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, 2016 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_pdata).
141 * The L1ARC's data pointer may or may not be uncompressed. The ARC has the
142 * ability to store the physical data (b_pdata) associated with the DVA of the
143 * arc_buf_hdr_t. Since the b_pdata 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_pdata 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_pdata +-+ |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_pdata buffer into a
203 * new data buffer, or shares the hdr's b_pdata 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_pdata +-+ |---------| |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_pdata
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_pdata. 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_pdata. The
253 * L2ARC will always write the contents of b_pdata 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/multilist.h>
276 #include <sys/vmsystm.h>
278 #include <sys/fs/swapnode.h>
280 #include <linux/mm_compat.h>
282 #include <sys/callb.h>
283 #include <sys/kstat.h>
284 #include <sys/dmu_tx.h>
285 #include <zfs_fletcher.h>
286 #include <sys/arc_impl.h>
287 #include <sys/trace_arc.h>
290 /* set with ZFS_DEBUG=watch, to enable watchpoints on frozen buffers */
291 boolean_t arc_watch
= B_FALSE
;
294 static kmutex_t arc_reclaim_lock
;
295 static kcondvar_t arc_reclaim_thread_cv
;
296 static boolean_t arc_reclaim_thread_exit
;
297 static kcondvar_t arc_reclaim_waiters_cv
;
300 * The number of headers to evict in arc_evict_state_impl() before
301 * dropping the sublist lock and evicting from another sublist. A lower
302 * value means we're more likely to evict the "correct" header (i.e. the
303 * oldest header in the arc state), but comes with higher overhead
304 * (i.e. more invocations of arc_evict_state_impl()).
306 int zfs_arc_evict_batch_limit
= 10;
309 * The number of sublists used for each of the arc state lists. If this
310 * is not set to a suitable value by the user, it will be configured to
311 * the number of CPUs on the system in arc_init().
313 int zfs_arc_num_sublists_per_state
= 0;
315 /* number of seconds before growing cache again */
316 static int arc_grow_retry
= 5;
318 /* shift of arc_c for calculating overflow limit in arc_get_data_buf */
319 int zfs_arc_overflow_shift
= 8;
321 /* shift of arc_c for calculating both min and max arc_p */
322 static int arc_p_min_shift
= 4;
324 /* log2(fraction of arc to reclaim) */
325 static int arc_shrink_shift
= 7;
328 * log2(fraction of ARC which must be free to allow growing).
329 * I.e. If there is less than arc_c >> arc_no_grow_shift free memory,
330 * when reading a new block into the ARC, we will evict an equal-sized block
333 * This must be less than arc_shrink_shift, so that when we shrink the ARC,
334 * we will still not allow it to grow.
336 int arc_no_grow_shift
= 5;
340 * minimum lifespan of a prefetch block in clock ticks
341 * (initialized in arc_init())
343 static int arc_min_prefetch_lifespan
;
346 * If this percent of memory is free, don't throttle.
348 int arc_lotsfree_percent
= 10;
353 * The arc has filled available memory and has now warmed up.
355 static boolean_t arc_warm
;
358 * log2 fraction of the zio arena to keep free.
360 int arc_zio_arena_free_shift
= 2;
363 * These tunables are for performance analysis.
365 unsigned long zfs_arc_max
= 0;
366 unsigned long zfs_arc_min
= 0;
367 unsigned long zfs_arc_meta_limit
= 0;
368 unsigned long zfs_arc_meta_min
= 0;
369 unsigned long zfs_arc_dnode_limit
= 0;
370 unsigned long zfs_arc_dnode_reduce_percent
= 10;
371 int zfs_arc_grow_retry
= 0;
372 int zfs_arc_shrink_shift
= 0;
373 int zfs_arc_p_min_shift
= 0;
374 int zfs_arc_average_blocksize
= 8 * 1024; /* 8KB */
376 int zfs_compressed_arc_enabled
= B_TRUE
;
379 * ARC will evict meta buffers that exceed arc_meta_limit. This
380 * tunable make arc_meta_limit adjustable for different workloads.
382 unsigned long zfs_arc_meta_limit_percent
= 75;
385 * Percentage that can be consumed by dnodes of ARC meta buffers.
387 unsigned long zfs_arc_dnode_limit_percent
= 10;
390 * These tunables are Linux specific
392 unsigned long zfs_arc_sys_free
= 0;
393 int zfs_arc_min_prefetch_lifespan
= 0;
394 int zfs_arc_p_aggressive_disable
= 1;
395 int zfs_arc_p_dampener_disable
= 1;
396 int zfs_arc_meta_prune
= 10000;
397 int zfs_arc_meta_strategy
= ARC_STRATEGY_META_BALANCED
;
398 int zfs_arc_meta_adjust_restarts
= 4096;
399 int zfs_arc_lotsfree_percent
= 10;
402 static arc_state_t ARC_anon
;
403 static arc_state_t ARC_mru
;
404 static arc_state_t ARC_mru_ghost
;
405 static arc_state_t ARC_mfu
;
406 static arc_state_t ARC_mfu_ghost
;
407 static arc_state_t ARC_l2c_only
;
409 typedef struct arc_stats
{
410 kstat_named_t arcstat_hits
;
411 kstat_named_t arcstat_misses
;
412 kstat_named_t arcstat_demand_data_hits
;
413 kstat_named_t arcstat_demand_data_misses
;
414 kstat_named_t arcstat_demand_metadata_hits
;
415 kstat_named_t arcstat_demand_metadata_misses
;
416 kstat_named_t arcstat_prefetch_data_hits
;
417 kstat_named_t arcstat_prefetch_data_misses
;
418 kstat_named_t arcstat_prefetch_metadata_hits
;
419 kstat_named_t arcstat_prefetch_metadata_misses
;
420 kstat_named_t arcstat_mru_hits
;
421 kstat_named_t arcstat_mru_ghost_hits
;
422 kstat_named_t arcstat_mfu_hits
;
423 kstat_named_t arcstat_mfu_ghost_hits
;
424 kstat_named_t arcstat_deleted
;
426 * Number of buffers that could not be evicted because the hash lock
427 * was held by another thread. The lock may not necessarily be held
428 * by something using the same buffer, since hash locks are shared
429 * by multiple buffers.
431 kstat_named_t arcstat_mutex_miss
;
433 * Number of buffers skipped because they have I/O in progress, are
434 * indrect prefetch buffers that have not lived long enough, or are
435 * not from the spa we're trying to evict from.
437 kstat_named_t arcstat_evict_skip
;
439 * Number of times arc_evict_state() was unable to evict enough
440 * buffers to reach its target amount.
442 kstat_named_t arcstat_evict_not_enough
;
443 kstat_named_t arcstat_evict_l2_cached
;
444 kstat_named_t arcstat_evict_l2_eligible
;
445 kstat_named_t arcstat_evict_l2_ineligible
;
446 kstat_named_t arcstat_evict_l2_skip
;
447 kstat_named_t arcstat_hash_elements
;
448 kstat_named_t arcstat_hash_elements_max
;
449 kstat_named_t arcstat_hash_collisions
;
450 kstat_named_t arcstat_hash_chains
;
451 kstat_named_t arcstat_hash_chain_max
;
452 kstat_named_t arcstat_p
;
453 kstat_named_t arcstat_c
;
454 kstat_named_t arcstat_c_min
;
455 kstat_named_t arcstat_c_max
;
456 kstat_named_t arcstat_size
;
458 * Number of compressed bytes stored in the arc_buf_hdr_t's b_pdata.
459 * Note that the compressed bytes may match the uncompressed bytes
460 * if the block is either not compressed or compressed arc is disabled.
462 kstat_named_t arcstat_compressed_size
;
464 * Uncompressed size of the data stored in b_pdata. If compressed
465 * arc is disabled then this value will be identical to the stat
468 kstat_named_t arcstat_uncompressed_size
;
470 * Number of bytes stored in all the arc_buf_t's. This is classified
471 * as "overhead" since this data is typically short-lived and will
472 * be evicted from the arc when it becomes unreferenced unless the
473 * zfs_keep_uncompressed_metadata or zfs_keep_uncompressed_level
474 * values have been set (see comment in dbuf.c for more information).
476 kstat_named_t arcstat_overhead_size
;
478 * Number of bytes consumed by internal ARC structures necessary
479 * for tracking purposes; these structures are not actually
480 * backed by ARC buffers. This includes arc_buf_hdr_t structures
481 * (allocated via arc_buf_hdr_t_full and arc_buf_hdr_t_l2only
482 * caches), and arc_buf_t structures (allocated via arc_buf_t
485 kstat_named_t arcstat_hdr_size
;
487 * Number of bytes consumed by ARC buffers of type equal to
488 * ARC_BUFC_DATA. This is generally consumed by buffers backing
489 * on disk user data (e.g. plain file contents).
491 kstat_named_t arcstat_data_size
;
493 * Number of bytes consumed by ARC buffers of type equal to
494 * ARC_BUFC_METADATA. This is generally consumed by buffers
495 * backing on disk data that is used for internal ZFS
496 * structures (e.g. ZAP, dnode, indirect blocks, etc).
498 kstat_named_t arcstat_metadata_size
;
500 * Number of bytes consumed by dmu_buf_impl_t objects.
502 kstat_named_t arcstat_dbuf_size
;
504 * Number of bytes consumed by dnode_t objects.
506 kstat_named_t arcstat_dnode_size
;
508 * Number of bytes consumed by bonus buffers.
510 kstat_named_t arcstat_bonus_size
;
512 * Total number of bytes consumed by ARC buffers residing in the
513 * arc_anon state. This includes *all* buffers in the arc_anon
514 * state; e.g. data, metadata, evictable, and unevictable buffers
515 * are all included in this value.
517 kstat_named_t arcstat_anon_size
;
519 * Number of bytes consumed by ARC buffers that meet the
520 * following criteria: backing buffers of type ARC_BUFC_DATA,
521 * residing in the arc_anon state, and are eligible for eviction
522 * (e.g. have no outstanding holds on the buffer).
524 kstat_named_t arcstat_anon_evictable_data
;
526 * Number of bytes consumed by ARC buffers that meet the
527 * following criteria: backing buffers of type ARC_BUFC_METADATA,
528 * residing in the arc_anon state, and are eligible for eviction
529 * (e.g. have no outstanding holds on the buffer).
531 kstat_named_t arcstat_anon_evictable_metadata
;
533 * Total number of bytes consumed by ARC buffers residing in the
534 * arc_mru state. This includes *all* buffers in the arc_mru
535 * state; e.g. data, metadata, evictable, and unevictable buffers
536 * are all included in this value.
538 kstat_named_t arcstat_mru_size
;
540 * Number of bytes consumed by ARC buffers that meet the
541 * following criteria: backing buffers of type ARC_BUFC_DATA,
542 * residing in the arc_mru state, and are eligible for eviction
543 * (e.g. have no outstanding holds on the buffer).
545 kstat_named_t arcstat_mru_evictable_data
;
547 * Number of bytes consumed by ARC buffers that meet the
548 * following criteria: backing buffers of type ARC_BUFC_METADATA,
549 * residing in the arc_mru state, and are eligible for eviction
550 * (e.g. have no outstanding holds on the buffer).
552 kstat_named_t arcstat_mru_evictable_metadata
;
554 * Total number of bytes that *would have been* consumed by ARC
555 * buffers in the arc_mru_ghost state. The key thing to note
556 * here, is the fact that this size doesn't actually indicate
557 * RAM consumption. The ghost lists only consist of headers and
558 * don't actually have ARC buffers linked off of these headers.
559 * Thus, *if* the headers had associated ARC buffers, these
560 * buffers *would have* consumed this number of bytes.
562 kstat_named_t arcstat_mru_ghost_size
;
564 * Number of bytes that *would have been* consumed by ARC
565 * buffers that are eligible for eviction, of type
566 * ARC_BUFC_DATA, and linked off the arc_mru_ghost state.
568 kstat_named_t arcstat_mru_ghost_evictable_data
;
570 * Number of bytes that *would have been* consumed by ARC
571 * buffers that are eligible for eviction, of type
572 * ARC_BUFC_METADATA, and linked off the arc_mru_ghost state.
574 kstat_named_t arcstat_mru_ghost_evictable_metadata
;
576 * Total number of bytes consumed by ARC buffers residing in the
577 * arc_mfu state. This includes *all* buffers in the arc_mfu
578 * state; e.g. data, metadata, evictable, and unevictable buffers
579 * are all included in this value.
581 kstat_named_t arcstat_mfu_size
;
583 * Number of bytes consumed by ARC buffers that are eligible for
584 * eviction, of type ARC_BUFC_DATA, and reside in the arc_mfu
587 kstat_named_t arcstat_mfu_evictable_data
;
589 * Number of bytes consumed by ARC buffers that are eligible for
590 * eviction, of type ARC_BUFC_METADATA, and reside in the
593 kstat_named_t arcstat_mfu_evictable_metadata
;
595 * Total number of bytes that *would have been* consumed by ARC
596 * buffers in the arc_mfu_ghost state. See the comment above
597 * arcstat_mru_ghost_size for more details.
599 kstat_named_t arcstat_mfu_ghost_size
;
601 * Number of bytes that *would have been* consumed by ARC
602 * buffers that are eligible for eviction, of type
603 * ARC_BUFC_DATA, and linked off the arc_mfu_ghost state.
605 kstat_named_t arcstat_mfu_ghost_evictable_data
;
607 * Number of bytes that *would have been* consumed by ARC
608 * buffers that are eligible for eviction, of type
609 * ARC_BUFC_METADATA, and linked off the arc_mru_ghost state.
611 kstat_named_t arcstat_mfu_ghost_evictable_metadata
;
612 kstat_named_t arcstat_l2_hits
;
613 kstat_named_t arcstat_l2_misses
;
614 kstat_named_t arcstat_l2_feeds
;
615 kstat_named_t arcstat_l2_rw_clash
;
616 kstat_named_t arcstat_l2_read_bytes
;
617 kstat_named_t arcstat_l2_write_bytes
;
618 kstat_named_t arcstat_l2_writes_sent
;
619 kstat_named_t arcstat_l2_writes_done
;
620 kstat_named_t arcstat_l2_writes_error
;
621 kstat_named_t arcstat_l2_writes_lock_retry
;
622 kstat_named_t arcstat_l2_evict_lock_retry
;
623 kstat_named_t arcstat_l2_evict_reading
;
624 kstat_named_t arcstat_l2_evict_l1cached
;
625 kstat_named_t arcstat_l2_free_on_write
;
626 kstat_named_t arcstat_l2_abort_lowmem
;
627 kstat_named_t arcstat_l2_cksum_bad
;
628 kstat_named_t arcstat_l2_io_error
;
629 kstat_named_t arcstat_l2_size
;
630 kstat_named_t arcstat_l2_asize
;
631 kstat_named_t arcstat_l2_hdr_size
;
632 kstat_named_t arcstat_memory_throttle_count
;
633 kstat_named_t arcstat_memory_direct_count
;
634 kstat_named_t arcstat_memory_indirect_count
;
635 kstat_named_t arcstat_no_grow
;
636 kstat_named_t arcstat_tempreserve
;
637 kstat_named_t arcstat_loaned_bytes
;
638 kstat_named_t arcstat_prune
;
639 kstat_named_t arcstat_meta_used
;
640 kstat_named_t arcstat_meta_limit
;
641 kstat_named_t arcstat_dnode_limit
;
642 kstat_named_t arcstat_meta_max
;
643 kstat_named_t arcstat_meta_min
;
644 kstat_named_t arcstat_sync_wait_for_async
;
645 kstat_named_t arcstat_demand_hit_predictive_prefetch
;
646 kstat_named_t arcstat_need_free
;
647 kstat_named_t arcstat_sys_free
;
650 static arc_stats_t arc_stats
= {
651 { "hits", KSTAT_DATA_UINT64
},
652 { "misses", KSTAT_DATA_UINT64
},
653 { "demand_data_hits", KSTAT_DATA_UINT64
},
654 { "demand_data_misses", KSTAT_DATA_UINT64
},
655 { "demand_metadata_hits", KSTAT_DATA_UINT64
},
656 { "demand_metadata_misses", KSTAT_DATA_UINT64
},
657 { "prefetch_data_hits", KSTAT_DATA_UINT64
},
658 { "prefetch_data_misses", KSTAT_DATA_UINT64
},
659 { "prefetch_metadata_hits", KSTAT_DATA_UINT64
},
660 { "prefetch_metadata_misses", KSTAT_DATA_UINT64
},
661 { "mru_hits", KSTAT_DATA_UINT64
},
662 { "mru_ghost_hits", KSTAT_DATA_UINT64
},
663 { "mfu_hits", KSTAT_DATA_UINT64
},
664 { "mfu_ghost_hits", KSTAT_DATA_UINT64
},
665 { "deleted", KSTAT_DATA_UINT64
},
666 { "mutex_miss", KSTAT_DATA_UINT64
},
667 { "evict_skip", KSTAT_DATA_UINT64
},
668 { "evict_not_enough", KSTAT_DATA_UINT64
},
669 { "evict_l2_cached", KSTAT_DATA_UINT64
},
670 { "evict_l2_eligible", KSTAT_DATA_UINT64
},
671 { "evict_l2_ineligible", KSTAT_DATA_UINT64
},
672 { "evict_l2_skip", KSTAT_DATA_UINT64
},
673 { "hash_elements", KSTAT_DATA_UINT64
},
674 { "hash_elements_max", KSTAT_DATA_UINT64
},
675 { "hash_collisions", KSTAT_DATA_UINT64
},
676 { "hash_chains", KSTAT_DATA_UINT64
},
677 { "hash_chain_max", KSTAT_DATA_UINT64
},
678 { "p", KSTAT_DATA_UINT64
},
679 { "c", KSTAT_DATA_UINT64
},
680 { "c_min", KSTAT_DATA_UINT64
},
681 { "c_max", KSTAT_DATA_UINT64
},
682 { "size", KSTAT_DATA_UINT64
},
683 { "compressed_size", KSTAT_DATA_UINT64
},
684 { "uncompressed_size", KSTAT_DATA_UINT64
},
685 { "overhead_size", KSTAT_DATA_UINT64
},
686 { "hdr_size", KSTAT_DATA_UINT64
},
687 { "data_size", KSTAT_DATA_UINT64
},
688 { "metadata_size", KSTAT_DATA_UINT64
},
689 { "dbuf_size", KSTAT_DATA_UINT64
},
690 { "dnode_size", KSTAT_DATA_UINT64
},
691 { "bonus_size", KSTAT_DATA_UINT64
},
692 { "anon_size", KSTAT_DATA_UINT64
},
693 { "anon_evictable_data", KSTAT_DATA_UINT64
},
694 { "anon_evictable_metadata", KSTAT_DATA_UINT64
},
695 { "mru_size", KSTAT_DATA_UINT64
},
696 { "mru_evictable_data", KSTAT_DATA_UINT64
},
697 { "mru_evictable_metadata", KSTAT_DATA_UINT64
},
698 { "mru_ghost_size", KSTAT_DATA_UINT64
},
699 { "mru_ghost_evictable_data", KSTAT_DATA_UINT64
},
700 { "mru_ghost_evictable_metadata", KSTAT_DATA_UINT64
},
701 { "mfu_size", KSTAT_DATA_UINT64
},
702 { "mfu_evictable_data", KSTAT_DATA_UINT64
},
703 { "mfu_evictable_metadata", KSTAT_DATA_UINT64
},
704 { "mfu_ghost_size", KSTAT_DATA_UINT64
},
705 { "mfu_ghost_evictable_data", KSTAT_DATA_UINT64
},
706 { "mfu_ghost_evictable_metadata", KSTAT_DATA_UINT64
},
707 { "l2_hits", KSTAT_DATA_UINT64
},
708 { "l2_misses", KSTAT_DATA_UINT64
},
709 { "l2_feeds", KSTAT_DATA_UINT64
},
710 { "l2_rw_clash", KSTAT_DATA_UINT64
},
711 { "l2_read_bytes", KSTAT_DATA_UINT64
},
712 { "l2_write_bytes", KSTAT_DATA_UINT64
},
713 { "l2_writes_sent", KSTAT_DATA_UINT64
},
714 { "l2_writes_done", KSTAT_DATA_UINT64
},
715 { "l2_writes_error", KSTAT_DATA_UINT64
},
716 { "l2_writes_lock_retry", KSTAT_DATA_UINT64
},
717 { "l2_evict_lock_retry", KSTAT_DATA_UINT64
},
718 { "l2_evict_reading", KSTAT_DATA_UINT64
},
719 { "l2_evict_l1cached", KSTAT_DATA_UINT64
},
720 { "l2_free_on_write", KSTAT_DATA_UINT64
},
721 { "l2_abort_lowmem", KSTAT_DATA_UINT64
},
722 { "l2_cksum_bad", KSTAT_DATA_UINT64
},
723 { "l2_io_error", KSTAT_DATA_UINT64
},
724 { "l2_size", KSTAT_DATA_UINT64
},
725 { "l2_asize", KSTAT_DATA_UINT64
},
726 { "l2_hdr_size", KSTAT_DATA_UINT64
},
727 { "memory_throttle_count", KSTAT_DATA_UINT64
},
728 { "memory_direct_count", KSTAT_DATA_UINT64
},
729 { "memory_indirect_count", KSTAT_DATA_UINT64
},
730 { "arc_no_grow", KSTAT_DATA_UINT64
},
731 { "arc_tempreserve", KSTAT_DATA_UINT64
},
732 { "arc_loaned_bytes", KSTAT_DATA_UINT64
},
733 { "arc_prune", KSTAT_DATA_UINT64
},
734 { "arc_meta_used", KSTAT_DATA_UINT64
},
735 { "arc_meta_limit", KSTAT_DATA_UINT64
},
736 { "arc_dnode_limit", KSTAT_DATA_UINT64
},
737 { "arc_meta_max", KSTAT_DATA_UINT64
},
738 { "arc_meta_min", KSTAT_DATA_UINT64
},
739 { "sync_wait_for_async", KSTAT_DATA_UINT64
},
740 { "demand_hit_predictive_prefetch", KSTAT_DATA_UINT64
},
741 { "arc_need_free", KSTAT_DATA_UINT64
},
742 { "arc_sys_free", KSTAT_DATA_UINT64
}
745 #define ARCSTAT(stat) (arc_stats.stat.value.ui64)
747 #define ARCSTAT_INCR(stat, val) \
748 atomic_add_64(&arc_stats.stat.value.ui64, (val))
750 #define ARCSTAT_BUMP(stat) ARCSTAT_INCR(stat, 1)
751 #define ARCSTAT_BUMPDOWN(stat) ARCSTAT_INCR(stat, -1)
753 #define ARCSTAT_MAX(stat, val) { \
755 while ((val) > (m = arc_stats.stat.value.ui64) && \
756 (m != atomic_cas_64(&arc_stats.stat.value.ui64, m, (val)))) \
760 #define ARCSTAT_MAXSTAT(stat) \
761 ARCSTAT_MAX(stat##_max, arc_stats.stat.value.ui64)
764 * We define a macro to allow ARC hits/misses to be easily broken down by
765 * two separate conditions, giving a total of four different subtypes for
766 * each of hits and misses (so eight statistics total).
768 #define ARCSTAT_CONDSTAT(cond1, stat1, notstat1, cond2, stat2, notstat2, stat) \
771 ARCSTAT_BUMP(arcstat_##stat1##_##stat2##_##stat); \
773 ARCSTAT_BUMP(arcstat_##stat1##_##notstat2##_##stat); \
777 ARCSTAT_BUMP(arcstat_##notstat1##_##stat2##_##stat); \
779 ARCSTAT_BUMP(arcstat_##notstat1##_##notstat2##_##stat);\
784 static arc_state_t
*arc_anon
;
785 static arc_state_t
*arc_mru
;
786 static arc_state_t
*arc_mru_ghost
;
787 static arc_state_t
*arc_mfu
;
788 static arc_state_t
*arc_mfu_ghost
;
789 static arc_state_t
*arc_l2c_only
;
792 * There are several ARC variables that are critical to export as kstats --
793 * but we don't want to have to grovel around in the kstat whenever we wish to
794 * manipulate them. For these variables, we therefore define them to be in
795 * terms of the statistic variable. This assures that we are not introducing
796 * the possibility of inconsistency by having shadow copies of the variables,
797 * while still allowing the code to be readable.
799 #define arc_size ARCSTAT(arcstat_size) /* actual total arc size */
800 #define arc_p ARCSTAT(arcstat_p) /* target size of MRU */
801 #define arc_c ARCSTAT(arcstat_c) /* target size of cache */
802 #define arc_c_min ARCSTAT(arcstat_c_min) /* min target cache size */
803 #define arc_c_max ARCSTAT(arcstat_c_max) /* max target cache size */
804 #define arc_no_grow ARCSTAT(arcstat_no_grow) /* do not grow cache size */
805 #define arc_tempreserve ARCSTAT(arcstat_tempreserve)
806 #define arc_loaned_bytes ARCSTAT(arcstat_loaned_bytes)
807 #define arc_meta_limit ARCSTAT(arcstat_meta_limit) /* max size for metadata */
808 #define arc_dnode_limit ARCSTAT(arcstat_dnode_limit) /* max size for dnodes */
809 #define arc_meta_min ARCSTAT(arcstat_meta_min) /* min size for metadata */
810 #define arc_meta_used ARCSTAT(arcstat_meta_used) /* size of metadata */
811 #define arc_meta_max ARCSTAT(arcstat_meta_max) /* max size of metadata */
812 #define arc_dbuf_size ARCSTAT(arcstat_dbuf_size) /* dbuf metadata */
813 #define arc_dnode_size ARCSTAT(arcstat_dnode_size) /* dnode metadata */
814 #define arc_bonus_size ARCSTAT(arcstat_bonus_size) /* bonus buffer metadata */
815 #define arc_need_free ARCSTAT(arcstat_need_free) /* bytes to be freed */
816 #define arc_sys_free ARCSTAT(arcstat_sys_free) /* target system free bytes */
818 /* compressed size of entire arc */
819 #define arc_compressed_size ARCSTAT(arcstat_compressed_size)
820 /* uncompressed size of entire arc */
821 #define arc_uncompressed_size ARCSTAT(arcstat_uncompressed_size)
822 /* number of bytes in the arc from arc_buf_t's */
823 #define arc_overhead_size ARCSTAT(arcstat_overhead_size)
825 static list_t arc_prune_list
;
826 static kmutex_t arc_prune_mtx
;
827 static taskq_t
*arc_prune_taskq
;
829 #define GHOST_STATE(state) \
830 ((state) == arc_mru_ghost || (state) == arc_mfu_ghost || \
831 (state) == arc_l2c_only)
833 #define HDR_IN_HASH_TABLE(hdr) ((hdr)->b_flags & ARC_FLAG_IN_HASH_TABLE)
834 #define HDR_IO_IN_PROGRESS(hdr) ((hdr)->b_flags & ARC_FLAG_IO_IN_PROGRESS)
835 #define HDR_IO_ERROR(hdr) ((hdr)->b_flags & ARC_FLAG_IO_ERROR)
836 #define HDR_PREFETCH(hdr) ((hdr)->b_flags & ARC_FLAG_PREFETCH)
837 #define HDR_COMPRESSION_ENABLED(hdr) \
838 ((hdr)->b_flags & ARC_FLAG_COMPRESSED_ARC)
840 #define HDR_L2CACHE(hdr) ((hdr)->b_flags & ARC_FLAG_L2CACHE)
841 #define HDR_L2_READING(hdr) \
842 (((hdr)->b_flags & ARC_FLAG_IO_IN_PROGRESS) && \
843 ((hdr)->b_flags & ARC_FLAG_HAS_L2HDR))
844 #define HDR_L2_WRITING(hdr) ((hdr)->b_flags & ARC_FLAG_L2_WRITING)
845 #define HDR_L2_EVICTED(hdr) ((hdr)->b_flags & ARC_FLAG_L2_EVICTED)
846 #define HDR_L2_WRITE_HEAD(hdr) ((hdr)->b_flags & ARC_FLAG_L2_WRITE_HEAD)
847 #define HDR_SHARED_DATA(hdr) ((hdr)->b_flags & ARC_FLAG_SHARED_DATA)
849 #define HDR_ISTYPE_METADATA(hdr) \
850 ((hdr)->b_flags & ARC_FLAG_BUFC_METADATA)
851 #define HDR_ISTYPE_DATA(hdr) (!HDR_ISTYPE_METADATA(hdr))
853 #define HDR_HAS_L1HDR(hdr) ((hdr)->b_flags & ARC_FLAG_HAS_L1HDR)
854 #define HDR_HAS_L2HDR(hdr) ((hdr)->b_flags & ARC_FLAG_HAS_L2HDR)
856 /* For storing compression mode in b_flags */
857 #define HDR_COMPRESS_OFFSET (highbit64(ARC_FLAG_COMPRESS_0) - 1)
859 #define HDR_GET_COMPRESS(hdr) ((enum zio_compress)BF32_GET((hdr)->b_flags, \
860 HDR_COMPRESS_OFFSET, SPA_COMPRESSBITS))
861 #define HDR_SET_COMPRESS(hdr, cmp) BF32_SET((hdr)->b_flags, \
862 HDR_COMPRESS_OFFSET, SPA_COMPRESSBITS, (cmp));
864 #define ARC_BUF_LAST(buf) ((buf)->b_next == NULL)
865 #define ARC_BUF_SHARED(buf) ((buf)->b_flags & ARC_BUF_FLAG_SHARED)
866 #define ARC_BUF_COMPRESSED(buf) ((buf)->b_flags & ARC_BUF_FLAG_COMPRESSED)
872 #define HDR_FULL_SIZE ((int64_t)sizeof (arc_buf_hdr_t))
873 #define HDR_L2ONLY_SIZE ((int64_t)offsetof(arc_buf_hdr_t, b_l1hdr))
876 * Hash table routines
879 #define HT_LOCK_ALIGN 64
880 #define HT_LOCK_PAD (P2NPHASE(sizeof (kmutex_t), (HT_LOCK_ALIGN)))
885 unsigned char pad
[HT_LOCK_PAD
];
889 #define BUF_LOCKS 8192
890 typedef struct buf_hash_table
{
892 arc_buf_hdr_t
**ht_table
;
893 struct ht_lock ht_locks
[BUF_LOCKS
];
896 static buf_hash_table_t buf_hash_table
;
898 #define BUF_HASH_INDEX(spa, dva, birth) \
899 (buf_hash(spa, dva, birth) & buf_hash_table.ht_mask)
900 #define BUF_HASH_LOCK_NTRY(idx) (buf_hash_table.ht_locks[idx & (BUF_LOCKS-1)])
901 #define BUF_HASH_LOCK(idx) (&(BUF_HASH_LOCK_NTRY(idx).ht_lock))
902 #define HDR_LOCK(hdr) \
903 (BUF_HASH_LOCK(BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth)))
905 uint64_t zfs_crc64_table
[256];
911 #define L2ARC_WRITE_SIZE (8 * 1024 * 1024) /* initial write max */
912 #define L2ARC_HEADROOM 2 /* num of writes */
915 * If we discover during ARC scan any buffers to be compressed, we boost
916 * our headroom for the next scanning cycle by this percentage multiple.
918 #define L2ARC_HEADROOM_BOOST 200
919 #define L2ARC_FEED_SECS 1 /* caching interval secs */
920 #define L2ARC_FEED_MIN_MS 200 /* min caching interval ms */
923 * We can feed L2ARC from two states of ARC buffers, mru and mfu,
924 * and each of the state has two types: data and metadata.
926 #define L2ARC_FEED_TYPES 4
928 #define l2arc_writes_sent ARCSTAT(arcstat_l2_writes_sent)
929 #define l2arc_writes_done ARCSTAT(arcstat_l2_writes_done)
931 /* L2ARC Performance Tunables */
932 unsigned long l2arc_write_max
= L2ARC_WRITE_SIZE
; /* def max write size */
933 unsigned long l2arc_write_boost
= L2ARC_WRITE_SIZE
; /* extra warmup write */
934 unsigned long l2arc_headroom
= L2ARC_HEADROOM
; /* # of dev writes */
935 unsigned long l2arc_headroom_boost
= L2ARC_HEADROOM_BOOST
;
936 unsigned long l2arc_feed_secs
= L2ARC_FEED_SECS
; /* interval seconds */
937 unsigned long l2arc_feed_min_ms
= L2ARC_FEED_MIN_MS
; /* min interval msecs */
938 int l2arc_noprefetch
= B_TRUE
; /* don't cache prefetch bufs */
939 int l2arc_feed_again
= B_TRUE
; /* turbo warmup */
940 int l2arc_norw
= B_FALSE
; /* no reads during writes */
945 static list_t L2ARC_dev_list
; /* device list */
946 static list_t
*l2arc_dev_list
; /* device list pointer */
947 static kmutex_t l2arc_dev_mtx
; /* device list mutex */
948 static l2arc_dev_t
*l2arc_dev_last
; /* last device used */
949 static list_t L2ARC_free_on_write
; /* free after write buf list */
950 static list_t
*l2arc_free_on_write
; /* free after write list ptr */
951 static kmutex_t l2arc_free_on_write_mtx
; /* mutex for list */
952 static uint64_t l2arc_ndev
; /* number of devices */
954 typedef struct l2arc_read_callback
{
955 arc_buf_hdr_t
*l2rcb_hdr
; /* read header */
956 blkptr_t l2rcb_bp
; /* original blkptr */
957 zbookmark_phys_t l2rcb_zb
; /* original bookmark */
958 int l2rcb_flags
; /* original flags */
959 } l2arc_read_callback_t
;
961 typedef struct l2arc_data_free
{
962 /* protected by l2arc_free_on_write_mtx */
965 arc_buf_contents_t l2df_type
;
966 list_node_t l2df_list_node
;
969 static kmutex_t l2arc_feed_thr_lock
;
970 static kcondvar_t l2arc_feed_thr_cv
;
971 static uint8_t l2arc_thread_exit
;
973 static void *arc_get_data_buf(arc_buf_hdr_t
*, uint64_t, void *);
974 static void arc_free_data_buf(arc_buf_hdr_t
*, void *, uint64_t, void *);
975 static void arc_hdr_free_pdata(arc_buf_hdr_t
*hdr
);
976 static void arc_hdr_alloc_pdata(arc_buf_hdr_t
*);
977 static void arc_access(arc_buf_hdr_t
*, kmutex_t
*);
978 static boolean_t
arc_is_overflowing(void);
979 static void arc_buf_watch(arc_buf_t
*);
980 static void arc_tuning_update(void);
981 static void arc_prune_async(int64_t);
982 static uint64_t arc_all_memory(void);
984 static arc_buf_contents_t
arc_buf_type(arc_buf_hdr_t
*);
985 static uint32_t arc_bufc_to_flags(arc_buf_contents_t
);
986 static inline void arc_hdr_set_flags(arc_buf_hdr_t
*hdr
, arc_flags_t flags
);
987 static inline void arc_hdr_clear_flags(arc_buf_hdr_t
*hdr
, arc_flags_t flags
);
989 static boolean_t
l2arc_write_eligible(uint64_t, arc_buf_hdr_t
*);
990 static void l2arc_read_done(zio_t
*);
993 buf_hash(uint64_t spa
, const dva_t
*dva
, uint64_t birth
)
995 uint8_t *vdva
= (uint8_t *)dva
;
996 uint64_t crc
= -1ULL;
999 ASSERT(zfs_crc64_table
[128] == ZFS_CRC64_POLY
);
1001 for (i
= 0; i
< sizeof (dva_t
); i
++)
1002 crc
= (crc
>> 8) ^ zfs_crc64_table
[(crc
^ vdva
[i
]) & 0xFF];
1004 crc
^= (spa
>>8) ^ birth
;
1009 #define HDR_EMPTY(hdr) \
1010 ((hdr)->b_dva.dva_word[0] == 0 && \
1011 (hdr)->b_dva.dva_word[1] == 0)
1013 #define HDR_EQUAL(spa, dva, birth, hdr) \
1014 ((hdr)->b_dva.dva_word[0] == (dva)->dva_word[0]) && \
1015 ((hdr)->b_dva.dva_word[1] == (dva)->dva_word[1]) && \
1016 ((hdr)->b_birth == birth) && ((hdr)->b_spa == spa)
1019 buf_discard_identity(arc_buf_hdr_t
*hdr
)
1021 hdr
->b_dva
.dva_word
[0] = 0;
1022 hdr
->b_dva
.dva_word
[1] = 0;
1026 static arc_buf_hdr_t
*
1027 buf_hash_find(uint64_t spa
, const blkptr_t
*bp
, kmutex_t
**lockp
)
1029 const dva_t
*dva
= BP_IDENTITY(bp
);
1030 uint64_t birth
= BP_PHYSICAL_BIRTH(bp
);
1031 uint64_t idx
= BUF_HASH_INDEX(spa
, dva
, birth
);
1032 kmutex_t
*hash_lock
= BUF_HASH_LOCK(idx
);
1035 mutex_enter(hash_lock
);
1036 for (hdr
= buf_hash_table
.ht_table
[idx
]; hdr
!= NULL
;
1037 hdr
= hdr
->b_hash_next
) {
1038 if (HDR_EQUAL(spa
, dva
, birth
, hdr
)) {
1043 mutex_exit(hash_lock
);
1049 * Insert an entry into the hash table. If there is already an element
1050 * equal to elem in the hash table, then the already existing element
1051 * will be returned and the new element will not be inserted.
