4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
24 * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
25 * Copyright (c) 2014 by Saso Kiselkov. All rights reserved.
26 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
30 * DVA-based Adjustable Replacement Cache
32 * While much of the theory of operation used here is
33 * based on the self-tuning, low overhead replacement cache
34 * presented by Megiddo and Modha at FAST 2003, there are some
35 * significant differences:
37 * 1. The Megiddo and Modha model assumes any page is evictable.
38 * Pages in its cache cannot be "locked" into memory. This makes
39 * the eviction algorithm simple: evict the last page in the list.
40 * This also make the performance characteristics easy to reason
41 * about. Our cache is not so simple. At any given moment, some
42 * subset of the blocks in the cache are un-evictable because we
43 * have handed out a reference to them. Blocks are only evictable
44 * when there are no external references active. This makes
45 * eviction far more problematic: we choose to evict the evictable
46 * blocks that are the "lowest" in the list.
48 * There are times when it is not possible to evict the requested
49 * space. In these circumstances we are unable to adjust the cache
50 * size. To prevent the cache growing unbounded at these times we
51 * implement a "cache throttle" that slows the flow of new data
52 * into the cache until we can make space available.
54 * 2. The Megiddo and Modha model assumes a fixed cache size.
55 * Pages are evicted when the cache is full and there is a cache
56 * miss. Our model has a variable sized cache. It grows with
57 * high use, but also tries to react to memory pressure from the
58 * operating system: decreasing its size when system memory is
61 * 3. The Megiddo and Modha model assumes a fixed page size. All
62 * elements of the cache are therefore exactly the same size. So
63 * when adjusting the cache size following a cache miss, its simply
64 * a matter of choosing a single page to evict. In our model, we
65 * have variable sized cache blocks (rangeing from 512 bytes to
66 * 128K bytes). We therefore choose a set of blocks to evict to make
67 * space for a cache miss that approximates as closely as possible
68 * the space used by the new block.
70 * See also: "ARC: A Self-Tuning, Low Overhead Replacement Cache"
71 * by N. Megiddo & D. Modha, FAST 2003
77 * A new reference to a cache buffer can be obtained in two
78 * ways: 1) via a hash table lookup using the DVA as a key,
79 * or 2) via one of the ARC lists. The arc_read() interface
80 * uses method 1, while the internal ARC algorithms for
81 * adjusting the cache use method 2. We therefore provide two
82 * types of locks: 1) the hash table lock array, and 2) the
85 * Buffers do not have their own mutexes, rather they rely on the
86 * hash table mutexes for the bulk of their protection (i.e. most
87 * fields in the arc_buf_hdr_t are protected by these mutexes).
89 * buf_hash_find() returns the appropriate mutex (held) when it
90 * locates the requested buffer in the hash table. It returns
91 * NULL for the mutex if the buffer was not in the table.
93 * buf_hash_remove() expects the appropriate hash mutex to be
94 * already held before it is invoked.
96 * Each ARC state also has a mutex which is used to protect the
97 * buffer list associated with the state. When attempting to
98 * obtain a hash table lock while holding an ARC list lock you
99 * must use: mutex_tryenter() to avoid deadlock. Also note that
100 * the active state mutex must be held before the ghost state mutex.
102 * It as also possible to register a callback which is run when the
103 * arc_meta_limit is reached and no buffers can be safely evicted. In
104 * this case the arc user should drop a reference on some arc buffers so
105 * they can be reclaimed and the arc_meta_limit honored. For example,
106 * when using the ZPL each dentry holds a references on a znode. These
107 * dentries must be pruned before the arc buffer holding the znode can
110 * Note that the majority of the performance stats are manipulated
111 * with atomic operations.
113 * The L2ARC uses the l2ad_mtx on each vdev for the following:
115 * - L2ARC buflist creation
116 * - L2ARC buflist eviction
117 * - L2ARC write completion, which walks L2ARC buflists
118 * - ARC header destruction, as it removes from L2ARC buflists
119 * - ARC header release, as it removes from L2ARC buflists
125 * Every block that is in the ARC is tracked by an arc_buf_hdr_t structure.
126 * This structure can point either to a block that is still in the cache or to
127 * one that is only accessible in an L2 ARC device, or it can provide
128 * information about a block that was recently evicted. If a block is
129 * only accessible in the L2ARC, then the arc_buf_hdr_t only has enough
130 * information to retrieve it from the L2ARC device. This information is
131 * stored in the l2arc_buf_hdr_t sub-structure of the arc_buf_hdr_t. A block
132 * that is in this state cannot access the data directly.
134 * Blocks that are actively being referenced or have not been evicted
135 * are cached in the L1ARC. The L1ARC (l1arc_buf_hdr_t) is a structure within
136 * the arc_buf_hdr_t that will point to the data block in memory. A block can
137 * only be read by a consumer if it has an l1arc_buf_hdr_t. The L1ARC
138 * caches data in two ways -- in a list of ARC buffers (arc_buf_t) and
139 * also in the arc_buf_hdr_t's private physical data block pointer (b_pabd).
141 * The L1ARC's data pointer may or may not be uncompressed. The ARC has the
142 * ability to store the physical data (b_pabd) associated with the DVA of the
143 * arc_buf_hdr_t. Since the b_pabd is a copy of the on-disk physical block,
144 * it will match its on-disk compression characteristics. This behavior can be
145 * disabled by setting 'zfs_compressed_arc_enabled' to B_FALSE. When the
146 * compressed ARC functionality is disabled, the b_pabd will point to an
147 * uncompressed version of the on-disk data.
149 * Data in the L1ARC is not accessed by consumers of the ARC directly. Each
150 * arc_buf_hdr_t can have multiple ARC buffers (arc_buf_t) which reference it.
151 * Each ARC buffer (arc_buf_t) is being actively accessed by a specific ARC
152 * consumer. The ARC will provide references to this data and will keep it
153 * cached until it is no longer in use. The ARC caches only the L1ARC's physical
154 * data block and will evict any arc_buf_t that is no longer referenced. The
155 * amount of memory consumed by the arc_buf_ts' data buffers can be seen via the
156 * "overhead_size" kstat.
158 * Depending on the consumer, an arc_buf_t can be requested in uncompressed or
159 * compressed form. The typical case is that consumers will want uncompressed
160 * data, and when that happens a new data buffer is allocated where the data is
161 * decompressed for them to use. Currently the only consumer who wants
162 * compressed arc_buf_t's is "zfs send", when it streams data exactly as it
163 * exists on disk. When this happens, the arc_buf_t's data buffer is shared
164 * with the arc_buf_hdr_t.
166 * Here is a diagram showing an arc_buf_hdr_t referenced by two arc_buf_t's. The
167 * first one is owned by a compressed send consumer (and therefore references
168 * the same compressed data buffer as the arc_buf_hdr_t) and the second could be
169 * used by any other consumer (and has its own uncompressed copy of the data
184 * | b_buf +------------>+-----------+ arc_buf_t
185 * | b_pabd +-+ |b_next +---->+-----------+
186 * +-----------+ | |-----------| |b_next +-->NULL
187 * | |b_comp = T | +-----------+
188 * | |b_data +-+ |b_comp = F |
189 * | +-----------+ | |b_data +-+
190 * +->+------+ | +-----------+ |
192 * data | |<--------------+ | uncompressed
193 * +------+ compressed, | data
194 * shared +-->+------+
199 * When a consumer reads a block, the ARC must first look to see if the
200 * arc_buf_hdr_t is cached. If the hdr is cached then the ARC allocates a new
201 * arc_buf_t and either copies uncompressed data into a new data buffer from an
202 * existing uncompressed arc_buf_t, decompresses the hdr's b_pabd buffer into a
203 * new data buffer, or shares the hdr's b_pabd buffer, depending on whether the
204 * hdr is compressed and the desired compression characteristics of the
205 * arc_buf_t consumer. If the arc_buf_t ends up sharing data with the
206 * arc_buf_hdr_t and both of them are uncompressed then the arc_buf_t must be
207 * the last buffer in the hdr's b_buf list, however a shared compressed buf can
208 * be anywhere in the hdr's list.
210 * The diagram below shows an example of an uncompressed ARC hdr that is
211 * sharing its data with an arc_buf_t (note that the shared uncompressed buf is
212 * the last element in the buf list):
224 * | | arc_buf_t (shared)
225 * | b_buf +------------>+---------+ arc_buf_t
226 * | | |b_next +---->+---------+
227 * | b_pabd +-+ |---------| |b_next +-->NULL
228 * +-----------+ | | | +---------+
230 * | +---------+ | |b_data +-+
231 * +->+------+ | +---------+ |
233 * uncompressed | | | |
236 * | uncompressed | | |
239 * +---------------------------------+
241 * Writing to the ARC requires that the ARC first discard the hdr's b_pabd
242 * since the physical block is about to be rewritten. The new data contents
243 * will be contained in the arc_buf_t. As the I/O pipeline performs the write,
244 * it may compress the data before writing it to disk. The ARC will be called
245 * with the transformed data and will bcopy the transformed on-disk block into
246 * a newly allocated b_pabd. Writes are always done into buffers which have
247 * either been loaned (and hence are new and don't have other readers) or
248 * buffers which have been released (and hence have their own hdr, if there
249 * were originally other readers of the buf's original hdr). This ensures that
250 * the ARC only needs to update a single buf and its hdr after a write occurs.
252 * When the L2ARC is in use, it will also take advantage of the b_pabd. The
253 * L2ARC will always write the contents of b_pabd to the L2ARC. This means
254 * that when compressed ARC is enabled that the L2ARC blocks are identical
255 * to the on-disk block in the main data pool. This provides a significant
256 * advantage since the ARC can leverage the bp's checksum when reading from the
257 * L2ARC to determine if the contents are valid. However, if the compressed
258 * ARC is disabled, then the L2ARC's block must be transformed to look
259 * like the physical block in the main data pool before comparing the
260 * checksum and determining its validity.
265 #include <sys/spa_impl.h>
266 #include <sys/zio_compress.h>
267 #include <sys/zio_checksum.h>
268 #include <sys/zfs_context.h>
270 #include <sys/refcount.h>
271 #include <sys/vdev.h>
272 #include <sys/vdev_impl.h>
273 #include <sys/dsl_pool.h>
274 #include <sys/zio_checksum.h>
275 #include <sys/multilist.h>
278 #include <sys/vmsystm.h>
280 #include <sys/fs/swapnode.h>
282 #include <linux/mm_compat.h>
284 #include <sys/callb.h>
285 #include <sys/kstat.h>
286 #include <sys/dmu_tx.h>
287 #include <zfs_fletcher.h>
288 #include <sys/arc_impl.h>
289 #include <sys/trace_arc.h>
292 /* set with ZFS_DEBUG=watch, to enable watchpoints on frozen buffers */
293 boolean_t arc_watch
= B_FALSE
;
296 static kmutex_t arc_reclaim_lock
;
297 static kcondvar_t arc_reclaim_thread_cv
;
298 static boolean_t arc_reclaim_thread_exit
;
299 static kcondvar_t arc_reclaim_waiters_cv
;
302 * The number of headers to evict in arc_evict_state_impl() before
303 * dropping the sublist lock and evicting from another sublist. A lower
304 * value means we're more likely to evict the "correct" header (i.e. the
305 * oldest header in the arc state), but comes with higher overhead
306 * (i.e. more invocations of arc_evict_state_impl()).
308 int zfs_arc_evict_batch_limit
= 10;
310 /* number of seconds before growing cache again */
311 static int arc_grow_retry
= 5;
313 /* shift of arc_c for calculating overflow limit in arc_get_data_impl */
314 int zfs_arc_overflow_shift
= 8;
316 /* shift of arc_c for calculating both min and max arc_p */
317 static int arc_p_min_shift
= 4;
319 /* log2(fraction of arc to reclaim) */
320 static int arc_shrink_shift
= 7;
323 * log2(fraction of ARC which must be free to allow growing).
324 * I.e. If there is less than arc_c >> arc_no_grow_shift free memory,
325 * when reading a new block into the ARC, we will evict an equal-sized block
328 * This must be less than arc_shrink_shift, so that when we shrink the ARC,
329 * we will still not allow it to grow.
331 int arc_no_grow_shift
= 5;
335 * minimum lifespan of a prefetch block in clock ticks
336 * (initialized in arc_init())
338 static int arc_min_prefetch_lifespan
;
341 * If this percent of memory is free, don't throttle.
343 int arc_lotsfree_percent
= 10;
348 * The arc has filled available memory and has now warmed up.
350 static boolean_t arc_warm
;
353 * log2 fraction of the zio arena to keep free.
355 int arc_zio_arena_free_shift
= 2;
358 * These tunables are for performance analysis.
360 unsigned long zfs_arc_max
= 0;
361 unsigned long zfs_arc_min
= 0;
362 unsigned long zfs_arc_meta_limit
= 0;
363 unsigned long zfs_arc_meta_min
= 0;
364 unsigned long zfs_arc_dnode_limit
= 0;
365 unsigned long zfs_arc_dnode_reduce_percent
= 10;
366 int zfs_arc_grow_retry
= 0;
367 int zfs_arc_shrink_shift
= 0;
368 int zfs_arc_p_min_shift
= 0;
369 int zfs_arc_average_blocksize
= 8 * 1024; /* 8KB */
371 int zfs_compressed_arc_enabled
= B_TRUE
;
374 * ARC will evict meta buffers that exceed arc_meta_limit. This
375 * tunable make arc_meta_limit adjustable for different workloads.
377 unsigned long zfs_arc_meta_limit_percent
= 75;
380 * Percentage that can be consumed by dnodes of ARC meta buffers.
382 unsigned long zfs_arc_dnode_limit_percent
= 10;
385 * These tunables are Linux specific
387 unsigned long zfs_arc_sys_free
= 0;
388 int zfs_arc_min_prefetch_lifespan
= 0;
389 int zfs_arc_p_aggressive_disable
= 1;
390 int zfs_arc_p_dampener_disable
= 1;
391 int zfs_arc_meta_prune
= 10000;
392 int zfs_arc_meta_strategy
= ARC_STRATEGY_META_BALANCED
;
393 int zfs_arc_meta_adjust_restarts
= 4096;
394 int zfs_arc_lotsfree_percent
= 10;
397 static arc_state_t ARC_anon
;
398 static arc_state_t ARC_mru
;
399 static arc_state_t ARC_mru_ghost
;
400 static arc_state_t ARC_mfu
;
401 static arc_state_t ARC_mfu_ghost
;
402 static arc_state_t ARC_l2c_only
;
404 typedef struct arc_stats
{
405 kstat_named_t arcstat_hits
;
406 kstat_named_t arcstat_misses
;
407 kstat_named_t arcstat_demand_data_hits
;
408 kstat_named_t arcstat_demand_data_misses
;
409 kstat_named_t arcstat_demand_metadata_hits
;
410 kstat_named_t arcstat_demand_metadata_misses
;
411 kstat_named_t arcstat_prefetch_data_hits
;
412 kstat_named_t arcstat_prefetch_data_misses
;
413 kstat_named_t arcstat_prefetch_metadata_hits
;
414 kstat_named_t arcstat_prefetch_metadata_misses
;
415 kstat_named_t arcstat_mru_hits
;
416 kstat_named_t arcstat_mru_ghost_hits
;
417 kstat_named_t arcstat_mfu_hits
;
418 kstat_named_t arcstat_mfu_ghost_hits
;
419 kstat_named_t arcstat_deleted
;
421 * Number of buffers that could not be evicted because the hash lock
422 * was held by another thread. The lock may not necessarily be held
423 * by something using the same buffer, since hash locks are shared
424 * by multiple buffers.
426 kstat_named_t arcstat_mutex_miss
;
428 * Number of buffers skipped because they have I/O in progress, are
429 * indrect prefetch buffers that have not lived long enough, or are
430 * not from the spa we're trying to evict from.
432 kstat_named_t arcstat_evict_skip
;
434 * Number of times arc_evict_state() was unable to evict enough
435 * buffers to reach its target amount.
437 kstat_named_t arcstat_evict_not_enough
;
438 kstat_named_t arcstat_evict_l2_cached
;
439 kstat_named_t arcstat_evict_l2_eligible
;
440 kstat_named_t arcstat_evict_l2_ineligible
;
441 kstat_named_t arcstat_evict_l2_skip
;
442 kstat_named_t arcstat_hash_elements
;
443 kstat_named_t arcstat_hash_elements_max
;
444 kstat_named_t arcstat_hash_collisions
;
445 kstat_named_t arcstat_hash_chains
;
446 kstat_named_t arcstat_hash_chain_max
;
447 kstat_named_t arcstat_p
;
448 kstat_named_t arcstat_c
;
449 kstat_named_t arcstat_c_min
;
450 kstat_named_t arcstat_c_max
;
451 kstat_named_t arcstat_size
;
453 * Number of compressed bytes stored in the arc_buf_hdr_t's b_pabd.
454 * Note that the compressed bytes may match the uncompressed bytes
455 * if the block is either not compressed or compressed arc is disabled.
457 kstat_named_t arcstat_compressed_size
;
459 * Uncompressed size of the data stored in b_pabd. If compressed
460 * arc is disabled then this value will be identical to the stat
463 kstat_named_t arcstat_uncompressed_size
;
465 * Number of bytes stored in all the arc_buf_t's. This is classified
466 * as "overhead" since this data is typically short-lived and will
467 * be evicted from the arc when it becomes unreferenced unless the
468 * zfs_keep_uncompressed_metadata or zfs_keep_uncompressed_level
469 * values have been set (see comment in dbuf.c for more information).
471 kstat_named_t arcstat_overhead_size
;
473 * Number of bytes consumed by internal ARC structures necessary
474 * for tracking purposes; these structures are not actually
475 * backed by ARC buffers. This includes arc_buf_hdr_t structures
476 * (allocated via arc_buf_hdr_t_full and arc_buf_hdr_t_l2only
477 * caches), and arc_buf_t structures (allocated via arc_buf_t
480 kstat_named_t arcstat_hdr_size
;
482 * Number of bytes consumed by ARC buffers of type equal to
483 * ARC_BUFC_DATA. This is generally consumed by buffers backing
484 * on disk user data (e.g. plain file contents).
486 kstat_named_t arcstat_data_size
;
488 * Number of bytes consumed by ARC buffers of type equal to
489 * ARC_BUFC_METADATA. This is generally consumed by buffers
490 * backing on disk data that is used for internal ZFS
491 * structures (e.g. ZAP, dnode, indirect blocks, etc).
493 kstat_named_t arcstat_metadata_size
;
495 * Number of bytes consumed by dmu_buf_impl_t objects.
497 kstat_named_t arcstat_dbuf_size
;
499 * Number of bytes consumed by dnode_t objects.
501 kstat_named_t arcstat_dnode_size
;
503 * Number of bytes consumed by bonus buffers.
505 kstat_named_t arcstat_bonus_size
;
507 * Total number of bytes consumed by ARC buffers residing in the
508 * arc_anon state. This includes *all* buffers in the arc_anon
509 * state; e.g. data, metadata, evictable, and unevictable buffers
510 * are all included in this value.
512 kstat_named_t arcstat_anon_size
;
514 * Number of bytes consumed by ARC buffers that meet the
515 * following criteria: backing buffers of type ARC_BUFC_DATA,
516 * residing in the arc_anon state, and are eligible for eviction
517 * (e.g. have no outstanding holds on the buffer).
519 kstat_named_t arcstat_anon_evictable_data
;
521 * Number of bytes consumed by ARC buffers that meet the
522 * following criteria: backing buffers of type ARC_BUFC_METADATA,
523 * residing in the arc_anon state, and are eligible for eviction
524 * (e.g. have no outstanding holds on the buffer).
526 kstat_named_t arcstat_anon_evictable_metadata
;
528 * Total number of bytes consumed by ARC buffers residing in the
529 * arc_mru state. This includes *all* buffers in the arc_mru
530 * state; e.g. data, metadata, evictable, and unevictable buffers
531 * are all included in this value.
533 kstat_named_t arcstat_mru_size
;
535 * Number of bytes consumed by ARC buffers that meet the
536 * following criteria: backing buffers of type ARC_BUFC_DATA,
537 * residing in the arc_mru state, and are eligible for eviction
538 * (e.g. have no outstanding holds on the buffer).
540 kstat_named_t arcstat_mru_evictable_data
;
542 * Number of bytes consumed by ARC buffers that meet the
543 * following criteria: backing buffers of type ARC_BUFC_METADATA,
544 * residing in the arc_mru state, and are eligible for eviction
545 * (e.g. have no outstanding holds on the buffer).
547 kstat_named_t arcstat_mru_evictable_metadata
;
549 * Total number of bytes that *would have been* consumed by ARC
550 * buffers in the arc_mru_ghost state. The key thing to note
551 * here, is the fact that this size doesn't actually indicate
552 * RAM consumption. The ghost lists only consist of headers and
553 * don't actually have ARC buffers linked off of these headers.
554 * Thus, *if* the headers had associated ARC buffers, these
555 * buffers *would have* consumed this number of bytes.
557 kstat_named_t arcstat_mru_ghost_size
;
559 * Number of bytes that *would have been* consumed by ARC
560 * buffers that are eligible for eviction, of type
561 * ARC_BUFC_DATA, and linked off the arc_mru_ghost state.
563 kstat_named_t arcstat_mru_ghost_evictable_data
;
565 * Number of bytes that *would have been* consumed by ARC
566 * buffers that are eligible for eviction, of type
567 * ARC_BUFC_METADATA, and linked off the arc_mru_ghost state.
569 kstat_named_t arcstat_mru_ghost_evictable_metadata
;
571 * Total number of bytes consumed by ARC buffers residing in the
572 * arc_mfu state. This includes *all* buffers in the arc_mfu
573 * state; e.g. data, metadata, evictable, and unevictable buffers
574 * are all included in this value.
576 kstat_named_t arcstat_mfu_size
;
578 * Number of bytes consumed by ARC buffers that are eligible for
579 * eviction, of type ARC_BUFC_DATA, and reside in the arc_mfu
582 kstat_named_t arcstat_mfu_evictable_data
;
584 * Number of bytes consumed by ARC buffers that are eligible for
585 * eviction, of type ARC_BUFC_METADATA, and reside in the
588 kstat_named_t arcstat_mfu_evictable_metadata
;
590 * Total number of bytes that *would have been* consumed by ARC
591 * buffers in the arc_mfu_ghost state. See the comment above
592 * arcstat_mru_ghost_size for more details.
594 kstat_named_t arcstat_mfu_ghost_size
;
596 * Number of bytes that *would have been* consumed by ARC
597 * buffers that are eligible for eviction, of type
598 * ARC_BUFC_DATA, and linked off the arc_mfu_ghost state.
600 kstat_named_t arcstat_mfu_ghost_evictable_data
;
602 * Number of bytes that *would have been* consumed by ARC
603 * buffers that are eligible for eviction, of type
604 * ARC_BUFC_METADATA, and linked off the arc_mru_ghost state.
606 kstat_named_t arcstat_mfu_ghost_evictable_metadata
;
607 kstat_named_t arcstat_l2_hits
;
608 kstat_named_t arcstat_l2_misses
;
609 kstat_named_t arcstat_l2_feeds
;
610 kstat_named_t arcstat_l2_rw_clash
;
611 kstat_named_t arcstat_l2_read_bytes
;
612 kstat_named_t arcstat_l2_write_bytes
;
613 kstat_named_t arcstat_l2_writes_sent
;
614 kstat_named_t arcstat_l2_writes_done
;
615 kstat_named_t arcstat_l2_writes_error
;
616 kstat_named_t arcstat_l2_writes_lock_retry
;
617 kstat_named_t arcstat_l2_evict_lock_retry
;
618 kstat_named_t arcstat_l2_evict_reading
;
619 kstat_named_t arcstat_l2_evict_l1cached
;
620 kstat_named_t arcstat_l2_free_on_write
;
621 kstat_named_t arcstat_l2_abort_lowmem
;
622 kstat_named_t arcstat_l2_cksum_bad
;
623 kstat_named_t arcstat_l2_io_error
;
624 kstat_named_t arcstat_l2_size
;
625 kstat_named_t arcstat_l2_asize
;
626 kstat_named_t arcstat_l2_hdr_size
;
627 kstat_named_t arcstat_memory_throttle_count
;
628 kstat_named_t arcstat_memory_direct_count
;
629 kstat_named_t arcstat_memory_indirect_count
;
630 kstat_named_t arcstat_no_grow
;
631 kstat_named_t arcstat_tempreserve
;
632 kstat_named_t arcstat_loaned_bytes
;
633 kstat_named_t arcstat_prune
;
634 kstat_named_t arcstat_meta_used
;
635 kstat_named_t arcstat_meta_limit
;
636 kstat_named_t arcstat_dnode_limit
;
637 kstat_named_t arcstat_meta_max
;
638 kstat_named_t arcstat_meta_min
;
639 kstat_named_t arcstat_sync_wait_for_async
;
640 kstat_named_t arcstat_demand_hit_predictive_prefetch
;
641 kstat_named_t arcstat_need_free
;
642 kstat_named_t arcstat_sys_free
;
645 static arc_stats_t arc_stats
= {
646 { "hits", KSTAT_DATA_UINT64
},
647 { "misses", KSTAT_DATA_UINT64
},
648 { "demand_data_hits", KSTAT_DATA_UINT64
},
649 { "demand_data_misses", KSTAT_DATA_UINT64
},
650 { "demand_metadata_hits", KSTAT_DATA_UINT64
},
651 { "demand_metadata_misses", KSTAT_DATA_UINT64
},
652 { "prefetch_data_hits", KSTAT_DATA_UINT64
},
653 { "prefetch_data_misses", KSTAT_DATA_UINT64
},
654 { "prefetch_metadata_hits", KSTAT_DATA_UINT64
},
655 { "prefetch_metadata_misses", KSTAT_DATA_UINT64
},
656 { "mru_hits", KSTAT_DATA_UINT64
},
657 { "mru_ghost_hits", KSTAT_DATA_UINT64
},
658 { "mfu_hits", KSTAT_DATA_UINT64
},
659 { "mfu_ghost_hits", KSTAT_DATA_UINT64
},
660 { "deleted", KSTAT_DATA_UINT64
},
661 { "mutex_miss", KSTAT_DATA_UINT64
},
662 { "evict_skip", KSTAT_DATA_UINT64
},
663 { "evict_not_enough", KSTAT_DATA_UINT64
},
664 { "evict_l2_cached", KSTAT_DATA_UINT64
},
665 { "evict_l2_eligible", KSTAT_DATA_UINT64
},
666 { "evict_l2_ineligible", KSTAT_DATA_UINT64
},
667 { "evict_l2_skip", KSTAT_DATA_UINT64
},
668 { "hash_elements", KSTAT_DATA_UINT64
},
669 { "hash_elements_max", KSTAT_DATA_UINT64
},
670 { "hash_collisions", KSTAT_DATA_UINT64
},
671 { "hash_chains", KSTAT_DATA_UINT64
},
672 { "hash_chain_max", KSTAT_DATA_UINT64
},
673 { "p", KSTAT_DATA_UINT64
},
674 { "c", KSTAT_DATA_UINT64
},
675 { "c_min", KSTAT_DATA_UINT64
},
676 { "c_max", KSTAT_DATA_UINT64
},
677 { "size", KSTAT_DATA_UINT64
},
678 { "compressed_size", KSTAT_DATA_UINT64
},
679 { "uncompressed_size", KSTAT_DATA_UINT64
},
680 { "overhead_size", KSTAT_DATA_UINT64
},
681 { "hdr_size", KSTAT_DATA_UINT64
},
682 { "data_size", KSTAT_DATA_UINT64
},
683 { "metadata_size", KSTAT_DATA_UINT64
},
684 { "dbuf_size", KSTAT_DATA_UINT64
},
685 { "dnode_size", KSTAT_DATA_UINT64
},
686 { "bonus_size", KSTAT_DATA_UINT64
},
687 { "anon_size", KSTAT_DATA_UINT64
},
688 { "anon_evictable_data", KSTAT_DATA_UINT64
},
689 { "anon_evictable_metadata", KSTAT_DATA_UINT64
},
690 { "mru_size", KSTAT_DATA_UINT64
},
691 { "mru_evictable_data", KSTAT_DATA_UINT64
},
692 { "mru_evictable_metadata", KSTAT_DATA_UINT64
},
693 { "mru_ghost_size", KSTAT_DATA_UINT64
},
694 { "mru_ghost_evictable_data", KSTAT_DATA_UINT64
},
695 { "mru_ghost_evictable_metadata", KSTAT_DATA_UINT64
},
696 { "mfu_size", KSTAT_DATA_UINT64
},
697 { "mfu_evictable_data", KSTAT_DATA_UINT64
},
698 { "mfu_evictable_metadata", KSTAT_DATA_UINT64
},
699 { "mfu_ghost_size", KSTAT_DATA_UINT64
},
700 { "mfu_ghost_evictable_data", KSTAT_DATA_UINT64
},
701 { "mfu_ghost_evictable_metadata", KSTAT_DATA_UINT64
},
702 { "l2_hits", KSTAT_DATA_UINT64
},
703 { "l2_misses", KSTAT_DATA_UINT64
},
704 { "l2_feeds", KSTAT_DATA_UINT64
},
705 { "l2_rw_clash", KSTAT_DATA_UINT64
},
706 { "l2_read_bytes", KSTAT_DATA_UINT64
},
707 { "l2_write_bytes", KSTAT_DATA_UINT64
},
708 { "l2_writes_sent", KSTAT_DATA_UINT64
},
709 { "l2_writes_done", KSTAT_DATA_UINT64
},
710 { "l2_writes_error", KSTAT_DATA_UINT64
},
711 { "l2_writes_lock_retry", KSTAT_DATA_UINT64
},
712 { "l2_evict_lock_retry", KSTAT_DATA_UINT64
},
713 { "l2_evict_reading", KSTAT_DATA_UINT64
},
714 { "l2_evict_l1cached", KSTAT_DATA_UINT64
},
715 { "l2_free_on_write", KSTAT_DATA_UINT64
},
716 { "l2_abort_lowmem", KSTAT_DATA_UINT64
},
717 { "l2_cksum_bad", KSTAT_DATA_UINT64
},
718 { "l2_io_error", KSTAT_DATA_UINT64
},
719 { "l2_size", KSTAT_DATA_UINT64
},
720 { "l2_asize", KSTAT_DATA_UINT64
},
721 { "l2_hdr_size", KSTAT_DATA_UINT64
},
722 { "memory_throttle_count", KSTAT_DATA_UINT64
},
723 { "memory_direct_count", KSTAT_DATA_UINT64
},
724 { "memory_indirect_count", KSTAT_DATA_UINT64
},
725 { "arc_no_grow", KSTAT_DATA_UINT64
},
726 { "arc_tempreserve", KSTAT_DATA_UINT64
},
727 { "arc_loaned_bytes", KSTAT_DATA_UINT64
},
728 { "arc_prune", KSTAT_DATA_UINT64
},
729 { "arc_meta_used", KSTAT_DATA_UINT64
},
730 { "arc_meta_limit", KSTAT_DATA_UINT64
},
731 { "arc_dnode_limit", KSTAT_DATA_UINT64
},
732 { "arc_meta_max", KSTAT_DATA_UINT64
},
733 { "arc_meta_min", KSTAT_DATA_UINT64
},
734 { "sync_wait_for_async", KSTAT_DATA_UINT64
},
735 { "demand_hit_predictive_prefetch", KSTAT_DATA_UINT64
},
736 { "arc_need_free", KSTAT_DATA_UINT64
},
737 { "arc_sys_free", KSTAT_DATA_UINT64
}
740 #define ARCSTAT(stat) (arc_stats.stat.value.ui64)
742 #define ARCSTAT_INCR(stat, val) \
743 atomic_add_64(&arc_stats.stat.value.ui64, (val))
745 #define ARCSTAT_BUMP(stat) ARCSTAT_INCR(stat, 1)
746 #define ARCSTAT_BUMPDOWN(stat) ARCSTAT_INCR(stat, -1)
748 #define ARCSTAT_MAX(stat, val) { \
750 while ((val) > (m = arc_stats.stat.value.ui64) && \
751 (m != atomic_cas_64(&arc_stats.stat.value.ui64, m, (val)))) \
755 #define ARCSTAT_MAXSTAT(stat) \
756 ARCSTAT_MAX(stat##_max, arc_stats.stat.value.ui64)
759 * We define a macro to allow ARC hits/misses to be easily broken down by
760 * two separate conditions, giving a total of four different subtypes for
761 * each of hits and misses (so eight statistics total).
763 #define ARCSTAT_CONDSTAT(cond1, stat1, notstat1, cond2, stat2, notstat2, stat) \
766 ARCSTAT_BUMP(arcstat_##stat1##_##stat2##_##stat); \
768 ARCSTAT_BUMP(arcstat_##stat1##_##notstat2##_##stat); \
772 ARCSTAT_BUMP(arcstat_##notstat1##_##stat2##_##stat); \
774 ARCSTAT_BUMP(arcstat_##notstat1##_##notstat2##_##stat);\
779 static arc_state_t
*arc_anon
;
780 static arc_state_t
*arc_mru
;
781 static arc_state_t
*arc_mru_ghost
;
782 static arc_state_t
*arc_mfu
;
783 static arc_state_t
*arc_mfu_ghost
;
784 static arc_state_t
*arc_l2c_only
;
787 * There are several ARC variables that are critical to export as kstats --
788 * but we don't want to have to grovel around in the kstat whenever we wish to
789 * manipulate them. For these variables, we therefore define them to be in
790 * terms of the statistic variable. This assures that we are not introducing
791 * the possibility of inconsistency by having shadow copies of the variables,
792 * while still allowing the code to be readable.
794 #define arc_size ARCSTAT(arcstat_size) /* actual total arc size */
795 #define arc_p ARCSTAT(arcstat_p) /* target size of MRU */
796 #define arc_c ARCSTAT(arcstat_c) /* target size of cache */
797 #define arc_c_min ARCSTAT(arcstat_c_min) /* min target cache size */
798 #define arc_c_max ARCSTAT(arcstat_c_max) /* max target cache size */
799 #define arc_no_grow ARCSTAT(arcstat_no_grow) /* do not grow cache size */
800 #define arc_tempreserve ARCSTAT(arcstat_tempreserve)
801 #define arc_loaned_bytes ARCSTAT(arcstat_loaned_bytes)
802 #define arc_meta_limit ARCSTAT(arcstat_meta_limit) /* max size for metadata */
803 #define arc_dnode_limit ARCSTAT(arcstat_dnode_limit) /* max size for dnodes */
804 #define arc_meta_min ARCSTAT(arcstat_meta_min) /* min size for metadata */
805 #define arc_meta_used ARCSTAT(arcstat_meta_used) /* size of metadata */
806 #define arc_meta_max ARCSTAT(arcstat_meta_max) /* max size of metadata */
807 #define arc_dbuf_size ARCSTAT(arcstat_dbuf_size) /* dbuf metadata */
808 #define arc_dnode_size ARCSTAT(arcstat_dnode_size) /* dnode metadata */
809 #define arc_bonus_size ARCSTAT(arcstat_bonus_size) /* bonus buffer metadata */
810 #define arc_need_free ARCSTAT(arcstat_need_free) /* bytes to be freed */
811 #define arc_sys_free ARCSTAT(arcstat_sys_free) /* target system free bytes */
813 /* compressed size of entire arc */
814 #define arc_compressed_size ARCSTAT(arcstat_compressed_size)
815 /* uncompressed size of entire arc */
816 #define arc_uncompressed_size ARCSTAT(arcstat_uncompressed_size)
817 /* number of bytes in the arc from arc_buf_t's */
818 #define arc_overhead_size ARCSTAT(arcstat_overhead_size)
820 static list_t arc_prune_list
;
821 static kmutex_t arc_prune_mtx
;
822 static taskq_t
*arc_prune_taskq
;
824 #define GHOST_STATE(state) \
825 ((state) == arc_mru_ghost || (state) == arc_mfu_ghost || \
826 (state) == arc_l2c_only)
828 #define HDR_IN_HASH_TABLE(hdr) ((hdr)->b_flags & ARC_FLAG_IN_HASH_TABLE)
829 #define HDR_IO_IN_PROGRESS(hdr) ((hdr)->b_flags & ARC_FLAG_IO_IN_PROGRESS)
830 #define HDR_IO_ERROR(hdr) ((hdr)->b_flags & ARC_FLAG_IO_ERROR)
831 #define HDR_PREFETCH(hdr) ((hdr)->b_flags & ARC_FLAG_PREFETCH)
832 #define HDR_COMPRESSION_ENABLED(hdr) \
833 ((hdr)->b_flags & ARC_FLAG_COMPRESSED_ARC)
835 #define HDR_L2CACHE(hdr) ((hdr)->b_flags & ARC_FLAG_L2CACHE)
836 #define HDR_L2_READING(hdr) \
837 (((hdr)->b_flags & ARC_FLAG_IO_IN_PROGRESS) && \
838 ((hdr)->b_flags & ARC_FLAG_HAS_L2HDR))
839 #define HDR_L2_WRITING(hdr) ((hdr)->b_flags & ARC_FLAG_L2_WRITING)
840 #define HDR_L2_EVICTED(hdr) ((hdr)->b_flags & ARC_FLAG_L2_EVICTED)
841 #define HDR_L2_WRITE_HEAD(hdr) ((hdr)->b_flags & ARC_FLAG_L2_WRITE_HEAD)
842 #define HDR_SHARED_DATA(hdr) ((hdr)->b_flags & ARC_FLAG_SHARED_DATA)
844 #define HDR_ISTYPE_METADATA(hdr) \
845 ((hdr)->b_flags & ARC_FLAG_BUFC_METADATA)
846 #define HDR_ISTYPE_DATA(hdr) (!HDR_ISTYPE_METADATA(hdr))
848 #define HDR_HAS_L1HDR(hdr) ((hdr)->b_flags & ARC_FLAG_HAS_L1HDR)
849 #define HDR_HAS_L2HDR(hdr) ((hdr)->b_flags & ARC_FLAG_HAS_L2HDR)
851 /* For storing compression mode in b_flags */
852 #define HDR_COMPRESS_OFFSET (highbit64(ARC_FLAG_COMPRESS_0) - 1)
854 #define HDR_GET_COMPRESS(hdr) ((enum zio_compress)BF32_GET((hdr)->b_flags, \
855 HDR_COMPRESS_OFFSET, SPA_COMPRESSBITS))
856 #define HDR_SET_COMPRESS(hdr, cmp) BF32_SET((hdr)->b_flags, \
857 HDR_COMPRESS_OFFSET, SPA_COMPRESSBITS, (cmp));
859 #define ARC_BUF_LAST(buf) ((buf)->b_next == NULL)
860 #define ARC_BUF_SHARED(buf) ((buf)->b_flags & ARC_BUF_FLAG_SHARED)
861 #define ARC_BUF_COMPRESSED(buf) ((buf)->b_flags & ARC_BUF_FLAG_COMPRESSED)
867 #define HDR_FULL_SIZE ((int64_t)sizeof (arc_buf_hdr_t))
868 #define HDR_L2ONLY_SIZE ((int64_t)offsetof(arc_buf_hdr_t, b_l1hdr))
871 * Hash table routines
874 #define HT_LOCK_ALIGN 64
875 #define HT_LOCK_PAD (P2NPHASE(sizeof (kmutex_t), (HT_LOCK_ALIGN)))
880 unsigned char pad
[HT_LOCK_PAD
];
884 #define BUF_LOCKS 8192
885 typedef struct buf_hash_table
{
887 arc_buf_hdr_t
**ht_table
;
888 struct ht_lock ht_locks
[BUF_LOCKS
];
891 static buf_hash_table_t buf_hash_table
;
893 #define BUF_HASH_INDEX(spa, dva, birth) \
894 (buf_hash(spa, dva, birth) & buf_hash_table.ht_mask)
895 #define BUF_HASH_LOCK_NTRY(idx) (buf_hash_table.ht_locks[idx & (BUF_LOCKS-1)])
896 #define BUF_HASH_LOCK(idx) (&(BUF_HASH_LOCK_NTRY(idx).ht_lock))
897 #define HDR_LOCK(hdr) \
898 (BUF_HASH_LOCK(BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth)))
900 uint64_t zfs_crc64_table
[256];
906 #define L2ARC_WRITE_SIZE (8 * 1024 * 1024) /* initial write max */
907 #define L2ARC_HEADROOM 2 /* num of writes */
910 * If we discover during ARC scan any buffers to be compressed, we boost
911 * our headroom for the next scanning cycle by this percentage multiple.
