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 2011 Nexenta Systems, Inc. All rights reserved.
24 * Copyright (c) 2011 by Delphix. All rights reserved.
28 * DVA-based Adjustable Replacement Cache
30 * While much of the theory of operation used here is
31 * based on the self-tuning, low overhead replacement cache
32 * presented by Megiddo and Modha at FAST 2003, there are some
33 * significant differences:
35 * 1. The Megiddo and Modha model assumes any page is evictable.
36 * Pages in its cache cannot be "locked" into memory. This makes
37 * the eviction algorithm simple: evict the last page in the list.
38 * This also make the performance characteristics easy to reason
39 * about. Our cache is not so simple. At any given moment, some
40 * subset of the blocks in the cache are un-evictable because we
41 * have handed out a reference to them. Blocks are only evictable
42 * when there are no external references active. This makes
43 * eviction far more problematic: we choose to evict the evictable
44 * blocks that are the "lowest" in the list.
46 * There are times when it is not possible to evict the requested
47 * space. In these circumstances we are unable to adjust the cache
48 * size. To prevent the cache growing unbounded at these times we
49 * implement a "cache throttle" that slows the flow of new data
50 * into the cache until we can make space available.
52 * 2. The Megiddo and Modha model assumes a fixed cache size.
53 * Pages are evicted when the cache is full and there is a cache
54 * miss. Our model has a variable sized cache. It grows with
55 * high use, but also tries to react to memory pressure from the
56 * operating system: decreasing its size when system memory is
59 * 3. The Megiddo and Modha model assumes a fixed page size. All
60 * elements of the cache are therefor exactly the same size. So
61 * when adjusting the cache size following a cache miss, its simply
62 * a matter of choosing a single page to evict. In our model, we
63 * have variable sized cache blocks (rangeing from 512 bytes to
64 * 128K bytes). We therefor choose a set of blocks to evict to make
65 * space for a cache miss that approximates as closely as possible
66 * the space used by the new block.
68 * See also: "ARC: A Self-Tuning, Low Overhead Replacement Cache"
69 * by N. Megiddo & D. Modha, FAST 2003
75 * A new reference to a cache buffer can be obtained in two
76 * ways: 1) via a hash table lookup using the DVA as a key,
77 * or 2) via one of the ARC lists. The arc_read() interface
78 * uses method 1, while the internal arc algorithms for
79 * adjusting the cache use method 2. We therefor provide two
80 * types of locks: 1) the hash table lock array, and 2) the
83 * Buffers do not have their own mutexes, rather they rely on the
84 * hash table mutexes for the bulk of their protection (i.e. most
85 * fields in the arc_buf_hdr_t are protected by these mutexes).
87 * buf_hash_find() returns the appropriate mutex (held) when it
88 * locates the requested buffer in the hash table. It returns
89 * NULL for the mutex if the buffer was not in the table.
91 * buf_hash_remove() expects the appropriate hash mutex to be
92 * already held before it is invoked.
94 * Each arc state also has a mutex which is used to protect the
95 * buffer list associated with the state. When attempting to
96 * obtain a hash table lock while holding an arc list lock you
97 * must use: mutex_tryenter() to avoid deadlock. Also note that
98 * the active state mutex must be held before the ghost state mutex.
100 * Arc buffers may have an associated eviction callback function.
101 * This function will be invoked prior to removing the buffer (e.g.
102 * in arc_do_user_evicts()). Note however that the data associated
103 * with the buffer may be evicted prior to the callback. The callback
104 * must be made with *no locks held* (to prevent deadlock). Additionally,
105 * the users of callbacks must ensure that their private data is
106 * protected from simultaneous callbacks from arc_buf_evict()
107 * and arc_do_user_evicts().
109 * It as also possible to register a callback which is run when the
110 * arc_meta_limit is reached and no buffers can be safely evicted. In
111 * this case the arc user should drop a reference on some arc buffers so
112 * they can be reclaimed and the arc_meta_limit honored. For example,
113 * when using the ZPL each dentry holds a references on a znode. These
114 * dentries must be pruned before the arc buffer holding the znode can
117 * Note that the majority of the performance stats are manipulated
118 * with atomic operations.
120 * The L2ARC uses the l2arc_buflist_mtx global mutex for the following:
122 * - L2ARC buflist creation
123 * - L2ARC buflist eviction
124 * - L2ARC write completion, which walks L2ARC buflists
125 * - ARC header destruction, as it removes from L2ARC buflists
126 * - ARC header release, as it removes from L2ARC buflists
131 #include <sys/zfs_context.h>
133 #include <sys/vdev.h>
134 #include <sys/vdev_impl.h>
136 #include <sys/vmsystm.h>
138 #include <sys/fs/swapnode.h>
141 #include <sys/callb.h>
142 #include <sys/kstat.h>
143 #include <sys/dmu_tx.h>
144 #include <zfs_fletcher.h>
146 static kmutex_t arc_reclaim_thr_lock
;
147 static kcondvar_t arc_reclaim_thr_cv
; /* used to signal reclaim thr */
148 static uint8_t arc_thread_exit
;
150 /* number of bytes to prune from caches when at arc_meta_limit is reached */
151 int zfs_arc_meta_prune
= 1048576;
153 typedef enum arc_reclaim_strategy
{
154 ARC_RECLAIM_AGGR
, /* Aggressive reclaim strategy */
155 ARC_RECLAIM_CONS
/* Conservative reclaim strategy */
156 } arc_reclaim_strategy_t
;
158 /* number of seconds before growing cache again */
159 int zfs_arc_grow_retry
= 5;
161 /* shift of arc_c for calculating both min and max arc_p */
162 int zfs_arc_p_min_shift
= 4;
164 /* log2(fraction of arc to reclaim) */
165 int zfs_arc_shrink_shift
= 5;
168 * minimum lifespan of a prefetch block in clock ticks
169 * (initialized in arc_init())
171 int zfs_arc_min_prefetch_lifespan
= HZ
;
173 /* disable arc proactive arc throttle due to low memory */
174 int zfs_arc_memory_throttle_disable
= 1;
176 /* disable duplicate buffer eviction */
177 int zfs_disable_dup_eviction
= 0;
181 /* expiration time for arc_no_grow */
182 static clock_t arc_grow_time
= 0;
185 * The arc has filled available memory and has now warmed up.
187 static boolean_t arc_warm
;
190 * These tunables are for performance analysis.
192 unsigned long zfs_arc_max
= 0;
193 unsigned long zfs_arc_min
= 0;
194 unsigned long zfs_arc_meta_limit
= 0;
197 * Note that buffers can be in one of 6 states:
198 * ARC_anon - anonymous (discussed below)
199 * ARC_mru - recently used, currently cached
200 * ARC_mru_ghost - recentely used, no longer in cache
201 * ARC_mfu - frequently used, currently cached
202 * ARC_mfu_ghost - frequently used, no longer in cache
203 * ARC_l2c_only - exists in L2ARC but not other states
204 * When there are no active references to the buffer, they are
205 * are linked onto a list in one of these arc states. These are
206 * the only buffers that can be evicted or deleted. Within each
207 * state there are multiple lists, one for meta-data and one for
208 * non-meta-data. Meta-data (indirect blocks, blocks of dnodes,
209 * etc.) is tracked separately so that it can be managed more
210 * explicitly: favored over data, limited explicitly.
212 * Anonymous buffers are buffers that are not associated with
213 * a DVA. These are buffers that hold dirty block copies
214 * before they are written to stable storage. By definition,
215 * they are "ref'd" and are considered part of arc_mru
216 * that cannot be freed. Generally, they will aquire a DVA
217 * as they are written and migrate onto the arc_mru list.
219 * The ARC_l2c_only state is for buffers that are in the second
220 * level ARC but no longer in any of the ARC_m* lists. The second
221 * level ARC itself may also contain buffers that are in any of
222 * the ARC_m* states - meaning that a buffer can exist in two
223 * places. The reason for the ARC_l2c_only state is to keep the
224 * buffer header in the hash table, so that reads that hit the
225 * second level ARC benefit from these fast lookups.
228 typedef struct arc_state
{
229 list_t arcs_list
[ARC_BUFC_NUMTYPES
]; /* list of evictable buffers */
230 uint64_t arcs_lsize
[ARC_BUFC_NUMTYPES
]; /* amount of evictable data */
231 uint64_t arcs_size
; /* total amount of data in this state */
236 static arc_state_t ARC_anon
;
237 static arc_state_t ARC_mru
;
238 static arc_state_t ARC_mru_ghost
;
239 static arc_state_t ARC_mfu
;
240 static arc_state_t ARC_mfu_ghost
;
241 static arc_state_t ARC_l2c_only
;
243 typedef struct arc_stats
{
244 kstat_named_t arcstat_hits
;
245 kstat_named_t arcstat_misses
;
246 kstat_named_t arcstat_demand_data_hits
;
247 kstat_named_t arcstat_demand_data_misses
;
248 kstat_named_t arcstat_demand_metadata_hits
;
249 kstat_named_t arcstat_demand_metadata_misses
;
250 kstat_named_t arcstat_prefetch_data_hits
;
251 kstat_named_t arcstat_prefetch_data_misses
;
252 kstat_named_t arcstat_prefetch_metadata_hits
;
253 kstat_named_t arcstat_prefetch_metadata_misses
;
254 kstat_named_t arcstat_mru_hits
;
255 kstat_named_t arcstat_mru_ghost_hits
;
256 kstat_named_t arcstat_mfu_hits
;
257 kstat_named_t arcstat_mfu_ghost_hits
;
258 kstat_named_t arcstat_deleted
;
259 kstat_named_t arcstat_recycle_miss
;
260 kstat_named_t arcstat_mutex_miss
;
261 kstat_named_t arcstat_evict_skip
;
262 kstat_named_t arcstat_evict_l2_cached
;
263 kstat_named_t arcstat_evict_l2_eligible
;
264 kstat_named_t arcstat_evict_l2_ineligible
;
265 kstat_named_t arcstat_hash_elements
;
266 kstat_named_t arcstat_hash_elements_max
;
267 kstat_named_t arcstat_hash_collisions
;
268 kstat_named_t arcstat_hash_chains
;
269 kstat_named_t arcstat_hash_chain_max
;
270 kstat_named_t arcstat_p
;
271 kstat_named_t arcstat_c
;
272 kstat_named_t arcstat_c_min
;
273 kstat_named_t arcstat_c_max
;
274 kstat_named_t arcstat_size
;
275 kstat_named_t arcstat_hdr_size
;
276 kstat_named_t arcstat_data_size
;
277 kstat_named_t arcstat_other_size
;
278 kstat_named_t arcstat_anon_size
;
279 kstat_named_t arcstat_anon_evict_data
;
280 kstat_named_t arcstat_anon_evict_metadata
;
281 kstat_named_t arcstat_mru_size
;
282 kstat_named_t arcstat_mru_evict_data
;
283 kstat_named_t arcstat_mru_evict_metadata
;
284 kstat_named_t arcstat_mru_ghost_size
;
285 kstat_named_t arcstat_mru_ghost_evict_data
;
286 kstat_named_t arcstat_mru_ghost_evict_metadata
;
287 kstat_named_t arcstat_mfu_size
;
288 kstat_named_t arcstat_mfu_evict_data
;
289 kstat_named_t arcstat_mfu_evict_metadata
;
290 kstat_named_t arcstat_mfu_ghost_size
;
291 kstat_named_t arcstat_mfu_ghost_evict_data
;
292 kstat_named_t arcstat_mfu_ghost_evict_metadata
;
293 kstat_named_t arcstat_l2_hits
;
294 kstat_named_t arcstat_l2_misses
;
295 kstat_named_t arcstat_l2_feeds
;
296 kstat_named_t arcstat_l2_rw_clash
;
297 kstat_named_t arcstat_l2_read_bytes
;
298 kstat_named_t arcstat_l2_write_bytes
;
299 kstat_named_t arcstat_l2_writes_sent
;
300 kstat_named_t arcstat_l2_writes_done
;
301 kstat_named_t arcstat_l2_writes_error
;
302 kstat_named_t arcstat_l2_writes_hdr_miss
;
303 kstat_named_t arcstat_l2_evict_lock_retry
;
304 kstat_named_t arcstat_l2_evict_reading
;
305 kstat_named_t arcstat_l2_free_on_write
;
306 kstat_named_t arcstat_l2_abort_lowmem
;
307 kstat_named_t arcstat_l2_cksum_bad
;
308 kstat_named_t arcstat_l2_io_error
;
309 kstat_named_t arcstat_l2_size
;
310 kstat_named_t arcstat_l2_hdr_size
;
311 kstat_named_t arcstat_memory_throttle_count
;
312 kstat_named_t arcstat_duplicate_buffers
;
313 kstat_named_t arcstat_duplicate_buffers_size
;
314 kstat_named_t arcstat_duplicate_reads
;
315 kstat_named_t arcstat_memory_direct_count
;
316 kstat_named_t arcstat_memory_indirect_count
;
317 kstat_named_t arcstat_no_grow
;
318 kstat_named_t arcstat_tempreserve
;
319 kstat_named_t arcstat_loaned_bytes
;
320 kstat_named_t arcstat_prune
;
321 kstat_named_t arcstat_meta_used
;
322 kstat_named_t arcstat_meta_limit
;
323 kstat_named_t arcstat_meta_max
;
326 static arc_stats_t arc_stats
= {
327 { "hits", KSTAT_DATA_UINT64
},
328 { "misses", KSTAT_DATA_UINT64
},
329 { "demand_data_hits", KSTAT_DATA_UINT64
},
330 { "demand_data_misses", KSTAT_DATA_UINT64
},
331 { "demand_metadata_hits", KSTAT_DATA_UINT64
},
332 { "demand_metadata_misses", KSTAT_DATA_UINT64
},
333 { "prefetch_data_hits", KSTAT_DATA_UINT64
},
334 { "prefetch_data_misses", KSTAT_DATA_UINT64
},
335 { "prefetch_metadata_hits", KSTAT_DATA_UINT64
},
336 { "prefetch_metadata_misses", KSTAT_DATA_UINT64
},
337 { "mru_hits", KSTAT_DATA_UINT64
},
338 { "mru_ghost_hits", KSTAT_DATA_UINT64
},
339 { "mfu_hits", KSTAT_DATA_UINT64
},
340 { "mfu_ghost_hits", KSTAT_DATA_UINT64
},
341 { "deleted", KSTAT_DATA_UINT64
},
342 { "recycle_miss", KSTAT_DATA_UINT64
},
343 { "mutex_miss", KSTAT_DATA_UINT64
},
344 { "evict_skip", KSTAT_DATA_UINT64
},
345 { "evict_l2_cached", KSTAT_DATA_UINT64
},
346 { "evict_l2_eligible", KSTAT_DATA_UINT64
},
347 { "evict_l2_ineligible", KSTAT_DATA_UINT64
},
348 { "hash_elements", KSTAT_DATA_UINT64
},
349 { "hash_elements_max", KSTAT_DATA_UINT64
},
350 { "hash_collisions", KSTAT_DATA_UINT64
},
351 { "hash_chains", KSTAT_DATA_UINT64
},
352 { "hash_chain_max", KSTAT_DATA_UINT64
},
353 { "p", KSTAT_DATA_UINT64
},
354 { "c", KSTAT_DATA_UINT64
},
355 { "c_min", KSTAT_DATA_UINT64
},
356 { "c_max", KSTAT_DATA_UINT64
},
357 { "size", KSTAT_DATA_UINT64
},
358 { "hdr_size", KSTAT_DATA_UINT64
},
359 { "data_size", KSTAT_DATA_UINT64
},
360 { "other_size", KSTAT_DATA_UINT64
},
361 { "anon_size", KSTAT_DATA_UINT64
},
362 { "anon_evict_data", KSTAT_DATA_UINT64
},
363 { "anon_evict_metadata", KSTAT_DATA_UINT64
},
364 { "mru_size", KSTAT_DATA_UINT64
},
365 { "mru_evict_data", KSTAT_DATA_UINT64
},
366 { "mru_evict_metadata", KSTAT_DATA_UINT64
},
367 { "mru_ghost_size", KSTAT_DATA_UINT64
},
368 { "mru_ghost_evict_data", KSTAT_DATA_UINT64
},
369 { "mru_ghost_evict_metadata", KSTAT_DATA_UINT64
},
370 { "mfu_size", KSTAT_DATA_UINT64
},
371 { "mfu_evict_data", KSTAT_DATA_UINT64
},
372 { "mfu_evict_metadata", KSTAT_DATA_UINT64
},
373 { "mfu_ghost_size", KSTAT_DATA_UINT64
},
374 { "mfu_ghost_evict_data", KSTAT_DATA_UINT64
},
375 { "mfu_ghost_evict_metadata", KSTAT_DATA_UINT64
},
376 { "l2_hits", KSTAT_DATA_UINT64
},
377 { "l2_misses", KSTAT_DATA_UINT64
},
378 { "l2_feeds", KSTAT_DATA_UINT64
},
379 { "l2_rw_clash", KSTAT_DATA_UINT64
},
380 { "l2_read_bytes", KSTAT_DATA_UINT64
},
381 { "l2_write_bytes", KSTAT_DATA_UINT64
},
382 { "l2_writes_sent", KSTAT_DATA_UINT64
},
383 { "l2_writes_done", KSTAT_DATA_UINT64
},
384 { "l2_writes_error", KSTAT_DATA_UINT64
},
385 { "l2_writes_hdr_miss", KSTAT_DATA_UINT64
},
386 { "l2_evict_lock_retry", KSTAT_DATA_UINT64
},
387 { "l2_evict_reading", KSTAT_DATA_UINT64
},
388 { "l2_free_on_write", KSTAT_DATA_UINT64
},
389 { "l2_abort_lowmem", KSTAT_DATA_UINT64
},
390 { "l2_cksum_bad", KSTAT_DATA_UINT64
},
391 { "l2_io_error", KSTAT_DATA_UINT64
},
392 { "l2_size", KSTAT_DATA_UINT64
},
393 { "l2_hdr_size", KSTAT_DATA_UINT64
},
394 { "memory_throttle_count", KSTAT_DATA_UINT64
},
395 { "duplicate_buffers", KSTAT_DATA_UINT64
},
396 { "duplicate_buffers_size", KSTAT_DATA_UINT64
},
397 { "duplicate_reads", KSTAT_DATA_UINT64
},
398 { "memory_direct_count", KSTAT_DATA_UINT64
},
399 { "memory_indirect_count", KSTAT_DATA_UINT64
},
400 { "arc_no_grow", KSTAT_DATA_UINT64
},
401 { "arc_tempreserve", KSTAT_DATA_UINT64
},
402 { "arc_loaned_bytes", KSTAT_DATA_UINT64
},
403 { "arc_prune", KSTAT_DATA_UINT64
},
404 { "arc_meta_used", KSTAT_DATA_UINT64
},
405 { "arc_meta_limit", KSTAT_DATA_UINT64
},
406 { "arc_meta_max", KSTAT_DATA_UINT64
},
409 #define ARCSTAT(stat) (arc_stats.stat.value.ui64)
411 #define ARCSTAT_INCR(stat, val) \
412 atomic_add_64(&arc_stats.stat.value.ui64, (val));
414 #define ARCSTAT_BUMP(stat) ARCSTAT_INCR(stat, 1)
415 #define ARCSTAT_BUMPDOWN(stat) ARCSTAT_INCR(stat, -1)
417 #define ARCSTAT_MAX(stat, val) { \
419 while ((val) > (m = arc_stats.stat.value.ui64) && \
420 (m != atomic_cas_64(&arc_stats.stat.value.ui64, m, (val)))) \
424 #define ARCSTAT_MAXSTAT(stat) \
425 ARCSTAT_MAX(stat##_max, arc_stats.stat.value.ui64)
428 * We define a macro to allow ARC hits/misses to be easily broken down by
429 * two separate conditions, giving a total of four different subtypes for
430 * each of hits and misses (so eight statistics total).
432 #define ARCSTAT_CONDSTAT(cond1, stat1, notstat1, cond2, stat2, notstat2, stat) \
435 ARCSTAT_BUMP(arcstat_##stat1##_##stat2##_##stat); \
437 ARCSTAT_BUMP(arcstat_##stat1##_##notstat2##_##stat); \
441 ARCSTAT_BUMP(arcstat_##notstat1##_##stat2##_##stat); \
443 ARCSTAT_BUMP(arcstat_##notstat1##_##notstat2##_##stat);\
448 static arc_state_t
*arc_anon
;
449 static arc_state_t
*arc_mru
;
450 static arc_state_t
*arc_mru_ghost
;
451 static arc_state_t
*arc_mfu
;
452 static arc_state_t
*arc_mfu_ghost
;
453 static arc_state_t
*arc_l2c_only
;
456 * There are several ARC variables that are critical to export as kstats --
457 * but we don't want to have to grovel around in the kstat whenever we wish to
458 * manipulate them. For these variables, we therefore define them to be in
459 * terms of the statistic variable. This assures that we are not introducing
460 * the possibility of inconsistency by having shadow copies of the variables,
461 * while still allowing the code to be readable.
463 #define arc_size ARCSTAT(arcstat_size) /* actual total arc size */
464 #define arc_p ARCSTAT(arcstat_p) /* target size of MRU */
465 #define arc_c ARCSTAT(arcstat_c) /* target size of cache */
466 #define arc_c_min ARCSTAT(arcstat_c_min) /* min target cache size */
467 #define arc_c_max ARCSTAT(arcstat_c_max) /* max target cache size */
468 #define arc_no_grow ARCSTAT(arcstat_no_grow)
469 #define arc_tempreserve ARCSTAT(arcstat_tempreserve)
470 #define arc_loaned_bytes ARCSTAT(arcstat_loaned_bytes)
471 #define arc_meta_used ARCSTAT(arcstat_meta_used)
472 #define arc_meta_limit ARCSTAT(arcstat_meta_limit)
473 #define arc_meta_max ARCSTAT(arcstat_meta_max)
475 typedef struct l2arc_buf_hdr l2arc_buf_hdr_t
;
477 typedef struct arc_callback arc_callback_t
;
479 struct arc_callback
{
481 arc_done_func_t
*acb_done
;
483 zio_t
*acb_zio_dummy
;
484 arc_callback_t
*acb_next
;
487 typedef struct arc_write_callback arc_write_callback_t
;
489 struct arc_write_callback
{
491 arc_done_func_t
*awcb_ready
;
492 arc_done_func_t
*awcb_done
;
497 /* protected by hash lock */
502 kmutex_t b_freeze_lock
;
503 zio_cksum_t
*b_freeze_cksum
;
505 arc_buf_hdr_t
*b_hash_next
;
510 arc_callback_t
*b_acb
;
514 arc_buf_contents_t b_type
;
518 /* protected by arc state mutex */
519 arc_state_t
*b_state
;
520 list_node_t b_arc_node
;
522 /* updated atomically */
523 clock_t b_arc_access
;
525 /* self protecting */
528 l2arc_buf_hdr_t
*b_l2hdr
;
529 list_node_t b_l2node
;
532 static list_t arc_prune_list
;
533 static kmutex_t arc_prune_mtx
;
534 static arc_buf_t
*arc_eviction_list
;
535 static kmutex_t arc_eviction_mtx
;
536 static arc_buf_hdr_t arc_eviction_hdr
;
537 static void arc_get_data_buf(arc_buf_t
*buf
);
538 static void arc_access(arc_buf_hdr_t
*buf
, kmutex_t
*hash_lock
);
539 static int arc_evict_needed(arc_buf_contents_t type
);
540 static void arc_evict_ghost(arc_state_t
*state
, uint64_t spa
, int64_t bytes
);
542 static boolean_t
l2arc_write_eligible(uint64_t spa_guid
, arc_buf_hdr_t
*ab
);
544 #define GHOST_STATE(state) \
545 ((state) == arc_mru_ghost || (state) == arc_mfu_ghost || \
546 (state) == arc_l2c_only)
549 * Private ARC flags. These flags are private ARC only flags that will show up
550 * in b_flags in the arc_hdr_buf_t. Some flags are publicly declared, and can
551 * be passed in as arc_flags in things like arc_read. However, these flags
552 * should never be passed and should only be set by ARC code. When adding new
553 * public flags, make sure not to smash the private ones.