1052 * Otherwise returns NULL.
1053 * If lockp == NULL, the caller is assumed to already hold the hash lock.
1055 static arc_buf_hdr_t
*
1056 buf_hash_insert(arc_buf_hdr_t
*hdr
, kmutex_t
**lockp
)
1058 uint64_t idx
= BUF_HASH_INDEX(hdr
->b_spa
, &hdr
->b_dva
, hdr
->b_birth
);
1059 kmutex_t
*hash_lock
= BUF_HASH_LOCK(idx
);
1060 arc_buf_hdr_t
*fhdr
;
1063 ASSERT(!DVA_IS_EMPTY(&hdr
->b_dva
));
1064 ASSERT(hdr
->b_birth
!= 0);
1065 ASSERT(!HDR_IN_HASH_TABLE(hdr
));
1067 if (lockp
!= NULL
) {
1069 mutex_enter(hash_lock
);
1071 ASSERT(MUTEX_HELD(hash_lock
));
1074 for (fhdr
= buf_hash_table
.ht_table
[idx
], i
= 0; fhdr
!= NULL
;
1075 fhdr
= fhdr
->b_hash_next
, i
++) {
1076 if (HDR_EQUAL(hdr
->b_spa
, &hdr
->b_dva
, hdr
->b_birth
, fhdr
))
1080 hdr
->b_hash_next
= buf_hash_table
.ht_table
[idx
];
1081 buf_hash_table
.ht_table
[idx
] = hdr
;
1082 arc_hdr_set_flags(hdr
, ARC_FLAG_IN_HASH_TABLE
);
1084 /* collect some hash table performance data */
1086 ARCSTAT_BUMP(arcstat_hash_collisions
);
1088 ARCSTAT_BUMP(arcstat_hash_chains
);
1090 ARCSTAT_MAX(arcstat_hash_chain_max
, i
);
1093 ARCSTAT_BUMP(arcstat_hash_elements
);
1094 ARCSTAT_MAXSTAT(arcstat_hash_elements
);
1100 buf_hash_remove(arc_buf_hdr_t
*hdr
)
1102 arc_buf_hdr_t
*fhdr
, **hdrp
;
1103 uint64_t idx
= BUF_HASH_INDEX(hdr
->b_spa
, &hdr
->b_dva
, hdr
->b_birth
);
1105 ASSERT(MUTEX_HELD(BUF_HASH_LOCK(idx
)));
1106 ASSERT(HDR_IN_HASH_TABLE(hdr
));
1108 hdrp
= &buf_hash_table
.ht_table
[idx
];
1109 while ((fhdr
= *hdrp
) != hdr
) {
1110 ASSERT3P(fhdr
, !=, NULL
);
1111 hdrp
= &fhdr
->b_hash_next
;
1113 *hdrp
= hdr
->b_hash_next
;
1114 hdr
->b_hash_next
= NULL
;
1115 arc_hdr_clear_flags(hdr
, ARC_FLAG_IN_HASH_TABLE
);
1117 /* collect some hash table performance data */
1118 ARCSTAT_BUMPDOWN(arcstat_hash_elements
);
1120 if (buf_hash_table
.ht_table
[idx
] &&
1121 buf_hash_table
.ht_table
[idx
]->b_hash_next
== NULL
)
1122 ARCSTAT_BUMPDOWN(arcstat_hash_chains
);
1126 * Global data structures and functions for the buf kmem cache.
1128 static kmem_cache_t
*hdr_full_cache
;
1129 static kmem_cache_t
*hdr_l2only_cache
;
1130 static kmem_cache_t
*buf_cache
;
1137 #if defined(_KERNEL) && defined(HAVE_SPL)
1139 * Large allocations which do not require contiguous pages
1140 * should be using vmem_free() in the linux kernel\
1142 vmem_free(buf_hash_table
.ht_table
,
1143 (buf_hash_table
.ht_mask
+ 1) * sizeof (void *));
1145 kmem_free(buf_hash_table
.ht_table
,
1146 (buf_hash_table
.ht_mask
+ 1) * sizeof (void *));
1148 for (i
= 0; i
< BUF_LOCKS
; i
++)
1149 mutex_destroy(&buf_hash_table
.ht_locks
[i
].ht_lock
);
1150 kmem_cache_destroy(hdr_full_cache
);
1151 kmem_cache_destroy(hdr_l2only_cache
);
1152 kmem_cache_destroy(buf_cache
);
1156 * Constructor callback - called when the cache is empty
1157 * and a new buf is requested.
1161 hdr_full_cons(void *vbuf
, void *unused
, int kmflag
)
1163 arc_buf_hdr_t
*hdr
= vbuf
;
1165 bzero(hdr
, HDR_FULL_SIZE
);
1166 cv_init(&hdr
->b_l1hdr
.b_cv
, NULL
, CV_DEFAULT
, NULL
);
1167 refcount_create(&hdr
->b_l1hdr
.b_refcnt
);
1168 mutex_init(&hdr
->b_l1hdr
.b_freeze_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1169 list_link_init(&hdr
->b_l1hdr
.b_arc_node
);
1170 list_link_init(&hdr
->b_l2hdr
.b_l2node
);
1171 multilist_link_init(&hdr
->b_l1hdr
.b_arc_node
);
1172 arc_space_consume(HDR_FULL_SIZE
, ARC_SPACE_HDRS
);
1179 hdr_l2only_cons(void *vbuf
, void *unused
, int kmflag
)
1181 arc_buf_hdr_t
*hdr
= vbuf
;
1183 bzero(hdr
, HDR_L2ONLY_SIZE
);
1184 arc_space_consume(HDR_L2ONLY_SIZE
, ARC_SPACE_L2HDRS
);
1191 buf_cons(void *vbuf
, void *unused
, int kmflag
)
1193 arc_buf_t
*buf
= vbuf
;
1195 bzero(buf
, sizeof (arc_buf_t
));
1196 mutex_init(&buf
->b_evict_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1197 arc_space_consume(sizeof (arc_buf_t
), ARC_SPACE_HDRS
);
1203 * Destructor callback - called when a cached buf is
1204 * no longer required.
1208 hdr_full_dest(void *vbuf
, void *unused
)
1210 arc_buf_hdr_t
*hdr
= vbuf
;
1212 ASSERT(HDR_EMPTY(hdr
));
1213 cv_destroy(&hdr
->b_l1hdr
.b_cv
);
1214 refcount_destroy(&hdr
->b_l1hdr
.b_refcnt
);
1215 mutex_destroy(&hdr
->b_l1hdr
.b_freeze_lock
);
1216 ASSERT(!multilist_link_active(&hdr
->b_l1hdr
.b_arc_node
));
1217 arc_space_return(HDR_FULL_SIZE
, ARC_SPACE_HDRS
);
1222 hdr_l2only_dest(void *vbuf
, void *unused
)
1224 ASSERTV(arc_buf_hdr_t
*hdr
= vbuf
);
1226 ASSERT(HDR_EMPTY(hdr
));
1227 arc_space_return(HDR_L2ONLY_SIZE
, ARC_SPACE_L2HDRS
);
1232 buf_dest(void *vbuf
, void *unused
)
1234 arc_buf_t
*buf
= vbuf
;
1236 mutex_destroy(&buf
->b_evict_lock
);
1237 arc_space_return(sizeof (arc_buf_t
), ARC_SPACE_HDRS
);
1241 * Reclaim callback -- invoked when memory is low.
1245 hdr_recl(void *unused
)
1247 dprintf("hdr_recl called\n");
1249 * umem calls the reclaim func when we destroy the buf cache,
1250 * which is after we do arc_fini().
1253 cv_signal(&arc_reclaim_thread_cv
);
1259 uint64_t *ct
= NULL
;
1260 uint64_t hsize
= 1ULL << 12;
1264 * The hash table is big enough to fill all of physical memory
1265 * with an average block size of zfs_arc_average_blocksize (default 8K).
1266 * By default, the table will take up
1267 * totalmem * sizeof(void*) / 8K (1MB per GB with 8-byte pointers).
1269 while (hsize
* zfs_arc_average_blocksize
< arc_all_memory())
1272 buf_hash_table
.ht_mask
= hsize
- 1;
1273 #if defined(_KERNEL) && defined(HAVE_SPL)
1275 * Large allocations which do not require contiguous pages
1276 * should be using vmem_alloc() in the linux kernel
1278 buf_hash_table
.ht_table
=
1279 vmem_zalloc(hsize
* sizeof (void*), KM_SLEEP
);
1281 buf_hash_table
.ht_table
=
1282 kmem_zalloc(hsize
* sizeof (void*), KM_NOSLEEP
);
1284 if (buf_hash_table
.ht_table
== NULL
) {
1285 ASSERT(hsize
> (1ULL << 8));
1290 hdr_full_cache
= kmem_cache_create("arc_buf_hdr_t_full", HDR_FULL_SIZE
,
1291 0, hdr_full_cons
, hdr_full_dest
, hdr_recl
, NULL
, NULL
, 0);
1292 hdr_l2only_cache
= kmem_cache_create("arc_buf_hdr_t_l2only",
1293 HDR_L2ONLY_SIZE
, 0, hdr_l2only_cons
, hdr_l2only_dest
, hdr_recl
,
1295 buf_cache
= kmem_cache_create("arc_buf_t", sizeof (arc_buf_t
),
1296 0, buf_cons
, buf_dest
, NULL
, NULL
, NULL
, 0);
1298 for (i
= 0; i
< 256; i
++)
1299 for (ct
= zfs_crc64_table
+ i
, *ct
= i
, j
= 8; j
> 0; j
--)
1300 *ct
= (*ct
>> 1) ^ (-(*ct
& 1) & ZFS_CRC64_POLY
);
1302 for (i
= 0; i
< BUF_LOCKS
; i
++) {
1303 mutex_init(&buf_hash_table
.ht_locks
[i
].ht_lock
,
1304 NULL
, MUTEX_DEFAULT
, NULL
);
1308 #define ARC_MINTIME (hz>>4) /* 62 ms */
1311 * This is the size that the buf occupies in memory. If the buf is compressed,
1312 * it will correspond to the compressed size. You should use this method of
1313 * getting the buf size unless you explicitly need the logical size.
1316 arc_buf_size(arc_buf_t
*buf
)
1318 return (ARC_BUF_COMPRESSED(buf
) ?
1319 HDR_GET_PSIZE(buf
->b_hdr
) : HDR_GET_LSIZE(buf
->b_hdr
));
1323 arc_buf_lsize(arc_buf_t
*buf
)
1325 return (HDR_GET_LSIZE(buf
->b_hdr
));
1329 arc_get_compression(arc_buf_t
*buf
)
1331 return (ARC_BUF_COMPRESSED(buf
) ?
1332 HDR_GET_COMPRESS(buf
->b_hdr
) : ZIO_COMPRESS_OFF
);
1335 static inline boolean_t
1336 arc_buf_is_shared(arc_buf_t
*buf
)
1338 boolean_t shared
= (buf
->b_data
!= NULL
&&
1339 buf
->b_data
== buf
->b_hdr
->b_l1hdr
.b_pdata
);
1340 IMPLY(shared
, HDR_SHARED_DATA(buf
->b_hdr
));
1341 IMPLY(shared
, ARC_BUF_SHARED(buf
));
1342 IMPLY(shared
, ARC_BUF_COMPRESSED(buf
) || ARC_BUF_LAST(buf
));
1345 * It would be nice to assert arc_can_share() too, but the "hdr isn't
1346 * already being shared" requirement prevents us from doing that.
1353 arc_cksum_free(arc_buf_hdr_t
*hdr
)
1355 ASSERT(HDR_HAS_L1HDR(hdr
));
1356 mutex_enter(&hdr
->b_l1hdr
.b_freeze_lock
);
1357 if (hdr
->b_l1hdr
.b_freeze_cksum
!= NULL
) {
1358 kmem_free(hdr
->b_l1hdr
.b_freeze_cksum
, sizeof (zio_cksum_t
));
1359 hdr
->b_l1hdr
.b_freeze_cksum
= NULL
;
1361 mutex_exit(&hdr
->b_l1hdr
.b_freeze_lock
);
1365 * If we've turned on the ZFS_DEBUG_MODIFY flag, verify that the buf's data
1366 * matches the checksum that is stored in the hdr. If there is no checksum,
1367 * or if the buf is compressed, this is a no-op.
1370 arc_cksum_verify(arc_buf_t
*buf
)
1372 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
1375 if (!(zfs_flags
& ZFS_DEBUG_MODIFY
))
1378 if (ARC_BUF_COMPRESSED(buf
)) {
1379 ASSERT(hdr
->b_l1hdr
.b_freeze_cksum
== NULL
||
1380 hdr
->b_l1hdr
.b_bufcnt
> 1);
1384 ASSERT(HDR_HAS_L1HDR(hdr
));
1386 mutex_enter(&hdr
->b_l1hdr
.b_freeze_lock
);
1387 if (hdr
->b_l1hdr
.b_freeze_cksum
== NULL
|| HDR_IO_ERROR(hdr
)) {
1388 mutex_exit(&hdr
->b_l1hdr
.b_freeze_lock
);
1392 fletcher_2_native(buf
->b_data
, arc_buf_size(buf
), NULL
, &zc
);
1393 if (!ZIO_CHECKSUM_EQUAL(*hdr
->b_l1hdr
.b_freeze_cksum
, zc
))
1394 panic("buffer modified while frozen!");
1395 mutex_exit(&hdr
->b_l1hdr
.b_freeze_lock
);
1399 arc_cksum_is_equal(arc_buf_hdr_t
*hdr
, zio_t
*zio
)
1401 enum zio_compress compress
= BP_GET_COMPRESS(zio
->io_bp
);
1402 boolean_t valid_cksum
;
1404 ASSERT(!BP_IS_EMBEDDED(zio
->io_bp
));
1405 VERIFY3U(BP_GET_PSIZE(zio
->io_bp
), ==, HDR_GET_PSIZE(hdr
));
1408 * We rely on the blkptr's checksum to determine if the block
1409 * is valid or not. When compressed arc is enabled, the l2arc
1410 * writes the block to the l2arc just as it appears in the pool.
1411 * This allows us to use the blkptr's checksum to validate the
1412 * data that we just read off of the l2arc without having to store
1413 * a separate checksum in the arc_buf_hdr_t. However, if compressed
1414 * arc is disabled, then the data written to the l2arc is always
1415 * uncompressed and won't match the block as it exists in the main
1416 * pool. When this is the case, we must first compress it if it is
1417 * compressed on the main pool before we can validate the checksum.
1419 if (!HDR_COMPRESSION_ENABLED(hdr
) && compress
!= ZIO_COMPRESS_OFF
) {
1423 ASSERT3U(HDR_GET_COMPRESS(hdr
), ==, ZIO_COMPRESS_OFF
);
1425 cbuf
= zio_buf_alloc(HDR_GET_PSIZE(hdr
));
1426 lsize
= HDR_GET_LSIZE(hdr
);
1427 csize
= zio_compress_data(compress
, zio
->io_data
, 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_data
, 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 ASSERT(!ARC_BUF_COMPRESSED(buf
) || hdr
->b_l1hdr
.b_bufcnt
> 1);
1487 mutex_exit(&hdr
->b_l1hdr
.b_freeze_lock
);
1489 } else if (ARC_BUF_COMPRESSED(buf
)) {
1491 * Since the checksum doesn't apply to compressed buffers, we
1492 * only keep a checksum if there are uncompressed buffers.
1493 * Therefore there must be another buffer, which is
1496 IMPLY(hdr
->b_l1hdr
.b_freeze_cksum
!= NULL
,
1497 hdr
->b_l1hdr
.b_bufcnt
> 1);
1498 mutex_exit(&hdr
->b_l1hdr
.b_freeze_lock
);
1502 ASSERT(!ARC_BUF_COMPRESSED(buf
));
1503 hdr
->b_l1hdr
.b_freeze_cksum
= kmem_alloc(sizeof (zio_cksum_t
),
1505 fletcher_2_native(buf
->b_data
, arc_buf_size(buf
), NULL
,
1506 hdr
->b_l1hdr
.b_freeze_cksum
);
1507 mutex_exit(&hdr
->b_l1hdr
.b_freeze_lock
);
1513 arc_buf_sigsegv(int sig
, siginfo_t
*si
, void *unused
)
1515 panic("Got SIGSEGV at address: 0x%lx\n", (long) si
->si_addr
);
1521 arc_buf_unwatch(arc_buf_t
*buf
)
1525 ASSERT0(mprotect(buf
->b_data
, HDR_GET_LSIZE(buf
->b_hdr
),
1526 PROT_READ
| PROT_WRITE
));
1533 arc_buf_watch(arc_buf_t
*buf
)
1537 ASSERT0(mprotect(buf
->b_data
, arc_buf_size(buf
),
1542 static arc_buf_contents_t
1543 arc_buf_type(arc_buf_hdr_t
*hdr
)
1545 arc_buf_contents_t type
;
1546 if (HDR_ISTYPE_METADATA(hdr
)) {
1547 type
= ARC_BUFC_METADATA
;
1549 type
= ARC_BUFC_DATA
;
1551 VERIFY3U(hdr
->b_type
, ==, type
);
1556 arc_is_metadata(arc_buf_t
*buf
)
1558 return (HDR_ISTYPE_METADATA(buf
->b_hdr
) != 0);
1562 arc_bufc_to_flags(arc_buf_contents_t type
)
1566 /* metadata field is 0 if buffer contains normal data */
1568 case ARC_BUFC_METADATA
:
1569 return (ARC_FLAG_BUFC_METADATA
);
1573 panic("undefined ARC buffer type!");
1574 return ((uint32_t)-1);
1578 arc_buf_thaw(arc_buf_t
*buf
)
1580 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
1582 ASSERT3P(hdr
->b_l1hdr
.b_state
, ==, arc_anon
);
1583 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
1585 arc_cksum_verify(buf
);
1588 * Compressed buffers do not manipulate the b_freeze_cksum or
1589 * allocate b_thawed.
1591 if (ARC_BUF_COMPRESSED(buf
)) {
1592 ASSERT(hdr
->b_l1hdr
.b_freeze_cksum
== NULL
||
1593 hdr
->b_l1hdr
.b_bufcnt
> 1);
1597 ASSERT(HDR_HAS_L1HDR(hdr
));
1598 arc_cksum_free(hdr
);
1599 arc_buf_unwatch(buf
);
1603 arc_buf_freeze(arc_buf_t
*buf
)
1605 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
1606 kmutex_t
*hash_lock
;
1608 if (!(zfs_flags
& ZFS_DEBUG_MODIFY
))
1611 if (ARC_BUF_COMPRESSED(buf
)) {
1612 ASSERT(hdr
->b_l1hdr
.b_freeze_cksum
== NULL
||
1613 hdr
->b_l1hdr
.b_bufcnt
> 1);
1617 hash_lock
= HDR_LOCK(hdr
);
1618 mutex_enter(hash_lock
);
1620 ASSERT(HDR_HAS_L1HDR(hdr
));
1621 ASSERT(hdr
->b_l1hdr
.b_freeze_cksum
!= NULL
||
1622 hdr
->b_l1hdr
.b_state
== arc_anon
);
1623 arc_cksum_compute(buf
);
1624 mutex_exit(hash_lock
);
1628 * The arc_buf_hdr_t's b_flags should never be modified directly. Instead,
1629 * the following functions should be used to ensure that the flags are
1630 * updated in a thread-safe way. When manipulating the flags either
1631 * the hash_lock must be held or the hdr must be undiscoverable. This
1632 * ensures that we're not racing with any other threads when updating
1636 arc_hdr_set_flags(arc_buf_hdr_t
*hdr
, arc_flags_t flags
)
1638 ASSERT(MUTEX_HELD(HDR_LOCK(hdr
)) || HDR_EMPTY(hdr
));
1639 hdr
->b_flags
|= flags
;
1643 arc_hdr_clear_flags(arc_buf_hdr_t
*hdr
, arc_flags_t flags
)
1645 ASSERT(MUTEX_HELD(HDR_LOCK(hdr
)) || HDR_EMPTY(hdr
));
1646 hdr
->b_flags
&= ~flags
;
1650 * Setting the compression bits in the arc_buf_hdr_t's b_flags is
1651 * done in a special way since we have to clear and set bits
1652 * at the same time. Consumers that wish to set the compression bits
1653 * must use this function to ensure that the flags are updated in
1654 * thread-safe manner.
1657 arc_hdr_set_compress(arc_buf_hdr_t
*hdr
, enum zio_compress cmp
)
1659 ASSERT(MUTEX_HELD(HDR_LOCK(hdr
)) || HDR_EMPTY(hdr
));
1662 * Holes and embedded blocks will always have a psize = 0 so
1663 * we ignore the compression of the blkptr and set the
1664 * want to uncompress them. Mark them as uncompressed.
1666 if (!zfs_compressed_arc_enabled
|| HDR_GET_PSIZE(hdr
) == 0) {
1667 arc_hdr_clear_flags(hdr
, ARC_FLAG_COMPRESSED_ARC
);
1668 HDR_SET_COMPRESS(hdr
, ZIO_COMPRESS_OFF
);
1669 ASSERT(!HDR_COMPRESSION_ENABLED(hdr
));
1670 ASSERT3U(HDR_GET_COMPRESS(hdr
), ==, ZIO_COMPRESS_OFF
);
1672 arc_hdr_set_flags(hdr
, ARC_FLAG_COMPRESSED_ARC
);
1673 HDR_SET_COMPRESS(hdr
, cmp
);
1674 ASSERT3U(HDR_GET_COMPRESS(hdr
), ==, cmp
);
1675 ASSERT(HDR_COMPRESSION_ENABLED(hdr
));
1680 * Looks for another buf on the same hdr which has the data decompressed, copies
1681 * from it, and returns true. If no such buf exists, returns false.
1684 arc_buf_try_copy_decompressed_data(arc_buf_t
*buf
)
1686 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
1688 boolean_t copied
= B_FALSE
;
1690 ASSERT(HDR_HAS_L1HDR(hdr
));
1691 ASSERT3P(buf
->b_data
, !=, NULL
);
1692 ASSERT(!ARC_BUF_COMPRESSED(buf
));
1694 for (from
= hdr
->b_l1hdr
.b_buf
; from
!= NULL
;
1695 from
= from
->b_next
) {
1696 /* can't use our own data buffer */
1701 if (!ARC_BUF_COMPRESSED(from
)) {
1702 bcopy(from
->b_data
, buf
->b_data
, arc_buf_size(buf
));
1709 * There were no decompressed bufs, so there should not be a
1710 * checksum on the hdr either.
1712 EQUIV(!copied
, hdr
->b_l1hdr
.b_freeze_cksum
== NULL
);
1718 * Given a buf that has a data buffer attached to it, this function will
1719 * efficiently fill the buf with data of the specified compression setting from
1720 * the hdr and update the hdr's b_freeze_cksum if necessary. If the buf and hdr
1721 * are already sharing a data buf, no copy is performed.
1723 * If the buf is marked as compressed but uncompressed data was requested, this
1724 * will allocate a new data buffer for the buf, remove that flag, and fill the
1725 * buf with uncompressed data. You can't request a compressed buf on a hdr with
1726 * uncompressed data, and (since we haven't added support for it yet) if you
1727 * want compressed data your buf must already be marked as compressed and have
1728 * the correct-sized data buffer.
1731 arc_buf_fill(arc_buf_t
*buf
, boolean_t compressed
)
1733 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
1734 boolean_t hdr_compressed
= (HDR_GET_COMPRESS(hdr
) != ZIO_COMPRESS_OFF
);
1735 dmu_object_byteswap_t bswap
= hdr
->b_l1hdr
.b_byteswap
;
1737 ASSERT3P(buf
->b_data
, !=, NULL
);
1738 IMPLY(compressed
, hdr_compressed
);
1739 IMPLY(compressed
, ARC_BUF_COMPRESSED(buf
));
1741 if (hdr_compressed
== compressed
) {
1742 if (!arc_buf_is_shared(buf
)) {
1743 bcopy(hdr
->b_l1hdr
.b_pdata
, buf
->b_data
,
1747 ASSERT(hdr_compressed
);
1748 ASSERT(!compressed
);
1749 ASSERT3U(HDR_GET_LSIZE(hdr
), !=, HDR_GET_PSIZE(hdr
));
1752 * If the buf is sharing its data with the hdr, unlink it and
1753 * allocate a new data buffer for the buf.
1755 if (arc_buf_is_shared(buf
)) {
1756 ASSERT(ARC_BUF_COMPRESSED(buf
));
1758 /* We need to give the buf it's own b_data */
1759 buf
->b_flags
&= ~ARC_BUF_FLAG_SHARED
;
1761 arc_get_data_buf(hdr
, HDR_GET_LSIZE(hdr
), buf
);
1762 arc_hdr_clear_flags(hdr
, ARC_FLAG_SHARED_DATA
);
1764 /* Previously overhead was 0; just add new overhead */
1765 ARCSTAT_INCR(arcstat_overhead_size
, HDR_GET_LSIZE(hdr
));
1766 } else if (ARC_BUF_COMPRESSED(buf
)) {
1767 /* We need to reallocate the buf's b_data */
1768 arc_free_data_buf(hdr
, buf
->b_data
, HDR_GET_PSIZE(hdr
),
1771 arc_get_data_buf(hdr
, HDR_GET_LSIZE(hdr
), buf
);
1773 /* We increased the size of b_data; update overhead */
1774 ARCSTAT_INCR(arcstat_overhead_size
,
1775 HDR_GET_LSIZE(hdr
) - HDR_GET_PSIZE(hdr
));
1779 * Regardless of the buf's previous compression settings, it
1780 * should not be compressed at the end of this function.
1782 buf
->b_flags
&= ~ARC_BUF_FLAG_COMPRESSED
;
1785 * Try copying the data from another buf which already has a
1786 * decompressed version. If that's not possible, it's time to
1787 * bite the bullet and decompress the data from the hdr.
1789 if (arc_buf_try_copy_decompressed_data(buf
)) {
1790 /* Skip byteswapping and checksumming (already done) */
1791 ASSERT3P(hdr
->b_l1hdr
.b_freeze_cksum
, !=, NULL
);
1794 int error
= zio_decompress_data(HDR_GET_COMPRESS(hdr
),
1795 hdr
->b_l1hdr
.b_pdata
, buf
->b_data
,
1796 HDR_GET_PSIZE(hdr
), HDR_GET_LSIZE(hdr
));
1799 * Absent hardware errors or software bugs, this should
1800 * be impossible, but log it anyway so we can debug it.
1804 "hdr %p, compress %d, psize %d, lsize %d",
1805 hdr
, HDR_GET_COMPRESS(hdr
),
1806 HDR_GET_PSIZE(hdr
), HDR_GET_LSIZE(hdr
));
1807 return (SET_ERROR(EIO
));
1812 /* Byteswap the buf's data if necessary */
1813 if (bswap
!= DMU_BSWAP_NUMFUNCS
) {
1814 ASSERT(!HDR_SHARED_DATA(hdr
));
1815 ASSERT3U(bswap
, <, DMU_BSWAP_NUMFUNCS
);
1816 dmu_ot_byteswap
[bswap
].ob_func(buf
->b_data
, HDR_GET_LSIZE(hdr
));
1819 /* Compute the hdr's checksum if necessary */
1820 arc_cksum_compute(buf
);
1826 arc_decompress(arc_buf_t
*buf
)
1828 return (arc_buf_fill(buf
, B_FALSE
));
1832 * Return the size of the block, b_pdata, that is stored in the arc_buf_hdr_t.
1835 arc_hdr_size(arc_buf_hdr_t
*hdr
)
1839 if (HDR_GET_COMPRESS(hdr
) != ZIO_COMPRESS_OFF
&&
1840 HDR_GET_PSIZE(hdr
) > 0) {
1841 size
= HDR_GET_PSIZE(hdr
);
1843 ASSERT3U(HDR_GET_LSIZE(hdr
), !=, 0);
1844 size
= HDR_GET_LSIZE(hdr
);
1850 * Increment the amount of evictable space in the arc_state_t's refcount.
1851 * We account for the space used by the hdr and the arc buf individually
1852 * so that we can add and remove them from the refcount individually.
1855 arc_evictable_space_increment(arc_buf_hdr_t
*hdr
, arc_state_t
*state
)
1857 arc_buf_contents_t type
= arc_buf_type(hdr
);
1860 ASSERT(HDR_HAS_L1HDR(hdr
));
1862 if (GHOST_STATE(state
)) {
1863 ASSERT0(hdr
->b_l1hdr
.b_bufcnt
);
1864 ASSERT3P(hdr
->b_l1hdr
.b_buf
, ==, NULL
);
1865 ASSERT3P(hdr
->b_l1hdr
.b_pdata
, ==, NULL
);
1866 (void) refcount_add_many(&state
->arcs_esize
[type
],
1867 HDR_GET_LSIZE(hdr
), hdr
);
1871 ASSERT(!GHOST_STATE(state
));
1872 if (hdr
->b_l1hdr
.b_pdata
!= NULL
) {
1873 (void) refcount_add_many(&state
->arcs_esize
[type
],
1874 arc_hdr_size(hdr
), hdr
);
1876 for (buf
= hdr
->b_l1hdr
.b_buf
; buf
!= NULL
; buf
= buf
->b_next
) {
1877 if (arc_buf_is_shared(buf
))
1879 (void) refcount_add_many(&state
->arcs_esize
[type
],
1880 arc_buf_size(buf
), buf
);
1885 * Decrement the amount of evictable space in the arc_state_t's refcount.
1886 * We account for the space used by the hdr and the arc buf individually
1887 * so that we can add and remove them from the refcount individually.
1890 arc_evictable_space_decrement(arc_buf_hdr_t
*hdr
, arc_state_t
*state
)
1892 arc_buf_contents_t type
= arc_buf_type(hdr
);
1895 ASSERT(HDR_HAS_L1HDR(hdr
));
1897 if (GHOST_STATE(state
)) {
1898 ASSERT0(hdr
->b_l1hdr
.b_bufcnt
);
1899 ASSERT3P(hdr
->b_l1hdr
.b_buf
, ==, NULL
);
1900 ASSERT3P(hdr
->b_l1hdr
.b_pdata
, ==, NULL
);
1901 (void) refcount_remove_many(&state
->arcs_esize
[type
],
1902 HDR_GET_LSIZE(hdr
), hdr
);
1906 ASSERT(!GHOST_STATE(state
));
1907 if (hdr
->b_l1hdr
.b_pdata
!= NULL
) {
1908 (void) refcount_remove_many(&state
->arcs_esize
[type
],
1909 arc_hdr_size(hdr
), hdr
);
1911 for (buf
= hdr
->b_l1hdr
.b_buf
; buf
!= NULL
; buf
= buf
->b_next
) {
1912 if (arc_buf_is_shared(buf
))
1914 (void) refcount_remove_many(&state
->arcs_esize
[type
],
1915 arc_buf_size(buf
), buf
);
1920 * Add a reference to this hdr indicating that someone is actively
1921 * referencing that memory. When the refcount transitions from 0 to 1,
1922 * we remove it from the respective arc_state_t list to indicate that
1923 * it is not evictable.
1926 add_reference(arc_buf_hdr_t
*hdr
, void *tag
)
1930 ASSERT(HDR_HAS_L1HDR(hdr
));
1931 if (!MUTEX_HELD(HDR_LOCK(hdr
))) {
1932 ASSERT(hdr
->b_l1hdr
.b_state
== arc_anon
);
1933 ASSERT(refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
));
1934 ASSERT3P(hdr
->b_l1hdr
.b_buf
, ==, NULL
);
1937 state
= hdr
->b_l1hdr
.b_state
;
1939 if ((refcount_add(&hdr
->b_l1hdr
.b_refcnt
, tag
) == 1) &&
1940 (state
!= arc_anon
)) {
1941 /* We don't use the L2-only state list. */
1942 if (state
!= arc_l2c_only
) {
1943 multilist_remove(&state
->arcs_list
[arc_buf_type(hdr
)],
1945 arc_evictable_space_decrement(hdr
, state
);
1947 /* remove the prefetch flag if we get a reference */
1948 arc_hdr_clear_flags(hdr
, ARC_FLAG_PREFETCH
);
1953 * Remove a reference from this hdr. When the reference transitions from
1954 * 1 to 0 and we're not anonymous, then we add this hdr to the arc_state_t's
1955 * list making it eligible for eviction.
1958 remove_reference(arc_buf_hdr_t
*hdr
, kmutex_t
*hash_lock
, void *tag
)
1961 arc_state_t
*state
= hdr
->b_l1hdr
.b_state
;
1963 ASSERT(HDR_HAS_L1HDR(hdr
));
1964 ASSERT(state
== arc_anon
|| MUTEX_HELD(hash_lock
));
1965 ASSERT(!GHOST_STATE(state
));
1968 * arc_l2c_only counts as a ghost state so we don't need to explicitly
1969 * check to prevent usage of the arc_l2c_only list.
