PROC(_vfs_zfs, OID_AUTO, arc_meta_limit, CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t), sysctl_vfs_zfs_arc_meta_limit, "QU", "ARC metadata limit"); #endif /* * 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]; /* * total amount of evictable data in this state */ uint64_t arcs_lsize[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_allocated; 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 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_cdata_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_l2_compress_successes; kstat_named_t arcstat_l2_compress_zeros; kstat_named_t arcstat_l2_compress_failures; kstat_named_t arcstat_l2_padding_needed; kstat_named_t arcstat_l2_write_trylock_fail; kstat_named_t arcstat_l2_write_passed_headroom; kstat_named_t arcstat_l2_write_spa_mismatch; kstat_named_t arcstat_l2_write_in_l2; kstat_named_t arcstat_l2_write_hdr_io_in_progress; kstat_named_t arcstat_l2_write_not_cacheable; kstat_named_t arcstat_l2_write_full; kstat_named_t arcstat_l2_write_buffer_iter; kstat_named_t arcstat_l2_write_pios; kstat_named_t arcstat_l2_write_buffer_bytes_scanned; kstat_named_t arcstat_l2_write_buffer_list_iter; kstat_named_t arcstat_l2_write_buffer_list_null_iter; kstat_named_t arcstat_memory_throttle_count; kstat_named_t arcstat_duplicate_buffers; kstat_named_t arcstat_duplicate_buffers_size; kstat_named_t arcstat_duplicate_reads; 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 }, { "allocated", 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 }, { "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_cdata_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 }, { "l2_compress_successes", KSTAT_DATA_UINT64 }, { "l2_compress_zeros", KSTAT_DATA_UINT64 }, { "l2_compress_failures", KSTAT_DATA_UINT64 }, { "l2_padding_needed", KSTAT_DATA_UINT64 }, { "l2_write_trylock_fail", KSTAT_DATA_UINT64 }, { "l2_write_passed_headroom", KSTAT_DATA_UINT64 }, { "l2_write_spa_mismatch", KSTAT_DATA_UINT64 }, { "l2_write_in_l2", KSTAT_DATA_UINT64 }, { "l2_write_io_in_progress", KSTAT_DATA_UINT64 }, { "l2_write_not_cacheable", KSTAT_DATA_UINT64 }, { "l2_write_full", KSTAT_DATA_UINT64 }, { "l2_write_buffer_iter", KSTAT_DATA_UINT64 }, { "l2_write_pios", KSTAT_DATA_UINT64 }, { "l2_write_buffer_bytes_scanned", KSTAT_DATA_UINT64 }, { "l2_write_buffer_list_iter", KSTAT_DATA_UINT64 }, { "l2_write_buffer_list_null_iter", KSTAT_DATA_UINT64 }, { "memory_throttle_count", KSTAT_DATA_UINT64 }, { "duplicate_buffers", KSTAT_DATA_UINT64 }, { "duplicate_buffers_size", KSTAT_DATA_UINT64 }, { "duplicate_reads", 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 */ #define L2ARC_IS_VALID_COMPRESS(_c_) \ ((_c_) == ZIO_COMPRESS_LZ4 || (_c_) == ZIO_COMPRESS_EMPTY) 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; 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; #ifdef ZFS_DEBUG /* * used for debugging wtih 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_datacnt; /* for waiting on writes to complete */ kcondvar_t b_cv; /* 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; /* temporary buffer holder for in-flight compressed or padded data */ void *b_tmp_cdata; } 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 */ /* real alloc'd buffer size depending on b_compress applied */ int32_t b_asize; uint8_t b_compress; 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; /* * Even though this checksum is only set/verified when a buffer is in * the L1 cache, it needs to be in the set of common fields because it * must be preserved from the time before a buffer is written out to * L2ARC until after it is read back in. */ zio_cksum_t *b_freeze_cksum; arc_buf_hdr_t *b_hash_next; arc_flags_t b_flags; /* immutable */ int32_t b_size; uint64_t b_spa; /* 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; }; #if defined(__FreeBSD__) && defined(_KERNEL) static int sysctl_vfs_zfs_arc_meta_limit(SYSCTL_HANDLER_ARGS) { uint64_t val; int err; val = arc_meta_limit; err = sysctl_handle_64(oidp, &val, 0, req); if (err != 0 || req->newptr == NULL) return (err); if (val <= 0 || val > arc_c_max) return (EINVAL); arc_meta_limit = val; return (0); } static int sysctl_vfs_zfs_arc_max(SYSCTL_HANDLER_ARGS) { uint64_t val; int err; val = zfs_arc_max; err = sysctl_handle_64(oidp, &val, 0, req); if (err != 0 || req->newptr == NULL) return (err); if (zfs_arc_max == 0) { /* Loader tunable so blindly set */ zfs_arc_max = val; return (0); } if (val < arc_abs_min || val > kmem_size()) return (EINVAL); if (val < arc_c_min) return (EINVAL); if (zfs_arc_meta_limit > 0 && val < zfs_arc_meta_limit) return (EINVAL); arc_c_max = val; arc_c = arc_c_max; arc_p = (arc_c >> 1); if (zfs_arc_meta_limit == 0) { /* limit meta-data to 1/4 of the arc capacity */ arc_meta_limit = arc_c_max / 4; } /* if kmem_flags are set, lets try to use less memory */ if (kmem_debugging()) arc_c = arc_c / 2; zfs_arc_max = arc_c; return (0); } static int sysctl_vfs_zfs_arc_min(SYSCTL_HANDLER_ARGS) { uint64_t val; int err; val = zfs_arc_min; err = sysctl_handle_64(oidp, &val, 0, req); if (err != 0 || req->newptr == NULL) return (err); if (zfs_arc_min == 0) { /* Loader tunable so blindly set */ zfs_arc_min = val; return (0); } if (val < arc_abs_min || val > arc_c_max) return (EINVAL); arc_c_min = val; if (zfs_arc_meta_min == 0) arc_meta_min = arc_c_min / 2; if (arc_c < arc_c_min) arc_c = arc_c_min; zfs_arc_min = arc_c_min; return (0); } #endif static arc_buf_t *arc_eviction_list; static arc_buf_hdr_t arc_eviction_hdr; #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_FREED_IN_READ(hdr) ((hdr)->b_flags & ARC_FLAG_FREED_IN_READ) #define HDR_BUF_AVAILABLE(hdr) ((hdr)->b_flags & ARC_FLAG_BUF_AVAILABLE) #define HDR_L2CACHE(hdr) ((hdr)->b_flags & ARC_FLAG_L2CACHE) #define HDR_L2COMPRESS(hdr) ((hdr)->b_flags & ARC_FLAG_L2COMPRESS) #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_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) /* * 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 CACHE_LINE_SIZE 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] __aligned(CACHE_LINE_SIZE); } 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 */ /* * Used to distinguish headers that are being process by * l2arc_write_buffers(), but have yet to be assigned to a l2arc disk * address. This can happen when the header is added to the l2arc's list * of buffers to write in the first stage of l2arc_write_buffers(), but * has not yet been written out which happens in the second stage of * l2arc_write_buffers(). */ #define L2ARC_ADDR_UNSET ((uint64_t)(-1)) #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 */ SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, l2arc_write_max, CTLFLAG_RW, &l2arc_write_max, 0, "max write size"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, l2arc_write_boost, CTLFLAG_RW, &l2arc_write_boost, 0, "extra write during warmup"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, l2arc_headroom, CTLFLAG_RW, &l2arc_headroom, 0, "number of dev writes"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, l2arc_feed_secs, CTLFLAG_RW, &l2arc_feed_secs, 0, "interval seconds"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, l2arc_feed_min_ms, CTLFLAG_RW, &l2arc_feed_min_ms, 0, "min interval milliseconds"); SYSCTL_INT(_vfs_zfs, OID_AUTO, l2arc_noprefetch, CTLFLAG_RW, &l2arc_noprefetch, 0, "don't cache prefetch bufs"); SYSCTL_INT(_vfs_zfs, OID_AUTO, l2arc_feed_again, CTLFLAG_RW, &l2arc_feed_again, 0, "turbo warmup"); SYSCTL_INT(_vfs_zfs, OID_AUTO, l2arc_norw, CTLFLAG_RW, &l2arc_norw, 0, "no reads during writes"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, anon_size, CTLFLAG_RD, &ARC_anon.arcs_size.rc_count, 0, "size of anonymous state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, anon_metadata_lsize, CTLFLAG_RD, &ARC_anon.arcs_lsize[ARC_BUFC_METADATA], 0, "size of anonymous state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, anon_data_lsize, CTLFLAG_RD, &ARC_anon.arcs_lsize[ARC_BUFC_DATA], 0, "size of anonymous state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mru_size, CTLFLAG_RD, &ARC_mru.arcs_size.rc_count, 0, "size of mru state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mru_metadata_lsize, CTLFLAG_RD, &ARC_mru.arcs_lsize[ARC_BUFC_METADATA], 0, "size of metadata in mru state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mru_data_lsize, CTLFLAG_RD, &ARC_mru.arcs_lsize[ARC_BUFC_DATA], 0, "size of data in mru state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mru_ghost_size, CTLFLAG_RD, &ARC_mru_ghost.arcs_size.rc_count, 0, "size of mru ghost state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mru_ghost_metadata_lsize, CTLFLAG_RD, &ARC_mru_ghost.arcs_lsize[ARC_BUFC_METADATA], 0, "size of metadata in mru ghost state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mru_ghost_data_lsize, CTLFLAG_RD, &ARC_mru_ghost.arcs_lsize[ARC_BUFC_DATA], 0, "size of data in mru ghost state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mfu_size, CTLFLAG_RD, &ARC_mfu.arcs_size.rc_count, 0, "size of mfu state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mfu_metadata_lsize, CTLFLAG_RD, &ARC_mfu.arcs_lsize[ARC_BUFC_METADATA], 0, "size of metadata in mfu state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mfu_data_lsize, CTLFLAG_RD, &ARC_mfu.arcs_lsize[ARC_BUFC_DATA], 0, "size of data in mfu state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mfu_ghost_size, CTLFLAG_RD, &ARC_mfu_ghost.arcs_size.rc_count, 0, "size of mfu ghost state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mfu_ghost_metadata_lsize, CTLFLAG_RD, &ARC_mfu_ghost.arcs_lsize[ARC_BUFC_METADATA], 0, "size of metadata in mfu ghost state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mfu_ghost_data_lsize, CTLFLAG_RD, &ARC_mfu_ghost.arcs_lsize[ARC_BUFC_DATA], 0, "size of data in mfu ghost state"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, l2c_only_size, CTLFLAG_RD, &ARC_l2c_only.arcs_size.rc_count, 0, "size of mru state"); /* * 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_t *l2rcb_buf; /* read buffer */ spa_t *l2rcb_spa; /* spa */ blkptr_t l2rcb_bp; /* original blkptr */ zbookmark_phys_t l2rcb_zb; /* original bookmark */ int l2rcb_flags; /* original flags */ enum zio_compress l2rcb_compress; /* applied compress */ void *l2rcb_data; /* temporary buffer */ } 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; void (*l2df_func)(void *, size_t); 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_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 boolean_t l2arc_write_eligible(uint64_t, arc_buf_hdr_t *); static void l2arc_read_done(zio_t *); static boolean_t l2arc_transform_buf(arc_buf_hdr_t *, boolean_t); static void l2arc_decompress_zio(zio_t *, arc_buf_hdr_t *, enum zio_compress); static void l2arc_release_cdata_buf(arc_buf_hdr_t *); static void l2arc_trim(const arc_buf_hdr_t *hdr) { l2arc_dev_t *dev = hdr->b_l2hdr.b_dev; ASSERT(HDR_HAS_L2HDR(hdr)); ASSERT(MUTEX_HELD(&dev->l2ad_mtx)); if (hdr->b_l2hdr.b_daddr == L2ARC_ADDR_UNSET) return; if (hdr->b_l2hdr.b_asize != 0) { trim_map_free(dev->l2ad_vdev, hdr->b_l2hdr.b_daddr, hdr->b_l2hdr.b_asize, 0); } else { ASSERT3U(hdr->b_l2hdr.b_compress, ==, ZIO_COMPRESS_EMPTY); } } 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 BUF_EMPTY(buf) \ ((buf)->b_dva.dva_word[0] == 0 && \ (buf)->b_dva.dva_word[1] == 0) #define BUF_EQUAL(spa, dva, birth, buf) \ ((buf)->b_dva.dva_word[0] == (dva)->dva_word[0]) && \ ((buf)->b_dva.dva_word[1] == (dva)->dva_word[1]) && \ ((buf)->b_birth == birth) && ((buf)->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 (BUF_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 (BUF_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; hdr->b_flags |= 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) { ASSERT(fhdr != NULL); hdrp = &fhdr->b_hash_next; } *hdrp = hdr->b_hash_next; hdr->b_hash_next = NULL; hdr->b_flags &= ~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(BUF_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(BUF_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 < (uint64_t)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); } } /* * 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) { nhdr->b_flags |= 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_tmp_cdata, ==, NULL); } else { ASSERT(hdr->b_l1hdr.b_buf == NULL); ASSERT0(hdr->b_l1hdr.b_datacnt); /* * 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_tmp_cdata field * might try to be accessed, even though it was removed. */ VERIFY(!HDR_L2_WRITING(hdr)); VERIFY3P(hdr->b_l1hdr.b_tmp_cdata, ==, 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 nhdr->b_flags &= ~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, hdr->b_l2hdr.b_asize, hdr); (void) refcount_add_many(&dev->l2ad_alloc, nhdr->b_l2hdr.b_asize, nhdr); buf_discard_identity(hdr); hdr->b_freeze_cksum = NULL; kmem_cache_free(old, hdr); return (nhdr); } #define ARC_MINTIME (hz>>4) /* 62 ms */ static void arc_cksum_verify(arc_buf_t *buf) { zio_cksum_t zc; if (!(zfs_flags & ZFS_DEBUG_MODIFY)) return; mutex_enter(&buf->b_hdr->b_l1hdr.b_freeze_lock); if (buf->b_hdr->b_freeze_cksum == NULL || HDR_IO_ERROR(buf->b_hdr)) { mutex_exit(&buf->b_hdr->b_l1hdr.b_freeze_lock); return; } fletcher_2_native(buf->b_data, buf->b_hdr->b_size, NULL, &zc); if (!ZIO_CHECKSUM_EQUAL(*buf->b_hdr->b_freeze_cksum, zc)) panic("buffer modified while frozen!"); mutex_exit(&buf->b_hdr->b_l1hdr.b_freeze_lock); } static int arc_cksum_equal(arc_buf_t *buf) { zio_cksum_t zc; int equal; mutex_enter(&buf->b_hdr->b_l1hdr.b_freeze_lock); fletcher_2_native(buf->b_data, buf->b_hdr->b_size, NULL, &zc); equal = ZIO_CHECKSUM_EQUAL(*buf->b_hdr->b_freeze_cksum, zc); mutex_exit(&buf->b_hdr->b_l1hdr.b_freeze_lock); return (equal); } static void arc_cksum_compute(arc_buf_t *buf, boolean_t force) { if (!force && !(zfs_flags & ZFS_DEBUG_MODIFY)) return; mutex_enter(&buf->b_hdr->b_l1hdr.b_freeze_lock); if (buf->b_hdr->b_freeze_cksum != NULL) { mutex_exit(&buf->b_hdr->b_l1hdr.b_freeze_lock); return; } buf->b_hdr->b_freeze_cksum = kmem_alloc(sizeof (zio_cksum_t), KM_SLEEP); fletcher_2_native(buf->b_data, buf->b_hdr->b_size, NULL, buf->b_hdr->b_freeze_cksum); mutex_exit(&buf->b_hdr->b_l1hdr.b_freeze_lock); #ifdef illumos arc_buf_watch(buf); #endif } #ifdef illumos #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 = buf->b_hdr->b_size; ctl.prwatch.pr_wflags = WA_WRITE; result = write(arc_procfd, &ctl, sizeof (ctl)); ASSERT3U(result, ==, sizeof (ctl)); } #endif } #endif /* illumos */ static arc_buf_contents_t arc_buf_type(arc_buf_hdr_t *hdr) { if (HDR_ISTYPE_METADATA(hdr)) { return (ARC_BUFC_METADATA); } else { return (ARC_BUFC_DATA); } } 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) { if (zfs_flags & ZFS_DEBUG_MODIFY) { if (buf->b_hdr->b_l1hdr.b_state != arc_anon) panic("modifying non-anon buffer!"); if (HDR_IO_IN_PROGRESS(buf->b_hdr)) panic("modifying buffer while i/o in progress!"); arc_cksum_verify(buf); } mutex_enter(&buf->b_hdr->b_l1hdr.b_freeze_lock); if (buf->b_hdr->b_freeze_cksum != NULL) { kmem_free(buf->b_hdr->b_freeze_cksum, sizeof (zio_cksum_t)); buf->b_hdr->b_freeze_cksum = NULL; } #ifdef ZFS_DEBUG if (zfs_flags & ZFS_DEBUG_MODIFY) { if (buf->b_hdr->b_l1hdr.b_thawed != NULL) kmem_free(buf->b_hdr->b_l1hdr.b_thawed, 1); buf->b_hdr->b_l1hdr.b_thawed = kmem_alloc(1, KM_SLEEP); } #endif mutex_exit(&buf->b_hdr->b_l1hdr.b_freeze_lock); #ifdef illumos arc_buf_unwatch(buf); #endif } void arc_buf_freeze(arc_buf_t *buf) { kmutex_t *hash_lock; if (!(zfs_flags & ZFS_DEBUG_MODIFY)) return; hash_lock = HDR_LOCK(buf->b_hdr); mutex_enter(hash_lock); ASSERT(buf->b_hdr->b_freeze_cksum != NULL || buf->b_hdr->b_l1hdr.b_state == arc_anon); arc_cksum_compute(buf, B_FALSE); mutex_exit(hash_lock); } static void add_reference(arc_buf_hdr_t *hdr, kmutex_t *hash_lock, void *tag) { ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT(MUTEX_HELD(hash_lock)); 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) { arc_buf_contents_t type = arc_buf_type(hdr); uint64_t delta = hdr->b_size * hdr->b_l1hdr.b_datacnt; multilist_t *list = &state->arcs_list[type]; uint64_t *size = &state->arcs_lsize[type]; multilist_remove(list, hdr); if (GHOST_STATE(state)) { ASSERT0(hdr->b_l1hdr.b_datacnt); ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); delta = hdr->b_size; } ASSERT(delta > 0); ASSERT3U(*size, >=, delta); atomic_add_64(size, -delta); } /* remove the prefetch flag if we get a reference */ hdr->b_flags &= ~ARC_FLAG_PREFETCH; } } 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)) { arc_buf_contents_t type = arc_buf_type(hdr); multilist_t *list = &state->arcs_list[type]; uint64_t *size = &state->arcs_lsize[type]; multilist_insert(list, hdr); ASSERT(hdr->b_l1hdr.b_datacnt > 0); atomic_add_64(size, hdr->b_size * hdr->b_l1hdr.b_datacnt); } 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 datacnt; uint64_t from_delta, to_delta; 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); datacnt = hdr->b_l1hdr.b_datacnt; } else { old_state = arc_l2c_only; refcnt = 0; datacnt = 0; } ASSERT(MUTEX_HELD(hash_lock)); ASSERT3P(new_state, !=, old_state); ASSERT(refcnt == 0 || datacnt > 0); ASSERT(!GHOST_STATE(new_state) || datacnt == 0); ASSERT(old_state != arc_anon || datacnt <= 1); from_delta = to_delta = datacnt * hdr->b_size; /* * 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) { uint64_t *size = &old_state->arcs_lsize[buftype]; ASSERT(HDR_HAS_L1HDR(hdr)); multilist_remove(&old_state->arcs_list[buftype], hdr); /* * If prefetching out of the ghost cache, * we will have a non-zero datacnt. */ if (GHOST_STATE(old_state) && datacnt == 0) { /* ghost elements have a ghost size */ ASSERT(hdr->b_l1hdr.b_buf == NULL); from_delta = hdr->b_size; } ASSERT3U(*size, >=, from_delta); atomic_add_64(size, -from_delta); } if (new_state != arc_anon && new_state != arc_l2c_only) { uint64_t *size = &new_state->arcs_lsize[buftype]; /* * 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); /* ghost elements have a ghost size */ if (GHOST_STATE(new_state)) { ASSERT0(datacnt); ASSERT(hdr->b_l1hdr.b_buf == NULL); to_delta = hdr->b_size; } atomic_add_64(size, to_delta); } } ASSERT(!BUF_EMPTY(hdr)); if (new_state == arc_anon && HDR_IN_HASH_TABLE(hdr)) buf_hash_remove(hdr); /* adjust state sizes (ignore arc_l2c_only) */ if (to_delta && new_state != arc_l2c_only) { ASSERT(HDR_HAS_L1HDR(hdr)); if (GHOST_STATE(new_state)) { ASSERT0(datacnt); /* * We moving a header to a ghost state, we first * remove all arc buffers. Thus, we'll have a * datacnt 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->b_size, hdr); } else { ASSERT3U(datacnt, !=, 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) { (void) refcount_add_many(&new_state->arcs_size, hdr->b_size, buf); } } } if (from_delta && old_state != arc_l2c_only) { ASSERT(HDR_HAS_L1HDR(hdr)); if (GHOST_STATE(old_state)) { /* * When moving a header off of a ghost state, * there's the possibility for datacnt to be * non-zero. This is because we first add the * arc buffer to the header prior to changing * the header's state. Since we used the header * for the reference when putting the header on * the ghost state, we must balance that and use * the header when removing off the ghost state * (even though datacnt is non zero). */ IMPLY(datacnt == 0, new_state == arc_anon || new_state == arc_l2c_only); (void) refcount_remove_many(&old_state->arcs_size, hdr->b_size, hdr); } else { ASSERT3P(datacnt, !