]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
1 | /* |
2 | * CDDL HEADER START | |
3 | * | |
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. | |
7 | * | |
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. | |
12 | * | |
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] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | /* | |
428870ff | 22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
36da08ef | 23 | * Copyright (c) 2012, Joyent, Inc. All rights reserved. |
bc77ba73 | 24 | * Copyright (c) 2011, 2016 by Delphix. All rights reserved. |
36da08ef | 25 | * Copyright (c) 2014 by Saso Kiselkov. All rights reserved. |
d3c2ae1c | 26 | * Copyright 2015 Nexenta Systems, Inc. All rights reserved. |
34dc7c2f BB |
27 | */ |
28 | ||
34dc7c2f BB |
29 | /* |
30 | * DVA-based Adjustable Replacement Cache | |
31 | * | |
32 | * While much of the theory of operation used here is | |
33 | * based on the self-tuning, low overhead replacement cache | |
34 | * presented by Megiddo and Modha at FAST 2003, there are some | |
35 | * significant differences: | |
36 | * | |
37 | * 1. The Megiddo and Modha model assumes any page is evictable. | |
38 | * Pages in its cache cannot be "locked" into memory. This makes | |
39 | * the eviction algorithm simple: evict the last page in the list. | |
40 | * This also make the performance characteristics easy to reason | |
41 | * about. Our cache is not so simple. At any given moment, some | |
42 | * subset of the blocks in the cache are un-evictable because we | |
43 | * have handed out a reference to them. Blocks are only evictable | |
44 | * when there are no external references active. This makes | |
45 | * eviction far more problematic: we choose to evict the evictable | |
46 | * blocks that are the "lowest" in the list. | |
47 | * | |
48 | * There are times when it is not possible to evict the requested | |
49 | * space. In these circumstances we are unable to adjust the cache | |
50 | * size. To prevent the cache growing unbounded at these times we | |
51 | * implement a "cache throttle" that slows the flow of new data | |
52 | * into the cache until we can make space available. | |
53 | * | |
54 | * 2. The Megiddo and Modha model assumes a fixed cache size. | |
55 | * Pages are evicted when the cache is full and there is a cache | |
56 | * miss. Our model has a variable sized cache. It grows with | |
57 | * high use, but also tries to react to memory pressure from the | |
58 | * operating system: decreasing its size when system memory is | |
59 | * tight. | |
60 | * | |
61 | * 3. The Megiddo and Modha model assumes a fixed page size. All | |
d3cc8b15 | 62 | * elements of the cache are therefore exactly the same size. So |
34dc7c2f BB |
63 | * when adjusting the cache size following a cache miss, its simply |
64 | * a matter of choosing a single page to evict. In our model, we | |
65 | * have variable sized cache blocks (rangeing from 512 bytes to | |
d3cc8b15 | 66 | * 128K bytes). We therefore choose a set of blocks to evict to make |
34dc7c2f BB |
67 | * space for a cache miss that approximates as closely as possible |
68 | * the space used by the new block. | |
69 | * | |
70 | * See also: "ARC: A Self-Tuning, Low Overhead Replacement Cache" | |
71 | * by N. Megiddo & D. Modha, FAST 2003 | |
72 | */ | |
73 | ||
74 | /* | |
75 | * The locking model: | |
76 | * | |
77 | * A new reference to a cache buffer can be obtained in two | |
78 | * ways: 1) via a hash table lookup using the DVA as a key, | |
79 | * or 2) via one of the ARC lists. The arc_read() interface | |
2aa34383 | 80 | * uses method 1, while the internal ARC algorithms for |
d3cc8b15 | 81 | * adjusting the cache use method 2. We therefore provide two |
34dc7c2f | 82 | * types of locks: 1) the hash table lock array, and 2) the |
2aa34383 | 83 | * ARC list locks. |
34dc7c2f | 84 | * |
5c839890 BC |
85 | * Buffers do not have their own mutexes, rather they rely on the |
86 | * hash table mutexes for the bulk of their protection (i.e. most | |
87 | * fields in the arc_buf_hdr_t are protected by these mutexes). | |
34dc7c2f BB |
88 | * |
89 | * buf_hash_find() returns the appropriate mutex (held) when it | |
90 | * locates the requested buffer in the hash table. It returns | |
91 | * NULL for the mutex if the buffer was not in the table. | |
92 | * | |
93 | * buf_hash_remove() expects the appropriate hash mutex to be | |
94 | * already held before it is invoked. | |
95 | * | |
2aa34383 | 96 | * Each ARC state also has a mutex which is used to protect the |
34dc7c2f | 97 | * buffer list associated with the state. When attempting to |
2aa34383 | 98 | * obtain a hash table lock while holding an ARC list lock you |
34dc7c2f BB |
99 | * must use: mutex_tryenter() to avoid deadlock. Also note that |
100 | * the active state mutex must be held before the ghost state mutex. | |
101 | * | |
ab26409d BB |
102 | * It as also possible to register a callback which is run when the |
103 | * arc_meta_limit is reached and no buffers can be safely evicted. In | |
104 | * this case the arc user should drop a reference on some arc buffers so | |
105 | * they can be reclaimed and the arc_meta_limit honored. For example, | |
106 | * when using the ZPL each dentry holds a references on a znode. These | |
107 | * dentries must be pruned before the arc buffer holding the znode can | |
108 | * be safely evicted. | |
109 | * | |
34dc7c2f BB |
110 | * Note that the majority of the performance stats are manipulated |
111 | * with atomic operations. | |
112 | * | |
b9541d6b | 113 | * The L2ARC uses the l2ad_mtx on each vdev for the following: |
34dc7c2f BB |
114 | * |
115 | * - L2ARC buflist creation | |
116 | * - L2ARC buflist eviction | |
117 | * - L2ARC write completion, which walks L2ARC buflists | |
118 | * - ARC header destruction, as it removes from L2ARC buflists | |
119 | * - ARC header release, as it removes from L2ARC buflists | |
120 | */ | |
121 | ||
d3c2ae1c GW |
122 | /* |
123 | * ARC operation: | |
124 | * | |
125 | * Every block that is in the ARC is tracked by an arc_buf_hdr_t structure. | |
126 | * This structure can point either to a block that is still in the cache or to | |
127 | * one that is only accessible in an L2 ARC device, or it can provide | |
128 | * information about a block that was recently evicted. If a block is | |
129 | * only accessible in the L2ARC, then the arc_buf_hdr_t only has enough | |
130 | * information to retrieve it from the L2ARC device. This information is | |
131 | * stored in the l2arc_buf_hdr_t sub-structure of the arc_buf_hdr_t. A block | |
132 | * that is in this state cannot access the data directly. | |
133 | * | |
134 | * Blocks that are actively being referenced or have not been evicted | |
135 | * are cached in the L1ARC. The L1ARC (l1arc_buf_hdr_t) is a structure within | |
136 | * the arc_buf_hdr_t that will point to the data block in memory. A block can | |
137 | * only be read by a consumer if it has an l1arc_buf_hdr_t. The L1ARC | |
2aa34383 | 138 | * caches data in two ways -- in a list of ARC buffers (arc_buf_t) and |
a6255b7f | 139 | * also in the arc_buf_hdr_t's private physical data block pointer (b_pabd). |
2aa34383 DK |
140 | * |
141 | * The L1ARC's data pointer may or may not be uncompressed. The ARC has the | |
a6255b7f DQ |
142 | * ability to store the physical data (b_pabd) associated with the DVA of the |
143 | * arc_buf_hdr_t. Since the b_pabd is a copy of the on-disk physical block, | |
2aa34383 DK |
144 | * it will match its on-disk compression characteristics. This behavior can be |
145 | * disabled by setting 'zfs_compressed_arc_enabled' to B_FALSE. When the | |
a6255b7f | 146 | * compressed ARC functionality is disabled, the b_pabd will point to an |
2aa34383 DK |
147 | * uncompressed version of the on-disk data. |
148 | * | |
149 | * Data in the L1ARC is not accessed by consumers of the ARC directly. Each | |
150 | * arc_buf_hdr_t can have multiple ARC buffers (arc_buf_t) which reference it. | |
151 | * Each ARC buffer (arc_buf_t) is being actively accessed by a specific ARC | |
152 | * consumer. The ARC will provide references to this data and will keep it | |
153 | * cached until it is no longer in use. The ARC caches only the L1ARC's physical | |
154 | * data block and will evict any arc_buf_t that is no longer referenced. The | |
155 | * amount of memory consumed by the arc_buf_ts' data buffers can be seen via the | |
d3c2ae1c GW |
156 | * "overhead_size" kstat. |
157 | * | |
2aa34383 DK |
158 | * Depending on the consumer, an arc_buf_t can be requested in uncompressed or |
159 | * compressed form. The typical case is that consumers will want uncompressed | |
160 | * data, and when that happens a new data buffer is allocated where the data is | |
161 | * decompressed for them to use. Currently the only consumer who wants | |
162 | * compressed arc_buf_t's is "zfs send", when it streams data exactly as it | |
163 | * exists on disk. When this happens, the arc_buf_t's data buffer is shared | |
164 | * with the arc_buf_hdr_t. | |
d3c2ae1c | 165 | * |
2aa34383 DK |
166 | * Here is a diagram showing an arc_buf_hdr_t referenced by two arc_buf_t's. The |
167 | * first one is owned by a compressed send consumer (and therefore references | |
168 | * the same compressed data buffer as the arc_buf_hdr_t) and the second could be | |
169 | * used by any other consumer (and has its own uncompressed copy of the data | |
170 | * buffer). | |
d3c2ae1c | 171 | * |
2aa34383 DK |
172 | * arc_buf_hdr_t |
173 | * +-----------+ | |
174 | * | fields | | |
175 | * | common to | | |
176 | * | L1- and | | |
177 | * | L2ARC | | |
178 | * +-----------+ | |
179 | * | l2arc_buf_hdr_t | |
180 | * | | | |
181 | * +-----------+ | |
182 | * | l1arc_buf_hdr_t | |
183 | * | | arc_buf_t | |
184 | * | b_buf +------------>+-----------+ arc_buf_t | |
a6255b7f | 185 | * | b_pabd +-+ |b_next +---->+-----------+ |
2aa34383 DK |
186 | * +-----------+ | |-----------| |b_next +-->NULL |
187 | * | |b_comp = T | +-----------+ | |
188 | * | |b_data +-+ |b_comp = F | | |
189 | * | +-----------+ | |b_data +-+ | |
190 | * +->+------+ | +-----------+ | | |
191 | * compressed | | | | | |
192 | * data | |<--------------+ | uncompressed | |
193 | * +------+ compressed, | data | |
194 | * shared +-->+------+ | |
195 | * data | | | |
196 | * | | | |
197 | * +------+ | |
d3c2ae1c GW |
198 | * |
199 | * When a consumer reads a block, the ARC must first look to see if the | |
2aa34383 DK |
200 | * arc_buf_hdr_t is cached. If the hdr is cached then the ARC allocates a new |
201 | * arc_buf_t and either copies uncompressed data into a new data buffer from an | |
a6255b7f DQ |
202 | * existing uncompressed arc_buf_t, decompresses the hdr's b_pabd buffer into a |
203 | * new data buffer, or shares the hdr's b_pabd buffer, depending on whether the | |
2aa34383 DK |
204 | * hdr is compressed and the desired compression characteristics of the |
205 | * arc_buf_t consumer. If the arc_buf_t ends up sharing data with the | |
206 | * arc_buf_hdr_t and both of them are uncompressed then the arc_buf_t must be | |
207 | * the last buffer in the hdr's b_buf list, however a shared compressed buf can | |
208 | * be anywhere in the hdr's list. | |
d3c2ae1c GW |
209 | * |
210 | * The diagram below shows an example of an uncompressed ARC hdr that is | |
2aa34383 DK |
211 | * sharing its data with an arc_buf_t (note that the shared uncompressed buf is |
212 | * the last element in the buf list): | |
d3c2ae1c GW |
213 | * |
214 | * arc_buf_hdr_t | |
215 | * +-----------+ | |
216 | * | | | |
217 | * | | | |
218 | * | | | |
219 | * +-----------+ | |
220 | * l2arc_buf_hdr_t| | | |
221 | * | | | |
222 | * +-----------+ | |
223 | * l1arc_buf_hdr_t| | | |
224 | * | | arc_buf_t (shared) | |
225 | * | b_buf +------------>+---------+ arc_buf_t | |
226 | * | | |b_next +---->+---------+ | |
a6255b7f | 227 | * | b_pabd +-+ |---------| |b_next +-->NULL |
d3c2ae1c GW |
228 | * +-----------+ | | | +---------+ |
229 | * | |b_data +-+ | | | |
230 | * | +---------+ | |b_data +-+ | |
231 | * +->+------+ | +---------+ | | |
232 | * | | | | | |
233 | * uncompressed | | | | | |
234 | * data +------+ | | | |
235 | * ^ +->+------+ | | |
236 | * | uncompressed | | | | |
237 | * | data | | | | |
238 | * | +------+ | | |
239 | * +---------------------------------+ | |
240 | * | |
a6255b7f | 241 | * Writing to the ARC requires that the ARC first discard the hdr's b_pabd |
d3c2ae1c | 242 | * since the physical block is about to be rewritten. The new data contents |
2aa34383 DK |
243 | * will be contained in the arc_buf_t. As the I/O pipeline performs the write, |
244 | * it may compress the data before writing it to disk. The ARC will be called | |
245 | * with the transformed data and will bcopy the transformed on-disk block into | |
a6255b7f | 246 | * a newly allocated b_pabd. Writes are always done into buffers which have |
2aa34383 DK |
247 | * either been loaned (and hence are new and don't have other readers) or |
248 | * buffers which have been released (and hence have their own hdr, if there | |
249 | * were originally other readers of the buf's original hdr). This ensures that | |
250 | * the ARC only needs to update a single buf and its hdr after a write occurs. | |
d3c2ae1c | 251 | * |
a6255b7f DQ |
252 | * When the L2ARC is in use, it will also take advantage of the b_pabd. The |
253 | * L2ARC will always write the contents of b_pabd to the L2ARC. This means | |
2aa34383 | 254 | * that when compressed ARC is enabled that the L2ARC blocks are identical |
d3c2ae1c GW |
255 | * to the on-disk block in the main data pool. This provides a significant |
256 | * advantage since the ARC can leverage the bp's checksum when reading from the | |
257 | * L2ARC to determine if the contents are valid. However, if the compressed | |
2aa34383 | 258 | * ARC is disabled, then the L2ARC's block must be transformed to look |
d3c2ae1c GW |
259 | * like the physical block in the main data pool before comparing the |
260 | * checksum and determining its validity. | |
261 | */ | |
262 | ||
34dc7c2f BB |
263 | #include <sys/spa.h> |
264 | #include <sys/zio.h> | |
d3c2ae1c | 265 | #include <sys/spa_impl.h> |
3a17a7a9 | 266 | #include <sys/zio_compress.h> |
d3c2ae1c | 267 | #include <sys/zio_checksum.h> |
34dc7c2f BB |
268 | #include <sys/zfs_context.h> |
269 | #include <sys/arc.h> | |
36da08ef | 270 | #include <sys/refcount.h> |
b128c09f | 271 | #include <sys/vdev.h> |
9babb374 | 272 | #include <sys/vdev_impl.h> |
e8b96c60 | 273 | #include <sys/dsl_pool.h> |
a6255b7f | 274 | #include <sys/zio_checksum.h> |
ca0bf58d | 275 | #include <sys/multilist.h> |
a6255b7f | 276 | #include <sys/abd.h> |
34dc7c2f BB |
277 | #ifdef _KERNEL |
278 | #include <sys/vmsystm.h> | |
279 | #include <vm/anon.h> | |
280 | #include <sys/fs/swapnode.h> | |
ab26409d | 281 | #include <sys/zpl.h> |
aaed7c40 | 282 | #include <linux/mm_compat.h> |
34dc7c2f BB |
283 | #endif |
284 | #include <sys/callb.h> | |
285 | #include <sys/kstat.h> | |
570827e1 | 286 | #include <sys/dmu_tx.h> |
428870ff | 287 | #include <zfs_fletcher.h> |
59ec819a | 288 | #include <sys/arc_impl.h> |
49ee64e5 | 289 | #include <sys/trace_arc.h> |
34dc7c2f | 290 | |
498877ba MA |
291 | #ifndef _KERNEL |
292 | /* set with ZFS_DEBUG=watch, to enable watchpoints on frozen buffers */ | |
293 | boolean_t arc_watch = B_FALSE; | |
294 | #endif | |
295 | ||
ca0bf58d PS |
296 | static kmutex_t arc_reclaim_lock; |
297 | static kcondvar_t arc_reclaim_thread_cv; | |
298 | static boolean_t arc_reclaim_thread_exit; | |
299 | static kcondvar_t arc_reclaim_waiters_cv; | |
300 | ||
e8b96c60 | 301 | /* |
ca0bf58d PS |
302 | * The number of headers to evict in arc_evict_state_impl() before |
303 | * dropping the sublist lock and evicting from another sublist. A lower | |
304 | * value means we're more likely to evict the "correct" header (i.e. the | |
305 | * oldest header in the arc state), but comes with higher overhead | |
306 | * (i.e. more invocations of arc_evict_state_impl()). | |
307 | */ | |
308 | int zfs_arc_evict_batch_limit = 10; | |
309 | ||
310 | /* | |
311 | * The number of sublists used for each of the arc state lists. If this | |
312 | * is not set to a suitable value by the user, it will be configured to | |
313 | * the number of CPUs on the system in arc_init(). | |
e8b96c60 | 314 | */ |
ca0bf58d | 315 | int zfs_arc_num_sublists_per_state = 0; |
e8b96c60 | 316 | |
34dc7c2f | 317 | /* number of seconds before growing cache again */ |
ca67b33a | 318 | static int arc_grow_retry = 5; |
34dc7c2f | 319 | |
a6255b7f | 320 | /* shift of arc_c for calculating overflow limit in arc_get_data_impl */ |
ca67b33a | 321 | int zfs_arc_overflow_shift = 8; |
62422785 | 322 | |
728d6ae9 BB |
323 | /* shift of arc_c for calculating both min and max arc_p */ |
324 | static int arc_p_min_shift = 4; | |
325 | ||
d164b209 | 326 | /* log2(fraction of arc to reclaim) */ |
ca67b33a | 327 | static int arc_shrink_shift = 7; |
d164b209 | 328 | |
34dc7c2f | 329 | /* |
ca67b33a MA |
330 | * log2(fraction of ARC which must be free to allow growing). |
331 | * I.e. If there is less than arc_c >> arc_no_grow_shift free memory, | |
332 | * when reading a new block into the ARC, we will evict an equal-sized block | |
333 | * from the ARC. | |
334 | * | |
335 | * This must be less than arc_shrink_shift, so that when we shrink the ARC, | |
336 | * we will still not allow it to grow. | |
34dc7c2f | 337 | */ |
ca67b33a | 338 | int arc_no_grow_shift = 5; |
bce45ec9 | 339 | |
49ddb315 | 340 | |
ca0bf58d PS |
341 | /* |
342 | * minimum lifespan of a prefetch block in clock ticks | |
343 | * (initialized in arc_init()) | |
344 | */ | |
ca67b33a | 345 | static int arc_min_prefetch_lifespan; |
ca0bf58d | 346 | |
e8b96c60 MA |
347 | /* |
348 | * If this percent of memory is free, don't throttle. | |
349 | */ | |
350 | int arc_lotsfree_percent = 10; | |
351 | ||
34dc7c2f BB |
352 | static int arc_dead; |
353 | ||
b128c09f BB |
354 | /* |
355 | * The arc has filled available memory and has now warmed up. | |
356 | */ | |
357 | static boolean_t arc_warm; | |
358 | ||
d3c2ae1c GW |
359 | /* |
360 | * log2 fraction of the zio arena to keep free. | |
361 | */ | |
362 | int arc_zio_arena_free_shift = 2; | |
363 | ||
34dc7c2f BB |
364 | /* |
365 | * These tunables are for performance analysis. | |
366 | */ | |
c28b2279 BB |
367 | unsigned long zfs_arc_max = 0; |
368 | unsigned long zfs_arc_min = 0; | |
369 | unsigned long zfs_arc_meta_limit = 0; | |
ca0bf58d | 370 | unsigned long zfs_arc_meta_min = 0; |
25458cbe TC |
371 | unsigned long zfs_arc_dnode_limit = 0; |
372 | unsigned long zfs_arc_dnode_reduce_percent = 10; | |
ca67b33a MA |
373 | int zfs_arc_grow_retry = 0; |
374 | int zfs_arc_shrink_shift = 0; | |
728d6ae9 | 375 | int zfs_arc_p_min_shift = 0; |
ca67b33a | 376 | int zfs_arc_average_blocksize = 8 * 1024; /* 8KB */ |
34dc7c2f | 377 | |
d3c2ae1c GW |
378 | int zfs_compressed_arc_enabled = B_TRUE; |
379 | ||
9907cc1c G |
380 | /* |
381 | * ARC will evict meta buffers that exceed arc_meta_limit. This | |
382 | * tunable make arc_meta_limit adjustable for different workloads. | |
383 | */ | |
384 | unsigned long zfs_arc_meta_limit_percent = 75; | |
385 | ||
386 | /* | |
387 | * Percentage that can be consumed by dnodes of ARC meta buffers. | |
388 | */ | |
389 | unsigned long zfs_arc_dnode_limit_percent = 10; | |
390 | ||
bc888666 | 391 | /* |
ca67b33a | 392 | * These tunables are Linux specific |
bc888666 | 393 | */ |
11f552fa | 394 | unsigned long zfs_arc_sys_free = 0; |
ca67b33a MA |
395 | int zfs_arc_min_prefetch_lifespan = 0; |
396 | int zfs_arc_p_aggressive_disable = 1; | |
397 | int zfs_arc_p_dampener_disable = 1; | |
398 | int zfs_arc_meta_prune = 10000; | |
399 | int zfs_arc_meta_strategy = ARC_STRATEGY_META_BALANCED; | |
400 | int zfs_arc_meta_adjust_restarts = 4096; | |
7e8bddd0 | 401 | int zfs_arc_lotsfree_percent = 10; |
bc888666 | 402 | |
34dc7c2f BB |
403 | /* The 6 states: */ |
404 | static arc_state_t ARC_anon; | |
405 | static arc_state_t ARC_mru; | |
406 | static arc_state_t ARC_mru_ghost; | |
407 | static arc_state_t ARC_mfu; | |
408 | static arc_state_t ARC_mfu_ghost; | |
409 | static arc_state_t ARC_l2c_only; | |
410 | ||
411 | typedef struct arc_stats { | |
412 | kstat_named_t arcstat_hits; | |
413 | kstat_named_t arcstat_misses; | |
414 | kstat_named_t arcstat_demand_data_hits; | |
415 | kstat_named_t arcstat_demand_data_misses; | |
416 | kstat_named_t arcstat_demand_metadata_hits; | |
417 | kstat_named_t arcstat_demand_metadata_misses; | |
418 | kstat_named_t arcstat_prefetch_data_hits; | |
419 | kstat_named_t arcstat_prefetch_data_misses; | |
420 | kstat_named_t arcstat_prefetch_metadata_hits; | |
421 | kstat_named_t arcstat_prefetch_metadata_misses; | |
422 | kstat_named_t arcstat_mru_hits; | |
423 | kstat_named_t arcstat_mru_ghost_hits; | |
424 | kstat_named_t arcstat_mfu_hits; | |
425 | kstat_named_t arcstat_mfu_ghost_hits; | |
426 | kstat_named_t arcstat_deleted; | |
e49f1e20 WA |
427 | /* |
428 | * Number of buffers that could not be evicted because the hash lock | |
429 | * was held by another thread. The lock may not necessarily be held | |
430 | * by something using the same buffer, since hash locks are shared | |
431 | * by multiple buffers. | |
432 | */ | |
34dc7c2f | 433 | kstat_named_t arcstat_mutex_miss; |
e49f1e20 WA |
434 | /* |
435 | * Number of buffers skipped because they have I/O in progress, are | |
436 | * indrect prefetch buffers that have not lived long enough, or are | |
437 | * not from the spa we're trying to evict from. | |
438 | */ | |
34dc7c2f | 439 | kstat_named_t arcstat_evict_skip; |
ca0bf58d PS |
440 | /* |
441 | * Number of times arc_evict_state() was unable to evict enough | |
442 | * buffers to reach its target amount. | |
443 | */ | |
444 | kstat_named_t arcstat_evict_not_enough; | |
428870ff BB |
445 | kstat_named_t arcstat_evict_l2_cached; |
446 | kstat_named_t arcstat_evict_l2_eligible; | |
447 | kstat_named_t arcstat_evict_l2_ineligible; | |
ca0bf58d | 448 | kstat_named_t arcstat_evict_l2_skip; |
34dc7c2f BB |
449 | kstat_named_t arcstat_hash_elements; |
450 | kstat_named_t arcstat_hash_elements_max; | |
451 | kstat_named_t arcstat_hash_collisions; | |
452 | kstat_named_t arcstat_hash_chains; | |
453 | kstat_named_t arcstat_hash_chain_max; | |
454 | kstat_named_t arcstat_p; | |
455 | kstat_named_t arcstat_c; | |
456 | kstat_named_t arcstat_c_min; | |
457 | kstat_named_t arcstat_c_max; | |
458 | kstat_named_t arcstat_size; | |
d3c2ae1c | 459 | /* |
a6255b7f | 460 | * Number of compressed bytes stored in the arc_buf_hdr_t's b_pabd. |
d3c2ae1c GW |
461 | * Note that the compressed bytes may match the uncompressed bytes |
462 | * if the block is either not compressed or compressed arc is disabled. | |
463 | */ | |
464 | kstat_named_t arcstat_compressed_size; | |
465 | /* | |
a6255b7f | 466 | * Uncompressed size of the data stored in b_pabd. If compressed |