1971 if (((cnt
= refcount_remove(&hdr
->b_l1hdr
.b_refcnt
, tag
)) == 0) &&
1972 (state
!= arc_anon
)) {
1973 multilist_insert(&state
->arcs_list
[arc_buf_type(hdr
)], hdr
);
1974 ASSERT3U(hdr
->b_l1hdr
.b_bufcnt
, >, 0);
1975 arc_evictable_space_increment(hdr
, state
);
1981 * Returns detailed information about a specific arc buffer. When the
1982 * state_index argument is set the function will calculate the arc header
1983 * list position for its arc state. Since this requires a linear traversal
1984 * callers are strongly encourage not to do this. However, it can be helpful
1985 * for targeted analysis so the functionality is provided.
1988 arc_buf_info(arc_buf_t
*ab
, arc_buf_info_t
*abi
, int state_index
)
1990 arc_buf_hdr_t
*hdr
= ab
->b_hdr
;
1991 l1arc_buf_hdr_t
*l1hdr
= NULL
;
1992 l2arc_buf_hdr_t
*l2hdr
= NULL
;
1993 arc_state_t
*state
= NULL
;
1995 memset(abi
, 0, sizeof (arc_buf_info_t
));
2000 abi
->abi_flags
= hdr
->b_flags
;
2002 if (HDR_HAS_L1HDR(hdr
)) {
2003 l1hdr
= &hdr
->b_l1hdr
;
2004 state
= l1hdr
->b_state
;
2006 if (HDR_HAS_L2HDR(hdr
))
2007 l2hdr
= &hdr
->b_l2hdr
;
2010 abi
->abi_bufcnt
= l1hdr
->b_bufcnt
;
2011 abi
->abi_access
= l1hdr
->b_arc_access
;
2012 abi
->abi_mru_hits
= l1hdr
->b_mru_hits
;
2013 abi
->abi_mru_ghost_hits
= l1hdr
->b_mru_ghost_hits
;
2014 abi
->abi_mfu_hits
= l1hdr
->b_mfu_hits
;
2015 abi
->abi_mfu_ghost_hits
= l1hdr
->b_mfu_ghost_hits
;
2016 abi
->abi_holds
= refcount_count(&l1hdr
->b_refcnt
);
2020 abi
->abi_l2arc_dattr
= l2hdr
->b_daddr
;
2021 abi
->abi_l2arc_hits
= l2hdr
->b_hits
;
2024 abi
->abi_state_type
= state
? state
->arcs_state
: ARC_STATE_ANON
;
2025 abi
->abi_state_contents
= arc_buf_type(hdr
);
2026 abi
->abi_size
= arc_hdr_size(hdr
);
2030 * Move the supplied buffer to the indicated state. The hash lock
2031 * for the buffer must be held by the caller.
2034 arc_change_state(arc_state_t
*new_state
, arc_buf_hdr_t
*hdr
,
2035 kmutex_t
*hash_lock
)
2037 arc_state_t
*old_state
;
2040 boolean_t update_old
, update_new
;
2041 arc_buf_contents_t buftype
= arc_buf_type(hdr
);
2044 * We almost always have an L1 hdr here, since we call arc_hdr_realloc()
2045 * in arc_read() when bringing a buffer out of the L2ARC. However, the
2046 * L1 hdr doesn't always exist when we change state to arc_anon before
2047 * destroying a header, in which case reallocating to add the L1 hdr is
2050 if (HDR_HAS_L1HDR(hdr
)) {
2051 old_state
= hdr
->b_l1hdr
.b_state
;
2052 refcnt
= refcount_count(&hdr
->b_l1hdr
.b_refcnt
);
2053 bufcnt
= hdr
->b_l1hdr
.b_bufcnt
;
2054 update_old
= (bufcnt
> 0 || hdr
->b_l1hdr
.b_pdata
!= NULL
);
2056 old_state
= arc_l2c_only
;
2059 update_old
= B_FALSE
;
2061 update_new
= update_old
;
2063 ASSERT(MUTEX_HELD(hash_lock
));
2064 ASSERT3P(new_state
, !=, old_state
);
2065 ASSERT(!GHOST_STATE(new_state
) || bufcnt
== 0);
2066 ASSERT(old_state
!= arc_anon
|| bufcnt
<= 1);
2069 * If this buffer is evictable, transfer it from the
2070 * old state list to the new state list.
2073 if (old_state
!= arc_anon
&& old_state
!= arc_l2c_only
) {
2074 ASSERT(HDR_HAS_L1HDR(hdr
));
2075 multilist_remove(&old_state
->arcs_list
[buftype
], hdr
);
2077 if (GHOST_STATE(old_state
)) {
2079 ASSERT3P(hdr
->b_l1hdr
.b_buf
, ==, NULL
);
2080 update_old
= B_TRUE
;
2082 arc_evictable_space_decrement(hdr
, old_state
);
2084 if (new_state
!= arc_anon
&& new_state
!= arc_l2c_only
) {
2086 * An L1 header always exists here, since if we're
2087 * moving to some L1-cached state (i.e. not l2c_only or
2088 * anonymous), we realloc the header to add an L1hdr
2091 ASSERT(HDR_HAS_L1HDR(hdr
));
2092 multilist_insert(&new_state
->arcs_list
[buftype
], hdr
);
2094 if (GHOST_STATE(new_state
)) {
2096 ASSERT3P(hdr
->b_l1hdr
.b_buf
, ==, NULL
);
2097 update_new
= B_TRUE
;
2099 arc_evictable_space_increment(hdr
, new_state
);
2103 ASSERT(!HDR_EMPTY(hdr
));
2104 if (new_state
== arc_anon
&& HDR_IN_HASH_TABLE(hdr
))
2105 buf_hash_remove(hdr
);
2107 /* adjust state sizes (ignore arc_l2c_only) */
2109 if (update_new
&& new_state
!= arc_l2c_only
) {
2110 ASSERT(HDR_HAS_L1HDR(hdr
));
2111 if (GHOST_STATE(new_state
)) {
2115 * When moving a header to a ghost state, we first
2116 * remove all arc buffers. Thus, we'll have a
2117 * bufcnt of zero, and no arc buffer to use for
2118 * the reference. As a result, we use the arc
2119 * header pointer for the reference.
2121 (void) refcount_add_many(&new_state
->arcs_size
,
2122 HDR_GET_LSIZE(hdr
), hdr
);
2123 ASSERT3P(hdr
->b_l1hdr
.b_pdata
, ==, NULL
);
2126 uint32_t buffers
= 0;
2129 * Each individual buffer holds a unique reference,
2130 * thus we must remove each of these references one
2133 for (buf
= hdr
->b_l1hdr
.b_buf
; buf
!= NULL
;
2134 buf
= buf
->b_next
) {
2135 ASSERT3U(bufcnt
, !=, 0);
2139 * When the arc_buf_t is sharing the data
2140 * block with the hdr, the owner of the
2141 * reference belongs to the hdr. Only
2142 * add to the refcount if the arc_buf_t is
2145 if (arc_buf_is_shared(buf
))
2148 (void) refcount_add_many(&new_state
->arcs_size
,
2149 arc_buf_size(buf
), buf
);
2151 ASSERT3U(bufcnt
, ==, buffers
);
2153 if (hdr
->b_l1hdr
.b_pdata
!= NULL
) {
2154 (void) refcount_add_many(&new_state
->arcs_size
,
2155 arc_hdr_size(hdr
), hdr
);
2157 ASSERT(GHOST_STATE(old_state
));
2162 if (update_old
&& old_state
!= arc_l2c_only
) {
2163 ASSERT(HDR_HAS_L1HDR(hdr
));
2164 if (GHOST_STATE(old_state
)) {
2166 ASSERT3P(hdr
->b_l1hdr
.b_pdata
, ==, NULL
);
2169 * When moving a header off of a ghost state,
2170 * the header will not contain any arc buffers.
2171 * We use the arc header pointer for the reference
2172 * which is exactly what we did when we put the
2173 * header on the ghost state.
2176 (void) refcount_remove_many(&old_state
->arcs_size
,
2177 HDR_GET_LSIZE(hdr
), hdr
);
2180 uint32_t buffers
= 0;
2183 * Each individual buffer holds a unique reference,
2184 * thus we must remove each of these references one
2187 for (buf
= hdr
->b_l1hdr
.b_buf
; buf
!= NULL
;
2188 buf
= buf
->b_next
) {
2189 ASSERT3U(bufcnt
, !=, 0);
2193 * When the arc_buf_t is sharing the data
2194 * block with the hdr, the owner of the
2195 * reference belongs to the hdr. Only
2196 * add to the refcount if the arc_buf_t is
2199 if (arc_buf_is_shared(buf
))
2202 (void) refcount_remove_many(
2203 &old_state
->arcs_size
, arc_buf_size(buf
),
2206 ASSERT3U(bufcnt
, ==, buffers
);
2207 ASSERT3P(hdr
->b_l1hdr
.b_pdata
, !=, NULL
);
2208 (void) refcount_remove_many(
2209 &old_state
->arcs_size
, arc_hdr_size(hdr
), hdr
);
2213 if (HDR_HAS_L1HDR(hdr
))
2214 hdr
->b_l1hdr
.b_state
= new_state
;
2217 * L2 headers should never be on the L2 state list since they don't
2218 * have L1 headers allocated.
2220 ASSERT(multilist_is_empty(&arc_l2c_only
->arcs_list
[ARC_BUFC_DATA
]) &&
2221 multilist_is_empty(&arc_l2c_only
->arcs_list
[ARC_BUFC_METADATA
]));
2225 arc_space_consume(uint64_t space
, arc_space_type_t type
)
2227 ASSERT(type
>= 0 && type
< ARC_SPACE_NUMTYPES
);
2232 case ARC_SPACE_DATA
:
2233 ARCSTAT_INCR(arcstat_data_size
, space
);
2235 case ARC_SPACE_META
:
2236 ARCSTAT_INCR(arcstat_metadata_size
, space
);
2238 case ARC_SPACE_BONUS
:
2239 ARCSTAT_INCR(arcstat_bonus_size
, space
);
2241 case ARC_SPACE_DNODE
:
2242 ARCSTAT_INCR(arcstat_dnode_size
, space
);
2244 case ARC_SPACE_DBUF
:
2245 ARCSTAT_INCR(arcstat_dbuf_size
, space
);
2247 case ARC_SPACE_HDRS
:
2248 ARCSTAT_INCR(arcstat_hdr_size
, space
);
2250 case ARC_SPACE_L2HDRS
:
2251 ARCSTAT_INCR(arcstat_l2_hdr_size
, space
);
2255 if (type
!= ARC_SPACE_DATA
)
2256 ARCSTAT_INCR(arcstat_meta_used
, space
);
2258 atomic_add_64(&arc_size
, space
);
2262 arc_space_return(uint64_t space
, arc_space_type_t type
)
2264 ASSERT(type
>= 0 && type
< ARC_SPACE_NUMTYPES
);
2269 case ARC_SPACE_DATA
:
2270 ARCSTAT_INCR(arcstat_data_size
, -space
);
2272 case ARC_SPACE_META
:
2273 ARCSTAT_INCR(arcstat_metadata_size
, -space
);
2275 case ARC_SPACE_BONUS
:
2276 ARCSTAT_INCR(arcstat_bonus_size
, -space
);
2278 case ARC_SPACE_DNODE
:
2279 ARCSTAT_INCR(arcstat_dnode_size
, -space
);
2281 case ARC_SPACE_DBUF
:
2282 ARCSTAT_INCR(arcstat_dbuf_size
, -space
);
2284 case ARC_SPACE_HDRS
:
2285 ARCSTAT_INCR(arcstat_hdr_size
, -space
);
2287 case ARC_SPACE_L2HDRS
:
2288 ARCSTAT_INCR(arcstat_l2_hdr_size
, -space
);
2292 if (type
!= ARC_SPACE_DATA
) {
2293 ASSERT(arc_meta_used
>= space
);
2294 if (arc_meta_max
< arc_meta_used
)
2295 arc_meta_max
= arc_meta_used
;
2296 ARCSTAT_INCR(arcstat_meta_used
, -space
);
2299 ASSERT(arc_size
>= space
);
2300 atomic_add_64(&arc_size
, -space
);
2304 * Given a hdr and a buf, returns whether that buf can share its b_data buffer
2305 * with the hdr's b_pdata.
2308 arc_can_share(arc_buf_hdr_t
*hdr
, arc_buf_t
*buf
)
2310 boolean_t hdr_compressed
, buf_compressed
;
2312 * The criteria for sharing a hdr's data are:
2313 * 1. the hdr's compression matches the buf's compression
2314 * 2. the hdr doesn't need to be byteswapped
2315 * 3. the hdr isn't already being shared
2316 * 4. the buf is either compressed or it is the last buf in the hdr list
2318 * Criterion #4 maintains the invariant that shared uncompressed
2319 * bufs must be the final buf in the hdr's b_buf list. Reading this, you
2320 * might ask, "if a compressed buf is allocated first, won't that be the
2321 * last thing in the list?", but in that case it's impossible to create
2322 * a shared uncompressed buf anyway (because the hdr must be compressed
2323 * to have the compressed buf). You might also think that #3 is
2324 * sufficient to make this guarantee, however it's possible
2325 * (specifically in the rare L2ARC write race mentioned in
2326 * arc_buf_alloc_impl()) there will be an existing uncompressed buf that
2327 * is sharable, but wasn't at the time of its allocation. Rather than
2328 * allow a new shared uncompressed buf to be created and then shuffle
2329 * the list around to make it the last element, this simply disallows
2330 * sharing if the new buf isn't the first to be added.
2332 ASSERT3P(buf
->b_hdr
, ==, hdr
);
2333 hdr_compressed
= HDR_GET_COMPRESS(hdr
) != ZIO_COMPRESS_OFF
;
2334 buf_compressed
= ARC_BUF_COMPRESSED(buf
) != 0;
2335 return (buf_compressed
== hdr_compressed
&&
2336 hdr
->b_l1hdr
.b_byteswap
== DMU_BSWAP_NUMFUNCS
&&
2337 !HDR_SHARED_DATA(hdr
) &&
2338 (ARC_BUF_LAST(buf
) || ARC_BUF_COMPRESSED(buf
)));
2342 * Allocate a buf for this hdr. If you care about the data that's in the hdr,
2343 * or if you want a compressed buffer, pass those flags in. Returns 0 if the
2344 * copy was made successfully, or an error code otherwise.
2347 arc_buf_alloc_impl(arc_buf_hdr_t
*hdr
, void *tag
, boolean_t compressed
,
2348 boolean_t fill
, arc_buf_t
**ret
)
2351 boolean_t can_share
;
2353 ASSERT(HDR_HAS_L1HDR(hdr
));
2354 ASSERT3U(HDR_GET_LSIZE(hdr
), >, 0);
2355 VERIFY(hdr
->b_type
== ARC_BUFC_DATA
||
2356 hdr
->b_type
== ARC_BUFC_METADATA
);
2357 ASSERT3P(ret
, !=, NULL
);
2358 ASSERT3P(*ret
, ==, NULL
);
2360 hdr
->b_l1hdr
.b_mru_hits
= 0;
2361 hdr
->b_l1hdr
.b_mru_ghost_hits
= 0;
2362 hdr
->b_l1hdr
.b_mfu_hits
= 0;
2363 hdr
->b_l1hdr
.b_mfu_ghost_hits
= 0;
2364 hdr
->b_l1hdr
.b_l2_hits
= 0;
2366 buf
= *ret
= kmem_cache_alloc(buf_cache
, KM_PUSHPAGE
);
2369 buf
->b_next
= hdr
->b_l1hdr
.b_buf
;
2372 add_reference(hdr
, tag
);
2375 * We're about to change the hdr's b_flags. We must either
2376 * hold the hash_lock or be undiscoverable.
2378 ASSERT(MUTEX_HELD(HDR_LOCK(hdr
)) || HDR_EMPTY(hdr
));
2381 * Only honor requests for compressed bufs if the hdr is actually
2384 if (compressed
&& HDR_GET_COMPRESS(hdr
) != ZIO_COMPRESS_OFF
)
2385 buf
->b_flags
|= ARC_BUF_FLAG_COMPRESSED
;
2388 * Although the ARC should handle it correctly, levels above the ARC
2389 * should prevent us from having multiple compressed bufs off the same
2390 * hdr. To ensure we notice it if this behavior changes, we assert this
2391 * here the best we can.
2393 IMPLY(ARC_BUF_COMPRESSED(buf
), !HDR_SHARED_DATA(hdr
));
2396 * If the hdr's data can be shared then we share the data buffer and
2397 * set the appropriate bit in the hdr's b_flags to indicate the hdr is
2398 * allocate a new buffer to store the buf's data.
2400 * There is one additional restriction here because we're sharing
2401 * hdr -> buf instead of the usual buf -> hdr: the hdr can't be actively
2402 * involved in an L2ARC write, because if this buf is used by an
2403 * arc_write() then the hdr's data buffer will be released when the
2404 * write completes, even though the L2ARC write might still be using it.
2406 can_share
= arc_can_share(hdr
, buf
) && !HDR_L2_WRITING(hdr
);
2408 /* Set up b_data and sharing */
2410 buf
->b_data
= hdr
->b_l1hdr
.b_pdata
;
2411 buf
->b_flags
|= ARC_BUF_FLAG_SHARED
;
2412 arc_hdr_set_flags(hdr
, ARC_FLAG_SHARED_DATA
);
2415 arc_get_data_buf(hdr
, arc_buf_size(buf
), buf
);
2416 ARCSTAT_INCR(arcstat_overhead_size
, arc_buf_size(buf
));
2418 VERIFY3P(buf
->b_data
, !=, NULL
);
2420 hdr
->b_l1hdr
.b_buf
= buf
;
2421 hdr
->b_l1hdr
.b_bufcnt
+= 1;
2424 * If the user wants the data from the hdr, we need to either copy or
2425 * decompress the data.
2428 return (arc_buf_fill(buf
, ARC_BUF_COMPRESSED(buf
) != 0));
2434 static char *arc_onloan_tag
= "onloan";
2437 * Loan out an anonymous arc buffer. Loaned buffers are not counted as in
2438 * flight data by arc_tempreserve_space() until they are "returned". Loaned
2439 * buffers must be returned to the arc before they can be used by the DMU or
2443 arc_loan_buf(spa_t
*spa
, boolean_t is_metadata
, int size
)
2445 arc_buf_t
*buf
= arc_alloc_buf(spa
, arc_onloan_tag
,
2446 is_metadata
? ARC_BUFC_METADATA
: ARC_BUFC_DATA
, size
);
2448 atomic_add_64(&arc_loaned_bytes
, size
);
2453 arc_loan_compressed_buf(spa_t
*spa
, uint64_t psize
, uint64_t lsize
,
2454 enum zio_compress compression_type
)
2456 arc_buf_t
*buf
= arc_alloc_compressed_buf(spa
, arc_onloan_tag
,
2457 psize
, lsize
, compression_type
);
2459 atomic_add_64(&arc_loaned_bytes
, psize
);
2465 * Return a loaned arc buffer to the arc.
2468 arc_return_buf(arc_buf_t
*buf
, void *tag
)
2470 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
2472 ASSERT3P(buf
->b_data
, !=, NULL
);
2473 ASSERT(HDR_HAS_L1HDR(hdr
));
2474 (void) refcount_add(&hdr
->b_l1hdr
.b_refcnt
, tag
);
2475 (void) refcount_remove(&hdr
->b_l1hdr
.b_refcnt
, arc_onloan_tag
);
2477 atomic_add_64(&arc_loaned_bytes
, -arc_buf_size(buf
));
2480 /* Detach an arc_buf from a dbuf (tag) */
2482 arc_loan_inuse_buf(arc_buf_t
*buf
, void *tag
)
2484 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
2486 ASSERT3P(buf
->b_data
, !=, NULL
);
2487 ASSERT(HDR_HAS_L1HDR(hdr
));
2488 (void) refcount_add(&hdr
->b_l1hdr
.b_refcnt
, arc_onloan_tag
);
2489 (void) refcount_remove(&hdr
->b_l1hdr
.b_refcnt
, tag
);
2491 atomic_add_64(&arc_loaned_bytes
, -arc_buf_size(buf
));
2495 l2arc_free_data_on_write(void *data
, size_t size
, arc_buf_contents_t type
)
2497 l2arc_data_free_t
*df
= kmem_alloc(sizeof (*df
), KM_SLEEP
);
2499 df
->l2df_data
= data
;
2500 df
->l2df_size
= size
;
2501 df
->l2df_type
= type
;
2502 mutex_enter(&l2arc_free_on_write_mtx
);
2503 list_insert_head(l2arc_free_on_write
, df
);
2504 mutex_exit(&l2arc_free_on_write_mtx
);
2508 arc_hdr_free_on_write(arc_buf_hdr_t
*hdr
)
2510 arc_state_t
*state
= hdr
->b_l1hdr
.b_state
;
2511 arc_buf_contents_t type
= arc_buf_type(hdr
);
2512 uint64_t size
= arc_hdr_size(hdr
);
2514 /* protected by hash lock, if in the hash table */
2515 if (multilist_link_active(&hdr
->b_l1hdr
.b_arc_node
)) {
2516 ASSERT(refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
));
2517 ASSERT(state
!= arc_anon
&& state
!= arc_l2c_only
);
2519 (void) refcount_remove_many(&state
->arcs_esize
[type
],
2522 (void) refcount_remove_many(&state
->arcs_size
, size
, hdr
);
2524 l2arc_free_data_on_write(hdr
->b_l1hdr
.b_pdata
, size
, type
);
2528 * Share the arc_buf_t's data with the hdr. Whenever we are sharing the
2529 * data buffer, we transfer the refcount ownership to the hdr and update
2530 * the appropriate kstats.
2533 arc_share_buf(arc_buf_hdr_t
*hdr
, arc_buf_t
*buf
)
2535 ASSERT(arc_can_share(hdr
, buf
));
2536 ASSERT3P(hdr
->b_l1hdr
.b_pdata
, ==, NULL
);
2537 ASSERT(MUTEX_HELD(HDR_LOCK(hdr
)) || HDR_EMPTY(hdr
));
2540 * Start sharing the data buffer. We transfer the
2541 * refcount ownership to the hdr since it always owns
2542 * the refcount whenever an arc_buf_t is shared.
2544 refcount_transfer_ownership(&hdr
->b_l1hdr
.b_state
->arcs_size
, buf
, hdr
);
2545 hdr
->b_l1hdr
.b_pdata
= buf
->b_data
;
2546 arc_hdr_set_flags(hdr
, ARC_FLAG_SHARED_DATA
);
2547 buf
->b_flags
|= ARC_BUF_FLAG_SHARED
;
2550 * Since we've transferred ownership to the hdr we need
2551 * to increment its compressed and uncompressed kstats and
2552 * decrement the overhead size.
2554 ARCSTAT_INCR(arcstat_compressed_size
, arc_hdr_size(hdr
));
2555 ARCSTAT_INCR(arcstat_uncompressed_size
, HDR_GET_LSIZE(hdr
));
2556 ARCSTAT_INCR(arcstat_overhead_size
, -arc_buf_size(buf
));
2560 arc_unshare_buf(arc_buf_hdr_t
*hdr
, arc_buf_t
*buf
)
2562 ASSERT(arc_buf_is_shared(buf
));
2563 ASSERT3P(hdr
->b_l1hdr
.b_pdata
, !=, NULL
);
2564 ASSERT(MUTEX_HELD(HDR_LOCK(hdr
)) || HDR_EMPTY(hdr
));
2567 * We are no longer sharing this buffer so we need
2568 * to transfer its ownership to the rightful owner.
2570 refcount_transfer_ownership(&hdr
->b_l1hdr
.b_state
->arcs_size
, hdr
, buf
);
2571 arc_hdr_clear_flags(hdr
, ARC_FLAG_SHARED_DATA
);
2572 hdr
->b_l1hdr
.b_pdata
= 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_pdata 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_pdata
, !=, NULL
);
2684 arc_hdr_free_pdata(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_pdata(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_pdata
, ==, NULL
);
2724 hdr
->b_l1hdr
.b_pdata
= arc_get_data_buf(hdr
, arc_hdr_size(hdr
), hdr
);
2725 hdr
->b_l1hdr
.b_byteswap
= DMU_BSWAP_NUMFUNCS
;
2726 ASSERT3P(hdr
->b_l1hdr
.b_pdata
, !=, 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_pdata(arc_buf_hdr_t
*hdr
)
2735 ASSERT(HDR_HAS_L1HDR(hdr
));
2736 ASSERT3P(hdr
->b_l1hdr
.b_pdata
, !=, 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_buf(hdr
, hdr
->b_l1hdr
.b_pdata
,
2749 arc_hdr_size(hdr
), hdr
);
2751 hdr
->b_l1hdr
.b_pdata
= 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_pdata(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_pdata
, ==, 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_pdata field
2846 * might try to be accessed, even though it was removed.
2848 VERIFY(!HDR_L2_WRITING(hdr
));
2849 VERIFY3P(hdr
->b_l1hdr
.b_pdata
, ==, 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
);
2938 arc_hdr_l2hdr_destroy(arc_buf_hdr_t
*hdr
)
2940 l2arc_buf_hdr_t
*l2hdr
= &hdr
->b_l2hdr
;
2941 l2arc_dev_t
*dev
= l2hdr
->b_dev
;
2942 uint64_t asize
= arc_hdr_size(hdr
);
2944 ASSERT(MUTEX_HELD(&dev
->l2ad_mtx
));
2945 ASSERT(HDR_HAS_L2HDR(hdr
));
2947 list_remove(&dev
->l2ad_buflist
, hdr
);
2949 ARCSTAT_INCR(arcstat_l2_asize
, -asize
);
2950 ARCSTAT_INCR(arcstat_l2_size
, -HDR_GET_LSIZE(hdr
));
2952 vdev_space_update(dev
->l2ad_vdev
, -asize
, 0, 0);
2954 (void) refcount_remove_many(&dev
->l2ad_alloc
, asize
, hdr
);
2955 arc_hdr_clear_flags(hdr
, ARC_FLAG_HAS_L2HDR
);
2959 arc_hdr_destroy(arc_buf_hdr_t
*hdr
)
2961 if (HDR_HAS_L1HDR(hdr
)) {
2962 ASSERT(hdr
->b_l1hdr
.b_buf
== NULL
||
2963 hdr
->b_l1hdr
.b_bufcnt
> 0);
2964 ASSERT(refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
));
2965 ASSERT3P(hdr
->b_l1hdr
.b_state
, ==, arc_anon
);
2967 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
2968 ASSERT(!HDR_IN_HASH_TABLE(hdr
));
2970 if (!HDR_EMPTY(hdr
))
2971 buf_discard_identity(hdr
);
2973 if (HDR_HAS_L2HDR(hdr
)) {
2974 l2arc_dev_t
*dev
= hdr
->b_l2hdr
.b_dev
;
2975 boolean_t buflist_held
= MUTEX_HELD(&dev
->l2ad_mtx
);
2978 mutex_enter(&dev
->l2ad_mtx
);
2981 * Even though we checked this conditional above, we
2982 * need to check this again now that we have the
2983 * l2ad_mtx. This is because we could be racing with
2984 * another thread calling l2arc_evict() which might have
2985 * destroyed this header's L2 portion as we were waiting
2986 * to acquire the l2ad_mtx. If that happens, we don't
2987 * want to re-destroy the header's L2 portion.
2989 if (HDR_HAS_L2HDR(hdr
))
2990 arc_hdr_l2hdr_destroy(hdr
);
2993 mutex_exit(&dev
->l2ad_mtx
);
2996 if (HDR_HAS_L1HDR(hdr
)) {
2997 arc_cksum_free(hdr
);
2999 while (hdr
->b_l1hdr
.b_buf
!= NULL
)
3000 arc_buf_destroy_impl(hdr
->b_l1hdr
.b_buf
);
3002 if (hdr
->b_l1hdr
.b_pdata
!= NULL
) {
3003 arc_hdr_free_pdata(hdr
);
3007 ASSERT3P(hdr
->b_hash_next
, ==, NULL
);
3008 if (HDR_HAS_L1HDR(hdr
)) {
3009 ASSERT(!multilist_link_active(&hdr
->b_l1hdr
.b_arc_node
));
3010 ASSERT3P(hdr
->b_l1hdr
.b_acb
, ==, NULL
);
3011 kmem_cache_free(hdr_full_cache
, hdr
);
3013 kmem_cache_free(hdr_l2only_cache
, hdr
);
3018 arc_buf_destroy(arc_buf_t
*buf
, void* tag
)
3020 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
3021 kmutex_t
*hash_lock
= HDR_LOCK(hdr
);
3023 if (hdr
->b_l1hdr
.b_state
== arc_anon
) {
3024 ASSERT3U(hdr
->b_l1hdr
.b_bufcnt
, ==, 1);
3025 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
3026 VERIFY0(remove_reference(hdr
, NULL
, tag
));
3027 arc_hdr_destroy(hdr
);
3031 mutex_enter(hash_lock
);
3032 ASSERT3P(hdr
, ==, buf
->b_hdr
);
3033 ASSERT(hdr
->b_l1hdr
.b_bufcnt
> 0);
3034 ASSERT3P(hash_lock
, ==, HDR_LOCK(hdr
));
3035 ASSERT3P(hdr
->b_l1hdr
.b_state
, !=, arc_anon
);
3036 ASSERT3P(buf
->b_data
, !=, NULL
);
3038 (void) remove_reference(hdr
, hash_lock
, tag
);
3039 arc_buf_destroy_impl(buf
);
3040 mutex_exit(hash_lock
);
3044 * Evict the arc_buf_hdr that is provided as a parameter. The resultant
3045 * state of the header is dependent on its state prior to entering this
3046 * function. The following transitions are possible:
3048 * - arc_mru -> arc_mru_ghost
3049 * - arc_mfu -> arc_mfu_ghost
3050 * - arc_mru_ghost -> arc_l2c_only
3051 * - arc_mru_ghost -> deleted
3052 * - arc_mfu_ghost -> arc_l2c_only
3053 * - arc_mfu_ghost -> deleted
3056 arc_evict_hdr(arc_buf_hdr_t
*hdr
, kmutex_t
*hash_lock
)
3058 arc_state_t
*evicted_state
, *state
;
3059 int64_t bytes_evicted
= 0;
3061 ASSERT(MUTEX_HELD(hash_lock
));
3062 ASSERT(HDR_HAS_L1HDR(hdr
));
3064 state
= hdr
->b_l1hdr
.b_state
;
3065 if (GHOST_STATE(state
)) {
3066 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
3067 ASSERT3P(hdr
->b_l1hdr
.b_buf
, ==, NULL
);
3070 * l2arc_write_buffers() relies on a header's L1 portion
3071 * (i.e. its b_pdata field) during its write phase.
3072 * Thus, we cannot push a header onto the arc_l2c_only
3073 * state (removing its L1 piece) until the header is
3074 * done being written to the l2arc.
3076 if (HDR_HAS_L2HDR(hdr
) && HDR_L2_WRITING(hdr
)) {
3077 ARCSTAT_BUMP(arcstat_evict_l2_skip
);
3078 return (bytes_evicted
);
3081 ARCSTAT_BUMP(arcstat_deleted
);
3082 bytes_evicted
+= HDR_GET_LSIZE(hdr
);
3084 DTRACE_PROBE1(arc__delete
, arc_buf_hdr_t
*, hdr
);
3086 if (HDR_HAS_L2HDR(hdr
)) {
3087 ASSERT(hdr
->b_l1hdr
.b_pdata
== NULL
);
3089 * This buffer is cached on the 2nd Level ARC;
3090 * don't destroy the header.
3092 arc_change_state(arc_l2c_only
, hdr
, hash_lock
);
3094 * dropping from L1+L2 cached to L2-only,
3095 * realloc to remove the L1 header.
3097 hdr
= arc_hdr_realloc(hdr
, hdr_full_cache
,
3100 arc_change_state(arc_anon
, hdr
, hash_lock
);
3101 arc_hdr_destroy(hdr
);
3103 return (bytes_evicted
);
3106 ASSERT(state
== arc_mru
|| state
== arc_mfu
);
3107 evicted_state
= (state
== arc_mru
) ? arc_mru_ghost
: arc_mfu_ghost
;
3109 /* prefetch buffers have a minimum lifespan */
3110 if (HDR_IO_IN_PROGRESS(hdr
) ||
3111 ((hdr
->b_flags
& (ARC_FLAG_PREFETCH
| ARC_FLAG_INDIRECT
)) &&
3112 ddi_get_lbolt() - hdr
->b_l1hdr
.b_arc_access
<
3113 arc_min_prefetch_lifespan
)) {
3114 ARCSTAT_BUMP(arcstat_evict_skip
);
3115 return (bytes_evicted
);
3118 ASSERT0(refcount_count(&hdr
->b_l1hdr
.b_refcnt
));
3119 while (hdr
->b_l1hdr
.b_buf
) {
3120 arc_buf_t
*buf
= hdr
->b_l1hdr
.b_buf
;
3121 if (!mutex_tryenter(&buf
->b_evict_lock
)) {
3122 ARCSTAT_BUMP(arcstat_mutex_miss
);
3125 if (buf
->b_data
!= NULL
)
3126 bytes_evicted
+= HDR_GET_LSIZE(hdr
);
3127 mutex_exit(&buf
->b_evict_lock
);
3128 arc_buf_destroy_impl(buf
);
3131 if (HDR_HAS_L2HDR(hdr
)) {
3132 ARCSTAT_INCR(arcstat_evict_l2_cached
, HDR_GET_LSIZE(hdr
));
3134 if (l2arc_write_eligible(hdr
->b_spa
, hdr
)) {
3135 ARCSTAT_INCR(arcstat_evict_l2_eligible
,
3136 HDR_GET_LSIZE(hdr
));
3138 ARCSTAT_INCR(arcstat_evict_l2_ineligible
,
3139 HDR_GET_LSIZE(hdr
));
3143 if (hdr
->b_l1hdr
.b_bufcnt
== 0) {
3144 arc_cksum_free(hdr
);
3146 bytes_evicted
+= arc_hdr_size(hdr
);
3149 * If this hdr is being evicted and has a compressed
3150 * buffer then we discard it here before we change states.
3151 * This ensures that the accounting is updated correctly
3152 * in arc_free_data_buf().
3154 arc_hdr_free_pdata(hdr
);
3156 arc_change_state(evicted_state
, hdr
, hash_lock
);
3157 ASSERT(HDR_IN_HASH_TABLE(hdr
));
3158 arc_hdr_set_flags(hdr
, ARC_FLAG_IN_HASH_TABLE
);
3159 DTRACE_PROBE1(arc__evict
, arc_buf_hdr_t
*, hdr
);
3162 return (bytes_evicted
);
3166 arc_evict_state_impl(multilist_t
*ml
, int idx
, arc_buf_hdr_t
*marker
,
3167 uint64_t spa
, int64_t bytes
)
3169 multilist_sublist_t
*mls
;
3170 uint64_t bytes_evicted
= 0;
3172 kmutex_t
*hash_lock
;
3173 int evict_count
= 0;
3175 ASSERT3P(marker
, !=, NULL
);
3176 IMPLY(bytes
< 0, bytes
== ARC_EVICT_ALL
);
3178 mls
= multilist_sublist_lock(ml
, idx
);
3180 for (hdr
= multilist_sublist_prev(mls
, marker
); hdr
!= NULL
;
3181 hdr
= multilist_sublist_prev(mls
, marker
)) {
3182 if ((bytes
!= ARC_EVICT_ALL
&& bytes_evicted
>= bytes
) ||
3183 (evict_count
>= zfs_arc_evict_batch_limit
))
3187 * To keep our iteration location, move the marker
3188 * forward. Since we're not holding hdr's hash lock, we
3189 * must be very careful and not remove 'hdr' from the
3190 * sublist. Otherwise, other consumers might mistake the
3191 * 'hdr' as not being on a sublist when they call the
3192 * multilist_link_active() function (they all rely on
3193 * the hash lock protecting concurrent insertions and
3194 * removals). multilist_sublist_move_forward() was
3195 * specifically implemented to ensure this is the case
3196 * (only 'marker' will be removed and re-inserted).
