4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
24 * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
25 * Copyright (c) 2014 by Saso Kiselkov. All rights reserved.
26 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
30 * DVA-based Adjustable Replacement Cache
32 * While much of the theory of operation used here is
33 * based on the self-tuning, low overhead replacement cache
34 * presented by Megiddo and Modha at FAST 2003, there are some
35 * significant differences:
37 * 1. The Megiddo and Modha model assumes any page is evictable.
38 * Pages in its cache cannot be "locked" into memory. This makes
39 * the eviction algorithm simple: evict the last page in the list.
40 * This also make the performance characteristics easy to reason
41 * about. Our cache is not so simple. At any given moment, some
42 * subset of the blocks in the cache are un-evictable because we
43 * have handed out a reference to them. Blocks are only evictable
44 * when there are no external references active. This makes
45 * eviction far more problematic: we choose to evict the evictable
46 * blocks that are the "lowest" in the list.
48 * There are times when it is not possible to evict the requested
49 * space. In these circumstances we are unable to adjust the cache
50 * size. To prevent the cache growing unbounded at these times we
51 * implement a "cache throttle" that slows the flow of new data
52 * into the cache until we can make space available.
54 * 2. The Megiddo and Modha model assumes a fixed cache size.
55 * Pages are evicted when the cache is full and there is a cache
56 * miss. Our model has a variable sized cache. It grows with
57 * high use, but also tries to react to memory pressure from the
58 * operating system: decreasing its size when system memory is
61 * 3. The Megiddo and Modha model assumes a fixed page size. All
62 * elements of the cache are therefore exactly the same size. So
63 * when adjusting the cache size following a cache miss, its simply
64 * a matter of choosing a single page to evict. In our model, we
65 * have variable sized cache blocks (rangeing from 512 bytes to
66 * 128K bytes). We therefore choose a set of blocks to evict to make
67 * space for a cache miss that approximates as closely as possible
68 * the space used by the new block.
70 * See also: "ARC: A Self-Tuning, Low Overhead Replacement Cache"
71 * by N. Megiddo & D. Modha, FAST 2003
77 * A new reference to a cache buffer can be obtained in two
78 * ways: 1) via a hash table lookup using the DVA as a key,
79 * or 2) via one of the ARC lists. The arc_read() interface
80 * uses method 1, while the internal ARC algorithms for
81 * adjusting the cache use method 2. We therefore provide two
82 * types of locks: 1) the hash table lock array, and 2) the
85 * Buffers do not have their own mutexes, rather they rely on the
86 * hash table mutexes for the bulk of their protection (i.e. most
87 * fields in the arc_buf_hdr_t are protected by these mutexes).
89 * buf_hash_find() returns the appropriate mutex (held) when it
90 * locates the requested buffer in the hash table. It returns
91 * NULL for the mutex if the buffer was not in the table.
93 * buf_hash_remove() expects the appropriate hash mutex to be
94 * already held before it is invoked.
96 * Each ARC state also has a mutex which is used to protect the
97 * buffer list associated with the state. When attempting to
98 * obtain a hash table lock while holding an ARC list lock you
99 * must use: mutex_tryenter() to avoid deadlock. Also note that
100 * the active state mutex must be held before the ghost state mutex.
102 * It as also possible to register a callback which is run when the
103 * arc_meta_limit is reached and no buffers can be safely evicted. In
104 * this case the arc user should drop a reference on some arc buffers so
105 * they can be reclaimed and the arc_meta_limit honored. For example,
106 * when using the ZPL each dentry holds a references on a znode. These
107 * dentries must be pruned before the arc buffer holding the znode can
110 * Note that the majority of the performance stats are manipulated
111 * with atomic operations.
113 * The L2ARC uses the l2ad_mtx on each vdev for the following:
115 * - L2ARC buflist creation
116 * - L2ARC buflist eviction
117 * - L2ARC write completion, which walks L2ARC buflists
118 * - ARC header destruction, as it removes from L2ARC buflists
119 * - ARC header release, as it removes from L2ARC buflists
125 * Every block that is in the ARC is tracked by an arc_buf_hdr_t structure.
126 * This structure can point either to a block that is still in the cache or to
127 * one that is only accessible in an L2 ARC device, or it can provide
128 * information about a block that was recently evicted. If a block is
129 * only accessible in the L2ARC, then the arc_buf_hdr_t only has enough
130 * information to retrieve it from the L2ARC device. This information is
131 * stored in the l2arc_buf_hdr_t sub-structure of the arc_buf_hdr_t. A block
132 * that is in this state cannot access the data directly.
134 * Blocks that are actively being referenced or have not been evicted
135 * are cached in the L1ARC. The L1ARC (l1arc_buf_hdr_t) is a structure within
136 * the arc_buf_hdr_t that will point to the data block in memory. A block can
137 * only be read by a consumer if it has an l1arc_buf_hdr_t. The L1ARC
138 * caches data in two ways -- in a list of ARC buffers (arc_buf_t) and
139 * also in the arc_buf_hdr_t's private physical data block pointer (b_pabd).
141 * The L1ARC's data pointer may or may not be uncompressed. The ARC has the
142 * ability to store the physical data (b_pabd) associated with the DVA of the
143 * arc_buf_hdr_t. Since the b_pabd is a copy of the on-disk physical block,
144 * it will match its on-disk compression characteristics. This behavior can be
145 * disabled by setting 'zfs_compressed_arc_enabled' to B_FALSE. When the
146 * compressed ARC functionality is disabled, the b_pabd will point to an
147 * uncompressed version of the on-disk data.
149 * Data in the L1ARC is not accessed by consumers of the ARC directly. Each
150 * arc_buf_hdr_t can have multiple ARC buffers (arc_buf_t) which reference it.
151 * Each ARC buffer (arc_buf_t) is being actively accessed by a specific ARC
152 * consumer. The ARC will provide references to this data and will keep it
153 * cached until it is no longer in use. The ARC caches only the L1ARC's physical
154 * data block and will evict any arc_buf_t that is no longer referenced. The
155 * amount of memory consumed by the arc_buf_ts' data buffers can be seen via the
156 * "overhead_size" kstat.
158 * Depending on the consumer, an arc_buf_t can be requested in uncompressed or
159 * compressed form. The typical case is that consumers will want uncompressed
160 * data, and when that happens a new data buffer is allocated where the data is
161 * decompressed for them to use. Currently the only consumer who wants
162 * compressed arc_buf_t's is "zfs send", when it streams data exactly as it
163 * exists on disk. When this happens, the arc_buf_t's data buffer is shared
164 * with the arc_buf_hdr_t.
166 * Here is a diagram showing an arc_buf_hdr_t referenced by two arc_buf_t's. The
167 * first one is owned by a compressed send consumer (and therefore references
168 * the same compressed data buffer as the arc_buf_hdr_t) and the second could be
169 * used by any other consumer (and has its own uncompressed copy of the data
184 * | b_buf +------------>+-----------+ arc_buf_t
185 * | b_pabd +-+ |b_next +---->+-----------+
186 * +-----------+ | |-----------| |b_next +-->NULL
187 * | |b_comp = T | +-----------+
188 * | |b_data +-+ |b_comp = F |
189 * | +-----------+ | |b_data +-+
190 * +->+------+ | +-----------+ |
192 * data | |<--------------+ | uncompressed
193 * +------+ compressed, | data
194 * shared +-->+------+
199 * When a consumer reads a block, the ARC must first look to see if the
200 * arc_buf_hdr_t is cached. If the hdr is cached then the ARC allocates a new
201 * arc_buf_t and either copies uncompressed data into a new data buffer from an
202 * existing uncompressed arc_buf_t, decompresses the hdr's b_pabd buffer into a
203 * new data buffer, or shares the hdr's b_pabd buffer, depending on whether the
204 * hdr is compressed and the desired compression characteristics of the
205 * arc_buf_t consumer. If the arc_buf_t ends up sharing data with the
206 * arc_buf_hdr_t and both of them are uncompressed then the arc_buf_t must be
207 * the last buffer in the hdr's b_buf list, however a shared compressed buf can
208 * be anywhere in the hdr's list.
210 * The diagram below shows an example of an uncompressed ARC hdr that is
211 * sharing its data with an arc_buf_t (note that the shared uncompressed buf is
212 * the last element in the buf list):
224 * | | arc_buf_t (shared)
225 * | b_buf +------------>+---------+ arc_buf_t
226 * | | |b_next +---->+---------+
227 * | b_pabd +-+ |---------| |b_next +-->NULL
228 * +-----------+ | | | +---------+
230 * | +---------+ | |b_data +-+
231 * +->+------+ | +---------+ |
233 * uncompressed | | | |
236 * | uncompressed | | |
239 * +---------------------------------+
241 * Writing to the ARC requires that the ARC first discard the hdr's b_pabd
242 * since the physical block is about to be rewritten. The new data contents
243 * will be contained in the arc_buf_t. As the I/O pipeline performs the write,
244 * it may compress the data before writing it to disk. The ARC will be called
245 * with the transformed data and will bcopy the transformed on-disk block into
246 * a newly allocated b_pabd. Writes are always done into buffers which have
247 * either been loaned (and hence are new and don't have other readers) or
248 * buffers which have been released (and hence have their own hdr, if there
249 * were originally other readers of the buf's original hdr). This ensures that
250 * the ARC only needs to update a single buf and its hdr after a write occurs.
252 * When the L2ARC is in use, it will also take advantage of the b_pabd. The
253 * L2ARC will always write the contents of b_pabd to the L2ARC. This means
254 * that when compressed ARC is enabled that the L2ARC blocks are identical
255 * to the on-disk block in the main data pool. This provides a significant
256 * advantage since the ARC can leverage the bp's checksum when reading from the
257 * L2ARC to determine if the contents are valid. However, if the compressed
258 * ARC is disabled, then the L2ARC's block must be transformed to look
259 * like the physical block in the main data pool before comparing the
260 * checksum and determining its validity.
262 * The L1ARC has a slightly different system for storing encrypted data.
263 * Raw (encrypted + possibly compressed) data has a few subtle differences from
264 * data that is just compressed. The biggest difference is that it is not
265 * possible to decrypt encrypted data (or visa versa) if the keys aren't loaded.
266 * The other difference is that encryption cannot be treated as a suggestion.
267 * If a caller would prefer compressed data, but they actually wind up with
268 * uncompressed data the worst thing that could happen is there might be a
269 * performance hit. If the caller requests encrypted data, however, we must be
270 * sure they actually get it or else secret information could be leaked. Raw
271 * data is stored in hdr->b_crypt_hdr.b_rabd. An encrypted header, therefore,
272 * may have both an encrypted version and a decrypted version of its data at
273 * once. When a caller needs a raw arc_buf_t, it is allocated and the data is
274 * copied out of this header. To avoid complications with b_pabd, raw buffers
280 #include <sys/spa_impl.h>
281 #include <sys/zio_compress.h>
282 #include <sys/zio_checksum.h>
283 #include <sys/zfs_context.h>
285 #include <sys/refcount.h>
286 #include <sys/vdev.h>
287 #include <sys/vdev_impl.h>
288 #include <sys/dsl_pool.h>
289 #include <sys/zio_checksum.h>
290 #include <sys/multilist.h>
293 #include <sys/fm/fs/zfs.h>
295 #include <sys/vmsystm.h>
297 #include <sys/fs/swapnode.h>
299 #include <linux/mm_compat.h>
300 #include <linux/page_compat.h>
302 #include <sys/callb.h>
303 #include <sys/kstat.h>
304 #include <sys/dmu_tx.h>
305 #include <zfs_fletcher.h>
306 #include <sys/arc_impl.h>
307 #include <sys/trace_arc.h>
310 /* set with ZFS_DEBUG=watch, to enable watchpoints on frozen buffers */
311 boolean_t arc_watch
= B_FALSE
;
314 static kmutex_t arc_reclaim_lock
;
315 static kcondvar_t arc_reclaim_thread_cv
;
316 static boolean_t arc_reclaim_thread_exit
;
317 static kcondvar_t arc_reclaim_waiters_cv
;
320 * The number of headers to evict in arc_evict_state_impl() before
321 * dropping the sublist lock and evicting from another sublist. A lower
322 * value means we're more likely to evict the "correct" header (i.e. the
323 * oldest header in the arc state), but comes with higher overhead
324 * (i.e. more invocations of arc_evict_state_impl()).
326 int zfs_arc_evict_batch_limit
= 10;
328 /* number of seconds before growing cache again */
329 static int arc_grow_retry
= 5;
331 /* shift of arc_c for calculating overflow limit in arc_get_data_impl */
332 int zfs_arc_overflow_shift
= 8;
334 /* shift of arc_c for calculating both min and max arc_p */
335 static int arc_p_min_shift
= 4;
337 /* log2(fraction of arc to reclaim) */
338 static int arc_shrink_shift
= 7;
340 /* percent of pagecache to reclaim arc to */
342 static uint_t zfs_arc_pc_percent
= 0;
346 * log2(fraction of ARC which must be free to allow growing).
347 * I.e. If there is less than arc_c >> arc_no_grow_shift free memory,
348 * when reading a new block into the ARC, we will evict an equal-sized block
351 * This must be less than arc_shrink_shift, so that when we shrink the ARC,
352 * we will still not allow it to grow.
354 int arc_no_grow_shift
= 5;
358 * minimum lifespan of a prefetch block in clock ticks
359 * (initialized in arc_init())
361 static int arc_min_prefetch_ms
;
362 static int arc_min_prescient_prefetch_ms
;
365 * If this percent of memory is free, don't throttle.
367 int arc_lotsfree_percent
= 10;
372 * The arc has filled available memory and has now warmed up.
374 static boolean_t arc_warm
;
377 * log2 fraction of the zio arena to keep free.
379 int arc_zio_arena_free_shift
= 2;
382 * These tunables are for performance analysis.
384 unsigned long zfs_arc_max
= 0;
385 unsigned long zfs_arc_min
= 0;
386 unsigned long zfs_arc_meta_limit
= 0;
387 unsigned long zfs_arc_meta_min
= 0;
388 unsigned long zfs_arc_dnode_limit
= 0;
389 unsigned long zfs_arc_dnode_reduce_percent
= 10;
390 int zfs_arc_grow_retry
= 0;
391 int zfs_arc_shrink_shift
= 0;
392 int zfs_arc_p_min_shift
= 0;
393 int zfs_arc_average_blocksize
= 8 * 1024; /* 8KB */
395 int zfs_compressed_arc_enabled
= B_TRUE
;
398 * ARC will evict meta buffers that exceed arc_meta_limit. This
399 * tunable make arc_meta_limit adjustable for different workloads.
401 unsigned long zfs_arc_meta_limit_percent
= 75;
404 * Percentage that can be consumed by dnodes of ARC meta buffers.
406 unsigned long zfs_arc_dnode_limit_percent
= 10;
409 * These tunables are Linux specific
411 unsigned long zfs_arc_sys_free
= 0;
412 int zfs_arc_min_prefetch_ms
= 0;
413 int zfs_arc_min_prescient_prefetch_ms
= 0;
414 int zfs_arc_p_dampener_disable
= 1;
415 int zfs_arc_meta_prune
= 10000;
416 int zfs_arc_meta_strategy
= ARC_STRATEGY_META_BALANCED
;
417 int zfs_arc_meta_adjust_restarts
= 4096;
418 int zfs_arc_lotsfree_percent
= 10;
421 static arc_state_t ARC_anon
;
422 static arc_state_t ARC_mru
;
423 static arc_state_t ARC_mru_ghost
;
424 static arc_state_t ARC_mfu
;
425 static arc_state_t ARC_mfu_ghost
;
426 static arc_state_t ARC_l2c_only
;
428 typedef struct arc_stats
{
429 kstat_named_t arcstat_hits
;
430 kstat_named_t arcstat_misses
;
431 kstat_named_t arcstat_demand_data_hits
;
432 kstat_named_t arcstat_demand_data_misses
;
433 kstat_named_t arcstat_demand_metadata_hits
;
434 kstat_named_t arcstat_demand_metadata_misses
;
435 kstat_named_t arcstat_prefetch_data_hits
;
436 kstat_named_t arcstat_prefetch_data_misses
;
437 kstat_named_t arcstat_prefetch_metadata_hits
;
438 kstat_named_t arcstat_prefetch_metadata_misses
;
439 kstat_named_t arcstat_mru_hits
;
440 kstat_named_t arcstat_mru_ghost_hits
;
441 kstat_named_t arcstat_mfu_hits
;
442 kstat_named_t arcstat_mfu_ghost_hits
;
443 kstat_named_t arcstat_deleted
;
445 * Number of buffers that could not be evicted because the hash lock
446 * was held by another thread. The lock may not necessarily be held
447 * by something using the same buffer, since hash locks are shared
448 * by multiple buffers.
450 kstat_named_t arcstat_mutex_miss
;
452 * Number of buffers skipped when updating the access state due to the
453 * header having already been released after acquiring the hash lock.
455 kstat_named_t arcstat_access_skip
;
457 * Number of buffers skipped because they have I/O in progress, are
458 * indirect prefetch buffers that have not lived long enough, or are
459 * not from the spa we're trying to evict from.
461 kstat_named_t arcstat_evict_skip
;
463 * Number of times arc_evict_state() was unable to evict enough
464 * buffers to reach its target amount.
466 kstat_named_t arcstat_evict_not_enough
;
467 kstat_named_t arcstat_evict_l2_cached
;
468 kstat_named_t arcstat_evict_l2_eligible
;
469 kstat_named_t arcstat_evict_l2_ineligible
;
470 kstat_named_t arcstat_evict_l2_skip
;
471 kstat_named_t arcstat_hash_elements
;
472 kstat_named_t arcstat_hash_elements_max
;
473 kstat_named_t arcstat_hash_collisions
;
474 kstat_named_t arcstat_hash_chains
;
475 kstat_named_t arcstat_hash_chain_max
;
476 kstat_named_t arcstat_p
;
477 kstat_named_t arcstat_c
;
478 kstat_named_t arcstat_c_min
;
479 kstat_named_t arcstat_c_max
;
480 kstat_named_t arcstat_size
;
482 * Number of compressed bytes stored in the arc_buf_hdr_t's b_pabd.
483 * Note that the compressed bytes may match the uncompressed bytes
484 * if the block is either not compressed or compressed arc is disabled.
486 kstat_named_t arcstat_compressed_size
;
488 * Uncompressed size of the data stored in b_pabd. If compressed
489 * arc is disabled then this value will be identical to the stat
492 kstat_named_t arcstat_uncompressed_size
;
494 * Number of bytes stored in all the arc_buf_t's. This is classified
495 * as "overhead" since this data is typically short-lived and will
496 * be evicted from the arc when it becomes unreferenced unless the
497 * zfs_keep_uncompressed_metadata or zfs_keep_uncompressed_level
498 * values have been set (see comment in dbuf.c for more information).
500 kstat_named_t arcstat_overhead_size
;
502 * Number of bytes consumed by internal ARC structures necessary
503 * for tracking purposes; these structures are not actually
504 * backed by ARC buffers. This includes arc_buf_hdr_t structures
505 * (allocated via arc_buf_hdr_t_full and arc_buf_hdr_t_l2only
506 * caches), and arc_buf_t structures (allocated via arc_buf_t
509 kstat_named_t arcstat_hdr_size
;
511 * Number of bytes consumed by ARC buffers of type equal to
512 * ARC_BUFC_DATA. This is generally consumed by buffers backing
513 * on disk user data (e.g. plain file contents).
515 kstat_named_t arcstat_data_size
;
517 * Number of bytes consumed by ARC buffers of type equal to
518 * ARC_BUFC_METADATA. This is generally consumed by buffers
519 * backing on disk data that is used for internal ZFS
520 * structures (e.g. ZAP, dnode, indirect blocks, etc).
522 kstat_named_t arcstat_metadata_size
;
524 * Number of bytes consumed by dmu_buf_impl_t objects.
526 kstat_named_t arcstat_dbuf_size
;
528 * Number of bytes consumed by dnode_t objects.
530 kstat_named_t arcstat_dnode_size
;
532 * Number of bytes consumed by bonus buffers.
534 kstat_named_t arcstat_bonus_size
;
536 * Total number of bytes consumed by ARC buffers residing in the
537 * arc_anon state. This includes *all* buffers in the arc_anon
538 * state; e.g. data, metadata, evictable, and unevictable buffers
539 * are all included in this value.
541 kstat_named_t arcstat_anon_size
;
543 * Number of bytes consumed by ARC buffers that meet the
544 * following criteria: backing buffers of type ARC_BUFC_DATA,
545 * residing in the arc_anon state, and are eligible for eviction
546 * (e.g. have no outstanding holds on the buffer).
548 kstat_named_t arcstat_anon_evictable_data
;
550 * Number of bytes consumed by ARC buffers that meet the
551 * following criteria: backing buffers of type ARC_BUFC_METADATA,
552 * residing in the arc_anon state, and are eligible for eviction
553 * (e.g. have no outstanding holds on the buffer).
555 kstat_named_t arcstat_anon_evictable_metadata
;
557 * Total number of bytes consumed by ARC buffers residing in the
558 * arc_mru state. This includes *all* buffers in the arc_mru
559 * state; e.g. data, metadata, evictable, and unevictable buffers
560 * are all included in this value.
562 kstat_named_t arcstat_mru_size
;
564 * Number of bytes consumed by ARC buffers that meet the
565 * following criteria: backing buffers of type ARC_BUFC_DATA,
566 * residing in the arc_mru state, and are eligible for eviction
567 * (e.g. have no outstanding holds on the buffer).
569 kstat_named_t arcstat_mru_evictable_data
;
571 * Number of bytes consumed by ARC buffers that meet the
572 * following criteria: backing buffers of type ARC_BUFC_METADATA,
573 * residing in the arc_mru state, and are eligible for eviction
574 * (e.g. have no outstanding holds on the buffer).
576 kstat_named_t arcstat_mru_evictable_metadata
;
578 * Total number of bytes that *would have been* consumed by ARC
579 * buffers in the arc_mru_ghost state. The key thing to note
580 * here, is the fact that this size doesn't actually indicate
581 * RAM consumption. The ghost lists only consist of headers and
582 * don't actually have ARC buffers linked off of these headers.
583 * Thus, *if* the headers had associated ARC buffers, these
584 * buffers *would have* consumed this number of bytes.
586 kstat_named_t arcstat_mru_ghost_size
;
588 * Number of bytes that *would have been* consumed by ARC
589 * buffers that are eligible for eviction, of type
590 * ARC_BUFC_DATA, and linked off the arc_mru_ghost state.
592 kstat_named_t arcstat_mru_ghost_evictable_data
;
594 * Number of bytes that *would have been* consumed by ARC
595 * buffers that are eligible for eviction, of type
596 * ARC_BUFC_METADATA, and linked off the arc_mru_ghost state.
598 kstat_named_t arcstat_mru_ghost_evictable_metadata
;
600 * Total number of bytes consumed by ARC buffers residing in the
601 * arc_mfu state. This includes *all* buffers in the arc_mfu
602 * state; e.g. data, metadata, evictable, and unevictable buffers
603 * are all included in this value.
605 kstat_named_t arcstat_mfu_size
;
607 * Number of bytes consumed by ARC buffers that are eligible for
608 * eviction, of type ARC_BUFC_DATA, and reside in the arc_mfu
611 kstat_named_t arcstat_mfu_evictable_data
;
613 * Number of bytes consumed by ARC buffers that are eligible for
614 * eviction, of type ARC_BUFC_METADATA, and reside in the
617 kstat_named_t arcstat_mfu_evictable_metadata
;
619 * Total number of bytes that *would have been* consumed by ARC
620 * buffers in the arc_mfu_ghost state. See the comment above
621 * arcstat_mru_ghost_size for more details.
623 kstat_named_t arcstat_mfu_ghost_size
;
625 * Number of bytes that *would have been* consumed by ARC
626 * buffers that are eligible for eviction, of type
627 * ARC_BUFC_DATA, and linked off the arc_mfu_ghost state.
629 kstat_named_t arcstat_mfu_ghost_evictable_data
;
631 * Number of bytes that *would have been* consumed by ARC
632 * buffers that are eligible for eviction, of type
633 * ARC_BUFC_METADATA, and linked off the arc_mru_ghost state.
635 kstat_named_t arcstat_mfu_ghost_evictable_metadata
;
636 kstat_named_t arcstat_l2_hits
;
637 kstat_named_t arcstat_l2_misses
;
638 kstat_named_t arcstat_l2_feeds
;
639 kstat_named_t arcstat_l2_rw_clash
;
640 kstat_named_t arcstat_l2_read_bytes
;
641 kstat_named_t arcstat_l2_write_bytes
;
642 kstat_named_t arcstat_l2_writes_sent
;
643 kstat_named_t arcstat_l2_writes_done
;
644 kstat_named_t arcstat_l2_writes_error
;
645 kstat_named_t arcstat_l2_writes_lock_retry
;
646 kstat_named_t arcstat_l2_evict_lock_retry
;
647 kstat_named_t arcstat_l2_evict_reading
;
648 kstat_named_t arcstat_l2_evict_l1cached
;
649 kstat_named_t arcstat_l2_free_on_write
;
650 kstat_named_t arcstat_l2_abort_lowmem
;
651 kstat_named_t arcstat_l2_cksum_bad
;
652 kstat_named_t arcstat_l2_io_error
;
653 kstat_named_t arcstat_l2_lsize
;
654 kstat_named_t arcstat_l2_psize
;
655 kstat_named_t arcstat_l2_hdr_size
;
656 kstat_named_t arcstat_memory_throttle_count
;
657 kstat_named_t arcstat_memory_direct_count
;
658 kstat_named_t arcstat_memory_indirect_count
;
659 kstat_named_t arcstat_memory_all_bytes
;
660 kstat_named_t arcstat_memory_free_bytes
;
661 kstat_named_t arcstat_memory_available_bytes
;
662 kstat_named_t arcstat_no_grow
;
663 kstat_named_t arcstat_tempreserve
;
664 kstat_named_t arcstat_loaned_bytes
;
665 kstat_named_t arcstat_prune
;
666 kstat_named_t arcstat_meta_used
;
667 kstat_named_t arcstat_meta_limit
;
668 kstat_named_t arcstat_dnode_limit
;
669 kstat_named_t arcstat_meta_max
;
670 kstat_named_t arcstat_meta_min
;
671 kstat_named_t arcstat_async_upgrade_sync
;
672 kstat_named_t arcstat_demand_hit_predictive_prefetch
;
673 kstat_named_t arcstat_demand_hit_prescient_prefetch
;
674 kstat_named_t arcstat_need_free
;
675 kstat_named_t arcstat_sys_free
;
676 kstat_named_t arcstat_raw_size
;
679 static arc_stats_t arc_stats
= {
680 { "hits", KSTAT_DATA_UINT64
},
681 { "misses", KSTAT_DATA_UINT64
},
682 { "demand_data_hits", KSTAT_DATA_UINT64
},
683 { "demand_data_misses", KSTAT_DATA_UINT64
},
684 { "demand_metadata_hits", KSTAT_DATA_UINT64
},
685 { "demand_metadata_misses", KSTAT_DATA_UINT64
},
686 { "prefetch_data_hits", KSTAT_DATA_UINT64
},
687 { "prefetch_data_misses", KSTAT_DATA_UINT64
},
688 { "prefetch_metadata_hits", KSTAT_DATA_UINT64
},
689 { "prefetch_metadata_misses", KSTAT_DATA_UINT64
},
690 { "mru_hits", KSTAT_DATA_UINT64
},
691 { "mru_ghost_hits", KSTAT_DATA_UINT64
},
692 { "mfu_hits", KSTAT_DATA_UINT64
},
693 { "mfu_ghost_hits", KSTAT_DATA_UINT64
},
694 { "deleted", KSTAT_DATA_UINT64
},
695 { "mutex_miss", KSTAT_DATA_UINT64
},
696 { "access_skip", KSTAT_DATA_UINT64
},
697 { "evict_skip", KSTAT_DATA_UINT64
},
698 { "evict_not_enough", KSTAT_DATA_UINT64
},
699 { "evict_l2_cached", KSTAT_DATA_UINT64
},
700 { "evict_l2_eligible", KSTAT_DATA_UINT64
},
701 { "evict_l2_ineligible", KSTAT_DATA_UINT64
},
702 { "evict_l2_skip", KSTAT_DATA_UINT64
},
703 { "hash_elements", KSTAT_DATA_UINT64
},
704 { "hash_elements_max", KSTAT_DATA_UINT64
},
705 { "hash_collisions", KSTAT_DATA_UINT64
},
706 { "hash_chains", KSTAT_DATA_UINT64
},
707 { "hash_chain_max", KSTAT_DATA_UINT64
},
708 { "p", KSTAT_DATA_UINT64
},
709 { "c", KSTAT_DATA_UINT64
},
710 { "c_min", KSTAT_DATA_UINT64
},
711 { "c_max", KSTAT_DATA_UINT64
},
712 { "size", KSTAT_DATA_UINT64
},
713 { "compressed_size", KSTAT_DATA_UINT64
},
714 { "uncompressed_size", KSTAT_DATA_UINT64
},
715 { "overhead_size", KSTAT_DATA_UINT64
},
716 { "hdr_size", KSTAT_DATA_UINT64
},
717 { "data_size", KSTAT_DATA_UINT64
},
718 { "metadata_size", KSTAT_DATA_UINT64
},
719 { "dbuf_size", KSTAT_DATA_UINT64
},
720 { "dnode_size", KSTAT_DATA_UINT64
},
721 { "bonus_size", KSTAT_DATA_UINT64
},
722 { "anon_size", KSTAT_DATA_UINT64
},
723 { "anon_evictable_data", KSTAT_DATA_UINT64
},
724 { "anon_evictable_metadata", KSTAT_DATA_UINT64
},
725 { "mru_size", KSTAT_DATA_UINT64
},
726 { "mru_evictable_data", KSTAT_DATA_UINT64
},
727 { "mru_evictable_metadata", KSTAT_DATA_UINT64
},
728 { "mru_ghost_size", KSTAT_DATA_UINT64
},
729 { "mru_ghost_evictable_data", KSTAT_DATA_UINT64
},
730 { "mru_ghost_evictable_metadata", KSTAT_DATA_UINT64
},
731 { "mfu_size", KSTAT_DATA_UINT64
},
732 { "mfu_evictable_data", KSTAT_DATA_UINT64
},
733 { "mfu_evictable_metadata", KSTAT_DATA_UINT64
},
734 { "mfu_ghost_size", KSTAT_DATA_UINT64
},
735 { "mfu_ghost_evictable_data", KSTAT_DATA_UINT64
},
736 { "mfu_ghost_evictable_metadata", KSTAT_DATA_UINT64
},
737 { "l2_hits", KSTAT_DATA_UINT64
},
738 { "l2_misses", KSTAT_DATA_UINT64
},
739 { "l2_feeds", KSTAT_DATA_UINT64
},
740 { "l2_rw_clash", KSTAT_DATA_UINT64
},
741 { "l2_read_bytes", KSTAT_DATA_UINT64
},
742 { "l2_write_bytes", KSTAT_DATA_UINT64
},
743 { "l2_writes_sent", KSTAT_DATA_UINT64
},
744 { "l2_writes_done", KSTAT_DATA_UINT64
},
745 { "l2_writes_error", KSTAT_DATA_UINT64
},
746 { "l2_writes_lock_retry", KSTAT_DATA_UINT64
},
747 { "l2_evict_lock_retry", KSTAT_DATA_UINT64
},
748 { "l2_evict_reading", KSTAT_DATA_UINT64
},
749 { "l2_evict_l1cached", KSTAT_DATA_UINT64
},
750 { "l2_free_on_write", KSTAT_DATA_UINT64
},
751 { "l2_abort_lowmem", KSTAT_DATA_UINT64
},
752 { "l2_cksum_bad", KSTAT_DATA_UINT64
},
753 { "l2_io_error", KSTAT_DATA_UINT64
},
754 { "l2_size", KSTAT_DATA_UINT64
},
755 { "l2_asize", KSTAT_DATA_UINT64
},
756 { "l2_hdr_size", KSTAT_DATA_UINT64
},
757 { "memory_throttle_count", KSTAT_DATA_UINT64
},
758 { "memory_direct_count", KSTAT_DATA_UINT64
},
759 { "memory_indirect_count", KSTAT_DATA_UINT64
},
760 { "memory_all_bytes", KSTAT_DATA_UINT64
},
761 { "memory_free_bytes", KSTAT_DATA_UINT64
},
762 { "memory_available_bytes", KSTAT_DATA_INT64
},
763 { "arc_no_grow", KSTAT_DATA_UINT64
},
764 { "arc_tempreserve", KSTAT_DATA_UINT64
},
765 { "arc_loaned_bytes", KSTAT_DATA_UINT64
},
766 { "arc_prune", KSTAT_DATA_UINT64
},
767 { "arc_meta_used", KSTAT_DATA_UINT64
},
768 { "arc_meta_limit", KSTAT_DATA_UINT64
},
769 { "arc_dnode_limit", KSTAT_DATA_UINT64
},
770 { "arc_meta_max", KSTAT_DATA_UINT64
},
771 { "arc_meta_min", KSTAT_DATA_UINT64
},
772 { "async_upgrade_sync", KSTAT_DATA_UINT64
},
773 { "demand_hit_predictive_prefetch", KSTAT_DATA_UINT64
},
774 { "demand_hit_prescient_prefetch", KSTAT_DATA_UINT64
},
775 { "arc_need_free", KSTAT_DATA_UINT64
},
776 { "arc_sys_free", KSTAT_DATA_UINT64
},
777 { "arc_raw_size", KSTAT_DATA_UINT64
}
780 #define ARCSTAT(stat) (arc_stats.stat.value.ui64)
782 #define ARCSTAT_INCR(stat, val) \
783 atomic_add_64(&arc_stats.stat.value.ui64, (val))
785 #define ARCSTAT_BUMP(stat) ARCSTAT_INCR(stat, 1)
786 #define ARCSTAT_BUMPDOWN(stat) ARCSTAT_INCR(stat, -1)
788 #define ARCSTAT_MAX(stat, val) { \
790 while ((val) > (m = arc_stats.stat.value.ui64) && \
791 (m != atomic_cas_64(&arc_stats.stat.value.ui64, m, (val)))) \
795 #define ARCSTAT_MAXSTAT(stat) \
796 ARCSTAT_MAX(stat##_max, arc_stats.stat.value.ui64)
799 * We define a macro to allow ARC hits/misses to be easily broken down by
800 * two separate conditions, giving a total of four different subtypes for
801 * each of hits and misses (so eight statistics total).
803 #define ARCSTAT_CONDSTAT(cond1, stat1, notstat1, cond2, stat2, notstat2, stat) \
806 ARCSTAT_BUMP(arcstat_##stat1##_##stat2##_##stat); \
808 ARCSTAT_BUMP(arcstat_##stat1##_##notstat2##_##stat); \
812 ARCSTAT_BUMP(arcstat_##notstat1##_##stat2##_##stat); \
814 ARCSTAT_BUMP(arcstat_##notstat1##_##notstat2##_##stat);\
819 static arc_state_t
*arc_anon
;
820 static arc_state_t
*arc_mru
;
821 static arc_state_t
*arc_mru_ghost
;
822 static arc_state_t
*arc_mfu
;
823 static arc_state_t
*arc_mfu_ghost
;
824 static arc_state_t
*arc_l2c_only
;
827 * There are several ARC variables that are critical to export as kstats --
828 * but we don't want to have to grovel around in the kstat whenever we wish to
829 * manipulate them. For these variables, we therefore define them to be in
830 * terms of the statistic variable. This assures that we are not introducing
831 * the possibility of inconsistency by having shadow copies of the variables,
832 * while still allowing the code to be readable.
834 #define arc_size ARCSTAT(arcstat_size) /* actual total arc size */
835 #define arc_p ARCSTAT(arcstat_p) /* target size of MRU */
836 #define arc_c ARCSTAT(arcstat_c) /* target size of cache */
837 #define arc_c_min ARCSTAT(arcstat_c_min) /* min target cache size */
838 #define arc_c_max ARCSTAT(arcstat_c_max) /* max target cache size */
839 #define arc_no_grow ARCSTAT(arcstat_no_grow) /* do not grow cache size */
840 #define arc_tempreserve ARCSTAT(arcstat_tempreserve)
841 #define arc_loaned_bytes ARCSTAT(arcstat_loaned_bytes)
842 #define arc_meta_limit ARCSTAT(arcstat_meta_limit) /* max size for metadata */
843 #define arc_dnode_limit ARCSTAT(arcstat_dnode_limit) /* max size for dnodes */
844 #define arc_meta_min ARCSTAT(arcstat_meta_min) /* min size for metadata */
845 #define arc_meta_used ARCSTAT(arcstat_meta_used) /* size of metadata */
846 #define arc_meta_max ARCSTAT(arcstat_meta_max) /* max size of metadata */
847 #define arc_dbuf_size ARCSTAT(arcstat_dbuf_size) /* dbuf metadata */
848 #define arc_dnode_size ARCSTAT(arcstat_dnode_size) /* dnode metadata */
849 #define arc_bonus_size ARCSTAT(arcstat_bonus_size) /* bonus buffer metadata */
850 #define arc_need_free ARCSTAT(arcstat_need_free) /* bytes to be freed */
851 #define arc_sys_free ARCSTAT(arcstat_sys_free) /* target system free bytes */
853 /* size of all b_rabd's in entire arc */
854 #define arc_raw_size ARCSTAT(arcstat_raw_size)
855 /* compressed size of entire arc */
856 #define arc_compressed_size ARCSTAT(arcstat_compressed_size)
857 /* uncompressed size of entire arc */
858 #define arc_uncompressed_size ARCSTAT(arcstat_uncompressed_size)
859 /* number of bytes in the arc from arc_buf_t's */
860 #define arc_overhead_size ARCSTAT(arcstat_overhead_size)
862 static list_t arc_prune_list
;
863 static kmutex_t arc_prune_mtx
;
864 static taskq_t
*arc_prune_taskq
;
866 #define GHOST_STATE(state) \
867 ((state) == arc_mru_ghost || (state) == arc_mfu_ghost || \
868 (state) == arc_l2c_only)
870 #define HDR_IN_HASH_TABLE(hdr) ((hdr)->b_flags & ARC_FLAG_IN_HASH_TABLE)
871 #define HDR_IO_IN_PROGRESS(hdr) ((hdr)->b_flags & ARC_FLAG_IO_IN_PROGRESS)
872 #define HDR_IO_ERROR(hdr) ((hdr)->b_flags & ARC_FLAG_IO_ERROR)
873 #define HDR_PREFETCH(hdr) ((hdr)->b_flags & ARC_FLAG_PREFETCH)
874 #define HDR_PRESCIENT_PREFETCH(hdr) \
875 ((hdr)->b_flags & ARC_FLAG_PRESCIENT_PREFETCH)
876 #define HDR_COMPRESSION_ENABLED(hdr) \
877 ((hdr)->b_flags & ARC_FLAG_COMPRESSED_ARC)
879 #define HDR_L2CACHE(hdr) ((hdr)->b_flags & ARC_FLAG_L2CACHE)
880 #define HDR_L2_READING(hdr) \
881 (((hdr)->b_flags & ARC_FLAG_IO_IN_PROGRESS) && \
882 ((hdr)->b_flags & ARC_FLAG_HAS_L2HDR))
883 #define HDR_L2_WRITING(hdr) ((hdr)->b_flags & ARC_FLAG_L2_WRITING)
884 #define HDR_L2_EVICTED(hdr) ((hdr)->b_flags & ARC_FLAG_L2_EVICTED)
885 #define HDR_L2_WRITE_HEAD(hdr) ((hdr)->b_flags & ARC_FLAG_L2_WRITE_HEAD)
886 #define HDR_PROTECTED(hdr) ((hdr)->b_flags & ARC_FLAG_PROTECTED)
887 #define HDR_NOAUTH(hdr) ((hdr)->b_flags & ARC_FLAG_NOAUTH)
888 #define HDR_SHARED_DATA(hdr) ((hdr)->b_flags & ARC_FLAG_SHARED_DATA)
890 #define HDR_ISTYPE_METADATA(hdr) \
891 ((hdr)->b_flags & ARC_FLAG_BUFC_METADATA)
892 #define HDR_ISTYPE_DATA(hdr) (!HDR_ISTYPE_METADATA(hdr))
894 #define HDR_HAS_L1HDR(hdr) ((hdr)->b_flags & ARC_FLAG_HAS_L1HDR)
895 #define HDR_HAS_L2HDR(hdr) ((hdr)->b_flags & ARC_FLAG_HAS_L2HDR)
896 #define HDR_HAS_RABD(hdr) \
897 (HDR_HAS_L1HDR(hdr) && HDR_PROTECTED(hdr) && \
898 (hdr)->b_crypt_hdr.b_rabd != NULL)
899 #define HDR_ENCRYPTED(hdr) \
900 (HDR_PROTECTED(hdr) && DMU_OT_IS_ENCRYPTED((hdr)->b_crypt_hdr.b_ot))
901 #define HDR_AUTHENTICATED(hdr) \
902 (HDR_PROTECTED(hdr) && !DMU_OT_IS_ENCRYPTED((hdr)->b_crypt_hdr.b_ot))
904 /* For storing compression mode in b_flags */
905 #define HDR_COMPRESS_OFFSET (highbit64(ARC_FLAG_COMPRESS_0) - 1)
907 #define HDR_GET_COMPRESS(hdr) ((enum zio_compress)BF32_GET((hdr)->b_flags, \
908 HDR_COMPRESS_OFFSET, SPA_COMPRESSBITS))
909 #define HDR_SET_COMPRESS(hdr, cmp) BF32_SET((hdr)->b_flags, \
910 HDR_COMPRESS_OFFSET, SPA_COMPRESSBITS, (cmp));
912 #define ARC_BUF_LAST(buf) ((buf)->b_next == NULL)
913 #define ARC_BUF_SHARED(buf) ((buf)->b_flags & ARC_BUF_FLAG_SHARED)
914 #define ARC_BUF_COMPRESSED(buf) ((buf)->b_flags & ARC_BUF_FLAG_COMPRESSED)
915 #define ARC_BUF_ENCRYPTED(buf) ((buf)->b_flags & ARC_BUF_FLAG_ENCRYPTED)
921 #define HDR_FULL_CRYPT_SIZE ((int64_t)sizeof (arc_buf_hdr_t))
922 #define HDR_FULL_SIZE ((int64_t)offsetof(arc_buf_hdr_t, b_crypt_hdr))
923 #define HDR_L2ONLY_SIZE ((int64_t)offsetof(arc_buf_hdr_t, b_l1hdr))
926 * Hash table routines
929 #define HT_LOCK_ALIGN 64
930 #define HT_LOCK_PAD (P2NPHASE(sizeof (kmutex_t), (HT_LOCK_ALIGN)))
935 unsigned char pad
[HT_LOCK_PAD
];
939 #define BUF_LOCKS 8192
940 typedef struct buf_hash_table
{
942 arc_buf_hdr_t
**ht_table
;
943 struct ht_lock ht_locks
[BUF_LOCKS
];
946 static buf_hash_table_t buf_hash_table
;
948 #define BUF_HASH_INDEX(spa, dva, birth) \
949 (buf_hash(spa, dva, birth) & buf_hash_table.ht_mask)
950 #define BUF_HASH_LOCK_NTRY(idx) (buf_hash_table.ht_locks[idx & (BUF_LOCKS-1)])
951 #define BUF_HASH_LOCK(idx) (&(BUF_HASH_LOCK_NTRY(idx).ht_lock))
952 #define HDR_LOCK(hdr) \
953 (BUF_HASH_LOCK(BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth)))
955 uint64_t zfs_crc64_table
[256];
961 #define L2ARC_WRITE_SIZE (8 * 1024 * 1024) /* initial write max */
962 #define L2ARC_HEADROOM 2 /* num of writes */
965 * If we discover during ARC scan any buffers to be compressed, we boost
966 * our headroom for the next scanning cycle by this percentage multiple.
968 #define L2ARC_HEADROOM_BOOST 200
969 #define L2ARC_FEED_SECS 1 /* caching interval secs */
970 #define L2ARC_FEED_MIN_MS 200 /* min caching interval ms */
973 * We can feed L2ARC from two states of ARC buffers, mru and mfu,
974 * and each of the state has two types: data and metadata.