d3c2ae1c GW |
467 | * arc is disabled then this value will be identical to the stat |
468 | * above. | |
469 | */ | |
470 | kstat_named_t arcstat_uncompressed_size; | |
471 | /* | |
472 | * Number of bytes stored in all the arc_buf_t's. This is classified | |
473 | * as "overhead" since this data is typically short-lived and will | |
474 | * be evicted from the arc when it becomes unreferenced unless the | |
475 | * zfs_keep_uncompressed_metadata or zfs_keep_uncompressed_level | |
476 | * values have been set (see comment in dbuf.c for more information). | |
477 | */ | |
478 | kstat_named_t arcstat_overhead_size; | |
500445c0 PS |
479 | /* |
480 | * Number of bytes consumed by internal ARC structures necessary | |
481 | * for tracking purposes; these structures are not actually | |
482 | * backed by ARC buffers. This includes arc_buf_hdr_t structures | |
483 | * (allocated via arc_buf_hdr_t_full and arc_buf_hdr_t_l2only | |
484 | * caches), and arc_buf_t structures (allocated via arc_buf_t | |
485 | * cache). | |
486 | */ | |
34dc7c2f | 487 | kstat_named_t arcstat_hdr_size; |
500445c0 PS |
488 | /* |
489 | * Number of bytes consumed by ARC buffers of type equal to | |
490 | * ARC_BUFC_DATA. This is generally consumed by buffers backing | |
491 | * on disk user data (e.g. plain file contents). | |
492 | */ | |
d164b209 | 493 | kstat_named_t arcstat_data_size; |
500445c0 PS |
494 | /* |
495 | * Number of bytes consumed by ARC buffers of type equal to | |
496 | * ARC_BUFC_METADATA. This is generally consumed by buffers | |
497 | * backing on disk data that is used for internal ZFS | |
498 | * structures (e.g. ZAP, dnode, indirect blocks, etc). | |
499 | */ | |
500 | kstat_named_t arcstat_metadata_size; | |
501 | /* | |
25458cbe | 502 | * Number of bytes consumed by dmu_buf_impl_t objects. |
500445c0 | 503 | */ |
25458cbe TC |
504 | kstat_named_t arcstat_dbuf_size; |
505 | /* | |
506 | * Number of bytes consumed by dnode_t objects. | |
507 | */ | |
508 | kstat_named_t arcstat_dnode_size; | |
509 | /* | |
510 | * Number of bytes consumed by bonus buffers. | |
511 | */ | |
512 | kstat_named_t arcstat_bonus_size; | |
500445c0 PS |
513 | /* |
514 | * Total number of bytes consumed by ARC buffers residing in the | |
515 | * arc_anon state. This includes *all* buffers in the arc_anon | |
516 | * state; e.g. data, metadata, evictable, and unevictable buffers | |
517 | * are all included in this value. | |
518 | */ | |
13be560d | 519 | kstat_named_t arcstat_anon_size; |
500445c0 PS |
520 | /* |
521 | * Number of bytes consumed by ARC buffers that meet the | |
522 | * following criteria: backing buffers of type ARC_BUFC_DATA, | |
523 | * residing in the arc_anon state, and are eligible for eviction | |
524 | * (e.g. have no outstanding holds on the buffer). | |
525 | */ | |
526 | kstat_named_t arcstat_anon_evictable_data; | |
527 | /* | |
528 | * Number of bytes consumed by ARC buffers that meet the | |
529 | * following criteria: backing buffers of type ARC_BUFC_METADATA, | |
530 | * residing in the arc_anon state, and are eligible for eviction | |
531 | * (e.g. have no outstanding holds on the buffer). | |
532 | */ | |
533 | kstat_named_t arcstat_anon_evictable_metadata; | |
534 | /* | |
535 | * Total number of bytes consumed by ARC buffers residing in the | |
536 | * arc_mru state. This includes *all* buffers in the arc_mru | |
537 | * state; e.g. data, metadata, evictable, and unevictable buffers | |
538 | * are all included in this value. | |
539 | */ | |
13be560d | 540 | kstat_named_t arcstat_mru_size; |
500445c0 PS |
541 | /* |
542 | * Number of bytes consumed by ARC buffers that meet the | |
543 | * following criteria: backing buffers of type ARC_BUFC_DATA, | |
544 | * residing in the arc_mru state, and are eligible for eviction | |
545 | * (e.g. have no outstanding holds on the buffer). | |
546 | */ | |
547 | kstat_named_t arcstat_mru_evictable_data; | |
548 | /* | |
549 | * Number of bytes consumed by ARC buffers that meet the | |
550 | * following criteria: backing buffers of type ARC_BUFC_METADATA, | |
551 | * residing in the arc_mru state, and are eligible for eviction | |
552 | * (e.g. have no outstanding holds on the buffer). | |
553 | */ | |
554 | kstat_named_t arcstat_mru_evictable_metadata; | |
555 | /* | |
556 | * Total number of bytes that *would have been* consumed by ARC | |
557 | * buffers in the arc_mru_ghost state. The key thing to note | |
558 | * here, is the fact that this size doesn't actually indicate | |
559 | * RAM consumption. The ghost lists only consist of headers and | |
560 | * don't actually have ARC buffers linked off of these headers. | |
561 | * Thus, *if* the headers had associated ARC buffers, these | |
562 | * buffers *would have* consumed this number of bytes. | |
563 | */ | |
13be560d | 564 | kstat_named_t arcstat_mru_ghost_size; |
500445c0 PS |
565 | /* |
566 | * Number of bytes that *would have been* consumed by ARC | |
567 | * buffers that are eligible for eviction, of type | |
568 | * ARC_BUFC_DATA, and linked off the arc_mru_ghost state. | |
569 | */ | |
570 | kstat_named_t arcstat_mru_ghost_evictable_data; | |
571 | /* | |
572 | * Number of bytes that *would have been* consumed by ARC | |
573 | * buffers that are eligible for eviction, of type | |
574 | * ARC_BUFC_METADATA, and linked off the arc_mru_ghost state. | |
575 | */ | |
576 | kstat_named_t arcstat_mru_ghost_evictable_metadata; | |
577 | /* | |
578 | * Total number of bytes consumed by ARC buffers residing in the | |
579 | * arc_mfu state. This includes *all* buffers in the arc_mfu | |
580 | * state; e.g. data, metadata, evictable, and unevictable buffers | |
581 | * are all included in this value. | |
582 | */ | |
13be560d | 583 | kstat_named_t arcstat_mfu_size; |
500445c0 PS |
584 | /* |
585 | * Number of bytes consumed by ARC buffers that are eligible for | |
586 | * eviction, of type ARC_BUFC_DATA, and reside in the arc_mfu | |
587 | * state. | |
588 | */ | |
589 | kstat_named_t arcstat_mfu_evictable_data; | |
590 | /* | |
591 | * Number of bytes consumed by ARC buffers that are eligible for | |
592 | * eviction, of type ARC_BUFC_METADATA, and reside in the | |
593 | * arc_mfu state. | |
594 | */ | |
595 | kstat_named_t arcstat_mfu_evictable_metadata; | |
596 | /* | |
597 | * Total number of bytes that *would have been* consumed by ARC | |
598 | * buffers in the arc_mfu_ghost state. See the comment above | |
599 | * arcstat_mru_ghost_size for more details. | |
600 | */ | |
13be560d | 601 | kstat_named_t arcstat_mfu_ghost_size; |
500445c0 PS |
602 | /* |
603 | * Number of bytes that *would have been* consumed by ARC | |
604 | * buffers that are eligible for eviction, of type | |
605 | * ARC_BUFC_DATA, and linked off the arc_mfu_ghost state. | |
606 | */ | |
607 | kstat_named_t arcstat_mfu_ghost_evictable_data; | |
608 | /* | |
609 | * Number of bytes that *would have been* consumed by ARC | |
610 | * buffers that are eligible for eviction, of type | |
611 | * ARC_BUFC_METADATA, and linked off the arc_mru_ghost state. | |
612 | */ | |
613 | kstat_named_t arcstat_mfu_ghost_evictable_metadata; | |
34dc7c2f BB |
614 | kstat_named_t arcstat_l2_hits; |
615 | kstat_named_t arcstat_l2_misses; | |
616 | kstat_named_t arcstat_l2_feeds; | |
617 | kstat_named_t arcstat_l2_rw_clash; | |
d164b209 BB |
618 | kstat_named_t arcstat_l2_read_bytes; |
619 | kstat_named_t arcstat_l2_write_bytes; | |
34dc7c2f BB |
620 | kstat_named_t arcstat_l2_writes_sent; |
621 | kstat_named_t arcstat_l2_writes_done; | |
622 | kstat_named_t arcstat_l2_writes_error; | |
ca0bf58d | 623 | kstat_named_t arcstat_l2_writes_lock_retry; |
34dc7c2f BB |
624 | kstat_named_t arcstat_l2_evict_lock_retry; |
625 | kstat_named_t arcstat_l2_evict_reading; | |
b9541d6b | 626 | kstat_named_t arcstat_l2_evict_l1cached; |
34dc7c2f BB |
627 | kstat_named_t arcstat_l2_free_on_write; |
628 | kstat_named_t arcstat_l2_abort_lowmem; | |
629 | kstat_named_t arcstat_l2_cksum_bad; | |
630 | kstat_named_t arcstat_l2_io_error; | |
631 | kstat_named_t arcstat_l2_size; | |
3a17a7a9 | 632 | kstat_named_t arcstat_l2_asize; |
34dc7c2f BB |
633 | kstat_named_t arcstat_l2_hdr_size; |
634 | kstat_named_t arcstat_memory_throttle_count; | |
7cb67b45 BB |
635 | kstat_named_t arcstat_memory_direct_count; |
636 | kstat_named_t arcstat_memory_indirect_count; | |
1834f2d8 BB |
637 | kstat_named_t arcstat_no_grow; |
638 | kstat_named_t arcstat_tempreserve; | |
639 | kstat_named_t arcstat_loaned_bytes; | |
ab26409d | 640 | kstat_named_t arcstat_prune; |
1834f2d8 BB |
641 | kstat_named_t arcstat_meta_used; |
642 | kstat_named_t arcstat_meta_limit; | |
25458cbe | 643 | kstat_named_t arcstat_dnode_limit; |
1834f2d8 | 644 | kstat_named_t arcstat_meta_max; |
ca0bf58d | 645 | kstat_named_t arcstat_meta_min; |
7f60329a MA |
646 | kstat_named_t arcstat_sync_wait_for_async; |
647 | kstat_named_t arcstat_demand_hit_predictive_prefetch; | |
11f552fa BB |
648 | kstat_named_t arcstat_need_free; |
649 | kstat_named_t arcstat_sys_free; | |
34dc7c2f BB |
650 | } arc_stats_t; |
651 | ||
652 | static arc_stats_t arc_stats = { | |
653 | { "hits", KSTAT_DATA_UINT64 }, | |
654 | { "misses", KSTAT_DATA_UINT64 }, | |
655 | { "demand_data_hits", KSTAT_DATA_UINT64 }, | |
656 | { "demand_data_misses", KSTAT_DATA_UINT64 }, | |
657 | { "demand_metadata_hits", KSTAT_DATA_UINT64 }, | |
658 | { "demand_metadata_misses", KSTAT_DATA_UINT64 }, | |
659 | { "prefetch_data_hits", KSTAT_DATA_UINT64 }, | |
660 | { "prefetch_data_misses", KSTAT_DATA_UINT64 }, | |
661 | { "prefetch_metadata_hits", KSTAT_DATA_UINT64 }, | |
662 | { "prefetch_metadata_misses", KSTAT_DATA_UINT64 }, | |
663 | { "mru_hits", KSTAT_DATA_UINT64 }, | |
664 | { "mru_ghost_hits", KSTAT_DATA_UINT64 }, | |
665 | { "mfu_hits", KSTAT_DATA_UINT64 }, | |
666 | { "mfu_ghost_hits", KSTAT_DATA_UINT64 }, | |
667 | { "deleted", KSTAT_DATA_UINT64 }, | |
34dc7c2f BB |
668 | { "mutex_miss", KSTAT_DATA_UINT64 }, |
669 | { "evict_skip", KSTAT_DATA_UINT64 }, | |
ca0bf58d | 670 | { "evict_not_enough", KSTAT_DATA_UINT64 }, |
428870ff BB |
671 | { "evict_l2_cached", KSTAT_DATA_UINT64 }, |
672 | { "evict_l2_eligible", KSTAT_DATA_UINT64 }, | |
673 | { "evict_l2_ineligible", KSTAT_DATA_UINT64 }, | |
ca0bf58d | 674 | { "evict_l2_skip", KSTAT_DATA_UINT64 }, |
34dc7c2f BB |
675 | { "hash_elements", KSTAT_DATA_UINT64 }, |
676 | { "hash_elements_max", KSTAT_DATA_UINT64 }, | |
677 | { "hash_collisions", KSTAT_DATA_UINT64 }, | |
678 | { "hash_chains", KSTAT_DATA_UINT64 }, | |
679 | { "hash_chain_max", KSTAT_DATA_UINT64 }, | |
680 | { "p", KSTAT_DATA_UINT64 }, | |
681 | { "c", KSTAT_DATA_UINT64 }, | |
682 | { "c_min", KSTAT_DATA_UINT64 }, | |
683 | { "c_max", KSTAT_DATA_UINT64 }, | |
684 | { "size", KSTAT_DATA_UINT64 }, | |
d3c2ae1c GW |
685 | { "compressed_size", KSTAT_DATA_UINT64 }, |
686 | { "uncompressed_size", KSTAT_DATA_UINT64 }, | |
687 | { "overhead_size", KSTAT_DATA_UINT64 }, | |
34dc7c2f | 688 | { "hdr_size", KSTAT_DATA_UINT64 }, |
d164b209 | 689 | { "data_size", KSTAT_DATA_UINT64 }, |
500445c0 | 690 | { "metadata_size", KSTAT_DATA_UINT64 }, |
25458cbe TC |
691 | { "dbuf_size", KSTAT_DATA_UINT64 }, |
692 | { "dnode_size", KSTAT_DATA_UINT64 }, | |
693 | { "bonus_size", KSTAT_DATA_UINT64 }, | |
13be560d | 694 | { "anon_size", KSTAT_DATA_UINT64 }, |
500445c0 PS |
695 | { "anon_evictable_data", KSTAT_DATA_UINT64 }, |
696 | { "anon_evictable_metadata", KSTAT_DATA_UINT64 }, | |
13be560d | 697 | { "mru_size", KSTAT_DATA_UINT64 }, |
500445c0 PS |
698 | { "mru_evictable_data", KSTAT_DATA_UINT64 }, |
699 | { "mru_evictable_metadata", KSTAT_DATA_UINT64 }, | |
13be560d | 700 | { "mru_ghost_size", KSTAT_DATA_UINT64 }, |
500445c0 PS |
701 | { "mru_ghost_evictable_data", KSTAT_DATA_UINT64 }, |
702 | { "mru_ghost_evictable_metadata", KSTAT_DATA_UINT64 }, | |
13be560d | 703 | { "mfu_size", KSTAT_DATA_UINT64 }, |
500445c0 PS |
704 | { "mfu_evictable_data", KSTAT_DATA_UINT64 }, |
705 | { "mfu_evictable_metadata", KSTAT_DATA_UINT64 }, | |
13be560d | 706 | { "mfu_ghost_size", KSTAT_DATA_UINT64 }, |
500445c0 PS |
707 | { "mfu_ghost_evictable_data", KSTAT_DATA_UINT64 }, |
708 | { "mfu_ghost_evictable_metadata", KSTAT_DATA_UINT64 }, | |
34dc7c2f BB |
709 | { "l2_hits", KSTAT_DATA_UINT64 }, |
710 | { "l2_misses", KSTAT_DATA_UINT64 }, | |
711 | { "l2_feeds", KSTAT_DATA_UINT64 }, | |
712 | { "l2_rw_clash", KSTAT_DATA_UINT64 }, | |
d164b209 BB |
713 | { "l2_read_bytes", KSTAT_DATA_UINT64 }, |
714 | { "l2_write_bytes", KSTAT_DATA_UINT64 }, | |
34dc7c2f BB |
715 | { "l2_writes_sent", KSTAT_DATA_UINT64 }, |
716 | { "l2_writes_done", KSTAT_DATA_UINT64 }, | |
717 | { "l2_writes_error", KSTAT_DATA_UINT64 }, | |
ca0bf58d | 718 | { "l2_writes_lock_retry", KSTAT_DATA_UINT64 }, |
34dc7c2f BB |
719 | { "l2_evict_lock_retry", KSTAT_DATA_UINT64 }, |
720 | { "l2_evict_reading", KSTAT_DATA_UINT64 }, | |
b9541d6b | 721 | { "l2_evict_l1cached", KSTAT_DATA_UINT64 }, |
34dc7c2f BB |
722 | { "l2_free_on_write", KSTAT_DATA_UINT64 }, |
723 | { "l2_abort_lowmem", KSTAT_DATA_UINT64 }, | |
724 | { "l2_cksum_bad", KSTAT_DATA_UINT64 }, | |
725 | { "l2_io_error", KSTAT_DATA_UINT64 }, | |
726 | { "l2_size", KSTAT_DATA_UINT64 }, | |
3a17a7a9 | 727 | { "l2_asize", KSTAT_DATA_UINT64 }, |
34dc7c2f | 728 | { "l2_hdr_size", KSTAT_DATA_UINT64 }, |
1834f2d8 | 729 | { "memory_throttle_count", KSTAT_DATA_UINT64 }, |
7cb67b45 BB |
730 | { "memory_direct_count", KSTAT_DATA_UINT64 }, |
731 | { "memory_indirect_count", KSTAT_DATA_UINT64 }, | |
1834f2d8 BB |
732 | { "arc_no_grow", KSTAT_DATA_UINT64 }, |
733 | { "arc_tempreserve", KSTAT_DATA_UINT64 }, | |
734 | { "arc_loaned_bytes", KSTAT_DATA_UINT64 }, | |
ab26409d | 735 | { "arc_prune", KSTAT_DATA_UINT64 }, |
1834f2d8 BB |
736 | { "arc_meta_used", KSTAT_DATA_UINT64 }, |
737 | { "arc_meta_limit", KSTAT_DATA_UINT64 }, | |
25458cbe | 738 | { "arc_dnode_limit", KSTAT_DATA_UINT64 }, |
1834f2d8 | 739 | { "arc_meta_max", KSTAT_DATA_UINT64 }, |
11f552fa | 740 | { "arc_meta_min", KSTAT_DATA_UINT64 }, |
7f60329a MA |
741 | { "sync_wait_for_async", KSTAT_DATA_UINT64 }, |
742 | { "demand_hit_predictive_prefetch", KSTAT_DATA_UINT64 }, | |
11f552fa BB |
743 | { "arc_need_free", KSTAT_DATA_UINT64 }, |
744 | { "arc_sys_free", KSTAT_DATA_UINT64 } | |
34dc7c2f BB |
745 | }; |
746 | ||
747 | #define ARCSTAT(stat) (arc_stats.stat.value.ui64) | |
748 | ||
749 | #define ARCSTAT_INCR(stat, val) \ | |
d3cc8b15 | 750 | atomic_add_64(&arc_stats.stat.value.ui64, (val)) |
34dc7c2f | 751 | |
428870ff | 752 | #define ARCSTAT_BUMP(stat) ARCSTAT_INCR(stat, 1) |
34dc7c2f BB |
753 | #define ARCSTAT_BUMPDOWN(stat) ARCSTAT_INCR(stat, -1) |
754 | ||
755 | #define ARCSTAT_MAX(stat, val) { \ | |
756 | uint64_t m; \ | |
757 | while ((val) > (m = arc_stats.stat.value.ui64) && \ | |
758 | (m != atomic_cas_64(&arc_stats.stat.value.ui64, m, (val)))) \ | |
759 | continue; \ | |
760 | } | |
761 | ||
762 | #define ARCSTAT_MAXSTAT(stat) \ | |
763 | ARCSTAT_MAX(stat##_max, arc_stats.stat.value.ui64) | |
764 | ||
765 | /* | |
766 | * We define a macro to allow ARC hits/misses to be easily broken down by | |
767 | * two separate conditions, giving a total of four different subtypes for | |
768 | * each of hits and misses (so eight statistics total). | |
769 | */ | |
770 | #define ARCSTAT_CONDSTAT(cond1, stat1, notstat1, cond2, stat2, notstat2, stat) \ | |
771 | if (cond1) { \ | |
772 | if (cond2) { \ | |
773 | ARCSTAT_BUMP(arcstat_##stat1##_##stat2##_##stat); \ | |
774 | } else { \ | |
775 | ARCSTAT_BUMP(arcstat_##stat1##_##notstat2##_##stat); \ | |
776 | } \ | |
777 | } else { \ | |
778 | if (cond2) { \ | |
779 | ARCSTAT_BUMP(arcstat_##notstat1##_##stat2##_##stat); \ | |
780 | } else { \ | |
781 | ARCSTAT_BUMP(arcstat_##notstat1##_##notstat2##_##stat);\ | |
782 | } \ | |
783 | } | |
784 | ||
785 | kstat_t *arc_ksp; | |
428870ff | 786 | static arc_state_t *arc_anon; |
34dc7c2f BB |
787 | static arc_state_t *arc_mru; |
788 | static arc_state_t *arc_mru_ghost; | |
789 | static arc_state_t *arc_mfu; | |
790 | static arc_state_t *arc_mfu_ghost; | |
791 | static arc_state_t *arc_l2c_only; | |
792 | ||
793 | /* | |
794 | * There are several ARC variables that are critical to export as kstats -- | |
795 | * but we don't want to have to grovel around in the kstat whenever we wish to | |
796 | * manipulate them. For these variables, we therefore define them to be in | |
797 | * terms of the statistic variable. This assures that we are not introducing | |
798 | * the possibility of inconsistency by having shadow copies of the variables, | |
799 | * while still allowing the code to be readable. | |
800 | */ | |
801 | #define arc_size ARCSTAT(arcstat_size) /* actual total arc size */ | |
802 | #define arc_p ARCSTAT(arcstat_p) /* target size of MRU */ | |
803 | #define arc_c ARCSTAT(arcstat_c) /* target size of cache */ | |
804 | #define arc_c_min ARCSTAT(arcstat_c_min) /* min target cache size */ | |
805 | #define arc_c_max ARCSTAT(arcstat_c_max) /* max target cache size */ | |
d3c2ae1c | 806 | #define arc_no_grow ARCSTAT(arcstat_no_grow) /* do not grow cache size */ |
1834f2d8 BB |
807 | #define arc_tempreserve ARCSTAT(arcstat_tempreserve) |
808 | #define arc_loaned_bytes ARCSTAT(arcstat_loaned_bytes) | |
23c0a133 | 809 | #define arc_meta_limit ARCSTAT(arcstat_meta_limit) /* max size for metadata */ |
25458cbe | 810 | #define arc_dnode_limit ARCSTAT(arcstat_dnode_limit) /* max size for dnodes */ |
ca0bf58d | 811 | #define arc_meta_min ARCSTAT(arcstat_meta_min) /* min size for metadata */ |
23c0a133 GW |
812 | #define arc_meta_used ARCSTAT(arcstat_meta_used) /* size of metadata */ |
813 | #define arc_meta_max ARCSTAT(arcstat_meta_max) /* max size of metadata */ | |
25458cbe TC |
814 | #define arc_dbuf_size ARCSTAT(arcstat_dbuf_size) /* dbuf metadata */ |
815 | #define arc_dnode_size ARCSTAT(arcstat_dnode_size) /* dnode metadata */ | |
816 | #define arc_bonus_size ARCSTAT(arcstat_bonus_size) /* bonus buffer metadata */ | |
11f552fa BB |
817 | #define arc_need_free ARCSTAT(arcstat_need_free) /* bytes to be freed */ |
818 | #define arc_sys_free ARCSTAT(arcstat_sys_free) /* target system free bytes */ | |
34dc7c2f | 819 | |
d3c2ae1c GW |
820 | /* compressed size of entire arc */ |
821 | #define arc_compressed_size ARCSTAT(arcstat_compressed_size) | |
822 | /* uncompressed size of entire arc */ | |
823 | #define arc_uncompressed_size ARCSTAT(arcstat_uncompressed_size) | |
824 | /* number of bytes in the arc from arc_buf_t's */ | |
825 | #define arc_overhead_size ARCSTAT(arcstat_overhead_size) | |
3a17a7a9 | 826 | |
ab26409d BB |
827 | static list_t arc_prune_list; |
828 | static kmutex_t arc_prune_mtx; | |
f6046738 | 829 | static taskq_t *arc_prune_taskq; |
428870ff | 830 | |
34dc7c2f BB |
831 | #define GHOST_STATE(state) \ |
832 | ((state) == arc_mru_ghost || (state) == arc_mfu_ghost || \ | |
833 | (state) == arc_l2c_only) | |
834 | ||
2a432414 GW |
835 | #define HDR_IN_HASH_TABLE(hdr) ((hdr)->b_flags & ARC_FLAG_IN_HASH_TABLE) |
836 | #define HDR_IO_IN_PROGRESS(hdr) ((hdr)->b_flags & ARC_FLAG_IO_IN_PROGRESS) | |
837 | #define HDR_IO_ERROR(hdr) ((hdr)->b_flags & ARC_FLAG_IO_ERROR) | |
838 | #define HDR_PREFETCH(hdr) ((hdr)->b_flags & ARC_FLAG_PREFETCH) | |
d3c2ae1c GW |
839 | #define HDR_COMPRESSION_ENABLED(hdr) \ |
840 | ((hdr)->b_flags & ARC_FLAG_COMPRESSED_ARC) | |
b9541d6b | 841 | |
2a432414 GW |
842 | #define HDR_L2CACHE(hdr) ((hdr)->b_flags & ARC_FLAG_L2CACHE) |
843 | #define HDR_L2_READING(hdr) \ | |
d3c2ae1c GW |
844 | (((hdr)->b_flags & ARC_FLAG_IO_IN_PROGRESS) && \ |
845 | ((hdr)->b_flags & ARC_FLAG_HAS_L2HDR)) | |
2a432414 GW |
846 | #define HDR_L2_WRITING(hdr) ((hdr)->b_flags & ARC_FLAG_L2_WRITING) |
847 | #define HDR_L2_EVICTED(hdr) ((hdr)->b_flags & ARC_FLAG_L2_EVICTED) | |
848 | #define HDR_L2_WRITE_HEAD(hdr) ((hdr)->b_flags & ARC_FLAG_L2_WRITE_HEAD) | |
d3c2ae1c | 849 | #define HDR_SHARED_DATA(hdr) ((hdr)->b_flags & ARC_FLAG_SHARED_DATA) |
34dc7c2f | 850 | |
b9541d6b | 851 | #define HDR_ISTYPE_METADATA(hdr) \ |
d3c2ae1c | 852 | ((hdr)->b_flags & ARC_FLAG_BUFC_METADATA) |
b9541d6b CW |
853 | #define HDR_ISTYPE_DATA(hdr) (!HDR_ISTYPE_METADATA(hdr)) |
854 | ||
855 | #define HDR_HAS_L1HDR(hdr) ((hdr)->b_flags & ARC_FLAG_HAS_L1HDR) | |
856 | #define HDR_HAS_L2HDR(hdr) ((hdr)->b_flags & ARC_FLAG_HAS_L2HDR) | |
857 | ||
d3c2ae1c GW |
858 | /* For storing compression mode in b_flags */ |
859 | #define HDR_COMPRESS_OFFSET (highbit64(ARC_FLAG_COMPRESS_0) - 1) | |
860 | ||
861 | #define HDR_GET_COMPRESS(hdr) ((enum zio_compress)BF32_GET((hdr)->b_flags, \ | |
862 | HDR_COMPRESS_OFFSET, SPA_COMPRESSBITS)) | |
863 | #define HDR_SET_COMPRESS(hdr, cmp) BF32_SET((hdr)->b_flags, \ | |
864 | HDR_COMPRESS_OFFSET, SPA_COMPRESSBITS, (cmp)); | |
865 | ||
866 | #define ARC_BUF_LAST(buf) ((buf)->b_next == NULL) | |
524b4217 DK |
867 | #define ARC_BUF_SHARED(buf) ((buf)->b_flags & ARC_BUF_FLAG_SHARED) |
868 | #define ARC_BUF_COMPRESSED(buf) ((buf)->b_flags & ARC_BUF_FLAG_COMPRESSED) | |
d3c2ae1c | 869 | |
34dc7c2f BB |
870 | /* |
871 | * Other sizes | |
872 | */ | |
873 | ||
b9541d6b CW |
874 | #define HDR_FULL_SIZE ((int64_t)sizeof (arc_buf_hdr_t)) |
875 | #define HDR_L2ONLY_SIZE ((int64_t)offsetof(arc_buf_hdr_t, b_l1hdr)) | |
34dc7c2f BB |
876 | |
877 | /* | |
878 | * Hash table routines | |
879 | */ | |
880 | ||
00b46022 BB |
881 | #define HT_LOCK_ALIGN 64 |
882 | #define HT_LOCK_PAD (P2NPHASE(sizeof (kmutex_t), (HT_LOCK_ALIGN))) | |
34dc7c2f BB |
883 | |
884 | struct ht_lock { | |
885 | kmutex_t ht_lock; | |
886 | #ifdef _KERNEL | |
00b46022 | 887 | unsigned char pad[HT_LOCK_PAD]; |
34dc7c2f BB |
888 | #endif |
889 | }; | |
890 | ||
b31d8ea7 | 891 | #define BUF_LOCKS 8192 |
34dc7c2f BB |
892 | typedef struct buf_hash_table { |
893 | uint64_t ht_mask; | |
894 | arc_buf_hdr_t **ht_table; | |
895 | struct ht_lock ht_locks[BUF_LOCKS]; | |
896 | } buf_hash_table_t; | |
897 | ||
898 | static buf_hash_table_t buf_hash_table; | |
899 | ||
900 | #define BUF_HASH_INDEX(spa, dva, birth) \ | |
901 | (buf_hash(spa, dva, birth) & buf_hash_table.ht_mask) | |
902 | #define BUF_HASH_LOCK_NTRY(idx) (buf_hash_table.ht_locks[idx & (BUF_LOCKS-1)]) | |
903 | #define BUF_HASH_LOCK(idx) (&(BUF_HASH_LOCK_NTRY(idx).ht_lock)) | |
428870ff BB |
904 | #define HDR_LOCK(hdr) \ |
905 | (BUF_HASH_LOCK(BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth))) | |
34dc7c2f BB |
906 | |
907 | uint64_t zfs_crc64_table[256]; | |
908 | ||
909 | /* | |
910 | * Level 2 ARC | |
911 | */ | |
912 | ||
913 | #define L2ARC_WRITE_SIZE (8 * 1024 * 1024) /* initial write max */ | |
3a17a7a9 | 914 | #define L2ARC_HEADROOM 2 /* num of writes */ |
8a09d5fd | 915 | |
3a17a7a9 SK |
916 | /* |
917 | * If we discover during ARC scan any buffers to be compressed, we boost | |
918 | * our headroom for the next scanning cycle by this percentage multiple. | |
919 | */ | |
920 | #define L2ARC_HEADROOM_BOOST 200 | |
d164b209 BB |
921 | #define L2ARC_FEED_SECS 1 /* caching interval secs */ |
922 | #define L2ARC_FEED_MIN_MS 200 /* min caching interval ms */ | |