913 #define L2ARC_HEADROOM_BOOST 200
914 #define L2ARC_FEED_SECS 1 /* caching interval secs */
915 #define L2ARC_FEED_MIN_MS 200 /* min caching interval ms */
918 * We can feed L2ARC from two states of ARC buffers, mru and mfu,
919 * and each of the state has two types: data and metadata.
921 #define L2ARC_FEED_TYPES 4
923 #define l2arc_writes_sent ARCSTAT(arcstat_l2_writes_sent)
924 #define l2arc_writes_done ARCSTAT(arcstat_l2_writes_done)
926 /* L2ARC Performance Tunables */
927 unsigned long l2arc_write_max
= L2ARC_WRITE_SIZE
; /* def max write size */
928 unsigned long l2arc_write_boost
= L2ARC_WRITE_SIZE
; /* extra warmup write */
929 unsigned long l2arc_headroom
= L2ARC_HEADROOM
; /* # of dev writes */
930 unsigned long l2arc_headroom_boost
= L2ARC_HEADROOM_BOOST
;
931 unsigned long l2arc_feed_secs
= L2ARC_FEED_SECS
; /* interval seconds */
932 unsigned long l2arc_feed_min_ms
= L2ARC_FEED_MIN_MS
; /* min interval msecs */
933 int l2arc_noprefetch
= B_TRUE
; /* don't cache prefetch bufs */
934 int l2arc_feed_again
= B_TRUE
; /* turbo warmup */
935 int l2arc_norw
= B_FALSE
; /* no reads during writes */
940 static list_t L2ARC_dev_list
; /* device list */
941 static list_t
*l2arc_dev_list
; /* device list pointer */
942 static kmutex_t l2arc_dev_mtx
; /* device list mutex */
943 static l2arc_dev_t
*l2arc_dev_last
; /* last device used */
944 static list_t L2ARC_free_on_write
; /* free after write buf list */
945 static list_t
*l2arc_free_on_write
; /* free after write list ptr */
946 static kmutex_t l2arc_free_on_write_mtx
; /* mutex for list */
947 static uint64_t l2arc_ndev
; /* number of devices */
949 typedef struct l2arc_read_callback
{
950 arc_buf_hdr_t
*l2rcb_hdr
; /* read header */
951 blkptr_t l2rcb_bp
; /* original blkptr */
952 zbookmark_phys_t l2rcb_zb
; /* original bookmark */
953 int l2rcb_flags
; /* original flags */
954 } l2arc_read_callback_t
;
956 typedef struct l2arc_data_free
{
957 /* protected by l2arc_free_on_write_mtx */
960 arc_buf_contents_t l2df_type
;
961 list_node_t l2df_list_node
;
964 static kmutex_t l2arc_feed_thr_lock
;
965 static kcondvar_t l2arc_feed_thr_cv
;
966 static uint8_t l2arc_thread_exit
;
968 static abd_t
*arc_get_data_abd(arc_buf_hdr_t
*, uint64_t, void *);
969 static void *arc_get_data_buf(arc_buf_hdr_t
*, uint64_t, void *);
970 static void arc_get_data_impl(arc_buf_hdr_t
*, uint64_t, void *);
971 static void arc_free_data_abd(arc_buf_hdr_t
*, abd_t
*, uint64_t, void *);
972 static void arc_free_data_buf(arc_buf_hdr_t
*, void *, uint64_t, void *);
973 static void arc_free_data_impl(arc_buf_hdr_t
*hdr
, uint64_t size
, void *tag
);
974 static void arc_hdr_free_pabd(arc_buf_hdr_t
*);
975 static void arc_hdr_alloc_pabd(arc_buf_hdr_t
*);
976 static void arc_access(arc_buf_hdr_t
*, kmutex_t
*);
977 static boolean_t
arc_is_overflowing(void);
978 static void arc_buf_watch(arc_buf_t
*);
979 static void arc_tuning_update(void);
980 static void arc_prune_async(int64_t);
981 static uint64_t arc_all_memory(void);
983 static arc_buf_contents_t
arc_buf_type(arc_buf_hdr_t
*);
984 static uint32_t arc_bufc_to_flags(arc_buf_contents_t
);
985 static inline void arc_hdr_set_flags(arc_buf_hdr_t
*hdr
, arc_flags_t flags
);
986 static inline void arc_hdr_clear_flags(arc_buf_hdr_t
*hdr
, arc_flags_t flags
);
988 static boolean_t
l2arc_write_eligible(uint64_t, arc_buf_hdr_t
*);
989 static void l2arc_read_done(zio_t
*);
992 buf_hash(uint64_t spa
, const dva_t
*dva
, uint64_t birth
)
994 uint8_t *vdva
= (uint8_t *)dva
;
995 uint64_t crc
= -1ULL;
998 ASSERT(zfs_crc64_table
[128] == ZFS_CRC64_POLY
);
1000 for (i
= 0; i
< sizeof (dva_t
); i
++)
1001 crc
= (crc
>> 8) ^ zfs_crc64_table
[(crc
^ vdva
[i
]) & 0xFF];
1003 crc
^= (spa
>>8) ^ birth
;
1008 #define HDR_EMPTY(hdr) \
1009 ((hdr)->b_dva.dva_word[0] == 0 && \
1010 (hdr)->b_dva.dva_word[1] == 0)
1012 #define HDR_EQUAL(spa, dva, birth, hdr) \
1013 ((hdr)->b_dva.dva_word[0] == (dva)->dva_word[0]) && \
1014 ((hdr)->b_dva.dva_word[1] == (dva)->dva_word[1]) && \
1015 ((hdr)->b_birth == birth) && ((hdr)->b_spa == spa)
1018 buf_discard_identity(arc_buf_hdr_t
*hdr
)
1020 hdr
->b_dva
.dva_word
[0] = 0;
1021 hdr
->b_dva
.dva_word
[1] = 0;
1025 static arc_buf_hdr_t
*
1026 buf_hash_find(uint64_t spa
, const blkptr_t
*bp
, kmutex_t
**lockp
)
1028 const dva_t
*dva
= BP_IDENTITY(bp
);
1029 uint64_t birth
= BP_PHYSICAL_BIRTH(bp
);
1030 uint64_t idx
= BUF_HASH_INDEX(spa
, dva
, birth
);
1031 kmutex_t
*hash_lock
= BUF_HASH_LOCK(idx
);
1034 mutex_enter(hash_lock
);
1035 for (hdr
= buf_hash_table
.ht_table
[idx
]; hdr
!= NULL
;
1036 hdr
= hdr
->b_hash_next
) {
1037 if (HDR_EQUAL(spa
, dva
, birth
, hdr
)) {
1042 mutex_exit(hash_lock
);
1048 * Insert an entry into the hash table. If there is already an element
1049 * equal to elem in the hash table, then the already existing element
1050 * will be returned and the new element will not be inserted.
1051 * Otherwise returns NULL.
1052 * If lockp == NULL, the caller is assumed to already hold the hash lock.
1054 static arc_buf_hdr_t
*
1055 buf_hash_insert(arc_buf_hdr_t
*hdr
, kmutex_t
**lockp
)
1057 uint64_t idx
= BUF_HASH_INDEX(hdr
->b_spa
, &hdr
->b_dva
, hdr
->b_birth
);
1058 kmutex_t
*hash_lock
= BUF_HASH_LOCK(idx
);
1059 arc_buf_hdr_t
*fhdr
;
1062 ASSERT(!DVA_IS_EMPTY(&hdr
->b_dva
));
1063 ASSERT(hdr
->b_birth
!= 0);
1064 ASSERT(!HDR_IN_HASH_TABLE(hdr
));
1066 if (lockp
!= NULL
) {
1068 mutex_enter(hash_lock
);
1070 ASSERT(MUTEX_HELD(hash_lock
));
1073 for (fhdr
= buf_hash_table
.ht_table
[idx
], i
= 0; fhdr
!= NULL
;
1074 fhdr
= fhdr
->b_hash_next
, i
++) {
1075 if (HDR_EQUAL(hdr
->b_spa
, &hdr
->b_dva
, hdr
->b_birth
, fhdr
))
1079 hdr
->b_hash_next
= buf_hash_table
.ht_table
[idx
];
1080 buf_hash_table
.ht_table
[idx
] = hdr
;
1081 arc_hdr_set_flags(hdr
, ARC_FLAG_IN_HASH_TABLE
);
1083 /* collect some hash table performance data */
1085 ARCSTAT_BUMP(arcstat_hash_collisions
);
1087 ARCSTAT_BUMP(arcstat_hash_chains
);
1089 ARCSTAT_MAX(arcstat_hash_chain_max
, i
);
1092 ARCSTAT_BUMP(arcstat_hash_elements
);
1093 ARCSTAT_MAXSTAT(arcstat_hash_elements
);
1099 buf_hash_remove(arc_buf_hdr_t
*hdr
)
1101 arc_buf_hdr_t
*fhdr
, **hdrp
;
1102 uint64_t idx
= BUF_HASH_INDEX(hdr
->b_spa
, &hdr
->b_dva
, hdr
->b_birth
);
1104 ASSERT(MUTEX_HELD(BUF_HASH_LOCK(idx
)));
1105 ASSERT(HDR_IN_HASH_TABLE(hdr
));
1107 hdrp
= &buf_hash_table
.ht_table
[idx
];
1108 while ((fhdr
= *hdrp
) != hdr
) {
1109 ASSERT3P(fhdr
, !=, NULL
);
1110 hdrp
= &fhdr
->b_hash_next
;
1112 *hdrp
= hdr
->b_hash_next
;
1113 hdr
->b_hash_next
= NULL
;
1114 arc_hdr_clear_flags(hdr
, ARC_FLAG_IN_HASH_TABLE
);
1116 /* collect some hash table performance data */
1117 ARCSTAT_BUMPDOWN(arcstat_hash_elements
);
1119 if (buf_hash_table
.ht_table
[idx
] &&
1120 buf_hash_table
.ht_table
[idx
]->b_hash_next
== NULL
)
1121 ARCSTAT_BUMPDOWN(arcstat_hash_chains
);
1125 * Global data structures and functions for the buf kmem cache.
1127 static kmem_cache_t
*hdr_full_cache
;
1128 static kmem_cache_t
*hdr_l2only_cache
;
1129 static kmem_cache_t
*buf_cache
;
1136 #if defined(_KERNEL) && defined(HAVE_SPL)
1138 * Large allocations which do not require contiguous pages
1139 * should be using vmem_free() in the linux kernel\
1141 vmem_free(buf_hash_table
.ht_table
,
1142 (buf_hash_table
.ht_mask
+ 1) * sizeof (void *));
1144 kmem_free(buf_hash_table
.ht_table
,
1145 (buf_hash_table
.ht_mask
+ 1) * sizeof (void *));
1147 for (i
= 0; i
< BUF_LOCKS
; i
++)
1148 mutex_destroy(&buf_hash_table
.ht_locks
[i
].ht_lock
);
1149 kmem_cache_destroy(hdr_full_cache
);
1150 kmem_cache_destroy(hdr_l2only_cache
);
1151 kmem_cache_destroy(buf_cache
);
1155 * Constructor callback - called when the cache is empty
1156 * and a new buf is requested.
1160 hdr_full_cons(void *vbuf
, void *unused
, int kmflag
)
1162 arc_buf_hdr_t
*hdr
= vbuf
;
1164 bzero(hdr
, HDR_FULL_SIZE
);
1165 cv_init(&hdr
->b_l1hdr
.b_cv
, NULL
, CV_DEFAULT
, NULL
);
1166 refcount_create(&hdr
->b_l1hdr
.b_refcnt
);
1167 mutex_init(&hdr
->b_l1hdr
.b_freeze_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1168 list_link_init(&hdr
->b_l1hdr
.b_arc_node
);
1169 list_link_init(&hdr
->b_l2hdr
.b_l2node
);
1170 multilist_link_init(&hdr
->b_l1hdr
.b_arc_node
);
1171 arc_space_consume(HDR_FULL_SIZE
, ARC_SPACE_HDRS
);
1178 hdr_l2only_cons(void *vbuf
, void *unused
, int kmflag
)
1180 arc_buf_hdr_t
*hdr
= vbuf
;
1182 bzero(hdr
, HDR_L2ONLY_SIZE
);
1183 arc_space_consume(HDR_L2ONLY_SIZE
, ARC_SPACE_L2HDRS
);
1190 buf_cons(void *vbuf
, void *unused
, int kmflag
)
1192 arc_buf_t
*buf
= vbuf
;
1194 bzero(buf
, sizeof (arc_buf_t
));
1195 mutex_init(&buf
->b_evict_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1196 arc_space_consume(sizeof (arc_buf_t
), ARC_SPACE_HDRS
);
1202 * Destructor callback - called when a cached buf is
1203 * no longer required.
1207 hdr_full_dest(void *vbuf
, void *unused
)
1209 arc_buf_hdr_t
*hdr
= vbuf
;
1211 ASSERT(HDR_EMPTY(hdr
));
1212 cv_destroy(&hdr
->b_l1hdr
.b_cv
);
1213 refcount_destroy(&hdr
->b_l1hdr
.b_refcnt
);
1214 mutex_destroy(&hdr
->b_l1hdr
.b_freeze_lock
);
1215 ASSERT(!multilist_link_active(&hdr
->b_l1hdr
.b_arc_node
));
1216 arc_space_return(HDR_FULL_SIZE
, ARC_SPACE_HDRS
);
1221 hdr_l2only_dest(void *vbuf
, void *unused
)
1223 ASSERTV(arc_buf_hdr_t
*hdr
= vbuf
);
1225 ASSERT(HDR_EMPTY(hdr
));
1226 arc_space_return(HDR_L2ONLY_SIZE
, ARC_SPACE_L2HDRS
);
1231 buf_dest(void *vbuf
, void *unused
)
1233 arc_buf_t
*buf
= vbuf
;
1235 mutex_destroy(&buf
->b_evict_lock
);
1236 arc_space_return(sizeof (arc_buf_t
), ARC_SPACE_HDRS
);
1240 * Reclaim callback -- invoked when memory is low.
1244 hdr_recl(void *unused
)
1246 dprintf("hdr_recl called\n");
1248 * umem calls the reclaim func when we destroy the buf cache,
1249 * which is after we do arc_fini().
1252 cv_signal(&arc_reclaim_thread_cv
);
1258 uint64_t *ct
= NULL
;
1259 uint64_t hsize
= 1ULL << 12;
1263 * The hash table is big enough to fill all of physical memory
1264 * with an average block size of zfs_arc_average_blocksize (default 8K).
1265 * By default, the table will take up
1266 * totalmem * sizeof(void*) / 8K (1MB per GB with 8-byte pointers).
1268 while (hsize
* zfs_arc_average_blocksize
< arc_all_memory())
1271 buf_hash_table
.ht_mask
= hsize
- 1;
1272 #if defined(_KERNEL) && defined(HAVE_SPL)
1274 * Large allocations which do not require contiguous pages
1275 * should be using vmem_alloc() in the linux kernel
1277 buf_hash_table
.ht_table
=
1278 vmem_zalloc(hsize
* sizeof (void*), KM_SLEEP
);
1280 buf_hash_table
.ht_table
=
1281 kmem_zalloc(hsize
* sizeof (void*), KM_NOSLEEP
);
1283 if (buf_hash_table
.ht_table
== NULL
) {
1284 ASSERT(hsize
> (1ULL << 8));
1289 hdr_full_cache
= kmem_cache_create("arc_buf_hdr_t_full", HDR_FULL_SIZE
,
1290 0, hdr_full_cons
, hdr_full_dest
, hdr_recl
, NULL
, NULL
, 0);
1291 hdr_l2only_cache
= kmem_cache_create("arc_buf_hdr_t_l2only",
1292 HDR_L2ONLY_SIZE
, 0, hdr_l2only_cons
, hdr_l2only_dest
, hdr_recl
,
1294 buf_cache
= kmem_cache_create("arc_buf_t", sizeof (arc_buf_t
),
1295 0, buf_cons
, buf_dest
, NULL
, NULL
, NULL
, 0);
1297 for (i
= 0; i
< 256; i
++)
1298 for (ct
= zfs_crc64_table
+ i
, *ct
= i
, j
= 8; j
> 0; j
--)
1299 *ct
= (*ct
>> 1) ^ (-(*ct
& 1) & ZFS_CRC64_POLY
);
1301 for (i
= 0; i
< BUF_LOCKS
; i
++) {
1302 mutex_init(&buf_hash_table
.ht_locks
[i
].ht_lock
,
1303 NULL
, MUTEX_DEFAULT
, NULL
);
1307 #define ARC_MINTIME (hz>>4) /* 62 ms */
1310 * This is the size that the buf occupies in memory. If the buf is compressed,
1311 * it will correspond to the compressed size. You should use this method of
1312 * getting the buf size unless you explicitly need the logical size.
1315 arc_buf_size(arc_buf_t
*buf
)
1317 return (ARC_BUF_COMPRESSED(buf
) ?
1318 HDR_GET_PSIZE(buf
->b_hdr
) : HDR_GET_LSIZE(buf
->b_hdr
));
1322 arc_buf_lsize(arc_buf_t
*buf
)
1324 return (HDR_GET_LSIZE(buf
->b_hdr
));
1328 arc_get_compression(arc_buf_t
*buf
)
1330 return (ARC_BUF_COMPRESSED(buf
) ?
1331 HDR_GET_COMPRESS(buf
->b_hdr
) : ZIO_COMPRESS_OFF
);
1334 static inline boolean_t
1335 arc_buf_is_shared(arc_buf_t
*buf
)
1337 boolean_t shared
= (buf
->b_data
!= NULL
&&
1338 buf
->b_hdr
->b_l1hdr
.b_pabd
!= NULL
&&
1339 abd_is_linear(buf
->b_hdr
->b_l1hdr
.b_pabd
) &&
1340 buf
->b_data
== abd_to_buf(buf
->b_hdr
->b_l1hdr
.b_pabd
));
1341 IMPLY(shared
, HDR_SHARED_DATA(buf
->b_hdr
));
1342 IMPLY(shared
, ARC_BUF_SHARED(buf
));
1343 IMPLY(shared
, ARC_BUF_COMPRESSED(buf
) || ARC_BUF_LAST(buf
));
1346 * It would be nice to assert arc_can_share() too, but the "hdr isn't
1347 * already being shared" requirement prevents us from doing that.
1354 arc_cksum_free(arc_buf_hdr_t
*hdr
)
1356 ASSERT(HDR_HAS_L1HDR(hdr
));
1357 mutex_enter(&hdr
->b_l1hdr
.b_freeze_lock
);
1358 if (hdr
->b_l1hdr
.b_freeze_cksum
!= NULL
) {
1359 kmem_free(hdr
->b_l1hdr
.b_freeze_cksum
, sizeof (zio_cksum_t
));
1360 hdr
->b_l1hdr
.b_freeze_cksum
= NULL
;
1362 mutex_exit(&hdr
->b_l1hdr
.b_freeze_lock
);
1366 * If we've turned on the ZFS_DEBUG_MODIFY flag, verify that the buf's data
1367 * matches the checksum that is stored in the hdr. If there is no checksum,
1368 * or if the buf is compressed, this is a no-op.
1371 arc_cksum_verify(arc_buf_t
*buf
)
1373 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
1376 if (!(zfs_flags
& ZFS_DEBUG_MODIFY
))
1379 if (ARC_BUF_COMPRESSED(buf
)) {
1383 ASSERT(HDR_HAS_L1HDR(hdr
));
1385 mutex_enter(&hdr
->b_l1hdr
.b_freeze_lock
);
1386 if (hdr
->b_l1hdr
.b_freeze_cksum
== NULL
|| HDR_IO_ERROR(hdr
)) {
1387 mutex_exit(&hdr
->b_l1hdr
.b_freeze_lock
);
1391 fletcher_2_native(buf
->b_data
, arc_buf_size(buf
), NULL
, &zc
);
1392 if (!ZIO_CHECKSUM_EQUAL(*hdr
->b_l1hdr
.b_freeze_cksum
, zc
))
1393 panic("buffer modified while frozen!");
1394 mutex_exit(&hdr
->b_l1hdr
.b_freeze_lock
);
1398 arc_cksum_is_equal(arc_buf_hdr_t
*hdr
, zio_t
*zio
)
1400 enum zio_compress compress
= BP_GET_COMPRESS(zio
->io_bp
);
1401 boolean_t valid_cksum
;
1403 ASSERT(!BP_IS_EMBEDDED(zio
->io_bp
));
1404 VERIFY3U(BP_GET_PSIZE(zio
->io_bp
), ==, HDR_GET_PSIZE(hdr
));
1407 * We rely on the blkptr's checksum to determine if the block
1408 * is valid or not. When compressed arc is enabled, the l2arc
1409 * writes the block to the l2arc just as it appears in the pool.
1410 * This allows us to use the blkptr's checksum to validate the
1411 * data that we just read off of the l2arc without having to store
1412 * a separate checksum in the arc_buf_hdr_t. However, if compressed
1413 * arc is disabled, then the data written to the l2arc is always
1414 * uncompressed and won't match the block as it exists in the main
1415 * pool. When this is the case, we must first compress it if it is
1416 * compressed on the main pool before we can validate the checksum.
1418 if (!HDR_COMPRESSION_ENABLED(hdr
) && compress
!= ZIO_COMPRESS_OFF
) {
1422 ASSERT3U(HDR_GET_COMPRESS(hdr
), ==, ZIO_COMPRESS_OFF
);
1424 cbuf
= zio_buf_alloc(HDR_GET_PSIZE(hdr
));
1425 lsize
= HDR_GET_LSIZE(hdr
);
1426 csize
= zio_compress_data(compress
, zio
->io_abd
, cbuf
, lsize
);
1428 ASSERT3U(csize
, <=, HDR_GET_PSIZE(hdr
));
1429 if (csize
< HDR_GET_PSIZE(hdr
)) {
1431 * Compressed blocks are always a multiple of the
1432 * smallest ashift in the pool. Ideally, we would
1433 * like to round up the csize to the next
1434 * spa_min_ashift but that value may have changed
1435 * since the block was last written. Instead,
1436 * we rely on the fact that the hdr's psize
1437 * was set to the psize of the block when it was
1438 * last written. We set the csize to that value
1439 * and zero out any part that should not contain
1442 bzero((char *)cbuf
+ csize
, HDR_GET_PSIZE(hdr
) - csize
);
1443 csize
= HDR_GET_PSIZE(hdr
);
1445 zio_push_transform(zio
, cbuf
, csize
, HDR_GET_PSIZE(hdr
), NULL
);
1449 * Block pointers always store the checksum for the logical data.
1450 * If the block pointer has the gang bit set, then the checksum
1451 * it represents is for the reconstituted data and not for an
1452 * individual gang member. The zio pipeline, however, must be able to
1453 * determine the checksum of each of the gang constituents so it
1454 * treats the checksum comparison differently than what we need
1455 * for l2arc blocks. This prevents us from using the
1456 * zio_checksum_error() interface directly. Instead we must call the
1457 * zio_checksum_error_impl() so that we can ensure the checksum is
1458 * generated using the correct checksum algorithm and accounts for the
1459 * logical I/O size and not just a gang fragment.
1461 valid_cksum
= (zio_checksum_error_impl(zio
->io_spa
, zio
->io_bp
,
1462 BP_GET_CHECKSUM(zio
->io_bp
), zio
->io_abd
, zio
->io_size
,
1463 zio
->io_offset
, NULL
) == 0);
1464 zio_pop_transforms(zio
);
1465 return (valid_cksum
);
1469 * Given a buf full of data, if ZFS_DEBUG_MODIFY is enabled this computes a
1470 * checksum and attaches it to the buf's hdr so that we can ensure that the buf
1471 * isn't modified later on. If buf is compressed or there is already a checksum
1472 * on the hdr, this is a no-op (we only checksum uncompressed bufs).
1475 arc_cksum_compute(arc_buf_t
*buf
)
1477 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
1479 if (!(zfs_flags
& ZFS_DEBUG_MODIFY
))
1482 ASSERT(HDR_HAS_L1HDR(hdr
));
1484 mutex_enter(&buf
->b_hdr
->b_l1hdr
.b_freeze_lock
);
1485 if (hdr
->b_l1hdr
.b_freeze_cksum
!= NULL
) {
1486 mutex_exit(&hdr
->b_l1hdr
.b_freeze_lock
);
1488 } else if (ARC_BUF_COMPRESSED(buf
)) {
1489 mutex_exit(&hdr
->b_l1hdr
.b_freeze_lock
);
1493 ASSERT(!ARC_BUF_COMPRESSED(buf
));
1494 hdr
->b_l1hdr
.b_freeze_cksum
= kmem_alloc(sizeof (zio_cksum_t
),
1496 fletcher_2_native(buf
->b_data
, arc_buf_size(buf
), NULL
,
1497 hdr
->b_l1hdr
.b_freeze_cksum
);
1498 mutex_exit(&hdr
->b_l1hdr
.b_freeze_lock
);
1504 arc_buf_sigsegv(int sig
, siginfo_t
*si
, void *unused
)
1506 panic("Got SIGSEGV at address: 0x%lx\n", (long)si
->si_addr
);
1512 arc_buf_unwatch(arc_buf_t
*buf
)
1516 ASSERT0(mprotect(buf
->b_data
, HDR_GET_LSIZE(buf
->b_hdr
),
1517 PROT_READ
| PROT_WRITE
));
1524 arc_buf_watch(arc_buf_t
*buf
)
1528 ASSERT0(mprotect(buf
->b_data
, arc_buf_size(buf
),
1533 static arc_buf_contents_t
1534 arc_buf_type(arc_buf_hdr_t
*hdr
)
1536 arc_buf_contents_t type
;
1537 if (HDR_ISTYPE_METADATA(hdr
)) {
1538 type
= ARC_BUFC_METADATA
;
1540 type
= ARC_BUFC_DATA
;
1542 VERIFY3U(hdr
->b_type
, ==, type
);
1547 arc_is_metadata(arc_buf_t
*buf
)
1549 return (HDR_ISTYPE_METADATA(buf
->b_hdr
) != 0);
1553 arc_bufc_to_flags(arc_buf_contents_t type
)
1557 /* metadata field is 0 if buffer contains normal data */
1559 case ARC_BUFC_METADATA
:
1560 return (ARC_FLAG_BUFC_METADATA
);
1564 panic("undefined ARC buffer type!");
1565 return ((uint32_t)-1);
1569 arc_buf_thaw(arc_buf_t
*buf
)
1571 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
1573 ASSERT3P(hdr
->b_l1hdr
.b_state
, ==, arc_anon
);
1574 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
1576 arc_cksum_verify(buf
);
1579 * Compressed buffers do not manipulate the b_freeze_cksum or
1580 * allocate b_thawed.
1582 if (ARC_BUF_COMPRESSED(buf
)) {
1586 ASSERT(HDR_HAS_L1HDR(hdr
));
1587 arc_cksum_free(hdr
);
1588 arc_buf_unwatch(buf
);
1592 arc_buf_freeze(arc_buf_t
*buf
)
1594 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
1595 kmutex_t
*hash_lock
;
1597 if (!(zfs_flags
& ZFS_DEBUG_MODIFY
))
1600 if (ARC_BUF_COMPRESSED(buf
)) {
1604 hash_lock
= HDR_LOCK(hdr
);
1605 mutex_enter(hash_lock
);
1607 ASSERT(HDR_HAS_L1HDR(hdr
));
1608 ASSERT(hdr
->b_l1hdr
.b_freeze_cksum
!= NULL
||
1609 hdr
->b_l1hdr
.b_state
== arc_anon
);
1610 arc_cksum_compute(buf
);
1611 mutex_exit(hash_lock
);
1615 * The arc_buf_hdr_t's b_flags should never be modified directly. Instead,
1616 * the following functions should be used to ensure that the flags are
1617 * updated in a thread-safe way. When manipulating the flags either
1618 * the hash_lock must be held or the hdr must be undiscoverable. This
1619 * ensures that we're not racing with any other threads when updating
1623 arc_hdr_set_flags(arc_buf_hdr_t
*hdr
, arc_flags_t flags
)
1625 ASSERT(MUTEX_HELD(HDR_LOCK(hdr
)) || HDR_EMPTY(hdr
));
1626 hdr
->b_flags
|= flags
;
1630 arc_hdr_clear_flags(arc_buf_hdr_t
*hdr
, arc_flags_t flags
)
1632 ASSERT(MUTEX_HELD(HDR_LOCK(hdr
)) || HDR_EMPTY(hdr
));
1633 hdr
->b_flags
&= ~flags
;
1637 * Setting the compression bits in the arc_buf_hdr_t's b_flags is
1638 * done in a special way since we have to clear and set bits
1639 * at the same time. Consumers that wish to set the compression bits
1640 * must use this function to ensure that the flags are updated in
1641 * thread-safe manner.
1644 arc_hdr_set_compress(arc_buf_hdr_t
*hdr
, enum zio_compress cmp
)
1646 ASSERT(MUTEX_HELD(HDR_LOCK(hdr
)) || HDR_EMPTY(hdr
));
1649 * Holes and embedded blocks will always have a psize = 0 so
1650 * we ignore the compression of the blkptr and set the
1651 * want to uncompress them. Mark them as uncompressed.
1653 if (!zfs_compressed_arc_enabled
|| HDR_GET_PSIZE(hdr
) == 0) {
1654 arc_hdr_clear_flags(hdr
, ARC_FLAG_COMPRESSED_ARC
);
1655 HDR_SET_COMPRESS(hdr
, ZIO_COMPRESS_OFF
);
1656 ASSERT(!HDR_COMPRESSION_ENABLED(hdr
));
1657 ASSERT3U(HDR_GET_COMPRESS(hdr
), ==, ZIO_COMPRESS_OFF
);
1659 arc_hdr_set_flags(hdr
, ARC_FLAG_COMPRESSED_ARC
);
1660 HDR_SET_COMPRESS(hdr
, cmp
);
1661 ASSERT3U(HDR_GET_COMPRESS(hdr
), ==, cmp
);
1662 ASSERT(HDR_COMPRESSION_ENABLED(hdr
));
1667 * Looks for another buf on the same hdr which has the data decompressed, copies
1668 * from it, and returns true. If no such buf exists, returns false.
1671 arc_buf_try_copy_decompressed_data(arc_buf_t
*buf
)
1673 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
1675 boolean_t copied
= B_FALSE
;
1677 ASSERT(HDR_HAS_L1HDR(hdr
));
1678 ASSERT3P(buf
->b_data
, !=, NULL
);
1679 ASSERT(!ARC_BUF_COMPRESSED(buf
));
1681 for (from
= hdr
->b_l1hdr
.b_buf
; from
!= NULL
;
1682 from
= from
->b_next
) {
1683 /* can't use our own data buffer */
1688 if (!ARC_BUF_COMPRESSED(from
)) {
1689 bcopy(from
->b_data
, buf
->b_data
, arc_buf_size(buf
));
1696 * There were no decompressed bufs, so there should not be a
1697 * checksum on the hdr either.
1699 EQUIV(!copied
, hdr
->b_l1hdr
.b_freeze_cksum
== NULL
);
1705 * Given a buf that has a data buffer attached to it, this function will
1706 * efficiently fill the buf with data of the specified compression setting from
1707 * the hdr and update the hdr's b_freeze_cksum if necessary. If the buf and hdr
1708 * are already sharing a data buf, no copy is performed.
1710 * If the buf is marked as compressed but uncompressed data was requested, this
1711 * will allocate a new data buffer for the buf, remove that flag, and fill the
1712 * buf with uncompressed data. You can't request a compressed buf on a hdr with
1713 * uncompressed data, and (since we haven't added support for it yet) if you
1714 * want compressed data your buf must already be marked as compressed and have
1715 * the correct-sized data buffer.
1718 arc_buf_fill(arc_buf_t
*buf
, boolean_t compressed
)
1720 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
1721 boolean_t hdr_compressed
= (HDR_GET_COMPRESS(hdr
) != ZIO_COMPRESS_OFF
);
1722 dmu_object_byteswap_t bswap
= hdr
->b_l1hdr
.b_byteswap
;
1724 ASSERT3P(buf
->b_data
, !=, NULL
);
1725 IMPLY(compressed
, hdr_compressed
);
1726 IMPLY(compressed
, ARC_BUF_COMPRESSED(buf
));
1728 if (hdr_compressed
== compressed
) {
1729 if (!arc_buf_is_shared(buf
)) {
1730 abd_copy_to_buf(buf
->b_data
, hdr
->b_l1hdr
.b_pabd
,
1734 ASSERT(hdr_compressed
);
1735 ASSERT(!compressed
);
1736 ASSERT3U(HDR_GET_LSIZE(hdr
), !=, HDR_GET_PSIZE(hdr
));
1739 * If the buf is sharing its data with the hdr, unlink it and
1740 * allocate a new data buffer for the buf.
1742 if (arc_buf_is_shared(buf
)) {
1743 ASSERT(ARC_BUF_COMPRESSED(buf
));
1745 /* We need to give the buf it's own b_data */
1746 buf
->b_flags
&= ~ARC_BUF_FLAG_SHARED
;
1748 arc_get_data_buf(hdr
, HDR_GET_LSIZE(hdr
), buf
);
1749 arc_hdr_clear_flags(hdr
, ARC_FLAG_SHARED_DATA
);
1751 /* Previously overhead was 0; just add new overhead */
1752 ARCSTAT_INCR(arcstat_overhead_size
, HDR_GET_LSIZE(hdr
));
1753 } else if (ARC_BUF_COMPRESSED(buf
)) {
1754 /* We need to reallocate the buf's b_data */
1755 arc_free_data_buf(hdr
, buf
->b_data
, HDR_GET_PSIZE(hdr
),
1758 arc_get_data_buf(hdr
, HDR_GET_LSIZE(hdr
), buf
);
1760 /* We increased the size of b_data; update overhead */
1761 ARCSTAT_INCR(arcstat_overhead_size
,
1762 HDR_GET_LSIZE(hdr
) - HDR_GET_PSIZE(hdr
));
1766 * Regardless of the buf's previous compression settings, it
1767 * should not be compressed at the end of this function.
1769 buf
->b_flags
&= ~ARC_BUF_FLAG_COMPRESSED
;
1772 * Try copying the data from another buf which already has a
1773 * decompressed version. If that's not possible, it's time to
1774 * bite the bullet and decompress the data from the hdr.
1776 if (arc_buf_try_copy_decompressed_data(buf
)) {
1777 /* Skip byteswapping and checksumming (already done) */
1778 ASSERT3P(hdr
->b_l1hdr
.b_freeze_cksum
, !=, NULL
);
1781 int error
= zio_decompress_data(HDR_GET_COMPRESS(hdr
),
1782 hdr
->b_l1hdr
.b_pabd
, buf
->b_data
,
1783 HDR_GET_PSIZE(hdr
), HDR_GET_LSIZE(hdr
));
1786 * Absent hardware errors or software bugs, this should
1787 * be impossible, but log it anyway so we can debug it.
1791 "hdr %p, compress %d, psize %d, lsize %d",
1792 hdr
, HDR_GET_COMPRESS(hdr
),
1793 HDR_GET_PSIZE(hdr
), HDR_GET_LSIZE(hdr
));
1794 return (SET_ERROR(EIO
));
1799 /* Byteswap the buf's data if necessary */
1800 if (bswap
!= DMU_BSWAP_NUMFUNCS
) {
1801 ASSERT(!HDR_SHARED_DATA(hdr
));
1802 ASSERT3U(bswap
, <, DMU_BSWAP_NUMFUNCS
);
1803 dmu_ot_byteswap
[bswap
].ob_func(buf
->b_data
, HDR_GET_LSIZE(hdr
));
1806 /* Compute the hdr's checksum if necessary */
1807 arc_cksum_compute(buf
);
1813 arc_decompress(arc_buf_t
*buf
)
1815 return (arc_buf_fill(buf
, B_FALSE
));
1819 * Return the size of the block, b_pabd, that is stored in the arc_buf_hdr_t.
1822 arc_hdr_size(arc_buf_hdr_t
*hdr
)
1826 if (HDR_GET_COMPRESS(hdr
) != ZIO_COMPRESS_OFF
&&
1827 HDR_GET_PSIZE(hdr
) > 0) {
1828 size
= HDR_GET_PSIZE(hdr
);
1830 ASSERT3U(HDR_GET_LSIZE(hdr
), !=, 0);
1831 size
= HDR_GET_LSIZE(hdr
);
1837 * Increment the amount of evictable space in the arc_state_t's refcount.