976 #define L2ARC_FEED_TYPES 4
978 #define l2arc_writes_sent ARCSTAT(arcstat_l2_writes_sent)
979 #define l2arc_writes_done ARCSTAT(arcstat_l2_writes_done)
981 /* L2ARC Performance Tunables */
982 unsigned long l2arc_write_max
= L2ARC_WRITE_SIZE
; /* def max write size */
983 unsigned long l2arc_write_boost
= L2ARC_WRITE_SIZE
; /* extra warmup write */
984 unsigned long l2arc_headroom
= L2ARC_HEADROOM
; /* # of dev writes */
985 unsigned long l2arc_headroom_boost
= L2ARC_HEADROOM_BOOST
;
986 unsigned long l2arc_feed_secs
= L2ARC_FEED_SECS
; /* interval seconds */
987 unsigned long l2arc_feed_min_ms
= L2ARC_FEED_MIN_MS
; /* min interval msecs */
988 int l2arc_noprefetch
= B_TRUE
; /* don't cache prefetch bufs */
989 int l2arc_feed_again
= B_TRUE
; /* turbo warmup */
990 int l2arc_norw
= B_FALSE
; /* no reads during writes */
995 static list_t L2ARC_dev_list
; /* device list */
996 static list_t
*l2arc_dev_list
; /* device list pointer */
997 static kmutex_t l2arc_dev_mtx
; /* device list mutex */
998 static l2arc_dev_t
*l2arc_dev_last
; /* last device used */
999 static list_t L2ARC_free_on_write
; /* free after write buf list */
1000 static list_t
*l2arc_free_on_write
; /* free after write list ptr */
1001 static kmutex_t l2arc_free_on_write_mtx
; /* mutex for list */
1002 static uint64_t l2arc_ndev
; /* number of devices */
1004 typedef struct l2arc_read_callback
{
1005 arc_buf_hdr_t
*l2rcb_hdr
; /* read header */
1006 blkptr_t l2rcb_bp
; /* original blkptr */
1007 zbookmark_phys_t l2rcb_zb
; /* original bookmark */
1008 int l2rcb_flags
; /* original flags */
1009 abd_t
*l2rcb_abd
; /* temporary buffer */
1010 } l2arc_read_callback_t
;
1012 typedef struct l2arc_data_free
{
1013 /* protected by l2arc_free_on_write_mtx */
1016 arc_buf_contents_t l2df_type
;
1017 list_node_t l2df_list_node
;
1018 } l2arc_data_free_t
;
1020 typedef enum arc_fill_flags
{
1021 ARC_FILL_LOCKED
= 1 << 0, /* hdr lock is held */
1022 ARC_FILL_COMPRESSED
= 1 << 1, /* fill with compressed data */
1023 ARC_FILL_ENCRYPTED
= 1 << 2, /* fill with encrypted data */
1024 ARC_FILL_NOAUTH
= 1 << 3, /* don't attempt to authenticate */
1025 ARC_FILL_IN_PLACE
= 1 << 4 /* fill in place (special case) */
1028 static kmutex_t l2arc_feed_thr_lock
;
1029 static kcondvar_t l2arc_feed_thr_cv
;
1030 static uint8_t l2arc_thread_exit
;
1032 static abd_t
*arc_get_data_abd(arc_buf_hdr_t
*, uint64_t, void *);
1033 static void *arc_get_data_buf(arc_buf_hdr_t
*, uint64_t, void *);
1034 static void arc_get_data_impl(arc_buf_hdr_t
*, uint64_t, void *);
1035 static void arc_free_data_abd(arc_buf_hdr_t
*, abd_t
*, uint64_t, void *);
1036 static void arc_free_data_buf(arc_buf_hdr_t
*, void *, uint64_t, void *);
1037 static void arc_free_data_impl(arc_buf_hdr_t
*hdr
, uint64_t size
, void *tag
);
1038 static void arc_hdr_free_abd(arc_buf_hdr_t
*, boolean_t
);
1039 static void arc_hdr_alloc_abd(arc_buf_hdr_t
*, boolean_t
);
1040 static void arc_access(arc_buf_hdr_t
*, kmutex_t
*);
1041 static boolean_t
arc_is_overflowing(void);
1042 static void arc_buf_watch(arc_buf_t
*);
1043 static void arc_tuning_update(void);
1044 static void arc_prune_async(int64_t);
1045 static uint64_t arc_all_memory(void);
1047 static arc_buf_contents_t
arc_buf_type(arc_buf_hdr_t
*);
1048 static uint32_t arc_bufc_to_flags(arc_buf_contents_t
);
1049 static inline void arc_hdr_set_flags(arc_buf_hdr_t
*hdr
, arc_flags_t flags
);
1050 static inline void arc_hdr_clear_flags(arc_buf_hdr_t
*hdr
, arc_flags_t flags
);
1052 static boolean_t
l2arc_write_eligible(uint64_t, arc_buf_hdr_t
*);
1053 static void l2arc_read_done(zio_t
*);
1056 buf_hash(uint64_t spa
, const dva_t
*dva
, uint64_t birth
)
1058 uint8_t *vdva
= (uint8_t *)dva
;
1059 uint64_t crc
= -1ULL;
1062 ASSERT(zfs_crc64_table
[128] == ZFS_CRC64_POLY
);
1064 for (i
= 0; i
< sizeof (dva_t
); i
++)
1065 crc
= (crc
>> 8) ^ zfs_crc64_table
[(crc
^ vdva
[i
]) & 0xFF];
1067 crc
^= (spa
>>8) ^ birth
;
1072 #define HDR_EMPTY(hdr) \
1073 ((hdr)->b_dva.dva_word[0] == 0 && \
1074 (hdr)->b_dva.dva_word[1] == 0)
1076 #define HDR_EQUAL(spa, dva, birth, hdr) \
1077 ((hdr)->b_dva.dva_word[0] == (dva)->dva_word[0]) && \
1078 ((hdr)->b_dva.dva_word[1] == (dva)->dva_word[1]) && \
1079 ((hdr)->b_birth == birth) && ((hdr)->b_spa == spa)
1082 buf_discard_identity(arc_buf_hdr_t
*hdr
)
1084 hdr
->b_dva
.dva_word
[0] = 0;
1085 hdr
->b_dva
.dva_word
[1] = 0;
1089 static arc_buf_hdr_t
*
1090 buf_hash_find(uint64_t spa
, const blkptr_t
*bp
, kmutex_t
**lockp
)
1092 const dva_t
*dva
= BP_IDENTITY(bp
);
1093 uint64_t birth
= BP_PHYSICAL_BIRTH(bp
);
1094 uint64_t idx
= BUF_HASH_INDEX(spa
, dva
, birth
);
1095 kmutex_t
*hash_lock
= BUF_HASH_LOCK(idx
);
1098 mutex_enter(hash_lock
);
1099 for (hdr
= buf_hash_table
.ht_table
[idx
]; hdr
!= NULL
;
1100 hdr
= hdr
->b_hash_next
) {
1101 if (HDR_EQUAL(spa
, dva
, birth
, hdr
)) {
1106 mutex_exit(hash_lock
);
1112 * Insert an entry into the hash table. If there is already an element
1113 * equal to elem in the hash table, then the already existing element
1114 * will be returned and the new element will not be inserted.
1115 * Otherwise returns NULL.
1116 * If lockp == NULL, the caller is assumed to already hold the hash lock.
1118 static arc_buf_hdr_t
*
1119 buf_hash_insert(arc_buf_hdr_t
*hdr
, kmutex_t
**lockp
)
1121 uint64_t idx
= BUF_HASH_INDEX(hdr
->b_spa
, &hdr
->b_dva
, hdr
->b_birth
);
1122 kmutex_t
*hash_lock
= BUF_HASH_LOCK(idx
);
1123 arc_buf_hdr_t
*fhdr
;
1126 ASSERT(!DVA_IS_EMPTY(&hdr
->b_dva
));
1127 ASSERT(hdr
->b_birth
!= 0);
1128 ASSERT(!HDR_IN_HASH_TABLE(hdr
));
1130 if (lockp
!= NULL
) {
1132 mutex_enter(hash_lock
);
1134 ASSERT(MUTEX_HELD(hash_lock
));
1137 for (fhdr
= buf_hash_table
.ht_table
[idx
], i
= 0; fhdr
!= NULL
;
1138 fhdr
= fhdr
->b_hash_next
, i
++) {
1139 if (HDR_EQUAL(hdr
->b_spa
, &hdr
->b_dva
, hdr
->b_birth
, fhdr
))
1143 hdr
->b_hash_next
= buf_hash_table
.ht_table
[idx
];
1144 buf_hash_table
.ht_table
[idx
] = hdr
;
1145 arc_hdr_set_flags(hdr
, ARC_FLAG_IN_HASH_TABLE
);
1147 /* collect some hash table performance data */
1149 ARCSTAT_BUMP(arcstat_hash_collisions
);
1151 ARCSTAT_BUMP(arcstat_hash_chains
);
1153 ARCSTAT_MAX(arcstat_hash_chain_max
, i
);
1156 ARCSTAT_BUMP(arcstat_hash_elements
);
1157 ARCSTAT_MAXSTAT(arcstat_hash_elements
);
1163 buf_hash_remove(arc_buf_hdr_t
*hdr
)
1165 arc_buf_hdr_t
*fhdr
, **hdrp
;
1166 uint64_t idx
= BUF_HASH_INDEX(hdr
->b_spa
, &hdr
->b_dva
, hdr
->b_birth
);
1168 ASSERT(MUTEX_HELD(BUF_HASH_LOCK(idx
)));
1169 ASSERT(HDR_IN_HASH_TABLE(hdr
));
1171 hdrp
= &buf_hash_table
.ht_table
[idx
];
1172 while ((fhdr
= *hdrp
) != hdr
) {
1173 ASSERT3P(fhdr
, !=, NULL
);
1174 hdrp
= &fhdr
->b_hash_next
;
1176 *hdrp
= hdr
->b_hash_next
;
1177 hdr
->b_hash_next
= NULL
;
1178 arc_hdr_clear_flags(hdr
, ARC_FLAG_IN_HASH_TABLE
);
1180 /* collect some hash table performance data */
1181 ARCSTAT_BUMPDOWN(arcstat_hash_elements
);
1183 if (buf_hash_table
.ht_table
[idx
] &&
1184 buf_hash_table
.ht_table
[idx
]->b_hash_next
== NULL
)
1185 ARCSTAT_BUMPDOWN(arcstat_hash_chains
);
1189 * Global data structures and functions for the buf kmem cache.
1192 static kmem_cache_t
*hdr_full_cache
;
1193 static kmem_cache_t
*hdr_full_crypt_cache
;
1194 static kmem_cache_t
*hdr_l2only_cache
;
1195 static kmem_cache_t
*buf_cache
;
1202 #if defined(_KERNEL) && defined(HAVE_SPL)
1204 * Large allocations which do not require contiguous pages
1205 * should be using vmem_free() in the linux kernel\
1207 vmem_free(buf_hash_table
.ht_table
,
1208 (buf_hash_table
.ht_mask
+ 1) * sizeof (void *));
1210 kmem_free(buf_hash_table
.ht_table
,
1211 (buf_hash_table
.ht_mask
+ 1) * sizeof (void *));
1213 for (i
= 0; i
< BUF_LOCKS
; i
++)
1214 mutex_destroy(&buf_hash_table
.ht_locks
[i
].ht_lock
);
1215 kmem_cache_destroy(hdr_full_cache
);
1216 kmem_cache_destroy(hdr_full_crypt_cache
);
1217 kmem_cache_destroy(hdr_l2only_cache
);
1218 kmem_cache_destroy(buf_cache
);
1222 * Constructor callback - called when the cache is empty
1223 * and a new buf is requested.
1227 hdr_full_cons(void *vbuf
, void *unused
, int kmflag
)
1229 arc_buf_hdr_t
*hdr
= vbuf
;
1231 bzero(hdr
, HDR_FULL_SIZE
);
1232 hdr
->b_l1hdr
.b_byteswap
= DMU_BSWAP_NUMFUNCS
;
1233 cv_init(&hdr
->b_l1hdr
.b_cv
, NULL
, CV_DEFAULT
, NULL
);
1234 refcount_create(&hdr
->b_l1hdr
.b_refcnt
);
1235 mutex_init(&hdr
->b_l1hdr
.b_freeze_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1236 list_link_init(&hdr
->b_l1hdr
.b_arc_node
);
1237 list_link_init(&hdr
->b_l2hdr
.b_l2node
);
1238 multilist_link_init(&hdr
->b_l1hdr
.b_arc_node
);
1239 arc_space_consume(HDR_FULL_SIZE
, ARC_SPACE_HDRS
);
1246 hdr_full_crypt_cons(void *vbuf
, void *unused
, int kmflag
)
1248 arc_buf_hdr_t
*hdr
= vbuf
;
1250 hdr_full_cons(vbuf
, unused
, kmflag
);
1251 bzero(&hdr
->b_crypt_hdr
, sizeof (hdr
->b_crypt_hdr
));
1252 arc_space_consume(sizeof (hdr
->b_crypt_hdr
), ARC_SPACE_HDRS
);
1259 hdr_l2only_cons(void *vbuf
, void *unused
, int kmflag
)
1261 arc_buf_hdr_t
*hdr
= vbuf
;
1263 bzero(hdr
, HDR_L2ONLY_SIZE
);
1264 arc_space_consume(HDR_L2ONLY_SIZE
, ARC_SPACE_L2HDRS
);
1271 buf_cons(void *vbuf
, void *unused
, int kmflag
)
1273 arc_buf_t
*buf
= vbuf
;
1275 bzero(buf
, sizeof (arc_buf_t
));
1276 mutex_init(&buf
->b_evict_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1277 arc_space_consume(sizeof (arc_buf_t
), ARC_SPACE_HDRS
);
1283 * Destructor callback - called when a cached buf is
1284 * no longer required.
1288 hdr_full_dest(void *vbuf
, void *unused
)
1290 arc_buf_hdr_t
*hdr
= vbuf
;
1292 ASSERT(HDR_EMPTY(hdr
));
1293 cv_destroy(&hdr
->b_l1hdr
.b_cv
);
1294 refcount_destroy(&hdr
->b_l1hdr
.b_refcnt
);
1295 mutex_destroy(&hdr
->b_l1hdr
.b_freeze_lock
);
1296 ASSERT(!multilist_link_active(&hdr
->b_l1hdr
.b_arc_node
));
1297 arc_space_return(HDR_FULL_SIZE
, ARC_SPACE_HDRS
);
1302 hdr_full_crypt_dest(void *vbuf
, void *unused
)
1304 arc_buf_hdr_t
*hdr
= vbuf
;
1306 hdr_full_dest(vbuf
, unused
);
1307 arc_space_return(sizeof (hdr
->b_crypt_hdr
), ARC_SPACE_HDRS
);
1312 hdr_l2only_dest(void *vbuf
, void *unused
)
1314 ASSERTV(arc_buf_hdr_t
*hdr
= vbuf
);
1316 ASSERT(HDR_EMPTY(hdr
));
1317 arc_space_return(HDR_L2ONLY_SIZE
, ARC_SPACE_L2HDRS
);
1322 buf_dest(void *vbuf
, void *unused
)
1324 arc_buf_t
*buf
= vbuf
;
1326 mutex_destroy(&buf
->b_evict_lock
);
1327 arc_space_return(sizeof (arc_buf_t
), ARC_SPACE_HDRS
);
1331 * Reclaim callback -- invoked when memory is low.
1335 hdr_recl(void *unused
)
1337 dprintf("hdr_recl called\n");
1339 * umem calls the reclaim func when we destroy the buf cache,
1340 * which is after we do arc_fini().
1343 cv_signal(&arc_reclaim_thread_cv
);
1349 uint64_t *ct
= NULL
;
1350 uint64_t hsize
= 1ULL << 12;
1354 * The hash table is big enough to fill all of physical memory
1355 * with an average block size of zfs_arc_average_blocksize (default 8K).
1356 * By default, the table will take up
1357 * totalmem * sizeof(void*) / 8K (1MB per GB with 8-byte pointers).
1359 while (hsize
* zfs_arc_average_blocksize
< arc_all_memory())
1362 buf_hash_table
.ht_mask
= hsize
- 1;
1363 #if defined(_KERNEL) && defined(HAVE_SPL)
1365 * Large allocations which do not require contiguous pages
1366 * should be using vmem_alloc() in the linux kernel
1368 buf_hash_table
.ht_table
=
1369 vmem_zalloc(hsize
* sizeof (void*), KM_SLEEP
);
1371 buf_hash_table
.ht_table
=
1372 kmem_zalloc(hsize
* sizeof (void*), KM_NOSLEEP
);
1374 if (buf_hash_table
.ht_table
== NULL
) {
1375 ASSERT(hsize
> (1ULL << 8));
1380 hdr_full_cache
= kmem_cache_create("arc_buf_hdr_t_full", HDR_FULL_SIZE
,
1381 0, hdr_full_cons
, hdr_full_dest
, hdr_recl
, NULL
, NULL
, 0);
1382 hdr_full_crypt_cache
= kmem_cache_create("arc_buf_hdr_t_full_crypt",
1383 HDR_FULL_CRYPT_SIZE
, 0, hdr_full_crypt_cons
, hdr_full_crypt_dest
,
1384 hdr_recl
, NULL
, NULL
, 0);
1385 hdr_l2only_cache
= kmem_cache_create("arc_buf_hdr_t_l2only",
1386 HDR_L2ONLY_SIZE
, 0, hdr_l2only_cons
, hdr_l2only_dest
, hdr_recl
,
1388 buf_cache
= kmem_cache_create("arc_buf_t", sizeof (arc_buf_t
),
1389 0, buf_cons
, buf_dest
, NULL
, NULL
, NULL
, 0);
1391 for (i
= 0; i
< 256; i
++)
1392 for (ct
= zfs_crc64_table
+ i
, *ct
= i
, j
= 8; j
> 0; j
--)
1393 *ct
= (*ct
>> 1) ^ (-(*ct
& 1) & ZFS_CRC64_POLY
);
1395 for (i
= 0; i
< BUF_LOCKS
; i
++) {
1396 mutex_init(&buf_hash_table
.ht_locks
[i
].ht_lock
,
1397 NULL
, MUTEX_DEFAULT
, NULL
);
1401 #define ARC_MINTIME (hz>>4) /* 62 ms */
1404 * This is the size that the buf occupies in memory. If the buf is compressed,
1405 * it will correspond to the compressed size. You should use this method of
1406 * getting the buf size unless you explicitly need the logical size.
1409 arc_buf_size(arc_buf_t
*buf
)
1411 return (ARC_BUF_COMPRESSED(buf
) ?
1412 HDR_GET_PSIZE(buf
->b_hdr
) : HDR_GET_LSIZE(buf
->b_hdr
));
1416 arc_buf_lsize(arc_buf_t
*buf
)
1418 return (HDR_GET_LSIZE(buf
->b_hdr
));
1422 * This function will return B_TRUE if the buffer is encrypted in memory.
1423 * This buffer can be decrypted by calling arc_untransform().
1426 arc_is_encrypted(arc_buf_t
*buf
)
1428 return (ARC_BUF_ENCRYPTED(buf
) != 0);
1432 * Returns B_TRUE if the buffer represents data that has not had its MAC
1436 arc_is_unauthenticated(arc_buf_t
*buf
)
1438 return (HDR_NOAUTH(buf
->b_hdr
) != 0);
1442 arc_get_raw_params(arc_buf_t
*buf
, boolean_t
*byteorder
, uint8_t *salt
,
1443 uint8_t *iv
, uint8_t *mac
)
1445 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
1447 ASSERT(HDR_PROTECTED(hdr
));
1449 bcopy(hdr
->b_crypt_hdr
.b_salt
, salt
, ZIO_DATA_SALT_LEN
);
1450 bcopy(hdr
->b_crypt_hdr
.b_iv
, iv
, ZIO_DATA_IV_LEN
);
1451 bcopy(hdr
->b_crypt_hdr
.b_mac
, mac
, ZIO_DATA_MAC_LEN
);
1452 *byteorder
= (hdr
->b_l1hdr
.b_byteswap
== DMU_BSWAP_NUMFUNCS
) ?
1453 ZFS_HOST_BYTEORDER
: !ZFS_HOST_BYTEORDER
;
1457 * Indicates how this buffer is compressed in memory. If it is not compressed
1458 * the value will be ZIO_COMPRESS_OFF. It can be made normally readable with
1459 * arc_untransform() as long as it is also unencrypted.
1462 arc_get_compression(arc_buf_t
*buf
)
1464 return (ARC_BUF_COMPRESSED(buf
) ?
1465 HDR_GET_COMPRESS(buf
->b_hdr
) : ZIO_COMPRESS_OFF
);
1469 * Return the compression algorithm used to store this data in the ARC. If ARC
1470 * compression is enabled or this is an encrypted block, this will be the same
1471 * as what's used to store it on-disk. Otherwise, this will be ZIO_COMPRESS_OFF.
1473 static inline enum zio_compress
1474 arc_hdr_get_compress(arc_buf_hdr_t
*hdr
)
1476 return (HDR_COMPRESSION_ENABLED(hdr
) ?
1477 HDR_GET_COMPRESS(hdr
) : ZIO_COMPRESS_OFF
);
1480 static inline boolean_t
1481 arc_buf_is_shared(arc_buf_t
*buf
)
1483 boolean_t shared
= (buf
->b_data
!= NULL
&&
1484 buf
->b_hdr
->b_l1hdr
.b_pabd
!= NULL
&&
1485 abd_is_linear(buf
->b_hdr
->b_l1hdr
.b_pabd
) &&
1486 buf
->b_data
== abd_to_buf(buf
->b_hdr
->b_l1hdr
.b_pabd
));
1487 IMPLY(shared
, HDR_SHARED_DATA(buf
->b_hdr
));
1488 IMPLY(shared
, ARC_BUF_SHARED(buf
));
1489 IMPLY(shared
, ARC_BUF_COMPRESSED(buf
) || ARC_BUF_LAST(buf
));
1492 * It would be nice to assert arc_can_share() too, but the "hdr isn't
1493 * already being shared" requirement prevents us from doing that.
1500 * Free the checksum associated with this header. If there is no checksum, this
1504 arc_cksum_free(arc_buf_hdr_t
*hdr
)
1506 ASSERT(HDR_HAS_L1HDR(hdr
));
1508 mutex_enter(&hdr
->b_l1hdr
.b_freeze_lock
);
1509 if (hdr
->b_l1hdr
.b_freeze_cksum
!= NULL
) {
1510 kmem_free(hdr
->b_l1hdr
.b_freeze_cksum
, sizeof (zio_cksum_t
));
1511 hdr
->b_l1hdr
.b_freeze_cksum
= NULL
;
1513 mutex_exit(&hdr
->b_l1hdr
.b_freeze_lock
);
1517 * Return true iff at least one of the bufs on hdr is not compressed.
1518 * Encrypted buffers count as compressed.
1521 arc_hdr_has_uncompressed_buf(arc_buf_hdr_t
*hdr
)
1523 for (arc_buf_t
*b
= hdr
->b_l1hdr
.b_buf
; b
!= NULL
; b
= b
->b_next
) {
1524 if (!ARC_BUF_COMPRESSED(b
)) {
1533 * If we've turned on the ZFS_DEBUG_MODIFY flag, verify that the buf's data
1534 * matches the checksum that is stored in the hdr. If there is no checksum,
1535 * or if the buf is compressed, this is a no-op.
1538 arc_cksum_verify(arc_buf_t
*buf
)
1540 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
1543 if (!(zfs_flags
& ZFS_DEBUG_MODIFY
))
1546 if (ARC_BUF_COMPRESSED(buf
)) {
1547 ASSERT(hdr
->b_l1hdr
.b_freeze_cksum
== NULL
||
1548 arc_hdr_has_uncompressed_buf(hdr
));
1552 ASSERT(HDR_HAS_L1HDR(hdr
));
1554 mutex_enter(&hdr
->b_l1hdr
.b_freeze_lock
);
1555 if (hdr
->b_l1hdr
.b_freeze_cksum
== NULL
|| HDR_IO_ERROR(hdr
)) {
1556 mutex_exit(&hdr
->b_l1hdr
.b_freeze_lock
);
1560 fletcher_2_native(buf
->b_data
, arc_buf_size(buf
), NULL
, &zc
);
1561 if (!ZIO_CHECKSUM_EQUAL(*hdr
->b_l1hdr
.b_freeze_cksum
, zc
))
1562 panic("buffer modified while frozen!");
1563 mutex_exit(&hdr
->b_l1hdr
.b_freeze_lock
);
1567 * This function makes the assumption that data stored in the L2ARC
1568 * will be transformed exactly as it is in the main pool. Because of
1569 * this we can verify the checksum against the reading process's bp.
1572 arc_cksum_is_equal(arc_buf_hdr_t
*hdr
, zio_t
*zio
)
1574 ASSERT(!BP_IS_EMBEDDED(zio
->io_bp
));
1575 VERIFY3U(BP_GET_PSIZE(zio
->io_bp
), ==, HDR_GET_PSIZE(hdr
));
1578 * Block pointers always store the checksum for the logical data.
1579 * If the block pointer has the gang bit set, then the checksum
1580 * it represents is for the reconstituted data and not for an
1581 * individual gang member. The zio pipeline, however, must be able to
1582 * determine the checksum of each of the gang constituents so it
1583 * treats the checksum comparison differently than what we need
1584 * for l2arc blocks. This prevents us from using the
1585 * zio_checksum_error() interface directly. Instead we must call the
1586 * zio_checksum_error_impl() so that we can ensure the checksum is
1587 * generated using the correct checksum algorithm and accounts for the
1588 * logical I/O size and not just a gang fragment.
1590 return (zio_checksum_error_impl(zio
->io_spa
, zio
->io_bp
,
1591 BP_GET_CHECKSUM(zio
->io_bp
), zio
->io_abd
, zio
->io_size
,
1592 zio
->io_offset
, NULL
) == 0);
1596 * Given a buf full of data, if ZFS_DEBUG_MODIFY is enabled this computes a
1597 * checksum and attaches it to the buf's hdr so that we can ensure that the buf
1598 * isn't modified later on. If buf is compressed or there is already a checksum
1599 * on the hdr, this is a no-op (we only checksum uncompressed bufs).
1602 arc_cksum_compute(arc_buf_t
*buf
)
1604 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
1606 if (!(zfs_flags
& ZFS_DEBUG_MODIFY
))
1609 ASSERT(HDR_HAS_L1HDR(hdr
));
1611 mutex_enter(&buf
->b_hdr
->b_l1hdr
.b_freeze_lock
);
1612 if (hdr
->b_l1hdr
.b_freeze_cksum
!= NULL
) {
1613 ASSERT(arc_hdr_has_uncompressed_buf(hdr
));
1614 mutex_exit(&hdr
->b_l1hdr
.b_freeze_lock
);
1616 } else if (ARC_BUF_COMPRESSED(buf
)) {
1617 mutex_exit(&hdr
->b_l1hdr
.b_freeze_lock
);
1621 ASSERT(!ARC_BUF_ENCRYPTED(buf
));
1622 ASSERT(!ARC_BUF_COMPRESSED(buf
));
1623 hdr
->b_l1hdr
.b_freeze_cksum
= kmem_alloc(sizeof (zio_cksum_t
),
1625 fletcher_2_native(buf
->b_data
, arc_buf_size(buf
), NULL
,
1626 hdr
->b_l1hdr
.b_freeze_cksum
);
1627 mutex_exit(&hdr
->b_l1hdr
.b_freeze_lock
);
1633 arc_buf_sigsegv(int sig
, siginfo_t
*si
, void *unused
)
1635 panic("Got SIGSEGV at address: 0x%lx\n", (long)si
->si_addr
);
1641 arc_buf_unwatch(arc_buf_t
*buf
)
1645 ASSERT0(mprotect(buf
->b_data
, arc_buf_size(buf
),
1646 PROT_READ
| PROT_WRITE
));
1653 arc_buf_watch(arc_buf_t
*buf
)
1657 ASSERT0(mprotect(buf
->b_data
, arc_buf_size(buf
),
1662 static arc_buf_contents_t
1663 arc_buf_type(arc_buf_hdr_t
*hdr
)
1665 arc_buf_contents_t type
;
1666 if (HDR_ISTYPE_METADATA(hdr
)) {
1667 type
= ARC_BUFC_METADATA
;
1669 type
= ARC_BUFC_DATA
;
1671 VERIFY3U(hdr
->b_type
, ==, type
);
1676 arc_is_metadata(arc_buf_t
*buf
)
1678 return (HDR_ISTYPE_METADATA(buf
->b_hdr
) != 0);
1682 arc_bufc_to_flags(arc_buf_contents_t type
)
1686 /* metadata field is 0 if buffer contains normal data */
1688 case ARC_BUFC_METADATA
:
1689 return (ARC_FLAG_BUFC_METADATA
);
1693 panic("undefined ARC buffer type!");
1694 return ((uint32_t)-1);
1698 arc_buf_thaw(arc_buf_t
*buf
)
1700 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
1702 ASSERT3P(hdr
->b_l1hdr
.b_state
, ==, arc_anon
);
1703 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
1705 arc_cksum_verify(buf
);
1708 * Compressed buffers do not manipulate the b_freeze_cksum or
1709 * allocate b_thawed.
1711 if (ARC_BUF_COMPRESSED(buf
)) {
1712 ASSERT(hdr
->b_l1hdr
.b_freeze_cksum
== NULL
||
1713 arc_hdr_has_uncompressed_buf(hdr
));
1717 ASSERT(HDR_HAS_L1HDR(hdr
));
1718 arc_cksum_free(hdr
);
1719 arc_buf_unwatch(buf
);
1723 arc_buf_freeze(arc_buf_t
*buf
)
1725 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
1726 kmutex_t
*hash_lock
;
1728 if (!(zfs_flags
& ZFS_DEBUG_MODIFY
))
1731 if (ARC_BUF_COMPRESSED(buf
)) {
1732 ASSERT(hdr
->b_l1hdr
.b_freeze_cksum
== NULL
||
1733 arc_hdr_has_uncompressed_buf(hdr
));
1737 hash_lock
= HDR_LOCK(hdr
);
1738 mutex_enter(hash_lock
);
1740 ASSERT(HDR_HAS_L1HDR(hdr
));
1741 ASSERT(hdr
->b_l1hdr
.b_freeze_cksum
!= NULL
||
1742 hdr
->b_l1hdr
.b_state
== arc_anon
);
1743 arc_cksum_compute(buf
);
1744 mutex_exit(hash_lock
);
1748 * The arc_buf_hdr_t's b_flags should never be modified directly. Instead,
1749 * the following functions should be used to ensure that the flags are
1750 * updated in a thread-safe way. When manipulating the flags either
1751 * the hash_lock must be held or the hdr must be undiscoverable. This
1752 * ensures that we're not racing with any other threads when updating
1756 arc_hdr_set_flags(arc_buf_hdr_t
*hdr
, arc_flags_t flags
)
1758 ASSERT(MUTEX_HELD(HDR_LOCK(hdr
)) || HDR_EMPTY(hdr
));
1759 hdr
->b_flags
|= flags
;
1763 arc_hdr_clear_flags(arc_buf_hdr_t
*hdr
, arc_flags_t flags
)
1765 ASSERT(MUTEX_HELD(HDR_LOCK(hdr
)) || HDR_EMPTY(hdr
));
1766 hdr
->b_flags
&= ~flags
;
1770 * Setting the compression bits in the arc_buf_hdr_t's b_flags is
1771 * done in a special way since we have to clear and set bits
1772 * at the same time. Consumers that wish to set the compression bits
1773 * must use this function to ensure that the flags are updated in
1774 * thread-safe manner.
1777 arc_hdr_set_compress(arc_buf_hdr_t
*hdr
, enum zio_compress cmp
)
1779 ASSERT(MUTEX_HELD(HDR_LOCK(hdr
)) || HDR_EMPTY(hdr
));
1782 * Holes and embedded blocks will always have a psize = 0 so
1783 * we ignore the compression of the blkptr and set the
1784 * want to uncompress them. Mark them as uncompressed.
1786 if (!zfs_compressed_arc_enabled
|| HDR_GET_PSIZE(hdr
) == 0) {
1787 arc_hdr_clear_flags(hdr
, ARC_FLAG_COMPRESSED_ARC
);
1788 ASSERT(!HDR_COMPRESSION_ENABLED(hdr
));
1790 arc_hdr_set_flags(hdr
, ARC_FLAG_COMPRESSED_ARC
);
1791 ASSERT(HDR_COMPRESSION_ENABLED(hdr
));
1794 HDR_SET_COMPRESS(hdr
, cmp
);
1795 ASSERT3U(HDR_GET_COMPRESS(hdr
), ==, cmp
);
1799 * Looks for another buf on the same hdr which has the data decompressed, copies
1800 * from it, and returns true. If no such buf exists, returns false.
1803 arc_buf_try_copy_decompressed_data(arc_buf_t
*buf
)
1805 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
1806 boolean_t copied
= B_FALSE
;
1808 ASSERT(HDR_HAS_L1HDR(hdr
));
1809 ASSERT3P(buf
->b_data
, !=, NULL
);
1810 ASSERT(!ARC_BUF_COMPRESSED(buf
));
1812 for (arc_buf_t
*from
= hdr
->b_l1hdr
.b_buf
; from
!= NULL
;
1813 from
= from
->b_next
) {
1814 /* can't use our own data buffer */
1819 if (!ARC_BUF_COMPRESSED(from
)) {
1820 bcopy(from
->b_data
, buf
->b_data
, arc_buf_size(buf
));
1827 * There were no decompressed bufs, so there should not be a
1828 * checksum on the hdr either.
1830 EQUIV(!copied
, hdr
->b_l1hdr
.b_freeze_cksum
== NULL
);
1836 * Return the size of the block, b_pabd, that is stored in the arc_buf_hdr_t.
1839 arc_hdr_size(arc_buf_hdr_t
*hdr
)
1843 if (arc_hdr_get_compress(hdr
) != ZIO_COMPRESS_OFF
&&
1844 HDR_GET_PSIZE(hdr
) > 0) {
1845 size
= HDR_GET_PSIZE(hdr
);
1847 ASSERT3U(HDR_GET_LSIZE(hdr
), !=, 0);
1848 size
= HDR_GET_LSIZE(hdr
);
1854 arc_hdr_authenticate(arc_buf_hdr_t
*hdr
, spa_t
*spa
, uint64_t dsobj
)
1858 uint64_t lsize
= HDR_GET_LSIZE(hdr
);
1859 uint64_t psize
= HDR_GET_PSIZE(hdr
);
1860 void *tmpbuf
= NULL
;
1861 abd_t
*abd
= hdr
->b_l1hdr
.b_pabd
;
1863 ASSERT(HDR_LOCK(hdr
) == NULL
|| MUTEX_HELD(HDR_LOCK(hdr
)));
1864 ASSERT(HDR_AUTHENTICATED(hdr
));
1865 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, !=, NULL
);
1868 * The MAC is calculated on the compressed data that is stored on disk.
1869 * However, if compressed arc is disabled we will only have the
1870 * decompressed data available to us now. Compress it into a temporary
1871 * abd so we can verify the MAC. The performance overhead of this will
1872 * be relatively low, since most objects in an encrypted objset will
1873 * be encrypted (instead of authenticated) anyway.
1875 if (HDR_GET_COMPRESS(hdr
) != ZIO_COMPRESS_OFF
&&
1876 !HDR_COMPRESSION_ENABLED(hdr
)) {
1877 tmpbuf
= zio_buf_alloc(lsize
);
1878 abd
= abd_get_from_buf(tmpbuf
, lsize
);
1879 abd_take_ownership_of_buf(abd
, B_TRUE
);
1881 csize
= zio_compress_data(HDR_GET_COMPRESS(hdr
),
1882 hdr
->b_l1hdr
.b_pabd
, tmpbuf
, lsize
);
1883 ASSERT3U(csize
, <=, psize
);
1884 abd_zero_off(abd
, csize
, psize
- csize
);
1888 * Authentication is best effort. We authenticate whenever the key is
1889 * available. If we succeed we clear ARC_FLAG_NOAUTH.
1891 if (hdr
->b_crypt_hdr
.b_ot
== DMU_OT_OBJSET
) {
1892 ASSERT3U(HDR_GET_COMPRESS(hdr
), ==, ZIO_COMPRESS_OFF
);
1893 ASSERT3U(lsize
, ==, psize
);
1894 ret
= spa_do_crypt_objset_mac_abd(B_FALSE
, spa
, dsobj
, abd
,
1895 psize
, hdr
->b_l1hdr
.b_byteswap
!= DMU_BSWAP_NUMFUNCS
);
1897 ret
= spa_do_crypt_mac_abd(B_FALSE
, spa
, dsobj
, abd
, psize
,
1898 hdr
->b_crypt_hdr
.b_mac
);
1902 arc_hdr_clear_flags(hdr
, ARC_FLAG_NOAUTH
);
1903 else if (ret
!= ENOENT
)
1919 * This function will take a header that only has raw encrypted data in
1920 * b_crypt_hdr.b_rabd and decrypt it into a new buffer which is stored in
1921 * b_l1hdr.b_pabd. If designated in the header flags, this function will
1922 * also decompress the data.
1925 arc_hdr_decrypt(arc_buf_hdr_t
*hdr
, spa_t
*spa
, uint64_t dsobj
)
1928 dsl_crypto_key_t
*dck
= NULL
;
1931 boolean_t no_crypt
= B_FALSE
;
1932 boolean_t bswap
= (hdr
->b_l1hdr
.b_byteswap
!= DMU_BSWAP_NUMFUNCS
);
1934 ASSERT(HDR_LOCK(hdr
) == NULL
|| MUTEX_HELD(HDR_LOCK(hdr
)));
1935 ASSERT(HDR_ENCRYPTED(hdr
));
1937 arc_hdr_alloc_abd(hdr
, B_FALSE
);
1940 * We must be careful to use the passed-in dsobj value here and
1941 * not the value in b_dsobj. b_dsobj is meant to be a best guess for
1942 * the L2ARC, which has the luxury of being able to fail without real
1943 * consequences (the data simply won't make it to the L2ARC). In
1944 * reality, the dsobj stored in the header may belong to a dataset
1945 * that has been unmounted or otherwise disowned, meaning the key
1946 * won't be accessible via that dsobj anymore.
1948 ret
= spa_keystore_lookup_key(spa
, dsobj
, FTAG
, &dck
);
1950 ret
= SET_ERROR(EACCES
);
1954 ret
= zio_do_crypt_abd(B_FALSE
, &dck
->dck_key
,
1955 hdr
->b_crypt_hdr
.b_salt
, hdr
->b_crypt_hdr
.b_ot
,
1956 hdr
->b_crypt_hdr
.b_iv
, hdr
->b_crypt_hdr
.b_mac
,
1957 HDR_GET_PSIZE(hdr
), bswap
, hdr
->b_l1hdr
.b_pabd
,
1958 hdr
->b_crypt_hdr
.b_rabd
, &no_crypt
);
1963 abd_copy(hdr
->b_l1hdr
.b_pabd
, hdr
->b_crypt_hdr
.b_rabd
,
1964 HDR_GET_PSIZE(hdr
));
1968 * If this header has disabled arc compression but the b_pabd is
1969 * compressed after decrypting it, we need to decompress the newly
1972 if (HDR_GET_COMPRESS(hdr
) != ZIO_COMPRESS_OFF
&&
1973 !HDR_COMPRESSION_ENABLED(hdr
)) {
1975 * We want to make sure that we are correctly honoring the
1976 * zfs_abd_scatter_enabled setting, so we allocate an abd here
1977 * and then loan a buffer from it, rather than allocating a
1978 * linear buffer and wrapping it in an abd later.
1980 cabd
= arc_get_data_abd(hdr
, arc_hdr_size(hdr
), hdr
);
1981 tmp
= abd_borrow_buf(cabd
, arc_hdr_size(hdr
));
1983 ret
= zio_decompress_data(HDR_GET_COMPRESS(hdr
),
1984 hdr
->b_l1hdr
.b_pabd
, tmp
, HDR_GET_PSIZE(hdr
),
1985 HDR_GET_LSIZE(hdr
));
1987 abd_return_buf(cabd
, tmp
, arc_hdr_size(hdr
));
1991 abd_return_buf_copy(cabd
, tmp
, arc_hdr_size(hdr
));
1992 arc_free_data_abd(hdr
, hdr
->b_l1hdr
.b_pabd
,
1993 arc_hdr_size(hdr
), hdr
);
1994 hdr
->b_l1hdr
.b_pabd
= cabd
;
1997 spa_keystore_dsl_key_rele(spa
, dck
, FTAG
);
2002 arc_hdr_free_abd(hdr
, B_FALSE
);
2004 spa_keystore_dsl_key_rele(spa
, dck
, FTAG
);
2006 arc_free_data_buf(hdr
, cabd
, arc_hdr_size(hdr
), hdr
);
2012 * This function is called during arc_buf_fill() to prepare the header's
2013 * abd plaintext pointer for use. This involves authenticated protected
2014 * data and decrypting encrypted data into the plaintext abd.
2017 arc_fill_hdr_crypt(arc_buf_hdr_t
*hdr
, kmutex_t
*hash_lock
, spa_t
*spa
,
2018 uint64_t dsobj
, boolean_t noauth
)
2022 ASSERT(HDR_PROTECTED(hdr
));
2024 if (hash_lock
!= NULL
)
2025 mutex_enter(hash_lock
);
2027 if (HDR_NOAUTH(hdr
) && !noauth
) {
2029 * The caller requested authenticated data but our data has
2030 * not been authenticated yet. Verify the MAC now if we can.
2032 ret
= arc_hdr_authenticate(hdr
, spa
, dsobj
);
2035 } else if (HDR_HAS_RABD(hdr
) && hdr
->b_l1hdr
.b_pabd
== NULL
) {
2037 * If we only have the encrypted version of the data, but the
2038 * unencrypted version was requested we take this opportunity
2039 * to store the decrypted version in the header for future use.
2041 ret
= arc_hdr_decrypt(hdr
, spa
, dsobj
);
2046 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, !=, NULL
);
2048 if (hash_lock
!= NULL
)
2049 mutex_exit(hash_lock
);
2054 if (hash_lock
!= NULL
)
2055 mutex_exit(hash_lock
);
2061 * This function is used by the dbuf code to decrypt bonus buffers in place.
2062 * The dbuf code itself doesn't have any locking for decrypting a shared dnode
2063 * block, so we use the hash lock here to protect against concurrent calls to
2067 arc_buf_untransform_in_place(arc_buf_t
*buf
, kmutex_t
*hash_lock
)
2069 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
2071 ASSERT(HDR_ENCRYPTED(hdr
));
2072 ASSERT3U(hdr
->b_crypt_hdr
.b_ot
, ==, DMU_OT_DNODE
);
2073 ASSERT(HDR_LOCK(hdr
) == NULL
|| MUTEX_HELD(HDR_LOCK(hdr
)));
2074 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, !=, NULL
);
2076 zio_crypt_copy_dnode_bonus(hdr
->b_l1hdr
.b_pabd
, buf
->b_data
,
2078 buf
->b_flags
&= ~ARC_BUF_FLAG_ENCRYPTED
;
2079 buf
->b_flags
&= ~ARC_BUF_FLAG_COMPRESSED
;
2080 hdr
->b_crypt_hdr
.b_ebufcnt
-= 1;
2084 * Given a buf that has a data buffer attached to it, this function will
2085 * efficiently fill the buf with data of the specified compression setting from
2086 * the hdr and update the hdr's b_freeze_cksum if necessary. If the buf and hdr
2087 * are already sharing a data buf, no copy is performed.
2089 * If the buf is marked as compressed but uncompressed data was requested, this
2090 * will allocate a new data buffer for the buf, remove that flag, and fill the
2091 * buf with uncompressed data. You can't request a compressed buf on a hdr with
2092 * uncompressed data, and (since we haven't added support for it yet) if you
2093 * want compressed data your buf must already be marked as compressed and have
2094 * the correct-sized data buffer.
2097 arc_buf_fill(arc_buf_t
*buf
, spa_t
*spa
, uint64_t dsobj
, arc_fill_flags_t flags
)
2100 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
2101 boolean_t hdr_compressed
=
2102 (arc_hdr_get_compress(hdr
) != ZIO_COMPRESS_OFF
);
2103 boolean_t compressed
= (flags
& ARC_FILL_COMPRESSED
) != 0;
2104 boolean_t encrypted
= (flags
& ARC_FILL_ENCRYPTED
) != 0;
2105 dmu_object_byteswap_t bswap
= hdr
->b_l1hdr
.b_byteswap
;
2106 kmutex_t
*hash_lock
= (flags
& ARC_FILL_LOCKED
) ? NULL
: HDR_LOCK(hdr
);
2108 ASSERT3P(buf
->b_data
, !=, NULL
);
2109 IMPLY(compressed
, hdr_compressed
|| ARC_BUF_ENCRYPTED(buf
));
2110 IMPLY(compressed
, ARC_BUF_COMPRESSED(buf
));
2111 IMPLY(encrypted
, HDR_ENCRYPTED(hdr
));
2112 IMPLY(encrypted
, ARC_BUF_ENCRYPTED(buf
));
2113 IMPLY(encrypted
, ARC_BUF_COMPRESSED(buf
));
2114 IMPLY(encrypted
, !ARC_BUF_SHARED(buf
));
2117 * If the caller wanted encrypted data we just need to copy it from
2118 * b_rabd and potentially byteswap it. We won't be able to do any
2119 * further transforms on it.
2122 ASSERT(HDR_HAS_RABD(hdr
));
2123 abd_copy_to_buf(buf
->b_data
, hdr
->b_crypt_hdr
.b_rabd
,
2124 HDR_GET_PSIZE(hdr
));
2129 * Adjust encrypted and authenticated headers to accomodate the
2130 * request if needed.
2132 if (HDR_PROTECTED(hdr
)) {
2133 error
= arc_fill_hdr_crypt(hdr
, hash_lock
, spa
,
2134 dsobj
, !!(flags
& ARC_FILL_NOAUTH
));
2140 * There is a special case here for dnode blocks which are
2141 * decrypting their bonus buffers. These blocks may request to
2142 * be decrypted in-place. This is necessary because there may
2143 * be many dnodes pointing into this buffer and there is
2144 * currently no method to synchronize replacing the backing
2145 * b_data buffer and updating all of the pointers. Here we use
2146 * the hash lock to ensure there are no races. If the need
2147 * arises for other types to be decrypted in-place, they must
2148 * add handling here as well.
2150 if ((flags
& ARC_FILL_IN_PLACE
) != 0) {
2151 ASSERT(!hdr_compressed
);
2152 ASSERT(!compressed
);
2155 if (HDR_ENCRYPTED(hdr
) && ARC_BUF_ENCRYPTED(buf
)) {
2156 ASSERT3U(hdr
->b_crypt_hdr
.b_ot
, ==, DMU_OT_DNODE
);
2158 if (hash_lock
!= NULL
)
2159 mutex_enter(hash_lock
);
2160 arc_buf_untransform_in_place(buf
, hash_lock
);
2161 if (hash_lock
!= NULL
)
2162 mutex_exit(hash_lock
);
2164 /* Compute the hdr's checksum if necessary */
2165 arc_cksum_compute(buf
);
2171 if (hdr_compressed
== compressed
) {
2172 if (!arc_buf_is_shared(buf
)) {
2173 abd_copy_to_buf(buf
->b_data
, hdr
->b_l1hdr
.b_pabd
,
2177 ASSERT(hdr_compressed
);
2178 ASSERT(!compressed
);
2179 ASSERT3U(HDR_GET_LSIZE(hdr
), !=, HDR_GET_PSIZE(hdr
));
2182 * If the buf is sharing its data with the hdr, unlink it and
2183 * allocate a new data buffer for the buf.