556 #define ARC_IN_HASH_TABLE (1 << 9) /* this buffer is hashed */
557 #define ARC_IO_IN_PROGRESS (1 << 10) /* I/O in progress for buf */
558 #define ARC_IO_ERROR (1 << 11) /* I/O failed for buf */
559 #define ARC_FREED_IN_READ (1 << 12) /* buf freed while in read */
560 #define ARC_BUF_AVAILABLE (1 << 13) /* block not in active use */
561 #define ARC_INDIRECT (1 << 14) /* this is an indirect block */
562 #define ARC_FREE_IN_PROGRESS (1 << 15) /* hdr about to be freed */
563 #define ARC_L2_WRITING (1 << 16) /* L2ARC write in progress */
564 #define ARC_L2_EVICTED (1 << 17) /* evicted during I/O */
565 #define ARC_L2_WRITE_HEAD (1 << 18) /* head of write list */
567 #define HDR_IN_HASH_TABLE(hdr) ((hdr)->b_flags & ARC_IN_HASH_TABLE)
568 #define HDR_IO_IN_PROGRESS(hdr) ((hdr)->b_flags & ARC_IO_IN_PROGRESS)
569 #define HDR_IO_ERROR(hdr) ((hdr)->b_flags & ARC_IO_ERROR)
570 #define HDR_PREFETCH(hdr) ((hdr)->b_flags & ARC_PREFETCH)
571 #define HDR_FREED_IN_READ(hdr) ((hdr)->b_flags & ARC_FREED_IN_READ)
572 #define HDR_BUF_AVAILABLE(hdr) ((hdr)->b_flags & ARC_BUF_AVAILABLE)
573 #define HDR_FREE_IN_PROGRESS(hdr) ((hdr)->b_flags & ARC_FREE_IN_PROGRESS)
574 #define HDR_L2CACHE(hdr) ((hdr)->b_flags & ARC_L2CACHE)
575 #define HDR_L2_READING(hdr) ((hdr)->b_flags & ARC_IO_IN_PROGRESS && \
576 (hdr)->b_l2hdr != NULL)
577 #define HDR_L2_WRITING(hdr) ((hdr)->b_flags & ARC_L2_WRITING)
578 #define HDR_L2_EVICTED(hdr) ((hdr)->b_flags & ARC_L2_EVICTED)
579 #define HDR_L2_WRITE_HEAD(hdr) ((hdr)->b_flags & ARC_L2_WRITE_HEAD)
585 #define HDR_SIZE ((int64_t)sizeof (arc_buf_hdr_t))
586 #define L2HDR_SIZE ((int64_t)sizeof (l2arc_buf_hdr_t))
589 * Hash table routines
592 #define HT_LOCK_ALIGN 64
593 #define HT_LOCK_PAD (P2NPHASE(sizeof (kmutex_t), (HT_LOCK_ALIGN)))
598 unsigned char pad
[HT_LOCK_PAD
];
602 #define BUF_LOCKS 256
603 typedef struct buf_hash_table
{
605 arc_buf_hdr_t
**ht_table
;
606 struct ht_lock ht_locks
[BUF_LOCKS
];
609 static buf_hash_table_t buf_hash_table
;
611 #define BUF_HASH_INDEX(spa, dva, birth) \
612 (buf_hash(spa, dva, birth) & buf_hash_table.ht_mask)
613 #define BUF_HASH_LOCK_NTRY(idx) (buf_hash_table.ht_locks[idx & (BUF_LOCKS-1)])
614 #define BUF_HASH_LOCK(idx) (&(BUF_HASH_LOCK_NTRY(idx).ht_lock))
615 #define HDR_LOCK(hdr) \
616 (BUF_HASH_LOCK(BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth)))
618 uint64_t zfs_crc64_table
[256];
624 #define L2ARC_WRITE_SIZE (8 * 1024 * 1024) /* initial write max */
625 #define L2ARC_HEADROOM 2 /* num of writes */
626 #define L2ARC_FEED_SECS 1 /* caching interval secs */
627 #define L2ARC_FEED_MIN_MS 200 /* min caching interval ms */
629 #define l2arc_writes_sent ARCSTAT(arcstat_l2_writes_sent)
630 #define l2arc_writes_done ARCSTAT(arcstat_l2_writes_done)
633 * L2ARC Performance Tunables
635 unsigned long l2arc_write_max
= L2ARC_WRITE_SIZE
; /* def max write size */
636 unsigned long l2arc_write_boost
= L2ARC_WRITE_SIZE
; /* extra warmup write */
637 unsigned long l2arc_headroom
= L2ARC_HEADROOM
; /* # of dev writes */
638 unsigned long l2arc_feed_secs
= L2ARC_FEED_SECS
; /* interval seconds */
639 unsigned long l2arc_feed_min_ms
= L2ARC_FEED_MIN_MS
; /* min interval msecs */
640 int l2arc_noprefetch
= B_TRUE
; /* don't cache prefetch bufs */
641 int l2arc_feed_again
= B_TRUE
; /* turbo warmup */
642 int l2arc_norw
= B_FALSE
; /* no reads during writes */
647 typedef struct l2arc_dev
{
648 vdev_t
*l2ad_vdev
; /* vdev */
649 spa_t
*l2ad_spa
; /* spa */
650 uint64_t l2ad_hand
; /* next write location */
651 uint64_t l2ad_write
; /* desired write size, bytes */
652 uint64_t l2ad_boost
; /* warmup write boost, bytes */
653 uint64_t l2ad_start
; /* first addr on device */
654 uint64_t l2ad_end
; /* last addr on device */
655 uint64_t l2ad_evict
; /* last addr eviction reached */
656 boolean_t l2ad_first
; /* first sweep through */
657 boolean_t l2ad_writing
; /* currently writing */
658 list_t
*l2ad_buflist
; /* buffer list */
659 list_node_t l2ad_node
; /* device list node */
662 static list_t L2ARC_dev_list
; /* device list */
663 static list_t
*l2arc_dev_list
; /* device list pointer */
664 static kmutex_t l2arc_dev_mtx
; /* device list mutex */
665 static l2arc_dev_t
*l2arc_dev_last
; /* last device used */
666 static kmutex_t l2arc_buflist_mtx
; /* mutex for all buflists */
667 static list_t L2ARC_free_on_write
; /* free after write buf list */
668 static list_t
*l2arc_free_on_write
; /* free after write list ptr */
669 static kmutex_t l2arc_free_on_write_mtx
; /* mutex for list */
670 static uint64_t l2arc_ndev
; /* number of devices */
672 typedef struct l2arc_read_callback
{
673 arc_buf_t
*l2rcb_buf
; /* read buffer */
674 spa_t
*l2rcb_spa
; /* spa */
675 blkptr_t l2rcb_bp
; /* original blkptr */
676 zbookmark_t l2rcb_zb
; /* original bookmark */
677 int l2rcb_flags
; /* original flags */
678 } l2arc_read_callback_t
;
680 typedef struct l2arc_write_callback
{
681 l2arc_dev_t
*l2wcb_dev
; /* device info */
682 arc_buf_hdr_t
*l2wcb_head
; /* head of write buflist */
683 } l2arc_write_callback_t
;
685 struct l2arc_buf_hdr
{
686 /* protected by arc_buf_hdr mutex */
687 l2arc_dev_t
*b_dev
; /* L2ARC device */
688 uint64_t b_daddr
; /* disk address, offset byte */
691 typedef struct l2arc_data_free
{
692 /* protected by l2arc_free_on_write_mtx */
695 void (*l2df_func
)(void *, size_t);
696 list_node_t l2df_list_node
;
699 static kmutex_t l2arc_feed_thr_lock
;
700 static kcondvar_t l2arc_feed_thr_cv
;
701 static uint8_t l2arc_thread_exit
;
703 static void l2arc_read_done(zio_t
*zio
);
704 static void l2arc_hdr_stat_add(void);
705 static void l2arc_hdr_stat_remove(void);
708 buf_hash(uint64_t spa
, const dva_t
*dva
, uint64_t birth
)
710 uint8_t *vdva
= (uint8_t *)dva
;
711 uint64_t crc
= -1ULL;
714 ASSERT(zfs_crc64_table
[128] == ZFS_CRC64_POLY
);
716 for (i
= 0; i
< sizeof (dva_t
); i
++)
717 crc
= (crc
>> 8) ^ zfs_crc64_table
[(crc
^ vdva
[i
]) & 0xFF];
719 crc
^= (spa
>>8) ^ birth
;
724 #define BUF_EMPTY(buf) \
725 ((buf)->b_dva.dva_word[0] == 0 && \
726 (buf)->b_dva.dva_word[1] == 0 && \
729 #define BUF_EQUAL(spa, dva, birth, buf) \
730 ((buf)->b_dva.dva_word[0] == (dva)->dva_word[0]) && \
731 ((buf)->b_dva.dva_word[1] == (dva)->dva_word[1]) && \
732 ((buf)->b_birth == birth) && ((buf)->b_spa == spa)
735 buf_discard_identity(arc_buf_hdr_t
*hdr
)
737 hdr
->b_dva
.dva_word
[0] = 0;
738 hdr
->b_dva
.dva_word
[1] = 0;
743 static arc_buf_hdr_t
*
744 buf_hash_find(uint64_t spa
, const dva_t
*dva
, uint64_t birth
, kmutex_t
**lockp
)
746 uint64_t idx
= BUF_HASH_INDEX(spa
, dva
, birth
);
747 kmutex_t
*hash_lock
= BUF_HASH_LOCK(idx
);
750 mutex_enter(hash_lock
);
751 for (buf
= buf_hash_table
.ht_table
[idx
]; buf
!= NULL
;
752 buf
= buf
->b_hash_next
) {
753 if (BUF_EQUAL(spa
, dva
, birth
, buf
)) {
758 mutex_exit(hash_lock
);
764 * Insert an entry into the hash table. If there is already an element
765 * equal to elem in the hash table, then the already existing element
766 * will be returned and the new element will not be inserted.
767 * Otherwise returns NULL.
769 static arc_buf_hdr_t
*
770 buf_hash_insert(arc_buf_hdr_t
*buf
, kmutex_t
**lockp
)
772 uint64_t idx
= BUF_HASH_INDEX(buf
->b_spa
, &buf
->b_dva
, buf
->b_birth
);
773 kmutex_t
*hash_lock
= BUF_HASH_LOCK(idx
);
777 ASSERT(!HDR_IN_HASH_TABLE(buf
));
779 mutex_enter(hash_lock
);
780 for (fbuf
= buf_hash_table
.ht_table
[idx
], i
= 0; fbuf
!= NULL
;
781 fbuf
= fbuf
->b_hash_next
, i
++) {
782 if (BUF_EQUAL(buf
->b_spa
, &buf
->b_dva
, buf
->b_birth
, fbuf
))
786 buf
->b_hash_next
= buf_hash_table
.ht_table
[idx
];
787 buf_hash_table
.ht_table
[idx
] = buf
;
788 buf
->b_flags
|= ARC_IN_HASH_TABLE
;
790 /* collect some hash table performance data */
792 ARCSTAT_BUMP(arcstat_hash_collisions
);
794 ARCSTAT_BUMP(arcstat_hash_chains
);
796 ARCSTAT_MAX(arcstat_hash_chain_max
, i
);
799 ARCSTAT_BUMP(arcstat_hash_elements
);
800 ARCSTAT_MAXSTAT(arcstat_hash_elements
);
806 buf_hash_remove(arc_buf_hdr_t
*buf
)
808 arc_buf_hdr_t
*fbuf
, **bufp
;
809 uint64_t idx
= BUF_HASH_INDEX(buf
->b_spa
, &buf
->b_dva
, buf
->b_birth
);
811 ASSERT(MUTEX_HELD(BUF_HASH_LOCK(idx
)));
812 ASSERT(HDR_IN_HASH_TABLE(buf
));
814 bufp
= &buf_hash_table
.ht_table
[idx
];
815 while ((fbuf
= *bufp
) != buf
) {
816 ASSERT(fbuf
!= NULL
);
817 bufp
= &fbuf
->b_hash_next
;
819 *bufp
= buf
->b_hash_next
;
820 buf
->b_hash_next
= NULL
;
821 buf
->b_flags
&= ~ARC_IN_HASH_TABLE
;
823 /* collect some hash table performance data */
824 ARCSTAT_BUMPDOWN(arcstat_hash_elements
);
826 if (buf_hash_table
.ht_table
[idx
] &&
827 buf_hash_table
.ht_table
[idx
]->b_hash_next
== NULL
)
828 ARCSTAT_BUMPDOWN(arcstat_hash_chains
);
832 * Global data structures and functions for the buf kmem cache.
834 static kmem_cache_t
*hdr_cache
;
835 static kmem_cache_t
*buf_cache
;
842 #if defined(_KERNEL) && defined(HAVE_SPL)
843 /* Large allocations which do not require contiguous pages
844 * should be using vmem_free() in the linux kernel */
845 vmem_free(buf_hash_table
.ht_table
,
846 (buf_hash_table
.ht_mask
+ 1) * sizeof (void *));
848 kmem_free(buf_hash_table
.ht_table
,
849 (buf_hash_table
.ht_mask
+ 1) * sizeof (void *));
851 for (i
= 0; i
< BUF_LOCKS
; i
++)
852 mutex_destroy(&buf_hash_table
.ht_locks
[i
].ht_lock
);
853 kmem_cache_destroy(hdr_cache
);
854 kmem_cache_destroy(buf_cache
);
858 * Constructor callback - called when the cache is empty
859 * and a new buf is requested.
863 hdr_cons(void *vbuf
, void *unused
, int kmflag
)
865 arc_buf_hdr_t
*buf
= vbuf
;
867 bzero(buf
, sizeof (arc_buf_hdr_t
));
868 refcount_create(&buf
->b_refcnt
);
869 cv_init(&buf
->b_cv
, NULL
, CV_DEFAULT
, NULL
);
870 mutex_init(&buf
->b_freeze_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
871 list_link_init(&buf
->b_arc_node
);
872 list_link_init(&buf
->b_l2node
);
873 arc_space_consume(sizeof (arc_buf_hdr_t
), ARC_SPACE_HDRS
);
880 buf_cons(void *vbuf
, void *unused
, int kmflag
)
882 arc_buf_t
*buf
= vbuf
;
884 bzero(buf
, sizeof (arc_buf_t
));
885 mutex_init(&buf
->b_evict_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
886 arc_space_consume(sizeof (arc_buf_t
), ARC_SPACE_HDRS
);
892 * Destructor callback - called when a cached buf is
893 * no longer required.
897 hdr_dest(void *vbuf
, void *unused
)
899 arc_buf_hdr_t
*buf
= vbuf
;
901 ASSERT(BUF_EMPTY(buf
));
902 refcount_destroy(&buf
->b_refcnt
);
903 cv_destroy(&buf
->b_cv
);
904 mutex_destroy(&buf
->b_freeze_lock
);
905 arc_space_return(sizeof (arc_buf_hdr_t
), ARC_SPACE_HDRS
);
910 buf_dest(void *vbuf
, void *unused
)
912 arc_buf_t
*buf
= vbuf
;
914 mutex_destroy(&buf
->b_evict_lock
);
915 arc_space_return(sizeof (arc_buf_t
), ARC_SPACE_HDRS
);
922 uint64_t hsize
= 1ULL << 12;
926 * The hash table is big enough to fill all of physical memory
927 * with an average 64K block size. The table will take up
928 * totalmem*sizeof(void*)/64K (eg. 128KB/GB with 8-byte pointers).
930 while (hsize
* 65536 < physmem
* PAGESIZE
)
933 buf_hash_table
.ht_mask
= hsize
- 1;
934 #if defined(_KERNEL) && defined(HAVE_SPL)
935 /* Large allocations which do not require contiguous pages
936 * should be using vmem_alloc() in the linux kernel */
937 buf_hash_table
.ht_table
=
938 vmem_zalloc(hsize
* sizeof (void*), KM_SLEEP
);
940 buf_hash_table
.ht_table
=
941 kmem_zalloc(hsize
* sizeof (void*), KM_NOSLEEP
);
943 if (buf_hash_table
.ht_table
== NULL
) {
944 ASSERT(hsize
> (1ULL << 8));
949 hdr_cache
= kmem_cache_create("arc_buf_hdr_t", sizeof (arc_buf_hdr_t
),
950 0, hdr_cons
, hdr_dest
, NULL
, NULL
, NULL
, 0);
951 buf_cache
= kmem_cache_create("arc_buf_t", sizeof (arc_buf_t
),
952 0, buf_cons
, buf_dest
, NULL
, NULL
, NULL
, 0);
954 for (i
= 0; i
< 256; i
++)
955 for (ct
= zfs_crc64_table
+ i
, *ct
= i
, j
= 8; j
> 0; j
--)
956 *ct
= (*ct
>> 1) ^ (-(*ct
& 1) & ZFS_CRC64_POLY
);
958 for (i
= 0; i
< BUF_LOCKS
; i
++) {
959 mutex_init(&buf_hash_table
.ht_locks
[i
].ht_lock
,
960 NULL
, MUTEX_DEFAULT
, NULL
);
964 #define ARC_MINTIME (hz>>4) /* 62 ms */
967 arc_cksum_verify(arc_buf_t
*buf
)
971 if (!(zfs_flags
& ZFS_DEBUG_MODIFY
))
974 mutex_enter(&buf
->b_hdr
->b_freeze_lock
);
975 if (buf
->b_hdr
->b_freeze_cksum
== NULL
||
976 (buf
->b_hdr
->b_flags
& ARC_IO_ERROR
)) {
977 mutex_exit(&buf
->b_hdr
->b_freeze_lock
);
980 fletcher_2_native(buf
->b_data
, buf
->b_hdr
->b_size
, &zc
);
981 if (!ZIO_CHECKSUM_EQUAL(*buf
->b_hdr
->b_freeze_cksum
, zc
))
982 panic("buffer modified while frozen!");
983 mutex_exit(&buf
->b_hdr
->b_freeze_lock
);
987 arc_cksum_equal(arc_buf_t
*buf
)
992 mutex_enter(&buf
->b_hdr
->b_freeze_lock
);
993 fletcher_2_native(buf
->b_data
, buf
->b_hdr
->b_size
, &zc
);
994 equal
= ZIO_CHECKSUM_EQUAL(*buf
->b_hdr
->b_freeze_cksum
, zc
);
995 mutex_exit(&buf
->b_hdr
->b_freeze_lock
);
1001 arc_cksum_compute(arc_buf_t
*buf
, boolean_t force
)
1003 if (!force
&& !(zfs_flags
& ZFS_DEBUG_MODIFY
))
1006 mutex_enter(&buf
->b_hdr
->b_freeze_lock
);
1007 if (buf
->b_hdr
->b_freeze_cksum
!= NULL
) {
1008 mutex_exit(&buf
->b_hdr
->b_freeze_lock
);
1011 buf
->b_hdr
->b_freeze_cksum
= kmem_alloc(sizeof (zio_cksum_t
),
1013 fletcher_2_native(buf
->b_data
, buf
->b_hdr
->b_size
,
1014 buf
->b_hdr
->b_freeze_cksum
);
1015 mutex_exit(&buf
->b_hdr
->b_freeze_lock
);
1019 arc_buf_thaw(arc_buf_t
*buf
)
1021 if (zfs_flags
& ZFS_DEBUG_MODIFY
) {
1022 if (buf
->b_hdr
->b_state
!= arc_anon
)
1023 panic("modifying non-anon buffer!");
1024 if (buf
->b_hdr
->b_flags
& ARC_IO_IN_PROGRESS
)
1025 panic("modifying buffer while i/o in progress!");
1026 arc_cksum_verify(buf
);
1029 mutex_enter(&buf
->b_hdr
->b_freeze_lock
);
1030 if (buf
->b_hdr
->b_freeze_cksum
!= NULL
) {
1031 kmem_free(buf
->b_hdr
->b_freeze_cksum
, sizeof (zio_cksum_t
));
1032 buf
->b_hdr
->b_freeze_cksum
= NULL
;
1035 mutex_exit(&buf
->b_hdr
->b_freeze_lock
);
1039 arc_buf_freeze(arc_buf_t
*buf
)
1041 kmutex_t
*hash_lock
;
1043 if (!(zfs_flags
& ZFS_DEBUG_MODIFY
))
1046 hash_lock
= HDR_LOCK(buf
->b_hdr
);
1047 mutex_enter(hash_lock
);
1049 ASSERT(buf
->b_hdr
->b_freeze_cksum
!= NULL
||
1050 buf
->b_hdr
->b_state
== arc_anon
);
1051 arc_cksum_compute(buf
, B_FALSE
);
1052 mutex_exit(hash_lock
);
1056 add_reference(arc_buf_hdr_t
*ab
, kmutex_t
*hash_lock
, void *tag
)
1058 ASSERT(MUTEX_HELD(hash_lock
));
1060 if ((refcount_add(&ab
->b_refcnt
, tag
) == 1) &&
1061 (ab
->b_state
!= arc_anon
)) {
1062 uint64_t delta
= ab
->b_size
* ab
->b_datacnt
;
1063 list_t
*list
= &ab
->b_state
->arcs_list
[ab
->b_type
];
1064 uint64_t *size
= &ab
->b_state
->arcs_lsize
[ab
->b_type
];
1066 ASSERT(!MUTEX_HELD(&ab
->b_state
->arcs_mtx
));
1067 mutex_enter(&ab
->b_state
->arcs_mtx
);
1068 ASSERT(list_link_active(&ab
->b_arc_node
));
1069 list_remove(list
, ab
);
1070 if (GHOST_STATE(ab
->b_state
)) {
1071 ASSERT0(ab
->b_datacnt
);
1072 ASSERT3P(ab
->b_buf
, ==, NULL
);
1076 ASSERT3U(*size
, >=, delta
);
1077 atomic_add_64(size
, -delta
);
1078 mutex_exit(&ab
->b_state
->arcs_mtx
);
1079 /* remove the prefetch flag if we get a reference */
1080 if (ab
->b_flags
& ARC_PREFETCH
)
1081 ab
->b_flags
&= ~ARC_PREFETCH
;
1086 remove_reference(arc_buf_hdr_t
*ab
, kmutex_t
*hash_lock
, void *tag
)
1089 arc_state_t
*state
= ab
->b_state
;
1091 ASSERT(state
== arc_anon
|| MUTEX_HELD(hash_lock
));
1092 ASSERT(!GHOST_STATE(state
));
1094 if (((cnt
= refcount_remove(&ab
->b_refcnt
, tag
)) == 0) &&
1095 (state
!= arc_anon
)) {
1096 uint64_t *size
= &state
->arcs_lsize
[ab
->b_type
];
1098 ASSERT(!MUTEX_HELD(&state
->arcs_mtx
));
1099 mutex_enter(&state
->arcs_mtx
);
1100 ASSERT(!list_link_active(&ab
->b_arc_node
));
1101 list_insert_head(&state
->arcs_list
[ab
->b_type
], ab
);
1102 ASSERT(ab
->b_datacnt
> 0);
1103 atomic_add_64(size
, ab
->b_size
* ab
->b_datacnt
);
1104 mutex_exit(&state
->arcs_mtx
);
1110 * Move the supplied buffer to the indicated state. The mutex
1111 * for the buffer must be held by the caller.
1114 arc_change_state(arc_state_t
*new_state
, arc_buf_hdr_t
*ab
, kmutex_t
*hash_lock
)
1116 arc_state_t
*old_state
= ab
->b_state
;
1117 int64_t refcnt
= refcount_count(&ab
->b_refcnt
);
1118 uint64_t from_delta
, to_delta
;
1120 ASSERT(MUTEX_HELD(hash_lock
));
1121 ASSERT(new_state
!= old_state
);
1122 ASSERT(refcnt
== 0 || ab
->b_datacnt
> 0);
1123 ASSERT(ab
->b_datacnt
== 0 || !GHOST_STATE(new_state
));
1124 ASSERT(ab
->b_datacnt
<= 1 || old_state
!= arc_anon
);
1126 from_delta
= to_delta
= ab
->b_datacnt
* ab
->b_size
;
1129 * If this buffer is evictable, transfer it from the
1130 * old state list to the new state list.