1838 * We account for the space used by the hdr and the arc buf individually
1839 * so that we can add and remove them from the refcount individually.
1842 arc_evictable_space_increment(arc_buf_hdr_t
*hdr
, arc_state_t
*state
)
1844 arc_buf_contents_t type
= arc_buf_type(hdr
);
1847 ASSERT(HDR_HAS_L1HDR(hdr
));
1849 if (GHOST_STATE(state
)) {
1850 ASSERT0(hdr
->b_l1hdr
.b_bufcnt
);
1851 ASSERT3P(hdr
->b_l1hdr
.b_buf
, ==, NULL
);
1852 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, ==, NULL
);
1853 (void) refcount_add_many(&state
->arcs_esize
[type
],
1854 HDR_GET_LSIZE(hdr
), hdr
);
1858 ASSERT(!GHOST_STATE(state
));
1859 if (hdr
->b_l1hdr
.b_pabd
!= NULL
) {
1860 (void) refcount_add_many(&state
->arcs_esize
[type
],
1861 arc_hdr_size(hdr
), hdr
);
1863 for (buf
= hdr
->b_l1hdr
.b_buf
; buf
!= NULL
; buf
= buf
->b_next
) {
1864 if (arc_buf_is_shared(buf
))
1866 (void) refcount_add_many(&state
->arcs_esize
[type
],
1867 arc_buf_size(buf
), buf
);
1872 * Decrement the amount of evictable space in the arc_state_t's refcount.
1873 * We account for the space used by the hdr and the arc buf individually
1874 * so that we can add and remove them from the refcount individually.
1877 arc_evictable_space_decrement(arc_buf_hdr_t
*hdr
, arc_state_t
*state
)
1879 arc_buf_contents_t type
= arc_buf_type(hdr
);
1882 ASSERT(HDR_HAS_L1HDR(hdr
));
1884 if (GHOST_STATE(state
)) {
1885 ASSERT0(hdr
->b_l1hdr
.b_bufcnt
);
1886 ASSERT3P(hdr
->b_l1hdr
.b_buf
, ==, NULL
);
1887 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, ==, NULL
);
1888 (void) refcount_remove_many(&state
->arcs_esize
[type
],
1889 HDR_GET_LSIZE(hdr
), hdr
);
1893 ASSERT(!GHOST_STATE(state
));
1894 if (hdr
->b_l1hdr
.b_pabd
!= NULL
) {
1895 (void) refcount_remove_many(&state
->arcs_esize
[type
],
1896 arc_hdr_size(hdr
), hdr
);
1898 for (buf
= hdr
->b_l1hdr
.b_buf
; buf
!= NULL
; buf
= buf
->b_next
) {
1899 if (arc_buf_is_shared(buf
))
1901 (void) refcount_remove_many(&state
->arcs_esize
[type
],
1902 arc_buf_size(buf
), buf
);
1907 * Add a reference to this hdr indicating that someone is actively
1908 * referencing that memory. When the refcount transitions from 0 to 1,
1909 * we remove it from the respective arc_state_t list to indicate that
1910 * it is not evictable.
1913 add_reference(arc_buf_hdr_t
*hdr
, void *tag
)
1917 ASSERT(HDR_HAS_L1HDR(hdr
));
1918 if (!MUTEX_HELD(HDR_LOCK(hdr
))) {
1919 ASSERT(hdr
->b_l1hdr
.b_state
== arc_anon
);
1920 ASSERT(refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
));
1921 ASSERT3P(hdr
->b_l1hdr
.b_buf
, ==, NULL
);
1924 state
= hdr
->b_l1hdr
.b_state
;
1926 if ((refcount_add(&hdr
->b_l1hdr
.b_refcnt
, tag
) == 1) &&
1927 (state
!= arc_anon
)) {
1928 /* We don't use the L2-only state list. */
1929 if (state
!= arc_l2c_only
) {
1930 multilist_remove(state
->arcs_list
[arc_buf_type(hdr
)],
1932 arc_evictable_space_decrement(hdr
, state
);
1934 /* remove the prefetch flag if we get a reference */
1935 arc_hdr_clear_flags(hdr
, ARC_FLAG_PREFETCH
);
1940 * Remove a reference from this hdr. When the reference transitions from
1941 * 1 to 0 and we're not anonymous, then we add this hdr to the arc_state_t's
1942 * list making it eligible for eviction.
1945 remove_reference(arc_buf_hdr_t
*hdr
, kmutex_t
*hash_lock
, void *tag
)
1948 arc_state_t
*state
= hdr
->b_l1hdr
.b_state
;
1950 ASSERT(HDR_HAS_L1HDR(hdr
));
1951 ASSERT(state
== arc_anon
|| MUTEX_HELD(hash_lock
));
1952 ASSERT(!GHOST_STATE(state
));
1955 * arc_l2c_only counts as a ghost state so we don't need to explicitly
1956 * check to prevent usage of the arc_l2c_only list.
1958 if (((cnt
= refcount_remove(&hdr
->b_l1hdr
.b_refcnt
, tag
)) == 0) &&
1959 (state
!= arc_anon
)) {
1960 multilist_insert(state
->arcs_list
[arc_buf_type(hdr
)], hdr
);
1961 ASSERT3U(hdr
->b_l1hdr
.b_bufcnt
, >, 0);
1962 arc_evictable_space_increment(hdr
, state
);
1968 * Returns detailed information about a specific arc buffer. When the
1969 * state_index argument is set the function will calculate the arc header
1970 * list position for its arc state. Since this requires a linear traversal
1971 * callers are strongly encourage not to do this. However, it can be helpful
1972 * for targeted analysis so the functionality is provided.
1975 arc_buf_info(arc_buf_t
*ab
, arc_buf_info_t
*abi
, int state_index
)
1977 arc_buf_hdr_t
*hdr
= ab
->b_hdr
;
1978 l1arc_buf_hdr_t
*l1hdr
= NULL
;
1979 l2arc_buf_hdr_t
*l2hdr
= NULL
;
1980 arc_state_t
*state
= NULL
;
1982 memset(abi
, 0, sizeof (arc_buf_info_t
));
1987 abi
->abi_flags
= hdr
->b_flags
;
1989 if (HDR_HAS_L1HDR(hdr
)) {
1990 l1hdr
= &hdr
->b_l1hdr
;
1991 state
= l1hdr
->b_state
;
1993 if (HDR_HAS_L2HDR(hdr
))
1994 l2hdr
= &hdr
->b_l2hdr
;
1997 abi
->abi_bufcnt
= l1hdr
->b_bufcnt
;
1998 abi
->abi_access
= l1hdr
->b_arc_access
;
1999 abi
->abi_mru_hits
= l1hdr
->b_mru_hits
;
2000 abi
->abi_mru_ghost_hits
= l1hdr
->b_mru_ghost_hits
;
2001 abi
->abi_mfu_hits
= l1hdr
->b_mfu_hits
;
2002 abi
->abi_mfu_ghost_hits
= l1hdr
->b_mfu_ghost_hits
;
2003 abi
->abi_holds
= refcount_count(&l1hdr
->b_refcnt
);
2007 abi
->abi_l2arc_dattr
= l2hdr
->b_daddr
;
2008 abi
->abi_l2arc_hits
= l2hdr
->b_hits
;
2011 abi
->abi_state_type
= state
? state
->arcs_state
: ARC_STATE_ANON
;
2012 abi
->abi_state_contents
= arc_buf_type(hdr
);
2013 abi
->abi_size
= arc_hdr_size(hdr
);
2017 * Move the supplied buffer to the indicated state. The hash lock
2018 * for the buffer must be held by the caller.
2021 arc_change_state(arc_state_t
*new_state
, arc_buf_hdr_t
*hdr
,
2022 kmutex_t
*hash_lock
)
2024 arc_state_t
*old_state
;
2027 boolean_t update_old
, update_new
;
2028 arc_buf_contents_t buftype
= arc_buf_type(hdr
);
2031 * We almost always have an L1 hdr here, since we call arc_hdr_realloc()
2032 * in arc_read() when bringing a buffer out of the L2ARC. However, the
2033 * L1 hdr doesn't always exist when we change state to arc_anon before
2034 * destroying a header, in which case reallocating to add the L1 hdr is
2037 if (HDR_HAS_L1HDR(hdr
)) {
2038 old_state
= hdr
->b_l1hdr
.b_state
;
2039 refcnt
= refcount_count(&hdr
->b_l1hdr
.b_refcnt
);
2040 bufcnt
= hdr
->b_l1hdr
.b_bufcnt
;
2041 update_old
= (bufcnt
> 0 || hdr
->b_l1hdr
.b_pabd
!= NULL
);
2043 old_state
= arc_l2c_only
;
2046 update_old
= B_FALSE
;
2048 update_new
= update_old
;
2050 ASSERT(MUTEX_HELD(hash_lock
));
2051 ASSERT3P(new_state
, !=, old_state
);
2052 ASSERT(!GHOST_STATE(new_state
) || bufcnt
== 0);
2053 ASSERT(old_state
!= arc_anon
|| bufcnt
<= 1);
2056 * If this buffer is evictable, transfer it from the
2057 * old state list to the new state list.
2060 if (old_state
!= arc_anon
&& old_state
!= arc_l2c_only
) {
2061 ASSERT(HDR_HAS_L1HDR(hdr
));
2062 multilist_remove(old_state
->arcs_list
[buftype
], hdr
);
2064 if (GHOST_STATE(old_state
)) {
2066 ASSERT3P(hdr
->b_l1hdr
.b_buf
, ==, NULL
);
2067 update_old
= B_TRUE
;
2069 arc_evictable_space_decrement(hdr
, old_state
);
2071 if (new_state
!= arc_anon
&& new_state
!= arc_l2c_only
) {
2073 * An L1 header always exists here, since if we're
2074 * moving to some L1-cached state (i.e. not l2c_only or
2075 * anonymous), we realloc the header to add an L1hdr
2078 ASSERT(HDR_HAS_L1HDR(hdr
));
2079 multilist_insert(new_state
->arcs_list
[buftype
], hdr
);
2081 if (GHOST_STATE(new_state
)) {
2083 ASSERT3P(hdr
->b_l1hdr
.b_buf
, ==, NULL
);
2084 update_new
= B_TRUE
;
2086 arc_evictable_space_increment(hdr
, new_state
);
2090 ASSERT(!HDR_EMPTY(hdr
));
2091 if (new_state
== arc_anon
&& HDR_IN_HASH_TABLE(hdr
))
2092 buf_hash_remove(hdr
);
2094 /* adjust state sizes (ignore arc_l2c_only) */
2096 if (update_new
&& new_state
!= arc_l2c_only
) {
2097 ASSERT(HDR_HAS_L1HDR(hdr
));
2098 if (GHOST_STATE(new_state
)) {
2102 * When moving a header to a ghost state, we first
2103 * remove all arc buffers. Thus, we'll have a
2104 * bufcnt of zero, and no arc buffer to use for
2105 * the reference. As a result, we use the arc
2106 * header pointer for the reference.
2108 (void) refcount_add_many(&new_state
->arcs_size
,
2109 HDR_GET_LSIZE(hdr
), hdr
);
2110 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, ==, NULL
);
2113 uint32_t buffers
= 0;
2116 * Each individual buffer holds a unique reference,
2117 * thus we must remove each of these references one
2120 for (buf
= hdr
->b_l1hdr
.b_buf
; buf
!= NULL
;
2121 buf
= buf
->b_next
) {
2122 ASSERT3U(bufcnt
, !=, 0);
2126 * When the arc_buf_t is sharing the data
2127 * block with the hdr, the owner of the
2128 * reference belongs to the hdr. Only
2129 * add to the refcount if the arc_buf_t is
2132 if (arc_buf_is_shared(buf
))
2135 (void) refcount_add_many(&new_state
->arcs_size
,
2136 arc_buf_size(buf
), buf
);
2138 ASSERT3U(bufcnt
, ==, buffers
);
2140 if (hdr
->b_l1hdr
.b_pabd
!= NULL
) {
2141 (void) refcount_add_many(&new_state
->arcs_size
,
2142 arc_hdr_size(hdr
), hdr
);
2144 ASSERT(GHOST_STATE(old_state
));
2149 if (update_old
&& old_state
!= arc_l2c_only
) {
2150 ASSERT(HDR_HAS_L1HDR(hdr
));
2151 if (GHOST_STATE(old_state
)) {
2153 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, ==, NULL
);
2156 * When moving a header off of a ghost state,
2157 * the header will not contain any arc buffers.
2158 * We use the arc header pointer for the reference
2159 * which is exactly what we did when we put the
2160 * header on the ghost state.
2163 (void) refcount_remove_many(&old_state
->arcs_size
,
2164 HDR_GET_LSIZE(hdr
), hdr
);
2167 uint32_t buffers
= 0;
2170 * Each individual buffer holds a unique reference,
2171 * thus we must remove each of these references one
2174 for (buf
= hdr
->b_l1hdr
.b_buf
; buf
!= NULL
;
2175 buf
= buf
->b_next
) {
2176 ASSERT3U(bufcnt
, !=, 0);
2180 * When the arc_buf_t is sharing the data
2181 * block with the hdr, the owner of the
2182 * reference belongs to the hdr. Only
2183 * add to the refcount if the arc_buf_t is
2186 if (arc_buf_is_shared(buf
))
2189 (void) refcount_remove_many(
2190 &old_state
->arcs_size
, arc_buf_size(buf
),
2193 ASSERT3U(bufcnt
, ==, buffers
);
2194 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, !=, NULL
);
2195 (void) refcount_remove_many(
2196 &old_state
->arcs_size
, arc_hdr_size(hdr
), hdr
);
2200 if (HDR_HAS_L1HDR(hdr
))
2201 hdr
->b_l1hdr
.b_state
= new_state
;
2204 * L2 headers should never be on the L2 state list since they don't
2205 * have L1 headers allocated.
2207 ASSERT(multilist_is_empty(arc_l2c_only
->arcs_list
[ARC_BUFC_DATA
]) &&
2208 multilist_is_empty(arc_l2c_only
->arcs_list
[ARC_BUFC_METADATA
]));
2212 arc_space_consume(uint64_t space
, arc_space_type_t type
)
2214 ASSERT(type
>= 0 && type
< ARC_SPACE_NUMTYPES
);
2219 case ARC_SPACE_DATA
:
2220 ARCSTAT_INCR(arcstat_data_size
, space
);
2222 case ARC_SPACE_META
:
2223 ARCSTAT_INCR(arcstat_metadata_size
, space
);
2225 case ARC_SPACE_BONUS
:
2226 ARCSTAT_INCR(arcstat_bonus_size
, space
);
2228 case ARC_SPACE_DNODE
:
2229 ARCSTAT_INCR(arcstat_dnode_size
, space
);
2231 case ARC_SPACE_DBUF
:
2232 ARCSTAT_INCR(arcstat_dbuf_size
, space
);
2234 case ARC_SPACE_HDRS
:
2235 ARCSTAT_INCR(arcstat_hdr_size
, space
);
2237 case ARC_SPACE_L2HDRS
:
2238 ARCSTAT_INCR(arcstat_l2_hdr_size
, space
);
2242 if (type
!= ARC_SPACE_DATA
)
2243 ARCSTAT_INCR(arcstat_meta_used
, space
);
2245 atomic_add_64(&arc_size
, space
);
2249 arc_space_return(uint64_t space
, arc_space_type_t type
)
2251 ASSERT(type
>= 0 && type
< ARC_SPACE_NUMTYPES
);
2256 case ARC_SPACE_DATA
:
2257 ARCSTAT_INCR(arcstat_data_size
, -space
);
2259 case ARC_SPACE_META
:
2260 ARCSTAT_INCR(arcstat_metadata_size
, -space
);
2262 case ARC_SPACE_BONUS
:
2263 ARCSTAT_INCR(arcstat_bonus_size
, -space
);
2265 case ARC_SPACE_DNODE
:
2266 ARCSTAT_INCR(arcstat_dnode_size
, -space
);
2268 case ARC_SPACE_DBUF
:
2269 ARCSTAT_INCR(arcstat_dbuf_size
, -space
);
2271 case ARC_SPACE_HDRS
:
2272 ARCSTAT_INCR(arcstat_hdr_size
, -space
);
2274 case ARC_SPACE_L2HDRS
:
2275 ARCSTAT_INCR(arcstat_l2_hdr_size
, -space
);
2279 if (type
!= ARC_SPACE_DATA
) {
2280 ASSERT(arc_meta_used
>= space
);
2281 if (arc_meta_max
< arc_meta_used
)
2282 arc_meta_max
= arc_meta_used
;
2283 ARCSTAT_INCR(arcstat_meta_used
, -space
);
2286 ASSERT(arc_size
>= space
);
2287 atomic_add_64(&arc_size
, -space
);
2291 * Given a hdr and a buf, returns whether that buf can share its b_data buffer
2292 * with the hdr's b_pabd.
2295 arc_can_share(arc_buf_hdr_t
*hdr
, arc_buf_t
*buf
)
2297 boolean_t hdr_compressed
, buf_compressed
;
2299 * The criteria for sharing a hdr's data are:
2300 * 1. the hdr's compression matches the buf's compression
2301 * 2. the hdr doesn't need to be byteswapped
2302 * 3. the hdr isn't already being shared
2303 * 4. the buf is either compressed or it is the last buf in the hdr list
2305 * Criterion #4 maintains the invariant that shared uncompressed
2306 * bufs must be the final buf in the hdr's b_buf list. Reading this, you
2307 * might ask, "if a compressed buf is allocated first, won't that be the
2308 * last thing in the list?", but in that case it's impossible to create
2309 * a shared uncompressed buf anyway (because the hdr must be compressed
2310 * to have the compressed buf). You might also think that #3 is
2311 * sufficient to make this guarantee, however it's possible
2312 * (specifically in the rare L2ARC write race mentioned in
2313 * arc_buf_alloc_impl()) there will be an existing uncompressed buf that
2314 * is sharable, but wasn't at the time of its allocation. Rather than
2315 * allow a new shared uncompressed buf to be created and then shuffle
2316 * the list around to make it the last element, this simply disallows
2317 * sharing if the new buf isn't the first to be added.
2319 ASSERT3P(buf
->b_hdr
, ==, hdr
);
2320 hdr_compressed
= HDR_GET_COMPRESS(hdr
) != ZIO_COMPRESS_OFF
;
2321 buf_compressed
= ARC_BUF_COMPRESSED(buf
) != 0;
2322 return (buf_compressed
== hdr_compressed
&&
2323 hdr
->b_l1hdr
.b_byteswap
== DMU_BSWAP_NUMFUNCS
&&
2324 !HDR_SHARED_DATA(hdr
) &&
2325 (ARC_BUF_LAST(buf
) || ARC_BUF_COMPRESSED(buf
)));
2329 * Allocate a buf for this hdr. If you care about the data that's in the hdr,
2330 * or if you want a compressed buffer, pass those flags in. Returns 0 if the
2331 * copy was made successfully, or an error code otherwise.
2334 arc_buf_alloc_impl(arc_buf_hdr_t
*hdr
, void *tag
, boolean_t compressed
,
2335 boolean_t fill
, arc_buf_t
**ret
)
2338 boolean_t can_share
;
2340 ASSERT(HDR_HAS_L1HDR(hdr
));
2341 ASSERT3U(HDR_GET_LSIZE(hdr
), >, 0);
2342 VERIFY(hdr
->b_type
== ARC_BUFC_DATA
||
2343 hdr
->b_type
== ARC_BUFC_METADATA
);
2344 ASSERT3P(ret
, !=, NULL
);
2345 ASSERT3P(*ret
, ==, NULL
);
2347 hdr
->b_l1hdr
.b_mru_hits
= 0;
2348 hdr
->b_l1hdr
.b_mru_ghost_hits
= 0;
2349 hdr
->b_l1hdr
.b_mfu_hits
= 0;
2350 hdr
->b_l1hdr
.b_mfu_ghost_hits
= 0;
2351 hdr
->b_l1hdr
.b_l2_hits
= 0;
2353 buf
= *ret
= kmem_cache_alloc(buf_cache
, KM_PUSHPAGE
);
2356 buf
->b_next
= hdr
->b_l1hdr
.b_buf
;
2359 add_reference(hdr
, tag
);
2362 * We're about to change the hdr's b_flags. We must either
2363 * hold the hash_lock or be undiscoverable.
2365 ASSERT(MUTEX_HELD(HDR_LOCK(hdr
)) || HDR_EMPTY(hdr
));
2368 * Only honor requests for compressed bufs if the hdr is actually
2371 if (compressed
&& HDR_GET_COMPRESS(hdr
) != ZIO_COMPRESS_OFF
)
2372 buf
->b_flags
|= ARC_BUF_FLAG_COMPRESSED
;
2375 * Although the ARC should handle it correctly, levels above the ARC
2376 * should prevent us from having multiple compressed bufs off the same
2377 * hdr. To ensure we notice it if this behavior changes, we assert this
2378 * here the best we can.
2380 IMPLY(ARC_BUF_COMPRESSED(buf
), !HDR_SHARED_DATA(hdr
));
2383 * If the hdr's data can be shared then we share the data buffer and
2384 * set the appropriate bit in the hdr's b_flags to indicate the hdr is
2385 * allocate a new buffer to store the buf's data.
2387 * There are two additional restrictions here because we're sharing
2388 * hdr -> buf instead of the usual buf -> hdr. First, the hdr can't be
2389 * actively involved in an L2ARC write, because if this buf is used by
2390 * an arc_write() then the hdr's data buffer will be released when the
2391 * write completes, even though the L2ARC write might still be using it.
2392 * Second, the hdr's ABD must be linear so that the buf's user doesn't
2393 * need to be ABD-aware.
2395 can_share
= arc_can_share(hdr
, buf
) && !HDR_L2_WRITING(hdr
) &&
2396 abd_is_linear(hdr
->b_l1hdr
.b_pabd
);
2398 /* Set up b_data and sharing */
2400 buf
->b_data
= abd_to_buf(hdr
->b_l1hdr
.b_pabd
);
2401 buf
->b_flags
|= ARC_BUF_FLAG_SHARED
;
2402 arc_hdr_set_flags(hdr
, ARC_FLAG_SHARED_DATA
);
2405 arc_get_data_buf(hdr
, arc_buf_size(buf
), buf
);
2406 ARCSTAT_INCR(arcstat_overhead_size
, arc_buf_size(buf
));
2408 VERIFY3P(buf
->b_data
, !=, NULL
);
2410 hdr
->b_l1hdr
.b_buf
= buf
;
2411 hdr
->b_l1hdr
.b_bufcnt
+= 1;
2414 * If the user wants the data from the hdr, we need to either copy or
2415 * decompress the data.
2418 return (arc_buf_fill(buf
, ARC_BUF_COMPRESSED(buf
) != 0));
2424 static char *arc_onloan_tag
= "onloan";
2427 * Loan out an anonymous arc buffer. Loaned buffers are not counted as in
2428 * flight data by arc_tempreserve_space() until they are "returned". Loaned
2429 * buffers must be returned to the arc before they can be used by the DMU or
2433 arc_loan_buf(spa_t
*spa
, boolean_t is_metadata
, int size
)
2435 arc_buf_t
*buf
= arc_alloc_buf(spa
, arc_onloan_tag
,
2436 is_metadata
? ARC_BUFC_METADATA
: ARC_BUFC_DATA
, size
);
2438 atomic_add_64(&arc_loaned_bytes
, size
);
2443 arc_loan_compressed_buf(spa_t
*spa
, uint64_t psize
, uint64_t lsize
,
2444 enum zio_compress compression_type
)
2446 arc_buf_t
*buf
= arc_alloc_compressed_buf(spa
, arc_onloan_tag
,
2447 psize
, lsize
, compression_type
);
2449 atomic_add_64(&arc_loaned_bytes
, psize
);
2455 * Return a loaned arc buffer to the arc.
2458 arc_return_buf(arc_buf_t
*buf
, void *tag
)
2460 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
2462 ASSERT3P(buf
->b_data
, !=, NULL
);
2463 ASSERT(HDR_HAS_L1HDR(hdr
));
2464 (void) refcount_add(&hdr
->b_l1hdr
.b_refcnt
, tag
);
2465 (void) refcount_remove(&hdr
->b_l1hdr
.b_refcnt
, arc_onloan_tag
);
2467 atomic_add_64(&arc_loaned_bytes
, -arc_buf_size(buf
));
2470 /* Detach an arc_buf from a dbuf (tag) */
2472 arc_loan_inuse_buf(arc_buf_t
*buf
, void *tag
)
2474 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
2476 ASSERT3P(buf
->b_data
, !=, NULL
);
2477 ASSERT(HDR_HAS_L1HDR(hdr
));
2478 (void) refcount_add(&hdr
->b_l1hdr
.b_refcnt
, arc_onloan_tag
);
2479 (void) refcount_remove(&hdr
->b_l1hdr
.b_refcnt
, tag
);
2481 atomic_add_64(&arc_loaned_bytes
, -arc_buf_size(buf
));
2485 l2arc_free_abd_on_write(abd_t
*abd
, size_t size
, arc_buf_contents_t type
)
2487 l2arc_data_free_t
*df
= kmem_alloc(sizeof (*df
), KM_SLEEP
);
2490 df
->l2df_size
= size
;
2491 df
->l2df_type
= type
;
2492 mutex_enter(&l2arc_free_on_write_mtx
);
2493 list_insert_head(l2arc_free_on_write
, df
);
2494 mutex_exit(&l2arc_free_on_write_mtx
);
2498 arc_hdr_free_on_write(arc_buf_hdr_t
*hdr
)
2500 arc_state_t
*state
= hdr
->b_l1hdr
.b_state
;
2501 arc_buf_contents_t type
= arc_buf_type(hdr
);
2502 uint64_t size
= arc_hdr_size(hdr
);
2504 /* protected by hash lock, if in the hash table */
2505 if (multilist_link_active(&hdr
->b_l1hdr
.b_arc_node
)) {
2506 ASSERT(refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
));
2507 ASSERT(state
!= arc_anon
&& state
!= arc_l2c_only
);
2509 (void) refcount_remove_many(&state
->arcs_esize
[type
],
2512 (void) refcount_remove_many(&state
->arcs_size
, size
, hdr
);
2513 if (type
== ARC_BUFC_METADATA
) {
2514 arc_space_return(size
, ARC_SPACE_META
);
2516 ASSERT(type
== ARC_BUFC_DATA
);
2517 arc_space_return(size
, ARC_SPACE_DATA
);
2520 l2arc_free_abd_on_write(hdr
->b_l1hdr
.b_pabd
, size
, type
);
2524 * Share the arc_buf_t's data with the hdr. Whenever we are sharing the
2525 * data buffer, we transfer the refcount ownership to the hdr and update
2526 * the appropriate kstats.
2529 arc_share_buf(arc_buf_hdr_t
*hdr
, arc_buf_t
*buf
)
2531 ASSERT(arc_can_share(hdr
, buf
));
2532 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, ==, NULL
);
2533 ASSERT(MUTEX_HELD(HDR_LOCK(hdr
)) || HDR_EMPTY(hdr
));
2536 * Start sharing the data buffer. We transfer the
2537 * refcount ownership to the hdr since it always owns
2538 * the refcount whenever an arc_buf_t is shared.
2540 refcount_transfer_ownership(&hdr
->b_l1hdr
.b_state
->arcs_size
, buf
, hdr
);
2541 hdr
->b_l1hdr
.b_pabd
= abd_get_from_buf(buf
->b_data
, arc_buf_size(buf
));
2542 abd_take_ownership_of_buf(hdr
->b_l1hdr
.b_pabd
,
2543 HDR_ISTYPE_METADATA(hdr
));
2544 arc_hdr_set_flags(hdr
, ARC_FLAG_SHARED_DATA
);
2545 buf
->b_flags
|= ARC_BUF_FLAG_SHARED
;
2548 * Since we've transferred ownership to the hdr we need
2549 * to increment its compressed and uncompressed kstats and
2550 * decrement the overhead size.
2552 ARCSTAT_INCR(arcstat_compressed_size
, arc_hdr_size(hdr
));
2553 ARCSTAT_INCR(arcstat_uncompressed_size
, HDR_GET_LSIZE(hdr
));
2554 ARCSTAT_INCR(arcstat_overhead_size
, -arc_buf_size(buf
));
2558 arc_unshare_buf(arc_buf_hdr_t
*hdr
, arc_buf_t
*buf
)
2560 ASSERT(arc_buf_is_shared(buf
));
2561 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, !=, NULL
);
2562 ASSERT(MUTEX_HELD(HDR_LOCK(hdr
)) || HDR_EMPTY(hdr
));
2565 * We are no longer sharing this buffer so we need
2566 * to transfer its ownership to the rightful owner.
2568 refcount_transfer_ownership(&hdr
->b_l1hdr
.b_state
->arcs_size
, hdr
, buf
);
2569 arc_hdr_clear_flags(hdr
, ARC_FLAG_SHARED_DATA
);
2570 abd_release_ownership_of_buf(hdr
->b_l1hdr
.b_pabd
);
2571 abd_put(hdr
->b_l1hdr
.b_pabd
);
2572 hdr
->b_l1hdr
.b_pabd
= NULL
;
2573 buf
->b_flags
&= ~ARC_BUF_FLAG_SHARED
;
2576 * Since the buffer is no longer shared between
2577 * the arc buf and the hdr, count it as overhead.
2579 ARCSTAT_INCR(arcstat_compressed_size
, -arc_hdr_size(hdr
));
2580 ARCSTAT_INCR(arcstat_uncompressed_size
, -HDR_GET_LSIZE(hdr
));
2581 ARCSTAT_INCR(arcstat_overhead_size
, arc_buf_size(buf
));
2585 * Remove an arc_buf_t from the hdr's buf list and return the last
2586 * arc_buf_t on the list. If no buffers remain on the list then return
2590 arc_buf_remove(arc_buf_hdr_t
*hdr
, arc_buf_t
*buf
)
2592 arc_buf_t
**bufp
= &hdr
->b_l1hdr
.b_buf
;
2593 arc_buf_t
*lastbuf
= NULL
;
2595 ASSERT(HDR_HAS_L1HDR(hdr
));
2596 ASSERT(MUTEX_HELD(HDR_LOCK(hdr
)) || HDR_EMPTY(hdr
));
2599 * Remove the buf from the hdr list and locate the last
2600 * remaining buffer on the list.
2602 while (*bufp
!= NULL
) {
2604 *bufp
= buf
->b_next
;
2607 * If we've removed a buffer in the middle of
2608 * the list then update the lastbuf and update
2611 if (*bufp
!= NULL
) {
2613 bufp
= &(*bufp
)->b_next
;
2617 ASSERT3P(lastbuf
, !=, buf
);
2618 IMPLY(hdr
->b_l1hdr
.b_bufcnt
> 0, lastbuf
!= NULL
);
2619 IMPLY(hdr
->b_l1hdr
.b_bufcnt
> 0, hdr
->b_l1hdr
.b_buf
!= NULL
);
2620 IMPLY(lastbuf
!= NULL
, ARC_BUF_LAST(lastbuf
));
2626 * Free up buf->b_data and pull the arc_buf_t off of the the arc_buf_hdr_t's
2630 arc_buf_destroy_impl(arc_buf_t
*buf
)
2633 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
2636 * Free up the data associated with the buf but only if we're not
2637 * sharing this with the hdr. If we are sharing it with the hdr, the
2638 * hdr is responsible for doing the free.
2640 if (buf
->b_data
!= NULL
) {
2642 * We're about to change the hdr's b_flags. We must either
2643 * hold the hash_lock or be undiscoverable.
2645 ASSERT(MUTEX_HELD(HDR_LOCK(hdr
)) || HDR_EMPTY(hdr
));
2647 arc_cksum_verify(buf
);
2648 arc_buf_unwatch(buf
);
2650 if (arc_buf_is_shared(buf
)) {
2651 arc_hdr_clear_flags(hdr
, ARC_FLAG_SHARED_DATA
);
2653 uint64_t size
= arc_buf_size(buf
);
2654 arc_free_data_buf(hdr
, buf
->b_data
, size
, buf
);
2655 ARCSTAT_INCR(arcstat_overhead_size
, -size
);
2659 ASSERT(hdr
->b_l1hdr
.b_bufcnt
> 0);
2660 hdr
->b_l1hdr
.b_bufcnt
-= 1;
2663 lastbuf
= arc_buf_remove(hdr
, buf
);
2665 if (ARC_BUF_SHARED(buf
) && !ARC_BUF_COMPRESSED(buf
)) {
2667 * If the current arc_buf_t is sharing its data buffer with the
2668 * hdr, then reassign the hdr's b_pabd to share it with the new
2669 * buffer at the end of the list. The shared buffer is always
2670 * the last one on the hdr's buffer list.
2672 * There is an equivalent case for compressed bufs, but since
2673 * they aren't guaranteed to be the last buf in the list and
2674 * that is an exceedingly rare case, we just allow that space be
2675 * wasted temporarily.
2677 if (lastbuf
!= NULL
) {
2678 /* Only one buf can be shared at once */
2679 VERIFY(!arc_buf_is_shared(lastbuf
));
2680 /* hdr is uncompressed so can't have compressed buf */
2681 VERIFY(!ARC_BUF_COMPRESSED(lastbuf
));
2683 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, !=, NULL
);
2684 arc_hdr_free_pabd(hdr
);
2687 * We must setup a new shared block between the
2688 * last buffer and the hdr. The data would have
2689 * been allocated by the arc buf so we need to transfer
2690 * ownership to the hdr since it's now being shared.
2692 arc_share_buf(hdr
, lastbuf
);
2694 } else if (HDR_SHARED_DATA(hdr
)) {
2696 * Uncompressed shared buffers are always at the end
2697 * of the list. Compressed buffers don't have the
2698 * same requirements. This makes it hard to
2699 * simply assert that the lastbuf is shared so
2700 * we rely on the hdr's compression flags to determine
2701 * if we have a compressed, shared buffer.
2703 ASSERT3P(lastbuf
, !=, NULL
);
2704 ASSERT(arc_buf_is_shared(lastbuf
) ||
2705 HDR_GET_COMPRESS(hdr
) != ZIO_COMPRESS_OFF
);
2708 if (hdr
->b_l1hdr
.b_bufcnt
== 0)
2709 arc_cksum_free(hdr
);
2711 /* clean up the buf */
2713 kmem_cache_free(buf_cache
, buf
);
2717 arc_hdr_alloc_pabd(arc_buf_hdr_t
*hdr
)
2719 ASSERT3U(HDR_GET_LSIZE(hdr
), >, 0);
2720 ASSERT(HDR_HAS_L1HDR(hdr
));
2721 ASSERT(!HDR_SHARED_DATA(hdr
));
2723 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, ==, NULL
);
2724 hdr
->b_l1hdr
.b_pabd
= arc_get_data_abd(hdr
, arc_hdr_size(hdr
), hdr
);
2725 hdr
->b_l1hdr
.b_byteswap
= DMU_BSWAP_NUMFUNCS
;
2726 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, !=, NULL
);
2728 ARCSTAT_INCR(arcstat_compressed_size
, arc_hdr_size(hdr
));
2729 ARCSTAT_INCR(arcstat_uncompressed_size
, HDR_GET_LSIZE(hdr
));
2733 arc_hdr_free_pabd(arc_buf_hdr_t
*hdr
)
2735 ASSERT(HDR_HAS_L1HDR(hdr
));
2736 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, !=, NULL
);
2739 * If the hdr is currently being written to the l2arc then
2740 * we defer freeing the data by adding it to the l2arc_free_on_write
2741 * list. The l2arc will free the data once it's finished
2742 * writing it to the l2arc device.
2744 if (HDR_L2_WRITING(hdr
)) {
2745 arc_hdr_free_on_write(hdr
);
2746 ARCSTAT_BUMP(arcstat_l2_free_on_write
);
2748 arc_free_data_abd(hdr
, hdr
->b_l1hdr
.b_pabd
,
2749 arc_hdr_size(hdr
), hdr
);
2751 hdr
->b_l1hdr
.b_pabd
= NULL
;
2752 hdr
->b_l1hdr
.b_byteswap
= DMU_BSWAP_NUMFUNCS
;
2754 ARCSTAT_INCR(arcstat_compressed_size
, -arc_hdr_size(hdr
));
2755 ARCSTAT_INCR(arcstat_uncompressed_size
, -HDR_GET_LSIZE(hdr
));
2758 static arc_buf_hdr_t
*
2759 arc_hdr_alloc(uint64_t spa
, int32_t psize
, int32_t lsize
,
2760 enum zio_compress compression_type
, arc_buf_contents_t type
)
2764 VERIFY(type
== ARC_BUFC_DATA
|| type
== ARC_BUFC_METADATA
);
2766 hdr
= kmem_cache_alloc(hdr_full_cache
, KM_PUSHPAGE
);
2767 ASSERT(HDR_EMPTY(hdr
));
2768 ASSERT3P(hdr
->b_l1hdr
.b_freeze_cksum
, ==, NULL
);
2769 HDR_SET_PSIZE(hdr
, psize
);
2770 HDR_SET_LSIZE(hdr
, lsize
);
2774 arc_hdr_set_flags(hdr
, arc_bufc_to_flags(type
) | ARC_FLAG_HAS_L1HDR
);
2775 arc_hdr_set_compress(hdr
, compression_type
);
2777 hdr
->b_l1hdr
.b_state
= arc_anon
;
2778 hdr
->b_l1hdr
.b_arc_access
= 0;
2779 hdr
->b_l1hdr
.b_bufcnt
= 0;
2780 hdr
->b_l1hdr
.b_buf
= NULL
;
2783 * Allocate the hdr's buffer. This will contain either
2784 * the compressed or uncompressed data depending on the block
2785 * it references and compressed arc enablement.
2787 arc_hdr_alloc_pabd(hdr
);
2788 ASSERT(refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
));
2794 * Transition between the two allocation states for the arc_buf_hdr struct.