34dc7c2f | 923 | |
4aafab91 G |
924 | /* |
925 | * We can feed L2ARC from two states of ARC buffers, mru and mfu, | |
926 | * and each of the state has two types: data and metadata. | |
927 | */ | |
928 | #define L2ARC_FEED_TYPES 4 | |
929 | ||
34dc7c2f BB |
930 | #define l2arc_writes_sent ARCSTAT(arcstat_l2_writes_sent) |
931 | #define l2arc_writes_done ARCSTAT(arcstat_l2_writes_done) | |
932 | ||
d3cc8b15 | 933 | /* L2ARC Performance Tunables */ |
abd8610c BB |
934 | unsigned long l2arc_write_max = L2ARC_WRITE_SIZE; /* def max write size */ |
935 | unsigned long l2arc_write_boost = L2ARC_WRITE_SIZE; /* extra warmup write */ | |
936 | unsigned long l2arc_headroom = L2ARC_HEADROOM; /* # of dev writes */ | |
3a17a7a9 | 937 | unsigned long l2arc_headroom_boost = L2ARC_HEADROOM_BOOST; |
abd8610c BB |
938 | unsigned long l2arc_feed_secs = L2ARC_FEED_SECS; /* interval seconds */ |
939 | unsigned long l2arc_feed_min_ms = L2ARC_FEED_MIN_MS; /* min interval msecs */ | |
940 | int l2arc_noprefetch = B_TRUE; /* don't cache prefetch bufs */ | |
941 | int l2arc_feed_again = B_TRUE; /* turbo warmup */ | |
c93504f0 | 942 | int l2arc_norw = B_FALSE; /* no reads during writes */ |
34dc7c2f BB |
943 | |
944 | /* | |
945 | * L2ARC Internals | |
946 | */ | |
34dc7c2f BB |
947 | static list_t L2ARC_dev_list; /* device list */ |
948 | static list_t *l2arc_dev_list; /* device list pointer */ | |
949 | static kmutex_t l2arc_dev_mtx; /* device list mutex */ | |
950 | static l2arc_dev_t *l2arc_dev_last; /* last device used */ | |
34dc7c2f BB |
951 | static list_t L2ARC_free_on_write; /* free after write buf list */ |
952 | static list_t *l2arc_free_on_write; /* free after write list ptr */ | |
953 | static kmutex_t l2arc_free_on_write_mtx; /* mutex for list */ | |
954 | static uint64_t l2arc_ndev; /* number of devices */ | |
955 | ||
956 | typedef struct l2arc_read_callback { | |
2aa34383 | 957 | arc_buf_hdr_t *l2rcb_hdr; /* read header */ |
3a17a7a9 | 958 | blkptr_t l2rcb_bp; /* original blkptr */ |
5dbd68a3 | 959 | zbookmark_phys_t l2rcb_zb; /* original bookmark */ |
3a17a7a9 | 960 | int l2rcb_flags; /* original flags */ |
34dc7c2f BB |
961 | } l2arc_read_callback_t; |
962 | ||
34dc7c2f BB |
963 | typedef struct l2arc_data_free { |
964 | /* protected by l2arc_free_on_write_mtx */ | |
a6255b7f | 965 | abd_t *l2df_abd; |
34dc7c2f | 966 | size_t l2df_size; |
d3c2ae1c | 967 | arc_buf_contents_t l2df_type; |
34dc7c2f BB |
968 | list_node_t l2df_list_node; |
969 | } l2arc_data_free_t; | |
970 | ||
971 | static kmutex_t l2arc_feed_thr_lock; | |
972 | static kcondvar_t l2arc_feed_thr_cv; | |
973 | static uint8_t l2arc_thread_exit; | |
974 | ||
a6255b7f | 975 | static abd_t *arc_get_data_abd(arc_buf_hdr_t *, uint64_t, void *); |
d3c2ae1c | 976 | static void *arc_get_data_buf(arc_buf_hdr_t *, uint64_t, void *); |
a6255b7f DQ |
977 | static void arc_get_data_impl(arc_buf_hdr_t *, uint64_t, void *); |
978 | static void arc_free_data_abd(arc_buf_hdr_t *, abd_t *, uint64_t, void *); | |
d3c2ae1c | 979 | static void arc_free_data_buf(arc_buf_hdr_t *, void *, uint64_t, void *); |
a6255b7f DQ |
980 | static void arc_free_data_impl(arc_buf_hdr_t *hdr, uint64_t size, void *tag); |
981 | static void arc_hdr_free_pabd(arc_buf_hdr_t *); | |
982 | static void arc_hdr_alloc_pabd(arc_buf_hdr_t *); | |
2a432414 | 983 | static void arc_access(arc_buf_hdr_t *, kmutex_t *); |
ca0bf58d | 984 | static boolean_t arc_is_overflowing(void); |
2a432414 | 985 | static void arc_buf_watch(arc_buf_t *); |
ca67b33a | 986 | static void arc_tuning_update(void); |
25458cbe | 987 | static void arc_prune_async(int64_t); |
9edb3695 | 988 | static uint64_t arc_all_memory(void); |
2a432414 | 989 | |
b9541d6b CW |
990 | static arc_buf_contents_t arc_buf_type(arc_buf_hdr_t *); |
991 | static uint32_t arc_bufc_to_flags(arc_buf_contents_t); | |
d3c2ae1c GW |
992 | static inline void arc_hdr_set_flags(arc_buf_hdr_t *hdr, arc_flags_t flags); |
993 | static inline void arc_hdr_clear_flags(arc_buf_hdr_t *hdr, arc_flags_t flags); | |
b9541d6b | 994 | |
2a432414 GW |
995 | static boolean_t l2arc_write_eligible(uint64_t, arc_buf_hdr_t *); |
996 | static void l2arc_read_done(zio_t *); | |
34dc7c2f BB |
997 | |
998 | static uint64_t | |
d164b209 | 999 | buf_hash(uint64_t spa, const dva_t *dva, uint64_t birth) |
34dc7c2f | 1000 | { |
34dc7c2f BB |
1001 | uint8_t *vdva = (uint8_t *)dva; |
1002 | uint64_t crc = -1ULL; | |
1003 | int i; | |
1004 | ||
1005 | ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); | |
1006 | ||
1007 | for (i = 0; i < sizeof (dva_t); i++) | |
1008 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ vdva[i]) & 0xFF]; | |
1009 | ||
d164b209 | 1010 | crc ^= (spa>>8) ^ birth; |
34dc7c2f BB |
1011 | |
1012 | return (crc); | |
1013 | } | |
1014 | ||
d3c2ae1c GW |
1015 | #define HDR_EMPTY(hdr) \ |
1016 | ((hdr)->b_dva.dva_word[0] == 0 && \ | |
1017 | (hdr)->b_dva.dva_word[1] == 0) | |
34dc7c2f | 1018 | |
d3c2ae1c GW |
1019 | #define HDR_EQUAL(spa, dva, birth, hdr) \ |
1020 | ((hdr)->b_dva.dva_word[0] == (dva)->dva_word[0]) && \ | |
1021 | ((hdr)->b_dva.dva_word[1] == (dva)->dva_word[1]) && \ | |
1022 | ((hdr)->b_birth == birth) && ((hdr)->b_spa == spa) | |
34dc7c2f | 1023 | |
428870ff BB |
1024 | static void |
1025 | buf_discard_identity(arc_buf_hdr_t *hdr) | |
1026 | { | |
1027 | hdr->b_dva.dva_word[0] = 0; | |
1028 | hdr->b_dva.dva_word[1] = 0; | |
1029 | hdr->b_birth = 0; | |
428870ff BB |
1030 | } |
1031 | ||
34dc7c2f | 1032 | static arc_buf_hdr_t * |
9b67f605 | 1033 | buf_hash_find(uint64_t spa, const blkptr_t *bp, kmutex_t **lockp) |
34dc7c2f | 1034 | { |
9b67f605 MA |
1035 | const dva_t *dva = BP_IDENTITY(bp); |
1036 | uint64_t birth = BP_PHYSICAL_BIRTH(bp); | |
34dc7c2f BB |
1037 | uint64_t idx = BUF_HASH_INDEX(spa, dva, birth); |
1038 | kmutex_t *hash_lock = BUF_HASH_LOCK(idx); | |
2a432414 | 1039 | arc_buf_hdr_t *hdr; |
34dc7c2f BB |
1040 | |
1041 | mutex_enter(hash_lock); | |
2a432414 GW |
1042 | for (hdr = buf_hash_table.ht_table[idx]; hdr != NULL; |
1043 | hdr = hdr->b_hash_next) { | |
d3c2ae1c | 1044 | if (HDR_EQUAL(spa, dva, birth, hdr)) { |
34dc7c2f | 1045 | *lockp = hash_lock; |
2a432414 | 1046 | return (hdr); |
34dc7c2f BB |
1047 | } |
1048 | } | |
1049 | mutex_exit(hash_lock); | |
1050 | *lockp = NULL; | |
1051 | return (NULL); | |
1052 | } | |
1053 | ||
1054 | /* | |
1055 | * Insert an entry into the hash table. If there is already an element | |
1056 | * equal to elem in the hash table, then the already existing element | |
1057 | * will be returned and the new element will not be inserted. | |
1058 | * Otherwise returns NULL. | |
b9541d6b | 1059 | * If lockp == NULL, the caller is assumed to already hold the hash lock. |
34dc7c2f BB |
1060 | */ |
1061 | static arc_buf_hdr_t * | |
2a432414 | 1062 | buf_hash_insert(arc_buf_hdr_t *hdr, kmutex_t **lockp) |
34dc7c2f | 1063 | { |
2a432414 | 1064 | uint64_t idx = BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth); |
34dc7c2f | 1065 | kmutex_t *hash_lock = BUF_HASH_LOCK(idx); |
2a432414 | 1066 | arc_buf_hdr_t *fhdr; |
34dc7c2f BB |
1067 | uint32_t i; |
1068 | ||
2a432414 GW |
1069 | ASSERT(!DVA_IS_EMPTY(&hdr->b_dva)); |
1070 | ASSERT(hdr->b_birth != 0); | |
1071 | ASSERT(!HDR_IN_HASH_TABLE(hdr)); | |
b9541d6b CW |
1072 | |
1073 | if (lockp != NULL) { | |
1074 | *lockp = hash_lock; | |
1075 | mutex_enter(hash_lock); | |
1076 | } else { | |
1077 | ASSERT(MUTEX_HELD(hash_lock)); | |
1078 | } | |
1079 | ||
2a432414 GW |
1080 | for (fhdr = buf_hash_table.ht_table[idx], i = 0; fhdr != NULL; |
1081 | fhdr = fhdr->b_hash_next, i++) { | |
d3c2ae1c | 1082 | if (HDR_EQUAL(hdr->b_spa, &hdr->b_dva, hdr->b_birth, fhdr)) |
2a432414 | 1083 | return (fhdr); |
34dc7c2f BB |
1084 | } |
1085 | ||
2a432414 GW |
1086 | hdr->b_hash_next = buf_hash_table.ht_table[idx]; |
1087 | buf_hash_table.ht_table[idx] = hdr; | |
d3c2ae1c | 1088 | arc_hdr_set_flags(hdr, ARC_FLAG_IN_HASH_TABLE); |
34dc7c2f BB |
1089 | |
1090 | /* collect some hash table performance data */ | |
1091 | if (i > 0) { | |
1092 | ARCSTAT_BUMP(arcstat_hash_collisions); | |
1093 | if (i == 1) | |
1094 | ARCSTAT_BUMP(arcstat_hash_chains); | |
1095 | ||
1096 | ARCSTAT_MAX(arcstat_hash_chain_max, i); | |
1097 | } | |
1098 | ||
1099 | ARCSTAT_BUMP(arcstat_hash_elements); | |
1100 | ARCSTAT_MAXSTAT(arcstat_hash_elements); | |
1101 | ||
1102 | return (NULL); | |
1103 | } | |
1104 | ||
1105 | static void | |
2a432414 | 1106 | buf_hash_remove(arc_buf_hdr_t *hdr) |
34dc7c2f | 1107 | { |
2a432414 GW |
1108 | arc_buf_hdr_t *fhdr, **hdrp; |
1109 | uint64_t idx = BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth); | |
34dc7c2f BB |
1110 | |
1111 | ASSERT(MUTEX_HELD(BUF_HASH_LOCK(idx))); | |
2a432414 | 1112 | ASSERT(HDR_IN_HASH_TABLE(hdr)); |
34dc7c2f | 1113 | |
2a432414 GW |
1114 | hdrp = &buf_hash_table.ht_table[idx]; |
1115 | while ((fhdr = *hdrp) != hdr) { | |
d3c2ae1c | 1116 | ASSERT3P(fhdr, !=, NULL); |
2a432414 | 1117 | hdrp = &fhdr->b_hash_next; |
34dc7c2f | 1118 | } |
2a432414 GW |
1119 | *hdrp = hdr->b_hash_next; |
1120 | hdr->b_hash_next = NULL; | |
d3c2ae1c | 1121 | arc_hdr_clear_flags(hdr, ARC_FLAG_IN_HASH_TABLE); |
34dc7c2f BB |
1122 | |
1123 | /* collect some hash table performance data */ | |
1124 | ARCSTAT_BUMPDOWN(arcstat_hash_elements); | |
1125 | ||
1126 | if (buf_hash_table.ht_table[idx] && | |
1127 | buf_hash_table.ht_table[idx]->b_hash_next == NULL) | |
1128 | ARCSTAT_BUMPDOWN(arcstat_hash_chains); | |
1129 | } | |
1130 | ||
1131 | /* | |
1132 | * Global data structures and functions for the buf kmem cache. | |
1133 | */ | |
b9541d6b CW |
1134 | static kmem_cache_t *hdr_full_cache; |
1135 | static kmem_cache_t *hdr_l2only_cache; | |
34dc7c2f BB |
1136 | static kmem_cache_t *buf_cache; |
1137 | ||
1138 | static void | |
1139 | buf_fini(void) | |
1140 | { | |
1141 | int i; | |
1142 | ||
00b46022 | 1143 | #if defined(_KERNEL) && defined(HAVE_SPL) |
d1d7e268 MK |
1144 | /* |
1145 | * Large allocations which do not require contiguous pages | |
1146 | * should be using vmem_free() in the linux kernel\ | |
1147 | */ | |
00b46022 BB |
1148 | vmem_free(buf_hash_table.ht_table, |
1149 | (buf_hash_table.ht_mask + 1) * sizeof (void *)); | |
1150 | #else | |
34dc7c2f BB |
1151 | kmem_free(buf_hash_table.ht_table, |
1152 | (buf_hash_table.ht_mask + 1) * sizeof (void *)); | |
00b46022 | 1153 | #endif |
34dc7c2f BB |
1154 | for (i = 0; i < BUF_LOCKS; i++) |
1155 | mutex_destroy(&buf_hash_table.ht_locks[i].ht_lock); | |
b9541d6b CW |
1156 | kmem_cache_destroy(hdr_full_cache); |
1157 | kmem_cache_destroy(hdr_l2only_cache); | |
34dc7c2f BB |
1158 | kmem_cache_destroy(buf_cache); |
1159 | } | |
1160 | ||
1161 | /* | |
1162 | * Constructor callback - called when the cache is empty | |
1163 | * and a new buf is requested. | |
1164 | */ | |
1165 | /* ARGSUSED */ | |
1166 | static int | |
b9541d6b CW |
1167 | hdr_full_cons(void *vbuf, void *unused, int kmflag) |
1168 | { | |
1169 | arc_buf_hdr_t *hdr = vbuf; | |
1170 | ||
1171 | bzero(hdr, HDR_FULL_SIZE); | |
1172 | cv_init(&hdr->b_l1hdr.b_cv, NULL, CV_DEFAULT, NULL); | |
1173 | refcount_create(&hdr->b_l1hdr.b_refcnt); | |
1174 | mutex_init(&hdr->b_l1hdr.b_freeze_lock, NULL, MUTEX_DEFAULT, NULL); | |
1175 | list_link_init(&hdr->b_l1hdr.b_arc_node); | |
1176 | list_link_init(&hdr->b_l2hdr.b_l2node); | |
ca0bf58d | 1177 | multilist_link_init(&hdr->b_l1hdr.b_arc_node); |
b9541d6b CW |
1178 | arc_space_consume(HDR_FULL_SIZE, ARC_SPACE_HDRS); |
1179 | ||
1180 | return (0); | |
1181 | } | |
1182 | ||
1183 | /* ARGSUSED */ | |
1184 | static int | |
1185 | hdr_l2only_cons(void *vbuf, void *unused, int kmflag) | |
34dc7c2f | 1186 | { |
2a432414 GW |
1187 | arc_buf_hdr_t *hdr = vbuf; |
1188 | ||
b9541d6b CW |
1189 | bzero(hdr, HDR_L2ONLY_SIZE); |
1190 | arc_space_consume(HDR_L2ONLY_SIZE, ARC_SPACE_L2HDRS); | |
34dc7c2f | 1191 | |
34dc7c2f BB |
1192 | return (0); |
1193 | } | |
1194 | ||
b128c09f BB |
1195 | /* ARGSUSED */ |
1196 | static int | |
1197 | buf_cons(void *vbuf, void *unused, int kmflag) | |
1198 | { | |
1199 | arc_buf_t *buf = vbuf; | |
1200 | ||
1201 | bzero(buf, sizeof (arc_buf_t)); | |
428870ff | 1202 | mutex_init(&buf->b_evict_lock, NULL, MUTEX_DEFAULT, NULL); |
d164b209 BB |
1203 | arc_space_consume(sizeof (arc_buf_t), ARC_SPACE_HDRS); |
1204 | ||
b128c09f BB |
1205 | return (0); |
1206 | } | |
1207 | ||
34dc7c2f BB |
1208 | /* |
1209 | * Destructor callback - called when a cached buf is | |
1210 | * no longer required. | |
1211 | */ | |
1212 | /* ARGSUSED */ | |
1213 | static void | |
b9541d6b | 1214 | hdr_full_dest(void *vbuf, void *unused) |
34dc7c2f | 1215 | { |
2a432414 | 1216 | arc_buf_hdr_t *hdr = vbuf; |
34dc7c2f | 1217 | |
d3c2ae1c | 1218 | ASSERT(HDR_EMPTY(hdr)); |
b9541d6b CW |
1219 | cv_destroy(&hdr->b_l1hdr.b_cv); |
1220 | refcount_destroy(&hdr->b_l1hdr.b_refcnt); | |
1221 | mutex_destroy(&hdr->b_l1hdr.b_freeze_lock); | |
ca0bf58d | 1222 | ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); |
b9541d6b CW |
1223 | arc_space_return(HDR_FULL_SIZE, ARC_SPACE_HDRS); |
1224 | } | |
1225 | ||
1226 | /* ARGSUSED */ | |
1227 | static void | |
1228 | hdr_l2only_dest(void *vbuf, void *unused) | |
1229 | { | |
1230 | ASSERTV(arc_buf_hdr_t *hdr = vbuf); | |
1231 | ||
d3c2ae1c | 1232 | ASSERT(HDR_EMPTY(hdr)); |
b9541d6b | 1233 | arc_space_return(HDR_L2ONLY_SIZE, ARC_SPACE_L2HDRS); |
34dc7c2f BB |
1234 | } |
1235 | ||
b128c09f BB |
1236 | /* ARGSUSED */ |
1237 | static void | |
1238 | buf_dest(void *vbuf, void *unused) | |
1239 | { | |
1240 | arc_buf_t *buf = vbuf; | |
1241 | ||
428870ff | 1242 | mutex_destroy(&buf->b_evict_lock); |
d164b209 | 1243 | arc_space_return(sizeof (arc_buf_t), ARC_SPACE_HDRS); |
b128c09f BB |
1244 | } |
1245 | ||
8c8af9d8 BB |
1246 | /* |
1247 | * Reclaim callback -- invoked when memory is low. | |
1248 | */ | |
1249 | /* ARGSUSED */ | |
1250 | static void | |
1251 | hdr_recl(void *unused) | |
1252 | { | |
1253 | dprintf("hdr_recl called\n"); | |
1254 | /* | |
1255 | * umem calls the reclaim func when we destroy the buf cache, | |
1256 | * which is after we do arc_fini(). | |
1257 | */ | |
1258 | if (!arc_dead) | |
1259 | cv_signal(&arc_reclaim_thread_cv); | |
1260 | } | |
1261 | ||
34dc7c2f BB |
1262 | static void |
1263 | buf_init(void) | |
1264 | { | |
2db28197 | 1265 | uint64_t *ct = NULL; |
34dc7c2f BB |
1266 | uint64_t hsize = 1ULL << 12; |
1267 | int i, j; | |
1268 | ||
1269 | /* | |
1270 | * The hash table is big enough to fill all of physical memory | |
49ddb315 MA |
1271 | * with an average block size of zfs_arc_average_blocksize (default 8K). |
1272 | * By default, the table will take up | |
1273 | * totalmem * sizeof(void*) / 8K (1MB per GB with 8-byte pointers). | |
34dc7c2f | 1274 | */ |
9edb3695 | 1275 | while (hsize * zfs_arc_average_blocksize < arc_all_memory()) |
34dc7c2f BB |
1276 | hsize <<= 1; |
1277 | retry: | |
1278 | buf_hash_table.ht_mask = hsize - 1; | |
00b46022 | 1279 | #if defined(_KERNEL) && defined(HAVE_SPL) |
d1d7e268 MK |
1280 | /* |
1281 | * Large allocations which do not require contiguous pages | |
1282 | * should be using vmem_alloc() in the linux kernel | |
1283 | */ | |
00b46022 BB |
1284 | buf_hash_table.ht_table = |
1285 | vmem_zalloc(hsize * sizeof (void*), KM_SLEEP); | |
1286 | #else | |
34dc7c2f BB |
1287 | buf_hash_table.ht_table = |
1288 | kmem_zalloc(hsize * sizeof (void*), KM_NOSLEEP); | |
00b46022 | 1289 | #endif |
34dc7c2f BB |
1290 | if (buf_hash_table.ht_table == NULL) { |
1291 | ASSERT(hsize > (1ULL << 8)); | |
1292 | hsize >>= 1; | |
1293 | goto retry; | |
1294 | } | |
1295 | ||
b9541d6b | 1296 | hdr_full_cache = kmem_cache_create("arc_buf_hdr_t_full", HDR_FULL_SIZE, |
8c8af9d8 | 1297 | 0, hdr_full_cons, hdr_full_dest, hdr_recl, NULL, NULL, 0); |
b9541d6b | 1298 | hdr_l2only_cache = kmem_cache_create("arc_buf_hdr_t_l2only", |
8c8af9d8 | 1299 | HDR_L2ONLY_SIZE, 0, hdr_l2only_cons, hdr_l2only_dest, hdr_recl, |
b9541d6b | 1300 | NULL, NULL, 0); |
34dc7c2f | 1301 | buf_cache = kmem_cache_create("arc_buf_t", sizeof (arc_buf_t), |
b128c09f | 1302 | 0, buf_cons, buf_dest, NULL, NULL, NULL, 0); |
34dc7c2f BB |
1303 | |
1304 | for (i = 0; i < 256; i++) | |
1305 | for (ct = zfs_crc64_table + i, *ct = i, j = 8; j > 0; j--) | |
1306 | *ct = (*ct >> 1) ^ (-(*ct & 1) & ZFS_CRC64_POLY); | |
1307 | ||
1308 | for (i = 0; i < BUF_LOCKS; i++) { | |
1309 | mutex_init(&buf_hash_table.ht_locks[i].ht_lock, | |
40d06e3c | 1310 | NULL, MUTEX_DEFAULT, NULL); |
34dc7c2f BB |
1311 | } |
1312 | } | |
1313 | ||
d3c2ae1c | 1314 | #define ARC_MINTIME (hz>>4) /* 62 ms */ |
ca0bf58d | 1315 | |
2aa34383 DK |
1316 | /* |
1317 | * This is the size that the buf occupies in memory. If the buf is compressed, | |
1318 | * it will correspond to the compressed size. You should use this method of | |
1319 | * getting the buf size unless you explicitly need the logical size. | |
1320 | */ | |
1321 | uint64_t | |
1322 | arc_buf_size(arc_buf_t *buf) | |
1323 | { | |
1324 | return (ARC_BUF_COMPRESSED(buf) ? | |
1325 | HDR_GET_PSIZE(buf->b_hdr) : HDR_GET_LSIZE(buf->b_hdr)); | |
1326 | } | |
1327 | ||
1328 | uint64_t | |
1329 | arc_buf_lsize(arc_buf_t *buf) | |
1330 | { | |
1331 | return (HDR_GET_LSIZE(buf->b_hdr)); | |
1332 | } | |
1333 | ||
1334 | enum zio_compress | |
1335 | arc_get_compression(arc_buf_t *buf) | |
1336 | { | |
1337 | return (ARC_BUF_COMPRESSED(buf) ? | |
1338 | HDR_GET_COMPRESS(buf->b_hdr) : ZIO_COMPRESS_OFF); | |
1339 | } | |
1340 | ||
d3c2ae1c GW |
1341 | static inline boolean_t |
1342 | arc_buf_is_shared(arc_buf_t *buf) | |
1343 | { | |
1344 | boolean_t shared = (buf->b_data != NULL && | |
a6255b7f DQ |
1345 | buf->b_hdr->b_l1hdr.b_pabd != NULL && |
1346 | abd_is_linear(buf->b_hdr->b_l1hdr.b_pabd) && | |
1347 | buf->b_data == abd_to_buf(buf->b_hdr->b_l1hdr.b_pabd)); | |
d3c2ae1c | 1348 | IMPLY(shared, HDR_SHARED_DATA(buf->b_hdr)); |
2aa34383 DK |
1349 | IMPLY(shared, ARC_BUF_SHARED(buf)); |
1350 | IMPLY(shared, ARC_BUF_COMPRESSED(buf) || ARC_BUF_LAST(buf)); | |
524b4217 DK |
1351 | |
1352 | /* | |
1353 | * It would be nice to assert arc_can_share() too, but the "hdr isn't | |
1354 | * already being shared" requirement prevents us from doing that. | |
1355 | */ | |
1356 | ||
d3c2ae1c GW |
1357 | return (shared); |
1358 | } | |
ca0bf58d | 1359 | |
d3c2ae1c GW |
1360 | static inline void |
1361 | arc_cksum_free(arc_buf_hdr_t *hdr) | |
1362 | { | |
1363 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
1364 | mutex_enter(&hdr->b_l1hdr.b_freeze_lock); | |
1365 | if (hdr->b_l1hdr.b_freeze_cksum != NULL) { | |
1366 | kmem_free(hdr->b_l1hdr.b_freeze_cksum, sizeof (zio_cksum_t)); | |
1367 | hdr->b_l1hdr.b_freeze_cksum = NULL; | |
b9541d6b | 1368 | } |
d3c2ae1c | 1369 | mutex_exit(&hdr->b_l1hdr.b_freeze_lock); |
b9541d6b CW |
1370 | } |
1371 | ||
524b4217 DK |
1372 | /* |
1373 | * If we've turned on the ZFS_DEBUG_MODIFY flag, verify that the buf's data | |
1374 | * matches the checksum that is stored in the hdr. If there is no checksum, | |
1375 | * or if the buf is compressed, this is a no-op. | |
1376 | */ | |
34dc7c2f BB |
1377 | static void |
1378 | arc_cksum_verify(arc_buf_t *buf) | |
1379 | { | |
d3c2ae1c | 1380 | arc_buf_hdr_t *hdr = buf->b_hdr; |
34dc7c2f BB |
1381 | zio_cksum_t zc; |
1382 | ||
1383 | if (!(zfs_flags & ZFS_DEBUG_MODIFY)) | |
1384 | return; | |
1385 | ||
524b4217 | 1386 | if (ARC_BUF_COMPRESSED(buf)) { |
524b4217 DK |
1387 | return; |
1388 | } | |
1389 | ||
d3c2ae1c GW |
1390 | ASSERT(HDR_HAS_L1HDR(hdr)); |
1391 | ||
1392 | mutex_enter(&hdr->b_l1hdr.b_freeze_lock); | |
1393 | if (hdr->b_l1hdr.b_freeze_cksum == NULL || HDR_IO_ERROR(hdr)) { | |
1394 | mutex_exit(&hdr->b_l1hdr.b_freeze_lock); | |
34dc7c2f BB |
1395 | return; |
1396 | } | |
2aa34383 | 1397 | |
3c67d83a | 1398 | fletcher_2_native(buf->b_data, arc_buf_size(buf), NULL, &zc); |
d3c2ae1c | 1399 | if (!ZIO_CHECKSUM_EQUAL(*hdr->b_l1hdr.b_freeze_cksum, zc)) |
34dc7c2f | 1400 | panic("buffer modified while frozen!"); |
d3c2ae1c | 1401 | mutex_exit(&hdr->b_l1hdr.b_freeze_lock); |
34dc7c2f BB |
1402 | } |
1403 | ||
d3c2ae1c GW |
1404 | static boolean_t |
1405 | arc_cksum_is_equal(arc_buf_hdr_t *hdr, zio_t *zio) | |
34dc7c2f | 1406 | { |
d3c2ae1c GW |
1407 | enum zio_compress compress = BP_GET_COMPRESS(zio->io_bp); |
1408 | boolean_t valid_cksum; | |
34dc7c2f | 1409 | |
d3c2ae1c GW |
1410 | ASSERT(!BP_IS_EMBEDDED(zio->io_bp)); |
1411 | VERIFY3U(BP_GET_PSIZE(zio->io_bp), ==, HDR_GET_PSIZE(hdr)); | |
34dc7c2f | 1412 | |
d3c2ae1c GW |
1413 | /* |
1414 | * We rely on the blkptr's checksum to determine if the block | |
1415 | * is valid or not. When compressed arc is enabled, the l2arc | |
1416 | * writes the block to the l2arc just as it appears in the pool. | |
1417 | * This allows us to use the blkptr's checksum to validate the | |
1418 | * data that we just read off of the l2arc without having to store | |
1419 | * a separate checksum in the arc_buf_hdr_t. However, if compressed | |
1420 | * arc is disabled, then the data written to the l2arc is always | |
1421 | * uncompressed and won't match the block as it exists in the main | |
1422 | * pool. When this is the case, we must first compress it if it is | |
1423 | * compressed on the main pool before we can validate the checksum. | |
1424 | */ | |
1425 | if (!HDR_COMPRESSION_ENABLED(hdr) && compress != ZIO_COMPRESS_OFF) { | |
1426 | uint64_t lsize; | |
1427 | uint64_t csize; | |
1428 | void *cbuf; | |
1429 | ASSERT3U(HDR_GET_COMPRESS(hdr), ==, ZIO_COMPRESS_OFF); | |
1430 | ||
1431 | cbuf = zio_buf_alloc(HDR_GET_PSIZE(hdr)); | |
1432 | lsize = HDR_GET_LSIZE(hdr); | |
a6255b7f DQ |
1433 | csize = zio_compress_data(compress, zio->io_abd, cbuf, lsize); |
1434 | ||
d3c2ae1c GW |
1435 | ASSERT3U(csize, <=, HDR_GET_PSIZE(hdr)); |
1436 | if (csize < HDR_GET_PSIZE(hdr)) { | |
1437 | /* | |
1438 | * Compressed blocks are always a multiple of the | |
1439 | * smallest ashift in the pool. Ideally, we would | |
1440 | * like to round up the csize to the next | |
1441 | * spa_min_ashift but that value may have changed | |
1442 | * since the block was last written. Instead, | |
1443 | * we rely on the fact that the hdr's psize | |
1444 | * was set to the psize of the block when it was | |
1445 | * last written. We set the csize to that value | |
1446 | * and zero out any part that should not contain | |
1447 | * data. | |
1448 | */ | |
1449 | bzero((char *)cbuf + csize, HDR_GET_PSIZE(hdr) - csize); | |
1450 | csize = HDR_GET_PSIZE(hdr); | |
1451 | } | |
1452 | zio_push_transform(zio, cbuf, csize, HDR_GET_PSIZE(hdr), NULL); | |
1453 | } | |
1454 | ||
1455 | /* | |
1456 | * Block pointers always store the checksum for the logical data. | |
1457 | * If the block pointer has the gang bit set, then the checksum | |
1458 | * it represents is for the reconstituted data and not for an | |
1459 | * individual gang member. The zio pipeline, however, must be able to | |
1460 | * determine the checksum of each of the gang constituents so it | |