1133 if (old_state
!= arc_anon
) {
1134 int use_mutex
= !MUTEX_HELD(&old_state
->arcs_mtx
);
1135 uint64_t *size
= &old_state
->arcs_lsize
[ab
->b_type
];
1138 mutex_enter(&old_state
->arcs_mtx
);
1140 ASSERT(list_link_active(&ab
->b_arc_node
));
1141 list_remove(&old_state
->arcs_list
[ab
->b_type
], ab
);
1144 * If prefetching out of the ghost cache,
1145 * we will have a non-zero datacnt.
1147 if (GHOST_STATE(old_state
) && ab
->b_datacnt
== 0) {
1148 /* ghost elements have a ghost size */
1149 ASSERT(ab
->b_buf
== NULL
);
1150 from_delta
= ab
->b_size
;
1152 ASSERT3U(*size
, >=, from_delta
);
1153 atomic_add_64(size
, -from_delta
);
1156 mutex_exit(&old_state
->arcs_mtx
);
1158 if (new_state
!= arc_anon
) {
1159 int use_mutex
= !MUTEX_HELD(&new_state
->arcs_mtx
);
1160 uint64_t *size
= &new_state
->arcs_lsize
[ab
->b_type
];
1163 mutex_enter(&new_state
->arcs_mtx
);
1165 list_insert_head(&new_state
->arcs_list
[ab
->b_type
], ab
);
1167 /* ghost elements have a ghost size */
1168 if (GHOST_STATE(new_state
)) {
1169 ASSERT(ab
->b_datacnt
== 0);
1170 ASSERT(ab
->b_buf
== NULL
);
1171 to_delta
= ab
->b_size
;
1173 atomic_add_64(size
, to_delta
);
1176 mutex_exit(&new_state
->arcs_mtx
);
1180 ASSERT(!BUF_EMPTY(ab
));
1181 if (new_state
== arc_anon
&& HDR_IN_HASH_TABLE(ab
))
1182 buf_hash_remove(ab
);
1184 /* adjust state sizes */
1186 atomic_add_64(&new_state
->arcs_size
, to_delta
);
1188 ASSERT3U(old_state
->arcs_size
, >=, from_delta
);
1189 atomic_add_64(&old_state
->arcs_size
, -from_delta
);
1191 ab
->b_state
= new_state
;
1193 /* adjust l2arc hdr stats */
1194 if (new_state
== arc_l2c_only
)
1195 l2arc_hdr_stat_add();
1196 else if (old_state
== arc_l2c_only
)
1197 l2arc_hdr_stat_remove();
1201 arc_space_consume(uint64_t space
, arc_space_type_t type
)
1203 ASSERT(type
>= 0 && type
< ARC_SPACE_NUMTYPES
);
1208 case ARC_SPACE_DATA
:
1209 ARCSTAT_INCR(arcstat_data_size
, space
);
1211 case ARC_SPACE_OTHER
:
1212 ARCSTAT_INCR(arcstat_other_size
, space
);
1214 case ARC_SPACE_HDRS
:
1215 ARCSTAT_INCR(arcstat_hdr_size
, space
);
1217 case ARC_SPACE_L2HDRS
:
1218 ARCSTAT_INCR(arcstat_l2_hdr_size
, space
);
1222 atomic_add_64(&arc_meta_used
, space
);
1223 atomic_add_64(&arc_size
, space
);
1227 arc_space_return(uint64_t space
, arc_space_type_t type
)
1229 ASSERT(type
>= 0 && type
< ARC_SPACE_NUMTYPES
);
1234 case ARC_SPACE_DATA
:
1235 ARCSTAT_INCR(arcstat_data_size
, -space
);
1237 case ARC_SPACE_OTHER
:
1238 ARCSTAT_INCR(arcstat_other_size
, -space
);
1240 case ARC_SPACE_HDRS
:
1241 ARCSTAT_INCR(arcstat_hdr_size
, -space
);
1243 case ARC_SPACE_L2HDRS
:
1244 ARCSTAT_INCR(arcstat_l2_hdr_size
, -space
);
1248 ASSERT(arc_meta_used
>= space
);
1249 if (arc_meta_max
< arc_meta_used
)
1250 arc_meta_max
= arc_meta_used
;
1251 atomic_add_64(&arc_meta_used
, -space
);
1252 ASSERT(arc_size
>= space
);
1253 atomic_add_64(&arc_size
, -space
);
1257 arc_buf_alloc(spa_t
*spa
, int size
, void *tag
, arc_buf_contents_t type
)
1262 ASSERT3U(size
, >, 0);
1263 hdr
= kmem_cache_alloc(hdr_cache
, KM_PUSHPAGE
);
1264 ASSERT(BUF_EMPTY(hdr
));
1267 hdr
->b_spa
= spa_load_guid(spa
);
1268 hdr
->b_state
= arc_anon
;
1269 hdr
->b_arc_access
= 0;
1270 buf
= kmem_cache_alloc(buf_cache
, KM_PUSHPAGE
);
1273 buf
->b_efunc
= NULL
;
1274 buf
->b_private
= NULL
;
1277 arc_get_data_buf(buf
);
1280 ASSERT(refcount_is_zero(&hdr
->b_refcnt
));
1281 (void) refcount_add(&hdr
->b_refcnt
, tag
);
1286 static char *arc_onloan_tag
= "onloan";
1289 * Loan out an anonymous arc buffer. Loaned buffers are not counted as in
1290 * flight data by arc_tempreserve_space() until they are "returned". Loaned
1291 * buffers must be returned to the arc before they can be used by the DMU or
1295 arc_loan_buf(spa_t
*spa
, int size
)
1299 buf
= arc_buf_alloc(spa
, size
, arc_onloan_tag
, ARC_BUFC_DATA
);
1301 atomic_add_64(&arc_loaned_bytes
, size
);
1306 * Return a loaned arc buffer to the arc.
1309 arc_return_buf(arc_buf_t
*buf
, void *tag
)
1311 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
1313 ASSERT(buf
->b_data
!= NULL
);
1314 (void) refcount_add(&hdr
->b_refcnt
, tag
);
1315 (void) refcount_remove(&hdr
->b_refcnt
, arc_onloan_tag
);
1317 atomic_add_64(&arc_loaned_bytes
, -hdr
->b_size
);
1320 /* Detach an arc_buf from a dbuf (tag) */
1322 arc_loan_inuse_buf(arc_buf_t
*buf
, void *tag
)
1326 ASSERT(buf
->b_data
!= NULL
);
1328 (void) refcount_add(&hdr
->b_refcnt
, arc_onloan_tag
);
1329 (void) refcount_remove(&hdr
->b_refcnt
, tag
);
1330 buf
->b_efunc
= NULL
;
1331 buf
->b_private
= NULL
;
1333 atomic_add_64(&arc_loaned_bytes
, hdr
->b_size
);
1337 arc_buf_clone(arc_buf_t
*from
)
1340 arc_buf_hdr_t
*hdr
= from
->b_hdr
;
1341 uint64_t size
= hdr
->b_size
;
1343 ASSERT(hdr
->b_state
!= arc_anon
);
1345 buf
= kmem_cache_alloc(buf_cache
, KM_PUSHPAGE
);
1348 buf
->b_efunc
= NULL
;
1349 buf
->b_private
= NULL
;
1350 buf
->b_next
= hdr
->b_buf
;
1352 arc_get_data_buf(buf
);
1353 bcopy(from
->b_data
, buf
->b_data
, size
);
1356 * This buffer already exists in the arc so create a duplicate
1357 * copy for the caller. If the buffer is associated with user data
1358 * then track the size and number of duplicates. These stats will be
1359 * updated as duplicate buffers are created and destroyed.
1361 if (hdr
->b_type
== ARC_BUFC_DATA
) {
1362 ARCSTAT_BUMP(arcstat_duplicate_buffers
);
1363 ARCSTAT_INCR(arcstat_duplicate_buffers_size
, size
);
1365 hdr
->b_datacnt
+= 1;
1370 arc_buf_add_ref(arc_buf_t
*buf
, void* tag
)
1373 kmutex_t
*hash_lock
;
1376 * Check to see if this buffer is evicted. Callers
1377 * must verify b_data != NULL to know if the add_ref
1380 mutex_enter(&buf
->b_evict_lock
);
1381 if (buf
->b_data
== NULL
) {
1382 mutex_exit(&buf
->b_evict_lock
);
1385 hash_lock
= HDR_LOCK(buf
->b_hdr
);
1386 mutex_enter(hash_lock
);
1388 ASSERT3P(hash_lock
, ==, HDR_LOCK(hdr
));
1389 mutex_exit(&buf
->b_evict_lock
);
1391 ASSERT(hdr
->b_state
== arc_mru
|| hdr
->b_state
== arc_mfu
);
1392 add_reference(hdr
, hash_lock
, tag
);
1393 DTRACE_PROBE1(arc__hit
, arc_buf_hdr_t
*, hdr
);
1394 arc_access(hdr
, hash_lock
);
1395 mutex_exit(hash_lock
);
1396 ARCSTAT_BUMP(arcstat_hits
);
1397 ARCSTAT_CONDSTAT(!(hdr
->b_flags
& ARC_PREFETCH
),
1398 demand
, prefetch
, hdr
->b_type
!= ARC_BUFC_METADATA
,
1399 data
, metadata
, hits
);
1403 * Free the arc data buffer. If it is an l2arc write in progress,
1404 * the buffer is placed on l2arc_free_on_write to be freed later.
1407 arc_buf_data_free(arc_buf_hdr_t
*hdr
, void (*free_func
)(void *, size_t),
1408 void *data
, size_t size
)
1410 if (HDR_L2_WRITING(hdr
)) {
1411 l2arc_data_free_t
*df
;
1412 df
= kmem_alloc(sizeof (l2arc_data_free_t
), KM_PUSHPAGE
);
1413 df
->l2df_data
= data
;
1414 df
->l2df_size
= size
;
1415 df
->l2df_func
= free_func
;
1416 mutex_enter(&l2arc_free_on_write_mtx
);
1417 list_insert_head(l2arc_free_on_write
, df
);
1418 mutex_exit(&l2arc_free_on_write_mtx
);
1419 ARCSTAT_BUMP(arcstat_l2_free_on_write
);
1421 free_func(data
, size
);
1426 arc_buf_destroy(arc_buf_t
*buf
, boolean_t recycle
, boolean_t all
)
1430 /* free up data associated with the buf */
1432 arc_state_t
*state
= buf
->b_hdr
->b_state
;
1433 uint64_t size
= buf
->b_hdr
->b_size
;
1434 arc_buf_contents_t type
= buf
->b_hdr
->b_type
;
1436 arc_cksum_verify(buf
);
1439 if (type
== ARC_BUFC_METADATA
) {
1440 arc_buf_data_free(buf
->b_hdr
, zio_buf_free
,
1442 arc_space_return(size
, ARC_SPACE_DATA
);
1444 ASSERT(type
== ARC_BUFC_DATA
);
1445 arc_buf_data_free(buf
->b_hdr
,
1446 zio_data_buf_free
, buf
->b_data
, size
);
1447 ARCSTAT_INCR(arcstat_data_size
, -size
);
1448 atomic_add_64(&arc_size
, -size
);
1451 if (list_link_active(&buf
->b_hdr
->b_arc_node
)) {
1452 uint64_t *cnt
= &state
->arcs_lsize
[type
];
1454 ASSERT(refcount_is_zero(&buf
->b_hdr
->b_refcnt
));
1455 ASSERT(state
!= arc_anon
);
1457 ASSERT3U(*cnt
, >=, size
);
1458 atomic_add_64(cnt
, -size
);
1460 ASSERT3U(state
->arcs_size
, >=, size
);
1461 atomic_add_64(&state
->arcs_size
, -size
);
1465 * If we're destroying a duplicate buffer make sure
1466 * that the appropriate statistics are updated.
1468 if (buf
->b_hdr
->b_datacnt
> 1 &&
1469 buf
->b_hdr
->b_type
== ARC_BUFC_DATA
) {
1470 ARCSTAT_BUMPDOWN(arcstat_duplicate_buffers
);
1471 ARCSTAT_INCR(arcstat_duplicate_buffers_size
, -size
);
1473 ASSERT(buf
->b_hdr
->b_datacnt
> 0);
1474 buf
->b_hdr
->b_datacnt
-= 1;
1477 /* only remove the buf if requested */
1481 /* remove the buf from the hdr list */
1482 for (bufp
= &buf
->b_hdr
->b_buf
; *bufp
!= buf
; bufp
= &(*bufp
)->b_next
)
1484 *bufp
= buf
->b_next
;
1487 ASSERT(buf
->b_efunc
== NULL
);
1489 /* clean up the buf */
1491 kmem_cache_free(buf_cache
, buf
);
1495 arc_hdr_destroy(arc_buf_hdr_t
*hdr
)
1497 l2arc_buf_hdr_t
*l2hdr
= hdr
->b_l2hdr
;
1499 ASSERT(refcount_is_zero(&hdr
->b_refcnt
));
1500 ASSERT3P(hdr
->b_state
, ==, arc_anon
);
1501 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
1503 if (l2hdr
!= NULL
) {
1504 boolean_t buflist_held
= MUTEX_HELD(&l2arc_buflist_mtx
);
1506 * To prevent arc_free() and l2arc_evict() from
1507 * attempting to free the same buffer at the same time,
1508 * a FREE_IN_PROGRESS flag is given to arc_free() to
1509 * give it priority. l2arc_evict() can't destroy this
1510 * header while we are waiting on l2arc_buflist_mtx.
1512 * The hdr may be removed from l2ad_buflist before we
1513 * grab l2arc_buflist_mtx, so b_l2hdr is rechecked.
1515 if (!buflist_held
) {
1516 mutex_enter(&l2arc_buflist_mtx
);
1517 l2hdr
= hdr
->b_l2hdr
;
1520 if (l2hdr
!= NULL
) {
1521 list_remove(l2hdr
->b_dev
->l2ad_buflist
, hdr
);
1522 ARCSTAT_INCR(arcstat_l2_size
, -hdr
->b_size
);
1523 kmem_free(l2hdr
, sizeof (l2arc_buf_hdr_t
));
1524 arc_space_return(L2HDR_SIZE
, ARC_SPACE_L2HDRS
);
1525 if (hdr
->b_state
== arc_l2c_only
)
1526 l2arc_hdr_stat_remove();
1527 hdr
->b_l2hdr
= NULL
;
1531 mutex_exit(&l2arc_buflist_mtx
);
1534 if (!BUF_EMPTY(hdr
)) {
1535 ASSERT(!HDR_IN_HASH_TABLE(hdr
));
1536 buf_discard_identity(hdr
);
1538 while (hdr
->b_buf
) {
1539 arc_buf_t
*buf
= hdr
->b_buf
;
1542 mutex_enter(&arc_eviction_mtx
);
1543 mutex_enter(&buf
->b_evict_lock
);
1544 ASSERT(buf
->b_hdr
!= NULL
);
1545 arc_buf_destroy(hdr
->b_buf
, FALSE
, FALSE
);
1546 hdr
->b_buf
= buf
->b_next
;
1547 buf
->b_hdr
= &arc_eviction_hdr
;
1548 buf
->b_next
= arc_eviction_list
;
1549 arc_eviction_list
= buf
;
1550 mutex_exit(&buf
->b_evict_lock
);
1551 mutex_exit(&arc_eviction_mtx
);
1553 arc_buf_destroy(hdr
->b_buf
, FALSE
, TRUE
);
1556 if (hdr
->b_freeze_cksum
!= NULL
) {
1557 kmem_free(hdr
->b_freeze_cksum
, sizeof (zio_cksum_t
));
1558 hdr
->b_freeze_cksum
= NULL
;
1561 ASSERT(!list_link_active(&hdr
->b_arc_node
));
1562 ASSERT3P(hdr
->b_hash_next
, ==, NULL
);
1563 ASSERT3P(hdr
->b_acb
, ==, NULL
);
1564 kmem_cache_free(hdr_cache
, hdr
);
1568 arc_buf_free(arc_buf_t
*buf
, void *tag
)
1570 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
1571 int hashed
= hdr
->b_state
!= arc_anon
;
1573 ASSERT(buf
->b_efunc
== NULL
);
1574 ASSERT(buf
->b_data
!= NULL
);
1577 kmutex_t
*hash_lock
= HDR_LOCK(hdr
);
1579 mutex_enter(hash_lock
);
1581 ASSERT3P(hash_lock
, ==, HDR_LOCK(hdr
));
1583 (void) remove_reference(hdr
, hash_lock
, tag
);
1584 if (hdr
->b_datacnt
> 1) {
1585 arc_buf_destroy(buf
, FALSE
, TRUE
);
1587 ASSERT(buf
== hdr
->b_buf
);
1588 ASSERT(buf
->b_efunc
== NULL
);
1589 hdr
->b_flags
|= ARC_BUF_AVAILABLE
;
1591 mutex_exit(hash_lock
);
1592 } else if (HDR_IO_IN_PROGRESS(hdr
)) {
1595 * We are in the middle of an async write. Don't destroy
1596 * this buffer unless the write completes before we finish
1597 * decrementing the reference count.
1599 mutex_enter(&arc_eviction_mtx
);
1600 (void) remove_reference(hdr
, NULL
, tag
);
1601 ASSERT(refcount_is_zero(&hdr
->b_refcnt
));
1602 destroy_hdr
= !HDR_IO_IN_PROGRESS(hdr
);
1603 mutex_exit(&arc_eviction_mtx
);
1605 arc_hdr_destroy(hdr
);
1607 if (remove_reference(hdr
, NULL
, tag
) > 0)
1608 arc_buf_destroy(buf
, FALSE
, TRUE
);
1610 arc_hdr_destroy(hdr
);
1615 arc_buf_remove_ref(arc_buf_t
*buf
, void* tag
)
1617 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
1618 kmutex_t
*hash_lock
= NULL
;
1619 int no_callback
= (buf
->b_efunc
== NULL
);
1621 if (hdr
->b_state
== arc_anon
) {
1622 ASSERT(hdr
->b_datacnt
== 1);
1623 arc_buf_free(buf
, tag
);
1624 return (no_callback
);
1627 hash_lock
= HDR_LOCK(hdr
);
1628 mutex_enter(hash_lock
);
1630 ASSERT3P(hash_lock
, ==, HDR_LOCK(hdr
));
1631 ASSERT(hdr
->b_state
!= arc_anon
);
1632 ASSERT(buf
->b_data
!= NULL
);
1634 (void) remove_reference(hdr
, hash_lock
, tag
);
1635 if (hdr
->b_datacnt
> 1) {
1637 arc_buf_destroy(buf
, FALSE
, TRUE
);
1638 } else if (no_callback
) {
1639 ASSERT(hdr
->b_buf
== buf
&& buf
->b_next
== NULL
);
1640 ASSERT(buf
->b_efunc
== NULL
);
1641 hdr
->b_flags
|= ARC_BUF_AVAILABLE
;
1643 ASSERT(no_callback
|| hdr
->b_datacnt
> 1 ||
1644 refcount_is_zero(&hdr
->b_refcnt
));
1645 mutex_exit(hash_lock
);
1646 return (no_callback
);
1650 arc_buf_size(arc_buf_t
*buf
)
1652 return (buf
->b_hdr
->b_size
);
1656 * Called from the DMU to determine if the current buffer should be
1657 * evicted. In order to ensure proper locking, the eviction must be initiated
1658 * from the DMU. Return true if the buffer is associated with user data and
1659 * duplicate buffers still exist.
1662 arc_buf_eviction_needed(arc_buf_t
*buf
)
1665 boolean_t evict_needed
= B_FALSE
;
1667 if (zfs_disable_dup_eviction
)
1670 mutex_enter(&buf
->b_evict_lock
);
1674 * We are in arc_do_user_evicts(); let that function
1675 * perform the eviction.
1677 ASSERT(buf
->b_data
== NULL
);
1678 mutex_exit(&buf
->b_evict_lock
);
1680 } else if (buf
->b_data
== NULL
) {
1682 * We have already been added to the arc eviction list;
1683 * recommend eviction.
1685 ASSERT3P(hdr
, ==, &arc_eviction_hdr
);
1686 mutex_exit(&buf
->b_evict_lock
);
1690 if (hdr
->b_datacnt
> 1 && hdr
->b_type
== ARC_BUFC_DATA
)
1691 evict_needed
= B_TRUE
;
1693 mutex_exit(&buf
->b_evict_lock
);
1694 return (evict_needed
);
1698 * Evict buffers from list until we've removed the specified number of
1699 * bytes. Move the removed buffers to the appropriate evict state.
1700 * If the recycle flag is set, then attempt to "recycle" a buffer:
1701 * - look for a buffer to evict that is `bytes' long.
1702 * - return the data block from this buffer rather than freeing it.
1703 * This flag is used by callers that are trying to make space for a
1704 * new buffer in a full arc cache.
1706 * This function makes a "best effort". It skips over any buffers
1707 * it can't get a hash_lock on, and so may not catch all candidates.
1708 * It may also return without evicting as much space as requested.
1711 arc_evict(arc_state_t
*state
, uint64_t spa
, int64_t bytes
, boolean_t recycle
,
1712 arc_buf_contents_t type
)
1714 arc_state_t
*evicted_state
;
1715 uint64_t bytes_evicted
= 0, skipped
= 0, missed
= 0;
1716 arc_buf_hdr_t
*ab
, *ab_prev
= NULL
;
1717 list_t
*list
= &state
->arcs_list
[type
];
1718 kmutex_t
*hash_lock
;
1719 boolean_t have_lock
;
1720 void *stolen
= NULL
;
1722 ASSERT(state
== arc_mru
|| state
== arc_mfu
);
1724 evicted_state
= (state
== arc_mru
) ? arc_mru_ghost
: arc_mfu_ghost
;
1726 mutex_enter(&state
->arcs_mtx
);
1727 mutex_enter(&evicted_state
->arcs_mtx
);
1729 for (ab
= list_tail(list
); ab
; ab
= ab_prev
) {
1730 ab_prev
= list_prev(list
, ab
);
1731 /* prefetch buffers have a minimum lifespan */
1732 if (HDR_IO_IN_PROGRESS(ab
) ||
1733 (spa
&& ab
->b_spa
!= spa
) ||
1734 (ab
->b_flags
& (ARC_PREFETCH
|ARC_INDIRECT
) &&
1735 ddi_get_lbolt() - ab
->b_arc_access
<
1736 zfs_arc_min_prefetch_lifespan
)) {
1740 /* "lookahead" for better eviction candidate */
1741 if (recycle
&& ab
->b_size
!= bytes
&&
1742 ab_prev
&& ab_prev
->b_size
== bytes
)
1744 hash_lock
= HDR_LOCK(ab
);
1745 have_lock
= MUTEX_HELD(hash_lock
);
1746 if (have_lock
|| mutex_tryenter(hash_lock
)) {
1747 ASSERT0(refcount_count(&ab
->b_refcnt
));
1748 ASSERT(ab
->b_datacnt
> 0);
1750 arc_buf_t
*buf
= ab
->b_buf
;
1751 if (!mutex_tryenter(&buf
->b_evict_lock
)) {
1756 bytes_evicted
+= ab
->b_size
;
1757 if (recycle
&& ab
->b_type
== type
&&
1758 ab
->b_size
== bytes
&&
1759 !HDR_L2_WRITING(ab
)) {
1760 stolen
= buf
->b_data
;
1765 mutex_enter(&arc_eviction_mtx
);
1766 arc_buf_destroy(buf
,
1767 buf
->b_data
== stolen
, FALSE
);
1768 ab
->b_buf
= buf
->b_next
;
1769 buf
->b_hdr
= &arc_eviction_hdr
;
1770 buf
->b_next
= arc_eviction_list
;
1771 arc_eviction_list
= buf
;
1772 mutex_exit(&arc_eviction_mtx
);
1773 mutex_exit(&buf
->b_evict_lock
);
1775 mutex_exit(&buf
->b_evict_lock
);
1776 arc_buf_destroy(buf
,
1777 buf
->b_data
== stolen
, TRUE
);
1782 ARCSTAT_INCR(arcstat_evict_l2_cached
,
1785 if (l2arc_write_eligible(ab
->b_spa
, ab
)) {
1786 ARCSTAT_INCR(arcstat_evict_l2_eligible
,
1790 arcstat_evict_l2_ineligible
,
1795 if (ab
->b_datacnt
== 0) {
1796 arc_change_state(evicted_state
, ab
, hash_lock
);
1797 ASSERT(HDR_IN_HASH_TABLE(ab
));
1798 ab
->b_flags
|= ARC_IN_HASH_TABLE
;
1799 ab
->b_flags
&= ~ARC_BUF_AVAILABLE
;
1800 DTRACE_PROBE1(arc__evict
, arc_buf_hdr_t
*, ab
);
1803 mutex_exit(hash_lock
);
1804 if (bytes
>= 0 && bytes_evicted
>= bytes
)
1811 mutex_exit(&evicted_state
->arcs_mtx
);
1812 mutex_exit(&state
->arcs_mtx
);
1814 if (bytes_evicted
< bytes
)
1815 dprintf("only evicted %lld bytes from %x\n",
1816 (longlong_t
)bytes_evicted
, state
);
1819 ARCSTAT_INCR(arcstat_evict_skip
, skipped
);
1822 ARCSTAT_INCR(arcstat_mutex_miss
, missed
);
1825 * We have just evicted some date into the ghost state, make
1826 * sure we also adjust the ghost state size if necessary.