2795 * The arc_buf_hdr struct can be allocated with (hdr_full_cache) or without
2796 * (hdr_l2only_cache) the fields necessary for the L1 cache - the smaller
2797 * version is used when a cache buffer is only in the L2ARC in order to reduce
2800 static arc_buf_hdr_t
*
2801 arc_hdr_realloc(arc_buf_hdr_t
*hdr
, kmem_cache_t
*old
, kmem_cache_t
*new)
2803 arc_buf_hdr_t
*nhdr
;
2804 l2arc_dev_t
*dev
= hdr
->b_l2hdr
.b_dev
;
2806 ASSERT(HDR_HAS_L2HDR(hdr
));
2807 ASSERT((old
== hdr_full_cache
&& new == hdr_l2only_cache
) ||
2808 (old
== hdr_l2only_cache
&& new == hdr_full_cache
));
2810 nhdr
= kmem_cache_alloc(new, KM_PUSHPAGE
);
2812 ASSERT(MUTEX_HELD(HDR_LOCK(hdr
)));
2813 buf_hash_remove(hdr
);
2815 bcopy(hdr
, nhdr
, HDR_L2ONLY_SIZE
);
2817 if (new == hdr_full_cache
) {
2818 arc_hdr_set_flags(nhdr
, ARC_FLAG_HAS_L1HDR
);
2820 * arc_access and arc_change_state need to be aware that a
2821 * header has just come out of L2ARC, so we set its state to
2822 * l2c_only even though it's about to change.
2824 nhdr
->b_l1hdr
.b_state
= arc_l2c_only
;
2826 /* Verify previous threads set to NULL before freeing */
2827 ASSERT3P(nhdr
->b_l1hdr
.b_pabd
, ==, NULL
);
2829 ASSERT3P(hdr
->b_l1hdr
.b_buf
, ==, NULL
);
2830 ASSERT0(hdr
->b_l1hdr
.b_bufcnt
);
2831 ASSERT3P(hdr
->b_l1hdr
.b_freeze_cksum
, ==, NULL
);
2834 * If we've reached here, We must have been called from
2835 * arc_evict_hdr(), as such we should have already been
2836 * removed from any ghost list we were previously on
2837 * (which protects us from racing with arc_evict_state),
2838 * thus no locking is needed during this check.
2840 ASSERT(!multilist_link_active(&hdr
->b_l1hdr
.b_arc_node
));
2843 * A buffer must not be moved into the arc_l2c_only
2844 * state if it's not finished being written out to the
2845 * l2arc device. Otherwise, the b_l1hdr.b_pabd field
2846 * might try to be accessed, even though it was removed.
2848 VERIFY(!HDR_L2_WRITING(hdr
));
2849 VERIFY3P(hdr
->b_l1hdr
.b_pabd
, ==, NULL
);
2851 arc_hdr_clear_flags(nhdr
, ARC_FLAG_HAS_L1HDR
);
2854 * The header has been reallocated so we need to re-insert it into any
2857 (void) buf_hash_insert(nhdr
, NULL
);
2859 ASSERT(list_link_active(&hdr
->b_l2hdr
.b_l2node
));
2861 mutex_enter(&dev
->l2ad_mtx
);
2864 * We must place the realloc'ed header back into the list at
2865 * the same spot. Otherwise, if it's placed earlier in the list,
2866 * l2arc_write_buffers() could find it during the function's
2867 * write phase, and try to write it out to the l2arc.
2869 list_insert_after(&dev
->l2ad_buflist
, hdr
, nhdr
);
2870 list_remove(&dev
->l2ad_buflist
, hdr
);
2872 mutex_exit(&dev
->l2ad_mtx
);
2875 * Since we're using the pointer address as the tag when
2876 * incrementing and decrementing the l2ad_alloc refcount, we
2877 * must remove the old pointer (that we're about to destroy) and
2878 * add the new pointer to the refcount. Otherwise we'd remove
2879 * the wrong pointer address when calling arc_hdr_destroy() later.
2882 (void) refcount_remove_many(&dev
->l2ad_alloc
, arc_hdr_size(hdr
), hdr
);
2883 (void) refcount_add_many(&dev
->l2ad_alloc
, arc_hdr_size(nhdr
), nhdr
);
2885 buf_discard_identity(hdr
);
2886 kmem_cache_free(old
, hdr
);
2892 * Allocate a new arc_buf_hdr_t and arc_buf_t and return the buf to the caller.
2893 * The buf is returned thawed since we expect the consumer to modify it.
2896 arc_alloc_buf(spa_t
*spa
, void *tag
, arc_buf_contents_t type
, int32_t size
)
2899 arc_buf_hdr_t
*hdr
= arc_hdr_alloc(spa_load_guid(spa
), size
, size
,
2900 ZIO_COMPRESS_OFF
, type
);
2901 ASSERT(!MUTEX_HELD(HDR_LOCK(hdr
)));
2904 VERIFY0(arc_buf_alloc_impl(hdr
, tag
, B_FALSE
, B_FALSE
, &buf
));
2911 * Allocate a compressed buf in the same manner as arc_alloc_buf. Don't use this
2912 * for bufs containing metadata.
2915 arc_alloc_compressed_buf(spa_t
*spa
, void *tag
, uint64_t psize
, uint64_t lsize
,
2916 enum zio_compress compression_type
)
2920 ASSERT3U(lsize
, >, 0);
2921 ASSERT3U(lsize
, >=, psize
);
2922 ASSERT(compression_type
> ZIO_COMPRESS_OFF
);
2923 ASSERT(compression_type
< ZIO_COMPRESS_FUNCTIONS
);
2925 hdr
= arc_hdr_alloc(spa_load_guid(spa
), psize
, lsize
,
2926 compression_type
, ARC_BUFC_DATA
);
2927 ASSERT(!MUTEX_HELD(HDR_LOCK(hdr
)));
2930 VERIFY0(arc_buf_alloc_impl(hdr
, tag
, B_TRUE
, B_FALSE
, &buf
));
2932 ASSERT3P(hdr
->b_l1hdr
.b_freeze_cksum
, ==, NULL
);
2934 if (!arc_buf_is_shared(buf
)) {
2936 * To ensure that the hdr has the correct data in it if we call
2937 * arc_decompress() on this buf before it's been written to
2938 * disk, it's easiest if we just set up sharing between the
2941 ASSERT(!abd_is_linear(hdr
->b_l1hdr
.b_pabd
));
2942 arc_hdr_free_pabd(hdr
);
2943 arc_share_buf(hdr
, buf
);
2950 arc_hdr_l2hdr_destroy(arc_buf_hdr_t
*hdr
)
2952 l2arc_buf_hdr_t
*l2hdr
= &hdr
->b_l2hdr
;
2953 l2arc_dev_t
*dev
= l2hdr
->b_dev
;
2954 uint64_t asize
= arc_hdr_size(hdr
);
2956 ASSERT(MUTEX_HELD(&dev
->l2ad_mtx
));
2957 ASSERT(HDR_HAS_L2HDR(hdr
));
2959 list_remove(&dev
->l2ad_buflist
, hdr
);
2961 ARCSTAT_INCR(arcstat_l2_asize
, -asize
);
2962 ARCSTAT_INCR(arcstat_l2_size
, -HDR_GET_LSIZE(hdr
));
2964 vdev_space_update(dev
->l2ad_vdev
, -asize
, 0, 0);
2966 (void) refcount_remove_many(&dev
->l2ad_alloc
, asize
, hdr
);
2967 arc_hdr_clear_flags(hdr
, ARC_FLAG_HAS_L2HDR
);
2971 arc_hdr_destroy(arc_buf_hdr_t
*hdr
)
2973 if (HDR_HAS_L1HDR(hdr
)) {
2974 ASSERT(hdr
->b_l1hdr
.b_buf
== NULL
||
2975 hdr
->b_l1hdr
.b_bufcnt
> 0);
2976 ASSERT(refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
));
2977 ASSERT3P(hdr
->b_l1hdr
.b_state
, ==, arc_anon
);
2979 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
2980 ASSERT(!HDR_IN_HASH_TABLE(hdr
));
2982 if (!HDR_EMPTY(hdr
))
2983 buf_discard_identity(hdr
);
2985 if (HDR_HAS_L2HDR(hdr
)) {
2986 l2arc_dev_t
*dev
= hdr
->b_l2hdr
.b_dev
;
2987 boolean_t buflist_held
= MUTEX_HELD(&dev
->l2ad_mtx
);
2990 mutex_enter(&dev
->l2ad_mtx
);
2993 * Even though we checked this conditional above, we
2994 * need to check this again now that we have the
2995 * l2ad_mtx. This is because we could be racing with
2996 * another thread calling l2arc_evict() which might have
2997 * destroyed this header's L2 portion as we were waiting
2998 * to acquire the l2ad_mtx. If that happens, we don't
2999 * want to re-destroy the header's L2 portion.
3001 if (HDR_HAS_L2HDR(hdr
))
3002 arc_hdr_l2hdr_destroy(hdr
);
3005 mutex_exit(&dev
->l2ad_mtx
);
3008 if (HDR_HAS_L1HDR(hdr
)) {
3009 arc_cksum_free(hdr
);
3011 while (hdr
->b_l1hdr
.b_buf
!= NULL
)
3012 arc_buf_destroy_impl(hdr
->b_l1hdr
.b_buf
);
3014 if (hdr
->b_l1hdr
.b_pabd
!= NULL
)
3015 arc_hdr_free_pabd(hdr
);
3018 ASSERT3P(hdr
->b_hash_next
, ==, NULL
);
3019 if (HDR_HAS_L1HDR(hdr
)) {
3020 ASSERT(!multilist_link_active(&hdr
->b_l1hdr
.b_arc_node
));
3021 ASSERT3P(hdr
->b_l1hdr
.b_acb
, ==, NULL
);
3022 kmem_cache_free(hdr_full_cache
, hdr
);
3024 kmem_cache_free(hdr_l2only_cache
, hdr
);
3029 arc_buf_destroy(arc_buf_t
*buf
, void* tag
)
3031 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
3032 kmutex_t
*hash_lock
= HDR_LOCK(hdr
);
3034 if (hdr
->b_l1hdr
.b_state
== arc_anon
) {
3035 ASSERT3U(hdr
->b_l1hdr
.b_bufcnt
, ==, 1);
3036 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
3037 VERIFY0(remove_reference(hdr
, NULL
, tag
));
3038 arc_hdr_destroy(hdr
);
3042 mutex_enter(hash_lock
);
3043 ASSERT3P(hdr
, ==, buf
->b_hdr
);
3044 ASSERT(hdr
->b_l1hdr
.b_bufcnt
> 0);
3045 ASSERT3P(hash_lock
, ==, HDR_LOCK(hdr
));
3046 ASSERT3P(hdr
->b_l1hdr
.b_state
, !=, arc_anon
);
3047 ASSERT3P(buf
->b_data
, !=, NULL
);
3049 (void) remove_reference(hdr
, hash_lock
, tag
);
3050 arc_buf_destroy_impl(buf
);
3051 mutex_exit(hash_lock
);
3055 * Evict the arc_buf_hdr that is provided as a parameter. The resultant
3056 * state of the header is dependent on its state prior to entering this
3057 * function. The following transitions are possible:
3059 * - arc_mru -> arc_mru_ghost
3060 * - arc_mfu -> arc_mfu_ghost
3061 * - arc_mru_ghost -> arc_l2c_only
3062 * - arc_mru_ghost -> deleted
3063 * - arc_mfu_ghost -> arc_l2c_only
3064 * - arc_mfu_ghost -> deleted
3067 arc_evict_hdr(arc_buf_hdr_t
*hdr
, kmutex_t
*hash_lock
)
3069 arc_state_t
*evicted_state
, *state
;
3070 int64_t bytes_evicted
= 0;
3072 ASSERT(MUTEX_HELD(hash_lock
));
3073 ASSERT(HDR_HAS_L1HDR(hdr
));
3075 state
= hdr
->b_l1hdr
.b_state
;
3076 if (GHOST_STATE(state
)) {
3077 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
3078 ASSERT3P(hdr
->b_l1hdr
.b_buf
, ==, NULL
);
3081 * l2arc_write_buffers() relies on a header's L1 portion
3082 * (i.e. its b_pabd field) during it's write phase.
3083 * Thus, we cannot push a header onto the arc_l2c_only
3084 * state (removing its L1 piece) until the header is
3085 * done being written to the l2arc.
3087 if (HDR_HAS_L2HDR(hdr
) && HDR_L2_WRITING(hdr
)) {
3088 ARCSTAT_BUMP(arcstat_evict_l2_skip
);
3089 return (bytes_evicted
);
3092 ARCSTAT_BUMP(arcstat_deleted
);
3093 bytes_evicted
+= HDR_GET_LSIZE(hdr
);
3095 DTRACE_PROBE1(arc__delete
, arc_buf_hdr_t
*, hdr
);
3097 if (HDR_HAS_L2HDR(hdr
)) {
3098 ASSERT(hdr
->b_l1hdr
.b_pabd
== NULL
);
3100 * This buffer is cached on the 2nd Level ARC;
3101 * don't destroy the header.
3103 arc_change_state(arc_l2c_only
, hdr
, hash_lock
);
3105 * dropping from L1+L2 cached to L2-only,
3106 * realloc to remove the L1 header.
3108 hdr
= arc_hdr_realloc(hdr
, hdr_full_cache
,
3111 arc_change_state(arc_anon
, hdr
, hash_lock
);
3112 arc_hdr_destroy(hdr
);
3114 return (bytes_evicted
);
3117 ASSERT(state
== arc_mru
|| state
== arc_mfu
);
3118 evicted_state
= (state
== arc_mru
) ? arc_mru_ghost
: arc_mfu_ghost
;
3120 /* prefetch buffers have a minimum lifespan */
3121 if (HDR_IO_IN_PROGRESS(hdr
) ||
3122 ((hdr
->b_flags
& (ARC_FLAG_PREFETCH
| ARC_FLAG_INDIRECT
)) &&
3123 ddi_get_lbolt() - hdr
->b_l1hdr
.b_arc_access
<
3124 arc_min_prefetch_lifespan
)) {
3125 ARCSTAT_BUMP(arcstat_evict_skip
);
3126 return (bytes_evicted
);
3129 ASSERT0(refcount_count(&hdr
->b_l1hdr
.b_refcnt
));
3130 while (hdr
->b_l1hdr
.b_buf
) {
3131 arc_buf_t
*buf
= hdr
->b_l1hdr
.b_buf
;
3132 if (!mutex_tryenter(&buf
->b_evict_lock
)) {
3133 ARCSTAT_BUMP(arcstat_mutex_miss
);
3136 if (buf
->b_data
!= NULL
)
3137 bytes_evicted
+= HDR_GET_LSIZE(hdr
);
3138 mutex_exit(&buf
->b_evict_lock
);
3139 arc_buf_destroy_impl(buf
);
3142 if (HDR_HAS_L2HDR(hdr
)) {
3143 ARCSTAT_INCR(arcstat_evict_l2_cached
, HDR_GET_LSIZE(hdr
));
3145 if (l2arc_write_eligible(hdr
->b_spa
, hdr
)) {
3146 ARCSTAT_INCR(arcstat_evict_l2_eligible
,
3147 HDR_GET_LSIZE(hdr
));
3149 ARCSTAT_INCR(arcstat_evict_l2_ineligible
,
3150 HDR_GET_LSIZE(hdr
));
3154 if (hdr
->b_l1hdr
.b_bufcnt
== 0) {
3155 arc_cksum_free(hdr
);
3157 bytes_evicted
+= arc_hdr_size(hdr
);
3160 * If this hdr is being evicted and has a compressed
3161 * buffer then we discard it here before we change states.
3162 * This ensures that the accounting is updated correctly
3163 * in arc_free_data_impl().
3165 arc_hdr_free_pabd(hdr
);
3167 arc_change_state(evicted_state
, hdr
, hash_lock
);
3168 ASSERT(HDR_IN_HASH_TABLE(hdr
));
3169 arc_hdr_set_flags(hdr
, ARC_FLAG_IN_HASH_TABLE
);
3170 DTRACE_PROBE1(arc__evict
, arc_buf_hdr_t
*, hdr
);
3173 return (bytes_evicted
);
3177 arc_evict_state_impl(multilist_t
*ml
, int idx
, arc_buf_hdr_t
*marker
,
3178 uint64_t spa
, int64_t bytes
)
3180 multilist_sublist_t
*mls
;
3181 uint64_t bytes_evicted
= 0;
3183 kmutex_t
*hash_lock
;
3184 int evict_count
= 0;
3186 ASSERT3P(marker
, !=, NULL
);
3187 IMPLY(bytes
< 0, bytes
== ARC_EVICT_ALL
);
3189 mls
= multilist_sublist_lock(ml
, idx
);
3191 for (hdr
= multilist_sublist_prev(mls
, marker
); hdr
!= NULL
;
3192 hdr
= multilist_sublist_prev(mls
, marker
)) {
3193 if ((bytes
!= ARC_EVICT_ALL
&& bytes_evicted
>= bytes
) ||
3194 (evict_count
>= zfs_arc_evict_batch_limit
))
3198 * To keep our iteration location, move the marker
3199 * forward. Since we're not holding hdr's hash lock, we
3200 * must be very careful and not remove 'hdr' from the
3201 * sublist. Otherwise, other consumers might mistake the
3202 * 'hdr' as not being on a sublist when they call the
3203 * multilist_link_active() function (they all rely on
3204 * the hash lock protecting concurrent insertions and
3205 * removals). multilist_sublist_move_forward() was
3206 * specifically implemented to ensure this is the case
3207 * (only 'marker' will be removed and re-inserted).
3209 multilist_sublist_move_forward(mls
, marker
);
3212 * The only case where the b_spa field should ever be
3213 * zero, is the marker headers inserted by
3214 * arc_evict_state(). It's possible for multiple threads
3215 * to be calling arc_evict_state() concurrently (e.g.
3216 * dsl_pool_close() and zio_inject_fault()), so we must
3217 * skip any markers we see from these other threads.
3219 if (hdr
->b_spa
== 0)
3222 /* we're only interested in evicting buffers of a certain spa */
3223 if (spa
!= 0 && hdr
->b_spa
!= spa
) {
3224 ARCSTAT_BUMP(arcstat_evict_skip
);
3228 hash_lock
= HDR_LOCK(hdr
);
3231 * We aren't calling this function from any code path
3232 * that would already be holding a hash lock, so we're
3233 * asserting on this assumption to be defensive in case
3234 * this ever changes. Without this check, it would be
3235 * possible to incorrectly increment arcstat_mutex_miss
3236 * below (e.g. if the code changed such that we called
3237 * this function with a hash lock held).
3239 ASSERT(!MUTEX_HELD(hash_lock
));
3241 if (mutex_tryenter(hash_lock
)) {
3242 uint64_t evicted
= arc_evict_hdr(hdr
, hash_lock
);
3243 mutex_exit(hash_lock
);
3245 bytes_evicted
+= evicted
;
3248 * If evicted is zero, arc_evict_hdr() must have
3249 * decided to skip this header, don't increment
3250 * evict_count in this case.
3256 * If arc_size isn't overflowing, signal any
3257 * threads that might happen to be waiting.
3259 * For each header evicted, we wake up a single
3260 * thread. If we used cv_broadcast, we could
3261 * wake up "too many" threads causing arc_size
3262 * to significantly overflow arc_c; since
3263 * arc_get_data_impl() doesn't check for overflow
3264 * when it's woken up (it doesn't because it's
3265 * possible for the ARC to be overflowing while
3266 * full of un-evictable buffers, and the
3267 * function should proceed in this case).
3269 * If threads are left sleeping, due to not
3270 * using cv_broadcast, they will be woken up
3271 * just before arc_reclaim_thread() sleeps.
3273 mutex_enter(&arc_reclaim_lock
);
3274 if (!arc_is_overflowing())
3275 cv_signal(&arc_reclaim_waiters_cv
);
3276 mutex_exit(&arc_reclaim_lock
);
3278 ARCSTAT_BUMP(arcstat_mutex_miss
);
3282 multilist_sublist_unlock(mls
);
3284 return (bytes_evicted
);
3288 * Evict buffers from the given arc state, until we've removed the
3289 * specified number of bytes. Move the removed buffers to the
3290 * appropriate evict state.
3292 * This function makes a "best effort". It skips over any buffers
3293 * it can't get a hash_lock on, and so, may not catch all candidates.
3294 * It may also return without evicting as much space as requested.
3296 * If bytes is specified using the special value ARC_EVICT_ALL, this
3297 * will evict all available (i.e. unlocked and evictable) buffers from
3298 * the given arc state; which is used by arc_flush().
3301 arc_evict_state(arc_state_t
*state
, uint64_t spa
, int64_t bytes
,
3302 arc_buf_contents_t type
)
3304 uint64_t total_evicted
= 0;
3305 multilist_t
*ml
= state
->arcs_list
[type
];
3307 arc_buf_hdr_t
**markers
;
3310 IMPLY(bytes
< 0, bytes
== ARC_EVICT_ALL
);
3312 num_sublists
= multilist_get_num_sublists(ml
);
3315 * If we've tried to evict from each sublist, made some
3316 * progress, but still have not hit the target number of bytes
3317 * to evict, we want to keep trying. The markers allow us to
3318 * pick up where we left off for each individual sublist, rather
3319 * than starting from the tail each time.
3321 markers
= kmem_zalloc(sizeof (*markers
) * num_sublists
, KM_SLEEP
);
3322 for (i
= 0; i
< num_sublists
; i
++) {
3323 multilist_sublist_t
*mls
;
3325 markers
[i
] = kmem_cache_alloc(hdr_full_cache
, KM_SLEEP
);
3328 * A b_spa of 0 is used to indicate that this header is
3329 * a marker. This fact is used in arc_adjust_type() and
3330 * arc_evict_state_impl().
3332 markers
[i
]->b_spa
= 0;
3334 mls
= multilist_sublist_lock(ml
, i
);
3335 multilist_sublist_insert_tail(mls
, markers
[i
]);
3336 multilist_sublist_unlock(mls
);
3340 * While we haven't hit our target number of bytes to evict, or
3341 * we're evicting all available buffers.
3343 while (total_evicted
< bytes
|| bytes
== ARC_EVICT_ALL
) {
3344 int sublist_idx
= multilist_get_random_index(ml
);
3345 uint64_t scan_evicted
= 0;
3348 * Try to reduce pinned dnodes with a floor of arc_dnode_limit.
3349 * Request that 10% of the LRUs be scanned by the superblock
3352 if (type
== ARC_BUFC_DATA
&& arc_dnode_size
> arc_dnode_limit
)
3353 arc_prune_async((arc_dnode_size
- arc_dnode_limit
) /
3354 sizeof (dnode_t
) / zfs_arc_dnode_reduce_percent
);
3357 * Start eviction using a randomly selected sublist,
3358 * this is to try and evenly balance eviction across all
3359 * sublists. Always starting at the same sublist
3360 * (e.g. index 0) would cause evictions to favor certain
3361 * sublists over others.
3363 for (i
= 0; i
< num_sublists
; i
++) {
3364 uint64_t bytes_remaining
;
3365 uint64_t bytes_evicted
;
3367 if (bytes
== ARC_EVICT_ALL
)
3368 bytes_remaining
= ARC_EVICT_ALL
;
3369 else if (total_evicted
< bytes
)
3370 bytes_remaining
= bytes
- total_evicted
;
3374 bytes_evicted
= arc_evict_state_impl(ml
, sublist_idx
,
3375 markers
[sublist_idx
], spa
, bytes_remaining
);
3377 scan_evicted
+= bytes_evicted
;
3378 total_evicted
+= bytes_evicted
;
3380 /* we've reached the end, wrap to the beginning */
3381 if (++sublist_idx
>= num_sublists
)
3386 * If we didn't evict anything during this scan, we have
3387 * no reason to believe we'll evict more during another
3388 * scan, so break the loop.
3390 if (scan_evicted
== 0) {
3391 /* This isn't possible, let's make that obvious */
3392 ASSERT3S(bytes
, !=, 0);
3395 * When bytes is ARC_EVICT_ALL, the only way to
3396 * break the loop is when scan_evicted is zero.
3397 * In that case, we actually have evicted enough,
3398 * so we don't want to increment the kstat.
3400 if (bytes
!= ARC_EVICT_ALL
) {
3401 ASSERT3S(total_evicted
, <, bytes
);
3402 ARCSTAT_BUMP(arcstat_evict_not_enough
);
3409 for (i
= 0; i
< num_sublists
; i
++) {
3410 multilist_sublist_t
*mls
= multilist_sublist_lock(ml
, i
);
3411 multilist_sublist_remove(mls
, markers
[i
]);
3412 multilist_sublist_unlock(mls
);
3414 kmem_cache_free(hdr_full_cache
, markers
[i
]);
3416 kmem_free(markers
, sizeof (*markers
) * num_sublists
);
3418 return (total_evicted
);
3422 * Flush all "evictable" data of the given type from the arc state
3423 * specified. This will not evict any "active" buffers (i.e. referenced).
3425 * When 'retry' is set to B_FALSE, the function will make a single pass
3426 * over the state and evict any buffers that it can. Since it doesn't
3427 * continually retry the eviction, it might end up leaving some buffers
3428 * in the ARC due to lock misses.
3430 * When 'retry' is set to B_TRUE, the function will continually retry the
3431 * eviction until *all* evictable buffers have been removed from the
3432 * state. As a result, if concurrent insertions into the state are
3433 * allowed (e.g. if the ARC isn't shutting down), this function might
3434 * wind up in an infinite loop, continually trying to evict buffers.
3437 arc_flush_state(arc_state_t
*state
, uint64_t spa
, arc_buf_contents_t type
,
3440 uint64_t evicted
= 0;
3442 while (refcount_count(&state
->arcs_esize
[type
]) != 0) {
3443 evicted
+= arc_evict_state(state
, spa
, ARC_EVICT_ALL
, type
);
3453 * Helper function for arc_prune_async() it is responsible for safely
3454 * handling the execution of a registered arc_prune_func_t.
3457 arc_prune_task(void *ptr
)
3459 arc_prune_t
*ap
= (arc_prune_t
*)ptr
;
3460 arc_prune_func_t
*func
= ap
->p_pfunc
;
3463 func(ap
->p_adjust
, ap
->p_private
);
3465 refcount_remove(&ap
->p_refcnt
, func
);
3469 * Notify registered consumers they must drop holds on a portion of the ARC
3470 * buffered they reference. This provides a mechanism to ensure the ARC can
3471 * honor the arc_meta_limit and reclaim otherwise pinned ARC buffers. This
3472 * is analogous to dnlc_reduce_cache() but more generic.
3474 * This operation is performed asynchronously so it may be safely called
3475 * in the context of the arc_reclaim_thread(). A reference is taken here
3476 * for each registered arc_prune_t and the arc_prune_task() is responsible
3477 * for releasing it once the registered arc_prune_func_t has completed.
3480 arc_prune_async(int64_t adjust
)
3484 mutex_enter(&arc_prune_mtx
);
3485 for (ap
= list_head(&arc_prune_list
); ap
!= NULL
;
3486 ap
= list_next(&arc_prune_list
, ap
)) {
3488 if (refcount_count(&ap
->p_refcnt
) >= 2)
3491 refcount_add(&ap
->p_refcnt
, ap
->p_pfunc
);
3492 ap
->p_adjust
= adjust
;
3493 if (taskq_dispatch(arc_prune_taskq
, arc_prune_task
,
3494 ap
, TQ_SLEEP
) == TASKQID_INVALID
) {
3495 refcount_remove(&ap
->p_refcnt
, ap
->p_pfunc
);
3498 ARCSTAT_BUMP(arcstat_prune
);
3500 mutex_exit(&arc_prune_mtx
);
3504 * Evict the specified number of bytes from the state specified,
3505 * restricting eviction to the spa and type given. This function
3506 * prevents us from trying to evict more from a state's list than
3507 * is "evictable", and to skip evicting altogether when passed a
3508 * negative value for "bytes". In contrast, arc_evict_state() will
3509 * evict everything it can, when passed a negative value for "bytes".
3512 arc_adjust_impl(arc_state_t
*state
, uint64_t spa
, int64_t bytes
,
3513 arc_buf_contents_t type
)
3517 if (bytes
> 0 && refcount_count(&state
->arcs_esize
[type
]) > 0) {
3518 delta
= MIN(refcount_count(&state
->arcs_esize
[type
]), bytes
);
3519 return (arc_evict_state(state
, spa
, delta
, type
));
3526 * The goal of this function is to evict enough meta data buffers from the
3527 * ARC in order to enforce the arc_meta_limit. Achieving this is slightly
3528 * more complicated than it appears because it is common for data buffers
3529 * to have holds on meta data buffers. In addition, dnode meta data buffers
3530 * will be held by the dnodes in the block preventing them from being freed.
3531 * This means we can't simply traverse the ARC and expect to always find
3532 * enough unheld meta data buffer to release.
3534 * Therefore, this function has been updated to make alternating passes
3535 * over the ARC releasing data buffers and then newly unheld meta data
3536 * buffers. This ensures forward progress is maintained and arc_meta_used
3537 * will decrease. Normally this is sufficient, but if required the ARC
3538 * will call the registered prune callbacks causing dentry and inodes to
3539 * be dropped from the VFS cache. This will make dnode meta data buffers
3540 * available for reclaim.
3543 arc_adjust_meta_balanced(void)
3545 int64_t delta
, prune
= 0, adjustmnt
;
3546 uint64_t total_evicted
= 0;
3547 arc_buf_contents_t type
= ARC_BUFC_DATA
;
3548 int restarts
= MAX(zfs_arc_meta_adjust_restarts
, 0);
3552 * This slightly differs than the way we evict from the mru in
3553 * arc_adjust because we don't have a "target" value (i.e. no
3554 * "meta" arc_p). As a result, I think we can completely
3555 * cannibalize the metadata in the MRU before we evict the
3556 * metadata from the MFU. I think we probably need to implement a
3557 * "metadata arc_p" value to do this properly.
3559 adjustmnt
= arc_meta_used
- arc_meta_limit
;
3561 if (adjustmnt
> 0 && refcount_count(&arc_mru
->arcs_esize
[type
]) > 0) {
3562 delta
= MIN(refcount_count(&arc_mru
->arcs_esize
[type
]),
3564 total_evicted
+= arc_adjust_impl(arc_mru
, 0, delta
, type
);
3569 * We can't afford to recalculate adjustmnt here. If we do,
3570 * new metadata buffers can sneak into the MRU or ANON lists,
3571 * thus penalize the MFU metadata. Although the fudge factor is
3572 * small, it has been empirically shown to be significant for
3573 * certain workloads (e.g. creating many empty directories). As
3574 * such, we use the original calculation for adjustmnt, and
3575 * simply decrement the amount of data evicted from the MRU.
3578 if (adjustmnt
> 0 && refcount_count(&arc_mfu
->arcs_esize
[type
]) > 0) {
3579 delta
= MIN(refcount_count(&arc_mfu
->arcs_esize
[type
]),
3581 total_evicted
+= arc_adjust_impl(arc_mfu
, 0, delta
, type
);
3584 adjustmnt
= arc_meta_used
- arc_meta_limit
;
3586 if (adjustmnt
> 0 &&
3587 refcount_count(&arc_mru_ghost
->arcs_esize
[type
]) > 0) {
3588 delta
= MIN(adjustmnt
,
3589 refcount_count(&arc_mru_ghost
->arcs_esize
[type
]));
3590 total_evicted
+= arc_adjust_impl(arc_mru_ghost
, 0, delta
, type
);
3594 if (adjustmnt
> 0 &&
3595 refcount_count(&arc_mfu_ghost
->arcs_esize
[type
]) > 0) {
3596 delta
= MIN(adjustmnt
,
3597 refcount_count(&arc_mfu_ghost
->arcs_esize
[type
]));
3598 total_evicted
+= arc_adjust_impl(arc_mfu_ghost
, 0, delta
, type
);
3602 * If after attempting to make the requested adjustment to the ARC
3603 * the meta limit is still being exceeded then request that the
3604 * higher layers drop some cached objects which have holds on ARC
3605 * meta buffers. Requests to the upper layers will be made with
3606 * increasingly large scan sizes until the ARC is below the limit.
3608 if (arc_meta_used
> arc_meta_limit
) {
3609 if (type
== ARC_BUFC_DATA
) {
3610 type
= ARC_BUFC_METADATA
;
3612 type
= ARC_BUFC_DATA
;
3614 if (zfs_arc_meta_prune
) {
3615 prune
+= zfs_arc_meta_prune
;
3616 arc_prune_async(prune
);
3625 return (total_evicted
);
3629 * Evict metadata buffers from the cache, such that arc_meta_used is
3630 * capped by the arc_meta_limit tunable.
3633 arc_adjust_meta_only(void)
3635 uint64_t total_evicted
= 0;
3639 * If we're over the meta limit, we want to evict enough
3640 * metadata to get back under the meta limit. We don't want to
3641 * evict so much that we drop the MRU below arc_p, though. If
3642 * we're over the meta limit more than we're over arc_p, we
3643 * evict some from the MRU here, and some from the MFU below.
3645 target
= MIN((int64_t)(arc_meta_used
- arc_meta_limit
),
3646 (int64_t)(refcount_count(&arc_anon
->arcs_size
) +
3647 refcount_count(&arc_mru
->arcs_size
) - arc_p
));
3649 total_evicted
+= arc_adjust_impl(arc_mru
, 0, target
, ARC_BUFC_METADATA
);
3652 * Similar to the above, we want to evict enough bytes to get us
3653 * below the meta limit, but not so much as to drop us below the
3654 * space allotted to the MFU (which is defined as arc_c - arc_p).
3656 target
= MIN((int64_t)(arc_meta_used
- arc_meta_limit
),
3657 (int64_t)(refcount_count(&arc_mfu
->arcs_size
) - (arc_c
- arc_p
)));
3659 total_evicted
+= arc_adjust_impl(arc_mfu
, 0, target
, ARC_BUFC_METADATA
);
3661 return (total_evicted
);
3665 arc_adjust_meta(void)
3667 if (zfs_arc_meta_strategy
== ARC_STRATEGY_META_ONLY
)
3668 return (arc_adjust_meta_only());
3670 return (arc_adjust_meta_balanced());
3674 * Return the type of the oldest buffer in the given arc state
3676 * This function will select a random sublist of type ARC_BUFC_DATA and
3677 * a random sublist of type ARC_BUFC_METADATA. The tail of each sublist
3678 * is compared, and the type which contains the "older" buffer will be
3681 static arc_buf_contents_t
3682 arc_adjust_type(arc_state_t
*state
)
3684 multilist_t
*data_ml
= state
->arcs_list
[ARC_BUFC_DATA
];
3685 multilist_t
*meta_ml
= state
->arcs_list
[ARC_BUFC_METADATA
];
3686 int data_idx
= multilist_get_random_index(data_ml
);
3687 int meta_idx
= multilist_get_random_index(meta_ml
);
3688 multilist_sublist_t
*data_mls
;
3689 multilist_sublist_t
*meta_mls
;
3690 arc_buf_contents_t type
;
3691 arc_buf_hdr_t
*data_hdr
;
3692 arc_buf_hdr_t
*meta_hdr
;
3695 * We keep the sublist lock until we're finished, to prevent
3696 * the headers from being destroyed via arc_evict_state().
3698 data_mls
= multilist_sublist_lock(data_ml
, data_idx
);
3699 meta_mls
= multilist_sublist_lock(meta_ml
, meta_idx
);
3702 * These two loops are to ensure we skip any markers that
3703 * might be at the tail of the lists due to arc_evict_state().
3706 for (data_hdr
= multilist_sublist_tail(data_mls
); data_hdr
!= NULL
;
3707 data_hdr
= multilist_sublist_prev(data_mls
, data_hdr
)) {
3708 if (data_hdr
->b_spa
!= 0)
3712 for (meta_hdr
= multilist_sublist_tail(meta_mls
); meta_hdr
!= NULL
;
3713 meta_hdr
= multilist_sublist_prev(meta_mls
, meta_hdr
)) {
3714 if (meta_hdr
->b_spa
!= 0)
3718 if (data_hdr
== NULL
&& meta_hdr
== NULL
) {
3719 type
= ARC_BUFC_DATA
;
3720 } else if (data_hdr
== NULL
) {
3721 ASSERT3P(meta_hdr
, !=, NULL
);
3722 type
= ARC_BUFC_METADATA
;
3723 } else if (meta_hdr
== NULL
) {
3724 ASSERT3P(data_hdr
, !=, NULL
);
3725 type
= ARC_BUFC_DATA
;
3727 ASSERT3P(data_hdr
, !=, NULL
);
3728 ASSERT3P(meta_hdr
, !=, NULL
);
3730 /* The headers can't be on the sublist without an L1 header */
3731 ASSERT(HDR_HAS_L1HDR(data_hdr
));
3732 ASSERT(HDR_HAS_L1HDR(meta_hdr
));
3734 if (data_hdr
->b_l1hdr
.b_arc_access
<
3735 meta_hdr
->b_l1hdr
.b_arc_access
) {
3736 type
= ARC_BUFC_DATA
;
3738 type
= ARC_BUFC_METADATA
;
3742 multilist_sublist_unlock(meta_mls
);
3743 multilist_sublist_unlock(data_mls
);
3749 * Evict buffers from the cache, such that arc_size is capped by arc_c.
3754 uint64_t total_evicted
= 0;
3759 * If we're over arc_meta_limit, we want to correct that before
3760 * potentially evicting data buffers below.
3762 total_evicted
+= arc_adjust_meta();
3767 * If we're over the target cache size, we want to evict enough
3768 * from the list to get back to our target size. We don't want
3769 * to evict too much from the MRU, such that it drops below
3770 * arc_p. So, if we're over our target cache size more than
3771 * the MRU is over arc_p, we'll evict enough to get back to
3772 * arc_p here, and then evict more from the MFU below.
3774 target
= MIN((int64_t)(arc_size
- arc_c
),
3775 (int64_t)(refcount_count(&arc_anon
->arcs_size
) +
3776 refcount_count(&arc_mru
->arcs_size
) + arc_meta_used
- arc_p
));
3779 * If we're below arc_meta_min, always prefer to evict data.
3780 * Otherwise, try to satisfy the requested number of bytes to
3781 * evict from the type which contains older buffers; in an
3782 * effort to keep newer buffers in the cache regardless of their
3783 * type. If we cannot satisfy the number of bytes from this
3784 * type, spill over into the next type.
3786 if (arc_adjust_type(arc_mru
) == ARC_BUFC_METADATA
&&
3787 arc_meta_used
> arc_meta_min
) {
3788 bytes
= arc_adjust_impl(arc_mru
, 0, target
, ARC_BUFC_METADATA
);
3789 total_evicted
+= bytes
;
3792 * If we couldn't evict our target number of bytes from
3793 * metadata, we try to get the rest from data.
3798 arc_adjust_impl(arc_mru
, 0, target
, ARC_BUFC_DATA
);
3800 bytes
= arc_adjust_impl(arc_mru
, 0, target
, ARC_BUFC_DATA
);
3801 total_evicted
+= bytes
;
3804 * If we couldn't evict our target number of bytes from
3805 * data, we try to get the rest from metadata.