1461 | * treats the checksum comparison differently than what we need | |
1462 | * for l2arc blocks. This prevents us from using the | |
1463 | * zio_checksum_error() interface directly. Instead we must call the | |
1464 | * zio_checksum_error_impl() so that we can ensure the checksum is | |
1465 | * generated using the correct checksum algorithm and accounts for the | |
1466 | * logical I/O size and not just a gang fragment. | |
1467 | */ | |
1468 | valid_cksum = (zio_checksum_error_impl(zio->io_spa, zio->io_bp, | |
a6255b7f | 1469 | BP_GET_CHECKSUM(zio->io_bp), zio->io_abd, zio->io_size, |
d3c2ae1c GW |
1470 | zio->io_offset, NULL) == 0); |
1471 | zio_pop_transforms(zio); | |
1472 | return (valid_cksum); | |
34dc7c2f BB |
1473 | } |
1474 | ||
524b4217 DK |
1475 | /* |
1476 | * Given a buf full of data, if ZFS_DEBUG_MODIFY is enabled this computes a | |
1477 | * checksum and attaches it to the buf's hdr so that we can ensure that the buf | |
1478 | * isn't modified later on. If buf is compressed or there is already a checksum | |
1479 | * on the hdr, this is a no-op (we only checksum uncompressed bufs). | |
1480 | */ | |
34dc7c2f | 1481 | static void |
d3c2ae1c | 1482 | arc_cksum_compute(arc_buf_t *buf) |
34dc7c2f | 1483 | { |
d3c2ae1c GW |
1484 | arc_buf_hdr_t *hdr = buf->b_hdr; |
1485 | ||
1486 | if (!(zfs_flags & ZFS_DEBUG_MODIFY)) | |
34dc7c2f BB |
1487 | return; |
1488 | ||
d3c2ae1c | 1489 | ASSERT(HDR_HAS_L1HDR(hdr)); |
2aa34383 | 1490 | |
b9541d6b | 1491 | mutex_enter(&buf->b_hdr->b_l1hdr.b_freeze_lock); |
d3c2ae1c | 1492 | if (hdr->b_l1hdr.b_freeze_cksum != NULL) { |
2aa34383 DK |
1493 | mutex_exit(&hdr->b_l1hdr.b_freeze_lock); |
1494 | return; | |
1495 | } else if (ARC_BUF_COMPRESSED(buf)) { | |
d3c2ae1c | 1496 | mutex_exit(&hdr->b_l1hdr.b_freeze_lock); |
34dc7c2f BB |
1497 | return; |
1498 | } | |
2aa34383 DK |
1499 | |
1500 | ASSERT(!ARC_BUF_COMPRESSED(buf)); | |
d3c2ae1c GW |
1501 | hdr->b_l1hdr.b_freeze_cksum = kmem_alloc(sizeof (zio_cksum_t), |
1502 | KM_SLEEP); | |
3c67d83a | 1503 | fletcher_2_native(buf->b_data, arc_buf_size(buf), NULL, |
d3c2ae1c GW |
1504 | hdr->b_l1hdr.b_freeze_cksum); |
1505 | mutex_exit(&hdr->b_l1hdr.b_freeze_lock); | |
498877ba MA |
1506 | arc_buf_watch(buf); |
1507 | } | |
1508 | ||
1509 | #ifndef _KERNEL | |
1510 | void | |
1511 | arc_buf_sigsegv(int sig, siginfo_t *si, void *unused) | |
1512 | { | |
02730c33 | 1513 | panic("Got SIGSEGV at address: 0x%lx\n", (long)si->si_addr); |
498877ba MA |
1514 | } |
1515 | #endif | |
1516 | ||
1517 | /* ARGSUSED */ | |
1518 | static void | |
1519 | arc_buf_unwatch(arc_buf_t *buf) | |
1520 | { | |
1521 | #ifndef _KERNEL | |
1522 | if (arc_watch) { | |
d3c2ae1c | 1523 | ASSERT0(mprotect(buf->b_data, HDR_GET_LSIZE(buf->b_hdr), |
498877ba MA |
1524 | PROT_READ | PROT_WRITE)); |
1525 | } | |
1526 | #endif | |
1527 | } | |
1528 | ||
1529 | /* ARGSUSED */ | |
1530 | static void | |
1531 | arc_buf_watch(arc_buf_t *buf) | |
1532 | { | |
1533 | #ifndef _KERNEL | |
1534 | if (arc_watch) | |
2aa34383 | 1535 | ASSERT0(mprotect(buf->b_data, arc_buf_size(buf), |
d3c2ae1c | 1536 | PROT_READ)); |
498877ba | 1537 | #endif |
34dc7c2f BB |
1538 | } |
1539 | ||
b9541d6b CW |
1540 | static arc_buf_contents_t |
1541 | arc_buf_type(arc_buf_hdr_t *hdr) | |
1542 | { | |
d3c2ae1c | 1543 | arc_buf_contents_t type; |
b9541d6b | 1544 | if (HDR_ISTYPE_METADATA(hdr)) { |
d3c2ae1c | 1545 | type = ARC_BUFC_METADATA; |
b9541d6b | 1546 | } else { |
d3c2ae1c | 1547 | type = ARC_BUFC_DATA; |
b9541d6b | 1548 | } |
d3c2ae1c GW |
1549 | VERIFY3U(hdr->b_type, ==, type); |
1550 | return (type); | |
b9541d6b CW |
1551 | } |
1552 | ||
2aa34383 DK |
1553 | boolean_t |
1554 | arc_is_metadata(arc_buf_t *buf) | |
1555 | { | |
1556 | return (HDR_ISTYPE_METADATA(buf->b_hdr) != 0); | |
1557 | } | |
1558 | ||
b9541d6b CW |
1559 | static uint32_t |
1560 | arc_bufc_to_flags(arc_buf_contents_t type) | |
1561 | { | |
1562 | switch (type) { | |
1563 | case ARC_BUFC_DATA: | |
1564 | /* metadata field is 0 if buffer contains normal data */ | |
1565 | return (0); | |
1566 | case ARC_BUFC_METADATA: | |
1567 | return (ARC_FLAG_BUFC_METADATA); | |
1568 | default: | |
1569 | break; | |
1570 | } | |
1571 | panic("undefined ARC buffer type!"); | |
1572 | return ((uint32_t)-1); | |
1573 | } | |
1574 | ||
34dc7c2f BB |
1575 | void |
1576 | arc_buf_thaw(arc_buf_t *buf) | |
1577 | { | |
d3c2ae1c GW |
1578 | arc_buf_hdr_t *hdr = buf->b_hdr; |
1579 | ||
2aa34383 DK |
1580 | ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon); |
1581 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); | |
1582 | ||
524b4217 | 1583 | arc_cksum_verify(buf); |
34dc7c2f | 1584 | |
2aa34383 DK |
1585 | /* |
1586 | * Compressed buffers do not manipulate the b_freeze_cksum or | |
1587 | * allocate b_thawed. | |
1588 | */ | |
1589 | if (ARC_BUF_COMPRESSED(buf)) { | |
2aa34383 DK |
1590 | return; |
1591 | } | |
1592 | ||
d3c2ae1c GW |
1593 | ASSERT(HDR_HAS_L1HDR(hdr)); |
1594 | arc_cksum_free(hdr); | |
498877ba | 1595 | arc_buf_unwatch(buf); |
34dc7c2f BB |
1596 | } |
1597 | ||
1598 | void | |
1599 | arc_buf_freeze(arc_buf_t *buf) | |
1600 | { | |
d3c2ae1c | 1601 | arc_buf_hdr_t *hdr = buf->b_hdr; |
428870ff BB |
1602 | kmutex_t *hash_lock; |
1603 | ||
34dc7c2f BB |
1604 | if (!(zfs_flags & ZFS_DEBUG_MODIFY)) |
1605 | return; | |
1606 | ||
2aa34383 | 1607 | if (ARC_BUF_COMPRESSED(buf)) { |
2aa34383 DK |
1608 | return; |
1609 | } | |
1610 | ||
d3c2ae1c | 1611 | hash_lock = HDR_LOCK(hdr); |
428870ff BB |
1612 | mutex_enter(hash_lock); |
1613 | ||
d3c2ae1c GW |
1614 | ASSERT(HDR_HAS_L1HDR(hdr)); |
1615 | ASSERT(hdr->b_l1hdr.b_freeze_cksum != NULL || | |
1616 | hdr->b_l1hdr.b_state == arc_anon); | |
1617 | arc_cksum_compute(buf); | |
428870ff | 1618 | mutex_exit(hash_lock); |
34dc7c2f BB |
1619 | } |
1620 | ||
d3c2ae1c GW |
1621 | /* |
1622 | * The arc_buf_hdr_t's b_flags should never be modified directly. Instead, | |
1623 | * the following functions should be used to ensure that the flags are | |
1624 | * updated in a thread-safe way. When manipulating the flags either | |
1625 | * the hash_lock must be held or the hdr must be undiscoverable. This | |
1626 | * ensures that we're not racing with any other threads when updating | |
1627 | * the flags. | |
1628 | */ | |
1629 | static inline void | |
1630 | arc_hdr_set_flags(arc_buf_hdr_t *hdr, arc_flags_t flags) | |
1631 | { | |
1632 | ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); | |
1633 | hdr->b_flags |= flags; | |
1634 | } | |
1635 | ||
1636 | static inline void | |
1637 | arc_hdr_clear_flags(arc_buf_hdr_t *hdr, arc_flags_t flags) | |
1638 | { | |
1639 | ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); | |
1640 | hdr->b_flags &= ~flags; | |
1641 | } | |
1642 | ||
1643 | /* | |
1644 | * Setting the compression bits in the arc_buf_hdr_t's b_flags is | |
1645 | * done in a special way since we have to clear and set bits | |
1646 | * at the same time. Consumers that wish to set the compression bits | |
1647 | * must use this function to ensure that the flags are updated in | |
1648 | * thread-safe manner. | |
1649 | */ | |
1650 | static void | |
1651 | arc_hdr_set_compress(arc_buf_hdr_t *hdr, enum zio_compress cmp) | |
1652 | { | |
1653 | ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); | |
1654 | ||
1655 | /* | |
1656 | * Holes and embedded blocks will always have a psize = 0 so | |
1657 | * we ignore the compression of the blkptr and set the | |
d3c2ae1c GW |
1658 | * want to uncompress them. Mark them as uncompressed. |
1659 | */ | |
1660 | if (!zfs_compressed_arc_enabled || HDR_GET_PSIZE(hdr) == 0) { | |
1661 | arc_hdr_clear_flags(hdr, ARC_FLAG_COMPRESSED_ARC); | |
1662 | HDR_SET_COMPRESS(hdr, ZIO_COMPRESS_OFF); | |
1663 | ASSERT(!HDR_COMPRESSION_ENABLED(hdr)); | |
1664 | ASSERT3U(HDR_GET_COMPRESS(hdr), ==, ZIO_COMPRESS_OFF); | |
1665 | } else { | |
1666 | arc_hdr_set_flags(hdr, ARC_FLAG_COMPRESSED_ARC); | |
1667 | HDR_SET_COMPRESS(hdr, cmp); | |
1668 | ASSERT3U(HDR_GET_COMPRESS(hdr), ==, cmp); | |
1669 | ASSERT(HDR_COMPRESSION_ENABLED(hdr)); | |
1670 | } | |
1671 | } | |
1672 | ||
524b4217 DK |
1673 | /* |
1674 | * Looks for another buf on the same hdr which has the data decompressed, copies | |
1675 | * from it, and returns true. If no such buf exists, returns false. | |
1676 | */ | |
1677 | static boolean_t | |
1678 | arc_buf_try_copy_decompressed_data(arc_buf_t *buf) | |
1679 | { | |
1680 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
1681 | arc_buf_t *from; | |
1682 | boolean_t copied = B_FALSE; | |
1683 | ||
1684 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
1685 | ASSERT3P(buf->b_data, !=, NULL); | |
1686 | ASSERT(!ARC_BUF_COMPRESSED(buf)); | |
1687 | ||
1688 | for (from = hdr->b_l1hdr.b_buf; from != NULL; | |
1689 | from = from->b_next) { | |
1690 | /* can't use our own data buffer */ | |
1691 | if (from == buf) { | |
1692 | continue; | |
1693 | } | |
1694 | ||
1695 | if (!ARC_BUF_COMPRESSED(from)) { | |
1696 | bcopy(from->b_data, buf->b_data, arc_buf_size(buf)); | |
1697 | copied = B_TRUE; | |
1698 | break; | |
1699 | } | |
1700 | } | |
1701 | ||
1702 | /* | |
1703 | * There were no decompressed bufs, so there should not be a | |
1704 | * checksum on the hdr either. | |
1705 | */ | |
1706 | EQUIV(!copied, hdr->b_l1hdr.b_freeze_cksum == NULL); | |
1707 | ||
1708 | return (copied); | |
1709 | } | |
1710 | ||
1711 | /* | |
1712 | * Given a buf that has a data buffer attached to it, this function will | |
1713 | * efficiently fill the buf with data of the specified compression setting from | |
1714 | * the hdr and update the hdr's b_freeze_cksum if necessary. If the buf and hdr | |
1715 | * are already sharing a data buf, no copy is performed. | |
1716 | * | |
1717 | * If the buf is marked as compressed but uncompressed data was requested, this | |
1718 | * will allocate a new data buffer for the buf, remove that flag, and fill the | |
1719 | * buf with uncompressed data. You can't request a compressed buf on a hdr with | |
1720 | * uncompressed data, and (since we haven't added support for it yet) if you | |
1721 | * want compressed data your buf must already be marked as compressed and have | |
1722 | * the correct-sized data buffer. | |
1723 | */ | |
1724 | static int | |
1725 | arc_buf_fill(arc_buf_t *buf, boolean_t compressed) | |
d3c2ae1c GW |
1726 | { |
1727 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
524b4217 | 1728 | boolean_t hdr_compressed = (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF); |
d3c2ae1c | 1729 | dmu_object_byteswap_t bswap = hdr->b_l1hdr.b_byteswap; |
d3c2ae1c | 1730 | |
524b4217 DK |
1731 | ASSERT3P(buf->b_data, !=, NULL); |
1732 | IMPLY(compressed, hdr_compressed); | |
1733 | IMPLY(compressed, ARC_BUF_COMPRESSED(buf)); | |
1734 | ||
1735 | if (hdr_compressed == compressed) { | |
2aa34383 | 1736 | if (!arc_buf_is_shared(buf)) { |
a6255b7f | 1737 | abd_copy_to_buf(buf->b_data, hdr->b_l1hdr.b_pabd, |
524b4217 | 1738 | arc_buf_size(buf)); |
2aa34383 | 1739 | } |
d3c2ae1c | 1740 | } else { |
524b4217 DK |
1741 | ASSERT(hdr_compressed); |
1742 | ASSERT(!compressed); | |
d3c2ae1c | 1743 | ASSERT3U(HDR_GET_LSIZE(hdr), !=, HDR_GET_PSIZE(hdr)); |
2aa34383 DK |
1744 | |
1745 | /* | |
524b4217 DK |
1746 | * If the buf is sharing its data with the hdr, unlink it and |
1747 | * allocate a new data buffer for the buf. | |
2aa34383 | 1748 | */ |
524b4217 DK |
1749 | if (arc_buf_is_shared(buf)) { |
1750 | ASSERT(ARC_BUF_COMPRESSED(buf)); | |
1751 | ||
1752 | /* We need to give the buf it's own b_data */ | |
1753 | buf->b_flags &= ~ARC_BUF_FLAG_SHARED; | |
2aa34383 DK |
1754 | buf->b_data = |
1755 | arc_get_data_buf(hdr, HDR_GET_LSIZE(hdr), buf); | |
1756 | arc_hdr_clear_flags(hdr, ARC_FLAG_SHARED_DATA); | |
1757 | ||
524b4217 | 1758 | /* Previously overhead was 0; just add new overhead */ |
2aa34383 | 1759 | ARCSTAT_INCR(arcstat_overhead_size, HDR_GET_LSIZE(hdr)); |
524b4217 DK |
1760 | } else if (ARC_BUF_COMPRESSED(buf)) { |
1761 | /* We need to reallocate the buf's b_data */ | |
1762 | arc_free_data_buf(hdr, buf->b_data, HDR_GET_PSIZE(hdr), | |
1763 | buf); | |
1764 | buf->b_data = | |
1765 | arc_get_data_buf(hdr, HDR_GET_LSIZE(hdr), buf); | |
1766 | ||
1767 | /* We increased the size of b_data; update overhead */ | |
1768 | ARCSTAT_INCR(arcstat_overhead_size, | |
1769 | HDR_GET_LSIZE(hdr) - HDR_GET_PSIZE(hdr)); | |
2aa34383 DK |
1770 | } |
1771 | ||
524b4217 DK |
1772 | /* |
1773 | * Regardless of the buf's previous compression settings, it | |
1774 | * should not be compressed at the end of this function. | |
1775 | */ | |
1776 | buf->b_flags &= ~ARC_BUF_FLAG_COMPRESSED; | |
1777 | ||
1778 | /* | |
1779 | * Try copying the data from another buf which already has a | |
1780 | * decompressed version. If that's not possible, it's time to | |
1781 | * bite the bullet and decompress the data from the hdr. | |
1782 | */ | |
1783 | if (arc_buf_try_copy_decompressed_data(buf)) { | |
1784 | /* Skip byteswapping and checksumming (already done) */ | |
1785 | ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, !=, NULL); | |
1786 | return (0); | |
1787 | } else { | |
1788 | int error = zio_decompress_data(HDR_GET_COMPRESS(hdr), | |
a6255b7f | 1789 | hdr->b_l1hdr.b_pabd, buf->b_data, |
524b4217 DK |
1790 | HDR_GET_PSIZE(hdr), HDR_GET_LSIZE(hdr)); |
1791 | ||
1792 | /* | |
1793 | * Absent hardware errors or software bugs, this should | |
1794 | * be impossible, but log it anyway so we can debug it. | |
1795 | */ | |
1796 | if (error != 0) { | |
1797 | zfs_dbgmsg( | |
1798 | "hdr %p, compress %d, psize %d, lsize %d", | |
1799 | hdr, HDR_GET_COMPRESS(hdr), | |
1800 | HDR_GET_PSIZE(hdr), HDR_GET_LSIZE(hdr)); | |
1801 | return (SET_ERROR(EIO)); | |
1802 | } | |
d3c2ae1c GW |
1803 | } |
1804 | } | |
524b4217 DK |
1805 | |
1806 | /* Byteswap the buf's data if necessary */ | |
d3c2ae1c GW |
1807 | if (bswap != DMU_BSWAP_NUMFUNCS) { |
1808 | ASSERT(!HDR_SHARED_DATA(hdr)); | |
1809 | ASSERT3U(bswap, <, DMU_BSWAP_NUMFUNCS); | |
1810 | dmu_ot_byteswap[bswap].ob_func(buf->b_data, HDR_GET_LSIZE(hdr)); | |
1811 | } | |
524b4217 DK |
1812 | |
1813 | /* Compute the hdr's checksum if necessary */ | |
d3c2ae1c | 1814 | arc_cksum_compute(buf); |
524b4217 | 1815 | |
d3c2ae1c GW |
1816 | return (0); |
1817 | } | |
1818 | ||
524b4217 DK |
1819 | int |
1820 | arc_decompress(arc_buf_t *buf) | |
1821 | { | |
1822 | return (arc_buf_fill(buf, B_FALSE)); | |
1823 | } | |
1824 | ||
d3c2ae1c | 1825 | /* |
a6255b7f | 1826 | * Return the size of the block, b_pabd, that is stored in the arc_buf_hdr_t. |
d3c2ae1c GW |
1827 | */ |
1828 | static uint64_t | |
1829 | arc_hdr_size(arc_buf_hdr_t *hdr) | |
1830 | { | |
1831 | uint64_t size; | |
1832 | ||
1833 | if (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF && | |
1834 | HDR_GET_PSIZE(hdr) > 0) { | |
1835 | size = HDR_GET_PSIZE(hdr); | |
1836 | } else { | |
1837 | ASSERT3U(HDR_GET_LSIZE(hdr), !=, 0); | |
1838 | size = HDR_GET_LSIZE(hdr); | |
1839 | } | |
1840 | return (size); | |
1841 | } | |
1842 | ||
1843 | /* | |
1844 | * Increment the amount of evictable space in the arc_state_t's refcount. | |
1845 | * We account for the space used by the hdr and the arc buf individually | |
1846 | * so that we can add and remove them from the refcount individually. | |
1847 | */ | |
34dc7c2f | 1848 | static void |
d3c2ae1c GW |
1849 | arc_evictable_space_increment(arc_buf_hdr_t *hdr, arc_state_t *state) |
1850 | { | |
1851 | arc_buf_contents_t type = arc_buf_type(hdr); | |
d3c2ae1c GW |
1852 | arc_buf_t *buf; |
1853 | ||
1854 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
1855 | ||
1856 | if (GHOST_STATE(state)) { | |
1857 | ASSERT0(hdr->b_l1hdr.b_bufcnt); | |
1858 | ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); | |
a6255b7f | 1859 | ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); |
2aa34383 DK |
1860 | (void) refcount_add_many(&state->arcs_esize[type], |
1861 | HDR_GET_LSIZE(hdr), hdr); | |
d3c2ae1c GW |
1862 | return; |
1863 | } | |
1864 | ||
1865 | ASSERT(!GHOST_STATE(state)); | |
a6255b7f | 1866 | if (hdr->b_l1hdr.b_pabd != NULL) { |
d3c2ae1c GW |
1867 | (void) refcount_add_many(&state->arcs_esize[type], |
1868 | arc_hdr_size(hdr), hdr); | |
1869 | } | |
1870 | for (buf = hdr->b_l1hdr.b_buf; buf != NULL; buf = buf->b_next) { | |
2aa34383 | 1871 | if (arc_buf_is_shared(buf)) |
d3c2ae1c | 1872 | continue; |
2aa34383 DK |
1873 | (void) refcount_add_many(&state->arcs_esize[type], |
1874 | arc_buf_size(buf), buf); | |
d3c2ae1c GW |
1875 | } |
1876 | } | |
1877 | ||
1878 | /* | |
1879 | * Decrement the amount of evictable space in the arc_state_t's refcount. | |
1880 | * We account for the space used by the hdr and the arc buf individually | |
1881 | * so that we can add and remove them from the refcount individually. | |
1882 | */ | |
1883 | static void | |
2aa34383 | 1884 | arc_evictable_space_decrement(arc_buf_hdr_t *hdr, arc_state_t *state) |
d3c2ae1c GW |
1885 | { |
1886 | arc_buf_contents_t type = arc_buf_type(hdr); | |
d3c2ae1c GW |
1887 | arc_buf_t *buf; |
1888 | ||
1889 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
1890 | ||
1891 | if (GHOST_STATE(state)) { | |
1892 | ASSERT0(hdr->b_l1hdr.b_bufcnt); | |
1893 | ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); | |
a6255b7f | 1894 | ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); |
d3c2ae1c | 1895 | (void) refcount_remove_many(&state->arcs_esize[type], |
2aa34383 | 1896 | HDR_GET_LSIZE(hdr), hdr); |
d3c2ae1c GW |
1897 | return; |
1898 | } | |
1899 | ||
1900 | ASSERT(!GHOST_STATE(state)); | |
a6255b7f | 1901 | if (hdr->b_l1hdr.b_pabd != NULL) { |
d3c2ae1c GW |
1902 | (void) refcount_remove_many(&state->arcs_esize[type], |
1903 | arc_hdr_size(hdr), hdr); | |
1904 | } | |
1905 | for (buf = hdr->b_l1hdr.b_buf; buf != NULL; buf = buf->b_next) { | |
2aa34383 | 1906 | if (arc_buf_is_shared(buf)) |
d3c2ae1c | 1907 | continue; |
d3c2ae1c | 1908 | (void) refcount_remove_many(&state->arcs_esize[type], |
2aa34383 | 1909 | arc_buf_size(buf), buf); |
d3c2ae1c GW |
1910 | } |
1911 | } | |
1912 | ||
1913 | /* | |
1914 | * Add a reference to this hdr indicating that someone is actively | |
1915 | * referencing that memory. When the refcount transitions from 0 to 1, | |
1916 | * we remove it from the respective arc_state_t list to indicate that | |
1917 | * it is not evictable. | |
1918 | */ | |
1919 | static void | |
1920 | add_reference(arc_buf_hdr_t *hdr, void *tag) | |
34dc7c2f | 1921 | { |
b9541d6b CW |
1922 | arc_state_t *state; |
1923 | ||
1924 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
d3c2ae1c GW |
1925 | if (!MUTEX_HELD(HDR_LOCK(hdr))) { |
1926 | ASSERT(hdr->b_l1hdr.b_state == arc_anon); | |
1927 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); | |
1928 | ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); | |
1929 | } | |
34dc7c2f | 1930 | |
b9541d6b CW |
1931 | state = hdr->b_l1hdr.b_state; |
1932 | ||
1933 | if ((refcount_add(&hdr->b_l1hdr.b_refcnt, tag) == 1) && | |
1934 | (state != arc_anon)) { | |
1935 | /* We don't use the L2-only state list. */ | |
1936 | if (state != arc_l2c_only) { | |
d3c2ae1c GW |
1937 | multilist_remove(&state->arcs_list[arc_buf_type(hdr)], |
1938 | hdr); | |
2aa34383 | 1939 | arc_evictable_space_decrement(hdr, state); |
34dc7c2f | 1940 | } |
b128c09f | 1941 | /* remove the prefetch flag if we get a reference */ |
d3c2ae1c | 1942 | arc_hdr_clear_flags(hdr, ARC_FLAG_PREFETCH); |
34dc7c2f BB |
1943 | } |
1944 | } | |
1945 | ||
d3c2ae1c GW |
1946 | /* |
1947 | * Remove a reference from this hdr. When the reference transitions from | |
1948 | * 1 to 0 and we're not anonymous, then we add this hdr to the arc_state_t's | |
1949 | * list making it eligible for eviction. | |
1950 | */ | |
34dc7c2f | 1951 | static int |
2a432414 | 1952 | remove_reference(arc_buf_hdr_t *hdr, kmutex_t *hash_lock, void *tag) |
34dc7c2f BB |
1953 | { |
1954 | int cnt; | |
b9541d6b | 1955 | arc_state_t *state = hdr->b_l1hdr.b_state; |
34dc7c2f | 1956 | |
b9541d6b | 1957 | ASSERT(HDR_HAS_L1HDR(hdr)); |
34dc7c2f BB |
1958 | ASSERT(state == arc_anon || MUTEX_HELD(hash_lock)); |
1959 | ASSERT(!GHOST_STATE(state)); | |
1960 | ||
b9541d6b CW |
1961 | /* |
1962 | * arc_l2c_only counts as a ghost state so we don't need to explicitly | |
1963 | * check to prevent usage of the arc_l2c_only list. | |
1964 | */ | |
1965 | if (((cnt = refcount_remove(&hdr->b_l1hdr.b_refcnt, tag)) == 0) && | |
34dc7c2f | 1966 | (state != arc_anon)) { |
d3c2ae1c GW |
1967 | multilist_insert(&state->arcs_list[arc_buf_type(hdr)], hdr); |
1968 | ASSERT3U(hdr->b_l1hdr.b_bufcnt, >, 0); | |
1969 | arc_evictable_space_increment(hdr, state); | |
34dc7c2f BB |
1970 | } |
1971 | return (cnt); | |
1972 | } | |
1973 | ||
e0b0ca98 BB |
1974 | /* |
1975 | * Returns detailed information about a specific arc buffer. When the | |
1976 | * state_index argument is set the function will calculate the arc header | |
1977 | * list position for its arc state. Since this requires a linear traversal | |
1978 | * callers are strongly encourage not to do this. However, it can be helpful | |
1979 | * for targeted analysis so the functionality is provided. | |
1980 | */ | |
1981 | void | |
1982 | arc_buf_info(arc_buf_t *ab, arc_buf_info_t *abi, int state_index) | |
1983 | { | |
1984 | arc_buf_hdr_t *hdr = ab->b_hdr; | |
b9541d6b CW |
1985 | l1arc_buf_hdr_t *l1hdr = NULL; |
1986 | l2arc_buf_hdr_t *l2hdr = NULL; | |
1987 | arc_state_t *state = NULL; | |
1988 | ||
8887c7d7 TC |
1989 | memset(abi, 0, sizeof (arc_buf_info_t)); |
1990 | ||
1991 | if (hdr == NULL) | |
1992 | return; | |
1993 | ||
1994 | abi->abi_flags = hdr->b_flags; | |
1995 | ||
b9541d6b CW |
1996 | if (HDR_HAS_L1HDR(hdr)) { |
1997 | l1hdr = &hdr->b_l1hdr; | |
1998 | state = l1hdr->b_state; | |
1999 | } | |
2000 | if (HDR_HAS_L2HDR(hdr)) | |
2001 | l2hdr = &hdr->b_l2hdr; | |
e0b0ca98 | 2002 | |
b9541d6b | 2003 | if (l1hdr) { |
d3c2ae1c | 2004 | abi->abi_bufcnt = l1hdr->b_bufcnt; |
b9541d6b CW |
2005 | abi->abi_access = l1hdr->b_arc_access; |
2006 | abi->abi_mru_hits = l1hdr->b_mru_hits; | |
2007 | abi->abi_mru_ghost_hits = l1hdr->b_mru_ghost_hits; | |
2008 | abi->abi_mfu_hits = l1hdr->b_mfu_hits; | |
2009 | abi->abi_mfu_ghost_hits = l1hdr->b_mfu_ghost_hits; | |
2010 | abi->abi_holds = refcount_count(&l1hdr->b_refcnt); | |
2011 | } | |
2012 | ||
2013 | if (l2hdr) { | |
2014 | abi->abi_l2arc_dattr = l2hdr->b_daddr; | |
b9541d6b CW |
2015 | abi->abi_l2arc_hits = l2hdr->b_hits; |
2016 | } | |
2017 | ||
e0b0ca98 | 2018 | abi->abi_state_type = state ? state->arcs_state : ARC_STATE_ANON; |
b9541d6b | 2019 | abi->abi_state_contents = arc_buf_type(hdr); |
d3c2ae1c | 2020 | abi->abi_size = arc_hdr_size(hdr); |
e0b0ca98 BB |
2021 | } |
2022 | ||
34dc7c2f | 2023 | /* |
ca0bf58d | 2024 | * Move the supplied buffer to the indicated state. The hash lock |
34dc7c2f BB |
2025 | * for the buffer must be held by the caller. |
2026 | */ | |
2027 | static void | |
2a432414 GW |
2028 | arc_change_state(arc_state_t *new_state, arc_buf_hdr_t *hdr, |