1829 arc_mru_ghost
->arcs_size
+ arc_mfu_ghost
->arcs_size
> arc_c
) {
1830 int64_t mru_over
= arc_anon
->arcs_size
+ arc_mru
->arcs_size
+
1831 arc_mru_ghost
->arcs_size
- arc_c
;
1833 if (mru_over
> 0 && arc_mru_ghost
->arcs_lsize
[type
] > 0) {
1835 MIN(arc_mru_ghost
->arcs_lsize
[type
], mru_over
);
1836 arc_evict_ghost(arc_mru_ghost
, 0, todelete
);
1837 } else if (arc_mfu_ghost
->arcs_lsize
[type
] > 0) {
1838 int64_t todelete
= MIN(arc_mfu_ghost
->arcs_lsize
[type
],
1839 arc_mru_ghost
->arcs_size
+
1840 arc_mfu_ghost
->arcs_size
- arc_c
);
1841 arc_evict_ghost(arc_mfu_ghost
, 0, todelete
);
1849 * Remove buffers from list until we've removed the specified number of
1850 * bytes. Destroy the buffers that are removed.
1853 arc_evict_ghost(arc_state_t
*state
, uint64_t spa
, int64_t bytes
)
1855 arc_buf_hdr_t
*ab
, *ab_prev
;
1856 arc_buf_hdr_t marker
;
1857 list_t
*list
= &state
->arcs_list
[ARC_BUFC_DATA
];
1858 kmutex_t
*hash_lock
;
1859 uint64_t bytes_deleted
= 0;
1860 uint64_t bufs_skipped
= 0;
1862 ASSERT(GHOST_STATE(state
));
1863 bzero(&marker
, sizeof(marker
));
1865 mutex_enter(&state
->arcs_mtx
);
1866 for (ab
= list_tail(list
); ab
; ab
= ab_prev
) {
1867 ab_prev
= list_prev(list
, ab
);
1868 if (spa
&& ab
->b_spa
!= spa
)
1871 /* ignore markers */
1875 hash_lock
= HDR_LOCK(ab
);
1876 /* caller may be trying to modify this buffer, skip it */
1877 if (MUTEX_HELD(hash_lock
))
1879 if (mutex_tryenter(hash_lock
)) {
1880 ASSERT(!HDR_IO_IN_PROGRESS(ab
));
1881 ASSERT(ab
->b_buf
== NULL
);
1882 ARCSTAT_BUMP(arcstat_deleted
);
1883 bytes_deleted
+= ab
->b_size
;
1885 if (ab
->b_l2hdr
!= NULL
) {
1887 * This buffer is cached on the 2nd Level ARC;
1888 * don't destroy the header.
1890 arc_change_state(arc_l2c_only
, ab
, hash_lock
);
1891 mutex_exit(hash_lock
);
1893 arc_change_state(arc_anon
, ab
, hash_lock
);
1894 mutex_exit(hash_lock
);
1895 arc_hdr_destroy(ab
);
1898 DTRACE_PROBE1(arc__delete
, arc_buf_hdr_t
*, ab
);
1899 if (bytes
>= 0 && bytes_deleted
>= bytes
)
1901 } else if (bytes
< 0) {
1903 * Insert a list marker and then wait for the
1904 * hash lock to become available. Once its
1905 * available, restart from where we left off.
1907 list_insert_after(list
, ab
, &marker
);
1908 mutex_exit(&state
->arcs_mtx
);
1909 mutex_enter(hash_lock
);
1910 mutex_exit(hash_lock
);
1911 mutex_enter(&state
->arcs_mtx
);
1912 ab_prev
= list_prev(list
, &marker
);
1913 list_remove(list
, &marker
);
1917 mutex_exit(&state
->arcs_mtx
);
1919 if (list
== &state
->arcs_list
[ARC_BUFC_DATA
] &&
1920 (bytes
< 0 || bytes_deleted
< bytes
)) {
1921 list
= &state
->arcs_list
[ARC_BUFC_METADATA
];
1926 ARCSTAT_INCR(arcstat_mutex_miss
, bufs_skipped
);
1930 if (bytes_deleted
< bytes
)
1931 dprintf("only deleted %lld bytes from %p\n",
1932 (longlong_t
)bytes_deleted
, state
);
1938 int64_t adjustment
, delta
;
1944 adjustment
= MIN((int64_t)(arc_size
- arc_c
),
1945 (int64_t)(arc_anon
->arcs_size
+ arc_mru
->arcs_size
+ arc_meta_used
-
1948 if (adjustment
> 0 && arc_mru
->arcs_lsize
[ARC_BUFC_DATA
] > 0) {
1949 delta
= MIN(arc_mru
->arcs_lsize
[ARC_BUFC_DATA
], adjustment
);
1950 (void) arc_evict(arc_mru
, 0, delta
, FALSE
, ARC_BUFC_DATA
);
1951 adjustment
-= delta
;
1954 if (adjustment
> 0 && arc_mru
->arcs_lsize
[ARC_BUFC_METADATA
] > 0) {
1955 delta
= MIN(arc_mru
->arcs_lsize
[ARC_BUFC_METADATA
], adjustment
);
1956 (void) arc_evict(arc_mru
, 0, delta
, FALSE
,
1964 adjustment
= arc_size
- arc_c
;
1966 if (adjustment
> 0 && arc_mfu
->arcs_lsize
[ARC_BUFC_DATA
] > 0) {
1967 delta
= MIN(adjustment
, arc_mfu
->arcs_lsize
[ARC_BUFC_DATA
]);
1968 (void) arc_evict(arc_mfu
, 0, delta
, FALSE
, ARC_BUFC_DATA
);
1969 adjustment
-= delta
;
1972 if (adjustment
> 0 && arc_mfu
->arcs_lsize
[ARC_BUFC_METADATA
] > 0) {
1973 int64_t delta
= MIN(adjustment
,
1974 arc_mfu
->arcs_lsize
[ARC_BUFC_METADATA
]);
1975 (void) arc_evict(arc_mfu
, 0, delta
, FALSE
,
1980 * Adjust ghost lists
1983 adjustment
= arc_mru
->arcs_size
+ arc_mru_ghost
->arcs_size
- arc_c
;
1985 if (adjustment
> 0 && arc_mru_ghost
->arcs_size
> 0) {
1986 delta
= MIN(arc_mru_ghost
->arcs_size
, adjustment
);
1987 arc_evict_ghost(arc_mru_ghost
, 0, delta
);
1991 arc_mru_ghost
->arcs_size
+ arc_mfu_ghost
->arcs_size
- arc_c
;
1993 if (adjustment
> 0 && arc_mfu_ghost
->arcs_size
> 0) {
1994 delta
= MIN(arc_mfu_ghost
->arcs_size
, adjustment
);
1995 arc_evict_ghost(arc_mfu_ghost
, 0, delta
);
2000 * Request that arc user drop references so that N bytes can be released
2001 * from the cache. This provides a mechanism to ensure the arc can honor
2002 * the arc_meta_limit and reclaim buffers which are pinned in the cache
2003 * by higher layers. (i.e. the zpl)
2006 arc_do_user_prune(int64_t adjustment
)
2008 arc_prune_func_t
*func
;
2010 arc_prune_t
*cp
, *np
;
2012 mutex_enter(&arc_prune_mtx
);
2014 cp
= list_head(&arc_prune_list
);
2015 while (cp
!= NULL
) {
2017 private = cp
->p_private
;
2018 np
= list_next(&arc_prune_list
, cp
);
2019 refcount_add(&cp
->p_refcnt
, func
);
2020 mutex_exit(&arc_prune_mtx
);
2023 func(adjustment
, private);
2025 mutex_enter(&arc_prune_mtx
);
2027 /* User removed prune callback concurrently with execution */
2028 if (refcount_remove(&cp
->p_refcnt
, func
) == 0) {
2029 ASSERT(!list_link_active(&cp
->p_node
));
2030 refcount_destroy(&cp
->p_refcnt
);
2031 kmem_free(cp
, sizeof (*cp
));
2037 ARCSTAT_BUMP(arcstat_prune
);
2038 mutex_exit(&arc_prune_mtx
);
2042 arc_do_user_evicts(void)
2044 mutex_enter(&arc_eviction_mtx
);
2045 while (arc_eviction_list
!= NULL
) {
2046 arc_buf_t
*buf
= arc_eviction_list
;
2047 arc_eviction_list
= buf
->b_next
;
2048 mutex_enter(&buf
->b_evict_lock
);
2050 mutex_exit(&buf
->b_evict_lock
);
2051 mutex_exit(&arc_eviction_mtx
);
2053 if (buf
->b_efunc
!= NULL
)
2054 VERIFY(buf
->b_efunc(buf
) == 0);
2056 buf
->b_efunc
= NULL
;
2057 buf
->b_private
= NULL
;
2058 kmem_cache_free(buf_cache
, buf
);
2059 mutex_enter(&arc_eviction_mtx
);
2061 mutex_exit(&arc_eviction_mtx
);
2065 * Evict only meta data objects from the cache leaving the data objects.
2066 * This is only used to enforce the tunable arc_meta_limit, if we are
2067 * unable to evict enough buffers notify the user via the prune callback.
2070 arc_adjust_meta(int64_t adjustment
, boolean_t may_prune
)
2074 if (adjustment
> 0 && arc_mru
->arcs_lsize
[ARC_BUFC_METADATA
] > 0) {
2075 delta
= MIN(arc_mru
->arcs_lsize
[ARC_BUFC_METADATA
], adjustment
);
2076 arc_evict(arc_mru
, 0, delta
, FALSE
, ARC_BUFC_METADATA
);
2077 adjustment
-= delta
;
2080 if (adjustment
> 0 && arc_mfu
->arcs_lsize
[ARC_BUFC_METADATA
] > 0) {
2081 delta
= MIN(arc_mfu
->arcs_lsize
[ARC_BUFC_METADATA
], adjustment
);
2082 arc_evict(arc_mfu
, 0, delta
, FALSE
, ARC_BUFC_METADATA
);
2083 adjustment
-= delta
;
2086 if (may_prune
&& (adjustment
> 0) && (arc_meta_used
> arc_meta_limit
))
2087 arc_do_user_prune(zfs_arc_meta_prune
);
2091 * Flush all *evictable* data from the cache for the given spa.
2092 * NOTE: this will not touch "active" (i.e. referenced) data.
2095 arc_flush(spa_t
*spa
)
2100 guid
= spa_load_guid(spa
);
2102 while (list_head(&arc_mru
->arcs_list
[ARC_BUFC_DATA
])) {
2103 (void) arc_evict(arc_mru
, guid
, -1, FALSE
, ARC_BUFC_DATA
);
2107 while (list_head(&arc_mru
->arcs_list
[ARC_BUFC_METADATA
])) {
2108 (void) arc_evict(arc_mru
, guid
, -1, FALSE
, ARC_BUFC_METADATA
);
2112 while (list_head(&arc_mfu
->arcs_list
[ARC_BUFC_DATA
])) {
2113 (void) arc_evict(arc_mfu
, guid
, -1, FALSE
, ARC_BUFC_DATA
);
2117 while (list_head(&arc_mfu
->arcs_list
[ARC_BUFC_METADATA
])) {
2118 (void) arc_evict(arc_mfu
, guid
, -1, FALSE
, ARC_BUFC_METADATA
);
2123 arc_evict_ghost(arc_mru_ghost
, guid
, -1);
2124 arc_evict_ghost(arc_mfu_ghost
, guid
, -1);
2126 mutex_enter(&arc_reclaim_thr_lock
);
2127 arc_do_user_evicts();
2128 mutex_exit(&arc_reclaim_thr_lock
);
2129 ASSERT(spa
|| arc_eviction_list
== NULL
);
2133 arc_shrink(uint64_t bytes
)
2135 if (arc_c
> arc_c_min
) {
2138 to_free
= bytes
? bytes
: arc_c
>> zfs_arc_shrink_shift
;
2140 if (arc_c
> arc_c_min
+ to_free
)
2141 atomic_add_64(&arc_c
, -to_free
);
2145 atomic_add_64(&arc_p
, -(arc_p
>> zfs_arc_shrink_shift
));
2146 if (arc_c
> arc_size
)
2147 arc_c
= MAX(arc_size
, arc_c_min
);
2149 arc_p
= (arc_c
>> 1);
2150 ASSERT(arc_c
>= arc_c_min
);
2151 ASSERT((int64_t)arc_p
>= 0);
2154 if (arc_size
> arc_c
)
2159 arc_kmem_reap_now(arc_reclaim_strategy_t strat
, uint64_t bytes
)
2162 kmem_cache_t
*prev_cache
= NULL
;
2163 kmem_cache_t
*prev_data_cache
= NULL
;
2164 extern kmem_cache_t
*zio_buf_cache
[];
2165 extern kmem_cache_t
*zio_data_buf_cache
[];
2168 * An aggressive reclamation will shrink the cache size as well as
2169 * reap free buffers from the arc kmem caches.
2171 if (strat
== ARC_RECLAIM_AGGR
)
2174 for (i
= 0; i
< SPA_MAXBLOCKSIZE
>> SPA_MINBLOCKSHIFT
; i
++) {
2175 if (zio_buf_cache
[i
] != prev_cache
) {
2176 prev_cache
= zio_buf_cache
[i
];
2177 kmem_cache_reap_now(zio_buf_cache
[i
]);
2179 if (zio_data_buf_cache
[i
] != prev_data_cache
) {
2180 prev_data_cache
= zio_data_buf_cache
[i
];
2181 kmem_cache_reap_now(zio_data_buf_cache
[i
]);
2185 kmem_cache_reap_now(buf_cache
);
2186 kmem_cache_reap_now(hdr_cache
);
2190 * Unlike other ZFS implementations this thread is only responsible for
2191 * adapting the target ARC size on Linux. The responsibility for memory
2192 * reclamation has been entirely delegated to the arc_shrinker_func()
2193 * which is registered with the VM. To reflect this change in behavior
2194 * the arc_reclaim thread has been renamed to arc_adapt.
2197 arc_adapt_thread(void)
2202 CALLB_CPR_INIT(&cpr
, &arc_reclaim_thr_lock
, callb_generic_cpr
, FTAG
);
2204 mutex_enter(&arc_reclaim_thr_lock
);
2205 while (arc_thread_exit
== 0) {
2207 arc_reclaim_strategy_t last_reclaim
= ARC_RECLAIM_CONS
;
2209 if (spa_get_random(100) == 0) {
2212 if (last_reclaim
== ARC_RECLAIM_CONS
) {
2213 last_reclaim
= ARC_RECLAIM_AGGR
;
2215 last_reclaim
= ARC_RECLAIM_CONS
;
2219 last_reclaim
= ARC_RECLAIM_AGGR
;
2223 /* reset the growth delay for every reclaim */
2224 arc_grow_time
= ddi_get_lbolt()+(zfs_arc_grow_retry
* hz
);
2226 arc_kmem_reap_now(last_reclaim
, 0);
2229 #endif /* !_KERNEL */
2231 /* No recent memory pressure allow the ARC to grow. */
2232 if (arc_no_grow
&& ddi_get_lbolt() >= arc_grow_time
)
2233 arc_no_grow
= FALSE
;
2236 * Keep meta data usage within limits, arc_shrink() is not
2237 * used to avoid collapsing the arc_c value when only the
2238 * arc_meta_limit is being exceeded.
2240 prune
= (int64_t)arc_meta_used
- (int64_t)arc_meta_limit
;
2242 arc_adjust_meta(prune
, B_TRUE
);
2246 if (arc_eviction_list
!= NULL
)
2247 arc_do_user_evicts();
2249 /* block until needed, or one second, whichever is shorter */
2250 CALLB_CPR_SAFE_BEGIN(&cpr
);
2251 (void) cv_timedwait_interruptible(&arc_reclaim_thr_cv
,
2252 &arc_reclaim_thr_lock
, (ddi_get_lbolt() + hz
));
2253 CALLB_CPR_SAFE_END(&cpr
, &arc_reclaim_thr_lock
);
2256 /* Allow the module options to be changed */
2257 if (zfs_arc_max
> 64 << 20 &&
2258 zfs_arc_max
< physmem
* PAGESIZE
&&
2259 zfs_arc_max
!= arc_c_max
)
2260 arc_c_max
= zfs_arc_max
;
2262 if (zfs_arc_min
> 0 &&
2263 zfs_arc_min
< arc_c_max
&&
2264 zfs_arc_min
!= arc_c_min
)
2265 arc_c_min
= zfs_arc_min
;
2267 if (zfs_arc_meta_limit
> 0 &&
2268 zfs_arc_meta_limit
<= arc_c_max
&&
2269 zfs_arc_meta_limit
!= arc_meta_limit
)
2270 arc_meta_limit
= zfs_arc_meta_limit
;
2276 arc_thread_exit
= 0;
2277 cv_broadcast(&arc_reclaim_thr_cv
);
2278 CALLB_CPR_EXIT(&cpr
); /* drops arc_reclaim_thr_lock */
2284 * Determine the amount of memory eligible for eviction contained in the
2285 * ARC. All clean data reported by the ghost lists can always be safely
2286 * evicted. Due to arc_c_min, the same does not hold for all clean data
2287 * contained by the regular mru and mfu lists.
2289 * In the case of the regular mru and mfu lists, we need to report as
2290 * much clean data as possible, such that evicting that same reported
2291 * data will not bring arc_size below arc_c_min. Thus, in certain
2292 * circumstances, the total amount of clean data in the mru and mfu
2293 * lists might not actually be evictable.
2295 * The following two distinct cases are accounted for:
2297 * 1. The sum of the amount of dirty data contained by both the mru and
2298 * mfu lists, plus the ARC's other accounting (e.g. the anon list),
2299 * is greater than or equal to arc_c_min.
2300 * (i.e. amount of dirty data >= arc_c_min)
2302 * This is the easy case; all clean data contained by the mru and mfu
2303 * lists is evictable. Evicting all clean data can only drop arc_size
2304 * to the amount of dirty data, which is greater than arc_c_min.
2306 * 2. The sum of the amount of dirty data contained by both the mru and
2307 * mfu lists, plus the ARC's other accounting (e.g. the anon list),
2308 * is less than arc_c_min.
2309 * (i.e. arc_c_min > amount of dirty data)
2311 * 2.1. arc_size is greater than or equal arc_c_min.
2312 * (i.e. arc_size >= arc_c_min > amount of dirty data)
2314 * In this case, not all clean data from the regular mru and mfu
2315 * lists is actually evictable; we must leave enough clean data
2316 * to keep arc_size above arc_c_min. Thus, the maximum amount of
2317 * evictable data from the two lists combined, is exactly the
2318 * difference between arc_size and arc_c_min.
2320 * 2.2. arc_size is less than arc_c_min
2321 * (i.e. arc_c_min > arc_size > amount of dirty data)
2323 * In this case, none of the data contained in the mru and mfu
2324 * lists is evictable, even if it's clean. Since arc_size is
2325 * already below arc_c_min, evicting any more would only
2326 * increase this negative difference.
2329 arc_evictable_memory(void) {
2330 uint64_t arc_clean
=
2331 arc_mru
->arcs_lsize
[ARC_BUFC_DATA
] +
2332 arc_mru
->arcs_lsize
[ARC_BUFC_METADATA
] +
2333 arc_mfu
->arcs_lsize
[ARC_BUFC_DATA
] +
2334 arc_mfu
->arcs_lsize
[ARC_BUFC_METADATA
];
2335 uint64_t ghost_clean
=
2336 arc_mru_ghost
->arcs_lsize
[ARC_BUFC_DATA
] +
2337 arc_mru_ghost
->arcs_lsize
[ARC_BUFC_METADATA
] +
2338 arc_mfu_ghost
->arcs_lsize
[ARC_BUFC_DATA
] +
2339 arc_mfu_ghost
->arcs_lsize
[ARC_BUFC_METADATA
];
2340 uint64_t arc_dirty
= MAX((int64_t)arc_size
- (int64_t)arc_clean
, 0);
2342 if (arc_dirty
>= arc_c_min
)
2343 return (ghost_clean
+ arc_clean
);
2345 return (ghost_clean
+ MAX((int64_t)arc_size
- (int64_t)arc_c_min
, 0));
2349 __arc_shrinker_func(struct shrinker
*shrink
, struct shrink_control
*sc
)
2353 /* The arc is considered warm once reclaim has occurred */
2354 if (unlikely(arc_warm
== B_FALSE
))
2357 /* Return the potential number of reclaimable pages */
2358 pages
= btop(arc_evictable_memory());
2359 if (sc
->nr_to_scan
== 0)
2362 /* Not allowed to perform filesystem reclaim */
2363 if (!(sc
->gfp_mask
& __GFP_FS
))
2366 /* Reclaim in progress */
2367 if (mutex_tryenter(&arc_reclaim_thr_lock
) == 0)
2371 * Evict the requested number of pages by shrinking arc_c the
2372 * requested amount. If there is nothing left to evict just
2373 * reap whatever we can from the various arc slabs.
2376 arc_kmem_reap_now(ARC_RECLAIM_AGGR
, ptob(sc
->nr_to_scan
));
2378 arc_kmem_reap_now(ARC_RECLAIM_CONS
, ptob(sc
->nr_to_scan
));
2382 * When direct reclaim is observed it usually indicates a rapid
2383 * increase in memory pressure. This occurs because the kswapd
2384 * threads were unable to asynchronously keep enough free memory
2385 * available. In this case set arc_no_grow to briefly pause arc
2386 * growth to avoid compounding the memory pressure.
2388 if (current_is_kswapd()) {
2389 ARCSTAT_BUMP(arcstat_memory_indirect_count
);
2391 arc_no_grow
= B_TRUE
;
2392 arc_grow_time
= ddi_get_lbolt() + (zfs_arc_grow_retry
* hz
);
2393 ARCSTAT_BUMP(arcstat_memory_direct_count
);
2396 mutex_exit(&arc_reclaim_thr_lock
);
2400 SPL_SHRINKER_CALLBACK_WRAPPER(arc_shrinker_func
);
2402 SPL_SHRINKER_DECLARE(arc_shrinker
, arc_shrinker_func
, DEFAULT_SEEKS
);
2403 #endif /* _KERNEL */
2406 * Adapt arc info given the number of bytes we are trying to add and
2407 * the state that we are comming from. This function is only called
2408 * when we are adding new content to the cache.
2411 arc_adapt(int bytes
, arc_state_t
*state
)
2414 uint64_t arc_p_min
= (arc_c
>> zfs_arc_p_min_shift
);
2416 if (state
== arc_l2c_only
)
2421 * Adapt the target size of the MRU list:
2422 * - if we just hit in the MRU ghost list, then increase
2423 * the target size of the MRU list.
2424 * - if we just hit in the MFU ghost list, then increase
2425 * the target size of the MFU list by decreasing the
2426 * target size of the MRU list.
2428 if (state
== arc_mru_ghost
) {
2429 mult
= ((arc_mru_ghost
->arcs_size
>= arc_mfu_ghost
->arcs_size
) ?
2430 1 : (arc_mfu_ghost
->arcs_size
/arc_mru_ghost
->arcs_size
));
2431 mult
= MIN(mult
, 10); /* avoid wild arc_p adjustment */
2433 arc_p
= MIN(arc_c
- arc_p_min
, arc_p
+ bytes
* mult
);
2434 } else if (state
== arc_mfu_ghost
) {
2437 mult
= ((arc_mfu_ghost
->arcs_size
>= arc_mru_ghost
->arcs_size
) ?