3810 arc_adjust_impl(arc_mru
, 0, target
, ARC_BUFC_METADATA
);
3816 * Now that we've tried to evict enough from the MRU to get its
3817 * size back to arc_p, if we're still above the target cache
3818 * size, we evict the rest from the MFU.
3820 target
= arc_size
- arc_c
;
3822 if (arc_adjust_type(arc_mfu
) == ARC_BUFC_METADATA
&&
3823 arc_meta_used
> arc_meta_min
) {
3824 bytes
= arc_adjust_impl(arc_mfu
, 0, target
, ARC_BUFC_METADATA
);
3825 total_evicted
+= bytes
;
3828 * If we couldn't evict our target number of bytes from
3829 * metadata, we try to get the rest from data.
3834 arc_adjust_impl(arc_mfu
, 0, target
, ARC_BUFC_DATA
);
3836 bytes
= arc_adjust_impl(arc_mfu
, 0, target
, ARC_BUFC_DATA
);
3837 total_evicted
+= bytes
;
3840 * If we couldn't evict our target number of bytes from
3841 * data, we try to get the rest from data.
3846 arc_adjust_impl(arc_mfu
, 0, target
, ARC_BUFC_METADATA
);
3850 * Adjust ghost lists
3852 * In addition to the above, the ARC also defines target values
3853 * for the ghost lists. The sum of the mru list and mru ghost
3854 * list should never exceed the target size of the cache, and
3855 * the sum of the mru list, mfu list, mru ghost list, and mfu
3856 * ghost list should never exceed twice the target size of the
3857 * cache. The following logic enforces these limits on the ghost
3858 * caches, and evicts from them as needed.
3860 target
= refcount_count(&arc_mru
->arcs_size
) +
3861 refcount_count(&arc_mru_ghost
->arcs_size
) - arc_c
;
3863 bytes
= arc_adjust_impl(arc_mru_ghost
, 0, target
, ARC_BUFC_DATA
);
3864 total_evicted
+= bytes
;
3869 arc_adjust_impl(arc_mru_ghost
, 0, target
, ARC_BUFC_METADATA
);
3872 * We assume the sum of the mru list and mfu list is less than
3873 * or equal to arc_c (we enforced this above), which means we
3874 * can use the simpler of the two equations below:
3876 * mru + mfu + mru ghost + mfu ghost <= 2 * arc_c
3877 * mru ghost + mfu ghost <= arc_c
3879 target
= refcount_count(&arc_mru_ghost
->arcs_size
) +
3880 refcount_count(&arc_mfu_ghost
->arcs_size
) - arc_c
;
3882 bytes
= arc_adjust_impl(arc_mfu_ghost
, 0, target
, ARC_BUFC_DATA
);
3883 total_evicted
+= bytes
;
3888 arc_adjust_impl(arc_mfu_ghost
, 0, target
, ARC_BUFC_METADATA
);
3890 return (total_evicted
);
3894 arc_flush(spa_t
*spa
, boolean_t retry
)
3899 * If retry is B_TRUE, a spa must not be specified since we have
3900 * no good way to determine if all of a spa's buffers have been
3901 * evicted from an arc state.
3903 ASSERT(!retry
|| spa
== 0);
3906 guid
= spa_load_guid(spa
);
3908 (void) arc_flush_state(arc_mru
, guid
, ARC_BUFC_DATA
, retry
);
3909 (void) arc_flush_state(arc_mru
, guid
, ARC_BUFC_METADATA
, retry
);
3911 (void) arc_flush_state(arc_mfu
, guid
, ARC_BUFC_DATA
, retry
);
3912 (void) arc_flush_state(arc_mfu
, guid
, ARC_BUFC_METADATA
, retry
);
3914 (void) arc_flush_state(arc_mru_ghost
, guid
, ARC_BUFC_DATA
, retry
);
3915 (void) arc_flush_state(arc_mru_ghost
, guid
, ARC_BUFC_METADATA
, retry
);
3917 (void) arc_flush_state(arc_mfu_ghost
, guid
, ARC_BUFC_DATA
, retry
);
3918 (void) arc_flush_state(arc_mfu_ghost
, guid
, ARC_BUFC_METADATA
, retry
);
3922 arc_shrink(int64_t to_free
)
3926 if (c
> to_free
&& c
- to_free
> arc_c_min
) {
3927 arc_c
= c
- to_free
;
3928 atomic_add_64(&arc_p
, -(arc_p
>> arc_shrink_shift
));
3929 if (arc_c
> arc_size
)
3930 arc_c
= MAX(arc_size
, arc_c_min
);
3932 arc_p
= (arc_c
>> 1);
3933 ASSERT(arc_c
>= arc_c_min
);
3934 ASSERT((int64_t)arc_p
>= 0);
3939 if (arc_size
> arc_c
)
3940 (void) arc_adjust();
3944 * Return maximum amount of memory that we could possibly use. Reduced
3945 * to half of all memory in user space which is primarily used for testing.
3948 arc_all_memory(void)
3951 return (MIN(ptob(physmem
),
3952 vmem_size(heap_arena
, VMEM_FREE
| VMEM_ALLOC
)));
3954 return (ptob(physmem
) / 2);
3958 typedef enum free_memory_reason_t
{
3963 FMR_PAGES_PP_MAXIMUM
,
3966 } free_memory_reason_t
;
3968 int64_t last_free_memory
;
3969 free_memory_reason_t last_free_reason
;
3973 * Additional reserve of pages for pp_reserve.
3975 int64_t arc_pages_pp_reserve
= 64;
3978 * Additional reserve of pages for swapfs.
3980 int64_t arc_swapfs_reserve
= 64;
3981 #endif /* _KERNEL */
3984 * Return the amount of memory that can be consumed before reclaim will be
3985 * needed. Positive if there is sufficient free memory, negative indicates
3986 * the amount of memory that needs to be freed up.
3989 arc_available_memory(void)
3991 int64_t lowest
= INT64_MAX
;
3992 free_memory_reason_t r
= FMR_UNKNOWN
;
3994 uint64_t available_memory
= ptob(freemem
);
3997 pgcnt_t needfree
= btop(arc_need_free
);
3998 pgcnt_t lotsfree
= btop(arc_sys_free
);
3999 pgcnt_t desfree
= 0;
4004 MIN(available_memory
, vmem_size(heap_arena
, VMEM_FREE
));
4008 n
= PAGESIZE
* (-needfree
);
4016 * check that we're out of range of the pageout scanner. It starts to
4017 * schedule paging if freemem is less than lotsfree and needfree.
4018 * lotsfree is the high-water mark for pageout, and needfree is the
4019 * number of needed free pages. We add extra pages here to make sure
4020 * the scanner doesn't start up while we're freeing memory.
4022 n
= PAGESIZE
* (btop(available_memory
) - lotsfree
- needfree
- desfree
);
4030 * check to make sure that swapfs has enough space so that anon
4031 * reservations can still succeed. anon_resvmem() checks that the
4032 * availrmem is greater than swapfs_minfree, and the number of reserved
4033 * swap pages. We also add a bit of extra here just to prevent
4034 * circumstances from getting really dire.
4036 n
= PAGESIZE
* (availrmem
- swapfs_minfree
- swapfs_reserve
-
4037 desfree
- arc_swapfs_reserve
);
4040 r
= FMR_SWAPFS_MINFREE
;
4045 * Check that we have enough availrmem that memory locking (e.g., via
4046 * mlock(3C) or memcntl(2)) can still succeed. (pages_pp_maximum
4047 * stores the number of pages that cannot be locked; when availrmem
4048 * drops below pages_pp_maximum, page locking mechanisms such as
4049 * page_pp_lock() will fail.)
4051 n
= PAGESIZE
* (availrmem
- pages_pp_maximum
-
4052 arc_pages_pp_reserve
);
4055 r
= FMR_PAGES_PP_MAXIMUM
;
4061 * If we're on an i386 platform, it's possible that we'll exhaust the
4062 * kernel heap space before we ever run out of available physical
4063 * memory. Most checks of the size of the heap_area compare against
4064 * tune.t_minarmem, which is the minimum available real memory that we
4065 * can have in the system. However, this is generally fixed at 25 pages
4066 * which is so low that it's useless. In this comparison, we seek to
4067 * calculate the total heap-size, and reclaim if more than 3/4ths of the
4068 * heap is allocated. (Or, in the calculation, if less than 1/4th is
4071 n
= vmem_size(heap_arena
, VMEM_FREE
) -
4072 (vmem_size(heap_arena
, VMEM_FREE
| VMEM_ALLOC
) >> 2);
4080 * If zio data pages are being allocated out of a separate heap segment,
4081 * then enforce that the size of available vmem for this arena remains
4082 * above about 1/4th (1/(2^arc_zio_arena_free_shift)) free.
4084 * Note that reducing the arc_zio_arena_free_shift keeps more virtual
4085 * memory (in the zio_arena) free, which can avoid memory
4086 * fragmentation issues.
4088 if (zio_arena
!= NULL
) {
4089 n
= (int64_t)vmem_size(zio_arena
, VMEM_FREE
) -
4090 (vmem_size(zio_arena
, VMEM_ALLOC
) >>
4091 arc_zio_arena_free_shift
);
4098 /* Every 100 calls, free a small amount */
4099 if (spa_get_random(100) == 0)
4101 #endif /* _KERNEL */
4103 last_free_memory
= lowest
;
4104 last_free_reason
= r
;
4110 * Determine if the system is under memory pressure and is asking
4111 * to reclaim memory. A return value of B_TRUE indicates that the system
4112 * is under memory pressure and that the arc should adjust accordingly.
4115 arc_reclaim_needed(void)
4117 return (arc_available_memory() < 0);
4121 arc_kmem_reap_now(void)
4124 kmem_cache_t
*prev_cache
= NULL
;
4125 kmem_cache_t
*prev_data_cache
= NULL
;
4126 extern kmem_cache_t
*zio_buf_cache
[];
4127 extern kmem_cache_t
*zio_data_buf_cache
[];
4128 extern kmem_cache_t
*range_seg_cache
;
4130 if ((arc_meta_used
>= arc_meta_limit
) && zfs_arc_meta_prune
) {
4132 * We are exceeding our meta-data cache limit.
4133 * Prune some entries to release holds on meta-data.
4135 arc_prune_async(zfs_arc_meta_prune
);
4138 for (i
= 0; i
< SPA_MAXBLOCKSIZE
>> SPA_MINBLOCKSHIFT
; i
++) {
4140 /* reach upper limit of cache size on 32-bit */
4141 if (zio_buf_cache
[i
] == NULL
)
4144 if (zio_buf_cache
[i
] != prev_cache
) {
4145 prev_cache
= zio_buf_cache
[i
];
4146 kmem_cache_reap_now(zio_buf_cache
[i
]);
4148 if (zio_data_buf_cache
[i
] != prev_data_cache
) {
4149 prev_data_cache
= zio_data_buf_cache
[i
];
4150 kmem_cache_reap_now(zio_data_buf_cache
[i
]);
4153 kmem_cache_reap_now(buf_cache
);
4154 kmem_cache_reap_now(hdr_full_cache
);
4155 kmem_cache_reap_now(hdr_l2only_cache
);
4156 kmem_cache_reap_now(range_seg_cache
);
4158 if (zio_arena
!= NULL
) {
4160 * Ask the vmem arena to reclaim unused memory from its
4163 vmem_qcache_reap(zio_arena
);
4168 * Threads can block in arc_get_data_impl() waiting for this thread to evict
4169 * enough data and signal them to proceed. When this happens, the threads in
4170 * arc_get_data_impl() are sleeping while holding the hash lock for their
4171 * particular arc header. Thus, we must be careful to never sleep on a
4172 * hash lock in this thread. This is to prevent the following deadlock:
4174 * - Thread A sleeps on CV in arc_get_data_impl() holding hash lock "L",
4175 * waiting for the reclaim thread to signal it.
4177 * - arc_reclaim_thread() tries to acquire hash lock "L" using mutex_enter,
4178 * fails, and goes to sleep forever.
4180 * This possible deadlock is avoided by always acquiring a hash lock
4181 * using mutex_tryenter() from arc_reclaim_thread().
4184 arc_reclaim_thread(void)
4186 fstrans_cookie_t cookie
= spl_fstrans_mark();
4187 hrtime_t growtime
= 0;
4190 CALLB_CPR_INIT(&cpr
, &arc_reclaim_lock
, callb_generic_cpr
, FTAG
);
4192 mutex_enter(&arc_reclaim_lock
);
4193 while (!arc_reclaim_thread_exit
) {
4195 uint64_t evicted
= 0;
4197 arc_tuning_update();
4200 * This is necessary in order for the mdb ::arc dcmd to
4201 * show up to date information. Since the ::arc command
4202 * does not call the kstat's update function, without
4203 * this call, the command may show stale stats for the
4204 * anon, mru, mru_ghost, mfu, and mfu_ghost lists. Even
4205 * with this change, the data might be up to 1 second
4206 * out of date; but that should suffice. The arc_state_t
4207 * structures can be queried directly if more accurate
4208 * information is needed.
4211 if (arc_ksp
!= NULL
)
4212 arc_ksp
->ks_update(arc_ksp
, KSTAT_READ
);
4214 mutex_exit(&arc_reclaim_lock
);
4217 * We call arc_adjust() before (possibly) calling
4218 * arc_kmem_reap_now(), so that we can wake up
4219 * arc_get_data_buf() sooner.
4221 evicted
= arc_adjust();
4223 int64_t free_memory
= arc_available_memory();
4224 if (free_memory
< 0) {
4226 arc_no_grow
= B_TRUE
;
4230 * Wait at least zfs_grow_retry (default 5) seconds
4231 * before considering growing.
4233 growtime
= gethrtime() + SEC2NSEC(arc_grow_retry
);
4235 arc_kmem_reap_now();
4238 * If we are still low on memory, shrink the ARC
4239 * so that we have arc_shrink_min free space.
4241 free_memory
= arc_available_memory();
4243 to_free
= (arc_c
>> arc_shrink_shift
) - free_memory
;
4246 to_free
= MAX(to_free
, arc_need_free
);
4248 arc_shrink(to_free
);
4250 } else if (free_memory
< arc_c
>> arc_no_grow_shift
) {
4251 arc_no_grow
= B_TRUE
;
4252 } else if (gethrtime() >= growtime
) {
4253 arc_no_grow
= B_FALSE
;
4256 mutex_enter(&arc_reclaim_lock
);
4259 * If evicted is zero, we couldn't evict anything via
4260 * arc_adjust(). This could be due to hash lock
4261 * collisions, but more likely due to the majority of
4262 * arc buffers being unevictable. Therefore, even if
4263 * arc_size is above arc_c, another pass is unlikely to
4264 * be helpful and could potentially cause us to enter an
4267 if (arc_size
<= arc_c
|| evicted
== 0) {
4269 * We're either no longer overflowing, or we
4270 * can't evict anything more, so we should wake
4271 * up any threads before we go to sleep and clear
4272 * arc_need_free since nothing more can be done.
4274 cv_broadcast(&arc_reclaim_waiters_cv
);
4278 * Block until signaled, or after one second (we
4279 * might need to perform arc_kmem_reap_now()
4280 * even if we aren't being signalled)
4282 CALLB_CPR_SAFE_BEGIN(&cpr
);
4283 (void) cv_timedwait_sig_hires(&arc_reclaim_thread_cv
,
4284 &arc_reclaim_lock
, SEC2NSEC(1), MSEC2NSEC(1), 0);
4285 CALLB_CPR_SAFE_END(&cpr
, &arc_reclaim_lock
);
4289 arc_reclaim_thread_exit
= B_FALSE
;
4290 cv_broadcast(&arc_reclaim_thread_cv
);
4291 CALLB_CPR_EXIT(&cpr
); /* drops arc_reclaim_lock */
4292 spl_fstrans_unmark(cookie
);
4298 * Determine the amount of memory eligible for eviction contained in the
4299 * ARC. All clean data reported by the ghost lists can always be safely
4300 * evicted. Due to arc_c_min, the same does not hold for all clean data
4301 * contained by the regular mru and mfu lists.
4303 * In the case of the regular mru and mfu lists, we need to report as
4304 * much clean data as possible, such that evicting that same reported
4305 * data will not bring arc_size below arc_c_min. Thus, in certain
4306 * circumstances, the total amount of clean data in the mru and mfu
4307 * lists might not actually be evictable.
4309 * The following two distinct cases are accounted for:
4311 * 1. The sum of the amount of dirty data contained by both the mru and
4312 * mfu lists, plus the ARC's other accounting (e.g. the anon list),
4313 * is greater than or equal to arc_c_min.
4314 * (i.e. amount of dirty data >= arc_c_min)
4316 * This is the easy case; all clean data contained by the mru and mfu
4317 * lists is evictable. Evicting all clean data can only drop arc_size
4318 * to the amount of dirty data, which is greater than arc_c_min.
4320 * 2. The sum of the amount of dirty data contained by both the mru and
4321 * mfu lists, plus the ARC's other accounting (e.g. the anon list),
4322 * is less than arc_c_min.
4323 * (i.e. arc_c_min > amount of dirty data)
4325 * 2.1. arc_size is greater than or equal arc_c_min.
4326 * (i.e. arc_size >= arc_c_min > amount of dirty data)
4328 * In this case, not all clean data from the regular mru and mfu
4329 * lists is actually evictable; we must leave enough clean data
4330 * to keep arc_size above arc_c_min. Thus, the maximum amount of
4331 * evictable data from the two lists combined, is exactly the
4332 * difference between arc_size and arc_c_min.
4334 * 2.2. arc_size is less than arc_c_min
4335 * (i.e. arc_c_min > arc_size > amount of dirty data)
4337 * In this case, none of the data contained in the mru and mfu
4338 * lists is evictable, even if it's clean. Since arc_size is
4339 * already below arc_c_min, evicting any more would only
4340 * increase this negative difference.
4343 arc_evictable_memory(void)
4345 uint64_t arc_clean
=
4346 refcount_count(&arc_mru
->arcs_esize
[ARC_BUFC_DATA
]) +
4347 refcount_count(&arc_mru
->arcs_esize
[ARC_BUFC_METADATA
]) +
4348 refcount_count(&arc_mfu
->arcs_esize
[ARC_BUFC_DATA
]) +
4349 refcount_count(&arc_mfu
->arcs_esize
[ARC_BUFC_METADATA
]);
4350 uint64_t ghost_clean
=
4351 refcount_count(&arc_mru_ghost
->arcs_esize
[ARC_BUFC_DATA
]) +
4352 refcount_count(&arc_mru_ghost
->arcs_esize
[ARC_BUFC_METADATA
]) +
4353 refcount_count(&arc_mfu_ghost
->arcs_esize
[ARC_BUFC_DATA
]) +
4354 refcount_count(&arc_mfu_ghost
->arcs_esize
[ARC_BUFC_METADATA
]);
4355 uint64_t arc_dirty
= MAX((int64_t)arc_size
- (int64_t)arc_clean
, 0);
4357 if (arc_dirty
>= arc_c_min
)
4358 return (ghost_clean
+ arc_clean
);
4360 return (ghost_clean
+ MAX((int64_t)arc_size
- (int64_t)arc_c_min
, 0));
4364 * If sc->nr_to_scan is zero, the caller is requesting a query of the
4365 * number of objects which can potentially be freed. If it is nonzero,
4366 * the request is to free that many objects.
4368 * Linux kernels >= 3.12 have the count_objects and scan_objects callbacks
4369 * in struct shrinker and also require the shrinker to return the number
4372 * Older kernels require the shrinker to return the number of freeable
4373 * objects following the freeing of nr_to_free.
4375 static spl_shrinker_t
4376 __arc_shrinker_func(struct shrinker
*shrink
, struct shrink_control
*sc
)
4380 /* The arc is considered warm once reclaim has occurred */
4381 if (unlikely(arc_warm
== B_FALSE
))
4384 /* Return the potential number of reclaimable pages */
4385 pages
= btop((int64_t)arc_evictable_memory());
4386 if (sc
->nr_to_scan
== 0)
4389 /* Not allowed to perform filesystem reclaim */
4390 if (!(sc
->gfp_mask
& __GFP_FS
))
4391 return (SHRINK_STOP
);
4393 /* Reclaim in progress */
4394 if (mutex_tryenter(&arc_reclaim_lock
) == 0)
4395 return (SHRINK_STOP
);
4397 mutex_exit(&arc_reclaim_lock
);
4400 * Evict the requested number of pages by shrinking arc_c the
4401 * requested amount. If there is nothing left to evict just
4402 * reap whatever we can from the various arc slabs.
4405 arc_shrink(ptob(sc
->nr_to_scan
));
4406 arc_kmem_reap_now();
4407 #ifdef HAVE_SPLIT_SHRINKER_CALLBACK
4408 pages
= MAX(pages
- btop(arc_evictable_memory()), 0);
4410 pages
= btop(arc_evictable_memory());
4413 arc_kmem_reap_now();
4414 pages
= SHRINK_STOP
;
4418 * We've reaped what we can, wake up threads.
4420 cv_broadcast(&arc_reclaim_waiters_cv
);
4423 * When direct reclaim is observed it usually indicates a rapid
4424 * increase in memory pressure. This occurs because the kswapd
4425 * threads were unable to asynchronously keep enough free memory
4426 * available. In this case set arc_no_grow to briefly pause arc
4427 * growth to avoid compounding the memory pressure.
4429 if (current_is_kswapd()) {
4430 ARCSTAT_BUMP(arcstat_memory_indirect_count
);
4432 arc_no_grow
= B_TRUE
;
4433 arc_need_free
= ptob(sc
->nr_to_scan
);
4434 ARCSTAT_BUMP(arcstat_memory_direct_count
);
4439 SPL_SHRINKER_CALLBACK_WRAPPER(arc_shrinker_func
);
4441 SPL_SHRINKER_DECLARE(arc_shrinker
, arc_shrinker_func
, DEFAULT_SEEKS
);
4442 #endif /* _KERNEL */
4445 * Adapt arc info given the number of bytes we are trying to add and
4446 * the state that we are coming from. This function is only called
4447 * when we are adding new content to the cache.
4450 arc_adapt(int bytes
, arc_state_t
*state
)
4453 uint64_t arc_p_min
= (arc_c
>> arc_p_min_shift
);
4454 int64_t mrug_size
= refcount_count(&arc_mru_ghost
->arcs_size
);
4455 int64_t mfug_size
= refcount_count(&arc_mfu_ghost
->arcs_size
);
4457 if (state
== arc_l2c_only
)
4462 * Adapt the target size of the MRU list:
4463 * - if we just hit in the MRU ghost list, then increase
4464 * the target size of the MRU list.
4465 * - if we just hit in the MFU ghost list, then increase
4466 * the target size of the MFU list by decreasing the
4467 * target size of the MRU list.
4469 if (state
== arc_mru_ghost
) {
4470 mult
= (mrug_size
>= mfug_size
) ? 1 : (mfug_size
/ mrug_size
);
4471 if (!zfs_arc_p_dampener_disable
)
4472 mult
= MIN(mult
, 10); /* avoid wild arc_p adjustment */
4474 arc_p
= MIN(arc_c
- arc_p_min
, arc_p
+ bytes
* mult
);
4475 } else if (state
== arc_mfu_ghost
) {
4478 mult
= (mfug_size
>= mrug_size
) ? 1 : (mrug_size
/ mfug_size
);
4479 if (!zfs_arc_p_dampener_disable
)
4480 mult
= MIN(mult
, 10);
4482 delta
= MIN(bytes
* mult
, arc_p
);
4483 arc_p
= MAX(arc_p_min
, arc_p
- delta
);
4485 ASSERT((int64_t)arc_p
>= 0);
4487 if (arc_reclaim_needed()) {
4488 cv_signal(&arc_reclaim_thread_cv
);
4495 if (arc_c
>= arc_c_max
)
4499 * If we're within (2 * maxblocksize) bytes of the target
4500 * cache size, increment the target cache size
4502 ASSERT3U(arc_c
, >=, 2ULL << SPA_MAXBLOCKSHIFT
);
4503 if (arc_size
>= arc_c
- (2ULL << SPA_MAXBLOCKSHIFT
)) {
4504 atomic_add_64(&arc_c
, (int64_t)bytes
);
4505 if (arc_c
> arc_c_max
)
4507 else if (state
== arc_anon
)
4508 atomic_add_64(&arc_p
, (int64_t)bytes
);
4512 ASSERT((int64_t)arc_p
>= 0);
4516 * Check if arc_size has grown past our upper threshold, determined by
4517 * zfs_arc_overflow_shift.
4520 arc_is_overflowing(void)
4522 /* Always allow at least one block of overflow */
4523 uint64_t overflow
= MAX(SPA_MAXBLOCKSIZE
,
4524 arc_c
>> zfs_arc_overflow_shift
);
4526 return (arc_size
>= arc_c
+ overflow
);
4530 arc_get_data_abd(arc_buf_hdr_t
*hdr
, uint64_t size
, void *tag
)
4532 arc_buf_contents_t type
= arc_buf_type(hdr
);
4534 arc_get_data_impl(hdr
, size
, tag
);
4535 if (type
== ARC_BUFC_METADATA
) {
4536 return (abd_alloc(size
, B_TRUE
));
4538 ASSERT(type
== ARC_BUFC_DATA
);
4539 return (abd_alloc(size
, B_FALSE
));
4544 arc_get_data_buf(arc_buf_hdr_t
*hdr
, uint64_t size
, void *tag
)
4546 arc_buf_contents_t type
= arc_buf_type(hdr
);
4548 arc_get_data_impl(hdr
, size
, tag
);
4549 if (type
== ARC_BUFC_METADATA
) {
4550 return (zio_buf_alloc(size
));
4552 ASSERT(type
== ARC_BUFC_DATA
);
4553 return (zio_data_buf_alloc(size
));
4558 * Allocate a block and return it to the caller. If we are hitting the
4559 * hard limit for the cache size, we must sleep, waiting for the eviction
4560 * thread to catch up. If we're past the target size but below the hard
4561 * limit, we'll only signal the reclaim thread and continue on.
4564 arc_get_data_impl(arc_buf_hdr_t
*hdr
, uint64_t size
, void *tag
)
4566 arc_state_t
*state
= hdr
->b_l1hdr
.b_state
;
4567 arc_buf_contents_t type
= arc_buf_type(hdr
);
4569 arc_adapt(size
, state
);
4572 * If arc_size is currently overflowing, and has grown past our
4573 * upper limit, we must be adding data faster than the evict
4574 * thread can evict. Thus, to ensure we don't compound the
4575 * problem by adding more data and forcing arc_size to grow even
4576 * further past it's target size, we halt and wait for the
4577 * eviction thread to catch up.
4579 * It's also possible that the reclaim thread is unable to evict
4580 * enough buffers to get arc_size below the overflow limit (e.g.
4581 * due to buffers being un-evictable, or hash lock collisions).
4582 * In this case, we want to proceed regardless if we're
4583 * overflowing; thus we don't use a while loop here.
4585 if (arc_is_overflowing()) {
4586 mutex_enter(&arc_reclaim_lock
);
4589 * Now that we've acquired the lock, we may no longer be
4590 * over the overflow limit, lets check.
4592 * We're ignoring the case of spurious wake ups. If that
4593 * were to happen, it'd let this thread consume an ARC
4594 * buffer before it should have (i.e. before we're under
4595 * the overflow limit and were signalled by the reclaim
4596 * thread). As long as that is a rare occurrence, it
4597 * shouldn't cause any harm.
4599 if (arc_is_overflowing()) {
4600 cv_signal(&arc_reclaim_thread_cv
);
4601 cv_wait(&arc_reclaim_waiters_cv
, &arc_reclaim_lock
);
4604 mutex_exit(&arc_reclaim_lock
);
4607 VERIFY3U(hdr
->b_type
, ==, type
);
4608 if (type
== ARC_BUFC_METADATA
) {
4609 arc_space_consume(size
, ARC_SPACE_META
);
4611 arc_space_consume(size
, ARC_SPACE_DATA
);
4615 * Update the state size. Note that ghost states have a
4616 * "ghost size" and so don't need to be updated.
4618 if (!GHOST_STATE(state
)) {
4620 (void) refcount_add_many(&state
->arcs_size
, size
, tag
);
4623 * If this is reached via arc_read, the link is
4624 * protected by the hash lock. If reached via
4625 * arc_buf_alloc, the header should not be accessed by
4626 * any other thread. And, if reached via arc_read_done,
4627 * the hash lock will protect it if it's found in the
4628 * hash table; otherwise no other thread should be
4629 * trying to [add|remove]_reference it.
4631 if (multilist_link_active(&hdr
->b_l1hdr
.b_arc_node
)) {
4632 ASSERT(refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
));
4633 (void) refcount_add_many(&state
->arcs_esize
[type
],
4638 * If we are growing the cache, and we are adding anonymous
4639 * data, and we have outgrown arc_p, update arc_p
4641 if (arc_size
< arc_c
&& hdr
->b_l1hdr
.b_state
== arc_anon
&&
4642 (refcount_count(&arc_anon
->arcs_size
) +
4643 refcount_count(&arc_mru
->arcs_size
) > arc_p
))
4644 arc_p
= MIN(arc_c
, arc_p
+ size
);
4649 arc_free_data_abd(arc_buf_hdr_t
*hdr
, abd_t
*abd
, uint64_t size
, void *tag
)
4651 arc_free_data_impl(hdr
, size
, tag
);
4656 arc_free_data_buf(arc_buf_hdr_t
*hdr
, void *buf
, uint64_t size
, void *tag
)
4658 arc_buf_contents_t type
= arc_buf_type(hdr
);
4660 arc_free_data_impl(hdr
, size
, tag
);
4661 if (type
== ARC_BUFC_METADATA
) {
4662 zio_buf_free(buf
, size
);
4664 ASSERT(type
== ARC_BUFC_DATA
);
4665 zio_data_buf_free(buf
, size
);
4670 * Free the arc data buffer.
4673 arc_free_data_impl(arc_buf_hdr_t
*hdr
, uint64_t size
, void *tag
)
4675 arc_state_t
*state
= hdr
->b_l1hdr
.b_state
;
4676 arc_buf_contents_t type
= arc_buf_type(hdr
);
4678 /* protected by hash lock, if in the hash table */
4679 if (multilist_link_active(&hdr
->b_l1hdr
.b_arc_node
)) {
4680 ASSERT(refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
));
4681 ASSERT(state
!= arc_anon
&& state
!= arc_l2c_only
);
4683 (void) refcount_remove_many(&state
->arcs_esize
[type
],
4686 (void) refcount_remove_many(&state
->arcs_size
, size
, tag
);
4688 VERIFY3U(hdr
->b_type
, ==, type
);
4689 if (type
== ARC_BUFC_METADATA
) {
4690 arc_space_return(size
, ARC_SPACE_META
);
4692 ASSERT(type
== ARC_BUFC_DATA
);
4693 arc_space_return(size
, ARC_SPACE_DATA
);
4698 * This routine is called whenever a buffer is accessed.
4699 * NOTE: the hash lock is dropped in this function.
4702 arc_access(arc_buf_hdr_t
*hdr
, kmutex_t
*hash_lock
)
4706 ASSERT(MUTEX_HELD(hash_lock
));
4707 ASSERT(HDR_HAS_L1HDR(hdr
));
4709 if (hdr
->b_l1hdr
.b_state
== arc_anon
) {
4711 * This buffer is not in the cache, and does not
4712 * appear in our "ghost" list. Add the new buffer
4716 ASSERT0(hdr
->b_l1hdr
.b_arc_access
);
4717 hdr
->b_l1hdr
.b_arc_access
= ddi_get_lbolt();
4718 DTRACE_PROBE1(new_state__mru
, arc_buf_hdr_t
*, hdr
);
4719 arc_change_state(arc_mru
, hdr
, hash_lock
);
4721 } else if (hdr
->b_l1hdr
.b_state
== arc_mru
) {
4722 now
= ddi_get_lbolt();
4725 * If this buffer is here because of a prefetch, then either:
4726 * - clear the flag if this is a "referencing" read
4727 * (any subsequent access will bump this into the MFU state).
4729 * - move the buffer to the head of the list if this is
4730 * another prefetch (to make it less likely to be evicted).
4732 if (HDR_PREFETCH(hdr
)) {
4733 if (refcount_count(&hdr
->b_l1hdr
.b_refcnt
) == 0) {
4734 /* link protected by hash lock */
4735 ASSERT(multilist_link_active(
4736 &hdr
->b_l1hdr
.b_arc_node
));
4738 arc_hdr_clear_flags(hdr
, ARC_FLAG_PREFETCH
);
4739 atomic_inc_32(&hdr
->b_l1hdr
.b_mru_hits
);
4740 ARCSTAT_BUMP(arcstat_mru_hits
);
4742 hdr
->b_l1hdr
.b_arc_access
= now
;
4747 * This buffer has been "accessed" only once so far,
4748 * but it is still in the cache. Move it to the MFU
4751 if (ddi_time_after(now
, hdr
->b_l1hdr
.b_arc_access
+
4754 * More than 125ms have passed since we
4755 * instantiated this buffer. Move it to the
4756 * most frequently used state.
4758 hdr
->b_l1hdr
.b_arc_access
= now
;
4759 DTRACE_PROBE1(new_state__mfu
, arc_buf_hdr_t
*, hdr
);
4760 arc_change_state(arc_mfu
, hdr
, hash_lock
);
4762 atomic_inc_32(&hdr
->b_l1hdr
.b_mru_hits
);
4763 ARCSTAT_BUMP(arcstat_mru_hits
);
4764 } else if (hdr
->b_l1hdr
.b_state
== arc_mru_ghost
) {
4765 arc_state_t
*new_state
;
4767 * This buffer has been "accessed" recently, but
4768 * was evicted from the cache. Move it to the
4772 if (HDR_PREFETCH(hdr
)) {
4773 new_state
= arc_mru
;
4774 if (refcount_count(&hdr
->b_l1hdr
.b_refcnt
) > 0)
4775 arc_hdr_clear_flags(hdr
, ARC_FLAG_PREFETCH
);
4776 DTRACE_PROBE1(new_state__mru
, arc_buf_hdr_t
*, hdr
);
4778 new_state
= arc_mfu
;
4779 DTRACE_PROBE1(new_state__mfu
, arc_buf_hdr_t
*, hdr
);
4782 hdr
->b_l1hdr
.b_arc_access
= ddi_get_lbolt();
4783 arc_change_state(new_state
, hdr
, hash_lock
);
4785 atomic_inc_32(&hdr
->b_l1hdr
.b_mru_ghost_hits
);
4786 ARCSTAT_BUMP(arcstat_mru_ghost_hits
);
4787 } else if (hdr
->b_l1hdr
.b_state
== arc_mfu
) {
4789 * This buffer has been accessed more than once and is
4790 * still in the cache. Keep it in the MFU state.
4792 * NOTE: an add_reference() that occurred when we did
4793 * the arc_read() will have kicked this off the list.
4794 * If it was a prefetch, we will explicitly move it to
4795 * the head of the list now.
4797 if ((HDR_PREFETCH(hdr
)) != 0) {
4798 ASSERT(refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
));
4799 /* link protected by hash_lock */
4800 ASSERT(multilist_link_active(&hdr
->b_l1hdr
.b_arc_node
));
4802 atomic_inc_32(&hdr
->b_l1hdr
.b_mfu_hits
);
4803 ARCSTAT_BUMP(arcstat_mfu_hits
);
4804 hdr
->b_l1hdr
.b_arc_access
= ddi_get_lbolt();
4805 } else if (hdr
->b_l1hdr
.b_state
== arc_mfu_ghost
) {
4806 arc_state_t
*new_state
= arc_mfu
;
4808 * This buffer has been accessed more than once but has
4809 * been evicted from the cache. Move it back to the
4813 if (HDR_PREFETCH(hdr
)) {
4815 * This is a prefetch access...
4816 * move this block back to the MRU state.
4818 ASSERT0(refcount_count(&hdr
->b_l1hdr
.b_refcnt
));
4819 new_state
= arc_mru
;
4822 hdr
->b_l1hdr
.b_arc_access
= ddi_get_lbolt();
4823 DTRACE_PROBE1(new_state__mfu
, arc_buf_hdr_t
*, hdr
);
4824 arc_change_state(new_state
, hdr
, hash_lock
);
4826 atomic_inc_32(&hdr
->b_l1hdr
.b_mfu_ghost_hits
);
4827 ARCSTAT_BUMP(arcstat_mfu_ghost_hits
);
4828 } else if (hdr
->b_l1hdr
.b_state
== arc_l2c_only
) {
4830 * This buffer is on the 2nd Level ARC.