2185 if (arc_buf_is_shared(buf
)) {
2186 ASSERT(ARC_BUF_COMPRESSED(buf
));
2188 /* We need to give the buf it's own b_data */
2189 buf
->b_flags
&= ~ARC_BUF_FLAG_SHARED
;
2191 arc_get_data_buf(hdr
, HDR_GET_LSIZE(hdr
), buf
);
2192 arc_hdr_clear_flags(hdr
, ARC_FLAG_SHARED_DATA
);
2194 /* Previously overhead was 0; just add new overhead */
2195 ARCSTAT_INCR(arcstat_overhead_size
, HDR_GET_LSIZE(hdr
));
2196 } else if (ARC_BUF_COMPRESSED(buf
)) {
2197 /* We need to reallocate the buf's b_data */
2198 arc_free_data_buf(hdr
, buf
->b_data
, HDR_GET_PSIZE(hdr
),
2201 arc_get_data_buf(hdr
, HDR_GET_LSIZE(hdr
), buf
);
2203 /* We increased the size of b_data; update overhead */
2204 ARCSTAT_INCR(arcstat_overhead_size
,
2205 HDR_GET_LSIZE(hdr
) - HDR_GET_PSIZE(hdr
));
2209 * Regardless of the buf's previous compression settings, it
2210 * should not be compressed at the end of this function.
2212 buf
->b_flags
&= ~ARC_BUF_FLAG_COMPRESSED
;
2215 * Try copying the data from another buf which already has a
2216 * decompressed version. If that's not possible, it's time to
2217 * bite the bullet and decompress the data from the hdr.
2219 if (arc_buf_try_copy_decompressed_data(buf
)) {
2220 /* Skip byteswapping and checksumming (already done) */
2221 ASSERT3P(hdr
->b_l1hdr
.b_freeze_cksum
, !=, NULL
);
2224 error
= zio_decompress_data(HDR_GET_COMPRESS(hdr
),
2225 hdr
->b_l1hdr
.b_pabd
, buf
->b_data
,
2226 HDR_GET_PSIZE(hdr
), HDR_GET_LSIZE(hdr
));
2229 * Absent hardware errors or software bugs, this should
2230 * be impossible, but log it anyway so we can debug it.
2234 "hdr %p, compress %d, psize %d, lsize %d",
2235 hdr
, arc_hdr_get_compress(hdr
),
2236 HDR_GET_PSIZE(hdr
), HDR_GET_LSIZE(hdr
));
2237 return (SET_ERROR(EIO
));
2243 /* Byteswap the buf's data if necessary */
2244 if (bswap
!= DMU_BSWAP_NUMFUNCS
) {
2245 ASSERT(!HDR_SHARED_DATA(hdr
));
2246 ASSERT3U(bswap
, <, DMU_BSWAP_NUMFUNCS
);
2247 dmu_ot_byteswap
[bswap
].ob_func(buf
->b_data
, HDR_GET_LSIZE(hdr
));
2250 /* Compute the hdr's checksum if necessary */
2251 arc_cksum_compute(buf
);
2257 * If this function is being called to decrypt an encrypted buffer or verify an
2258 * authenticated one, the key must be loaded and a mapping must be made
2259 * available in the keystore via spa_keystore_create_mapping() or one of its
2263 arc_untransform(arc_buf_t
*buf
, spa_t
*spa
, const zbookmark_phys_t
*zb
,
2267 arc_fill_flags_t flags
= 0;
2270 flags
|= ARC_FILL_IN_PLACE
;
2272 ret
= arc_buf_fill(buf
, spa
, zb
->zb_objset
, flags
);
2273 if (ret
== ECKSUM
) {
2275 * Convert authentication and decryption errors to EIO
2276 * (and generate an ereport) before leaving the ARC.
2278 ret
= SET_ERROR(EIO
);
2279 zfs_ereport_post(FM_EREPORT_ZFS_AUTHENTICATION
,
2280 spa
, NULL
, zb
, NULL
, 0, 0);
2287 * Increment the amount of evictable space in the arc_state_t's refcount.
2288 * We account for the space used by the hdr and the arc buf individually
2289 * so that we can add and remove them from the refcount individually.
2292 arc_evictable_space_increment(arc_buf_hdr_t
*hdr
, arc_state_t
*state
)
2294 arc_buf_contents_t type
= arc_buf_type(hdr
);
2296 ASSERT(HDR_HAS_L1HDR(hdr
));
2298 if (GHOST_STATE(state
)) {
2299 ASSERT0(hdr
->b_l1hdr
.b_bufcnt
);
2300 ASSERT3P(hdr
->b_l1hdr
.b_buf
, ==, NULL
);
2301 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, ==, NULL
);
2302 ASSERT(!HDR_HAS_RABD(hdr
));
2303 (void) refcount_add_many(&state
->arcs_esize
[type
],
2304 HDR_GET_LSIZE(hdr
), hdr
);
2308 ASSERT(!GHOST_STATE(state
));
2309 if (hdr
->b_l1hdr
.b_pabd
!= NULL
) {
2310 (void) refcount_add_many(&state
->arcs_esize
[type
],
2311 arc_hdr_size(hdr
), hdr
);
2313 if (HDR_HAS_RABD(hdr
)) {
2314 (void) refcount_add_many(&state
->arcs_esize
[type
],
2315 HDR_GET_PSIZE(hdr
), hdr
);
2318 for (arc_buf_t
*buf
= hdr
->b_l1hdr
.b_buf
; buf
!= NULL
;
2319 buf
= buf
->b_next
) {
2320 if (arc_buf_is_shared(buf
))
2322 (void) refcount_add_many(&state
->arcs_esize
[type
],
2323 arc_buf_size(buf
), buf
);
2328 * Decrement the amount of evictable space in the arc_state_t's refcount.
2329 * We account for the space used by the hdr and the arc buf individually
2330 * so that we can add and remove them from the refcount individually.
2333 arc_evictable_space_decrement(arc_buf_hdr_t
*hdr
, arc_state_t
*state
)
2335 arc_buf_contents_t type
= arc_buf_type(hdr
);
2337 ASSERT(HDR_HAS_L1HDR(hdr
));
2339 if (GHOST_STATE(state
)) {
2340 ASSERT0(hdr
->b_l1hdr
.b_bufcnt
);
2341 ASSERT3P(hdr
->b_l1hdr
.b_buf
, ==, NULL
);
2342 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, ==, NULL
);
2343 ASSERT(!HDR_HAS_RABD(hdr
));
2344 (void) refcount_remove_many(&state
->arcs_esize
[type
],
2345 HDR_GET_LSIZE(hdr
), hdr
);
2349 ASSERT(!GHOST_STATE(state
));
2350 if (hdr
->b_l1hdr
.b_pabd
!= NULL
) {
2351 (void) refcount_remove_many(&state
->arcs_esize
[type
],
2352 arc_hdr_size(hdr
), hdr
);
2354 if (HDR_HAS_RABD(hdr
)) {
2355 (void) refcount_remove_many(&state
->arcs_esize
[type
],
2356 HDR_GET_PSIZE(hdr
), hdr
);
2359 for (arc_buf_t
*buf
= hdr
->b_l1hdr
.b_buf
; buf
!= NULL
;
2360 buf
= buf
->b_next
) {
2361 if (arc_buf_is_shared(buf
))
2363 (void) refcount_remove_many(&state
->arcs_esize
[type
],
2364 arc_buf_size(buf
), buf
);
2369 * Add a reference to this hdr indicating that someone is actively
2370 * referencing that memory. When the refcount transitions from 0 to 1,
2371 * we remove it from the respective arc_state_t list to indicate that
2372 * it is not evictable.
2375 add_reference(arc_buf_hdr_t
*hdr
, void *tag
)
2379 ASSERT(HDR_HAS_L1HDR(hdr
));
2380 if (!MUTEX_HELD(HDR_LOCK(hdr
))) {
2381 ASSERT(hdr
->b_l1hdr
.b_state
== arc_anon
);
2382 ASSERT(refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
));
2383 ASSERT3P(hdr
->b_l1hdr
.b_buf
, ==, NULL
);
2386 state
= hdr
->b_l1hdr
.b_state
;
2388 if ((refcount_add(&hdr
->b_l1hdr
.b_refcnt
, tag
) == 1) &&
2389 (state
!= arc_anon
)) {
2390 /* We don't use the L2-only state list. */
2391 if (state
!= arc_l2c_only
) {
2392 multilist_remove(state
->arcs_list
[arc_buf_type(hdr
)],
2394 arc_evictable_space_decrement(hdr
, state
);
2396 /* remove the prefetch flag if we get a reference */
2397 arc_hdr_clear_flags(hdr
, ARC_FLAG_PREFETCH
);
2402 * Remove a reference from this hdr. When the reference transitions from
2403 * 1 to 0 and we're not anonymous, then we add this hdr to the arc_state_t's
2404 * list making it eligible for eviction.
2407 remove_reference(arc_buf_hdr_t
*hdr
, kmutex_t
*hash_lock
, void *tag
)
2410 arc_state_t
*state
= hdr
->b_l1hdr
.b_state
;
2412 ASSERT(HDR_HAS_L1HDR(hdr
));
2413 ASSERT(state
== arc_anon
|| MUTEX_HELD(hash_lock
));
2414 ASSERT(!GHOST_STATE(state
));
2417 * arc_l2c_only counts as a ghost state so we don't need to explicitly
2418 * check to prevent usage of the arc_l2c_only list.
2420 if (((cnt
= refcount_remove(&hdr
->b_l1hdr
.b_refcnt
, tag
)) == 0) &&
2421 (state
!= arc_anon
)) {
2422 multilist_insert(state
->arcs_list
[arc_buf_type(hdr
)], hdr
);
2423 ASSERT3U(hdr
->b_l1hdr
.b_bufcnt
, >, 0);
2424 arc_evictable_space_increment(hdr
, state
);
2430 * Returns detailed information about a specific arc buffer. When the
2431 * state_index argument is set the function will calculate the arc header
2432 * list position for its arc state. Since this requires a linear traversal
2433 * callers are strongly encourage not to do this. However, it can be helpful
2434 * for targeted analysis so the functionality is provided.
2437 arc_buf_info(arc_buf_t
*ab
, arc_buf_info_t
*abi
, int state_index
)
2439 arc_buf_hdr_t
*hdr
= ab
->b_hdr
;
2440 l1arc_buf_hdr_t
*l1hdr
= NULL
;
2441 l2arc_buf_hdr_t
*l2hdr
= NULL
;
2442 arc_state_t
*state
= NULL
;
2444 memset(abi
, 0, sizeof (arc_buf_info_t
));
2449 abi
->abi_flags
= hdr
->b_flags
;
2451 if (HDR_HAS_L1HDR(hdr
)) {
2452 l1hdr
= &hdr
->b_l1hdr
;
2453 state
= l1hdr
->b_state
;
2455 if (HDR_HAS_L2HDR(hdr
))
2456 l2hdr
= &hdr
->b_l2hdr
;
2459 abi
->abi_bufcnt
= l1hdr
->b_bufcnt
;
2460 abi
->abi_access
= l1hdr
->b_arc_access
;
2461 abi
->abi_mru_hits
= l1hdr
->b_mru_hits
;
2462 abi
->abi_mru_ghost_hits
= l1hdr
->b_mru_ghost_hits
;
2463 abi
->abi_mfu_hits
= l1hdr
->b_mfu_hits
;
2464 abi
->abi_mfu_ghost_hits
= l1hdr
->b_mfu_ghost_hits
;
2465 abi
->abi_holds
= refcount_count(&l1hdr
->b_refcnt
);
2469 abi
->abi_l2arc_dattr
= l2hdr
->b_daddr
;
2470 abi
->abi_l2arc_hits
= l2hdr
->b_hits
;
2473 abi
->abi_state_type
= state
? state
->arcs_state
: ARC_STATE_ANON
;
2474 abi
->abi_state_contents
= arc_buf_type(hdr
);
2475 abi
->abi_size
= arc_hdr_size(hdr
);
2479 * Move the supplied buffer to the indicated state. The hash lock
2480 * for the buffer must be held by the caller.
2483 arc_change_state(arc_state_t
*new_state
, arc_buf_hdr_t
*hdr
,
2484 kmutex_t
*hash_lock
)
2486 arc_state_t
*old_state
;
2489 boolean_t update_old
, update_new
;
2490 arc_buf_contents_t buftype
= arc_buf_type(hdr
);
2493 * We almost always have an L1 hdr here, since we call arc_hdr_realloc()
2494 * in arc_read() when bringing a buffer out of the L2ARC. However, the
2495 * L1 hdr doesn't always exist when we change state to arc_anon before
2496 * destroying a header, in which case reallocating to add the L1 hdr is
2499 if (HDR_HAS_L1HDR(hdr
)) {
2500 old_state
= hdr
->b_l1hdr
.b_state
;
2501 refcnt
= refcount_count(&hdr
->b_l1hdr
.b_refcnt
);
2502 bufcnt
= hdr
->b_l1hdr
.b_bufcnt
;
2503 update_old
= (bufcnt
> 0 || hdr
->b_l1hdr
.b_pabd
!= NULL
||
2506 old_state
= arc_l2c_only
;
2509 update_old
= B_FALSE
;
2511 update_new
= update_old
;
2513 ASSERT(MUTEX_HELD(hash_lock
));
2514 ASSERT3P(new_state
, !=, old_state
);
2515 ASSERT(!GHOST_STATE(new_state
) || bufcnt
== 0);
2516 ASSERT(old_state
!= arc_anon
|| bufcnt
<= 1);
2519 * If this buffer is evictable, transfer it from the
2520 * old state list to the new state list.
2523 if (old_state
!= arc_anon
&& old_state
!= arc_l2c_only
) {
2524 ASSERT(HDR_HAS_L1HDR(hdr
));
2525 multilist_remove(old_state
->arcs_list
[buftype
], hdr
);
2527 if (GHOST_STATE(old_state
)) {
2529 ASSERT3P(hdr
->b_l1hdr
.b_buf
, ==, NULL
);
2530 update_old
= B_TRUE
;
2532 arc_evictable_space_decrement(hdr
, old_state
);
2534 if (new_state
!= arc_anon
&& new_state
!= arc_l2c_only
) {
2536 * An L1 header always exists here, since if we're
2537 * moving to some L1-cached state (i.e. not l2c_only or
2538 * anonymous), we realloc the header to add an L1hdr
2541 ASSERT(HDR_HAS_L1HDR(hdr
));
2542 multilist_insert(new_state
->arcs_list
[buftype
], hdr
);
2544 if (GHOST_STATE(new_state
)) {
2546 ASSERT3P(hdr
->b_l1hdr
.b_buf
, ==, NULL
);
2547 update_new
= B_TRUE
;
2549 arc_evictable_space_increment(hdr
, new_state
);
2553 ASSERT(!HDR_EMPTY(hdr
));
2554 if (new_state
== arc_anon
&& HDR_IN_HASH_TABLE(hdr
))
2555 buf_hash_remove(hdr
);
2557 /* adjust state sizes (ignore arc_l2c_only) */
2559 if (update_new
&& new_state
!= arc_l2c_only
) {
2560 ASSERT(HDR_HAS_L1HDR(hdr
));
2561 if (GHOST_STATE(new_state
)) {
2565 * When moving a header to a ghost state, we first
2566 * remove all arc buffers. Thus, we'll have a
2567 * bufcnt of zero, and no arc buffer to use for
2568 * the reference. As a result, we use the arc
2569 * header pointer for the reference.
2571 (void) refcount_add_many(&new_state
->arcs_size
,
2572 HDR_GET_LSIZE(hdr
), hdr
);
2573 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, ==, NULL
);
2574 ASSERT(!HDR_HAS_RABD(hdr
));
2576 uint32_t buffers
= 0;
2579 * Each individual buffer holds a unique reference,
2580 * thus we must remove each of these references one
2583 for (arc_buf_t
*buf
= hdr
->b_l1hdr
.b_buf
; buf
!= NULL
;
2584 buf
= buf
->b_next
) {
2585 ASSERT3U(bufcnt
, !=, 0);
2589 * When the arc_buf_t is sharing the data
2590 * block with the hdr, the owner of the
2591 * reference belongs to the hdr. Only
2592 * add to the refcount if the arc_buf_t is
2595 if (arc_buf_is_shared(buf
))
2598 (void) refcount_add_many(&new_state
->arcs_size
,
2599 arc_buf_size(buf
), buf
);
2601 ASSERT3U(bufcnt
, ==, buffers
);
2603 if (hdr
->b_l1hdr
.b_pabd
!= NULL
) {
2604 (void) refcount_add_many(&new_state
->arcs_size
,
2605 arc_hdr_size(hdr
), hdr
);
2608 if (HDR_HAS_RABD(hdr
)) {
2609 (void) refcount_add_many(&new_state
->arcs_size
,
2610 HDR_GET_PSIZE(hdr
), hdr
);
2615 if (update_old
&& old_state
!= arc_l2c_only
) {
2616 ASSERT(HDR_HAS_L1HDR(hdr
));
2617 if (GHOST_STATE(old_state
)) {
2619 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, ==, NULL
);
2620 ASSERT(!HDR_HAS_RABD(hdr
));
2623 * When moving a header off of a ghost state,
2624 * the header will not contain any arc buffers.
2625 * We use the arc header pointer for the reference
2626 * which is exactly what we did when we put the
2627 * header on the ghost state.
2630 (void) refcount_remove_many(&old_state
->arcs_size
,
2631 HDR_GET_LSIZE(hdr
), hdr
);
2633 uint32_t buffers
= 0;
2636 * Each individual buffer holds a unique reference,
2637 * thus we must remove each of these references one
2640 for (arc_buf_t
*buf
= hdr
->b_l1hdr
.b_buf
; buf
!= NULL
;
2641 buf
= buf
->b_next
) {
2642 ASSERT3U(bufcnt
, !=, 0);
2646 * When the arc_buf_t is sharing the data
2647 * block with the hdr, the owner of the
2648 * reference belongs to the hdr. Only
2649 * add to the refcount if the arc_buf_t is
2652 if (arc_buf_is_shared(buf
))
2655 (void) refcount_remove_many(
2656 &old_state
->arcs_size
, arc_buf_size(buf
),
2659 ASSERT3U(bufcnt
, ==, buffers
);
2660 ASSERT(hdr
->b_l1hdr
.b_pabd
!= NULL
||
2663 if (hdr
->b_l1hdr
.b_pabd
!= NULL
) {
2664 (void) refcount_remove_many(
2665 &old_state
->arcs_size
, arc_hdr_size(hdr
),
2669 if (HDR_HAS_RABD(hdr
)) {
2670 (void) refcount_remove_many(
2671 &old_state
->arcs_size
, HDR_GET_PSIZE(hdr
),
2677 if (HDR_HAS_L1HDR(hdr
))
2678 hdr
->b_l1hdr
.b_state
= new_state
;
2681 * L2 headers should never be on the L2 state list since they don't
2682 * have L1 headers allocated.
2684 ASSERT(multilist_is_empty(arc_l2c_only
->arcs_list
[ARC_BUFC_DATA
]) &&
2685 multilist_is_empty(arc_l2c_only
->arcs_list
[ARC_BUFC_METADATA
]));
2689 arc_space_consume(uint64_t space
, arc_space_type_t type
)
2691 ASSERT(type
>= 0 && type
< ARC_SPACE_NUMTYPES
);
2696 case ARC_SPACE_DATA
:
2697 ARCSTAT_INCR(arcstat_data_size
, space
);
2699 case ARC_SPACE_META
:
2700 ARCSTAT_INCR(arcstat_metadata_size
, space
);
2702 case ARC_SPACE_BONUS
:
2703 ARCSTAT_INCR(arcstat_bonus_size
, space
);
2705 case ARC_SPACE_DNODE
:
2706 ARCSTAT_INCR(arcstat_dnode_size
, space
);
2708 case ARC_SPACE_DBUF
:
2709 ARCSTAT_INCR(arcstat_dbuf_size
, space
);
2711 case ARC_SPACE_HDRS
:
2712 ARCSTAT_INCR(arcstat_hdr_size
, space
);
2714 case ARC_SPACE_L2HDRS
:
2715 ARCSTAT_INCR(arcstat_l2_hdr_size
, space
);
2719 if (type
!= ARC_SPACE_DATA
)
2720 ARCSTAT_INCR(arcstat_meta_used
, space
);
2722 atomic_add_64(&arc_size
, space
);
2726 arc_space_return(uint64_t space
, arc_space_type_t type
)
2728 ASSERT(type
>= 0 && type
< ARC_SPACE_NUMTYPES
);
2733 case ARC_SPACE_DATA
:
2734 ARCSTAT_INCR(arcstat_data_size
, -space
);
2736 case ARC_SPACE_META
:
2737 ARCSTAT_INCR(arcstat_metadata_size
, -space
);
2739 case ARC_SPACE_BONUS
:
2740 ARCSTAT_INCR(arcstat_bonus_size
, -space
);
2742 case ARC_SPACE_DNODE
:
2743 ARCSTAT_INCR(arcstat_dnode_size
, -space
);
2745 case ARC_SPACE_DBUF
:
2746 ARCSTAT_INCR(arcstat_dbuf_size
, -space
);
2748 case ARC_SPACE_HDRS
:
2749 ARCSTAT_INCR(arcstat_hdr_size
, -space
);
2751 case ARC_SPACE_L2HDRS
:
2752 ARCSTAT_INCR(arcstat_l2_hdr_size
, -space
);
2756 if (type
!= ARC_SPACE_DATA
) {
2757 ASSERT(arc_meta_used
>= space
);
2758 if (arc_meta_max
< arc_meta_used
)
2759 arc_meta_max
= arc_meta_used
;
2760 ARCSTAT_INCR(arcstat_meta_used
, -space
);
2763 ASSERT(arc_size
>= space
);
2764 atomic_add_64(&arc_size
, -space
);
2768 * Given a hdr and a buf, returns whether that buf can share its b_data buffer
2769 * with the hdr's b_pabd.
2772 arc_can_share(arc_buf_hdr_t
*hdr
, arc_buf_t
*buf
)
2775 * The criteria for sharing a hdr's data are:
2776 * 1. the buffer is not encrypted
2777 * 2. the hdr's compression matches the buf's compression
2778 * 3. the hdr doesn't need to be byteswapped
2779 * 4. the hdr isn't already being shared
2780 * 5. the buf is either compressed or it is the last buf in the hdr list
2782 * Criterion #5 maintains the invariant that shared uncompressed
2783 * bufs must be the final buf in the hdr's b_buf list. Reading this, you
2784 * might ask, "if a compressed buf is allocated first, won't that be the
2785 * last thing in the list?", but in that case it's impossible to create
2786 * a shared uncompressed buf anyway (because the hdr must be compressed
2787 * to have the compressed buf). You might also think that #3 is
2788 * sufficient to make this guarantee, however it's possible
2789 * (specifically in the rare L2ARC write race mentioned in
2790 * arc_buf_alloc_impl()) there will be an existing uncompressed buf that
2791 * is sharable, but wasn't at the time of its allocation. Rather than
2792 * allow a new shared uncompressed buf to be created and then shuffle
2793 * the list around to make it the last element, this simply disallows
2794 * sharing if the new buf isn't the first to be added.
2796 ASSERT3P(buf
->b_hdr
, ==, hdr
);
2797 boolean_t hdr_compressed
=
2798 arc_hdr_get_compress(hdr
) != ZIO_COMPRESS_OFF
;
2799 boolean_t buf_compressed
= ARC_BUF_COMPRESSED(buf
) != 0;
2800 return (!ARC_BUF_ENCRYPTED(buf
) &&
2801 buf_compressed
== hdr_compressed
&&
2802 hdr
->b_l1hdr
.b_byteswap
== DMU_BSWAP_NUMFUNCS
&&
2803 !HDR_SHARED_DATA(hdr
) &&
2804 (ARC_BUF_LAST(buf
) || ARC_BUF_COMPRESSED(buf
)));
2808 * Allocate a buf for this hdr. If you care about the data that's in the hdr,
2809 * or if you want a compressed buffer, pass those flags in. Returns 0 if the
2810 * copy was made successfully, or an error code otherwise.
2813 arc_buf_alloc_impl(arc_buf_hdr_t
*hdr
, spa_t
*spa
, uint64_t dsobj
, void *tag
,
2814 boolean_t encrypted
, boolean_t compressed
, boolean_t noauth
,
2815 boolean_t fill
, arc_buf_t
**ret
)
2818 arc_fill_flags_t flags
= ARC_FILL_LOCKED
;
2820 ASSERT(HDR_HAS_L1HDR(hdr
));
2821 ASSERT3U(HDR_GET_LSIZE(hdr
), >, 0);
2822 VERIFY(hdr
->b_type
== ARC_BUFC_DATA
||
2823 hdr
->b_type
== ARC_BUFC_METADATA
);
2824 ASSERT3P(ret
, !=, NULL
);
2825 ASSERT3P(*ret
, ==, NULL
);
2826 IMPLY(encrypted
, compressed
);
2828 hdr
->b_l1hdr
.b_mru_hits
= 0;
2829 hdr
->b_l1hdr
.b_mru_ghost_hits
= 0;
2830 hdr
->b_l1hdr
.b_mfu_hits
= 0;
2831 hdr
->b_l1hdr
.b_mfu_ghost_hits
= 0;
2832 hdr
->b_l1hdr
.b_l2_hits
= 0;
2834 buf
= *ret
= kmem_cache_alloc(buf_cache
, KM_PUSHPAGE
);
2837 buf
->b_next
= hdr
->b_l1hdr
.b_buf
;
2840 add_reference(hdr
, tag
);
2843 * We're about to change the hdr's b_flags. We must either
2844 * hold the hash_lock or be undiscoverable.
2846 ASSERT(MUTEX_HELD(HDR_LOCK(hdr
)) || HDR_EMPTY(hdr
));
2849 * Only honor requests for compressed bufs if the hdr is actually
2850 * compressed. This must be overriden if the buffer is encrypted since
2851 * encrypted buffers cannot be decompressed.
2854 buf
->b_flags
|= ARC_BUF_FLAG_COMPRESSED
;
2855 buf
->b_flags
|= ARC_BUF_FLAG_ENCRYPTED
;
2856 flags
|= ARC_FILL_COMPRESSED
| ARC_FILL_ENCRYPTED
;
2857 } else if (compressed
&&
2858 arc_hdr_get_compress(hdr
) != ZIO_COMPRESS_OFF
) {
2859 buf
->b_flags
|= ARC_BUF_FLAG_COMPRESSED
;
2860 flags
|= ARC_FILL_COMPRESSED
;
2865 flags
|= ARC_FILL_NOAUTH
;
2869 * If the hdr's data can be shared then we share the data buffer and
2870 * set the appropriate bit in the hdr's b_flags to indicate the hdr is
2871 * allocate a new buffer to store the buf's data.
2873 * There are two additional restrictions here because we're sharing
2874 * hdr -> buf instead of the usual buf -> hdr. First, the hdr can't be
2875 * actively involved in an L2ARC write, because if this buf is used by
2876 * an arc_write() then the hdr's data buffer will be released when the
2877 * write completes, even though the L2ARC write might still be using it.
2878 * Second, the hdr's ABD must be linear so that the buf's user doesn't
2879 * need to be ABD-aware.
2881 boolean_t can_share
= arc_can_share(hdr
, buf
) && !HDR_L2_WRITING(hdr
) &&
2882 hdr
->b_l1hdr
.b_pabd
!= NULL
&& abd_is_linear(hdr
->b_l1hdr
.b_pabd
);
2884 /* Set up b_data and sharing */
2886 buf
->b_data
= abd_to_buf(hdr
->b_l1hdr
.b_pabd
);
2887 buf
->b_flags
|= ARC_BUF_FLAG_SHARED
;
2888 arc_hdr_set_flags(hdr
, ARC_FLAG_SHARED_DATA
);
2891 arc_get_data_buf(hdr
, arc_buf_size(buf
), buf
);
2892 ARCSTAT_INCR(arcstat_overhead_size
, arc_buf_size(buf
));
2894 VERIFY3P(buf
->b_data
, !=, NULL
);
2896 hdr
->b_l1hdr
.b_buf
= buf
;
2897 hdr
->b_l1hdr
.b_bufcnt
+= 1;
2899 hdr
->b_crypt_hdr
.b_ebufcnt
+= 1;
2902 * If the user wants the data from the hdr, we need to either copy or
2903 * decompress the data.
2906 return (arc_buf_fill(buf
, spa
, dsobj
, flags
));
2912 static char *arc_onloan_tag
= "onloan";
2915 arc_loaned_bytes_update(int64_t delta
)
2917 atomic_add_64(&arc_loaned_bytes
, delta
);
2919 /* assert that it did not wrap around */
2920 ASSERT3S(atomic_add_64_nv(&arc_loaned_bytes
, 0), >=, 0);
2924 * Loan out an anonymous arc buffer. Loaned buffers are not counted as in
2925 * flight data by arc_tempreserve_space() until they are "returned". Loaned
2926 * buffers must be returned to the arc before they can be used by the DMU or
2930 arc_loan_buf(spa_t
*spa
, boolean_t is_metadata
, int size
)
2932 arc_buf_t
*buf
= arc_alloc_buf(spa
, arc_onloan_tag
,
2933 is_metadata
? ARC_BUFC_METADATA
: ARC_BUFC_DATA
, size
);
2935 arc_loaned_bytes_update(arc_buf_size(buf
));
2941 arc_loan_compressed_buf(spa_t
*spa
, uint64_t psize
, uint64_t lsize
,
2942 enum zio_compress compression_type
)
2944 arc_buf_t
*buf
= arc_alloc_compressed_buf(spa
, arc_onloan_tag
,
2945 psize
, lsize
, compression_type
);
2947 arc_loaned_bytes_update(arc_buf_size(buf
));
2953 arc_loan_raw_buf(spa_t
*spa
, uint64_t dsobj
, boolean_t byteorder
,
2954 const uint8_t *salt
, const uint8_t *iv
, const uint8_t *mac
,
2955 dmu_object_type_t ot
, uint64_t psize
, uint64_t lsize
,
2956 enum zio_compress compression_type
)
2958 arc_buf_t
*buf
= arc_alloc_raw_buf(spa
, arc_onloan_tag
, dsobj
,
2959 byteorder
, salt
, iv
, mac
, ot
, psize
, lsize
, compression_type
);
2961 atomic_add_64(&arc_loaned_bytes
, psize
);
2967 * Return a loaned arc buffer to the arc.
2970 arc_return_buf(arc_buf_t
*buf
, void *tag
)
2972 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
2974 ASSERT3P(buf
->b_data
, !=, NULL
);
2975 ASSERT(HDR_HAS_L1HDR(hdr
));
2976 (void) refcount_add(&hdr
->b_l1hdr
.b_refcnt
, tag
);
2977 (void) refcount_remove(&hdr
->b_l1hdr
.b_refcnt
, arc_onloan_tag
);
2979 arc_loaned_bytes_update(-arc_buf_size(buf
));
2982 /* Detach an arc_buf from a dbuf (tag) */
2984 arc_loan_inuse_buf(arc_buf_t
*buf
, void *tag
)
2986 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
2988 ASSERT3P(buf
->b_data
, !=, NULL
);
2989 ASSERT(HDR_HAS_L1HDR(hdr
));
2990 (void) refcount_add(&hdr
->b_l1hdr
.b_refcnt
, arc_onloan_tag
);
2991 (void) refcount_remove(&hdr
->b_l1hdr
.b_refcnt
, tag
);
2993 arc_loaned_bytes_update(arc_buf_size(buf
));
2997 l2arc_free_abd_on_write(abd_t
*abd
, size_t size
, arc_buf_contents_t type
)
2999 l2arc_data_free_t
*df
= kmem_alloc(sizeof (*df
), KM_SLEEP
);
3002 df
->l2df_size
= size
;
3003 df
->l2df_type
= type
;
3004 mutex_enter(&l2arc_free_on_write_mtx
);
3005 list_insert_head(l2arc_free_on_write
, df
);
3006 mutex_exit(&l2arc_free_on_write_mtx
);
3010 arc_hdr_free_on_write(arc_buf_hdr_t
*hdr
, boolean_t free_rdata
)
3012 arc_state_t
*state
= hdr
->b_l1hdr
.b_state
;
3013 arc_buf_contents_t type
= arc_buf_type(hdr
);
3014 uint64_t size
= (free_rdata
) ? HDR_GET_PSIZE(hdr
) : arc_hdr_size(hdr
);
3016 /* protected by hash lock, if in the hash table */
3017 if (multilist_link_active(&hdr
->b_l1hdr
.b_arc_node
)) {
3018 ASSERT(refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
));
3019 ASSERT(state
!= arc_anon
&& state
!= arc_l2c_only
);
3021 (void) refcount_remove_many(&state
->arcs_esize
[type
],
3024 (void) refcount_remove_many(&state
->arcs_size
, size
, hdr
);
3025 if (type
== ARC_BUFC_METADATA
) {
3026 arc_space_return(size
, ARC_SPACE_META
);
3028 ASSERT(type
== ARC_BUFC_DATA
);
3029 arc_space_return(size
, ARC_SPACE_DATA
);
3033 l2arc_free_abd_on_write(hdr
->b_crypt_hdr
.b_rabd
, size
, type
);
3035 l2arc_free_abd_on_write(hdr
->b_l1hdr
.b_pabd
, size
, type
);
3040 * Share the arc_buf_t's data with the hdr. Whenever we are sharing the
3041 * data buffer, we transfer the refcount ownership to the hdr and update
3042 * the appropriate kstats.
3045 arc_share_buf(arc_buf_hdr_t
*hdr
, arc_buf_t
*buf
)
3047 ASSERT(arc_can_share(hdr
, buf
));
3048 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, ==, NULL
);
3049 ASSERT(!ARC_BUF_ENCRYPTED(buf
));
3050 ASSERT(MUTEX_HELD(HDR_LOCK(hdr
)) || HDR_EMPTY(hdr
));
3053 * Start sharing the data buffer. We transfer the
3054 * refcount ownership to the hdr since it always owns
3055 * the refcount whenever an arc_buf_t is shared.
3057 refcount_transfer_ownership(&hdr
->b_l1hdr
.b_state
->arcs_size
, buf
, hdr
);
3058 hdr
->b_l1hdr
.b_pabd
= abd_get_from_buf(buf
->b_data
, arc_buf_size(buf
));
3059 abd_take_ownership_of_buf(hdr
->b_l1hdr
.b_pabd
,
3060 HDR_ISTYPE_METADATA(hdr
));
3061 arc_hdr_set_flags(hdr
, ARC_FLAG_SHARED_DATA
);
3062 buf
->b_flags
|= ARC_BUF_FLAG_SHARED
;
3065 * Since we've transferred ownership to the hdr we need
3066 * to increment its compressed and uncompressed kstats and
3067 * decrement the overhead size.
3069 ARCSTAT_INCR(arcstat_compressed_size
, arc_hdr_size(hdr
));
3070 ARCSTAT_INCR(arcstat_uncompressed_size
, HDR_GET_LSIZE(hdr
));
3071 ARCSTAT_INCR(arcstat_overhead_size
, -arc_buf_size(buf
));
3075 arc_unshare_buf(arc_buf_hdr_t
*hdr
, arc_buf_t
*buf
)
3077 ASSERT(arc_buf_is_shared(buf
));
3078 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, !=, NULL
);
3079 ASSERT(MUTEX_HELD(HDR_LOCK(hdr
)) || HDR_EMPTY(hdr
));
3082 * We are no longer sharing this buffer so we need
3083 * to transfer its ownership to the rightful owner.
3085 refcount_transfer_ownership(&hdr
->b_l1hdr
.b_state
->arcs_size
, hdr
, buf
);
3086 arc_hdr_clear_flags(hdr
, ARC_FLAG_SHARED_DATA
);
3087 abd_release_ownership_of_buf(hdr
->b_l1hdr
.b_pabd
);
3088 abd_put(hdr
->b_l1hdr
.b_pabd
);
3089 hdr
->b_l1hdr
.b_pabd
= NULL
;
3090 buf
->b_flags
&= ~ARC_BUF_FLAG_SHARED
;
3093 * Since the buffer is no longer shared between
3094 * the arc buf and the hdr, count it as overhead.
3096 ARCSTAT_INCR(arcstat_compressed_size
, -arc_hdr_size(hdr
));
3097 ARCSTAT_INCR(arcstat_uncompressed_size
, -HDR_GET_LSIZE(hdr
));
3098 ARCSTAT_INCR(arcstat_overhead_size
, arc_buf_size(buf
));
3102 * Remove an arc_buf_t from the hdr's buf list and return the last
3103 * arc_buf_t on the list. If no buffers remain on the list then return
3107 arc_buf_remove(arc_buf_hdr_t
*hdr
, arc_buf_t
*buf
)
3109 ASSERT(HDR_HAS_L1HDR(hdr
));
3110 ASSERT(MUTEX_HELD(HDR_LOCK(hdr
)) || HDR_EMPTY(hdr
));
3112 arc_buf_t
**bufp
= &hdr
->b_l1hdr
.b_buf
;
3113 arc_buf_t
*lastbuf
= NULL
;
3116 * Remove the buf from the hdr list and locate the last
3117 * remaining buffer on the list.
3119 while (*bufp
!= NULL
) {
3121 *bufp
= buf
->b_next
;
3124 * If we've removed a buffer in the middle of
3125 * the list then update the lastbuf and update
3128 if (*bufp
!= NULL
) {
3130 bufp
= &(*bufp
)->b_next
;
3134 ASSERT3P(lastbuf
, !=, buf
);
3135 IMPLY(hdr
->b_l1hdr
.b_bufcnt
> 0, lastbuf
!= NULL
);
3136 IMPLY(hdr
->b_l1hdr
.b_bufcnt
> 0, hdr
->b_l1hdr
.b_buf
!= NULL
);
3137 IMPLY(lastbuf
!= NULL
, ARC_BUF_LAST(lastbuf
));
3143 * Free up buf->b_data and pull the arc_buf_t off of the the arc_buf_hdr_t's
3147 arc_buf_destroy_impl(arc_buf_t
*buf
)
3149 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
3152 * Free up the data associated with the buf but only if we're not
3153 * sharing this with the hdr. If we are sharing it with the hdr, the
3154 * hdr is responsible for doing the free.
3156 if (buf
->b_data
!= NULL
) {
3158 * We're about to change the hdr's b_flags. We must either
3159 * hold the hash_lock or be undiscoverable.
3161 ASSERT(MUTEX_HELD(HDR_LOCK(hdr
)) || HDR_EMPTY(hdr
));
3163 arc_cksum_verify(buf
);
3164 arc_buf_unwatch(buf
);
3166 if (arc_buf_is_shared(buf
)) {
3167 arc_hdr_clear_flags(hdr
, ARC_FLAG_SHARED_DATA
);
3169 uint64_t size
= arc_buf_size(buf
);
3170 arc_free_data_buf(hdr
, buf
->b_data
, size
, buf
);
3171 ARCSTAT_INCR(arcstat_overhead_size
, -size
);
3175 ASSERT(hdr
->b_l1hdr
.b_bufcnt
> 0);
3176 hdr
->b_l1hdr
.b_bufcnt
-= 1;
3178 if (ARC_BUF_ENCRYPTED(buf
)) {
3179 hdr
->b_crypt_hdr
.b_ebufcnt
-= 1;
3182 * If we have no more encrypted buffers and we've
3183 * already gotten a copy of the decrypted data we can
3184 * free b_rabd to save some space.
3186 if (hdr
->b_crypt_hdr
.b_ebufcnt
== 0 &&
3187 HDR_HAS_RABD(hdr
) && hdr
->b_l1hdr
.b_pabd
!= NULL
&&
3188 !HDR_IO_IN_PROGRESS(hdr
)) {
3189 arc_hdr_free_abd(hdr
, B_TRUE
);
3194 arc_buf_t
*lastbuf
= arc_buf_remove(hdr
, buf
);
3196 if (ARC_BUF_SHARED(buf
) && !ARC_BUF_COMPRESSED(buf
)) {
3198 * If the current arc_buf_t is sharing its data buffer with the
3199 * hdr, then reassign the hdr's b_pabd to share it with the new
3200 * buffer at the end of the list. The shared buffer is always
3201 * the last one on the hdr's buffer list.
3203 * There is an equivalent case for compressed bufs, but since
3204 * they aren't guaranteed to be the last buf in the list and
3205 * that is an exceedingly rare case, we just allow that space be
3206 * wasted temporarily. We must also be careful not to share
3207 * encrypted buffers, since they cannot be shared.
3209 if (lastbuf
!= NULL
&& !ARC_BUF_ENCRYPTED(lastbuf
)) {
3210 /* Only one buf can be shared at once */
3211 VERIFY(!arc_buf_is_shared(lastbuf
));
3212 /* hdr is uncompressed so can't have compressed buf */
3213 VERIFY(!ARC_BUF_COMPRESSED(lastbuf
));
3215 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, !=, NULL
);
3216 arc_hdr_free_abd(hdr
, B_FALSE
);
3219 * We must setup a new shared block between the
3220 * last buffer and the hdr. The data would have
3221 * been allocated by the arc buf so we need to transfer
3222 * ownership to the hdr since it's now being shared.
3224 arc_share_buf(hdr
, lastbuf
);
3226 } else if (HDR_SHARED_DATA(hdr
)) {
3228 * Uncompressed shared buffers are always at the end
3229 * of the list. Compressed buffers don't have the
3230 * same requirements. This makes it hard to
3231 * simply assert that the lastbuf is shared so
3232 * we rely on the hdr's compression flags to determine
3233 * if we have a compressed, shared buffer.
3235 ASSERT3P(lastbuf
, !=, NULL
);
3236 ASSERT(arc_buf_is_shared(lastbuf
) ||
3237 arc_hdr_get_compress(hdr
) != ZIO_COMPRESS_OFF
);
3241 * Free the checksum if we're removing the last uncompressed buf from
3244 if (!arc_hdr_has_uncompressed_buf(hdr
)) {
3245 arc_cksum_free(hdr
);
3248 /* clean up the buf */
3250 kmem_cache_free(buf_cache
, buf
);
3254 arc_hdr_alloc_abd(arc_buf_hdr_t
*hdr
, boolean_t alloc_rdata
)
3258 ASSERT3U(HDR_GET_LSIZE(hdr
), >, 0);
3259 ASSERT(HDR_HAS_L1HDR(hdr
));
3260 ASSERT(!HDR_SHARED_DATA(hdr
) || alloc_rdata
);
3261 IMPLY(alloc_rdata
, HDR_PROTECTED(hdr
));
3264 size
= HDR_GET_PSIZE(hdr
);
3265 ASSERT3P(hdr
->b_crypt_hdr
.b_rabd
, ==, NULL
);
3266 hdr
->b_crypt_hdr
.b_rabd
= arc_get_data_abd(hdr
, size
, hdr
);
3267 ASSERT3P(hdr
->b_crypt_hdr
.b_rabd
, !=, NULL
);
3268 ARCSTAT_INCR(arcstat_raw_size
, size
);
3270 size
= arc_hdr_size(hdr
);
3271 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, ==, NULL
);
3272 hdr
->b_l1hdr
.b_pabd
= arc_get_data_abd(hdr
, size
, hdr
);
3273 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, !=, NULL
);
3276 ARCSTAT_INCR(arcstat_compressed_size
, size
);
3277 ARCSTAT_INCR(arcstat_uncompressed_size
, HDR_GET_LSIZE(hdr
));
3281 arc_hdr_free_abd(arc_buf_hdr_t
*hdr
, boolean_t free_rdata
)
3283 uint64_t size
= (free_rdata
) ? HDR_GET_PSIZE(hdr
) : arc_hdr_size(hdr
);
3285 ASSERT(HDR_HAS_L1HDR(hdr
));
3286 ASSERT(hdr
->b_l1hdr
.b_pabd
!= NULL
|| HDR_HAS_RABD(hdr
));
3287 IMPLY(free_rdata
, HDR_HAS_RABD(hdr
));
3290 * If the hdr is currently being written to the l2arc then
3291 * we defer freeing the data by adding it to the l2arc_free_on_write
3292 * list. The l2arc will free the data once it's finished
3293 * writing it to the l2arc device.
3295 if (HDR_L2_WRITING(hdr
)) {
3296 arc_hdr_free_on_write(hdr
, free_rdata
);
3297 ARCSTAT_BUMP(arcstat_l2_free_on_write
);
3298 } else if (free_rdata
) {
3299 arc_free_data_abd(hdr
, hdr
->b_crypt_hdr
.b_rabd
, size
, hdr
);
3301 arc_free_data_abd(hdr
, hdr
->b_l1hdr
.b_pabd
, size
, hdr
);
3305 hdr
->b_crypt_hdr
.b_rabd
= NULL
;
3306 ARCSTAT_INCR(arcstat_raw_size
, -size
);
3308 hdr
->b_l1hdr
.b_pabd
= NULL
;
3311 if (hdr
->b_l1hdr
.b_pabd
== NULL
&& !HDR_HAS_RABD(hdr
))
3312 hdr
->b_l1hdr
.b_byteswap
= DMU_BSWAP_NUMFUNCS
;
3314 ARCSTAT_INCR(arcstat_compressed_size
, -size
);
3315 ARCSTAT_INCR(arcstat_uncompressed_size
, -HDR_GET_LSIZE(hdr
));
3318 static arc_buf_hdr_t
*
3319 arc_hdr_alloc(uint64_t spa
, int32_t psize
, int32_t lsize
,
3320 boolean_t
protected, enum zio_compress compression_type
,
3321 arc_buf_contents_t type
, boolean_t alloc_rdata
)
3325 VERIFY(type
== ARC_BUFC_DATA
|| type
== ARC_BUFC_METADATA
);
3327 hdr
= kmem_cache_alloc(hdr_full_crypt_cache
, KM_PUSHPAGE
);
3329 hdr
= kmem_cache_alloc(hdr_full_cache
, KM_PUSHPAGE
);
3332 ASSERT(HDR_EMPTY(hdr
));
3333 ASSERT3P(hdr
->b_l1hdr
.b_freeze_cksum
, ==, NULL
);
3334 HDR_SET_PSIZE(hdr
, psize
);
3335 HDR_SET_LSIZE(hdr
, lsize
);
3339 arc_hdr_set_flags(hdr
, arc_bufc_to_flags(type
) | ARC_FLAG_HAS_L1HDR
);
3340 arc_hdr_set_compress(hdr
, compression_type
);
3342 arc_hdr_set_flags(hdr
, ARC_FLAG_PROTECTED
);
3344 hdr
->b_l1hdr
.b_state
= arc_anon
;
3345 hdr
->b_l1hdr
.b_arc_access
= 0;
3346 hdr
->b_l1hdr
.b_bufcnt
= 0;
3347 hdr
->b_l1hdr
.b_buf
= NULL
;
3350 * Allocate the hdr's buffer. This will contain either
3351 * the compressed or uncompressed data depending on the block
3352 * it references and compressed arc enablement.