2438 1 : (arc_mru_ghost
->arcs_size
/arc_mfu_ghost
->arcs_size
));
2439 mult
= MIN(mult
, 10);
2441 delta
= MIN(bytes
* mult
, arc_p
);
2442 arc_p
= MAX(arc_p_min
, arc_p
- delta
);
2444 ASSERT((int64_t)arc_p
>= 0);
2449 if (arc_c
>= arc_c_max
)
2453 * If we're within (2 * maxblocksize) bytes of the target
2454 * cache size, increment the target cache size
2456 if (arc_size
> arc_c
- (2ULL << SPA_MAXBLOCKSHIFT
)) {
2457 atomic_add_64(&arc_c
, (int64_t)bytes
);
2458 if (arc_c
> arc_c_max
)
2460 else if (state
== arc_anon
)
2461 atomic_add_64(&arc_p
, (int64_t)bytes
);
2465 ASSERT((int64_t)arc_p
>= 0);
2469 * Check if the cache has reached its limits and eviction is required
2473 arc_evict_needed(arc_buf_contents_t type
)
2475 if (type
== ARC_BUFC_METADATA
&& arc_meta_used
>= arc_meta_limit
)
2481 return (arc_size
> arc_c
);
2485 * The buffer, supplied as the first argument, needs a data block.
2486 * So, if we are at cache max, determine which cache should be victimized.
2487 * We have the following cases:
2489 * 1. Insert for MRU, p > sizeof(arc_anon + arc_mru) ->
2490 * In this situation if we're out of space, but the resident size of the MFU is
2491 * under the limit, victimize the MFU cache to satisfy this insertion request.
2493 * 2. Insert for MRU, p <= sizeof(arc_anon + arc_mru) ->
2494 * Here, we've used up all of the available space for the MRU, so we need to
2495 * evict from our own cache instead. Evict from the set of resident MRU
2498 * 3. Insert for MFU (c - p) > sizeof(arc_mfu) ->
2499 * c minus p represents the MFU space in the cache, since p is the size of the
2500 * cache that is dedicated to the MRU. In this situation there's still space on
2501 * the MFU side, so the MRU side needs to be victimized.
2503 * 4. Insert for MFU (c - p) < sizeof(arc_mfu) ->
2504 * MFU's resident set is consuming more space than it has been allotted. In
2505 * this situation, we must victimize our own cache, the MFU, for this insertion.
2508 arc_get_data_buf(arc_buf_t
*buf
)
2510 arc_state_t
*state
= buf
->b_hdr
->b_state
;
2511 uint64_t size
= buf
->b_hdr
->b_size
;
2512 arc_buf_contents_t type
= buf
->b_hdr
->b_type
;
2514 arc_adapt(size
, state
);
2517 * We have not yet reached cache maximum size,
2518 * just allocate a new buffer.
2520 if (!arc_evict_needed(type
)) {
2521 if (type
== ARC_BUFC_METADATA
) {
2522 buf
->b_data
= zio_buf_alloc(size
);
2523 arc_space_consume(size
, ARC_SPACE_DATA
);
2525 ASSERT(type
== ARC_BUFC_DATA
);
2526 buf
->b_data
= zio_data_buf_alloc(size
);
2527 ARCSTAT_INCR(arcstat_data_size
, size
);
2528 atomic_add_64(&arc_size
, size
);
2534 * If we are prefetching from the mfu ghost list, this buffer
2535 * will end up on the mru list; so steal space from there.
2537 if (state
== arc_mfu_ghost
)
2538 state
= buf
->b_hdr
->b_flags
& ARC_PREFETCH
? arc_mru
: arc_mfu
;
2539 else if (state
== arc_mru_ghost
)
2542 if (state
== arc_mru
|| state
== arc_anon
) {
2543 uint64_t mru_used
= arc_anon
->arcs_size
+ arc_mru
->arcs_size
;
2544 state
= (arc_mfu
->arcs_lsize
[type
] >= size
&&
2545 arc_p
> mru_used
) ? arc_mfu
: arc_mru
;
2548 uint64_t mfu_space
= arc_c
- arc_p
;
2549 state
= (arc_mru
->arcs_lsize
[type
] >= size
&&
2550 mfu_space
> arc_mfu
->arcs_size
) ? arc_mru
: arc_mfu
;
2553 if ((buf
->b_data
= arc_evict(state
, 0, size
, TRUE
, type
)) == NULL
) {
2554 if (type
== ARC_BUFC_METADATA
) {
2555 buf
->b_data
= zio_buf_alloc(size
);
2556 arc_space_consume(size
, ARC_SPACE_DATA
);
2559 * If we are unable to recycle an existing meta buffer
2560 * signal the reclaim thread. It will notify users
2561 * via the prune callback to drop references. The
2562 * prune callback in run in the context of the reclaim
2563 * thread to avoid deadlocking on the hash_lock.
2565 cv_signal(&arc_reclaim_thr_cv
);
2567 ASSERT(type
== ARC_BUFC_DATA
);
2568 buf
->b_data
= zio_data_buf_alloc(size
);
2569 ARCSTAT_INCR(arcstat_data_size
, size
);
2570 atomic_add_64(&arc_size
, size
);
2573 ARCSTAT_BUMP(arcstat_recycle_miss
);
2575 ASSERT(buf
->b_data
!= NULL
);
2578 * Update the state size. Note that ghost states have a
2579 * "ghost size" and so don't need to be updated.
2581 if (!GHOST_STATE(buf
->b_hdr
->b_state
)) {
2582 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
2584 atomic_add_64(&hdr
->b_state
->arcs_size
, size
);
2585 if (list_link_active(&hdr
->b_arc_node
)) {
2586 ASSERT(refcount_is_zero(&hdr
->b_refcnt
));
2587 atomic_add_64(&hdr
->b_state
->arcs_lsize
[type
], size
);
2590 * If we are growing the cache, and we are adding anonymous
2591 * data, and we have outgrown arc_p, update arc_p
2593 if (arc_size
< arc_c
&& hdr
->b_state
== arc_anon
&&
2594 arc_anon
->arcs_size
+ arc_mru
->arcs_size
> arc_p
)
2595 arc_p
= MIN(arc_c
, arc_p
+ size
);
2600 * This routine is called whenever a buffer is accessed.
2601 * NOTE: the hash lock is dropped in this function.
2604 arc_access(arc_buf_hdr_t
*buf
, kmutex_t
*hash_lock
)
2608 ASSERT(MUTEX_HELD(hash_lock
));
2610 if (buf
->b_state
== arc_anon
) {
2612 * This buffer is not in the cache, and does not
2613 * appear in our "ghost" list. Add the new buffer
2617 ASSERT(buf
->b_arc_access
== 0);
2618 buf
->b_arc_access
= ddi_get_lbolt();
2619 DTRACE_PROBE1(new_state__mru
, arc_buf_hdr_t
*, buf
);
2620 arc_change_state(arc_mru
, buf
, hash_lock
);
2622 } else if (buf
->b_state
== arc_mru
) {
2623 now
= ddi_get_lbolt();
2626 * If this buffer is here because of a prefetch, then either:
2627 * - clear the flag if this is a "referencing" read
2628 * (any subsequent access will bump this into the MFU state).
2630 * - move the buffer to the head of the list if this is
2631 * another prefetch (to make it less likely to be evicted).
2633 if ((buf
->b_flags
& ARC_PREFETCH
) != 0) {
2634 if (refcount_count(&buf
->b_refcnt
) == 0) {
2635 ASSERT(list_link_active(&buf
->b_arc_node
));
2637 buf
->b_flags
&= ~ARC_PREFETCH
;
2638 ARCSTAT_BUMP(arcstat_mru_hits
);
2640 buf
->b_arc_access
= now
;
2645 * This buffer has been "accessed" only once so far,
2646 * but it is still in the cache. Move it to the MFU
2649 if (now
> buf
->b_arc_access
+ ARC_MINTIME
) {
2651 * More than 125ms have passed since we
2652 * instantiated this buffer. Move it to the
2653 * most frequently used state.
2655 buf
->b_arc_access
= now
;
2656 DTRACE_PROBE1(new_state__mfu
, arc_buf_hdr_t
*, buf
);
2657 arc_change_state(arc_mfu
, buf
, hash_lock
);
2659 ARCSTAT_BUMP(arcstat_mru_hits
);
2660 } else if (buf
->b_state
== arc_mru_ghost
) {
2661 arc_state_t
*new_state
;
2663 * This buffer has been "accessed" recently, but
2664 * was evicted from the cache. Move it to the
2668 if (buf
->b_flags
& ARC_PREFETCH
) {
2669 new_state
= arc_mru
;
2670 if (refcount_count(&buf
->b_refcnt
) > 0)
2671 buf
->b_flags
&= ~ARC_PREFETCH
;
2672 DTRACE_PROBE1(new_state__mru
, arc_buf_hdr_t
*, buf
);
2674 new_state
= arc_mfu
;
2675 DTRACE_PROBE1(new_state__mfu
, arc_buf_hdr_t
*, buf
);
2678 buf
->b_arc_access
= ddi_get_lbolt();
2679 arc_change_state(new_state
, buf
, hash_lock
);
2681 ARCSTAT_BUMP(arcstat_mru_ghost_hits
);
2682 } else if (buf
->b_state
== arc_mfu
) {
2684 * This buffer has been accessed more than once and is
2685 * still in the cache. Keep it in the MFU state.
2687 * NOTE: an add_reference() that occurred when we did
2688 * the arc_read() will have kicked this off the list.
2689 * If it was a prefetch, we will explicitly move it to
2690 * the head of the list now.
2692 if ((buf
->b_flags
& ARC_PREFETCH
) != 0) {
2693 ASSERT(refcount_count(&buf
->b_refcnt
) == 0);
2694 ASSERT(list_link_active(&buf
->b_arc_node
));
2696 ARCSTAT_BUMP(arcstat_mfu_hits
);
2697 buf
->b_arc_access
= ddi_get_lbolt();
2698 } else if (buf
->b_state
== arc_mfu_ghost
) {
2699 arc_state_t
*new_state
= arc_mfu
;
2701 * This buffer has been accessed more than once but has
2702 * been evicted from the cache. Move it back to the
2706 if (buf
->b_flags
& ARC_PREFETCH
) {
2708 * This is a prefetch access...
2709 * move this block back to the MRU state.
2711 ASSERT0(refcount_count(&buf
->b_refcnt
));
2712 new_state
= arc_mru
;
2715 buf
->b_arc_access
= ddi_get_lbolt();
2716 DTRACE_PROBE1(new_state__mfu
, arc_buf_hdr_t
*, buf
);
2717 arc_change_state(new_state
, buf
, hash_lock
);
2719 ARCSTAT_BUMP(arcstat_mfu_ghost_hits
);
2720 } else if (buf
->b_state
== arc_l2c_only
) {
2722 * This buffer is on the 2nd Level ARC.
2725 buf
->b_arc_access
= ddi_get_lbolt();
2726 DTRACE_PROBE1(new_state__mfu
, arc_buf_hdr_t
*, buf
);
2727 arc_change_state(arc_mfu
, buf
, hash_lock
);
2729 ASSERT(!"invalid arc state");
2733 /* a generic arc_done_func_t which you can use */
2736 arc_bcopy_func(zio_t
*zio
, arc_buf_t
*buf
, void *arg
)
2738 if (zio
== NULL
|| zio
->io_error
== 0)
2739 bcopy(buf
->b_data
, arg
, buf
->b_hdr
->b_size
);
2740 VERIFY(arc_buf_remove_ref(buf
, arg
) == 1);
2743 /* a generic arc_done_func_t */
2745 arc_getbuf_func(zio_t
*zio
, arc_buf_t
*buf
, void *arg
)
2747 arc_buf_t
**bufp
= arg
;
2748 if (zio
&& zio
->io_error
) {
2749 VERIFY(arc_buf_remove_ref(buf
, arg
) == 1);
2753 ASSERT(buf
->b_data
);
2758 arc_read_done(zio_t
*zio
)
2760 arc_buf_hdr_t
*hdr
, *found
;
2762 arc_buf_t
*abuf
; /* buffer we're assigning to callback */
2763 kmutex_t
*hash_lock
;
2764 arc_callback_t
*callback_list
, *acb
;
2765 int freeable
= FALSE
;
2767 buf
= zio
->io_private
;
2771 * The hdr was inserted into hash-table and removed from lists
2772 * prior to starting I/O. We should find this header, since
2773 * it's in the hash table, and it should be legit since it's
2774 * not possible to evict it during the I/O. The only possible
2775 * reason for it not to be found is if we were freed during the
2778 found
= buf_hash_find(hdr
->b_spa
, &hdr
->b_dva
, hdr
->b_birth
,
2781 ASSERT((found
== NULL
&& HDR_FREED_IN_READ(hdr
) && hash_lock
== NULL
) ||
2782 (found
== hdr
&& DVA_EQUAL(&hdr
->b_dva
, BP_IDENTITY(zio
->io_bp
))) ||
2783 (found
== hdr
&& HDR_L2_READING(hdr
)));
2785 hdr
->b_flags
&= ~ARC_L2_EVICTED
;
2786 if (l2arc_noprefetch
&& (hdr
->b_flags
& ARC_PREFETCH
))
2787 hdr
->b_flags
&= ~ARC_L2CACHE
;
2789 /* byteswap if necessary */
2790 callback_list
= hdr
->b_acb
;
2791 ASSERT(callback_list
!= NULL
);
2792 if (BP_SHOULD_BYTESWAP(zio
->io_bp
) && zio
->io_error
== 0) {
2793 dmu_object_byteswap_t bswap
=
2794 DMU_OT_BYTESWAP(BP_GET_TYPE(zio
->io_bp
));
2795 if (BP_GET_LEVEL(zio
->io_bp
) > 0)
2796 byteswap_uint64_array(buf
->b_data
, hdr
->b_size
);
2798 dmu_ot_byteswap
[bswap
].ob_func(buf
->b_data
, hdr
->b_size
);
2801 arc_cksum_compute(buf
, B_FALSE
);
2803 if (hash_lock
&& zio
->io_error
== 0 && hdr
->b_state
== arc_anon
) {
2805 * Only call arc_access on anonymous buffers. This is because
2806 * if we've issued an I/O for an evicted buffer, we've already
2807 * called arc_access (to prevent any simultaneous readers from
2808 * getting confused).
2810 arc_access(hdr
, hash_lock
);
2813 /* create copies of the data buffer for the callers */
2815 for (acb
= callback_list
; acb
; acb
= acb
->acb_next
) {
2816 if (acb
->acb_done
) {
2818 ARCSTAT_BUMP(arcstat_duplicate_reads
);
2819 abuf
= arc_buf_clone(buf
);
2821 acb
->acb_buf
= abuf
;
2826 hdr
->b_flags
&= ~ARC_IO_IN_PROGRESS
;
2827 ASSERT(!HDR_BUF_AVAILABLE(hdr
));
2829 ASSERT(buf
->b_efunc
== NULL
);
2830 ASSERT(hdr
->b_datacnt
== 1);
2831 hdr
->b_flags
|= ARC_BUF_AVAILABLE
;
2834 ASSERT(refcount_is_zero(&hdr
->b_refcnt
) || callback_list
!= NULL
);
2836 if (zio
->io_error
!= 0) {
2837 hdr
->b_flags
|= ARC_IO_ERROR
;
2838 if (hdr
->b_state
!= arc_anon
)
2839 arc_change_state(arc_anon
, hdr
, hash_lock
);
2840 if (HDR_IN_HASH_TABLE(hdr
))
2841 buf_hash_remove(hdr
);
2842 freeable
= refcount_is_zero(&hdr
->b_refcnt
);
2846 * Broadcast before we drop the hash_lock to avoid the possibility
2847 * that the hdr (and hence the cv) might be freed before we get to
2848 * the cv_broadcast().
2850 cv_broadcast(&hdr
->b_cv
);
2853 mutex_exit(hash_lock
);
2856 * This block was freed while we waited for the read to
2857 * complete. It has been removed from the hash table and
2858 * moved to the anonymous state (so that it won't show up
2861 ASSERT3P(hdr
->b_state
, ==, arc_anon
);
2862 freeable
= refcount_is_zero(&hdr
->b_refcnt
);
2865 /* execute each callback and free its structure */
2866 while ((acb
= callback_list
) != NULL
) {
2868 acb
->acb_done(zio
, acb
->acb_buf
, acb
->acb_private
);
2870 if (acb
->acb_zio_dummy
!= NULL
) {
2871 acb
->acb_zio_dummy
->io_error
= zio
->io_error
;
2872 zio_nowait(acb
->acb_zio_dummy
);
2875 callback_list
= acb
->acb_next
;
2876 kmem_free(acb
, sizeof (arc_callback_t
));
2880 arc_hdr_destroy(hdr
);
2884 * "Read" the block at the specified DVA (in bp) via the
2885 * cache. If the block is found in the cache, invoke the provided
2886 * callback immediately and return. Note that the `zio' parameter
2887 * in the callback will be NULL in this case, since no IO was
2888 * required. If the block is not in the cache pass the read request
2889 * on to the spa with a substitute callback function, so that the
2890 * requested block will be added to the cache.
2892 * If a read request arrives for a block that has a read in-progress,
2893 * either wait for the in-progress read to complete (and return the
2894 * results); or, if this is a read with a "done" func, add a record
2895 * to the read to invoke the "done" func when the read completes,
2896 * and return; or just return.
2898 * arc_read_done() will invoke all the requested "done" functions
2899 * for readers of this block.
2902 arc_read(zio_t
*pio
, spa_t
*spa
, const blkptr_t
*bp
, arc_done_func_t
*done
,
2903 void *private, int priority
, int zio_flags
, uint32_t *arc_flags
,
2904 const zbookmark_t
*zb
)
2907 arc_buf_t
*buf
= NULL
;
2908 kmutex_t
*hash_lock
;
2910 uint64_t guid
= spa_load_guid(spa
);
2913 hdr
= buf_hash_find(guid
, BP_IDENTITY(bp
), BP_PHYSICAL_BIRTH(bp
),
2915 if (hdr
&& hdr
->b_datacnt
> 0) {
2917 *arc_flags
|= ARC_CACHED
;
2919 if (HDR_IO_IN_PROGRESS(hdr
)) {
2921 if (*arc_flags
& ARC_WAIT
) {
2922 cv_wait(&hdr
->b_cv
, hash_lock
);
2923 mutex_exit(hash_lock
);
2926 ASSERT(*arc_flags
& ARC_NOWAIT
);
2929 arc_callback_t
*acb
= NULL
;
2931 acb
= kmem_zalloc(sizeof (arc_callback_t
),
2933 acb
->acb_done
= done
;
2934 acb
->acb_private
= private;
2936 acb
->acb_zio_dummy
= zio_null(pio
,
2937 spa
, NULL
, NULL
, NULL
, zio_flags
);
2939 ASSERT(acb
->acb_done
!= NULL
);
2940 acb
->acb_next
= hdr
->b_acb
;
2942 add_reference(hdr
, hash_lock
, private);
2943 mutex_exit(hash_lock
);
2946 mutex_exit(hash_lock
);
2950 ASSERT(hdr
->b_state
== arc_mru
|| hdr
->b_state
== arc_mfu
);
2953 add_reference(hdr
, hash_lock
, private);
2955 * If this block is already in use, create a new
2956 * copy of the data so that we will be guaranteed
2957 * that arc_release() will always succeed.
2961 ASSERT(buf
->b_data
);
2962 if (HDR_BUF_AVAILABLE(hdr
)) {
2963 ASSERT(buf
->b_efunc
== NULL
);
2964 hdr
->b_flags
&= ~ARC_BUF_AVAILABLE
;
2966 buf
= arc_buf_clone(buf
);
2969 } else if (*arc_flags
& ARC_PREFETCH
&&
2970 refcount_count(&hdr
->b_refcnt
) == 0) {
2971 hdr
->b_flags
|= ARC_PREFETCH
;
2973 DTRACE_PROBE1(arc__hit
, arc_buf_hdr_t
*, hdr
);
2974 arc_access(hdr
, hash_lock
);
2975 if (*arc_flags
& ARC_L2CACHE
)
2976 hdr
->b_flags
|= ARC_L2CACHE
;
2977 mutex_exit(hash_lock
);
2978 ARCSTAT_BUMP(arcstat_hits
);
2979 ARCSTAT_CONDSTAT(!(hdr
->b_flags
& ARC_PREFETCH
),
2980 demand
, prefetch
, hdr
->b_type
!= ARC_BUFC_METADATA
,
2981 data
, metadata
, hits
);
2984 done(NULL
, buf
, private);
2986 uint64_t size
= BP_GET_LSIZE(bp
);
2987 arc_callback_t
*acb
;
2990 boolean_t devw
= B_FALSE
;
2993 /* this block is not in the cache */
2994 arc_buf_hdr_t
*exists
;
2995 arc_buf_contents_t type
= BP_GET_BUFC_TYPE(bp
);
2996 buf
= arc_buf_alloc(spa
, size
, private, type
);
2998 hdr
->b_dva
= *BP_IDENTITY(bp
);
2999 hdr
->b_birth
= BP_PHYSICAL_BIRTH(bp
);
3000 hdr
->b_cksum0
= bp
->blk_cksum
.zc_word
[0];
3001 exists
= buf_hash_insert(hdr
, &hash_lock
);
3003 /* somebody beat us to the hash insert */
3004 mutex_exit(hash_lock
);
3005 buf_discard_identity(hdr
);
3006 (void) arc_buf_remove_ref(buf
, private);
3007 goto top
; /* restart the IO request */
3009 /* if this is a prefetch, we don't have a reference */
3010 if (*arc_flags
& ARC_PREFETCH
) {
3011 (void) remove_reference(hdr
, hash_lock
,
3013 hdr
->b_flags
|= ARC_PREFETCH
;
3015 if (*arc_flags
& ARC_L2CACHE
)
3016 hdr
->b_flags
|= ARC_L2CACHE
;
3017 if (BP_GET_LEVEL(bp
) > 0)
3018 hdr
->b_flags
|= ARC_INDIRECT
;
3020 /* this block is in the ghost cache */
3021 ASSERT(GHOST_STATE(hdr
->b_state
));
3022 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
3023 ASSERT0(refcount_count(&hdr
->b_refcnt
));
3024 ASSERT(hdr
->b_buf
== NULL
);
3026 /* if this is a prefetch, we don't have a reference */
3027 if (*arc_flags
& ARC_PREFETCH
)
3028 hdr
->b_flags
|= ARC_PREFETCH
;
3030 add_reference(hdr
, hash_lock
, private);
3031 if (*arc_flags
& ARC_L2CACHE
)
3032 hdr
->b_flags
|= ARC_L2CACHE
;
3033 buf
= kmem_cache_alloc(buf_cache
, KM_PUSHPAGE
);
3036 buf
->b_efunc
= NULL
;
3037 buf
->b_private
= NULL
;
3040 ASSERT(hdr
->b_datacnt
== 0);
3042 arc_get_data_buf(buf
);
3043 arc_access(hdr
, hash_lock
);
3046 ASSERT(!GHOST_STATE(hdr
->b_state
));
3048 acb
= kmem_zalloc(sizeof (arc_callback_t
), KM_PUSHPAGE
);
3049 acb
->acb_done
= done
;
3050 acb
->acb_private
= private;
3052 ASSERT(hdr
->b_acb
== NULL
);
3054 hdr
->b_flags
|= ARC_IO_IN_PROGRESS
;
3056 if (HDR_L2CACHE(hdr
) && hdr
->b_l2hdr
!= NULL
&&
3057 (vd
= hdr
->b_l2hdr
->b_dev
->l2ad_vdev
) != NULL
) {
3058 devw
= hdr
->b_l2hdr
->b_dev
->l2ad_writing
;
3059 addr
= hdr
->b_l2hdr
->b_daddr
;
3061 * Lock out device removal.