=, 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) { (void) refcount_remove_many( &old_state->arcs_size, hdr->b_size, buf); } } } 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])); } 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); } arc_buf_t * arc_buf_alloc(spa_t *spa, int32_t size, void *tag, arc_buf_contents_t type) { arc_buf_hdr_t *hdr; arc_buf_t *buf; ASSERT3U(size, >, 0); hdr = kmem_cache_alloc(hdr_full_cache, KM_PUSHPAGE); ASSERT(BUF_EMPTY(hdr)); ASSERT3P(hdr->b_freeze_cksum, ==, NULL); hdr->b_size = size; hdr->b_spa = spa_load_guid(spa); buf = kmem_cache_alloc(buf_cache, KM_PUSHPAGE); buf->b_hdr = hdr; buf->b_data = NULL; buf->b_efunc = NULL; buf->b_private = NULL; buf->b_next = NULL; hdr->b_flags = arc_bufc_to_flags(type); hdr->b_flags |= ARC_FLAG_HAS_L1HDR; hdr->b_l1hdr.b_buf = buf; hdr->b_l1hdr.b_state = arc_anon; hdr->b_l1hdr.b_arc_access = 0; hdr->b_l1hdr.b_datacnt = 1; hdr->b_l1hdr.b_tmp_cdata = NULL; arc_get_data_buf(buf); ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); (void) refcount_add(&hdr->b_l1hdr.b_refcnt, tag); return (buf); } static char *arc_onloan_tag = "onloan"; /* * 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, int size) { arc_buf_t *buf; buf = arc_buf_alloc(spa, size, arc_onloan_tag, ARC_BUFC_DATA); atomic_add_64(&arc_loaned_bytes, size); 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; ASSERT(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); atomic_add_64(&arc_loaned_bytes, -hdr->b_size); } /* 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; ASSERT(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); buf->b_efunc = NULL; buf->b_private = NULL; atomic_add_64(&arc_loaned_bytes, hdr->b_size); } static arc_buf_t * arc_buf_clone(arc_buf_t *from) { arc_buf_t *buf; arc_buf_hdr_t *hdr = from->b_hdr; uint64_t size = hdr->b_size; ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT(hdr->b_l1hdr.b_state != arc_anon); buf = kmem_cache_alloc(buf_cache, KM_PUSHPAGE); buf->b_hdr = hdr; buf->b_data = NULL; buf->b_efunc = NULL; buf->b_private = NULL; buf->b_next = hdr->b_l1hdr.b_buf; hdr->b_l1hdr.b_buf = buf; arc_get_data_buf(buf); bcopy(from->b_data, buf->b_data, size); /* * This buffer already exists in the arc so create a duplicate * copy for the caller. If the buffer is associated with user data * then track the size and number of duplicates. These stats will be * updated as duplicate buffers are created and destroyed. */ if (HDR_ISTYPE_DATA(hdr)) { ARCSTAT_BUMP(arcstat_duplicate_buffers); ARCSTAT_INCR(arcstat_duplicate_buffers_size, size); } hdr->b_l1hdr.b_datacnt += 1; return (buf); } void arc_buf_add_ref(arc_buf_t *buf, void* tag) { arc_buf_hdr_t *hdr; kmutex_t *hash_lock; /* * Check to see if this buffer is evicted. Callers * must verify b_data != NULL to know if the add_ref * was successful. */ mutex_enter(&buf->b_evict_lock); if (buf->b_data == NULL) { mutex_exit(&buf->b_evict_lock); return; } hash_lock = HDR_LOCK(buf->b_hdr); mutex_enter(hash_lock); hdr = buf->b_hdr; ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); mutex_exit(&buf->b_evict_lock); ASSERT(hdr->b_l1hdr.b_state == arc_mru || hdr->b_l1hdr.b_state == arc_mfu); add_reference(hdr, hash_lock, tag); DTRACE_PROBE1(arc__hit, arc_buf_hdr_t *, hdr); arc_access(hdr, hash_lock); mutex_exit(hash_lock); ARCSTAT_BUMP(arcstat_hits); ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr), demand, prefetch, !HDR_ISTYPE_METADATA(hdr), data, metadata, hits); } static void arc_buf_free_on_write(void *data, size_t size, void (*free_func)(void *, size_t)) { l2arc_data_free_t *df; df = kmem_alloc(sizeof (*df), KM_SLEEP); df->l2df_data = data; df->l2df_size = size; df->l2df_func = free_func; mutex_enter(&l2arc_free_on_write_mtx); list_insert_head(l2arc_free_on_write, df); mutex_exit(&l2arc_free_on_write_mtx); } /* * Free the arc data buffer. If it is an l2arc write in progress, * the buffer is placed on l2arc_free_on_write to be freed later. */ static void arc_buf_data_free(arc_buf_t *buf, void (*free_func)(void *, size_t)) { arc_buf_hdr_t *hdr = buf->b_hdr; if (HDR_L2_WRITING(hdr)) { arc_buf_free_on_write(buf->b_data, hdr->b_size, free_func); ARCSTAT_BUMP(arcstat_l2_free_on_write); } else { free_func(buf->b_data, hdr->b_size); } } static void arc_buf_l2_cdata_free(arc_buf_hdr_t *hdr) { size_t align, asize, len; ASSERT(HDR_HAS_L2HDR(hdr)); ASSERT(MUTEX_HELD(&hdr->b_l2hdr.b_dev->l2ad_mtx)); /* * The b_tmp_cdata field is linked off of the b_l1hdr, so if * that doesn't exist, the header is in the arc_l2c_only state, * and there isn't anything to free (it's already been freed). */ if (!HDR_HAS_L1HDR(hdr)) return; /* * The header isn't being written to the l2arc device, thus it * shouldn't have a b_tmp_cdata to free. */ if (!HDR_L2_WRITING(hdr)) { ASSERT3P(hdr->b_l1hdr.b_tmp_cdata, ==, NULL); return; } /* * The bufer has been chosen for writing to L2ARC, but it's * not being written just yet. In other words, * b_tmp_cdata points to exactly the same buffer as b_data, * l2arc_transform_buf hasn't been called. */ if (hdr->b_l2hdr.b_daddr == L2ARC_ADDR_UNSET) { ASSERT3P(hdr->b_l1hdr.b_tmp_cdata, ==, hdr->b_l1hdr.b_buf->b_data); ASSERT3U(hdr->b_l2hdr.b_compress, ==, ZIO_COMPRESS_OFF); hdr->b_l1hdr.b_tmp_cdata = NULL; return; } /* * There's nothing to free since the buffer was all zero's and * compressed to a zero length buffer. */ if (hdr->b_l2hdr.b_compress == ZIO_COMPRESS_EMPTY) { ASSERT3P(hdr->b_l1hdr.b_tmp_cdata, ==, NULL); return; } /* * Nothing to do if the temporary buffer was not required. */ if (hdr->b_l1hdr.b_tmp_cdata == NULL) return; ARCSTAT_BUMP(arcstat_l2_cdata_free_on_write); len = hdr->b_size; align = (size_t)1 << hdr->b_l2hdr.b_dev->l2ad_vdev->vdev_ashift; asize = P2ROUNDUP(len, align); arc_buf_free_on_write(hdr->b_l1hdr.b_tmp_cdata, asize, zio_data_buf_free); hdr->b_l1hdr.b_tmp_cdata = NULL; } /* * Free up buf->b_data and if 'remove' is set, then pull the * arc_buf_t off of the the arc_buf_hdr_t's list and free it. */ static void arc_buf_destroy(arc_buf_t *buf, boolean_t remove) { arc_buf_t **bufp; /* free up data associated with the buf */ if (buf->b_data != NULL) { arc_state_t *state = buf->b_hdr->b_l1hdr.b_state; uint64_t size = buf->b_hdr->b_size; arc_buf_contents_t type = arc_buf_type(buf->b_hdr); arc_cksum_verify(buf); #ifdef illumos arc_buf_unwatch(buf); #endif if (type == ARC_BUFC_METADATA) { arc_buf_data_free(buf, zio_buf_free); arc_space_return(size, ARC_SPACE_META); } else { ASSERT(type == ARC_BUFC_DATA); arc_buf_data_free(buf, zio_data_buf_free); arc_space_return(size, ARC_SPACE_DATA); } /* protected by hash lock, if in the hash table */ if (multilist_link_active(&buf->b_hdr->b_l1hdr.b_arc_node)) { uint64_t *cnt = &state->arcs_lsize[type]; ASSERT(refcount_is_zero( &buf->b_hdr->b_l1hdr.b_refcnt)); ASSERT(state != arc_anon && state != arc_l2c_only); ASSERT3U(*cnt, >=, size); atomic_add_64(cnt, -size); } (void) refcount_remove_many(&state->arcs_size, size, buf); buf->b_data = NULL; /* * If we're destroying a duplicate buffer make sure * that the appropriate statistics are updated. */ if (buf->b_hdr->b_l1hdr.b_datacnt > 1 && HDR_ISTYPE_DATA(buf->b_hdr)) { ARCSTAT_BUMPDOWN(arcstat_duplicate_buffers); ARCSTAT_INCR(arcstat_duplicate_buffers_size, -size); } ASSERT(buf->b_hdr->b_l1hdr.b_datacnt > 0); buf->b_hdr->b_l1hdr.b_datacnt -= 1; } /* only remove the buf if requested */ if (!remove) return; /* remove the buf from the hdr list */ for (bufp = &buf->b_hdr->b_l1hdr.b_buf; *bufp != buf; bufp = &(*bufp)->b_next) continue; *bufp = buf->b_next; buf->b_next = NULL; ASSERT(buf->b_efunc == NULL); /* clean up the buf */ buf->b_hdr = NULL; kmem_cache_free(buf_cache, 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; ASSERT(MUTEX_HELD(&dev->l2ad_mtx)); ASSERT(HDR_HAS_L2HDR(hdr)); list_remove(&dev->l2ad_buflist, hdr); /* * We don't want to leak the b_tmp_cdata buffer that was * allocated in l2arc_write_buffers() */ arc_buf_l2_cdata_free(hdr); /* * If the l2hdr's b_daddr is equal to L2ARC_ADDR_UNSET, then * this header is being processed by l2arc_write_buffers() (i.e. * it's in the first stage of l2arc_write_buffers()). * Re-affirming that truth here, just to serve as a reminder. If * b_daddr does not equal L2ARC_ADDR_UNSET, then the header may or * may not have its HDR_L2_WRITING flag set. (the write may have * completed, in which case HDR_L2_WRITING will be false and the * b_daddr field will point to the address of the buffer on disk). */ IMPLY(l2hdr->b_daddr == L2ARC_ADDR_UNSET, HDR_L2_WRITING(hdr)); /* * If b_daddr is equal to L2ARC_ADDR_UNSET, we're racing with * l2arc_write_buffers(). Since we've just removed this header * from the l2arc buffer list, this header will never reach the * second stage of l2arc_write_buffers(), which increments the * accounting stats for this header. Thus, we must be careful * not to decrement them for this header either. */ if (l2hdr->b_daddr != L2ARC_ADDR_UNSET) { ARCSTAT_INCR(arcstat_l2_asize, -l2hdr->b_asize); ARCSTAT_INCR(arcstat_l2_size, -hdr->b_size); vdev_space_update(dev->l2ad_vdev, -l2hdr->b_asize, 0, 0); (void) refcount_remove_many(&dev->l2ad_alloc, l2hdr->b_asize, hdr); } hdr->b_flags &= ~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_datacnt > 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_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)) { l2arc_trim(hdr); arc_hdr_l2hdr_destroy(hdr); } if (!buflist_held) mutex_exit(&dev->l2ad_mtx); } if (!BUF_EMPTY(hdr)) buf_discard_identity(hdr); if (hdr->b_freeze_cksum != NULL) { kmem_free(hdr->b_freeze_cksum, sizeof (zio_cksum_t)); hdr->b_freeze_cksum = NULL; } if (HDR_HAS_L1HDR(hdr)) { while (hdr->b_l1hdr.b_buf) { arc_buf_t *buf = hdr->b_l1hdr.b_buf; if (buf->b_efunc != NULL) { mutex_enter(&arc_user_evicts_lock); mutex_enter(&buf->b_evict_lock); ASSERT(buf->b_hdr != NULL); arc_buf_destroy(hdr->b_l1hdr.b_buf, FALSE); hdr->b_l1hdr.b_buf = buf->b_next; buf->b_hdr = &arc_eviction_hdr; buf->b_next = arc_eviction_list; arc_eviction_list = buf; mutex_exit(&buf->b_evict_lock); cv_signal(&arc_user_evicts_cv); mutex_exit(&arc_user_evicts_lock); } else { arc_buf_destroy(hdr->b_l1hdr.b_buf, TRUE); } } #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 } 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_free(arc_buf_t *buf, void *tag) { arc_buf_hdr_t *hdr = buf->b_hdr; int hashed = hdr->b_l1hdr.b_state != arc_anon; ASSERT(buf->b_efunc == NULL); ASSERT(buf->b_data != NULL); if (hashed) { kmutex_t *hash_lock = HDR_LOCK(hdr); mutex_enter(hash_lock); hdr = buf->b_hdr; ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); (void) remove_reference(hdr, hash_lock, tag); if (hdr->b_l1hdr.b_datacnt > 1) { arc_buf_destroy(buf, TRUE); } else { ASSERT(buf == hdr->b_l1hdr.b_buf); ASSERT(buf->b_efunc == NULL); hdr->b_flags |= ARC_FLAG_BUF_AVAILABLE; } mutex_exit(hash_lock); } else if (HDR_IO_IN_PROGRESS(hdr)) { int destroy_hdr; /* * We are in the middle of an async write. Don't destroy * this buffer unless the write completes before we finish * decrementing the reference count. */ mutex_enter(&arc_user_evicts_lock); (void) remove_reference(hdr, NULL, tag); ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); destroy_hdr = !HDR_IO_IN_PROGRESS(hdr); mutex_exit(&arc_user_evicts_lock); if (destroy_hdr) arc_hdr_destroy(hdr); } else { if (remove_reference(hdr, NULL, tag) > 0) arc_buf_destroy(buf, TRUE); else arc_hdr_destroy(hdr); } } boolean_t arc_buf_remove_ref(arc_buf_t *buf, void* tag) { arc_buf_hdr_t *hdr = buf->b_hdr; kmutex_t *hash_lock = HDR_LOCK(hdr); boolean_t no_callback = (buf->b_efunc == NULL); if (hdr->b_l1hdr.b_state == arc_anon) { ASSERT(hdr->b_l1hdr.b_datacnt == 1); arc_buf_free(buf, tag); return (no_callback); } mutex_enter(hash_lock); hdr = buf->b_hdr; ASSERT(hdr->b_l1hdr.b_datacnt > 0); ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); ASSERT(hdr->b_l1hdr.b_state != arc_anon); ASSERT(buf->b_data != NULL); (void) remove_reference(hdr, hash_lock, tag); if (hdr->b_l1hdr.b_datacnt > 1) { if (no_callback) arc_buf_destroy(buf, TRUE); } else if (no_callback) { ASSERT(hdr->b_l1hdr.b_buf == buf && buf->b_next == NULL); ASSERT(buf->b_efunc == NULL); hdr->b_flags |= ARC_FLAG_BUF_AVAILABLE; } ASSERT(no_callback || hdr->b_l1hdr.b_datacnt > 1 || refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); mutex_exit(hash_lock); return (no_callback); } int32_t arc_buf_size(arc_buf_t *buf) { return (buf->b_hdr->b_size); } /* * Called from the DMU to determine if the current buffer should be * evicted. In order to ensure proper locking, the eviction must be initiated * from the DMU. Return true if the buffer is associated with user data and * duplicate buffers still exist. */ boolean_t arc_buf_eviction_needed(arc_buf_t *buf) { arc_buf_hdr_t *hdr; boolean_t evict_needed = B_FALSE; if (zfs_disable_dup_eviction) return (B_FALSE); mutex_enter(&buf->b_evict_lock); hdr = buf->b_hdr; if (hdr == NULL) { /* * We are in arc_do_user_evicts(); let that function * perform the eviction. */ ASSERT(buf->b_data == NULL); mutex_exit(&buf->b_evict_lock); return (B_FALSE); } else if (buf->b_data == NULL) { /* * We have already been added to the arc eviction list; * recommend eviction. */ ASSERT3P(hdr, ==, &arc_eviction_hdr); mutex_exit(&buf->b_evict_lock); return (B_TRUE); } if (hdr->b_l1hdr.b_datacnt > 1 && HDR_ISTYPE_DATA(hdr)) evict_needed = B_TRUE; mutex_exit(&buf->b_evict_lock); return (evict_needed); } /* * 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)); ASSERT(hdr->b_l1hdr.b_buf == NULL); /* * l2arc_write_buffers() relies on a header's L1 portion * (i.e. it's b_tmp_cdata field) during it's 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->b_size; DTRACE_PROBE1(arc__delete, arc_buf_hdr_t *, hdr); 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)); ASSERT3U(hdr->b_l1hdr.b_datacnt, >, 0); 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->b_size; if (buf->b_efunc != NULL) { mutex_enter(&arc_user_evicts_lock); arc_buf_destroy(buf, FALSE); hdr->b_l1hdr.b_buf = buf->b_next; buf->b_hdr = &arc_eviction_hdr; buf->b_next = arc_eviction_list; arc_eviction_list = buf; cv_signal(&arc_user_evicts_cv); mutex_exit(&arc_user_evicts_lock); mutex_exit(&buf->b_evict_lock); } else { mutex_exit(&buf->b_evict_lock); arc_buf_destroy(buf, TRUE); } } if (HDR_HAS_L2HDR(hdr)) { ARCSTAT_INCR(arcstat_evict_l2_cached, hdr->b_size); } else { if (l2arc_write_eligible(hdr->b_spa, hdr)) ARCSTAT_INCR(arcstat_evict_l2_eligible, hdr->b_size); else ARCSTAT_INCR(arcstat_evict_l2_ineligible, hdr->b_size); } if (hdr->b_l1hdr.b_datacnt == 0) { arc_change_state(evicted_state, hdr, hash_lock); ASSERT(HDR_IN_HASH_TABLE(hdr)); hdr->b_flags |= ARC_FLAG_IN_HASH_TABLE; hdr->b_flags &= ~ARC_FLAG_BUF_AVAILABLE; 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]; 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 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 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 (state->arcs_lsize[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 && state->arcs_lsize[type] > 0) { delta = MIN(state->arcs_lsize[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 alloted 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]; 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); } static void arc_do_user_evicts(void) { mutex_enter(&arc_user_evicts_lock); while (arc_eviction_list != NULL) { arc_buf_t *buf = arc_eviction_list; arc_eviction_list = buf->b_next; mutex_enter(&buf->b_evict_lock); buf->b_hdr = NULL; mutex_exit(&buf->b_evict_lock); mutex_exit(&arc_user_evicts_lock); if (buf->b_efunc != NULL) VERIFY0(buf->b_efunc(buf->b_private)); buf->b_efunc = NULL; buf->b_private = NULL; kmem_cache_free(buf_cache, buf); mutex_enter(&arc_user_evicts_lock); } mutex_exit(&arc_user_evicts_lock); } void arc_flush(spa_t *spa, boolean_t retry) { uint64_t guid = 0; /* * If retry is 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); arc_do_user_evicts(); ASSERT(spa || arc_eviction_list == NULL); } void arc_shrink(int64_t to_free) { if (arc_c > arc_c_min) { DTRACE_PROBE4(arc__shrink, uint64_t, arc_c, uint64_t, arc_c_min, uint64_t, arc_p, uint64_t, to_free); 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); DTRACE_PROBE2(arc__shrunk, uint64_t, arc_c, uint64_t, arc_p); ASSERT(arc_c >= arc_c_min); ASSERT((int64_t)arc_p >= 0); } if (arc_size > arc_c) { DTRACE_PROBE2(arc__shrink_adjust, uint64_t, arc_size, uint64_t, arc_c); (void) arc_adjust(); } } static long needfree = 0; 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, FMR_ZIO_FRAG, } 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; } } /* * Cooperate with pagedaemon when it's time for it to scan * and reclaim some pages. */ n = PAGESIZE * ((int64_t)freemem - zfs_arc_free_target); if (n < lowest) { lowest = n; r = FMR_LOTSFREE; } #ifdef illumos /* * 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; } #endif /* illumos */ #if defined(__i386) || !defined(UMA_MD_SMALL_ALLOC) /* * 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; } #define zio_arena NULL #else #define zio_arena 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 1/16th free. * * Note: The 1/16th arena free requirement was put in place * to aggressively evict memory from the arc in order to avoid * memory fragmentation issues. */ if (zio_arena != NULL) { n = (int64_t)vmem_size(zio_arena, VMEM_FREE) - (vmem_size(zio_arena, VMEM_ALLOC) >> 4); if (n < lowest) { lowest = n; r = FMR_ZIO_ARENA; } } /* * Above limits know nothing about real level of KVA fragmentation. * Start aggressive reclamation if too little sequential KVA left. */ if (lowest > 0) { n = (vmem_size(heap_arena, VMEM_MAXFREE) < zfs_max_recordsize) ? -((int64_t)vmem_size(heap_arena, VMEM_ALLOC) >> 4) : INT64_MAX; if (n < lowest) { lowest = n; r = FMR_ZIO_FRAG; } } #else /* _KERNEL */ /* Every 100 calls, free a small amount */ if (spa_get_random(100) == 0) lowest = -1024; #endif /* _KERNEL */ last_free_memory = lowest; last_free_reason = r; DTRACE_PROBE2(arc__available_memory, int64_t, lowest, int, r); return (lowest); } /* * Determine if the system is under memory pressure and is asking * to reclaim memory. A return value of 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); } extern kmem_cache_t *zio_buf_cache[]; extern kmem_cache_t *zio_data_buf_cache[]; extern kmem_cache_t *range_seg_cache; static __noinline void arc_kmem_reap_now(void) { size_t i; kmem_cache_t *prev_cache = NULL; kmem_cache_t *prev_data_cache = NULL; DTRACE_PROBE(arc__kmem_reap_start); #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); #ifdef illumos if (zio_arena != NULL) { /* * Ask the vmem arena to reclaim unused memory from its * quantum caches. */ vmem_qcache_reap(zio_arena); } #endif DTRACE_PROBE(arc__kmem_reap_end); } /* * 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 *dummy __unused) { 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) { int64_t free_memory = arc_available_memory(); uint64_t evicted = 0; mutex_exit(&arc_reclaim_lock); 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; } evicted = arc_adjust(); 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) { #ifdef _KERNEL needfree = 0; #endif /* * 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 = FALSE; cv_broadcast(&arc_reclaim_thread_cv); CALLB_CPR_EXIT(&cpr); /* drops arc_reclaim_lock */ thread_exit(); } static void arc_user_evicts_thread(void *dummy __unused) { callb_cpr_t cpr; CALLB_CPR_INIT(&cpr, &arc_user_evicts_lock, callb_generic_cpr, FTAG); mutex_enter(&arc_user_evicts_lock); while (!arc_user_evicts_thread_exit) { mutex_exit(&arc_user_evicts_lock); arc_do_user_evicts(); /* * 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_enter(&arc_user_evicts_lock); /* * Block until signaled, or after one second (we need to * call the arc's kstat update function regularly). */ CALLB_CPR_SAFE_BEGIN(&cpr); (void) cv_timedwait(&arc_user_evicts_cv, &arc_user_evicts_lock, hz); CALLB_CPR_SAFE_END(&cpr, &arc_user_evicts_lock); } arc_user_evicts_thread_exit = FALSE; cv_broadcast(&arc_user_evicts_cv); CALLB_CPR_EXIT(&cpr); /* drops arc_user_evicts_lock */ thread_exit(); } static u_int arc_dnlc_evicts_arg; extern struct vfsops zfs_vfsops; static void arc_dnlc_evicts_thread(void *dummy __unused) { callb_cpr_t cpr; u_int percent; CALLB_CPR_INIT(&cpr, &arc_dnlc_evicts_lock, callb_generic_cpr, FTAG); mutex_enter(&arc_dnlc_evicts_lock); while (!arc_dnlc_evicts_thread_exit) { CALLB_CPR_SAFE_BEGIN(&cpr); (void) cv_wait(&arc_dnlc_evicts_cv, &arc_dnlc_evicts_lock); CALLB_CPR_SAFE_END(&cpr, &arc_dnlc_evicts_lock); if (arc_dnlc_evicts_arg != 0) { percent = arc_dnlc_evicts_arg; mutex_exit(&arc_dnlc_evicts_lock); #ifdef _KERNEL vnlru_free(desiredvnodes * percent / 100, &zfs_vfsops); #endif mutex_enter(&arc_dnlc_evicts_lock); /* * Clear our token only after vnlru_free() * pass is done, to avoid false queueing of * the requests. */ arc_dnlc_evicts_arg = 0; } } arc_dnlc_evicts_thread_exit = FALSE; cv_broadcast(&arc_dnlc_evicts_cv); CALLB_CPR_EXIT(&cpr); thread_exit(); } void dnlc_reduce_cache(void *arg) { u_int percent; percent = (u_int)(uintptr_t)arg; mutex_enter(&arc_dnlc_evicts_lock); if (arc_dnlc_evicts_arg == 0) { arc_dnlc_evicts_arg = percent; cv_broadcast(&arc_dnlc_evicts_cv); } mutex_exit(&arc_dnlc_evicts_lock); } /* * 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)) { DTRACE_PROBE1(arc__inc_adapt, int, bytes); 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); } /* * The buffer, supplied as the first argument, needs a data block. 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_t *buf) { arc_state_t *state = buf->b_hdr->b_l1hdr.b_state; uint64_t size = buf->b_hdr->b_size; arc_buf_contents_t type = arc_buf_type(buf->b_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); } if (type == ARC_BUFC_METADATA) { buf->b_data = zio_buf_alloc(size); arc_space_consume(size, ARC_SPACE_META); } else { ASSERT(type == ARC_BUFC_DATA); buf->b_data = 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(buf->b_hdr->b_l1hdr.b_state)) { arc_buf_hdr_t *hdr = buf->b_hdr; arc_state_t *state = hdr->b_l1hdr.b_state; (void) refcount_add_many(&state->arcs_size, size, buf); /* * 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)); atomic_add_64(&hdr->b_l1hdr.b_state->arcs_lsize[type], size); } /* * 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); } ARCSTAT_BUMP(arcstat_allocated); } /* * 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 { hdr->b_flags &= ~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) hdr->b_flags &= ~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, buf->b_hdr->b_size); VERIFY(arc_buf_remove_ref(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) { VERIFY(arc_buf_remove_ref(buf, arg)); *bufp = NULL; } else { *bufp = buf; ASSERT(buf->b_data); } } static void arc_read_done(zio_t *zio) { arc_buf_hdr_t *hdr; arc_buf_t *buf; arc_buf_t *abuf; /* buffer we're assigning to callback */ kmutex_t *hash_lock = NULL; arc_callback_t *callback_list, *acb; int freeable = FALSE; buf = zio->io_private; hdr = buf->b_hdr; /* * 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 == NULL && HDR_FREED_IN_READ(hdr) && hash_lock == NULL) || (found == hdr && DVA_EQUAL(&hdr->b_dva, BP_IDENTITY(zio->io_bp))) || (found == hdr && HDR_L2_READING(hdr))); } hdr->b_flags &= ~ARC_FLAG_L2_EVICTED; if (l2arc_noprefetch && HDR_PREFETCH(hdr)) hdr->b_flags &= ~ARC_FLAG_L2CACHE; /* byteswap if necessary */ callback_list = hdr->b_l1hdr.b_acb; ASSERT(callback_list != NULL); if (BP_SHOULD_BYTESWAP(zio->io_bp) && zio->io_error == 0) { dmu_object_byteswap_t bswap = DMU_OT_BYTESWAP(BP_GET_TYPE(zio->io_bp)); arc_byteswap_func_t *func = BP_GET_LEVEL(zio->io_bp) > 0 ? byteswap_uint64_array : dmu_ot_byteswap[bswap].ob_func; func(buf->b_data, hdr->b_size); } arc_cksum_compute(buf, B_FALSE); #ifdef illumos arc_buf_watch(buf); #endif if (hash_lock && zio->io_error == 0 && 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); } /* create copies of the data buffer for the callers */ abuf = buf; for (acb = callback_list; acb; acb = acb->acb_next) { if (acb->acb_done) { if (abuf == NULL) { ARCSTAT_BUMP(arcstat_duplicate_reads); abuf = arc_buf_clone(buf); } acb->acb_buf = abuf; abuf = NULL; } } hdr->b_l1hdr.b_acb = NULL; hdr->b_flags &= ~ARC_FLAG_IO_IN_PROGRESS; ASSERT(!HDR_BUF_AVAILABLE(hdr)); if (abuf == buf) { ASSERT(buf->b_efunc == NULL); ASSERT(hdr->b_l1hdr.b_datacnt == 1); hdr->b_flags |= ARC_FLAG_BUF_AVAILABLE; } ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt) || callback_list != NULL); if (zio->io_error != 0) { hdr->b_flags |= 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; arc_buf_t *buf = NULL; kmutex_t *hash_lock = NULL; zio_t *rzio; uint64_t guid = spa_load_guid(spa); 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_datacnt > 0) { *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) { hdr->b_flags &= ~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; if (pio != NULL) acb->acb_zio_dummy = zio_null(pio, spa, NULL, NULL, NULL, zio_flags); ASSERT(acb->acb_done != NULL); acb->acb_next = hdr->b_l1hdr.b_acb; hdr->b_l1hdr.b_acb = acb; add_reference(hdr, hash_lock, private); 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); hdr->b_flags &= ~ARC_FLAG_PREDICTIVE_PREFETCH; } add_reference(hdr, hash_lock, private); /* * If this block is already in use, create a new * copy of the data so that we will be guaranteed * that arc_release() will always succeed. */ buf = hdr->b_l1hdr.b_buf; ASSERT(buf); ASSERT(buf->b_data); if (HDR_BUF_AVAILABLE(hdr)) { ASSERT(buf->b_efunc == NULL); hdr->b_flags &= ~ARC_FLAG_BUF_AVAILABLE; } else { buf = arc_buf_clone(buf); } } else if (*arc_flags & ARC_FLAG_PREFETCH && refcount_count(&hdr->b_l1hdr.b_refcnt) == 0) { hdr->b_flags |= ARC_FLAG_PREFETCH; } DTRACE_PROBE1(arc__hit, arc_buf_hdr_t *, hdr); arc_access(hdr, hash_lock); if (*arc_flags & ARC_FLAG_L2CACHE) hdr->b_flags |= ARC_FLAG_L2CACHE; if (*arc_flags & ARC_FLAG_L2COMPRESS) hdr->b_flags |= ARC_FLAG_L2COMPRESS; 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 size = BP_GET_LSIZE(bp); arc_callback_t *acb; vdev_t *vd = NULL; uint64_t addr = 0; boolean_t devw = B_FALSE; enum zio_compress b_compress = ZIO_COMPRESS_OFF; int32_t b_asize = 0; 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); buf = arc_buf_alloc(spa, size, private, type); hdr = buf->b_hdr; 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); (void) arc_buf_remove_ref(buf, private); goto top; /* restart the IO request */ } /* * If there is a callback, we pass our reference to * it; otherwise we remove our reference. */ if (done == NULL) { (void) remove_reference(hdr, hash_lock, private); } if (*arc_flags & ARC_FLAG_PREFETCH) hdr->b_flags |= ARC_FLAG_PREFETCH; if (*arc_flags & ARC_FLAG_L2CACHE) hdr->b_flags |= ARC_FLAG_L2CACHE; if (*arc_flags & ARC_FLAG_L2COMPRESS) hdr->b_flags |= ARC_FLAG_L2COMPRESS; if (BP_GET_LEVEL(bp) > 0) hdr->b_flags |= ARC_FLAG_INDIRECT; } 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); } 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); /* * If there is a callback, we pass a reference to it. */ if (done != NULL) add_reference(hdr, hash_lock, private); if (*arc_flags & ARC_FLAG_PREFETCH) hdr->b_flags |= ARC_FLAG_PREFETCH; if (*arc_flags & ARC_FLAG_L2CACHE) hdr->b_flags |= ARC_FLAG_L2CACHE; if (*arc_flags & ARC_FLAG_L2COMPRESS) hdr->b_flags |= ARC_FLAG_L2COMPRESS; buf = kmem_cache_alloc(buf_cache, KM_PUSHPAGE); buf->b_hdr = hdr; buf->b_data = NULL; buf->b_efunc = NULL; buf->b_private = NULL; buf->b_next = NULL; hdr->b_l1hdr.b_buf = buf; ASSERT0(hdr->b_l1hdr.b_datacnt); hdr->b_l1hdr.b_datacnt = 1; arc_get_data_buf(buf); arc_access(hdr, hash_lock); } if (*arc_flags & ARC_FLAG_PREDICTIVE_PREFETCH) hdr->b_flags |= 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; ASSERT(hdr->b_l1hdr.b_acb == NULL); hdr->b_l1hdr.b_acb = acb; hdr->b_flags |= 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; b_compress = hdr->b_l2hdr.b_compress; b_asize = hdr->b_l2hdr.b_asize; /* * Lock out device removal. */ if (vdev_is_dead(vd) || !spa_config_tryenter(spa, SCL_L2ARC, vd, RW_READER)) vd = NULL; } 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->b_size, ==, size); DTRACE_PROBE4(arc__miss, arc_buf_hdr_t *, hdr, blkptr_t *, bp, uint64_t, size, zbookmark_phys_t *, zb); ARCSTAT_BUMP(arcstat_misses); ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr), demand, prefetch, !HDR_ISTYPE_METADATA(hdr), data, metadata, misses); #ifdef _KERNEL #ifdef RACCT if (racct_enable) { PROC_LOCK(curproc); racct_add_force(curproc, RACCT_READBPS, size); racct_add_force(curproc, RACCT_READIOPS, 1); PROC_UNLOCK(curproc); } #endif /* RACCT */ curthread->td_ru.ru_inblock++; #endif if (priority == ZIO_PRIORITY_ASYNC_READ) hdr->b_flags |= ARC_FLAG_PRIO_ASYNC_READ; else hdr->b_flags &= ~ARC_FLAG_PRIO_ASYNC_READ; 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; void* b_data; 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_buf = buf; cb->l2rcb_spa = spa; cb->l2rcb_bp = *bp; cb->l2rcb_zb = *zb; cb->l2rcb_flags = zio_flags; cb->l2rcb_compress = b_compress; if (b_asize > hdr->b_size) { ASSERT3U(b_compress, ==, ZIO_COMPRESS_OFF); b_data = zio_data_buf_alloc(b_asize); cb->l2rcb_data = b_data; } else { b_data = buf->b_data; } ASSERT(addr >= VDEV_LABEL_START_SIZE && addr + size < 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. */ if (b_compress == ZIO_COMPRESS_EMPTY) { ASSERT3U(b_asize, ==, 0); rzio = zio_null(pio, spa, vd, l2arc_read_done, cb, zio_flags | ZIO_FLAG_DONT_CACHE | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY); } else { rzio = zio_read_phys(pio, vd, addr, b_asize, b_data, 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, b_asize); 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, buf->b_data, size, arc_read_done, buf, 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); } void arc_set_callback(arc_buf_t *buf, arc_evict_func_t *func, void *private) { ASSERT(buf->b_hdr != NULL); ASSERT(buf->b_hdr->b_l1hdr.b_state != arc_anon); ASSERT(!refcount_is_zero(&buf->b_hdr->b_l1hdr.b_refcnt) || func == NULL); ASSERT(buf->b_efunc == NULL); ASSERT(!HDR_BUF_AVAILABLE(buf->b_hdr)); buf->b_efunc = func; buf->b_private = private; } /* * 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; if (HDR_BUF_AVAILABLE(hdr)) { arc_buf_t *buf = hdr->b_l1hdr.b_buf; add_reference(hdr, hash_lock, FTAG); hdr->b_flags &= ~ARC_FLAG_BUF_AVAILABLE; mutex_exit(hash_lock); arc_release(buf, FTAG); (void) arc_buf_remove_ref(buf, FTAG); } else { mutex_exit(hash_lock); } } /* * Clear the user eviction callback set by arc_set_callback(), first calling * it if it exists. Because the presence of a callback keeps an arc_buf cached * clearing the callback may result in the arc_buf being destroyed. However, * it will not result in the *last* arc_buf being destroyed, hence the data * will remain cached in the ARC. We make a copy of the arc buffer here so * that we can process the callback without holding any locks. * * It's possible that the callback is already in the process of being cleared * by another thread. In this case we can not clear the callback. * * Returns B_TRUE if the callback was successfully called and cleared. */ boolean_t arc_clear_callback(arc_buf_t *buf) { arc_buf_hdr_t *hdr; kmutex_t *hash_lock; arc_evict_func_t *efunc = buf->b_efunc; void *private = buf->b_private; mutex_enter(&buf->b_evict_lock); hdr = buf->b_hdr; if (hdr == NULL) { /* * We are in arc_do_user_evicts(). */ ASSERT(buf->b_data == NULL); mutex_exit(&buf->b_evict_lock); return (B_FALSE); } else if (buf->b_data == NULL) { /* * We are on the eviction list; process this buffer now * but let arc_do_user_evicts() do the reaping. */ buf->b_efunc = NULL; mutex_exit(&buf->b_evict_lock); VERIFY0(efunc(private)); return (B_TRUE); } hash_lock = HDR_LOCK(hdr); mutex_enter(hash_lock); hdr = buf->b_hdr; ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); ASSERT3U(refcount_count(&hdr->b_l1hdr.b_refcnt), <, hdr->b_l1hdr.b_datacnt); ASSERT(hdr->b_l1hdr.b_state == arc_mru || hdr->b_l1hdr.b_state == arc_mfu); buf->b_efunc = NULL; buf->b_private = NULL; if (hdr->b_l1hdr.b_datacnt > 1) { mutex_exit(&buf->b_evict_lock); arc_buf_destroy(buf, TRUE); } else { ASSERT(buf == hdr->b_l1hdr.b_buf); hdr->b_flags |= ARC_FLAG_BUF_AVAILABLE; mutex_exit(&buf->b_evict_lock); } mutex_exit(hash_lock); VERIFY0(efunc(private)); return (B_TRUE); } /* * 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(BUF_EMPTY(hdr)); ASSERT3U(hdr->b_l1hdr.b_datacnt, ==, 1); ASSERT3S(refcount_count(&hdr->b_l1hdr.b_refcnt), ==, 1); ASSERT(!list_link_active(&hdr->b_l1hdr.b_arc_node)); ASSERT3P(buf->b_efunc, ==, NULL); ASSERT3P(buf->b_private, ==, NULL); hdr->b_l1hdr.b_arc_access = 0; 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 */ ASSERT(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)) { l2arc_trim(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_datacnt > 1) { arc_buf_hdr_t *nhdr; arc_buf_t **bufp; uint64_t blksz = hdr->b_size; uint64_t spa = hdr->b_spa; arc_buf_contents_t type = arc_buf_type(hdr); uint32_t flags = hdr->b_flags; ASSERT(hdr->b_l1hdr.b_buf != buf || buf->b_next != NULL); /* * Pull the data off of this hdr and attach it to * a new anonymous hdr. */ (void) remove_reference(hdr, hash_lock, tag); bufp = &hdr->b_l1hdr.b_buf; while (*bufp != buf) bufp = &(*bufp)->b_next; *bufp = buf->b_next; buf->b_next = NULL; ASSERT3P(state, !=, arc_l2c_only); (void) refcount_remove_many( &state->arcs_size, hdr->b_size, buf); if (refcount_is_zero(&hdr->b_l1hdr.b_refcnt)) { ASSERT3P(state, !=, arc_l2c_only); uint64_t *size = &state->arcs_lsize[type]; ASSERT3U(*size, >=, hdr->b_size); atomic_add_64(size, -hdr->b_size); } /* * We're releasing a duplicate user data buffer, update * our statistics accordingly. */ if (HDR_ISTYPE_DATA(hdr)) { ARCSTAT_BUMPDOWN(arcstat_duplicate_buffers); ARCSTAT_INCR(arcstat_duplicate_buffers_size, -hdr->b_size); } hdr->b_l1hdr.b_datacnt -= 1; arc_cksum_verify(buf); #ifdef illumos arc_buf_unwatch(buf); #endif mutex_exit(hash_lock); nhdr = kmem_cache_alloc(hdr_full_cache, KM_PUSHPAGE); nhdr->b_size = blksz; nhdr->b_spa = spa; nhdr->b_flags = flags & ARC_FLAG_L2_WRITING; nhdr->b_flags |= arc_bufc_to_flags(type); nhdr->b_flags |= ARC_FLAG_HAS_L1HDR; nhdr->b_l1hdr.b_buf = buf; nhdr->b_l1hdr.b_datacnt = 1; nhdr->b_l1hdr.b_state = arc_anon; nhdr->b_l1hdr.b_arc_access = 0; nhdr->b_l1hdr.b_tmp_cdata = NULL; nhdr->b_freeze_cksum = NULL; (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, blksz, 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); } buf->b_efunc = NULL; buf->b_private = NULL; } 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; ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT(!refcount_is_zero(&buf->b_hdr->b_l1hdr.b_refcnt)); ASSERT(hdr->b_l1hdr.b_datacnt > 0); callback->awcb_ready(zio, buf, callback->awcb_private); /* * If the IO is already in progress, then this is a re-write * attempt, so we need to thaw and re-compute the cksum. * It is the responsibility of the callback to handle the * accounting for any re-write attempt. */ if (HDR_IO_IN_PROGRESS(hdr)) { mutex_enter(&hdr->b_l1hdr.b_freeze_lock); if (hdr->b_freeze_cksum != NULL) { kmem_free(hdr->b_freeze_cksum, sizeof (zio_cksum_t)); hdr->b_freeze_cksum = NULL; } mutex_exit(&hdr->b_l1hdr.b_freeze_lock); } arc_cksum_compute(buf, B_FALSE); hdr->b_flags |= ARC_FLAG_IO_IN_PROGRESS; } +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; ASSERT(hdr->b_l1hdr.b_acb == NULL); if (zio->io_error == 0) { 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(BUF_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 (!BUF_EMPTY(hdr)) { arc_buf_hdr_t *exists; kmutex_t *hash_lock; ASSERT(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_datacnt == 1); ASSERT(hdr->b_l1hdr.b_state == arc_anon); ASSERT(BP_GET_DEDUP(zio->io_bp)); ASSERT(BP_GET_LEVEL(zio->io_bp) == 0); } } hdr->b_flags &= ~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 { hdr->b_flags &= ~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, boolean_t l2arc_compress, - const zio_prop_t *zp, arc_done_func_t *ready, arc_done_func_t *physdone, + 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; ASSERT(ready != NULL); ASSERT(done != NULL); ASSERT(!HDR_IO_ERROR(hdr)); ASSERT(!HDR_IO_IN_PROGRESS(hdr)); ASSERT(hdr->b_l1hdr.b_acb == NULL); ASSERT(hdr->b_l1hdr.b_datacnt > 0); if (l2arc) hdr->b_flags |= ARC_FLAG_L2CACHE; if (l2arc_compress) hdr->b_flags |= ARC_FLAG_L2COMPRESS; 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; zio = zio_write(pio, spa, txg, bp, buf->b_data, hdr->b_size, zp, - arc_write_ready, arc_write_physdone, arc_write_done, callback, + 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) || !defined(UMA_MD_SMALL_ALLOC) available_memory = MIN(available_memory, ptob(vmem_size(heap_arena, VMEM_FREE))); #endif if (freemem > (uint64_t)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 == pageproc) { 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); DTRACE_PROBE1(arc__set_reserve, uint64_t, arc_c); } 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. */ 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) { dprintf("failing, arc_tempreserve=%lluK anon_meta=%lluK " "anon_data=%lluK tempreserve=%lluK arc_c=%lluK\n", arc_tempreserve>>10, arc_anon->arcs_lsize[ARC_BUFC_METADATA]>>10, arc_anon->arcs_lsize[ARC_BUFC_DATA]>>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 = state->arcs_lsize[ARC_BUFC_DATA]; evict_metadata->value.ui64 = state->arcs_lsize[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(!BUF_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)); } #ifdef _KERNEL static eventhandler_tag arc_event_lowmem = NULL; static void arc_lowmem(void *arg __unused, int howto __unused) { mutex_enter(&arc_reclaim_lock); /* XXX: Memory deficit should be passed as argument. */ needfree = btoc(arc_c >> arc_shrink_shift); DTRACE_PROBE(arc__needfree); cv_signal(&arc_reclaim_thread_cv); /* * It is unsafe to block here in arbitrary threads, because we can come * here from ARC itself and may hold ARC locks and thus risk a deadlock * with ARC reclaim thread. */ if (curproc == pageproc) (void) cv_wait(&arc_reclaim_waiters_cv, &arc_reclaim_lock); mutex_exit(&arc_reclaim_lock); } #endif void arc_init(void) { int i, prefetch_tunable_set = 0; 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); mutex_init(&arc_user_evicts_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&arc_user_evicts_cv, NULL, CV_DEFAULT, NULL); mutex_init(&arc_dnlc_evicts_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&arc_dnlc_evicts_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 = kmem_size() / 8; #ifdef illumos #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 #endif /* illumos */ /* set min cache to 1/32 of all memory, or arc_abs_min, whichever is more */ arc_c_min = MAX(arc_c / 4, arc_abs_min); /* set max to 1/2 of all memory, or all but 1GB, whichever is more */ if (arc_c * 8 >= 1 << 30) arc_c_max = (arc_c * 8) - (1 << 30); else arc_c_max = arc_c_min; arc_c_max = MAX(arc_c * 5, 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 #ifdef _KERNEL /* * Allow the tunables to override our calculations if they are * reasonable. */ if (zfs_arc_max > arc_abs_min && zfs_arc_max < kmem_size()) arc_c_max = zfs_arc_max; if (zfs_arc_min > arc_abs_min && zfs_arc_min <= arc_c_max) arc_c_min = zfs_arc_min; #endif arc_c = arc_c_max; arc_p = (arc_c >> 1); /* 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 (zfs_arc_num_sublists_per_state < 1) zfs_arc_num_sublists_per_state = MAX(max_ncpus, 1); /* 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; zfs_arc_min = arc_c_min; zfs_arc_max = arc_c_max; 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; arc_size = 0; multilist_create(&arc_mru->arcs_list[ARC_BUFC_METADATA], sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); multilist_create(&arc_mru->arcs_list[ARC_BUFC_DATA], sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); multilist_create(&arc_mru_ghost->arcs_list[ARC_BUFC_METADATA], sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); multilist_create(&arc_mru_ghost->arcs_list[ARC_BUFC_DATA], sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); multilist_create(&arc_mfu->arcs_list[ARC_BUFC_METADATA], sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); multilist_create(&arc_mfu->arcs_list[ARC_BUFC_DATA], sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); multilist_create(&arc_mfu_ghost->arcs_list[ARC_BUFC_METADATA], sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); multilist_create(&arc_mfu_ghost->arcs_list[ARC_BUFC_DATA], sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); multilist_create(&arc_l2c_only->arcs_list[ARC_BUFC_METADATA], sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); multilist_create(&arc_l2c_only->arcs_list[ARC_BUFC_DATA], sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); 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); buf_init(); arc_reclaim_thread_exit = FALSE; arc_user_evicts_thread_exit = FALSE; arc_dnlc_evicts_thread_exit = FALSE; arc_eviction_list = NULL; bzero(&arc_eviction_hdr, sizeof (arc_buf_hdr_t)); 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); #ifdef _KERNEL arc_event_lowmem = EVENTHANDLER_REGISTER(vm_lowmem, arc_lowmem, NULL, EVENTHANDLER_PRI_FIRST); #endif (void) thread_create(NULL, 0, arc_user_evicts_thread, NULL, 0, &p0, TS_RUN, minclsyspri); (void) thread_create(NULL, 0, arc_dnlc_evicts_thread, NULL, 0, &p0, TS_RUN, minclsyspri); arc_dead = 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 = ptob(physmem) * zfs_dirty_data_max_percent / 100; zfs_dirty_data_max = MIN(zfs_dirty_data_max, zfs_dirty_data_max_max); } #ifdef _KERNEL if (TUNABLE_INT_FETCH("vfs.zfs.prefetch_disable", &zfs_prefetch_disable)) prefetch_tunable_set = 1; #ifdef __i386__ if (prefetch_tunable_set == 0) { printf("ZFS NOTICE: Prefetch is disabled by default on i386 " "-- to enable,\n"); printf(" add \"vfs.zfs.prefetch_disable=0\" " "to /boot/loader.conf.\n"); zfs_prefetch_disable = 1; } #else if ((((uint64_t)physmem * PAGESIZE) < (1ULL << 32)) && prefetch_tunable_set == 0) { printf("ZFS NOTICE: Prefetch is disabled by default if less " "than 4GB of RAM is present;\n" " to enable, add \"vfs.zfs.prefetch_disable=0\" " "to /boot/loader.conf.\n"); zfs_prefetch_disable = 1; } #endif /* Warn about ZFS memory and address space requirements. */ if (((uint64_t)physmem * PAGESIZE) < (256 + 128 + 64) * (1 << 20)) { printf("ZFS WARNING: Recommended minimum RAM size is 512MB; " "expect unstable behavior.\n"); } if (kmem_size() < 512 * (1 << 20)) { printf("ZFS WARNING: Recommended minimum kmem_size is 512MB; " "expect unstable behavior.\n"); printf(" Consider tuning vm.kmem_size and " "vm.kmem_size_max\n"); printf(" in /boot/loader.conf.\n"); } #endif } void arc_fini(void) { mutex_enter(&arc_reclaim_lock); arc_reclaim_thread_exit = TRUE; /* * The reclaim thread will set arc_reclaim_thread_exit back to * 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); mutex_enter(&arc_user_evicts_lock); arc_user_evicts_thread_exit = TRUE; /* * The user evicts thread will set arc_user_evicts_thread_exit * to FALSE when it is finished exiting; we're waiting for that. */ while (arc_user_evicts_thread_exit) { cv_signal(&arc_user_evicts_cv); cv_wait(&arc_user_evicts_cv, &arc_user_evicts_lock); } mutex_exit(&arc_user_evicts_lock); mutex_enter(&arc_dnlc_evicts_lock); arc_dnlc_evicts_thread_exit = TRUE; /* * The user evicts thread will set arc_user_evicts_thread_exit * to FALSE when it is finished exiting; we're waiting for that. */ while (arc_dnlc_evicts_thread_exit) { cv_signal(&arc_dnlc_evicts_cv); cv_wait(&arc_dnlc_evicts_cv, &arc_dnlc_evicts_lock); } mutex_exit(&arc_dnlc_evicts_lock); /* Use TRUE to ensure *all* buffers are evicted */ arc_flush(NULL, TRUE); arc_dead = 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); mutex_destroy(&arc_user_evicts_lock); cv_destroy(&arc_user_evicts_cv); mutex_destroy(&arc_dnlc_evicts_lock); cv_destroy(&arc_dnlc_evicts_cv); 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_l2c_only->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_l2c_only->arcs_list[ARC_BUFC_DATA]); buf_fini(); ASSERT0(arc_loaned_bytes); #ifdef _KERNEL if (arc_event_lowmem != NULL) EVENTHANDLER_DEREGISTER(vm_lowmem, arc_event_lowmem); #endif } /* * 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) { ARCSTAT_BUMP(arcstat_l2_write_spa_mismatch); return (B_FALSE); } if (HDR_HAS_L2HDR(hdr)) { ARCSTAT_BUMP(arcstat_l2_write_in_l2); return (B_FALSE); } if (HDR_IO_IN_PROGRESS(hdr)) { ARCSTAT_BUMP(arcstat_l2_write_hdr_io_in_progress); return (B_FALSE); } if (!HDR_L2CACHE(hdr)) { ARCSTAT_BUMP(arcstat_l2_write_not_cacheable); 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); ASSERT(df->l2df_data != NULL); ASSERT(df->l2df_func != NULL); df->l2df_func(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; ASSERT(cb != NULL); dev = cb->l2wcb_dev; ASSERT(dev != NULL); head = cb->l2wcb_head; ASSERT(head != NULL); buflist = &dev->l2ad_buflist; ASSERT(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)); /* * We may have allocated a buffer for L2ARC compression, * we must release it to avoid leaking this data. */ l2arc_release_cdata_buf(hdr); if (zio->io_error != 0) { /* * Error - drop L2ARC entry. */ list_remove(buflist, hdr); l2arc_trim(hdr); hdr->b_flags &= ~ARC_FLAG_HAS_L2HDR; ARCSTAT_INCR(arcstat_l2_asize, -hdr->b_l2hdr.b_asize); ARCSTAT_INCR(arcstat_l2_size, -hdr->b_size); bytes_dropped += hdr->b_l2hdr.b_asize; (void) refcount_remove_many(&dev->l2ad_alloc, hdr->b_l2hdr.b_asize, hdr); } /* * Allow ARC to begin reads and ghost list evictions to * this L2ARC entry. */ hdr->b_flags &= ~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; arc_buf_t *buf; kmutex_t *hash_lock; int equal; ASSERT(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; ASSERT(cb != NULL); buf = cb->l2rcb_buf; ASSERT(buf != NULL); hash_lock = HDR_LOCK(buf->b_hdr); mutex_enter(hash_lock); hdr = buf->b_hdr; ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); /* * If the data was read into a temporary buffer, * move it and free the buffer. */ if (cb->l2rcb_data != NULL) { ASSERT3U(hdr->b_size, <, zio->io_size); ASSERT3U(cb->l2rcb_compress, ==, ZIO_COMPRESS_OFF); if (zio->io_error == 0) bcopy(cb->l2rcb_data, buf->b_data, hdr->b_size); /* * The following must be done regardless of whether * there was an error: * - free the temporary buffer * - point zio to the real ARC buffer * - set zio size accordingly * These are required because zio is either re-used for * an I/O of the block in the case of the error * or the zio is passed to arc_read_done() and it * needs real data. */ zio_data_buf_free(cb->l2rcb_data, zio->io_size); zio->io_size = zio->io_orig_size = hdr->b_size; zio->io_data = zio->io_orig_data = buf->b_data; } /* * If the buffer was compressed, decompress it first. */ if (cb->l2rcb_compress != ZIO_COMPRESS_OFF) l2arc_decompress_zio(zio, hdr, cb->l2rcb_compress); ASSERT(zio->io_data != NULL); ASSERT3U(zio->io_size, ==, hdr->b_size); ASSERT3U(BP_GET_LSIZE(&cb->l2rcb_bp), ==, hdr->b_size); /* * Check this survived the L2ARC journey. */ equal = arc_cksum_equal(buf); if (equal && zio->io_error == 0 && !HDR_L2_EVICTED(hdr)) { mutex_exit(hash_lock); zio->io_private = buf; 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 */ 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 (!equal) 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, cb->l2rcb_spa, &cb->l2rcb_bp, buf->b_data, hdr->b_size, arc_read_done, buf, 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]; break; case 1: ml = &arc_mru->arcs_list[ARC_BUFC_METADATA]; break; case 2: ml = &arc_mfu->arcs_list[ARC_BUFC_DATA]; break; case 3: 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); hdr->b_flags |= ARC_FLAG_L2_EVICTED; } /* Ensure this header has finished being written */ ASSERT(!HDR_L2_WRITING(hdr)); ASSERT3P(hdr->b_l1hdr.b_tmp_cdata, ==, NULL); 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, boolean_t *headroom_boost) { arc_buf_hdr_t *hdr, *hdr_prev, *head; uint64_t write_asize, write_sz, headroom, buf_compress_minsz; void *buf_data; boolean_t full; l2arc_write_callback_t *cb; zio_t *pio, *wzio; uint64_t guid = spa_load_guid(spa); const boolean_t do_headroom_boost = *headroom_boost; int try; ASSERT(dev->l2ad_vdev != NULL); /* Lower the flag now, we might want to raise it again later. */ *headroom_boost = B_FALSE; pio = NULL; write_sz = write_asize = 0; full = B_FALSE; head = kmem_cache_alloc(hdr_l2only_cache, KM_PUSHPAGE); head->b_flags |= ARC_FLAG_L2_WRITE_HEAD; head->b_flags |= ARC_FLAG_HAS_L2HDR; ARCSTAT_BUMP(arcstat_l2_write_buffer_iter); /* * We will want to try to compress buffers that are at least 2x the * device sector size. */ buf_compress_minsz = 2 << dev->l2ad_vdev->vdev_ashift; /* * Copy