3354 arc_hdr_alloc_abd(hdr
, alloc_rdata
);
3355 ASSERT(refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
));
3361 * Transition between the two allocation states for the arc_buf_hdr struct.
3362 * The arc_buf_hdr struct can be allocated with (hdr_full_cache) or without
3363 * (hdr_l2only_cache) the fields necessary for the L1 cache - the smaller
3364 * version is used when a cache buffer is only in the L2ARC in order to reduce
3367 static arc_buf_hdr_t
*
3368 arc_hdr_realloc(arc_buf_hdr_t
*hdr
, kmem_cache_t
*old
, kmem_cache_t
*new)
3370 ASSERT(HDR_HAS_L2HDR(hdr
));
3372 arc_buf_hdr_t
*nhdr
;
3373 l2arc_dev_t
*dev
= hdr
->b_l2hdr
.b_dev
;
3375 ASSERT((old
== hdr_full_cache
&& new == hdr_l2only_cache
) ||
3376 (old
== hdr_l2only_cache
&& new == hdr_full_cache
));
3379 * if the caller wanted a new full header and the header is to be
3380 * encrypted we will actually allocate the header from the full crypt
3381 * cache instead. The same applies to freeing from the old cache.
3383 if (HDR_PROTECTED(hdr
) && new == hdr_full_cache
)
3384 new = hdr_full_crypt_cache
;
3385 if (HDR_PROTECTED(hdr
) && old
== hdr_full_cache
)
3386 old
= hdr_full_crypt_cache
;
3388 nhdr
= kmem_cache_alloc(new, KM_PUSHPAGE
);
3390 ASSERT(MUTEX_HELD(HDR_LOCK(hdr
)));
3391 buf_hash_remove(hdr
);
3393 bcopy(hdr
, nhdr
, HDR_L2ONLY_SIZE
);
3395 if (new == hdr_full_cache
|| new == hdr_full_crypt_cache
) {
3396 arc_hdr_set_flags(nhdr
, ARC_FLAG_HAS_L1HDR
);
3398 * arc_access and arc_change_state need to be aware that a
3399 * header has just come out of L2ARC, so we set its state to
3400 * l2c_only even though it's about to change.
3402 nhdr
->b_l1hdr
.b_state
= arc_l2c_only
;
3404 /* Verify previous threads set to NULL before freeing */
3405 ASSERT3P(nhdr
->b_l1hdr
.b_pabd
, ==, NULL
);
3406 ASSERT(!HDR_HAS_RABD(hdr
));
3408 ASSERT3P(hdr
->b_l1hdr
.b_buf
, ==, NULL
);
3409 ASSERT0(hdr
->b_l1hdr
.b_bufcnt
);
3410 ASSERT3P(hdr
->b_l1hdr
.b_freeze_cksum
, ==, NULL
);
3413 * If we've reached here, We must have been called from
3414 * arc_evict_hdr(), as such we should have already been
3415 * removed from any ghost list we were previously on
3416 * (which protects us from racing with arc_evict_state),
3417 * thus no locking is needed during this check.
3419 ASSERT(!multilist_link_active(&hdr
->b_l1hdr
.b_arc_node
));
3422 * A buffer must not be moved into the arc_l2c_only
3423 * state if it's not finished being written out to the
3424 * l2arc device. Otherwise, the b_l1hdr.b_pabd field
3425 * might try to be accessed, even though it was removed.
3427 VERIFY(!HDR_L2_WRITING(hdr
));
3428 VERIFY3P(hdr
->b_l1hdr
.b_pabd
, ==, NULL
);
3429 ASSERT(!HDR_HAS_RABD(hdr
));
3431 arc_hdr_clear_flags(nhdr
, ARC_FLAG_HAS_L1HDR
);
3434 * The header has been reallocated so we need to re-insert it into any
3437 (void) buf_hash_insert(nhdr
, NULL
);
3439 ASSERT(list_link_active(&hdr
->b_l2hdr
.b_l2node
));
3441 mutex_enter(&dev
->l2ad_mtx
);
3444 * We must place the realloc'ed header back into the list at
3445 * the same spot. Otherwise, if it's placed earlier in the list,
3446 * l2arc_write_buffers() could find it during the function's
3447 * write phase, and try to write it out to the l2arc.
3449 list_insert_after(&dev
->l2ad_buflist
, hdr
, nhdr
);
3450 list_remove(&dev
->l2ad_buflist
, hdr
);
3452 mutex_exit(&dev
->l2ad_mtx
);
3455 * Since we're using the pointer address as the tag when
3456 * incrementing and decrementing the l2ad_alloc refcount, we
3457 * must remove the old pointer (that we're about to destroy) and
3458 * add the new pointer to the refcount. Otherwise we'd remove
3459 * the wrong pointer address when calling arc_hdr_destroy() later.
3462 (void) refcount_remove_many(&dev
->l2ad_alloc
, arc_hdr_size(hdr
), hdr
);
3463 (void) refcount_add_many(&dev
->l2ad_alloc
, arc_hdr_size(nhdr
), nhdr
);
3465 buf_discard_identity(hdr
);
3466 kmem_cache_free(old
, hdr
);
3472 * This function allows an L1 header to be reallocated as a crypt
3473 * header and vice versa. If we are going to a crypt header, the
3474 * new fields will be zeroed out.
3476 static arc_buf_hdr_t
*
3477 arc_hdr_realloc_crypt(arc_buf_hdr_t
*hdr
, boolean_t need_crypt
)
3479 arc_buf_hdr_t
*nhdr
;
3481 kmem_cache_t
*ncache
, *ocache
;
3483 ASSERT(HDR_HAS_L1HDR(hdr
));
3484 ASSERT3U(!!HDR_PROTECTED(hdr
), !=, need_crypt
);
3485 ASSERT3P(hdr
->b_l1hdr
.b_state
, ==, arc_anon
);
3486 ASSERT(!multilist_link_active(&hdr
->b_l1hdr
.b_arc_node
));
3489 ncache
= hdr_full_crypt_cache
;
3490 ocache
= hdr_full_cache
;
3492 ncache
= hdr_full_cache
;
3493 ocache
= hdr_full_crypt_cache
;
3496 nhdr
= kmem_cache_alloc(ncache
, KM_PUSHPAGE
);
3497 bcopy(hdr
, nhdr
, HDR_L2ONLY_SIZE
);
3498 nhdr
->b_l1hdr
.b_freeze_cksum
= hdr
->b_l1hdr
.b_freeze_cksum
;
3499 nhdr
->b_l1hdr
.b_bufcnt
= hdr
->b_l1hdr
.b_bufcnt
;
3500 nhdr
->b_l1hdr
.b_byteswap
= hdr
->b_l1hdr
.b_byteswap
;
3501 nhdr
->b_l1hdr
.b_state
= hdr
->b_l1hdr
.b_state
;
3502 nhdr
->b_l1hdr
.b_arc_access
= hdr
->b_l1hdr
.b_arc_access
;
3503 nhdr
->b_l1hdr
.b_mru_hits
= hdr
->b_l1hdr
.b_mru_hits
;
3504 nhdr
->b_l1hdr
.b_mru_ghost_hits
= hdr
->b_l1hdr
.b_mru_ghost_hits
;
3505 nhdr
->b_l1hdr
.b_mfu_hits
= hdr
->b_l1hdr
.b_mfu_hits
;
3506 nhdr
->b_l1hdr
.b_mfu_ghost_hits
= hdr
->b_l1hdr
.b_mfu_ghost_hits
;
3507 nhdr
->b_l1hdr
.b_l2_hits
= hdr
->b_l1hdr
.b_l2_hits
;
3508 nhdr
->b_l1hdr
.b_acb
= hdr
->b_l1hdr
.b_acb
;
3509 nhdr
->b_l1hdr
.b_pabd
= hdr
->b_l1hdr
.b_pabd
;
3510 nhdr
->b_l1hdr
.b_buf
= hdr
->b_l1hdr
.b_buf
;
3513 * This refcount_add() exists only to ensure that the individual
3514 * arc buffers always point to a header that is referenced, avoiding
3515 * a small race condition that could trigger ASSERTs.
3517 (void) refcount_add(&nhdr
->b_l1hdr
.b_refcnt
, FTAG
);
3519 for (buf
= nhdr
->b_l1hdr
.b_buf
; buf
!= NULL
; buf
= buf
->b_next
) {
3520 mutex_enter(&buf
->b_evict_lock
);
3522 mutex_exit(&buf
->b_evict_lock
);
3525 refcount_transfer(&nhdr
->b_l1hdr
.b_refcnt
, &hdr
->b_l1hdr
.b_refcnt
);
3526 (void) refcount_remove(&nhdr
->b_l1hdr
.b_refcnt
, FTAG
);
3529 arc_hdr_set_flags(nhdr
, ARC_FLAG_PROTECTED
);
3531 arc_hdr_clear_flags(nhdr
, ARC_FLAG_PROTECTED
);
3534 buf_discard_identity(hdr
);
3535 kmem_cache_free(ocache
, hdr
);
3541 * This function is used by the send / receive code to convert a newly
3542 * allocated arc_buf_t to one that is suitable for a raw encrypted write. It
3543 * is also used to allow the root objset block to be uupdated without altering
3544 * its embedded MACs. Both block types will always be uncompressed so we do not
3545 * have to worry about compression type or psize.
3548 arc_convert_to_raw(arc_buf_t
*buf
, uint64_t dsobj
, boolean_t byteorder
,
3549 dmu_object_type_t ot
, const uint8_t *salt
, const uint8_t *iv
,
3552 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
3554 ASSERT(ot
== DMU_OT_DNODE
|| ot
== DMU_OT_OBJSET
);
3555 ASSERT(HDR_HAS_L1HDR(hdr
));
3556 ASSERT3P(hdr
->b_l1hdr
.b_state
, ==, arc_anon
);
3558 buf
->b_flags
|= (ARC_BUF_FLAG_COMPRESSED
| ARC_BUF_FLAG_ENCRYPTED
);
3559 if (!HDR_PROTECTED(hdr
))
3560 hdr
= arc_hdr_realloc_crypt(hdr
, B_TRUE
);
3561 hdr
->b_crypt_hdr
.b_dsobj
= dsobj
;
3562 hdr
->b_crypt_hdr
.b_ot
= ot
;
3563 hdr
->b_l1hdr
.b_byteswap
= (byteorder
== ZFS_HOST_BYTEORDER
) ?
3564 DMU_BSWAP_NUMFUNCS
: DMU_OT_BYTESWAP(ot
);
3565 if (!arc_hdr_has_uncompressed_buf(hdr
))
3566 arc_cksum_free(hdr
);
3569 bcopy(salt
, hdr
->b_crypt_hdr
.b_salt
, ZIO_DATA_SALT_LEN
);
3571 bcopy(iv
, hdr
->b_crypt_hdr
.b_iv
, ZIO_DATA_IV_LEN
);
3573 bcopy(mac
, hdr
->b_crypt_hdr
.b_mac
, ZIO_DATA_MAC_LEN
);
3577 * Allocate a new arc_buf_hdr_t and arc_buf_t and return the buf to the caller.
3578 * The buf is returned thawed since we expect the consumer to modify it.
3581 arc_alloc_buf(spa_t
*spa
, void *tag
, arc_buf_contents_t type
, int32_t size
)
3583 arc_buf_hdr_t
*hdr
= arc_hdr_alloc(spa_load_guid(spa
), size
, size
,
3584 B_FALSE
, ZIO_COMPRESS_OFF
, type
, B_FALSE
);
3585 ASSERT(!MUTEX_HELD(HDR_LOCK(hdr
)));
3587 arc_buf_t
*buf
= NULL
;
3588 VERIFY0(arc_buf_alloc_impl(hdr
, spa
, 0, tag
, B_FALSE
, B_FALSE
,
3589 B_FALSE
, B_FALSE
, &buf
));
3596 * Allocate a compressed buf in the same manner as arc_alloc_buf. Don't use this
3597 * for bufs containing metadata.
3600 arc_alloc_compressed_buf(spa_t
*spa
, void *tag
, uint64_t psize
, uint64_t lsize
,
3601 enum zio_compress compression_type
)
3603 ASSERT3U(lsize
, >, 0);
3604 ASSERT3U(lsize
, >=, psize
);
3605 ASSERT3U(compression_type
, >, ZIO_COMPRESS_OFF
);
3606 ASSERT3U(compression_type
, <, ZIO_COMPRESS_FUNCTIONS
);
3608 arc_buf_hdr_t
*hdr
= arc_hdr_alloc(spa_load_guid(spa
), psize
, lsize
,
3609 B_FALSE
, compression_type
, ARC_BUFC_DATA
, B_FALSE
);
3610 ASSERT(!MUTEX_HELD(HDR_LOCK(hdr
)));
3612 arc_buf_t
*buf
= NULL
;
3613 VERIFY0(arc_buf_alloc_impl(hdr
, spa
, 0, tag
, B_FALSE
,
3614 B_TRUE
, B_FALSE
, B_FALSE
, &buf
));
3616 ASSERT3P(hdr
->b_l1hdr
.b_freeze_cksum
, ==, NULL
);
3618 if (!arc_buf_is_shared(buf
)) {
3620 * To ensure that the hdr has the correct data in it if we call
3621 * arc_untransform() on this buf before it's been written to
3622 * disk, it's easiest if we just set up sharing between the
3625 ASSERT(!abd_is_linear(hdr
->b_l1hdr
.b_pabd
));
3626 arc_hdr_free_abd(hdr
, B_FALSE
);
3627 arc_share_buf(hdr
, buf
);
3634 arc_alloc_raw_buf(spa_t
*spa
, void *tag
, uint64_t dsobj
, boolean_t byteorder
,
3635 const uint8_t *salt
, const uint8_t *iv
, const uint8_t *mac
,
3636 dmu_object_type_t ot
, uint64_t psize
, uint64_t lsize
,
3637 enum zio_compress compression_type
)
3641 arc_buf_contents_t type
= DMU_OT_IS_METADATA(ot
) ?
3642 ARC_BUFC_METADATA
: ARC_BUFC_DATA
;
3644 ASSERT3U(lsize
, >, 0);
3645 ASSERT3U(lsize
, >=, psize
);
3646 ASSERT3U(compression_type
, >=, ZIO_COMPRESS_OFF
);
3647 ASSERT3U(compression_type
, <, ZIO_COMPRESS_FUNCTIONS
);
3649 hdr
= arc_hdr_alloc(spa_load_guid(spa
), psize
, lsize
, B_TRUE
,
3650 compression_type
, type
, B_TRUE
);
3651 ASSERT(!MUTEX_HELD(HDR_LOCK(hdr
)));
3653 hdr
->b_crypt_hdr
.b_dsobj
= dsobj
;
3654 hdr
->b_crypt_hdr
.b_ot
= ot
;
3655 hdr
->b_l1hdr
.b_byteswap
= (byteorder
== ZFS_HOST_BYTEORDER
) ?
3656 DMU_BSWAP_NUMFUNCS
: DMU_OT_BYTESWAP(ot
);
3657 bcopy(salt
, hdr
->b_crypt_hdr
.b_salt
, ZIO_DATA_SALT_LEN
);
3658 bcopy(iv
, hdr
->b_crypt_hdr
.b_iv
, ZIO_DATA_IV_LEN
);
3659 bcopy(mac
, hdr
->b_crypt_hdr
.b_mac
, ZIO_DATA_MAC_LEN
);
3662 * This buffer will be considered encrypted even if the ot is not an
3663 * encrypted type. It will become authenticated instead in
3664 * arc_write_ready().
3667 VERIFY0(arc_buf_alloc_impl(hdr
, spa
, dsobj
, tag
, B_TRUE
, B_TRUE
,
3668 B_FALSE
, B_FALSE
, &buf
));
3670 ASSERT3P(hdr
->b_l1hdr
.b_freeze_cksum
, ==, NULL
);
3676 arc_hdr_l2hdr_destroy(arc_buf_hdr_t
*hdr
)
3678 l2arc_buf_hdr_t
*l2hdr
= &hdr
->b_l2hdr
;
3679 l2arc_dev_t
*dev
= l2hdr
->b_dev
;
3680 uint64_t psize
= arc_hdr_size(hdr
);
3682 ASSERT(MUTEX_HELD(&dev
->l2ad_mtx
));
3683 ASSERT(HDR_HAS_L2HDR(hdr
));
3685 list_remove(&dev
->l2ad_buflist
, hdr
);
3687 ARCSTAT_INCR(arcstat_l2_psize
, -psize
);
3688 ARCSTAT_INCR(arcstat_l2_lsize
, -HDR_GET_LSIZE(hdr
));
3690 vdev_space_update(dev
->l2ad_vdev
, -psize
, 0, 0);
3692 (void) refcount_remove_many(&dev
->l2ad_alloc
, psize
, hdr
);
3693 arc_hdr_clear_flags(hdr
, ARC_FLAG_HAS_L2HDR
);
3697 arc_hdr_destroy(arc_buf_hdr_t
*hdr
)
3699 if (HDR_HAS_L1HDR(hdr
)) {
3700 ASSERT(hdr
->b_l1hdr
.b_buf
== NULL
||
3701 hdr
->b_l1hdr
.b_bufcnt
> 0);
3702 ASSERT(refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
));
3703 ASSERT3P(hdr
->b_l1hdr
.b_state
, ==, arc_anon
);
3705 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
3706 ASSERT(!HDR_IN_HASH_TABLE(hdr
));
3708 if (!HDR_EMPTY(hdr
))
3709 buf_discard_identity(hdr
);
3711 if (HDR_HAS_L2HDR(hdr
)) {
3712 l2arc_dev_t
*dev
= hdr
->b_l2hdr
.b_dev
;
3713 boolean_t buflist_held
= MUTEX_HELD(&dev
->l2ad_mtx
);
3716 mutex_enter(&dev
->l2ad_mtx
);
3719 * Even though we checked this conditional above, we
3720 * need to check this again now that we have the
3721 * l2ad_mtx. This is because we could be racing with
3722 * another thread calling l2arc_evict() which might have
3723 * destroyed this header's L2 portion as we were waiting
3724 * to acquire the l2ad_mtx. If that happens, we don't
3725 * want to re-destroy the header's L2 portion.
3727 if (HDR_HAS_L2HDR(hdr
))
3728 arc_hdr_l2hdr_destroy(hdr
);
3731 mutex_exit(&dev
->l2ad_mtx
);
3734 if (HDR_HAS_L1HDR(hdr
)) {
3735 arc_cksum_free(hdr
);
3737 while (hdr
->b_l1hdr
.b_buf
!= NULL
)
3738 arc_buf_destroy_impl(hdr
->b_l1hdr
.b_buf
);
3740 if (hdr
->b_l1hdr
.b_pabd
!= NULL
) {
3741 arc_hdr_free_abd(hdr
, B_FALSE
);
3744 if (HDR_HAS_RABD(hdr
))
3745 arc_hdr_free_abd(hdr
, B_TRUE
);
3748 ASSERT3P(hdr
->b_hash_next
, ==, NULL
);
3749 if (HDR_HAS_L1HDR(hdr
)) {
3750 ASSERT(!multilist_link_active(&hdr
->b_l1hdr
.b_arc_node
));
3751 ASSERT3P(hdr
->b_l1hdr
.b_acb
, ==, NULL
);
3753 if (!HDR_PROTECTED(hdr
)) {
3754 kmem_cache_free(hdr_full_cache
, hdr
);
3756 kmem_cache_free(hdr_full_crypt_cache
, hdr
);
3759 kmem_cache_free(hdr_l2only_cache
, hdr
);
3764 arc_buf_destroy(arc_buf_t
*buf
, void* tag
)
3766 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
3767 kmutex_t
*hash_lock
= HDR_LOCK(hdr
);
3769 if (hdr
->b_l1hdr
.b_state
== arc_anon
) {
3770 ASSERT3U(hdr
->b_l1hdr
.b_bufcnt
, ==, 1);
3771 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
3772 VERIFY0(remove_reference(hdr
, NULL
, tag
));
3773 arc_hdr_destroy(hdr
);
3777 mutex_enter(hash_lock
);
3778 ASSERT3P(hdr
, ==, buf
->b_hdr
);
3779 ASSERT(hdr
->b_l1hdr
.b_bufcnt
> 0);
3780 ASSERT3P(hash_lock
, ==, HDR_LOCK(hdr
));
3781 ASSERT3P(hdr
->b_l1hdr
.b_state
, !=, arc_anon
);
3782 ASSERT3P(buf
->b_data
, !=, NULL
);
3784 (void) remove_reference(hdr
, hash_lock
, tag
);
3785 arc_buf_destroy_impl(buf
);
3786 mutex_exit(hash_lock
);
3790 * Evict the arc_buf_hdr that is provided as a parameter. The resultant
3791 * state of the header is dependent on its state prior to entering this
3792 * function. The following transitions are possible:
3794 * - arc_mru -> arc_mru_ghost
3795 * - arc_mfu -> arc_mfu_ghost
3796 * - arc_mru_ghost -> arc_l2c_only
3797 * - arc_mru_ghost -> deleted
3798 * - arc_mfu_ghost -> arc_l2c_only
3799 * - arc_mfu_ghost -> deleted
3802 arc_evict_hdr(arc_buf_hdr_t
*hdr
, kmutex_t
*hash_lock
)
3804 arc_state_t
*evicted_state
, *state
;
3805 int64_t bytes_evicted
= 0;
3806 int min_lifetime
= HDR_PRESCIENT_PREFETCH(hdr
) ?
3807 arc_min_prescient_prefetch_ms
: arc_min_prefetch_ms
;
3809 ASSERT(MUTEX_HELD(hash_lock
));
3810 ASSERT(HDR_HAS_L1HDR(hdr
));
3812 state
= hdr
->b_l1hdr
.b_state
;
3813 if (GHOST_STATE(state
)) {
3814 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
3815 ASSERT3P(hdr
->b_l1hdr
.b_buf
, ==, NULL
);
3818 * l2arc_write_buffers() relies on a header's L1 portion
3819 * (i.e. its b_pabd field) during it's write phase.
3820 * Thus, we cannot push a header onto the arc_l2c_only
3821 * state (removing its L1 piece) until the header is
3822 * done being written to the l2arc.
3824 if (HDR_HAS_L2HDR(hdr
) && HDR_L2_WRITING(hdr
)) {
3825 ARCSTAT_BUMP(arcstat_evict_l2_skip
);
3826 return (bytes_evicted
);
3829 ARCSTAT_BUMP(arcstat_deleted
);
3830 bytes_evicted
+= HDR_GET_LSIZE(hdr
);
3832 DTRACE_PROBE1(arc__delete
, arc_buf_hdr_t
*, hdr
);
3834 if (HDR_HAS_L2HDR(hdr
)) {
3835 ASSERT(hdr
->b_l1hdr
.b_pabd
== NULL
);
3836 ASSERT(!HDR_HAS_RABD(hdr
));
3838 * This buffer is cached on the 2nd Level ARC;
3839 * don't destroy the header.
3841 arc_change_state(arc_l2c_only
, hdr
, hash_lock
);
3843 * dropping from L1+L2 cached to L2-only,
3844 * realloc to remove the L1 header.
3846 hdr
= arc_hdr_realloc(hdr
, hdr_full_cache
,
3849 arc_change_state(arc_anon
, hdr
, hash_lock
);
3850 arc_hdr_destroy(hdr
);
3852 return (bytes_evicted
);
3855 ASSERT(state
== arc_mru
|| state
== arc_mfu
);
3856 evicted_state
= (state
== arc_mru
) ? arc_mru_ghost
: arc_mfu_ghost
;
3858 /* prefetch buffers have a minimum lifespan */
3859 if (HDR_IO_IN_PROGRESS(hdr
) ||
3860 ((hdr
->b_flags
& (ARC_FLAG_PREFETCH
| ARC_FLAG_INDIRECT
)) &&
3861 ddi_get_lbolt() - hdr
->b_l1hdr
.b_arc_access
<
3862 MSEC_TO_TICK(min_lifetime
))) {
3863 ARCSTAT_BUMP(arcstat_evict_skip
);
3864 return (bytes_evicted
);
3867 ASSERT0(refcount_count(&hdr
->b_l1hdr
.b_refcnt
));
3868 while (hdr
->b_l1hdr
.b_buf
) {
3869 arc_buf_t
*buf
= hdr
->b_l1hdr
.b_buf
;
3870 if (!mutex_tryenter(&buf
->b_evict_lock
)) {
3871 ARCSTAT_BUMP(arcstat_mutex_miss
);
3874 if (buf
->b_data
!= NULL
)
3875 bytes_evicted
+= HDR_GET_LSIZE(hdr
);
3876 mutex_exit(&buf
->b_evict_lock
);
3877 arc_buf_destroy_impl(buf
);
3880 if (HDR_HAS_L2HDR(hdr
)) {
3881 ARCSTAT_INCR(arcstat_evict_l2_cached
, HDR_GET_LSIZE(hdr
));
3883 if (l2arc_write_eligible(hdr
->b_spa
, hdr
)) {
3884 ARCSTAT_INCR(arcstat_evict_l2_eligible
,
3885 HDR_GET_LSIZE(hdr
));
3887 ARCSTAT_INCR(arcstat_evict_l2_ineligible
,
3888 HDR_GET_LSIZE(hdr
));
3892 if (hdr
->b_l1hdr
.b_bufcnt
== 0) {
3893 arc_cksum_free(hdr
);
3895 bytes_evicted
+= arc_hdr_size(hdr
);
3898 * If this hdr is being evicted and has a compressed
3899 * buffer then we discard it here before we change states.
3900 * This ensures that the accounting is updated correctly
3901 * in arc_free_data_impl().
3903 if (hdr
->b_l1hdr
.b_pabd
!= NULL
)
3904 arc_hdr_free_abd(hdr
, B_FALSE
);
3906 if (HDR_HAS_RABD(hdr
))
3907 arc_hdr_free_abd(hdr
, B_TRUE
);
3909 arc_change_state(evicted_state
, hdr
, hash_lock
);
3910 ASSERT(HDR_IN_HASH_TABLE(hdr
));
3911 arc_hdr_set_flags(hdr
, ARC_FLAG_IN_HASH_TABLE
);
3912 DTRACE_PROBE1(arc__evict
, arc_buf_hdr_t
*, hdr
);
3915 return (bytes_evicted
);
3919 arc_evict_state_impl(multilist_t
*ml
, int idx
, arc_buf_hdr_t
*marker
,
3920 uint64_t spa
, int64_t bytes
)
3922 multilist_sublist_t
*mls
;
3923 uint64_t bytes_evicted
= 0;
3925 kmutex_t
*hash_lock
;
3926 int evict_count
= 0;
3928 ASSERT3P(marker
, !=, NULL
);
3929 IMPLY(bytes
< 0, bytes
== ARC_EVICT_ALL
);
3931 mls
= multilist_sublist_lock(ml
, idx
);
3933 for (hdr
= multilist_sublist_prev(mls
, marker
); hdr
!= NULL
;
3934 hdr
= multilist_sublist_prev(mls
, marker
)) {
3935 if ((bytes
!= ARC_EVICT_ALL
&& bytes_evicted
>= bytes
) ||
3936 (evict_count
>= zfs_arc_evict_batch_limit
))
3940 * To keep our iteration location, move the marker
3941 * forward. Since we're not holding hdr's hash lock, we
3942 * must be very careful and not remove 'hdr' from the
3943 * sublist. Otherwise, other consumers might mistake the
3944 * 'hdr' as not being on a sublist when they call the
3945 * multilist_link_active() function (they all rely on
3946 * the hash lock protecting concurrent insertions and
3947 * removals). multilist_sublist_move_forward() was
3948 * specifically implemented to ensure this is the case
3949 * (only 'marker' will be removed and re-inserted).
3951 multilist_sublist_move_forward(mls
, marker
);
3954 * The only case where the b_spa field should ever be
3955 * zero, is the marker headers inserted by
3956 * arc_evict_state(). It's possible for multiple threads
3957 * to be calling arc_evict_state() concurrently (e.g.
3958 * dsl_pool_close() and zio_inject_fault()), so we must
3959 * skip any markers we see from these other threads.
3961 if (hdr
->b_spa
== 0)
3964 /* we're only interested in evicting buffers of a certain spa */
3965 if (spa
!= 0 && hdr
->b_spa
!= spa
) {
3966 ARCSTAT_BUMP(arcstat_evict_skip
);
3970 hash_lock
= HDR_LOCK(hdr
);
3973 * We aren't calling this function from any code path
3974 * that would already be holding a hash lock, so we're
3975 * asserting on this assumption to be defensive in case
3976 * this ever changes. Without this check, it would be
3977 * possible to incorrectly increment arcstat_mutex_miss
3978 * below (e.g. if the code changed such that we called
3979 * this function with a hash lock held).
3981 ASSERT(!MUTEX_HELD(hash_lock
));
3983 if (mutex_tryenter(hash_lock
)) {
3984 uint64_t evicted
= arc_evict_hdr(hdr
, hash_lock
);
3985 mutex_exit(hash_lock
);
3987 bytes_evicted
+= evicted
;
3990 * If evicted is zero, arc_evict_hdr() must have
3991 * decided to skip this header, don't increment
3992 * evict_count in this case.
3998 * If arc_size isn't overflowing, signal any
3999 * threads that might happen to be waiting.
4001 * For each header evicted, we wake up a single
4002 * thread. If we used cv_broadcast, we could
4003 * wake up "too many" threads causing arc_size
4004 * to significantly overflow arc_c; since
4005 * arc_get_data_impl() doesn't check for overflow
4006 * when it's woken up (it doesn't because it's
4007 * possible for the ARC to be overflowing while
4008 * full of un-evictable buffers, and the
4009 * function should proceed in this case).
4011 * If threads are left sleeping, due to not
4012 * using cv_broadcast, they will be woken up
4013 * just before arc_reclaim_thread() sleeps.
4015 mutex_enter(&arc_reclaim_lock
);
4016 if (!arc_is_overflowing())
4017 cv_signal(&arc_reclaim_waiters_cv
);
4018 mutex_exit(&arc_reclaim_lock
);
4020 ARCSTAT_BUMP(arcstat_mutex_miss
);
4024 multilist_sublist_unlock(mls
);
4026 return (bytes_evicted
);
4030 * Evict buffers from the given arc state, until we've removed the
4031 * specified number of bytes. Move the removed buffers to the
4032 * appropriate evict state.
4034 * This function makes a "best effort". It skips over any buffers
4035 * it can't get a hash_lock on, and so, may not catch all candidates.
4036 * It may also return without evicting as much space as requested.
4038 * If bytes is specified using the special value ARC_EVICT_ALL, this
4039 * will evict all available (i.e. unlocked and evictable) buffers from
4040 * the given arc state; which is used by arc_flush().
4043 arc_evict_state(arc_state_t
*state
, uint64_t spa
, int64_t bytes
,
4044 arc_buf_contents_t type
)
4046 uint64_t total_evicted
= 0;
4047 multilist_t
*ml
= state
->arcs_list
[type
];
4049 arc_buf_hdr_t
**markers
;
4051 IMPLY(bytes
< 0, bytes
== ARC_EVICT_ALL
);
4053 num_sublists
= multilist_get_num_sublists(ml
);
4056 * If we've tried to evict from each sublist, made some
4057 * progress, but still have not hit the target number of bytes
4058 * to evict, we want to keep trying. The markers allow us to
4059 * pick up where we left off for each individual sublist, rather
4060 * than starting from the tail each time.
4062 markers
= kmem_zalloc(sizeof (*markers
) * num_sublists
, KM_SLEEP
);
4063 for (int i
= 0; i
< num_sublists
; i
++) {
4064 multilist_sublist_t
*mls
;
4066 markers
[i
] = kmem_cache_alloc(hdr_full_cache
, KM_SLEEP
);
4069 * A b_spa of 0 is used to indicate that this header is
4070 * a marker. This fact is used in arc_adjust_type() and
4071 * arc_evict_state_impl().
4073 markers
[i
]->b_spa
= 0;
4075 mls
= multilist_sublist_lock(ml
, i
);
4076 multilist_sublist_insert_tail(mls
, markers
[i
]);
4077 multilist_sublist_unlock(mls
);
4081 * While we haven't hit our target number of bytes to evict, or
4082 * we're evicting all available buffers.
4084 while (total_evicted
< bytes
|| bytes
== ARC_EVICT_ALL
) {
4085 int sublist_idx
= multilist_get_random_index(ml
);
4086 uint64_t scan_evicted
= 0;
4089 * Try to reduce pinned dnodes with a floor of arc_dnode_limit.
4090 * Request that 10% of the LRUs be scanned by the superblock
4093 if (type
== ARC_BUFC_DATA
&& arc_dnode_size
> arc_dnode_limit
)
4094 arc_prune_async((arc_dnode_size
- arc_dnode_limit
) /
4095 sizeof (dnode_t
) / zfs_arc_dnode_reduce_percent
);
4098 * Start eviction using a randomly selected sublist,
4099 * this is to try and evenly balance eviction across all
4100 * sublists. Always starting at the same sublist
4101 * (e.g. index 0) would cause evictions to favor certain
4102 * sublists over others.
4104 for (int i
= 0; i
< num_sublists
; i
++) {
4105 uint64_t bytes_remaining
;
4106 uint64_t bytes_evicted
;
4108 if (bytes
== ARC_EVICT_ALL
)
4109 bytes_remaining
= ARC_EVICT_ALL
;
4110 else if (total_evicted
< bytes
)
4111 bytes_remaining
= bytes
- total_evicted
;
4115 bytes_evicted
= arc_evict_state_impl(ml
, sublist_idx
,
4116 markers
[sublist_idx
], spa
, bytes_remaining
);
4118 scan_evicted
+= bytes_evicted
;
4119 total_evicted
+= bytes_evicted
;
4121 /* we've reached the end, wrap to the beginning */
4122 if (++sublist_idx
>= num_sublists
)
4127 * If we didn't evict anything during this scan, we have
4128 * no reason to believe we'll evict more during another
4129 * scan, so break the loop.
4131 if (scan_evicted
== 0) {
4132 /* This isn't possible, let's make that obvious */
4133 ASSERT3S(bytes
, !=, 0);
4136 * When bytes is ARC_EVICT_ALL, the only way to
4137 * break the loop is when scan_evicted is zero.
4138 * In that case, we actually have evicted enough,
4139 * so we don't want to increment the kstat.
4141 if (bytes
!= ARC_EVICT_ALL
) {
4142 ASSERT3S(total_evicted
, <, bytes
);
4143 ARCSTAT_BUMP(arcstat_evict_not_enough
);
4150 for (int i
= 0; i
< num_sublists
; i
++) {
4151 multilist_sublist_t
*mls
= multilist_sublist_lock(ml
, i
);
4152 multilist_sublist_remove(mls
, markers
[i
]);
4153 multilist_sublist_unlock(mls
);
4155 kmem_cache_free(hdr_full_cache
, markers
[i
]);
4157 kmem_free(markers
, sizeof (*markers
) * num_sublists
);
4159 return (total_evicted
);
4163 * Flush all "evictable" data of the given type from the arc state
4164 * specified. This will not evict any "active" buffers (i.e. referenced).
4166 * When 'retry' is set to B_FALSE, the function will make a single pass
4167 * over the state and evict any buffers that it can. Since it doesn't
4168 * continually retry the eviction, it might end up leaving some buffers
4169 * in the ARC due to lock misses.
4171 * When 'retry' is set to B_TRUE, the function will continually retry the
4172 * eviction until *all* evictable buffers have been removed from the
4173 * state. As a result, if concurrent insertions into the state are
4174 * allowed (e.g. if the ARC isn't shutting down), this function might
4175 * wind up in an infinite loop, continually trying to evict buffers.
4178 arc_flush_state(arc_state_t
*state
, uint64_t spa
, arc_buf_contents_t type
,
4181 uint64_t evicted
= 0;
4183 while (refcount_count(&state
->arcs_esize
[type
]) != 0) {
4184 evicted
+= arc_evict_state(state
, spa
, ARC_EVICT_ALL
, type
);
4194 * Helper function for arc_prune_async() it is responsible for safely
4195 * handling the execution of a registered arc_prune_func_t.
4198 arc_prune_task(void *ptr
)
4200 arc_prune_t
*ap
= (arc_prune_t
*)ptr
;
4201 arc_prune_func_t
*func
= ap
->p_pfunc
;
4204 func(ap
->p_adjust
, ap
->p_private
);
4206 refcount_remove(&ap
->p_refcnt
, func
);
4210 * Notify registered consumers they must drop holds on a portion of the ARC
4211 * buffered they reference. This provides a mechanism to ensure the ARC can
4212 * honor the arc_meta_limit and reclaim otherwise pinned ARC buffers. This
4213 * is analogous to dnlc_reduce_cache() but more generic.
4215 * This operation is performed asynchronously so it may be safely called
4216 * in the context of the arc_reclaim_thread(). A reference is taken here
4217 * for each registered arc_prune_t and the arc_prune_task() is responsible
4218 * for releasing it once the registered arc_prune_func_t has completed.
4221 arc_prune_async(int64_t adjust
)
4225 mutex_enter(&arc_prune_mtx
);
4226 for (ap
= list_head(&arc_prune_list
); ap
!= NULL
;
4227 ap
= list_next(&arc_prune_list
, ap
)) {
4229 if (refcount_count(&ap
->p_refcnt
) >= 2)
4232 refcount_add(&ap
->p_refcnt
, ap
->p_pfunc
);
4233 ap
->p_adjust
= adjust
;
4234 if (taskq_dispatch(arc_prune_taskq
, arc_prune_task
,
4235 ap
, TQ_SLEEP
) == TASKQID_INVALID
) {
4236 refcount_remove(&ap
->p_refcnt
, ap
->p_pfunc
);
4239 ARCSTAT_BUMP(arcstat_prune
);
4241 mutex_exit(&arc_prune_mtx
);
4245 * Evict the specified number of bytes from the state specified,
4246 * restricting eviction to the spa and type given. This function
4247 * prevents us from trying to evict more from a state's list than
4248 * is "evictable", and to skip evicting altogether when passed a
4249 * negative value for "bytes". In contrast, arc_evict_state() will
4250 * evict everything it can, when passed a negative value for "bytes".
4253 arc_adjust_impl(arc_state_t
*state
, uint64_t spa
, int64_t bytes
,
4254 arc_buf_contents_t type
)
4258 if (bytes
> 0 && refcount_count(&state
->arcs_esize
[type
]) > 0) {
4259 delta
= MIN(refcount_count(&state
->arcs_esize
[type
]), bytes
);
4260 return (arc_evict_state(state
, spa
, delta
, type
));
4267 * The goal of this function is to evict enough meta data buffers from the
4268 * ARC in order to enforce the arc_meta_limit. Achieving this is slightly
4269 * more complicated than it appears because it is common for data buffers
4270 * to have holds on meta data buffers. In addition, dnode meta data buffers
4271 * will be held by the dnodes in the block preventing them from being freed.
4272 * This means we can't simply traverse the ARC and expect to always find
4273 * enough unheld meta data buffer to release.
4275 * Therefore, this function has been updated to make alternating passes
4276 * over the ARC releasing data buffers and then newly unheld meta data
4277 * buffers. This ensures forward progress is maintained and arc_meta_used
4278 * will decrease. Normally this is sufficient, but if required the ARC
4279 * will call the registered prune callbacks causing dentry and inodes to
4280 * be dropped from the VFS cache. This will make dnode meta data buffers
4281 * available for reclaim.
4284 arc_adjust_meta_balanced(void)
4286 int64_t delta
, prune
= 0, adjustmnt
;
4287 uint64_t total_evicted
= 0;
4288 arc_buf_contents_t type
= ARC_BUFC_DATA
;
4289 int restarts
= MAX(zfs_arc_meta_adjust_restarts
, 0);
4293 * This slightly differs than the way we evict from the mru in
4294 * arc_adjust because we don't have a "target" value (i.e. no
4295 * "meta" arc_p). As a result, I think we can completely
4296 * cannibalize the metadata in the MRU before we evict the
4297 * metadata from the MFU. I think we probably need to implement a
4298 * "metadata arc_p" value to do this properly.
4300 adjustmnt
= arc_meta_used
- arc_meta_limit
;
4302 if (adjustmnt
> 0 && refcount_count(&arc_mru
->arcs_esize
[type
]) > 0) {
4303 delta
= MIN(refcount_count(&arc_mru
->arcs_esize
[type
]),
4305 total_evicted
+= arc_adjust_impl(arc_mru
, 0, delta
, type
);
4310 * We can't afford to recalculate adjustmnt here. If we do,
4311 * new metadata buffers can sneak into the MRU or ANON lists,
4312 * thus penalize the MFU metadata. Although the fudge factor is
4313 * small, it has been empirically shown to be significant for
4314 * certain workloads (e.g. creating many empty directories). As
4315 * such, we use the original calculation for adjustmnt, and
4316 * simply decrement the amount of data evicted from the MRU.
4319 if (adjustmnt
> 0 && refcount_count(&arc_mfu
->arcs_esize
[type
]) > 0) {
4320 delta
= MIN(refcount_count(&arc_mfu
->arcs_esize
[type
]),
4322 total_evicted
+= arc_adjust_impl(arc_mfu
, 0, delta
, type
);
4325 adjustmnt
= arc_meta_used
- arc_meta_limit
;
4327 if (adjustmnt
> 0 &&
4328 refcount_count(&arc_mru_ghost
->arcs_esize
[type
]) > 0) {
4329 delta
= MIN(adjustmnt
,
4330 refcount_count(&arc_mru_ghost
->arcs_esize
[type
]));
4331 total_evicted
+= arc_adjust_impl(arc_mru_ghost
, 0, delta
, type
);
4335 if (adjustmnt
> 0 &&
4336 refcount_count(&arc_mfu_ghost
->arcs_esize
[type
]) > 0) {
4337 delta
= MIN(adjustmnt
,
4338 refcount_count(&arc_mfu_ghost
->arcs_esize
[type
]));
4339 total_evicted
+= arc_adjust_impl(arc_mfu_ghost
, 0, delta
, type
);
4343 * If after attempting to make the requested adjustment to the ARC
4344 * the meta limit is still being exceeded then request that the
4345 * higher layers drop some cached objects which have holds on ARC
4346 * meta buffers. Requests to the upper layers will be made with
4347 * increasingly large scan sizes until the ARC is below the limit.
4349 if (arc_meta_used
> arc_meta_limit
) {
4350 if (type
== ARC_BUFC_DATA
) {
4351 type
= ARC_BUFC_METADATA
;
4353 type
= ARC_BUFC_DATA
;
4355 if (zfs_arc_meta_prune
) {
4356 prune
+= zfs_arc_meta_prune
;
4357 arc_prune_async(prune
);
4366 return (total_evicted
);
4370 * Evict metadata buffers from the cache, such that arc_meta_used is
4371 * capped by the arc_meta_limit tunable.
4374 arc_adjust_meta_only(void)
4376 uint64_t total_evicted
= 0;
4380 * If we're over the meta limit, we want to evict enough
4381 * metadata to get back under the meta limit. We don't want to
4382 * evict so much that we drop the MRU below arc_p, though. If
4383 * we're over the meta limit more than we're over arc_p, we
4384 * evict some from the MRU here, and some from the MFU below.
4386 target
= MIN((int64_t)(arc_meta_used
- arc_meta_limit
),
4387 (int64_t)(refcount_count(&arc_anon
->arcs_size
) +
4388 refcount_count(&arc_mru
->arcs_size
) - arc_p
));
4390 total_evicted
+= arc_adjust_impl(arc_mru
, 0, target
, ARC_BUFC_METADATA
);
4393 * Similar to the above, we want to evict enough bytes to get us
4394 * below the meta limit, but not so much as to drop us below the
4395 * space allotted to the MFU (which is defined as arc_c - arc_p).
4397 target
= MIN((int64_t)(arc_meta_used
- arc_meta_limit
),
4398 (int64_t)(refcount_count(&arc_mfu
->arcs_size
) - (arc_c
- arc_p
)));
4400 total_evicted
+= arc_adjust_impl(arc_mfu
, 0, target
, ARC_BUFC_METADATA
);
4402 return (total_evicted
);
4406 arc_adjust_meta(void)
4408 if (zfs_arc_meta_strategy
== ARC_STRATEGY_META_ONLY
)
4409 return (arc_adjust_meta_only());
4411 return (arc_adjust_meta_balanced());
4415 * Return the type of the oldest buffer in the given arc state
4417 * This function will select a random sublist of type ARC_BUFC_DATA and
4418 * a random sublist of type ARC_BUFC_METADATA. The tail of each sublist
4419 * is compared, and the type which contains the "older" buffer will be
4422 static arc_buf_contents_t
4423 arc_adjust_type(arc_state_t
*state
)
4425 multilist_t
*data_ml
= state
->arcs_list
[ARC_BUFC_DATA
];
4426 multilist_t
*meta_ml
= state
->arcs_list
[ARC_BUFC_METADATA
];
4427 int data_idx
= multilist_get_random_index(data_ml
);
4428 int meta_idx
= multilist_get_random_index(meta_ml
);
4429 multilist_sublist_t
*data_mls
;
4430 multilist_sublist_t
*meta_mls
;
4431 arc_buf_contents_t type
;
4432 arc_buf_hdr_t
*data_hdr
;
4433 arc_buf_hdr_t
*meta_hdr
;
4436 * We keep the sublist lock until we're finished, to prevent
4437 * the headers from being destroyed via arc_evict_state().
4439 data_mls
= multilist_sublist_lock(data_ml
, data_idx
);
4440 meta_mls
= multilist_sublist_lock(meta_ml
, meta_idx
);
4443 * These two loops are to ensure we skip any markers that
4444 * might be at the tail of the lists due to arc_evict_state().