3198 multilist_sublist_move_forward(mls
, marker
);
3201 * The only case where the b_spa field should ever be
3202 * zero, is the marker headers inserted by
3203 * arc_evict_state(). It's possible for multiple threads
3204 * to be calling arc_evict_state() concurrently (e.g.
3205 * dsl_pool_close() and zio_inject_fault()), so we must
3206 * skip any markers we see from these other threads.
3208 if (hdr
->b_spa
== 0)
3211 /* we're only interested in evicting buffers of a certain spa */
3212 if (spa
!= 0 && hdr
->b_spa
!= spa
) {
3213 ARCSTAT_BUMP(arcstat_evict_skip
);
3217 hash_lock
= HDR_LOCK(hdr
);
3220 * We aren't calling this function from any code path
3221 * that would already be holding a hash lock, so we're
3222 * asserting on this assumption to be defensive in case
3223 * this ever changes. Without this check, it would be
3224 * possible to incorrectly increment arcstat_mutex_miss
3225 * below (e.g. if the code changed such that we called
3226 * this function with a hash lock held).
3228 ASSERT(!MUTEX_HELD(hash_lock
));
3230 if (mutex_tryenter(hash_lock
)) {
3231 uint64_t evicted
= arc_evict_hdr(hdr
, hash_lock
);
3232 mutex_exit(hash_lock
);
3234 bytes_evicted
+= evicted
;
3237 * If evicted is zero, arc_evict_hdr() must have
3238 * decided to skip this header, don't increment
3239 * evict_count in this case.
3245 * If arc_size isn't overflowing, signal any
3246 * threads that might happen to be waiting.
3248 * For each header evicted, we wake up a single
3249 * thread. If we used cv_broadcast, we could
3250 * wake up "too many" threads causing arc_size
3251 * to significantly overflow arc_c; since
3252 * arc_get_data_buf() doesn't check for overflow
3253 * when it's woken up (it doesn't because it's
3254 * possible for the ARC to be overflowing while
3255 * full of un-evictable buffers, and the
3256 * function should proceed in this case).
3258 * If threads are left sleeping, due to not
3259 * using cv_broadcast, they will be woken up
3260 * just before arc_reclaim_thread() sleeps.
3262 mutex_enter(&arc_reclaim_lock
);
3263 if (!arc_is_overflowing())
3264 cv_signal(&arc_reclaim_waiters_cv
);
3265 mutex_exit(&arc_reclaim_lock
);
3267 ARCSTAT_BUMP(arcstat_mutex_miss
);
3271 multilist_sublist_unlock(mls
);
3273 return (bytes_evicted
);
3277 * Evict buffers from the given arc state, until we've removed the
3278 * specified number of bytes. Move the removed buffers to the
3279 * appropriate evict state.
3281 * This function makes a "best effort". It skips over any buffers
3282 * it can't get a hash_lock on, and so, may not catch all candidates.
3283 * It may also return without evicting as much space as requested.
3285 * If bytes is specified using the special value ARC_EVICT_ALL, this
3286 * will evict all available (i.e. unlocked and evictable) buffers from
3287 * the given arc state; which is used by arc_flush().
3290 arc_evict_state(arc_state_t
*state
, uint64_t spa
, int64_t bytes
,
3291 arc_buf_contents_t type
)
3293 uint64_t total_evicted
= 0;
3294 multilist_t
*ml
= &state
->arcs_list
[type
];
3296 arc_buf_hdr_t
**markers
;
3299 IMPLY(bytes
< 0, bytes
== ARC_EVICT_ALL
);
3301 num_sublists
= multilist_get_num_sublists(ml
);
3304 * If we've tried to evict from each sublist, made some
3305 * progress, but still have not hit the target number of bytes
3306 * to evict, we want to keep trying. The markers allow us to
3307 * pick up where we left off for each individual sublist, rather
3308 * than starting from the tail each time.
3310 markers
= kmem_zalloc(sizeof (*markers
) * num_sublists
, KM_SLEEP
);
3311 for (i
= 0; i
< num_sublists
; i
++) {
3312 multilist_sublist_t
*mls
;
3314 markers
[i
] = kmem_cache_alloc(hdr_full_cache
, KM_SLEEP
);
3317 * A b_spa of 0 is used to indicate that this header is
3318 * a marker. This fact is used in arc_adjust_type() and
3319 * arc_evict_state_impl().
3321 markers
[i
]->b_spa
= 0;
3323 mls
= multilist_sublist_lock(ml
, i
);
3324 multilist_sublist_insert_tail(mls
, markers
[i
]);
3325 multilist_sublist_unlock(mls
);
3329 * While we haven't hit our target number of bytes to evict, or
3330 * we're evicting all available buffers.
3332 while (total_evicted
< bytes
|| bytes
== ARC_EVICT_ALL
) {
3333 int sublist_idx
= multilist_get_random_index(ml
);
3334 uint64_t scan_evicted
= 0;
3337 * Try to reduce pinned dnodes with a floor of arc_dnode_limit.
3338 * Request that 10% of the LRUs be scanned by the superblock
3341 if (type
== ARC_BUFC_DATA
&& arc_dnode_size
> arc_dnode_limit
)
3342 arc_prune_async((arc_dnode_size
- arc_dnode_limit
) /
3343 sizeof (dnode_t
) / zfs_arc_dnode_reduce_percent
);
3346 * Start eviction using a randomly selected sublist,
3347 * this is to try and evenly balance eviction across all
3348 * sublists. Always starting at the same sublist
3349 * (e.g. index 0) would cause evictions to favor certain
3350 * sublists over others.
3352 for (i
= 0; i
< num_sublists
; i
++) {
3353 uint64_t bytes_remaining
;
3354 uint64_t bytes_evicted
;
3356 if (bytes
== ARC_EVICT_ALL
)
3357 bytes_remaining
= ARC_EVICT_ALL
;
3358 else if (total_evicted
< bytes
)
3359 bytes_remaining
= bytes
- total_evicted
;
3363 bytes_evicted
= arc_evict_state_impl(ml
, sublist_idx
,
3364 markers
[sublist_idx
], spa
, bytes_remaining
);
3366 scan_evicted
+= bytes_evicted
;
3367 total_evicted
+= bytes_evicted
;
3369 /* we've reached the end, wrap to the beginning */
3370 if (++sublist_idx
>= num_sublists
)
3375 * If we didn't evict anything during this scan, we have
3376 * no reason to believe we'll evict more during another
3377 * scan, so break the loop.
3379 if (scan_evicted
== 0) {
3380 /* This isn't possible, let's make that obvious */
3381 ASSERT3S(bytes
, !=, 0);
3384 * When bytes is ARC_EVICT_ALL, the only way to
3385 * break the loop is when scan_evicted is zero.
3386 * In that case, we actually have evicted enough,
3387 * so we don't want to increment the kstat.
3389 if (bytes
!= ARC_EVICT_ALL
) {
3390 ASSERT3S(total_evicted
, <, bytes
);
3391 ARCSTAT_BUMP(arcstat_evict_not_enough
);
3398 for (i
= 0; i
< num_sublists
; i
++) {
3399 multilist_sublist_t
*mls
= multilist_sublist_lock(ml
, i
);
3400 multilist_sublist_remove(mls
, markers
[i
]);
3401 multilist_sublist_unlock(mls
);
3403 kmem_cache_free(hdr_full_cache
, markers
[i
]);
3405 kmem_free(markers
, sizeof (*markers
) * num_sublists
);
3407 return (total_evicted
);
3411 * Flush all "evictable" data of the given type from the arc state
3412 * specified. This will not evict any "active" buffers (i.e. referenced).
3414 * When 'retry' is set to B_FALSE, the function will make a single pass
3415 * over the state and evict any buffers that it can. Since it doesn't
3416 * continually retry the eviction, it might end up leaving some buffers
3417 * in the ARC due to lock misses.
3419 * When 'retry' is set to B_TRUE, the function will continually retry the
3420 * eviction until *all* evictable buffers have been removed from the
3421 * state. As a result, if concurrent insertions into the state are
3422 * allowed (e.g. if the ARC isn't shutting down), this function might
3423 * wind up in an infinite loop, continually trying to evict buffers.
3426 arc_flush_state(arc_state_t
*state
, uint64_t spa
, arc_buf_contents_t type
,
3429 uint64_t evicted
= 0;
3431 while (refcount_count(&state
->arcs_esize
[type
]) != 0) {
3432 evicted
+= arc_evict_state(state
, spa
, ARC_EVICT_ALL
, type
);
3442 * Helper function for arc_prune_async() it is responsible for safely
3443 * handling the execution of a registered arc_prune_func_t.
3446 arc_prune_task(void *ptr
)
3448 arc_prune_t
*ap
= (arc_prune_t
*)ptr
;
3449 arc_prune_func_t
*func
= ap
->p_pfunc
;
3452 func(ap
->p_adjust
, ap
->p_private
);
3454 refcount_remove(&ap
->p_refcnt
, func
);
3458 * Notify registered consumers they must drop holds on a portion of the ARC
3459 * buffered they reference. This provides a mechanism to ensure the ARC can
3460 * honor the arc_meta_limit and reclaim otherwise pinned ARC buffers. This
3461 * is analogous to dnlc_reduce_cache() but more generic.
3463 * This operation is performed asynchronously so it may be safely called
3464 * in the context of the arc_reclaim_thread(). A reference is taken here
3465 * for each registered arc_prune_t and the arc_prune_task() is responsible
3466 * for releasing it once the registered arc_prune_func_t has completed.
3469 arc_prune_async(int64_t adjust
)
3473 mutex_enter(&arc_prune_mtx
);
3474 for (ap
= list_head(&arc_prune_list
); ap
!= NULL
;
3475 ap
= list_next(&arc_prune_list
, ap
)) {
3477 if (refcount_count(&ap
->p_refcnt
) >= 2)
3480 refcount_add(&ap
->p_refcnt
, ap
->p_pfunc
);
3481 ap
->p_adjust
= adjust
;
3482 if (taskq_dispatch(arc_prune_taskq
, arc_prune_task
,
3483 ap
, TQ_SLEEP
) == TASKQID_INVALID
) {
3484 refcount_remove(&ap
->p_refcnt
, ap
->p_pfunc
);
3487 ARCSTAT_BUMP(arcstat_prune
);
3489 mutex_exit(&arc_prune_mtx
);
3493 * Evict the specified number of bytes from the state specified,
3494 * restricting eviction to the spa and type given. This function
3495 * prevents us from trying to evict more from a state's list than
3496 * is "evictable", and to skip evicting altogether when passed a
3497 * negative value for "bytes". In contrast, arc_evict_state() will
3498 * evict everything it can, when passed a negative value for "bytes".
3501 arc_adjust_impl(arc_state_t
*state
, uint64_t spa
, int64_t bytes
,
3502 arc_buf_contents_t type
)
3506 if (bytes
> 0 && refcount_count(&state
->arcs_esize
[type
]) > 0) {
3507 delta
= MIN(refcount_count(&state
->arcs_esize
[type
]), bytes
);
3508 return (arc_evict_state(state
, spa
, delta
, type
));
3515 * The goal of this function is to evict enough meta data buffers from the
3516 * ARC in order to enforce the arc_meta_limit. Achieving this is slightly
3517 * more complicated than it appears because it is common for data buffers
3518 * to have holds on meta data buffers. In addition, dnode meta data buffers
3519 * will be held by the dnodes in the block preventing them from being freed.
3520 * This means we can't simply traverse the ARC and expect to always find
3521 * enough unheld meta data buffer to release.
3523 * Therefore, this function has been updated to make alternating passes
3524 * over the ARC releasing data buffers and then newly unheld meta data
3525 * buffers. This ensures forward progress is maintained and arc_meta_used
3526 * will decrease. Normally this is sufficient, but if required the ARC
3527 * will call the registered prune callbacks causing dentry and inodes to
3528 * be dropped from the VFS cache. This will make dnode meta data buffers
3529 * available for reclaim.
3532 arc_adjust_meta_balanced(void)
3534 int64_t delta
, prune
= 0, adjustmnt
;
3535 uint64_t total_evicted
= 0;
3536 arc_buf_contents_t type
= ARC_BUFC_DATA
;
3537 int restarts
= MAX(zfs_arc_meta_adjust_restarts
, 0);
3541 * This slightly differs than the way we evict from the mru in
3542 * arc_adjust because we don't have a "target" value (i.e. no
3543 * "meta" arc_p). As a result, I think we can completely
3544 * cannibalize the metadata in the MRU before we evict the
3545 * metadata from the MFU. I think we probably need to implement a
3546 * "metadata arc_p" value to do this properly.
3548 adjustmnt
= arc_meta_used
- arc_meta_limit
;
3550 if (adjustmnt
> 0 && refcount_count(&arc_mru
->arcs_esize
[type
]) > 0) {
3551 delta
= MIN(refcount_count(&arc_mru
->arcs_esize
[type
]),
3553 total_evicted
+= arc_adjust_impl(arc_mru
, 0, delta
, type
);
3558 * We can't afford to recalculate adjustmnt here. If we do,
3559 * new metadata buffers can sneak into the MRU or ANON lists,
3560 * thus penalize the MFU metadata. Although the fudge factor is
3561 * small, it has been empirically shown to be significant for
3562 * certain workloads (e.g. creating many empty directories). As
3563 * such, we use the original calculation for adjustmnt, and
3564 * simply decrement the amount of data evicted from the MRU.
3567 if (adjustmnt
> 0 && refcount_count(&arc_mfu
->arcs_esize
[type
]) > 0) {
3568 delta
= MIN(refcount_count(&arc_mfu
->arcs_esize
[type
]),
3570 total_evicted
+= arc_adjust_impl(arc_mfu
, 0, delta
, type
);
3573 adjustmnt
= arc_meta_used
- arc_meta_limit
;
3575 if (adjustmnt
> 0 &&
3576 refcount_count(&arc_mru_ghost
->arcs_esize
[type
]) > 0) {
3577 delta
= MIN(adjustmnt
,
3578 refcount_count(&arc_mru_ghost
->arcs_esize
[type
]));
3579 total_evicted
+= arc_adjust_impl(arc_mru_ghost
, 0, delta
, type
);
3583 if (adjustmnt
> 0 &&
3584 refcount_count(&arc_mfu_ghost
->arcs_esize
[type
]) > 0) {
3585 delta
= MIN(adjustmnt
,
3586 refcount_count(&arc_mfu_ghost
->arcs_esize
[type
]));
3587 total_evicted
+= arc_adjust_impl(arc_mfu_ghost
, 0, delta
, type
);
3591 * If after attempting to make the requested adjustment to the ARC
3592 * the meta limit is still being exceeded then request that the
3593 * higher layers drop some cached objects which have holds on ARC
3594 * meta buffers. Requests to the upper layers will be made with
3595 * increasingly large scan sizes until the ARC is below the limit.
3597 if (arc_meta_used
> arc_meta_limit
) {
3598 if (type
== ARC_BUFC_DATA
) {
3599 type
= ARC_BUFC_METADATA
;
3601 type
= ARC_BUFC_DATA
;
3603 if (zfs_arc_meta_prune
) {
3604 prune
+= zfs_arc_meta_prune
;
3605 arc_prune_async(prune
);
3614 return (total_evicted
);
3618 * Evict metadata buffers from the cache, such that arc_meta_used is
3619 * capped by the arc_meta_limit tunable.
3622 arc_adjust_meta_only(void)
3624 uint64_t total_evicted
= 0;
3628 * If we're over the meta limit, we want to evict enough
3629 * metadata to get back under the meta limit. We don't want to
3630 * evict so much that we drop the MRU below arc_p, though. If
3631 * we're over the meta limit more than we're over arc_p, we
3632 * evict some from the MRU here, and some from the MFU below.
3634 target
= MIN((int64_t)(arc_meta_used
- arc_meta_limit
),
3635 (int64_t)(refcount_count(&arc_anon
->arcs_size
) +
3636 refcount_count(&arc_mru
->arcs_size
) - arc_p
));
3638 total_evicted
+= arc_adjust_impl(arc_mru
, 0, target
, ARC_BUFC_METADATA
);
3641 * Similar to the above, we want to evict enough bytes to get us
3642 * below the meta limit, but not so much as to drop us below the
3643 * space allotted to the MFU (which is defined as arc_c - arc_p).
3645 target
= MIN((int64_t)(arc_meta_used
- arc_meta_limit
),
3646 (int64_t)(refcount_count(&arc_mfu
->arcs_size
) - (arc_c
- arc_p
)));
3648 total_evicted
+= arc_adjust_impl(arc_mfu
, 0, target
, ARC_BUFC_METADATA
);
3650 return (total_evicted
);
3654 arc_adjust_meta(void)
3656 if (zfs_arc_meta_strategy
== ARC_STRATEGY_META_ONLY
)
3657 return (arc_adjust_meta_only());
3659 return (arc_adjust_meta_balanced());
3663 * Return the type of the oldest buffer in the given arc state
3665 * This function will select a random sublist of type ARC_BUFC_DATA and
3666 * a random sublist of type ARC_BUFC_METADATA. The tail of each sublist
3667 * is compared, and the type which contains the "older" buffer will be
3670 static arc_buf_contents_t
3671 arc_adjust_type(arc_state_t
*state
)
3673 multilist_t
*data_ml
= &state
->arcs_list
[ARC_BUFC_DATA
];
3674 multilist_t
*meta_ml
= &state
->arcs_list
[ARC_BUFC_METADATA
];
3675 int data_idx
= multilist_get_random_index(data_ml
);
3676 int meta_idx
= multilist_get_random_index(meta_ml
);
3677 multilist_sublist_t
*data_mls
;
3678 multilist_sublist_t
*meta_mls
;
3679 arc_buf_contents_t type
;
3680 arc_buf_hdr_t
*data_hdr
;
3681 arc_buf_hdr_t
*meta_hdr
;
3684 * We keep the sublist lock until we're finished, to prevent
3685 * the headers from being destroyed via arc_evict_state().
3687 data_mls
= multilist_sublist_lock(data_ml
, data_idx
);
3688 meta_mls
= multilist_sublist_lock(meta_ml
, meta_idx
);
3691 * These two loops are to ensure we skip any markers that
3692 * might be at the tail of the lists due to arc_evict_state().
3695 for (data_hdr
= multilist_sublist_tail(data_mls
); data_hdr
!= NULL
;
3696 data_hdr
= multilist_sublist_prev(data_mls
, data_hdr
)) {
3697 if (data_hdr
->b_spa
!= 0)
3701 for (meta_hdr
= multilist_sublist_tail(meta_mls
); meta_hdr
!= NULL
;
3702 meta_hdr
= multilist_sublist_prev(meta_mls
, meta_hdr
)) {
3703 if (meta_hdr
->b_spa
!= 0)
3707 if (data_hdr
== NULL
&& meta_hdr
== NULL
) {
3708 type
= ARC_BUFC_DATA
;
3709 } else if (data_hdr
== NULL
) {
3710 ASSERT3P(meta_hdr
, !=, NULL
);
3711 type
= ARC_BUFC_METADATA
;
3712 } else if (meta_hdr
== NULL
) {
3713 ASSERT3P(data_hdr
, !=, NULL
);
3714 type
= ARC_BUFC_DATA
;
3716 ASSERT3P(data_hdr
, !=, NULL
);
3717 ASSERT3P(meta_hdr
, !=, NULL
);
3719 /* The headers can't be on the sublist without an L1 header */
3720 ASSERT(HDR_HAS_L1HDR(data_hdr
));
3721 ASSERT(HDR_HAS_L1HDR(meta_hdr
));
3723 if (data_hdr
->b_l1hdr
.b_arc_access
<
3724 meta_hdr
->b_l1hdr
.b_arc_access
) {
3725 type
= ARC_BUFC_DATA
;
3727 type
= ARC_BUFC_METADATA
;
3731 multilist_sublist_unlock(meta_mls
);
3732 multilist_sublist_unlock(data_mls
);
3738 * Evict buffers from the cache, such that arc_size is capped by arc_c.
3743 uint64_t total_evicted
= 0;
3748 * If we're over arc_meta_limit, we want to correct that before
3749 * potentially evicting data buffers below.
3751 total_evicted
+= arc_adjust_meta();
3756 * If we're over the target cache size, we want to evict enough
3757 * from the list to get back to our target size. We don't want
3758 * to evict too much from the MRU, such that it drops below
3759 * arc_p. So, if we're over our target cache size more than
3760 * the MRU is over arc_p, we'll evict enough to get back to
3761 * arc_p here, and then evict more from the MFU below.
3763 target
= MIN((int64_t)(arc_size
- arc_c
),
3764 (int64_t)(refcount_count(&arc_anon
->arcs_size
) +
3765 refcount_count(&arc_mru
->arcs_size
) + arc_meta_used
- arc_p
));
3768 * If we're below arc_meta_min, always prefer to evict data.
3769 * Otherwise, try to satisfy the requested number of bytes to
3770 * evict from the type which contains older buffers; in an
3771 * effort to keep newer buffers in the cache regardless of their
3772 * type. If we cannot satisfy the number of bytes from this
3773 * type, spill over into the next type.
3775 if (arc_adjust_type(arc_mru
) == ARC_BUFC_METADATA
&&
3776 arc_meta_used
> arc_meta_min
) {
3777 bytes
= arc_adjust_impl(arc_mru
, 0, target
, ARC_BUFC_METADATA
);
3778 total_evicted
+= bytes
;
3781 * If we couldn't evict our target number of bytes from
3782 * metadata, we try to get the rest from data.
3787 arc_adjust_impl(arc_mru
, 0, target
, ARC_BUFC_DATA
);
3789 bytes
= arc_adjust_impl(arc_mru
, 0, target
, ARC_BUFC_DATA
);
3790 total_evicted
+= bytes
;
3793 * If we couldn't evict our target number of bytes from
3794 * data, we try to get the rest from metadata.
3799 arc_adjust_impl(arc_mru
, 0, target
, ARC_BUFC_METADATA
);
3805 * Now that we've tried to evict enough from the MRU to get its
3806 * size back to arc_p, if we're still above the target cache
3807 * size, we evict the rest from the MFU.
3809 target
= arc_size
- arc_c
;
3811 if (arc_adjust_type(arc_mfu
) == ARC_BUFC_METADATA
&&
3812 arc_meta_used
> arc_meta_min
) {
3813 bytes
= arc_adjust_impl(arc_mfu
, 0, target
, ARC_BUFC_METADATA
);
3814 total_evicted
+= bytes
;
3817 * If we couldn't evict our target number of bytes from
3818 * metadata, we try to get the rest from data.
3823 arc_adjust_impl(arc_mfu
, 0, target
, ARC_BUFC_DATA
);
3825 bytes
= arc_adjust_impl(arc_mfu
, 0, target
, ARC_BUFC_DATA
);
3826 total_evicted
+= bytes
;
3829 * If we couldn't evict our target number of bytes from
3830 * data, we try to get the rest from data.
3835 arc_adjust_impl(arc_mfu
, 0, target
, ARC_BUFC_METADATA
);
3839 * Adjust ghost lists
3841 * In addition to the above, the ARC also defines target values
3842 * for the ghost lists. The sum of the mru list and mru ghost
3843 * list should never exceed the target size of the cache, and
3844 * the sum of the mru list, mfu list, mru ghost list, and mfu
3845 * ghost list should never exceed twice the target size of the
3846 * cache. The following logic enforces these limits on the ghost
3847 * caches, and evicts from them as needed.
3849 target
= refcount_count(&arc_mru
->arcs_size
) +
3850 refcount_count(&arc_mru_ghost
->arcs_size
) - arc_c
;
3852 bytes
= arc_adjust_impl(arc_mru_ghost
, 0, target
, ARC_BUFC_DATA
);
3853 total_evicted
+= bytes
;
3858 arc_adjust_impl(arc_mru_ghost
, 0, target
, ARC_BUFC_METADATA
);
3861 * We assume the sum of the mru list and mfu list is less than
3862 * or equal to arc_c (we enforced this above), which means we
3863 * can use the simpler of the two equations below:
3865 * mru + mfu + mru ghost + mfu ghost <= 2 * arc_c
3866 * mru ghost + mfu ghost <= arc_c
3868 target
= refcount_count(&arc_mru_ghost
->arcs_size
) +
3869 refcount_count(&arc_mfu_ghost
->arcs_size
) - arc_c
;
3871 bytes
= arc_adjust_impl(arc_mfu_ghost
, 0, target
, ARC_BUFC_DATA
);
3872 total_evicted
+= bytes
;
3877 arc_adjust_impl(arc_mfu_ghost
, 0, target
, ARC_BUFC_METADATA
);
3879 return (total_evicted
);
3883 arc_flush(spa_t
*spa
, boolean_t retry
)
3888 * If retry is B_TRUE, a spa must not be specified since we have
3889 * no good way to determine if all of a spa's buffers have been
3890 * evicted from an arc state.
3892 ASSERT(!retry
|| spa
== 0);
3895 guid
= spa_load_guid(spa
);
3897 (void) arc_flush_state(arc_mru
, guid
, ARC_BUFC_DATA
, retry
);
3898 (void) arc_flush_state(arc_mru
, guid
, ARC_BUFC_METADATA
, retry
);
3900 (void) arc_flush_state(arc_mfu
, guid
, ARC_BUFC_DATA
, retry
);
3901 (void) arc_flush_state(arc_mfu
, guid
, ARC_BUFC_METADATA
, retry
);
3903 (void) arc_flush_state(arc_mru_ghost
, guid
, ARC_BUFC_DATA
, retry
);
3904 (void) arc_flush_state(arc_mru_ghost
, guid
, ARC_BUFC_METADATA
, retry
);
3906 (void) arc_flush_state(arc_mfu_ghost
, guid
, ARC_BUFC_DATA
, retry
);
3907 (void) arc_flush_state(arc_mfu_ghost
, guid
, ARC_BUFC_METADATA
, retry
);
3911 arc_shrink(int64_t to_free
)
3915 if (c
> to_free
&& c
- to_free
> arc_c_min
) {
3916 arc_c
= c
- to_free
;
3917 atomic_add_64(&arc_p
, -(arc_p
>> arc_shrink_shift
));
3918 if (arc_c
> arc_size
)
3919 arc_c
= MAX(arc_size
, arc_c_min
);
3921 arc_p
= (arc_c
>> 1);
3922 ASSERT(arc_c
>= arc_c_min
);
3923 ASSERT((int64_t)arc_p
>= 0);
3928 if (arc_size
> arc_c
)
3929 (void) arc_adjust();
3933 * Return maximum amount of memory that we could possibly use. Reduced
3934 * to half of all memory in user space which is primarily used for testing.
3937 arc_all_memory(void)
3940 return (MIN(ptob(physmem
),
3941 vmem_size(heap_arena
, VMEM_FREE
| VMEM_ALLOC
)));
3943 return (ptob(physmem
) / 2);
3947 typedef enum free_memory_reason_t
{
3952 FMR_PAGES_PP_MAXIMUM
,
3955 } free_memory_reason_t
;
3957 int64_t last_free_memory
;
3958 free_memory_reason_t last_free_reason
;
3962 * Additional reserve of pages for pp_reserve.
3964 int64_t arc_pages_pp_reserve
= 64;
3967 * Additional reserve of pages for swapfs.
3969 int64_t arc_swapfs_reserve
= 64;
3970 #endif /* _KERNEL */
3973 * Return the amount of memory that can be consumed before reclaim will be
3974 * needed. Positive if there is sufficient free memory, negative indicates
3975 * the amount of memory that needs to be freed up.
3978 arc_available_memory(void)
3980 int64_t lowest
= INT64_MAX
;
3981 free_memory_reason_t r
= FMR_UNKNOWN
;
3983 uint64_t available_memory
= ptob(freemem
);
3986 pgcnt_t needfree
= btop(arc_need_free
);
3987 pgcnt_t lotsfree
= btop(arc_sys_free
);
3988 pgcnt_t desfree
= 0;
3993 MIN(available_memory
, vmem_size(heap_arena
, VMEM_FREE
));
3997 n
= PAGESIZE
* (-needfree
);
4005 * check that we're out of range of the pageout scanner. It starts to
4006 * schedule paging if freemem is less than lotsfree and needfree.
4007 * lotsfree is the high-water mark for pageout, and needfree is the
4008 * number of needed free pages. We add extra pages here to make sure
4009 * the scanner doesn't start up while we're freeing memory.
4011 n
= PAGESIZE
* (btop(available_memory
) - lotsfree
- needfree
- desfree
);
4019 * check to make sure that swapfs has enough space so that anon
4020 * reservations can still succeed. anon_resvmem() checks that the
4021 * availrmem is greater than swapfs_minfree, and the number of reserved
4022 * swap pages. We also add a bit of extra here just to prevent
4023 * circumstances from getting really dire.
4025 n
= PAGESIZE
* (availrmem
- swapfs_minfree
- swapfs_reserve
-
4026 desfree
- arc_swapfs_reserve
);
4029 r
= FMR_SWAPFS_MINFREE
;
4034 * Check that we have enough availrmem that memory locking (e.g., via
4035 * mlock(3C) or memcntl(2)) can still succeed. (pages_pp_maximum
4036 * stores the number of pages that cannot be locked; when availrmem
4037 * drops below pages_pp_maximum, page locking mechanisms such as
4038 * page_pp_lock() will fail.)
4040 n
= PAGESIZE
* (availrmem
- pages_pp_maximum
-
4041 arc_pages_pp_reserve
);
4044 r
= FMR_PAGES_PP_MAXIMUM
;
4050 * If we're on an i386 platform, it's possible that we'll exhaust the
4051 * kernel heap space before we ever run out of available physical
4052 * memory. Most checks of the size of the heap_area compare against
4053 * tune.t_minarmem, which is the minimum available real memory that we
4054 * can have in the system. However, this is generally fixed at 25 pages
4055 * which is so low that it's useless. In this comparison, we seek to
4056 * calculate the total heap-size, and reclaim if more than 3/4ths of the
4057 * heap is allocated. (Or, in the calculation, if less than 1/4th is
4060 n
= vmem_size(heap_arena
, VMEM_FREE
) -
4061 (vmem_size(heap_arena
, VMEM_FREE
| VMEM_ALLOC
) >> 2);
4069 * If zio data pages are being allocated out of a separate heap segment,
4070 * then enforce that the size of available vmem for this arena remains
4071 * above about 1/4th (1/(2^arc_zio_arena_free_shift)) free.
4073 * Note that reducing the arc_zio_arena_free_shift keeps more virtual
4074 * memory (in the zio_arena) free, which can avoid memory
4075 * fragmentation issues.
4077 if (zio_arena
!= NULL
) {
4078 n
= (int64_t)vmem_size(zio_arena
, VMEM_FREE
) -
4079 (vmem_size(zio_arena
, VMEM_ALLOC
) >>
4080 arc_zio_arena_free_shift
);
4087 /* Every 100 calls, free a small amount */
4088 if (spa_get_random(100) == 0)
4090 #endif /* _KERNEL */
4092 last_free_memory
= lowest
;
4093 last_free_reason
= r
;
4099 * Determine if the system is under memory pressure and is asking
4100 * to reclaim memory. A return value of B_TRUE indicates that the system
4101 * is under memory pressure and that the arc should adjust accordingly.
4104 arc_reclaim_needed(void)
4106 return (arc_available_memory() < 0);
4110 arc_kmem_reap_now(void)
4113 kmem_cache_t
*prev_cache
= NULL
;
4114 kmem_cache_t
*prev_data_cache
= NULL
;
4115 extern kmem_cache_t
*zio_buf_cache
[];
4116 extern kmem_cache_t
*zio_data_buf_cache
[];
4117 extern kmem_cache_t
*range_seg_cache
;
4119 if ((arc_meta_used
>= arc_meta_limit
) && zfs_arc_meta_prune
) {
4121 * We are exceeding our meta-data cache limit.
4122 * Prune some entries to release holds on meta-data.
4124 arc_prune_async(zfs_arc_meta_prune
);
4127 for (i
= 0; i
< SPA_MAXBLOCKSIZE
>> SPA_MINBLOCKSHIFT
; i
++) {
4129 /* reach upper limit of cache size on 32-bit */
4130 if (zio_buf_cache
[i
] == NULL
)
4133 if (zio_buf_cache
[i
] != prev_cache
) {
4134 prev_cache
= zio_buf_cache
[i
];
4135 kmem_cache_reap_now(zio_buf_cache
[i
]);
4137 if (zio_data_buf_cache
[i
] != prev_data_cache
) {
4138 prev_data_cache
= zio_data_buf_cache
[i
];
4139 kmem_cache_reap_now(zio_data_buf_cache
[i
]);
4142 kmem_cache_reap_now(buf_cache
);
4143 kmem_cache_reap_now(hdr_full_cache
);
4144 kmem_cache_reap_now(hdr_l2only_cache
);
4145 kmem_cache_reap_now(range_seg_cache
);
4147 if (zio_arena
!= NULL
) {
4149 * Ask the vmem arena to reclaim unused memory from its
4152 vmem_qcache_reap(zio_arena
);
4157 * Threads can block in arc_get_data_buf() waiting for this thread to evict
4158 * enough data and signal them to proceed. When this happens, the threads in
4159 * arc_get_data_buf() are sleeping while holding the hash lock for their
4160 * particular arc header. Thus, we must be careful to never sleep on a
4161 * hash lock in this thread. This is to prevent the following deadlock:
4163 * - Thread A sleeps on CV in arc_get_data_buf() holding hash lock "L",
4164 * waiting for the reclaim thread to signal it.
4166 * - arc_reclaim_thread() tries to acquire hash lock "L" using mutex_enter,
4167 * fails, and goes to sleep forever.
4169 * This possible deadlock is avoided by always acquiring a hash lock
4170 * using mutex_tryenter() from arc_reclaim_thread().
4173 arc_reclaim_thread(void)
4175 fstrans_cookie_t cookie
= spl_fstrans_mark();
4176 hrtime_t growtime
= 0;
4179 CALLB_CPR_INIT(&cpr
, &arc_reclaim_lock
, callb_generic_cpr
, FTAG
);
4181 mutex_enter(&arc_reclaim_lock
);
4182 while (!arc_reclaim_thread_exit
) {
4184 int64_t free_memory
= arc_available_memory();
4185 uint64_t evicted
= 0;
4187 arc_tuning_update();
4190 * This is necessary in order for the mdb ::arc dcmd to
4191 * show up to date information. Since the ::arc command
4192 * does not call the kstat's update function, without
4193 * this call, the command may show stale stats for the
4194 * anon, mru, mru_ghost, mfu, and mfu_ghost lists. Even
4195 * with this change, the data might be up to 1 second
4196 * out of date; but that should suffice. The arc_state_t
4197 * structures can be queried directly if more accurate
4198 * information is needed.