2029 | kmutex_t *hash_lock) | |
34dc7c2f | 2030 | { |
b9541d6b CW |
2031 | arc_state_t *old_state; |
2032 | int64_t refcnt; | |
d3c2ae1c GW |
2033 | uint32_t bufcnt; |
2034 | boolean_t update_old, update_new; | |
b9541d6b CW |
2035 | arc_buf_contents_t buftype = arc_buf_type(hdr); |
2036 | ||
2037 | /* | |
2038 | * We almost always have an L1 hdr here, since we call arc_hdr_realloc() | |
2039 | * in arc_read() when bringing a buffer out of the L2ARC. However, the | |
2040 | * L1 hdr doesn't always exist when we change state to arc_anon before | |
2041 | * destroying a header, in which case reallocating to add the L1 hdr is | |
2042 | * pointless. | |
2043 | */ | |
2044 | if (HDR_HAS_L1HDR(hdr)) { | |
2045 | old_state = hdr->b_l1hdr.b_state; | |
2046 | refcnt = refcount_count(&hdr->b_l1hdr.b_refcnt); | |
d3c2ae1c | 2047 | bufcnt = hdr->b_l1hdr.b_bufcnt; |
a6255b7f | 2048 | update_old = (bufcnt > 0 || hdr->b_l1hdr.b_pabd != NULL); |
b9541d6b CW |
2049 | } else { |
2050 | old_state = arc_l2c_only; | |
2051 | refcnt = 0; | |
d3c2ae1c GW |
2052 | bufcnt = 0; |
2053 | update_old = B_FALSE; | |
b9541d6b | 2054 | } |
d3c2ae1c | 2055 | update_new = update_old; |
34dc7c2f BB |
2056 | |
2057 | ASSERT(MUTEX_HELD(hash_lock)); | |
e8b96c60 | 2058 | ASSERT3P(new_state, !=, old_state); |
d3c2ae1c GW |
2059 | ASSERT(!GHOST_STATE(new_state) || bufcnt == 0); |
2060 | ASSERT(old_state != arc_anon || bufcnt <= 1); | |
34dc7c2f BB |
2061 | |
2062 | /* | |
2063 | * If this buffer is evictable, transfer it from the | |
2064 | * old state list to the new state list. | |
2065 | */ | |
2066 | if (refcnt == 0) { | |
b9541d6b | 2067 | if (old_state != arc_anon && old_state != arc_l2c_only) { |
b9541d6b | 2068 | ASSERT(HDR_HAS_L1HDR(hdr)); |
ca0bf58d | 2069 | multilist_remove(&old_state->arcs_list[buftype], hdr); |
34dc7c2f | 2070 | |
d3c2ae1c GW |
2071 | if (GHOST_STATE(old_state)) { |
2072 | ASSERT0(bufcnt); | |
2073 | ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); | |
2074 | update_old = B_TRUE; | |
34dc7c2f | 2075 | } |
2aa34383 | 2076 | arc_evictable_space_decrement(hdr, old_state); |
34dc7c2f | 2077 | } |
b9541d6b | 2078 | if (new_state != arc_anon && new_state != arc_l2c_only) { |
b9541d6b CW |
2079 | /* |
2080 | * An L1 header always exists here, since if we're | |
2081 | * moving to some L1-cached state (i.e. not l2c_only or | |
2082 | * anonymous), we realloc the header to add an L1hdr | |
2083 | * beforehand. | |
2084 | */ | |
2085 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
ca0bf58d | 2086 | multilist_insert(&new_state->arcs_list[buftype], hdr); |
34dc7c2f | 2087 | |
34dc7c2f | 2088 | if (GHOST_STATE(new_state)) { |
d3c2ae1c GW |
2089 | ASSERT0(bufcnt); |
2090 | ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); | |
2091 | update_new = B_TRUE; | |
34dc7c2f | 2092 | } |
d3c2ae1c | 2093 | arc_evictable_space_increment(hdr, new_state); |
34dc7c2f BB |
2094 | } |
2095 | } | |
2096 | ||
d3c2ae1c | 2097 | ASSERT(!HDR_EMPTY(hdr)); |
2a432414 GW |
2098 | if (new_state == arc_anon && HDR_IN_HASH_TABLE(hdr)) |
2099 | buf_hash_remove(hdr); | |
34dc7c2f | 2100 | |
b9541d6b | 2101 | /* adjust state sizes (ignore arc_l2c_only) */ |
36da08ef | 2102 | |
d3c2ae1c | 2103 | if (update_new && new_state != arc_l2c_only) { |
36da08ef PS |
2104 | ASSERT(HDR_HAS_L1HDR(hdr)); |
2105 | if (GHOST_STATE(new_state)) { | |
d3c2ae1c | 2106 | ASSERT0(bufcnt); |
36da08ef PS |
2107 | |
2108 | /* | |
d3c2ae1c | 2109 | * When moving a header to a ghost state, we first |
36da08ef | 2110 | * remove all arc buffers. Thus, we'll have a |
d3c2ae1c | 2111 | * bufcnt of zero, and no arc buffer to use for |
36da08ef PS |
2112 | * the reference. As a result, we use the arc |
2113 | * header pointer for the reference. | |
2114 | */ | |
2115 | (void) refcount_add_many(&new_state->arcs_size, | |
d3c2ae1c | 2116 | HDR_GET_LSIZE(hdr), hdr); |
a6255b7f | 2117 | ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); |
36da08ef PS |
2118 | } else { |
2119 | arc_buf_t *buf; | |
d3c2ae1c | 2120 | uint32_t buffers = 0; |
36da08ef PS |
2121 | |
2122 | /* | |
2123 | * Each individual buffer holds a unique reference, | |
2124 | * thus we must remove each of these references one | |
2125 | * at a time. | |
2126 | */ | |
2127 | for (buf = hdr->b_l1hdr.b_buf; buf != NULL; | |
2128 | buf = buf->b_next) { | |
d3c2ae1c GW |
2129 | ASSERT3U(bufcnt, !=, 0); |
2130 | buffers++; | |
2131 | ||
2132 | /* | |
2133 | * When the arc_buf_t is sharing the data | |
2134 | * block with the hdr, the owner of the | |
2135 | * reference belongs to the hdr. Only | |
2136 | * add to the refcount if the arc_buf_t is | |
2137 | * not shared. | |
2138 | */ | |
2aa34383 | 2139 | if (arc_buf_is_shared(buf)) |
d3c2ae1c | 2140 | continue; |
d3c2ae1c | 2141 | |
36da08ef | 2142 | (void) refcount_add_many(&new_state->arcs_size, |
2aa34383 | 2143 | arc_buf_size(buf), buf); |
d3c2ae1c GW |
2144 | } |
2145 | ASSERT3U(bufcnt, ==, buffers); | |
2146 | ||
a6255b7f | 2147 | if (hdr->b_l1hdr.b_pabd != NULL) { |
d3c2ae1c GW |
2148 | (void) refcount_add_many(&new_state->arcs_size, |
2149 | arc_hdr_size(hdr), hdr); | |
2150 | } else { | |
2151 | ASSERT(GHOST_STATE(old_state)); | |
36da08ef PS |
2152 | } |
2153 | } | |
2154 | } | |
2155 | ||
d3c2ae1c | 2156 | if (update_old && old_state != arc_l2c_only) { |
36da08ef PS |
2157 | ASSERT(HDR_HAS_L1HDR(hdr)); |
2158 | if (GHOST_STATE(old_state)) { | |
d3c2ae1c | 2159 | ASSERT0(bufcnt); |
a6255b7f | 2160 | ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); |
d3c2ae1c | 2161 | |
36da08ef PS |
2162 | /* |
2163 | * When moving a header off of a ghost state, | |
d3c2ae1c GW |
2164 | * the header will not contain any arc buffers. |
2165 | * We use the arc header pointer for the reference | |
2166 | * which is exactly what we did when we put the | |
2167 | * header on the ghost state. | |
36da08ef PS |
2168 | */ |
2169 | ||
36da08ef | 2170 | (void) refcount_remove_many(&old_state->arcs_size, |
d3c2ae1c | 2171 | HDR_GET_LSIZE(hdr), hdr); |
36da08ef PS |
2172 | } else { |
2173 | arc_buf_t *buf; | |
d3c2ae1c | 2174 | uint32_t buffers = 0; |
36da08ef PS |
2175 | |
2176 | /* | |
2177 | * Each individual buffer holds a unique reference, | |
2178 | * thus we must remove each of these references one | |
2179 | * at a time. | |
2180 | */ | |
2181 | for (buf = hdr->b_l1hdr.b_buf; buf != NULL; | |
2182 | buf = buf->b_next) { | |
d3c2ae1c GW |
2183 | ASSERT3U(bufcnt, !=, 0); |
2184 | buffers++; | |
2185 | ||
2186 | /* | |
2187 | * When the arc_buf_t is sharing the data | |
2188 | * block with the hdr, the owner of the | |
2189 | * reference belongs to the hdr. Only | |
2190 | * add to the refcount if the arc_buf_t is | |
2191 | * not shared. | |
2192 | */ | |
2aa34383 | 2193 | if (arc_buf_is_shared(buf)) |
d3c2ae1c | 2194 | continue; |
d3c2ae1c | 2195 | |
36da08ef | 2196 | (void) refcount_remove_many( |
2aa34383 | 2197 | &old_state->arcs_size, arc_buf_size(buf), |
d3c2ae1c | 2198 | buf); |
36da08ef | 2199 | } |
d3c2ae1c | 2200 | ASSERT3U(bufcnt, ==, buffers); |
a6255b7f | 2201 | ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); |
d3c2ae1c GW |
2202 | (void) refcount_remove_many( |
2203 | &old_state->arcs_size, arc_hdr_size(hdr), hdr); | |
36da08ef | 2204 | } |
34dc7c2f | 2205 | } |
36da08ef | 2206 | |
b9541d6b CW |
2207 | if (HDR_HAS_L1HDR(hdr)) |
2208 | hdr->b_l1hdr.b_state = new_state; | |
34dc7c2f | 2209 | |
b9541d6b CW |
2210 | /* |
2211 | * L2 headers should never be on the L2 state list since they don't | |
2212 | * have L1 headers allocated. | |
2213 | */ | |
ca0bf58d PS |
2214 | ASSERT(multilist_is_empty(&arc_l2c_only->arcs_list[ARC_BUFC_DATA]) && |
2215 | multilist_is_empty(&arc_l2c_only->arcs_list[ARC_BUFC_METADATA])); | |
34dc7c2f BB |
2216 | } |
2217 | ||
2218 | void | |
d164b209 | 2219 | arc_space_consume(uint64_t space, arc_space_type_t type) |
34dc7c2f | 2220 | { |
d164b209 BB |
2221 | ASSERT(type >= 0 && type < ARC_SPACE_NUMTYPES); |
2222 | ||
2223 | switch (type) { | |
e75c13c3 BB |
2224 | default: |
2225 | break; | |
d164b209 BB |
2226 | case ARC_SPACE_DATA: |
2227 | ARCSTAT_INCR(arcstat_data_size, space); | |
2228 | break; | |
cc7f677c | 2229 | case ARC_SPACE_META: |
500445c0 | 2230 | ARCSTAT_INCR(arcstat_metadata_size, space); |
cc7f677c | 2231 | break; |
25458cbe TC |
2232 | case ARC_SPACE_BONUS: |
2233 | ARCSTAT_INCR(arcstat_bonus_size, space); | |
2234 | break; | |
2235 | case ARC_SPACE_DNODE: | |
2236 | ARCSTAT_INCR(arcstat_dnode_size, space); | |
2237 | break; | |
2238 | case ARC_SPACE_DBUF: | |
2239 | ARCSTAT_INCR(arcstat_dbuf_size, space); | |
d164b209 BB |
2240 | break; |
2241 | case ARC_SPACE_HDRS: | |
2242 | ARCSTAT_INCR(arcstat_hdr_size, space); | |
2243 | break; | |
2244 | case ARC_SPACE_L2HDRS: | |
2245 | ARCSTAT_INCR(arcstat_l2_hdr_size, space); | |
2246 | break; | |
2247 | } | |
2248 | ||
500445c0 | 2249 | if (type != ARC_SPACE_DATA) |
cc7f677c PS |
2250 | ARCSTAT_INCR(arcstat_meta_used, space); |
2251 | ||
34dc7c2f BB |
2252 | atomic_add_64(&arc_size, space); |
2253 | } | |
2254 | ||
2255 | void | |
d164b209 | 2256 | arc_space_return(uint64_t space, arc_space_type_t type) |
34dc7c2f | 2257 | { |
d164b209 BB |
2258 | ASSERT(type >= 0 && type < ARC_SPACE_NUMTYPES); |
2259 | ||
2260 | switch (type) { | |
e75c13c3 BB |
2261 | default: |
2262 | break; | |
d164b209 BB |
2263 | case ARC_SPACE_DATA: |
2264 | ARCSTAT_INCR(arcstat_data_size, -space); | |
2265 | break; | |
cc7f677c | 2266 | case ARC_SPACE_META: |
500445c0 | 2267 | ARCSTAT_INCR(arcstat_metadata_size, -space); |
cc7f677c | 2268 | break; |
25458cbe TC |
2269 | case ARC_SPACE_BONUS: |
2270 | ARCSTAT_INCR(arcstat_bonus_size, -space); | |
2271 | break; | |
2272 | case ARC_SPACE_DNODE: | |
2273 | ARCSTAT_INCR(arcstat_dnode_size, -space); | |
2274 | break; | |
2275 | case ARC_SPACE_DBUF: | |
2276 | ARCSTAT_INCR(arcstat_dbuf_size, -space); | |
d164b209 BB |
2277 | break; |
2278 | case ARC_SPACE_HDRS: | |
2279 | ARCSTAT_INCR(arcstat_hdr_size, -space); | |
2280 | break; | |
2281 | case ARC_SPACE_L2HDRS: | |
2282 | ARCSTAT_INCR(arcstat_l2_hdr_size, -space); | |
2283 | break; | |
2284 | } | |
2285 | ||
cc7f677c PS |
2286 | if (type != ARC_SPACE_DATA) { |
2287 | ASSERT(arc_meta_used >= space); | |
500445c0 PS |
2288 | if (arc_meta_max < arc_meta_used) |
2289 | arc_meta_max = arc_meta_used; | |
cc7f677c PS |
2290 | ARCSTAT_INCR(arcstat_meta_used, -space); |
2291 | } | |
2292 | ||
34dc7c2f BB |
2293 | ASSERT(arc_size >= space); |
2294 | atomic_add_64(&arc_size, -space); | |
2295 | } | |
2296 | ||
d3c2ae1c | 2297 | /* |
524b4217 | 2298 | * Given a hdr and a buf, returns whether that buf can share its b_data buffer |
a6255b7f | 2299 | * with the hdr's b_pabd. |
d3c2ae1c | 2300 | */ |
524b4217 DK |
2301 | static boolean_t |
2302 | arc_can_share(arc_buf_hdr_t *hdr, arc_buf_t *buf) | |
2303 | { | |
2304 | boolean_t hdr_compressed, buf_compressed; | |
2305 | /* | |
2306 | * The criteria for sharing a hdr's data are: | |
2307 | * 1. the hdr's compression matches the buf's compression | |
2308 | * 2. the hdr doesn't need to be byteswapped | |
2309 | * 3. the hdr isn't already being shared | |
2310 | * 4. the buf is either compressed or it is the last buf in the hdr list | |
2311 | * | |
2312 | * Criterion #4 maintains the invariant that shared uncompressed | |
2313 | * bufs must be the final buf in the hdr's b_buf list. Reading this, you | |
2314 | * might ask, "if a compressed buf is allocated first, won't that be the | |
2315 | * last thing in the list?", but in that case it's impossible to create | |
2316 | * a shared uncompressed buf anyway (because the hdr must be compressed | |
2317 | * to have the compressed buf). You might also think that #3 is | |
2318 | * sufficient to make this guarantee, however it's possible | |
2319 | * (specifically in the rare L2ARC write race mentioned in | |
2320 | * arc_buf_alloc_impl()) there will be an existing uncompressed buf that | |
2321 | * is sharable, but wasn't at the time of its allocation. Rather than | |
2322 | * allow a new shared uncompressed buf to be created and then shuffle | |
2323 | * the list around to make it the last element, this simply disallows | |
2324 | * sharing if the new buf isn't the first to be added. | |
2325 | */ | |
2326 | ASSERT3P(buf->b_hdr, ==, hdr); | |
2327 | hdr_compressed = HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF; | |
2328 | buf_compressed = ARC_BUF_COMPRESSED(buf) != 0; | |
2329 | return (buf_compressed == hdr_compressed && | |
2330 | hdr->b_l1hdr.b_byteswap == DMU_BSWAP_NUMFUNCS && | |
2331 | !HDR_SHARED_DATA(hdr) && | |
2332 | (ARC_BUF_LAST(buf) || ARC_BUF_COMPRESSED(buf))); | |
2333 | } | |
2334 | ||
2335 | /* | |
2336 | * Allocate a buf for this hdr. If you care about the data that's in the hdr, | |
2337 | * or if you want a compressed buffer, pass those flags in. Returns 0 if the | |
2338 | * copy was made successfully, or an error code otherwise. | |
2339 | */ | |
2340 | static int | |
2341 | arc_buf_alloc_impl(arc_buf_hdr_t *hdr, void *tag, boolean_t compressed, | |
2342 | boolean_t fill, arc_buf_t **ret) | |
34dc7c2f | 2343 | { |
34dc7c2f | 2344 | arc_buf_t *buf; |
524b4217 | 2345 | boolean_t can_share; |
34dc7c2f | 2346 | |
d3c2ae1c GW |
2347 | ASSERT(HDR_HAS_L1HDR(hdr)); |
2348 | ASSERT3U(HDR_GET_LSIZE(hdr), >, 0); | |
2349 | VERIFY(hdr->b_type == ARC_BUFC_DATA || | |
2350 | hdr->b_type == ARC_BUFC_METADATA); | |
524b4217 DK |
2351 | ASSERT3P(ret, !=, NULL); |
2352 | ASSERT3P(*ret, ==, NULL); | |
d3c2ae1c | 2353 | |
b9541d6b CW |
2354 | hdr->b_l1hdr.b_mru_hits = 0; |
2355 | hdr->b_l1hdr.b_mru_ghost_hits = 0; | |
2356 | hdr->b_l1hdr.b_mfu_hits = 0; | |
2357 | hdr->b_l1hdr.b_mfu_ghost_hits = 0; | |
2358 | hdr->b_l1hdr.b_l2_hits = 0; | |
2359 | ||
524b4217 | 2360 | buf = *ret = kmem_cache_alloc(buf_cache, KM_PUSHPAGE); |
34dc7c2f BB |
2361 | buf->b_hdr = hdr; |
2362 | buf->b_data = NULL; | |
2aa34383 | 2363 | buf->b_next = hdr->b_l1hdr.b_buf; |
524b4217 | 2364 | buf->b_flags = 0; |
b9541d6b | 2365 | |
d3c2ae1c GW |
2366 | add_reference(hdr, tag); |
2367 | ||
2368 | /* | |
2369 | * We're about to change the hdr's b_flags. We must either | |
2370 | * hold the hash_lock or be undiscoverable. | |
2371 | */ | |
2372 | ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); | |
2373 | ||
2374 | /* | |
524b4217 DK |
2375 | * Only honor requests for compressed bufs if the hdr is actually |
2376 | * compressed. | |
2377 | */ | |
2378 | if (compressed && HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF) | |
2379 | buf->b_flags |= ARC_BUF_FLAG_COMPRESSED; | |
2380 | ||
2381 | /* | |
2382 | * Although the ARC should handle it correctly, levels above the ARC | |
2383 | * should prevent us from having multiple compressed bufs off the same | |
2384 | * hdr. To ensure we notice it if this behavior changes, we assert this | |
2385 | * here the best we can. | |
2386 | */ | |
2387 | IMPLY(ARC_BUF_COMPRESSED(buf), !HDR_SHARED_DATA(hdr)); | |
2388 | ||
2389 | /* | |
2390 | * If the hdr's data can be shared then we share the data buffer and | |
2391 | * set the appropriate bit in the hdr's b_flags to indicate the hdr is | |
2aa34383 | 2392 | * allocate a new buffer to store the buf's data. |
524b4217 | 2393 | * |
a6255b7f DQ |
2394 | * There are two additional restrictions here because we're sharing |
2395 | * hdr -> buf instead of the usual buf -> hdr. First, the hdr can't be | |
2396 | * actively involved in an L2ARC write, because if this buf is used by | |
2397 | * an arc_write() then the hdr's data buffer will be released when the | |
524b4217 | 2398 | * write completes, even though the L2ARC write might still be using it. |
a6255b7f DQ |
2399 | * Second, the hdr's ABD must be linear so that the buf's user doesn't |
2400 | * need to be ABD-aware. | |
d3c2ae1c | 2401 | */ |
a6255b7f DQ |
2402 | can_share = arc_can_share(hdr, buf) && !HDR_L2_WRITING(hdr) && |
2403 | abd_is_linear(hdr->b_l1hdr.b_pabd); | |
524b4217 DK |
2404 | |
2405 | /* Set up b_data and sharing */ | |
2406 | if (can_share) { | |
a6255b7f | 2407 | buf->b_data = abd_to_buf(hdr->b_l1hdr.b_pabd); |
524b4217 | 2408 | buf->b_flags |= ARC_BUF_FLAG_SHARED; |
d3c2ae1c GW |
2409 | arc_hdr_set_flags(hdr, ARC_FLAG_SHARED_DATA); |
2410 | } else { | |
524b4217 DK |
2411 | buf->b_data = |
2412 | arc_get_data_buf(hdr, arc_buf_size(buf), buf); | |
2413 | ARCSTAT_INCR(arcstat_overhead_size, arc_buf_size(buf)); | |
d3c2ae1c GW |
2414 | } |
2415 | VERIFY3P(buf->b_data, !=, NULL); | |
b9541d6b CW |
2416 | |
2417 | hdr->b_l1hdr.b_buf = buf; | |
d3c2ae1c | 2418 | hdr->b_l1hdr.b_bufcnt += 1; |
b9541d6b | 2419 | |
524b4217 DK |
2420 | /* |
2421 | * If the user wants the data from the hdr, we need to either copy or | |
2422 | * decompress the data. | |
2423 | */ | |
2424 | if (fill) { | |
2425 | return (arc_buf_fill(buf, ARC_BUF_COMPRESSED(buf) != 0)); | |
2426 | } | |
d3c2ae1c | 2427 | |
524b4217 | 2428 | return (0); |
34dc7c2f BB |
2429 | } |
2430 | ||
9babb374 BB |
2431 | static char *arc_onloan_tag = "onloan"; |
2432 | ||
2433 | /* | |
2434 | * Loan out an anonymous arc buffer. Loaned buffers are not counted as in | |
2435 | * flight data by arc_tempreserve_space() until they are "returned". Loaned | |
2436 | * buffers must be returned to the arc before they can be used by the DMU or | |
2437 | * freed. | |
2438 | */ | |
2439 | arc_buf_t * | |
2aa34383 | 2440 | arc_loan_buf(spa_t *spa, boolean_t is_metadata, int size) |
9babb374 | 2441 | { |
2aa34383 DK |
2442 | arc_buf_t *buf = arc_alloc_buf(spa, arc_onloan_tag, |
2443 | is_metadata ? ARC_BUFC_METADATA : ARC_BUFC_DATA, size); | |
9babb374 BB |
2444 | |
2445 | atomic_add_64(&arc_loaned_bytes, size); | |
2446 | return (buf); | |
2447 | } | |
2448 | ||
2aa34383 DK |
2449 | arc_buf_t * |
2450 | arc_loan_compressed_buf(spa_t *spa, uint64_t psize, uint64_t lsize, | |
2451 | enum zio_compress compression_type) | |
2452 | { | |
2453 | arc_buf_t *buf = arc_alloc_compressed_buf(spa, arc_onloan_tag, | |
2454 | psize, lsize, compression_type); | |
2455 | ||
2456 | atomic_add_64(&arc_loaned_bytes, psize); | |
2457 | return (buf); | |
2458 | } | |
2459 | ||
2460 | ||
9babb374 BB |
2461 | /* |
2462 | * Return a loaned arc buffer to the arc. | |
2463 | */ | |
2464 | void | |
2465 | arc_return_buf(arc_buf_t *buf, void *tag) | |
2466 | { | |
2467 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
2468 | ||
d3c2ae1c | 2469 | ASSERT3P(buf->b_data, !=, NULL); |
b9541d6b CW |
2470 | ASSERT(HDR_HAS_L1HDR(hdr)); |
2471 | (void) refcount_add(&hdr->b_l1hdr.b_refcnt, tag); | |
2472 | (void) refcount_remove(&hdr->b_l1hdr.b_refcnt, arc_onloan_tag); | |
9babb374 | 2473 | |
2aa34383 | 2474 | atomic_add_64(&arc_loaned_bytes, -arc_buf_size(buf)); |
9babb374 BB |
2475 | } |
2476 | ||
428870ff BB |
2477 | /* Detach an arc_buf from a dbuf (tag) */ |
2478 | void | |
2479 | arc_loan_inuse_buf(arc_buf_t *buf, void *tag) | |
2480 | { | |
b9541d6b | 2481 | arc_buf_hdr_t *hdr = buf->b_hdr; |
428870ff | 2482 | |
d3c2ae1c | 2483 | ASSERT3P(buf->b_data, !=, NULL); |
b9541d6b CW |
2484 | ASSERT(HDR_HAS_L1HDR(hdr)); |
2485 | (void) refcount_add(&hdr->b_l1hdr.b_refcnt, arc_onloan_tag); | |
2486 | (void) refcount_remove(&hdr->b_l1hdr.b_refcnt, tag); | |
428870ff | 2487 | |
2aa34383 | 2488 | atomic_add_64(&arc_loaned_bytes, -arc_buf_size(buf)); |
428870ff BB |
2489 | } |
2490 | ||
d3c2ae1c | 2491 | static void |
a6255b7f | 2492 | l2arc_free_abd_on_write(abd_t *abd, size_t size, arc_buf_contents_t type) |
34dc7c2f | 2493 | { |
d3c2ae1c | 2494 | l2arc_data_free_t *df = kmem_alloc(sizeof (*df), KM_SLEEP); |
34dc7c2f | 2495 | |
a6255b7f | 2496 | df->l2df_abd = abd; |
d3c2ae1c GW |
2497 | df->l2df_size = size; |
2498 | df->l2df_type = type; | |
2499 | mutex_enter(&l2arc_free_on_write_mtx); | |
2500 | list_insert_head(l2arc_free_on_write, df); | |
2501 | mutex_exit(&l2arc_free_on_write_mtx); | |
2502 | } | |
428870ff | 2503 | |
d3c2ae1c GW |
2504 | static void |
2505 | arc_hdr_free_on_write(arc_buf_hdr_t *hdr) | |
2506 | { | |
2507 | arc_state_t *state = hdr->b_l1hdr.b_state; | |
2508 | arc_buf_contents_t type = arc_buf_type(hdr); | |
2509 | uint64_t size = arc_hdr_size(hdr); | |
1eb5bfa3 | 2510 | |
d3c2ae1c GW |
2511 | /* protected by hash lock, if in the hash table */ |
2512 | if (multilist_link_active(&hdr->b_l1hdr.b_arc_node)) { | |
2513 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); | |
2514 | ASSERT(state != arc_anon && state != arc_l2c_only); | |
2515 | ||
2516 | (void) refcount_remove_many(&state->arcs_esize[type], | |
2517 | size, hdr); | |
1eb5bfa3 | 2518 | } |
d3c2ae1c GW |
2519 | (void) refcount_remove_many(&state->arcs_size, size, hdr); |
2520 | ||
a6255b7f | 2521 | l2arc_free_abd_on_write(hdr->b_l1hdr.b_pabd, size, type); |
34dc7c2f BB |
2522 | } |
2523 | ||
d3c2ae1c GW |
2524 | /* |
2525 | * Share the arc_buf_t's data with the hdr. Whenever we are sharing the | |
2526 | * data buffer, we transfer the refcount ownership to the hdr and update | |
2527 | * the appropriate kstats. | |
2528 | */ | |
2529 | static void | |
2530 | arc_share_buf(arc_buf_hdr_t *hdr, arc_buf_t *buf) | |
34dc7c2f | 2531 | { |
524b4217 | 2532 | ASSERT(arc_can_share(hdr, buf)); |
a6255b7f | 2533 | ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); |
d3c2ae1c | 2534 | ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); |
34dc7c2f BB |
2535 | |
2536 | /* | |
d3c2ae1c GW |
2537 | * Start sharing the data buffer. We transfer the |
2538 | * refcount ownership to the hdr since it always owns | |
2539 | * the refcount whenever an arc_buf_t is shared. | |
34dc7c2f | 2540 | */ |
d3c2ae1c | 2541 | refcount_transfer_ownership(&hdr->b_l1hdr.b_state->arcs_size, buf, hdr); |
a6255b7f DQ |
2542 | hdr->b_l1hdr.b_pabd = abd_get_from_buf(buf->b_data, arc_buf_size(buf)); |
2543 | abd_take_ownership_of_buf(hdr->b_l1hdr.b_pabd, | |
2544 | HDR_ISTYPE_METADATA(hdr)); | |
d3c2ae1c | 2545 | arc_hdr_set_flags(hdr, ARC_FLAG_SHARED_DATA); |
524b4217 | 2546 | buf->b_flags |= ARC_BUF_FLAG_SHARED; |
34dc7c2f | 2547 | |
d3c2ae1c GW |
2548 | /* |
2549 | * Since we've transferred ownership to the hdr we need | |
2550 | * to increment its compressed and uncompressed kstats and | |
2551 | * decrement the overhead size. | |
2552 | */ | |
2553 | ARCSTAT_INCR(arcstat_compressed_size, arc_hdr_size(hdr)); | |
2554 | ARCSTAT_INCR(arcstat_uncompressed_size, HDR_GET_LSIZE(hdr)); | |
2aa34383 | 2555 | ARCSTAT_INCR(arcstat_overhead_size, -arc_buf_size(buf)); |
34dc7c2f BB |
2556 | } |
2557 | ||
ca0bf58d | 2558 | static void |
d3c2ae1c | 2559 | arc_unshare_buf(arc_buf_hdr_t *hdr, arc_buf_t *buf) |
ca0bf58d | 2560 | { |
d3c2ae1c | 2561 | ASSERT(arc_buf_is_shared(buf)); |
a6255b7f | 2562 | ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); |
d3c2ae1c | 2563 | ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); |
ca0bf58d | 2564 | |
d3c2ae1c GW |
2565 | /* |
2566 | * We are no longer sharing this buffer so we need | |
2567 | * to transfer its ownership to the rightful owner. | |
2568 | */ | |