3063 if (vdev_is_dead(vd
) ||
3064 !spa_config_tryenter(spa
, SCL_L2ARC
, vd
, RW_READER
))
3068 mutex_exit(hash_lock
);
3070 ASSERT3U(hdr
->b_size
, ==, size
);
3071 DTRACE_PROBE4(arc__miss
, arc_buf_hdr_t
*, hdr
, blkptr_t
*, bp
,
3072 uint64_t, size
, zbookmark_t
*, zb
);
3073 ARCSTAT_BUMP(arcstat_misses
);
3074 ARCSTAT_CONDSTAT(!(hdr
->b_flags
& ARC_PREFETCH
),
3075 demand
, prefetch
, hdr
->b_type
!= ARC_BUFC_METADATA
,
3076 data
, metadata
, misses
);
3078 if (vd
!= NULL
&& l2arc_ndev
!= 0 && !(l2arc_norw
&& devw
)) {
3080 * Read from the L2ARC if the following are true:
3081 * 1. The L2ARC vdev was previously cached.
3082 * 2. This buffer still has L2ARC metadata.
3083 * 3. This buffer isn't currently writing to the L2ARC.
3084 * 4. The L2ARC entry wasn't evicted, which may
3085 * also have invalidated the vdev.
3086 * 5. This isn't prefetch and l2arc_noprefetch is set.
3088 if (hdr
->b_l2hdr
!= NULL
&&
3089 !HDR_L2_WRITING(hdr
) && !HDR_L2_EVICTED(hdr
) &&
3090 !(l2arc_noprefetch
&& HDR_PREFETCH(hdr
))) {
3091 l2arc_read_callback_t
*cb
;
3093 DTRACE_PROBE1(l2arc__hit
, arc_buf_hdr_t
*, hdr
);
3094 ARCSTAT_BUMP(arcstat_l2_hits
);
3096 cb
= kmem_zalloc(sizeof (l2arc_read_callback_t
),
3098 cb
->l2rcb_buf
= buf
;
3099 cb
->l2rcb_spa
= spa
;
3102 cb
->l2rcb_flags
= zio_flags
;
3105 * l2arc read. The SCL_L2ARC lock will be
3106 * released by l2arc_read_done().
3108 rzio
= zio_read_phys(pio
, vd
, addr
, size
,
3109 buf
->b_data
, ZIO_CHECKSUM_OFF
,
3110 l2arc_read_done
, cb
, priority
, zio_flags
|
3111 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_CANFAIL
|
3112 ZIO_FLAG_DONT_PROPAGATE
|
3113 ZIO_FLAG_DONT_RETRY
, B_FALSE
);
3114 DTRACE_PROBE2(l2arc__read
, vdev_t
*, vd
,
3116 ARCSTAT_INCR(arcstat_l2_read_bytes
, size
);
3118 if (*arc_flags
& ARC_NOWAIT
) {
3123 ASSERT(*arc_flags
& ARC_WAIT
);
3124 if (zio_wait(rzio
) == 0)
3127 /* l2arc read error; goto zio_read() */
3129 DTRACE_PROBE1(l2arc__miss
,
3130 arc_buf_hdr_t
*, hdr
);
3131 ARCSTAT_BUMP(arcstat_l2_misses
);
3132 if (HDR_L2_WRITING(hdr
))
3133 ARCSTAT_BUMP(arcstat_l2_rw_clash
);
3134 spa_config_exit(spa
, SCL_L2ARC
, vd
);
3138 spa_config_exit(spa
, SCL_L2ARC
, vd
);
3139 if (l2arc_ndev
!= 0) {
3140 DTRACE_PROBE1(l2arc__miss
,
3141 arc_buf_hdr_t
*, hdr
);
3142 ARCSTAT_BUMP(arcstat_l2_misses
);
3146 rzio
= zio_read(pio
, spa
, bp
, buf
->b_data
, size
,
3147 arc_read_done
, buf
, priority
, zio_flags
, zb
);
3149 if (*arc_flags
& ARC_WAIT
)
3150 return (zio_wait(rzio
));
3152 ASSERT(*arc_flags
& ARC_NOWAIT
);
3159 arc_add_prune_callback(arc_prune_func_t
*func
, void *private)
3163 p
= kmem_alloc(sizeof(*p
), KM_SLEEP
);
3165 p
->p_private
= private;
3166 list_link_init(&p
->p_node
);
3167 refcount_create(&p
->p_refcnt
);
3169 mutex_enter(&arc_prune_mtx
);
3170 refcount_add(&p
->p_refcnt
, &arc_prune_list
);
3171 list_insert_head(&arc_prune_list
, p
);
3172 mutex_exit(&arc_prune_mtx
);
3178 arc_remove_prune_callback(arc_prune_t
*p
)
3180 mutex_enter(&arc_prune_mtx
);
3181 list_remove(&arc_prune_list
, p
);
3182 if (refcount_remove(&p
->p_refcnt
, &arc_prune_list
) == 0) {
3183 refcount_destroy(&p
->p_refcnt
);
3184 kmem_free(p
, sizeof (*p
));
3186 mutex_exit(&arc_prune_mtx
);
3190 arc_set_callback(arc_buf_t
*buf
, arc_evict_func_t
*func
, void *private)
3192 ASSERT(buf
->b_hdr
!= NULL
);
3193 ASSERT(buf
->b_hdr
->b_state
!= arc_anon
);
3194 ASSERT(!refcount_is_zero(&buf
->b_hdr
->b_refcnt
) || func
== NULL
);
3195 ASSERT(buf
->b_efunc
== NULL
);
3196 ASSERT(!HDR_BUF_AVAILABLE(buf
->b_hdr
));
3198 buf
->b_efunc
= func
;
3199 buf
->b_private
= private;
3203 * Notify the arc that a block was freed, and thus will never be used again.
3206 arc_freed(spa_t
*spa
, const blkptr_t
*bp
)
3209 kmutex_t
*hash_lock
;
3210 uint64_t guid
= spa_load_guid(spa
);
3212 hdr
= buf_hash_find(guid
, BP_IDENTITY(bp
), BP_PHYSICAL_BIRTH(bp
),
3216 if (HDR_BUF_AVAILABLE(hdr
)) {
3217 arc_buf_t
*buf
= hdr
->b_buf
;
3218 add_reference(hdr
, hash_lock
, FTAG
);
3219 hdr
->b_flags
&= ~ARC_BUF_AVAILABLE
;
3220 mutex_exit(hash_lock
);
3222 arc_release(buf
, FTAG
);
3223 (void) arc_buf_remove_ref(buf
, FTAG
);
3225 mutex_exit(hash_lock
);
3231 * This is used by the DMU to let the ARC know that a buffer is
3232 * being evicted, so the ARC should clean up. If this arc buf
3233 * is not yet in the evicted state, it will be put there.
3236 arc_buf_evict(arc_buf_t
*buf
)
3239 kmutex_t
*hash_lock
;
3242 mutex_enter(&buf
->b_evict_lock
);
3246 * We are in arc_do_user_evicts().
3248 ASSERT(buf
->b_data
== NULL
);
3249 mutex_exit(&buf
->b_evict_lock
);
3251 } else if (buf
->b_data
== NULL
) {
3252 arc_buf_t copy
= *buf
; /* structure assignment */
3254 * We are on the eviction list; process this buffer now
3255 * but let arc_do_user_evicts() do the reaping.
3257 buf
->b_efunc
= NULL
;
3258 mutex_exit(&buf
->b_evict_lock
);
3259 VERIFY(copy
.b_efunc(©
) == 0);
3262 hash_lock
= HDR_LOCK(hdr
);
3263 mutex_enter(hash_lock
);
3265 ASSERT3P(hash_lock
, ==, HDR_LOCK(hdr
));
3267 ASSERT3U(refcount_count(&hdr
->b_refcnt
), <, hdr
->b_datacnt
);
3268 ASSERT(hdr
->b_state
== arc_mru
|| hdr
->b_state
== arc_mfu
);
3271 * Pull this buffer off of the hdr
3274 while (*bufp
!= buf
)
3275 bufp
= &(*bufp
)->b_next
;
3276 *bufp
= buf
->b_next
;
3278 ASSERT(buf
->b_data
!= NULL
);
3279 arc_buf_destroy(buf
, FALSE
, FALSE
);
3281 if (hdr
->b_datacnt
== 0) {
3282 arc_state_t
*old_state
= hdr
->b_state
;
3283 arc_state_t
*evicted_state
;
3285 ASSERT(hdr
->b_buf
== NULL
);
3286 ASSERT(refcount_is_zero(&hdr
->b_refcnt
));
3289 (old_state
== arc_mru
) ? arc_mru_ghost
: arc_mfu_ghost
;
3291 mutex_enter(&old_state
->arcs_mtx
);
3292 mutex_enter(&evicted_state
->arcs_mtx
);
3294 arc_change_state(evicted_state
, hdr
, hash_lock
);
3295 ASSERT(HDR_IN_HASH_TABLE(hdr
));
3296 hdr
->b_flags
|= ARC_IN_HASH_TABLE
;
3297 hdr
->b_flags
&= ~ARC_BUF_AVAILABLE
;
3299 mutex_exit(&evicted_state
->arcs_mtx
);
3300 mutex_exit(&old_state
->arcs_mtx
);
3302 mutex_exit(hash_lock
);
3303 mutex_exit(&buf
->b_evict_lock
);
3305 VERIFY(buf
->b_efunc(buf
) == 0);
3306 buf
->b_efunc
= NULL
;
3307 buf
->b_private
= NULL
;
3310 kmem_cache_free(buf_cache
, buf
);
3315 * Release this buffer from the cache. This must be done
3316 * after a read and prior to modifying the buffer contents.
3317 * If the buffer has more than one reference, we must make
3318 * a new hdr for the buffer.
3321 arc_release(arc_buf_t
*buf
, void *tag
)
3324 kmutex_t
*hash_lock
= NULL
;
3325 l2arc_buf_hdr_t
*l2hdr
;
3326 uint64_t buf_size
= 0;
3329 * It would be nice to assert that if it's DMU metadata (level >
3330 * 0 || it's the dnode file), then it must be syncing context.
3331 * But we don't know that information at this level.
3334 mutex_enter(&buf
->b_evict_lock
);
3337 /* this buffer is not on any list */
3338 ASSERT(refcount_count(&hdr
->b_refcnt
) > 0);
3340 if (hdr
->b_state
== arc_anon
) {
3341 /* this buffer is already released */
3342 ASSERT(buf
->b_efunc
== NULL
);
3344 hash_lock
= HDR_LOCK(hdr
);
3345 mutex_enter(hash_lock
);
3347 ASSERT3P(hash_lock
, ==, HDR_LOCK(hdr
));
3350 l2hdr
= hdr
->b_l2hdr
;
3352 mutex_enter(&l2arc_buflist_mtx
);
3353 hdr
->b_l2hdr
= NULL
;
3354 buf_size
= hdr
->b_size
;
3358 * Do we have more than one buf?
3360 if (hdr
->b_datacnt
> 1) {
3361 arc_buf_hdr_t
*nhdr
;
3363 uint64_t blksz
= hdr
->b_size
;
3364 uint64_t spa
= hdr
->b_spa
;
3365 arc_buf_contents_t type
= hdr
->b_type
;
3366 uint32_t flags
= hdr
->b_flags
;
3368 ASSERT(hdr
->b_buf
!= buf
|| buf
->b_next
!= NULL
);
3370 * Pull the data off of this hdr and attach it to
3371 * a new anonymous hdr.
3373 (void) remove_reference(hdr
, hash_lock
, tag
);
3375 while (*bufp
!= buf
)
3376 bufp
= &(*bufp
)->b_next
;
3377 *bufp
= buf
->b_next
;
3380 ASSERT3U(hdr
->b_state
->arcs_size
, >=, hdr
->b_size
);
3381 atomic_add_64(&hdr
->b_state
->arcs_size
, -hdr
->b_size
);
3382 if (refcount_is_zero(&hdr
->b_refcnt
)) {
3383 uint64_t *size
= &hdr
->b_state
->arcs_lsize
[hdr
->b_type
];
3384 ASSERT3U(*size
, >=, hdr
->b_size
);
3385 atomic_add_64(size
, -hdr
->b_size
);
3389 * We're releasing a duplicate user data buffer, update
3390 * our statistics accordingly.
3392 if (hdr
->b_type
== ARC_BUFC_DATA
) {
3393 ARCSTAT_BUMPDOWN(arcstat_duplicate_buffers
);
3394 ARCSTAT_INCR(arcstat_duplicate_buffers_size
,
3397 hdr
->b_datacnt
-= 1;
3398 arc_cksum_verify(buf
);
3400 mutex_exit(hash_lock
);
3402 nhdr
= kmem_cache_alloc(hdr_cache
, KM_PUSHPAGE
);
3403 nhdr
->b_size
= blksz
;
3405 nhdr
->b_type
= type
;
3407 nhdr
->b_state
= arc_anon
;
3408 nhdr
->b_arc_access
= 0;
3409 nhdr
->b_flags
= flags
& ARC_L2_WRITING
;
3410 nhdr
->b_l2hdr
= NULL
;
3411 nhdr
->b_datacnt
= 1;
3412 nhdr
->b_freeze_cksum
= NULL
;
3413 (void) refcount_add(&nhdr
->b_refcnt
, tag
);
3415 mutex_exit(&buf
->b_evict_lock
);
3416 atomic_add_64(&arc_anon
->arcs_size
, blksz
);
3418 mutex_exit(&buf
->b_evict_lock
);
3419 ASSERT(refcount_count(&hdr
->b_refcnt
) == 1);
3420 ASSERT(!list_link_active(&hdr
->b_arc_node
));
3421 ASSERT(!HDR_IO_IN_PROGRESS(hdr
));
3422 if (hdr
->b_state
!= arc_anon
)
3423 arc_change_state(arc_anon
, hdr
, hash_lock
);
3424 hdr
->b_arc_access
= 0;
3426 mutex_exit(hash_lock
);
3428 buf_discard_identity(hdr
);
3431 buf
->b_efunc
= NULL
;
3432 buf
->b_private
= NULL
;
3435 list_remove(l2hdr
->b_dev
->l2ad_buflist
, hdr
);
3436 kmem_free(l2hdr
, sizeof (l2arc_buf_hdr_t
));
3437 arc_space_return(L2HDR_SIZE
, ARC_SPACE_L2HDRS
);
3438 ARCSTAT_INCR(arcstat_l2_size
, -buf_size
);
3439 mutex_exit(&l2arc_buflist_mtx
);
3444 arc_released(arc_buf_t
*buf
)
3448 mutex_enter(&buf
->b_evict_lock
);
3449 released
= (buf
->b_data
!= NULL
&& buf
->b_hdr
->b_state
== arc_anon
);
3450 mutex_exit(&buf
->b_evict_lock
);
3455 arc_has_callback(arc_buf_t
*buf
)
3459 mutex_enter(&buf
->b_evict_lock
);
3460 callback
= (buf
->b_efunc
!= NULL
);
3461 mutex_exit(&buf
->b_evict_lock
);
3467 arc_referenced(arc_buf_t
*buf
)
3471 mutex_enter(&buf
->b_evict_lock
);
3472 referenced
= (refcount_count(&buf
->b_hdr
->b_refcnt
));
3473 mutex_exit(&buf
->b_evict_lock
);
3474 return (referenced
);
3479 arc_write_ready(zio_t
*zio
)
3481 arc_write_callback_t
*callback
= zio
->io_private
;
3482 arc_buf_t
*buf
= callback
->awcb_buf
;
3483 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
3485 ASSERT(!refcount_is_zero(&buf
->b_hdr
->b_refcnt
));
3486 callback
->awcb_ready(zio
, buf
, callback
->awcb_private
);
3489 * If the IO is already in progress, then this is a re-write
3490 * attempt, so we need to thaw and re-compute the cksum.
3491 * It is the responsibility of the callback to handle the
3492 * accounting for any re-write attempt.
3494 if (HDR_IO_IN_PROGRESS(hdr
)) {
3495 mutex_enter(&hdr
->b_freeze_lock
);
3496 if (hdr
->b_freeze_cksum
!= NULL
) {
3497 kmem_free(hdr
->b_freeze_cksum
, sizeof (zio_cksum_t
));
3498 hdr
->b_freeze_cksum
= NULL
;
3500 mutex_exit(&hdr
->b_freeze_lock
);
3502 arc_cksum_compute(buf
, B_FALSE
);
3503 hdr
->b_flags
|= ARC_IO_IN_PROGRESS
;
3507 arc_write_done(zio_t
*zio
)
3509 arc_write_callback_t
*callback
= zio
->io_private
;
3510 arc_buf_t
*buf
= callback
->awcb_buf
;
3511 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
3513 ASSERT(hdr
->b_acb
== NULL
);
3515 if (zio
->io_error
== 0) {
3516 hdr
->b_dva
= *BP_IDENTITY(zio
->io_bp
);
3517 hdr
->b_birth
= BP_PHYSICAL_BIRTH(zio
->io_bp
);
3518 hdr
->b_cksum0
= zio
->io_bp
->blk_cksum
.zc_word
[0];
3520 ASSERT(BUF_EMPTY(hdr
));
3524 * If the block to be written was all-zero, we may have
3525 * compressed it away. In this case no write was performed
3526 * so there will be no dva/birth/checksum. The buffer must
3527 * therefore remain anonymous (and uncached).
3529 if (!BUF_EMPTY(hdr
)) {
3530 arc_buf_hdr_t
*exists
;
3531 kmutex_t
*hash_lock
;
3533 ASSERT(zio
->io_error
== 0);
3535 arc_cksum_verify(buf
);
3537 exists
= buf_hash_insert(hdr
, &hash_lock
);
3540 * This can only happen if we overwrite for
3541 * sync-to-convergence, because we remove
3542 * buffers from the hash table when we arc_free().
3544 if (zio
->io_flags
& ZIO_FLAG_IO_REWRITE
) {
3545 if (!BP_EQUAL(&zio
->io_bp_orig
, zio
->io_bp
))
3546 panic("bad overwrite, hdr=%p exists=%p",
3547 (void *)hdr
, (void *)exists
);
3548 ASSERT(refcount_is_zero(&exists
->b_refcnt
));
3549 arc_change_state(arc_anon
, exists
, hash_lock
);
3550 mutex_exit(hash_lock
);
3551 arc_hdr_destroy(exists
);
3552 exists
= buf_hash_insert(hdr
, &hash_lock
);
3553 ASSERT3P(exists
, ==, NULL
);
3556 ASSERT(hdr
->b_datacnt
== 1);
3557 ASSERT(hdr
->b_state
== arc_anon
);
3558 ASSERT(BP_GET_DEDUP(zio
->io_bp
));
3559 ASSERT(BP_GET_LEVEL(zio
->io_bp
) == 0);
3562 hdr
->b_flags
&= ~ARC_IO_IN_PROGRESS
;
3563 /* if it's not anon, we are doing a scrub */
3564 if (!exists
&& hdr
->b_state
== arc_anon
)
3565 arc_access(hdr
, hash_lock
);
3566 mutex_exit(hash_lock
);
3568 hdr
->b_flags
&= ~ARC_IO_IN_PROGRESS
;
3571 ASSERT(!refcount_is_zero(&hdr
->b_refcnt
));
3572 callback
->awcb_done(zio
, buf
, callback
->awcb_private
);
3574 kmem_free(callback
, sizeof (arc_write_callback_t
));
3578 arc_write(zio_t
*pio
, spa_t
*spa
, uint64_t txg
,
3579 blkptr_t
*bp
, arc_buf_t
*buf
, boolean_t l2arc
, const zio_prop_t
*zp
,
3580 arc_done_func_t
*ready
, arc_done_func_t
*done
, void *private,
3581 int priority
, int zio_flags
, const zbookmark_t
*zb
)
3583 arc_buf_hdr_t
*hdr
= buf
->b_hdr
;
3584 arc_write_callback_t
*callback
;
3587 ASSERT(ready
!= NULL
);
3588 ASSERT(done
!= NULL
);
3589 ASSERT(!HDR_IO_ERROR(hdr
));
3590 ASSERT((hdr
->b_flags
& ARC_IO_IN_PROGRESS
) == 0);
3591 ASSERT(hdr
->b_acb
== NULL
);
3593 hdr
->b_flags
|= ARC_L2CACHE
;
3594 callback
= kmem_zalloc(sizeof (arc_write_callback_t
), KM_PUSHPAGE
);
3595 callback
->awcb_ready
= ready
;
3596 callback
->awcb_done
= done
;
3597 callback
->awcb_private
= private;
3598 callback
->awcb_buf
= buf
;
3600 zio
= zio_write(pio
, spa
, txg
, bp
, buf
->b_data
, hdr
->b_size
, zp
,
3601 arc_write_ready
, arc_write_done
, callback
, priority
, zio_flags
, zb
);
3607 arc_memory_throttle(uint64_t reserve
, uint64_t inflight_data
, uint64_t txg
)
3610 uint64_t available_memory
;
3612 if (zfs_arc_memory_throttle_disable
)
3615 /* Easily reclaimable memory (free + inactive + arc-evictable) */
3616 available_memory
= ptob(spl_kmem_availrmem()) + arc_evictable_memory();
3618 if (available_memory
<= zfs_write_limit_max
) {
3619 ARCSTAT_INCR(arcstat_memory_throttle_count
, 1);
3620 DMU_TX_STAT_BUMP(dmu_tx_memory_reclaim
);
3624 if (inflight_data
> available_memory
/ 4) {
3625 ARCSTAT_INCR(arcstat_memory_throttle_count
, 1);
3626 DMU_TX_STAT_BUMP(dmu_tx_memory_inflight
);
3634 arc_tempreserve_clear(uint64_t reserve
)
3636 atomic_add_64(&arc_tempreserve
, -reserve
);
3637 ASSERT((int64_t)arc_tempreserve
>= 0);
3641 arc_tempreserve_space(uint64_t reserve
, uint64_t txg
)
3648 * Once in a while, fail for no reason. Everything should cope.
3650 if (spa_get_random(10000) == 0) {
3651 dprintf("forcing random failure\n");
3655 if (reserve
> arc_c
/4 && !arc_no_grow
)
3656 arc_c
= MIN(arc_c_max
, reserve
* 4);
3657 if (reserve
> arc_c
) {
3658 DMU_TX_STAT_BUMP(dmu_tx_memory_reserve
);
3663 * Don't count loaned bufs as in flight dirty data to prevent long
3664 * network delays from blocking transactions that are ready to be
3665 * assigned to a txg.
3667 anon_size
= MAX((int64_t)(arc_anon
->arcs_size
- arc_loaned_bytes
), 0);
3670 * Writes will, almost always, require additional memory allocations
3671 * in order to compress/encrypt/etc the data. We therefor need to
3672 * make sure that there is sufficient available memory for this.
3674 if ((error
= arc_memory_throttle(reserve
, anon_size
, txg
)))
3678 * Throttle writes when the amount of dirty data in the cache
3679 * gets too large. We try to keep the cache less than half full
3680 * of dirty blocks so that our sync times don't grow too large.
3681 * Note: if two requests come in concurrently, we might let them
3682 * both succeed, when one of them should fail. Not a huge deal.