4201 if (arc_ksp
!= NULL
)
4202 arc_ksp
->ks_update(arc_ksp
, KSTAT_READ
);
4204 mutex_exit(&arc_reclaim_lock
);
4206 if (free_memory
< 0) {
4208 arc_no_grow
= B_TRUE
;
4212 * Wait at least zfs_grow_retry (default 5) seconds
4213 * before considering growing.
4215 growtime
= gethrtime() + SEC2NSEC(arc_grow_retry
);
4217 arc_kmem_reap_now();
4220 * If we are still low on memory, shrink the ARC
4221 * so that we have arc_shrink_min free space.
4223 free_memory
= arc_available_memory();
4225 to_free
= (arc_c
>> arc_shrink_shift
) - free_memory
;
4228 to_free
= MAX(to_free
, arc_need_free
);
4230 arc_shrink(to_free
);
4232 } else if (free_memory
< arc_c
>> arc_no_grow_shift
) {
4233 arc_no_grow
= B_TRUE
;
4234 } else if (gethrtime() >= growtime
) {
4235 arc_no_grow
= B_FALSE
;
4238 evicted
= arc_adjust();
4240 mutex_enter(&arc_reclaim_lock
);
4243 * If evicted is zero, we couldn't evict anything via
4244 * arc_adjust(). This could be due to hash lock
4245 * collisions, but more likely due to the majority of
4246 * arc buffers being unevictable. Therefore, even if
4247 * arc_size is above arc_c, another pass is unlikely to
4248 * be helpful and could potentially cause us to enter an
4251 if (arc_size
<= arc_c
|| evicted
== 0) {
4253 * We're either no longer overflowing, or we
4254 * can't evict anything more, so we should wake
4255 * up any threads before we go to sleep and clear
4256 * arc_need_free since nothing more can be done.
4258 cv_broadcast(&arc_reclaim_waiters_cv
);
4262 * Block until signaled, or after one second (we
4263 * might need to perform arc_kmem_reap_now()
4264 * even if we aren't being signalled)
4266 CALLB_CPR_SAFE_BEGIN(&cpr
);
4267 (void) cv_timedwait_sig_hires(&arc_reclaim_thread_cv
,
4268 &arc_reclaim_lock
, SEC2NSEC(1), MSEC2NSEC(1), 0);
4269 CALLB_CPR_SAFE_END(&cpr
, &arc_reclaim_lock
);
4273 arc_reclaim_thread_exit
= B_FALSE
;
4274 cv_broadcast(&arc_reclaim_thread_cv
);
4275 CALLB_CPR_EXIT(&cpr
); /* drops arc_reclaim_lock */
4276 spl_fstrans_unmark(cookie
);
4282 * Determine the amount of memory eligible for eviction contained in the
4283 * ARC. All clean data reported by the ghost lists can always be safely
4284 * evicted. Due to arc_c_min, the same does not hold for all clean data
4285 * contained by the regular mru and mfu lists.
4287 * In the case of the regular mru and mfu lists, we need to report as
4288 * much clean data as possible, such that evicting that same reported
4289 * data will not bring arc_size below arc_c_min. Thus, in certain
4290 * circumstances, the total amount of clean data in the mru and mfu
4291 * lists might not actually be evictable.
4293 * The following two distinct cases are accounted for:
4295 * 1. The sum of the amount of dirty data contained by both the mru and
4296 * mfu lists, plus the ARC's other accounting (e.g. the anon list),
4297 * is greater than or equal to arc_c_min.
4298 * (i.e. amount of dirty data >= arc_c_min)
4300 * This is the easy case; all clean data contained by the mru and mfu
4301 * lists is evictable. Evicting all clean data can only drop arc_size
4302 * to the amount of dirty data, which is greater than arc_c_min.
4304 * 2. The sum of the amount of dirty data contained by both the mru and
4305 * mfu lists, plus the ARC's other accounting (e.g. the anon list),
4306 * is less than arc_c_min.
4307 * (i.e. arc_c_min > amount of dirty data)
4309 * 2.1. arc_size is greater than or equal arc_c_min.
4310 * (i.e. arc_size >= arc_c_min > amount of dirty data)
4312 * In this case, not all clean data from the regular mru and mfu
4313 * lists is actually evictable; we must leave enough clean data
4314 * to keep arc_size above arc_c_min. Thus, the maximum amount of
4315 * evictable data from the two lists combined, is exactly the
4316 * difference between arc_size and arc_c_min.
4318 * 2.2. arc_size is less than arc_c_min
4319 * (i.e. arc_c_min > arc_size > amount of dirty data)
4321 * In this case, none of the data contained in the mru and mfu
4322 * lists is evictable, even if it's clean. Since arc_size is
4323 * already below arc_c_min, evicting any more would only
4324 * increase this negative difference.
4327 arc_evictable_memory(void) {
4328 uint64_t arc_clean
=
4329 refcount_count(&arc_mru
->arcs_esize
[ARC_BUFC_DATA
]) +
4330 refcount_count(&arc_mru
->arcs_esize
[ARC_BUFC_METADATA
]) +
4331 refcount_count(&arc_mfu
->arcs_esize
[ARC_BUFC_DATA
]) +
4332 refcount_count(&arc_mfu
->arcs_esize
[ARC_BUFC_METADATA
]);
4333 uint64_t ghost_clean
=
4334 refcount_count(&arc_mru_ghost
->arcs_esize
[ARC_BUFC_DATA
]) +
4335 refcount_count(&arc_mru_ghost
->arcs_esize
[ARC_BUFC_METADATA
]) +
4336 refcount_count(&arc_mfu_ghost
->arcs_esize
[ARC_BUFC_DATA
]) +
4337 refcount_count(&arc_mfu_ghost
->arcs_esize
[ARC_BUFC_METADATA
]);
4338 uint64_t arc_dirty
= MAX((int64_t)arc_size
- (int64_t)arc_clean
, 0);
4340 if (arc_dirty
>= arc_c_min
)
4341 return (ghost_clean
+ arc_clean
);
4343 return (ghost_clean
+ MAX((int64_t)arc_size
- (int64_t)arc_c_min
, 0));
4347 * If sc->nr_to_scan is zero, the caller is requesting a query of the
4348 * number of objects which can potentially be freed. If it is nonzero,
4349 * the request is to free that many objects.
4351 * Linux kernels >= 3.12 have the count_objects and scan_objects callbacks
4352 * in struct shrinker and also require the shrinker to return the number
4355 * Older kernels require the shrinker to return the number of freeable
4356 * objects following the freeing of nr_to_free.
4358 static spl_shrinker_t
4359 __arc_shrinker_func(struct shrinker
*shrink
, struct shrink_control
*sc
)
4363 /* The arc is considered warm once reclaim has occurred */
4364 if (unlikely(arc_warm
== B_FALSE
))
4367 /* Return the potential number of reclaimable pages */
4368 pages
= btop((int64_t)arc_evictable_memory());
4369 if (sc
->nr_to_scan
== 0)
4372 /* Not allowed to perform filesystem reclaim */
4373 if (!(sc
->gfp_mask
& __GFP_FS
))
4374 return (SHRINK_STOP
);
4376 /* Reclaim in progress */
4377 if (mutex_tryenter(&arc_reclaim_lock
) == 0)
4378 return (SHRINK_STOP
);
4380 mutex_exit(&arc_reclaim_lock
);
4383 * Evict the requested number of pages by shrinking arc_c the
4384 * requested amount. If there is nothing left to evict just
4385 * reap whatever we can from the various arc slabs.
4388 arc_shrink(ptob(sc
->nr_to_scan
));
4389 arc_kmem_reap_now();
4390 #ifdef HAVE_SPLIT_SHRINKER_CALLBACK
4391 pages
= MAX(pages
- btop(arc_evictable_memory()), 0);
4393 pages
= btop(arc_evictable_memory());
4396 arc_kmem_reap_now();
4397 pages
= SHRINK_STOP
;
4401 * We've reaped what we can, wake up threads.
4403 cv_broadcast(&arc_reclaim_waiters_cv
);
4406 * When direct reclaim is observed it usually indicates a rapid
4407 * increase in memory pressure. This occurs because the kswapd
4408 * threads were unable to asynchronously keep enough free memory
4409 * available. In this case set arc_no_grow to briefly pause arc
4410 * growth to avoid compounding the memory pressure.
4412 if (current_is_kswapd()) {
4413 ARCSTAT_BUMP(arcstat_memory_indirect_count
);
4415 arc_no_grow
= B_TRUE
;
4416 arc_need_free
= ptob(sc
->nr_to_scan
);
4417 ARCSTAT_BUMP(arcstat_memory_direct_count
);
4422 SPL_SHRINKER_CALLBACK_WRAPPER(arc_shrinker_func
);
4424 SPL_SHRINKER_DECLARE(arc_shrinker
, arc_shrinker_func
, DEFAULT_SEEKS
);
4425 #endif /* _KERNEL */
4428 * Adapt arc info given the number of bytes we are trying to add and
4429 * the state that we are comming from. This function is only called
4430 * when we are adding new content to the cache.
4433 arc_adapt(int bytes
, arc_state_t
*state
)
4436 uint64_t arc_p_min
= (arc_c
>> arc_p_min_shift
);
4437 int64_t mrug_size
= refcount_count(&arc_mru_ghost
->arcs_size
);
4438 int64_t mfug_size
= refcount_count(&arc_mfu_ghost
->arcs_size
);
4440 if (state
== arc_l2c_only
)
4445 * Adapt the target size of the MRU list:
4446 * - if we just hit in the MRU ghost list, then increase
4447 * the target size of the MRU list.
4448 * - if we just hit in the MFU ghost list, then increase
4449 * the target size of the MFU list by decreasing the
4450 * target size of the MRU list.
4452 if (state
== arc_mru_ghost
) {
4453 mult
= (mrug_size
>= mfug_size
) ? 1 : (mfug_size
/ mrug_size
);
4454 if (!zfs_arc_p_dampener_disable
)
4455 mult
= MIN(mult
, 10); /* avoid wild arc_p adjustment */
4457 arc_p
= MIN(arc_c
- arc_p_min
, arc_p
+ bytes
* mult
);
4458 } else if (state
== arc_mfu_ghost
) {
4461 mult
= (mfug_size
>= mrug_size
) ? 1 : (mrug_size
/ mfug_size
);
4462 if (!zfs_arc_p_dampener_disable
)
4463 mult
= MIN(mult
, 10);
4465 delta
= MIN(bytes
* mult
, arc_p
);
4466 arc_p
= MAX(arc_p_min
, arc_p
- delta
);
4468 ASSERT((int64_t)arc_p
>= 0);
4470 if (arc_reclaim_needed()) {
4471 cv_signal(&arc_reclaim_thread_cv
);
4478 if (arc_c
>= arc_c_max
)
4482 * If we're within (2 * maxblocksize) bytes of the target
4483 * cache size, increment the target cache size
4485 ASSERT3U(arc_c
, >=, 2ULL << SPA_MAXBLOCKSHIFT
);
4486 if (arc_size
>= arc_c
- (2ULL << SPA_MAXBLOCKSHIFT
)) {
4487 atomic_add_64(&arc_c
, (int64_t)bytes
);
4488 if (arc_c
> arc_c_max
)
4490 else if (state
== arc_anon
)
4491 atomic_add_64(&arc_p
, (int64_t)bytes
);
4495 ASSERT((int64_t)arc_p
>= 0);
4499 * Check if arc_size has grown past our upper threshold, determined by
4500 * zfs_arc_overflow_shift.
4503 arc_is_overflowing(void)
4505 /* Always allow at least one block of overflow */
4506 uint64_t overflow
= MAX(SPA_MAXBLOCKSIZE
,
4507 arc_c
>> zfs_arc_overflow_shift
);
4509 return (arc_size
>= arc_c
+ overflow
);
4513 * Allocate a block and return it to the caller. If we are hitting the
4514 * hard limit for the cache size, we must sleep, waiting for the eviction
4515 * thread to catch up. If we're past the target size but below the hard
4516 * limit, we'll only signal the reclaim thread and continue on.
4519 arc_get_data_buf(arc_buf_hdr_t
*hdr
, uint64_t size
, void *tag
)
4522 arc_state_t
*state
= hdr
->b_l1hdr
.b_state
;
4523 arc_buf_contents_t type
= arc_buf_type(hdr
);
4525 arc_adapt(size
, state
);
4528 * If arc_size is currently overflowing, and has grown past our
4529 * upper limit, we must be adding data faster than the evict
4530 * thread can evict. Thus, to ensure we don't compound the
4531 * problem by adding more data and forcing arc_size to grow even
4532 * further past it's target size, we halt and wait for the
4533 * eviction thread to catch up.
4535 * It's also possible that the reclaim thread is unable to evict
4536 * enough buffers to get arc_size below the overflow limit (e.g.
4537 * due to buffers being un-evictable, or hash lock collisions).
4538 * In this case, we want to proceed regardless if we're
4539 * overflowing; thus we don't use a while loop here.
4541 if (arc_is_overflowing()) {
4542 mutex_enter(&arc_reclaim_lock
);
4545 * Now that we've acquired the lock, we may no longer be
4546 * over the overflow limit, lets check.
4548 * We're ignoring the case of spurious wake ups. If that
4549 * were to happen, it'd let this thread consume an ARC
4550 * buffer before it should have (i.e. before we're under
4551 * the overflow limit and were signalled by the reclaim
4552 * thread). As long as that is a rare occurrence, it
4553 * shouldn't cause any harm.
4555 if (arc_is_overflowing()) {
4556 cv_signal(&arc_reclaim_thread_cv
);
4557 cv_wait(&arc_reclaim_waiters_cv
, &arc_reclaim_lock
);
4560 mutex_exit(&arc_reclaim_lock
);
4563 VERIFY3U(hdr
->b_type
, ==, type
);
4564 if (type
== ARC_BUFC_METADATA
) {
4565 datap
= zio_buf_alloc(size
);
4566 arc_space_consume(size
, ARC_SPACE_META
);
4568 ASSERT(type
== ARC_BUFC_DATA
);
4569 datap
= zio_data_buf_alloc(size
);
4570 arc_space_consume(size
, ARC_SPACE_DATA
);
4574 * Update the state size. Note that ghost states have a
4575 * "ghost size" and so don't need to be updated.
4577 if (!GHOST_STATE(state
)) {
4579 (void) refcount_add_many(&state
->arcs_size
, size
, tag
);
4582 * If this is reached via arc_read, the link is
4583 * protected by the hash lock. If reached via
4584 * arc_buf_alloc, the header should not be accessed by
4585 * any other thread. And, if reached via arc_read_done,
4586 * the hash lock will protect it if it's found in the
4587 * hash table; otherwise no other thread should be
4588 * trying to [add|remove]_reference it.
4590 if (multilist_link_active(&hdr
->b_l1hdr
.b_arc_node
)) {
4591 ASSERT(refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
));
4592 (void) refcount_add_many(&state
->arcs_esize
[type
],
4597 * If we are growing the cache, and we are adding anonymous
4598 * data, and we have outgrown arc_p, update arc_p
4600 if (arc_size
< arc_c
&& hdr
->b_l1hdr
.b_state
== arc_anon
&&
4601 (refcount_count(&arc_anon
->arcs_size
) +
4602 refcount_count(&arc_mru
->arcs_size
) > arc_p
))
4603 arc_p
= MIN(arc_c
, arc_p
+ size
);
4609 * Free the arc data buffer.
4612 arc_free_data_buf(arc_buf_hdr_t
*hdr
, void *data
, uint64_t size
, void *tag
)
4614 arc_state_t
*state
= hdr
->b_l1hdr
.b_state
;
4615 arc_buf_contents_t type
= arc_buf_type(hdr
);
4617 /* protected by hash lock, if in the hash table */
4618 if (multilist_link_active(&hdr
->b_l1hdr
.b_arc_node
)) {
4619 ASSERT(refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
));
4620 ASSERT(state
!= arc_anon
&& state
!= arc_l2c_only
);
4622 (void) refcount_remove_many(&state
->arcs_esize
[type
],
4625 (void) refcount_remove_many(&state
->arcs_size
, size
, tag
);
4627 VERIFY3U(hdr
->b_type
, ==, type
);
4628 if (type
== ARC_BUFC_METADATA
) {
4629 zio_buf_free(data
, size
);
4630 arc_space_return(size
, ARC_SPACE_META
);
4632 ASSERT(type
== ARC_BUFC_DATA
);
4633 zio_data_buf_free(data
, size
);
4634 arc_space_return(size
, ARC_SPACE_DATA
);
4639 * This routine is called whenever a buffer is accessed.
4640 * NOTE: the hash lock is dropped in this function.
4643 arc_access(arc_buf_hdr_t
*hdr
, kmutex_t
*hash_lock
)
4647 ASSERT(MUTEX_HELD(hash_lock
));
4648 ASSERT(HDR_HAS_L1HDR(hdr
));
4650 if (hdr
->b_l1hdr
.b_state
== arc_anon
) {
4652 * This buffer is not in the cache, and does not
4653 * appear in our "ghost" list. Add the new buffer
4657 ASSERT0(hdr
->b_l1hdr
.b_arc_access
);
4658 hdr
->b_l1hdr
.b_arc_access
= ddi_get_lbolt();
4659 DTRACE_PROBE1(new_state__mru
, arc_buf_hdr_t
*, hdr
);
4660 arc_change_state(arc_mru
, hdr
, hash_lock
);
4662 } else if (hdr
->b_l1hdr
.b_state
== arc_mru
) {
4663 now
= ddi_get_lbolt();
4666 * If this buffer is here because of a prefetch, then either:
4667 * - clear the flag if this is a "referencing" read
4668 * (any subsequent access will bump this into the MFU state).
4670 * - move the buffer to the head of the list if this is
4671 * another prefetch (to make it less likely to be evicted).
4673 if (HDR_PREFETCH(hdr
)) {
4674 if (refcount_count(&hdr
->b_l1hdr
.b_refcnt
) == 0) {
4675 /* link protected by hash lock */
4676 ASSERT(multilist_link_active(
4677 &hdr
->b_l1hdr
.b_arc_node
));
4679 arc_hdr_clear_flags(hdr
, ARC_FLAG_PREFETCH
);
4680 atomic_inc_32(&hdr
->b_l1hdr
.b_mru_hits
);
4681 ARCSTAT_BUMP(arcstat_mru_hits
);
4683 hdr
->b_l1hdr
.b_arc_access
= now
;
4688 * This buffer has been "accessed" only once so far,
4689 * but it is still in the cache. Move it to the MFU
4692 if (ddi_time_after(now
, hdr
->b_l1hdr
.b_arc_access
+
4695 * More than 125ms have passed since we
4696 * instantiated this buffer. Move it to the
4697 * most frequently used state.
4699 hdr
->b_l1hdr
.b_arc_access
= now
;
4700 DTRACE_PROBE1(new_state__mfu
, arc_buf_hdr_t
*, hdr
);
4701 arc_change_state(arc_mfu
, hdr
, hash_lock
);
4703 atomic_inc_32(&hdr
->b_l1hdr
.b_mru_hits
);
4704 ARCSTAT_BUMP(arcstat_mru_hits
);
4705 } else if (hdr
->b_l1hdr
.b_state
== arc_mru_ghost
) {
4706 arc_state_t
*new_state
;
4708 * This buffer has been "accessed" recently, but
4709 * was evicted from the cache. Move it to the
4713 if (HDR_PREFETCH(hdr
)) {
4714 new_state
= arc_mru
;
4715 if (refcount_count(&hdr
->b_l1hdr
.b_refcnt
) > 0)
4716 arc_hdr_clear_flags(hdr
, ARC_FLAG_PREFETCH
);
4717 DTRACE_PROBE1(new_state__mru
, arc_buf_hdr_t
*, hdr
);
4719 new_state
= arc_mfu
;
4720 DTRACE_PROBE1(new_state__mfu
, arc_buf_hdr_t
*, hdr
);
4723 hdr
->b_l1hdr
.b_arc_access
= ddi_get_lbolt();
4724 arc_change_state(new_state
, hdr
, hash_lock
);
4726 atomic_inc_32(&hdr
->b_l1hdr
.b_mru_ghost_hits
);
4727 ARCSTAT_BUMP(arcstat_mru_ghost_hits
);
4728 } else if (hdr
->b_l1hdr
.b_state
== arc_mfu
) {
4730 * This buffer has been accessed more than once and is
4731 * still in the cache. Keep it in the MFU state.
4733 * NOTE: an add_reference() that occurred when we did
4734 * the arc_read() will have kicked this off the list.
4735 * If it was a prefetch, we will explicitly move it to
4736 * the head of the list now.
4738 if ((HDR_PREFETCH(hdr
)) != 0) {
4739 ASSERT(refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
));
4740 /* link protected by hash_lock */
4741 ASSERT(multilist_link_active(&hdr
->b_l1hdr
.b_arc_node
));
4743 atomic_inc_32(&hdr
->b_l1hdr
.b_mfu_hits
);
4744 ARCSTAT_BUMP(arcstat_mfu_hits
);
4745 hdr
->b_l1hdr
.b_arc_access
= ddi_get_lbolt();
4746 } else if (hdr
->b_l1hdr
.b_state
== arc_mfu_ghost
) {
4747 arc_state_t
*new_state
= arc_mfu
;
4749 * This buffer has been accessed more than once but has
4750 * been evicted from the cache. Move it back to the
4754 if (HDR_PREFETCH(hdr
)) {
4756 * This is a prefetch access...
4757 * move this block back to the MRU state.
4759 ASSERT0(refcount_count(&hdr
->b_l1hdr
.b_refcnt
));
4760 new_state
= arc_mru
;
4763 hdr
->b_l1hdr
.b_arc_access
= ddi_get_lbolt();
4764 DTRACE_PROBE1(new_state__mfu
, arc_buf_hdr_t
*, hdr
);
4765 arc_change_state(new_state
, hdr
, hash_lock
);
4767 atomic_inc_32(&hdr
->b_l1hdr
.b_mfu_ghost_hits
);
4768 ARCSTAT_BUMP(arcstat_mfu_ghost_hits
);
4769 } else if (hdr
->b_l1hdr
.b_state
== arc_l2c_only
) {
4771 * This buffer is on the 2nd Level ARC.
4774 hdr
->b_l1hdr
.b_arc_access
= ddi_get_lbolt();
4775 DTRACE_PROBE1(new_state__mfu
, arc_buf_hdr_t
*, hdr
);
4776 arc_change_state(arc_mfu
, hdr
, hash_lock
);
4778 cmn_err(CE_PANIC
, "invalid arc state 0x%p",
4779 hdr
->b_l1hdr
.b_state
);
4783 /* a generic arc_done_func_t which you can use */
4786 arc_bcopy_func(zio_t
*zio
, arc_buf_t
*buf
, void *arg
)
4788 if (zio
== NULL
|| zio
->io_error
== 0)
4789 bcopy(buf
->b_data
, arg
, arc_buf_size(buf
));
4790 arc_buf_destroy(buf
, arg
);
4793 /* a generic arc_done_func_t */
4795 arc_getbuf_func(zio_t
*zio
, arc_buf_t
*buf
, void *arg
)
4797 arc_buf_t
**bufp
= arg
;
4798 if (zio
&& zio
->io_error
) {
4799 arc_buf_destroy(buf
, arg
);
4803 ASSERT(buf
->b_data
);
4808 arc_hdr_verify(arc_buf_hdr_t
*hdr
, blkptr_t
*bp
)
4810 if (BP_IS_HOLE(bp
) || BP_IS_EMBEDDED(bp
)) {
4811 ASSERT3U(HDR_GET_PSIZE(hdr
), ==, 0);
4812 ASSERT3U(HDR_GET_COMPRESS(hdr
), ==, ZIO_COMPRESS_OFF
);
4814 if (HDR_COMPRESSION_ENABLED(hdr
)) {
4815 ASSERT3U(HDR_GET_COMPRESS(hdr
), ==,
4816 BP_GET_COMPRESS(bp
));
4818 ASSERT3U(HDR_GET_LSIZE(hdr
), ==, BP_GET_LSIZE(bp
));
4819 ASSERT3U(HDR_GET_PSIZE(hdr
), ==, BP_GET_PSIZE(bp
));
4824 arc_read_done(zio_t
*zio
)
4826 arc_buf_hdr_t
*hdr
= zio
->io_private
;
4827 kmutex_t
*hash_lock
= NULL
;
4828 arc_callback_t
*callback_list
;
4829 arc_callback_t
*acb
;
4830 boolean_t freeable
= B_FALSE
;
4831 boolean_t no_zio_error
= (zio
->io_error
== 0);
4832 int callback_cnt
= 0;
4834 * The hdr was inserted into hash-table and removed from lists
4835 * prior to starting I/O. We should find this header, since
4836 * it's in the hash table, and it should be legit since it's
4837 * not possible to evict it during the I/O. The only possible
4838 * reason for it not to be found is if we were freed during the
4841 if (HDR_IN_HASH_TABLE(hdr
)) {
4842 arc_buf_hdr_t
*found
;
4844 ASSERT3U(hdr
->b_birth
, ==, BP_PHYSICAL_BIRTH(zio
->io_bp
));
4845 ASSERT3U(hdr
->b_dva
.dva_word
[0], ==,
4846 BP_IDENTITY(zio
->io_bp
)->dva_word
[0]);
4847 ASSERT3U(hdr
->b_dva
.dva_word
[1], ==,
4848 BP_IDENTITY(zio
->io_bp
)->dva_word
[1]);
4850 found
= buf_hash_find(hdr
->b_spa
, zio
->io_bp
, &hash_lock
);
4852 ASSERT((found
== hdr
&&
4853 DVA_EQUAL(&hdr
->b_dva
, BP_IDENTITY(zio
->io_bp
))) ||
4854 (found
== hdr
&& HDR_L2_READING(hdr
)));
4855 ASSERT3P(hash_lock
, !=, NULL
);
4859 /* byteswap if necessary */
4860 if (BP_SHOULD_BYTESWAP(zio
->io_bp
)) {
4861 if (BP_GET_LEVEL(zio
->io_bp
) > 0) {
4862 hdr
->b_l1hdr
.b_byteswap
= DMU_BSWAP_UINT64
;
4864 hdr
->b_l1hdr
.b_byteswap
=
4865 DMU_OT_BYTESWAP(BP_GET_TYPE(zio
->io_bp
));
4868 hdr
->b_l1hdr
.b_byteswap
= DMU_BSWAP_NUMFUNCS
;
4872 arc_hdr_clear_flags(hdr
, ARC_FLAG_L2_EVICTED
);
4873 if (l2arc_noprefetch
&& HDR_PREFETCH(hdr
))
4874 arc_hdr_clear_flags(hdr
, ARC_FLAG_L2CACHE
);
4876 callback_list
= hdr
->b_l1hdr
.b_acb
;
4877 ASSERT3P(callback_list
, !=, NULL
);
4879 if (hash_lock
&& no_zio_error
&& hdr
->b_l1hdr
.b_state
== arc_anon
) {
4881 * Only call arc_access on anonymous buffers. This is because
4882 * if we've issued an I/O for an evicted buffer, we've already
4883 * called arc_access (to prevent any simultaneous readers from
4884 * getting confused).
4886 arc_access(hdr
, hash_lock
);
4890 * If a read request has a callback (i.e. acb_done is not NULL), then we
4891 * make a buf containing the data according to the parameters which were
4892 * passed in. The implementation of arc_buf_alloc_impl() ensures that we
4893 * aren't needlessly decompressing the data multiple times.
4895 for (acb
= callback_list
; acb
!= NULL
; acb
= acb
->acb_next
) {
4900 /* This is a demand read since prefetches don't use callbacks */
4904 error
= arc_buf_alloc_impl(hdr
, acb
->acb_private
,
4905 acb
->acb_compressed
, no_zio_error
, &acb
->acb_buf
);
4907 zio
->io_error
= error
;
4910 hdr
->b_l1hdr
.b_acb
= NULL
;
4911 arc_hdr_clear_flags(hdr
, ARC_FLAG_IO_IN_PROGRESS
);
4912 if (callback_cnt
== 0) {
4913 ASSERT(HDR_PREFETCH(hdr
));
4914 ASSERT0(hdr
->b_l1hdr
.b_bufcnt
);
4915 ASSERT3P(hdr
->b_l1hdr
.b_pdata
, !=, NULL
);
4918 ASSERT(refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
) ||
4919 callback_list
!= NULL
);
4922 arc_hdr_verify(hdr
, zio
->io_bp
);
4924 arc_hdr_set_flags(hdr
, ARC_FLAG_IO_ERROR
);
4925 if (hdr
->b_l1hdr
.b_state
!= arc_anon
)
4926 arc_change_state(arc_anon
, hdr
, hash_lock
);
4927 if (HDR_IN_HASH_TABLE(hdr
))
4928 buf_hash_remove(hdr
);
4929 freeable
= refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
);
4933 * Broadcast before we drop the hash_lock to avoid the possibility
4934 * that the hdr (and hence the cv) might be freed before we get to
4935 * the cv_broadcast().
4937 cv_broadcast(&hdr
->b_l1hdr
.b_cv
);
4939 if (hash_lock
!= NULL
) {
4940 mutex_exit(hash_lock
);
4943 * This block was freed while we waited for the read to
4944 * complete. It has been removed from the hash table and
4945 * moved to the anonymous state (so that it won't show up
4948 ASSERT3P(hdr
->b_l1hdr
.b_state
, ==, arc_anon
);
4949 freeable
= refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
);
4952 /* execute each callback and free its structure */
4953 while ((acb
= callback_list
) != NULL
) {
4955 acb
->acb_done(zio
, acb
->acb_buf
, acb
->acb_private
);
4957 if (acb
->acb_zio_dummy
!= NULL
) {
4958 acb
->acb_zio_dummy
->io_error
= zio
->io_error
;
4959 zio_nowait(acb
->acb_zio_dummy
);
4962 callback_list
= acb
->acb_next
;
4963 kmem_free(acb
, sizeof (arc_callback_t
));
4967 arc_hdr_destroy(hdr
);
4971 * "Read" the block at the specified DVA (in bp) via the
4972 * cache. If the block is found in the cache, invoke the provided
4973 * callback immediately and return. Note that the `zio' parameter
4974 * in the callback will be NULL in this case, since no IO was
4975 * required. If the block is not in the cache pass the read request
4976 * on to the spa with a substitute callback function, so that the
4977 * requested block will be added to the cache.
4979 * If a read request arrives for a block that has a read in-progress,
4980 * either wait for the in-progress read to complete (and return the
4981 * results); or, if this is a read with a "done" func, add a record
4982 * to the read to invoke the "done" func when the read completes,
4983 * and return; or just return.
4985 * arc_read_done() will invoke all the requested "done" functions
4986 * for readers of this block.
4989 arc_read(zio_t
*pio
, spa_t
*spa
, const blkptr_t
*bp
, arc_done_func_t
*done
,
4990 void *private, zio_priority_t priority
, int zio_flags
,
4991 arc_flags_t
*arc_flags
, const zbookmark_phys_t
*zb
)
4993 arc_buf_hdr_t
*hdr
= NULL
;
4994 kmutex_t
*hash_lock
= NULL
;
4996 uint64_t guid
= spa_load_guid(spa
);
4997 boolean_t compressed_read
= (zio_flags
& ZIO_FLAG_RAW
) != 0;
5000 ASSERT(!BP_IS_EMBEDDED(bp
) ||
5001 BPE_GET_ETYPE(bp
) == BP_EMBEDDED_TYPE_DATA
);
5004 if (!BP_IS_EMBEDDED(bp
)) {
5006 * Embedded BP's have no DVA and require no I/O to "read".
5007 * Create an anonymous arc buf to back it.
5009 hdr
= buf_hash_find(guid
, bp
, &hash_lock
);
5012 if (hdr
!= NULL
&& HDR_HAS_L1HDR(hdr
) && hdr
->b_l1hdr
.b_pdata
!= NULL
) {
5013 arc_buf_t
*buf
= NULL
;
5014 *arc_flags
|= ARC_FLAG_CACHED
;
5016 if (HDR_IO_IN_PROGRESS(hdr
)) {
5018 if ((hdr
->b_flags
& ARC_FLAG_PRIO_ASYNC_READ
) &&
5019 priority
== ZIO_PRIORITY_SYNC_READ
) {
5021 * This sync read must wait for an
5022 * in-progress async read (e.g. a predictive
5023 * prefetch). Async reads are queued
5024 * separately at the vdev_queue layer, so
5025 * this is a form of priority inversion.
5026 * Ideally, we would "inherit" the demand
5027 * i/o's priority by moving the i/o from
5028 * the async queue to the synchronous queue,
5029 * but there is currently no mechanism to do
5030 * so. Track this so that we can evaluate
5031 * the magnitude of this potential performance
5034 * Note that if the prefetch i/o is already
5035 * active (has been issued to the device),
5036 * the prefetch improved performance, because
5037 * we issued it sooner than we would have
5038 * without the prefetch.
5040 DTRACE_PROBE1(arc__sync__wait__for__async
,
5041 arc_buf_hdr_t
*, hdr
);
5042 ARCSTAT_BUMP(arcstat_sync_wait_for_async
);
5044 if (hdr
->b_flags
& ARC_FLAG_PREDICTIVE_PREFETCH
) {
5045 arc_hdr_clear_flags(hdr
,
5046 ARC_FLAG_PREDICTIVE_PREFETCH
);
5049 if (*arc_flags
& ARC_FLAG_WAIT
) {
5050 cv_wait(&hdr
->b_l1hdr
.b_cv
, hash_lock
);
5051 mutex_exit(hash_lock
);
5054 ASSERT(*arc_flags
& ARC_FLAG_NOWAIT
);
5057 arc_callback_t
*acb
= NULL
;
5059 acb
= kmem_zalloc(sizeof (arc_callback_t
),
5061 acb
->acb_done
= done
;
5062 acb
->acb_private
= private;
5064 acb
->acb_zio_dummy
= zio_null(pio
,
5065 spa
, NULL
, NULL
, NULL
, zio_flags
);
5067 ASSERT3P(acb
->acb_done
, !=, NULL
);
5068 acb
->acb_next
= hdr
->b_l1hdr
.b_acb
;
5069 hdr
->b_l1hdr
.b_acb
= acb
;
5070 mutex_exit(hash_lock
);
5073 mutex_exit(hash_lock
);
5077 ASSERT(hdr
->b_l1hdr
.b_state
== arc_mru
||
5078 hdr
->b_l1hdr
.b_state
== arc_mfu
);
5081 if (hdr
->b_flags
& ARC_FLAG_PREDICTIVE_PREFETCH
) {
5083 * This is a demand read which does not have to
5084 * wait for i/o because we did a predictive
5085 * prefetch i/o for it, which has completed.