2569 | refcount_transfer_ownership(&hdr->b_l1hdr.b_state->arcs_size, hdr, buf); | |
2570 | arc_hdr_clear_flags(hdr, ARC_FLAG_SHARED_DATA); | |
a6255b7f DQ |
2571 | abd_release_ownership_of_buf(hdr->b_l1hdr.b_pabd); |
2572 | abd_put(hdr->b_l1hdr.b_pabd); | |
2573 | hdr->b_l1hdr.b_pabd = NULL; | |
524b4217 | 2574 | buf->b_flags &= ~ARC_BUF_FLAG_SHARED; |
d3c2ae1c GW |
2575 | |
2576 | /* | |
2577 | * Since the buffer is no longer shared between | |
2578 | * the arc buf and the hdr, count it as overhead. | |
2579 | */ | |
2580 | ARCSTAT_INCR(arcstat_compressed_size, -arc_hdr_size(hdr)); | |
2581 | ARCSTAT_INCR(arcstat_uncompressed_size, -HDR_GET_LSIZE(hdr)); | |
2aa34383 | 2582 | ARCSTAT_INCR(arcstat_overhead_size, arc_buf_size(buf)); |
ca0bf58d PS |
2583 | } |
2584 | ||
34dc7c2f | 2585 | /* |
2aa34383 DK |
2586 | * Remove an arc_buf_t from the hdr's buf list and return the last |
2587 | * arc_buf_t on the list. If no buffers remain on the list then return | |
2588 | * NULL. | |
2589 | */ | |
2590 | static arc_buf_t * | |
2591 | arc_buf_remove(arc_buf_hdr_t *hdr, arc_buf_t *buf) | |
2592 | { | |
2593 | arc_buf_t **bufp = &hdr->b_l1hdr.b_buf; | |
2594 | arc_buf_t *lastbuf = NULL; | |
2595 | ||
2596 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
2597 | ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); | |
2598 | ||
2599 | /* | |
2600 | * Remove the buf from the hdr list and locate the last | |
2601 | * remaining buffer on the list. | |
2602 | */ | |
2603 | while (*bufp != NULL) { | |
2604 | if (*bufp == buf) | |
2605 | *bufp = buf->b_next; | |
2606 | ||
2607 | /* | |
2608 | * If we've removed a buffer in the middle of | |
2609 | * the list then update the lastbuf and update | |
2610 | * bufp. | |
2611 | */ | |
2612 | if (*bufp != NULL) { | |
2613 | lastbuf = *bufp; | |
2614 | bufp = &(*bufp)->b_next; | |
2615 | } | |
2616 | } | |
2617 | buf->b_next = NULL; | |
2618 | ASSERT3P(lastbuf, !=, buf); | |
2619 | IMPLY(hdr->b_l1hdr.b_bufcnt > 0, lastbuf != NULL); | |
2620 | IMPLY(hdr->b_l1hdr.b_bufcnt > 0, hdr->b_l1hdr.b_buf != NULL); | |
2621 | IMPLY(lastbuf != NULL, ARC_BUF_LAST(lastbuf)); | |
2622 | ||
2623 | return (lastbuf); | |
2624 | } | |
2625 | ||
2626 | /* | |
2627 | * Free up buf->b_data and pull the arc_buf_t off of the the arc_buf_hdr_t's | |
2628 | * list and free it. | |
34dc7c2f BB |
2629 | */ |
2630 | static void | |
2aa34383 | 2631 | arc_buf_destroy_impl(arc_buf_t *buf) |
34dc7c2f | 2632 | { |
2aa34383 | 2633 | arc_buf_t *lastbuf; |
498877ba | 2634 | arc_buf_hdr_t *hdr = buf->b_hdr; |
ca0bf58d PS |
2635 | |
2636 | /* | |
524b4217 DK |
2637 | * Free up the data associated with the buf but only if we're not |
2638 | * sharing this with the hdr. If we are sharing it with the hdr, the | |
2639 | * hdr is responsible for doing the free. | |
ca0bf58d | 2640 | */ |
d3c2ae1c GW |
2641 | if (buf->b_data != NULL) { |
2642 | /* | |
2643 | * We're about to change the hdr's b_flags. We must either | |
2644 | * hold the hash_lock or be undiscoverable. | |
2645 | */ | |
2646 | ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); | |
2647 | ||
524b4217 | 2648 | arc_cksum_verify(buf); |
d3c2ae1c GW |
2649 | arc_buf_unwatch(buf); |
2650 | ||
2aa34383 | 2651 | if (arc_buf_is_shared(buf)) { |
d3c2ae1c GW |
2652 | arc_hdr_clear_flags(hdr, ARC_FLAG_SHARED_DATA); |
2653 | } else { | |
2aa34383 | 2654 | uint64_t size = arc_buf_size(buf); |
d3c2ae1c GW |
2655 | arc_free_data_buf(hdr, buf->b_data, size, buf); |
2656 | ARCSTAT_INCR(arcstat_overhead_size, -size); | |
2657 | } | |
2658 | buf->b_data = NULL; | |
2659 | ||
2660 | ASSERT(hdr->b_l1hdr.b_bufcnt > 0); | |
2661 | hdr->b_l1hdr.b_bufcnt -= 1; | |
2662 | } | |
2663 | ||
2aa34383 | 2664 | lastbuf = arc_buf_remove(hdr, buf); |
d3c2ae1c | 2665 | |
524b4217 | 2666 | if (ARC_BUF_SHARED(buf) && !ARC_BUF_COMPRESSED(buf)) { |
2aa34383 | 2667 | /* |
524b4217 | 2668 | * If the current arc_buf_t is sharing its data buffer with the |
a6255b7f | 2669 | * hdr, then reassign the hdr's b_pabd to share it with the new |
524b4217 DK |
2670 | * buffer at the end of the list. The shared buffer is always |
2671 | * the last one on the hdr's buffer list. | |
2672 | * | |
2673 | * There is an equivalent case for compressed bufs, but since | |
2674 | * they aren't guaranteed to be the last buf in the list and | |
2675 | * that is an exceedingly rare case, we just allow that space be | |
2676 | * wasted temporarily. | |
2aa34383 DK |
2677 | */ |
2678 | if (lastbuf != NULL) { | |
524b4217 | 2679 | /* Only one buf can be shared at once */ |
2aa34383 | 2680 | VERIFY(!arc_buf_is_shared(lastbuf)); |
524b4217 DK |
2681 | /* hdr is uncompressed so can't have compressed buf */ |
2682 | VERIFY(!ARC_BUF_COMPRESSED(lastbuf)); | |
d3c2ae1c | 2683 | |
a6255b7f DQ |
2684 | ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); |
2685 | arc_hdr_free_pabd(hdr); | |
d3c2ae1c | 2686 | |
2aa34383 DK |
2687 | /* |
2688 | * We must setup a new shared block between the | |
2689 | * last buffer and the hdr. The data would have | |
2690 | * been allocated by the arc buf so we need to transfer | |
2691 | * ownership to the hdr since it's now being shared. | |
2692 | */ | |
2693 | arc_share_buf(hdr, lastbuf); | |
2694 | } | |
2695 | } else if (HDR_SHARED_DATA(hdr)) { | |
d3c2ae1c | 2696 | /* |
2aa34383 DK |
2697 | * Uncompressed shared buffers are always at the end |
2698 | * of the list. Compressed buffers don't have the | |
2699 | * same requirements. This makes it hard to | |
2700 | * simply assert that the lastbuf is shared so | |
2701 | * we rely on the hdr's compression flags to determine | |
2702 | * if we have a compressed, shared buffer. | |
d3c2ae1c | 2703 | */ |
2aa34383 DK |
2704 | ASSERT3P(lastbuf, !=, NULL); |
2705 | ASSERT(arc_buf_is_shared(lastbuf) || | |
2706 | HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF); | |
ca0bf58d PS |
2707 | } |
2708 | ||
d3c2ae1c GW |
2709 | if (hdr->b_l1hdr.b_bufcnt == 0) |
2710 | arc_cksum_free(hdr); | |
2711 | ||
2712 | /* clean up the buf */ | |
2713 | buf->b_hdr = NULL; | |
2714 | kmem_cache_free(buf_cache, buf); | |
2715 | } | |
2716 | ||
2717 | static void | |
a6255b7f | 2718 | arc_hdr_alloc_pabd(arc_buf_hdr_t *hdr) |
d3c2ae1c GW |
2719 | { |
2720 | ASSERT3U(HDR_GET_LSIZE(hdr), >, 0); | |
2721 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
2722 | ASSERT(!HDR_SHARED_DATA(hdr)); | |
2723 | ||
a6255b7f DQ |
2724 | ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); |
2725 | hdr->b_l1hdr.b_pabd = arc_get_data_abd(hdr, arc_hdr_size(hdr), hdr); | |
d3c2ae1c | 2726 | hdr->b_l1hdr.b_byteswap = DMU_BSWAP_NUMFUNCS; |
a6255b7f | 2727 | ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); |
d3c2ae1c GW |
2728 | |
2729 | ARCSTAT_INCR(arcstat_compressed_size, arc_hdr_size(hdr)); | |
2730 | ARCSTAT_INCR(arcstat_uncompressed_size, HDR_GET_LSIZE(hdr)); | |
2731 | } | |
2732 | ||
2733 | static void | |
a6255b7f | 2734 | arc_hdr_free_pabd(arc_buf_hdr_t *hdr) |
d3c2ae1c GW |
2735 | { |
2736 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
a6255b7f | 2737 | ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); |
d3c2ae1c | 2738 | |
ca0bf58d | 2739 | /* |
d3c2ae1c GW |
2740 | * If the hdr is currently being written to the l2arc then |
2741 | * we defer freeing the data by adding it to the l2arc_free_on_write | |
2742 | * list. The l2arc will free the data once it's finished | |
2743 | * writing it to the l2arc device. | |
ca0bf58d | 2744 | */ |
d3c2ae1c GW |
2745 | if (HDR_L2_WRITING(hdr)) { |
2746 | arc_hdr_free_on_write(hdr); | |
2747 | ARCSTAT_BUMP(arcstat_l2_free_on_write); | |
2748 | } else { | |
a6255b7f | 2749 | arc_free_data_abd(hdr, hdr->b_l1hdr.b_pabd, |
d3c2ae1c | 2750 | arc_hdr_size(hdr), hdr); |
ca0bf58d | 2751 | } |
a6255b7f | 2752 | hdr->b_l1hdr.b_pabd = NULL; |
d3c2ae1c | 2753 | hdr->b_l1hdr.b_byteswap = DMU_BSWAP_NUMFUNCS; |
ca0bf58d | 2754 | |
d3c2ae1c GW |
2755 | ARCSTAT_INCR(arcstat_compressed_size, -arc_hdr_size(hdr)); |
2756 | ARCSTAT_INCR(arcstat_uncompressed_size, -HDR_GET_LSIZE(hdr)); | |
2757 | } | |
2758 | ||
2759 | static arc_buf_hdr_t * | |
2760 | arc_hdr_alloc(uint64_t spa, int32_t psize, int32_t lsize, | |
2aa34383 | 2761 | enum zio_compress compression_type, arc_buf_contents_t type) |
d3c2ae1c GW |
2762 | { |
2763 | arc_buf_hdr_t *hdr; | |
2764 | ||
d3c2ae1c GW |
2765 | VERIFY(type == ARC_BUFC_DATA || type == ARC_BUFC_METADATA); |
2766 | ||
2767 | hdr = kmem_cache_alloc(hdr_full_cache, KM_PUSHPAGE); | |
2768 | ASSERT(HDR_EMPTY(hdr)); | |
2769 | ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL); | |
2770 | HDR_SET_PSIZE(hdr, psize); | |
2771 | HDR_SET_LSIZE(hdr, lsize); | |
2772 | hdr->b_spa = spa; | |
2773 | hdr->b_type = type; | |
2774 | hdr->b_flags = 0; | |
2775 | arc_hdr_set_flags(hdr, arc_bufc_to_flags(type) | ARC_FLAG_HAS_L1HDR); | |
2aa34383 | 2776 | arc_hdr_set_compress(hdr, compression_type); |
ca0bf58d | 2777 | |
d3c2ae1c GW |
2778 | hdr->b_l1hdr.b_state = arc_anon; |
2779 | hdr->b_l1hdr.b_arc_access = 0; | |
2780 | hdr->b_l1hdr.b_bufcnt = 0; | |
2781 | hdr->b_l1hdr.b_buf = NULL; | |
ca0bf58d | 2782 | |
d3c2ae1c GW |
2783 | /* |
2784 | * Allocate the hdr's buffer. This will contain either | |
2785 | * the compressed or uncompressed data depending on the block | |
2786 | * it references and compressed arc enablement. | |
2787 | */ | |
a6255b7f | 2788 | arc_hdr_alloc_pabd(hdr); |
d3c2ae1c | 2789 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); |
ca0bf58d | 2790 | |
d3c2ae1c | 2791 | return (hdr); |
ca0bf58d PS |
2792 | } |
2793 | ||
bd089c54 | 2794 | /* |
d3c2ae1c GW |
2795 | * Transition between the two allocation states for the arc_buf_hdr struct. |
2796 | * The arc_buf_hdr struct can be allocated with (hdr_full_cache) or without | |
2797 | * (hdr_l2only_cache) the fields necessary for the L1 cache - the smaller | |
2798 | * version is used when a cache buffer is only in the L2ARC in order to reduce | |
2799 | * memory usage. | |
bd089c54 | 2800 | */ |
d3c2ae1c GW |
2801 | static arc_buf_hdr_t * |
2802 | arc_hdr_realloc(arc_buf_hdr_t *hdr, kmem_cache_t *old, kmem_cache_t *new) | |
34dc7c2f | 2803 | { |
d3c2ae1c GW |
2804 | arc_buf_hdr_t *nhdr; |
2805 | l2arc_dev_t *dev = hdr->b_l2hdr.b_dev; | |
34dc7c2f | 2806 | |
d3c2ae1c GW |
2807 | ASSERT(HDR_HAS_L2HDR(hdr)); |
2808 | ASSERT((old == hdr_full_cache && new == hdr_l2only_cache) || | |
2809 | (old == hdr_l2only_cache && new == hdr_full_cache)); | |
34dc7c2f | 2810 | |
d3c2ae1c | 2811 | nhdr = kmem_cache_alloc(new, KM_PUSHPAGE); |
428870ff | 2812 | |
d3c2ae1c GW |
2813 | ASSERT(MUTEX_HELD(HDR_LOCK(hdr))); |
2814 | buf_hash_remove(hdr); | |
ca0bf58d | 2815 | |
d3c2ae1c | 2816 | bcopy(hdr, nhdr, HDR_L2ONLY_SIZE); |
34dc7c2f | 2817 | |
d3c2ae1c GW |
2818 | if (new == hdr_full_cache) { |
2819 | arc_hdr_set_flags(nhdr, ARC_FLAG_HAS_L1HDR); | |
2820 | /* | |
2821 | * arc_access and arc_change_state need to be aware that a | |
2822 | * header has just come out of L2ARC, so we set its state to | |
2823 | * l2c_only even though it's about to change. | |
2824 | */ | |
2825 | nhdr->b_l1hdr.b_state = arc_l2c_only; | |
34dc7c2f | 2826 | |
d3c2ae1c | 2827 | /* Verify previous threads set to NULL before freeing */ |
a6255b7f | 2828 | ASSERT3P(nhdr->b_l1hdr.b_pabd, ==, NULL); |
d3c2ae1c GW |
2829 | } else { |
2830 | ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); | |
2831 | ASSERT0(hdr->b_l1hdr.b_bufcnt); | |
2832 | ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL); | |
36da08ef | 2833 | |
d3c2ae1c GW |
2834 | /* |
2835 | * If we've reached here, We must have been called from | |
2836 | * arc_evict_hdr(), as such we should have already been | |
2837 | * removed from any ghost list we were previously on | |
2838 | * (which protects us from racing with arc_evict_state), | |
2839 | * thus no locking is needed during this check. | |
2840 | */ | |
2841 | ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); | |
1eb5bfa3 GW |
2842 | |
2843 | /* | |
d3c2ae1c GW |
2844 | * A buffer must not be moved into the arc_l2c_only |
2845 | * state if it's not finished being written out to the | |
a6255b7f | 2846 | * l2arc device. Otherwise, the b_l1hdr.b_pabd field |
d3c2ae1c | 2847 | * might try to be accessed, even though it was removed. |
1eb5bfa3 | 2848 | */ |
d3c2ae1c | 2849 | VERIFY(!HDR_L2_WRITING(hdr)); |
a6255b7f | 2850 | VERIFY3P(hdr->b_l1hdr.b_pabd, ==, NULL); |
d3c2ae1c GW |
2851 | |
2852 | arc_hdr_clear_flags(nhdr, ARC_FLAG_HAS_L1HDR); | |
34dc7c2f | 2853 | } |
d3c2ae1c GW |
2854 | /* |
2855 | * The header has been reallocated so we need to re-insert it into any | |
2856 | * lists it was on. | |
2857 | */ | |
2858 | (void) buf_hash_insert(nhdr, NULL); | |
34dc7c2f | 2859 | |
d3c2ae1c | 2860 | ASSERT(list_link_active(&hdr->b_l2hdr.b_l2node)); |
34dc7c2f | 2861 | |
d3c2ae1c GW |
2862 | mutex_enter(&dev->l2ad_mtx); |
2863 | ||
2864 | /* | |
2865 | * We must place the realloc'ed header back into the list at | |
2866 | * the same spot. Otherwise, if it's placed earlier in the list, | |
2867 | * l2arc_write_buffers() could find it during the function's | |
2868 | * write phase, and try to write it out to the l2arc. | |
2869 | */ | |
2870 | list_insert_after(&dev->l2ad_buflist, hdr, nhdr); | |
2871 | list_remove(&dev->l2ad_buflist, hdr); | |
34dc7c2f | 2872 | |
d3c2ae1c | 2873 | mutex_exit(&dev->l2ad_mtx); |
34dc7c2f | 2874 | |
d3c2ae1c GW |
2875 | /* |
2876 | * Since we're using the pointer address as the tag when | |
2877 | * incrementing and decrementing the l2ad_alloc refcount, we | |
2878 | * must remove the old pointer (that we're about to destroy) and | |
2879 | * add the new pointer to the refcount. Otherwise we'd remove | |
2880 | * the wrong pointer address when calling arc_hdr_destroy() later. | |
2881 | */ | |
2882 | ||
2883 | (void) refcount_remove_many(&dev->l2ad_alloc, arc_hdr_size(hdr), hdr); | |
2884 | (void) refcount_add_many(&dev->l2ad_alloc, arc_hdr_size(nhdr), nhdr); | |
2885 | ||
2886 | buf_discard_identity(hdr); | |
2887 | kmem_cache_free(old, hdr); | |
2888 | ||
2889 | return (nhdr); | |
2890 | } | |
2891 | ||
2892 | /* | |
2893 | * Allocate a new arc_buf_hdr_t and arc_buf_t and return the buf to the caller. | |
2894 | * The buf is returned thawed since we expect the consumer to modify it. | |
2895 | */ | |
2896 | arc_buf_t * | |
2aa34383 | 2897 | arc_alloc_buf(spa_t *spa, void *tag, arc_buf_contents_t type, int32_t size) |
d3c2ae1c GW |
2898 | { |
2899 | arc_buf_t *buf; | |
2900 | arc_buf_hdr_t *hdr = arc_hdr_alloc(spa_load_guid(spa), size, size, | |
2901 | ZIO_COMPRESS_OFF, type); | |
2902 | ASSERT(!MUTEX_HELD(HDR_LOCK(hdr))); | |
2aa34383 | 2903 | |
524b4217 DK |
2904 | buf = NULL; |
2905 | VERIFY0(arc_buf_alloc_impl(hdr, tag, B_FALSE, B_FALSE, &buf)); | |
d3c2ae1c | 2906 | arc_buf_thaw(buf); |
2aa34383 DK |
2907 | |
2908 | return (buf); | |
2909 | } | |
2910 | ||
2911 | /* | |
2912 | * Allocate a compressed buf in the same manner as arc_alloc_buf. Don't use this | |
2913 | * for bufs containing metadata. | |
2914 | */ | |
2915 | arc_buf_t * | |
2916 | arc_alloc_compressed_buf(spa_t *spa, void *tag, uint64_t psize, uint64_t lsize, | |
2917 | enum zio_compress compression_type) | |
2918 | { | |
2919 | arc_buf_hdr_t *hdr; | |
2920 | arc_buf_t *buf; | |
2921 | ASSERT3U(lsize, >, 0); | |
2922 | ASSERT3U(lsize, >=, psize); | |
2923 | ASSERT(compression_type > ZIO_COMPRESS_OFF); | |
2924 | ASSERT(compression_type < ZIO_COMPRESS_FUNCTIONS); | |
2925 | ||
2926 | hdr = arc_hdr_alloc(spa_load_guid(spa), psize, lsize, | |
2927 | compression_type, ARC_BUFC_DATA); | |
2928 | ASSERT(!MUTEX_HELD(HDR_LOCK(hdr))); | |
2929 | ||
524b4217 DK |
2930 | buf = NULL; |
2931 | VERIFY0(arc_buf_alloc_impl(hdr, tag, B_TRUE, B_FALSE, &buf)); | |
2aa34383 DK |
2932 | arc_buf_thaw(buf); |
2933 | ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL); | |
2934 | ||
a6255b7f DQ |
2935 | if (!arc_buf_is_shared(buf)) { |
2936 | /* | |
2937 | * To ensure that the hdr has the correct data in it if we call | |
2938 | * arc_decompress() on this buf before it's been written to | |
2939 | * disk, it's easiest if we just set up sharing between the | |
2940 | * buf and the hdr. | |
2941 | */ | |
2942 | ASSERT(!abd_is_linear(hdr->b_l1hdr.b_pabd)); | |
2943 | arc_hdr_free_pabd(hdr); | |
2944 | arc_share_buf(hdr, buf); | |
2945 | } | |
2946 | ||
d3c2ae1c | 2947 | return (buf); |
34dc7c2f BB |
2948 | } |
2949 | ||
d962d5da PS |
2950 | static void |
2951 | arc_hdr_l2hdr_destroy(arc_buf_hdr_t *hdr) | |
2952 | { | |
2953 | l2arc_buf_hdr_t *l2hdr = &hdr->b_l2hdr; | |
2954 | l2arc_dev_t *dev = l2hdr->b_dev; | |
d3c2ae1c | 2955 | uint64_t asize = arc_hdr_size(hdr); |
d962d5da PS |
2956 | |
2957 | ASSERT(MUTEX_HELD(&dev->l2ad_mtx)); | |
2958 | ASSERT(HDR_HAS_L2HDR(hdr)); | |
2959 | ||
2960 | list_remove(&dev->l2ad_buflist, hdr); | |
2961 | ||
d3c2ae1c GW |
2962 | ARCSTAT_INCR(arcstat_l2_asize, -asize); |
2963 | ARCSTAT_INCR(arcstat_l2_size, -HDR_GET_LSIZE(hdr)); | |
d962d5da | 2964 | |
d3c2ae1c | 2965 | vdev_space_update(dev->l2ad_vdev, -asize, 0, 0); |
d962d5da | 2966 | |
d3c2ae1c GW |
2967 | (void) refcount_remove_many(&dev->l2ad_alloc, asize, hdr); |
2968 | arc_hdr_clear_flags(hdr, ARC_FLAG_HAS_L2HDR); | |
d962d5da PS |
2969 | } |
2970 | ||
34dc7c2f BB |
2971 | static void |
2972 | arc_hdr_destroy(arc_buf_hdr_t *hdr) | |
2973 | { | |
b9541d6b CW |
2974 | if (HDR_HAS_L1HDR(hdr)) { |
2975 | ASSERT(hdr->b_l1hdr.b_buf == NULL || | |
d3c2ae1c | 2976 | hdr->b_l1hdr.b_bufcnt > 0); |
b9541d6b CW |
2977 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); |
2978 | ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon); | |
2979 | } | |
34dc7c2f | 2980 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); |
b9541d6b CW |
2981 | ASSERT(!HDR_IN_HASH_TABLE(hdr)); |
2982 | ||
d3c2ae1c GW |
2983 | if (!HDR_EMPTY(hdr)) |
2984 | buf_discard_identity(hdr); | |
2985 | ||
b9541d6b | 2986 | if (HDR_HAS_L2HDR(hdr)) { |
d962d5da PS |
2987 | l2arc_dev_t *dev = hdr->b_l2hdr.b_dev; |
2988 | boolean_t buflist_held = MUTEX_HELD(&dev->l2ad_mtx); | |
428870ff | 2989 | |
d962d5da PS |
2990 | if (!buflist_held) |
2991 | mutex_enter(&dev->l2ad_mtx); | |
b9541d6b | 2992 | |
ca0bf58d | 2993 | /* |
d962d5da PS |
2994 | * Even though we checked this conditional above, we |
2995 | * need to check this again now that we have the | |
2996 | * l2ad_mtx. This is because we could be racing with | |
2997 | * another thread calling l2arc_evict() which might have | |
2998 | * destroyed this header's L2 portion as we were waiting | |
2999 | * to acquire the l2ad_mtx. If that happens, we don't | |
3000 | * want to re-destroy the header's L2 portion. | |
ca0bf58d | 3001 | */ |
d962d5da PS |
3002 | if (HDR_HAS_L2HDR(hdr)) |
3003 | arc_hdr_l2hdr_destroy(hdr); | |
428870ff BB |
3004 | |
3005 | if (!buflist_held) | |
d962d5da | 3006 | mutex_exit(&dev->l2ad_mtx); |
34dc7c2f BB |
3007 | } |
3008 | ||
d3c2ae1c GW |
3009 | if (HDR_HAS_L1HDR(hdr)) { |
3010 | arc_cksum_free(hdr); | |
b9541d6b | 3011 | |
d3c2ae1c | 3012 | while (hdr->b_l1hdr.b_buf != NULL) |
2aa34383 | 3013 | arc_buf_destroy_impl(hdr->b_l1hdr.b_buf); |
34dc7c2f | 3014 | |
a6255b7f DQ |
3015 | if (hdr->b_l1hdr.b_pabd != NULL) |
3016 | arc_hdr_free_pabd(hdr); | |
b9541d6b CW |
3017 | } |
3018 | ||
34dc7c2f | 3019 | ASSERT3P(hdr->b_hash_next, ==, NULL); |
b9541d6b | 3020 | if (HDR_HAS_L1HDR(hdr)) { |
ca0bf58d | 3021 | ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); |
b9541d6b CW |
3022 | ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL); |
3023 | kmem_cache_free(hdr_full_cache, hdr); | |
3024 | } else { | |
3025 | kmem_cache_free(hdr_l2only_cache, hdr); | |
3026 | } | |
34dc7c2f BB |
3027 | } |
3028 | ||
3029 | void | |
d3c2ae1c | 3030 | arc_buf_destroy(arc_buf_t *buf, void* tag) |
34dc7c2f BB |
3031 | { |
3032 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
96c080cb | 3033 | kmutex_t *hash_lock = HDR_LOCK(hdr); |
34dc7c2f | 3034 | |
b9541d6b | 3035 | if (hdr->b_l1hdr.b_state == arc_anon) { |
d3c2ae1c GW |
3036 | ASSERT3U(hdr->b_l1hdr.b_bufcnt, ==, 1); |
3037 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); | |
3038 | VERIFY0(remove_reference(hdr, NULL, tag)); | |
3039 | arc_hdr_destroy(hdr); | |
3040 | return; | |
34dc7c2f BB |
3041 | } |
3042 | ||
3043 | mutex_enter(hash_lock); | |
d3c2ae1c GW |
3044 | ASSERT3P(hdr, ==, buf->b_hdr); |
3045 | ASSERT(hdr->b_l1hdr.b_bufcnt > 0); | |
428870ff | 3046 | ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); |
d3c2ae1c GW |
3047 | ASSERT3P(hdr->b_l1hdr.b_state, !=, arc_anon); |
3048 | ASSERT3P(buf->b_data, !=, NULL); | |
34dc7c2f BB |
3049 | |
3050 | (void) remove_reference(hdr, hash_lock, tag); | |
2aa34383 | 3051 | arc_buf_destroy_impl(buf); |
34dc7c2f | 3052 | mutex_exit(hash_lock); |
34dc7c2f BB |
3053 | } |
3054 | ||
34dc7c2f | 3055 | /* |
ca0bf58d PS |
3056 | * Evict the arc_buf_hdr that is provided as a parameter. The resultant |
3057 | * state of the header is dependent on its state prior to entering this | |
3058 | * function. The following transitions are possible: | |
34dc7c2f | 3059 | * |
ca0bf58d PS |
3060 | * - arc_mru -> arc_mru_ghost |
3061 | * - arc_mfu -> arc_mfu_ghost | |
3062 | * - arc_mru_ghost -> arc_l2c_only | |
3063 | * - arc_mru_ghost -> deleted | |
3064 | * - arc_mfu_ghost -> arc_l2c_only | |
3065 | * - arc_mfu_ghost -> deleted | |
34dc7c2f | 3066 | */ |
ca0bf58d PS |
3067 | static int64_t |
3068 | arc_evict_hdr(arc_buf_hdr_t *hdr, kmutex_t *hash_lock) | |
34dc7c2f | 3069 | { |
ca0bf58d PS |
3070 | arc_state_t *evicted_state, *state; |
3071 | int64_t bytes_evicted = 0; | |
34dc7c2f | 3072 | |
ca0bf58d PS |
3073 | ASSERT(MUTEX_HELD(hash_lock)); |
3074 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
e8b96c60 | 3075 | |
ca0bf58d PS |
3076 | state = hdr->b_l1hdr.b_state; |
3077 | if (GHOST_STATE(state)) { | |
3078 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); | |
d3c2ae1c | 3079 | ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); |
e8b96c60 MA |
3080 | |
3081 | /* | |
ca0bf58d | 3082 | * l2arc_write_buffers() relies on a header's L1 portion |
a6255b7f | 3083 | * (i.e. its b_pabd field) during it's write phase. |
ca0bf58d PS |
3084 | * Thus, we cannot push a header onto the arc_l2c_only |
3085 | * state (removing its L1 piece) until the header is | |
3086 | * done being written to the l2arc. | |
e8b96c60 | 3087 | */ |
ca0bf58d PS |
3088 | if (HDR_HAS_L2HDR(hdr) && HDR_L2_WRITING(hdr)) { |
3089 | ARCSTAT_BUMP(arcstat_evict_l2_skip); | |
3090 | return (bytes_evicted); | |
e8b96c60 MA |
3091 | } |
3092 | ||
ca0bf58d | 3093 | ARCSTAT_BUMP(arcstat_deleted); |
d3c2ae1c | 3094 | bytes_evicted += HDR_GET_LSIZE(hdr); |
428870ff | 3095 | |
ca0bf58d | 3096 | DTRACE_PROBE1(arc__delete, arc_buf_hdr_t *, hdr); |
428870ff | 3097 | |
ca0bf58d | 3098 | if (HDR_HAS_L2HDR(hdr)) { |
a6255b7f | 3099 | ASSERT(hdr->b_l1hdr.b_pabd == NULL); |
ca0bf58d PS |
3100 | /* |
3101 | * This buffer is cached on the 2nd Level ARC; | |
3102 | * don't destroy the header. | |
3103 | */ | |