4447 for (data_hdr
= multilist_sublist_tail(data_mls
); data_hdr
!= NULL
;
4448 data_hdr
= multilist_sublist_prev(data_mls
, data_hdr
)) {
4449 if (data_hdr
->b_spa
!= 0)
4453 for (meta_hdr
= multilist_sublist_tail(meta_mls
); meta_hdr
!= NULL
;
4454 meta_hdr
= multilist_sublist_prev(meta_mls
, meta_hdr
)) {
4455 if (meta_hdr
->b_spa
!= 0)
4459 if (data_hdr
== NULL
&& meta_hdr
== NULL
) {
4460 type
= ARC_BUFC_DATA
;
4461 } else if (data_hdr
== NULL
) {
4462 ASSERT3P(meta_hdr
, !=, NULL
);
4463 type
= ARC_BUFC_METADATA
;
4464 } else if (meta_hdr
== NULL
) {
4465 ASSERT3P(data_hdr
, !=, NULL
);
4466 type
= ARC_BUFC_DATA
;
4468 ASSERT3P(data_hdr
, !=, NULL
);
4469 ASSERT3P(meta_hdr
, !=, NULL
);
4471 /* The headers can't be on the sublist without an L1 header */
4472 ASSERT(HDR_HAS_L1HDR(data_hdr
));
4473 ASSERT(HDR_HAS_L1HDR(meta_hdr
));
4475 if (data_hdr
->b_l1hdr
.b_arc_access
<
4476 meta_hdr
->b_l1hdr
.b_arc_access
) {
4477 type
= ARC_BUFC_DATA
;
4479 type
= ARC_BUFC_METADATA
;
4483 multilist_sublist_unlock(meta_mls
);
4484 multilist_sublist_unlock(data_mls
);
4490 * Evict buffers from the cache, such that arc_size is capped by arc_c.
4495 uint64_t total_evicted
= 0;
4500 * If we're over arc_meta_limit, we want to correct that before
4501 * potentially evicting data buffers below.
4503 total_evicted
+= arc_adjust_meta();
4508 * If we're over the target cache size, we want to evict enough
4509 * from the list to get back to our target size. We don't want
4510 * to evict too much from the MRU, such that it drops below
4511 * arc_p. So, if we're over our target cache size more than
4512 * the MRU is over arc_p, we'll evict enough to get back to
4513 * arc_p here, and then evict more from the MFU below.
4515 target
= MIN((int64_t)(arc_size
- arc_c
),
4516 (int64_t)(refcount_count(&arc_anon
->arcs_size
) +
4517 refcount_count(&arc_mru
->arcs_size
) + arc_meta_used
- arc_p
));
4520 * If we're below arc_meta_min, always prefer to evict data.
4521 * Otherwise, try to satisfy the requested number of bytes to
4522 * evict from the type which contains older buffers; in an
4523 * effort to keep newer buffers in the cache regardless of their
4524 * type. If we cannot satisfy the number of bytes from this
4525 * type, spill over into the next type.
4527 if (arc_adjust_type(arc_mru
) == ARC_BUFC_METADATA
&&
4528 arc_meta_used
> arc_meta_min
) {
4529 bytes
= arc_adjust_impl(arc_mru
, 0, target
, ARC_BUFC_METADATA
);
4530 total_evicted
+= bytes
;
4533 * If we couldn't evict our target number of bytes from
4534 * metadata, we try to get the rest from data.
4539 arc_adjust_impl(arc_mru
, 0, target
, ARC_BUFC_DATA
);
4541 bytes
= arc_adjust_impl(arc_mru
, 0, target
, ARC_BUFC_DATA
);
4542 total_evicted
+= bytes
;
4545 * If we couldn't evict our target number of bytes from
4546 * data, we try to get the rest from metadata.
4551 arc_adjust_impl(arc_mru
, 0, target
, ARC_BUFC_METADATA
);
4557 * Now that we've tried to evict enough from the MRU to get its
4558 * size back to arc_p, if we're still above the target cache
4559 * size, we evict the rest from the MFU.
4561 target
= arc_size
- arc_c
;
4563 if (arc_adjust_type(arc_mfu
) == ARC_BUFC_METADATA
&&
4564 arc_meta_used
> arc_meta_min
) {
4565 bytes
= arc_adjust_impl(arc_mfu
, 0, target
, ARC_BUFC_METADATA
);
4566 total_evicted
+= bytes
;
4569 * If we couldn't evict our target number of bytes from
4570 * metadata, we try to get the rest from data.
4575 arc_adjust_impl(arc_mfu
, 0, target
, ARC_BUFC_DATA
);
4577 bytes
= arc_adjust_impl(arc_mfu
, 0, target
, ARC_BUFC_DATA
);
4578 total_evicted
+= bytes
;
4581 * If we couldn't evict our target number of bytes from
4582 * data, we try to get the rest from data.
4587 arc_adjust_impl(arc_mfu
, 0, target
, ARC_BUFC_METADATA
);
4591 * Adjust ghost lists
4593 * In addition to the above, the ARC also defines target values
4594 * for the ghost lists. The sum of the mru list and mru ghost
4595 * list should never exceed the target size of the cache, and
4596 * the sum of the mru list, mfu list, mru ghost list, and mfu
4597 * ghost list should never exceed twice the target size of the
4598 * cache. The following logic enforces these limits on the ghost
4599 * caches, and evicts from them as needed.
4601 target
= refcount_count(&arc_mru
->arcs_size
) +
4602 refcount_count(&arc_mru_ghost
->arcs_size
) - arc_c
;
4604 bytes
= arc_adjust_impl(arc_mru_ghost
, 0, target
, ARC_BUFC_DATA
);
4605 total_evicted
+= bytes
;
4610 arc_adjust_impl(arc_mru_ghost
, 0, target
, ARC_BUFC_METADATA
);
4613 * We assume the sum of the mru list and mfu list is less than
4614 * or equal to arc_c (we enforced this above), which means we
4615 * can use the simpler of the two equations below:
4617 * mru + mfu + mru ghost + mfu ghost <= 2 * arc_c
4618 * mru ghost + mfu ghost <= arc_c
4620 target
= refcount_count(&arc_mru_ghost
->arcs_size
) +
4621 refcount_count(&arc_mfu_ghost
->arcs_size
) - arc_c
;
4623 bytes
= arc_adjust_impl(arc_mfu_ghost
, 0, target
, ARC_BUFC_DATA
);
4624 total_evicted
+= bytes
;
4629 arc_adjust_impl(arc_mfu_ghost
, 0, target
, ARC_BUFC_METADATA
);
4631 return (total_evicted
);
4635 arc_flush(spa_t
*spa
, boolean_t retry
)
4640 * If retry is B_TRUE, a spa must not be specified since we have
4641 * no good way to determine if all of a spa's buffers have been
4642 * evicted from an arc state.
4644 ASSERT(!retry
|| spa
== 0);
4647 guid
= spa_load_guid(spa
);
4649 (void) arc_flush_state(arc_mru
, guid
, ARC_BUFC_DATA
, retry
);
4650 (void) arc_flush_state(arc_mru
, guid
, ARC_BUFC_METADATA
, retry
);
4652 (void) arc_flush_state(arc_mfu
, guid
, ARC_BUFC_DATA
, retry
);
4653 (void) arc_flush_state(arc_mfu
, guid
, ARC_BUFC_METADATA
, retry
);
4655 (void) arc_flush_state(arc_mru_ghost
, guid
, ARC_BUFC_DATA
, retry
);
4656 (void) arc_flush_state(arc_mru_ghost
, guid
, ARC_BUFC_METADATA
, retry
);
4658 (void) arc_flush_state(arc_mfu_ghost
, guid
, ARC_BUFC_DATA
, retry
);
4659 (void) arc_flush_state(arc_mfu_ghost
, guid
, ARC_BUFC_METADATA
, retry
);
4663 arc_shrink(int64_t to_free
)
4667 if (c
> to_free
&& c
- to_free
> arc_c_min
) {
4668 arc_c
= c
- to_free
;
4669 atomic_add_64(&arc_p
, -(arc_p
>> arc_shrink_shift
));
4670 if (arc_c
> arc_size
)
4671 arc_c
= MAX(arc_size
, arc_c_min
);
4673 arc_p
= (arc_c
>> 1);
4674 ASSERT(arc_c
>= arc_c_min
);
4675 ASSERT((int64_t)arc_p
>= 0);
4680 if (arc_size
> arc_c
)
4681 (void) arc_adjust();
4685 * Return maximum amount of memory that we could possibly use. Reduced
4686 * to half of all memory in user space which is primarily used for testing.
4689 arc_all_memory(void)
4692 #ifdef CONFIG_HIGHMEM
4693 return (ptob(totalram_pages
- totalhigh_pages
));
4695 return (ptob(totalram_pages
));
4696 #endif /* CONFIG_HIGHMEM */
4698 return (ptob(physmem
) / 2);
4699 #endif /* _KERNEL */
4703 * Return the amount of memory that is considered free. In user space
4704 * which is primarily used for testing we pretend that free memory ranges
4705 * from 0-20% of all memory.
4708 arc_free_memory(void)
4711 #ifdef CONFIG_HIGHMEM
4714 return (ptob(si
.freeram
- si
.freehigh
));
4716 return (ptob(nr_free_pages() +
4717 nr_inactive_file_pages() +
4718 nr_inactive_anon_pages() +
4719 nr_slab_reclaimable_pages()));
4721 #endif /* CONFIG_HIGHMEM */
4723 return (spa_get_random(arc_all_memory() * 20 / 100));
4724 #endif /* _KERNEL */
4727 typedef enum free_memory_reason_t
{
4732 FMR_PAGES_PP_MAXIMUM
,
4735 } free_memory_reason_t
;
4737 int64_t last_free_memory
;
4738 free_memory_reason_t last_free_reason
;
4742 * Additional reserve of pages for pp_reserve.
4744 int64_t arc_pages_pp_reserve
= 64;
4747 * Additional reserve of pages for swapfs.
4749 int64_t arc_swapfs_reserve
= 64;
4750 #endif /* _KERNEL */
4753 * Return the amount of memory that can be consumed before reclaim will be
4754 * needed. Positive if there is sufficient free memory, negative indicates
4755 * the amount of memory that needs to be freed up.
4758 arc_available_memory(void)
4760 int64_t lowest
= INT64_MAX
;
4761 free_memory_reason_t r
= FMR_UNKNOWN
;
4768 pgcnt_t needfree
= btop(arc_need_free
);
4769 pgcnt_t lotsfree
= btop(arc_sys_free
);
4770 pgcnt_t desfree
= 0;
4771 pgcnt_t freemem
= btop(arc_free_memory());
4775 n
= PAGESIZE
* (-needfree
);
4783 * check that we're out of range of the pageout scanner. It starts to
4784 * schedule paging if freemem is less than lotsfree and needfree.
4785 * lotsfree is the high-water mark for pageout, and needfree is the
4786 * number of needed free pages. We add extra pages here to make sure
4787 * the scanner doesn't start up while we're freeing memory.
4789 n
= PAGESIZE
* (freemem
- lotsfree
- needfree
- desfree
);
4797 * check to make sure that swapfs has enough space so that anon
4798 * reservations can still succeed. anon_resvmem() checks that the
4799 * availrmem is greater than swapfs_minfree, and the number of reserved
4800 * swap pages. We also add a bit of extra here just to prevent
4801 * circumstances from getting really dire.
4803 n
= PAGESIZE
* (availrmem
- swapfs_minfree
- swapfs_reserve
-
4804 desfree
- arc_swapfs_reserve
);
4807 r
= FMR_SWAPFS_MINFREE
;
4811 * Check that we have enough availrmem that memory locking (e.g., via
4812 * mlock(3C) or memcntl(2)) can still succeed. (pages_pp_maximum
4813 * stores the number of pages that cannot be locked; when availrmem
4814 * drops below pages_pp_maximum, page locking mechanisms such as
4815 * page_pp_lock() will fail.)
4817 n
= PAGESIZE
* (availrmem
- pages_pp_maximum
-
4818 arc_pages_pp_reserve
);
4821 r
= FMR_PAGES_PP_MAXIMUM
;
4827 * If we're on a 32-bit platform, it's possible that we'll exhaust the
4828 * kernel heap space before we ever run out of available physical
4829 * memory. Most checks of the size of the heap_area compare against
4830 * tune.t_minarmem, which is the minimum available real memory that we
4831 * can have in the system. However, this is generally fixed at 25 pages
4832 * which is so low that it's useless. In this comparison, we seek to
4833 * calculate the total heap-size, and reclaim if more than 3/4ths of the
4834 * heap is allocated. (Or, in the calculation, if less than 1/4th is
4837 n
= vmem_size(heap_arena
, VMEM_FREE
) -
4838 (vmem_size(heap_arena
, VMEM_FREE
| VMEM_ALLOC
) >> 2);
4846 * If zio data pages are being allocated out of a separate heap segment,
4847 * then enforce that the size of available vmem for this arena remains
4848 * above about 1/4th (1/(2^arc_zio_arena_free_shift)) free.
4850 * Note that reducing the arc_zio_arena_free_shift keeps more virtual
4851 * memory (in the zio_arena) free, which can avoid memory
4852 * fragmentation issues.
4854 if (zio_arena
!= NULL
) {
4855 n
= (int64_t)vmem_size(zio_arena
, VMEM_FREE
) -
4856 (vmem_size(zio_arena
, VMEM_ALLOC
) >>
4857 arc_zio_arena_free_shift
);
4864 /* Every 100 calls, free a small amount */
4865 if (spa_get_random(100) == 0)
4867 #endif /* _KERNEL */
4869 last_free_memory
= lowest
;
4870 last_free_reason
= r
;
4876 * Determine if the system is under memory pressure and is asking
4877 * to reclaim memory. A return value of B_TRUE indicates that the system
4878 * is under memory pressure and that the arc should adjust accordingly.
4881 arc_reclaim_needed(void)
4883 return (arc_available_memory() < 0);
4887 arc_kmem_reap_now(void)
4890 kmem_cache_t
*prev_cache
= NULL
;
4891 kmem_cache_t
*prev_data_cache
= NULL
;
4892 extern kmem_cache_t
*zio_buf_cache
[];
4893 extern kmem_cache_t
*zio_data_buf_cache
[];
4894 extern kmem_cache_t
*range_seg_cache
;
4897 if ((arc_meta_used
>= arc_meta_limit
) && zfs_arc_meta_prune
) {
4899 * We are exceeding our meta-data cache limit.
4900 * Prune some entries to release holds on meta-data.
4902 arc_prune_async(zfs_arc_meta_prune
);
4906 * Reclaim unused memory from all kmem caches.
4912 for (i
= 0; i
< SPA_MAXBLOCKSIZE
>> SPA_MINBLOCKSHIFT
; i
++) {
4914 /* reach upper limit of cache size on 32-bit */
4915 if (zio_buf_cache
[i
] == NULL
)
4918 if (zio_buf_cache
[i
] != prev_cache
) {
4919 prev_cache
= zio_buf_cache
[i
];
4920 kmem_cache_reap_now(zio_buf_cache
[i
]);
4922 if (zio_data_buf_cache
[i
] != prev_data_cache
) {
4923 prev_data_cache
= zio_data_buf_cache
[i
];
4924 kmem_cache_reap_now(zio_data_buf_cache
[i
]);
4927 kmem_cache_reap_now(buf_cache
);
4928 kmem_cache_reap_now(hdr_full_cache
);
4929 kmem_cache_reap_now(hdr_l2only_cache
);
4930 kmem_cache_reap_now(range_seg_cache
);
4932 if (zio_arena
!= NULL
) {
4934 * Ask the vmem arena to reclaim unused memory from its
4937 vmem_qcache_reap(zio_arena
);
4942 * Threads can block in arc_get_data_impl() waiting for this thread to evict
4943 * enough data and signal them to proceed. When this happens, the threads in
4944 * arc_get_data_impl() are sleeping while holding the hash lock for their
4945 * particular arc header. Thus, we must be careful to never sleep on a
4946 * hash lock in this thread. This is to prevent the following deadlock:
4948 * - Thread A sleeps on CV in arc_get_data_impl() holding hash lock "L",
4949 * waiting for the reclaim thread to signal it.
4951 * - arc_reclaim_thread() tries to acquire hash lock "L" using mutex_enter,
4952 * fails, and goes to sleep forever.
4954 * This possible deadlock is avoided by always acquiring a hash lock
4955 * using mutex_tryenter() from arc_reclaim_thread().
4959 arc_reclaim_thread(void *unused
)
4961 fstrans_cookie_t cookie
= spl_fstrans_mark();
4962 hrtime_t growtime
= 0;
4965 CALLB_CPR_INIT(&cpr
, &arc_reclaim_lock
, callb_generic_cpr
, FTAG
);
4967 mutex_enter(&arc_reclaim_lock
);
4968 while (!arc_reclaim_thread_exit
) {
4969 uint64_t evicted
= 0;
4970 uint64_t need_free
= arc_need_free
;
4971 arc_tuning_update();
4974 * This is necessary in order for the mdb ::arc dcmd to
4975 * show up to date information. Since the ::arc command
4976 * does not call the kstat's update function, without
4977 * this call, the command may show stale stats for the
4978 * anon, mru, mru_ghost, mfu, and mfu_ghost lists. Even
4979 * with this change, the data might be up to 1 second
4980 * out of date; but that should suffice. The arc_state_t
4981 * structures can be queried directly if more accurate
4982 * information is needed.
4985 if (arc_ksp
!= NULL
)
4986 arc_ksp
->ks_update(arc_ksp
, KSTAT_READ
);
4988 mutex_exit(&arc_reclaim_lock
);
4991 * We call arc_adjust() before (possibly) calling
4992 * arc_kmem_reap_now(), so that we can wake up
4993 * arc_get_data_buf() sooner.
4995 evicted
= arc_adjust();
4997 int64_t free_memory
= arc_available_memory();
4998 if (free_memory
< 0) {
5000 arc_no_grow
= B_TRUE
;
5004 * Wait at least zfs_grow_retry (default 5) seconds
5005 * before considering growing.
5007 growtime
= gethrtime() + SEC2NSEC(arc_grow_retry
);
5009 arc_kmem_reap_now();
5012 * If we are still low on memory, shrink the ARC
5013 * so that we have arc_shrink_min free space.
5015 free_memory
= arc_available_memory();
5018 (arc_c
>> arc_shrink_shift
) - free_memory
;
5021 to_free
= MAX(to_free
, need_free
);
5023 arc_shrink(to_free
);
5025 } else if (free_memory
< arc_c
>> arc_no_grow_shift
) {
5026 arc_no_grow
= B_TRUE
;
5027 } else if (gethrtime() >= growtime
) {
5028 arc_no_grow
= B_FALSE
;
5031 mutex_enter(&arc_reclaim_lock
);
5034 * If evicted is zero, we couldn't evict anything via
5035 * arc_adjust(). This could be due to hash lock
5036 * collisions, but more likely due to the majority of
5037 * arc buffers being unevictable. Therefore, even if
5038 * arc_size is above arc_c, another pass is unlikely to
5039 * be helpful and could potentially cause us to enter an
5042 if (arc_size
<= arc_c
|| evicted
== 0) {
5044 * We're either no longer overflowing, or we
5045 * can't evict anything more, so we should wake
5046 * up any threads before we go to sleep and remove
5047 * the bytes we were working on from arc_need_free
5048 * since nothing more will be done here.
5050 cv_broadcast(&arc_reclaim_waiters_cv
);
5051 ARCSTAT_INCR(arcstat_need_free
, -need_free
);
5054 * Block until signaled, or after one second (we
5055 * might need to perform arc_kmem_reap_now()
5056 * even if we aren't being signalled)
5058 CALLB_CPR_SAFE_BEGIN(&cpr
);
5059 (void) cv_timedwait_sig_hires(&arc_reclaim_thread_cv
,
5060 &arc_reclaim_lock
, SEC2NSEC(1), MSEC2NSEC(1), 0);
5061 CALLB_CPR_SAFE_END(&cpr
, &arc_reclaim_lock
);
5065 arc_reclaim_thread_exit
= B_FALSE
;
5066 cv_broadcast(&arc_reclaim_thread_cv
);
5067 CALLB_CPR_EXIT(&cpr
); /* drops arc_reclaim_lock */
5068 spl_fstrans_unmark(cookie
);
5074 * Determine the amount of memory eligible for eviction contained in the
5075 * ARC. All clean data reported by the ghost lists can always be safely
5076 * evicted. Due to arc_c_min, the same does not hold for all clean data
5077 * contained by the regular mru and mfu lists.
5079 * In the case of the regular mru and mfu lists, we need to report as
5080 * much clean data as possible, such that evicting that same reported
5081 * data will not bring arc_size below arc_c_min. Thus, in certain
5082 * circumstances, the total amount of clean data in the mru and mfu
5083 * lists might not actually be evictable.
5085 * The following two distinct cases are accounted for:
5087 * 1. The sum of the amount of dirty data contained by both the mru and
5088 * mfu lists, plus the ARC's other accounting (e.g. the anon list),
5089 * is greater than or equal to arc_c_min.
5090 * (i.e. amount of dirty data >= arc_c_min)
5092 * This is the easy case; all clean data contained by the mru and mfu
5093 * lists is evictable. Evicting all clean data can only drop arc_size
5094 * to the amount of dirty data, which is greater than arc_c_min.
5096 * 2. The sum of the amount of dirty data contained by both the mru and
5097 * mfu lists, plus the ARC's other accounting (e.g. the anon list),
5098 * is less than arc_c_min.
5099 * (i.e. arc_c_min > amount of dirty data)
5101 * 2.1. arc_size is greater than or equal arc_c_min.
5102 * (i.e. arc_size >= arc_c_min > amount of dirty data)
5104 * In this case, not all clean data from the regular mru and mfu
5105 * lists is actually evictable; we must leave enough clean data
5106 * to keep arc_size above arc_c_min. Thus, the maximum amount of
5107 * evictable data from the two lists combined, is exactly the
5108 * difference between arc_size and arc_c_min.
5110 * 2.2. arc_size is less than arc_c_min
5111 * (i.e. arc_c_min > arc_size > amount of dirty data)
5113 * In this case, none of the data contained in the mru and mfu
5114 * lists is evictable, even if it's clean. Since arc_size is
5115 * already below arc_c_min, evicting any more would only
5116 * increase this negative difference.
5119 arc_evictable_memory(void)
5121 uint64_t arc_clean
=
5122 refcount_count(&arc_mru
->arcs_esize
[ARC_BUFC_DATA
]) +
5123 refcount_count(&arc_mru
->arcs_esize
[ARC_BUFC_METADATA
]) +
5124 refcount_count(&arc_mfu
->arcs_esize
[ARC_BUFC_DATA
]) +
5125 refcount_count(&arc_mfu
->arcs_esize
[ARC_BUFC_METADATA
]);
5126 uint64_t arc_dirty
= MAX((int64_t)arc_size
- (int64_t)arc_clean
, 0);
5129 * Scale reported evictable memory in proportion to page cache, cap
5130 * at specified min/max.
5132 uint64_t min
= (ptob(nr_file_pages()) / 100) * zfs_arc_pc_percent
;
5133 min
= MAX(arc_c_min
, MIN(arc_c_max
, min
));
5135 if (arc_dirty
>= min
)
5138 return (MAX((int64_t)arc_size
- (int64_t)min
, 0));
5142 * If sc->nr_to_scan is zero, the caller is requesting a query of the
5143 * number of objects which can potentially be freed. If it is nonzero,
5144 * the request is to free that many objects.
5146 * Linux kernels >= 3.12 have the count_objects and scan_objects callbacks
5147 * in struct shrinker and also require the shrinker to return the number
5150 * Older kernels require the shrinker to return the number of freeable
5151 * objects following the freeing of nr_to_free.
5153 static spl_shrinker_t
5154 __arc_shrinker_func(struct shrinker
*shrink
, struct shrink_control
*sc
)
5158 /* The arc is considered warm once reclaim has occurred */
5159 if (unlikely(arc_warm
== B_FALSE
))
5162 /* Return the potential number of reclaimable pages */
5163 pages
= btop((int64_t)arc_evictable_memory());
5164 if (sc
->nr_to_scan
== 0)
5167 /* Not allowed to perform filesystem reclaim */
5168 if (!(sc
->gfp_mask
& __GFP_FS
))
5169 return (SHRINK_STOP
);
5171 /* Reclaim in progress */
5172 if (mutex_tryenter(&arc_reclaim_lock
) == 0) {
5173 ARCSTAT_INCR(arcstat_need_free
, ptob(sc
->nr_to_scan
));
5177 mutex_exit(&arc_reclaim_lock
);
5180 * Evict the requested number of pages by shrinking arc_c the
5184 arc_shrink(ptob(sc
->nr_to_scan
));
5185 if (current_is_kswapd())
5186 arc_kmem_reap_now();
5187 #ifdef HAVE_SPLIT_SHRINKER_CALLBACK
5188 pages
= MAX((int64_t)pages
-
5189 (int64_t)btop(arc_evictable_memory()), 0);
5191 pages
= btop(arc_evictable_memory());
5194 * We've shrunk what we can, wake up threads.
5196 cv_broadcast(&arc_reclaim_waiters_cv
);
5198 pages
= SHRINK_STOP
;
5201 * When direct reclaim is observed it usually indicates a rapid
5202 * increase in memory pressure. This occurs because the kswapd
5203 * threads were unable to asynchronously keep enough free memory
5204 * available. In this case set arc_no_grow to briefly pause arc
5205 * growth to avoid compounding the memory pressure.
5207 if (current_is_kswapd()) {
5208 ARCSTAT_BUMP(arcstat_memory_indirect_count
);
5210 arc_no_grow
= B_TRUE
;
5211 arc_kmem_reap_now();
5212 ARCSTAT_BUMP(arcstat_memory_direct_count
);
5217 SPL_SHRINKER_CALLBACK_WRAPPER(arc_shrinker_func
);
5219 SPL_SHRINKER_DECLARE(arc_shrinker
, arc_shrinker_func
, DEFAULT_SEEKS
);
5220 #endif /* _KERNEL */
5223 * Adapt arc info given the number of bytes we are trying to add and
5224 * the state that we are coming from. This function is only called
5225 * when we are adding new content to the cache.
5228 arc_adapt(int bytes
, arc_state_t
*state
)
5231 uint64_t arc_p_min
= (arc_c
>> arc_p_min_shift
);
5232 int64_t mrug_size
= refcount_count(&arc_mru_ghost
->arcs_size
);
5233 int64_t mfug_size
= refcount_count(&arc_mfu_ghost
->arcs_size
);
5235 if (state
== arc_l2c_only
)
5240 * Adapt the target size of the MRU list:
5241 * - if we just hit in the MRU ghost list, then increase
5242 * the target size of the MRU list.
5243 * - if we just hit in the MFU ghost list, then increase
5244 * the target size of the MFU list by decreasing the
5245 * target size of the MRU list.
5247 if (state
== arc_mru_ghost
) {
5248 mult
= (mrug_size
>= mfug_size
) ? 1 : (mfug_size
/ mrug_size
);
5249 if (!zfs_arc_p_dampener_disable
)
5250 mult
= MIN(mult
, 10); /* avoid wild arc_p adjustment */
5252 arc_p
= MIN(arc_c
- arc_p_min
, arc_p
+ bytes
* mult
);
5253 } else if (state
== arc_mfu_ghost
) {
5256 mult
= (mfug_size
>= mrug_size
) ? 1 : (mrug_size
/ mfug_size
);
5257 if (!zfs_arc_p_dampener_disable
)
5258 mult
= MIN(mult
, 10);
5260 delta
= MIN(bytes
* mult
, arc_p
);
5261 arc_p
= MAX(arc_p_min
, arc_p
- delta
);
5263 ASSERT((int64_t)arc_p
>= 0);
5265 if (arc_reclaim_needed()) {
5266 cv_signal(&arc_reclaim_thread_cv
);
5273 if (arc_c
>= arc_c_max
)
5277 * If we're within (2 * maxblocksize) bytes of the target
5278 * cache size, increment the target cache size
5280 ASSERT3U(arc_c
, >=, 2ULL << SPA_MAXBLOCKSHIFT
);
5281 if (arc_size
>= arc_c
- (2ULL << SPA_MAXBLOCKSHIFT
)) {
5282 atomic_add_64(&arc_c
, (int64_t)bytes
);
5283 if (arc_c
> arc_c_max
)
5285 else if (state
== arc_anon
)
5286 atomic_add_64(&arc_p
, (int64_t)bytes
);
5290 ASSERT((int64_t)arc_p
>= 0);
5294 * Check if arc_size has grown past our upper threshold, determined by
5295 * zfs_arc_overflow_shift.
5298 arc_is_overflowing(void)
5300 /* Always allow at least one block of overflow */
5301 uint64_t overflow
= MAX(SPA_MAXBLOCKSIZE
,
5302 arc_c
>> zfs_arc_overflow_shift
);
5304 return (arc_size
>= arc_c
+ overflow
);
5308 arc_get_data_abd(arc_buf_hdr_t
*hdr
, uint64_t size
, void *tag
)
5310 arc_buf_contents_t type
= arc_buf_type(hdr
);
5312 arc_get_data_impl(hdr
, size
, tag
);
5313 if (type
== ARC_BUFC_METADATA
) {
5314 return (abd_alloc(size
, B_TRUE
));
5316 ASSERT(type
== ARC_BUFC_DATA
);
5317 return (abd_alloc(size
, B_FALSE
));
5322 arc_get_data_buf(arc_buf_hdr_t
*hdr
, uint64_t size
, void *tag
)
5324 arc_buf_contents_t type
= arc_buf_type(hdr
);
5326 arc_get_data_impl(hdr
, size
, tag
);
5327 if (type
== ARC_BUFC_METADATA
) {
5328 return (zio_buf_alloc(size
));
5330 ASSERT(type
== ARC_BUFC_DATA
);
5331 return (zio_data_buf_alloc(size
));
5336 * Allocate a block and return it to the caller. If we are hitting the
5337 * hard limit for the cache size, we must sleep, waiting for the eviction
5338 * thread to catch up. If we're past the target size but below the hard
5339 * limit, we'll only signal the reclaim thread and continue on.
5342 arc_get_data_impl(arc_buf_hdr_t
*hdr
, uint64_t size
, void *tag
)
5344 arc_state_t
*state
= hdr
->b_l1hdr
.b_state
;
5345 arc_buf_contents_t type
= arc_buf_type(hdr
);
5347 arc_adapt(size
, state
);
5350 * If arc_size is currently overflowing, and has grown past our
5351 * upper limit, we must be adding data faster than the evict
5352 * thread can evict. Thus, to ensure we don't compound the
5353 * problem by adding more data and forcing arc_size to grow even
5354 * further past it's target size, we halt and wait for the
5355 * eviction thread to catch up.
5357 * It's also possible that the reclaim thread is unable to evict
5358 * enough buffers to get arc_size below the overflow limit (e.g.
5359 * due to buffers being un-evictable, or hash lock collisions).
5360 * In this case, we want to proceed regardless if we're
5361 * overflowing; thus we don't use a while loop here.
5363 if (arc_is_overflowing()) {
5364 mutex_enter(&arc_reclaim_lock
);
5367 * Now that we've acquired the lock, we may no longer be
5368 * over the overflow limit, lets check.
5370 * We're ignoring the case of spurious wake ups. If that
5371 * were to happen, it'd let this thread consume an ARC
5372 * buffer before it should have (i.e. before we're under
5373 * the overflow limit and were signalled by the reclaim
5374 * thread). As long as that is a rare occurrence, it
5375 * shouldn't cause any harm.
5377 if (arc_is_overflowing()) {
5378 cv_signal(&arc_reclaim_thread_cv
);
5379 cv_wait(&arc_reclaim_waiters_cv
, &arc_reclaim_lock
);
5382 mutex_exit(&arc_reclaim_lock
);
5385 VERIFY3U(hdr
->b_type
, ==, type
);
5386 if (type
== ARC_BUFC_METADATA
) {
5387 arc_space_consume(size
, ARC_SPACE_META
);
5389 arc_space_consume(size
, ARC_SPACE_DATA
);
5393 * Update the state size. Note that ghost states have a
5394 * "ghost size" and so don't need to be updated.
5396 if (!GHOST_STATE(state
)) {
5398 (void) refcount_add_many(&state
->arcs_size
, size
, tag
);
5401 * If this is reached via arc_read, the link is
5402 * protected by the hash lock. If reached via
5403 * arc_buf_alloc, the header should not be accessed by
5404 * any other thread. And, if reached via arc_read_done,
5405 * the hash lock will protect it if it's found in the
5406 * hash table; otherwise no other thread should be
5407 * trying to [add|remove]_reference it.
5409 if (multilist_link_active(&hdr
->b_l1hdr
.b_arc_node
)) {
5410 ASSERT(refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
));
5411 (void) refcount_add_many(&state
->arcs_esize
[type
],
5416 * If we are growing the cache, and we are adding anonymous
5417 * data, and we have outgrown arc_p, update arc_p
5419 if (arc_size
< arc_c
&& hdr
->b_l1hdr
.b_state
== arc_anon
&&
5420 (refcount_count(&arc_anon
->arcs_size
) +
5421 refcount_count(&arc_mru
->arcs_size
) > arc_p
))
5422 arc_p
= MIN(arc_c
, arc_p
+ size
);
5427 arc_free_data_abd(arc_buf_hdr_t
*hdr
, abd_t
*abd
, uint64_t size
, void *tag
)
5429 arc_free_data_impl(hdr
, size
, tag
);
5434 arc_free_data_buf(arc_buf_hdr_t
*hdr
, void *buf
, uint64_t size
, void *tag
)
5436 arc_buf_contents_t type
= arc_buf_type(hdr
);
5438 arc_free_data_impl(hdr
, size
, tag
);
5439 if (type
== ARC_BUFC_METADATA
) {
5440 zio_buf_free(buf
, size
);
5442 ASSERT(type
== ARC_BUFC_DATA
);
5443 zio_data_buf_free(buf
, size
);
5448 * Free the arc data buffer.
5451 arc_free_data_impl(arc_buf_hdr_t
*hdr
, uint64_t size
, void *tag
)
5453 arc_state_t
*state
= hdr
->b_l1hdr
.b_state
;
5454 arc_buf_contents_t type
= arc_buf_type(hdr
);
5456 /* protected by hash lock, if in the hash table */
5457 if (multilist_link_active(&hdr
->b_l1hdr
.b_arc_node
)) {
5458 ASSERT(refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
));
5459 ASSERT(state
!= arc_anon
&& state
!= arc_l2c_only
);
5461 (void) refcount_remove_many(&state
->arcs_esize
[type
],
5464 (void) refcount_remove_many(&state
->arcs_size
, size
, tag
);
5466 VERIFY3U(hdr
->b_type
, ==, type
);
5467 if (type
== ARC_BUFC_METADATA
) {
5468 arc_space_return(size
, ARC_SPACE_META
);
5470 ASSERT(type
== ARC_BUFC_DATA
);
5471 arc_space_return(size
, ARC_SPACE_DATA
);
5476 * This routine is called whenever a buffer is accessed.
5477 * NOTE: the hash lock is dropped in this function.
5480 arc_access(arc_buf_hdr_t
*hdr
, kmutex_t
*hash_lock
)
5484 ASSERT(MUTEX_HELD(hash_lock
));
5485 ASSERT(HDR_HAS_L1HDR(hdr
));
5487 if (hdr
->b_l1hdr
.b_state
== arc_anon
) {
5489 * This buffer is not in the cache, and does not
5490 * appear in our "ghost" list. Add the new buffer
5494 ASSERT0(hdr
->b_l1hdr
.b_arc_access
);
5495 hdr
->b_l1hdr
.b_arc_access
= ddi_get_lbolt();
5496 DTRACE_PROBE1(new_state__mru
, arc_buf_hdr_t
*, hdr
);
5497 arc_change_state(arc_mru
, hdr
, hash_lock
);
5499 } else if (hdr
->b_l1hdr
.b_state
== arc_mru
) {
5500 now
= ddi_get_lbolt();
5503 * If this buffer is here because of a prefetch, then either:
5504 * - clear the flag if this is a "referencing" read
5505 * (any subsequent access will bump this into the MFU state).
5507 * - move the buffer to the head of the list if this is
5508 * another prefetch (to make it less likely to be evicted).
5510 if (HDR_PREFETCH(hdr
) || HDR_PRESCIENT_PREFETCH(hdr
)) {
5511 if (refcount_count(&hdr
->b_l1hdr
.b_refcnt
) == 0) {
5512 /* link protected by hash lock */
5513 ASSERT(multilist_link_active(
5514 &hdr
->b_l1hdr
.b_arc_node
));
5516 arc_hdr_clear_flags(hdr
,
5518 ARC_FLAG_PRESCIENT_PREFETCH
);
5519 atomic_inc_32(&hdr
->b_l1hdr
.b_mru_hits
);
5520 ARCSTAT_BUMP(arcstat_mru_hits
);
5522 hdr
->b_l1hdr
.b_arc_access
= now
;
5527 * This buffer has been "accessed" only once so far,
5528 * but it is still in the cache. Move it to the MFU
5531 if (ddi_time_after(now
, hdr
->b_l1hdr
.b_arc_access
+
5534 * More than 125ms have passed since we
5535 * instantiated this buffer. Move it to the
5536 * most frequently used state.
5538 hdr
->b_l1hdr
.b_arc_access
= now
;
5539 DTRACE_PROBE1(new_state__mfu
, arc_buf_hdr_t
*, hdr
);
5540 arc_change_state(arc_mfu
, hdr
, hash_lock
);
5542 atomic_inc_32(&hdr
->b_l1hdr
.b_mru_hits
);
5543 ARCSTAT_BUMP(arcstat_mru_hits
);
5544 } else if (hdr
->b_l1hdr
.b_state
== arc_mru_ghost
) {
5545 arc_state_t
*new_state
;
5547 * This buffer has been "accessed" recently, but
5548 * was evicted from the cache. Move it to the
5552 if (HDR_PREFETCH(hdr
) || HDR_PRESCIENT_PREFETCH(hdr
)) {
5553 new_state
= arc_mru
;
5554 if (refcount_count(&hdr
->b_l1hdr
.b_refcnt
) > 0) {
5555 arc_hdr_clear_flags(hdr
,
5557 ARC_FLAG_PRESCIENT_PREFETCH
);
5559 DTRACE_PROBE1(new_state__mru
, arc_buf_hdr_t
*, hdr
);
5561 new_state
= arc_mfu
;
5562 DTRACE_PROBE1(new_state__mfu
, arc_buf_hdr_t
*, hdr
);
5565 hdr
->b_l1hdr
.b_arc_access
= ddi_get_lbolt();
5566 arc_change_state(new_state
, hdr
, hash_lock
);
5568 atomic_inc_32(&hdr
->b_l1hdr
.b_mru_ghost_hits
);
5569 ARCSTAT_BUMP(arcstat_mru_ghost_hits
);
5570 } else if (hdr
->b_l1hdr
.b_state
== arc_mfu
) {
5572 * This buffer has been accessed more than once and is
5573 * still in the cache. Keep it in the MFU state.
5575 * NOTE: an add_reference() that occurred when we did
5576 * the arc_read() will have kicked this off the list.
5577 * If it was a prefetch, we will explicitly move it to
5578 * the head of the list now.
5581 atomic_inc_32(&hdr
->b_l1hdr
.b_mfu_hits
);
5582 ARCSTAT_BUMP(arcstat_mfu_hits
);
5583 hdr
->b_l1hdr
.b_arc_access
= ddi_get_lbolt();
5584 } else if (hdr
->b_l1hdr
.b_state
== arc_mfu_ghost
) {
5585 arc_state_t
*new_state
= arc_mfu
;
5587 * This buffer has been accessed more than once but has
5588 * been evicted from the cache. Move it back to the
5592 if (HDR_PREFETCH(hdr
) || HDR_PRESCIENT_PREFETCH(hdr
)) {
5594 * This is a prefetch access...
5595 * move this block back to the MRU state.
5597 new_state
= arc_mru
;
5600 hdr
->b_l1hdr
.b_arc_access
= ddi_get_lbolt();
5601 DTRACE_PROBE1(new_state__mfu
, arc_buf_hdr_t
*, hdr
);
5602 arc_change_state(new_state
, hdr
, hash_lock
);
5604 atomic_inc_32(&hdr
->b_l1hdr
.b_mfu_ghost_hits
);
5605 ARCSTAT_BUMP(arcstat_mfu_ghost_hits
);
5606 } else if (hdr
->b_l1hdr
.b_state
== arc_l2c_only
) {
5608 * This buffer is on the 2nd Level ARC.
5611 hdr
->b_l1hdr
.b_arc_access
= ddi_get_lbolt();
5612 DTRACE_PROBE1(new_state__mfu
, arc_buf_hdr_t
*, hdr
);
5613 arc_change_state(arc_mfu
, hdr
, hash_lock
);
5615 cmn_err(CE_PANIC
, "invalid arc state 0x%p",
5616 hdr
->b_l1hdr
.b_state
);
5621 * This routine is called by dbuf_hold() to update the arc_access() state
5622 * which otherwise would be skipped for entries in the dbuf cache.
5625 arc_buf_access(arc_buf_t
*buf
)
5627 mutex_enter(&buf
->b_evict_lock
);
5628 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
5631 * Avoid taking the hash_lock when possible as an optimization.
5632 * The header must be checked again under the hash_lock in order
5633 * to handle the case where it is concurrently being released.
5635 if (hdr
->b_l1hdr
.b_state
== arc_anon
|| HDR_EMPTY(hdr
)) {
5636 mutex_exit(&buf
->b_evict_lock
);
5640 kmutex_t
*hash_lock
= HDR_LOCK(hdr
);
5641 mutex_enter(hash_lock
);
5643 if (hdr
->b_l1hdr
.b_state
== arc_anon
|| HDR_EMPTY(hdr
)) {
5644 mutex_exit(hash_lock
);
5645 mutex_exit(&buf
->b_evict_lock
);
5646 ARCSTAT_BUMP(arcstat_access_skip
);
5650 mutex_exit(&buf
->b_evict_lock
);
5652 ASSERT(hdr
->b_l1hdr
.b_state
== arc_mru
||
5653 hdr
->b_l1hdr
.b_state
== arc_mfu
);
5655 DTRACE_PROBE1(arc__hit
, arc_buf_hdr_t
*, hdr
);
5656 arc_access(hdr
, hash_lock
);
5657 mutex_exit(hash_lock
);
5659 ARCSTAT_BUMP(arcstat_hits
);
5660 ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr
) && !HDR_PRESCIENT_PREFETCH(hdr
),
5661 demand
, prefetch
, !HDR_ISTYPE_METADATA(hdr
), data
, metadata
, hits
);
5664 /* a generic arc_read_done_func_t which you can use */
5667 arc_bcopy_func(zio_t
*zio
, const zbookmark_phys_t
*zb
, const blkptr_t
*bp
,
5668 arc_buf_t
*buf
, void *arg
)
5673 bcopy(buf
->b_data
, arg
, arc_buf_size(buf
));
5674 arc_buf_destroy(buf
, arg
);
5677 /* a generic arc_read_done_func_t */
5680 arc_getbuf_func(zio_t
*zio
, const zbookmark_phys_t
*zb
, const blkptr_t
*bp
,
5681 arc_buf_t
*buf
, void *arg
)
5683 arc_buf_t
**bufp
= arg
;
5689 ASSERT(buf
->b_data
);
5694 arc_hdr_verify(arc_buf_hdr_t
*hdr
, blkptr_t
*bp
)
5696 if (BP_IS_HOLE(bp
) || BP_IS_EMBEDDED(bp
)) {
5697 ASSERT3U(HDR_GET_PSIZE(hdr
), ==, 0);
5698 ASSERT3U(arc_hdr_get_compress(hdr
), ==, ZIO_COMPRESS_OFF
);
5700 if (HDR_COMPRESSION_ENABLED(hdr
)) {
5701 ASSERT3U(arc_hdr_get_compress(hdr
), ==,
5702 BP_GET_COMPRESS(bp
));
5704 ASSERT3U(HDR_GET_LSIZE(hdr
), ==, BP_GET_LSIZE(bp
));
5705 ASSERT3U(HDR_GET_PSIZE(hdr
), ==, BP_GET_PSIZE(bp
));
5706 ASSERT3U(!!HDR_PROTECTED(hdr
), ==, BP_IS_PROTECTED(bp
));
5711 arc_read_done(zio_t
*zio
)
5713 blkptr_t
*bp
= zio
->io_bp
;
5714 arc_buf_hdr_t
*hdr
= zio
->io_private
;
5715 kmutex_t
*hash_lock
= NULL
;
5716 arc_callback_t
*callback_list
;
5717 arc_callback_t
*acb
;
5718 boolean_t freeable
= B_FALSE
;
5721 * The hdr was inserted into hash-table and removed from lists
5722 * prior to starting I/O. We should find this header, since
5723 * it's in the hash table, and it should be legit since it's
5724 * not possible to evict it during the I/O. The only possible
5725 * reason for it not to be found is if we were freed during the
5728 if (HDR_IN_HASH_TABLE(hdr
)) {
5729 arc_buf_hdr_t
*found
;
5731 ASSERT3U(hdr
->b_birth
, ==, BP_PHYSICAL_BIRTH(zio
->io_bp
));
5732 ASSERT3U(hdr
->b_dva
.dva_word
[0], ==,
5733 BP_IDENTITY(zio
->io_bp
)->dva_word
[0]);
5734 ASSERT3U(hdr
->b_dva
.dva_word
[1], ==,
5735 BP_IDENTITY(zio
->io_bp
)->dva_word
[1]);
5737 found
= buf_hash_find(hdr
->b_spa
, zio
->io_bp
, &hash_lock
);
5739 ASSERT((found
== hdr
&&
5740 DVA_EQUAL(&hdr
->b_dva
, BP_IDENTITY(zio
->io_bp
))) ||
5741 (found
== hdr
&& HDR_L2_READING(hdr
)));
5742 ASSERT3P(hash_lock
, !=, NULL
);
5745 if (BP_IS_PROTECTED(bp
)) {
5746 hdr
->b_crypt_hdr
.b_ot
= BP_GET_TYPE(bp
);
5747 hdr
->b_crypt_hdr
.b_dsobj
= zio
->io_bookmark
.zb_objset
;
5748 zio_crypt_decode_params_bp(bp
, hdr
->b_crypt_hdr
.b_salt
,
5749 hdr
->b_crypt_hdr
.b_iv
);
5751 if (BP_GET_TYPE(bp
) == DMU_OT_INTENT_LOG
) {
5754 tmpbuf
= abd_borrow_buf_copy(zio
->io_abd
,
5755 sizeof (zil_chain_t
));
5756 zio_crypt_decode_mac_zil(tmpbuf
,
5757 hdr
->b_crypt_hdr
.b_mac
);
5758 abd_return_buf(zio
->io_abd
, tmpbuf
,
5759 sizeof (zil_chain_t
));
5761 zio_crypt_decode_mac_bp(bp
, hdr
->b_crypt_hdr
.b_mac
);
5765 if (zio
->io_error
== 0) {
5766 /* byteswap if necessary */
5767 if (BP_SHOULD_BYTESWAP(zio
->io_bp
)) {
5768 if (BP_GET_LEVEL(zio
->io_bp
) > 0) {
5769 hdr
->b_l1hdr
.b_byteswap
= DMU_BSWAP_UINT64
;
5771 hdr
->b_l1hdr
.b_byteswap
=
5772 DMU_OT_BYTESWAP(BP_GET_TYPE(zio
->io_bp
));
5775 hdr
->b_l1hdr
.b_byteswap
= DMU_BSWAP_NUMFUNCS
;
5779 arc_hdr_clear_flags(hdr
, ARC_FLAG_L2_EVICTED
);
5780 if (l2arc_noprefetch
&& HDR_PREFETCH(hdr
))
5781 arc_hdr_clear_flags(hdr
, ARC_FLAG_L2CACHE
);
5783 callback_list
= hdr
->b_l1hdr
.b_acb
;
5784 ASSERT3P(callback_list
, !=, NULL
);
5786 if (hash_lock
&& zio
->io_error
== 0 &&
5787 hdr
->b_l1hdr
.b_state
== arc_anon
) {
5789 * Only call arc_access on anonymous buffers. This is because
5790 * if we've issued an I/O for an evicted buffer, we've already
5791 * called arc_access (to prevent any simultaneous readers from
5792 * getting confused).