5088 arc__demand__hit__predictive__prefetch
,
5089 arc_buf_hdr_t
*, hdr
);
5091 arcstat_demand_hit_predictive_prefetch
);
5092 arc_hdr_clear_flags(hdr
,
5093 ARC_FLAG_PREDICTIVE_PREFETCH
);
5095 ASSERT(!BP_IS_EMBEDDED(bp
) || !BP_IS_HOLE(bp
));
5097 /* Get a buf with the desired data in it. */
5098 VERIFY0(arc_buf_alloc_impl(hdr
, private,
5099 compressed_read
, B_TRUE
, &buf
));
5100 } else if (*arc_flags
& ARC_FLAG_PREFETCH
&&
5101 refcount_count(&hdr
->b_l1hdr
.b_refcnt
) == 0) {
5102 arc_hdr_set_flags(hdr
, ARC_FLAG_PREFETCH
);
5104 DTRACE_PROBE1(arc__hit
, arc_buf_hdr_t
*, hdr
);
5105 arc_access(hdr
, hash_lock
);
5106 if (*arc_flags
& ARC_FLAG_L2CACHE
)
5107 arc_hdr_set_flags(hdr
, ARC_FLAG_L2CACHE
);
5108 mutex_exit(hash_lock
);
5109 ARCSTAT_BUMP(arcstat_hits
);
5110 ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr
),
5111 demand
, prefetch
, !HDR_ISTYPE_METADATA(hdr
),
5112 data
, metadata
, hits
);
5115 done(NULL
, buf
, private);
5117 uint64_t lsize
= BP_GET_LSIZE(bp
);
5118 uint64_t psize
= BP_GET_PSIZE(bp
);
5119 arc_callback_t
*acb
;
5122 boolean_t devw
= B_FALSE
;
5126 * Gracefully handle a damaged logical block size as a
5129 if (lsize
> spa_maxblocksize(spa
)) {
5130 rc
= SET_ERROR(ECKSUM
);
5135 /* this block is not in the cache */
5136 arc_buf_hdr_t
*exists
= NULL
;
5137 arc_buf_contents_t type
= BP_GET_BUFC_TYPE(bp
);
5138 hdr
= arc_hdr_alloc(spa_load_guid(spa
), psize
, lsize
,
5139 BP_GET_COMPRESS(bp
), type
);
5141 if (!BP_IS_EMBEDDED(bp
)) {
5142 hdr
->b_dva
= *BP_IDENTITY(bp
);
5143 hdr
->b_birth
= BP_PHYSICAL_BIRTH(bp
);
5144 exists
= buf_hash_insert(hdr
, &hash_lock
);
5146 if (exists
!= NULL
) {
5147 /* somebody beat us to the hash insert */
5148 mutex_exit(hash_lock
);
5149 buf_discard_identity(hdr
);
5150 arc_hdr_destroy(hdr
);
5151 goto top
; /* restart the IO request */
5155 * This block is in the ghost cache. If it was L2-only
5156 * (and thus didn't have an L1 hdr), we realloc the
5157 * header to add an L1 hdr.
5159 if (!HDR_HAS_L1HDR(hdr
)) {
5160 hdr
= arc_hdr_realloc(hdr
, hdr_l2only_cache
,
5164 ASSERT3P(hdr
->b_l1hdr
.b_pdata
, ==, NULL
);
5165 ASSERT(GHOST_STATE(hdr
->b_l1hdr
.b_state
));
5166 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
5167 ASSERT(refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
));
5168 ASSERT3P(hdr
->b_l1hdr
.b_buf
, ==, NULL
);
5169 ASSERT3P(hdr
->b_l1hdr
.b_freeze_cksum
, ==, NULL
);
5172 * This is a delicate dance that we play here.
5173 * This hdr is in the ghost list so we access it
5174 * to move it out of the ghost list before we
5175 * initiate the read. If it's a prefetch then
5176 * it won't have a callback so we'll remove the
5177 * reference that arc_buf_alloc_impl() created. We
5178 * do this after we've called arc_access() to
5179 * avoid hitting an assert in remove_reference().
5181 arc_access(hdr
, hash_lock
);
5182 arc_hdr_alloc_pdata(hdr
);
5184 ASSERT3P(hdr
->b_l1hdr
.b_pdata
, !=, NULL
);
5185 size
= arc_hdr_size(hdr
);
5188 * If compression is enabled on the hdr, then will do
5189 * RAW I/O and will store the compressed data in the hdr's
5190 * data block. Otherwise, the hdr's data block will contain
5191 * the uncompressed data.
5193 if (HDR_GET_COMPRESS(hdr
) != ZIO_COMPRESS_OFF
) {
5194 zio_flags
|= ZIO_FLAG_RAW
;
5197 if (*arc_flags
& ARC_FLAG_PREFETCH
)
5198 arc_hdr_set_flags(hdr
, ARC_FLAG_PREFETCH
);
5199 if (*arc_flags
& ARC_FLAG_L2CACHE
)
5200 arc_hdr_set_flags(hdr
, ARC_FLAG_L2CACHE
);
5201 if (BP_GET_LEVEL(bp
) > 0)
5202 arc_hdr_set_flags(hdr
, ARC_FLAG_INDIRECT
);
5203 if (*arc_flags
& ARC_FLAG_PREDICTIVE_PREFETCH
)
5204 arc_hdr_set_flags(hdr
, ARC_FLAG_PREDICTIVE_PREFETCH
);
5205 ASSERT(!GHOST_STATE(hdr
->b_l1hdr
.b_state
));
5207 acb
= kmem_zalloc(sizeof (arc_callback_t
), KM_SLEEP
);
5208 acb
->acb_done
= done
;
5209 acb
->acb_private
= private;
5210 acb
->acb_compressed
= compressed_read
;
5212 ASSERT3P(hdr
->b_l1hdr
.b_acb
, ==, NULL
);
5213 hdr
->b_l1hdr
.b_acb
= acb
;
5214 arc_hdr_set_flags(hdr
, ARC_FLAG_IO_IN_PROGRESS
);
5216 if (HDR_HAS_L2HDR(hdr
) &&
5217 (vd
= hdr
->b_l2hdr
.b_dev
->l2ad_vdev
) != NULL
) {
5218 devw
= hdr
->b_l2hdr
.b_dev
->l2ad_writing
;
5219 addr
= hdr
->b_l2hdr
.b_daddr
;
5221 * Lock out device removal.
5223 if (vdev_is_dead(vd
) ||
5224 !spa_config_tryenter(spa
, SCL_L2ARC
, vd
, RW_READER
))
5228 if (priority
== ZIO_PRIORITY_ASYNC_READ
)
5229 arc_hdr_set_flags(hdr
, ARC_FLAG_PRIO_ASYNC_READ
);
5231 arc_hdr_clear_flags(hdr
, ARC_FLAG_PRIO_ASYNC_READ
);
5233 if (hash_lock
!= NULL
)
5234 mutex_exit(hash_lock
);
5237 * At this point, we have a level 1 cache miss. Try again in
5238 * L2ARC if possible.
5240 ASSERT3U(HDR_GET_LSIZE(hdr
), ==, lsize
);
5242 DTRACE_PROBE4(arc__miss
, arc_buf_hdr_t
*, hdr
, blkptr_t
*, bp
,
5243 uint64_t, lsize
, zbookmark_phys_t
*, zb
);
5244 ARCSTAT_BUMP(arcstat_misses
);
5245 ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr
),
5246 demand
, prefetch
, !HDR_ISTYPE_METADATA(hdr
),
5247 data
, metadata
, misses
);
5249 if (vd
!= NULL
&& l2arc_ndev
!= 0 && !(l2arc_norw
&& devw
)) {
5251 * Read from the L2ARC if the following are true:
5252 * 1. The L2ARC vdev was previously cached.
5253 * 2. This buffer still has L2ARC metadata.
5254 * 3. This buffer isn't currently writing to the L2ARC.
5255 * 4. The L2ARC entry wasn't evicted, which may
5256 * also have invalidated the vdev.
5257 * 5. This isn't prefetch and l2arc_noprefetch is set.
5259 if (HDR_HAS_L2HDR(hdr
) &&
5260 !HDR_L2_WRITING(hdr
) && !HDR_L2_EVICTED(hdr
) &&
5261 !(l2arc_noprefetch
&& HDR_PREFETCH(hdr
))) {
5262 l2arc_read_callback_t
*cb
;
5264 DTRACE_PROBE1(l2arc__hit
, arc_buf_hdr_t
*, hdr
);
5265 ARCSTAT_BUMP(arcstat_l2_hits
);
5266 atomic_inc_32(&hdr
->b_l2hdr
.b_hits
);
5268 cb
= kmem_zalloc(sizeof (l2arc_read_callback_t
),
5270 cb
->l2rcb_hdr
= hdr
;
5273 cb
->l2rcb_flags
= zio_flags
;
5275 ASSERT(addr
>= VDEV_LABEL_START_SIZE
&&
5276 addr
+ lsize
< vd
->vdev_psize
-
5277 VDEV_LABEL_END_SIZE
);
5280 * l2arc read. The SCL_L2ARC lock will be
5281 * released by l2arc_read_done().
5282 * Issue a null zio if the underlying buffer
5283 * was squashed to zero size by compression.
5285 ASSERT3U(HDR_GET_COMPRESS(hdr
), !=,
5286 ZIO_COMPRESS_EMPTY
);
5287 rzio
= zio_read_phys(pio
, vd
, addr
,
5288 size
, hdr
->b_l1hdr
.b_pdata
,
5290 l2arc_read_done
, cb
, priority
,
5291 zio_flags
| ZIO_FLAG_DONT_CACHE
|
5293 ZIO_FLAG_DONT_PROPAGATE
|
5294 ZIO_FLAG_DONT_RETRY
, B_FALSE
);
5296 DTRACE_PROBE2(l2arc__read
, vdev_t
*, vd
,
5298 ARCSTAT_INCR(arcstat_l2_read_bytes
, size
);
5300 if (*arc_flags
& ARC_FLAG_NOWAIT
) {
5305 ASSERT(*arc_flags
& ARC_FLAG_WAIT
);
5306 if (zio_wait(rzio
) == 0)
5309 /* l2arc read error; goto zio_read() */
5311 DTRACE_PROBE1(l2arc__miss
,
5312 arc_buf_hdr_t
*, hdr
);
5313 ARCSTAT_BUMP(arcstat_l2_misses
);
5314 if (HDR_L2_WRITING(hdr
))
5315 ARCSTAT_BUMP(arcstat_l2_rw_clash
);
5316 spa_config_exit(spa
, SCL_L2ARC
, vd
);
5320 spa_config_exit(spa
, SCL_L2ARC
, vd
);
5321 if (l2arc_ndev
!= 0) {
5322 DTRACE_PROBE1(l2arc__miss
,
5323 arc_buf_hdr_t
*, hdr
);
5324 ARCSTAT_BUMP(arcstat_l2_misses
);
5328 rzio
= zio_read(pio
, spa
, bp
, hdr
->b_l1hdr
.b_pdata
, size
,
5329 arc_read_done
, hdr
, priority
, zio_flags
, zb
);
5331 if (*arc_flags
& ARC_FLAG_WAIT
) {
5332 rc
= zio_wait(rzio
);
5336 ASSERT(*arc_flags
& ARC_FLAG_NOWAIT
);
5341 spa_read_history_add(spa
, zb
, *arc_flags
);
5346 arc_add_prune_callback(arc_prune_func_t
*func
, void *private)
5350 p
= kmem_alloc(sizeof (*p
), KM_SLEEP
);
5352 p
->p_private
= private;
5353 list_link_init(&p
->p_node
);
5354 refcount_create(&p
->p_refcnt
);
5356 mutex_enter(&arc_prune_mtx
);
5357 refcount_add(&p
->p_refcnt
, &arc_prune_list
);
5358 list_insert_head(&arc_prune_list
, p
);
5359 mutex_exit(&arc_prune_mtx
);
5365 arc_remove_prune_callback(arc_prune_t
*p
)
5367 boolean_t wait
= B_FALSE
;
5368 mutex_enter(&arc_prune_mtx
);
5369 list_remove(&arc_prune_list
, p
);
5370 if (refcount_remove(&p
->p_refcnt
, &arc_prune_list
) > 0)
5372 mutex_exit(&arc_prune_mtx
);
5374 /* wait for arc_prune_task to finish */
5376 taskq_wait_outstanding(arc_prune_taskq
, 0);
5377 ASSERT0(refcount_count(&p
->p_refcnt
));
5378 refcount_destroy(&p
->p_refcnt
);
5379 kmem_free(p
, sizeof (*p
));
5383 * Notify the arc that a block was freed, and thus will never be used again.
5386 arc_freed(spa_t
*spa
, const blkptr_t
*bp
)
5389 kmutex_t
*hash_lock
;
5390 uint64_t guid
= spa_load_guid(spa
);
5392 ASSERT(!BP_IS_EMBEDDED(bp
));
5394 hdr
= buf_hash_find(guid
, bp
, &hash_lock
);
5399 * We might be trying to free a block that is still doing I/O
5400 * (i.e. prefetch) or has a reference (i.e. a dedup-ed,
5401 * dmu_sync-ed block). If this block is being prefetched, then it
5402 * would still have the ARC_FLAG_IO_IN_PROGRESS flag set on the hdr
5403 * until the I/O completes. A block may also have a reference if it is
5404 * part of a dedup-ed, dmu_synced write. The dmu_sync() function would
5405 * have written the new block to its final resting place on disk but
5406 * without the dedup flag set. This would have left the hdr in the MRU
5407 * state and discoverable. When the txg finally syncs it detects that
5408 * the block was overridden in open context and issues an override I/O.
5409 * Since this is a dedup block, the override I/O will determine if the
5410 * block is already in the DDT. If so, then it will replace the io_bp
5411 * with the bp from the DDT and allow the I/O to finish. When the I/O
5412 * reaches the done callback, dbuf_write_override_done, it will
5413 * check to see if the io_bp and io_bp_override are identical.
5414 * If they are not, then it indicates that the bp was replaced with
5415 * the bp in the DDT and the override bp is freed. This allows
5416 * us to arrive here with a reference on a block that is being
5417 * freed. So if we have an I/O in progress, or a reference to
5418 * this hdr, then we don't destroy the hdr.
5420 if (!HDR_HAS_L1HDR(hdr
) || (!HDR_IO_IN_PROGRESS(hdr
) &&
5421 refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
))) {
5422 arc_change_state(arc_anon
, hdr
, hash_lock
);
5423 arc_hdr_destroy(hdr
);
5424 mutex_exit(hash_lock
);
5426 mutex_exit(hash_lock
);
5432 * Release this buffer from the cache, making it an anonymous buffer. This
5433 * must be done after a read and prior to modifying the buffer contents.
5434 * If the buffer has more than one reference, we must make
5435 * a new hdr for the buffer.
5438 arc_release(arc_buf_t
*buf
, void *tag
)
5440 kmutex_t
*hash_lock
;
5442 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
5445 * It would be nice to assert that if its DMU metadata (level >
5446 * 0 || it's the dnode file), then it must be syncing context.
5447 * But we don't know that information at this level.
5450 mutex_enter(&buf
->b_evict_lock
);
5452 ASSERT(HDR_HAS_L1HDR(hdr
));
5455 * We don't grab the hash lock prior to this check, because if
5456 * the buffer's header is in the arc_anon state, it won't be
5457 * linked into the hash table.
5459 if (hdr
->b_l1hdr
.b_state
== arc_anon
) {
5460 mutex_exit(&buf
->b_evict_lock
);
5461 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
5462 ASSERT(!HDR_IN_HASH_TABLE(hdr
));
5463 ASSERT(!HDR_HAS_L2HDR(hdr
));
5464 ASSERT(HDR_EMPTY(hdr
));
5466 ASSERT3U(hdr
->b_l1hdr
.b_bufcnt
, ==, 1);
5467 ASSERT3S(refcount_count(&hdr
->b_l1hdr
.b_refcnt
), ==, 1);
5468 ASSERT(!list_link_active(&hdr
->b_l1hdr
.b_arc_node
));
5470 hdr
->b_l1hdr
.b_arc_access
= 0;
5473 * If the buf is being overridden then it may already
5474 * have a hdr that is not empty.
5476 buf_discard_identity(hdr
);
5482 hash_lock
= HDR_LOCK(hdr
);
5483 mutex_enter(hash_lock
);
5486 * This assignment is only valid as long as the hash_lock is
5487 * held, we must be careful not to reference state or the
5488 * b_state field after dropping the lock.
5490 state
= hdr
->b_l1hdr
.b_state
;
5491 ASSERT3P(hash_lock
, ==, HDR_LOCK(hdr
));
5492 ASSERT3P(state
, !=, arc_anon
);
5494 /* this buffer is not on any list */
5495 ASSERT3S(refcount_count(&hdr
->b_l1hdr
.b_refcnt
), >, 0);
5497 if (HDR_HAS_L2HDR(hdr
)) {
5498 mutex_enter(&hdr
->b_l2hdr
.b_dev
->l2ad_mtx
);
5501 * We have to recheck this conditional again now that
5502 * we're holding the l2ad_mtx to prevent a race with
5503 * another thread which might be concurrently calling
5504 * l2arc_evict(). In that case, l2arc_evict() might have
5505 * destroyed the header's L2 portion as we were waiting
5506 * to acquire the l2ad_mtx.
5508 if (HDR_HAS_L2HDR(hdr
))
5509 arc_hdr_l2hdr_destroy(hdr
);
5511 mutex_exit(&hdr
->b_l2hdr
.b_dev
->l2ad_mtx
);
5515 * Do we have more than one buf?
5517 if (hdr
->b_l1hdr
.b_bufcnt
> 1) {
5518 arc_buf_hdr_t
*nhdr
;
5519 uint64_t spa
= hdr
->b_spa
;
5520 uint64_t psize
= HDR_GET_PSIZE(hdr
);
5521 uint64_t lsize
= HDR_GET_LSIZE(hdr
);
5522 enum zio_compress compress
= HDR_GET_COMPRESS(hdr
);
5523 arc_buf_contents_t type
= arc_buf_type(hdr
);
5524 arc_buf_t
*lastbuf
= NULL
;
5525 VERIFY3U(hdr
->b_type
, ==, type
);
5527 ASSERT(hdr
->b_l1hdr
.b_buf
!= buf
|| buf
->b_next
!= NULL
);
5528 (void) remove_reference(hdr
, hash_lock
, tag
);
5530 if (arc_buf_is_shared(buf
) && !ARC_BUF_COMPRESSED(buf
)) {
5531 ASSERT3P(hdr
->b_l1hdr
.b_buf
, !=, buf
);
5532 ASSERT(ARC_BUF_LAST(buf
));
5536 * Pull the data off of this hdr and attach it to
5537 * a new anonymous hdr. Also find the last buffer
5538 * in the hdr's buffer list.
5540 lastbuf
= arc_buf_remove(hdr
, buf
);
5541 ASSERT3P(lastbuf
, !=, NULL
);
5544 * If the current arc_buf_t and the hdr are sharing their data
5545 * buffer, then we must stop sharing that block.
5547 if (arc_buf_is_shared(buf
)) {
5548 ASSERT3P(hdr
->b_l1hdr
.b_buf
, !=, buf
);
5549 VERIFY(!arc_buf_is_shared(lastbuf
));
5552 * First, sever the block sharing relationship between
5553 * buf and the arc_buf_hdr_t. Then, setup a new
5554 * block sharing relationship with the last buffer
5555 * on the arc_buf_t list.
5557 arc_unshare_buf(hdr
, buf
);
5560 * Now we need to recreate the hdr's b_pdata. Since we
5561 * have lastbuf handy, we try to share with it, but if
5562 * we can't then we allocate a new b_pdata and copy the
5563 * data from buf into it.
5565 if (arc_can_share(hdr
, lastbuf
)) {
5566 arc_share_buf(hdr
, lastbuf
);
5568 arc_hdr_alloc_pdata(hdr
);
5569 bcopy(buf
->b_data
, hdr
->b_l1hdr
.b_pdata
, psize
);
5571 VERIFY3P(lastbuf
->b_data
, !=, NULL
);
5572 } else if (HDR_SHARED_DATA(hdr
)) {
5574 * Uncompressed shared buffers are always at the end
5575 * of the list. Compressed buffers don't have the
5576 * same requirements. This makes it hard to
5577 * simply assert that the lastbuf is shared so
5578 * we rely on the hdr's compression flags to determine
5579 * if we have a compressed, shared buffer.
5581 ASSERT(arc_buf_is_shared(lastbuf
) ||
5582 HDR_GET_COMPRESS(hdr
) != ZIO_COMPRESS_OFF
);
5583 ASSERT(!ARC_BUF_SHARED(buf
));
5585 ASSERT3P(hdr
->b_l1hdr
.b_pdata
, !=, NULL
);
5586 ASSERT3P(state
, !=, arc_l2c_only
);
5588 (void) refcount_remove_many(&state
->arcs_size
,
5589 arc_buf_size(buf
), buf
);
5591 if (refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
)) {
5592 ASSERT3P(state
, !=, arc_l2c_only
);
5593 (void) refcount_remove_many(&state
->arcs_esize
[type
],
5594 arc_buf_size(buf
), buf
);
5597 hdr
->b_l1hdr
.b_bufcnt
-= 1;
5598 arc_cksum_verify(buf
);
5599 arc_buf_unwatch(buf
);
5601 mutex_exit(hash_lock
);
5604 * Allocate a new hdr. The new hdr will contain a b_pdata
5605 * buffer which will be freed in arc_write().
5607 nhdr
= arc_hdr_alloc(spa
, psize
, lsize
, compress
, type
);
5608 ASSERT3P(nhdr
->b_l1hdr
.b_buf
, ==, NULL
);
5609 ASSERT0(nhdr
->b_l1hdr
.b_bufcnt
);
5610 ASSERT0(refcount_count(&nhdr
->b_l1hdr
.b_refcnt
));
5611 VERIFY3U(nhdr
->b_type
, ==, type
);
5612 ASSERT(!HDR_SHARED_DATA(nhdr
));
5614 nhdr
->b_l1hdr
.b_buf
= buf
;
5615 nhdr
->b_l1hdr
.b_bufcnt
= 1;
5616 nhdr
->b_l1hdr
.b_mru_hits
= 0;
5617 nhdr
->b_l1hdr
.b_mru_ghost_hits
= 0;
5618 nhdr
->b_l1hdr
.b_mfu_hits
= 0;
5619 nhdr
->b_l1hdr
.b_mfu_ghost_hits
= 0;
5620 nhdr
->b_l1hdr
.b_l2_hits
= 0;
5621 (void) refcount_add(&nhdr
->b_l1hdr
.b_refcnt
, tag
);
5624 mutex_exit(&buf
->b_evict_lock
);
5625 (void) refcount_add_many(&arc_anon
->arcs_size
,
5626 HDR_GET_LSIZE(nhdr
), buf
);
5628 mutex_exit(&buf
->b_evict_lock
);
5629 ASSERT(refcount_count(&hdr
->b_l1hdr
.b_refcnt
) == 1);
5630 /* protected by hash lock, or hdr is on arc_anon */
5631 ASSERT(!multilist_link_active(&hdr
->b_l1hdr
.b_arc_node
));
5632 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
5633 hdr
->b_l1hdr
.b_mru_hits
= 0;
5634 hdr
->b_l1hdr
.b_mru_ghost_hits
= 0;
5635 hdr
->b_l1hdr
.b_mfu_hits
= 0;
5636 hdr
->b_l1hdr
.b_mfu_ghost_hits
= 0;
5637 hdr
->b_l1hdr
.b_l2_hits
= 0;
5638 arc_change_state(arc_anon
, hdr
, hash_lock
);
5639 hdr
->b_l1hdr
.b_arc_access
= 0;
5640 mutex_exit(hash_lock
);
5642 buf_discard_identity(hdr
);
5648 arc_released(arc_buf_t
*buf
)
5652 mutex_enter(&buf
->b_evict_lock
);
5653 released
= (buf
->b_data
!= NULL
&&
5654 buf
->b_hdr
->b_l1hdr
.b_state
== arc_anon
);
5655 mutex_exit(&buf
->b_evict_lock
);
5661 arc_referenced(arc_buf_t
*buf
)
5665 mutex_enter(&buf
->b_evict_lock
);
5666 referenced
= (refcount_count(&buf
->b_hdr
->b_l1hdr
.b_refcnt
));
5667 mutex_exit(&buf
->b_evict_lock
);
5668 return (referenced
);
5673 arc_write_ready(zio_t
*zio
)
5675 arc_write_callback_t
*callback
= zio
->io_private
;
5676 arc_buf_t
*buf
= callback
->awcb_buf
;
5677 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
5678 uint64_t psize
= BP_IS_HOLE(zio
->io_bp
) ? 0 : BP_GET_PSIZE(zio
->io_bp
);
5679 enum zio_compress compress
;
5681 ASSERT(HDR_HAS_L1HDR(hdr
));
5682 ASSERT(!refcount_is_zero(&buf
->b_hdr
->b_l1hdr
.b_refcnt
));
5683 ASSERT(hdr
->b_l1hdr
.b_bufcnt
> 0);
5686 * If we're reexecuting this zio because the pool suspended, then
5687 * cleanup any state that was previously set the first time the
5688 * callback was invoked.
5690 if (zio
->io_flags
& ZIO_FLAG_REEXECUTED
) {
5691 arc_cksum_free(hdr
);
5692 arc_buf_unwatch(buf
);
5693 if (hdr
->b_l1hdr
.b_pdata
!= NULL
) {
5694 if (arc_buf_is_shared(buf
)) {
5695 arc_unshare_buf(hdr
, buf
);
5697 arc_hdr_free_pdata(hdr
);
5701 ASSERT3P(hdr
->b_l1hdr
.b_pdata
, ==, NULL
);
5702 ASSERT(!HDR_SHARED_DATA(hdr
));
5703 ASSERT(!arc_buf_is_shared(buf
));
5705 callback
->awcb_ready(zio
, buf
, callback
->awcb_private
);
5707 if (HDR_IO_IN_PROGRESS(hdr
))
5708 ASSERT(zio
->io_flags
& ZIO_FLAG_REEXECUTED
);
5710 arc_cksum_compute(buf
);
5711 arc_hdr_set_flags(hdr
, ARC_FLAG_IO_IN_PROGRESS
);
5713 if (BP_IS_HOLE(zio
->io_bp
) || BP_IS_EMBEDDED(zio
->io_bp
)) {
5714 compress
= ZIO_COMPRESS_OFF
;
5716 ASSERT3U(HDR_GET_LSIZE(hdr
), ==, BP_GET_LSIZE(zio
->io_bp
));
5717 compress
= BP_GET_COMPRESS(zio
->io_bp
);
5719 HDR_SET_PSIZE(hdr
, psize
);
5720 arc_hdr_set_compress(hdr
, compress
);
5723 * If the hdr is compressed, then copy the compressed
5724 * zio contents into arc_buf_hdr_t. Otherwise, copy the original
5725 * data buf into the hdr. Ideally, we would like to always copy the
5726 * io_data into b_pdata but the user may have disabled compressed
5727 * arc thus the on-disk block may or may not match what we maintain
5728 * in the hdr's b_pdata field.
5730 if (HDR_GET_COMPRESS(hdr
) != ZIO_COMPRESS_OFF
&&
5731 !ARC_BUF_COMPRESSED(buf
)) {
5732 ASSERT3U(BP_GET_COMPRESS(zio
->io_bp
), !=, ZIO_COMPRESS_OFF
);
5733 ASSERT3U(psize
, >, 0);
5734 arc_hdr_alloc_pdata(hdr
);
5735 bcopy(zio
->io_data
, hdr
->b_l1hdr
.b_pdata
, psize
);
5737 ASSERT3P(buf
->b_data
, ==, zio
->io_orig_data
);
5738 ASSERT3U(zio
->io_orig_size
, ==, arc_buf_size(buf
));
5739 ASSERT3U(hdr
->b_l1hdr
.b_bufcnt
, ==, 1);
5742 * This hdr is not compressed so we're able to share
5743 * the arc_buf_t data buffer with the hdr.
5745 arc_share_buf(hdr
, buf
);
5746 ASSERT0(bcmp(zio
->io_orig_data
, hdr
->b_l1hdr
.b_pdata
,
5747 HDR_GET_LSIZE(hdr
)));
5749 arc_hdr_verify(hdr
, zio
->io_bp
);
5753 arc_write_children_ready(zio_t
*zio
)
5755 arc_write_callback_t
*callback
= zio
->io_private
;
5756 arc_buf_t
*buf
= callback
->awcb_buf
;
5758 callback
->awcb_children_ready(zio
, buf
, callback
->awcb_private
);
5762 * The SPA calls this callback for each physical write that happens on behalf
5763 * of a logical write. See the comment in dbuf_write_physdone() for details.
5766 arc_write_physdone(zio_t
*zio
)
5768 arc_write_callback_t
*cb
= zio
->io_private
;
5769 if (cb
->awcb_physdone
!= NULL
)
5770 cb
->awcb_physdone(zio
, cb
->awcb_buf
, cb
->awcb_private
);
5774 arc_write_done(zio_t
*zio
)
5776 arc_write_callback_t
*callback
= zio
->io_private
;
5777 arc_buf_t
*buf
= callback
->awcb_buf
;
5778 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
5780 ASSERT3P(hdr
->b_l1hdr
.b_acb
, ==, NULL
);
5782 if (zio
->io_error
== 0) {
5783 arc_hdr_verify(hdr
, zio
->io_bp
);
5785 if (BP_IS_HOLE(zio
->io_bp
) || BP_IS_EMBEDDED(zio
->io_bp
)) {
5786 buf_discard_identity(hdr
);
5788 hdr
->b_dva
= *BP_IDENTITY(zio
->io_bp
);
5789 hdr
->b_birth
= BP_PHYSICAL_BIRTH(zio
->io_bp
);
5792 ASSERT(HDR_EMPTY(hdr
));
5796 * If the block to be written was all-zero or compressed enough to be
5797 * embedded in the BP, no write was performed so there will be no
5798 * dva/birth/checksum. The buffer must therefore remain anonymous
5801 if (!HDR_EMPTY(hdr
)) {
5802 arc_buf_hdr_t
*exists
;
5803 kmutex_t
*hash_lock
;
5805 ASSERT3U(zio
->io_error
, ==, 0);
5807 arc_cksum_verify(buf
);
5809 exists
= buf_hash_insert(hdr
, &hash_lock
);
5810 if (exists
!= NULL
) {
5812 * This can only happen if we overwrite for
5813 * sync-to-convergence, because we remove
5814 * buffers from the hash table when we arc_free().
5816 if (zio
->io_flags
& ZIO_FLAG_IO_REWRITE
) {
5817 if (!BP_EQUAL(&zio
->io_bp_orig
, zio
->io_bp
))
5818 panic("bad overwrite, hdr=%p exists=%p",
5819 (void *)hdr
, (void *)exists
);
5820 ASSERT(refcount_is_zero(
5821 &exists
->b_l1hdr
.b_refcnt
));
5822 arc_change_state(arc_anon
, exists
, hash_lock
);
5823 mutex_exit(hash_lock
);
5824 arc_hdr_destroy(exists
);
5825 exists
= buf_hash_insert(hdr
, &hash_lock
);
5826 ASSERT3P(exists
, ==, NULL
);
5827 } else if (zio
->io_flags
& ZIO_FLAG_NOPWRITE
) {
5829 ASSERT(zio
->io_prop
.zp_nopwrite
);
5830 if (!BP_EQUAL(&zio
->io_bp_orig
, zio
->io_bp
))
5831 panic("bad nopwrite, hdr=%p exists=%p",
5832 (void *)hdr
, (void *)exists
);
5835 ASSERT(hdr
->b_l1hdr
.b_bufcnt
== 1);
5836 ASSERT(hdr
->b_l1hdr
.b_state
== arc_anon
);
5837 ASSERT(BP_GET_DEDUP(zio
->io_bp
));
5838 ASSERT(BP_GET_LEVEL(zio
->io_bp
) == 0);
5841 arc_hdr_clear_flags(hdr
, ARC_FLAG_IO_IN_PROGRESS
);
5842 /* if it's not anon, we are doing a scrub */
5843 if (exists
== NULL
&& hdr
->b_l1hdr
.b_state
== arc_anon
)
5844 arc_access(hdr
, hash_lock
);
5845 mutex_exit(hash_lock
);
5847 arc_hdr_clear_flags(hdr
, ARC_FLAG_IO_IN_PROGRESS
);
5850 ASSERT(!refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
));
5851 callback
->awcb_done(zio
, buf
, callback
->awcb_private
);
5853 kmem_free(callback
, sizeof (arc_write_callback_t
));
5857 arc_write(zio_t
*pio
, spa_t
*spa
, uint64_t txg
,
5858 blkptr_t
*bp
, arc_buf_t
*buf
, boolean_t l2arc
,
5859 const zio_prop_t
*zp
, arc_done_func_t
*ready
,
5860 arc_done_func_t
*children_ready
, arc_done_func_t
*physdone
,
5861 arc_done_func_t
*done
, void *private, zio_priority_t priority
,
5862 int zio_flags
, const zbookmark_phys_t
*zb
)
5864 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
5865 arc_write_callback_t
*callback
;
5868 ASSERT3P(ready
, !=, NULL
);
5869 ASSERT3P(done
, !=, NULL
);
5870 ASSERT(!HDR_IO_ERROR(hdr
));
5871 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
5872 ASSERT3P(hdr
->b_l1hdr
.b_acb
, ==, NULL
);
5873 ASSERT3U(hdr
->b_l1hdr
.b_bufcnt
, >, 0);
5875 arc_hdr_set_flags(hdr
, ARC_FLAG_L2CACHE
);
5876 if (ARC_BUF_COMPRESSED(buf
)) {
5877 ASSERT3U(zp
->zp_compress
, !=, ZIO_COMPRESS_OFF
);
5878 zio_flags
|= ZIO_FLAG_RAW
;
5880 callback
= kmem_zalloc(sizeof (arc_write_callback_t
), KM_SLEEP
);
5881 callback
->awcb_ready
= ready
;
5882 callback
->awcb_children_ready
= children_ready
;
5883 callback
->awcb_physdone
= physdone
;
5884 callback
->awcb_done
= done
;
5885 callback
->awcb_private
= private;
5886 callback
->awcb_buf
= buf
;
5889 * The hdr's b_pdata is now stale, free it now. A new data block
5890 * will be allocated when the zio pipeline calls arc_write_ready().
5892 if (hdr
->b_l1hdr
.b_pdata
!= NULL
) {
5894 * If the buf is currently sharing the data block with
5895 * the hdr then we need to break that relationship here.
5896 * The hdr will remain with a NULL data pointer and the
5897 * buf will take sole ownership of the block.
5899 if (arc_buf_is_shared(buf
)) {
5900 arc_unshare_buf(hdr
, buf
);
5902 arc_hdr_free_pdata(hdr
);
5904 VERIFY3P(buf
->b_data
, !=, NULL
);
5905 arc_hdr_set_compress(hdr
, ZIO_COMPRESS_OFF
);
5907 ASSERT(!arc_buf_is_shared(buf
));
5908 ASSERT3P(hdr
->b_l1hdr
.b_pdata
, ==, NULL
);
5910 zio
= zio_write(pio
, spa
, txg
, bp
, buf
->b_data
,
5911 HDR_GET_LSIZE(hdr
), arc_buf_size(buf
), zp
,
5913 (children_ready
!= NULL
) ? arc_write_children_ready
: NULL
,
5914 arc_write_physdone
, arc_write_done
, callback
,
5915 priority
, zio_flags
, zb
);
5921 arc_memory_throttle(uint64_t reserve
, uint64_t txg
)
5924 uint64_t available_memory
= ptob(freemem
);
5925 static uint64_t page_load
= 0;
5926 static uint64_t last_txg
= 0;
5928 pgcnt_t minfree
= btop(arc_sys_free
/ 4);
5933 MIN(available_memory
, vmem_size(heap_arena
, VMEM_FREE
));
5936 if (available_memory
> arc_all_memory() * arc_lotsfree_percent
/ 100)
5939 if (txg
> last_txg
) {
5944 * If we are in pageout, we know that memory is already tight,
5945 * the arc is already going to be evicting, so we just want to
5946 * continue to let page writes occur as quickly as possible.