4833 hdr
->b_l1hdr
.b_arc_access
= ddi_get_lbolt();
4834 DTRACE_PROBE1(new_state__mfu
, arc_buf_hdr_t
*, hdr
);
4835 arc_change_state(arc_mfu
, hdr
, hash_lock
);
4837 cmn_err(CE_PANIC
, "invalid arc state 0x%p",
4838 hdr
->b_l1hdr
.b_state
);
4842 /* a generic arc_done_func_t which you can use */
4845 arc_bcopy_func(zio_t
*zio
, arc_buf_t
*buf
, void *arg
)
4847 if (zio
== NULL
|| zio
->io_error
== 0)
4848 bcopy(buf
->b_data
, arg
, arc_buf_size(buf
));
4849 arc_buf_destroy(buf
, arg
);
4852 /* a generic arc_done_func_t */
4854 arc_getbuf_func(zio_t
*zio
, arc_buf_t
*buf
, void *arg
)
4856 arc_buf_t
**bufp
= arg
;
4857 if (zio
&& zio
->io_error
) {
4858 arc_buf_destroy(buf
, arg
);
4862 ASSERT(buf
->b_data
);
4867 arc_hdr_verify(arc_buf_hdr_t
*hdr
, blkptr_t
*bp
)
4869 if (BP_IS_HOLE(bp
) || BP_IS_EMBEDDED(bp
)) {
4870 ASSERT3U(HDR_GET_PSIZE(hdr
), ==, 0);
4871 ASSERT3U(HDR_GET_COMPRESS(hdr
), ==, ZIO_COMPRESS_OFF
);
4873 if (HDR_COMPRESSION_ENABLED(hdr
)) {
4874 ASSERT3U(HDR_GET_COMPRESS(hdr
), ==,
4875 BP_GET_COMPRESS(bp
));
4877 ASSERT3U(HDR_GET_LSIZE(hdr
), ==, BP_GET_LSIZE(bp
));
4878 ASSERT3U(HDR_GET_PSIZE(hdr
), ==, BP_GET_PSIZE(bp
));
4883 arc_read_done(zio_t
*zio
)
4885 arc_buf_hdr_t
*hdr
= zio
->io_private
;
4886 kmutex_t
*hash_lock
= NULL
;
4887 arc_callback_t
*callback_list
;
4888 arc_callback_t
*acb
;
4889 boolean_t freeable
= B_FALSE
;
4890 boolean_t no_zio_error
= (zio
->io_error
== 0);
4891 int callback_cnt
= 0;
4893 * The hdr was inserted into hash-table and removed from lists
4894 * prior to starting I/O. We should find this header, since
4895 * it's in the hash table, and it should be legit since it's
4896 * not possible to evict it during the I/O. The only possible
4897 * reason for it not to be found is if we were freed during the
4900 if (HDR_IN_HASH_TABLE(hdr
)) {
4901 arc_buf_hdr_t
*found
;
4903 ASSERT3U(hdr
->b_birth
, ==, BP_PHYSICAL_BIRTH(zio
->io_bp
));
4904 ASSERT3U(hdr
->b_dva
.dva_word
[0], ==,
4905 BP_IDENTITY(zio
->io_bp
)->dva_word
[0]);
4906 ASSERT3U(hdr
->b_dva
.dva_word
[1], ==,
4907 BP_IDENTITY(zio
->io_bp
)->dva_word
[1]);
4909 found
= buf_hash_find(hdr
->b_spa
, zio
->io_bp
, &hash_lock
);
4911 ASSERT((found
== hdr
&&
4912 DVA_EQUAL(&hdr
->b_dva
, BP_IDENTITY(zio
->io_bp
))) ||
4913 (found
== hdr
&& HDR_L2_READING(hdr
)));
4914 ASSERT3P(hash_lock
, !=, NULL
);
4918 /* byteswap if necessary */
4919 if (BP_SHOULD_BYTESWAP(zio
->io_bp
)) {
4920 if (BP_GET_LEVEL(zio
->io_bp
) > 0) {
4921 hdr
->b_l1hdr
.b_byteswap
= DMU_BSWAP_UINT64
;
4923 hdr
->b_l1hdr
.b_byteswap
=
4924 DMU_OT_BYTESWAP(BP_GET_TYPE(zio
->io_bp
));
4927 hdr
->b_l1hdr
.b_byteswap
= DMU_BSWAP_NUMFUNCS
;
4931 arc_hdr_clear_flags(hdr
, ARC_FLAG_L2_EVICTED
);
4932 if (l2arc_noprefetch
&& HDR_PREFETCH(hdr
))
4933 arc_hdr_clear_flags(hdr
, ARC_FLAG_L2CACHE
);
4935 callback_list
= hdr
->b_l1hdr
.b_acb
;
4936 ASSERT3P(callback_list
, !=, NULL
);
4938 if (hash_lock
&& no_zio_error
&& hdr
->b_l1hdr
.b_state
== arc_anon
) {
4940 * Only call arc_access on anonymous buffers. This is because
4941 * if we've issued an I/O for an evicted buffer, we've already
4942 * called arc_access (to prevent any simultaneous readers from
4943 * getting confused).
4945 arc_access(hdr
, hash_lock
);
4949 * If a read request has a callback (i.e. acb_done is not NULL), then we
4950 * make a buf containing the data according to the parameters which were
4951 * passed in. The implementation of arc_buf_alloc_impl() ensures that we
4952 * aren't needlessly decompressing the data multiple times.
4954 for (acb
= callback_list
; acb
!= NULL
; acb
= acb
->acb_next
) {
4959 /* This is a demand read since prefetches don't use callbacks */
4963 error
= arc_buf_alloc_impl(hdr
, acb
->acb_private
,
4964 acb
->acb_compressed
, no_zio_error
, &acb
->acb_buf
);
4966 zio
->io_error
= error
;
4969 hdr
->b_l1hdr
.b_acb
= NULL
;
4970 arc_hdr_clear_flags(hdr
, ARC_FLAG_IO_IN_PROGRESS
);
4971 if (callback_cnt
== 0) {
4972 ASSERT(HDR_PREFETCH(hdr
));
4973 ASSERT0(hdr
->b_l1hdr
.b_bufcnt
);
4974 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, !=, NULL
);
4977 ASSERT(refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
) ||
4978 callback_list
!= NULL
);
4981 arc_hdr_verify(hdr
, zio
->io_bp
);
4983 arc_hdr_set_flags(hdr
, ARC_FLAG_IO_ERROR
);
4984 if (hdr
->b_l1hdr
.b_state
!= arc_anon
)
4985 arc_change_state(arc_anon
, hdr
, hash_lock
);
4986 if (HDR_IN_HASH_TABLE(hdr
))
4987 buf_hash_remove(hdr
);
4988 freeable
= refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
);
4992 * Broadcast before we drop the hash_lock to avoid the possibility
4993 * that the hdr (and hence the cv) might be freed before we get to
4994 * the cv_broadcast().
4996 cv_broadcast(&hdr
->b_l1hdr
.b_cv
);
4998 if (hash_lock
!= NULL
) {
4999 mutex_exit(hash_lock
);
5002 * This block was freed while we waited for the read to
5003 * complete. It has been removed from the hash table and
5004 * moved to the anonymous state (so that it won't show up
5007 ASSERT3P(hdr
->b_l1hdr
.b_state
, ==, arc_anon
);
5008 freeable
= refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
);
5011 /* execute each callback and free its structure */
5012 while ((acb
= callback_list
) != NULL
) {
5014 acb
->acb_done(zio
, acb
->acb_buf
, acb
->acb_private
);
5016 if (acb
->acb_zio_dummy
!= NULL
) {
5017 acb
->acb_zio_dummy
->io_error
= zio
->io_error
;
5018 zio_nowait(acb
->acb_zio_dummy
);
5021 callback_list
= acb
->acb_next
;
5022 kmem_free(acb
, sizeof (arc_callback_t
));
5026 arc_hdr_destroy(hdr
);
5030 * "Read" the block at the specified DVA (in bp) via the
5031 * cache. If the block is found in the cache, invoke the provided
5032 * callback immediately and return. Note that the `zio' parameter
5033 * in the callback will be NULL in this case, since no IO was
5034 * required. If the block is not in the cache pass the read request
5035 * on to the spa with a substitute callback function, so that the
5036 * requested block will be added to the cache.
5038 * If a read request arrives for a block that has a read in-progress,
5039 * either wait for the in-progress read to complete (and return the
5040 * results); or, if this is a read with a "done" func, add a record
5041 * to the read to invoke the "done" func when the read completes,
5042 * and return; or just return.
5044 * arc_read_done() will invoke all the requested "done" functions
5045 * for readers of this block.
5048 arc_read(zio_t
*pio
, spa_t
*spa
, const blkptr_t
*bp
, arc_done_func_t
*done
,
5049 void *private, zio_priority_t priority
, int zio_flags
,
5050 arc_flags_t
*arc_flags
, const zbookmark_phys_t
*zb
)
5052 arc_buf_hdr_t
*hdr
= NULL
;
5053 kmutex_t
*hash_lock
= NULL
;
5055 uint64_t guid
= spa_load_guid(spa
);
5056 boolean_t compressed_read
= (zio_flags
& ZIO_FLAG_RAW
) != 0;
5059 ASSERT(!BP_IS_EMBEDDED(bp
) ||
5060 BPE_GET_ETYPE(bp
) == BP_EMBEDDED_TYPE_DATA
);
5063 if (!BP_IS_EMBEDDED(bp
)) {
5065 * Embedded BP's have no DVA and require no I/O to "read".
5066 * Create an anonymous arc buf to back it.
5068 hdr
= buf_hash_find(guid
, bp
, &hash_lock
);
5071 if (hdr
!= NULL
&& HDR_HAS_L1HDR(hdr
) && hdr
->b_l1hdr
.b_pabd
!= NULL
) {
5072 arc_buf_t
*buf
= NULL
;
5073 *arc_flags
|= ARC_FLAG_CACHED
;
5075 if (HDR_IO_IN_PROGRESS(hdr
)) {
5077 if ((hdr
->b_flags
& ARC_FLAG_PRIO_ASYNC_READ
) &&
5078 priority
== ZIO_PRIORITY_SYNC_READ
) {
5080 * This sync read must wait for an
5081 * in-progress async read (e.g. a predictive
5082 * prefetch). Async reads are queued
5083 * separately at the vdev_queue layer, so
5084 * this is a form of priority inversion.
5085 * Ideally, we would "inherit" the demand
5086 * i/o's priority by moving the i/o from
5087 * the async queue to the synchronous queue,
5088 * but there is currently no mechanism to do
5089 * so. Track this so that we can evaluate
5090 * the magnitude of this potential performance
5093 * Note that if the prefetch i/o is already
5094 * active (has been issued to the device),
5095 * the prefetch improved performance, because
5096 * we issued it sooner than we would have
5097 * without the prefetch.
5099 DTRACE_PROBE1(arc__sync__wait__for__async
,
5100 arc_buf_hdr_t
*, hdr
);
5101 ARCSTAT_BUMP(arcstat_sync_wait_for_async
);
5103 if (hdr
->b_flags
& ARC_FLAG_PREDICTIVE_PREFETCH
) {
5104 arc_hdr_clear_flags(hdr
,
5105 ARC_FLAG_PREDICTIVE_PREFETCH
);
5108 if (*arc_flags
& ARC_FLAG_WAIT
) {
5109 cv_wait(&hdr
->b_l1hdr
.b_cv
, hash_lock
);
5110 mutex_exit(hash_lock
);
5113 ASSERT(*arc_flags
& ARC_FLAG_NOWAIT
);
5116 arc_callback_t
*acb
= NULL
;
5118 acb
= kmem_zalloc(sizeof (arc_callback_t
),
5120 acb
->acb_done
= done
;
5121 acb
->acb_private
= private;
5123 acb
->acb_zio_dummy
= zio_null(pio
,
5124 spa
, NULL
, NULL
, NULL
, zio_flags
);
5126 ASSERT3P(acb
->acb_done
, !=, NULL
);
5127 acb
->acb_next
= hdr
->b_l1hdr
.b_acb
;
5128 hdr
->b_l1hdr
.b_acb
= acb
;
5129 mutex_exit(hash_lock
);
5132 mutex_exit(hash_lock
);
5136 ASSERT(hdr
->b_l1hdr
.b_state
== arc_mru
||
5137 hdr
->b_l1hdr
.b_state
== arc_mfu
);
5140 if (hdr
->b_flags
& ARC_FLAG_PREDICTIVE_PREFETCH
) {
5142 * This is a demand read which does not have to
5143 * wait for i/o because we did a predictive
5144 * prefetch i/o for it, which has completed.
5147 arc__demand__hit__predictive__prefetch
,
5148 arc_buf_hdr_t
*, hdr
);
5150 arcstat_demand_hit_predictive_prefetch
);
5151 arc_hdr_clear_flags(hdr
,
5152 ARC_FLAG_PREDICTIVE_PREFETCH
);
5154 ASSERT(!BP_IS_EMBEDDED(bp
) || !BP_IS_HOLE(bp
));
5156 /* Get a buf with the desired data in it. */
5157 VERIFY0(arc_buf_alloc_impl(hdr
, private,
5158 compressed_read
, B_TRUE
, &buf
));
5159 } else if (*arc_flags
& ARC_FLAG_PREFETCH
&&
5160 refcount_count(&hdr
->b_l1hdr
.b_refcnt
) == 0) {
5161 arc_hdr_set_flags(hdr
, ARC_FLAG_PREFETCH
);
5163 DTRACE_PROBE1(arc__hit
, arc_buf_hdr_t
*, hdr
);
5164 arc_access(hdr
, hash_lock
);
5165 if (*arc_flags
& ARC_FLAG_L2CACHE
)
5166 arc_hdr_set_flags(hdr
, ARC_FLAG_L2CACHE
);
5167 mutex_exit(hash_lock
);
5168 ARCSTAT_BUMP(arcstat_hits
);
5169 ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr
),
5170 demand
, prefetch
, !HDR_ISTYPE_METADATA(hdr
),
5171 data
, metadata
, hits
);
5174 done(NULL
, buf
, private);
5176 uint64_t lsize
= BP_GET_LSIZE(bp
);
5177 uint64_t psize
= BP_GET_PSIZE(bp
);
5178 arc_callback_t
*acb
;
5181 boolean_t devw
= B_FALSE
;
5185 * Gracefully handle a damaged logical block size as a
5188 if (lsize
> spa_maxblocksize(spa
)) {
5189 rc
= SET_ERROR(ECKSUM
);
5194 /* this block is not in the cache */
5195 arc_buf_hdr_t
*exists
= NULL
;
5196 arc_buf_contents_t type
= BP_GET_BUFC_TYPE(bp
);
5197 hdr
= arc_hdr_alloc(spa_load_guid(spa
), psize
, lsize
,
5198 BP_GET_COMPRESS(bp
), type
);
5200 if (!BP_IS_EMBEDDED(bp
)) {
5201 hdr
->b_dva
= *BP_IDENTITY(bp
);
5202 hdr
->b_birth
= BP_PHYSICAL_BIRTH(bp
);
5203 exists
= buf_hash_insert(hdr
, &hash_lock
);
5205 if (exists
!= NULL
) {
5206 /* somebody beat us to the hash insert */
5207 mutex_exit(hash_lock
);
5208 buf_discard_identity(hdr
);
5209 arc_hdr_destroy(hdr
);
5210 goto top
; /* restart the IO request */
5214 * This block is in the ghost cache. If it was L2-only
5215 * (and thus didn't have an L1 hdr), we realloc the
5216 * header to add an L1 hdr.
5218 if (!HDR_HAS_L1HDR(hdr
)) {
5219 hdr
= arc_hdr_realloc(hdr
, hdr_l2only_cache
,
5223 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, ==, NULL
);
5224 ASSERT(GHOST_STATE(hdr
->b_l1hdr
.b_state
));
5225 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
5226 ASSERT(refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
));
5227 ASSERT3P(hdr
->b_l1hdr
.b_buf
, ==, NULL
);
5228 ASSERT3P(hdr
->b_l1hdr
.b_freeze_cksum
, ==, NULL
);
5231 * This is a delicate dance that we play here.
5232 * This hdr is in the ghost list so we access it
5233 * to move it out of the ghost list before we
5234 * initiate the read. If it's a prefetch then
5235 * it won't have a callback so we'll remove the
5236 * reference that arc_buf_alloc_impl() created. We
5237 * do this after we've called arc_access() to
5238 * avoid hitting an assert in remove_reference().
5240 arc_access(hdr
, hash_lock
);
5241 arc_hdr_alloc_pabd(hdr
);
5243 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, !=, NULL
);
5244 size
= arc_hdr_size(hdr
);
5247 * If compression is enabled on the hdr, then will do
5248 * RAW I/O and will store the compressed data in the hdr's
5249 * data block. Otherwise, the hdr's data block will contain
5250 * the uncompressed data.
5252 if (HDR_GET_COMPRESS(hdr
) != ZIO_COMPRESS_OFF
) {
5253 zio_flags
|= ZIO_FLAG_RAW
;
5256 if (*arc_flags
& ARC_FLAG_PREFETCH
)
5257 arc_hdr_set_flags(hdr
, ARC_FLAG_PREFETCH
);
5258 if (*arc_flags
& ARC_FLAG_L2CACHE
)
5259 arc_hdr_set_flags(hdr
, ARC_FLAG_L2CACHE
);
5260 if (BP_GET_LEVEL(bp
) > 0)
5261 arc_hdr_set_flags(hdr
, ARC_FLAG_INDIRECT
);
5262 if (*arc_flags
& ARC_FLAG_PREDICTIVE_PREFETCH
)
5263 arc_hdr_set_flags(hdr
, ARC_FLAG_PREDICTIVE_PREFETCH
);
5264 ASSERT(!GHOST_STATE(hdr
->b_l1hdr
.b_state
));
5266 acb
= kmem_zalloc(sizeof (arc_callback_t
), KM_SLEEP
);
5267 acb
->acb_done
= done
;
5268 acb
->acb_private
= private;
5269 acb
->acb_compressed
= compressed_read
;
5271 ASSERT3P(hdr
->b_l1hdr
.b_acb
, ==, NULL
);
5272 hdr
->b_l1hdr
.b_acb
= acb
;
5273 arc_hdr_set_flags(hdr
, ARC_FLAG_IO_IN_PROGRESS
);
5275 if (HDR_HAS_L2HDR(hdr
) &&
5276 (vd
= hdr
->b_l2hdr
.b_dev
->l2ad_vdev
) != NULL
) {
5277 devw
= hdr
->b_l2hdr
.b_dev
->l2ad_writing
;
5278 addr
= hdr
->b_l2hdr
.b_daddr
;
5280 * Lock out device removal.
5282 if (vdev_is_dead(vd
) ||
5283 !spa_config_tryenter(spa
, SCL_L2ARC
, vd
, RW_READER
))
5287 if (priority
== ZIO_PRIORITY_ASYNC_READ
)
5288 arc_hdr_set_flags(hdr
, ARC_FLAG_PRIO_ASYNC_READ
);
5290 arc_hdr_clear_flags(hdr
, ARC_FLAG_PRIO_ASYNC_READ
);
5292 if (hash_lock
!= NULL
)
5293 mutex_exit(hash_lock
);
5296 * At this point, we have a level 1 cache miss. Try again in
5297 * L2ARC if possible.
5299 ASSERT3U(HDR_GET_LSIZE(hdr
), ==, lsize
);
5301 DTRACE_PROBE4(arc__miss
, arc_buf_hdr_t
*, hdr
, blkptr_t
*, bp
,
5302 uint64_t, lsize
, zbookmark_phys_t
*, zb
);
5303 ARCSTAT_BUMP(arcstat_misses
);
5304 ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr
),
5305 demand
, prefetch
, !HDR_ISTYPE_METADATA(hdr
),
5306 data
, metadata
, misses
);
5308 if (vd
!= NULL
&& l2arc_ndev
!= 0 && !(l2arc_norw
&& devw
)) {
5310 * Read from the L2ARC if the following are true:
5311 * 1. The L2ARC vdev was previously cached.
5312 * 2. This buffer still has L2ARC metadata.
5313 * 3. This buffer isn't currently writing to the L2ARC.
5314 * 4. The L2ARC entry wasn't evicted, which may
5315 * also have invalidated the vdev.
5316 * 5. This isn't prefetch and l2arc_noprefetch is set.
5318 if (HDR_HAS_L2HDR(hdr
) &&
5319 !HDR_L2_WRITING(hdr
) && !HDR_L2_EVICTED(hdr
) &&
5320 !(l2arc_noprefetch
&& HDR_PREFETCH(hdr
))) {
5321 l2arc_read_callback_t
*cb
;
5323 DTRACE_PROBE1(l2arc__hit
, arc_buf_hdr_t
*, hdr
);
5324 ARCSTAT_BUMP(arcstat_l2_hits
);
5325 atomic_inc_32(&hdr
->b_l2hdr
.b_hits
);
5327 cb
= kmem_zalloc(sizeof (l2arc_read_callback_t
),
5329 cb
->l2rcb_hdr
= hdr
;
5332 cb
->l2rcb_flags
= zio_flags
;
5334 ASSERT(addr
>= VDEV_LABEL_START_SIZE
&&
5335 addr
+ lsize
< vd
->vdev_psize
-
5336 VDEV_LABEL_END_SIZE
);
5339 * l2arc read. The SCL_L2ARC lock will be
5340 * released by l2arc_read_done().
5341 * Issue a null zio if the underlying buffer
5342 * was squashed to zero size by compression.
5344 ASSERT3U(HDR_GET_COMPRESS(hdr
), !=,
5345 ZIO_COMPRESS_EMPTY
);
5346 rzio
= zio_read_phys(pio
, vd
, addr
,
5347 size
, hdr
->b_l1hdr
.b_pabd
,
5349 l2arc_read_done
, cb
, priority
,
5350 zio_flags
| ZIO_FLAG_DONT_CACHE
|
5352 ZIO_FLAG_DONT_PROPAGATE
|
5353 ZIO_FLAG_DONT_RETRY
, B_FALSE
);
5355 DTRACE_PROBE2(l2arc__read
, vdev_t
*, vd
,
5357 ARCSTAT_INCR(arcstat_l2_read_bytes
, size
);
5359 if (*arc_flags
& ARC_FLAG_NOWAIT
) {
5364 ASSERT(*arc_flags
& ARC_FLAG_WAIT
);
5365 if (zio_wait(rzio
) == 0)
5368 /* l2arc read error; goto zio_read() */
5370 DTRACE_PROBE1(l2arc__miss
,
5371 arc_buf_hdr_t
*, hdr
);
5372 ARCSTAT_BUMP(arcstat_l2_misses
);
5373 if (HDR_L2_WRITING(hdr
))
5374 ARCSTAT_BUMP(arcstat_l2_rw_clash
);
5375 spa_config_exit(spa
, SCL_L2ARC
, vd
);
5379 spa_config_exit(spa
, SCL_L2ARC
, vd
);
5380 if (l2arc_ndev
!= 0) {
5381 DTRACE_PROBE1(l2arc__miss
,
5382 arc_buf_hdr_t
*, hdr
);
5383 ARCSTAT_BUMP(arcstat_l2_misses
);
5387 rzio
= zio_read(pio
, spa
, bp
, hdr
->b_l1hdr
.b_pabd
, size
,
5388 arc_read_done
, hdr
, priority
, zio_flags
, zb
);
5390 if (*arc_flags
& ARC_FLAG_WAIT
) {
5391 rc
= zio_wait(rzio
);
5395 ASSERT(*arc_flags
& ARC_FLAG_NOWAIT
);
5400 spa_read_history_add(spa
, zb
, *arc_flags
);
5405 arc_add_prune_callback(arc_prune_func_t
*func
, void *private)
5409 p
= kmem_alloc(sizeof (*p
), KM_SLEEP
);
5411 p
->p_private
= private;
5412 list_link_init(&p
->p_node
);
5413 refcount_create(&p
->p_refcnt
);
5415 mutex_enter(&arc_prune_mtx
);
5416 refcount_add(&p
->p_refcnt
, &arc_prune_list
);
5417 list_insert_head(&arc_prune_list
, p
);
5418 mutex_exit(&arc_prune_mtx
);
5424 arc_remove_prune_callback(arc_prune_t
*p
)
5426 boolean_t wait
= B_FALSE
;
5427 mutex_enter(&arc_prune_mtx
);
5428 list_remove(&arc_prune_list
, p
);
5429 if (refcount_remove(&p
->p_refcnt
, &arc_prune_list
) > 0)
5431 mutex_exit(&arc_prune_mtx
);
5433 /* wait for arc_prune_task to finish */
5435 taskq_wait_outstanding(arc_prune_taskq
, 0);
5436 ASSERT0(refcount_count(&p
->p_refcnt
));
5437 refcount_destroy(&p
->p_refcnt
);
5438 kmem_free(p
, sizeof (*p
));
5442 * Notify the arc that a block was freed, and thus will never be used again.
5445 arc_freed(spa_t
*spa
, const blkptr_t
*bp
)
5448 kmutex_t
*hash_lock
;
5449 uint64_t guid
= spa_load_guid(spa
);
5451 ASSERT(!BP_IS_EMBEDDED(bp
));
5453 hdr
= buf_hash_find(guid
, bp
, &hash_lock
);
5458 * We might be trying to free a block that is still doing I/O
5459 * (i.e. prefetch) or has a reference (i.e. a dedup-ed,
5460 * dmu_sync-ed block). If this block is being prefetched, then it
5461 * would still have the ARC_FLAG_IO_IN_PROGRESS flag set on the hdr
5462 * until the I/O completes. A block may also have a reference if it is
5463 * part of a dedup-ed, dmu_synced write. The dmu_sync() function would
5464 * have written the new block to its final resting place on disk but
5465 * without the dedup flag set. This would have left the hdr in the MRU
5466 * state and discoverable. When the txg finally syncs it detects that
5467 * the block was overridden in open context and issues an override I/O.
5468 * Since this is a dedup block, the override I/O will determine if the
5469 * block is already in the DDT. If so, then it will replace the io_bp
5470 * with the bp from the DDT and allow the I/O to finish. When the I/O
5471 * reaches the done callback, dbuf_write_override_done, it will
5472 * check to see if the io_bp and io_bp_override are identical.
5473 * If they are not, then it indicates that the bp was replaced with
5474 * the bp in the DDT and the override bp is freed. This allows
5475 * us to arrive here with a reference on a block that is being
5476 * freed. So if we have an I/O in progress, or a reference to
5477 * this hdr, then we don't destroy the hdr.
5479 if (!HDR_HAS_L1HDR(hdr
) || (!HDR_IO_IN_PROGRESS(hdr
) &&
5480 refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
))) {
5481 arc_change_state(arc_anon
, hdr
, hash_lock
);
5482 arc_hdr_destroy(hdr
);
5483 mutex_exit(hash_lock
);
5485 mutex_exit(hash_lock
);
5491 * Release this buffer from the cache, making it an anonymous buffer. This
5492 * must be done after a read and prior to modifying the buffer contents.
5493 * If the buffer has more than one reference, we must make
5494 * a new hdr for the buffer.
5497 arc_release(arc_buf_t
*buf
, void *tag
)
5499 kmutex_t
*hash_lock
;
5501 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
5504 * It would be nice to assert that if its DMU metadata (level >
5505 * 0 || it's the dnode file), then it must be syncing context.
5506 * But we don't know that information at this level.
5509 mutex_enter(&buf
->b_evict_lock
);
5511 ASSERT(HDR_HAS_L1HDR(hdr
));
5514 * We don't grab the hash lock prior to this check, because if
5515 * the buffer's header is in the arc_anon state, it won't be
5516 * linked into the hash table.
5518 if (hdr
->b_l1hdr
.b_state
== arc_anon
) {
5519 mutex_exit(&buf
->b_evict_lock
);
5520 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
5521 ASSERT(!HDR_IN_HASH_TABLE(hdr
));
5522 ASSERT(!HDR_HAS_L2HDR(hdr
));
5523 ASSERT(HDR_EMPTY(hdr
));
5525 ASSERT3U(hdr
->b_l1hdr
.b_bufcnt
, ==, 1);
5526 ASSERT3S(refcount_count(&hdr
->b_l1hdr
.b_refcnt
), ==, 1);
5527 ASSERT(!list_link_active(&hdr
->b_l1hdr
.b_arc_node
));
5529 hdr
->b_l1hdr
.b_arc_access
= 0;
5532 * If the buf is being overridden then it may already
5533 * have a hdr that is not empty.
5535 buf_discard_identity(hdr
);
5541 hash_lock
= HDR_LOCK(hdr
);
5542 mutex_enter(hash_lock
);
5545 * This assignment is only valid as long as the hash_lock is
5546 * held, we must be careful not to reference state or the
5547 * b_state field after dropping the lock.
5549 state
= hdr
->b_l1hdr
.b_state
;
5550 ASSERT3P(hash_lock
, ==, HDR_LOCK(hdr
));
5551 ASSERT3P(state
, !=, arc_anon
);
5553 /* this buffer is not on any list */
5554 ASSERT3S(refcount_count(&hdr
->b_l1hdr
.b_refcnt
), >, 0);
5556 if (HDR_HAS_L2HDR(hdr
)) {
5557 mutex_enter(&hdr
->b_l2hdr
.b_dev
->l2ad_mtx
);
5560 * We have to recheck this conditional again now that
5561 * we're holding the l2ad_mtx to prevent a race with
5562 * another thread which might be concurrently calling
5563 * l2arc_evict(). In that case, l2arc_evict() might have
5564 * destroyed the header's L2 portion as we were waiting
5565 * to acquire the l2ad_mtx.
5567 if (HDR_HAS_L2HDR(hdr
))
5568 arc_hdr_l2hdr_destroy(hdr
);
5570 mutex_exit(&hdr
->b_l2hdr
.b_dev
->l2ad_mtx
);
5574 * Do we have more than one buf?
5576 if (hdr
->b_l1hdr
.b_bufcnt
> 1) {
5577 arc_buf_hdr_t
*nhdr
;
5578 uint64_t spa
= hdr
->b_spa
;
5579 uint64_t psize
= HDR_GET_PSIZE(hdr
);
5580 uint64_t lsize
= HDR_GET_LSIZE(hdr
);
5581 enum zio_compress compress
= HDR_GET_COMPRESS(hdr
);
5582 arc_buf_contents_t type
= arc_buf_type(hdr
);
5583 arc_buf_t
*lastbuf
= NULL
;
5584 VERIFY3U(hdr
->b_type
, ==, type
);
5586 ASSERT(hdr
->b_l1hdr
.b_buf
!= buf
|| buf
->b_next
!= NULL
);
5587 (void) remove_reference(hdr
, hash_lock
, tag
);
5589 if (arc_buf_is_shared(buf
) && !ARC_BUF_COMPRESSED(buf
)) {
5590 ASSERT3P(hdr
->b_l1hdr
.b_buf
, !=, buf
);
5591 ASSERT(ARC_BUF_LAST(buf
));
5595 * Pull the data off of this hdr and attach it to
5596 * a new anonymous hdr. Also find the last buffer
5597 * in the hdr's buffer list.
5599 lastbuf
= arc_buf_remove(hdr
, buf
);
5600 ASSERT3P(lastbuf
, !=, NULL
);
5603 * If the current arc_buf_t and the hdr are sharing their data
5604 * buffer, then we must stop sharing that block.
5606 if (arc_buf_is_shared(buf
)) {
5607 ASSERT3P(hdr
->b_l1hdr
.b_buf
, !=, buf
);
5608 VERIFY(!arc_buf_is_shared(lastbuf
));
5611 * First, sever the block sharing relationship between
5612 * buf and the arc_buf_hdr_t. Then, setup a new
5613 * block sharing relationship with the last buffer
5614 * on the arc_buf_t list.
5616 arc_unshare_buf(hdr
, buf
);
5619 * Now we need to recreate the hdr's b_pabd. Since we
5620 * have lastbuf handy, we try to share with it, but if
5621 * we can't then we allocate a new b_pabd and copy the
5622 * data from buf into it.
5624 if (arc_can_share(hdr
, lastbuf
)) {
5625 arc_share_buf(hdr
, lastbuf
);
5627 arc_hdr_alloc_pabd(hdr
);
5628 abd_copy_from_buf(hdr
->b_l1hdr
.b_pabd
,
5629 buf
->b_data
, psize
);
5631 VERIFY3P(lastbuf
->b_data
, !=, NULL
);
5632 } else if (HDR_SHARED_DATA(hdr
)) {
5634 * Uncompressed shared buffers are always at the end
5635 * of the list. Compressed buffers don't have the
5636 * same requirements. This makes it hard to
5637 * simply assert that the lastbuf is shared so
5638 * we rely on the hdr's compression flags to determine
5639 * if we have a compressed, shared buffer.
5641 ASSERT(arc_buf_is_shared(lastbuf
) ||
5642 HDR_GET_COMPRESS(hdr
) != ZIO_COMPRESS_OFF
);
5643 ASSERT(!ARC_BUF_SHARED(buf
));
5645 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, !=, NULL
);
5646 ASSERT3P(state
, !=, arc_l2c_only
);
5648 (void) refcount_remove_many(&state
->arcs_size
,
5649 arc_buf_size(buf
), buf
);
5651 if (refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
)) {
5652 ASSERT3P(state
, !=, arc_l2c_only
);
5653 (void) refcount_remove_many(&state
->arcs_esize
[type
],
5654 arc_buf_size(buf
), buf
);
5657 hdr
->b_l1hdr
.b_bufcnt
-= 1;
5658 arc_cksum_verify(buf
);
5659 arc_buf_unwatch(buf
);
5661 mutex_exit(hash_lock
);
5664 * Allocate a new hdr. The new hdr will contain a b_pabd
5665 * buffer which will be freed in arc_write().
5667 nhdr
= arc_hdr_alloc(spa
, psize
, lsize
, compress
, type
);
5668 ASSERT3P(nhdr
->b_l1hdr
.b_buf
, ==, NULL
);
5669 ASSERT0(nhdr
->b_l1hdr
.b_bufcnt
);
5670 ASSERT0(refcount_count(&nhdr
->b_l1hdr
.b_refcnt
));
5671 VERIFY3U(nhdr
->b_type
, ==, type
);
5672 ASSERT(!HDR_SHARED_DATA(nhdr
));
5674 nhdr
->b_l1hdr
.b_buf
= buf
;
5675 nhdr
->b_l1hdr
.b_bufcnt
= 1;
5676 nhdr
->b_l1hdr
.b_mru_hits
= 0;
5677 nhdr
->b_l1hdr
.b_mru_ghost_hits
= 0;
5678 nhdr
->b_l1hdr
.b_mfu_hits
= 0;
5679 nhdr
->b_l1hdr
.b_mfu_ghost_hits
= 0;
5680 nhdr
->b_l1hdr
.b_l2_hits
= 0;
5681 (void) refcount_add(&nhdr
->b_l1hdr
.b_refcnt
, tag
);
5684 mutex_exit(&buf
->b_evict_lock
);
5685 (void) refcount_add_many(&arc_anon
->arcs_size
,
5686 HDR_GET_LSIZE(nhdr
), buf
);
5688 mutex_exit(&buf
->b_evict_lock
);
5689 ASSERT(refcount_count(&hdr
->b_l1hdr
.b_refcnt
) == 1);
5690 /* protected by hash lock, or hdr is on arc_anon */
5691 ASSERT(!multilist_link_active(&hdr
->b_l1hdr
.b_arc_node
));
5692 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
5693 hdr
->b_l1hdr
.b_mru_hits
= 0;
5694 hdr
->b_l1hdr
.b_mru_ghost_hits
= 0;
5695 hdr
->b_l1hdr
.b_mfu_hits
= 0;
5696 hdr
->b_l1hdr
.b_mfu_ghost_hits
= 0;
5697 hdr
->b_l1hdr
.b_l2_hits
= 0;
5698 arc_change_state(arc_anon
, hdr
, hash_lock
);
5699 hdr
->b_l1hdr
.b_arc_access
= 0;
5700 mutex_exit(hash_lock
);
5702 buf_discard_identity(hdr
);
5708 arc_released(arc_buf_t
*buf
)
5712 mutex_enter(&buf
->b_evict_lock
);
5713 released
= (buf
->b_data
!= NULL
&&
5714 buf
->b_hdr
->b_l1hdr
.b_state
== arc_anon
);
5715 mutex_exit(&buf
->b_evict_lock
);
5721 arc_referenced(arc_buf_t
*buf
)
5725 mutex_enter(&buf
->b_evict_lock
);
5726 referenced
= (refcount_count(&buf
->b_hdr
->b_l1hdr
.b_refcnt
));
5727 mutex_exit(&buf
->b_evict_lock
);
5728 return (referenced
);
5733 arc_write_ready(zio_t
*zio
)
5735 arc_write_callback_t
*callback
= zio
->io_private
;
5736 arc_buf_t
*buf
= callback
->awcb_buf
;
5737 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
5738 uint64_t psize
= BP_IS_HOLE(zio
->io_bp
) ? 0 : BP_GET_PSIZE(zio
->io_bp
);
5739 enum zio_compress compress
;
5740 fstrans_cookie_t cookie
= spl_fstrans_mark();
5742 ASSERT(HDR_HAS_L1HDR(hdr
));
5743 ASSERT(!refcount_is_zero(&buf
->b_hdr
->b_l1hdr
.b_refcnt
));
5744 ASSERT(hdr
->b_l1hdr
.b_bufcnt
> 0);
5747 * If we're reexecuting this zio because the pool suspended, then
5748 * cleanup any state that was previously set the first time the
5749 * callback was invoked.
5751 if (zio
->io_flags
& ZIO_FLAG_REEXECUTED
) {
5752 arc_cksum_free(hdr
);
5753 arc_buf_unwatch(buf
);
5754 if (hdr
->b_l1hdr
.b_pabd
!= NULL
) {
5755 if (arc_buf_is_shared(buf
)) {
5756 arc_unshare_buf(hdr
, buf
);
5758 arc_hdr_free_pabd(hdr
);
5762 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, ==, NULL
);
5763 ASSERT(!HDR_SHARED_DATA(hdr
));
5764 ASSERT(!arc_buf_is_shared(buf
));
5766 callback
->awcb_ready(zio
, buf
, callback
->awcb_private
);
5768 if (HDR_IO_IN_PROGRESS(hdr
))
5769 ASSERT(zio
->io_flags
& ZIO_FLAG_REEXECUTED
);
5771 arc_cksum_compute(buf
);
5772 arc_hdr_set_flags(hdr
, ARC_FLAG_IO_IN_PROGRESS
);
5774 if (BP_IS_HOLE(zio
->io_bp
) || BP_IS_EMBEDDED(zio
->io_bp
)) {
5775 compress
= ZIO_COMPRESS_OFF
;
5777 ASSERT3U(HDR_GET_LSIZE(hdr
), ==, BP_GET_LSIZE(zio
->io_bp
));
5778 compress
= BP_GET_COMPRESS(zio
->io_bp
);
5780 HDR_SET_PSIZE(hdr
, psize
);
5781 arc_hdr_set_compress(hdr
, compress
);
5784 * Fill the hdr with data. If the hdr is compressed, the data we want
5785 * is available from the zio, otherwise we can take it from the buf.
5787 * We might be able to share the buf's data with the hdr here. However,
5788 * doing so would cause the ARC to be full of linear ABDs if we write a
5789 * lot of shareable data. As a compromise, we check whether scattered
5790 * ABDs are allowed, and assume that if they are then the user wants
5791 * the ARC to be primarily filled with them regardless of the data being
5792 * written. Therefore, if they're allowed then we allocate one and copy
5793 * the data into it; otherwise, we share the data directly if we can.
5795 if (zfs_abd_scatter_enabled
|| !arc_can_share(hdr
, buf
)) {
5796 arc_hdr_alloc_pabd(hdr
);
5799 * Ideally, we would always copy the io_abd into b_pabd, but the
5800 * user may have disabled compressed ARC, thus we must check the
5801 * hdr's compression setting rather than the io_bp's.