3685 if (reserve
+ arc_tempreserve
+ anon_size
> arc_c
/ 2 &&
3686 anon_size
> arc_c
/ 4) {
3687 dprintf("failing, arc_tempreserve=%lluK anon_meta=%lluK "
3688 "anon_data=%lluK tempreserve=%lluK arc_c=%lluK\n",
3689 arc_tempreserve
>>10,
3690 arc_anon
->arcs_lsize
[ARC_BUFC_METADATA
]>>10,
3691 arc_anon
->arcs_lsize
[ARC_BUFC_DATA
]>>10,
3692 reserve
>>10, arc_c
>>10);
3693 DMU_TX_STAT_BUMP(dmu_tx_dirty_throttle
);
3696 atomic_add_64(&arc_tempreserve
, reserve
);
3701 arc_kstat_update_state(arc_state_t
*state
, kstat_named_t
*size
,
3702 kstat_named_t
*evict_data
, kstat_named_t
*evict_metadata
)
3704 size
->value
.ui64
= state
->arcs_size
;
3705 evict_data
->value
.ui64
= state
->arcs_lsize
[ARC_BUFC_DATA
];
3706 evict_metadata
->value
.ui64
= state
->arcs_lsize
[ARC_BUFC_METADATA
];
3710 arc_kstat_update(kstat_t
*ksp
, int rw
)
3712 arc_stats_t
*as
= ksp
->ks_data
;
3714 if (rw
== KSTAT_WRITE
) {
3717 arc_kstat_update_state(arc_anon
,
3718 &as
->arcstat_anon_size
,
3719 &as
->arcstat_anon_evict_data
,
3720 &as
->arcstat_anon_evict_metadata
);
3721 arc_kstat_update_state(arc_mru
,
3722 &as
->arcstat_mru_size
,
3723 &as
->arcstat_mru_evict_data
,
3724 &as
->arcstat_mru_evict_metadata
);
3725 arc_kstat_update_state(arc_mru_ghost
,
3726 &as
->arcstat_mru_ghost_size
,
3727 &as
->arcstat_mru_ghost_evict_data
,
3728 &as
->arcstat_mru_ghost_evict_metadata
);
3729 arc_kstat_update_state(arc_mfu
,
3730 &as
->arcstat_mfu_size
,
3731 &as
->arcstat_mfu_evict_data
,
3732 &as
->arcstat_mfu_evict_metadata
);
3733 arc_kstat_update_state(arc_mfu_ghost
,
3734 &as
->arcstat_mfu_ghost_size
,
3735 &as
->arcstat_mfu_ghost_evict_data
,
3736 &as
->arcstat_mfu_ghost_evict_metadata
);
3745 mutex_init(&arc_reclaim_thr_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
3746 cv_init(&arc_reclaim_thr_cv
, NULL
, CV_DEFAULT
, NULL
);
3748 /* Convert seconds to clock ticks */
3749 zfs_arc_min_prefetch_lifespan
= 1 * hz
;
3751 /* Start out with 1/8 of all memory */
3752 arc_c
= physmem
* PAGESIZE
/ 8;
3756 * On architectures where the physical memory can be larger
3757 * than the addressable space (intel in 32-bit mode), we may
3758 * need to limit the cache to 1/8 of VM size.
3760 arc_c
= MIN(arc_c
, vmem_size(heap_arena
, VMEM_ALLOC
| VMEM_FREE
) / 8);
3762 * Register a shrinker to support synchronous (direct) memory
3763 * reclaim from the arc. This is done to prevent kswapd from
3764 * swapping out pages when it is preferable to shrink the arc.
3766 spl_register_shrinker(&arc_shrinker
);
3769 /* set min cache to 1/32 of all memory, or 64MB, whichever is more */
3770 arc_c_min
= MAX(arc_c
/ 4, 64<<20);
3771 /* set max to 1/2 of all memory */
3772 arc_c_max
= MAX(arc_c
* 4, arc_c_max
);
3775 * Allow the tunables to override our calculations if they are
3776 * reasonable (ie. over 64MB)
3778 if (zfs_arc_max
> 64<<20 && zfs_arc_max
< physmem
* PAGESIZE
)
3779 arc_c_max
= zfs_arc_max
;
3780 if (zfs_arc_min
> 64<<20 && zfs_arc_min
<= arc_c_max
)
3781 arc_c_min
= zfs_arc_min
;
3784 arc_p
= (arc_c
>> 1);
3786 /* limit meta-data to 1/4 of the arc capacity */
3787 arc_meta_limit
= arc_c_max
/ 4;
3790 /* Allow the tunable to override if it is reasonable */
3791 if (zfs_arc_meta_limit
> 0 && zfs_arc_meta_limit
<= arc_c_max
)
3792 arc_meta_limit
= zfs_arc_meta_limit
;
3794 if (arc_c_min
< arc_meta_limit
/ 2 && zfs_arc_min
== 0)
3795 arc_c_min
= arc_meta_limit
/ 2;
3797 /* if kmem_flags are set, lets try to use less memory */
3798 if (kmem_debugging())
3800 if (arc_c
< arc_c_min
)
3803 arc_anon
= &ARC_anon
;
3805 arc_mru_ghost
= &ARC_mru_ghost
;
3807 arc_mfu_ghost
= &ARC_mfu_ghost
;
3808 arc_l2c_only
= &ARC_l2c_only
;
3811 mutex_init(&arc_anon
->arcs_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
3812 mutex_init(&arc_mru
->arcs_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
3813 mutex_init(&arc_mru_ghost
->arcs_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
3814 mutex_init(&arc_mfu
->arcs_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
3815 mutex_init(&arc_mfu_ghost
->arcs_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
3816 mutex_init(&arc_l2c_only
->arcs_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
3818 list_create(&arc_mru
->arcs_list
[ARC_BUFC_METADATA
],
3819 sizeof (arc_buf_hdr_t
), offsetof(arc_buf_hdr_t
, b_arc_node
));
3820 list_create(&arc_mru
->arcs_list
[ARC_BUFC_DATA
],
3821 sizeof (arc_buf_hdr_t
), offsetof(arc_buf_hdr_t
, b_arc_node
));
3822 list_create(&arc_mru_ghost
->arcs_list
[ARC_BUFC_METADATA
],
3823 sizeof (arc_buf_hdr_t
), offsetof(arc_buf_hdr_t
, b_arc_node
));
3824 list_create(&arc_mru_ghost
->arcs_list
[ARC_BUFC_DATA
],
3825 sizeof (arc_buf_hdr_t
), offsetof(arc_buf_hdr_t
, b_arc_node
));
3826 list_create(&arc_mfu
->arcs_list
[ARC_BUFC_METADATA
],
3827 sizeof (arc_buf_hdr_t
), offsetof(arc_buf_hdr_t
, b_arc_node
));
3828 list_create(&arc_mfu
->arcs_list
[ARC_BUFC_DATA
],
3829 sizeof (arc_buf_hdr_t
), offsetof(arc_buf_hdr_t
, b_arc_node
));
3830 list_create(&arc_mfu_ghost
->arcs_list
[ARC_BUFC_METADATA
],
3831 sizeof (arc_buf_hdr_t
), offsetof(arc_buf_hdr_t
, b_arc_node
));
3832 list_create(&arc_mfu_ghost
->arcs_list
[ARC_BUFC_DATA
],
3833 sizeof (arc_buf_hdr_t
), offsetof(arc_buf_hdr_t
, b_arc_node
));
3834 list_create(&arc_l2c_only
->arcs_list
[ARC_BUFC_METADATA
],
3835 sizeof (arc_buf_hdr_t
), offsetof(arc_buf_hdr_t
, b_arc_node
));
3836 list_create(&arc_l2c_only
->arcs_list
[ARC_BUFC_DATA
],
3837 sizeof (arc_buf_hdr_t
), offsetof(arc_buf_hdr_t
, b_arc_node
));
3841 arc_thread_exit
= 0;
3842 list_create(&arc_prune_list
, sizeof (arc_prune_t
),
3843 offsetof(arc_prune_t
, p_node
));
3844 arc_eviction_list
= NULL
;
3845 mutex_init(&arc_prune_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
3846 mutex_init(&arc_eviction_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
3847 bzero(&arc_eviction_hdr
, sizeof (arc_buf_hdr_t
));
3849 arc_ksp
= kstat_create("zfs", 0, "arcstats", "misc", KSTAT_TYPE_NAMED
,
3850 sizeof (arc_stats
) / sizeof (kstat_named_t
), KSTAT_FLAG_VIRTUAL
);
3852 if (arc_ksp
!= NULL
) {
3853 arc_ksp
->ks_data
= &arc_stats
;
3854 arc_ksp
->ks_update
= arc_kstat_update
;
3855 kstat_install(arc_ksp
);
3858 (void) thread_create(NULL
, 0, arc_adapt_thread
, NULL
, 0, &p0
,
3859 TS_RUN
, minclsyspri
);
3864 if (zfs_write_limit_max
== 0)
3865 zfs_write_limit_max
= ptob(physmem
) >> zfs_write_limit_shift
;
3867 zfs_write_limit_shift
= 0;
3868 mutex_init(&zfs_write_limit_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
3876 mutex_enter(&arc_reclaim_thr_lock
);
3878 spl_unregister_shrinker(&arc_shrinker
);
3879 #endif /* _KERNEL */
3881 arc_thread_exit
= 1;
3882 while (arc_thread_exit
!= 0)
3883 cv_wait(&arc_reclaim_thr_cv
, &arc_reclaim_thr_lock
);
3884 mutex_exit(&arc_reclaim_thr_lock
);
3890 if (arc_ksp
!= NULL
) {
3891 kstat_delete(arc_ksp
);
3895 mutex_enter(&arc_prune_mtx
);
3896 while ((p
= list_head(&arc_prune_list
)) != NULL
) {
3897 list_remove(&arc_prune_list
, p
);
3898 refcount_remove(&p
->p_refcnt
, &arc_prune_list
);
3899 refcount_destroy(&p
->p_refcnt
);
3900 kmem_free(p
, sizeof (*p
));
3902 mutex_exit(&arc_prune_mtx
);
3904 list_destroy(&arc_prune_list
);
3905 mutex_destroy(&arc_prune_mtx
);
3906 mutex_destroy(&arc_eviction_mtx
);
3907 mutex_destroy(&arc_reclaim_thr_lock
);
3908 cv_destroy(&arc_reclaim_thr_cv
);
3910 list_destroy(&arc_mru
->arcs_list
[ARC_BUFC_METADATA
]);
3911 list_destroy(&arc_mru_ghost
->arcs_list
[ARC_BUFC_METADATA
]);
3912 list_destroy(&arc_mfu
->arcs_list
[ARC_BUFC_METADATA
]);
3913 list_destroy(&arc_mfu_ghost
->arcs_list
[ARC_BUFC_METADATA
]);
3914 list_destroy(&arc_mru
->arcs_list
[ARC_BUFC_DATA
]);
3915 list_destroy(&arc_mru_ghost
->arcs_list
[ARC_BUFC_DATA
]);
3916 list_destroy(&arc_mfu
->arcs_list
[ARC_BUFC_DATA
]);
3917 list_destroy(&arc_mfu_ghost
->arcs_list
[ARC_BUFC_DATA
]);
3919 mutex_destroy(&arc_anon
->arcs_mtx
);
3920 mutex_destroy(&arc_mru
->arcs_mtx
);
3921 mutex_destroy(&arc_mru_ghost
->arcs_mtx
);
3922 mutex_destroy(&arc_mfu
->arcs_mtx
);
3923 mutex_destroy(&arc_mfu_ghost
->arcs_mtx
);
3924 mutex_destroy(&arc_l2c_only
->arcs_mtx
);
3926 mutex_destroy(&zfs_write_limit_lock
);
3930 ASSERT(arc_loaned_bytes
== 0);
3936 * The level 2 ARC (L2ARC) is a cache layer in-between main memory and disk.
3937 * It uses dedicated storage devices to hold cached data, which are populated
3938 * using large infrequent writes. The main role of this cache is to boost
3939 * the performance of random read workloads. The intended L2ARC devices
3940 * include short-stroked disks, solid state disks, and other media with
3941 * substantially faster read latency than disk.
3943 * +-----------------------+
3945 * +-----------------------+
3948 * l2arc_feed_thread() arc_read()
3952 * +---------------+ |
3954 * +---------------+ |
3959 * +-------+ +-------+
3961 * | cache | | cache |
3962 * +-------+ +-------+
3963 * +=========+ .-----.
3964 * : L2ARC : |-_____-|
3965 * : devices : | Disks |
3966 * +=========+ `-_____-'
3968 * Read requests are satisfied from the following sources, in order:
3971 * 2) vdev cache of L2ARC devices
3973 * 4) vdev cache of disks
3976 * Some L2ARC device types exhibit extremely slow write performance.
3977 * To accommodate for this there are some significant differences between
3978 * the L2ARC and traditional cache design:
3980 * 1. There is no eviction path from the ARC to the L2ARC. Evictions from
3981 * the ARC behave as usual, freeing buffers and placing headers on ghost
3982 * lists. The ARC does not send buffers to the L2ARC during eviction as
3983 * this would add inflated write latencies for all ARC memory pressure.
3985 * 2. The L2ARC attempts to cache data from the ARC before it is evicted.
3986 * It does this by periodically scanning buffers from the eviction-end of
3987 * the MFU and MRU ARC lists, copying them to the L2ARC devices if they are
3988 * not already there. It scans until a headroom of buffers is satisfied,
3989 * which itself is a buffer for ARC eviction. The thread that does this is
3990 * l2arc_feed_thread(), illustrated below; example sizes are included to
3991 * provide a better sense of ratio than this diagram:
3994 * +---------------------+----------+
3995 * ARC_mfu |:::::#:::::::::::::::|o#o###o###|-->. # already on L2ARC
3996 * +---------------------+----------+ | o L2ARC eligible
3997 * ARC_mru |:#:::::::::::::::::::|#o#ooo####|-->| : ARC buffer
3998 * +---------------------+----------+ |
3999 * 15.9 Gbytes ^ 32 Mbytes |
4001 * l2arc_feed_thread()
4003 * l2arc write hand <--[oooo]--'
4007 * +==============================+
4008 * L2ARC dev |####|#|###|###| |####| ... |
4009 * +==============================+
4012 * 3. If an ARC buffer is copied to the L2ARC but then hit instead of
4013 * evicted, then the L2ARC has cached a buffer much sooner than it probably
4014 * needed to, potentially wasting L2ARC device bandwidth and storage. It is
4015 * safe to say that this is an uncommon case, since buffers at the end of
4016 * the ARC lists have moved there due to inactivity.
4018 * 4. If the ARC evicts faster than the L2ARC can maintain a headroom,
4019 * then the L2ARC simply misses copying some buffers. This serves as a
4020 * pressure valve to prevent heavy read workloads from both stalling the ARC
4021 * with waits and clogging the L2ARC with writes. This also helps prevent
4022 * the potential for the L2ARC to churn if it attempts to cache content too
4023 * quickly, such as during backups of the entire pool.
4025 * 5. After system boot and before the ARC has filled main memory, there are
4026 * no evictions from the ARC and so the tails of the ARC_mfu and ARC_mru
4027 * lists can remain mostly static. Instead of searching from tail of these
4028 * lists as pictured, the l2arc_feed_thread() will search from the list heads
4029 * for eligible buffers, greatly increasing its chance of finding them.
4031 * The L2ARC device write speed is also boosted during this time so that
4032 * the L2ARC warms up faster. Since there have been no ARC evictions yet,
4033 * there are no L2ARC reads, and no fear of degrading read performance
4034 * through increased writes.
4036 * 6. Writes to the L2ARC devices are grouped and sent in-sequence, so that
4037 * the vdev queue can aggregate them into larger and fewer writes. Each
4038 * device is written to in a rotor fashion, sweeping writes through
4039 * available space then repeating.
4041 * 7. The L2ARC does not store dirty content. It never needs to flush
4042 * write buffers back to disk based storage.
4044 * 8. If an ARC buffer is written (and dirtied) which also exists in the
4045 * L2ARC, the now stale L2ARC buffer is immediately dropped.
4047 * The performance of the L2ARC can be tweaked by a number of tunables, which
4048 * may be necessary for different workloads:
4050 * l2arc_write_max max write bytes per interval
4051 * l2arc_write_boost extra write bytes during device warmup
4052 * l2arc_noprefetch skip caching prefetched buffers
4053 * l2arc_headroom number of max device writes to precache
4054 * l2arc_feed_secs seconds between L2ARC writing
4056 * Tunables may be removed or added as future performance improvements are
4057 * integrated, and also may become zpool properties.
4059 * There are three key functions that control how the L2ARC warms up:
4061 * l2arc_write_eligible() check if a buffer is eligible to cache
4062 * l2arc_write_size() calculate how much to write
4063 * l2arc_write_interval() calculate sleep delay between writes
4065 * These three functions determine what to write, how much, and how quickly
4070 l2arc_write_eligible(uint64_t spa_guid
, arc_buf_hdr_t
*ab
)
4073 * A buffer is *not* eligible for the L2ARC if it:
4074 * 1. belongs to a different spa.
4075 * 2. is already cached on the L2ARC.
4076 * 3. has an I/O in progress (it may be an incomplete read).
4077 * 4. is flagged not eligible (zfs property).
4079 if (ab
->b_spa
!= spa_guid
|| ab
->b_l2hdr
!= NULL
||
4080 HDR_IO_IN_PROGRESS(ab
) || !HDR_L2CACHE(ab
))
4087 l2arc_write_size(l2arc_dev_t
*dev
)
4091 size
= dev
->l2ad_write
;
4093 if (arc_warm
== B_FALSE
)
4094 size
+= dev
->l2ad_boost
;
4101 l2arc_write_interval(clock_t began
, uint64_t wanted
, uint64_t wrote
)
4103 clock_t interval
, next
, now
;
4106 * If the ARC lists are busy, increase our write rate; if the
4107 * lists are stale, idle back. This is achieved by checking
4108 * how much we previously wrote - if it was more than half of
4109 * what we wanted, schedule the next write much sooner.
4111 if (l2arc_feed_again
&& wrote
> (wanted
/ 2))
4112 interval
= (hz
* l2arc_feed_min_ms
) / 1000;
4114 interval
= hz
* l2arc_feed_secs
;
4116 now
= ddi_get_lbolt();
4117 next
= MAX(now
, MIN(now
+ interval
, began
+ interval
));
4123 l2arc_hdr_stat_add(void)
4125 ARCSTAT_INCR(arcstat_l2_hdr_size
, HDR_SIZE
);
4126 ARCSTAT_INCR(arcstat_hdr_size
, -HDR_SIZE
);
4130 l2arc_hdr_stat_remove(void)
4132 ARCSTAT_INCR(arcstat_l2_hdr_size
, -HDR_SIZE
);
4133 ARCSTAT_INCR(arcstat_hdr_size
, HDR_SIZE
);
4137 * Cycle through L2ARC devices. This is how L2ARC load balances.
4138 * If a device is returned, this also returns holding the spa config lock.
4140 static l2arc_dev_t
*
4141 l2arc_dev_get_next(void)
4143 l2arc_dev_t
*first
, *next
= NULL
;
4146 * Lock out the removal of spas (spa_namespace_lock), then removal
4147 * of cache devices (l2arc_dev_mtx). Once a device has been selected,
4148 * both locks will be dropped and a spa config lock held instead.
4150 mutex_enter(&spa_namespace_lock
);
4151 mutex_enter(&l2arc_dev_mtx
);
4153 /* if there are no vdevs, there is nothing to do */
4154 if (l2arc_ndev
== 0)
4158 next
= l2arc_dev_last
;
4160 /* loop around the list looking for a non-faulted vdev */
4162 next
= list_head(l2arc_dev_list
);
4164 next
= list_next(l2arc_dev_list
, next
);
4166 next
= list_head(l2arc_dev_list
);
4169 /* if we have come back to the start, bail out */
4172 else if (next
== first
)
4175 } while (vdev_is_dead(next
->l2ad_vdev
));
4177 /* if we were unable to find any usable vdevs, return NULL */
4178 if (vdev_is_dead(next
->l2ad_vdev
))
4181 l2arc_dev_last
= next
;
4184 mutex_exit(&l2arc_dev_mtx
);
4187 * Grab the config lock to prevent the 'next' device from being
4188 * removed while we are writing to it.
4191 spa_config_enter(next
->l2ad_spa
, SCL_L2ARC
, next
, RW_READER
);
4192 mutex_exit(&spa_namespace_lock
);
4198 * Free buffers that were tagged for destruction.
4201 l2arc_do_free_on_write(void)
4204 l2arc_data_free_t
*df
, *df_prev
;
4206 mutex_enter(&l2arc_free_on_write_mtx
);
4207 buflist
= l2arc_free_on_write
;
4209 for (df
= list_tail(buflist
); df
; df
= df_prev
) {
4210 df_prev
= list_prev(buflist
, df
);
4211 ASSERT(df
->l2df_data
!= NULL
);
4212 ASSERT(df
->l2df_func
!= NULL
);
4213 df
->l2df_func(df
->l2df_data
, df
->l2df_size
);
4214 list_remove(buflist
, df
);
4215 kmem_free(df
, sizeof (l2arc_data_free_t
));
4218 mutex_exit(&l2arc_free_on_write_mtx
);
4222 * A write to a cache device has completed. Update all headers to allow
4223 * reads from these buffers to begin.
4226 l2arc_write_done(zio_t
*zio
)
4228 l2arc_write_callback_t
*cb
;
4231 arc_buf_hdr_t
*head
, *ab
, *ab_prev
;
4232 l2arc_buf_hdr_t
*abl2
;
4233 kmutex_t
*hash_lock
;
4235 cb
= zio
->io_private
;
4237 dev
= cb
->l2wcb_dev
;
4238 ASSERT(dev
!= NULL
);
4239 head
= cb
->l2wcb_head
;
4240 ASSERT(head
!= NULL
);
4241 buflist
= dev
->l2ad_buflist
;
4242 ASSERT(buflist
!= NULL
);
4243 DTRACE_PROBE2(l2arc__iodone
, zio_t
*, zio
,
4244 l2arc_write_callback_t
*, cb
);
4246 if (zio
->io_error
!= 0)
4247 ARCSTAT_BUMP(arcstat_l2_writes_error
);
4249 mutex_enter(&l2arc_buflist_mtx
);
4252 * All writes completed, or an error was hit.
4254 for (ab
= list_prev(buflist
, head
); ab
; ab
= ab_prev
) {
4255 ab_prev
= list_prev(buflist
, ab
);
4257 hash_lock
= HDR_LOCK(ab
);
4258 if (!mutex_tryenter(hash_lock
)) {
4260 * This buffer misses out. It may be in a stage
4261 * of eviction. Its ARC_L2_WRITING flag will be
4262 * left set, denying reads to this buffer.
4264 ARCSTAT_BUMP(arcstat_l2_writes_hdr_miss
);
4268 if (zio
->io_error
!= 0) {
4270 * Error - drop L2ARC entry.
4272 list_remove(buflist
, ab
);
4275 kmem_free(abl2
, sizeof (l2arc_buf_hdr_t
));
4276 arc_space_return(L2HDR_SIZE
, ARC_SPACE_L2HDRS
);
4277 ARCSTAT_INCR(arcstat_l2_size
, -ab
->b_size
);
4281 * Allow ARC to begin reads to this L2ARC entry.
4283 ab
->b_flags
&= ~ARC_L2_WRITING
;
4285 mutex_exit(hash_lock
);
4288 atomic_inc_64(&l2arc_writes_done
);
4289 list_remove(buflist
, head
);
4290 kmem_cache_free(hdr_cache
, head
);
4291 mutex_exit(&l2arc_buflist_mtx
);
4293 l2arc_do_free_on_write();
4295 kmem_free(cb
, sizeof (l2arc_write_callback_t
));
4299 * A read to a cache device completed. Validate buffer contents before
4300 * handing over to the regular ARC routines.
4303 l2arc_read_done(zio_t
*zio
)
4305 l2arc_read_callback_t
*cb
;
4308 kmutex_t
*hash_lock
;
4311 ASSERT(zio
->io_vd
!= NULL
);
4312 ASSERT(zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
);
4314 spa_config_exit(zio
->io_spa
, SCL_L2ARC
, zio
->io_vd
);
4316 cb
= zio
->io_private
;
4318 buf
= cb
->l2rcb_buf
;
4319 ASSERT(buf
!= NULL
);
4321 hash_lock
= HDR_LOCK(buf
->b_hdr
);
4322 mutex_enter(hash_lock
);
4324 ASSERT3P(hash_lock
, ==, HDR_LOCK(hdr
));
4327 * Check this survived the L2ARC journey.
4329 equal
= arc_cksum_equal(buf
);
4330 if (equal
&& zio
->io_error
== 0 && !HDR_L2_EVICTED(hdr
)) {
4331 mutex_exit(hash_lock
);
4332 zio
->io_private
= buf
;
4333 zio
->io_bp_copy
= cb
->l2rcb_bp
; /* XXX fix in L2ARC 2.0 */
4334 zio
->io_bp
= &zio
->io_bp_copy
; /* XXX fix in L2ARC 2.0 */
4337 mutex_exit(hash_lock
);
4339 * Buffer didn't survive caching. Increment stats and
4340 * reissue to the original storage device.
4342 if (zio
->io_error
!= 0) {
4343 ARCSTAT_BUMP(arcstat_l2_io_error
);
4345 zio
->io_error
= EIO
;
4348 ARCSTAT_BUMP(arcstat_l2_cksum_bad
);
4351 * If there's no waiter, issue an async i/o to the primary
4352 * storage now. If there *is* a waiter, the caller must
4353 * issue the i/o in a context where it's OK to block.