5948 if (current_is_kswapd()) {
5949 if (page_load
> MAX(ptob(minfree
), available_memory
) / 4) {
5950 DMU_TX_STAT_BUMP(dmu_tx_memory_reclaim
);
5951 return (SET_ERROR(ERESTART
));
5953 /* Note: reserve is inflated, so we deflate */
5954 page_load
+= reserve
/ 8;
5956 } else if (page_load
> 0 && arc_reclaim_needed()) {
5957 /* memory is low, delay before restarting */
5958 ARCSTAT_INCR(arcstat_memory_throttle_count
, 1);
5959 DMU_TX_STAT_BUMP(dmu_tx_memory_reclaim
);
5960 return (SET_ERROR(EAGAIN
));
5968 arc_tempreserve_clear(uint64_t reserve
)
5970 atomic_add_64(&arc_tempreserve
, -reserve
);
5971 ASSERT((int64_t)arc_tempreserve
>= 0);
5975 arc_tempreserve_space(uint64_t reserve
, uint64_t txg
)
5981 reserve
> arc_c
/4 &&
5982 reserve
* 4 > (2ULL << SPA_MAXBLOCKSHIFT
))
5983 arc_c
= MIN(arc_c_max
, reserve
* 4);
5986 * Throttle when the calculated memory footprint for the TXG
5987 * exceeds the target ARC size.
5989 if (reserve
> arc_c
) {
5990 DMU_TX_STAT_BUMP(dmu_tx_memory_reserve
);
5991 return (SET_ERROR(ERESTART
));
5995 * Don't count loaned bufs as in flight dirty data to prevent long
5996 * network delays from blocking transactions that are ready to be
5997 * assigned to a txg.
5999 anon_size
= MAX((int64_t)(refcount_count(&arc_anon
->arcs_size
) -
6000 arc_loaned_bytes
), 0);
6003 * Writes will, almost always, require additional memory allocations
6004 * in order to compress/encrypt/etc the data. We therefore need to
6005 * make sure that there is sufficient available memory for this.
6007 error
= arc_memory_throttle(reserve
, txg
);
6012 * Throttle writes when the amount of dirty data in the cache
6013 * gets too large. We try to keep the cache less than half full
6014 * of dirty blocks so that our sync times don't grow too large.
6015 * Note: if two requests come in concurrently, we might let them
6016 * both succeed, when one of them should fail. Not a huge deal.
6019 if (reserve
+ arc_tempreserve
+ anon_size
> arc_c
/ 2 &&
6020 anon_size
> arc_c
/ 4) {
6021 uint64_t meta_esize
=
6022 refcount_count(&arc_anon
->arcs_esize
[ARC_BUFC_METADATA
]);
6023 uint64_t data_esize
=
6024 refcount_count(&arc_anon
->arcs_esize
[ARC_BUFC_DATA
]);
6025 dprintf("failing, arc_tempreserve=%lluK anon_meta=%lluK "
6026 "anon_data=%lluK tempreserve=%lluK arc_c=%lluK\n",
6027 arc_tempreserve
>> 10, meta_esize
>> 10,
6028 data_esize
>> 10, reserve
>> 10, arc_c
>> 10);
6029 DMU_TX_STAT_BUMP(dmu_tx_dirty_throttle
);
6030 return (SET_ERROR(ERESTART
));
6032 atomic_add_64(&arc_tempreserve
, reserve
);
6037 arc_kstat_update_state(arc_state_t
*state
, kstat_named_t
*size
,
6038 kstat_named_t
*evict_data
, kstat_named_t
*evict_metadata
)
6040 size
->value
.ui64
= refcount_count(&state
->arcs_size
);
6041 evict_data
->value
.ui64
=
6042 refcount_count(&state
->arcs_esize
[ARC_BUFC_DATA
]);
6043 evict_metadata
->value
.ui64
=
6044 refcount_count(&state
->arcs_esize
[ARC_BUFC_METADATA
]);
6048 arc_kstat_update(kstat_t
*ksp
, int rw
)
6050 arc_stats_t
*as
= ksp
->ks_data
;
6052 if (rw
== KSTAT_WRITE
) {
6055 arc_kstat_update_state(arc_anon
,
6056 &as
->arcstat_anon_size
,
6057 &as
->arcstat_anon_evictable_data
,
6058 &as
->arcstat_anon_evictable_metadata
);
6059 arc_kstat_update_state(arc_mru
,
6060 &as
->arcstat_mru_size
,
6061 &as
->arcstat_mru_evictable_data
,
6062 &as
->arcstat_mru_evictable_metadata
);
6063 arc_kstat_update_state(arc_mru_ghost
,
6064 &as
->arcstat_mru_ghost_size
,
6065 &as
->arcstat_mru_ghost_evictable_data
,
6066 &as
->arcstat_mru_ghost_evictable_metadata
);
6067 arc_kstat_update_state(arc_mfu
,
6068 &as
->arcstat_mfu_size
,
6069 &as
->arcstat_mfu_evictable_data
,
6070 &as
->arcstat_mfu_evictable_metadata
);
6071 arc_kstat_update_state(arc_mfu_ghost
,
6072 &as
->arcstat_mfu_ghost_size
,
6073 &as
->arcstat_mfu_ghost_evictable_data
,
6074 &as
->arcstat_mfu_ghost_evictable_metadata
);
6081 * This function *must* return indices evenly distributed between all
6082 * sublists of the multilist. This is needed due to how the ARC eviction
6083 * code is laid out; arc_evict_state() assumes ARC buffers are evenly
6084 * distributed between all sublists and uses this assumption when
6085 * deciding which sublist to evict from and how much to evict from it.
6088 arc_state_multilist_index_func(multilist_t
*ml
, void *obj
)
6090 arc_buf_hdr_t
*hdr
= obj
;
6093 * We rely on b_dva to generate evenly distributed index
6094 * numbers using buf_hash below. So, as an added precaution,
6095 * let's make sure we never add empty buffers to the arc lists.
6097 ASSERT(!HDR_EMPTY(hdr
));
6100 * The assumption here, is the hash value for a given
6101 * arc_buf_hdr_t will remain constant throughout its lifetime
6102 * (i.e. its b_spa, b_dva, and b_birth fields don't change).
6103 * Thus, we don't need to store the header's sublist index
6104 * on insertion, as this index can be recalculated on removal.
6106 * Also, the low order bits of the hash value are thought to be
6107 * distributed evenly. Otherwise, in the case that the multilist
6108 * has a power of two number of sublists, each sublists' usage
6109 * would not be evenly distributed.
6111 return (buf_hash(hdr
->b_spa
, &hdr
->b_dva
, hdr
->b_birth
) %
6112 multilist_get_num_sublists(ml
));
6116 * Called during module initialization and periodically thereafter to
6117 * apply reasonable changes to the exposed performance tunings. Non-zero
6118 * zfs_* values which differ from the currently set values will be applied.
6121 arc_tuning_update(void)
6123 uint64_t percent
, allmem
= arc_all_memory();
6125 /* Valid range: 64M - <all physical memory> */
6126 if ((zfs_arc_max
) && (zfs_arc_max
!= arc_c_max
) &&
6127 (zfs_arc_max
> 64 << 20) && (zfs_arc_max
< allmem
) &&
6128 (zfs_arc_max
> arc_c_min
)) {
6129 arc_c_max
= zfs_arc_max
;
6131 arc_p
= (arc_c
>> 1);
6132 /* Valid range of arc_meta_limit: arc_meta_min - arc_c_max */
6133 percent
= MIN(zfs_arc_meta_limit_percent
, 100);
6134 arc_meta_limit
= MAX(arc_meta_min
, (percent
* arc_c_max
) / 100);
6135 percent
= MIN(zfs_arc_dnode_limit_percent
, 100);
6136 arc_dnode_limit
= (percent
* arc_meta_limit
) / 100;
6139 /* Valid range: 32M - <arc_c_max> */
6140 if ((zfs_arc_min
) && (zfs_arc_min
!= arc_c_min
) &&
6141 (zfs_arc_min
>= 2ULL << SPA_MAXBLOCKSHIFT
) &&
6142 (zfs_arc_min
<= arc_c_max
)) {
6143 arc_c_min
= zfs_arc_min
;
6144 arc_c
= MAX(arc_c
, arc_c_min
);
6147 /* Valid range: 16M - <arc_c_max> */
6148 if ((zfs_arc_meta_min
) && (zfs_arc_meta_min
!= arc_meta_min
) &&
6149 (zfs_arc_meta_min
>= 1ULL << SPA_MAXBLOCKSHIFT
) &&
6150 (zfs_arc_meta_min
<= arc_c_max
)) {
6151 arc_meta_min
= zfs_arc_meta_min
;
6152 arc_meta_limit
= MAX(arc_meta_limit
, arc_meta_min
);
6153 arc_dnode_limit
= arc_meta_limit
/ 10;
6156 /* Valid range: <arc_meta_min> - <arc_c_max> */
6157 if ((zfs_arc_meta_limit
) && (zfs_arc_meta_limit
!= arc_meta_limit
) &&
6158 (zfs_arc_meta_limit
>= zfs_arc_meta_min
) &&
6159 (zfs_arc_meta_limit
<= arc_c_max
))
6160 arc_meta_limit
= zfs_arc_meta_limit
;
6162 /* Valid range: <arc_meta_min> - <arc_c_max> */
6163 if ((zfs_arc_dnode_limit
) && (zfs_arc_dnode_limit
!= arc_dnode_limit
) &&
6164 (zfs_arc_dnode_limit
>= zfs_arc_meta_min
) &&
6165 (zfs_arc_dnode_limit
<= arc_c_max
))
6166 arc_dnode_limit
= zfs_arc_dnode_limit
;
6168 /* Valid range: 1 - N */
6169 if (zfs_arc_grow_retry
)
6170 arc_grow_retry
= zfs_arc_grow_retry
;
6172 /* Valid range: 1 - N */
6173 if (zfs_arc_shrink_shift
) {
6174 arc_shrink_shift
= zfs_arc_shrink_shift
;
6175 arc_no_grow_shift
= MIN(arc_no_grow_shift
, arc_shrink_shift
-1);
6178 /* Valid range: 1 - N */
6179 if (zfs_arc_p_min_shift
)
6180 arc_p_min_shift
= zfs_arc_p_min_shift
;
6182 /* Valid range: 1 - N ticks */
6183 if (zfs_arc_min_prefetch_lifespan
)
6184 arc_min_prefetch_lifespan
= zfs_arc_min_prefetch_lifespan
;
6186 /* Valid range: 0 - 100 */
6187 if ((zfs_arc_lotsfree_percent
>= 0) &&
6188 (zfs_arc_lotsfree_percent
<= 100))
6189 arc_lotsfree_percent
= zfs_arc_lotsfree_percent
;
6191 /* Valid range: 0 - <all physical memory> */
6192 if ((zfs_arc_sys_free
) && (zfs_arc_sys_free
!= arc_sys_free
))
6193 arc_sys_free
= MIN(MAX(zfs_arc_sys_free
, 0), allmem
);
6198 arc_state_init(void)
6200 arc_anon
= &ARC_anon
;
6202 arc_mru_ghost
= &ARC_mru_ghost
;
6204 arc_mfu_ghost
= &ARC_mfu_ghost
;
6205 arc_l2c_only
= &ARC_l2c_only
;
6207 multilist_create(&arc_mru
->arcs_list
[ARC_BUFC_METADATA
],
6208 sizeof (arc_buf_hdr_t
),
6209 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
6210 zfs_arc_num_sublists_per_state
, arc_state_multilist_index_func
);
6211 multilist_create(&arc_mru
->arcs_list
[ARC_BUFC_DATA
],
6212 sizeof (arc_buf_hdr_t
),
6213 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
6214 zfs_arc_num_sublists_per_state
, arc_state_multilist_index_func
);
6215 multilist_create(&arc_mru_ghost
->arcs_list
[ARC_BUFC_METADATA
],
6216 sizeof (arc_buf_hdr_t
),
6217 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
6218 zfs_arc_num_sublists_per_state
, arc_state_multilist_index_func
);
6219 multilist_create(&arc_mru_ghost
->arcs_list
[ARC_BUFC_DATA
],
6220 sizeof (arc_buf_hdr_t
),
6221 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
6222 zfs_arc_num_sublists_per_state
, arc_state_multilist_index_func
);
6223 multilist_create(&arc_mfu
->arcs_list
[ARC_BUFC_METADATA
],
6224 sizeof (arc_buf_hdr_t
),
6225 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
6226 zfs_arc_num_sublists_per_state
, arc_state_multilist_index_func
);
6227 multilist_create(&arc_mfu
->arcs_list
[ARC_BUFC_DATA
],
6228 sizeof (arc_buf_hdr_t
),
6229 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
6230 zfs_arc_num_sublists_per_state
, arc_state_multilist_index_func
);
6231 multilist_create(&arc_mfu_ghost
->arcs_list
[ARC_BUFC_METADATA
],
6232 sizeof (arc_buf_hdr_t
),
6233 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
6234 zfs_arc_num_sublists_per_state
, arc_state_multilist_index_func
);
6235 multilist_create(&arc_mfu_ghost
->arcs_list
[ARC_BUFC_DATA
],
6236 sizeof (arc_buf_hdr_t
),
6237 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
6238 zfs_arc_num_sublists_per_state
, arc_state_multilist_index_func
);
6239 multilist_create(&arc_l2c_only
->arcs_list
[ARC_BUFC_METADATA
],
6240 sizeof (arc_buf_hdr_t
),
6241 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
6242 zfs_arc_num_sublists_per_state
, arc_state_multilist_index_func
);
6243 multilist_create(&arc_l2c_only
->arcs_list
[ARC_BUFC_DATA
],
6244 sizeof (arc_buf_hdr_t
),
6245 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
6246 zfs_arc_num_sublists_per_state
, arc_state_multilist_index_func
);
6248 refcount_create(&arc_anon
->arcs_esize
[ARC_BUFC_METADATA
]);
6249 refcount_create(&arc_anon
->arcs_esize
[ARC_BUFC_DATA
]);
6250 refcount_create(&arc_mru
->arcs_esize
[ARC_BUFC_METADATA
]);
6251 refcount_create(&arc_mru
->arcs_esize
[ARC_BUFC_DATA
]);
6252 refcount_create(&arc_mru_ghost
->arcs_esize
[ARC_BUFC_METADATA
]);
6253 refcount_create(&arc_mru_ghost
->arcs_esize
[ARC_BUFC_DATA
]);
6254 refcount_create(&arc_mfu
->arcs_esize
[ARC_BUFC_METADATA
]);
6255 refcount_create(&arc_mfu
->arcs_esize
[ARC_BUFC_DATA
]);
6256 refcount_create(&arc_mfu_ghost
->arcs_esize
[ARC_BUFC_METADATA
]);
6257 refcount_create(&arc_mfu_ghost
->arcs_esize
[ARC_BUFC_DATA
]);
6258 refcount_create(&arc_l2c_only
->arcs_esize
[ARC_BUFC_METADATA
]);
6259 refcount_create(&arc_l2c_only
->arcs_esize
[ARC_BUFC_DATA
]);
6261 refcount_create(&arc_anon
->arcs_size
);
6262 refcount_create(&arc_mru
->arcs_size
);
6263 refcount_create(&arc_mru_ghost
->arcs_size
);
6264 refcount_create(&arc_mfu
->arcs_size
);
6265 refcount_create(&arc_mfu_ghost
->arcs_size
);
6266 refcount_create(&arc_l2c_only
->arcs_size
);
6268 arc_anon
->arcs_state
= ARC_STATE_ANON
;
6269 arc_mru
->arcs_state
= ARC_STATE_MRU
;
6270 arc_mru_ghost
->arcs_state
= ARC_STATE_MRU_GHOST
;
6271 arc_mfu
->arcs_state
= ARC_STATE_MFU
;
6272 arc_mfu_ghost
->arcs_state
= ARC_STATE_MFU_GHOST
;
6273 arc_l2c_only
->arcs_state
= ARC_STATE_L2C_ONLY
;
6277 arc_state_fini(void)
6279 refcount_destroy(&arc_anon
->arcs_esize
[ARC_BUFC_METADATA
]);
6280 refcount_destroy(&arc_anon
->arcs_esize
[ARC_BUFC_DATA
]);
6281 refcount_destroy(&arc_mru
->arcs_esize
[ARC_BUFC_METADATA
]);
6282 refcount_destroy(&arc_mru
->arcs_esize
[ARC_BUFC_DATA
]);
6283 refcount_destroy(&arc_mru_ghost
->arcs_esize
[ARC_BUFC_METADATA
]);
6284 refcount_destroy(&arc_mru_ghost
->arcs_esize
[ARC_BUFC_DATA
]);
6285 refcount_destroy(&arc_mfu
->arcs_esize
[ARC_BUFC_METADATA
]);
6286 refcount_destroy(&arc_mfu
->arcs_esize
[ARC_BUFC_DATA
]);
6287 refcount_destroy(&arc_mfu_ghost
->arcs_esize
[ARC_BUFC_METADATA
]);
6288 refcount_destroy(&arc_mfu_ghost
->arcs_esize
[ARC_BUFC_DATA
]);
6289 refcount_destroy(&arc_l2c_only
->arcs_esize
[ARC_BUFC_METADATA
]);
6290 refcount_destroy(&arc_l2c_only
->arcs_esize
[ARC_BUFC_DATA
]);
6292 refcount_destroy(&arc_anon
->arcs_size
);
6293 refcount_destroy(&arc_mru
->arcs_size
);
6294 refcount_destroy(&arc_mru_ghost
->arcs_size
);
6295 refcount_destroy(&arc_mfu
->arcs_size
);
6296 refcount_destroy(&arc_mfu_ghost
->arcs_size
);
6297 refcount_destroy(&arc_l2c_only
->arcs_size
);
6299 multilist_destroy(&arc_mru
->arcs_list
[ARC_BUFC_METADATA
]);
6300 multilist_destroy(&arc_mru_ghost
->arcs_list
[ARC_BUFC_METADATA
]);
6301 multilist_destroy(&arc_mfu
->arcs_list
[ARC_BUFC_METADATA
]);
6302 multilist_destroy(&arc_mfu_ghost
->arcs_list
[ARC_BUFC_METADATA
]);
6303 multilist_destroy(&arc_mru
->arcs_list
[ARC_BUFC_DATA
]);
6304 multilist_destroy(&arc_mru_ghost
->arcs_list
[ARC_BUFC_DATA
]);
6305 multilist_destroy(&arc_mfu
->arcs_list
[ARC_BUFC_DATA
]);
6306 multilist_destroy(&arc_mfu_ghost
->arcs_list
[ARC_BUFC_DATA
]);
6307 multilist_destroy(&arc_l2c_only
->arcs_list
[ARC_BUFC_METADATA
]);
6308 multilist_destroy(&arc_l2c_only
->arcs_list
[ARC_BUFC_DATA
]);
6320 uint64_t percent
, allmem
= arc_all_memory();
6322 mutex_init(&arc_reclaim_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
6323 cv_init(&arc_reclaim_thread_cv
, NULL
, CV_DEFAULT
, NULL
);
6324 cv_init(&arc_reclaim_waiters_cv
, NULL
, CV_DEFAULT
, NULL
);
6326 /* Convert seconds to clock ticks */
6327 arc_min_prefetch_lifespan
= 1 * hz
;
6331 * Register a shrinker to support synchronous (direct) memory
6332 * reclaim from the arc. This is done to prevent kswapd from
6333 * swapping out pages when it is preferable to shrink the arc.
6335 spl_register_shrinker(&arc_shrinker
);
6337 /* Set to 1/64 of all memory or a minimum of 512K */
6338 arc_sys_free
= MAX(allmem
/ 64, (512 * 1024));
6342 /* Set max to 1/2 of all memory */
6343 arc_c_max
= allmem
/ 2;
6346 * In userland, there's only the memory pressure that we artificially
6347 * create (see arc_available_memory()). Don't let arc_c get too
6348 * small, because it can cause transactions to be larger than
6349 * arc_c, causing arc_tempreserve_space() to fail.
6352 arc_c_min
= MAX(arc_c_max
/ 2, 2ULL << SPA_MAXBLOCKSHIFT
);
6354 arc_c_min
= 2ULL << SPA_MAXBLOCKSHIFT
;
6358 arc_p
= (arc_c
>> 1);
6361 /* Set min to 1/2 of arc_c_min */
6362 arc_meta_min
= 1ULL << SPA_MAXBLOCKSHIFT
;
6363 /* Initialize maximum observed usage to zero */
6366 * Set arc_meta_limit to a percent of arc_c_max with a floor of
6367 * arc_meta_min, and a ceiling of arc_c_max.
6369 percent
= MIN(zfs_arc_meta_limit_percent
, 100);
6370 arc_meta_limit
= MAX(arc_meta_min
, (percent
* arc_c_max
) / 100);
6371 percent
= MIN(zfs_arc_dnode_limit_percent
, 100);
6372 arc_dnode_limit
= (percent
* arc_meta_limit
) / 100;
6374 /* Apply user specified tunings */
6375 arc_tuning_update();
6377 if (zfs_arc_num_sublists_per_state
< 1)
6378 zfs_arc_num_sublists_per_state
= MAX(boot_ncpus
, 1);
6380 /* if kmem_flags are set, lets try to use less memory */
6381 if (kmem_debugging())
6383 if (arc_c
< arc_c_min
)
6389 list_create(&arc_prune_list
, sizeof (arc_prune_t
),
6390 offsetof(arc_prune_t
, p_node
));
6391 mutex_init(&arc_prune_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
6393 arc_prune_taskq
= taskq_create("arc_prune", max_ncpus
, defclsyspri
,
6394 max_ncpus
, INT_MAX
, TASKQ_PREPOPULATE
| TASKQ_DYNAMIC
);
6396 arc_reclaim_thread_exit
= B_FALSE
;
6398 arc_ksp
= kstat_create("zfs", 0, "arcstats", "misc", KSTAT_TYPE_NAMED
,
6399 sizeof (arc_stats
) / sizeof (kstat_named_t
), KSTAT_FLAG_VIRTUAL
);
6401 if (arc_ksp
!= NULL
) {
6402 arc_ksp
->ks_data
= &arc_stats
;
6403 arc_ksp
->ks_update
= arc_kstat_update
;
6404 kstat_install(arc_ksp
);
6407 (void) thread_create(NULL
, 0, arc_reclaim_thread
, NULL
, 0, &p0
,
6408 TS_RUN
, defclsyspri
);
6414 * Calculate maximum amount of dirty data per pool.
6416 * If it has been set by a module parameter, take that.
6417 * Otherwise, use a percentage of physical memory defined by
6418 * zfs_dirty_data_max_percent (default 10%) with a cap at
6419 * zfs_dirty_data_max_max (default 25% of physical memory).
6421 if (zfs_dirty_data_max_max
== 0)
6422 zfs_dirty_data_max_max
= allmem
*
6423 zfs_dirty_data_max_max_percent
/ 100;
6425 if (zfs_dirty_data_max
== 0) {
6426 zfs_dirty_data_max
= allmem
*
6427 zfs_dirty_data_max_percent
/ 100;
6428 zfs_dirty_data_max
= MIN(zfs_dirty_data_max
,
6429 zfs_dirty_data_max_max
);
6439 spl_unregister_shrinker(&arc_shrinker
);
6440 #endif /* _KERNEL */
6442 mutex_enter(&arc_reclaim_lock
);
6443 arc_reclaim_thread_exit
= B_TRUE
;
6445 * The reclaim thread will set arc_reclaim_thread_exit back to
6446 * B_FALSE when it is finished exiting; we're waiting for that.
6448 while (arc_reclaim_thread_exit
) {
6449 cv_signal(&arc_reclaim_thread_cv
);
6450 cv_wait(&arc_reclaim_thread_cv
, &arc_reclaim_lock
);
6452 mutex_exit(&arc_reclaim_lock
);
6454 /* Use B_TRUE to ensure *all* buffers are evicted */
6455 arc_flush(NULL
, B_TRUE
);
6459 if (arc_ksp
!= NULL
) {
6460 kstat_delete(arc_ksp
);
6464 taskq_wait(arc_prune_taskq
);
6465 taskq_destroy(arc_prune_taskq
);
6467 mutex_enter(&arc_prune_mtx
);
6468 while ((p
= list_head(&arc_prune_list
)) != NULL
) {
6469 list_remove(&arc_prune_list
, p
);
6470 refcount_remove(&p
->p_refcnt
, &arc_prune_list
);
6471 refcount_destroy(&p
->p_refcnt
);
6472 kmem_free(p
, sizeof (*p
));
6474 mutex_exit(&arc_prune_mtx
);
6476 list_destroy(&arc_prune_list
);
6477 mutex_destroy(&arc_prune_mtx
);
6478 mutex_destroy(&arc_reclaim_lock
);
6479 cv_destroy(&arc_reclaim_thread_cv
);
6480 cv_destroy(&arc_reclaim_waiters_cv
);
6485 ASSERT0(arc_loaned_bytes
);
6491 * The level 2 ARC (L2ARC) is a cache layer in-between main memory and disk.
6492 * It uses dedicated storage devices to hold cached data, which are populated
6493 * using large infrequent writes. The main role of this cache is to boost
6494 * the performance of random read workloads. The intended L2ARC devices
6495 * include short-stroked disks, solid state disks, and other media with
6496 * substantially faster read latency than disk.
6498 * +-----------------------+
6500 * +-----------------------+
6503 * l2arc_feed_thread() arc_read()
6507 * +---------------+ |
6509 * +---------------+ |
6514 * +-------+ +-------+
6516 * | cache | | cache |
6517 * +-------+ +-------+
6518 * +=========+ .-----.
6519 * : L2ARC : |-_____-|
6520 * : devices : | Disks |
6521 * +=========+ `-_____-'
6523 * Read requests are satisfied from the following sources, in order:
6526 * 2) vdev cache of L2ARC devices
6528 * 4) vdev cache of disks
6531 * Some L2ARC device types exhibit extremely slow write performance.
6532 * To accommodate for this there are some significant differences between
6533 * the L2ARC and traditional cache design:
6535 * 1. There is no eviction path from the ARC to the L2ARC. Evictions from
6536 * the ARC behave as usual, freeing buffers and placing headers on ghost
6537 * lists. The ARC does not send buffers to the L2ARC during eviction as
6538 * this would add inflated write latencies for all ARC memory pressure.
6540 * 2. The L2ARC attempts to cache data from the ARC before it is evicted.
6541 * It does this by periodically scanning buffers from the eviction-end of
6542 * the MFU and MRU ARC lists, copying them to the L2ARC devices if they are
6543 * not already there. It scans until a headroom of buffers is satisfied,
6544 * which itself is a buffer for ARC eviction. If a compressible buffer is
6545 * found during scanning and selected for writing to an L2ARC device, we
6546 * temporarily boost scanning headroom during the next scan cycle to make
6547 * sure we adapt to compression effects (which might significantly reduce
6548 * the data volume we write to L2ARC). The thread that does this is
6549 * l2arc_feed_thread(), illustrated below; example sizes are included to
6550 * provide a better sense of ratio than this diagram:
6553 * +---------------------+----------+
6554 * ARC_mfu |:::::#:::::::::::::::|o#o###o###|-->. # already on L2ARC
6555 * +---------------------+----------+ | o L2ARC eligible
6556 * ARC_mru |:#:::::::::::::::::::|#o#ooo####|-->| : ARC buffer
6557 * +---------------------+----------+ |
6558 * 15.9 Gbytes ^ 32 Mbytes |
6560 * l2arc_feed_thread()
6562 * l2arc write hand <--[oooo]--'
6566 * +==============================+
6567 * L2ARC dev |####|#|###|###| |####| ... |
6568 * +==============================+
6571 * 3. If an ARC buffer is copied to the L2ARC but then hit instead of
6572 * evicted, then the L2ARC has cached a buffer much sooner than it probably
6573 * needed to, potentially wasting L2ARC device bandwidth and storage. It is
6574 * safe to say that this is an uncommon case, since buffers at the end of
6575 * the ARC lists have moved there due to inactivity.
6577 * 4. If the ARC evicts faster than the L2ARC can maintain a headroom,
6578 * then the L2ARC simply misses copying some buffers. This serves as a
6579 * pressure valve to prevent heavy read workloads from both stalling the ARC
6580 * with waits and clogging the L2ARC with writes. This also helps prevent
6581 * the potential for the L2ARC to churn if it attempts to cache content too
6582 * quickly, such as during backups of the entire pool.
6584 * 5. After system boot and before the ARC has filled main memory, there are
6585 * no evictions from the ARC and so the tails of the ARC_mfu and ARC_mru
6586 * lists can remain mostly static. Instead of searching from tail of these
6587 * lists as pictured, the l2arc_feed_thread() will search from the list heads
6588 * for eligible buffers, greatly increasing its chance of finding them.
6590 * The L2ARC device write speed is also boosted during this time so that
6591 * the L2ARC warms up faster. Since there have been no ARC evictions yet,
6592 * there are no L2ARC reads, and no fear of degrading read performance
6593 * through increased writes.
6595 * 6. Writes to the L2ARC devices are grouped and sent in-sequence, so that
6596 * the vdev queue can aggregate them into larger and fewer writes. Each
6597 * device is written to in a rotor fashion, sweeping writes through
6598 * available space then repeating.
6600 * 7. The L2ARC does not store dirty content. It never needs to flush
6601 * write buffers back to disk based storage.
6603 * 8. If an ARC buffer is written (and dirtied) which also exists in the
6604 * L2ARC, the now stale L2ARC buffer is immediately dropped.
6606 * The performance of the L2ARC can be tweaked by a number of tunables, which
6607 * may be necessary for different workloads:
6609 * l2arc_write_max max write bytes per interval
6610 * l2arc_write_boost extra write bytes during device warmup
6611 * l2arc_noprefetch skip caching prefetched buffers
6612 * l2arc_headroom number of max device writes to precache
6613 * l2arc_headroom_boost when we find compressed buffers during ARC
6614 * scanning, we multiply headroom by this
6615 * percentage factor for the next scan cycle,
6616 * since more compressed buffers are likely to
6618 * l2arc_feed_secs seconds between L2ARC writing
6620 * Tunables may be removed or added as future performance improvements are
6621 * integrated, and also may become zpool properties.
6623 * There are three key functions that control how the L2ARC warms up:
6625 * l2arc_write_eligible() check if a buffer is eligible to cache
6626 * l2arc_write_size() calculate how much to write
6627 * l2arc_write_interval() calculate sleep delay between writes
6629 * These three functions determine what to write, how much, and how quickly
6634 l2arc_write_eligible(uint64_t spa_guid
, arc_buf_hdr_t
*hdr
)
6637 * A buffer is *not* eligible for the L2ARC if it:
6638 * 1. belongs to a different spa.
6639 * 2. is already cached on the L2ARC.
6640 * 3. has an I/O in progress (it may be an incomplete read).
6641 * 4. is flagged not eligible (zfs property).
6643 if (hdr
->b_spa
!= spa_guid
|| HDR_HAS_L2HDR(hdr
) ||
6644 HDR_IO_IN_PROGRESS(hdr
) || !HDR_L2CACHE(hdr
))
6651 l2arc_write_size(void)
6656 * Make sure our globals have meaningful values in case the user
6659 size
= l2arc_write_max
;
6661 cmn_err(CE_NOTE
, "Bad value for l2arc_write_max, value must "
6662 "be greater than zero, resetting it to the default (%d)",
6664 size
= l2arc_write_max
= L2ARC_WRITE_SIZE
;
6667 if (arc_warm
== B_FALSE
)
6668 size
+= l2arc_write_boost
;
6675 l2arc_write_interval(clock_t began
, uint64_t wanted
, uint64_t wrote
)
6677 clock_t interval
, next
, now
;
6680 * If the ARC lists are busy, increase our write rate; if the
6681 * lists are stale, idle back. This is achieved by checking
6682 * how much we previously wrote - if it was more than half of
6683 * what we wanted, schedule the next write much sooner.
6685 if (l2arc_feed_again
&& wrote
> (wanted
/ 2))
6686 interval
= (hz
* l2arc_feed_min_ms
) / 1000;
6688 interval
= hz
* l2arc_feed_secs
;
6690 now
= ddi_get_lbolt();
6691 next
= MAX(now
, MIN(now
+ interval
, began
+ interval
));
6697 * Cycle through L2ARC devices. This is how L2ARC load balances.
6698 * If a device is returned, this also returns holding the spa config lock.
6700 static l2arc_dev_t
*
6701 l2arc_dev_get_next(void)
6703 l2arc_dev_t
*first
, *next
= NULL
;
6706 * Lock out the removal of spas (spa_namespace_lock), then removal
6707 * of cache devices (l2arc_dev_mtx). Once a device has been selected,
6708 * both locks will be dropped and a spa config lock held instead.
6710 mutex_enter(&spa_namespace_lock
);
6711 mutex_enter(&l2arc_dev_mtx
);
6713 /* if there are no vdevs, there is nothing to do */
6714 if (l2arc_ndev
== 0)
6718 next
= l2arc_dev_last
;
6720 /* loop around the list looking for a non-faulted vdev */
6722 next
= list_head(l2arc_dev_list
);
6724 next
= list_next(l2arc_dev_list
, next
);
6726 next
= list_head(l2arc_dev_list
);
6729 /* if we have come back to the start, bail out */
6732 else if (next
== first
)
6735 } while (vdev_is_dead(next
->l2ad_vdev
));
6737 /* if we were unable to find any usable vdevs, return NULL */
6738 if (vdev_is_dead(next
->l2ad_vdev
))
6741 l2arc_dev_last
= next
;
6744 mutex_exit(&l2arc_dev_mtx
);
6747 * Grab the config lock to prevent the 'next' device from being
6748 * removed while we are writing to it.
6751 spa_config_enter(next
->l2ad_spa
, SCL_L2ARC
, next
, RW_READER
);
6752 mutex_exit(&spa_namespace_lock
);
6758 * Free buffers that were tagged for destruction.
6761 l2arc_do_free_on_write(void)
6764 l2arc_data_free_t
*df
, *df_prev
;
6766 mutex_enter(&l2arc_free_on_write_mtx
);
6767 buflist
= l2arc_free_on_write
;
6769 for (df
= list_tail(buflist
); df
; df
= df_prev
) {
6770 df_prev
= list_prev(buflist
, df
);
6771 ASSERT3P(df
->l2df_data
, !=, NULL
);
6772 if (df
->l2df_type
== ARC_BUFC_METADATA
) {
6773 zio_buf_free(df
->l2df_data
, df
->l2df_size
);
6775 ASSERT(df
->l2df_type
== ARC_BUFC_DATA
);
6776 zio_data_buf_free(df
->l2df_data
, df
->l2df_size
);
6778 list_remove(buflist
, df
);
6779 kmem_free(df
, sizeof (l2arc_data_free_t
));
6782 mutex_exit(&l2arc_free_on_write_mtx
);
6786 * A write to a cache device has completed. Update all headers to allow
6787 * reads from these buffers to begin.