buffers for L2ARC writing. */ for (try = 0; try <= 3; try++) { multilist_sublist_t *mls = l2arc_sublist_lock(try); uint64_t passed_sz = 0; ARCSTAT_BUMP(arcstat_l2_write_buffer_list_iter); /* * 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); if (hdr == NULL) ARCSTAT_BUMP(arcstat_l2_write_buffer_list_null_iter); headroom = target_sz * l2arc_headroom; if (do_headroom_boost) headroom = (headroom * l2arc_headroom_boost) / 100; for (; hdr; hdr = hdr_prev) { kmutex_t *hash_lock; uint64_t buf_sz; uint64_t buf_a_sz; size_t align; if (arc_warm == B_FALSE) hdr_prev = multilist_sublist_next(mls, hdr); else hdr_prev = multilist_sublist_prev(mls, hdr); ARCSTAT_INCR(arcstat_l2_write_buffer_bytes_scanned, hdr->b_size); hash_lock = HDR_LOCK(hdr); if (!mutex_tryenter(hash_lock)) { ARCSTAT_BUMP(arcstat_l2_write_trylock_fail); /* * Skip this buffer rather than waiting. */ continue; } passed_sz += hdr->b_size; if (passed_sz > headroom) { /* * Searched too far. */ mutex_exit(hash_lock); ARCSTAT_BUMP(arcstat_l2_write_passed_headroom); break; } if (!l2arc_write_eligible(guid, hdr)) { mutex_exit(hash_lock); continue; } /* * Assume that the buffer is not going to be compressed * and could take more space on disk because of a larger * disk block size. */ buf_sz = hdr->b_size; align = (size_t)1 << dev->l2ad_vdev->vdev_ashift; buf_a_sz = P2ROUNDUP(buf_sz, align); if ((write_asize + buf_a_sz) > target_sz) { full = B_TRUE; mutex_exit(hash_lock); ARCSTAT_BUMP(arcstat_l2_write_full); 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); ARCSTAT_BUMP(arcstat_l2_write_pios); } /* * Create and add a new L2ARC header. */ hdr->b_l2hdr.b_dev = dev; hdr->b_flags |= ARC_FLAG_L2_WRITING; /* * Temporarily stash the data buffer in b_tmp_cdata. * The subsequent write step will pick it up from * there. This is because can't access b_l1hdr.b_buf * without holding the hash_lock, which we in turn * can't access without holding the ARC list locks * (which we want to avoid during compression/writing). */ hdr->b_l2hdr.b_compress = ZIO_COMPRESS_OFF; hdr->b_l2hdr.b_asize = hdr->b_size; hdr->b_l1hdr.b_tmp_cdata = hdr->b_l1hdr.b_buf->b_data; /* * Explicitly set the b_daddr field to a known * value which means "invalid address". This * enables us to differentiate which stage of * l2arc_write_buffers() the particular header * is in (e.g. this loop, or the one below). * ARC_FLAG_L2_WRITING is not enough to make * this distinction, and we need to know in * order to do proper l2arc vdev accounting in * arc_release() and arc_hdr_destroy(). * * Note, we can't use a new flag to distinguish * the two stages because we don't hold the * header's hash_lock below, in the second stage * of this function. Thus, we can't simply * change the b_flags field to denote that the * IO has been sent. We can change the b_daddr * field of the L2 portion, though, since we'll * be holding the l2ad_mtx; which is why we're * using it to denote the header's state change. */ hdr->b_l2hdr.b_daddr = L2ARC_ADDR_UNSET; hdr->b_flags |= ARC_FLAG_HAS_L2HDR; mutex_enter(&dev->l2ad_mtx); list_insert_head(&dev->l2ad_buflist, hdr); mutex_exit(&dev->l2ad_mtx); /* * Compute and store the buffer cksum before * writing. On debug the cksum is verified first. */ arc_cksum_verify(hdr->b_l1hdr.b_buf); arc_cksum_compute(hdr->b_l1hdr.b_buf, B_TRUE); mutex_exit(hash_lock); write_sz += buf_sz; write_asize += buf_a_sz; } 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); } mutex_enter(&dev->l2ad_mtx); /* * Now start writing the buffers. We're starting at the write head * and work backwards, retracing the course of the buffer selector * loop above. */ write_asize = 0; for (hdr = list_prev(&dev->l2ad_buflist, head); hdr; hdr = list_prev(&dev->l2ad_buflist, hdr)) { uint64_t buf_sz; boolean_t compress; /* * 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)); /* * We shouldn't need to lock the buffer here, since we flagged * it as ARC_FLAG_L2_WRITING in the previous step, but we must * take care to only access its L2 cache parameters. In * particular, hdr->l1hdr.b_buf may be invalid by now due to * ARC eviction. */ hdr->b_l2hdr.b_daddr = dev->l2ad_hand; /* * Save a pointer to the original buffer data we had previously * stashed away. */ buf_data = hdr->b_l1hdr.b_tmp_cdata; compress = HDR_L2COMPRESS(hdr) && hdr->b_l2hdr.b_asize >= buf_compress_minsz; if (l2arc_transform_buf(hdr, compress)) { /* * If compression succeeded, enable headroom * boost on the next scan cycle. */ *headroom_boost = B_TRUE; } /* * Get the new buffer size that accounts for compression * and padding. */ buf_sz = hdr->b_l2hdr.b_asize; /* * We need to do this regardless if buf_sz is zero or * not, otherwise, when this l2hdr is evicted we'll * remove a reference that was never added. */ (void) refcount_add_many(&dev->l2ad_alloc, buf_sz, hdr); /* Compression may have squashed the buffer to zero length. */ if (buf_sz != 0) { /* * If the data was padded or compressed, then it * it is in a new buffer. */ if (hdr->b_l1hdr.b_tmp_cdata != NULL) buf_data = hdr->b_l1hdr.b_tmp_cdata; wzio = zio_write_phys(pio, dev->l2ad_vdev, dev->l2ad_hand, buf_sz, buf_data, ZIO_CHECKSUM_OFF, NULL, NULL, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_CANFAIL, B_FALSE); DTRACE_PROBE2(l2arc__write, vdev_t *, dev->l2ad_vdev, zio_t *, wzio); (void) zio_nowait(wzio); write_asize += buf_sz; dev->l2ad_hand += buf_sz; } } mutex_exit(&dev->l2ad_mtx); 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); } /* * Transforms, possibly compresses and pads, an L2ARC buffer. * The data to be compressed must be prefilled in l1hdr.b_tmp_cdata and its * size in l2hdr->b_asize. This routine tries to compress the data and * depending on the compression result there are three possible outcomes: * *) The buffer was incompressible. The buffer size was already ashift aligned. * The original hdr contents were left untouched except for b_tmp_cdata, * which is reset to NULL. The caller must keep a pointer to the original * data. * *) The buffer was incompressible. The buffer size was not ashift aligned. * b_tmp_cdata was replaced with a temporary data buffer which holds a padded * (aligned) copy of the data. Once writing is done, invoke * l2arc_release_cdata_buf on this hdr to free the temporary buffer. * *) The buffer was all-zeros, so there is no need to write it to an L2 * device. To indicate this situation b_tmp_cdata is NULL'ed, b_asize is * set to zero and b_compress is set to ZIO_COMPRESS_EMPTY. * *) Compression succeeded and b_tmp_cdata was replaced with a temporary * data buffer which holds the compressed data to be written, and b_asize * tells us how much data there is. b_compress is set to the appropriate * compression algorithm. Once writing is done, invoke * l2arc_release_cdata_buf on this l2hdr to free this temporary buffer. * * Returns B_TRUE if compression succeeded, or B_FALSE if it didn't (the * buffer was incompressible). */ static boolean_t l2arc_transform_buf(arc_buf_hdr_t *hdr, boolean_t compress) { void *cdata; size_t align, asize, csize, len, rounded; ASSERT(HDR_HAS_L2HDR(hdr)); l2arc_buf_hdr_t *l2hdr = &hdr->b_l2hdr; ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT3S(l2hdr->b_compress, ==, ZIO_COMPRESS_OFF); ASSERT(hdr->b_l1hdr.b_tmp_cdata != NULL); len = l2hdr->b_asize; align = (size_t)1 << l2hdr->b_dev->l2ad_vdev->vdev_ashift; asize = P2ROUNDUP(len, align); cdata = zio_data_buf_alloc(asize); ASSERT3P(cdata, !=, NULL); if (compress) csize = zio_compress_data(ZIO_COMPRESS_LZ4, hdr->b_l1hdr.b_tmp_cdata, cdata, len); else csize = len; if (csize == 0) { /* zero block, indicate that there's nothing to write */ zio_data_buf_free(cdata, asize); l2hdr->b_compress = ZIO_COMPRESS_EMPTY; l2hdr->b_asize = 0; hdr->b_l1hdr.b_tmp_cdata = NULL; ARCSTAT_BUMP(arcstat_l2_compress_zeros); return (B_TRUE); } rounded = P2ROUNDUP(csize, align); ASSERT3U(rounded, <=, asize); if (rounded < len) { /* * Compression succeeded, we'll keep the cdata around for * writing and release it afterwards. */ if (rounded > csize) { bzero((char *)cdata + csize, rounded - csize); csize = rounded; } l2hdr->b_compress = ZIO_COMPRESS_LZ4; l2hdr->b_asize = csize; hdr->b_l1hdr.b_tmp_cdata = cdata; ARCSTAT_BUMP(arcstat_l2_compress_successes); return (B_TRUE); } else { /* * Compression did not save space. */ if (P2PHASE(len, align) != 0) { /* * Use compression buffer for a copy of data padded to * the proper size. Compression algorithm remains set * to ZIO_COMPRESS_OFF. */ ASSERT3U(len, <, asize); bcopy(hdr->b_l1hdr.b_tmp_cdata, cdata, len); bzero((char *)cdata + len, asize - len); l2hdr->b_asize = asize; hdr->b_l1hdr.b_tmp_cdata = cdata; ARCSTAT_BUMP(arcstat_l2_padding_needed); } else { ASSERT3U(len, ==, asize); /* * The original buffer is good as is, * release the compressed buffer. * l2hdr will be left unmodified except for b_tmp_cdata. */ zio_data_buf_free(cdata, asize); hdr->b_l1hdr.b_tmp_cdata = NULL; } if (compress) ARCSTAT_BUMP(arcstat_l2_compress_failures); return (B_FALSE); } } /* * Decompresses a zio read back from an l2arc device. On success, the * underlying zio's io_data buffer is overwritten by the uncompressed * version. On decompression error (corrupt compressed stream), the * zio->io_error value is set to signal an I/O error. * * Please note that the compressed data stream is not checksummed, so * if the underlying device is experiencing data corruption, we may feed * corrupt data to the decompressor, so the decompressor needs to be * able to handle this situation (LZ4 does). */ static void l2arc_decompress_zio(zio_t *zio, arc_buf_hdr_t *hdr, enum zio_compress c) { ASSERT(L2ARC_IS_VALID_COMPRESS(c)); if (zio->io_error != 0) { /* * An io error has occured, just restore the original io * size in preparation for a main pool read. */ zio->io_orig_size = zio->io_size = hdr->b_size; return; } if (c == ZIO_COMPRESS_EMPTY) { /* * An empty buffer results in a null zio, which means we * need to fill its io_data after we're done restoring the * buffer's contents. */ ASSERT(hdr->b_l1hdr.b_buf != NULL); bzero(hdr->b_l1hdr.b_buf->b_data, hdr->b_size); zio->io_data = zio->io_orig_data = hdr->b_l1hdr.b_buf->b_data; } else { ASSERT(zio->io_data != NULL); /* * We copy the compressed data from the start of the arc buffer * (the zio_read will have pulled in only what we need, the * rest is garbage which we will overwrite at decompression) * and then decompress back to the ARC data buffer. This way we * can minimize copying by simply decompressing back over the * original compressed data (rather than decompressing to an * aux buffer and then copying back the uncompressed buffer, * which is likely to be much larger). */ uint64_t csize; void *cdata; csize = zio->io_size; cdata = zio_data_buf_alloc(csize); bcopy(zio->io_data, cdata, csize); if (zio_decompress_data(c, cdata, zio->io_data, csize, hdr->b_size) != 0) zio->io_error = EIO; zio_data_buf_free(cdata, csize); } /* Restore the expected uncompressed IO size. */ zio->io_orig_size = zio->io_size = hdr->b_size; } /* * Releases the temporary b_tmp_cdata buffer in an l2arc header structure. * This buffer serves as a temporary holder of compressed or padded data while * the buffer entry is being written to an l2arc device. Once that is * done, we can dispose of it. */ static void l2arc_release_cdata_buf(arc_buf_hdr_t *hdr) { size_t align, asize, len; enum zio_compress comp = hdr->b_l2hdr.b_compress; ASSERT(HDR_HAS_L2HDR(hdr)); ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT(comp == ZIO_COMPRESS_OFF || L2ARC_IS_VALID_COMPRESS(comp)); if (hdr->b_l1hdr.b_tmp_cdata != NULL) { ASSERT(comp != ZIO_COMPRESS_EMPTY); len = hdr->b_size; align = (size_t)1 << hdr->b_l2hdr.b_dev->l2ad_vdev->vdev_ashift; asize = P2ROUNDUP(len, align); zio_data_buf_free(hdr->b_l1hdr.b_tmp_cdata, asize); hdr->b_l1hdr.b_tmp_cdata = NULL; } else { ASSERT(comp == ZIO_COMPRESS_OFF || comp == ZIO_COMPRESS_EMPTY); } } /* * This thread feeds the L2ARC at regular intervals. This is the beating * heart of the L2ARC. */ static void l2arc_feed_thread(void *dummy __unused) { callb_cpr_t cpr; l2arc_dev_t *dev; spa_t *spa; uint64_t size, wrote; clock_t begin, next = ddi_get_lbolt(); boolean_t headroom_boost = B_FALSE; 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 - ddi_get_lbolt()); 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; ASSERT(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, &headroom_boost); /* * 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)); vdev_ashift_optimize(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; } } ASSERT(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: stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dbuf.c =================================================================== --- stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dbuf.c (revision 304137) +++ stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dbuf.c (revision 304138) @@ -1,3233 +1,3342 @@ /* * 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, 2015 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 /* * Number of times that zfs_free_range() took the slow path while doing * a zfs receive. A nonzero value indicates a potential performance problem. */ uint64_t zfs_free_range_recv_miss; static void dbuf_destroy(dmu_buf_impl_t *db); 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, dmu_buf_t **clear_on_evict_dbufp); #endif /* ! __lint */ /* * Global data structures and functions for the dbuf cache. */ static kmem_cache_t *dbuf_cache; static taskq_t *dbu_evict_taskq; /* 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); 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); 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_HASH(os, obj, level, blkid) dbuf_hash(os, obj, level, blkid); #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); } static arc_evict_func_t dbuf_do_evict; 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 /* * Invoke the callback from a taskq to avoid lock order reversals * and limit stack depth. */ taskq_dispatch_ent(dbu_evict_taskq, dbu->dbu_evict_func, 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); } } void dbuf_evict(dmu_buf_impl_t *db) { ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(db->db_buf == NULL); ASSERT(db->db_data_pending == NULL); dbuf_clear(db); dbuf_destroy(db); } 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 < (uint64_t)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_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); /* * 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); } 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_cache); taskq_destroy(dbu_evict_taskq); } /* * 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) { - uint64_t *buf = db->db.db_data; - int i; + 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); + 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); 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; if (!arc_released(buf)) arc_set_callback(buf, dbuf_do_evict, db); } /* * 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; 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, blksz); bcopy(db->db.db_data, abuf->b_data, blksz); } else { abuf = db->db_buf; arc_loan_inuse_buf(abuf, db); 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); VERIFY(arc_buf_remove_ref(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); - DB_DNODE_EXIT(db); dbuf_set_data(db, arc_buf_alloc(db->db_objset->os_spa, db->db.db_size, db, type)); 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; if (DBUF_IS_L2COMPRESSIBLE(db)) aflags |= ARC_FLAG_L2COMPRESS; 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); } 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) { 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_buf_alloc(spa, db->db.db_size, db, type)); 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); } /* * 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 modify them in the current active group. * * This function is used in two places: when we are dirtying a * buffer for the first time in a txg, and when we are freeing * a range in a dnode that includes this buffer. * * 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 = db->db.db_size; arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db); spa_t *spa = db->db_objset->os_spa; dr->dt.dl.dr_data = arc_buf_alloc(spa, size, db, type); bcopy(db->db.db_data, dr->dt.dl.dr_data->b_data, size); } else { dbuf_clear_data(db); } } 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. * * This is a no-op if the dataset is in the middle of an incremental * receive; see comment below for details. */ 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; boolean_t freespill = (start_blkid == DMU_SPILL_BLKID || end_blkid == DMU_SPILL_BLKID); if (end_blkid > dn->dn_maxblkid && !freespill) 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); if (start_blkid >= dn->dn_unlisted_l0_blkid && !freespill) { /* There can't be any dbufs in this range; no need to search. */ #ifdef DEBUG db = avl_find(&dn->dn_dbufs, &db_search, &where); ASSERT3P(db, ==, NULL); db = avl_nearest(&dn->dn_dbufs, where, AVL_AFTER); ASSERT(db == NULL || db->db_level > 0); #endif mutex_exit(&dn->dn_dbufs_mtx); return; } else if (dmu_objset_is_receiving(dn->dn_objset)) { /* * If we are receiving, we expect there to be no dbufs in * the range to be freed, because receive modifies each * block at most once, and in offset order. If this is * not the case, it can lead to performance problems, * so note that we unexpectedly took the slow path. */ atomic_inc_64(&zfs_free_range_recv_miss); } 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_clear(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); } static int dbuf_block_freeable(dmu_buf_impl_t *db) { dsl_dataset_t *ds = db->db_objset->os_dsl_dataset; uint64_t birth_txg = 0; /* * We don't need any locking to protect db_blkptr: * If it's syncing, then db_last_dirty will be set * so we'll ignore db_blkptr. * * This logic ensures that only block births for * filled blocks are considered. */ ASSERT(MUTEX_HELD(&db->db_mtx)); if (db->db_last_dirty && (db->db_blkptr == NULL || !BP_IS_HOLE(db->db_blkptr))) { birth_txg = db->db_last_dirty->dr_txg; } else if (db->db_blkptr != NULL && !BP_IS_HOLE(db->db_blkptr)) { birth_txg = db->db_blkptr->blk_birth; } /* * If this block don't exist or is in a snapshot, it can't be freed. * Don't pass the bp to dsl_dataset_block_freeable() since we * are holding the db_mtx lock and might deadlock if we are * prefetching a dedup-ed block. */ if (birth_txg != 0) return (ds == NULL || dsl_dataset_block_freeable(ds, NULL, birth_txg)); else return (B_FALSE); } 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_buf_alloc(dn->dn_objset->os_spa, size, db, type); /* 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); VERIFY(arc_buf_remove_ref(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); dnode_willuse_space(dn, 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; boolean_t do_free_accounting = B_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. */ 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); /* * 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 && !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); } 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; ASSERT(!dmu_tx_is_syncing(tx) || DMU_OBJECT_IS_SPECIAL(dn->dn_object) || os->os_dsl_dataset == NULL || BP_IS_HOLE(os->os_rootbp)); 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) { /* * Update the accounting. * Note: we delay "free accounting" until after we drop * the db_mtx. This keeps us from grabbing other locks * (and possibly deadlocking) in bp_get_dsize() while * also holding the db_mtx. */ dnode_willuse_space(dn, db->db.db_size, tx); do_free_accounting = dbuf_block_freeable(db); } /* * 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); } else if (do_free_accounting) { blkptr_t *bp = db->db_blkptr; int64_t willfree = (bp && !BP_IS_HOLE(bp)) ? bp_get_dsize(os->os_spa, bp) : db->db.db_size; /* * This is only a guess -- if the dbuf is dirty * in a previous txg, we don't know how much * space it will use on disk yet. We should * really have the struct_rwlock to access * db_blkptr, but since this is just a guess, * it's OK if we get an odd answer. */ ddt_prefetch(os->os_spa, bp); dnode_willuse_space(dn, -willfree, tx); } if (!RW_WRITE_HELD(&dn->dn_struct_rwlock)) { rw_enter(&dn->dn_struct_rwlock, RW_READER); drop_struct_lock = TRUE; } 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) VERIFY(arc_buf_remove_ref(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) { arc_buf_t *buf = db->db_buf; ASSERT(db->db_state == DB_NOFILL || arc_released(buf)); dbuf_clear_data(db); VERIFY(arc_buf_remove_ref(buf, db)); dbuf_evict(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); ASSERT(DBUF_GET_BUFC_TYPE(db) == ARC_BUFC_DATA); ASSERT(buf != NULL); ASSERT(arc_buf_size(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); VERIFY(arc_buf_remove_ref(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; VERIFY(arc_buf_remove_ref(db->db_buf, db)); } else if (dr == NULL || dr->dt.dl.dr_data != db->db_buf) { arc_release(db->db_buf, db); VERIFY(arc_buf_remove_ref(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); } /* * "Clear" the contents of this dbuf. This will mark the dbuf * EVICTING and clear *most* of its references. Unfortunately, * when we are not holding the dn_dbufs_mtx, we can't clear the * entry in the dn_dbufs list. We have to wait until dbuf_destroy() * in this case. For callers from the DMU we will usually see: * dbuf_clear()->arc_clear_callback()->dbuf_do_evict()->dbuf_destroy() * For the arc callback, we will usually see: * dbuf_do_evict()->dbuf_clear();dbuf_destroy() * Sometimes, though, we will get a mix of these two: * DMU: dbuf_clear()->arc_clear_callback() * ARC: dbuf_do_evict()->dbuf_destroy() * * This routine will dissociate the dbuf from the arc, by calling * arc_clear_callback(), but will not evict the data from the ARC. */ void dbuf_clear(dmu_buf_impl_t *db) { dnode_t *dn; dmu_buf_impl_t *parent = db->db_parent; dmu_buf_impl_t *dndb; boolean_t dbuf_gone = B_FALSE; ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(refcount_is_zero(&db->db_holds)); dbuf_evict_user(db); if (db->db_state == DB_CACHED) { ASSERT(db->db.db_data != NULL); if (db->db_blkid == DMU_BONUS_BLKID) { zio_buf_free(db->db.db_data, DN_MAX_BONUSLEN); arc_space_return(DN_MAX_BONUSLEN, ARC_SPACE_OTHER); } db->db.db_data = NULL; db->db_state = DB_UNCACHED; } 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; DB_DNODE_ENTER(db); dn = DB_DNODE(db); dndb = dn->dn_dbuf; if (db->db_blkid != DMU_BONUS_BLKID && MUTEX_HELD(&dn->dn_dbufs_mtx)) { avl_remove(&dn->dn_dbufs, db); atomic_dec_32(&dn->dn_dbufs_count); membar_producer(); DB_DNODE_EXIT(db); /* * 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; } else { DB_DNODE_EXIT(db); } if (db->db_buf) dbuf_gone = arc_clear_callback(db->db_buf); if (!dbuf_gone) mutex_exit(&db->db_mtx); /* * 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) { int nlevels, epbs; *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); } if (dn->dn_phys->dn_nlevels == 0) nlevels = 1; else nlevels = dn->dn_phys->dn_nlevels; epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; ASSERT3U(level * epbs, <, 64); ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock)); if (level >= nlevels || (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)); 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_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_cache, db); mutex_exit(&dn->dn_dbufs_mtx); return (odb); } avl_add(&dn->dn_dbufs, db); if (db->db_level == 0 && db->db_blkid >= dn->dn_unlisted_l0_blkid) dn->dn_unlisted_l0_blkid = db->db_blkid + 1; 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); } static int dbuf_do_evict(void *private) { dmu_buf_impl_t *db = private; if (!MUTEX_HELD(&db->db_mtx)) mutex_enter(&db->db_mtx); ASSERT(refcount_is_zero(&db->db_holds)); if (db->db_state != DB_EVICTING) { ASSERT(db->db_state == DB_CACHED); DBUF_VERIFY(db); db->db_buf = NULL; dbuf_evict(db); } else { mutex_exit(&db->db_mtx); dbuf_destroy(db); } return (0); } static void dbuf_destroy(dmu_buf_impl_t *db) { ASSERT(refcount_is_zero(&db->db_holds)); if (db->db_blkid != DMU_BONUS_BLKID) { /* * If this dbuf is still on the dn_dbufs list, * remove it from that list. */ if (db->db_dnode_handle != NULL) { dnode_t *dn; DB_DNODE_ENTER(db); dn = DB_DNODE(db); mutex_enter(&dn->dn_dbufs_mtx); avl_remove(&dn->dn_dbufs, db); atomic_dec_32(&dn->dn_dbufs_count); mutex_exit(&dn->dn_dbufs_mtx); DB_DNODE_EXIT(db); /* * 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. */ dnode_rele(dn, db); db->db_dnode_handle = NULL; } dbuf_hash_remove(db); } db->db_parent = NULL; 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); kmem_cache_free(dbuf_cache, db); arc_space_return(sizeof (dmu_buf_impl_t), ARC_SPACE_OTHER); } 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 */ 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); if (zio != NULL) { ASSERT3S(BP_GET_LEVEL(zio->io_bp), ==, dpa->dpa_curlevel); ASSERT3U(BP_GET_LSIZE(zio->io_bp), ==, zio->io_size); ASSERT3P(zio->io_spa, ==, dpa->dpa_spa); } 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); } (void) arc_buf_remove_ref(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_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 && refcount_is_zero(&db->db_holds)) { arc_buf_add_ref(db->db_buf, db); if (db->db_buf->b_data == NULL) { dbuf_clear(db); if (parent) { dbuf_rele(parent, NULL); parent = NULL; } goto top; } 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_buf_alloc(dn->dn_objset->os_spa, db->db.db_size, db, type)); bcopy(dr->dt.dl.dr_data->b_data, db->db.db_data, db->db.db_size); } } (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); ASSERT(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 && 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_evict(db); } else if (arc_released(db->db_buf)) { arc_buf_t *buf = db->db_buf; /* * This dbuf has anonymous data associated with it. */ dbuf_clear_data(db); VERIFY(arc_buf_remove_ref(buf, db)); dbuf_evict(db); } else { VERIFY(!arc_buf_remove_ref(db->db_buf, db)); /* * A dbuf will be eligible for eviction if either the * 'primarycache' property is set or a duplicate * copy of this buffer is already cached in the arc. * * In the case of the 'primarycache' a buffer * is considered for eviction if it matches the * criteria set in the property. * * To decide if our buffer is considered a * duplicate, we must call into the arc to determine * if multiple buffers are referencing the same * block on-disk. If so, then we simply evict * ourselves. */ if (!DBUF_IS_CACHEABLE(db)) { if (db->db_blkptr != NULL && !BP_IS_HOLE(db->db_blkptr) && !BP_IS_EMBEDDED(db->db_blkptr)) { spa_t *spa = dmu_objset_spa(db->db_objset); blkptr_t bp = *db->db_blkptr; dbuf_clear(db); arc_freed(spa, &bp); } else { dbuf_clear(db); } } else if (db->db_pending_evict || arc_buf_eviction_needed(db->db_buf)) { dbuf_clear(db); } else { mutex_exit(&db->db_mtx); } } } 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); } boolean_t dmu_buf_freeable(dmu_buf_t *dbuf) { boolean_t res = B_FALSE; dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbuf; if (db->db_blkptr) res = dsl_dataset_block_freeable(db->db_objset->os_dsl_dataset, db->db_blkptr, db->db_blkptr->blk_birth); return (res); } blkptr_t * dmu_buf_get_blkptr(dmu_buf_t *db) { dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; return (dbi->db_blkptr); } 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; if (dr->dr_dbuf->db_level != 0) { list_destroy(&dr->dt.di.dr_children); mutex_destroy(&dr->dt.di.dr_mtx); } 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 blksz = arc_buf_size(*datap); arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db); *datap = arc_buf_alloc(os->os_spa, blksz, db, type); bcopy(db->db.db_data, (*datap)->b_data, blksz); } 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; + uint64_t i; + int epbs; + + ASSERT3U(db->db_level, >, 0); + DB_DNODE_ENTER(db); + dn = DB_DNODE(db); + epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; + + /* 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) { + /* didn't find any non-holes */ + bzero(db->db.db_data, db->db.db_size); + } + 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) VERIFY(arc_buf_remove_ref(dr->dt.dl.dr_data, db)); else if (!arc_released(db->db_buf)) arc_set_callback(db->db_buf, dbuf_do_evict, 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); if (!arc_released(db->db_buf)) arc_set_callback(db->db_buf, dbuf_do_evict, db); } 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, &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, &zp, - dbuf_write_override_ready, NULL, dbuf_write_override_done, + 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, &zp, - dbuf_write_nofill_ready, NULL, dbuf_write_nofill_done, db, + 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), DBUF_IS_L2COMPRESSIBLE(db), &zp, dbuf_write_ready, + children_ready_cb, dbuf_write_physdone, dbuf_write_done, db, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); } } Index: stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu.c =================================================================== --- stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu.c (revision 304137) +++ stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu.c (revision 304138) @@ -1,2159 +1,2160 @@ /* * 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, 2016 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, 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 #ifdef _KERNEL #include #include #include #endif /* * Enable/disable nopwrite feature. */ int zfs_nopwrite_enabled = 1; SYSCTL_DECL(_vfs_zfs); SYSCTL_INT(_vfs_zfs, OID_AUTO, nopwrite_enabled, CTLFLAG_RDTUN, &zfs_nopwrite_enabled, 0, "Enable nopwrite feature"); const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = { { DMU_BSWAP_UINT8, TRUE, "unallocated" }, { DMU_BSWAP_ZAP, TRUE, "object directory" }, { DMU_BSWAP_UINT64, TRUE, "object array" }, { DMU_BSWAP_UINT8, TRUE, "packed nvlist" }, { DMU_BSWAP_UINT64, TRUE, "packed nvlist size" }, { DMU_BSWAP_UINT64, TRUE, "bpobj" }, { DMU_BSWAP_UINT64, TRUE, "bpobj header" }, { DMU_BSWAP_UINT64, TRUE, "SPA space map header" }, { DMU_BSWAP_UINT64, TRUE, "SPA space map" }, { DMU_BSWAP_UINT64, TRUE, "ZIL intent log" }, { DMU_BSWAP_DNODE, TRUE, "DMU dnode" }, { DMU_BSWAP_OBJSET, TRUE, "DMU objset" }, { DMU_BSWAP_UINT64, TRUE, "DSL directory" }, { DMU_BSWAP_ZAP, TRUE, "DSL directory child map"}, { DMU_BSWAP_ZAP, TRUE, "DSL dataset snap map" }, { DMU_BSWAP_ZAP, TRUE, "DSL props" }, { DMU_BSWAP_UINT64, TRUE, "DSL dataset" }, { DMU_BSWAP_ZNODE, TRUE, "ZFS znode" }, { DMU_BSWAP_OLDACL, TRUE, "ZFS V0 ACL" }, { DMU_BSWAP_UINT8, FALSE, "ZFS plain file" }, { DMU_BSWAP_ZAP, TRUE, "ZFS directory" }, { DMU_BSWAP_ZAP, TRUE, "ZFS master node" }, { DMU_BSWAP_ZAP, TRUE, "ZFS delete queue" }, { DMU_BSWAP_UINT8, FALSE, "zvol object" }, { DMU_BSWAP_ZAP, TRUE, "zvol prop" }, { DMU_BSWAP_UINT8, FALSE, "other uint8[]" }, { DMU_BSWAP_UINT64, FALSE, "other uint64[]" }, { DMU_BSWAP_ZAP, TRUE, "other ZAP" }, { DMU_BSWAP_ZAP, TRUE, "persistent error log" }, { DMU_BSWAP_UINT8, TRUE, "SPA history" }, { DMU_BSWAP_UINT64, TRUE, "SPA history offsets" }, { DMU_BSWAP_ZAP, TRUE, "Pool properties" }, { DMU_BSWAP_ZAP, TRUE, "DSL permissions" }, { DMU_BSWAP_ACL, TRUE, "ZFS ACL" }, { DMU_BSWAP_UINT8, TRUE, "ZFS SYSACL" }, { DMU_BSWAP_UINT8, TRUE, "FUID table" }, { DMU_BSWAP_UINT64, TRUE, "FUID table size" }, { DMU_BSWAP_ZAP, TRUE, "DSL dataset next clones"}, { DMU_BSWAP_ZAP, TRUE, "scan work queue" }, { DMU_BSWAP_ZAP, TRUE, "ZFS user/group used" }, { DMU_BSWAP_ZAP, TRUE, "ZFS user/group quota" }, { DMU_BSWAP_ZAP, TRUE, "snapshot refcount tags"}, { DMU_BSWAP_ZAP, TRUE, "DDT ZAP algorithm" }, { DMU_BSWAP_ZAP, TRUE, "DDT statistics" }, { DMU_BSWAP_UINT8, TRUE, "System attributes" }, { DMU_BSWAP_ZAP, TRUE, "SA master node" }, { DMU_BSWAP_ZAP, TRUE, "SA attr registration" }, { DMU_BSWAP_ZAP, TRUE, "SA attr layouts" }, { DMU_BSWAP_ZAP, TRUE, "scan translations" }, { DMU_BSWAP_UINT8, FALSE, "deduplicated block" }, { DMU_BSWAP_ZAP, TRUE, "DSL deadlist map" }, { DMU_BSWAP_UINT64, TRUE, "DSL deadlist map hdr" }, { DMU_BSWAP_ZAP, TRUE, "DSL dir clones" }, { DMU_BSWAP_UINT64, TRUE, "bpobj subobj" } }; const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = { { byteswap_uint8_array, "uint8" }, { byteswap_uint16_array, "uint16" }, { byteswap_uint32_array, "uint32" }, { byteswap_uint64_array, "uint64" }, { zap_byteswap, "zap" }, { dnode_buf_byteswap, "dnode" }, { dmu_objset_byteswap, "objset" }, { zfs_znode_byteswap, "znode" }, { zfs_oldacl_byteswap, "oldacl" }, { zfs_acl_byteswap, "acl" } }; int dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset, void *tag, dmu_buf_t **dbp) { dnode_t *dn; uint64_t blkid; dmu_buf_impl_t *db; int err; err = dnode_hold(os, object, FTAG, &dn); if (err) return (err); blkid = dbuf_whichblock(dn, 0, offset); rw_enter(&dn->dn_struct_rwlock, RW_READER); db = dbuf_hold(dn, blkid, tag); rw_exit(&dn->dn_struct_rwlock); dnode_rele(dn, FTAG); if (db == NULL) { *dbp = NULL; return (SET_ERROR(EIO)); } *dbp = &db->db; return (err); } int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, void *tag, dmu_buf_t **dbp, int flags) { int err; int db_flags = DB_RF_CANFAIL; if (flags & DMU_READ_NO_PREFETCH) db_flags |= DB_RF_NOPREFETCH; err = dmu_buf_hold_noread(os, object, offset, tag, dbp); if (err == 0) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); err = dbuf_read(db, NULL, db_flags); if (err != 0) { dbuf_rele(db, tag); *dbp = NULL; } } return (err); } int dmu_bonus_max(void) { return (DN_MAX_BONUSLEN); } int dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; dnode_t *dn; int error; DB_DNODE_ENTER(db); dn = DB_DNODE(db); if (dn->dn_bonus != db) { error = SET_ERROR(EINVAL); } else if (newsize < 0 || newsize > db_fake->db_size) { error = SET_ERROR(EINVAL); } else { dnode_setbonuslen(dn, newsize, tx); error = 0; } DB_DNODE_EXIT(db); return (error); } int dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; dnode_t *dn; int error; DB_DNODE_ENTER(db); dn = DB_DNODE(db); if (!DMU_OT_IS_VALID(type)) { error = SET_ERROR(EINVAL); } else if (dn->dn_bonus != db) { error = SET_ERROR(EINVAL); } else { dnode_setbonus_type(dn, type, tx); error = 0; } DB_DNODE_EXIT(db); return (error); } dmu_object_type_t dmu_get_bonustype(dmu_buf_t *db_fake) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; dnode_t *dn; dmu_object_type_t type; DB_DNODE_ENTER(db); dn = DB_DNODE(db); type = dn->dn_bonustype; DB_DNODE_EXIT(db); return (type); } int dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx) { dnode_t *dn; int error; error = dnode_hold(os, object, FTAG, &dn); dbuf_rm_spill(dn, tx); rw_enter(&dn->dn_struct_rwlock, RW_WRITER); dnode_rm_spill(dn, tx); rw_exit(&dn->dn_struct_rwlock); dnode_rele(dn, FTAG); return (error); } /* * returns ENOENT, EIO, or 0. */ int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp) { dnode_t *dn; dmu_buf_impl_t *db; int error; error = dnode_hold(os, object, FTAG, &dn); if (error) return (error); rw_enter(&dn->dn_struct_rwlock, RW_READER); if (dn->dn_bonus == NULL) { rw_exit(&dn->dn_struct_rwlock); rw_enter(&dn->dn_struct_rwlock, RW_WRITER); if (dn->dn_bonus == NULL) dbuf_create_bonus(dn); } db = dn->dn_bonus; /* as long as the bonus buf is held, the dnode will be held */ if (refcount_add(&db->db_holds, tag) == 1) { VERIFY(dnode_add_ref(dn, db)); atomic_inc_32(&dn->dn_dbufs_count); } /* * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's * hold and incrementing the dbuf count to ensure that dnode_move() sees * a dnode hold for every dbuf. */ rw_exit(&dn->dn_struct_rwlock); dnode_rele(dn, FTAG); VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH)); *dbp = &db->db; return (0); } /* * returns ENOENT, EIO, or 0. * * This interface will allocate a blank spill dbuf when a spill blk * doesn't already exist on the dnode. * * if you only want to find an already existing spill db, then * dmu_spill_hold_existing() should be used. */ int dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp) { dmu_buf_impl_t *db = NULL; int err; if ((flags & DB_RF_HAVESTRUCT) == 0) rw_enter(&dn->dn_struct_rwlock, RW_READER); db = dbuf_hold(dn, DMU_SPILL_BLKID, tag); if ((flags & DB_RF_HAVESTRUCT) == 0) rw_exit(&dn->dn_struct_rwlock); ASSERT(db != NULL); err = dbuf_read(db, NULL, flags); if (err == 0) *dbp = &db->db; else dbuf_rele(db, tag); return (err); } int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; dnode_t *dn; int err; DB_DNODE_ENTER(db); dn = DB_DNODE(db); if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) { err = SET_ERROR(EINVAL); } else { rw_enter(&dn->dn_struct_rwlock, RW_READER); if (!dn->dn_have_spill) { err = SET_ERROR(ENOENT); } else { err = dmu_spill_hold_by_dnode(dn, DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp); } rw_exit(&dn->dn_struct_rwlock); } DB_DNODE_EXIT(db); return (err); } int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; dnode_t *dn; int err; DB_DNODE_ENTER(db); dn = DB_DNODE(db); err = dmu_spill_hold_by_dnode(dn, DB_RF_CANFAIL, tag, dbp); DB_DNODE_EXIT(db); return (err); } /* * Note: longer-term, we should modify all of the dmu_buf_*() interfaces * to take a held dnode rather than -- the lookup is wasteful, * and can induce severe lock contention when writing to several files * whose dnodes are in the same block. */ static int dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length, boolean_t read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags) { dmu_buf_t **dbp; uint64_t blkid, nblks, i; uint32_t dbuf_flags; int err; zio_t *zio; ASSERT(length <= DMU_MAX_ACCESS); /* * Note: We directly notify the prefetch code of this read, so that * we can tell it about the multi-block read. dbuf_read() only knows * about the one block it is accessing. */ dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT | DB_RF_NOPREFETCH; rw_enter(&dn->dn_struct_rwlock, RW_READER); if (dn->dn_datablkshift) { int blkshift = dn->dn_datablkshift; nblks = (P2ROUNDUP(offset + length, 1ULL << blkshift) - P2ALIGN(offset, 1ULL << blkshift)) >> blkshift; } else { if (offset + length > dn->dn_datablksz) { zfs_panic_recover("zfs: accessing past end of object " "%llx/%llx (size=%u access=%llu+%llu)", (longlong_t)dn->dn_objset-> os_dsl_dataset->ds_object, (longlong_t)dn->dn_object, dn->dn_datablksz, (longlong_t)offset, (longlong_t)length); rw_exit(&dn->dn_struct_rwlock); return (SET_ERROR(EIO)); } nblks = 1; } dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP); #if defined(_KERNEL) && defined(RACCT) if (racct_enable && !read) { PROC_LOCK(curproc); racct_add_force(curproc, RACCT_WRITEBPS, length); racct_add_force(curproc, RACCT_WRITEIOPS, nblks); PROC_UNLOCK(curproc); } #endif zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL); blkid = dbuf_whichblock(dn, 0, offset); for (i = 0; i < nblks; i++) { dmu_buf_impl_t *db = dbuf_hold(dn, blkid + i, tag); if (db == NULL) { rw_exit(&dn->dn_struct_rwlock); dmu_buf_rele_array(dbp, nblks, tag); zio_nowait(zio); return (SET_ERROR(EIO)); } /* initiate async i/o */ if (read) (void) dbuf_read(db, zio, dbuf_flags); #ifdef _KERNEL else curthread->td_ru.ru_oublock++; #endif dbp[i] = &db->db; } if ((flags & DMU_READ_NO_PREFETCH) == 0 && DNODE_META_IS_CACHEABLE(dn) && length <= zfetch_array_rd_sz) { dmu_zfetch(&dn->dn_zfetch, blkid, nblks, read && DNODE_IS_CACHEABLE(dn)); } rw_exit(&dn->dn_struct_rwlock); /* wait for async i/o */ err = zio_wait(zio); if (err) { dmu_buf_rele_array(dbp, nblks, tag); return (err); } /* wait for other io to complete */ if (read) { for (i = 0; i < nblks; i++) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i]; 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) err = SET_ERROR(EIO); mutex_exit(&db->db_mtx); if (err) { dmu_buf_rele_array(dbp, nblks, tag); return (err); } } } *numbufsp = nblks; *dbpp = dbp; return (0); } static int dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset, uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) { dnode_t *dn; int err; err = dnode_hold(os, object, FTAG, &dn); if (err) return (err); err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, numbufsp, dbpp, DMU_READ_PREFETCH); dnode_rele(dn, FTAG); return (err); } int dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset, uint64_t length, boolean_t read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; dnode_t *dn; int err; DB_DNODE_ENTER(db); dn = DB_DNODE(db); err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, numbufsp, dbpp, DMU_READ_PREFETCH); DB_DNODE_EXIT(db); return (err); } void dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag) { int i; dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake; if (numbufs == 0) return; for (i = 0; i < numbufs; i++) { if (dbp[i]) dbuf_rele(dbp[i], tag); } kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs); } /* * Issue prefetch i/os for the given blocks. If level is greater than 0, the * indirect blocks prefeteched will be those that point to the blocks containing * the data starting at offset, and continuing to offset + len. * * Note that if the indirect blocks above the blocks being prefetched are not in * cache, they will be asychronously read in. */ void dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset, uint64_t len, zio_priority_t pri) { dnode_t *dn; uint64_t blkid; int nblks, err; if (len == 0) { /* they're interested in the bonus buffer */ dn = DMU_META_DNODE(os); if (object == 0 || object >= DN_MAX_OBJECT) return; rw_enter(&dn->dn_struct_rwlock, RW_READER); blkid = dbuf_whichblock(dn, level, object * sizeof (dnode_phys_t)); dbuf_prefetch(dn, level, blkid, pri, 0); rw_exit(&dn->dn_struct_rwlock); return; } /* * XXX - Note, if the dnode for the requested object is not * already cached, we will do a *synchronous* read in the * dnode_hold() call. The same is true for any indirects. */ err = dnode_hold(os, object, FTAG, &dn); if (err != 0) return; rw_enter(&dn->dn_struct_rwlock, RW_READER); /* * offset + len - 1 is the last byte we want to prefetch for, and offset * is the first. Then dbuf_whichblk(dn, level, off + len - 1) is the * last block we want to prefetch, and dbuf_whichblock(dn, level, * offset) is the first. Then the number we need to prefetch is the * last - first + 1. */ if (level > 0 || dn->dn_datablkshift != 0) { nblks = dbuf_whichblock(dn, level, offset + len - 1) - dbuf_whichblock(dn, level, offset) + 1; } else { nblks = (offset < dn->dn_datablksz); } if (nblks != 0) { blkid = dbuf_whichblock(dn, level, offset); for (int i = 0; i < nblks; i++) dbuf_prefetch(dn, level, blkid + i, pri, 0); } rw_exit(&dn->dn_struct_rwlock); dnode_rele(dn, FTAG); } /* * Get the next "chunk" of file data to free. We traverse the file from * the end so that the file gets shorter over time (if we crashes in the * middle, this will leave us in a better state). We find allocated file * data by simply searching the allocated level 1 indirects. * * On input, *start should be the first offset that does not need to be * freed (e.g. "offset + length"). On return, *start will be the first * offset that should be freed. */ static int get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum) { uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1); /* bytes of data covered by a level-1 indirect block */ uint64_t iblkrange = dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT); ASSERT3U(minimum, <=, *start); if (*start - minimum <= iblkrange * maxblks) { *start = minimum; return (0); } ASSERT(ISP2(iblkrange)); for (uint64_t blks = 0; *start > minimum && blks < maxblks; blks++) { int err; /* * dnode_next_offset(BACKWARDS) will find an allocated L1 * indirect block at or before the input offset. We must * decrement *start so that it is at the end of the region * to search. */ (*start)--; err = dnode_next_offset(dn, DNODE_FIND_BACKWARDS, start, 2, 1, 0); /* if there are no indirect blocks before start, we are done */ if (err == ESRCH) { *start = minimum; break; } else if (err != 0) { return (err); } /* set start to the beginning of this L1 indirect */ *start = P2ALIGN(*start, iblkrange); } if (*start < minimum) *start = minimum; return (0); } static int dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset, uint64_t length) { uint64_t object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz; int err; if (offset >= object_size) return (0); if (length == DMU_OBJECT_END || offset + length > object_size) length = object_size - offset; while (length != 0) { uint64_t chunk_end, chunk_begin; chunk_end = chunk_begin = offset + length; /* move chunk_begin backwards to the beginning of this chunk */ err = get_next_chunk(dn, &chunk_begin, offset); if (err) return (err); ASSERT3U(chunk_begin, >=, offset); ASSERT3U(chunk_begin, <=, chunk_end); dmu_tx_t *tx = dmu_tx_create(os); dmu_tx_hold_free(tx, dn->dn_object, chunk_begin, chunk_end - chunk_begin); /* * Mark this transaction as typically resulting in a net * reduction in space used. */ dmu_tx_mark_netfree(tx); err = dmu_tx_assign(tx, TXG_WAIT); if (err) { dmu_tx_abort(tx); return (err); } dnode_free_range(dn, chunk_begin, chunk_end - chunk_begin, tx); dmu_tx_commit(tx); length -= chunk_end - chunk_begin; } return (0); } int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset, uint64_t length) { dnode_t *dn; int err; err = dnode_hold(os, object, FTAG, &dn); if (err != 0) return (err); err = dmu_free_long_range_impl(os, dn, offset, length); /* * It is important to zero out the maxblkid when freeing the entire * file, so that (a) subsequent calls to dmu_free_long_range_impl() * will take the fast path, and (b) dnode_reallocate() can verify * that the entire file has been freed. */ if (err == 0 && offset == 0 && length == DMU_OBJECT_END) dn->dn_maxblkid = 0; dnode_rele(dn, FTAG); return (err); } int dmu_free_long_object(objset_t *os, uint64_t object) { dmu_tx_t *tx; int err; err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END); if (err != 0) return (err); tx = dmu_tx_create(os); dmu_tx_hold_bonus(tx, object); dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); dmu_tx_mark_netfree(tx); err = dmu_tx_assign(tx, TXG_WAIT); if (err == 0) { err = dmu_object_free(os, object, tx); dmu_tx_commit(tx); } else { dmu_tx_abort(tx); } return (err); } int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, dmu_tx_t *tx) { dnode_t *dn; int err = dnode_hold(os, object, FTAG, &dn); if (err) return (err); ASSERT(offset < UINT64_MAX); ASSERT(size == -1ULL || size <= UINT64_MAX - offset); dnode_free_range(dn, offset, size, tx); dnode_rele(dn, FTAG); return (0); } int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, void *buf, uint32_t flags) { dnode_t *dn; dmu_buf_t **dbp; int numbufs, err; err = dnode_hold(os, object, FTAG, &dn); if (err) return (err); /* * Deal with odd block sizes, where there can't be data past the first * block. If we ever do the tail block optimization, we will need to * handle that here as well. */ if (dn->dn_maxblkid == 0) { int newsz = offset > dn->dn_datablksz ? 