5794 arc_access(hdr
, hash_lock
);
5798 * If a read request has a callback (i.e. acb_done is not NULL), then we
5799 * make a buf containing the data according to the parameters which were
5800 * passed in. The implementation of arc_buf_alloc_impl() ensures that we
5801 * aren't needlessly decompressing the data multiple times.
5803 int callback_cnt
= 0;
5804 for (acb
= callback_list
; acb
!= NULL
; acb
= acb
->acb_next
) {
5810 if (zio
->io_error
!= 0)
5813 int error
= arc_buf_alloc_impl(hdr
, zio
->io_spa
,
5814 acb
->acb_dsobj
, acb
->acb_private
, acb
->acb_encrypted
,
5815 acb
->acb_compressed
, acb
->acb_noauth
, B_TRUE
,
5818 arc_buf_destroy(acb
->acb_buf
, acb
->acb_private
);
5819 acb
->acb_buf
= NULL
;
5823 * Assert non-speculative zios didn't fail because an
5824 * encryption key wasn't loaded
5826 ASSERT((zio
->io_flags
& ZIO_FLAG_SPECULATIVE
) ||
5830 * If we failed to decrypt, report an error now (as the zio
5831 * layer would have done if it had done the transforms).
5833 if (error
== ECKSUM
) {
5834 ASSERT(BP_IS_PROTECTED(bp
));
5835 error
= SET_ERROR(EIO
);
5836 spa_log_error(zio
->io_spa
, &zio
->io_bookmark
);
5837 if ((zio
->io_flags
& ZIO_FLAG_SPECULATIVE
) == 0) {
5838 zfs_ereport_post(FM_EREPORT_ZFS_AUTHENTICATION
,
5839 zio
->io_spa
, NULL
, &zio
->io_bookmark
, zio
,
5844 if (zio
->io_error
== 0)
5845 zio
->io_error
= error
;
5847 hdr
->b_l1hdr
.b_acb
= NULL
;
5848 arc_hdr_clear_flags(hdr
, ARC_FLAG_IO_IN_PROGRESS
);
5849 if (callback_cnt
== 0)
5850 ASSERT(hdr
->b_l1hdr
.b_pabd
!= NULL
|| HDR_HAS_RABD(hdr
));
5852 ASSERT(refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
) ||
5853 callback_list
!= NULL
);
5855 if (zio
->io_error
== 0) {
5856 arc_hdr_verify(hdr
, zio
->io_bp
);
5858 arc_hdr_set_flags(hdr
, ARC_FLAG_IO_ERROR
);
5859 if (hdr
->b_l1hdr
.b_state
!= arc_anon
)
5860 arc_change_state(arc_anon
, hdr
, hash_lock
);
5861 if (HDR_IN_HASH_TABLE(hdr
))
5862 buf_hash_remove(hdr
);
5863 freeable
= refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
);
5867 * Broadcast before we drop the hash_lock to avoid the possibility
5868 * that the hdr (and hence the cv) might be freed before we get to
5869 * the cv_broadcast().
5871 cv_broadcast(&hdr
->b_l1hdr
.b_cv
);
5873 if (hash_lock
!= NULL
) {
5874 mutex_exit(hash_lock
);
5877 * This block was freed while we waited for the read to
5878 * complete. It has been removed from the hash table and
5879 * moved to the anonymous state (so that it won't show up
5882 ASSERT3P(hdr
->b_l1hdr
.b_state
, ==, arc_anon
);
5883 freeable
= refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
);
5886 /* execute each callback and free its structure */
5887 while ((acb
= callback_list
) != NULL
) {
5888 if (acb
->acb_done
) {
5889 acb
->acb_done(zio
, &zio
->io_bookmark
, zio
->io_bp
,
5890 acb
->acb_buf
, acb
->acb_private
);
5893 if (acb
->acb_zio_dummy
!= NULL
) {
5894 acb
->acb_zio_dummy
->io_error
= zio
->io_error
;
5895 zio_nowait(acb
->acb_zio_dummy
);
5898 callback_list
= acb
->acb_next
;
5899 kmem_free(acb
, sizeof (arc_callback_t
));
5903 arc_hdr_destroy(hdr
);
5907 * "Read" the block at the specified DVA (in bp) via the
5908 * cache. If the block is found in the cache, invoke the provided
5909 * callback immediately and return. Note that the `zio' parameter
5910 * in the callback will be NULL in this case, since no IO was
5911 * required. If the block is not in the cache pass the read request
5912 * on to the spa with a substitute callback function, so that the
5913 * requested block will be added to the cache.
5915 * If a read request arrives for a block that has a read in-progress,
5916 * either wait for the in-progress read to complete (and return the
5917 * results); or, if this is a read with a "done" func, add a record
5918 * to the read to invoke the "done" func when the read completes,
5919 * and return; or just return.
5921 * arc_read_done() will invoke all the requested "done" functions
5922 * for readers of this block.
5925 arc_read(zio_t
*pio
, spa_t
*spa
, const blkptr_t
*bp
,
5926 arc_read_done_func_t
*done
, void *private, zio_priority_t priority
,
5927 int zio_flags
, arc_flags_t
*arc_flags
, const zbookmark_phys_t
*zb
)
5929 arc_buf_hdr_t
*hdr
= NULL
;
5930 kmutex_t
*hash_lock
= NULL
;
5932 uint64_t guid
= spa_load_guid(spa
);
5933 boolean_t compressed_read
= (zio_flags
& ZIO_FLAG_RAW_COMPRESS
) != 0;
5934 boolean_t encrypted_read
= BP_IS_ENCRYPTED(bp
) &&
5935 (zio_flags
& ZIO_FLAG_RAW_ENCRYPT
) != 0;
5936 boolean_t noauth_read
= BP_IS_AUTHENTICATED(bp
) &&
5937 (zio_flags
& ZIO_FLAG_RAW_ENCRYPT
) != 0;
5940 ASSERT(!BP_IS_EMBEDDED(bp
) ||
5941 BPE_GET_ETYPE(bp
) == BP_EMBEDDED_TYPE_DATA
);
5944 if (!BP_IS_EMBEDDED(bp
)) {
5946 * Embedded BP's have no DVA and require no I/O to "read".
5947 * Create an anonymous arc buf to back it.
5949 hdr
= buf_hash_find(guid
, bp
, &hash_lock
);
5953 * Determine if we have an L1 cache hit or a cache miss. For simplicity
5954 * we maintain encrypted data seperately from compressed / uncompressed
5955 * data. If the user is requesting raw encrypted data and we don't have
5956 * that in the header we will read from disk to guarantee that we can
5957 * get it even if the encryption keys aren't loaded.
5959 if (hdr
!= NULL
&& HDR_HAS_L1HDR(hdr
) && (HDR_HAS_RABD(hdr
) ||
5960 (hdr
->b_l1hdr
.b_pabd
!= NULL
&& !encrypted_read
))) {
5961 arc_buf_t
*buf
= NULL
;
5962 *arc_flags
|= ARC_FLAG_CACHED
;
5964 if (HDR_IO_IN_PROGRESS(hdr
)) {
5965 zio_t
*head_zio
= hdr
->b_l1hdr
.b_acb
->acb_zio_head
;
5967 ASSERT3P(head_zio
, !=, NULL
);
5968 if ((hdr
->b_flags
& ARC_FLAG_PRIO_ASYNC_READ
) &&
5969 priority
== ZIO_PRIORITY_SYNC_READ
) {
5971 * This is a sync read that needs to wait for
5972 * an in-flight async read. Request that the
5973 * zio have its priority upgraded.
5975 zio_change_priority(head_zio
, priority
);
5976 DTRACE_PROBE1(arc__async__upgrade__sync
,
5977 arc_buf_hdr_t
*, hdr
);
5978 ARCSTAT_BUMP(arcstat_async_upgrade_sync
);
5980 if (hdr
->b_flags
& ARC_FLAG_PREDICTIVE_PREFETCH
) {
5981 arc_hdr_clear_flags(hdr
,
5982 ARC_FLAG_PREDICTIVE_PREFETCH
);
5985 if (*arc_flags
& ARC_FLAG_WAIT
) {
5986 cv_wait(&hdr
->b_l1hdr
.b_cv
, hash_lock
);
5987 mutex_exit(hash_lock
);
5990 ASSERT(*arc_flags
& ARC_FLAG_NOWAIT
);
5993 arc_callback_t
*acb
= NULL
;
5995 acb
= kmem_zalloc(sizeof (arc_callback_t
),
5997 acb
->acb_done
= done
;
5998 acb
->acb_private
= private;
5999 acb
->acb_compressed
= compressed_read
;
6000 acb
->acb_encrypted
= encrypted_read
;
6001 acb
->acb_noauth
= noauth_read
;
6002 acb
->acb_dsobj
= zb
->zb_objset
;
6004 acb
->acb_zio_dummy
= zio_null(pio
,
6005 spa
, NULL
, NULL
, NULL
, zio_flags
);
6007 ASSERT3P(acb
->acb_done
, !=, NULL
);
6008 acb
->acb_zio_head
= head_zio
;
6009 acb
->acb_next
= hdr
->b_l1hdr
.b_acb
;
6010 hdr
->b_l1hdr
.b_acb
= acb
;
6011 mutex_exit(hash_lock
);
6014 mutex_exit(hash_lock
);
6018 ASSERT(hdr
->b_l1hdr
.b_state
== arc_mru
||
6019 hdr
->b_l1hdr
.b_state
== arc_mfu
);
6022 if (hdr
->b_flags
& ARC_FLAG_PREDICTIVE_PREFETCH
) {
6024 * This is a demand read which does not have to
6025 * wait for i/o because we did a predictive
6026 * prefetch i/o for it, which has completed.
6029 arc__demand__hit__predictive__prefetch
,
6030 arc_buf_hdr_t
*, hdr
);
6032 arcstat_demand_hit_predictive_prefetch
);
6033 arc_hdr_clear_flags(hdr
,
6034 ARC_FLAG_PREDICTIVE_PREFETCH
);
6037 if (hdr
->b_flags
& ARC_FLAG_PRESCIENT_PREFETCH
) {
6039 arcstat_demand_hit_prescient_prefetch
);
6040 arc_hdr_clear_flags(hdr
,
6041 ARC_FLAG_PRESCIENT_PREFETCH
);
6044 ASSERT(!BP_IS_EMBEDDED(bp
) || !BP_IS_HOLE(bp
));
6046 /* Get a buf with the desired data in it. */
6047 rc
= arc_buf_alloc_impl(hdr
, spa
, zb
->zb_objset
,
6048 private, encrypted_read
, compressed_read
,
6049 noauth_read
, B_TRUE
, &buf
);
6052 * Convert authentication and decryption errors
6053 * to EIO (and generate an ereport) before
6056 rc
= SET_ERROR(EIO
);
6057 zfs_ereport_post(FM_EREPORT_ZFS_AUTHENTICATION
,
6058 spa
, NULL
, zb
, NULL
, 0, 0);
6061 arc_buf_destroy(buf
, private);
6065 /* assert any errors weren't due to unloaded keys */
6066 ASSERT((zio_flags
& ZIO_FLAG_SPECULATIVE
) ||
6068 } else if (*arc_flags
& ARC_FLAG_PREFETCH
&&
6069 refcount_count(&hdr
->b_l1hdr
.b_refcnt
) == 0) {
6070 arc_hdr_set_flags(hdr
, ARC_FLAG_PREFETCH
);
6072 DTRACE_PROBE1(arc__hit
, arc_buf_hdr_t
*, hdr
);
6073 arc_access(hdr
, hash_lock
);
6074 if (*arc_flags
& ARC_FLAG_PRESCIENT_PREFETCH
)
6075 arc_hdr_set_flags(hdr
, ARC_FLAG_PRESCIENT_PREFETCH
);
6076 if (*arc_flags
& ARC_FLAG_L2CACHE
)
6077 arc_hdr_set_flags(hdr
, ARC_FLAG_L2CACHE
);
6078 mutex_exit(hash_lock
);
6079 ARCSTAT_BUMP(arcstat_hits
);
6080 ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr
),
6081 demand
, prefetch
, !HDR_ISTYPE_METADATA(hdr
),
6082 data
, metadata
, hits
);
6085 done(NULL
, zb
, bp
, buf
, private);
6087 uint64_t lsize
= BP_GET_LSIZE(bp
);
6088 uint64_t psize
= BP_GET_PSIZE(bp
);
6089 arc_callback_t
*acb
;
6092 boolean_t devw
= B_FALSE
;
6097 * Gracefully handle a damaged logical block size as a
6100 if (lsize
> spa_maxblocksize(spa
)) {
6101 rc
= SET_ERROR(ECKSUM
);
6106 /* this block is not in the cache */
6107 arc_buf_hdr_t
*exists
= NULL
;
6108 arc_buf_contents_t type
= BP_GET_BUFC_TYPE(bp
);
6109 hdr
= arc_hdr_alloc(spa_load_guid(spa
), psize
, lsize
,
6110 BP_IS_PROTECTED(bp
), BP_GET_COMPRESS(bp
), type
,
6113 if (!BP_IS_EMBEDDED(bp
)) {
6114 hdr
->b_dva
= *BP_IDENTITY(bp
);
6115 hdr
->b_birth
= BP_PHYSICAL_BIRTH(bp
);
6116 exists
= buf_hash_insert(hdr
, &hash_lock
);
6118 if (exists
!= NULL
) {
6119 /* somebody beat us to the hash insert */
6120 mutex_exit(hash_lock
);
6121 buf_discard_identity(hdr
);
6122 arc_hdr_destroy(hdr
);
6123 goto top
; /* restart the IO request */
6127 * This block is in the ghost cache or encrypted data
6128 * was requested and we didn't have it. If it was
6129 * L2-only (and thus didn't have an L1 hdr),
6130 * we realloc the header to add an L1 hdr.
6132 if (!HDR_HAS_L1HDR(hdr
)) {
6133 hdr
= arc_hdr_realloc(hdr
, hdr_l2only_cache
,
6137 if (GHOST_STATE(hdr
->b_l1hdr
.b_state
)) {
6138 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, ==, NULL
);
6139 ASSERT(!HDR_HAS_RABD(hdr
));
6140 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
6141 ASSERT0(refcount_count(&hdr
->b_l1hdr
.b_refcnt
));
6142 ASSERT3P(hdr
->b_l1hdr
.b_buf
, ==, NULL
);
6143 ASSERT3P(hdr
->b_l1hdr
.b_freeze_cksum
, ==, NULL
);
6144 } else if (HDR_IO_IN_PROGRESS(hdr
)) {
6146 * If this header already had an IO in progress
6147 * and we are performing another IO to fetch
6148 * encrypted data we must wait until the first
6149 * IO completes so as not to confuse
6150 * arc_read_done(). This should be very rare
6151 * and so the performance impact shouldn't
6154 cv_wait(&hdr
->b_l1hdr
.b_cv
, hash_lock
);
6155 mutex_exit(hash_lock
);
6160 * This is a delicate dance that we play here.
6161 * This hdr might be in the ghost list so we access
6162 * it to move it out of the ghost list before we
6163 * initiate the read. If it's a prefetch then
6164 * it won't have a callback so we'll remove the
6165 * reference that arc_buf_alloc_impl() created. We
6166 * do this after we've called arc_access() to
6167 * avoid hitting an assert in remove_reference().
6169 arc_access(hdr
, hash_lock
);
6170 arc_hdr_alloc_abd(hdr
, encrypted_read
);
6173 if (encrypted_read
) {
6174 ASSERT(HDR_HAS_RABD(hdr
));
6175 size
= HDR_GET_PSIZE(hdr
);
6176 hdr_abd
= hdr
->b_crypt_hdr
.b_rabd
;
6177 zio_flags
|= ZIO_FLAG_RAW
;
6179 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, !=, NULL
);
6180 size
= arc_hdr_size(hdr
);
6181 hdr_abd
= hdr
->b_l1hdr
.b_pabd
;
6183 if (arc_hdr_get_compress(hdr
) != ZIO_COMPRESS_OFF
) {
6184 zio_flags
|= ZIO_FLAG_RAW_COMPRESS
;
6188 * For authenticated bp's, we do not ask the ZIO layer
6189 * to authenticate them since this will cause the entire
6190 * IO to fail if the key isn't loaded. Instead, we
6191 * defer authentication until arc_buf_fill(), which will
6192 * verify the data when the key is available.
6194 if (BP_IS_AUTHENTICATED(bp
))
6195 zio_flags
|= ZIO_FLAG_RAW_ENCRYPT
;
6198 if (*arc_flags
& ARC_FLAG_PREFETCH
&&
6199 refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
))
6200 arc_hdr_set_flags(hdr
, ARC_FLAG_PREFETCH
);
6201 if (*arc_flags
& ARC_FLAG_PRESCIENT_PREFETCH
)
6202 arc_hdr_set_flags(hdr
, ARC_FLAG_PRESCIENT_PREFETCH
);
6203 if (*arc_flags
& ARC_FLAG_L2CACHE
)
6204 arc_hdr_set_flags(hdr
, ARC_FLAG_L2CACHE
);
6205 if (BP_IS_AUTHENTICATED(bp
))
6206 arc_hdr_set_flags(hdr
, ARC_FLAG_NOAUTH
);
6207 if (BP_GET_LEVEL(bp
) > 0)
6208 arc_hdr_set_flags(hdr
, ARC_FLAG_INDIRECT
);
6209 if (*arc_flags
& ARC_FLAG_PREDICTIVE_PREFETCH
)
6210 arc_hdr_set_flags(hdr
, ARC_FLAG_PREDICTIVE_PREFETCH
);
6211 ASSERT(!GHOST_STATE(hdr
->b_l1hdr
.b_state
));
6213 acb
= kmem_zalloc(sizeof (arc_callback_t
), KM_SLEEP
);
6214 acb
->acb_done
= done
;
6215 acb
->acb_private
= private;
6216 acb
->acb_compressed
= compressed_read
;
6217 acb
->acb_encrypted
= encrypted_read
;
6218 acb
->acb_noauth
= noauth_read
;
6219 acb
->acb_dsobj
= zb
->zb_objset
;
6221 ASSERT3P(hdr
->b_l1hdr
.b_acb
, ==, NULL
);
6222 hdr
->b_l1hdr
.b_acb
= acb
;
6223 arc_hdr_set_flags(hdr
, ARC_FLAG_IO_IN_PROGRESS
);
6225 if (HDR_HAS_L2HDR(hdr
) &&
6226 (vd
= hdr
->b_l2hdr
.b_dev
->l2ad_vdev
) != NULL
) {
6227 devw
= hdr
->b_l2hdr
.b_dev
->l2ad_writing
;
6228 addr
= hdr
->b_l2hdr
.b_daddr
;
6230 * Lock out L2ARC device removal.
6232 if (vdev_is_dead(vd
) ||
6233 !spa_config_tryenter(spa
, SCL_L2ARC
, vd
, RW_READER
))
6238 * We count both async reads and scrub IOs as asynchronous so
6239 * that both can be upgraded in the event of a cache hit while
6240 * the read IO is still in-flight.
6242 if (priority
== ZIO_PRIORITY_ASYNC_READ
||
6243 priority
== ZIO_PRIORITY_SCRUB
)
6244 arc_hdr_set_flags(hdr
, ARC_FLAG_PRIO_ASYNC_READ
);
6246 arc_hdr_clear_flags(hdr
, ARC_FLAG_PRIO_ASYNC_READ
);
6249 * At this point, we have a level 1 cache miss. Try again in
6250 * L2ARC if possible.
6252 ASSERT3U(HDR_GET_LSIZE(hdr
), ==, lsize
);
6254 DTRACE_PROBE4(arc__miss
, arc_buf_hdr_t
*, hdr
, blkptr_t
*, bp
,
6255 uint64_t, lsize
, zbookmark_phys_t
*, zb
);
6256 ARCSTAT_BUMP(arcstat_misses
);
6257 ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr
),
6258 demand
, prefetch
, !HDR_ISTYPE_METADATA(hdr
),
6259 data
, metadata
, misses
);
6261 if (vd
!= NULL
&& l2arc_ndev
!= 0 && !(l2arc_norw
&& devw
)) {
6263 * Read from the L2ARC if the following are true:
6264 * 1. The L2ARC vdev was previously cached.
6265 * 2. This buffer still has L2ARC metadata.
6266 * 3. This buffer isn't currently writing to the L2ARC.
6267 * 4. The L2ARC entry wasn't evicted, which may
6268 * also have invalidated the vdev.
6269 * 5. This isn't prefetch and l2arc_noprefetch is set.
6271 if (HDR_HAS_L2HDR(hdr
) &&
6272 !HDR_L2_WRITING(hdr
) && !HDR_L2_EVICTED(hdr
) &&
6273 !(l2arc_noprefetch
&& HDR_PREFETCH(hdr
))) {
6274 l2arc_read_callback_t
*cb
;
6278 DTRACE_PROBE1(l2arc__hit
, arc_buf_hdr_t
*, hdr
);
6279 ARCSTAT_BUMP(arcstat_l2_hits
);
6280 atomic_inc_32(&hdr
->b_l2hdr
.b_hits
);
6282 cb
= kmem_zalloc(sizeof (l2arc_read_callback_t
),
6284 cb
->l2rcb_hdr
= hdr
;
6287 cb
->l2rcb_flags
= zio_flags
;
6289 asize
= vdev_psize_to_asize(vd
, size
);
6290 if (asize
!= size
) {
6291 abd
= abd_alloc_for_io(asize
,
6292 HDR_ISTYPE_METADATA(hdr
));
6293 cb
->l2rcb_abd
= abd
;
6298 ASSERT(addr
>= VDEV_LABEL_START_SIZE
&&
6299 addr
+ asize
<= vd
->vdev_psize
-
6300 VDEV_LABEL_END_SIZE
);
6303 * l2arc read. The SCL_L2ARC lock will be
6304 * released by l2arc_read_done().
6305 * Issue a null zio if the underlying buffer
6306 * was squashed to zero size by compression.
6308 ASSERT3U(arc_hdr_get_compress(hdr
), !=,
6309 ZIO_COMPRESS_EMPTY
);
6310 rzio
= zio_read_phys(pio
, vd
, addr
,
6313 l2arc_read_done
, cb
, priority
,
6314 zio_flags
| ZIO_FLAG_DONT_CACHE
|
6316 ZIO_FLAG_DONT_PROPAGATE
|
6317 ZIO_FLAG_DONT_RETRY
, B_FALSE
);
6318 acb
->acb_zio_head
= rzio
;
6320 if (hash_lock
!= NULL
)
6321 mutex_exit(hash_lock
);
6323 DTRACE_PROBE2(l2arc__read
, vdev_t
*, vd
,
6325 ARCSTAT_INCR(arcstat_l2_read_bytes
,
6326 HDR_GET_PSIZE(hdr
));
6328 if (*arc_flags
& ARC_FLAG_NOWAIT
) {
6333 ASSERT(*arc_flags
& ARC_FLAG_WAIT
);
6334 if (zio_wait(rzio
) == 0)
6337 /* l2arc read error; goto zio_read() */
6338 if (hash_lock
!= NULL
)
6339 mutex_enter(hash_lock
);
6341 DTRACE_PROBE1(l2arc__miss
,
6342 arc_buf_hdr_t
*, hdr
);
6343 ARCSTAT_BUMP(arcstat_l2_misses
);
6344 if (HDR_L2_WRITING(hdr
))
6345 ARCSTAT_BUMP(arcstat_l2_rw_clash
);
6346 spa_config_exit(spa
, SCL_L2ARC
, vd
);
6350 spa_config_exit(spa
, SCL_L2ARC
, vd
);
6351 if (l2arc_ndev
!= 0) {
6352 DTRACE_PROBE1(l2arc__miss
,
6353 arc_buf_hdr_t
*, hdr
);
6354 ARCSTAT_BUMP(arcstat_l2_misses
);
6358 rzio
= zio_read(pio
, spa
, bp
, hdr_abd
, size
,
6359 arc_read_done
, hdr
, priority
, zio_flags
, zb
);
6360 acb
->acb_zio_head
= rzio
;
6362 if (hash_lock
!= NULL
)
6363 mutex_exit(hash_lock
);
6365 if (*arc_flags
& ARC_FLAG_WAIT
) {
6366 rc
= zio_wait(rzio
);
6370 ASSERT(*arc_flags
& ARC_FLAG_NOWAIT
);
6375 /* embedded bps don't actually go to disk */
6376 if (!BP_IS_EMBEDDED(bp
))
6377 spa_read_history_add(spa
, zb
, *arc_flags
);
6382 arc_add_prune_callback(arc_prune_func_t
*func
, void *private)
6386 p
= kmem_alloc(sizeof (*p
), KM_SLEEP
);
6388 p
->p_private
= private;
6389 list_link_init(&p
->p_node
);
6390 refcount_create(&p
->p_refcnt
);
6392 mutex_enter(&arc_prune_mtx
);
6393 refcount_add(&p
->p_refcnt
, &arc_prune_list
);
6394 list_insert_head(&arc_prune_list
, p
);
6395 mutex_exit(&arc_prune_mtx
);
6401 arc_remove_prune_callback(arc_prune_t
*p
)
6403 boolean_t wait
= B_FALSE
;
6404 mutex_enter(&arc_prune_mtx
);
6405 list_remove(&arc_prune_list
, p
);
6406 if (refcount_remove(&p
->p_refcnt
, &arc_prune_list
) > 0)
6408 mutex_exit(&arc_prune_mtx
);
6410 /* wait for arc_prune_task to finish */
6412 taskq_wait_outstanding(arc_prune_taskq
, 0);
6413 ASSERT0(refcount_count(&p
->p_refcnt
));
6414 refcount_destroy(&p
->p_refcnt
);
6415 kmem_free(p
, sizeof (*p
));
6419 * Notify the arc that a block was freed, and thus will never be used again.
6422 arc_freed(spa_t
*spa
, const blkptr_t
*bp
)
6425 kmutex_t
*hash_lock
;
6426 uint64_t guid
= spa_load_guid(spa
);
6428 ASSERT(!BP_IS_EMBEDDED(bp
));
6430 hdr
= buf_hash_find(guid
, bp
, &hash_lock
);
6435 * We might be trying to free a block that is still doing I/O
6436 * (i.e. prefetch) or has a reference (i.e. a dedup-ed,
6437 * dmu_sync-ed block). If this block is being prefetched, then it
6438 * would still have the ARC_FLAG_IO_IN_PROGRESS flag set on the hdr
6439 * until the I/O completes. A block may also have a reference if it is
6440 * part of a dedup-ed, dmu_synced write. The dmu_sync() function would
6441 * have written the new block to its final resting place on disk but
6442 * without the dedup flag set. This would have left the hdr in the MRU
6443 * state and discoverable. When the txg finally syncs it detects that
6444 * the block was overridden in open context and issues an override I/O.
6445 * Since this is a dedup block, the override I/O will determine if the
6446 * block is already in the DDT. If so, then it will replace the io_bp
6447 * with the bp from the DDT and allow the I/O to finish. When the I/O
6448 * reaches the done callback, dbuf_write_override_done, it will
6449 * check to see if the io_bp and io_bp_override are identical.
6450 * If they are not, then it indicates that the bp was replaced with
6451 * the bp in the DDT and the override bp is freed. This allows
6452 * us to arrive here with a reference on a block that is being
6453 * freed. So if we have an I/O in progress, or a reference to
6454 * this hdr, then we don't destroy the hdr.
6456 if (!HDR_HAS_L1HDR(hdr
) || (!HDR_IO_IN_PROGRESS(hdr
) &&
6457 refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
))) {
6458 arc_change_state(arc_anon
, hdr
, hash_lock
);
6459 arc_hdr_destroy(hdr
);
6460 mutex_exit(hash_lock
);
6462 mutex_exit(hash_lock
);
6468 * Release this buffer from the cache, making it an anonymous buffer. This
6469 * must be done after a read and prior to modifying the buffer contents.
6470 * If the buffer has more than one reference, we must make
6471 * a new hdr for the buffer.
6474 arc_release(arc_buf_t
*buf
, void *tag
)
6476 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
6479 * It would be nice to assert that if its DMU metadata (level >
6480 * 0 || it's the dnode file), then it must be syncing context.
6481 * But we don't know that information at this level.
6484 mutex_enter(&buf
->b_evict_lock
);
6486 ASSERT(HDR_HAS_L1HDR(hdr
));
6489 * We don't grab the hash lock prior to this check, because if
6490 * the buffer's header is in the arc_anon state, it won't be
6491 * linked into the hash table.
6493 if (hdr
->b_l1hdr
.b_state
== arc_anon
) {
6494 mutex_exit(&buf
->b_evict_lock
);
6495 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
6496 ASSERT(!HDR_IN_HASH_TABLE(hdr
));
6497 ASSERT(!HDR_HAS_L2HDR(hdr
));
6498 ASSERT(HDR_EMPTY(hdr
));
6500 ASSERT3U(hdr
->b_l1hdr
.b_bufcnt
, ==, 1);
6501 ASSERT3S(refcount_count(&hdr
->b_l1hdr
.b_refcnt
), ==, 1);
6502 ASSERT(!list_link_active(&hdr
->b_l1hdr
.b_arc_node
));
6504 hdr
->b_l1hdr
.b_arc_access
= 0;
6507 * If the buf is being overridden then it may already
6508 * have a hdr that is not empty.
6510 buf_discard_identity(hdr
);
6516 kmutex_t
*hash_lock
= HDR_LOCK(hdr
);
6517 mutex_enter(hash_lock
);
6520 * This assignment is only valid as long as the hash_lock is
6521 * held, we must be careful not to reference state or the
6522 * b_state field after dropping the lock.
6524 arc_state_t
*state
= hdr
->b_l1hdr
.b_state
;
6525 ASSERT3P(hash_lock
, ==, HDR_LOCK(hdr
));
6526 ASSERT3P(state
, !=, arc_anon
);
6528 /* this buffer is not on any list */
6529 ASSERT3S(refcount_count(&hdr
->b_l1hdr
.b_refcnt
), >, 0);
6531 if (HDR_HAS_L2HDR(hdr
)) {
6532 mutex_enter(&hdr
->b_l2hdr
.b_dev
->l2ad_mtx
);
6535 * We have to recheck this conditional again now that
6536 * we're holding the l2ad_mtx to prevent a race with
6537 * another thread which might be concurrently calling
6538 * l2arc_evict(). In that case, l2arc_evict() might have
6539 * destroyed the header's L2 portion as we were waiting
6540 * to acquire the l2ad_mtx.
6542 if (HDR_HAS_L2HDR(hdr
))
6543 arc_hdr_l2hdr_destroy(hdr
);
6545 mutex_exit(&hdr
->b_l2hdr
.b_dev
->l2ad_mtx
);
6549 * Do we have more than one buf?
6551 if (hdr
->b_l1hdr
.b_bufcnt
> 1) {
6552 arc_buf_hdr_t
*nhdr
;
6553 uint64_t spa
= hdr
->b_spa
;
6554 uint64_t psize
= HDR_GET_PSIZE(hdr
);
6555 uint64_t lsize
= HDR_GET_LSIZE(hdr
);
6556 boolean_t
protected = HDR_PROTECTED(hdr
);
6557 enum zio_compress compress
= arc_hdr_get_compress(hdr
);
6558 arc_buf_contents_t type
= arc_buf_type(hdr
);
6559 VERIFY3U(hdr
->b_type
, ==, type
);
6561 ASSERT(hdr
->b_l1hdr
.b_buf
!= buf
|| buf
->b_next
!= NULL
);
6562 (void) remove_reference(hdr
, hash_lock
, tag
);
6564 if (arc_buf_is_shared(buf
) && !ARC_BUF_COMPRESSED(buf
)) {
6565 ASSERT3P(hdr
->b_l1hdr
.b_buf
, !=, buf
);
6566 ASSERT(ARC_BUF_LAST(buf
));
6570 * Pull the data off of this hdr and attach it to
6571 * a new anonymous hdr. Also find the last buffer
6572 * in the hdr's buffer list.
6574 arc_buf_t
*lastbuf
= arc_buf_remove(hdr
, buf
);
6575 ASSERT3P(lastbuf
, !=, NULL
);
6578 * If the current arc_buf_t and the hdr are sharing their data
6579 * buffer, then we must stop sharing that block.
6581 if (arc_buf_is_shared(buf
)) {
6582 ASSERT3P(hdr
->b_l1hdr
.b_buf
, !=, buf
);
6583 VERIFY(!arc_buf_is_shared(lastbuf
));
6586 * First, sever the block sharing relationship between
6587 * buf and the arc_buf_hdr_t.
6589 arc_unshare_buf(hdr
, buf
);
6592 * Now we need to recreate the hdr's b_pabd. Since we
6593 * have lastbuf handy, we try to share with it, but if
6594 * we can't then we allocate a new b_pabd and copy the
6595 * data from buf into it.
6597 if (arc_can_share(hdr
, lastbuf
)) {
6598 arc_share_buf(hdr
, lastbuf
);
6600 arc_hdr_alloc_abd(hdr
, B_FALSE
);
6601 abd_copy_from_buf(hdr
->b_l1hdr
.b_pabd
,
6602 buf
->b_data
, psize
);
6604 VERIFY3P(lastbuf
->b_data
, !=, NULL
);
6605 } else if (HDR_SHARED_DATA(hdr
)) {
6607 * Uncompressed shared buffers are always at the end
6608 * of the list. Compressed buffers don't have the
6609 * same requirements. This makes it hard to
6610 * simply assert that the lastbuf is shared so
6611 * we rely on the hdr's compression flags to determine
6612 * if we have a compressed, shared buffer.
6614 ASSERT(arc_buf_is_shared(lastbuf
) ||
6615 arc_hdr_get_compress(hdr
) != ZIO_COMPRESS_OFF
);
6616 ASSERT(!ARC_BUF_SHARED(buf
));
6619 ASSERT(hdr
->b_l1hdr
.b_pabd
!= NULL
|| HDR_HAS_RABD(hdr
));
6620 ASSERT3P(state
, !=, arc_l2c_only
);
6622 (void) refcount_remove_many(&state
->arcs_size
,
6623 arc_buf_size(buf
), buf
);
6625 if (refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
)) {
6626 ASSERT3P(state
, !=, arc_l2c_only
);
6627 (void) refcount_remove_many(&state
->arcs_esize
[type
],
6628 arc_buf_size(buf
), buf
);
6631 hdr
->b_l1hdr
.b_bufcnt
-= 1;
6632 if (ARC_BUF_ENCRYPTED(buf
))
6633 hdr
->b_crypt_hdr
.b_ebufcnt
-= 1;
6635 arc_cksum_verify(buf
);
6636 arc_buf_unwatch(buf
);
6638 /* if this is the last uncompressed buf free the checksum */
6639 if (!arc_hdr_has_uncompressed_buf(hdr
))
6640 arc_cksum_free(hdr
);
6642 mutex_exit(hash_lock
);
6645 * Allocate a new hdr. The new hdr will contain a b_pabd
6646 * buffer which will be freed in arc_write().
6648 nhdr
= arc_hdr_alloc(spa
, psize
, lsize
, protected,
6649 compress
, type
, HDR_HAS_RABD(hdr
));
6650 ASSERT3P(nhdr
->b_l1hdr
.b_buf
, ==, NULL
);
6651 ASSERT0(nhdr
->b_l1hdr
.b_bufcnt
);
6652 ASSERT0(refcount_count(&nhdr
->b_l1hdr
.b_refcnt
));
6653 VERIFY3U(nhdr
->b_type
, ==, type
);
6654 ASSERT(!HDR_SHARED_DATA(nhdr
));
6656 nhdr
->b_l1hdr
.b_buf
= buf
;
6657 nhdr
->b_l1hdr
.b_bufcnt
= 1;
6658 if (ARC_BUF_ENCRYPTED(buf
))
6659 nhdr
->b_crypt_hdr
.b_ebufcnt
= 1;
6660 nhdr
->b_l1hdr
.b_mru_hits
= 0;
6661 nhdr
->b_l1hdr
.b_mru_ghost_hits
= 0;
6662 nhdr
->b_l1hdr
.b_mfu_hits
= 0;
6663 nhdr
->b_l1hdr
.b_mfu_ghost_hits
= 0;
6664 nhdr
->b_l1hdr
.b_l2_hits
= 0;
6665 (void) refcount_add(&nhdr
->b_l1hdr
.b_refcnt
, tag
);
6668 mutex_exit(&buf
->b_evict_lock
);
6669 (void) refcount_add_many(&arc_anon
->arcs_size
,
6670 HDR_GET_LSIZE(nhdr
), buf
);
6672 mutex_exit(&buf
->b_evict_lock
);
6673 ASSERT(refcount_count(&hdr
->b_l1hdr
.b_refcnt
) == 1);
6674 /* protected by hash lock, or hdr is on arc_anon */
6675 ASSERT(!multilist_link_active(&hdr
->b_l1hdr
.b_arc_node
));
6676 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
6677 hdr
->b_l1hdr
.b_mru_hits
= 0;
6678 hdr
->b_l1hdr
.b_mru_ghost_hits
= 0;
6679 hdr
->b_l1hdr
.b_mfu_hits
= 0;
6680 hdr
->b_l1hdr
.b_mfu_ghost_hits
= 0;
6681 hdr
->b_l1hdr
.b_l2_hits
= 0;
6682 arc_change_state(arc_anon
, hdr
, hash_lock
);
6683 hdr
->b_l1hdr
.b_arc_access
= 0;
6685 mutex_exit(hash_lock
);
6686 buf_discard_identity(hdr
);
6692 arc_released(arc_buf_t
*buf
)
6696 mutex_enter(&buf
->b_evict_lock
);
6697 released
= (buf
->b_data
!= NULL
&&
6698 buf
->b_hdr
->b_l1hdr
.b_state
== arc_anon
);
6699 mutex_exit(&buf
->b_evict_lock
);
6705 arc_referenced(arc_buf_t
*buf
)
6709 mutex_enter(&buf
->b_evict_lock
);
6710 referenced
= (refcount_count(&buf
->b_hdr
->b_l1hdr
.b_refcnt
));
6711 mutex_exit(&buf
->b_evict_lock
);
6712 return (referenced
);
6717 arc_write_ready(zio_t
*zio
)
6719 arc_write_callback_t
*callback
= zio
->io_private
;
6720 arc_buf_t
*buf
= callback
->awcb_buf
;
6721 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
6722 blkptr_t
*bp
= zio
->io_bp
;
6723 uint64_t psize
= BP_IS_HOLE(bp
) ? 0 : BP_GET_PSIZE(bp
);
6724 fstrans_cookie_t cookie
= spl_fstrans_mark();
6726 ASSERT(HDR_HAS_L1HDR(hdr
));
6727 ASSERT(!refcount_is_zero(&buf
->b_hdr
->b_l1hdr
.b_refcnt
));
6728 ASSERT(hdr
->b_l1hdr
.b_bufcnt
> 0);
6731 * If we're reexecuting this zio because the pool suspended, then
6732 * cleanup any state that was previously set the first time the
6733 * callback was invoked.
6735 if (zio
->io_flags
& ZIO_FLAG_REEXECUTED
) {
6736 arc_cksum_free(hdr
);
6737 arc_buf_unwatch(buf
);
6738 if (hdr
->b_l1hdr
.b_pabd
!= NULL
) {
6739 if (arc_buf_is_shared(buf
)) {
6740 arc_unshare_buf(hdr
, buf
);
6742 arc_hdr_free_abd(hdr
, B_FALSE
);
6746 if (HDR_HAS_RABD(hdr
))
6747 arc_hdr_free_abd(hdr
, B_TRUE
);
6749 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, ==, NULL
);
6750 ASSERT(!HDR_HAS_RABD(hdr
));
6751 ASSERT(!HDR_SHARED_DATA(hdr
));
6752 ASSERT(!arc_buf_is_shared(buf
));
6754 callback
->awcb_ready(zio
, buf
, callback
->awcb_private
);
6756 if (HDR_IO_IN_PROGRESS(hdr
))
6757 ASSERT(zio
->io_flags
& ZIO_FLAG_REEXECUTED
);
6759 arc_hdr_set_flags(hdr
, ARC_FLAG_IO_IN_PROGRESS
);
6761 if (BP_IS_PROTECTED(bp
) != !!HDR_PROTECTED(hdr
))
6762 hdr
= arc_hdr_realloc_crypt(hdr
, BP_IS_PROTECTED(bp
));
6764 if (BP_IS_PROTECTED(bp
)) {
6765 /* ZIL blocks are written through zio_rewrite */
6766 ASSERT3U(BP_GET_TYPE(bp
), !=, DMU_OT_INTENT_LOG
);
6767 ASSERT(HDR_PROTECTED(hdr
));
6769 if (BP_SHOULD_BYTESWAP(bp
)) {
6770 if (BP_GET_LEVEL(bp
) > 0) {
6771 hdr
->b_l1hdr
.b_byteswap
= DMU_BSWAP_UINT64
;
6773 hdr
->b_l1hdr
.b_byteswap
=
6774 DMU_OT_BYTESWAP(BP_GET_TYPE(bp
));
6777 hdr
->b_l1hdr
.b_byteswap
= DMU_BSWAP_NUMFUNCS
;
6780 hdr
->b_crypt_hdr
.b_ot
= BP_GET_TYPE(bp
);
6781 hdr
->b_crypt_hdr
.b_dsobj
= zio
->io_bookmark
.zb_objset
;
6782 zio_crypt_decode_params_bp(bp
, hdr
->b_crypt_hdr
.b_salt
,
6783 hdr
->b_crypt_hdr
.b_iv
);
6784 zio_crypt_decode_mac_bp(bp
, hdr
->b_crypt_hdr
.b_mac
);
6788 * If this block was written for raw encryption but the zio layer
6789 * ended up only authenticating it, adjust the buffer flags now.
6791 if (BP_IS_AUTHENTICATED(bp
) && ARC_BUF_ENCRYPTED(buf
)) {
6792 arc_hdr_set_flags(hdr
, ARC_FLAG_NOAUTH
);
6793 buf
->b_flags
&= ~ARC_BUF_FLAG_ENCRYPTED
;
6794 if (BP_GET_COMPRESS(bp
) == ZIO_COMPRESS_OFF
)
6795 buf
->b_flags
&= ~ARC_BUF_FLAG_COMPRESSED
;
6796 } else if (BP_IS_HOLE(bp
) && ARC_BUF_ENCRYPTED(buf
)) {
6797 buf
->b_flags
&= ~ARC_BUF_FLAG_ENCRYPTED
;
6798 buf
->b_flags
&= ~ARC_BUF_FLAG_COMPRESSED
;
6801 /* this must be done after the buffer flags are adjusted */
6802 arc_cksum_compute(buf
);
6804 enum zio_compress compress
;
6805 if (BP_IS_HOLE(bp
) || BP_IS_EMBEDDED(bp
)) {
6806 compress
= ZIO_COMPRESS_OFF
;
6808 ASSERT3U(HDR_GET_LSIZE(hdr
), ==, BP_GET_LSIZE(bp
));
6809 compress
= BP_GET_COMPRESS(bp
);
6811 HDR_SET_PSIZE(hdr
, psize
);
6812 arc_hdr_set_compress(hdr
, compress
);
6814 if (zio
->io_error
!= 0 || psize
== 0)
6818 * Fill the hdr with data. If the buffer is encrypted we have no choice
6819 * but to copy the data into b_radb. If the hdr is compressed, the data
6820 * we want is available from the zio, otherwise we can take it from
6823 * We might be able to share the buf's data with the hdr here. However,
6824 * doing so would cause the ARC to be full of linear ABDs if we write a
6825 * lot of shareable data. As a compromise, we check whether scattered
6826 * ABDs are allowed, and assume that if they are then the user wants
6827 * the ARC to be primarily filled with them regardless of the data being
6828 * written. Therefore, if they're allowed then we allocate one and copy
6829 * the data into it; otherwise, we share the data directly if we can.
6831 if (ARC_BUF_ENCRYPTED(buf
)) {
6832 ASSERT3U(psize
, >, 0);
6833 ASSERT(ARC_BUF_COMPRESSED(buf
));
6834 arc_hdr_alloc_abd(hdr
, B_TRUE
);
6835 abd_copy(hdr
->b_crypt_hdr
.b_rabd
, zio
->io_abd
, psize
);
6836 } else if (zfs_abd_scatter_enabled
|| !arc_can_share(hdr
, buf
)) {
6838 * Ideally, we would always copy the io_abd into b_pabd, but the
6839 * user may have disabled compressed ARC, thus we must check the
6840 * hdr's compression setting rather than the io_bp's.