3104 | arc_change_state(arc_l2c_only, hdr, hash_lock); | |
3105 | /* | |
3106 | * dropping from L1+L2 cached to L2-only, | |
3107 | * realloc to remove the L1 header. | |
3108 | */ | |
3109 | hdr = arc_hdr_realloc(hdr, hdr_full_cache, | |
3110 | hdr_l2only_cache); | |
34dc7c2f | 3111 | } else { |
ca0bf58d PS |
3112 | arc_change_state(arc_anon, hdr, hash_lock); |
3113 | arc_hdr_destroy(hdr); | |
34dc7c2f | 3114 | } |
ca0bf58d | 3115 | return (bytes_evicted); |
34dc7c2f BB |
3116 | } |
3117 | ||
ca0bf58d PS |
3118 | ASSERT(state == arc_mru || state == arc_mfu); |
3119 | evicted_state = (state == arc_mru) ? arc_mru_ghost : arc_mfu_ghost; | |
34dc7c2f | 3120 | |
ca0bf58d PS |
3121 | /* prefetch buffers have a minimum lifespan */ |
3122 | if (HDR_IO_IN_PROGRESS(hdr) || | |
3123 | ((hdr->b_flags & (ARC_FLAG_PREFETCH | ARC_FLAG_INDIRECT)) && | |
3124 | ddi_get_lbolt() - hdr->b_l1hdr.b_arc_access < | |
3125 | arc_min_prefetch_lifespan)) { | |
3126 | ARCSTAT_BUMP(arcstat_evict_skip); | |
3127 | return (bytes_evicted); | |
da8ccd0e PS |
3128 | } |
3129 | ||
ca0bf58d | 3130 | ASSERT0(refcount_count(&hdr->b_l1hdr.b_refcnt)); |
ca0bf58d PS |
3131 | while (hdr->b_l1hdr.b_buf) { |
3132 | arc_buf_t *buf = hdr->b_l1hdr.b_buf; | |
3133 | if (!mutex_tryenter(&buf->b_evict_lock)) { | |
3134 | ARCSTAT_BUMP(arcstat_mutex_miss); | |
3135 | break; | |
3136 | } | |
3137 | if (buf->b_data != NULL) | |
d3c2ae1c GW |
3138 | bytes_evicted += HDR_GET_LSIZE(hdr); |
3139 | mutex_exit(&buf->b_evict_lock); | |
2aa34383 | 3140 | arc_buf_destroy_impl(buf); |
ca0bf58d | 3141 | } |
34dc7c2f | 3142 | |
ca0bf58d | 3143 | if (HDR_HAS_L2HDR(hdr)) { |
d3c2ae1c | 3144 | ARCSTAT_INCR(arcstat_evict_l2_cached, HDR_GET_LSIZE(hdr)); |
ca0bf58d | 3145 | } else { |
d3c2ae1c GW |
3146 | if (l2arc_write_eligible(hdr->b_spa, hdr)) { |
3147 | ARCSTAT_INCR(arcstat_evict_l2_eligible, | |
3148 | HDR_GET_LSIZE(hdr)); | |
3149 | } else { | |
3150 | ARCSTAT_INCR(arcstat_evict_l2_ineligible, | |
3151 | HDR_GET_LSIZE(hdr)); | |
3152 | } | |
ca0bf58d | 3153 | } |
34dc7c2f | 3154 | |
d3c2ae1c GW |
3155 | if (hdr->b_l1hdr.b_bufcnt == 0) { |
3156 | arc_cksum_free(hdr); | |
3157 | ||
3158 | bytes_evicted += arc_hdr_size(hdr); | |
3159 | ||
3160 | /* | |
3161 | * If this hdr is being evicted and has a compressed | |
3162 | * buffer then we discard it here before we change states. | |
3163 | * This ensures that the accounting is updated correctly | |
a6255b7f | 3164 | * in arc_free_data_impl(). |
d3c2ae1c | 3165 | */ |
a6255b7f | 3166 | arc_hdr_free_pabd(hdr); |
d3c2ae1c | 3167 | |
ca0bf58d PS |
3168 | arc_change_state(evicted_state, hdr, hash_lock); |
3169 | ASSERT(HDR_IN_HASH_TABLE(hdr)); | |
d3c2ae1c | 3170 | arc_hdr_set_flags(hdr, ARC_FLAG_IN_HASH_TABLE); |
ca0bf58d PS |
3171 | DTRACE_PROBE1(arc__evict, arc_buf_hdr_t *, hdr); |
3172 | } | |
34dc7c2f | 3173 | |
ca0bf58d | 3174 | return (bytes_evicted); |
34dc7c2f BB |
3175 | } |
3176 | ||
ca0bf58d PS |
3177 | static uint64_t |
3178 | arc_evict_state_impl(multilist_t *ml, int idx, arc_buf_hdr_t *marker, | |
3179 | uint64_t spa, int64_t bytes) | |
34dc7c2f | 3180 | { |
ca0bf58d PS |
3181 | multilist_sublist_t *mls; |
3182 | uint64_t bytes_evicted = 0; | |
3183 | arc_buf_hdr_t *hdr; | |
34dc7c2f | 3184 | kmutex_t *hash_lock; |
ca0bf58d | 3185 | int evict_count = 0; |
34dc7c2f | 3186 | |
ca0bf58d | 3187 | ASSERT3P(marker, !=, NULL); |
96c080cb | 3188 | IMPLY(bytes < 0, bytes == ARC_EVICT_ALL); |
ca0bf58d PS |
3189 | |
3190 | mls = multilist_sublist_lock(ml, idx); | |
572e2857 | 3191 | |
ca0bf58d PS |
3192 | for (hdr = multilist_sublist_prev(mls, marker); hdr != NULL; |
3193 | hdr = multilist_sublist_prev(mls, marker)) { | |
3194 | if ((bytes != ARC_EVICT_ALL && bytes_evicted >= bytes) || | |
3195 | (evict_count >= zfs_arc_evict_batch_limit)) | |
3196 | break; | |
3197 | ||
3198 | /* | |
3199 | * To keep our iteration location, move the marker | |
3200 | * forward. Since we're not holding hdr's hash lock, we | |
3201 | * must be very careful and not remove 'hdr' from the | |
3202 | * sublist. Otherwise, other consumers might mistake the | |
3203 | * 'hdr' as not being on a sublist when they call the | |
3204 | * multilist_link_active() function (they all rely on | |
3205 | * the hash lock protecting concurrent insertions and | |
3206 | * removals). multilist_sublist_move_forward() was | |
3207 | * specifically implemented to ensure this is the case | |
3208 | * (only 'marker' will be removed and re-inserted). | |
3209 | */ | |
3210 | multilist_sublist_move_forward(mls, marker); | |
3211 | ||
3212 | /* | |
3213 | * The only case where the b_spa field should ever be | |
3214 | * zero, is the marker headers inserted by | |
3215 | * arc_evict_state(). It's possible for multiple threads | |
3216 | * to be calling arc_evict_state() concurrently (e.g. | |
3217 | * dsl_pool_close() and zio_inject_fault()), so we must | |
3218 | * skip any markers we see from these other threads. | |
3219 | */ | |
2a432414 | 3220 | if (hdr->b_spa == 0) |
572e2857 BB |
3221 | continue; |
3222 | ||
ca0bf58d PS |
3223 | /* we're only interested in evicting buffers of a certain spa */ |
3224 | if (spa != 0 && hdr->b_spa != spa) { | |
3225 | ARCSTAT_BUMP(arcstat_evict_skip); | |
428870ff | 3226 | continue; |
ca0bf58d PS |
3227 | } |
3228 | ||
3229 | hash_lock = HDR_LOCK(hdr); | |
e8b96c60 MA |
3230 | |
3231 | /* | |
ca0bf58d PS |
3232 | * We aren't calling this function from any code path |
3233 | * that would already be holding a hash lock, so we're | |
3234 | * asserting on this assumption to be defensive in case | |
3235 | * this ever changes. Without this check, it would be | |
3236 | * possible to incorrectly increment arcstat_mutex_miss | |
3237 | * below (e.g. if the code changed such that we called | |
3238 | * this function with a hash lock held). | |
e8b96c60 | 3239 | */ |
ca0bf58d PS |
3240 | ASSERT(!MUTEX_HELD(hash_lock)); |
3241 | ||
34dc7c2f | 3242 | if (mutex_tryenter(hash_lock)) { |
ca0bf58d PS |
3243 | uint64_t evicted = arc_evict_hdr(hdr, hash_lock); |
3244 | mutex_exit(hash_lock); | |
34dc7c2f | 3245 | |
ca0bf58d | 3246 | bytes_evicted += evicted; |
34dc7c2f | 3247 | |
572e2857 | 3248 | /* |
ca0bf58d PS |
3249 | * If evicted is zero, arc_evict_hdr() must have |
3250 | * decided to skip this header, don't increment | |
3251 | * evict_count in this case. | |
572e2857 | 3252 | */ |
ca0bf58d PS |
3253 | if (evicted != 0) |
3254 | evict_count++; | |
3255 | ||
3256 | /* | |
3257 | * If arc_size isn't overflowing, signal any | |
3258 | * threads that might happen to be waiting. | |
3259 | * | |
3260 | * For each header evicted, we wake up a single | |
3261 | * thread. If we used cv_broadcast, we could | |
3262 | * wake up "too many" threads causing arc_size | |
3263 | * to significantly overflow arc_c; since | |
a6255b7f | 3264 | * arc_get_data_impl() doesn't check for overflow |
ca0bf58d PS |
3265 | * when it's woken up (it doesn't because it's |
3266 | * possible for the ARC to be overflowing while | |
3267 | * full of un-evictable buffers, and the | |
3268 | * function should proceed in this case). | |
3269 | * | |
3270 | * If threads are left sleeping, due to not | |
3271 | * using cv_broadcast, they will be woken up | |
3272 | * just before arc_reclaim_thread() sleeps. | |
3273 | */ | |
3274 | mutex_enter(&arc_reclaim_lock); | |
3275 | if (!arc_is_overflowing()) | |
3276 | cv_signal(&arc_reclaim_waiters_cv); | |
3277 | mutex_exit(&arc_reclaim_lock); | |
e8b96c60 | 3278 | } else { |
ca0bf58d | 3279 | ARCSTAT_BUMP(arcstat_mutex_miss); |
e8b96c60 | 3280 | } |
34dc7c2f | 3281 | } |
34dc7c2f | 3282 | |
ca0bf58d | 3283 | multilist_sublist_unlock(mls); |
34dc7c2f | 3284 | |
ca0bf58d | 3285 | return (bytes_evicted); |
34dc7c2f BB |
3286 | } |
3287 | ||
ca0bf58d PS |
3288 | /* |
3289 | * Evict buffers from the given arc state, until we've removed the | |
3290 | * specified number of bytes. Move the removed buffers to the | |
3291 | * appropriate evict state. | |
3292 | * | |
3293 | * This function makes a "best effort". It skips over any buffers | |
3294 | * it can't get a hash_lock on, and so, may not catch all candidates. | |
3295 | * It may also return without evicting as much space as requested. | |
3296 | * | |
3297 | * If bytes is specified using the special value ARC_EVICT_ALL, this | |
3298 | * will evict all available (i.e. unlocked and evictable) buffers from | |
3299 | * the given arc state; which is used by arc_flush(). | |
3300 | */ | |
3301 | static uint64_t | |
3302 | arc_evict_state(arc_state_t *state, uint64_t spa, int64_t bytes, | |
3303 | arc_buf_contents_t type) | |
34dc7c2f | 3304 | { |
ca0bf58d PS |
3305 | uint64_t total_evicted = 0; |
3306 | multilist_t *ml = &state->arcs_list[type]; | |
3307 | int num_sublists; | |
3308 | arc_buf_hdr_t **markers; | |
3309 | int i; | |
3310 | ||
96c080cb | 3311 | IMPLY(bytes < 0, bytes == ARC_EVICT_ALL); |
ca0bf58d PS |
3312 | |
3313 | num_sublists = multilist_get_num_sublists(ml); | |
d164b209 BB |
3314 | |
3315 | /* | |
ca0bf58d PS |
3316 | * If we've tried to evict from each sublist, made some |
3317 | * progress, but still have not hit the target number of bytes | |
3318 | * to evict, we want to keep trying. The markers allow us to | |
3319 | * pick up where we left off for each individual sublist, rather | |
3320 | * than starting from the tail each time. | |
d164b209 | 3321 | */ |
ca0bf58d PS |
3322 | markers = kmem_zalloc(sizeof (*markers) * num_sublists, KM_SLEEP); |
3323 | for (i = 0; i < num_sublists; i++) { | |
3324 | multilist_sublist_t *mls; | |
34dc7c2f | 3325 | |
ca0bf58d PS |
3326 | markers[i] = kmem_cache_alloc(hdr_full_cache, KM_SLEEP); |
3327 | ||
3328 | /* | |
3329 | * A b_spa of 0 is used to indicate that this header is | |
3330 | * a marker. This fact is used in arc_adjust_type() and | |
3331 | * arc_evict_state_impl(). | |
3332 | */ | |
3333 | markers[i]->b_spa = 0; | |
34dc7c2f | 3334 | |
ca0bf58d PS |
3335 | mls = multilist_sublist_lock(ml, i); |
3336 | multilist_sublist_insert_tail(mls, markers[i]); | |
3337 | multilist_sublist_unlock(mls); | |
34dc7c2f BB |
3338 | } |
3339 | ||
d164b209 | 3340 | /* |
ca0bf58d PS |
3341 | * While we haven't hit our target number of bytes to evict, or |
3342 | * we're evicting all available buffers. | |
d164b209 | 3343 | */ |
ca0bf58d | 3344 | while (total_evicted < bytes || bytes == ARC_EVICT_ALL) { |
25458cbe TC |
3345 | int sublist_idx = multilist_get_random_index(ml); |
3346 | uint64_t scan_evicted = 0; | |
3347 | ||
3348 | /* | |
3349 | * Try to reduce pinned dnodes with a floor of arc_dnode_limit. | |
3350 | * Request that 10% of the LRUs be scanned by the superblock | |
3351 | * shrinker. | |
3352 | */ | |
3353 | if (type == ARC_BUFC_DATA && arc_dnode_size > arc_dnode_limit) | |
3354 | arc_prune_async((arc_dnode_size - arc_dnode_limit) / | |
3355 | sizeof (dnode_t) / zfs_arc_dnode_reduce_percent); | |
3356 | ||
ca0bf58d PS |
3357 | /* |
3358 | * Start eviction using a randomly selected sublist, | |
3359 | * this is to try and evenly balance eviction across all | |
3360 | * sublists. Always starting at the same sublist | |
3361 | * (e.g. index 0) would cause evictions to favor certain | |
3362 | * sublists over others. | |
3363 | */ | |
ca0bf58d PS |
3364 | for (i = 0; i < num_sublists; i++) { |
3365 | uint64_t bytes_remaining; | |
3366 | uint64_t bytes_evicted; | |
d164b209 | 3367 | |
ca0bf58d PS |
3368 | if (bytes == ARC_EVICT_ALL) |
3369 | bytes_remaining = ARC_EVICT_ALL; | |
3370 | else if (total_evicted < bytes) | |
3371 | bytes_remaining = bytes - total_evicted; | |
3372 | else | |
3373 | break; | |
34dc7c2f | 3374 | |
ca0bf58d PS |
3375 | bytes_evicted = arc_evict_state_impl(ml, sublist_idx, |
3376 | markers[sublist_idx], spa, bytes_remaining); | |
3377 | ||
3378 | scan_evicted += bytes_evicted; | |
3379 | total_evicted += bytes_evicted; | |
3380 | ||
3381 | /* we've reached the end, wrap to the beginning */ | |
3382 | if (++sublist_idx >= num_sublists) | |
3383 | sublist_idx = 0; | |
3384 | } | |
3385 | ||
3386 | /* | |
3387 | * If we didn't evict anything during this scan, we have | |
3388 | * no reason to believe we'll evict more during another | |
3389 | * scan, so break the loop. | |
3390 | */ | |
3391 | if (scan_evicted == 0) { | |
3392 | /* This isn't possible, let's make that obvious */ | |
3393 | ASSERT3S(bytes, !=, 0); | |
34dc7c2f | 3394 | |
ca0bf58d PS |
3395 | /* |
3396 | * When bytes is ARC_EVICT_ALL, the only way to | |
3397 | * break the loop is when scan_evicted is zero. | |
3398 | * In that case, we actually have evicted enough, | |
3399 | * so we don't want to increment the kstat. | |
3400 | */ | |
3401 | if (bytes != ARC_EVICT_ALL) { | |
3402 | ASSERT3S(total_evicted, <, bytes); | |
3403 | ARCSTAT_BUMP(arcstat_evict_not_enough); | |
3404 | } | |
d164b209 | 3405 | |
ca0bf58d PS |
3406 | break; |
3407 | } | |
d164b209 | 3408 | } |
34dc7c2f | 3409 | |
ca0bf58d PS |
3410 | for (i = 0; i < num_sublists; i++) { |
3411 | multilist_sublist_t *mls = multilist_sublist_lock(ml, i); | |
3412 | multilist_sublist_remove(mls, markers[i]); | |
3413 | multilist_sublist_unlock(mls); | |
34dc7c2f | 3414 | |
ca0bf58d | 3415 | kmem_cache_free(hdr_full_cache, markers[i]); |
34dc7c2f | 3416 | } |
ca0bf58d PS |
3417 | kmem_free(markers, sizeof (*markers) * num_sublists); |
3418 | ||
3419 | return (total_evicted); | |
3420 | } | |
3421 | ||
3422 | /* | |
3423 | * Flush all "evictable" data of the given type from the arc state | |
3424 | * specified. This will not evict any "active" buffers (i.e. referenced). | |
3425 | * | |
d3c2ae1c | 3426 | * When 'retry' is set to B_FALSE, the function will make a single pass |
ca0bf58d PS |
3427 | * over the state and evict any buffers that it can. Since it doesn't |
3428 | * continually retry the eviction, it might end up leaving some buffers | |
3429 | * in the ARC due to lock misses. | |
3430 | * | |
d3c2ae1c | 3431 | * When 'retry' is set to B_TRUE, the function will continually retry the |
ca0bf58d PS |
3432 | * eviction until *all* evictable buffers have been removed from the |
3433 | * state. As a result, if concurrent insertions into the state are | |
3434 | * allowed (e.g. if the ARC isn't shutting down), this function might | |
3435 | * wind up in an infinite loop, continually trying to evict buffers. | |
3436 | */ | |
3437 | static uint64_t | |
3438 | arc_flush_state(arc_state_t *state, uint64_t spa, arc_buf_contents_t type, | |
3439 | boolean_t retry) | |
3440 | { | |
3441 | uint64_t evicted = 0; | |
3442 | ||
d3c2ae1c | 3443 | while (refcount_count(&state->arcs_esize[type]) != 0) { |
ca0bf58d PS |
3444 | evicted += arc_evict_state(state, spa, ARC_EVICT_ALL, type); |
3445 | ||
3446 | if (!retry) | |
3447 | break; | |
3448 | } | |
3449 | ||
3450 | return (evicted); | |
34dc7c2f BB |
3451 | } |
3452 | ||
ab26409d | 3453 | /* |
ef5b2e10 BB |
3454 | * Helper function for arc_prune_async() it is responsible for safely |
3455 | * handling the execution of a registered arc_prune_func_t. | |
ab26409d BB |
3456 | */ |
3457 | static void | |
f6046738 | 3458 | arc_prune_task(void *ptr) |
ab26409d | 3459 | { |
f6046738 BB |
3460 | arc_prune_t *ap = (arc_prune_t *)ptr; |
3461 | arc_prune_func_t *func = ap->p_pfunc; | |
ab26409d | 3462 | |
f6046738 BB |
3463 | if (func != NULL) |
3464 | func(ap->p_adjust, ap->p_private); | |
ab26409d | 3465 | |
4442f60d | 3466 | refcount_remove(&ap->p_refcnt, func); |
f6046738 | 3467 | } |
ab26409d | 3468 | |
f6046738 BB |
3469 | /* |
3470 | * Notify registered consumers they must drop holds on a portion of the ARC | |
3471 | * buffered they reference. This provides a mechanism to ensure the ARC can | |
3472 | * honor the arc_meta_limit and reclaim otherwise pinned ARC buffers. This | |
3473 | * is analogous to dnlc_reduce_cache() but more generic. | |
3474 | * | |
ef5b2e10 | 3475 | * This operation is performed asynchronously so it may be safely called |
ca67b33a | 3476 | * in the context of the arc_reclaim_thread(). A reference is taken here |
f6046738 BB |
3477 | * for each registered arc_prune_t and the arc_prune_task() is responsible |
3478 | * for releasing it once the registered arc_prune_func_t has completed. | |
3479 | */ | |
3480 | static void | |
3481 | arc_prune_async(int64_t adjust) | |
3482 | { | |
3483 | arc_prune_t *ap; | |
ab26409d | 3484 | |
f6046738 BB |
3485 | mutex_enter(&arc_prune_mtx); |
3486 | for (ap = list_head(&arc_prune_list); ap != NULL; | |
3487 | ap = list_next(&arc_prune_list, ap)) { | |
ab26409d | 3488 | |
f6046738 BB |
3489 | if (refcount_count(&ap->p_refcnt) >= 2) |
3490 | continue; | |
ab26409d | 3491 | |
f6046738 BB |
3492 | refcount_add(&ap->p_refcnt, ap->p_pfunc); |
3493 | ap->p_adjust = adjust; | |
b60eac3d | 3494 | if (taskq_dispatch(arc_prune_taskq, arc_prune_task, |
48d3eb40 | 3495 | ap, TQ_SLEEP) == TASKQID_INVALID) { |
b60eac3d | 3496 | refcount_remove(&ap->p_refcnt, ap->p_pfunc); |
3497 | continue; | |
3498 | } | |
f6046738 | 3499 | ARCSTAT_BUMP(arcstat_prune); |
ab26409d | 3500 | } |
ab26409d BB |
3501 | mutex_exit(&arc_prune_mtx); |
3502 | } | |
3503 | ||
ca0bf58d PS |
3504 | /* |
3505 | * Evict the specified number of bytes from the state specified, | |
3506 | * restricting eviction to the spa and type given. This function | |
3507 | * prevents us from trying to evict more from a state's list than | |
3508 | * is "evictable", and to skip evicting altogether when passed a | |
3509 | * negative value for "bytes". In contrast, arc_evict_state() will | |
3510 | * evict everything it can, when passed a negative value for "bytes". | |
3511 | */ | |
3512 | static uint64_t | |
3513 | arc_adjust_impl(arc_state_t *state, uint64_t spa, int64_t bytes, | |
3514 | arc_buf_contents_t type) | |
3515 | { | |
3516 | int64_t delta; | |
3517 | ||
d3c2ae1c GW |
3518 | if (bytes > 0 && refcount_count(&state->arcs_esize[type]) > 0) { |
3519 | delta = MIN(refcount_count(&state->arcs_esize[type]), bytes); | |
ca0bf58d PS |
3520 | return (arc_evict_state(state, spa, delta, type)); |
3521 | } | |
3522 | ||
3523 | return (0); | |
3524 | } | |
3525 | ||
3526 | /* | |
3527 | * The goal of this function is to evict enough meta data buffers from the | |
3528 | * ARC in order to enforce the arc_meta_limit. Achieving this is slightly | |
3529 | * more complicated than it appears because it is common for data buffers | |
3530 | * to have holds on meta data buffers. In addition, dnode meta data buffers | |
3531 | * will be held by the dnodes in the block preventing them from being freed. | |
3532 | * This means we can't simply traverse the ARC and expect to always find | |
3533 | * enough unheld meta data buffer to release. | |
3534 | * | |
3535 | * Therefore, this function has been updated to make alternating passes | |
3536 | * over the ARC releasing data buffers and then newly unheld meta data | |
3537 | * buffers. This ensures forward progress is maintained and arc_meta_used | |
3538 | * will decrease. Normally this is sufficient, but if required the ARC | |
3539 | * will call the registered prune callbacks causing dentry and inodes to | |
3540 | * be dropped from the VFS cache. This will make dnode meta data buffers | |
3541 | * available for reclaim. | |
3542 | */ | |
3543 | static uint64_t | |
f6046738 | 3544 | arc_adjust_meta_balanced(void) |
ca0bf58d | 3545 | { |
25e2ab16 TC |
3546 | int64_t delta, prune = 0, adjustmnt; |
3547 | uint64_t total_evicted = 0; | |
ca0bf58d | 3548 | arc_buf_contents_t type = ARC_BUFC_DATA; |
ca67b33a | 3549 | int restarts = MAX(zfs_arc_meta_adjust_restarts, 0); |
ca0bf58d PS |
3550 | |
3551 | restart: | |
3552 | /* | |
3553 | * This slightly differs than the way we evict from the mru in | |
3554 | * arc_adjust because we don't have a "target" value (i.e. no | |
3555 | * "meta" arc_p). As a result, I think we can completely | |
3556 | * cannibalize the metadata in the MRU before we evict the | |
3557 | * metadata from the MFU. I think we probably need to implement a | |
3558 | * "metadata arc_p" value to do this properly. | |
3559 | */ | |
3560 | adjustmnt = arc_meta_used - arc_meta_limit; | |
3561 | ||
d3c2ae1c GW |
3562 | if (adjustmnt > 0 && refcount_count(&arc_mru->arcs_esize[type]) > 0) { |
3563 | delta = MIN(refcount_count(&arc_mru->arcs_esize[type]), | |
3564 | adjustmnt); | |
ca0bf58d PS |
3565 | total_evicted += arc_adjust_impl(arc_mru, 0, delta, type); |
3566 | adjustmnt -= delta; | |
3567 | } | |
3568 | ||
3569 | /* | |
3570 | * We can't afford to recalculate adjustmnt here. If we do, | |
3571 | * new metadata buffers can sneak into the MRU or ANON lists, | |
3572 | * thus penalize the MFU metadata. Although the fudge factor is | |
3573 | * small, it has been empirically shown to be significant for | |
3574 | * certain workloads (e.g. creating many empty directories). As | |
3575 | * such, we use the original calculation for adjustmnt, and | |
3576 | * simply decrement the amount of data evicted from the MRU. | |
3577 | */ | |
3578 | ||
d3c2ae1c GW |
3579 | if (adjustmnt > 0 && refcount_count(&arc_mfu->arcs_esize[type]) > 0) { |
3580 | delta = MIN(refcount_count(&arc_mfu->arcs_esize[type]), | |
3581 | adjustmnt); | |
ca0bf58d PS |
3582 | total_evicted += arc_adjust_impl(arc_mfu, 0, delta, type); |
3583 | } | |
3584 | ||
3585 | adjustmnt = arc_meta_used - arc_meta_limit; | |
3586 | ||
d3c2ae1c GW |
3587 | if (adjustmnt > 0 && |
3588 | refcount_count(&arc_mru_ghost->arcs_esize[type]) > 0) { | |
ca0bf58d | 3589 | delta = MIN(adjustmnt, |
d3c2ae1c | 3590 | refcount_count(&arc_mru_ghost->arcs_esize[type])); |
ca0bf58d PS |
3591 | total_evicted += arc_adjust_impl(arc_mru_ghost, 0, delta, type); |
3592 | adjustmnt -= delta; | |
3593 | } | |
3594 | ||
d3c2ae1c GW |
3595 | if (adjustmnt > 0 && |
3596 | refcount_count(&arc_mfu_ghost->arcs_esize[type]) > 0) { | |
ca0bf58d | 3597 | delta = MIN(adjustmnt, |
d3c2ae1c | 3598 | refcount_count(&arc_mfu_ghost->arcs_esize[type])); |
ca0bf58d PS |
3599 | total_evicted += arc_adjust_impl(arc_mfu_ghost, 0, delta, type); |
3600 | } | |
3601 | ||
3602 | /* | |
3603 | * If after attempting to make the requested adjustment to the ARC | |
3604 | * the meta limit is still being exceeded then request that the | |
3605 | * higher layers drop some cached objects which have holds on ARC | |
3606 | * meta buffers. Requests to the upper layers will be made with | |
3607 | * increasingly large scan sizes until the ARC is below the limit. | |
3608 | */ | |
3609 | if (arc_meta_used > arc_meta_limit) { | |
3610 | if (type == ARC_BUFC_DATA) { | |
3611 | type = ARC_BUFC_METADATA; | |
3612 | } else { | |
3613 | type = ARC_BUFC_DATA; | |
3614 | ||
3615 | if (zfs_arc_meta_prune) { | |
3616 | prune += zfs_arc_meta_prune; | |
f6046738 | 3617 | arc_prune_async(prune); |
ca0bf58d PS |
3618 | } |
3619 | } | |
3620 | ||
3621 | if (restarts > 0) { | |
3622 | restarts--; | |
3623 | goto restart; | |
3624 | } | |
3625 | } | |
3626 | return (total_evicted); | |
3627 | } | |
3628 | ||
f6046738 BB |
3629 | /* |