0 : MIN(size, dn->dn_datablksz - offset); bzero((char *)buf + newsz, size - newsz); size = newsz; } while (size > 0) { uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2); int i; /* * NB: we could do this block-at-a-time, but it's nice * to be reading in parallel. */ err = dmu_buf_hold_array_by_dnode(dn, offset, mylen, TRUE, FTAG, &numbufs, &dbp, flags); if (err) break; for (i = 0; i < numbufs; i++) { int tocpy; int bufoff; dmu_buf_t *db = dbp[i]; ASSERT(size > 0); bufoff = offset - db->db_offset; tocpy = (int)MIN(db->db_size - bufoff, size); bcopy((char *)db->db_data + bufoff, buf, tocpy); offset += tocpy; size -= tocpy; buf = (char *)buf + tocpy; } dmu_buf_rele_array(dbp, numbufs, FTAG); } dnode_rele(dn, FTAG); return (err); } void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, const void *buf, dmu_tx_t *tx) { dmu_buf_t **dbp; int numbufs, i; if (size == 0) return; VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, FALSE, FTAG, &numbufs, &dbp)); for (i = 0; i < numbufs; i++) { int tocpy; int bufoff; dmu_buf_t *db = dbp[i]; ASSERT(size > 0); bufoff = offset - db->db_offset; tocpy = (int)MIN(db->db_size - bufoff, size); ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); if (tocpy == db->db_size) dmu_buf_will_fill(db, tx); else dmu_buf_will_dirty(db, tx); bcopy(buf, (char *)db->db_data + bufoff, tocpy); if (tocpy == db->db_size) dmu_buf_fill_done(db, tx); offset += tocpy; size -= tocpy; buf = (char *)buf + tocpy; } dmu_buf_rele_array(dbp, numbufs, FTAG); } void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, dmu_tx_t *tx) { dmu_buf_t **dbp; int numbufs, i; if (size == 0) return; VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, FALSE, FTAG, &numbufs, &dbp)); for (i = 0; i < numbufs; i++) { dmu_buf_t *db = dbp[i]; dmu_buf_will_not_fill(db, tx); } dmu_buf_rele_array(dbp, numbufs, FTAG); } void dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset, void *data, uint8_t etype, uint8_t comp, int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx) { dmu_buf_t *db; ASSERT3U(etype, <, NUM_BP_EMBEDDED_TYPES); ASSERT3U(comp, <, ZIO_COMPRESS_FUNCTIONS); VERIFY0(dmu_buf_hold_noread(os, object, offset, FTAG, &db)); dmu_buf_write_embedded(db, data, (bp_embedded_type_t)etype, (enum zio_compress)comp, uncompressed_size, compressed_size, byteorder, tx); dmu_buf_rele(db, FTAG); } /* * DMU support for xuio */ kstat_t *xuio_ksp = NULL; int dmu_xuio_init(xuio_t *xuio, int nblk) { dmu_xuio_t *priv; uio_t *uio = &xuio->xu_uio; uio->uio_iovcnt = nblk; uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP); priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP); priv->cnt = nblk; priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP); priv->iovp = uio->uio_iov; XUIO_XUZC_PRIV(xuio) = priv; if (XUIO_XUZC_RW(xuio) == UIO_READ) XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk); else XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk); return (0); } void dmu_xuio_fini(xuio_t *xuio) { dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); int nblk = priv->cnt; kmem_free(priv->iovp, nblk * sizeof (iovec_t)); kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *)); kmem_free(priv, sizeof (dmu_xuio_t)); if (XUIO_XUZC_RW(xuio) == UIO_READ) XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk); else XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk); } /* * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf } * and increase priv->next by 1. */ int dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n) { struct iovec *iov; uio_t *uio = &xuio->xu_uio; dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); int i = priv->next++; ASSERT(i < priv->cnt); ASSERT(off + n <= arc_buf_size(abuf)); iov = uio->uio_iov + i; iov->iov_base = (char *)abuf->b_data + off; iov->iov_len = n; priv->bufs[i] = abuf; return (0); } int dmu_xuio_cnt(xuio_t *xuio) { dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); return (priv->cnt); } arc_buf_t * dmu_xuio_arcbuf(xuio_t *xuio, int i) { dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); ASSERT(i < priv->cnt); return (priv->bufs[i]); } void dmu_xuio_clear(xuio_t *xuio, int i) { dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); ASSERT(i < priv->cnt); priv->bufs[i] = NULL; } static void xuio_stat_init(void) { xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc", KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL); if (xuio_ksp != NULL) { xuio_ksp->ks_data = &xuio_stats; kstat_install(xuio_ksp); } } static void xuio_stat_fini(void) { if (xuio_ksp != NULL) { kstat_delete(xuio_ksp); xuio_ksp = NULL; } } void xuio_stat_wbuf_copied() { XUIOSTAT_BUMP(xuiostat_wbuf_copied); } void xuio_stat_wbuf_nocopy() { XUIOSTAT_BUMP(xuiostat_wbuf_nocopy); } #ifdef _KERNEL static int dmu_read_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size) { dmu_buf_t **dbp; int numbufs, i, err; xuio_t *xuio = NULL; /* * NB: we could do this block-at-a-time, but it's nice * to be reading in parallel. */ err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size, TRUE, FTAG, &numbufs, &dbp, 0); if (err) return (err); #ifdef UIO_XUIO if (uio->uio_extflg == UIO_XUIO) xuio = (xuio_t *)uio; #endif for (i = 0; i < numbufs; i++) { int tocpy; int bufoff; dmu_buf_t *db = dbp[i]; ASSERT(size > 0); bufoff = uio->uio_loffset - db->db_offset; tocpy = (int)MIN(db->db_size - bufoff, size); if (xuio) { dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; arc_buf_t *dbuf_abuf = dbi->db_buf; arc_buf_t *abuf = dbuf_loan_arcbuf(dbi); err = dmu_xuio_add(xuio, abuf, bufoff, tocpy); if (!err) { uio->uio_resid -= tocpy; uio->uio_loffset += tocpy; } if (abuf == dbuf_abuf) XUIOSTAT_BUMP(xuiostat_rbuf_nocopy); else XUIOSTAT_BUMP(xuiostat_rbuf_copied); } else { #ifdef illumos err = uiomove((char *)db->db_data + bufoff, tocpy, UIO_READ, uio); #else err = vn_io_fault_uiomove((char *)db->db_data + bufoff, tocpy, uio); #endif } if (err) break; size -= tocpy; } dmu_buf_rele_array(dbp, numbufs, FTAG); return (err); } /* * Read 'size' bytes into the uio buffer. * From object zdb->db_object. * Starting at offset uio->uio_loffset. * * If the caller already has a dbuf in the target object * (e.g. its bonus buffer), this routine is faster than dmu_read_uio(), * because we don't have to find the dnode_t for the object. */ int dmu_read_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; dnode_t *dn; int err; if (size == 0) return (0); DB_DNODE_ENTER(db); dn = DB_DNODE(db); err = dmu_read_uio_dnode(dn, uio, size); DB_DNODE_EXIT(db); return (err); } /* * Read 'size' bytes into the uio buffer. * From the specified object * Starting at offset uio->uio_loffset. */ int dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size) { dnode_t *dn; int err; if (size == 0) return (0); err = dnode_hold(os, object, FTAG, &dn); if (err) return (err); err = dmu_read_uio_dnode(dn, uio, size); dnode_rele(dn, FTAG); return (err); } static int dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx) { dmu_buf_t **dbp; int numbufs; int err = 0; int i; err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size, FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH); if (err) return (err); for (i = 0; i < numbufs; i++) { int tocpy; int bufoff; dmu_buf_t *db = dbp[i]; ASSERT(size > 0); bufoff = uio->uio_loffset - db->db_offset; tocpy = (int)MIN(db->db_size - bufoff, size); ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); if (tocpy == db->db_size) dmu_buf_will_fill(db, tx); else dmu_buf_will_dirty(db, tx); #ifdef illumos /* * XXX uiomove could block forever (eg. nfs-backed * pages). There needs to be a uiolockdown() function * to lock the pages in memory, so that uiomove won't * block. */ err = uiomove((char *)db->db_data + bufoff, tocpy, UIO_WRITE, uio); #else err = vn_io_fault_uiomove((char *)db->db_data + bufoff, tocpy, uio); #endif if (tocpy == db->db_size) dmu_buf_fill_done(db, tx); if (err) break; size -= tocpy; } dmu_buf_rele_array(dbp, numbufs, FTAG); return (err); } /* * Write 'size' bytes from the uio buffer. * To object zdb->db_object. * Starting at offset uio->uio_loffset. * * If the caller already has a dbuf in the target object * (e.g. its bonus buffer), this routine is faster than dmu_write_uio(), * because we don't have to find the dnode_t for the object. */ int dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; dnode_t *dn; int err; if (size == 0) return (0); DB_DNODE_ENTER(db); dn = DB_DNODE(db); err = dmu_write_uio_dnode(dn, uio, size, tx); DB_DNODE_EXIT(db); return (err); } /* * Write 'size' bytes from the uio buffer. * To the specified object. * Starting at offset uio->uio_loffset. */ int dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size, dmu_tx_t *tx) { dnode_t *dn; int err; if (size == 0) return (0); err = dnode_hold(os, object, FTAG, &dn); if (err) return (err); err = dmu_write_uio_dnode(dn, uio, size, tx); dnode_rele(dn, FTAG); return (err); } #ifdef illumos int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, page_t *pp, dmu_tx_t *tx) { dmu_buf_t **dbp; int numbufs, i; int err; if (size == 0) return (0); err = dmu_buf_hold_array(os, object, offset, size, FALSE, FTAG, &numbufs, &dbp); if (err) return (err); for (i = 0; i < numbufs; i++) { int tocpy, copied, thiscpy; int bufoff; dmu_buf_t *db = dbp[i]; caddr_t va; ASSERT(size > 0); ASSERT3U(db->db_size, >=, PAGESIZE); bufoff = offset - db->db_offset; tocpy = (int)MIN(db->db_size - bufoff, size); ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); if (tocpy == db->db_size) dmu_buf_will_fill(db, tx); else dmu_buf_will_dirty(db, tx); for (copied = 0; copied < tocpy; copied += PAGESIZE) { ASSERT3U(pp->p_offset, ==, db->db_offset + bufoff); thiscpy = MIN(PAGESIZE, tocpy - copied); va = zfs_map_page(pp, S_READ); bcopy(va, (char *)db->db_data + bufoff, thiscpy); zfs_unmap_page(pp, va); pp = pp->p_next; bufoff += PAGESIZE; } if (tocpy == db->db_size) dmu_buf_fill_done(db, tx); offset += tocpy; size -= tocpy; } dmu_buf_rele_array(dbp, numbufs, FTAG); return (err); } #else /* !illumos */ int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, vm_page_t *ma, dmu_tx_t *tx) { dmu_buf_t **dbp; struct sf_buf *sf; int numbufs, i; int err; if (size == 0) return (0); err = dmu_buf_hold_array(os, object, offset, size, FALSE, FTAG, &numbufs, &dbp); if (err) return (err); for (i = 0; i < numbufs; i++) { int tocpy, copied, thiscpy; int bufoff; dmu_buf_t *db = dbp[i]; caddr_t va; ASSERT(size > 0); ASSERT3U(db->db_size, >=, PAGESIZE); bufoff = offset - db->db_offset; tocpy = (int)MIN(db->db_size - bufoff, size); ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); if (tocpy == db->db_size) dmu_buf_will_fill(db, tx); else dmu_buf_will_dirty(db, tx); for (copied = 0; copied < tocpy; copied += PAGESIZE) { ASSERT3U(ptoa((*ma)->pindex), ==, db->db_offset + bufoff); thiscpy = MIN(PAGESIZE, tocpy - copied); va = zfs_map_page(*ma, &sf); bcopy(va, (char *)db->db_data + bufoff, thiscpy); zfs_unmap_page(sf); ma += 1; bufoff += PAGESIZE; } if (tocpy == db->db_size) dmu_buf_fill_done(db, tx); offset += tocpy; size -= tocpy; } dmu_buf_rele_array(dbp, numbufs, FTAG); return (err); } #endif /* illumos */ #endif /* _KERNEL */ /* * Allocate a loaned anonymous arc buffer. */ arc_buf_t * dmu_request_arcbuf(dmu_buf_t *handle, int size) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle; return (arc_loan_buf(db->db_objset->os_spa, size)); } /* * Free a loaned arc buffer. */ void dmu_return_arcbuf(arc_buf_t *buf) { arc_return_buf(buf, FTAG); VERIFY(arc_buf_remove_ref(buf, FTAG)); } /* * When possible directly assign passed loaned arc buffer to a dbuf. * If this is not possible copy the contents of passed arc buf via * dmu_write(). */ void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf, dmu_tx_t *tx) { dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle; dnode_t *dn; dmu_buf_impl_t *db; uint32_t blksz = (uint32_t)arc_buf_size(buf); uint64_t blkid; DB_DNODE_ENTER(dbuf); dn = DB_DNODE(dbuf); rw_enter(&dn->dn_struct_rwlock, RW_READER); blkid = dbuf_whichblock(dn, 0, offset); VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL); rw_exit(&dn->dn_struct_rwlock); DB_DNODE_EXIT(dbuf); /* * We can only assign if the offset is aligned, the arc buf is the * same size as the dbuf, and the dbuf is not metadata. It * can't be metadata because the loaned arc buf comes from the * user-data kmem arena. */ if (offset == db->db.db_offset && blksz == db->db.db_size && DBUF_GET_BUFC_TYPE(db) == ARC_BUFC_DATA) { #ifdef _KERNEL curthread->td_ru.ru_oublock++; #ifdef RACCT if (racct_enable) { PROC_LOCK(curproc); racct_add_force(curproc, RACCT_WRITEBPS, blksz); racct_add_force(curproc, RACCT_WRITEIOPS, 1); PROC_UNLOCK(curproc); } #endif /* RACCT */ #endif /* _KERNEL */ dbuf_assign_arcbuf(db, buf, tx); dbuf_rele(db, FTAG); } else { objset_t *os; uint64_t object; DB_DNODE_ENTER(dbuf); dn = DB_DNODE(dbuf); os = dn->dn_objset; object = dn->dn_object; DB_DNODE_EXIT(dbuf); dbuf_rele(db, FTAG); dmu_write(os, object, offset, blksz, buf->b_data, tx); dmu_return_arcbuf(buf); XUIOSTAT_BUMP(xuiostat_wbuf_copied); } } typedef struct { dbuf_dirty_record_t *dsa_dr; dmu_sync_cb_t *dsa_done; zgd_t *dsa_zgd; dmu_tx_t *dsa_tx; } dmu_sync_arg_t; /* ARGSUSED */ static void dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg) { dmu_sync_arg_t *dsa = varg; dmu_buf_t *db = dsa->dsa_zgd->zgd_db; blkptr_t *bp = zio->io_bp; if (zio->io_error == 0) { if (BP_IS_HOLE(bp)) { /* * A block of zeros may compress to a hole, but the * block size still needs to be known for replay. */ BP_SET_LSIZE(bp, db->db_size); } else if (!BP_IS_EMBEDDED(bp)) { ASSERT(BP_GET_LEVEL(bp) == 0); bp->blk_fill = 1; } } } static void dmu_sync_late_arrival_ready(zio_t *zio) { dmu_sync_ready(zio, NULL, zio->io_private); } /* ARGSUSED */ static void dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg) { dmu_sync_arg_t *dsa = varg; dbuf_dirty_record_t *dr = dsa->dsa_dr; dmu_buf_impl_t *db = dr->dr_dbuf; mutex_enter(&db->db_mtx); ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC); if (zio->io_error == 0) { dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE); if (dr->dt.dl.dr_nopwrite) { blkptr_t *bp = zio->io_bp; blkptr_t *bp_orig = &zio->io_bp_orig; uint8_t chksum = BP_GET_CHECKSUM(bp_orig); ASSERT(BP_EQUAL(bp, bp_orig)); ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF); ASSERT(zio_checksum_table[chksum].ci_flags & ZCHECKSUM_FLAG_NOPWRITE); } dr->dt.dl.dr_overridden_by = *zio->io_bp; dr->dt.dl.dr_override_state = DR_OVERRIDDEN; dr->dt.dl.dr_copies = zio->io_prop.zp_copies; /* * Old style holes are filled with all zeros, whereas * new-style holes maintain their lsize, type, level, * and birth time (see zio_write_compress). While we * need to reset the BP_SET_LSIZE() call that happened * in dmu_sync_ready for old style holes, we do *not* * want to wipe out the information contained in new * style holes. Thus, only zero out the block pointer if * it's an old style hole. */ if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by) && dr->dt.dl.dr_overridden_by.blk_birth == 0) BP_ZERO(&dr->dt.dl.dr_overridden_by); } else { dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; } cv_broadcast(&db->db_changed); mutex_exit(&db->db_mtx); dsa->dsa_done(dsa->dsa_zgd, zio->io_error); kmem_free(dsa, sizeof (*dsa)); } static void dmu_sync_late_arrival_done(zio_t *zio) { blkptr_t *bp = zio->io_bp; dmu_sync_arg_t *dsa = zio->io_private; blkptr_t *bp_orig = &zio->io_bp_orig; if (zio->io_error == 0 && !BP_IS_HOLE(bp)) { /* * If we didn't allocate a new block (i.e. ZIO_FLAG_NOPWRITE) * then there is nothing to do here. Otherwise, free the * newly allocated block in this txg. */ if (zio->io_flags & ZIO_FLAG_NOPWRITE) { ASSERT(BP_EQUAL(bp, bp_orig)); } else { ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig)); ASSERT(zio->io_bp->blk_birth == zio->io_txg); ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa)); zio_free(zio->io_spa, zio->io_txg, zio->io_bp); } } dmu_tx_commit(dsa->dsa_tx); dsa->dsa_done(dsa->dsa_zgd, zio->io_error); kmem_free(dsa, sizeof (*dsa)); } static int dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd, zio_prop_t *zp, zbookmark_phys_t *zb) { dmu_sync_arg_t *dsa; dmu_tx_t *tx; tx = dmu_tx_create(os); dmu_tx_hold_space(tx, zgd->zgd_db->db_size); if (dmu_tx_assign(tx, TXG_WAIT) != 0) { dmu_tx_abort(tx); /* Make zl_get_data do txg_waited_synced() */ return (SET_ERROR(EIO)); } dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); dsa->dsa_dr = NULL; dsa->dsa_done = done; dsa->dsa_zgd = zgd; dsa->dsa_tx = tx; - zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp, - zgd->zgd_db->db_data, zgd->zgd_db->db_size, zp, - dmu_sync_late_arrival_ready, NULL, dmu_sync_late_arrival_done, dsa, - ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb)); + zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), + zgd->zgd_bp, zgd->zgd_db->db_data, zgd->zgd_db->db_size, + zp, dmu_sync_late_arrival_ready, NULL, + NULL, dmu_sync_late_arrival_done, dsa, ZIO_PRIORITY_SYNC_WRITE, + ZIO_FLAG_CANFAIL, zb)); return (0); } /* * Intent log support: sync the block associated with db to disk. * N.B. and XXX: the caller is responsible for making sure that the * data isn't changing while dmu_sync() is writing it. * * Return values: * * EEXIST: this txg has already been synced, so there's nothing to do. * The caller should not log the write. * * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do. * The caller should not log the write. * * EALREADY: this block is already in the process of being synced. * The caller should track its progress (somehow). * * EIO: could not do the I/O. * The caller should do a txg_wait_synced(). * * 0: the I/O has been initiated. * The caller should log this blkptr in the done callback. * It is possible that the I/O will fail, in which case * the error will be reported to the done callback and * propagated to pio from zio_done(). */ int dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd) { blkptr_t *bp = zgd->zgd_bp; dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db; objset_t *os = db->db_objset; dsl_dataset_t *ds = os->os_dsl_dataset; dbuf_dirty_record_t *dr; dmu_sync_arg_t *dsa; zbookmark_phys_t zb; zio_prop_t zp; dnode_t *dn; ASSERT(pio != NULL); ASSERT(txg != 0); SET_BOOKMARK(&zb, ds->ds_object, db->db.db_object, db->db_level, db->db_blkid); DB_DNODE_ENTER(db); dn = DB_DNODE(db); dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp); DB_DNODE_EXIT(db); /* * If we're frozen (running ziltest), we always need to generate a bp. */ if (txg > spa_freeze_txg(os->os_spa)) return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); /* * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf() * and us. If we determine that this txg is not yet syncing, * but it begins to sync a moment later, that's OK because the * sync thread will block in dbuf_sync_leaf() until we drop db_mtx. */ mutex_enter(&db->db_mtx); if (txg <= spa_last_synced_txg(os->os_spa)) { /* * This txg has already synced. There's nothing to do. */ mutex_exit(&db->db_mtx); return (SET_ERROR(EEXIST)); } if (txg <= spa_syncing_txg(os->os_spa)) { /* * This txg is currently syncing, so we can't mess with * the dirty record anymore; just write a new log block. */ mutex_exit(&db->db_mtx); return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); } dr = db->db_last_dirty; while (dr && dr->dr_txg != txg) dr = dr->dr_next; if (dr == NULL) { /* * There's no dr for this dbuf, so it must have been freed. * There's no need to log writes to freed blocks, so we're done. */ mutex_exit(&db->db_mtx); return (SET_ERROR(ENOENT)); } ASSERT(dr->dr_next == NULL || dr->dr_next->dr_txg < txg); /* * Assume the on-disk data is X, the current syncing data (in * txg - 1) is Y, and the current in-memory data is Z (currently * in dmu_sync). * * We usually want to perform a nopwrite if X and Z are the * same. However, if Y is different (i.e. the BP is going to * change before this write takes effect), then a nopwrite will * be incorrect - we would override with X, which could have * been freed when Y was written. * * (Note that this is not a concern when we are nop-writing from * syncing context, because X and Y must be identical, because * all previous txgs have been synced.) * * Therefore, we disable nopwrite if the current BP could change * before this TXG. There are two ways it could change: by * being dirty (dr_next is non-NULL), or by being freed * (dnode_block_freed()). This behavior is verified by * zio_done(), which VERIFYs that the override BP is identical * to the on-disk BP. */ DB_DNODE_ENTER(db); dn = DB_DNODE(db); if (dr->dr_next != NULL || dnode_block_freed(dn, db->db_blkid)) zp.zp_nopwrite = B_FALSE; DB_DNODE_EXIT(db); ASSERT(dr->dr_txg == txg); if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC || dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { /* * We have already issued a sync write for this buffer, * or this buffer has already been synced. It could not * have been dirtied since, or we would have cleared the state. */ mutex_exit(&db->db_mtx); return (SET_ERROR(EALREADY)); } ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC; mutex_exit(&db->db_mtx); dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); dsa->dsa_dr = dr; dsa->dsa_done = done; dsa->dsa_zgd = zgd; dsa->dsa_tx = NULL; zio_nowait(arc_write(pio, os->os_spa, txg, bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db), DBUF_IS_L2COMPRESSIBLE(db), &zp, dmu_sync_ready, - NULL, dmu_sync_done, dsa, ZIO_PRIORITY_SYNC_WRITE, - ZIO_FLAG_CANFAIL, &zb)); + NULL, NULL, dmu_sync_done, dsa, + ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb)); return (0); } int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs, dmu_tx_t *tx) { dnode_t *dn; int err; err = dnode_hold(os, object, FTAG, &dn); if (err) return (err); err = dnode_set_blksz(dn, size, ibs, tx); dnode_rele(dn, FTAG); return (err); } void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, dmu_tx_t *tx) { dnode_t *dn; /* * Send streams include each object's checksum function. This * check ensures that the receiving system can understand the * checksum function transmitted. */ ASSERT3U(checksum, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS); VERIFY0(dnode_hold(os, object, FTAG, &dn)); ASSERT3U(checksum, <, ZIO_CHECKSUM_FUNCTIONS); dn->dn_checksum = checksum; dnode_setdirty(dn, tx); dnode_rele(dn, FTAG); } void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, dmu_tx_t *tx) { dnode_t *dn; /* * Send streams include each object's compression function. This * check ensures that the receiving system can understand the * compression function transmitted. */ ASSERT3U(compress, <, ZIO_COMPRESS_LEGACY_FUNCTIONS); VERIFY0(dnode_hold(os, object, FTAG, &dn)); dn->dn_compress = compress; dnode_setdirty(dn, tx); dnode_rele(dn, FTAG); } int zfs_mdcomp_disable = 0; SYSCTL_INT(_vfs_zfs, OID_AUTO, mdcomp_disable, CTLFLAG_RWTUN, &zfs_mdcomp_disable, 0, "Disable metadata compression"); /* * When the "redundant_metadata" property is set to "most", only indirect * blocks of this level and higher will have an additional ditto block. */ int zfs_redundant_metadata_most_ditto_level = 2; void dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp) { dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET; boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) || (wp & WP_SPILL)); enum zio_checksum checksum = os->os_checksum; enum zio_compress compress = os->os_compress; enum zio_checksum dedup_checksum = os->os_dedup_checksum; boolean_t dedup = B_FALSE; boolean_t nopwrite = B_FALSE; boolean_t dedup_verify = os->os_dedup_verify; int copies = os->os_copies; /* * We maintain different write policies for each of the following * types of data: * 1. metadata * 2. preallocated blocks (i.e. level-0 blocks of a dump device) * 3. all other level 0 blocks */ if (ismd) { if (zfs_mdcomp_disable) { compress = ZIO_COMPRESS_EMPTY; } else { /* * XXX -- we should design a compression algorithm * that specializes in arrays of bps. */ compress = zio_compress_select(os->os_spa, ZIO_COMPRESS_ON, ZIO_COMPRESS_ON); } /* * Metadata always gets checksummed. If the data * checksum is multi-bit correctable, and it's not a * ZBT-style checksum, then it's suitable for metadata * as well. Otherwise, the metadata checksum defaults * to fletcher4. */ if (!