5803 if (HDR_GET_COMPRESS(hdr
) != ZIO_COMPRESS_OFF
) {
5804 ASSERT3U(BP_GET_COMPRESS(zio
->io_bp
), !=,
5806 ASSERT3U(psize
, >, 0);
5808 abd_copy(hdr
->b_l1hdr
.b_pabd
, zio
->io_abd
, psize
);
5810 ASSERT3U(zio
->io_orig_size
, ==, arc_hdr_size(hdr
));
5812 abd_copy_from_buf(hdr
->b_l1hdr
.b_pabd
, buf
->b_data
,
5816 ASSERT3P(buf
->b_data
, ==, abd_to_buf(zio
->io_orig_abd
));
5817 ASSERT3U(zio
->io_orig_size
, ==, arc_buf_size(buf
));
5818 ASSERT3U(hdr
->b_l1hdr
.b_bufcnt
, ==, 1);
5820 arc_share_buf(hdr
, buf
);
5823 arc_hdr_verify(hdr
, zio
->io_bp
);
5824 spl_fstrans_unmark(cookie
);
5828 arc_write_children_ready(zio_t
*zio
)
5830 arc_write_callback_t
*callback
= zio
->io_private
;
5831 arc_buf_t
*buf
= callback
->awcb_buf
;
5833 callback
->awcb_children_ready(zio
, buf
, callback
->awcb_private
);
5837 * The SPA calls this callback for each physical write that happens on behalf
5838 * of a logical write. See the comment in dbuf_write_physdone() for details.
5841 arc_write_physdone(zio_t
*zio
)
5843 arc_write_callback_t
*cb
= zio
->io_private
;
5844 if (cb
->awcb_physdone
!= NULL
)
5845 cb
->awcb_physdone(zio
, cb
->awcb_buf
, cb
->awcb_private
);
5849 arc_write_done(zio_t
*zio
)
5851 arc_write_callback_t
*callback
= zio
->io_private
;
5852 arc_buf_t
*buf
= callback
->awcb_buf
;
5853 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
5855 ASSERT3P(hdr
->b_l1hdr
.b_acb
, ==, NULL
);
5857 if (zio
->io_error
== 0) {
5858 arc_hdr_verify(hdr
, zio
->io_bp
);
5860 if (BP_IS_HOLE(zio
->io_bp
) || BP_IS_EMBEDDED(zio
->io_bp
)) {
5861 buf_discard_identity(hdr
);
5863 hdr
->b_dva
= *BP_IDENTITY(zio
->io_bp
);
5864 hdr
->b_birth
= BP_PHYSICAL_BIRTH(zio
->io_bp
);
5867 ASSERT(HDR_EMPTY(hdr
));
5871 * If the block to be written was all-zero or compressed enough to be
5872 * embedded in the BP, no write was performed so there will be no
5873 * dva/birth/checksum. The buffer must therefore remain anonymous
5876 if (!HDR_EMPTY(hdr
)) {
5877 arc_buf_hdr_t
*exists
;
5878 kmutex_t
*hash_lock
;
5880 ASSERT3U(zio
->io_error
, ==, 0);
5882 arc_cksum_verify(buf
);
5884 exists
= buf_hash_insert(hdr
, &hash_lock
);
5885 if (exists
!= NULL
) {
5887 * This can only happen if we overwrite for
5888 * sync-to-convergence, because we remove
5889 * buffers from the hash table when we arc_free().
5891 if (zio
->io_flags
& ZIO_FLAG_IO_REWRITE
) {
5892 if (!BP_EQUAL(&zio
->io_bp_orig
, zio
->io_bp
))
5893 panic("bad overwrite, hdr=%p exists=%p",
5894 (void *)hdr
, (void *)exists
);
5895 ASSERT(refcount_is_zero(
5896 &exists
->b_l1hdr
.b_refcnt
));
5897 arc_change_state(arc_anon
, exists
, hash_lock
);
5898 mutex_exit(hash_lock
);
5899 arc_hdr_destroy(exists
);
5900 exists
= buf_hash_insert(hdr
, &hash_lock
);
5901 ASSERT3P(exists
, ==, NULL
);
5902 } else if (zio
->io_flags
& ZIO_FLAG_NOPWRITE
) {
5904 ASSERT(zio
->io_prop
.zp_nopwrite
);
5905 if (!BP_EQUAL(&zio
->io_bp_orig
, zio
->io_bp
))
5906 panic("bad nopwrite, hdr=%p exists=%p",
5907 (void *)hdr
, (void *)exists
);
5910 ASSERT(hdr
->b_l1hdr
.b_bufcnt
== 1);
5911 ASSERT(hdr
->b_l1hdr
.b_state
== arc_anon
);
5912 ASSERT(BP_GET_DEDUP(zio
->io_bp
));
5913 ASSERT(BP_GET_LEVEL(zio
->io_bp
) == 0);
5916 arc_hdr_clear_flags(hdr
, ARC_FLAG_IO_IN_PROGRESS
);
5917 /* if it's not anon, we are doing a scrub */
5918 if (exists
== NULL
&& hdr
->b_l1hdr
.b_state
== arc_anon
)
5919 arc_access(hdr
, hash_lock
);
5920 mutex_exit(hash_lock
);
5922 arc_hdr_clear_flags(hdr
, ARC_FLAG_IO_IN_PROGRESS
);
5925 ASSERT(!refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
));
5926 callback
->awcb_done(zio
, buf
, callback
->awcb_private
);
5928 abd_put(zio
->io_abd
);
5929 kmem_free(callback
, sizeof (arc_write_callback_t
));
5933 arc_write(zio_t
*pio
, spa_t
*spa
, uint64_t txg
,
5934 blkptr_t
*bp
, arc_buf_t
*buf
, boolean_t l2arc
,
5935 const zio_prop_t
*zp
, arc_done_func_t
*ready
,
5936 arc_done_func_t
*children_ready
, arc_done_func_t
*physdone
,
5937 arc_done_func_t
*done
, void *private, zio_priority_t priority
,
5938 int zio_flags
, const zbookmark_phys_t
*zb
)
5940 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
5941 arc_write_callback_t
*callback
;
5944 ASSERT3P(ready
, !=, NULL
);
5945 ASSERT3P(done
, !=, NULL
);
5946 ASSERT(!HDR_IO_ERROR(hdr
));
5947 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
5948 ASSERT3P(hdr
->b_l1hdr
.b_acb
, ==, NULL
);
5949 ASSERT3U(hdr
->b_l1hdr
.b_bufcnt
, >, 0);
5951 arc_hdr_set_flags(hdr
, ARC_FLAG_L2CACHE
);
5952 if (ARC_BUF_COMPRESSED(buf
)) {
5953 ASSERT3U(zp
->zp_compress
, !=, ZIO_COMPRESS_OFF
);
5954 zio_flags
|= ZIO_FLAG_RAW
;
5956 callback
= kmem_zalloc(sizeof (arc_write_callback_t
), KM_SLEEP
);
5957 callback
->awcb_ready
= ready
;
5958 callback
->awcb_children_ready
= children_ready
;
5959 callback
->awcb_physdone
= physdone
;
5960 callback
->awcb_done
= done
;
5961 callback
->awcb_private
= private;
5962 callback
->awcb_buf
= buf
;
5965 * The hdr's b_pabd is now stale, free it now. A new data block
5966 * will be allocated when the zio pipeline calls arc_write_ready().
5968 if (hdr
->b_l1hdr
.b_pabd
!= NULL
) {
5970 * If the buf is currently sharing the data block with
5971 * the hdr then we need to break that relationship here.
5972 * The hdr will remain with a NULL data pointer and the
5973 * buf will take sole ownership of the block.
5975 if (arc_buf_is_shared(buf
)) {
5976 arc_unshare_buf(hdr
, buf
);
5978 arc_hdr_free_pabd(hdr
);
5980 VERIFY3P(buf
->b_data
, !=, NULL
);
5981 arc_hdr_set_compress(hdr
, ZIO_COMPRESS_OFF
);
5983 ASSERT(!arc_buf_is_shared(buf
));
5984 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, ==, NULL
);
5986 zio
= zio_write(pio
, spa
, txg
, bp
,
5987 abd_get_from_buf(buf
->b_data
, HDR_GET_LSIZE(hdr
)),
5988 HDR_GET_LSIZE(hdr
), arc_buf_size(buf
), zp
,
5990 (children_ready
!= NULL
) ? arc_write_children_ready
: NULL
,
5991 arc_write_physdone
, arc_write_done
, callback
,
5992 priority
, zio_flags
, zb
);
5998 arc_memory_throttle(uint64_t reserve
, uint64_t txg
)
6001 uint64_t available_memory
= ptob(freemem
);
6002 static uint64_t page_load
= 0;
6003 static uint64_t last_txg
= 0;
6005 pgcnt_t minfree
= btop(arc_sys_free
/ 4);
6010 MIN(available_memory
, vmem_size(heap_arena
, VMEM_FREE
));
6013 if (available_memory
> arc_all_memory() * arc_lotsfree_percent
/ 100)
6016 if (txg
> last_txg
) {
6021 * If we are in pageout, we know that memory is already tight,
6022 * the arc is already going to be evicting, so we just want to
6023 * continue to let page writes occur as quickly as possible.
6025 if (current_is_kswapd()) {
6026 if (page_load
> MAX(ptob(minfree
), available_memory
) / 4) {
6027 DMU_TX_STAT_BUMP(dmu_tx_memory_reclaim
);
6028 return (SET_ERROR(ERESTART
));
6030 /* Note: reserve is inflated, so we deflate */
6031 page_load
+= reserve
/ 8;
6033 } else if (page_load
> 0 && arc_reclaim_needed()) {
6034 /* memory is low, delay before restarting */
6035 ARCSTAT_INCR(arcstat_memory_throttle_count
, 1);
6036 DMU_TX_STAT_BUMP(dmu_tx_memory_reclaim
);
6037 return (SET_ERROR(EAGAIN
));
6045 arc_tempreserve_clear(uint64_t reserve
)
6047 atomic_add_64(&arc_tempreserve
, -reserve
);
6048 ASSERT((int64_t)arc_tempreserve
>= 0);
6052 arc_tempreserve_space(uint64_t reserve
, uint64_t txg
)
6058 reserve
> arc_c
/4 &&
6059 reserve
* 4 > (2ULL << SPA_MAXBLOCKSHIFT
))
6060 arc_c
= MIN(arc_c_max
, reserve
* 4);
6063 * Throttle when the calculated memory footprint for the TXG
6064 * exceeds the target ARC size.
6066 if (reserve
> arc_c
) {
6067 DMU_TX_STAT_BUMP(dmu_tx_memory_reserve
);
6068 return (SET_ERROR(ERESTART
));
6072 * Don't count loaned bufs as in flight dirty data to prevent long
6073 * network delays from blocking transactions that are ready to be
6074 * assigned to a txg.
6076 anon_size
= MAX((int64_t)(refcount_count(&arc_anon
->arcs_size
) -
6077 arc_loaned_bytes
), 0);
6080 * Writes will, almost always, require additional memory allocations
6081 * in order to compress/encrypt/etc the data. We therefore need to
6082 * make sure that there is sufficient available memory for this.
6084 error
= arc_memory_throttle(reserve
, txg
);
6089 * Throttle writes when the amount of dirty data in the cache
6090 * gets too large. We try to keep the cache less than half full
6091 * of dirty blocks so that our sync times don't grow too large.
6092 * Note: if two requests come in concurrently, we might let them
6093 * both succeed, when one of them should fail. Not a huge deal.
6096 if (reserve
+ arc_tempreserve
+ anon_size
> arc_c
/ 2 &&
6097 anon_size
> arc_c
/ 4) {
6098 uint64_t meta_esize
=
6099 refcount_count(&arc_anon
->arcs_esize
[ARC_BUFC_METADATA
]);
6100 uint64_t data_esize
=
6101 refcount_count(&arc_anon
->arcs_esize
[ARC_BUFC_DATA
]);
6102 dprintf("failing, arc_tempreserve=%lluK anon_meta=%lluK "
6103 "anon_data=%lluK tempreserve=%lluK arc_c=%lluK\n",
6104 arc_tempreserve
>> 10, meta_esize
>> 10,
6105 data_esize
>> 10, reserve
>> 10, arc_c
>> 10);
6106 DMU_TX_STAT_BUMP(dmu_tx_dirty_throttle
);
6107 return (SET_ERROR(ERESTART
));
6109 atomic_add_64(&arc_tempreserve
, reserve
);
6114 arc_kstat_update_state(arc_state_t
*state
, kstat_named_t
*size
,
6115 kstat_named_t
*evict_data
, kstat_named_t
*evict_metadata
)
6117 size
->value
.ui64
= refcount_count(&state
->arcs_size
);
6118 evict_data
->value
.ui64
=
6119 refcount_count(&state
->arcs_esize
[ARC_BUFC_DATA
]);
6120 evict_metadata
->value
.ui64
=
6121 refcount_count(&state
->arcs_esize
[ARC_BUFC_METADATA
]);
6125 arc_kstat_update(kstat_t
*ksp
, int rw
)
6127 arc_stats_t
*as
= ksp
->ks_data
;
6129 if (rw
== KSTAT_WRITE
) {
6132 arc_kstat_update_state(arc_anon
,
6133 &as
->arcstat_anon_size
,
6134 &as
->arcstat_anon_evictable_data
,
6135 &as
->arcstat_anon_evictable_metadata
);
6136 arc_kstat_update_state(arc_mru
,
6137 &as
->arcstat_mru_size
,
6138 &as
->arcstat_mru_evictable_data
,
6139 &as
->arcstat_mru_evictable_metadata
);
6140 arc_kstat_update_state(arc_mru_ghost
,
6141 &as
->arcstat_mru_ghost_size
,
6142 &as
->arcstat_mru_ghost_evictable_data
,
6143 &as
->arcstat_mru_ghost_evictable_metadata
);
6144 arc_kstat_update_state(arc_mfu
,
6145 &as
->arcstat_mfu_size
,
6146 &as
->arcstat_mfu_evictable_data
,
6147 &as
->arcstat_mfu_evictable_metadata
);
6148 arc_kstat_update_state(arc_mfu_ghost
,
6149 &as
->arcstat_mfu_ghost_size
,
6150 &as
->arcstat_mfu_ghost_evictable_data
,
6151 &as
->arcstat_mfu_ghost_evictable_metadata
);
6158 * This function *must* return indices evenly distributed between all
6159 * sublists of the multilist. This is needed due to how the ARC eviction
6160 * code is laid out; arc_evict_state() assumes ARC buffers are evenly
6161 * distributed between all sublists and uses this assumption when
6162 * deciding which sublist to evict from and how much to evict from it.
6165 arc_state_multilist_index_func(multilist_t
*ml
, void *obj
)
6167 arc_buf_hdr_t
*hdr
= obj
;
6170 * We rely on b_dva to generate evenly distributed index
6171 * numbers using buf_hash below. So, as an added precaution,
6172 * let's make sure we never add empty buffers to the arc lists.
6174 ASSERT(!HDR_EMPTY(hdr
));
6177 * The assumption here, is the hash value for a given
6178 * arc_buf_hdr_t will remain constant throughout its lifetime
6179 * (i.e. its b_spa, b_dva, and b_birth fields don't change).
6180 * Thus, we don't need to store the header's sublist index
6181 * on insertion, as this index can be recalculated on removal.
6183 * Also, the low order bits of the hash value are thought to be
6184 * distributed evenly. Otherwise, in the case that the multilist
6185 * has a power of two number of sublists, each sublists' usage
6186 * would not be evenly distributed.
6188 return (buf_hash(hdr
->b_spa
, &hdr
->b_dva
, hdr
->b_birth
) %
6189 multilist_get_num_sublists(ml
));
6193 * Called during module initialization and periodically thereafter to
6194 * apply reasonable changes to the exposed performance tunings. Non-zero
6195 * zfs_* values which differ from the currently set values will be applied.
6198 arc_tuning_update(void)
6200 uint64_t percent
, allmem
= arc_all_memory();
6202 /* Valid range: 64M - <all physical memory> */
6203 if ((zfs_arc_max
) && (zfs_arc_max
!= arc_c_max
) &&
6204 (zfs_arc_max
> 64 << 20) && (zfs_arc_max
< allmem
) &&
6205 (zfs_arc_max
> arc_c_min
)) {
6206 arc_c_max
= zfs_arc_max
;
6208 arc_p
= (arc_c
>> 1);
6209 /* Valid range of arc_meta_limit: arc_meta_min - arc_c_max */
6210 percent
= MIN(zfs_arc_meta_limit_percent
, 100);
6211 arc_meta_limit
= MAX(arc_meta_min
, (percent
* arc_c_max
) / 100);
6212 percent
= MIN(zfs_arc_dnode_limit_percent
, 100);
6213 arc_dnode_limit
= (percent
* arc_meta_limit
) / 100;
6216 /* Valid range: 32M - <arc_c_max> */
6217 if ((zfs_arc_min
) && (zfs_arc_min
!= arc_c_min
) &&
6218 (zfs_arc_min
>= 2ULL << SPA_MAXBLOCKSHIFT
) &&
6219 (zfs_arc_min
<= arc_c_max
)) {
6220 arc_c_min
= zfs_arc_min
;
6221 arc_c
= MAX(arc_c
, arc_c_min
);
6224 /* Valid range: 16M - <arc_c_max> */
6225 if ((zfs_arc_meta_min
) && (zfs_arc_meta_min
!= arc_meta_min
) &&
6226 (zfs_arc_meta_min
>= 1ULL << SPA_MAXBLOCKSHIFT
) &&
6227 (zfs_arc_meta_min
<= arc_c_max
)) {
6228 arc_meta_min
= zfs_arc_meta_min
;
6229 arc_meta_limit
= MAX(arc_meta_limit
, arc_meta_min
);
6230 arc_dnode_limit
= arc_meta_limit
/ 10;
6233 /* Valid range: <arc_meta_min> - <arc_c_max> */
6234 if ((zfs_arc_meta_limit
) && (zfs_arc_meta_limit
!= arc_meta_limit
) &&
6235 (zfs_arc_meta_limit
>= zfs_arc_meta_min
) &&
6236 (zfs_arc_meta_limit
<= arc_c_max
))
6237 arc_meta_limit
= zfs_arc_meta_limit
;
6239 /* Valid range: <arc_meta_min> - <arc_c_max> */
6240 if ((zfs_arc_dnode_limit
) && (zfs_arc_dnode_limit
!= arc_dnode_limit
) &&
6241 (zfs_arc_dnode_limit
>= zfs_arc_meta_min
) &&
6242 (zfs_arc_dnode_limit
<= arc_c_max
))
6243 arc_dnode_limit
= zfs_arc_dnode_limit
;
6245 /* Valid range: 1 - N */
6246 if (zfs_arc_grow_retry
)
6247 arc_grow_retry
= zfs_arc_grow_retry
;
6249 /* Valid range: 1 - N */
6250 if (zfs_arc_shrink_shift
) {
6251 arc_shrink_shift
= zfs_arc_shrink_shift
;
6252 arc_no_grow_shift
= MIN(arc_no_grow_shift
, arc_shrink_shift
-1);
6255 /* Valid range: 1 - N */
6256 if (zfs_arc_p_min_shift
)
6257 arc_p_min_shift
= zfs_arc_p_min_shift
;
6259 /* Valid range: 1 - N ticks */
6260 if (zfs_arc_min_prefetch_lifespan
)
6261 arc_min_prefetch_lifespan
= zfs_arc_min_prefetch_lifespan
;
6263 /* Valid range: 0 - 100 */
6264 if ((zfs_arc_lotsfree_percent
>= 0) &&
6265 (zfs_arc_lotsfree_percent
<= 100))
6266 arc_lotsfree_percent
= zfs_arc_lotsfree_percent
;
6268 /* Valid range: 0 - <all physical memory> */
6269 if ((zfs_arc_sys_free
) && (zfs_arc_sys_free
!= arc_sys_free
))
6270 arc_sys_free
= MIN(MAX(zfs_arc_sys_free
, 0), allmem
);
6275 arc_state_init(void)
6277 arc_anon
= &ARC_anon
;
6279 arc_mru_ghost
= &ARC_mru_ghost
;
6281 arc_mfu_ghost
= &ARC_mfu_ghost
;
6282 arc_l2c_only
= &ARC_l2c_only
;
6284 arc_mru
->arcs_list
[ARC_BUFC_METADATA
] =
6285 multilist_create(sizeof (arc_buf_hdr_t
),
6286 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
6287 arc_state_multilist_index_func
);
6288 arc_mru
->arcs_list
[ARC_BUFC_DATA
] =
6289 multilist_create(sizeof (arc_buf_hdr_t
),
6290 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
6291 arc_state_multilist_index_func
);
6292 arc_mru_ghost
->arcs_list
[ARC_BUFC_METADATA
] =
6293 multilist_create(sizeof (arc_buf_hdr_t
),
6294 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
6295 arc_state_multilist_index_func
);
6296 arc_mru_ghost
->arcs_list
[ARC_BUFC_DATA
] =
6297 multilist_create(sizeof (arc_buf_hdr_t
),
6298 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
6299 arc_state_multilist_index_func
);
6300 arc_mfu
->arcs_list
[ARC_BUFC_METADATA
] =
6301 multilist_create(sizeof (arc_buf_hdr_t
),
6302 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
6303 arc_state_multilist_index_func
);
6304 arc_mfu
->arcs_list
[ARC_BUFC_DATA
] =
6305 multilist_create(sizeof (arc_buf_hdr_t
),
6306 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
6307 arc_state_multilist_index_func
);
6308 arc_mfu_ghost
->arcs_list
[ARC_BUFC_METADATA
] =
6309 multilist_create(sizeof (arc_buf_hdr_t
),
6310 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
6311 arc_state_multilist_index_func
);
6312 arc_mfu_ghost
->arcs_list
[ARC_BUFC_DATA
] =
6313 multilist_create(sizeof (arc_buf_hdr_t
),
6314 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
6315 arc_state_multilist_index_func
);
6316 arc_l2c_only
->arcs_list
[ARC_BUFC_METADATA
] =
6317 multilist_create(sizeof (arc_buf_hdr_t
),
6318 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
6319 arc_state_multilist_index_func
);
6320 arc_l2c_only
->arcs_list
[ARC_BUFC_DATA
] =
6321 multilist_create(sizeof (arc_buf_hdr_t
),
6322 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
6323 arc_state_multilist_index_func
);
6325 refcount_create(&arc_anon
->arcs_esize
[ARC_BUFC_METADATA
]);
6326 refcount_create(&arc_anon
->arcs_esize
[ARC_BUFC_DATA
]);
6327 refcount_create(&arc_mru
->arcs_esize
[ARC_BUFC_METADATA
]);
6328 refcount_create(&arc_mru
->arcs_esize
[ARC_BUFC_DATA
]);
6329 refcount_create(&arc_mru_ghost
->arcs_esize
[ARC_BUFC_METADATA
]);
6330 refcount_create(&arc_mru_ghost
->arcs_esize
[ARC_BUFC_DATA
]);
6331 refcount_create(&arc_mfu
->arcs_esize
[ARC_BUFC_METADATA
]);
6332 refcount_create(&arc_mfu
->arcs_esize
[ARC_BUFC_DATA
]);
6333 refcount_create(&arc_mfu_ghost
->arcs_esize
[ARC_BUFC_METADATA
]);
6334 refcount_create(&arc_mfu_ghost
->arcs_esize
[ARC_BUFC_DATA
]);
6335 refcount_create(&arc_l2c_only
->arcs_esize
[ARC_BUFC_METADATA
]);
6336 refcount_create(&arc_l2c_only
->arcs_esize
[ARC_BUFC_DATA
]);
6338 refcount_create(&arc_anon
->arcs_size
);
6339 refcount_create(&arc_mru
->arcs_size
);
6340 refcount_create(&arc_mru_ghost
->arcs_size
);
6341 refcount_create(&arc_mfu
->arcs_size
);
6342 refcount_create(&arc_mfu_ghost
->arcs_size
);
6343 refcount_create(&arc_l2c_only
->arcs_size
);
6345 arc_anon
->arcs_state
= ARC_STATE_ANON
;
6346 arc_mru
->arcs_state
= ARC_STATE_MRU
;
6347 arc_mru_ghost
->arcs_state
= ARC_STATE_MRU_GHOST
;
6348 arc_mfu
->arcs_state
= ARC_STATE_MFU
;
6349 arc_mfu_ghost
->arcs_state
= ARC_STATE_MFU_GHOST
;
6350 arc_l2c_only
->arcs_state
= ARC_STATE_L2C_ONLY
;
6354 arc_state_fini(void)
6356 refcount_destroy(&arc_anon
->arcs_esize
[ARC_BUFC_METADATA
]);
6357 refcount_destroy(&arc_anon
->arcs_esize
[ARC_BUFC_DATA
]);
6358 refcount_destroy(&arc_mru
->arcs_esize
[ARC_BUFC_METADATA
]);
6359 refcount_destroy(&arc_mru
->arcs_esize
[ARC_BUFC_DATA
]);
6360 refcount_destroy(&arc_mru_ghost
->arcs_esize
[ARC_BUFC_METADATA
]);
6361 refcount_destroy(&arc_mru_ghost
->arcs_esize
[ARC_BUFC_DATA
]);
6362 refcount_destroy(&arc_mfu
->arcs_esize
[ARC_BUFC_METADATA
]);
6363 refcount_destroy(&arc_mfu
->arcs_esize
[ARC_BUFC_DATA
]);
6364 refcount_destroy(&arc_mfu_ghost
->arcs_esize
[ARC_BUFC_METADATA
]);
6365 refcount_destroy(&arc_mfu_ghost
->arcs_esize
[ARC_BUFC_DATA
]);
6366 refcount_destroy(&arc_l2c_only
->arcs_esize
[ARC_BUFC_METADATA
]);
6367 refcount_destroy(&arc_l2c_only
->arcs_esize
[ARC_BUFC_DATA
]);
6369 refcount_destroy(&arc_anon
->arcs_size
);
6370 refcount_destroy(&arc_mru
->arcs_size
);
6371 refcount_destroy(&arc_mru_ghost
->arcs_size
);
6372 refcount_destroy(&arc_mfu
->arcs_size
);
6373 refcount_destroy(&arc_mfu_ghost
->arcs_size
);
6374 refcount_destroy(&arc_l2c_only
->arcs_size
);
6376 multilist_destroy(arc_mru
->arcs_list
[ARC_BUFC_METADATA
]);
6377 multilist_destroy(arc_mru_ghost
->arcs_list
[ARC_BUFC_METADATA
]);
6378 multilist_destroy(arc_mfu
->arcs_list
[ARC_BUFC_METADATA
]);
6379 multilist_destroy(arc_mfu_ghost
->arcs_list
[ARC_BUFC_METADATA
]);
6380 multilist_destroy(arc_mru
->arcs_list
[ARC_BUFC_DATA
]);
6381 multilist_destroy(arc_mru_ghost
->arcs_list
[ARC_BUFC_DATA
]);
6382 multilist_destroy(arc_mfu
->arcs_list
[ARC_BUFC_DATA
]);
6383 multilist_destroy(arc_mfu_ghost
->arcs_list
[ARC_BUFC_DATA
]);
6384 multilist_destroy(arc_l2c_only
->arcs_list
[ARC_BUFC_METADATA
]);
6385 multilist_destroy(arc_l2c_only
->arcs_list
[ARC_BUFC_DATA
]);
6397 uint64_t percent
, allmem
= arc_all_memory();
6399 mutex_init(&arc_reclaim_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
6400 cv_init(&arc_reclaim_thread_cv
, NULL
, CV_DEFAULT
, NULL
);
6401 cv_init(&arc_reclaim_waiters_cv
, NULL
, CV_DEFAULT
, NULL
);
6403 /* Convert seconds to clock ticks */
6404 arc_min_prefetch_lifespan
= 1 * hz
;
6408 * Register a shrinker to support synchronous (direct) memory
6409 * reclaim from the arc. This is done to prevent kswapd from
6410 * swapping out pages when it is preferable to shrink the arc.
6412 spl_register_shrinker(&arc_shrinker
);
6414 /* Set to 1/64 of all memory or a minimum of 512K */
6415 arc_sys_free
= MAX(allmem
/ 64, (512 * 1024));
6419 /* Set max to 1/2 of all memory */
6420 arc_c_max
= allmem
/ 2;
6423 * In userland, there's only the memory pressure that we artificially
6424 * create (see arc_available_memory()). Don't let arc_c get too
6425 * small, because it can cause transactions to be larger than
6426 * arc_c, causing arc_tempreserve_space() to fail.
6429 arc_c_min
= MAX(arc_c_max
/ 2, 2ULL << SPA_MAXBLOCKSHIFT
);
6431 arc_c_min
= 2ULL << SPA_MAXBLOCKSHIFT
;
6435 arc_p
= (arc_c
>> 1);
6438 /* Set min to 1/2 of arc_c_min */
6439 arc_meta_min
= 1ULL << SPA_MAXBLOCKSHIFT
;
6440 /* Initialize maximum observed usage to zero */
6443 * Set arc_meta_limit to a percent of arc_c_max with a floor of
6444 * arc_meta_min, and a ceiling of arc_c_max.
6446 percent
= MIN(zfs_arc_meta_limit_percent
, 100);
6447 arc_meta_limit
= MAX(arc_meta_min
, (percent
* arc_c_max
) / 100);
6448 percent
= MIN(zfs_arc_dnode_limit_percent
, 100);
6449 arc_dnode_limit
= (percent
* arc_meta_limit
) / 100;
6451 /* Apply user specified tunings */
6452 arc_tuning_update();
6454 /* if kmem_flags are set, lets try to use less memory */
6455 if (kmem_debugging())
6457 if (arc_c
< arc_c_min
)
6463 list_create(&arc_prune_list
, sizeof (arc_prune_t
),
6464 offsetof(arc_prune_t
, p_node
));
6465 mutex_init(&arc_prune_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
6467 arc_prune_taskq
= taskq_create("arc_prune", max_ncpus
, defclsyspri
,
6468 max_ncpus
, INT_MAX
, TASKQ_PREPOPULATE
| TASKQ_DYNAMIC
);
6470 arc_reclaim_thread_exit
= B_FALSE
;
6472 arc_ksp
= kstat_create("zfs", 0, "arcstats", "misc", KSTAT_TYPE_NAMED
,
6473 sizeof (arc_stats
) / sizeof (kstat_named_t
), KSTAT_FLAG_VIRTUAL
);
6475 if (arc_ksp
!= NULL
) {
6476 arc_ksp
->ks_data
= &arc_stats
;
6477 arc_ksp
->ks_update
= arc_kstat_update
;
6478 kstat_install(arc_ksp
);
6481 (void) thread_create(NULL
, 0, arc_reclaim_thread
, NULL
, 0, &p0
,
6482 TS_RUN
, defclsyspri
);
6488 * Calculate maximum amount of dirty data per pool.
6490 * If it has been set by a module parameter, take that.
6491 * Otherwise, use a percentage of physical memory defined by
6492 * zfs_dirty_data_max_percent (default 10%) with a cap at
6493 * zfs_dirty_data_max_max (default 25% of physical memory).
6495 if (zfs_dirty_data_max_max
== 0)
6496 zfs_dirty_data_max_max
= allmem
*
6497 zfs_dirty_data_max_max_percent
/ 100;
6499 if (zfs_dirty_data_max
== 0) {
6500 zfs_dirty_data_max
= allmem
*
6501 zfs_dirty_data_max_percent
/ 100;
6502 zfs_dirty_data_max
= MIN(zfs_dirty_data_max
,
6503 zfs_dirty_data_max_max
);
6513 spl_unregister_shrinker(&arc_shrinker
);
6514 #endif /* _KERNEL */
6516 mutex_enter(&arc_reclaim_lock
);
6517 arc_reclaim_thread_exit
= B_TRUE
;
6519 * The reclaim thread will set arc_reclaim_thread_exit back to
6520 * B_FALSE when it is finished exiting; we're waiting for that.
6522 while (arc_reclaim_thread_exit
) {
6523 cv_signal(&arc_reclaim_thread_cv
);
6524 cv_wait(&arc_reclaim_thread_cv
, &arc_reclaim_lock
);
6526 mutex_exit(&arc_reclaim_lock
);
6528 /* Use B_TRUE to ensure *all* buffers are evicted */
6529 arc_flush(NULL
, B_TRUE
);
6533 if (arc_ksp
!= NULL
) {
6534 kstat_delete(arc_ksp
);
6538 taskq_wait(arc_prune_taskq
);
6539 taskq_destroy(arc_prune_taskq
);
6541 mutex_enter(&arc_prune_mtx
);
6542 while ((p
= list_head(&arc_prune_list
)) != NULL
) {
6543 list_remove(&arc_prune_list
, p
);
6544 refcount_remove(&p
->p_refcnt
, &arc_prune_list
);
6545 refcount_destroy(&p
->p_refcnt
);
6546 kmem_free(p
, sizeof (*p
));
6548 mutex_exit(&arc_prune_mtx
);
6550 list_destroy(&arc_prune_list
);
6551 mutex_destroy(&arc_prune_mtx
);
6552 mutex_destroy(&arc_reclaim_lock
);
6553 cv_destroy(&arc_reclaim_thread_cv
);
6554 cv_destroy(&arc_reclaim_waiters_cv
);
6559 ASSERT0(arc_loaned_bytes
);
6565 * The level 2 ARC (L2ARC) is a cache layer in-between main memory and disk.
6566 * It uses dedicated storage devices to hold cached data, which are populated
6567 * using large infrequent writes. The main role of this cache is to boost
6568 * the performance of random read workloads. The intended L2ARC devices
6569 * include short-stroked disks, solid state disks, and other media with
6570 * substantially faster read latency than disk.
6572 * +-----------------------+
6574 * +-----------------------+
6577 * l2arc_feed_thread() arc_read()
6581 * +---------------+ |
6583 * +---------------+ |
6588 * +-------+ +-------+
6590 * | cache | | cache |
6591 * +-------+ +-------+
6592 * +=========+ .-----.
6593 * : L2ARC : |-_____-|
6594 * : devices : | Disks |
6595 * +=========+ `-_____-'
6597 * Read requests are satisfied from the following sources, in order:
6600 * 2) vdev cache of L2ARC devices
6602 * 4) vdev cache of disks
6605 * Some L2ARC device types exhibit extremely slow write performance.
6606 * To accommodate for this there are some significant differences between
6607 * the L2ARC and traditional cache design:
6609 * 1. There is no eviction path from the ARC to the L2ARC. Evictions from
6610 * the ARC behave as usual, freeing buffers and placing headers on ghost
6611 * lists. The ARC does not send buffers to the L2ARC during eviction as
6612 * this would add inflated write latencies for all ARC memory pressure.
6614 * 2. The L2ARC attempts to cache data from the ARC before it is evicted.
6615 * It does this by periodically scanning buffers from the eviction-end of
6616 * the MFU and MRU ARC lists, copying them to the L2ARC devices if they are
6617 * not already there. It scans until a headroom of buffers is satisfied,
6618 * which itself is a buffer for ARC eviction. If a compressible buffer is
6619 * found during scanning and selected for writing to an L2ARC device, we
6620 * temporarily boost scanning headroom during the next scan cycle to make
6621 * sure we adapt to compression effects (which might significantly reduce
6622 * the data volume we write to L2ARC). The thread that does this is
6623 * l2arc_feed_thread(), illustrated below; example sizes are included to
6624 * provide a better sense of ratio than this diagram:
6627 * +---------------------+----------+
6628 * ARC_mfu |:::::#:::::::::::::::|o#o###o###|-->. # already on L2ARC
6629 * +---------------------+----------+ | o L2ARC eligible
6630 * ARC_mru |:#:::::::::::::::::::|#o#ooo####|-->| : ARC buffer
6631 * +---------------------+----------+ |
6632 * 15.9 Gbytes ^ 32 Mbytes |
6634 * l2arc_feed_thread()
6636 * l2arc write hand <--[oooo]--'
6640 * +==============================+
6641 * L2ARC dev |####|#|###|###| |####| ... |
6642 * +==============================+
6645 * 3. If an ARC buffer is copied to the L2ARC but then hit instead of
6646 * evicted, then the L2ARC has cached a buffer much sooner than it probably
6647 * needed to, potentially wasting L2ARC device bandwidth and storage. It is
6648 * safe to say that this is an uncommon case, since buffers at the end of
6649 * the ARC lists have moved there due to inactivity.
6651 * 4. If the ARC evicts faster than the L2ARC can maintain a headroom,
6652 * then the L2ARC simply misses copying some buffers. This serves as a
6653 * pressure valve to prevent heavy read workloads from both stalling the ARC
6654 * with waits and clogging the L2ARC with writes. This also helps prevent
6655 * the potential for the L2ARC to churn if it attempts to cache content too
6656 * quickly, such as during backups of the entire pool.
6658 * 5. After system boot and before the ARC has filled main memory, there are
6659 * no evictions from the ARC and so the tails of the ARC_mfu and ARC_mru
6660 * lists can remain mostly static. Instead of searching from tail of these
6661 * lists as pictured, the l2arc_feed_thread() will search from the list heads
6662 * for eligible buffers, greatly increasing its chance of finding them.
6664 * The L2ARC device write speed is also boosted during this time so that
6665 * the L2ARC warms up faster. Since there have been no ARC evictions yet,
6666 * there are no L2ARC reads, and no fear of degrading read performance
6667 * through increased writes.
6669 * 6. Writes to the L2ARC devices are grouped and sent in-sequence, so that
6670 * the vdev queue can aggregate them into larger and fewer writes. Each
6671 * device is written to in a rotor fashion, sweeping writes through
6672 * available space then repeating.
6674 * 7. The L2ARC does not store dirty content. It never needs to flush
6675 * write buffers back to disk based storage.
6677 * 8. If an ARC buffer is written (and dirtied) which also exists in the
6678 * L2ARC, the now stale L2ARC buffer is immediately dropped.
6680 * The performance of the L2ARC can be tweaked by a number of tunables, which
6681 * may be necessary for different workloads:
6683 * l2arc_write_max max write bytes per interval
6684 * l2arc_write_boost extra write bytes during device warmup
6685 * l2arc_noprefetch skip caching prefetched buffers
6686 * l2arc_headroom number of max device writes to precache
6687 * l2arc_headroom_boost when we find compressed buffers during ARC
6688 * scanning, we multiply headroom by this
6689 * percentage factor for the next scan cycle,
6690 * since more compressed buffers are likely to
6692 * l2arc_feed_secs seconds between L2ARC writing
6694 * Tunables may be removed or added as future performance improvements are
6695 * integrated, and also may become zpool properties.
6697 * There are three key functions that control how the L2ARC warms up:
6699 * l2arc_write_eligible() check if a buffer is eligible to cache
6700 * l2arc_write_size() calculate how much to write
6701 * l2arc_write_interval() calculate sleep delay between writes
6703 * These three functions determine what to write, how much, and how quickly
6708 l2arc_write_eligible(uint64_t spa_guid
, arc_buf_hdr_t
*hdr
)
6711 * A buffer is *not* eligible for the L2ARC if it:
6712 * 1. belongs to a different spa.