6842 if (BP_IS_ENCRYPTED(bp
)) {
6843 ASSERT3U(psize
, >, 0);
6844 arc_hdr_alloc_abd(hdr
, B_TRUE
);
6845 abd_copy(hdr
->b_crypt_hdr
.b_rabd
, zio
->io_abd
, psize
);
6846 } else if (arc_hdr_get_compress(hdr
) != ZIO_COMPRESS_OFF
&&
6847 !ARC_BUF_COMPRESSED(buf
)) {
6848 ASSERT3U(psize
, >, 0);
6849 arc_hdr_alloc_abd(hdr
, B_FALSE
);
6850 abd_copy(hdr
->b_l1hdr
.b_pabd
, zio
->io_abd
, psize
);
6852 ASSERT3U(zio
->io_orig_size
, ==, arc_hdr_size(hdr
));
6853 arc_hdr_alloc_abd(hdr
, B_FALSE
);
6854 abd_copy_from_buf(hdr
->b_l1hdr
.b_pabd
, buf
->b_data
,
6858 ASSERT3P(buf
->b_data
, ==, abd_to_buf(zio
->io_orig_abd
));
6859 ASSERT3U(zio
->io_orig_size
, ==, arc_buf_size(buf
));
6860 ASSERT3U(hdr
->b_l1hdr
.b_bufcnt
, ==, 1);
6862 arc_share_buf(hdr
, buf
);
6866 arc_hdr_verify(hdr
, bp
);
6867 spl_fstrans_unmark(cookie
);
6871 arc_write_children_ready(zio_t
*zio
)
6873 arc_write_callback_t
*callback
= zio
->io_private
;
6874 arc_buf_t
*buf
= callback
->awcb_buf
;
6876 callback
->awcb_children_ready(zio
, buf
, callback
->awcb_private
);
6880 * The SPA calls this callback for each physical write that happens on behalf
6881 * of a logical write. See the comment in dbuf_write_physdone() for details.
6884 arc_write_physdone(zio_t
*zio
)
6886 arc_write_callback_t
*cb
= zio
->io_private
;
6887 if (cb
->awcb_physdone
!= NULL
)
6888 cb
->awcb_physdone(zio
, cb
->awcb_buf
, cb
->awcb_private
);
6892 arc_write_done(zio_t
*zio
)
6894 arc_write_callback_t
*callback
= zio
->io_private
;
6895 arc_buf_t
*buf
= callback
->awcb_buf
;
6896 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
6898 ASSERT3P(hdr
->b_l1hdr
.b_acb
, ==, NULL
);
6900 if (zio
->io_error
== 0) {
6901 arc_hdr_verify(hdr
, zio
->io_bp
);
6903 if (BP_IS_HOLE(zio
->io_bp
) || BP_IS_EMBEDDED(zio
->io_bp
)) {
6904 buf_discard_identity(hdr
);
6906 hdr
->b_dva
= *BP_IDENTITY(zio
->io_bp
);
6907 hdr
->b_birth
= BP_PHYSICAL_BIRTH(zio
->io_bp
);
6910 ASSERT(HDR_EMPTY(hdr
));
6914 * If the block to be written was all-zero or compressed enough to be
6915 * embedded in the BP, no write was performed so there will be no
6916 * dva/birth/checksum. The buffer must therefore remain anonymous
6919 if (!HDR_EMPTY(hdr
)) {
6920 arc_buf_hdr_t
*exists
;
6921 kmutex_t
*hash_lock
;
6923 ASSERT3U(zio
->io_error
, ==, 0);
6925 arc_cksum_verify(buf
);
6927 exists
= buf_hash_insert(hdr
, &hash_lock
);
6928 if (exists
!= NULL
) {
6930 * This can only happen if we overwrite for
6931 * sync-to-convergence, because we remove
6932 * buffers from the hash table when we arc_free().
6934 if (zio
->io_flags
& ZIO_FLAG_IO_REWRITE
) {
6935 if (!BP_EQUAL(&zio
->io_bp_orig
, zio
->io_bp
))
6936 panic("bad overwrite, hdr=%p exists=%p",
6937 (void *)hdr
, (void *)exists
);
6938 ASSERT(refcount_is_zero(
6939 &exists
->b_l1hdr
.b_refcnt
));
6940 arc_change_state(arc_anon
, exists
, hash_lock
);
6941 mutex_exit(hash_lock
);
6942 arc_hdr_destroy(exists
);
6943 exists
= buf_hash_insert(hdr
, &hash_lock
);
6944 ASSERT3P(exists
, ==, NULL
);
6945 } else if (zio
->io_flags
& ZIO_FLAG_NOPWRITE
) {
6947 ASSERT(zio
->io_prop
.zp_nopwrite
);
6948 if (!BP_EQUAL(&zio
->io_bp_orig
, zio
->io_bp
))
6949 panic("bad nopwrite, hdr=%p exists=%p",
6950 (void *)hdr
, (void *)exists
);
6953 ASSERT(hdr
->b_l1hdr
.b_bufcnt
== 1);
6954 ASSERT(hdr
->b_l1hdr
.b_state
== arc_anon
);
6955 ASSERT(BP_GET_DEDUP(zio
->io_bp
));
6956 ASSERT(BP_GET_LEVEL(zio
->io_bp
) == 0);
6959 arc_hdr_clear_flags(hdr
, ARC_FLAG_IO_IN_PROGRESS
);
6960 /* if it's not anon, we are doing a scrub */
6961 if (exists
== NULL
&& hdr
->b_l1hdr
.b_state
== arc_anon
)
6962 arc_access(hdr
, hash_lock
);
6963 mutex_exit(hash_lock
);
6965 arc_hdr_clear_flags(hdr
, ARC_FLAG_IO_IN_PROGRESS
);
6968 ASSERT(!refcount_is_zero(&hdr
->b_l1hdr
.b_refcnt
));
6969 callback
->awcb_done(zio
, buf
, callback
->awcb_private
);
6971 abd_put(zio
->io_abd
);
6972 kmem_free(callback
, sizeof (arc_write_callback_t
));
6976 arc_write(zio_t
*pio
, spa_t
*spa
, uint64_t txg
,
6977 blkptr_t
*bp
, arc_buf_t
*buf
, boolean_t l2arc
,
6978 const zio_prop_t
*zp
, arc_write_done_func_t
*ready
,
6979 arc_write_done_func_t
*children_ready
, arc_write_done_func_t
*physdone
,
6980 arc_write_done_func_t
*done
, void *private, zio_priority_t priority
,
6981 int zio_flags
, const zbookmark_phys_t
*zb
)
6983 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
6984 arc_write_callback_t
*callback
;
6986 zio_prop_t localprop
= *zp
;
6988 ASSERT3P(ready
, !=, NULL
);
6989 ASSERT3P(done
, !=, NULL
);
6990 ASSERT(!HDR_IO_ERROR(hdr
));
6991 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
6992 ASSERT3P(hdr
->b_l1hdr
.b_acb
, ==, NULL
);
6993 ASSERT3U(hdr
->b_l1hdr
.b_bufcnt
, >, 0);
6995 arc_hdr_set_flags(hdr
, ARC_FLAG_L2CACHE
);
6997 if (ARC_BUF_ENCRYPTED(buf
)) {
6998 ASSERT(ARC_BUF_COMPRESSED(buf
));
6999 localprop
.zp_encrypt
= B_TRUE
;
7000 localprop
.zp_compress
= HDR_GET_COMPRESS(hdr
);
7001 localprop
.zp_byteorder
=
7002 (hdr
->b_l1hdr
.b_byteswap
== DMU_BSWAP_NUMFUNCS
) ?
7003 ZFS_HOST_BYTEORDER
: !ZFS_HOST_BYTEORDER
;
7004 bcopy(hdr
->b_crypt_hdr
.b_salt
, localprop
.zp_salt
,
7006 bcopy(hdr
->b_crypt_hdr
.b_iv
, localprop
.zp_iv
,
7008 bcopy(hdr
->b_crypt_hdr
.b_mac
, localprop
.zp_mac
,
7010 if (DMU_OT_IS_ENCRYPTED(localprop
.zp_type
)) {
7011 localprop
.zp_nopwrite
= B_FALSE
;
7012 localprop
.zp_copies
=
7013 MIN(localprop
.zp_copies
, SPA_DVAS_PER_BP
- 1);
7015 zio_flags
|= ZIO_FLAG_RAW
;
7016 } else if (ARC_BUF_COMPRESSED(buf
)) {
7017 ASSERT3U(HDR_GET_LSIZE(hdr
), !=, arc_buf_size(buf
));
7018 localprop
.zp_compress
= HDR_GET_COMPRESS(hdr
);
7019 zio_flags
|= ZIO_FLAG_RAW_COMPRESS
;
7021 callback
= kmem_zalloc(sizeof (arc_write_callback_t
), KM_SLEEP
);
7022 callback
->awcb_ready
= ready
;
7023 callback
->awcb_children_ready
= children_ready
;
7024 callback
->awcb_physdone
= physdone
;
7025 callback
->awcb_done
= done
;
7026 callback
->awcb_private
= private;
7027 callback
->awcb_buf
= buf
;
7030 * The hdr's b_pabd is now stale, free it now. A new data block
7031 * will be allocated when the zio pipeline calls arc_write_ready().
7033 if (hdr
->b_l1hdr
.b_pabd
!= NULL
) {
7035 * If the buf is currently sharing the data block with
7036 * the hdr then we need to break that relationship here.
7037 * The hdr will remain with a NULL data pointer and the
7038 * buf will take sole ownership of the block.
7040 if (arc_buf_is_shared(buf
)) {
7041 arc_unshare_buf(hdr
, buf
);
7043 arc_hdr_free_abd(hdr
, B_FALSE
);
7045 VERIFY3P(buf
->b_data
, !=, NULL
);
7048 if (HDR_HAS_RABD(hdr
))
7049 arc_hdr_free_abd(hdr
, B_TRUE
);
7051 if (!(zio_flags
& ZIO_FLAG_RAW
))
7052 arc_hdr_set_compress(hdr
, ZIO_COMPRESS_OFF
);
7054 ASSERT(!arc_buf_is_shared(buf
));
7055 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, ==, NULL
);
7057 zio
= zio_write(pio
, spa
, txg
, bp
,
7058 abd_get_from_buf(buf
->b_data
, HDR_GET_LSIZE(hdr
)),
7059 HDR_GET_LSIZE(hdr
), arc_buf_size(buf
), &localprop
, arc_write_ready
,
7060 (children_ready
!= NULL
) ? arc_write_children_ready
: NULL
,
7061 arc_write_physdone
, arc_write_done
, callback
,
7062 priority
, zio_flags
, zb
);
7068 arc_memory_throttle(uint64_t reserve
, uint64_t txg
)
7071 uint64_t available_memory
= arc_free_memory();
7072 static uint64_t page_load
= 0;
7073 static uint64_t last_txg
= 0;
7077 MIN(available_memory
, vmem_size(heap_arena
, VMEM_FREE
));
7080 if (available_memory
> arc_all_memory() * arc_lotsfree_percent
/ 100)
7083 if (txg
> last_txg
) {
7088 * If we are in pageout, we know that memory is already tight,
7089 * the arc is already going to be evicting, so we just want to
7090 * continue to let page writes occur as quickly as possible.
7092 if (current_is_kswapd()) {
7093 if (page_load
> MAX(arc_sys_free
/ 4, available_memory
) / 4) {
7094 DMU_TX_STAT_BUMP(dmu_tx_memory_reclaim
);
7095 return (SET_ERROR(ERESTART
));
7097 /* Note: reserve is inflated, so we deflate */
7098 page_load
+= reserve
/ 8;
7100 } else if (page_load
> 0 && arc_reclaim_needed()) {
7101 /* memory is low, delay before restarting */
7102 ARCSTAT_INCR(arcstat_memory_throttle_count
, 1);
7103 DMU_TX_STAT_BUMP(dmu_tx_memory_reclaim
);
7104 return (SET_ERROR(EAGAIN
));
7112 arc_tempreserve_clear(uint64_t reserve
)
7114 atomic_add_64(&arc_tempreserve
, -reserve
);
7115 ASSERT((int64_t)arc_tempreserve
>= 0);
7119 arc_tempreserve_space(uint64_t reserve
, uint64_t txg
)
7125 reserve
> arc_c
/4 &&
7126 reserve
* 4 > (2ULL << SPA_MAXBLOCKSHIFT
))
7127 arc_c
= MIN(arc_c_max
, reserve
* 4);
7130 * Throttle when the calculated memory footprint for the TXG
7131 * exceeds the target ARC size.
7133 if (reserve
> arc_c
) {
7134 DMU_TX_STAT_BUMP(dmu_tx_memory_reserve
);
7135 return (SET_ERROR(ERESTART
));
7139 * Don't count loaned bufs as in flight dirty data to prevent long
7140 * network delays from blocking transactions that are ready to be
7141 * assigned to a txg.
7144 /* assert that it has not wrapped around */
7145 ASSERT3S(atomic_add_64_nv(&arc_loaned_bytes
, 0), >=, 0);
7147 anon_size
= MAX((int64_t)(refcount_count(&arc_anon
->arcs_size
) -
7148 arc_loaned_bytes
), 0);
7151 * Writes will, almost always, require additional memory allocations
7152 * in order to compress/encrypt/etc the data. We therefore need to
7153 * make sure that there is sufficient available memory for this.
7155 error
= arc_memory_throttle(reserve
, txg
);
7160 * Throttle writes when the amount of dirty data in the cache
7161 * gets too large. We try to keep the cache less than half full
7162 * of dirty blocks so that our sync times don't grow too large.
7163 * Note: if two requests come in concurrently, we might let them
7164 * both succeed, when one of them should fail. Not a huge deal.
7167 if (reserve
+ arc_tempreserve
+ anon_size
> arc_c
/ 2 &&
7168 anon_size
> arc_c
/ 4) {
7170 uint64_t meta_esize
=
7171 refcount_count(&arc_anon
->arcs_esize
[ARC_BUFC_METADATA
]);
7172 uint64_t data_esize
=
7173 refcount_count(&arc_anon
->arcs_esize
[ARC_BUFC_DATA
]);
7174 dprintf("failing, arc_tempreserve=%lluK anon_meta=%lluK "
7175 "anon_data=%lluK tempreserve=%lluK arc_c=%lluK\n",
7176 arc_tempreserve
>> 10, meta_esize
>> 10,
7177 data_esize
>> 10, reserve
>> 10, arc_c
>> 10);
7179 DMU_TX_STAT_BUMP(dmu_tx_dirty_throttle
);
7180 return (SET_ERROR(ERESTART
));
7182 atomic_add_64(&arc_tempreserve
, reserve
);
7187 arc_kstat_update_state(arc_state_t
*state
, kstat_named_t
*size
,
7188 kstat_named_t
*evict_data
, kstat_named_t
*evict_metadata
)
7190 size
->value
.ui64
= refcount_count(&state
->arcs_size
);
7191 evict_data
->value
.ui64
=
7192 refcount_count(&state
->arcs_esize
[ARC_BUFC_DATA
]);
7193 evict_metadata
->value
.ui64
=
7194 refcount_count(&state
->arcs_esize
[ARC_BUFC_METADATA
]);
7198 arc_kstat_update(kstat_t
*ksp
, int rw
)
7200 arc_stats_t
*as
= ksp
->ks_data
;
7202 if (rw
== KSTAT_WRITE
) {
7203 return (SET_ERROR(EACCES
));
7205 arc_kstat_update_state(arc_anon
,
7206 &as
->arcstat_anon_size
,
7207 &as
->arcstat_anon_evictable_data
,
7208 &as
->arcstat_anon_evictable_metadata
);
7209 arc_kstat_update_state(arc_mru
,
7210 &as
->arcstat_mru_size
,
7211 &as
->arcstat_mru_evictable_data
,
7212 &as
->arcstat_mru_evictable_metadata
);
7213 arc_kstat_update_state(arc_mru_ghost
,
7214 &as
->arcstat_mru_ghost_size
,
7215 &as
->arcstat_mru_ghost_evictable_data
,
7216 &as
->arcstat_mru_ghost_evictable_metadata
);
7217 arc_kstat_update_state(arc_mfu
,
7218 &as
->arcstat_mfu_size
,
7219 &as
->arcstat_mfu_evictable_data
,
7220 &as
->arcstat_mfu_evictable_metadata
);
7221 arc_kstat_update_state(arc_mfu_ghost
,
7222 &as
->arcstat_mfu_ghost_size
,
7223 &as
->arcstat_mfu_ghost_evictable_data
,
7224 &as
->arcstat_mfu_ghost_evictable_metadata
);
7226 as
->arcstat_memory_all_bytes
.value
.ui64
=
7228 as
->arcstat_memory_free_bytes
.value
.ui64
=
7230 as
->arcstat_memory_available_bytes
.value
.i64
=
7231 arc_available_memory();
7238 * This function *must* return indices evenly distributed between all
7239 * sublists of the multilist. This is needed due to how the ARC eviction
7240 * code is laid out; arc_evict_state() assumes ARC buffers are evenly
7241 * distributed between all sublists and uses this assumption when
7242 * deciding which sublist to evict from and how much to evict from it.
7245 arc_state_multilist_index_func(multilist_t
*ml
, void *obj
)
7247 arc_buf_hdr_t
*hdr
= obj
;
7250 * We rely on b_dva to generate evenly distributed index
7251 * numbers using buf_hash below. So, as an added precaution,
7252 * let's make sure we never add empty buffers to the arc lists.
7254 ASSERT(!HDR_EMPTY(hdr
));
7257 * The assumption here, is the hash value for a given
7258 * arc_buf_hdr_t will remain constant throughout its lifetime
7259 * (i.e. its b_spa, b_dva, and b_birth fields don't change).
7260 * Thus, we don't need to store the header's sublist index
7261 * on insertion, as this index can be recalculated on removal.
7263 * Also, the low order bits of the hash value are thought to be
7264 * distributed evenly. Otherwise, in the case that the multilist
7265 * has a power of two number of sublists, each sublists' usage
7266 * would not be evenly distributed.
7268 return (buf_hash(hdr
->b_spa
, &hdr
->b_dva
, hdr
->b_birth
) %
7269 multilist_get_num_sublists(ml
));
7273 * Called during module initialization and periodically thereafter to
7274 * apply reasonable changes to the exposed performance tunings. Non-zero
7275 * zfs_* values which differ from the currently set values will be applied.
7278 arc_tuning_update(void)
7280 uint64_t allmem
= arc_all_memory();
7281 unsigned long limit
;
7283 /* Valid range: 64M - <all physical memory> */
7284 if ((zfs_arc_max
) && (zfs_arc_max
!= arc_c_max
) &&
7285 (zfs_arc_max
>= 64 << 20) && (zfs_arc_max
< allmem
) &&
7286 (zfs_arc_max
> arc_c_min
)) {
7287 arc_c_max
= zfs_arc_max
;
7289 arc_p
= (arc_c
>> 1);
7290 if (arc_meta_limit
> arc_c_max
)
7291 arc_meta_limit
= arc_c_max
;
7292 if (arc_dnode_limit
> arc_meta_limit
)
7293 arc_dnode_limit
= arc_meta_limit
;
7296 /* Valid range: 32M - <arc_c_max> */
7297 if ((zfs_arc_min
) && (zfs_arc_min
!= arc_c_min
) &&
7298 (zfs_arc_min
>= 2ULL << SPA_MAXBLOCKSHIFT
) &&
7299 (zfs_arc_min
<= arc_c_max
)) {
7300 arc_c_min
= zfs_arc_min
;
7301 arc_c
= MAX(arc_c
, arc_c_min
);
7304 /* Valid range: 16M - <arc_c_max> */
7305 if ((zfs_arc_meta_min
) && (zfs_arc_meta_min
!= arc_meta_min
) &&
7306 (zfs_arc_meta_min
>= 1ULL << SPA_MAXBLOCKSHIFT
) &&
7307 (zfs_arc_meta_min
<= arc_c_max
)) {
7308 arc_meta_min
= zfs_arc_meta_min
;
7309 if (arc_meta_limit
< arc_meta_min
)
7310 arc_meta_limit
= arc_meta_min
;
7311 if (arc_dnode_limit
< arc_meta_min
)
7312 arc_dnode_limit
= arc_meta_min
;
7315 /* Valid range: <arc_meta_min> - <arc_c_max> */
7316 limit
= zfs_arc_meta_limit
? zfs_arc_meta_limit
:
7317 MIN(zfs_arc_meta_limit_percent
, 100) * arc_c_max
/ 100;
7318 if ((limit
!= arc_meta_limit
) &&
7319 (limit
>= arc_meta_min
) &&
7320 (limit
<= arc_c_max
))
7321 arc_meta_limit
= limit
;
7323 /* Valid range: <arc_meta_min> - <arc_meta_limit> */
7324 limit
= zfs_arc_dnode_limit
? zfs_arc_dnode_limit
:
7325 MIN(zfs_arc_dnode_limit_percent
, 100) * arc_meta_limit
/ 100;
7326 if ((limit
!= arc_dnode_limit
) &&
7327 (limit
>= arc_meta_min
) &&
7328 (limit
<= arc_meta_limit
))
7329 arc_dnode_limit
= limit
;
7331 /* Valid range: 1 - N */
7332 if (zfs_arc_grow_retry
)
7333 arc_grow_retry
= zfs_arc_grow_retry
;
7335 /* Valid range: 1 - N */
7336 if (zfs_arc_shrink_shift
) {
7337 arc_shrink_shift
= zfs_arc_shrink_shift
;
7338 arc_no_grow_shift
= MIN(arc_no_grow_shift
, arc_shrink_shift
-1);
7341 /* Valid range: 1 - N */
7342 if (zfs_arc_p_min_shift
)
7343 arc_p_min_shift
= zfs_arc_p_min_shift
;
7345 /* Valid range: 1 - N ms */
7346 if (zfs_arc_min_prefetch_ms
)
7347 arc_min_prefetch_ms
= zfs_arc_min_prefetch_ms
;
7349 /* Valid range: 1 - N ms */
7350 if (zfs_arc_min_prescient_prefetch_ms
) {
7351 arc_min_prescient_prefetch_ms
=
7352 zfs_arc_min_prescient_prefetch_ms
;
7355 /* Valid range: 0 - 100 */
7356 if ((zfs_arc_lotsfree_percent
>= 0) &&
7357 (zfs_arc_lotsfree_percent
<= 100))
7358 arc_lotsfree_percent
= zfs_arc_lotsfree_percent
;
7360 /* Valid range: 0 - <all physical memory> */
7361 if ((zfs_arc_sys_free
) && (zfs_arc_sys_free
!= arc_sys_free
))
7362 arc_sys_free
= MIN(MAX(zfs_arc_sys_free
, 0), allmem
);
7367 arc_state_init(void)
7369 arc_anon
= &ARC_anon
;
7371 arc_mru_ghost
= &ARC_mru_ghost
;
7373 arc_mfu_ghost
= &ARC_mfu_ghost
;
7374 arc_l2c_only
= &ARC_l2c_only
;
7376 arc_mru
->arcs_list
[ARC_BUFC_METADATA
] =
7377 multilist_create(sizeof (arc_buf_hdr_t
),
7378 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
7379 arc_state_multilist_index_func
);
7380 arc_mru
->arcs_list
[ARC_BUFC_DATA
] =
7381 multilist_create(sizeof (arc_buf_hdr_t
),
7382 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
7383 arc_state_multilist_index_func
);
7384 arc_mru_ghost
->arcs_list
[ARC_BUFC_METADATA
] =
7385 multilist_create(sizeof (arc_buf_hdr_t
),
7386 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
7387 arc_state_multilist_index_func
);
7388 arc_mru_ghost
->arcs_list
[ARC_BUFC_DATA
] =
7389 multilist_create(sizeof (arc_buf_hdr_t
),
7390 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
7391 arc_state_multilist_index_func
);
7392 arc_mfu
->arcs_list
[ARC_BUFC_METADATA
] =
7393 multilist_create(sizeof (arc_buf_hdr_t
),
7394 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
7395 arc_state_multilist_index_func
);
7396 arc_mfu
->arcs_list
[ARC_BUFC_DATA
] =
7397 multilist_create(sizeof (arc_buf_hdr_t
),
7398 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
7399 arc_state_multilist_index_func
);
7400 arc_mfu_ghost
->arcs_list
[ARC_BUFC_METADATA
] =
7401 multilist_create(sizeof (arc_buf_hdr_t
),
7402 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
7403 arc_state_multilist_index_func
);
7404 arc_mfu_ghost
->arcs_list
[ARC_BUFC_DATA
] =
7405 multilist_create(sizeof (arc_buf_hdr_t
),
7406 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
7407 arc_state_multilist_index_func
);
7408 arc_l2c_only
->arcs_list
[ARC_BUFC_METADATA
] =
7409 multilist_create(sizeof (arc_buf_hdr_t
),
7410 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
7411 arc_state_multilist_index_func
);
7412 arc_l2c_only
->arcs_list
[ARC_BUFC_DATA
] =
7413 multilist_create(sizeof (arc_buf_hdr_t
),
7414 offsetof(arc_buf_hdr_t
, b_l1hdr
.b_arc_node
),
7415 arc_state_multilist_index_func
);
7417 refcount_create(&arc_anon
->arcs_esize
[ARC_BUFC_METADATA
]);
7418 refcount_create(&arc_anon
->arcs_esize
[ARC_BUFC_DATA
]);
7419 refcount_create(&arc_mru
->arcs_esize
[ARC_BUFC_METADATA
]);
7420 refcount_create(&arc_mru
->arcs_esize
[ARC_BUFC_DATA
]);
7421 refcount_create(&arc_mru_ghost
->arcs_esize
[ARC_BUFC_METADATA
]);
7422 refcount_create(&arc_mru_ghost
->arcs_esize
[ARC_BUFC_DATA
]);
7423 refcount_create(&arc_mfu
->arcs_esize
[ARC_BUFC_METADATA
]);
7424 refcount_create(&arc_mfu
->arcs_esize
[ARC_BUFC_DATA
]);
7425 refcount_create(&arc_mfu_ghost
->arcs_esize
[ARC_BUFC_METADATA
]);
7426 refcount_create(&arc_mfu_ghost
->arcs_esize
[ARC_BUFC_DATA
]);
7427 refcount_create(&arc_l2c_only
->arcs_esize
[ARC_BUFC_METADATA
]);
7428 refcount_create(&arc_l2c_only
->arcs_esize
[ARC_BUFC_DATA
]);
7430 refcount_create(&arc_anon
->arcs_size
);
7431 refcount_create(&arc_mru
->arcs_size
);
7432 refcount_create(&arc_mru_ghost
->arcs_size
);
7433 refcount_create(&arc_mfu
->arcs_size
);
7434 refcount_create(&arc_mfu_ghost
->arcs_size
);
7435 refcount_create(&arc_l2c_only
->arcs_size
);
7437 arc_anon
->arcs_state
= ARC_STATE_ANON
;
7438 arc_mru
->arcs_state
= ARC_STATE_MRU
;
7439 arc_mru_ghost
->arcs_state
= ARC_STATE_MRU_GHOST
;
7440 arc_mfu
->arcs_state
= ARC_STATE_MFU
;
7441 arc_mfu_ghost
->arcs_state
= ARC_STATE_MFU_GHOST
;
7442 arc_l2c_only
->arcs_state
= ARC_STATE_L2C_ONLY
;
7446 arc_state_fini(void)
7448 refcount_destroy(&arc_anon
->arcs_esize
[ARC_BUFC_METADATA
]);
7449 refcount_destroy(&arc_anon
->arcs_esize
[ARC_BUFC_DATA
]);
7450 refcount_destroy(&arc_mru
->arcs_esize
[ARC_BUFC_METADATA
]);
7451 refcount_destroy(&arc_mru
->arcs_esize
[ARC_BUFC_DATA
]);
7452 refcount_destroy(&arc_mru_ghost
->arcs_esize
[ARC_BUFC_METADATA
]);
7453 refcount_destroy(&arc_mru_ghost
->arcs_esize
[ARC_BUFC_DATA
]);
7454 refcount_destroy(&arc_mfu
->arcs_esize
[ARC_BUFC_METADATA
]);
7455 refcount_destroy(&arc_mfu
->arcs_esize
[ARC_BUFC_DATA
]);
7456 refcount_destroy(&arc_mfu_ghost
->arcs_esize
[ARC_BUFC_METADATA
]);
7457 refcount_destroy(&arc_mfu_ghost
->arcs_esize
[ARC_BUFC_DATA
]);
7458 refcount_destroy(&arc_l2c_only
->arcs_esize
[ARC_BUFC_METADATA
]);
7459 refcount_destroy(&arc_l2c_only
->arcs_esize
[ARC_BUFC_DATA
]);
7461 refcount_destroy(&arc_anon
->arcs_size
);
7462 refcount_destroy(&arc_mru
->arcs_size
);
7463 refcount_destroy(&arc_mru_ghost
->arcs_size
);
7464 refcount_destroy(&arc_mfu
->arcs_size
);
7465 refcount_destroy(&arc_mfu_ghost
->arcs_size
);
7466 refcount_destroy(&arc_l2c_only
->arcs_size
);
7468 multilist_destroy(arc_mru
->arcs_list
[ARC_BUFC_METADATA
]);
7469 multilist_destroy(arc_mru_ghost
->arcs_list
[ARC_BUFC_METADATA
]);
7470 multilist_destroy(arc_mfu
->arcs_list
[ARC_BUFC_METADATA
]);
7471 multilist_destroy(arc_mfu_ghost
->arcs_list
[ARC_BUFC_METADATA
]);
7472 multilist_destroy(arc_mru
->arcs_list
[ARC_BUFC_DATA
]);
7473 multilist_destroy(arc_mru_ghost
->arcs_list
[ARC_BUFC_DATA
]);
7474 multilist_destroy(arc_mfu
->arcs_list
[ARC_BUFC_DATA
]);
7475 multilist_destroy(arc_mfu_ghost
->arcs_list
[ARC_BUFC_DATA
]);
7476 multilist_destroy(arc_l2c_only
->arcs_list
[ARC_BUFC_METADATA
]);
7477 multilist_destroy(arc_l2c_only
->arcs_list
[ARC_BUFC_DATA
]);
7481 arc_target_bytes(void)
7489 uint64_t percent
, allmem
= arc_all_memory();
7491 mutex_init(&arc_reclaim_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
7492 cv_init(&arc_reclaim_thread_cv
, NULL
, CV_DEFAULT
, NULL
);
7493 cv_init(&arc_reclaim_waiters_cv
, NULL
, CV_DEFAULT
, NULL
);
7495 arc_min_prefetch_ms
= 1000;
7496 arc_min_prescient_prefetch_ms
= 6000;
7500 * Register a shrinker to support synchronous (direct) memory
7501 * reclaim from the arc. This is done to prevent kswapd from
7502 * swapping out pages when it is preferable to shrink the arc.
7504 spl_register_shrinker(&arc_shrinker
);
7506 /* Set to 1/64 of all memory or a minimum of 512K */
7507 arc_sys_free
= MAX(allmem
/ 64, (512 * 1024));
7511 /* Set max to 1/2 of all memory */
7512 arc_c_max
= allmem
/ 2;
7515 /* Set min cache to 1/32 of all memory, or 32MB, whichever is more */
7516 arc_c_min
= MAX(allmem
/ 32, 2ULL << SPA_MAXBLOCKSHIFT
);
7519 * In userland, there's only the memory pressure that we artificially
7520 * create (see arc_available_memory()). Don't let arc_c get too
7521 * small, because it can cause transactions to be larger than
7522 * arc_c, causing arc_tempreserve_space() to fail.
7524 arc_c_min
= MAX(arc_c_max
/ 2, 2ULL << SPA_MAXBLOCKSHIFT
);
7528 arc_p
= (arc_c
>> 1);
7531 /* Set min to 1/2 of arc_c_min */
7532 arc_meta_min
= 1ULL << SPA_MAXBLOCKSHIFT
;
7533 /* Initialize maximum observed usage to zero */
7536 * Set arc_meta_limit to a percent of arc_c_max with a floor of
7537 * arc_meta_min, and a ceiling of arc_c_max.
7539 percent
= MIN(zfs_arc_meta_limit_percent
, 100);
7540 arc_meta_limit
= MAX(arc_meta_min
, (percent
* arc_c_max
) / 100);
7541 percent
= MIN(zfs_arc_dnode_limit_percent
, 100);
7542 arc_dnode_limit
= (percent
* arc_meta_limit
) / 100;
7544 /* Apply user specified tunings */
7545 arc_tuning_update();
7547 /* if kmem_flags are set, lets try to use less memory */
7548 if (kmem_debugging())
7550 if (arc_c
< arc_c_min
)
7556 list_create(&arc_prune_list
, sizeof (arc_prune_t
),
7557 offsetof(arc_prune_t
, p_node
));
7558 mutex_init(&arc_prune_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
7560 arc_prune_taskq
= taskq_create("arc_prune", max_ncpus
, defclsyspri
,
7561 max_ncpus
, INT_MAX
, TASKQ_PREPOPULATE
| TASKQ_DYNAMIC
);
7563 arc_reclaim_thread_exit
= B_FALSE
;
7565 arc_ksp
= kstat_create("zfs", 0, "arcstats", "misc", KSTAT_TYPE_NAMED
,
7566 sizeof (arc_stats
) / sizeof (kstat_named_t
), KSTAT_FLAG_VIRTUAL
);
7568 if (arc_ksp
!= NULL
) {
7569 arc_ksp
->ks_data
= &arc_stats
;
7570 arc_ksp
->ks_update
= arc_kstat_update
;
7571 kstat_install(arc_ksp
);
7574 (void) thread_create(NULL
, 0, arc_reclaim_thread
, NULL
, 0, &p0
,
7575 TS_RUN
, defclsyspri
);
7581 * Calculate maximum amount of dirty data per pool.
7583 * If it has been set by a module parameter, take that.
7584 * Otherwise, use a percentage of physical memory defined by
7585 * zfs_dirty_data_max_percent (default 10%) with a cap at
7586 * zfs_dirty_data_max_max (default 4G or 25% of physical memory).
7588 if (zfs_dirty_data_max_max
== 0)
7589 zfs_dirty_data_max_max
= MIN(4ULL * 1024 * 1024 * 1024,
7590 allmem
* zfs_dirty_data_max_max_percent
/ 100);
7592 if (zfs_dirty_data_max
== 0) {
7593 zfs_dirty_data_max
= allmem
*
7594 zfs_dirty_data_max_percent
/ 100;
7595 zfs_dirty_data_max
= MIN(zfs_dirty_data_max
,
7596 zfs_dirty_data_max_max
);
7606 spl_unregister_shrinker(&arc_shrinker
);
7607 #endif /* _KERNEL */
7609 mutex_enter(&arc_reclaim_lock
);
7610 arc_reclaim_thread_exit
= B_TRUE
;
7612 * The reclaim thread will set arc_reclaim_thread_exit back to
7613 * B_FALSE when it is finished exiting; we're waiting for that.
7615 while (arc_reclaim_thread_exit
) {
7616 cv_signal(&arc_reclaim_thread_cv
);
7617 cv_wait(&arc_reclaim_thread_cv
, &arc_reclaim_lock
);
7619 mutex_exit(&arc_reclaim_lock
);
7621 /* Use B_TRUE to ensure *all* buffers are evicted */
7622 arc_flush(NULL
, B_TRUE
);
7626 if (arc_ksp
!= NULL
) {
7627 kstat_delete(arc_ksp
);
7631 taskq_wait(arc_prune_taskq
);
7632 taskq_destroy(arc_prune_taskq
);
7634 mutex_enter(&arc_prune_mtx
);
7635 while ((p
= list_head(&arc_prune_list
)) != NULL
) {
7636 list_remove(&arc_prune_list
, p
);
7637 refcount_remove(&p
->p_refcnt
, &arc_prune_list
);
7638 refcount_destroy(&p
->p_refcnt
);
7639 kmem_free(p
, sizeof (*p
));
7641 mutex_exit(&arc_prune_mtx
);
7643 list_destroy(&arc_prune_list
);
7644 mutex_destroy(&arc_prune_mtx
);
7645 mutex_destroy(&arc_reclaim_lock
);
7646 cv_destroy(&arc_reclaim_thread_cv
);
7647 cv_destroy(&arc_reclaim_waiters_cv
);
7652 ASSERT0(arc_loaned_bytes
);
7658 * The level 2 ARC (L2ARC) is a cache layer in-between main memory and disk.
7659 * It uses dedicated storage devices to hold cached data, which are populated
7660 * using large infrequent writes. The main role of this cache is to boost
7661 * the performance of random read workloads. The intended L2ARC devices
7662 * include short-stroked disks, solid state disks, and other media with
7663 * substantially faster read latency than disk.
7665 * +-----------------------+
7667 * +-----------------------+
7670 * l2arc_feed_thread() arc_read()
7674 * +---------------+ |
7676 * +---------------+ |
7681 * +-------+ +-------+
7683 * | cache | | cache |
7684 * +-------+ +-------+
7685 * +=========+ .-----.
7686 * : L2ARC : |-_____-|
7687 * : devices : | Disks |
7688 * +=========+ `-_____-'
7690 * Read requests are satisfied from the following sources, in order:
7693 * 2) vdev cache of L2ARC devices
7695 * 4) vdev cache of disks
7698 * Some L2ARC device types exhibit extremely slow write performance.
7699 * To accommodate for this there are some significant differences between
7700 * the L2ARC and traditional cache design:
7702 * 1. There is no eviction path from the ARC to the L2ARC. Evictions from
7703 * the ARC behave as usual, freeing buffers and placing headers on ghost
7704 * lists. The ARC does not send buffers to the L2ARC during eviction as
7705 * this would add inflated write latencies for all ARC memory pressure.
7707 * 2. The L2ARC attempts to cache data from the ARC before it is evicted.
7708 * It does this by periodically scanning buffers from the eviction-end of
7709 * the MFU and MRU ARC lists, copying them to the L2ARC devices if they are
7710 * not already there. It scans until a headroom of buffers is satisfied,
7711 * which itself is a buffer for ARC eviction. If a compressible buffer is
7712 * found during scanning and selected for writing to an L2ARC device, we
7713 * temporarily boost scanning headroom during the next scan cycle to make
7714 * sure we adapt to compression effects (which might significantly reduce
7715 * the data volume we write to L2ARC). The thread that does this is
7716 * l2arc_feed_thread(), illustrated below; example sizes are included to
7717 * provide a better sense of ratio than this diagram:
7720 * +---------------------+----------+
7721 * ARC_mfu |:::::#:::::::::::::::|o#o###o###|-->. # already on L2ARC
7722 * +---------------------+----------+ | o L2ARC eligible
7723 * ARC_mru |:#:::::::::::::::::::|#o#ooo####|-->| : ARC buffer
7724 * +---------------------+----------+ |
7725 * 15.9 Gbytes ^ 32 Mbytes |
7727 * l2arc_feed_thread()
7729 * l2arc write hand <--[oooo]--'
7733 * +==============================+
7734 * L2ARC dev |####|#|###|###| |####| ... |
7735 * +==============================+
7738 * 3. If an ARC buffer is copied to the L2ARC but then hit instead of
7739 * evicted, then the L2ARC has cached a buffer much sooner than it probably
7740 * needed to, potentially wasting L2ARC device bandwidth and storage. It is
7741 * safe to say that this is an uncommon case, since buffers at the end of
7742 * the ARC lists have moved there due to inactivity.
7744 * 4. If the ARC evicts faster than the L2ARC can maintain a headroom,
7745 * then the L2ARC simply misses copying some buffers. This serves as a
7746 * pressure valve to prevent heavy read workloads from both stalling the ARC
7747 * with waits and clogging the L2ARC with writes. This also helps prevent
7748 * the potential for the L2ARC to churn if it attempts to cache content too
7749 * quickly, such as during backups of the entire pool.
7751 * 5. After system boot and before the ARC has filled main memory, there are
7752 * no evictions from the ARC and so the tails of the ARC_mfu and ARC_mru
7753 * lists can remain mostly static. Instead of searching from tail of these
7754 * lists as pictured, the l2arc_feed_thread() will search from the list heads
7755 * for eligible buffers, greatly increasing its chance of finding them.
7757 * The L2ARC device write speed is also boosted during this time so that
7758 * the L2ARC warms up faster. Since there have been no ARC evictions yet,
7759 * there are no L2ARC reads, and no fear of degrading read performance
7760 * through increased writes.
7762 * 6. Writes to the L2ARC devices are grouped and sent in-sequence, so that
7763 * the vdev queue can aggregate them into larger and fewer writes. Each
7764 * device is written to in a rotor fashion, sweeping writes through
7765 * available space then repeating.
7767 * 7. The L2ARC does not store dirty content. It never needs to flush
7768 * write buffers back to disk based storage.
7770 * 8. If an ARC buffer is written (and dirtied) which also exists in the
7771 * L2ARC, the now stale L2ARC buffer is immediately dropped.
7773 * The performance of the L2ARC can be tweaked by a number of tunables, which
7774 * may be necessary for different workloads:
7776 * l2arc_write_max max write bytes per interval
7777 * l2arc_write_boost extra write bytes during device warmup
7778 * l2arc_noprefetch skip caching prefetched buffers
7779 * l2arc_headroom number of max device writes to precache
7780 * l2arc_headroom_boost when we find compressed buffers during ARC
7781 * scanning, we multiply headroom by this
7782 * percentage factor for the next scan cycle,
7783 * since more compressed buffers are likely to
7785 * l2arc_feed_secs seconds between L2ARC writing
7787 * Tunables may be removed or added as future performance improvements are
7788 * integrated, and also may become zpool properties.
7790 * There are three key functions that control how the L2ARC warms up:
7792 * l2arc_write_eligible() check if a buffer is eligible to cache
7793 * l2arc_write_size() calculate how much to write
7794 * l2arc_write_interval() calculate sleep delay between writes
7796 * These three functions determine what to write, how much, and how quickly
7801 l2arc_write_eligible(uint64_t spa_guid
, arc_buf_hdr_t
*hdr
)
7804 * A buffer is *not* eligible for the L2ARC if it:
7805 * 1. belongs to a different spa.
7806 * 2. is already cached on the L2ARC.
7807 * 3. has an I/O in progress (it may be an incomplete read).
7808 * 4. is flagged not eligible (zfs property).
7810 if (hdr
->b_spa
!= spa_guid
|| HDR_HAS_L2HDR(hdr
) ||
7811 HDR_IO_IN_PROGRESS(hdr
) || !HDR_L2CACHE(hdr
))
7818 l2arc_write_size(void)
7823 * Make sure our globals have meaningful values in case the user
7826 size
= l2arc_write_max
;
7828 cmn_err(CE_NOTE
, "Bad value for l2arc_write_max, value must "
7829 "be greater than zero, resetting it to the default (%d)",
7831 size
= l2arc_write_max
= L2ARC_WRITE_SIZE
;
7834 if (arc_warm
== B_FALSE
)
7835 size
+= l2arc_write_boost
;
7842 l2arc_write_interval(clock_t began
, uint64_t wanted
, uint64_t wrote
)
7844 clock_t interval
, next
, now
;
7847 * If the ARC lists are busy, increase our write rate; if the
7848 * lists are stale, idle back. This is achieved by checking
7849 * how much we previously wrote - if it was more than half of
7850 * what we wanted, schedule the next write much sooner.
7852 if (l2arc_feed_again
&& wrote
> (wanted
/ 2))
7853 interval
= (hz
* l2arc_feed_min_ms
) / 1000;
7855 interval
= hz
* l2arc_feed_secs
;
7857 now
= ddi_get_lbolt();
7858 next
= MAX(now
, MIN(now
+ interval
, began
+ interval
));
7864 * Cycle through L2ARC devices. This is how L2ARC load balances.
7865 * If a device is returned, this also returns holding the spa config lock.
7867 static l2arc_dev_t
*
7868 l2arc_dev_get_next(void)
7870 l2arc_dev_t
*first
, *next
= NULL
;
7873 * Lock out the removal of spas (spa_namespace_lock), then removal
7874 * of cache devices (l2arc_dev_mtx). Once a device has been selected,
7875 * both locks will be dropped and a spa config lock held instead.
7877 mutex_enter(&spa_namespace_lock
);
7878 mutex_enter(&l2arc_dev_mtx
);
7880 /* if there are no vdevs, there is nothing to do */
7881 if (l2arc_ndev
== 0)
7885 next
= l2arc_dev_last
;
7887 /* loop around the list looking for a non-faulted vdev */
7889 next
= list_head(l2arc_dev_list
);
7891 next
= list_next(l2arc_dev_list
, next
);
7893 next
= list_head(l2arc_dev_list
);
7896 /* if we have come back to the start, bail out */
7899 else if (next
== first
)
7902 } while (vdev_is_dead(next
->l2ad_vdev
));
7904 /* if we were unable to find any usable vdevs, return NULL */
7905 if (vdev_is_dead(next
->l2ad_vdev
))
7908 l2arc_dev_last
= next
;
7911 mutex_exit(&l2arc_dev_mtx
);
7914 * Grab the config lock to prevent the 'next' device from being
7915 * removed while we are writing to it.
7918 spa_config_enter(next
->l2ad_spa
, SCL_L2ARC
, next
, RW_READER
);
7919 mutex_exit(&spa_namespace_lock
);
7925 * Free buffers that were tagged for destruction.
7928 l2arc_do_free_on_write(void)
7931 l2arc_data_free_t
*df
, *df_prev
;
7933 mutex_enter(&l2arc_free_on_write_mtx
);
7934 buflist
= l2arc_free_on_write
;
7936 for (df
= list_tail(buflist
); df
; df
= df_prev
) {
7937 df_prev
= list_prev(buflist
, df
);
7938 ASSERT3P(df
->l2df_abd
, !=, NULL
);
7939 abd_free(df
->l2df_abd
);
7940 list_remove(buflist
, df
);
7941 kmem_free(df
, sizeof (l2arc_data_free_t
));
7944 mutex_exit(&l2arc_free_on_write_mtx
);
7948 * A write to a cache device has completed. Update all headers to allow
7949 * reads from these buffers to begin.
7952 l2arc_write_done(zio_t
*zio
)
7954 l2arc_write_callback_t
*cb
;
7957 arc_buf_hdr_t
*head
, *hdr
, *hdr_prev
;
7958 kmutex_t
*hash_lock
;
7959 int64_t bytes_dropped
= 0;
7961 cb
= zio
->io_private
;
7962 ASSERT3P(cb
, !=, NULL
);
7963 dev
= cb
->l2wcb_dev
;
7964 ASSERT3P(dev
, !=, NULL
);
7965 head
= cb
->l2wcb_head
;
7966 ASSERT3P(head
, !=, NULL
);
7967 buflist
= &dev
->l2ad_buflist
;
7968 ASSERT3P(buflist
, !=, NULL
);
7969 DTRACE_PROBE2(l2arc__iodone
, zio_t
*, zio
,
7970 l2arc_write_callback_t
*, cb
);
7972 if (zio
->io_error
!= 0)
7973 ARCSTAT_BUMP(arcstat_l2_writes_error
);
7976 * All writes completed, or an error was hit.