(zio_checksum_table[checksum].ci_flags & ZCHECKSUM_FLAG_METADATA) || (zio_checksum_table[checksum].ci_flags & ZCHECKSUM_FLAG_EMBEDDED)) checksum = ZIO_CHECKSUM_FLETCHER_4; if (os->os_redundant_metadata == ZFS_REDUNDANT_METADATA_ALL || (os->os_redundant_metadata == ZFS_REDUNDANT_METADATA_MOST && (level >= zfs_redundant_metadata_most_ditto_level || DMU_OT_IS_METADATA(type) || (wp & WP_SPILL)))) copies++; } else if (wp & WP_NOFILL) { ASSERT(level == 0); /* * If we're writing preallocated blocks, we aren't actually * writing them so don't set any policy properties. These * blocks are currently only used by an external subsystem * outside of zfs (i.e. dump) and not written by the zio * pipeline. */ compress = ZIO_COMPRESS_OFF; checksum = ZIO_CHECKSUM_NOPARITY; } else { compress = zio_compress_select(os->os_spa, dn->dn_compress, compress); checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ? zio_checksum_select(dn->dn_checksum, checksum) : dedup_checksum; /* * Determine dedup setting. If we are in dmu_sync(), * we won't actually dedup now because that's all * done in syncing context; but we do want to use the * dedup checkum. If the checksum is not strong * enough to ensure unique signatures, force * dedup_verify. */ if (dedup_checksum != ZIO_CHECKSUM_OFF) { dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE; if (!(zio_checksum_table[checksum].ci_flags & ZCHECKSUM_FLAG_DEDUP)) dedup_verify = B_TRUE; } /* * Enable nopwrite if we have secure enough checksum * algorithm (see comment in zio_nop_write) and * compression is enabled. We don't enable nopwrite if * dedup is enabled as the two features are mutually * exclusive. */ nopwrite = (!dedup && (zio_checksum_table[checksum].ci_flags & ZCHECKSUM_FLAG_NOPWRITE) && compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled); } zp->zp_checksum = checksum; zp->zp_compress = compress; zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type; zp->zp_level = level; zp->zp_copies = MIN(copies, spa_max_replication(os->os_spa)); zp->zp_dedup = dedup; zp->zp_dedup_verify = dedup && dedup_verify; zp->zp_nopwrite = nopwrite; } int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off) { dnode_t *dn; int err; /* * Sync any current changes before * we go trundling through the block pointers. */ err = dmu_object_wait_synced(os, object); if (err) { return (err); } err = dnode_hold(os, object, FTAG, &dn); if (err) { return (err); } err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0); dnode_rele(dn, FTAG); return (err); } /* * Given the ZFS object, if it contains any dirty nodes * this function flushes all dirty blocks to disk. This * ensures the DMU object info is updated. A more efficient * future version might just find the TXG with the maximum * ID and wait for that to be synced. */ int dmu_object_wait_synced(objset_t *os, uint64_t object) { dnode_t *dn; int error, i; error = dnode_hold(os, object, FTAG, &dn); if (error) { return (error); } for (i = 0; i < TXG_SIZE; i++) { if (list_link_active(&dn->dn_dirty_link[i])) { break; } } dnode_rele(dn, FTAG); if (i != TXG_SIZE) { txg_wait_synced(dmu_objset_pool(os), 0); } return (0); } void dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) { dnode_phys_t *dnp; rw_enter(&dn->dn_struct_rwlock, RW_READER); mutex_enter(&dn->dn_mtx); dnp = dn->dn_phys; doi->doi_data_block_size = dn->dn_datablksz; doi->doi_metadata_block_size = dn->dn_indblkshift ? 1ULL << dn->dn_indblkshift : 0; doi->doi_type = dn->dn_type; doi->doi_bonus_type = dn->dn_bonustype; doi->doi_bonus_size = dn->dn_bonuslen; doi->doi_indirection = dn->dn_nlevels; doi->doi_checksum = dn->dn_checksum; doi->doi_compress = dn->dn_compress; doi->doi_nblkptr = dn->dn_nblkptr; doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9; doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz; doi->doi_fill_count = 0; for (int i = 0; i < dnp->dn_nblkptr; i++) doi->doi_fill_count += BP_GET_FILL(&dnp->dn_blkptr[i]); mutex_exit(&dn->dn_mtx); rw_exit(&dn->dn_struct_rwlock); } /* * Get information on a DMU object. * If doi is NULL, just indicates whether the object exists. */ int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi) { dnode_t *dn; int err = dnode_hold(os, object, FTAG, &dn); if (err) return (err); if (doi != NULL) dmu_object_info_from_dnode(dn, doi); dnode_rele(dn, FTAG); return (0); } /* * As above, but faster; can be used when you have a held dbuf in hand. */ void dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; DB_DNODE_ENTER(db); dmu_object_info_from_dnode(DB_DNODE(db), doi); DB_DNODE_EXIT(db); } /* * Faster still when you only care about the size. * This is specifically optimized for zfs_getattr(). */ void dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize, u_longlong_t *nblk512) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; dnode_t *dn; DB_DNODE_ENTER(db); dn = DB_DNODE(db); *blksize = dn->dn_datablksz; /* add 1 for dnode space */ *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >> SPA_MINBLOCKSHIFT) + 1; DB_DNODE_EXIT(db); } void byteswap_uint64_array(void *vbuf, size_t size) { uint64_t *buf = vbuf; size_t count = size >> 3; int i; ASSERT((size & 7) == 0); for (i = 0; i < count; i++) buf[i] = BSWAP_64(buf[i]); } void byteswap_uint32_array(void *vbuf, size_t size) { uint32_t *buf = vbuf; size_t count = size >> 2; int i; ASSERT((size & 3) == 0); for (i = 0; i < count; i++) buf[i] = BSWAP_32(buf[i]); } void byteswap_uint16_array(void *vbuf, size_t size) { uint16_t *buf = vbuf; size_t count = size >> 1; int i; ASSERT((size & 1) == 0); for (i = 0; i < count; i++) buf[i] = BSWAP_16(buf[i]); } /* ARGSUSED */ void byteswap_uint8_array(void *vbuf, size_t size) { } void dmu_init(void) { zfs_dbgmsg_init(); sa_cache_init(); xuio_stat_init(); dmu_objset_init(); dnode_init(); dbuf_init(); zfetch_init(); zio_compress_init(); l2arc_init(); arc_init(); } void dmu_fini(void) { arc_fini(); /* arc depends on l2arc, so arc must go first */ l2arc_fini(); zfetch_fini(); zio_compress_fini(); dbuf_fini(); dnode_fini(); dmu_objset_fini(); xuio_stat_fini(); sa_cache_fini(); zfs_dbgmsg_fini(); } Index: stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_objset.c =================================================================== --- stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_objset.c (revision 304137) +++ stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_objset.c (revision 304138) @@ -1,1988 +1,1988 @@ /* * 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, 2014 by Delphix. All rights reserved. + * Copyright (c) 2012, 2016 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 2015 Nexenta Systems, Inc. All rights reserved. * Copyright (c) 2015, STRATO AG, Inc. All rights reserved. * Copyright (c) 2014 Integros [integros.com] */ /* 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; 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); } } 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; if (DMU_OS_IS_L2COMPRESSIBLE(os)) aflags |= ARC_FLAG_L2COMPRESS; 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_buf_alloc(spa, sizeof (objset_phys_t), &os->os_phys_buf, ARC_BUFC_METADATA); 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)); (void) arc_buf_remove_ref(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_buf_alloc(spa, size, &os->os_phys_buf, ARC_BUFC_METADATA); 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) { VERIFY(arc_buf_remove_ref(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])); } 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_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; err = dmu_objset_open_impl(dsl_dataset_get_spa(ds), ds, dsl_dataset_get_blkptr(ds), &os); 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[MAXNAMELEN]; 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); VERIFY(arc_buf_remove_ref(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_obj_lock); mutex_destroy(&os->os_user_ptr_lock); 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. */ while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift + (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) < DN_MAX_OBJECT * sizeof (dnode_phys_t)) 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)); 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)); bp = dsl_dataset_get_blkptr(ds); os = dmu_objset_create_impl(pdd->dd_pool->dp_spa, ds, bp, doca->doca_type, tx); 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)); 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[MAXNAMELEN]; 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) { dnode_t *dn; while (dn = list_head(list)) { 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); if (newlist) { (void) dnode_add_ref(dn, newlist); list_insert_tail(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)); ASSERT3P(bp, ==, os->os_rootbp); 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]); } /* 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); } } /* 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; 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, &zp); zio = arc_write(pio, os->os_spa, tx->tx_txg, os->os_rootbp, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os), - DMU_OS_IS_L2COMPRESSIBLE(os), &zp, dmu_objset_write_ready, - NULL, dmu_objset_write_done, os, ZIO_PRIORITY_ASYNC_WRITE, - ZIO_FLAG_MUSTSUCCEED, &zb); + DMU_OS_IS_L2COMPRESSIBLE(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. */ list_create(newlist, sizeof (dnode_t), 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); list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff]; while (dr = list_head(list)) { ASSERT0(dr->dr_dbuf->db_level); list_remove(list, dr); if (dr->dr_zio) zio_nowait(dr->dr_zio); } /* * 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])); } 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); } static void do_userquota_update(objset_t *os, uint64_t used, uint64_t flags, uint64_t user, uint64_t group, boolean_t subtract, dmu_tx_t *tx) { if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) { int64_t delta = DNODE_SIZE + used; if (subtract) delta = -delta; VERIFY3U(0, ==, zap_increment_int(os, DMU_USERUSED_OBJECT, user, delta, tx)); VERIFY3U(0, ==, zap_increment_int(os, DMU_GROUPUSED_OBJECT, group, delta, tx)); } } void dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx) { dnode_t *dn; list_t *list = &os->os_synced_dnodes; ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os)); while (dn = list_head(list)) { 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) { VERIFY(0 == zap_create_claim(os, DMU_USERUSED_OBJECT, DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); VERIFY(0 == zap_create_claim(os, DMU_GROUPUSED_OBJECT, DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); } /* * We intentionally modify the zap object even if the * net delta is zero. Otherwise * the block of the zap obj could be shared between * datasets but need to be different between them after * a bprewrite. */ flags = dn->dn_id_flags; ASSERT(flags); if (flags & DN_ID_OLD_EXIST) { do_userquota_update(os, dn->dn_oldused, dn->dn_oldflags, dn->dn_olduid, dn->dn_oldgid, B_TRUE, tx); } if (flags & DN_ID_NEW_EXIST) { do_userquota_update(os, DN_USED_BYTES(dn->dn_phys), dn->dn_phys->dn_flags, dn->dn_newuid, dn->dn_newgid, B_FALSE, tx); } 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); } } /* * 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_FIRST, 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; 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; dmu_objset_find_ctx_t *child_dcp; 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; err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, NULL, 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); child_dcp = kmem_alloc(sizeof (*child_dcp), KM_SLEEP); *child_dcp = *dcp; child_dcp->dc_ddobj = attr->za_first_integer; 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); } } dsl_dir_rele(dd, FTAG); kmem_free(attr, sizeof (zap_attribute_t)); if (err != 0) goto out; /* * Apply to self. */ err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 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); } 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_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 MAXNAMELEN bytes */ int dmu_fsname(const char *snapname, char *buf) { char *atp = strchr(snapname, '@'); if (atp == NULL) return (SET_ERROR(EINVAL)); if (atp - snapname >= MAXNAMELEN) return (SET_ERROR(ENAMETOOLONG)); (void) strlcpy(buf, snapname, atp - snapname + 1); return (0); } Index: stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dnode_sync.c =================================================================== --- stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dnode_sync.c (revision 304137) +++ stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dnode_sync.c (revision 304138) @@ -1,726 +1,720 @@ /* * 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, 2015 by Delphix. All rights reserved. + * Copyright (c) 2012, 2016 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); - /* check for existing blkptrs in the dnode */ - for (i = 0; i < nblkptr; i++) - if (!BP_IS_HOLE(&dn->dn_phys->dn_blkptr[i])) - break; - if (i != nblkptr) { - /* 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); - } + /* 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, i; int epbs, shift; /* * 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; 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 (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, db->db_level - 1, i, 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) { /* didn't find any non-holes */ bzero(db->db.db_data, db->db.db_size); 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_clear(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_evict(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) { 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: stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/arc.h =================================================================== --- stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/arc.h (revision 304137) +++ stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/arc.h (revision 304138) @@ -1,182 +1,184 @@ /* * 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, 2014 by Delphix. All rights reserved. + * Copyright (c) 2012, 2016 by Delphix. All rights reserved. * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. */ #ifndef _SYS_ARC_H #define _SYS_ARC_H #include #ifdef __cplusplus extern "C" { #endif #include #include #include /* * Used by arc_flush() to inform arc_evict_state() that it should evict * all available buffers from the arc state being passed in. */ #define ARC_EVICT_ALL -1ULL typedef struct arc_buf_hdr arc_buf_hdr_t; typedef struct arc_buf arc_buf_t; typedef void arc_done_func_t(zio_t *zio, arc_buf_t *buf, void *priv); typedef int arc_evict_func_t(void *priv); /* generic arc_done_func_t's which you can use */ arc_done_func_t arc_bcopy_func; arc_done_func_t arc_getbuf_func; typedef enum arc_flags { /* * Public flags that can be passed into the ARC by external consumers. */ ARC_FLAG_NONE = 1 << 0, /* No flags set */ ARC_FLAG_WAIT = 1 << 1, /* perform sync I/O */ ARC_FLAG_NOWAIT = 1 << 2, /* perform async I/O */ ARC_FLAG_PREFETCH = 1 << 3, /* I/O is a prefetch */ ARC_FLAG_CACHED = 1 << 4, /* I/O was in cache */ ARC_FLAG_L2CACHE = 1 << 5, /* cache in L2ARC */ ARC_FLAG_L2COMPRESS = 1 << 6, /* compress in L2ARC */ ARC_FLAG_PREDICTIVE_PREFETCH = 1 << 7, /* I/O from zfetch */ /* * Private ARC flags. These flags are private ARC only flags that * will show up in b_flags in the arc_hdr_buf_t. These flags should * only be set by ARC code. */ ARC_FLAG_IN_HASH_TABLE = 1 << 8, /* buffer is hashed */ ARC_FLAG_IO_IN_PROGRESS = 1 << 9, /* I/O in progress */ ARC_FLAG_IO_ERROR = 1 << 10, /* I/O failed for buf */ ARC_FLAG_FREED_IN_READ = 1 << 11, /* freed during read */ ARC_FLAG_BUF_AVAILABLE = 1 << 12, /* block not in use */ ARC_FLAG_INDIRECT = 1 << 13, /* indirect block */ /* Indicates that block was read with ASYNC priority. */ ARC_FLAG_PRIO_ASYNC_READ = 1 << 14, ARC_FLAG_L2_WRITING = 1 << 15, /* write in progress */ ARC_FLAG_L2_EVICTED = 1 << 16, /* evicted during I/O */ ARC_FLAG_L2_WRITE_HEAD = 1 << 17, /* head of write list */ /* indicates that the buffer contains metadata (otherwise, data) */ ARC_FLAG_BUFC_METADATA = 1 << 18, /* Flags specifying whether optional hdr struct fields are defined */ ARC_FLAG_HAS_L1HDR = 1 << 19, ARC_FLAG_HAS_L2HDR = 1 << 20, } arc_flags_t; struct arc_buf { arc_buf_hdr_t *b_hdr; arc_buf_t *b_next; kmutex_t b_evict_lock; void *b_data; arc_evict_func_t *b_efunc; void *b_private; }; typedef enum arc_buf_contents { ARC_BUFC_DATA, /* buffer contains data */ ARC_BUFC_METADATA, /* buffer contains metadata */ ARC_BUFC_NUMTYPES } arc_buf_contents_t; /* * The following breakdows of arc_size exist for kstat only. */ typedef enum arc_space_type { ARC_SPACE_DATA, ARC_SPACE_META, ARC_SPACE_HDRS, ARC_SPACE_L2HDRS, ARC_SPACE_OTHER, ARC_SPACE_NUMTYPES } arc_space_type_t; void arc_space_consume(uint64_t space, arc_space_type_t type); void arc_space_return(uint64_t space, arc_space_type_t type); arc_buf_t *arc_buf_alloc(spa_t *spa, int size, void *tag, arc_buf_contents_t type); arc_buf_t *arc_loan_buf(spa_t *spa, int size); void arc_return_buf(arc_buf_t *buf, void *tag); void arc_loan_inuse_buf(arc_buf_t *buf, void *tag); void arc_buf_add_ref(arc_buf_t *buf, void *tag); boolean_t arc_buf_remove_ref(arc_buf_t *buf, void *tag); int arc_buf_size(arc_buf_t *buf); void arc_release(arc_buf_t *buf, void *tag); int arc_released(arc_buf_t *buf); void arc_buf_freeze(arc_buf_t *buf); void arc_buf_thaw(arc_buf_t *buf); boolean_t arc_buf_eviction_needed(arc_buf_t *buf); #ifdef ZFS_DEBUG int arc_referenced(arc_buf_t *buf); #endif int arc_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, arc_done_func_t *done, void *priv, zio_priority_t priority, int flags, arc_flags_t *arc_flags, const zbookmark_phys_t *zb); zio_t *arc_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, arc_buf_t *buf, boolean_t l2arc, boolean_t l2arc_compress, - const zio_prop_t *zp, arc_done_func_t *ready, arc_done_func_t *physdone, - arc_done_func_t *done, void *priv, zio_priority_t priority, - int zio_flags, const zbookmark_phys_t *zb); + const zio_prop_t *zp, + arc_done_func_t *ready, arc_done_func_t *child_ready, + arc_done_func_t *physdone, arc_done_func_t *done, + void *priv, zio_priority_t priority, int zio_flags, + const zbookmark_phys_t *zb); void arc_freed(spa_t *spa, const blkptr_t *bp); void arc_set_callback(arc_buf_t *buf, arc_evict_func_t *func, void *priv); boolean_t arc_clear_callback(arc_buf_t *buf); void arc_flush(spa_t *spa, boolean_t retry); void arc_tempreserve_clear(uint64_t reserve); int arc_tempreserve_space(uint64_t reserve, uint64_t txg); void arc_init(void); void arc_fini(void); /* * Level 2 ARC */ void l2arc_add_vdev(spa_t *spa, vdev_t *vd); void l2arc_remove_vdev(vdev_t *vd); boolean_t l2arc_vdev_present(vdev_t *vd); void l2arc_init(void); void l2arc_fini(void); void l2arc_start(void); void l2arc_stop(void); #ifdef illumos #ifndef _KERNEL extern boolean_t arc_watch; extern int arc_procfd; #endif #endif /* illumos */ #ifdef __cplusplus } #endif #endif /* _SYS_ARC_H */ Index: stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zio.h =================================================================== --- stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zio.h (revision 304137) +++ stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zio.h (revision 304138) @@ -1,646 +1,648 @@ /* * 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, 2014 by Delphix. All rights reserved. + * Copyright (c) 2012, 2016 by Delphix. All rights reserved. * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. */ #ifndef _ZIO_H #define _ZIO_H #include #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, #ifdef illumos ZIO_CHECKSUM_EDONR, #endif 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 enum zio_compress { ZIO_COMPRESS_INHERIT = 0, ZIO_COMPRESS_ON, ZIO_COMPRESS_OFF, ZIO_COMPRESS_LZJB, ZIO_COMPRESS_EMPTY, ZIO_COMPRESS_GZIP_1, ZIO_COMPRESS_GZIP_2, ZIO_COMPRESS_GZIP_3, ZIO_COMPRESS_GZIP_4, ZIO_COMPRESS_GZIP_5, ZIO_COMPRESS_GZIP_6, ZIO_COMPRESS_GZIP_7, ZIO_COMPRESS_GZIP_8, ZIO_COMPRESS_GZIP_9, ZIO_COMPRESS_ZLE, ZIO_COMPRESS_LZ4, ZIO_COMPRESS_FUNCTIONS }; /* * 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_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, #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 << 14, /* must be first for INHERIT */ ZIO_FLAG_PROBE = 1 << 15, ZIO_FLAG_TRYHARD = 1 << 16, ZIO_FLAG_OPTIONAL = 1 << 17, #define ZIO_FLAG_VDEV_INHERIT (ZIO_FLAG_DONT_QUEUE - 1) /* * Flags not inherited by any children. */ ZIO_FLAG_DONT_QUEUE = 1 << 18, /* must be first for INHERIT */ ZIO_FLAG_DONT_PROPAGATE = 1 << 19, ZIO_FLAG_IO_BYPASS = 1 << 20, ZIO_FLAG_IO_REWRITE = 1 << 21, ZIO_FLAG_RAW = 1 << 22, ZIO_FLAG_GANG_CHILD = 1 << 23, ZIO_FLAG_DDT_CHILD = 1 << 24, ZIO_FLAG_GODFATHER = 1 << 25, ZIO_FLAG_NOPWRITE = 1 << 26, ZIO_FLAG_REEXECUTED = 1 << 27, ZIO_FLAG_DELEGATED = 1 << 28, }; #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 number 122 and 123 to indicate checksum errors and * fragmentation. Those doesn't collide with any errno values as they * are greater than ELAST. */ #define ECKSUM 122 #define EFRAGS 123 typedef void zio_done_func_t(zio_t *zio); 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_link { zio_t *zl_parent; zio_t *zl_child; list_node_t zl_parent_node; list_node_t zl_child_node; } zio_link_t; /* * Used for TRIM kstat. */ typedef struct zio_trim_stats { /* * Number of bytes successfully TRIMmed. */ kstat_named_t bytes; /* * Number of successful TRIM requests. */ kstat_named_t success; /* * Number of TRIM requests that failed because TRIM is not * supported. */ kstat_named_t unsupported; /* * Number of TRIM requests that failed for other reasons. */ kstat_named_t failed; } zio_trim_stats_t; extern zio_trim_stats_t zio_trim_stats; #define ZIO_TRIM_STAT_INCR(stat, val) \ atomic_add_64(&zio_trim_stats.stat.value.ui64, (val)); #define ZIO_TRIM_STAT_BUMP(stat) \ ZIO_TRIM_STAT_INCR(stat, 1); 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_link_t *io_walk_link; 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; /* 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_target_timestamp; avl_node_t io_queue_node; avl_node_t io_offset_node; /* 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; 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; avl_node_t io_trim_node; list_node_t io_trim_link; }; extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd, zio_done_func_t *done, void *priv, enum zio_flag flags); extern zio_t *zio_root(spa_t *spa, zio_done_func_t *done, void *priv, enum zio_flag flags); extern 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 *priv, 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, const zio_prop_t *zp, - zio_done_func_t *ready, zio_done_func_t *physdone, zio_done_func_t *done, - void *priv, - zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb); + zio_done_func_t *ready, zio_done_func_t *children_ready, + zio_done_func_t *physdone, zio_done_func_t *done, + void *priv, 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 *priv, 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 *priv, enum zio_flag flags); extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd, uint64_t offset, uint64_t size, zio_done_func_t *done, void *priv, zio_priority_t priority, 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 *priv, 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 *priv, 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, uint64_t size, 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 zio_t *zio_trim(zio_t *zio, spa_t *spa, vdev_t *vd, uint64_t offset, uint64_t size); 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); extern zio_t *zio_walk_children(zio_t *pio); 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_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 *priv); 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 *priv); 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: stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zio.c =================================================================== --- stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zio.c (revision 304137) +++ stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zio.c (revision 304138) @@ -1,3662 +1,3675 @@ /* * 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, 2016 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 SYSCTL_DECL(_vfs_zfs); SYSCTL_NODE(_vfs_zfs, OID_AUTO, zio, CTLFLAG_RW, 0, "ZFS ZIO"); #if defined(__amd64__) static int zio_use_uma = 1; #else static int zio_use_uma = 0; #endif SYSCTL_INT(_vfs_zfs_zio, OID_AUTO, use_uma, CTLFLAG_RDTUN, &zio_use_uma, 0, "Use uma(9) for ZIO allocations"); static int zio_exclude_metadata = 0; SYSCTL_INT(_vfs_zfs_zio, OID_AUTO, exclude_metadata, CTLFLAG_RDTUN, &zio_exclude_metadata, 0, "Exclude metadata buffers from dumps as well"); zio_trim_stats_t zio_trim_stats = { { "bytes", KSTAT_DATA_UINT64, "Number of bytes successfully TRIMmed" }, { "success", KSTAT_DATA_UINT64, "Number of successful TRIM requests" }, { "unsupported", KSTAT_DATA_UINT64, "Number of TRIM requests that failed because TRIM is not supported" }, { "failed", KSTAT_DATA_UINT64, "Number of TRIM requests that failed for reasons other than not supported" }, }; static kstat_t *zio_trim_ksp; /* * ========================================================================== * I/O type descriptions * ========================================================================== */ const char *zio_type_name[ZIO_TYPES] = { "zio_null", "zio_read", "zio_write", "zio_free", "zio_claim", "zio_ioctl" }; /* * ========================================================================== * 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 */ SYSCTL_INT(_vfs_zfs, OID_AUTO, sync_pass_deferred_free, CTLFLAG_RDTUN, &zfs_sync_pass_deferred_free, 0, "defer frees starting in this pass"); int zfs_sync_pass_dont_compress = 5; /* don't compress starting in this pass */ SYSCTL_INT(_vfs_zfs, OID_AUTO, sync_pass_dont_compress, CTLFLAG_RDTUN, &zfs_sync_pass_dont_compress, 0, "don't compress starting in this pass"); int zfs_sync_pass_rewrite = 2; /* rewrite new bps starting in this pass */ SYSCTL_INT(_vfs_zfs, OID_AUTO, sync_pass_rewrite, CTLFLAG_RDTUN, &zfs_sync_pass_rewrite, 0, "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 illumos #ifdef ZFS_DEBUG int zio_buf_debug_limit = 16384; #else int zio_buf_debug_limit = 0; #endif #endif void zio_init(void) { size_t c; 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); if (!zio_use_uma) goto out; /* * 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; int cflags = zio_exclude_metadata ? KMC_NODEBUG : 0; while (!ISP2(p2)) p2 &= p2 - 1; #ifdef illumos #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 #endif /* illumos */ 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, NULL, cflags | KMC_NOTOUCH | KMC_NODEBUG); } } 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]; } out: zio_inject_init(); zio_trim_ksp = kstat_create("zfs", 0, "zio_trim", "misc", KSTAT_TYPE_NAMED, sizeof(zio_trim_stats) / sizeof(kstat_named_t), KSTAT_FLAG_VIRTUAL); if (zio_trim_ksp != NULL) { zio_trim_ksp->ks_data = &zio_trim_stats; kstat_install(zio_trim_ksp); } } 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(); if (zio_trim_ksp != NULL) { kstat_delete(zio_trim_ksp); zio_trim_ksp = NULL; } } /* * ========================================================================== * 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; int flags = zio_exclude_metadata ? KM_NODEBUG : 0; VERIFY3U(c, <, SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); if (zio_use_uma) return (kmem_cache_alloc(zio_buf_cache[c], KM_PUSHPAGE)); else return (kmem_alloc(size, KM_SLEEP|flags)); } /* * 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); if (zio_use_uma) return (kmem_cache_alloc(zio_data_buf_cache[c], KM_PUSHPAGE)); else return (kmem_alloc(size, KM_SLEEP | KM_NODEBUG)); } void zio_buf_free(void *buf, size_t size) { size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; VERIFY3U(c, <, SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); if (zio_use_uma) kmem_cache_free(zio_buf_cache[c], buf); else kmem_free(buf, size); } void zio_data_buf_free(void *buf, size_t size) { size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; VERIFY3U(c, <, SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); if (zio_use_uma) kmem_cache_free(zio_data_buf_cache[c], buf); else kmem_free(buf, size); } /* * ========================================================================== * Push and pop I/O transform buffers * ========================================================================== */ static 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; } static 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 * ========================================================================== */ /* * NOTE - Callers to zio_walk_parents() and zio_walk_children must * continue calling these functions until they return NULL. * Otherwise, the next caller will pick up the list walk in * some indeterminate state. (Otherwise every caller would * have to pass in a cookie to keep the state represented by * io_walk_link, which gets annoying.) */ zio_t * zio_walk_parents(zio_t *cio) { zio_link_t *zl = cio->io_walk_link; list_t *pl = &cio->io_parent_list; zl = (zl == NULL) ? list_head(pl) : list_next(pl, zl); cio->io_walk_link = 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 = pio->io_walk_link; list_t *cl = &pio->io_child_list; zl = (zl == NULL) ? list_head(cl) : list_next(cl, zl); pio->io_walk_link = zl; if (zl == NULL) return (NULL); ASSERT(zl->zl_parent == pio); return (zl->zl_child); } zio_t * zio_unique_parent(zio_t *cio) { zio_t *pio = zio_walk_parents(cio); VERIFY(zio_walk_parents(cio) == 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; 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) { pio->io_stall = NULL; mutex_exit(&pio->io_lock); zio_execute(pio); } 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]; } /* * ========================================================================== * 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 size, 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(type == ZIO_TYPE_FREE || size, <=, SPA_MAXBLOCKSIZE); ASSERT(P2PHASE(size, 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); 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)); 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 = size; 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_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) { 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, 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, 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 size, const zio_prop_t *zp, - zio_done_func_t *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_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, size, 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, done, private, ZIO_TYPE_WRITE, priority, flags, 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 (zfs_trim_enabled || 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, BP_GET_PSIZE(bp), 0))); } } zio_t * zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, uint64_t size, 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); if (zfs_trim_enabled) stage |= ZIO_STAGE_ISSUE_ASYNC | ZIO_STAGE_VDEV_IO_START | ZIO_STAGE_VDEV_IO_ASSESS; /* * 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. */ else if (BP_IS_GANG(bp) || BP_GET_DEDUP(bp)) stage |= ZIO_STAGE_ISSUE_ASYNC; flags |= ZIO_FLAG_DONT_QUEUE; zio = zio_create(pio, spa, txg, bp, NULL, size, 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), done, private, ZIO_TYPE_CLAIM, ZIO_PRIORITY_NOW, flags, NULL, 0, NULL, ZIO_STAGE_OPEN, ZIO_CLAIM_PIPELINE); return (zio); } zio_t * zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd, uint64_t offset, uint64_t size, zio_done_func_t *done, void *private, zio_priority_t priority, enum zio_flag flags) { zio_t *zio; int c; if (vd->vdev_children == 0) { zio = zio_create(pio, spa, 0, NULL, NULL, size, done, private, ZIO_TYPE_IOCTL, priority, flags, vd, offset, 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, offset, size, done, private, priority, 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, 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, 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; } /* Not all IO types require vdev io done stage e.g. free */ if (!(pio->io_pipeline & ZIO_STAGE_VDEV_IO_DONE)) pipeline &= ~ZIO_STAGE_VDEV_IO_DONE; 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; zio = zio_create(pio, pio->io_spa, pio->io_txg, bp, data, size, done, private, type, priority, flags, vd, offset, &pio->io_bookmark, ZIO_STAGE_VDEV_IO_START >> 1, pipeline); 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, 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, 0, 0, NULL, NULL, ZIO_PRIORITY_NOW, ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY)); } zio_t * zio_trim(zio_t *zio, spa_t *spa, vdev_t *vd, uint64_t offset, uint64_t size) { ASSERT(vd->vdev_ops->vdev_op_leaf); return (zio_create(zio, spa, 0, NULL, NULL, size, NULL, NULL, ZIO_TYPE_FREE, ZIO_PRIORITY_TRIM, ZIO_FLAG_DONT_AGGREGATE | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY, vd, offset, NULL, ZIO_STAGE_OPEN, ZIO_FREE_PHYS_PIPELINE)); } 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)) zio->io_orig_size = zio->io_size = 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) { 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_size; uint64_t psize = lsize; int pass = 1; /* * 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); if (zio->io_bp_override) { 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); 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); } zio->io_bp_override = NULL; BP_ZERO(bp); } 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 (compress != ZIO_COMPRESS_OFF) { 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); } } } /* * 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); ASSERT(q == ZIO_TASKQ_ISSUE || q == ZIO_TASKQ_INTERRUPT); /* * 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. */ #if defined(illumos) || !defined(_KERNEL) ASSERT(zio->io_tqent.tqent_next == NULL); #else ASSERT(zio->io_tqent.tqent_task.ta_pending == 0); #endif 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; 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; 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; 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); } 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_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'. */ for (cio = zio_walk_children(pio); cio != NULL; cio = cio_next) { cio_next = zio_walk_children(pio); 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 him. */ if (!(pio->io_flags & ZIO_FLAG_GODFATHER)) 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, BP_IS_GANG(bp) ? SPA_GANGBLOCKSIZE : BP_GET_PSIZE(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 && gio->io_data != NULL) 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; 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; error = metaslab_alloc(spa, spa_normal_class(spa), SPA_GANGBLOCKSIZE, bp, gbh_copies, txg, pio == gio ? NULL : gio->io_bp, METASLAB_HINTBP_FAVOR | METASLAB_GANG_HEADER); if (error) { 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_nowait(zio_write(zio, spa, txg, &gbh->zg_blkptr[g], (char *)pio->io_data + (pio->io_size - resid), lsize, &zp, - zio_write_gang_member_ready, NULL, NULL, &gn->gn_child[g], - pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), - &pio->io_bookmark)); + zio_write_gang_member_ready, NULL, NULL, NULL, + &gn->gn_child[g], pio->io_priority, + ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark)); } /* * 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; /* * 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; blkptr_t blk = *zio->io_bp; int error; ddt_bp_fill(ddp, &blk, ddp->ddp_phys_birth); ddt_exit(ddt); error = arc_read(NULL, spa, &blk, arc_getbuf_func, &abuf, ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, &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); VERIFY(arc_buf_remove_ref(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); while ((pio = zio_walk_parents(zio)) != 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) { while (zio_walk_parents(zio) != 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); 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; } 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, &czp, NULL, NULL, - zio_ddt_ditto_write_done, dde, zio->io_priority, + 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, zp, zio_ddt_child_write_ready, NULL, + 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 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)); /* * The dump device does not support gang blocks so allocation on * behalf of the dump device (i.e. ZIO_FLAG_NODATA) must avoid * the "fast" gang feature. */ flags |= (zio->io_flags & ZIO_FLAG_NODATA) ? METASLAB_GANG_AVOID : 0; flags |= (zio->io_flags & ZIO_FLAG_GANG_CHILD) ? METASLAB_GANG_CHILD : 0; error = metaslab_alloc(spa, mc, zio->io_size, bp, zio->io_prop.zp_copies, zio->io_txg, NULL, flags); if (error) { 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; ASSERT(txg > spa_syncing_txg(spa)); /* * ZIL blocks are always contiguous (i.e. not gang blocks) so we * set the METASLAB_GANG_AVOID flag so that they don't "fast gang" * when allocating them. */ if (use_slog) { error = metaslab_alloc(spa, spa_log_class(spa), size, new_bp, 1, txg, old_bp, METASLAB_HINTBP_AVOID | METASLAB_GANG_AVOID); } if (error) { error = metaslab_alloc(spa, spa_normal_class(spa), size, new_bp, 1, txg, old_bp, METASLAB_HINTBP_AVOID); } 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, write and delete 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; int ret; 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); } if (vd->vdev_ops->vdev_op_leaf && zio->io_type == ZIO_TYPE_FREE && zio->io_priority == ZIO_PRIORITY_NOW) { trim_map_free(vd, zio->io_offset, zio->io_size, zio->io_txg); return (ZIO_PIPELINE_CONTINUE); } /* * 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 = NULL; if (zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE) 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, abuf ? asize : 0, 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 the physical io we allow alignment * to a logical block size. */ uint64_t log_align = 1ULL << vd->vdev_top->vdev_logical_ashift; ASSERT0(P2PHASE(zio->io_offset, log_align)); ASSERT0(P2PHASE(zio->io_size, log_align)); } VERIFY(zio->io_type == ZIO_TYPE_READ || 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) { switch (zio->io_type) { case ZIO_TYPE_READ: if (vdev_cache_read(zio)) return (ZIO_PIPELINE_CONTINUE); /* FALLTHROUGH */ case ZIO_TYPE_WRITE: case ZIO_TYPE_FREE: 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); } break; } /* * Note that we ignore repair writes for TRIM because they can * conflict with normal writes. This isn't an issue because, by * definition, we only repair blocks that aren't freed. */ if (zio->io_type == ZIO_TYPE_WRITE && !(zio->io_flags & ZIO_FLAG_IO_REPAIR) && !trim_map_write_start(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 || zio->io_type == ZIO_TYPE_FREE); if (vd != NULL && vd->vdev_ops->vdev_op_leaf && (zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE || zio->io_type == ZIO_TYPE_FREE)) { if (zio->io_type == ZIO_TYPE_WRITE && !(zio->io_flags & ZIO_FLAG_IO_REPAIR)) trim_map_write_done(zio); 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 (zio->io_error == ENOTSUP && zio->io_type == ZIO_TYPE_FREE) { /* Not all devices support TRIM. */ } else 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 (zio->io_type == ZIO_TYPE_FREE && zio->io_priority != ZIO_PRIORITY_NOW) { switch (zio->io_error) { case 0: ZIO_TRIM_STAT_INCR(bytes, zio->io_size); ZIO_TRIM_STAT_BUMP(success); break; case EOPNOTSUPP: ZIO_TRIM_STAT_BUMP(unsupported); break; default: ZIO_TRIM_STAT_BUMP(failed); break; } } /* * 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 (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; 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) zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; mutex_enter(&zio->io_lock); zio->io_state[ZIO_WAIT_READY] = 1; pio = zio_walk_parents(zio); 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); 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); } 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; /* * 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); 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. */ for (pio = zio_walk_parents(zio); pio != NULL; pio = pio_next) { zio_link_t *zl = zio->io_walk_link; pio_next = zio_walk_parents(zio); if ((pio->io_flags & ZIO_FLAG_GODFATHER) && (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND)) { zio_remove_child(pio, zio, 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. */ #if defined(illumos) || !defined(_KERNEL) ASSERT(zio->io_tqent.tqent_next == NULL); #else ASSERT(zio->io_tqent.tqent_task.ta_pending == 0); #endif 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); for (pio = zio_walk_parents(zio); pio != NULL; pio = pio_next) { zio_link_t *zl = zio->io_walk_link; pio_next = zio_walk_parents(zio); zio_remove_child(pio, zio, 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_free_bp_init, zio_issue_async, zio_write_bp_init, 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_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); } Index: stable/11 =================================================================== --- stable/11 (revision 304137) +++ stable/11 (revision 304138) Property changes on: stable/11 ___________________________________________________________________ Modified: svn:mergeinfo ## -0,0 +0,1 ## Merged /head:r302838