6790 l2arc_write_done(zio_t
*zio
)
6792 l2arc_write_callback_t
*cb
;
6795 arc_buf_hdr_t
*head
, *hdr
, *hdr_prev
;
6796 kmutex_t
*hash_lock
;
6797 int64_t bytes_dropped
= 0;
6799 cb
= zio
->io_private
;
6800 ASSERT3P(cb
, !=, NULL
);
6801 dev
= cb
->l2wcb_dev
;
6802 ASSERT3P(dev
, !=, NULL
);
6803 head
= cb
->l2wcb_head
;
6804 ASSERT3P(head
, !=, NULL
);
6805 buflist
= &dev
->l2ad_buflist
;
6806 ASSERT3P(buflist
, !=, NULL
);
6807 DTRACE_PROBE2(l2arc__iodone
, zio_t
*, zio
,
6808 l2arc_write_callback_t
*, cb
);
6810 if (zio
->io_error
!= 0)
6811 ARCSTAT_BUMP(arcstat_l2_writes_error
);
6814 * All writes completed, or an error was hit.
6817 mutex_enter(&dev
->l2ad_mtx
);
6818 for (hdr
= list_prev(buflist
, head
); hdr
; hdr
= hdr_prev
) {
6819 hdr_prev
= list_prev(buflist
, hdr
);
6821 hash_lock
= HDR_LOCK(hdr
);
6824 * We cannot use mutex_enter or else we can deadlock
6825 * with l2arc_write_buffers (due to swapping the order
6826 * the hash lock and l2ad_mtx are taken).
6828 if (!mutex_tryenter(hash_lock
)) {
6830 * Missed the hash lock. We must retry so we
6831 * don't leave the ARC_FLAG_L2_WRITING bit set.
6833 ARCSTAT_BUMP(arcstat_l2_writes_lock_retry
);
6836 * We don't want to rescan the headers we've
6837 * already marked as having been written out, so
6838 * we reinsert the head node so we can pick up
6839 * where we left off.
6841 list_remove(buflist
, head
);
6842 list_insert_after(buflist
, hdr
, head
);
6844 mutex_exit(&dev
->l2ad_mtx
);
6847 * We wait for the hash lock to become available
6848 * to try and prevent busy waiting, and increase
6849 * the chance we'll be able to acquire the lock
6850 * the next time around.
6852 mutex_enter(hash_lock
);
6853 mutex_exit(hash_lock
);
6858 * We could not have been moved into the arc_l2c_only
6859 * state while in-flight due to our ARC_FLAG_L2_WRITING
6860 * bit being set. Let's just ensure that's being enforced.
6862 ASSERT(HDR_HAS_L1HDR(hdr
));
6865 * Skipped - drop L2ARC entry and mark the header as no
6866 * longer L2 eligibile.
6868 if (zio
->io_error
!= 0) {
6870 * Error - drop L2ARC entry.
6872 list_remove(buflist
, hdr
);
6873 arc_hdr_clear_flags(hdr
, ARC_FLAG_HAS_L2HDR
);
6875 ARCSTAT_INCR(arcstat_l2_asize
, -arc_hdr_size(hdr
));
6876 ARCSTAT_INCR(arcstat_l2_size
, -HDR_GET_LSIZE(hdr
));
6878 bytes_dropped
+= arc_hdr_size(hdr
);
6879 (void) refcount_remove_many(&dev
->l2ad_alloc
,
6880 arc_hdr_size(hdr
), hdr
);
6884 * Allow ARC to begin reads and ghost list evictions to
6887 arc_hdr_clear_flags(hdr
, ARC_FLAG_L2_WRITING
);
6889 mutex_exit(hash_lock
);
6892 atomic_inc_64(&l2arc_writes_done
);
6893 list_remove(buflist
, head
);
6894 ASSERT(!HDR_HAS_L1HDR(head
));
6895 kmem_cache_free(hdr_l2only_cache
, head
);
6896 mutex_exit(&dev
->l2ad_mtx
);
6898 vdev_space_update(dev
->l2ad_vdev
, -bytes_dropped
, 0, 0);
6900 l2arc_do_free_on_write();
6902 kmem_free(cb
, sizeof (l2arc_write_callback_t
));
6906 * A read to a cache device completed. Validate buffer contents before
6907 * handing over to the regular ARC routines.
6910 l2arc_read_done(zio_t
*zio
)
6912 l2arc_read_callback_t
*cb
;
6914 kmutex_t
*hash_lock
;
6915 boolean_t valid_cksum
;
6917 ASSERT3P(zio
->io_vd
, !=, NULL
);
6918 ASSERT(zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
);
6920 spa_config_exit(zio
->io_spa
, SCL_L2ARC
, zio
->io_vd
);
6922 cb
= zio
->io_private
;
6923 ASSERT3P(cb
, !=, NULL
);
6924 hdr
= cb
->l2rcb_hdr
;
6925 ASSERT3P(hdr
, !=, NULL
);
6927 hash_lock
= HDR_LOCK(hdr
);
6928 mutex_enter(hash_lock
);
6929 ASSERT3P(hash_lock
, ==, HDR_LOCK(hdr
));
6931 ASSERT3P(zio
->io_data
, !=, NULL
);
6934 * Check this survived the L2ARC journey.
6936 ASSERT3P(zio
->io_data
, ==, hdr
->b_l1hdr
.b_pdata
);
6937 zio
->io_bp_copy
= cb
->l2rcb_bp
; /* XXX fix in L2ARC 2.0 */
6938 zio
->io_bp
= &zio
->io_bp_copy
; /* XXX fix in L2ARC 2.0 */
6940 valid_cksum
= arc_cksum_is_equal(hdr
, zio
);
6941 if (valid_cksum
&& zio
->io_error
== 0 && !HDR_L2_EVICTED(hdr
)) {
6942 mutex_exit(hash_lock
);
6943 zio
->io_private
= hdr
;
6946 mutex_exit(hash_lock
);
6948 * Buffer didn't survive caching. Increment stats and
6949 * reissue to the original storage device.
6951 if (zio
->io_error
!= 0) {
6952 ARCSTAT_BUMP(arcstat_l2_io_error
);
6954 zio
->io_error
= SET_ERROR(EIO
);
6957 ARCSTAT_BUMP(arcstat_l2_cksum_bad
);
6960 * If there's no waiter, issue an async i/o to the primary
6961 * storage now. If there *is* a waiter, the caller must
6962 * issue the i/o in a context where it's OK to block.
6964 if (zio
->io_waiter
== NULL
) {
6965 zio_t
*pio
= zio_unique_parent(zio
);
6967 ASSERT(!pio
|| pio
->io_child_type
== ZIO_CHILD_LOGICAL
);
6969 zio_nowait(zio_read(pio
, zio
->io_spa
, zio
->io_bp
,
6970 hdr
->b_l1hdr
.b_pdata
, zio
->io_size
, arc_read_done
,
6971 hdr
, zio
->io_priority
, cb
->l2rcb_flags
,
6976 kmem_free(cb
, sizeof (l2arc_read_callback_t
));
6980 * This is the list priority from which the L2ARC will search for pages to
6981 * cache. This is used within loops (0..3) to cycle through lists in the
6982 * desired order. This order can have a significant effect on cache
6985 * Currently the metadata lists are hit first, MFU then MRU, followed by
6986 * the data lists. This function returns a locked list, and also returns
6989 static multilist_sublist_t
*
6990 l2arc_sublist_lock(int list_num
)
6992 multilist_t
*ml
= NULL
;
6995 ASSERT(list_num
>= 0 && list_num
< L2ARC_FEED_TYPES
);
6999 ml
= &arc_mfu
->arcs_list
[ARC_BUFC_METADATA
];
7002 ml
= &arc_mru
->arcs_list
[ARC_BUFC_METADATA
];
7005 ml
= &arc_mfu
->arcs_list
[ARC_BUFC_DATA
];
7008 ml
= &arc_mru
->arcs_list
[ARC_BUFC_DATA
];
7015 * Return a randomly-selected sublist. This is acceptable
7016 * because the caller feeds only a little bit of data for each
7017 * call (8MB). Subsequent calls will result in different
7018 * sublists being selected.
7020 idx
= multilist_get_random_index(ml
);
7021 return (multilist_sublist_lock(ml
, idx
));
7025 * Evict buffers from the device write hand to the distance specified in
7026 * bytes. This distance may span populated buffers, it may span nothing.
7027 * This is clearing a region on the L2ARC device ready for writing.
7028 * If the 'all' boolean is set, every buffer is evicted.
7031 l2arc_evict(l2arc_dev_t
*dev
, uint64_t distance
, boolean_t all
)
7034 arc_buf_hdr_t
*hdr
, *hdr_prev
;
7035 kmutex_t
*hash_lock
;
7038 buflist
= &dev
->l2ad_buflist
;
7040 if (!all
&& dev
->l2ad_first
) {
7042 * This is the first sweep through the device. There is
7048 if (dev
->l2ad_hand
>= (dev
->l2ad_end
- (2 * distance
))) {
7050 * When nearing the end of the device, evict to the end
7051 * before the device write hand jumps to the start.
7053 taddr
= dev
->l2ad_end
;
7055 taddr
= dev
->l2ad_hand
+ distance
;
7057 DTRACE_PROBE4(l2arc__evict
, l2arc_dev_t
*, dev
, list_t
*, buflist
,
7058 uint64_t, taddr
, boolean_t
, all
);
7061 mutex_enter(&dev
->l2ad_mtx
);
7062 for (hdr
= list_tail(buflist
); hdr
; hdr
= hdr_prev
) {
7063 hdr_prev
= list_prev(buflist
, hdr
);
7065 hash_lock
= HDR_LOCK(hdr
);
7068 * We cannot use mutex_enter or else we can deadlock
7069 * with l2arc_write_buffers (due to swapping the order
7070 * the hash lock and l2ad_mtx are taken).
7072 if (!mutex_tryenter(hash_lock
)) {
7074 * Missed the hash lock. Retry.
7076 ARCSTAT_BUMP(arcstat_l2_evict_lock_retry
);
7077 mutex_exit(&dev
->l2ad_mtx
);
7078 mutex_enter(hash_lock
);
7079 mutex_exit(hash_lock
);
7083 if (HDR_L2_WRITE_HEAD(hdr
)) {
7085 * We hit a write head node. Leave it for
7086 * l2arc_write_done().
7088 list_remove(buflist
, hdr
);
7089 mutex_exit(hash_lock
);
7093 if (!all
&& HDR_HAS_L2HDR(hdr
) &&
7094 (hdr
->b_l2hdr
.b_daddr
> taddr
||
7095 hdr
->b_l2hdr
.b_daddr
< dev
->l2ad_hand
)) {
7097 * We've evicted to the target address,
7098 * or the end of the device.
7100 mutex_exit(hash_lock
);
7104 ASSERT(HDR_HAS_L2HDR(hdr
));
7105 if (!HDR_HAS_L1HDR(hdr
)) {
7106 ASSERT(!HDR_L2_READING(hdr
));
7108 * This doesn't exist in the ARC. Destroy.
7109 * arc_hdr_destroy() will call list_remove()
7110 * and decrement arcstat_l2_size.
7112 arc_change_state(arc_anon
, hdr
, hash_lock
);
7113 arc_hdr_destroy(hdr
);
7115 ASSERT(hdr
->b_l1hdr
.b_state
!= arc_l2c_only
);
7116 ARCSTAT_BUMP(arcstat_l2_evict_l1cached
);
7118 * Invalidate issued or about to be issued
7119 * reads, since we may be about to write
7120 * over this location.
7122 if (HDR_L2_READING(hdr
)) {
7123 ARCSTAT_BUMP(arcstat_l2_evict_reading
);
7124 arc_hdr_set_flags(hdr
, ARC_FLAG_L2_EVICTED
);
7127 /* Ensure this header has finished being written */
7128 ASSERT(!HDR_L2_WRITING(hdr
));
7130 arc_hdr_l2hdr_destroy(hdr
);
7132 mutex_exit(hash_lock
);
7134 mutex_exit(&dev
->l2ad_mtx
);
7138 * Find and write ARC buffers to the L2ARC device.
7140 * An ARC_FLAG_L2_WRITING flag is set so that the L2ARC buffers are not valid
7141 * for reading until they have completed writing.
7142 * The headroom_boost is an in-out parameter used to maintain headroom boost
7143 * state between calls to this function.
7145 * Returns the number of bytes actually written (which may be smaller than
7146 * the delta by which the device hand has changed due to alignment).
7149 l2arc_write_buffers(spa_t
*spa
, l2arc_dev_t
*dev
, uint64_t target_sz
)
7151 arc_buf_hdr_t
*hdr
, *hdr_prev
, *head
;
7152 uint64_t write_asize
, write_psize
, write_sz
, headroom
;
7154 l2arc_write_callback_t
*cb
;
7156 uint64_t guid
= spa_load_guid(spa
);
7159 ASSERT3P(dev
->l2ad_vdev
, !=, NULL
);
7162 write_sz
= write_asize
= write_psize
= 0;
7164 head
= kmem_cache_alloc(hdr_l2only_cache
, KM_PUSHPAGE
);
7165 arc_hdr_set_flags(head
, ARC_FLAG_L2_WRITE_HEAD
| ARC_FLAG_HAS_L2HDR
);
7168 * Copy buffers for L2ARC writing.
7170 for (try = 0; try < L2ARC_FEED_TYPES
; try++) {
7171 multilist_sublist_t
*mls
= l2arc_sublist_lock(try);
7172 uint64_t passed_sz
= 0;
7174 VERIFY3P(mls
, !=, NULL
);
7177 * L2ARC fast warmup.
7179 * Until the ARC is warm and starts to evict, read from the
7180 * head of the ARC lists rather than the tail.
7182 if (arc_warm
== B_FALSE
)
7183 hdr
= multilist_sublist_head(mls
);
7185 hdr
= multilist_sublist_tail(mls
);
7187 headroom
= target_sz
* l2arc_headroom
;
7188 if (zfs_compressed_arc_enabled
)
7189 headroom
= (headroom
* l2arc_headroom_boost
) / 100;
7191 for (; hdr
; hdr
= hdr_prev
) {
7192 kmutex_t
*hash_lock
;
7193 uint64_t asize
, size
;
7196 if (arc_warm
== B_FALSE
)
7197 hdr_prev
= multilist_sublist_next(mls
, hdr
);
7199 hdr_prev
= multilist_sublist_prev(mls
, hdr
);
7201 hash_lock
= HDR_LOCK(hdr
);
7202 if (!mutex_tryenter(hash_lock
)) {
7204 * Skip this buffer rather than waiting.
7209 passed_sz
+= HDR_GET_LSIZE(hdr
);
7210 if (passed_sz
> headroom
) {
7214 mutex_exit(hash_lock
);
7218 if (!l2arc_write_eligible(guid
, hdr
)) {
7219 mutex_exit(hash_lock
);
7223 if ((write_asize
+ HDR_GET_LSIZE(hdr
)) > target_sz
) {
7225 mutex_exit(hash_lock
);
7231 * Insert a dummy header on the buflist so
7232 * l2arc_write_done() can find where the
7233 * write buffers begin without searching.
7235 mutex_enter(&dev
->l2ad_mtx
);
7236 list_insert_head(&dev
->l2ad_buflist
, head
);
7237 mutex_exit(&dev
->l2ad_mtx
);
7240 sizeof (l2arc_write_callback_t
), KM_SLEEP
);
7241 cb
->l2wcb_dev
= dev
;
7242 cb
->l2wcb_head
= head
;
7243 pio
= zio_root(spa
, l2arc_write_done
, cb
,
7247 hdr
->b_l2hdr
.b_dev
= dev
;
7248 hdr
->b_l2hdr
.b_hits
= 0;
7250 hdr
->b_l2hdr
.b_daddr
= dev
->l2ad_hand
;
7251 arc_hdr_set_flags(hdr
,
7252 ARC_FLAG_L2_WRITING
| ARC_FLAG_HAS_L2HDR
);
7254 mutex_enter(&dev
->l2ad_mtx
);
7255 list_insert_head(&dev
->l2ad_buflist
, hdr
);
7256 mutex_exit(&dev
->l2ad_mtx
);
7259 * We rely on the L1 portion of the header below, so
7260 * it's invalid for this header to have been evicted out
7261 * of the ghost cache, prior to being written out. The
7262 * ARC_FLAG_L2_WRITING bit ensures this won't happen.
7264 ASSERT(HDR_HAS_L1HDR(hdr
));
7266 ASSERT3U(HDR_GET_PSIZE(hdr
), >, 0);
7267 ASSERT3P(hdr
->b_l1hdr
.b_pdata
, !=, NULL
);
7268 ASSERT3U(arc_hdr_size(hdr
), >, 0);
7269 size
= arc_hdr_size(hdr
);
7271 (void) refcount_add_many(&dev
->l2ad_alloc
, size
, hdr
);
7274 * Normally the L2ARC can use the hdr's data, but if
7275 * we're sharing data between the hdr and one of its
7276 * bufs, L2ARC needs its own copy of the data so that
7277 * the ZIO below can't race with the buf consumer. To
7278 * ensure that this copy will be available for the
7279 * lifetime of the ZIO and be cleaned up afterwards, we
7280 * add it to the l2arc_free_on_write queue.
7282 if (!HDR_SHARED_DATA(hdr
)) {
7283 to_write
= hdr
->b_l1hdr
.b_pdata
;
7285 arc_buf_contents_t type
= arc_buf_type(hdr
);
7286 if (type
== ARC_BUFC_METADATA
) {
7287 to_write
= zio_buf_alloc(size
);
7289 ASSERT3U(type
, ==, ARC_BUFC_DATA
);
7290 to_write
= zio_data_buf_alloc(size
);
7293 bcopy(hdr
->b_l1hdr
.b_pdata
, to_write
, size
);
7294 l2arc_free_data_on_write(to_write
, size
, type
);
7296 wzio
= zio_write_phys(pio
, dev
->l2ad_vdev
,
7297 hdr
->b_l2hdr
.b_daddr
, size
, to_write
,
7298 ZIO_CHECKSUM_OFF
, NULL
, hdr
,
7299 ZIO_PRIORITY_ASYNC_WRITE
,
7300 ZIO_FLAG_CANFAIL
, B_FALSE
);
7302 write_sz
+= HDR_GET_LSIZE(hdr
);
7303 DTRACE_PROBE2(l2arc__write
, vdev_t
*, dev
->l2ad_vdev
,
7306 write_asize
+= size
;
7308 * Keep the clock hand suitably device-aligned.
7310 asize
= vdev_psize_to_asize(dev
->l2ad_vdev
, size
);
7311 write_psize
+= asize
;
7312 dev
->l2ad_hand
+= asize
;
7314 mutex_exit(hash_lock
);
7316 (void) zio_nowait(wzio
);
7319 multilist_sublist_unlock(mls
);
7325 /* No buffers selected for writing? */
7328 ASSERT(!HDR_HAS_L1HDR(head
));
7329 kmem_cache_free(hdr_l2only_cache
, head
);
7333 ASSERT3U(write_asize
, <=, target_sz
);
7334 ARCSTAT_BUMP(arcstat_l2_writes_sent
);
7335 ARCSTAT_INCR(arcstat_l2_write_bytes
, write_asize
);
7336 ARCSTAT_INCR(arcstat_l2_size
, write_sz
);
7337 ARCSTAT_INCR(arcstat_l2_asize
, write_asize
);
7338 vdev_space_update(dev
->l2ad_vdev
, write_asize
, 0, 0);
7341 * Bump device hand to the device start if it is approaching the end.
7342 * l2arc_evict() will already have evicted ahead for this case.
7344 if (dev
->l2ad_hand
>= (dev
->l2ad_end
- target_sz
)) {
7345 dev
->l2ad_hand
= dev
->l2ad_start
;
7346 dev
->l2ad_first
= B_FALSE
;
7349 dev
->l2ad_writing
= B_TRUE
;
7350 (void) zio_wait(pio
);
7351 dev
->l2ad_writing
= B_FALSE
;
7353 return (write_asize
);
7357 * This thread feeds the L2ARC at regular intervals. This is the beating
7358 * heart of the L2ARC.
7361 l2arc_feed_thread(void)
7366 uint64_t size
, wrote
;
7367 clock_t begin
, next
= ddi_get_lbolt();
7368 fstrans_cookie_t cookie
;
7370 CALLB_CPR_INIT(&cpr
, &l2arc_feed_thr_lock
, callb_generic_cpr
, FTAG
);
7372 mutex_enter(&l2arc_feed_thr_lock
);
7374 cookie
= spl_fstrans_mark();
7375 while (l2arc_thread_exit
== 0) {
7376 CALLB_CPR_SAFE_BEGIN(&cpr
);
7377 (void) cv_timedwait_sig(&l2arc_feed_thr_cv
,
7378 &l2arc_feed_thr_lock
, next
);
7379 CALLB_CPR_SAFE_END(&cpr
, &l2arc_feed_thr_lock
);
7380 next
= ddi_get_lbolt() + hz
;
7383 * Quick check for L2ARC devices.
7385 mutex_enter(&l2arc_dev_mtx
);
7386 if (l2arc_ndev
== 0) {
7387 mutex_exit(&l2arc_dev_mtx
);
7390 mutex_exit(&l2arc_dev_mtx
);
7391 begin
= ddi_get_lbolt();
7394 * This selects the next l2arc device to write to, and in
7395 * doing so the next spa to feed from: dev->l2ad_spa. This
7396 * will return NULL if there are now no l2arc devices or if
7397 * they are all faulted.
7399 * If a device is returned, its spa's config lock is also
7400 * held to prevent device removal. l2arc_dev_get_next()
7401 * will grab and release l2arc_dev_mtx.
7403 if ((dev
= l2arc_dev_get_next()) == NULL
)
7406 spa
= dev
->l2ad_spa
;
7407 ASSERT3P(spa
, !=, NULL
);
7410 * If the pool is read-only then force the feed thread to
7411 * sleep a little longer.
7413 if (!spa_writeable(spa
)) {
7414 next
= ddi_get_lbolt() + 5 * l2arc_feed_secs
* hz
;
7415 spa_config_exit(spa
, SCL_L2ARC
, dev
);
7420 * Avoid contributing to memory pressure.
7422 if (arc_reclaim_needed()) {
7423 ARCSTAT_BUMP(arcstat_l2_abort_lowmem
);
7424 spa_config_exit(spa
, SCL_L2ARC
, dev
);
7428 ARCSTAT_BUMP(arcstat_l2_feeds
);
7430 size
= l2arc_write_size();
7433 * Evict L2ARC buffers that will be overwritten.
7435 l2arc_evict(dev
, size
, B_FALSE
);
7438 * Write ARC buffers.
7440 wrote
= l2arc_write_buffers(spa
, dev
, size
);
7443 * Calculate interval between writes.
7445 next
= l2arc_write_interval(begin
, size
, wrote
);
7446 spa_config_exit(spa
, SCL_L2ARC
, dev
);
7448 spl_fstrans_unmark(cookie
);
7450 l2arc_thread_exit
= 0;
7451 cv_broadcast(&l2arc_feed_thr_cv
);
7452 CALLB_CPR_EXIT(&cpr
); /* drops l2arc_feed_thr_lock */
7457 l2arc_vdev_present(vdev_t
*vd
)
7461 mutex_enter(&l2arc_dev_mtx
);
7462 for (dev
= list_head(l2arc_dev_list
); dev
!= NULL
;
7463 dev
= list_next(l2arc_dev_list
, dev
)) {
7464 if (dev
->l2ad_vdev
== vd
)
7467 mutex_exit(&l2arc_dev_mtx
);
7469 return (dev
!= NULL
);
7473 * Add a vdev for use by the L2ARC. By this point the spa has already
7474 * validated the vdev and opened it.
7477 l2arc_add_vdev(spa_t
*spa
, vdev_t
*vd
)
7479 l2arc_dev_t
*adddev
;
7481 ASSERT(!l2arc_vdev_present(vd
));
7484 * Create a new l2arc device entry.
7486 adddev
= kmem_zalloc(sizeof (l2arc_dev_t
), KM_SLEEP
);
7487 adddev
->l2ad_spa
= spa
;
7488 adddev
->l2ad_vdev
= vd
;
7489 adddev
->l2ad_start
= VDEV_LABEL_START_SIZE
;
7490 adddev
->l2ad_end
= VDEV_LABEL_START_SIZE
+ vdev_get_min_asize(vd
);
7491 adddev
->l2ad_hand
= adddev
->l2ad_start
;
7492 adddev
->l2ad_first
= B_TRUE
;
7493 adddev
->l2ad_writing
= B_FALSE
;
7494 list_link_init(&adddev
->l2ad_node
);
7496 mutex_init(&adddev
->l2ad_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
7498 * This is a list of all ARC buffers that are still valid on the
7501 list_create(&adddev
->l2ad_buflist
, sizeof (arc_buf_hdr_t
),
7502 offsetof(arc_buf_hdr_t
, b_l2hdr
.b_l2node
));
7504 vdev_space_update(vd
, 0, 0, adddev
->l2ad_end
- adddev
->l2ad_hand
);
7505 refcount_create(&adddev
->l2ad_alloc
);
7508 * Add device to global list
7510 mutex_enter(&l2arc_dev_mtx
);
7511 list_insert_head(l2arc_dev_list
, adddev
);
7512 atomic_inc_64(&l2arc_ndev
);
7513 mutex_exit(&l2arc_dev_mtx
);
7517 * Remove a vdev from the L2ARC.
7520 l2arc_remove_vdev(vdev_t
*vd
)
7522 l2arc_dev_t
*dev
, *nextdev
, *remdev
= NULL
;
7525 * Find the device by vdev
7527 mutex_enter(&l2arc_dev_mtx
);
7528 for (dev
= list_head(l2arc_dev_list
); dev
; dev
= nextdev
) {
7529 nextdev
= list_next(l2arc_dev_list
, dev
);
7530 if (vd
== dev
->l2ad_vdev
) {
7535 ASSERT3P(remdev
, !=, NULL
);
7538 * Remove device from global list
7540 list_remove(l2arc_dev_list
, remdev
);
7541 l2arc_dev_last
= NULL
; /* may have been invalidated */
7542 atomic_dec_64(&l2arc_ndev
);
7543 mutex_exit(&l2arc_dev_mtx
);
7546 * Clear all buflists and ARC references. L2ARC device flush.
7548 l2arc_evict(remdev
, 0, B_TRUE
);
7549 list_destroy(&remdev
->l2ad_buflist
);
7550 mutex_destroy(&remdev
->l2ad_mtx
);
7551 refcount_destroy(&remdev
->l2ad_alloc
);
7552 kmem_free(remdev
, sizeof (l2arc_dev_t
));
7558 l2arc_thread_exit
= 0;
7560 l2arc_writes_sent
= 0;
7561 l2arc_writes_done
= 0;
7563 mutex_init(&l2arc_feed_thr_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
7564 cv_init(&l2arc_feed_thr_cv
, NULL
, CV_DEFAULT
, NULL
);
7565 mutex_init(&l2arc_dev_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
7566 mutex_init(&l2arc_free_on_write_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
7568 l2arc_dev_list
= &L2ARC_dev_list
;
7569 l2arc_free_on_write
= &L2ARC_free_on_write
;
7570 list_create(l2arc_dev_list
, sizeof (l2arc_dev_t
),
7571 offsetof(l2arc_dev_t
, l2ad_node
));
7572 list_create(l2arc_free_on_write
, sizeof (l2arc_data_free_t
),
7573 offsetof(l2arc_data_free_t
, l2df_list_node
));
7580 * This is called from dmu_fini(), which is called from spa_fini();
7581 * Because of this, we can assume that all l2arc devices have
7582 * already been removed when the pools themselves were removed.
7585 l2arc_do_free_on_write();
7587 mutex_destroy(&l2arc_feed_thr_lock
);
7588 cv_destroy(&l2arc_feed_thr_cv
);
7589 mutex_destroy(&l2arc_dev_mtx
);
7590 mutex_destroy(&l2arc_free_on_write_mtx
);
7592 list_destroy(l2arc_dev_list
);
7593 list_destroy(l2arc_free_on_write
);
7599 if (!(spa_mode_global
& FWRITE
))
7602 (void) thread_create(NULL
, 0, l2arc_feed_thread
, NULL
, 0, &p0
,
7603 TS_RUN
, defclsyspri
);
7609 if (!(spa_mode_global
& FWRITE
))
7612 mutex_enter(&l2arc_feed_thr_lock
);
7613 cv_signal(&l2arc_feed_thr_cv
); /* kick thread out of startup */
7614 l2arc_thread_exit
= 1;
7615 while (l2arc_thread_exit
!= 0)
7616 cv_wait(&l2arc_feed_thr_cv
, &l2arc_feed_thr_lock
);
7617 mutex_exit(&l2arc_feed_thr_lock
);
7620 #if defined(_KERNEL) && defined(HAVE_SPL)
7621 EXPORT_SYMBOL(arc_buf_size
);
7622 EXPORT_SYMBOL(arc_write
);
7623 EXPORT_SYMBOL(arc_read
);
7624 EXPORT_SYMBOL(arc_buf_info
);
7625 EXPORT_SYMBOL(arc_getbuf_func
);
7626 EXPORT_SYMBOL(arc_add_prune_callback
);
7627 EXPORT_SYMBOL(arc_remove_prune_callback
);
7629 module_param(zfs_arc_min
, ulong
, 0644);
7630 MODULE_PARM_DESC(zfs_arc_min
, "Min arc size");
7632 module_param(zfs_arc_max
, ulong
, 0644);
7633 MODULE_PARM_DESC(zfs_arc_max
, "Max arc size");
7635 module_param(zfs_arc_meta_limit
, ulong
, 0644);
7636 MODULE_PARM_DESC(zfs_arc_meta_limit
, "Meta limit for arc size");
7638 module_param(zfs_arc_meta_limit_percent
, ulong
, 0644);
7639 MODULE_PARM_DESC(zfs_arc_meta_limit_percent
,
7640 "Percent of arc size for arc meta limit");
7642 module_param(zfs_arc_meta_min
, ulong
, 0644);
7643 MODULE_PARM_DESC(zfs_arc_meta_min
, "Min arc metadata");
7645 module_param(zfs_arc_meta_prune
, int, 0644);
7646 MODULE_PARM_DESC(zfs_arc_meta_prune
, "Meta objects to scan for prune");
7648 module_param(zfs_arc_meta_adjust_restarts
, int, 0644);
7649 MODULE_PARM_DESC(zfs_arc_meta_adjust_restarts
,
7650 "Limit number of restarts in arc_adjust_meta");
7652 module_param(zfs_arc_meta_strategy
, int, 0644);
7653 MODULE_PARM_DESC(zfs_arc_meta_strategy
, "Meta reclaim strategy");
7655 module_param(zfs_arc_grow_retry
, int, 0644);
7656 MODULE_PARM_DESC(zfs_arc_grow_retry
, "Seconds before growing arc size");
7658 module_param(zfs_arc_p_aggressive_disable
, int, 0644);
7659 MODULE_PARM_DESC(zfs_arc_p_aggressive_disable
, "disable aggressive arc_p grow");
7661 module_param(zfs_arc_p_dampener_disable
, int, 0644);
7662 MODULE_PARM_DESC(zfs_arc_p_dampener_disable
, "disable arc_p adapt dampener");
7664 module_param(zfs_arc_shrink_shift
, int, 0644);
7665 MODULE_PARM_DESC(zfs_arc_shrink_shift
, "log2(fraction of arc to reclaim)");
7667 module_param(zfs_arc_p_min_shift
, int, 0644);
7668 MODULE_PARM_DESC(zfs_arc_p_min_shift
, "arc_c shift to calc min/max arc_p");
7670 module_param(zfs_arc_average_blocksize
, int, 0444);
7671 MODULE_PARM_DESC(zfs_arc_average_blocksize
, "Target average block size");
7673 module_param(zfs_compressed_arc_enabled
, int, 0644);
7674 MODULE_PARM_DESC(zfs_arc_average_blocksize
, "Disable compressed arc buffers");
7676 module_param(zfs_arc_min_prefetch_lifespan
, int, 0644);
7677 MODULE_PARM_DESC(zfs_arc_min_prefetch_lifespan
, "Min life of prefetch block");
7679 module_param(zfs_arc_num_sublists_per_state
, int, 0644);
7680 MODULE_PARM_DESC(zfs_arc_num_sublists_per_state
,
7681 "Number of sublists used in each of the ARC state lists");
7683 module_param(l2arc_write_max
, ulong
, 0644);
7684 MODULE_PARM_DESC(l2arc_write_max
, "Max write bytes per interval");
7686 module_param(l2arc_write_boost
, ulong
, 0644);
7687 MODULE_PARM_DESC(l2arc_write_boost
, "Extra write bytes during device warmup");
7689 module_param(l2arc_headroom
, ulong
, 0644);
7690 MODULE_PARM_DESC(l2arc_headroom
, "Number of max device writes to precache");
7692 module_param(l2arc_headroom_boost
, ulong
, 0644);
7693 MODULE_PARM_DESC(l2arc_headroom_boost
, "Compressed l2arc_headroom multiplier");
7695 module_param(l2arc_feed_secs
, ulong
, 0644);
7696 MODULE_PARM_DESC(l2arc_feed_secs
, "Seconds between L2ARC writing");
7698 module_param(l2arc_feed_min_ms
, ulong
, 0644);
7699 MODULE_PARM_DESC(l2arc_feed_min_ms
, "Min feed interval in milliseconds");
7701 module_param(l2arc_noprefetch
, int, 0644);
7702 MODULE_PARM_DESC(l2arc_noprefetch
, "Skip caching prefetched buffers");
7704 module_param(l2arc_feed_again
, int, 0644);
7705 MODULE_PARM_DESC(l2arc_feed_again
, "Turbo L2ARC warmup");
7707 module_param(l2arc_norw
, int, 0644);
7708 MODULE_PARM_DESC(l2arc_norw
, "No reads during writes");
7710 module_param(zfs_arc_lotsfree_percent
, int, 0644);
7711 MODULE_PARM_DESC(zfs_arc_lotsfree_percent
,
7712 "System free memory I/O throttle in bytes");
7714 module_param(zfs_arc_sys_free
, ulong
, 0644);
7715 MODULE_PARM_DESC(zfs_arc_sys_free
, "System free memory target size in bytes");
7717 module_param(zfs_arc_dnode_limit
, ulong
, 0644);
7718 MODULE_PARM_DESC(zfs_arc_dnode_limit
, "Minimum bytes of dnodes in arc");
7720 module_param(zfs_arc_dnode_limit_percent
, ulong
, 0644);
7721 MODULE_PARM_DESC(zfs_arc_dnode_limit_percent
,
7722 "Percent of ARC meta buffers for dnodes");
7724 module_param(zfs_arc_dnode_reduce_percent
, ulong
, 0644);
7725 MODULE_PARM_DESC(zfs_arc_dnode_reduce_percent
,
7726 "Percentage of excess dnodes to try to unpin");