7979 mutex_enter(&dev
->l2ad_mtx
);
7980 for (hdr
= list_prev(buflist
, head
); hdr
; hdr
= hdr_prev
) {
7981 hdr_prev
= list_prev(buflist
, hdr
);
7983 hash_lock
= HDR_LOCK(hdr
);
7986 * We cannot use mutex_enter or else we can deadlock
7987 * with l2arc_write_buffers (due to swapping the order
7988 * the hash lock and l2ad_mtx are taken).
7990 if (!mutex_tryenter(hash_lock
)) {
7992 * Missed the hash lock. We must retry so we
7993 * don't leave the ARC_FLAG_L2_WRITING bit set.
7995 ARCSTAT_BUMP(arcstat_l2_writes_lock_retry
);
7998 * We don't want to rescan the headers we've
7999 * already marked as having been written out, so
8000 * we reinsert the head node so we can pick up
8001 * where we left off.
8003 list_remove(buflist
, head
);
8004 list_insert_after(buflist
, hdr
, head
);
8006 mutex_exit(&dev
->l2ad_mtx
);
8009 * We wait for the hash lock to become available
8010 * to try and prevent busy waiting, and increase
8011 * the chance we'll be able to acquire the lock
8012 * the next time around.
8014 mutex_enter(hash_lock
);
8015 mutex_exit(hash_lock
);
8020 * We could not have been moved into the arc_l2c_only
8021 * state while in-flight due to our ARC_FLAG_L2_WRITING
8022 * bit being set. Let's just ensure that's being enforced.
8024 ASSERT(HDR_HAS_L1HDR(hdr
));
8027 * Skipped - drop L2ARC entry and mark the header as no
8028 * longer L2 eligibile.
8030 if (zio
->io_error
!= 0) {
8032 * Error - drop L2ARC entry.
8034 list_remove(buflist
, hdr
);
8035 arc_hdr_clear_flags(hdr
, ARC_FLAG_HAS_L2HDR
);
8037 ARCSTAT_INCR(arcstat_l2_psize
, -arc_hdr_size(hdr
));
8038 ARCSTAT_INCR(arcstat_l2_lsize
, -HDR_GET_LSIZE(hdr
));
8040 bytes_dropped
+= arc_hdr_size(hdr
);
8041 (void) refcount_remove_many(&dev
->l2ad_alloc
,
8042 arc_hdr_size(hdr
), hdr
);
8046 * Allow ARC to begin reads and ghost list evictions to
8049 arc_hdr_clear_flags(hdr
, ARC_FLAG_L2_WRITING
);
8051 mutex_exit(hash_lock
);
8054 atomic_inc_64(&l2arc_writes_done
);
8055 list_remove(buflist
, head
);
8056 ASSERT(!HDR_HAS_L1HDR(head
));
8057 kmem_cache_free(hdr_l2only_cache
, head
);
8058 mutex_exit(&dev
->l2ad_mtx
);
8060 vdev_space_update(dev
->l2ad_vdev
, -bytes_dropped
, 0, 0);
8062 l2arc_do_free_on_write();
8064 kmem_free(cb
, sizeof (l2arc_write_callback_t
));
8068 l2arc_untransform(zio_t
*zio
, l2arc_read_callback_t
*cb
)
8071 spa_t
*spa
= zio
->io_spa
;
8072 arc_buf_hdr_t
*hdr
= cb
->l2rcb_hdr
;
8073 blkptr_t
*bp
= zio
->io_bp
;
8074 dsl_crypto_key_t
*dck
= NULL
;
8075 uint8_t salt
[ZIO_DATA_SALT_LEN
];
8076 uint8_t iv
[ZIO_DATA_IV_LEN
];
8077 uint8_t mac
[ZIO_DATA_MAC_LEN
];
8078 boolean_t no_crypt
= B_FALSE
;
8081 * ZIL data is never be written to the L2ARC, so we don't need
8082 * special handling for its unique MAC storage.
8084 ASSERT3U(BP_GET_TYPE(bp
), !=, DMU_OT_INTENT_LOG
);
8085 ASSERT(MUTEX_HELD(HDR_LOCK(hdr
)));
8086 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, !=, NULL
);
8089 * If the data was encrypted, decrypt it now. Note that
8090 * we must check the bp here and not the hdr, since the
8091 * hdr does not have its encryption parameters updated
8092 * until arc_read_done().
8094 if (BP_IS_ENCRYPTED(bp
)) {
8095 abd_t
*eabd
= arc_get_data_abd(hdr
,
8096 arc_hdr_size(hdr
), hdr
);
8098 zio_crypt_decode_params_bp(bp
, salt
, iv
);
8099 zio_crypt_decode_mac_bp(bp
, mac
);
8101 ret
= spa_keystore_lookup_key(spa
,
8102 cb
->l2rcb_zb
.zb_objset
, FTAG
, &dck
);
8104 arc_free_data_abd(hdr
, eabd
, arc_hdr_size(hdr
), hdr
);
8108 ret
= zio_do_crypt_abd(B_FALSE
, &dck
->dck_key
,
8109 salt
, BP_GET_TYPE(bp
), iv
, mac
, HDR_GET_PSIZE(hdr
),
8110 BP_SHOULD_BYTESWAP(bp
), eabd
, hdr
->b_l1hdr
.b_pabd
,
8113 arc_free_data_abd(hdr
, eabd
, arc_hdr_size(hdr
), hdr
);
8114 spa_keystore_dsl_key_rele(spa
, dck
, FTAG
);
8118 spa_keystore_dsl_key_rele(spa
, dck
, FTAG
);
8121 * If we actually performed decryption, replace b_pabd
8122 * with the decrypted data. Otherwise we can just throw
8123 * our decryption buffer away.
8126 arc_free_data_abd(hdr
, hdr
->b_l1hdr
.b_pabd
,
8127 arc_hdr_size(hdr
), hdr
);
8128 hdr
->b_l1hdr
.b_pabd
= eabd
;
8131 arc_free_data_abd(hdr
, eabd
, arc_hdr_size(hdr
), hdr
);
8136 * If the L2ARC block was compressed, but ARC compression
8137 * is disabled we decompress the data into a new buffer and
8138 * replace the existing data.
8140 if (HDR_GET_COMPRESS(hdr
) != ZIO_COMPRESS_OFF
&&
8141 !HDR_COMPRESSION_ENABLED(hdr
)) {
8142 abd_t
*cabd
= arc_get_data_abd(hdr
, arc_hdr_size(hdr
), hdr
);
8143 void *tmp
= abd_borrow_buf(cabd
, arc_hdr_size(hdr
));
8145 ret
= zio_decompress_data(HDR_GET_COMPRESS(hdr
),
8146 hdr
->b_l1hdr
.b_pabd
, tmp
, HDR_GET_PSIZE(hdr
),
8147 HDR_GET_LSIZE(hdr
));
8149 abd_return_buf_copy(cabd
, tmp
, arc_hdr_size(hdr
));
8150 arc_free_data_abd(hdr
, cabd
, arc_hdr_size(hdr
), hdr
);
8154 abd_return_buf_copy(cabd
, tmp
, arc_hdr_size(hdr
));
8155 arc_free_data_abd(hdr
, hdr
->b_l1hdr
.b_pabd
,
8156 arc_hdr_size(hdr
), hdr
);
8157 hdr
->b_l1hdr
.b_pabd
= cabd
;
8159 zio
->io_size
= HDR_GET_LSIZE(hdr
);
8170 * A read to a cache device completed. Validate buffer contents before
8171 * handing over to the regular ARC routines.
8174 l2arc_read_done(zio_t
*zio
)
8177 l2arc_read_callback_t
*cb
;
8179 kmutex_t
*hash_lock
;
8180 boolean_t valid_cksum
, using_rdata
;
8182 ASSERT3P(zio
->io_vd
, !=, NULL
);
8183 ASSERT(zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
);
8185 spa_config_exit(zio
->io_spa
, SCL_L2ARC
, zio
->io_vd
);
8187 cb
= zio
->io_private
;
8188 ASSERT3P(cb
, !=, NULL
);
8189 hdr
= cb
->l2rcb_hdr
;
8190 ASSERT3P(hdr
, !=, NULL
);
8192 hash_lock
= HDR_LOCK(hdr
);
8193 mutex_enter(hash_lock
);
8194 ASSERT3P(hash_lock
, ==, HDR_LOCK(hdr
));
8197 * If the data was read into a temporary buffer,
8198 * move it and free the buffer.
8200 if (cb
->l2rcb_abd
!= NULL
) {
8201 ASSERT3U(arc_hdr_size(hdr
), <, zio
->io_size
);
8202 if (zio
->io_error
== 0) {
8203 abd_copy(hdr
->b_l1hdr
.b_pabd
, cb
->l2rcb_abd
,
8208 * The following must be done regardless of whether
8209 * there was an error:
8210 * - free the temporary buffer
8211 * - point zio to the real ARC buffer
8212 * - set zio size accordingly
8213 * These are required because zio is either re-used for
8214 * an I/O of the block in the case of the error
8215 * or the zio is passed to arc_read_done() and it
8218 abd_free(cb
->l2rcb_abd
);
8219 zio
->io_size
= zio
->io_orig_size
= arc_hdr_size(hdr
);
8221 if (BP_IS_ENCRYPTED(&cb
->l2rcb_bp
) &&
8222 (cb
->l2rcb_flags
& ZIO_FLAG_RAW_ENCRYPT
)) {
8223 ASSERT(HDR_HAS_RABD(hdr
));
8224 zio
->io_abd
= zio
->io_orig_abd
=
8225 hdr
->b_crypt_hdr
.b_rabd
;
8227 ASSERT3P(hdr
->b_l1hdr
.b_pabd
, !=, NULL
);
8228 zio
->io_abd
= zio
->io_orig_abd
= hdr
->b_l1hdr
.b_pabd
;
8232 ASSERT3P(zio
->io_abd
, !=, NULL
);
8235 * Check this survived the L2ARC journey.
8237 ASSERT(zio
->io_abd
== hdr
->b_l1hdr
.b_pabd
||
8238 (HDR_HAS_RABD(hdr
) && zio
->io_abd
== hdr
->b_crypt_hdr
.b_rabd
));
8239 zio
->io_bp_copy
= cb
->l2rcb_bp
; /* XXX fix in L2ARC 2.0 */
8240 zio
->io_bp
= &zio
->io_bp_copy
; /* XXX fix in L2ARC 2.0 */
8242 valid_cksum
= arc_cksum_is_equal(hdr
, zio
);
8243 using_rdata
= (HDR_HAS_RABD(hdr
) &&
8244 zio
->io_abd
== hdr
->b_crypt_hdr
.b_rabd
);
8247 * b_rabd will always match the data as it exists on disk if it is
8248 * being used. Therefore if we are reading into b_rabd we do not
8249 * attempt to untransform the data.
8251 if (valid_cksum
&& !using_rdata
)
8252 tfm_error
= l2arc_untransform(zio
, cb
);
8254 if (valid_cksum
&& tfm_error
== 0 && zio
->io_error
== 0 &&
8255 !HDR_L2_EVICTED(hdr
)) {
8256 mutex_exit(hash_lock
);
8257 zio
->io_private
= hdr
;
8260 mutex_exit(hash_lock
);
8262 * Buffer didn't survive caching. Increment stats and
8263 * reissue to the original storage device.
8265 if (zio
->io_error
!= 0) {
8266 ARCSTAT_BUMP(arcstat_l2_io_error
);
8268 zio
->io_error
= SET_ERROR(EIO
);
8270 if (!valid_cksum
|| tfm_error
!= 0)
8271 ARCSTAT_BUMP(arcstat_l2_cksum_bad
);
8274 * If there's no waiter, issue an async i/o to the primary
8275 * storage now. If there *is* a waiter, the caller must
8276 * issue the i/o in a context where it's OK to block.
8278 if (zio
->io_waiter
== NULL
) {
8279 zio_t
*pio
= zio_unique_parent(zio
);
8280 void *abd
= (using_rdata
) ?
8281 hdr
->b_crypt_hdr
.b_rabd
: hdr
->b_l1hdr
.b_pabd
;
8283 ASSERT(!pio
|| pio
->io_child_type
== ZIO_CHILD_LOGICAL
);
8285 zio_nowait(zio_read(pio
, zio
->io_spa
, zio
->io_bp
,
8286 abd
, zio
->io_size
, arc_read_done
,
8287 hdr
, zio
->io_priority
, cb
->l2rcb_flags
,
8292 kmem_free(cb
, sizeof (l2arc_read_callback_t
));
8296 * This is the list priority from which the L2ARC will search for pages to
8297 * cache. This is used within loops (0..3) to cycle through lists in the
8298 * desired order. This order can have a significant effect on cache
8301 * Currently the metadata lists are hit first, MFU then MRU, followed by
8302 * the data lists. This function returns a locked list, and also returns
8305 static multilist_sublist_t
*
8306 l2arc_sublist_lock(int list_num
)
8308 multilist_t
*ml
= NULL
;
8311 ASSERT(list_num
>= 0 && list_num
< L2ARC_FEED_TYPES
);
8315 ml
= arc_mfu
->arcs_list
[ARC_BUFC_METADATA
];
8318 ml
= arc_mru
->arcs_list
[ARC_BUFC_METADATA
];
8321 ml
= arc_mfu
->arcs_list
[ARC_BUFC_DATA
];
8324 ml
= arc_mru
->arcs_list
[ARC_BUFC_DATA
];
8331 * Return a randomly-selected sublist. This is acceptable
8332 * because the caller feeds only a little bit of data for each
8333 * call (8MB). Subsequent calls will result in different
8334 * sublists being selected.
8336 idx
= multilist_get_random_index(ml
);
8337 return (multilist_sublist_lock(ml
, idx
));
8341 * Evict buffers from the device write hand to the distance specified in
8342 * bytes. This distance may span populated buffers, it may span nothing.
8343 * This is clearing a region on the L2ARC device ready for writing.
8344 * If the 'all' boolean is set, every buffer is evicted.
8347 l2arc_evict(l2arc_dev_t
*dev
, uint64_t distance
, boolean_t all
)
8350 arc_buf_hdr_t
*hdr
, *hdr_prev
;
8351 kmutex_t
*hash_lock
;
8354 buflist
= &dev
->l2ad_buflist
;
8356 if (!all
&& dev
->l2ad_first
) {
8358 * This is the first sweep through the device. There is
8364 if (dev
->l2ad_hand
>= (dev
->l2ad_end
- (2 * distance
))) {
8366 * When nearing the end of the device, evict to the end
8367 * before the device write hand jumps to the start.
8369 taddr
= dev
->l2ad_end
;
8371 taddr
= dev
->l2ad_hand
+ distance
;
8373 DTRACE_PROBE4(l2arc__evict
, l2arc_dev_t
*, dev
, list_t
*, buflist
,
8374 uint64_t, taddr
, boolean_t
, all
);
8377 mutex_enter(&dev
->l2ad_mtx
);
8378 for (hdr
= list_tail(buflist
); hdr
; hdr
= hdr_prev
) {
8379 hdr_prev
= list_prev(buflist
, hdr
);
8381 hash_lock
= HDR_LOCK(hdr
);
8384 * We cannot use mutex_enter or else we can deadlock
8385 * with l2arc_write_buffers (due to swapping the order
8386 * the hash lock and l2ad_mtx are taken).
8388 if (!mutex_tryenter(hash_lock
)) {
8390 * Missed the hash lock. Retry.
8392 ARCSTAT_BUMP(arcstat_l2_evict_lock_retry
);
8393 mutex_exit(&dev
->l2ad_mtx
);
8394 mutex_enter(hash_lock
);
8395 mutex_exit(hash_lock
);
8400 * A header can't be on this list if it doesn't have L2 header.
8402 ASSERT(HDR_HAS_L2HDR(hdr
));
8404 /* Ensure this header has finished being written. */
8405 ASSERT(!HDR_L2_WRITING(hdr
));
8406 ASSERT(!HDR_L2_WRITE_HEAD(hdr
));
8408 if (!all
&& (hdr
->b_l2hdr
.b_daddr
>= taddr
||
8409 hdr
->b_l2hdr
.b_daddr
< dev
->l2ad_hand
)) {
8411 * We've evicted to the target address,
8412 * or the end of the device.
8414 mutex_exit(hash_lock
);
8418 if (!HDR_HAS_L1HDR(hdr
)) {
8419 ASSERT(!HDR_L2_READING(hdr
));
8421 * This doesn't exist in the ARC. Destroy.
8422 * arc_hdr_destroy() will call list_remove()
8423 * and decrement arcstat_l2_lsize.
8425 arc_change_state(arc_anon
, hdr
, hash_lock
);
8426 arc_hdr_destroy(hdr
);
8428 ASSERT(hdr
->b_l1hdr
.b_state
!= arc_l2c_only
);
8429 ARCSTAT_BUMP(arcstat_l2_evict_l1cached
);
8431 * Invalidate issued or about to be issued
8432 * reads, since we may be about to write
8433 * over this location.
8435 if (HDR_L2_READING(hdr
)) {
8436 ARCSTAT_BUMP(arcstat_l2_evict_reading
);
8437 arc_hdr_set_flags(hdr
, ARC_FLAG_L2_EVICTED
);
8440 arc_hdr_l2hdr_destroy(hdr
);
8442 mutex_exit(hash_lock
);
8444 mutex_exit(&dev
->l2ad_mtx
);
8448 * Handle any abd transforms that might be required for writing to the L2ARC.
8449 * If successful, this function will always return an abd with the data
8450 * transformed as it is on disk in a new abd of asize bytes.
8453 l2arc_apply_transforms(spa_t
*spa
, arc_buf_hdr_t
*hdr
, uint64_t asize
,
8458 abd_t
*cabd
= NULL
, *eabd
= NULL
, *to_write
= hdr
->b_l1hdr
.b_pabd
;
8459 enum zio_compress compress
= HDR_GET_COMPRESS(hdr
);
8460 uint64_t psize
= HDR_GET_PSIZE(hdr
);
8461 uint64_t size
= arc_hdr_size(hdr
);
8462 boolean_t ismd
= HDR_ISTYPE_METADATA(hdr
);
8463 boolean_t bswap
= (hdr
->b_l1hdr
.b_byteswap
!= DMU_BSWAP_NUMFUNCS
);
8464 dsl_crypto_key_t
*dck
= NULL
;
8465 uint8_t mac
[ZIO_DATA_MAC_LEN
] = { 0 };
8466 boolean_t no_crypt
= B_FALSE
;
8468 ASSERT((HDR_GET_COMPRESS(hdr
) != ZIO_COMPRESS_OFF
&&
8469 !HDR_COMPRESSION_ENABLED(hdr
)) ||
8470 HDR_ENCRYPTED(hdr
) || HDR_SHARED_DATA(hdr
) || psize
!= asize
);
8471 ASSERT3U(psize
, <=, asize
);
8474 * If this data simply needs its own buffer, we simply allocate it
8475 * and copy the data. This may be done to elimiate a depedency on a
8476 * shared buffer or to reallocate the buffer to match asize.
8478 if (HDR_HAS_RABD(hdr
) && asize
!= psize
) {
8479 ASSERT3U(asize
, >=, psize
);
8480 to_write
= abd_alloc_for_io(asize
, ismd
);
8481 abd_copy(to_write
, hdr
->b_crypt_hdr
.b_rabd
, psize
);
8483 abd_zero_off(to_write
, psize
, asize
- psize
);
8487 if ((compress
== ZIO_COMPRESS_OFF
|| HDR_COMPRESSION_ENABLED(hdr
)) &&
8488 !HDR_ENCRYPTED(hdr
)) {
8489 ASSERT3U(size
, ==, psize
);
8490 to_write
= abd_alloc_for_io(asize
, ismd
);
8491 abd_copy(to_write
, hdr
->b_l1hdr
.b_pabd
, size
);
8493 abd_zero_off(to_write
, size
, asize
- size
);
8497 if (compress
!= ZIO_COMPRESS_OFF
&& !HDR_COMPRESSION_ENABLED(hdr
)) {
8498 cabd
= abd_alloc_for_io(asize
, ismd
);
8499 tmp
= abd_borrow_buf(cabd
, asize
);
8501 psize
= zio_compress_data(compress
, to_write
, tmp
, size
);
8502 ASSERT3U(psize
, <=, HDR_GET_PSIZE(hdr
));
8504 bzero((char *)tmp
+ psize
, asize
- psize
);
8505 psize
= HDR_GET_PSIZE(hdr
);
8506 abd_return_buf_copy(cabd
, tmp
, asize
);
8510 if (HDR_ENCRYPTED(hdr
)) {
8511 eabd
= abd_alloc_for_io(asize
, ismd
);
8514 * If the dataset was disowned before the buffer
8515 * made it to this point, the key to re-encrypt
8516 * it won't be available. In this case we simply
8517 * won't write the buffer to the L2ARC.
8519 ret
= spa_keystore_lookup_key(spa
, hdr
->b_crypt_hdr
.b_dsobj
,
8524 ret
= zio_do_crypt_abd(B_TRUE
, &dck
->dck_key
,
8525 hdr
->b_crypt_hdr
.b_salt
, hdr
->b_crypt_hdr
.b_ot
,
8526 hdr
->b_crypt_hdr
.b_iv
, mac
, psize
, bswap
, to_write
,
8532 abd_copy(eabd
, to_write
, psize
);
8535 abd_zero_off(eabd
, psize
, asize
- psize
);
8537 /* assert that the MAC we got here matches the one we saved */
8538 ASSERT0(bcmp(mac
, hdr
->b_crypt_hdr
.b_mac
, ZIO_DATA_MAC_LEN
));
8539 spa_keystore_dsl_key_rele(spa
, dck
, FTAG
);
8541 if (to_write
== cabd
)
8548 ASSERT3P(to_write
, !=, hdr
->b_l1hdr
.b_pabd
);
8549 *abd_out
= to_write
;
8554 spa_keystore_dsl_key_rele(spa
, dck
, FTAG
);
8565 * Find and write ARC buffers to the L2ARC device.
8567 * An ARC_FLAG_L2_WRITING flag is set so that the L2ARC buffers are not valid
8568 * for reading until they have completed writing.
8569 * The headroom_boost is an in-out parameter used to maintain headroom boost
8570 * state between calls to this function.
8572 * Returns the number of bytes actually written (which may be smaller than
8573 * the delta by which the device hand has changed due to alignment).
8576 l2arc_write_buffers(spa_t
*spa
, l2arc_dev_t
*dev
, uint64_t target_sz
)
8578 arc_buf_hdr_t
*hdr
, *hdr_prev
, *head
;
8579 uint64_t write_asize
, write_psize
, write_lsize
, headroom
;
8581 l2arc_write_callback_t
*cb
;
8583 uint64_t guid
= spa_load_guid(spa
);
8585 ASSERT3P(dev
->l2ad_vdev
, !=, NULL
);
8588 write_lsize
= write_asize
= write_psize
= 0;
8590 head
= kmem_cache_alloc(hdr_l2only_cache
, KM_PUSHPAGE
);
8591 arc_hdr_set_flags(head
, ARC_FLAG_L2_WRITE_HEAD
| ARC_FLAG_HAS_L2HDR
);
8594 * Copy buffers for L2ARC writing.
8596 for (int try = 0; try < L2ARC_FEED_TYPES
; try++) {
8597 multilist_sublist_t
*mls
= l2arc_sublist_lock(try);
8598 uint64_t passed_sz
= 0;
8600 VERIFY3P(mls
, !=, NULL
);
8603 * L2ARC fast warmup.
8605 * Until the ARC is warm and starts to evict, read from the
8606 * head of the ARC lists rather than the tail.
8608 if (arc_warm
== B_FALSE
)
8609 hdr
= multilist_sublist_head(mls
);
8611 hdr
= multilist_sublist_tail(mls
);
8613 headroom
= target_sz
* l2arc_headroom
;
8614 if (zfs_compressed_arc_enabled
)
8615 headroom
= (headroom
* l2arc_headroom_boost
) / 100;
8617 for (; hdr
; hdr
= hdr_prev
) {
8618 kmutex_t
*hash_lock
;
8619 abd_t
*to_write
= NULL
;
8621 if (arc_warm
== B_FALSE
)
8622 hdr_prev
= multilist_sublist_next(mls
, hdr
);
8624 hdr_prev
= multilist_sublist_prev(mls
, hdr
);
8626 hash_lock
= HDR_LOCK(hdr
);
8627 if (!mutex_tryenter(hash_lock
)) {
8629 * Skip this buffer rather than waiting.
8634 passed_sz
+= HDR_GET_LSIZE(hdr
);
8635 if (passed_sz
> headroom
) {
8639 mutex_exit(hash_lock
);
8643 if (!l2arc_write_eligible(guid
, hdr
)) {
8644 mutex_exit(hash_lock
);
8649 * We rely on the L1 portion of the header below, so
8650 * it's invalid for this header to have been evicted out
8651 * of the ghost cache, prior to being written out. The
8652 * ARC_FLAG_L2_WRITING bit ensures this won't happen.
8654 ASSERT(HDR_HAS_L1HDR(hdr
));
8656 ASSERT3U(HDR_GET_PSIZE(hdr
), >, 0);
8657 ASSERT3U(arc_hdr_size(hdr
), >, 0);
8658 ASSERT(hdr
->b_l1hdr
.b_pabd
!= NULL
||
8660 uint64_t psize
= HDR_GET_PSIZE(hdr
);
8661 uint64_t asize
= vdev_psize_to_asize(dev
->l2ad_vdev
,
8664 if ((write_asize
+ asize
) > target_sz
) {
8666 mutex_exit(hash_lock
);
8671 * We rely on the L1 portion of the header below, so
8672 * it's invalid for this header to have been evicted out
8673 * of the ghost cache, prior to being written out. The
8674 * ARC_FLAG_L2_WRITING bit ensures this won't happen.
8676 arc_hdr_set_flags(hdr
, ARC_FLAG_L2_WRITING
);
8677 ASSERT(HDR_HAS_L1HDR(hdr
));
8679 ASSERT3U(HDR_GET_PSIZE(hdr
), >, 0);
8680 ASSERT(hdr
->b_l1hdr
.b_pabd
!= NULL
||
8682 ASSERT3U(arc_hdr_size(hdr
), >, 0);
8685 * If this header has b_rabd, we can use this since it
8686 * must always match the data exactly as it exists on
8687 * disk. Otherwise, the L2ARC can normally use the
8688 * hdr's data, but if we're sharing data between the
8689 * hdr and one of its bufs, L2ARC needs its own copy of
8690 * the data so that the ZIO below can't race with the
8691 * buf consumer. To ensure that this copy will be
8692 * available for the lifetime of the ZIO and be cleaned
8693 * up afterwards, we add it to the l2arc_free_on_write
8694 * queue. If we need to apply any transforms to the
8695 * data (compression, encryption) we will also need the
8698 if (HDR_HAS_RABD(hdr
) && psize
== asize
) {
8699 to_write
= hdr
->b_crypt_hdr
.b_rabd
;
8700 } else if ((HDR_COMPRESSION_ENABLED(hdr
) ||
8701 HDR_GET_COMPRESS(hdr
) == ZIO_COMPRESS_OFF
) &&
8702 !HDR_ENCRYPTED(hdr
) && !HDR_SHARED_DATA(hdr
) &&
8704 to_write
= hdr
->b_l1hdr
.b_pabd
;
8707 arc_buf_contents_t type
= arc_buf_type(hdr
);
8709 ret
= l2arc_apply_transforms(spa
, hdr
, asize
,
8712 arc_hdr_clear_flags(hdr
,
8713 ARC_FLAG_L2_WRITING
);
8714 mutex_exit(hash_lock
);
8718 l2arc_free_abd_on_write(to_write
, asize
, type
);
8723 * Insert a dummy header on the buflist so
8724 * l2arc_write_done() can find where the
8725 * write buffers begin without searching.
8727 mutex_enter(&dev
->l2ad_mtx
);
8728 list_insert_head(&dev
->l2ad_buflist
, head
);
8729 mutex_exit(&dev
->l2ad_mtx
);
8732 sizeof (l2arc_write_callback_t
), KM_SLEEP
);
8733 cb
->l2wcb_dev
= dev
;
8734 cb
->l2wcb_head
= head
;
8735 pio
= zio_root(spa
, l2arc_write_done
, cb
,
8739 hdr
->b_l2hdr
.b_dev
= dev
;
8740 hdr
->b_l2hdr
.b_hits
= 0;
8742 hdr
->b_l2hdr
.b_daddr
= dev
->l2ad_hand
;
8743 arc_hdr_set_flags(hdr
, ARC_FLAG_HAS_L2HDR
);
8745 mutex_enter(&dev
->l2ad_mtx
);
8746 list_insert_head(&dev
->l2ad_buflist
, hdr
);
8747 mutex_exit(&dev
->l2ad_mtx
);
8749 (void) refcount_add_many(&dev
->l2ad_alloc
,
8750 arc_hdr_size(hdr
), hdr
);
8752 wzio
= zio_write_phys(pio
, dev
->l2ad_vdev
,
8753 hdr
->b_l2hdr
.b_daddr
, asize
, to_write
,
8754 ZIO_CHECKSUM_OFF
, NULL
, hdr
,
8755 ZIO_PRIORITY_ASYNC_WRITE
,
8756 ZIO_FLAG_CANFAIL
, B_FALSE
);
8758 write_lsize
+= HDR_GET_LSIZE(hdr
);
8759 DTRACE_PROBE2(l2arc__write
, vdev_t
*, dev
->l2ad_vdev
,
8762 write_psize
+= psize
;
8763 write_asize
+= asize
;
8764 dev
->l2ad_hand
+= asize
;
8766 mutex_exit(hash_lock
);
8768 (void) zio_nowait(wzio
);
8771 multilist_sublist_unlock(mls
);
8777 /* No buffers selected for writing? */
8779 ASSERT0(write_lsize
);
8780 ASSERT(!HDR_HAS_L1HDR(head
));
8781 kmem_cache_free(hdr_l2only_cache
, head
);
8785 ASSERT3U(write_asize
, <=, target_sz
);
8786 ARCSTAT_BUMP(arcstat_l2_writes_sent
);
8787 ARCSTAT_INCR(arcstat_l2_write_bytes
, write_psize
);
8788 ARCSTAT_INCR(arcstat_l2_lsize
, write_lsize
);
8789 ARCSTAT_INCR(arcstat_l2_psize
, write_psize
);
8790 vdev_space_update(dev
->l2ad_vdev
, write_psize
, 0, 0);
8793 * Bump device hand to the device start if it is approaching the end.
8794 * l2arc_evict() will already have evicted ahead for this case.
8796 if (dev
->l2ad_hand
>= (dev
->l2ad_end
- target_sz
)) {
8797 dev
->l2ad_hand
= dev
->l2ad_start
;
8798 dev
->l2ad_first
= B_FALSE
;
8801 dev
->l2ad_writing
= B_TRUE
;
8802 (void) zio_wait(pio
);
8803 dev
->l2ad_writing
= B_FALSE
;
8805 return (write_asize
);
8809 * This thread feeds the L2ARC at regular intervals. This is the beating
8810 * heart of the L2ARC.
8814 l2arc_feed_thread(void *unused
)
8819 uint64_t size
, wrote
;
8820 clock_t begin
, next
= ddi_get_lbolt();
8821 fstrans_cookie_t cookie
;
8823 CALLB_CPR_INIT(&cpr
, &l2arc_feed_thr_lock
, callb_generic_cpr
, FTAG
);
8825 mutex_enter(&l2arc_feed_thr_lock
);
8827 cookie
= spl_fstrans_mark();
8828 while (l2arc_thread_exit
== 0) {
8829 CALLB_CPR_SAFE_BEGIN(&cpr
);
8830 (void) cv_timedwait_sig(&l2arc_feed_thr_cv
,
8831 &l2arc_feed_thr_lock
, next
);
8832 CALLB_CPR_SAFE_END(&cpr
, &l2arc_feed_thr_lock
);
8833 next
= ddi_get_lbolt() + hz
;
8836 * Quick check for L2ARC devices.
8838 mutex_enter(&l2arc_dev_mtx
);
8839 if (l2arc_ndev
== 0) {
8840 mutex_exit(&l2arc_dev_mtx
);
8843 mutex_exit(&l2arc_dev_mtx
);
8844 begin
= ddi_get_lbolt();
8847 * This selects the next l2arc device to write to, and in
8848 * doing so the next spa to feed from: dev->l2ad_spa. This
8849 * will return NULL if there are now no l2arc devices or if
8850 * they are all faulted.
8852 * If a device is returned, its spa's config lock is also
8853 * held to prevent device removal. l2arc_dev_get_next()
8854 * will grab and release l2arc_dev_mtx.
8856 if ((dev
= l2arc_dev_get_next()) == NULL
)
8859 spa
= dev
->l2ad_spa
;
8860 ASSERT3P(spa
, !=, NULL
);
8863 * If the pool is read-only then force the feed thread to
8864 * sleep a little longer.
8866 if (!spa_writeable(spa
)) {
8867 next
= ddi_get_lbolt() + 5 * l2arc_feed_secs
* hz
;
8868 spa_config_exit(spa
, SCL_L2ARC
, dev
);
8873 * Avoid contributing to memory pressure.
8875 if (arc_reclaim_needed()) {
8876 ARCSTAT_BUMP(arcstat_l2_abort_lowmem
);
8877 spa_config_exit(spa
, SCL_L2ARC
, dev
);
8881 ARCSTAT_BUMP(arcstat_l2_feeds
);
8883 size
= l2arc_write_size();
8886 * Evict L2ARC buffers that will be overwritten.
8888 l2arc_evict(dev
, size
, B_FALSE
);
8891 * Write ARC buffers.
8893 wrote
= l2arc_write_buffers(spa
, dev
, size
);
8896 * Calculate interval between writes.
8898 next
= l2arc_write_interval(begin
, size
, wrote
);
8899 spa_config_exit(spa
, SCL_L2ARC
, dev
);
8901 spl_fstrans_unmark(cookie
);
8903 l2arc_thread_exit
= 0;
8904 cv_broadcast(&l2arc_feed_thr_cv
);
8905 CALLB_CPR_EXIT(&cpr
); /* drops l2arc_feed_thr_lock */
8910 l2arc_vdev_present(vdev_t
*vd
)
8914 mutex_enter(&l2arc_dev_mtx
);
8915 for (dev
= list_head(l2arc_dev_list
); dev
!= NULL
;
8916 dev
= list_next(l2arc_dev_list
, dev
)) {
8917 if (dev
->l2ad_vdev
== vd
)
8920 mutex_exit(&l2arc_dev_mtx
);
8922 return (dev
!= NULL
);
8926 * Add a vdev for use by the L2ARC. By this point the spa has already
8927 * validated the vdev and opened it.
8930 l2arc_add_vdev(spa_t
*spa
, vdev_t
*vd
)
8932 l2arc_dev_t
*adddev
;
8934 ASSERT(!l2arc_vdev_present(vd
));
8937 * Create a new l2arc device entry.
8939 adddev
= kmem_zalloc(sizeof (l2arc_dev_t
), KM_SLEEP
);
8940 adddev
->l2ad_spa
= spa
;
8941 adddev
->l2ad_vdev
= vd
;
8942 adddev
->l2ad_start
= VDEV_LABEL_START_SIZE
;
8943 adddev
->l2ad_end
= VDEV_LABEL_START_SIZE
+ vdev_get_min_asize(vd
);
8944 adddev
->l2ad_hand
= adddev
->l2ad_start
;
8945 adddev
->l2ad_first
= B_TRUE
;
8946 adddev
->l2ad_writing
= B_FALSE
;
8947 list_link_init(&adddev
->l2ad_node
);
8949 mutex_init(&adddev
->l2ad_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
8951 * This is a list of all ARC buffers that are still valid on the
8954 list_create(&adddev
->l2ad_buflist
, sizeof (arc_buf_hdr_t
),
8955 offsetof(arc_buf_hdr_t
, b_l2hdr
.b_l2node
));
8957 vdev_space_update(vd
, 0, 0, adddev
->l2ad_end
- adddev
->l2ad_hand
);
8958 refcount_create(&adddev
->l2ad_alloc
);
8961 * Add device to global list
8963 mutex_enter(&l2arc_dev_mtx
);
8964 list_insert_head(l2arc_dev_list
, adddev
);
8965 atomic_inc_64(&l2arc_ndev
);
8966 mutex_exit(&l2arc_dev_mtx
);
8970 * Remove a vdev from the L2ARC.
8973 l2arc_remove_vdev(vdev_t
*vd
)
8975 l2arc_dev_t
*dev
, *nextdev
, *remdev
= NULL
;
8978 * Find the device by vdev
8980 mutex_enter(&l2arc_dev_mtx
);
8981 for (dev
= list_head(l2arc_dev_list
); dev
; dev
= nextdev
) {
8982 nextdev
= list_next(l2arc_dev_list
, dev
);
8983 if (vd
== dev
->l2ad_vdev
) {
8988 ASSERT3P(remdev
, !=, NULL
);
8991 * Remove device from global list
8993 list_remove(l2arc_dev_list
, remdev
);
8994 l2arc_dev_last
= NULL
; /* may have been invalidated */
8995 atomic_dec_64(&l2arc_ndev
);
8996 mutex_exit(&l2arc_dev_mtx
);
8999 * Clear all buflists and ARC references. L2ARC device flush.
9001 l2arc_evict(remdev
, 0, B_TRUE
);
9002 list_destroy(&remdev
->l2ad_buflist
);
9003 mutex_destroy(&remdev
->l2ad_mtx
);
9004 refcount_destroy(&remdev
->l2ad_alloc
);
9005 kmem_free(remdev
, sizeof (l2arc_dev_t
));
9011 l2arc_thread_exit
= 0;
9013 l2arc_writes_sent
= 0;
9014 l2arc_writes_done
= 0;
9016 mutex_init(&l2arc_feed_thr_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
9017 cv_init(&l2arc_feed_thr_cv
, NULL
, CV_DEFAULT
, NULL
);
9018 mutex_init(&l2arc_dev_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
9019 mutex_init(&l2arc_free_on_write_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
9021 l2arc_dev_list
= &L2ARC_dev_list
;
9022 l2arc_free_on_write
= &L2ARC_free_on_write
;
9023 list_create(l2arc_dev_list
, sizeof (l2arc_dev_t
),
9024 offsetof(l2arc_dev_t
, l2ad_node
));
9025 list_create(l2arc_free_on_write
, sizeof (l2arc_data_free_t
),
9026 offsetof(l2arc_data_free_t
, l2df_list_node
));
9033 * This is called from dmu_fini(), which is called from spa_fini();
9034 * Because of this, we can assume that all l2arc devices have
9035 * already been removed when the pools themselves were removed.
9038 l2arc_do_free_on_write();
9040 mutex_destroy(&l2arc_feed_thr_lock
);
9041 cv_destroy(&l2arc_feed_thr_cv
);
9042 mutex_destroy(&l2arc_dev_mtx
);
9043 mutex_destroy(&l2arc_free_on_write_mtx
);
9045 list_destroy(l2arc_dev_list
);
9046 list_destroy(l2arc_free_on_write
);
9052 if (!(spa_mode_global
& FWRITE
))
9055 (void) thread_create(NULL
, 0, l2arc_feed_thread
, NULL
, 0, &p0
,
9056 TS_RUN
, defclsyspri
);
9062 if (!(spa_mode_global
& FWRITE
))
9065 mutex_enter(&l2arc_feed_thr_lock
);
9066 cv_signal(&l2arc_feed_thr_cv
); /* kick thread out of startup */
9067 l2arc_thread_exit
= 1;
9068 while (l2arc_thread_exit
!= 0)
9069 cv_wait(&l2arc_feed_thr_cv
, &l2arc_feed_thr_lock
);
9070 mutex_exit(&l2arc_feed_thr_lock
);
9073 #if defined(_KERNEL) && defined(HAVE_SPL)
9074 EXPORT_SYMBOL(arc_buf_size
);
9075 EXPORT_SYMBOL(arc_write
);
9076 EXPORT_SYMBOL(arc_read
);
9077 EXPORT_SYMBOL(arc_buf_info
);
9078 EXPORT_SYMBOL(arc_getbuf_func
);
9079 EXPORT_SYMBOL(arc_add_prune_callback
);
9080 EXPORT_SYMBOL(arc_remove_prune_callback
);
9083 module_param(zfs_arc_min
, ulong
, 0644);
9084 MODULE_PARM_DESC(zfs_arc_min
, "Min arc size");
9086 module_param(zfs_arc_max
, ulong
, 0644);
9087 MODULE_PARM_DESC(zfs_arc_max
, "Max arc size");
9089 module_param(zfs_arc_meta_limit
, ulong
, 0644);
9090 MODULE_PARM_DESC(zfs_arc_meta_limit
, "Meta limit for arc size");
9092 module_param(zfs_arc_meta_limit_percent
, ulong
, 0644);
9093 MODULE_PARM_DESC(zfs_arc_meta_limit_percent
,
9094 "Percent of arc size for arc meta limit");
9096 module_param(zfs_arc_meta_min
, ulong
, 0644);
9097 MODULE_PARM_DESC(zfs_arc_meta_min
, "Min arc metadata");
9099 module_param(zfs_arc_meta_prune
, int, 0644);
9100 MODULE_PARM_DESC(zfs_arc_meta_prune
, "Meta objects to scan for prune");
9102 module_param(zfs_arc_meta_adjust_restarts
, int, 0644);
9103 MODULE_PARM_DESC(zfs_arc_meta_adjust_restarts
,
9104 "Limit number of restarts in arc_adjust_meta");
9106 module_param(zfs_arc_meta_strategy
, int, 0644);
9107 MODULE_PARM_DESC(zfs_arc_meta_strategy
, "Meta reclaim strategy");
9109 module_param(zfs_arc_grow_retry
, int, 0644);
9110 MODULE_PARM_DESC(zfs_arc_grow_retry
, "Seconds before growing arc size");
9112 module_param(zfs_arc_p_dampener_disable
, int, 0644);
9113 MODULE_PARM_DESC(zfs_arc_p_dampener_disable
, "disable arc_p adapt dampener");
9115 module_param(zfs_arc_shrink_shift
, int, 0644);
9116 MODULE_PARM_DESC(zfs_arc_shrink_shift
, "log2(fraction of arc to reclaim)");
9118 module_param(zfs_arc_pc_percent
, uint
, 0644);
9119 MODULE_PARM_DESC(zfs_arc_pc_percent
,
9120 "Percent of pagecache to reclaim arc to");
9122 module_param(zfs_arc_p_min_shift
, int, 0644);
9123 MODULE_PARM_DESC(zfs_arc_p_min_shift
, "arc_c shift to calc min/max arc_p");
9125 module_param(zfs_arc_average_blocksize
, int, 0444);
9126 MODULE_PARM_DESC(zfs_arc_average_blocksize
, "Target average block size");
9128 module_param(zfs_compressed_arc_enabled
, int, 0644);
9129 MODULE_PARM_DESC(zfs_compressed_arc_enabled
, "Disable compressed arc buffers");
9131 module_param(zfs_arc_min_prefetch_ms
, int, 0644);
9132 MODULE_PARM_DESC(zfs_arc_min_prefetch_ms
, "Min life of prefetch block in ms");
9134 module_param(zfs_arc_min_prescient_prefetch_ms
, int, 0644);
9135 MODULE_PARM_DESC(zfs_arc_min_prescient_prefetch_ms
,
9136 "Min life of prescient prefetched block in ms");
9138 module_param(l2arc_write_max
, ulong
, 0644);
9139 MODULE_PARM_DESC(l2arc_write_max
, "Max write bytes per interval");
9141 module_param(l2arc_write_boost
, ulong
, 0644);
9142 MODULE_PARM_DESC(l2arc_write_boost
, "Extra write bytes during device warmup");
9144 module_param(l2arc_headroom
, ulong
, 0644);
9145 MODULE_PARM_DESC(l2arc_headroom
, "Number of max device writes to precache");
9147 module_param(l2arc_headroom_boost
, ulong
, 0644);
9148 MODULE_PARM_DESC(l2arc_headroom_boost
, "Compressed l2arc_headroom multiplier");
9150 module_param(l2arc_feed_secs
, ulong
, 0644);
9151 MODULE_PARM_DESC(l2arc_feed_secs
, "Seconds between L2ARC writing");
9153 module_param(l2arc_feed_min_ms
, ulong
, 0644);
9154 MODULE_PARM_DESC(l2arc_feed_min_ms
, "Min feed interval in milliseconds");
9156 module_param(l2arc_noprefetch
, int, 0644);
9157 MODULE_PARM_DESC(l2arc_noprefetch
, "Skip caching prefetched buffers");
9159 module_param(l2arc_feed_again
, int, 0644);
9160 MODULE_PARM_DESC(l2arc_feed_again
, "Turbo L2ARC warmup");
9162 module_param(l2arc_norw
, int, 0644);
9163 MODULE_PARM_DESC(l2arc_norw
, "No reads during writes");
9165 module_param(zfs_arc_lotsfree_percent
, int, 0644);
9166 MODULE_PARM_DESC(zfs_arc_lotsfree_percent
,
9167 "System free memory I/O throttle in bytes");
9169 module_param(zfs_arc_sys_free
, ulong
, 0644);
9170 MODULE_PARM_DESC(zfs_arc_sys_free
, "System free memory target size in bytes");
9172 module_param(zfs_arc_dnode_limit
, ulong
, 0644);
9173 MODULE_PARM_DESC(zfs_arc_dnode_limit
, "Minimum bytes of dnodes in arc");
9175 module_param(zfs_arc_dnode_limit_percent
, ulong
, 0644);
9176 MODULE_PARM_DESC(zfs_arc_dnode_limit_percent
,
9177 "Percent of ARC meta buffers for dnodes");
9179 module_param(zfs_arc_dnode_reduce_percent
, ulong
, 0644);
9180 MODULE_PARM_DESC(zfs_arc_dnode_reduce_percent
,
9181 "Percentage of excess dnodes to try to unpin");