4355 if (zio
->io_waiter
== NULL
) {
4356 zio_t
*pio
= zio_unique_parent(zio
);
4358 ASSERT(!pio
|| pio
->io_child_type
== ZIO_CHILD_LOGICAL
);
4360 zio_nowait(zio_read(pio
, cb
->l2rcb_spa
, &cb
->l2rcb_bp
,
4361 buf
->b_data
, zio
->io_size
, arc_read_done
, buf
,
4362 zio
->io_priority
, cb
->l2rcb_flags
, &cb
->l2rcb_zb
));
4366 kmem_free(cb
, sizeof (l2arc_read_callback_t
));
4370 * This is the list priority from which the L2ARC will search for pages to
4371 * cache. This is used within loops (0..3) to cycle through lists in the
4372 * desired order. This order can have a significant effect on cache
4375 * Currently the metadata lists are hit first, MFU then MRU, followed by
4376 * the data lists. This function returns a locked list, and also returns
4380 l2arc_list_locked(int list_num
, kmutex_t
**lock
)
4382 list_t
*list
= NULL
;
4384 ASSERT(list_num
>= 0 && list_num
<= 3);
4388 list
= &arc_mfu
->arcs_list
[ARC_BUFC_METADATA
];
4389 *lock
= &arc_mfu
->arcs_mtx
;
4392 list
= &arc_mru
->arcs_list
[ARC_BUFC_METADATA
];
4393 *lock
= &arc_mru
->arcs_mtx
;
4396 list
= &arc_mfu
->arcs_list
[ARC_BUFC_DATA
];
4397 *lock
= &arc_mfu
->arcs_mtx
;
4400 list
= &arc_mru
->arcs_list
[ARC_BUFC_DATA
];
4401 *lock
= &arc_mru
->arcs_mtx
;
4405 ASSERT(!(MUTEX_HELD(*lock
)));
4411 * Evict buffers from the device write hand to the distance specified in
4412 * bytes. This distance may span populated buffers, it may span nothing.
4413 * This is clearing a region on the L2ARC device ready for writing.
4414 * If the 'all' boolean is set, every buffer is evicted.
4417 l2arc_evict(l2arc_dev_t
*dev
, uint64_t distance
, boolean_t all
)
4420 l2arc_buf_hdr_t
*abl2
;
4421 arc_buf_hdr_t
*ab
, *ab_prev
;
4422 kmutex_t
*hash_lock
;
4425 buflist
= dev
->l2ad_buflist
;
4427 if (buflist
== NULL
)
4430 if (!all
&& dev
->l2ad_first
) {
4432 * This is the first sweep through the device. There is
4438 if (dev
->l2ad_hand
>= (dev
->l2ad_end
- (2 * distance
))) {
4440 * When nearing the end of the device, evict to the end
4441 * before the device write hand jumps to the start.
4443 taddr
= dev
->l2ad_end
;
4445 taddr
= dev
->l2ad_hand
+ distance
;
4447 DTRACE_PROBE4(l2arc__evict
, l2arc_dev_t
*, dev
, list_t
*, buflist
,
4448 uint64_t, taddr
, boolean_t
, all
);
4451 mutex_enter(&l2arc_buflist_mtx
);
4452 for (ab
= list_tail(buflist
); ab
; ab
= ab_prev
) {
4453 ab_prev
= list_prev(buflist
, ab
);
4455 hash_lock
= HDR_LOCK(ab
);
4456 if (!mutex_tryenter(hash_lock
)) {
4458 * Missed the hash lock. Retry.
4460 ARCSTAT_BUMP(arcstat_l2_evict_lock_retry
);
4461 mutex_exit(&l2arc_buflist_mtx
);
4462 mutex_enter(hash_lock
);
4463 mutex_exit(hash_lock
);
4467 if (HDR_L2_WRITE_HEAD(ab
)) {
4469 * We hit a write head node. Leave it for
4470 * l2arc_write_done().
4472 list_remove(buflist
, ab
);
4473 mutex_exit(hash_lock
);
4477 if (!all
&& ab
->b_l2hdr
!= NULL
&&
4478 (ab
->b_l2hdr
->b_daddr
> taddr
||
4479 ab
->b_l2hdr
->b_daddr
< dev
->l2ad_hand
)) {
4481 * We've evicted to the target address,
4482 * or the end of the device.
4484 mutex_exit(hash_lock
);
4488 if (HDR_FREE_IN_PROGRESS(ab
)) {
4490 * Already on the path to destruction.
4492 mutex_exit(hash_lock
);
4496 if (ab
->b_state
== arc_l2c_only
) {
4497 ASSERT(!HDR_L2_READING(ab
));
4499 * This doesn't exist in the ARC. Destroy.
4500 * arc_hdr_destroy() will call list_remove()
4501 * and decrement arcstat_l2_size.
4503 arc_change_state(arc_anon
, ab
, hash_lock
);
4504 arc_hdr_destroy(ab
);
4507 * Invalidate issued or about to be issued
4508 * reads, since we may be about to write
4509 * over this location.
4511 if (HDR_L2_READING(ab
)) {
4512 ARCSTAT_BUMP(arcstat_l2_evict_reading
);
4513 ab
->b_flags
|= ARC_L2_EVICTED
;
4517 * Tell ARC this no longer exists in L2ARC.
4519 if (ab
->b_l2hdr
!= NULL
) {
4522 kmem_free(abl2
, sizeof (l2arc_buf_hdr_t
));
4523 arc_space_return(L2HDR_SIZE
, ARC_SPACE_L2HDRS
);
4524 ARCSTAT_INCR(arcstat_l2_size
, -ab
->b_size
);
4526 list_remove(buflist
, ab
);
4529 * This may have been leftover after a
4532 ab
->b_flags
&= ~ARC_L2_WRITING
;
4534 mutex_exit(hash_lock
);
4536 mutex_exit(&l2arc_buflist_mtx
);
4538 vdev_space_update(dev
->l2ad_vdev
, -(taddr
- dev
->l2ad_evict
), 0, 0);
4539 dev
->l2ad_evict
= taddr
;
4543 * Find and write ARC buffers to the L2ARC device.
4545 * An ARC_L2_WRITING flag is set so that the L2ARC buffers are not valid
4546 * for reading until they have completed writing.
4549 l2arc_write_buffers(spa_t
*spa
, l2arc_dev_t
*dev
, uint64_t target_sz
)
4551 arc_buf_hdr_t
*ab
, *ab_prev
, *head
;
4552 l2arc_buf_hdr_t
*hdrl2
;
4554 uint64_t passed_sz
, write_sz
, buf_sz
, headroom
;
4556 kmutex_t
*hash_lock
, *list_lock
= NULL
;
4557 boolean_t have_lock
, full
;
4558 l2arc_write_callback_t
*cb
;
4560 uint64_t guid
= spa_load_guid(spa
);
4563 ASSERT(dev
->l2ad_vdev
!= NULL
);
4568 head
= kmem_cache_alloc(hdr_cache
, KM_PUSHPAGE
);
4569 head
->b_flags
|= ARC_L2_WRITE_HEAD
;
4572 * Copy buffers for L2ARC writing.
4574 mutex_enter(&l2arc_buflist_mtx
);
4575 for (try = 0; try <= 3; try++) {
4576 list
= l2arc_list_locked(try, &list_lock
);
4580 * L2ARC fast warmup.
4582 * Until the ARC is warm and starts to evict, read from the
4583 * head of the ARC lists rather than the tail.
4585 headroom
= target_sz
* l2arc_headroom
;
4586 if (arc_warm
== B_FALSE
)
4587 ab
= list_head(list
);
4589 ab
= list_tail(list
);
4591 for (; ab
; ab
= ab_prev
) {
4592 if (arc_warm
== B_FALSE
)
4593 ab_prev
= list_next(list
, ab
);
4595 ab_prev
= list_prev(list
, ab
);
4597 hash_lock
= HDR_LOCK(ab
);
4598 have_lock
= MUTEX_HELD(hash_lock
);
4599 if (!have_lock
&& !mutex_tryenter(hash_lock
)) {
4601 * Skip this buffer rather than waiting.
4606 passed_sz
+= ab
->b_size
;
4607 if (passed_sz
> headroom
) {
4611 mutex_exit(hash_lock
);
4615 if (!l2arc_write_eligible(guid
, ab
)) {
4616 mutex_exit(hash_lock
);
4620 if ((write_sz
+ ab
->b_size
) > target_sz
) {
4622 mutex_exit(hash_lock
);
4628 * Insert a dummy header on the buflist so
4629 * l2arc_write_done() can find where the
4630 * write buffers begin without searching.
4632 list_insert_head(dev
->l2ad_buflist
, head
);
4634 cb
= kmem_alloc(sizeof (l2arc_write_callback_t
),
4636 cb
->l2wcb_dev
= dev
;
4637 cb
->l2wcb_head
= head
;
4638 pio
= zio_root(spa
, l2arc_write_done
, cb
,
4643 * Create and add a new L2ARC header.
4645 hdrl2
= kmem_zalloc(sizeof (l2arc_buf_hdr_t
),
4648 hdrl2
->b_daddr
= dev
->l2ad_hand
;
4649 arc_space_consume(L2HDR_SIZE
, ARC_SPACE_L2HDRS
);
4651 ab
->b_flags
|= ARC_L2_WRITING
;
4652 ab
->b_l2hdr
= hdrl2
;
4653 list_insert_head(dev
->l2ad_buflist
, ab
);
4654 buf_data
= ab
->b_buf
->b_data
;
4655 buf_sz
= ab
->b_size
;
4658 * Compute and store the buffer cksum before
4659 * writing. On debug the cksum is verified first.
4661 arc_cksum_verify(ab
->b_buf
);
4662 arc_cksum_compute(ab
->b_buf
, B_TRUE
);
4664 mutex_exit(hash_lock
);
4666 wzio
= zio_write_phys(pio
, dev
->l2ad_vdev
,
4667 dev
->l2ad_hand
, buf_sz
, buf_data
, ZIO_CHECKSUM_OFF
,
4668 NULL
, NULL
, ZIO_PRIORITY_ASYNC_WRITE
,
4669 ZIO_FLAG_CANFAIL
, B_FALSE
);
4671 DTRACE_PROBE2(l2arc__write
, vdev_t
*, dev
->l2ad_vdev
,
4673 (void) zio_nowait(wzio
);
4676 * Keep the clock hand suitably device-aligned.
4678 buf_sz
= vdev_psize_to_asize(dev
->l2ad_vdev
, buf_sz
);
4681 dev
->l2ad_hand
+= buf_sz
;
4684 mutex_exit(list_lock
);
4689 mutex_exit(&l2arc_buflist_mtx
);
4693 kmem_cache_free(hdr_cache
, head
);
4697 ASSERT3U(write_sz
, <=, target_sz
);
4698 ARCSTAT_BUMP(arcstat_l2_writes_sent
);
4699 ARCSTAT_INCR(arcstat_l2_write_bytes
, write_sz
);
4700 ARCSTAT_INCR(arcstat_l2_size
, write_sz
);
4701 vdev_space_update(dev
->l2ad_vdev
, write_sz
, 0, 0);
4704 * Bump device hand to the device start if it is approaching the end.
4705 * l2arc_evict() will already have evicted ahead for this case.
4707 if (dev
->l2ad_hand
>= (dev
->l2ad_end
- target_sz
)) {
4708 vdev_space_update(dev
->l2ad_vdev
,
4709 dev
->l2ad_end
- dev
->l2ad_hand
, 0, 0);
4710 dev
->l2ad_hand
= dev
->l2ad_start
;
4711 dev
->l2ad_evict
= dev
->l2ad_start
;
4712 dev
->l2ad_first
= B_FALSE
;
4715 dev
->l2ad_writing
= B_TRUE
;
4716 (void) zio_wait(pio
);
4717 dev
->l2ad_writing
= B_FALSE
;
4723 * This thread feeds the L2ARC at regular intervals. This is the beating
4724 * heart of the L2ARC.
4727 l2arc_feed_thread(void)
4732 uint64_t size
, wrote
;
4733 clock_t begin
, next
= ddi_get_lbolt();
4735 CALLB_CPR_INIT(&cpr
, &l2arc_feed_thr_lock
, callb_generic_cpr
, FTAG
);
4737 mutex_enter(&l2arc_feed_thr_lock
);
4739 while (l2arc_thread_exit
== 0) {
4740 CALLB_CPR_SAFE_BEGIN(&cpr
);
4741 (void) cv_timedwait_interruptible(&l2arc_feed_thr_cv
,
4742 &l2arc_feed_thr_lock
, next
);
4743 CALLB_CPR_SAFE_END(&cpr
, &l2arc_feed_thr_lock
);
4744 next
= ddi_get_lbolt() + hz
;
4747 * Quick check for L2ARC devices.
4749 mutex_enter(&l2arc_dev_mtx
);
4750 if (l2arc_ndev
== 0) {
4751 mutex_exit(&l2arc_dev_mtx
);
4754 mutex_exit(&l2arc_dev_mtx
);
4755 begin
= ddi_get_lbolt();
4758 * This selects the next l2arc device to write to, and in
4759 * doing so the next spa to feed from: dev->l2ad_spa. This
4760 * will return NULL if there are now no l2arc devices or if
4761 * they are all faulted.
4763 * If a device is returned, its spa's config lock is also
4764 * held to prevent device removal. l2arc_dev_get_next()
4765 * will grab and release l2arc_dev_mtx.
4767 if ((dev
= l2arc_dev_get_next()) == NULL
)
4770 spa
= dev
->l2ad_spa
;
4771 ASSERT(spa
!= NULL
);
4774 * If the pool is read-only then force the feed thread to
4775 * sleep a little longer.
4777 if (!spa_writeable(spa
)) {
4778 next
= ddi_get_lbolt() + 5 * l2arc_feed_secs
* hz
;
4779 spa_config_exit(spa
, SCL_L2ARC
, dev
);
4784 * Avoid contributing to memory pressure.
4787 ARCSTAT_BUMP(arcstat_l2_abort_lowmem
);
4788 spa_config_exit(spa
, SCL_L2ARC
, dev
);
4792 ARCSTAT_BUMP(arcstat_l2_feeds
);
4794 size
= l2arc_write_size(dev
);
4797 * Evict L2ARC buffers that will be overwritten.
4799 l2arc_evict(dev
, size
, B_FALSE
);
4802 * Write ARC buffers.
4804 wrote
= l2arc_write_buffers(spa
, dev
, size
);
4807 * Calculate interval between writes.
4809 next
= l2arc_write_interval(begin
, size
, wrote
);
4810 spa_config_exit(spa
, SCL_L2ARC
, dev
);
4813 l2arc_thread_exit
= 0;
4814 cv_broadcast(&l2arc_feed_thr_cv
);
4815 CALLB_CPR_EXIT(&cpr
); /* drops l2arc_feed_thr_lock */
4820 l2arc_vdev_present(vdev_t
*vd
)
4824 mutex_enter(&l2arc_dev_mtx
);
4825 for (dev
= list_head(l2arc_dev_list
); dev
!= NULL
;
4826 dev
= list_next(l2arc_dev_list
, dev
)) {
4827 if (dev
->l2ad_vdev
== vd
)
4830 mutex_exit(&l2arc_dev_mtx
);
4832 return (dev
!= NULL
);
4836 * Add a vdev for use by the L2ARC. By this point the spa has already
4837 * validated the vdev and opened it.
4840 l2arc_add_vdev(spa_t
*spa
, vdev_t
*vd
)
4842 l2arc_dev_t
*adddev
;
4844 ASSERT(!l2arc_vdev_present(vd
));
4847 * Create a new l2arc device entry.
4849 adddev
= kmem_zalloc(sizeof (l2arc_dev_t
), KM_SLEEP
);
4850 adddev
->l2ad_spa
= spa
;
4851 adddev
->l2ad_vdev
= vd
;
4852 adddev
->l2ad_write
= l2arc_write_max
;
4853 adddev
->l2ad_boost
= l2arc_write_boost
;
4854 adddev
->l2ad_start
= VDEV_LABEL_START_SIZE
;
4855 adddev
->l2ad_end
= VDEV_LABEL_START_SIZE
+ vdev_get_min_asize(vd
);
4856 adddev
->l2ad_hand
= adddev
->l2ad_start
;
4857 adddev
->l2ad_evict
= adddev
->l2ad_start
;
4858 adddev
->l2ad_first
= B_TRUE
;
4859 adddev
->l2ad_writing
= B_FALSE
;
4860 list_link_init(&adddev
->l2ad_node
);
4861 ASSERT3U(adddev
->l2ad_write
, >, 0);
4864 * This is a list of all ARC buffers that are still valid on the
4867 adddev
->l2ad_buflist
= kmem_zalloc(sizeof (list_t
), KM_SLEEP
);
4868 list_create(adddev
->l2ad_buflist
, sizeof (arc_buf_hdr_t
),
4869 offsetof(arc_buf_hdr_t
, b_l2node
));
4871 vdev_space_update(vd
, 0, 0, adddev
->l2ad_end
- adddev
->l2ad_hand
);
4874 * Add device to global list
4876 mutex_enter(&l2arc_dev_mtx
);
4877 list_insert_head(l2arc_dev_list
, adddev
);
4878 atomic_inc_64(&l2arc_ndev
);
4879 mutex_exit(&l2arc_dev_mtx
);
4883 * Remove a vdev from the L2ARC.
4886 l2arc_remove_vdev(vdev_t
*vd
)
4888 l2arc_dev_t
*dev
, *nextdev
, *remdev
= NULL
;
4891 * Find the device by vdev
4893 mutex_enter(&l2arc_dev_mtx
);
4894 for (dev
= list_head(l2arc_dev_list
); dev
; dev
= nextdev
) {
4895 nextdev
= list_next(l2arc_dev_list
, dev
);
4896 if (vd
== dev
->l2ad_vdev
) {
4901 ASSERT(remdev
!= NULL
);
4904 * Remove device from global list
4906 list_remove(l2arc_dev_list
, remdev
);
4907 l2arc_dev_last
= NULL
; /* may have been invalidated */
4908 atomic_dec_64(&l2arc_ndev
);
4909 mutex_exit(&l2arc_dev_mtx
);
4912 * Clear all buflists and ARC references. L2ARC device flush.
4914 l2arc_evict(remdev
, 0, B_TRUE
);
4915 list_destroy(remdev
->l2ad_buflist
);
4916 kmem_free(remdev
->l2ad_buflist
, sizeof (list_t
));
4917 kmem_free(remdev
, sizeof (l2arc_dev_t
));
4923 l2arc_thread_exit
= 0;
4925 l2arc_writes_sent
= 0;
4926 l2arc_writes_done
= 0;
4928 mutex_init(&l2arc_feed_thr_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
4929 cv_init(&l2arc_feed_thr_cv
, NULL
, CV_DEFAULT
, NULL
);
4930 mutex_init(&l2arc_dev_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
4931 mutex_init(&l2arc_buflist_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
4932 mutex_init(&l2arc_free_on_write_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
4934 l2arc_dev_list
= &L2ARC_dev_list
;
4935 l2arc_free_on_write
= &L2ARC_free_on_write
;
4936 list_create(l2arc_dev_list
, sizeof (l2arc_dev_t
),
4937 offsetof(l2arc_dev_t
, l2ad_node
));
4938 list_create(l2arc_free_on_write
, sizeof (l2arc_data_free_t
),
4939 offsetof(l2arc_data_free_t
, l2df_list_node
));
4946 * This is called from dmu_fini(), which is called from spa_fini();
4947 * Because of this, we can assume that all l2arc devices have
4948 * already been removed when the pools themselves were removed.
4951 l2arc_do_free_on_write();
4953 mutex_destroy(&l2arc_feed_thr_lock
);
4954 cv_destroy(&l2arc_feed_thr_cv
);
4955 mutex_destroy(&l2arc_dev_mtx
);
4956 mutex_destroy(&l2arc_buflist_mtx
);
4957 mutex_destroy(&l2arc_free_on_write_mtx
);
4959 list_destroy(l2arc_dev_list
);
4960 list_destroy(l2arc_free_on_write
);
4966 if (!(spa_mode_global
& FWRITE
))
4969 (void) thread_create(NULL
, 0, l2arc_feed_thread
, NULL
, 0, &p0
,
4970 TS_RUN
, minclsyspri
);
4976 if (!(spa_mode_global
& FWRITE
))
4979 mutex_enter(&l2arc_feed_thr_lock
);
4980 cv_signal(&l2arc_feed_thr_cv
); /* kick thread out of startup */
4981 l2arc_thread_exit
= 1;
4982 while (l2arc_thread_exit
!= 0)
4983 cv_wait(&l2arc_feed_thr_cv
, &l2arc_feed_thr_lock
);
4984 mutex_exit(&l2arc_feed_thr_lock
);
4987 #if defined(_KERNEL) && defined(HAVE_SPL)
4988 EXPORT_SYMBOL(arc_read
);
4989 EXPORT_SYMBOL(arc_buf_remove_ref
);
4990 EXPORT_SYMBOL(arc_getbuf_func
);
4991 EXPORT_SYMBOL(arc_add_prune_callback
);
4992 EXPORT_SYMBOL(arc_remove_prune_callback
);
4994 module_param(zfs_arc_min
, ulong
, 0644);
4995 MODULE_PARM_DESC(zfs_arc_min
, "Min arc size");
4997 module_param(zfs_arc_max
, ulong
, 0644);
4998 MODULE_PARM_DESC(zfs_arc_max
, "Max arc size");
5000 module_param(zfs_arc_meta_limit
, ulong
, 0644);
5001 MODULE_PARM_DESC(zfs_arc_meta_limit
, "Meta limit for arc size");
5003 module_param(zfs_arc_meta_prune
, int, 0644);
5004 MODULE_PARM_DESC(zfs_arc_meta_prune
, "Bytes of meta data to prune");
5006 module_param(zfs_arc_grow_retry
, int, 0644);
5007 MODULE_PARM_DESC(zfs_arc_grow_retry
, "Seconds before growing arc size");
5009 module_param(zfs_arc_shrink_shift
, int, 0644);
5010 MODULE_PARM_DESC(zfs_arc_shrink_shift
, "log2(fraction of arc to reclaim)");
5012 module_param(zfs_arc_p_min_shift
, int, 0644);
5013 MODULE_PARM_DESC(zfs_arc_p_min_shift
, "arc_c shift to calc min/max arc_p");
5015 module_param(zfs_disable_dup_eviction
, int, 0644);
5016 MODULE_PARM_DESC(zfs_disable_dup_eviction
, "disable duplicate buffer eviction");
5018 module_param(zfs_arc_memory_throttle_disable
, int, 0644);
5019 MODULE_PARM_DESC(zfs_arc_memory_throttle_disable
, "disable memory throttle");
5021 module_param(zfs_arc_min_prefetch_lifespan
, int, 0644);
5022 MODULE_PARM_DESC(zfs_arc_min_prefetch_lifespan
, "Min life of prefetch block");
5024 module_param(l2arc_write_max
, ulong
, 0644);
5025 MODULE_PARM_DESC(l2arc_write_max
, "Max write bytes per interval");
5027 module_param(l2arc_write_boost
, ulong
, 0644);
5028 MODULE_PARM_DESC(l2arc_write_boost
, "Extra write bytes during device warmup");
5030 module_param(l2arc_headroom
, ulong
, 0644);
5031 MODULE_PARM_DESC(l2arc_headroom
, "Number of max device writes to precache");
5033 module_param(l2arc_feed_secs
, ulong
, 0644);
5034 MODULE_PARM_DESC(l2arc_feed_secs
, "Seconds between L2ARC writing");
5036 module_param(l2arc_feed_min_ms
, ulong
, 0644);
5037 MODULE_PARM_DESC(l2arc_feed_min_ms
, "Min feed interval in milliseconds");
5039 module_param(l2arc_noprefetch
, int, 0644);
5040 MODULE_PARM_DESC(l2arc_noprefetch
, "Skip caching prefetched buffers");
5042 module_param(l2arc_feed_again
, int, 0644);
5043 MODULE_PARM_DESC(l2arc_feed_again
, "Turbo L2ARC warmup");
5045 module_param(l2arc_norw
, int, 0644);
5046 MODULE_PARM_DESC(l2arc_norw
, "No reads during writes");