6713 * 2. is already cached on the L2ARC.
6714 * 3. has an I/O in progress (it may be an incomplete read).
6715 * 4. is flagged not eligible (zfs property).
6717 if (hdr
->b_spa
!= spa_guid
|| HDR_HAS_L2HDR(hdr
) ||
6718 HDR_IO_IN_PROGRESS(hdr
) || !HDR_L2CACHE(hdr
))
6725 l2arc_write_size(void)
6730 * Make sure our globals have meaningful values in case the user
6733 size
= l2arc_write_max
;
6735 cmn_err(CE_NOTE
, "Bad value for l2arc_write_max, value must "
6736 "be greater than zero, resetting it to the default (%d)",
6738 size
= l2arc_write_max
= L2ARC_WRITE_SIZE
;
6741 if (arc_warm
== B_FALSE
)
6742 size
+= l2arc_write_boost
;
6749 l2arc_write_interval(clock_t began
, uint64_t wanted
, uint64_t wrote
)
6751 clock_t interval
, next
, now
;
6754 * If the ARC lists are busy, increase our write rate; if the
6755 * lists are stale, idle back. This is achieved by checking
6756 * how much we previously wrote - if it was more than half of
6757 * what we wanted, schedule the next write much sooner.
6759 if (l2arc_feed_again
&& wrote
> (wanted
/ 2))
6760 interval
= (hz
* l2arc_feed_min_ms
) / 1000;
6762 interval
= hz
* l2arc_feed_secs
;
6764 now
= ddi_get_lbolt();
6765 next
= MAX(now
, MIN(now
+ interval
, began
+ interval
));
6771 * Cycle through L2ARC devices. This is how L2ARC load balances.
6772 * If a device is returned, this also returns holding the spa config lock.
6774 static l2arc_dev_t
*
6775 l2arc_dev_get_next(void)
6777 l2arc_dev_t
*first
, *next
= NULL
;
6780 * Lock out the removal of spas (spa_namespace_lock), then removal
6781 * of cache devices (l2arc_dev_mtx). Once a device has been selected,
6782 * both locks will be dropped and a spa config lock held instead.
6784 mutex_enter(&spa_namespace_lock
);
6785 mutex_enter(&l2arc_dev_mtx
);
6787 /* if there are no vdevs, there is nothing to do */
6788 if (l2arc_ndev
== 0)
6792 next
= l2arc_dev_last
;
6794 /* loop around the list looking for a non-faulted vdev */
6796 next
= list_head(l2arc_dev_list
);
6798 next
= list_next(l2arc_dev_list
, next
);
6800 next
= list_head(l2arc_dev_list
);
6803 /* if we have come back to the start, bail out */
6806 else if (next
== first
)
6809 } while (vdev_is_dead(next
->l2ad_vdev
));
6811 /* if we were unable to find any usable vdevs, return NULL */
6812 if (vdev_is_dead(next
->l2ad_vdev
))
6815 l2arc_dev_last
= next
;
6818 mutex_exit(&l2arc_dev_mtx
);
6821 * Grab the config lock to prevent the 'next' device from being
6822 * removed while we are writing to it.
6825 spa_config_enter(next
->l2ad_spa
, SCL_L2ARC
, next
, RW_READER
);
6826 mutex_exit(&spa_namespace_lock
);
6832 * Free buffers that were tagged for destruction.
6835 l2arc_do_free_on_write(void)
6838 l2arc_data_free_t
*df
, *df_prev
;
6840 mutex_enter(&l2arc_free_on_write_mtx
);
6841 buflist
= l2arc_free_on_write
;
6843 for (df
= list_tail(buflist
); df
; df
= df_prev
) {
6844 df_prev
= list_prev(buflist
, df
);
6845 ASSERT3P(df
->l2df_abd
, !=, NULL
);
6846 abd_free(df
->l2df_abd
);
6847 list_remove(buflist
, df
);
6848 kmem_free(df
, sizeof (l2arc_data_free_t
));
6851 mutex_exit(&l2arc_free_on_write_mtx
);
6855 * A write to a cache device has completed. Update all headers to allow
6856 * reads from these buffers to begin.
6859 l2arc_write_done(zio_t
*zio
)
6861 l2arc_write_callback_t
*cb
;
6864 arc_buf_hdr_t
*head
, *hdr
, *hdr_prev
;
6865 kmutex_t
*hash_lock
;
6866 int64_t bytes_dropped
= 0;
6868 cb
= zio
->io_private
;
6869 ASSERT3P(cb
, !=, NULL
);
6870 dev
= cb
->l2wcb_dev
;
6871 ASSERT3P(dev
, !=, NULL
);
6872 head
= cb
->l2wcb_head
;
6873 ASSERT3P(head
, !=, NULL
);
6874 buflist
= &dev
->l2ad_buflist
;
6875 ASSERT3P(buflist
, !=, NULL
);
6876 DTRACE_PROBE2(l2arc__iodone
, zio_t
*, zio
,
6877 l2arc_write_callback_t
*, cb
);
6879 if (zio
->io_error
!= 0)
6880 ARCSTAT_BUMP(arcstat_l2_writes_error
);
6883 * All writes completed, or an error was hit.
6886 mutex_enter(&dev
->l2ad_mtx
);
6887 for (hdr
= list_prev(buflist
, head
); hdr
; hdr
= hdr_prev
) {
6888 hdr_prev
= list_prev(buflist
, hdr
);
6890 hash_lock
= HDR_LOCK(hdr
);
6893 * We cannot use mutex_enter or else we can deadlock
6894 * with l2arc_write_buffers (due to swapping the order
6895 * the hash lock and l2ad_mtx are taken).
6897 if (!mutex_tryenter(hash_lock
)) {
6899 * Missed the hash lock. We must retry so we
6900 * don't leave the ARC_FLAG_L2_WRITING bit set.
6902 ARCSTAT_BUMP(arcstat_l2_writes_lock_retry
);
6905 * We don't want to rescan the headers we've
6906 * already marked as having been written out, so
6907 * we reinsert the head node so we can pick up
6908 * where we left off.
6910 list_remove(buflist
, head
);
6911 list_insert_after(buflist
, hdr
, head
);
6913 mutex_exit(&dev
->l2ad_mtx
);
6916 * We wait for the hash lock to become available
6917 * to try and prevent busy waiting, and increase
6918 * the chance we'll be able to acquire the lock
6919 * the next time around.
6921 mutex_enter(hash_lock
);
6922 mutex_exit(hash_lock
);
6927 * We could not have been moved into the arc_l2c_only
6928 * state while in-flight due to our ARC_FLAG_L2_WRITING
6929 * bit being set. Let's just ensure that's being enforced.
6931 ASSERT(HDR_HAS_L1HDR(hdr
));
6934 * Skipped - drop L2ARC entry and mark the header as no
6935 * longer L2 eligibile.
6937 if (zio
->io_error
!= 0) {
6939 * Error - drop L2ARC entry.
6941 list_remove(buflist
, hdr
);
6942 arc_hdr_clear_flags(hdr
, ARC_FLAG_HAS_L2HDR
);
6944 ARCSTAT_INCR(arcstat_l2_asize
, -arc_hdr_size(hdr
));
6945 ARCSTAT_INCR(arcstat_l2_size
, -HDR_GET_LSIZE(hdr
));
6947 bytes_dropped
+= arc_hdr_size(hdr
);
6948 (void) refcount_remove_many(&dev
->l2ad_alloc
,
6949 arc_hdr_size(hdr
), hdr
);
6953 * Allow ARC to begin reads and ghost list evictions to
6956 arc_hdr_clear_flags(hdr
, ARC_FLAG_L2_WRITING
);
6958 mutex_exit(hash_lock
);
6961 atomic_inc_64(&l2arc_writes_done
);
6962 list_remove(buflist
, head
);
6963 ASSERT(!HDR_HAS_L1HDR(head
));
6964 kmem_cache_free(hdr_l2only_cache
, head
);
6965 mutex_exit(&dev
->l2ad_mtx
);
6967 vdev_space_update(dev
->l2ad_vdev
, -bytes_dropped
, 0, 0);
6969 l2arc_do_free_on_write();
6971 kmem_free(cb
, sizeof (l2arc_write_callback_t
));
6975 * A read to a cache device completed. Validate buffer contents before
6976 * handing over to the regular ARC routines.
6979 l2arc_read_done(zio_t
*zio
)
6981 l2arc_read_callback_t
*cb
;
6983 kmutex_t
*hash_lock
;
6984 boolean_t valid_cksum
;
6986 ASSERT3P(zio
->io_vd
, !=, NULL
);
6987 ASSERT(zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
);
6989 spa_config_exit(zio
->io_spa
, SCL_L2ARC
, zio
->io_vd
);
6991 cb
= zio
->io_private
;
6992 ASSERT3P(cb
, !=, NULL
);
6993 hdr
= cb
->l2rcb_hdr
;
6994 ASSERT3P(hdr
, !=, NULL
);
6996 hash_lock
= HDR_LOCK(hdr
);
6997 mutex_enter(hash_lock
);
6998 ASSERT3P(hash_lock
, ==, HDR_LOCK(hdr
));
7000 ASSERT3P(zio
->io_abd
, !=, NULL
);
7003 * Check this survived the L2ARC journey.
7005 ASSERT3P(zio
->io_abd
, ==, hdr
->b_l1hdr
.b_pabd
);
7006 zio
->io_bp_copy
= cb
->l2rcb_bp
; /* XXX fix in L2ARC 2.0 */
7007 zio
->io_bp
= &zio
->io_bp_copy
; /* XXX fix in L2ARC 2.0 */
7009 valid_cksum
= arc_cksum_is_equal(hdr
, zio
);
7010 if (valid_cksum
&& zio
->io_error
== 0 && !HDR_L2_EVICTED(hdr
)) {
7011 mutex_exit(hash_lock
);
7012 zio
->io_private
= hdr
;
7015 mutex_exit(hash_lock
);
7017 * Buffer didn't survive caching. Increment stats and
7018 * reissue to the original storage device.
7020 if (zio
->io_error
!= 0) {
7021 ARCSTAT_BUMP(arcstat_l2_io_error
);
7023 zio
->io_error
= SET_ERROR(EIO
);
7026 ARCSTAT_BUMP(arcstat_l2_cksum_bad
);
7029 * If there's no waiter, issue an async i/o to the primary
7030 * storage now. If there *is* a waiter, the caller must
7031 * issue the i/o in a context where it's OK to block.
7033 if (zio
->io_waiter
== NULL
) {
7034 zio_t
*pio
= zio_unique_parent(zio
);
7036 ASSERT(!pio
|| pio
->io_child_type
== ZIO_CHILD_LOGICAL
);
7038 zio_nowait(zio_read(pio
, zio
->io_spa
, zio
->io_bp
,
7039 hdr
->b_l1hdr
.b_pabd
, zio
->io_size
, arc_read_done
,
7040 hdr
, zio
->io_priority
, cb
->l2rcb_flags
,
7045 kmem_free(cb
, sizeof (l2arc_read_callback_t
));
7049 * This is the list priority from which the L2ARC will search for pages to
7050 * cache. This is used within loops (0..3) to cycle through lists in the
7051 * desired order. This order can have a significant effect on cache
7054 * Currently the metadata lists are hit first, MFU then MRU, followed by
7055 * the data lists. This function returns a locked list, and also returns
7058 static multilist_sublist_t
*
7059 l2arc_sublist_lock(int list_num
)
7061 multilist_t
*ml
= NULL
;
7064 ASSERT(list_num
>= 0 && list_num
< L2ARC_FEED_TYPES
);
7068 ml
= arc_mfu
->arcs_list
[ARC_BUFC_METADATA
];
7071 ml
= arc_mru
->arcs_list
[ARC_BUFC_METADATA
];
7074 ml
= arc_mfu
->arcs_list
[ARC_BUFC_DATA
];
7077 ml
= arc_mru
->arcs_list
[ARC_BUFC_DATA
];
7084 * Return a randomly-selected sublist. This is acceptable
7085 * because the caller feeds only a little bit of data for each
7086 * call (8MB). Subsequent calls will result in different
7087 * sublists being selected.
7089 idx
= multilist_get_random_index(ml
);
7090 return (multilist_sublist_lock(ml
, idx
));
7094 * Evict buffers from the device write hand to the distance specified in
7095 * bytes. This distance may span populated buffers, it may span nothing.
7096 * This is clearing a region on the L2ARC device ready for writing.
7097 * If the 'all' boolean is set, every buffer is evicted.
7100 l2arc_evict(l2arc_dev_t
*dev
, uint64_t distance
, boolean_t all
)
7103 arc_buf_hdr_t
*hdr
, *hdr_prev
;
7104 kmutex_t
*hash_lock
;
7107 buflist
= &dev
->l2ad_buflist
;
7109 if (!all
&& dev
->l2ad_first
) {
7111 * This is the first sweep through the device. There is
7117 if (dev
->l2ad_hand
>= (dev
->l2ad_end
- (2 * distance
))) {
7119 * When nearing the end of the device, evict to the end
7120 * before the device write hand jumps to the start.
7122 taddr
= dev
->l2ad_end
;
7124 taddr
= dev
->l2ad_hand
+ distance
;
7126 DTRACE_PROBE4(l2arc__evict
, l2arc_dev_t
*, dev
, list_t
*, buflist
,
7127 uint64_t, taddr
, boolean_t
, all
);
7130 mutex_enter(&dev
->l2ad_mtx
);
7131 for (hdr
= list_tail(buflist
); hdr
; hdr
= hdr_prev
) {
7132 hdr_prev
= list_prev(buflist
, hdr
);
7134 hash_lock
= HDR_LOCK(hdr
);
7137 * We cannot use mutex_enter or else we can deadlock
7138 * with l2arc_write_buffers (due to swapping the order
7139 * the hash lock and l2ad_mtx are taken).
7141 if (!mutex_tryenter(hash_lock
)) {
7143 * Missed the hash lock. Retry.
7145 ARCSTAT_BUMP(arcstat_l2_evict_lock_retry
);
7146 mutex_exit(&dev
->l2ad_mtx
);
7147 mutex_enter(hash_lock
);
7148 mutex_exit(hash_lock
);
7152 if (HDR_L2_WRITE_HEAD(hdr
)) {
7154 * We hit a write head node. Leave it for
7155 * l2arc_write_done().
7157 list_remove(buflist
, hdr
);
7158 mutex_exit(hash_lock
);
7162 if (!all
&& HDR_HAS_L2HDR(hdr
) &&
7163 (hdr
->b_l2hdr
.b_daddr
> taddr
||
7164 hdr
->b_l2hdr
.b_daddr
< dev
->l2ad_hand
)) {
7166 * We've evicted to the target address,
7167 * or the end of the device.
7169 mutex_exit(hash_lock
);
7173 ASSERT(HDR_HAS_L2HDR(hdr
));
7174 if (!HDR_HAS_L1HDR(hdr
)) {
7175 ASSERT(!HDR_L2_READING(hdr
));
7177 * This doesn't exist in the ARC. Destroy.
7178 * arc_hdr_destroy() will call list_remove()
7179 * and decrement arcstat_l2_size.
7181 arc_change_state(arc_anon
, hdr
, hash_lock
);
7182 arc_hdr_destroy(hdr
);
7184 ASSERT(hdr
->b_l1hdr
.b_state
!= arc_l2c_only
);
7185 ARCSTAT_BUMP(arcstat_l2_evict_l1cached
);
7187 * Invalidate issued or about to be issued
7188 * reads, since we may be about to write
7189 * over this location.
7191 if (HDR_L2_READING(hdr
)) {
7192 ARCSTAT_BUMP(arcstat_l2_evict_reading
);
7193 arc_hdr_set_flags(hdr
, ARC_FLAG_L2_EVICTED
);
7196 /* Ensure this header has finished being written */
7197 ASSERT(!HDR_L2_WRITING(hdr
));
7199 arc_hdr_l2hdr_destroy(hdr
);
7201 mutex_exit(hash_lock
);
7203 mutex_exit(&dev
->l2ad_mtx
);
7207 * Find and write ARC buffers to the L2ARC device.
7209 * An ARC_FLAG_L2_WRITING flag is set so that the L2ARC buffers are not valid
7210 * for reading until they have completed writing.
7211 * The headroom_boost is an in-out parameter used to maintain headroom boost
7212 * state between calls to this function.
7214 * Returns the number of bytes actually written (which may be smaller than
7215 * the delta by which the device hand has changed due to alignment).
7218 l2arc_write_buffers(spa_t
*spa
, l2arc_dev_t
*dev
, uint64_t target_sz
)
7220 arc_buf_hdr_t
*hdr
, *hdr_prev
, *head
;
7221 uint64_t write_asize
, write_psize
, write_sz
, headroom
;
7223 l2arc_write_callback_t
*cb
;
7225 uint64_t guid
= spa_load_guid(spa
);
7228 ASSERT3P(dev
->l2ad_vdev
, !=, NULL
);
7231 write_sz
= write_asize
= write_psize
= 0;
7233 head
= kmem_cache_alloc(hdr_l2only_cache
, KM_PUSHPAGE
);
7234 arc_hdr_set_flags(head
, ARC_FLAG_L2_WRITE_HEAD
| ARC_FLAG_HAS_L2HDR
);
7237 * Copy buffers for L2ARC writing.
7239 for (try = 0; try < L2ARC_FEED_TYPES
; try++) {
7240 multilist_sublist_t
*mls
= l2arc_sublist_lock(try);
7241 uint64_t passed_sz
= 0;
7243 VERIFY3P(mls
, !=, NULL
);
7246 * L2ARC fast warmup.
7248 * Until the ARC is warm and starts to evict, read from the
7249 * head of the ARC lists rather than the tail.
7251 if (arc_warm
== B_FALSE
)
7252 hdr
= multilist_sublist_head(mls
);
7254 hdr
= multilist_sublist_tail(mls
);
7256 headroom
= target_sz
* l2arc_headroom
;
7257 if (zfs_compressed_arc_enabled
)
7258 headroom
= (headroom
* l2arc_headroom_boost
) / 100;
7260 for (; hdr
; hdr
= hdr_prev
) {
7261 kmutex_t
*hash_lock
;
7262 uint64_t asize
, size
;
7265 if (arc_warm
== B_FALSE
)
7266 hdr_prev
= multilist_sublist_next(mls
, hdr
);
7268 hdr_prev
= multilist_sublist_prev(mls
, hdr
);
7270 hash_lock
= HDR_LOCK(hdr
);
7271 if (!mutex_tryenter(hash_lock
)) {
7273 * Skip this buffer rather than waiting.
7278 passed_sz
+= HDR_GET_LSIZE(hdr
);
7279 if (passed_sz
> headroom
) {
7283 mutex_exit(hash_lock
);
7287 if (!l2arc_write_eligible(guid
, hdr
)) {
7288 mutex_exit(hash_lock
);
7292 if ((write_asize
+ HDR_GET_LSIZE(hdr
)) > target_sz
) {
7294 mutex_exit(hash_lock
);
7300 * Insert a dummy header on the buflist so
7301 * l2arc_write_done() can find where the
7302 * write buffers begin without searching.
7304 mutex_enter(&dev
->l2ad_mtx
);
7305 list_insert_head(&dev
->l2ad_buflist
, head
);
7306 mutex_exit(&dev
->l2ad_mtx
);
7309 sizeof (l2arc_write_callback_t
), KM_SLEEP
);
7310 cb
->l2wcb_dev
= dev
;
7311 cb
->l2wcb_head
= head
;
7312 pio
= zio_root(spa
, l2arc_write_done
, cb
,
7316 hdr
->b_l2hdr
.b_dev
= dev
;
7317 hdr
->b_l2hdr
.b_hits
= 0;
7319 hdr
->b_l2hdr
.b_daddr
= dev
->l2ad_hand
;
7320 arc_hdr_set_flags(hdr
,
7321 ARC_FLAG_L2_WRITING
| ARC_FLAG_HAS_L2HDR
);
7323 mutex_enter(&dev
->l2ad_mtx
);
7324 list_insert_head(&dev
->l2ad_buflist
, hdr
);
7325 mutex_exit(&dev
->l2ad_mtx
);
7328 * We rely on the L1 portion of the header below, so
7329 * it's invalid for this header to have been evicted out
7330 * of the ghost cache, prior to being written out. The
7331 * ARC_FLAG_L2_WRITING bit ensures this won't happen.
7333 ASSERT(HDR_HAS_L1HDR(hdr
));
7335 ASSERT3U(HDR_GET_PSIZE(hdr
), >, 0);
7336 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, !=, NULL
);
7337 ASSERT3U(arc_hdr_size(hdr
), >, 0);
7338 size
= arc_hdr_size(hdr
);
7340 (void) refcount_add_many(&dev
->l2ad_alloc
, size
, hdr
);
7343 * Normally the L2ARC can use the hdr's data, but if
7344 * we're sharing data between the hdr and one of its
7345 * bufs, L2ARC needs its own copy of the data so that
7346 * the ZIO below can't race with the buf consumer. To
7347 * ensure that this copy will be available for the
7348 * lifetime of the ZIO and be cleaned up afterwards, we
7349 * add it to the l2arc_free_on_write queue.
7351 if (!HDR_SHARED_DATA(hdr
)) {
7352 to_write
= hdr
->b_l1hdr
.b_pabd
;
7354 to_write
= abd_alloc_for_io(size
,
7355 HDR_ISTYPE_METADATA(hdr
));
7356 abd_copy(to_write
, hdr
->b_l1hdr
.b_pabd
, size
);
7357 l2arc_free_abd_on_write(to_write
, size
,
7360 wzio
= zio_write_phys(pio
, dev
->l2ad_vdev
,
7361 hdr
->b_l2hdr
.b_daddr
, size
, to_write
,
7362 ZIO_CHECKSUM_OFF
, NULL
, hdr
,
7363 ZIO_PRIORITY_ASYNC_WRITE
,
7364 ZIO_FLAG_CANFAIL
, B_FALSE
);
7366 write_sz
+= HDR_GET_LSIZE(hdr
);
7367 DTRACE_PROBE2(l2arc__write
, vdev_t
*, dev
->l2ad_vdev
,
7370 write_asize
+= size
;
7372 * Keep the clock hand suitably device-aligned.
7374 asize
= vdev_psize_to_asize(dev
->l2ad_vdev
, size
);
7375 write_psize
+= asize
;
7376 dev
->l2ad_hand
+= asize
;
7378 mutex_exit(hash_lock
);
7380 (void) zio_nowait(wzio
);
7383 multilist_sublist_unlock(mls
);
7389 /* No buffers selected for writing? */
7392 ASSERT(!HDR_HAS_L1HDR(head
));
7393 kmem_cache_free(hdr_l2only_cache
, head
);
7397 ASSERT3U(write_asize
, <=, target_sz
);
7398 ARCSTAT_BUMP(arcstat_l2_writes_sent
);
7399 ARCSTAT_INCR(arcstat_l2_write_bytes
, write_asize
);
7400 ARCSTAT_INCR(arcstat_l2_size
, write_sz
);
7401 ARCSTAT_INCR(arcstat_l2_asize
, write_asize
);
7402 vdev_space_update(dev
->l2ad_vdev
, write_asize
, 0, 0);
7405 * Bump device hand to the device start if it is approaching the end.
7406 * l2arc_evict() will already have evicted ahead for this case.
7408 if (dev
->l2ad_hand
>= (dev
->l2ad_end
- target_sz
)) {
7409 dev
->l2ad_hand
= dev
->l2ad_start
;
7410 dev
->l2ad_first
= B_FALSE
;
7413 dev
->l2ad_writing
= B_TRUE
;
7414 (void) zio_wait(pio
);
7415 dev
->l2ad_writing
= B_FALSE
;
7417 return (write_asize
);
7421 * This thread feeds the L2ARC at regular intervals. This is the beating
7422 * heart of the L2ARC.
7425 l2arc_feed_thread(void)
7430 uint64_t size
, wrote
;
7431 clock_t begin
, next
= ddi_get_lbolt();
7432 fstrans_cookie_t cookie
;
7434 CALLB_CPR_INIT(&cpr
, &l2arc_feed_thr_lock
, callb_generic_cpr
, FTAG
);
7436 mutex_enter(&l2arc_feed_thr_lock
);
7438 cookie
= spl_fstrans_mark();
7439 while (l2arc_thread_exit
== 0) {
7440 CALLB_CPR_SAFE_BEGIN(&cpr
);
7441 (void) cv_timedwait_sig(&l2arc_feed_thr_cv
,
7442 &l2arc_feed_thr_lock
, next
);
7443 CALLB_CPR_SAFE_END(&cpr
, &l2arc_feed_thr_lock
);
7444 next
= ddi_get_lbolt() + hz
;
7447 * Quick check for L2ARC devices.
7449 mutex_enter(&l2arc_dev_mtx
);
7450 if (l2arc_ndev
== 0) {
7451 mutex_exit(&l2arc_dev_mtx
);
7454 mutex_exit(&l2arc_dev_mtx
);
7455 begin
= ddi_get_lbolt();
7458 * This selects the next l2arc device to write to, and in
7459 * doing so the next spa to feed from: dev->l2ad_spa. This
7460 * will return NULL if there are now no l2arc devices or if
7461 * they are all faulted.
7463 * If a device is returned, its spa's config lock is also
7464 * held to prevent device removal. l2arc_dev_get_next()
7465 * will grab and release l2arc_dev_mtx.
7467 if ((dev
= l2arc_dev_get_next()) == NULL
)
7470 spa
= dev
->l2ad_spa
;
7471 ASSERT3P(spa
, !=, NULL
);
7474 * If the pool is read-only then force the feed thread to
7475 * sleep a little longer.
7477 if (!spa_writeable(spa
)) {
7478 next
= ddi_get_lbolt() + 5 * l2arc_feed_secs
* hz
;
7479 spa_config_exit(spa
, SCL_L2ARC
, dev
);
7484 * Avoid contributing to memory pressure.
7486 if (arc_reclaim_needed()) {
7487 ARCSTAT_BUMP(arcstat_l2_abort_lowmem
);
7488 spa_config_exit(spa
, SCL_L2ARC
, dev
);
7492 ARCSTAT_BUMP(arcstat_l2_feeds
);
7494 size
= l2arc_write_size();
7497 * Evict L2ARC buffers that will be overwritten.
7499 l2arc_evict(dev
, size
, B_FALSE
);
7502 * Write ARC buffers.
7504 wrote
= l2arc_write_buffers(spa
, dev
, size
);
7507 * Calculate interval between writes.
7509 next
= l2arc_write_interval(begin
, size
, wrote
);
7510 spa_config_exit(spa
, SCL_L2ARC
, dev
);
7512 spl_fstrans_unmark(cookie
);
7514 l2arc_thread_exit
= 0;
7515 cv_broadcast(&l2arc_feed_thr_cv
);
7516 CALLB_CPR_EXIT(&cpr
); /* drops l2arc_feed_thr_lock */
7521 l2arc_vdev_present(vdev_t
*vd
)
7525 mutex_enter(&l2arc_dev_mtx
);
7526 for (dev
= list_head(l2arc_dev_list
); dev
!= NULL
;
7527 dev
= list_next(l2arc_dev_list
, dev
)) {
7528 if (dev
->l2ad_vdev
== vd
)
7531 mutex_exit(&l2arc_dev_mtx
);
7533 return (dev
!= NULL
);
7537 * Add a vdev for use by the L2ARC. By this point the spa has already
7538 * validated the vdev and opened it.
7541 l2arc_add_vdev(spa_t
*spa
, vdev_t
*vd
)
7543 l2arc_dev_t
*adddev
;
7545 ASSERT(!l2arc_vdev_present(vd
));
7548 * Create a new l2arc device entry.
7550 adddev
= kmem_zalloc(sizeof (l2arc_dev_t
), KM_SLEEP
);
7551 adddev
->l2ad_spa
= spa
;
7552 adddev
->l2ad_vdev
= vd
;
7553 adddev
->l2ad_start
= VDEV_LABEL_START_SIZE
;
7554 adddev
->l2ad_end
= VDEV_LABEL_START_SIZE
+ vdev_get_min_asize(vd
);
7555 adddev
->l2ad_hand
= adddev
->l2ad_start
;
7556 adddev
->l2ad_first
= B_TRUE
;
7557 adddev
->l2ad_writing
= B_FALSE
;
7558 list_link_init(&adddev
->l2ad_node
);
7560 mutex_init(&adddev
->l2ad_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
7562 * This is a list of all ARC buffers that are still valid on the
7565 list_create(&adddev
->l2ad_buflist
, sizeof (arc_buf_hdr_t
),
7566 offsetof(arc_buf_hdr_t
, b_l2hdr
.b_l2node
));
7568 vdev_space_update(vd
, 0, 0, adddev
->l2ad_end
- adddev
->l2ad_hand
);
7569 refcount_create(&adddev
->l2ad_alloc
);
7572 * Add device to global list
7574 mutex_enter(&l2arc_dev_mtx
);
7575 list_insert_head(l2arc_dev_list
, adddev
);
7576 atomic_inc_64(&l2arc_ndev
);
7577 mutex_exit(&l2arc_dev_mtx
);
7581 * Remove a vdev from the L2ARC.
7584 l2arc_remove_vdev(vdev_t
*vd
)
7586 l2arc_dev_t
*dev
, *nextdev
, *remdev
= NULL
;
7589 * Find the device by vdev
7591 mutex_enter(&l2arc_dev_mtx
);
7592 for (dev
= list_head(l2arc_dev_list
); dev
; dev
= nextdev
) {
7593 nextdev
= list_next(l2arc_dev_list
, dev
);
7594 if (vd
== dev
->l2ad_vdev
) {
7599 ASSERT3P(remdev
, !=, NULL
);
7602 * Remove device from global list
7604 list_remove(l2arc_dev_list
, remdev
);
7605 l2arc_dev_last
= NULL
; /* may have been invalidated */
7606 atomic_dec_64(&l2arc_ndev
);
7607 mutex_exit(&l2arc_dev_mtx
);
7610 * Clear all buflists and ARC references. L2ARC device flush.
7612 l2arc_evict(remdev
, 0, B_TRUE
);
7613 list_destroy(&remdev
->l2ad_buflist
);
7614 mutex_destroy(&remdev
->l2ad_mtx
);
7615 refcount_destroy(&remdev
->l2ad_alloc
);
7616 kmem_free(remdev
, sizeof (l2arc_dev_t
));
7622 l2arc_thread_exit
= 0;
7624 l2arc_writes_sent
= 0;
7625 l2arc_writes_done
= 0;
7627 mutex_init(&l2arc_feed_thr_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
7628 cv_init(&l2arc_feed_thr_cv
, NULL
, CV_DEFAULT
, NULL
);
7629 mutex_init(&l2arc_dev_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
7630 mutex_init(&l2arc_free_on_write_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
7632 l2arc_dev_list
= &L2ARC_dev_list
;
7633 l2arc_free_on_write
= &L2ARC_free_on_write
;
7634 list_create(l2arc_dev_list
, sizeof (l2arc_dev_t
),
7635 offsetof(l2arc_dev_t
, l2ad_node
));
7636 list_create(l2arc_free_on_write
, sizeof (l2arc_data_free_t
),
7637 offsetof(l2arc_data_free_t
, l2df_list_node
));
7644 * This is called from dmu_fini(), which is called from spa_fini();
7645 * Because of this, we can assume that all l2arc devices have
7646 * already been removed when the pools themselves were removed.
7649 l2arc_do_free_on_write();
7651 mutex_destroy(&l2arc_feed_thr_lock
);
7652 cv_destroy(&l2arc_feed_thr_cv
);
7653 mutex_destroy(&l2arc_dev_mtx
);
7654 mutex_destroy(&l2arc_free_on_write_mtx
);
7656 list_destroy(l2arc_dev_list
);
7657 list_destroy(l2arc_free_on_write
);
7663 if (!(spa_mode_global
& FWRITE
))
7666 (void) thread_create(NULL
, 0, l2arc_feed_thread
, NULL
, 0, &p0
,
7667 TS_RUN
, defclsyspri
);
7673 if (!(spa_mode_global
& FWRITE
))
7676 mutex_enter(&l2arc_feed_thr_lock
);
7677 cv_signal(&l2arc_feed_thr_cv
); /* kick thread out of startup */
7678 l2arc_thread_exit
= 1;
7679 while (l2arc_thread_exit
!= 0)
7680 cv_wait(&l2arc_feed_thr_cv
, &l2arc_feed_thr_lock
);
7681 mutex_exit(&l2arc_feed_thr_lock
);
7684 #if defined(_KERNEL) && defined(HAVE_SPL)
7685 EXPORT_SYMBOL(arc_buf_size
);
7686 EXPORT_SYMBOL(arc_write
);
7687 EXPORT_SYMBOL(arc_read
);
7688 EXPORT_SYMBOL(arc_buf_info
);
7689 EXPORT_SYMBOL(arc_getbuf_func
);
7690 EXPORT_SYMBOL(arc_add_prune_callback
);
7691 EXPORT_SYMBOL(arc_remove_prune_callback
);
7694 module_param(zfs_arc_min
, ulong
, 0644);
7695 MODULE_PARM_DESC(zfs_arc_min
, "Min arc size");
7697 module_param(zfs_arc_max
, ulong
, 0644);
7698 MODULE_PARM_DESC(zfs_arc_max
, "Max arc size");
7700 module_param(zfs_arc_meta_limit
, ulong
, 0644);
7701 MODULE_PARM_DESC(zfs_arc_meta_limit
, "Meta limit for arc size");
7703 module_param(zfs_arc_meta_limit_percent
, ulong
, 0644);
7704 MODULE_PARM_DESC(zfs_arc_meta_limit_percent
,
7705 "Percent of arc size for arc meta limit");
7707 module_param(zfs_arc_meta_min
, ulong
, 0644);
7708 MODULE_PARM_DESC(zfs_arc_meta_min
, "Min arc metadata");
7710 module_param(zfs_arc_meta_prune
, int, 0644);
7711 MODULE_PARM_DESC(zfs_arc_meta_prune
, "Meta objects to scan for prune");
7713 module_param(zfs_arc_meta_adjust_restarts
, int, 0644);
7714 MODULE_PARM_DESC(zfs_arc_meta_adjust_restarts
,
7715 "Limit number of restarts in arc_adjust_meta");
7717 module_param(zfs_arc_meta_strategy
, int, 0644);
7718 MODULE_PARM_DESC(zfs_arc_meta_strategy
, "Meta reclaim strategy");
7720 module_param(zfs_arc_grow_retry
, int, 0644);
7721 MODULE_PARM_DESC(zfs_arc_grow_retry
, "Seconds before growing arc size");
7723 module_param(zfs_arc_p_aggressive_disable
, int, 0644);
7724 MODULE_PARM_DESC(zfs_arc_p_aggressive_disable
, "disable aggressive arc_p grow");
7726 module_param(zfs_arc_p_dampener_disable
, int, 0644);
7727 MODULE_PARM_DESC(zfs_arc_p_dampener_disable
, "disable arc_p adapt dampener");
7729 module_param(zfs_arc_shrink_shift
, int, 0644);
7730 MODULE_PARM_DESC(zfs_arc_shrink_shift
, "log2(fraction of arc to reclaim)");
7732 module_param(zfs_arc_p_min_shift
, int, 0644);
7733 MODULE_PARM_DESC(zfs_arc_p_min_shift
, "arc_c shift to calc min/max arc_p");
7735 module_param(zfs_arc_average_blocksize
, int, 0444);
7736 MODULE_PARM_DESC(zfs_arc_average_blocksize
, "Target average block size");
7738 module_param(zfs_compressed_arc_enabled
, int, 0644);
7739 MODULE_PARM_DESC(zfs_compressed_arc_enabled
, "Disable compressed arc buffers");
7741 module_param(zfs_arc_min_prefetch_lifespan
, int, 0644);
7742 MODULE_PARM_DESC(zfs_arc_min_prefetch_lifespan
, "Min life of prefetch block");
7744 module_param(l2arc_write_max
, ulong
, 0644);
7745 MODULE_PARM_DESC(l2arc_write_max
, "Max write bytes per interval");
7747 module_param(l2arc_write_boost
, ulong
, 0644);
7748 MODULE_PARM_DESC(l2arc_write_boost
, "Extra write bytes during device warmup");
7750 module_param(l2arc_headroom
, ulong
, 0644);
7751 MODULE_PARM_DESC(l2arc_headroom
, "Number of max device writes to precache");
7753 module_param(l2arc_headroom_boost
, ulong
, 0644);
7754 MODULE_PARM_DESC(l2arc_headroom_boost
, "Compressed l2arc_headroom multiplier");
7756 module_param(l2arc_feed_secs
, ulong
, 0644);
7757 MODULE_PARM_DESC(l2arc_feed_secs
, "Seconds between L2ARC writing");
7759 module_param(l2arc_feed_min_ms
, ulong
, 0644);
7760 MODULE_PARM_DESC(l2arc_feed_min_ms
, "Min feed interval in milliseconds");
7762 module_param(l2arc_noprefetch
, int, 0644);
7763 MODULE_PARM_DESC(l2arc_noprefetch
, "Skip caching prefetched buffers");
7765 module_param(l2arc_feed_again
, int, 0644);
7766 MODULE_PARM_DESC(l2arc_feed_again
, "Turbo L2ARC warmup");
7768 module_param(l2arc_norw
, int, 0644);
7769 MODULE_PARM_DESC(l2arc_norw
, "No reads during writes");
7771 module_param(zfs_arc_lotsfree_percent
, int, 0644);
7772 MODULE_PARM_DESC(zfs_arc_lotsfree_percent
,
7773 "System free memory I/O throttle in bytes");
7775 module_param(zfs_arc_sys_free
, ulong
, 0644);
7776 MODULE_PARM_DESC(zfs_arc_sys_free
, "System free memory target size in bytes");
7778 module_param(zfs_arc_dnode_limit
, ulong
, 0644);
7779 MODULE_PARM_DESC(zfs_arc_dnode_limit
, "Minimum bytes of dnodes in arc");
7781 module_param(zfs_arc_dnode_limit_percent
, ulong
, 0644);
7782 MODULE_PARM_DESC(zfs_arc_dnode_limit_percent
,
7783 "Percent of ARC meta buffers for dnodes");
7785 module_param(zfs_arc_dnode_reduce_percent
, ulong
, 0644);
7786 MODULE_PARM_DESC(zfs_arc_dnode_reduce_percent
,
7787 "Percentage of excess dnodes to try to unpin");