3630 | * Evict metadata buffers from the cache, such that arc_meta_used is | |
3631 | * capped by the arc_meta_limit tunable. | |
3632 | */ | |
3633 | static uint64_t | |
3634 | arc_adjust_meta_only(void) | |
3635 | { | |
3636 | uint64_t total_evicted = 0; | |
3637 | int64_t target; | |
3638 | ||
3639 | /* | |
3640 | * If we're over the meta limit, we want to evict enough | |
3641 | * metadata to get back under the meta limit. We don't want to | |
3642 | * evict so much that we drop the MRU below arc_p, though. If | |
3643 | * we're over the meta limit more than we're over arc_p, we | |
3644 | * evict some from the MRU here, and some from the MFU below. | |
3645 | */ | |
3646 | target = MIN((int64_t)(arc_meta_used - arc_meta_limit), | |
36da08ef PS |
3647 | (int64_t)(refcount_count(&arc_anon->arcs_size) + |
3648 | refcount_count(&arc_mru->arcs_size) - arc_p)); | |
f6046738 BB |
3649 | |
3650 | total_evicted += arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_METADATA); | |
3651 | ||
3652 | /* | |
3653 | * Similar to the above, we want to evict enough bytes to get us | |
3654 | * below the meta limit, but not so much as to drop us below the | |
2aa34383 | 3655 | * space allotted to the MFU (which is defined as arc_c - arc_p). |
f6046738 BB |
3656 | */ |
3657 | target = MIN((int64_t)(arc_meta_used - arc_meta_limit), | |
36da08ef | 3658 | (int64_t)(refcount_count(&arc_mfu->arcs_size) - (arc_c - arc_p))); |
f6046738 BB |
3659 | |
3660 | total_evicted += arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_METADATA); | |
3661 | ||
3662 | return (total_evicted); | |
3663 | } | |
3664 | ||
3665 | static uint64_t | |
3666 | arc_adjust_meta(void) | |
3667 | { | |
3668 | if (zfs_arc_meta_strategy == ARC_STRATEGY_META_ONLY) | |
3669 | return (arc_adjust_meta_only()); | |
3670 | else | |
3671 | return (arc_adjust_meta_balanced()); | |
3672 | } | |
3673 | ||
ca0bf58d PS |
3674 | /* |
3675 | * Return the type of the oldest buffer in the given arc state | |
3676 | * | |
3677 | * This function will select a random sublist of type ARC_BUFC_DATA and | |
3678 | * a random sublist of type ARC_BUFC_METADATA. The tail of each sublist | |
3679 | * is compared, and the type which contains the "older" buffer will be | |
3680 | * returned. | |
3681 | */ | |
3682 | static arc_buf_contents_t | |
3683 | arc_adjust_type(arc_state_t *state) | |
3684 | { | |
3685 | multilist_t *data_ml = &state->arcs_list[ARC_BUFC_DATA]; | |
3686 | multilist_t *meta_ml = &state->arcs_list[ARC_BUFC_METADATA]; | |
3687 | int data_idx = multilist_get_random_index(data_ml); | |
3688 | int meta_idx = multilist_get_random_index(meta_ml); | |
3689 | multilist_sublist_t *data_mls; | |
3690 | multilist_sublist_t *meta_mls; | |
3691 | arc_buf_contents_t type; | |
3692 | arc_buf_hdr_t *data_hdr; | |
3693 | arc_buf_hdr_t *meta_hdr; | |
3694 | ||
3695 | /* | |
3696 | * We keep the sublist lock until we're finished, to prevent | |
3697 | * the headers from being destroyed via arc_evict_state(). | |
3698 | */ | |
3699 | data_mls = multilist_sublist_lock(data_ml, data_idx); | |
3700 | meta_mls = multilist_sublist_lock(meta_ml, meta_idx); | |
3701 | ||
3702 | /* | |
3703 | * These two loops are to ensure we skip any markers that | |
3704 | * might be at the tail of the lists due to arc_evict_state(). | |
3705 | */ | |
3706 | ||
3707 | for (data_hdr = multilist_sublist_tail(data_mls); data_hdr != NULL; | |
3708 | data_hdr = multilist_sublist_prev(data_mls, data_hdr)) { | |
3709 | if (data_hdr->b_spa != 0) | |
3710 | break; | |
3711 | } | |
3712 | ||
3713 | for (meta_hdr = multilist_sublist_tail(meta_mls); meta_hdr != NULL; | |
3714 | meta_hdr = multilist_sublist_prev(meta_mls, meta_hdr)) { | |
3715 | if (meta_hdr->b_spa != 0) | |
3716 | break; | |
3717 | } | |
3718 | ||
3719 | if (data_hdr == NULL && meta_hdr == NULL) { | |
3720 | type = ARC_BUFC_DATA; | |
3721 | } else if (data_hdr == NULL) { | |
3722 | ASSERT3P(meta_hdr, !=, NULL); | |
3723 | type = ARC_BUFC_METADATA; | |
3724 | } else if (meta_hdr == NULL) { | |
3725 | ASSERT3P(data_hdr, !=, NULL); | |
3726 | type = ARC_BUFC_DATA; | |
3727 | } else { | |
3728 | ASSERT3P(data_hdr, !=, NULL); | |
3729 | ASSERT3P(meta_hdr, !=, NULL); | |
3730 | ||
3731 | /* The headers can't be on the sublist without an L1 header */ | |
3732 | ASSERT(HDR_HAS_L1HDR(data_hdr)); | |
3733 | ASSERT(HDR_HAS_L1HDR(meta_hdr)); | |
3734 | ||
3735 | if (data_hdr->b_l1hdr.b_arc_access < | |
3736 | meta_hdr->b_l1hdr.b_arc_access) { | |
3737 | type = ARC_BUFC_DATA; | |
3738 | } else { | |
3739 | type = ARC_BUFC_METADATA; | |
3740 | } | |
3741 | } | |
3742 | ||
3743 | multilist_sublist_unlock(meta_mls); | |
3744 | multilist_sublist_unlock(data_mls); | |
3745 | ||
3746 | return (type); | |
3747 | } | |
3748 | ||
3749 | /* | |
3750 | * Evict buffers from the cache, such that arc_size is capped by arc_c. | |
3751 | */ | |
3752 | static uint64_t | |
3753 | arc_adjust(void) | |
3754 | { | |
3755 | uint64_t total_evicted = 0; | |
3756 | uint64_t bytes; | |
3757 | int64_t target; | |
3758 | ||
3759 | /* | |
3760 | * If we're over arc_meta_limit, we want to correct that before | |
3761 | * potentially evicting data buffers below. | |
3762 | */ | |
3763 | total_evicted += arc_adjust_meta(); | |
3764 | ||
3765 | /* | |
3766 | * Adjust MRU size | |
3767 | * | |
3768 | * If we're over the target cache size, we want to evict enough | |
3769 | * from the list to get back to our target size. We don't want | |
3770 | * to evict too much from the MRU, such that it drops below | |
3771 | * arc_p. So, if we're over our target cache size more than | |
3772 | * the MRU is over arc_p, we'll evict enough to get back to | |
3773 | * arc_p here, and then evict more from the MFU below. | |
3774 | */ | |
3775 | target = MIN((int64_t)(arc_size - arc_c), | |
36da08ef PS |
3776 | (int64_t)(refcount_count(&arc_anon->arcs_size) + |
3777 | refcount_count(&arc_mru->arcs_size) + arc_meta_used - arc_p)); | |
ca0bf58d PS |
3778 | |
3779 | /* | |
3780 | * If we're below arc_meta_min, always prefer to evict data. | |
3781 | * Otherwise, try to satisfy the requested number of bytes to | |
3782 | * evict from the type which contains older buffers; in an | |
3783 | * effort to keep newer buffers in the cache regardless of their | |
3784 | * type. If we cannot satisfy the number of bytes from this | |
3785 | * type, spill over into the next type. | |
3786 | */ | |
3787 | if (arc_adjust_type(arc_mru) == ARC_BUFC_METADATA && | |
3788 | arc_meta_used > arc_meta_min) { | |
3789 | bytes = arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_METADATA); | |
3790 | total_evicted += bytes; | |
3791 | ||
3792 | /* | |
3793 | * If we couldn't evict our target number of bytes from | |
3794 | * metadata, we try to get the rest from data. | |
3795 | */ | |
3796 | target -= bytes; | |
3797 | ||
3798 | total_evicted += | |
3799 | arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_DATA); | |
3800 | } else { | |
3801 | bytes = arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_DATA); | |
3802 | total_evicted += bytes; | |
3803 | ||
3804 | /* | |
3805 | * If we couldn't evict our target number of bytes from | |
3806 | * data, we try to get the rest from metadata. | |
3807 | */ | |
3808 | target -= bytes; | |
3809 | ||
3810 | total_evicted += | |
3811 | arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_METADATA); | |
3812 | } | |
3813 | ||
3814 | /* | |
3815 | * Adjust MFU size | |
3816 | * | |
3817 | * Now that we've tried to evict enough from the MRU to get its | |
3818 | * size back to arc_p, if we're still above the target cache | |
3819 | * size, we evict the rest from the MFU. | |
3820 | */ | |
3821 | target = arc_size - arc_c; | |
3822 | ||
a7b10a93 | 3823 | if (arc_adjust_type(arc_mfu) == ARC_BUFC_METADATA && |
ca0bf58d PS |
3824 | arc_meta_used > arc_meta_min) { |
3825 | bytes = arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_METADATA); | |
3826 | total_evicted += bytes; | |
3827 | ||
3828 | /* | |
3829 | * If we couldn't evict our target number of bytes from | |
3830 | * metadata, we try to get the rest from data. | |
3831 | */ | |
3832 | target -= bytes; | |
3833 | ||
3834 | total_evicted += | |
3835 | arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_DATA); | |
3836 | } else { | |
3837 | bytes = arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_DATA); | |
3838 | total_evicted += bytes; | |
3839 | ||
3840 | /* | |
3841 | * If we couldn't evict our target number of bytes from | |
3842 | * data, we try to get the rest from data. | |
3843 | */ | |
3844 | target -= bytes; | |
3845 | ||
3846 | total_evicted += | |
3847 | arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_METADATA); | |
3848 | } | |
3849 | ||
3850 | /* | |
3851 | * Adjust ghost lists | |
3852 | * | |
3853 | * In addition to the above, the ARC also defines target values | |
3854 | * for the ghost lists. The sum of the mru list and mru ghost | |
3855 | * list should never exceed the target size of the cache, and | |
3856 | * the sum of the mru list, mfu list, mru ghost list, and mfu | |
3857 | * ghost list should never exceed twice the target size of the | |
3858 | * cache. The following logic enforces these limits on the ghost | |
3859 | * caches, and evicts from them as needed. | |
3860 | */ | |
36da08ef PS |
3861 | target = refcount_count(&arc_mru->arcs_size) + |
3862 | refcount_count(&arc_mru_ghost->arcs_size) - arc_c; | |
ca0bf58d PS |
3863 | |
3864 | bytes = arc_adjust_impl(arc_mru_ghost, 0, target, ARC_BUFC_DATA); | |
3865 | total_evicted += bytes; | |
3866 | ||
3867 | target -= bytes; | |
3868 | ||
3869 | total_evicted += | |
3870 | arc_adjust_impl(arc_mru_ghost, 0, target, ARC_BUFC_METADATA); | |
3871 | ||
3872 | /* | |
3873 | * We assume the sum of the mru list and mfu list is less than | |
3874 | * or equal to arc_c (we enforced this above), which means we | |
3875 | * can use the simpler of the two equations below: | |
3876 | * | |
3877 | * mru + mfu + mru ghost + mfu ghost <= 2 * arc_c | |
3878 | * mru ghost + mfu ghost <= arc_c | |
3879 | */ | |
36da08ef PS |
3880 | target = refcount_count(&arc_mru_ghost->arcs_size) + |
3881 | refcount_count(&arc_mfu_ghost->arcs_size) - arc_c; | |
ca0bf58d PS |
3882 | |
3883 | bytes = arc_adjust_impl(arc_mfu_ghost, 0, target, ARC_BUFC_DATA); | |
3884 | total_evicted += bytes; | |
3885 | ||
3886 | target -= bytes; | |
3887 | ||
3888 | total_evicted += | |
3889 | arc_adjust_impl(arc_mfu_ghost, 0, target, ARC_BUFC_METADATA); | |
3890 | ||
3891 | return (total_evicted); | |
3892 | } | |
3893 | ||
ca0bf58d PS |
3894 | void |
3895 | arc_flush(spa_t *spa, boolean_t retry) | |
ab26409d | 3896 | { |
ca0bf58d | 3897 | uint64_t guid = 0; |
94520ca4 | 3898 | |
bc888666 | 3899 | /* |
d3c2ae1c | 3900 | * If retry is B_TRUE, a spa must not be specified since we have |
ca0bf58d PS |
3901 | * no good way to determine if all of a spa's buffers have been |
3902 | * evicted from an arc state. | |
bc888666 | 3903 | */ |
ca0bf58d | 3904 | ASSERT(!retry || spa == 0); |
d164b209 | 3905 | |
b9541d6b | 3906 | if (spa != NULL) |
3541dc6d | 3907 | guid = spa_load_guid(spa); |
d164b209 | 3908 | |
ca0bf58d PS |
3909 | (void) arc_flush_state(arc_mru, guid, ARC_BUFC_DATA, retry); |
3910 | (void) arc_flush_state(arc_mru, guid, ARC_BUFC_METADATA, retry); | |
3911 | ||
3912 | (void) arc_flush_state(arc_mfu, guid, ARC_BUFC_DATA, retry); | |
3913 | (void) arc_flush_state(arc_mfu, guid, ARC_BUFC_METADATA, retry); | |
3914 | ||
3915 | (void) arc_flush_state(arc_mru_ghost, guid, ARC_BUFC_DATA, retry); | |
3916 | (void) arc_flush_state(arc_mru_ghost, guid, ARC_BUFC_METADATA, retry); | |
34dc7c2f | 3917 | |
ca0bf58d PS |
3918 | (void) arc_flush_state(arc_mfu_ghost, guid, ARC_BUFC_DATA, retry); |
3919 | (void) arc_flush_state(arc_mfu_ghost, guid, ARC_BUFC_METADATA, retry); | |
34dc7c2f BB |
3920 | } |
3921 | ||
34dc7c2f | 3922 | void |
ca67b33a | 3923 | arc_shrink(int64_t to_free) |
34dc7c2f | 3924 | { |
1b8951b3 | 3925 | uint64_t c = arc_c; |
34dc7c2f | 3926 | |
1b8951b3 TC |
3927 | if (c > to_free && c - to_free > arc_c_min) { |
3928 | arc_c = c - to_free; | |
ca67b33a | 3929 | atomic_add_64(&arc_p, -(arc_p >> arc_shrink_shift)); |
34dc7c2f BB |
3930 | if (arc_c > arc_size) |
3931 | arc_c = MAX(arc_size, arc_c_min); | |
3932 | if (arc_p > arc_c) | |
3933 | arc_p = (arc_c >> 1); | |
3934 | ASSERT(arc_c >= arc_c_min); | |
3935 | ASSERT((int64_t)arc_p >= 0); | |
1b8951b3 TC |
3936 | } else { |
3937 | arc_c = arc_c_min; | |
34dc7c2f BB |
3938 | } |
3939 | ||
3940 | if (arc_size > arc_c) | |
ca0bf58d | 3941 | (void) arc_adjust(); |
34dc7c2f BB |
3942 | } |
3943 | ||
9edb3695 BB |
3944 | /* |
3945 | * Return maximum amount of memory that we could possibly use. Reduced | |
3946 | * to half of all memory in user space which is primarily used for testing. | |
3947 | */ | |
3948 | static uint64_t | |
3949 | arc_all_memory(void) | |
3950 | { | |
3951 | #ifdef _KERNEL | |
3952 | return (MIN(ptob(physmem), | |
3953 | vmem_size(heap_arena, VMEM_FREE | VMEM_ALLOC))); | |
3954 | #else | |
3955 | return (ptob(physmem) / 2); | |
3956 | #endif | |
3957 | } | |
3958 | ||
ca67b33a MA |
3959 | typedef enum free_memory_reason_t { |
3960 | FMR_UNKNOWN, | |
3961 | FMR_NEEDFREE, | |
3962 | FMR_LOTSFREE, | |
3963 | FMR_SWAPFS_MINFREE, | |
3964 | FMR_PAGES_PP_MAXIMUM, | |
3965 | FMR_HEAP_ARENA, | |
3966 | FMR_ZIO_ARENA, | |
3967 | } free_memory_reason_t; | |
3968 | ||
3969 | int64_t last_free_memory; | |
3970 | free_memory_reason_t last_free_reason; | |
3971 | ||
3972 | #ifdef _KERNEL | |
ca67b33a MA |
3973 | /* |
3974 | * Additional reserve of pages for pp_reserve. | |
3975 | */ | |
3976 | int64_t arc_pages_pp_reserve = 64; | |
3977 | ||
3978 | /* | |
3979 | * Additional reserve of pages for swapfs. | |
3980 | */ | |
3981 | int64_t arc_swapfs_reserve = 64; | |
ca67b33a MA |
3982 | #endif /* _KERNEL */ |
3983 | ||
3984 | /* | |
3985 | * Return the amount of memory that can be consumed before reclaim will be | |
3986 | * needed. Positive if there is sufficient free memory, negative indicates | |
3987 | * the amount of memory that needs to be freed up. | |
3988 | */ | |
3989 | static int64_t | |
3990 | arc_available_memory(void) | |
3991 | { | |
3992 | int64_t lowest = INT64_MAX; | |
3993 | free_memory_reason_t r = FMR_UNKNOWN; | |
ca67b33a | 3994 | #ifdef _KERNEL |
9edb3695 | 3995 | uint64_t available_memory = ptob(freemem); |
ca67b33a | 3996 | int64_t n; |
11f552fa BB |
3997 | #ifdef __linux__ |
3998 | pgcnt_t needfree = btop(arc_need_free); | |
3999 | pgcnt_t lotsfree = btop(arc_sys_free); | |
4000 | pgcnt_t desfree = 0; | |
4001 | #endif | |
ca67b33a | 4002 | |
9edb3695 BB |
4003 | #if defined(__i386) |
4004 | available_memory = | |
4005 | MIN(available_memory, vmem_size(heap_arena, VMEM_FREE)); | |
4006 | #endif | |
4007 | ||
ca67b33a MA |
4008 | if (needfree > 0) { |
4009 | n = PAGESIZE * (-needfree); | |
4010 | if (n < lowest) { | |
4011 | lowest = n; | |
4012 | r = FMR_NEEDFREE; | |
4013 | } | |
4014 | } | |
4015 | ||
4016 | /* | |
4017 | * check that we're out of range of the pageout scanner. It starts to | |
4018 | * schedule paging if freemem is less than lotsfree and needfree. | |
4019 | * lotsfree is the high-water mark for pageout, and needfree is the | |
4020 | * number of needed free pages. We add extra pages here to make sure | |
4021 | * the scanner doesn't start up while we're freeing memory. | |
4022 | */ | |
9edb3695 | 4023 | n = PAGESIZE * (btop(available_memory) - lotsfree - needfree - desfree); |
ca67b33a MA |
4024 | if (n < lowest) { |
4025 | lowest = n; | |
4026 | r = FMR_LOTSFREE; | |
4027 | } | |
4028 | ||
11f552fa | 4029 | #ifndef __linux__ |
ca67b33a MA |
4030 | /* |
4031 | * check to make sure that swapfs has enough space so that anon | |
4032 | * reservations can still succeed. anon_resvmem() checks that the | |
4033 | * availrmem is greater than swapfs_minfree, and the number of reserved | |
4034 | * swap pages. We also add a bit of extra here just to prevent | |
4035 | * circumstances from getting really dire. | |
4036 | */ | |
4037 | n = PAGESIZE * (availrmem - swapfs_minfree - swapfs_reserve - | |
4038 | desfree - arc_swapfs_reserve); | |
4039 | if (n < lowest) { | |
4040 | lowest = n; | |
4041 | r = FMR_SWAPFS_MINFREE; | |
4042 | } | |
4043 | ||
4044 | ||
4045 | /* | |
4046 | * Check that we have enough availrmem that memory locking (e.g., via | |
4047 | * mlock(3C) or memcntl(2)) can still succeed. (pages_pp_maximum | |
4048 | * stores the number of pages that cannot be locked; when availrmem | |
4049 | * drops below pages_pp_maximum, page locking mechanisms such as | |
4050 | * page_pp_lock() will fail.) | |
4051 | */ | |
4052 | n = PAGESIZE * (availrmem - pages_pp_maximum - | |
4053 | arc_pages_pp_reserve); | |
4054 | if (n < lowest) { | |
4055 | lowest = n; | |
4056 | r = FMR_PAGES_PP_MAXIMUM; | |
4057 | } | |
11f552fa | 4058 | #endif |
ca67b33a MA |
4059 | |
4060 | #if defined(__i386) | |
4061 | /* | |
4062 | * If we're on an i386 platform, it's possible that we'll exhaust the | |
4063 | * kernel heap space before we ever run out of available physical | |
4064 | * memory. Most checks of the size of the heap_area compare against | |
4065 | * tune.t_minarmem, which is the minimum available real memory that we | |
4066 | * can have in the system. However, this is generally fixed at 25 pages | |
4067 | * which is so low that it's useless. In this comparison, we seek to | |
4068 | * calculate the total heap-size, and reclaim if more than 3/4ths of the | |
4069 | * heap is allocated. (Or, in the calculation, if less than 1/4th is | |
4070 | * free) | |
4071 | */ | |
4072 | n = vmem_size(heap_arena, VMEM_FREE) - | |
4073 | (vmem_size(heap_arena, VMEM_FREE | VMEM_ALLOC) >> 2); | |
4074 | if (n < lowest) { | |
4075 | lowest = n; | |
4076 | r = FMR_HEAP_ARENA; | |
4077 | } | |
4078 | #endif | |
4079 | ||
4080 | /* | |
4081 | * If zio data pages are being allocated out of a separate heap segment, | |
4082 | * then enforce that the size of available vmem for this arena remains | |
d3c2ae1c | 4083 | * above about 1/4th (1/(2^arc_zio_arena_free_shift)) free. |
ca67b33a | 4084 | * |
d3c2ae1c GW |
4085 | * Note that reducing the arc_zio_arena_free_shift keeps more virtual |
4086 | * memory (in the zio_arena) free, which can avoid memory | |
4087 | * fragmentation issues. | |
ca67b33a MA |
4088 | */ |
4089 | if (zio_arena != NULL) { | |
9edb3695 BB |
4090 | n = (int64_t)vmem_size(zio_arena, VMEM_FREE) - |
4091 | (vmem_size(zio_arena, VMEM_ALLOC) >> | |
4092 | arc_zio_arena_free_shift); | |
ca67b33a MA |
4093 | if (n < lowest) { |
4094 | lowest = n; | |
4095 | r = FMR_ZIO_ARENA; | |
4096 | } | |
4097 | } | |
11f552fa | 4098 | #else /* _KERNEL */ |
ca67b33a MA |
4099 | /* Every 100 calls, free a small amount */ |
4100 | if (spa_get_random(100) == 0) | |
4101 | lowest = -1024; | |
11f552fa | 4102 | #endif /* _KERNEL */ |
ca67b33a MA |
4103 | |
4104 | last_free_memory = lowest; | |
4105 | last_free_reason = r; | |
4106 | ||
4107 | return (lowest); | |
4108 | } | |
4109 | ||
4110 | /* | |
4111 | * Determine if the system is under memory pressure and is asking | |
d3c2ae1c | 4112 | * to reclaim memory. A return value of B_TRUE indicates that the system |
ca67b33a MA |
4113 | * is under memory pressure and that the arc should adjust accordingly. |
4114 | */ | |
4115 | static boolean_t | |
4116 | arc_reclaim_needed(void) | |
4117 | { | |
4118 | return (arc_available_memory() < 0); | |
4119 | } | |
4120 | ||
34dc7c2f | 4121 | static void |
ca67b33a | 4122 | arc_kmem_reap_now(void) |
34dc7c2f BB |
4123 | { |
4124 | size_t i; | |
4125 | kmem_cache_t *prev_cache = NULL; | |
4126 | kmem_cache_t *prev_data_cache = NULL; | |
4127 | extern kmem_cache_t *zio_buf_cache[]; | |
4128 | extern kmem_cache_t *zio_data_buf_cache[]; | |
669dedb3 | 4129 | extern kmem_cache_t *range_seg_cache; |
34dc7c2f | 4130 | |
f6046738 BB |
4131 | if ((arc_meta_used >= arc_meta_limit) && zfs_arc_meta_prune) { |
4132 | /* | |
4133 | * We are exceeding our meta-data cache limit. | |
4134 | * Prune some entries to release holds on meta-data. | |
4135 | */ | |
ef5b2e10 | 4136 | arc_prune_async(zfs_arc_meta_prune); |
f6046738 BB |
4137 | } |
4138 | ||
34dc7c2f | 4139 | for (i = 0; i < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; i++) { |
d0c614ec | 4140 | #ifdef _ILP32 |
4141 | /* reach upper limit of cache size on 32-bit */ | |
4142 | if (zio_buf_cache[i] == NULL) | |
4143 | break; | |
4144 | #endif | |
34dc7c2f BB |
4145 | if (zio_buf_cache[i] != prev_cache) { |
4146 | prev_cache = zio_buf_cache[i]; | |
4147 | kmem_cache_reap_now(zio_buf_cache[i]); | |
4148 | } | |
4149 | if (zio_data_buf_cache[i] != prev_data_cache) { | |
4150 | prev_data_cache = zio_data_buf_cache[i]; | |
4151 | kmem_cache_reap_now(zio_data_buf_cache[i]); | |
4152 | } | |
4153 | } | |
ca0bf58d | 4154 | kmem_cache_reap_now(buf_cache); |
b9541d6b CW |
4155 | kmem_cache_reap_now(hdr_full_cache); |
4156 | kmem_cache_reap_now(hdr_l2only_cache); | |
669dedb3 | 4157 | kmem_cache_reap_now(range_seg_cache); |
ca67b33a MA |
4158 | |
4159 | if (zio_arena != NULL) { | |
4160 | /* | |
4161 | * Ask the vmem arena to reclaim unused memory from its | |
4162 | * quantum caches. | |
4163 | */ | |
4164 | vmem_qcache_reap(zio_arena); | |
4165 | } | |
34dc7c2f BB |
4166 | } |
4167 | ||
302f753f | 4168 | /* |
a6255b7f | 4169 | * Threads can block in arc_get_data_impl() waiting for this thread to evict |
ca0bf58d | 4170 | * enough data and signal them to proceed. When this happens, the threads in |
a6255b7f | 4171 | * arc_get_data_impl() are sleeping while holding the hash lock for their |
ca0bf58d PS |
4172 | * particular arc header. Thus, we must be careful to never sleep on a |
4173 | * hash lock in this thread. This is to prevent the following deadlock: | |
4174 | * | |
a6255b7f | 4175 | * - Thread A sleeps on CV in arc_get_data_impl() holding hash lock "L", |
ca0bf58d PS |
4176 | * waiting for the reclaim thread to signal it. |
4177 | * | |
4178 | * - arc_reclaim_thread() tries to acquire hash lock "L" using mutex_enter, | |
4179 | * fails, and goes to sleep forever. | |
4180 | * | |
4181 | * This possible deadlock is avoided by always acquiring a hash lock | |
4182 | * using mutex_tryenter() from arc_reclaim_thread(). | |
302f753f | 4183 | */ |
34dc7c2f | 4184 | static void |
ca67b33a | 4185 | arc_reclaim_thread(void) |
34dc7c2f | 4186 | { |
ca67b33a | 4187 | fstrans_cookie_t cookie = spl_fstrans_mark(); |
ae6d0c60 | 4188 | hrtime_t growtime = 0; |
34dc7c2f BB |
4189 | callb_cpr_t cpr; |
4190 | ||
ca0bf58d | 4191 | CALLB_CPR_INIT(&cpr, &arc_reclaim_lock, callb_generic_cpr, FTAG); |
34dc7c2f | 4192 | |
ca0bf58d | 4193 | mutex_enter(&arc_reclaim_lock); |
ca67b33a MA |
4194 | while (!arc_reclaim_thread_exit) { |
4195 | int64_t to_free; | |
ca67b33a | 4196 | uint64_t evicted = 0; |
302f753f | 4197 | |
ca67b33a | 4198 | arc_tuning_update(); |
34dc7c2f | 4199 | |
d3c2ae1c GW |
4200 | /* |
4201 | * This is necessary in order for the mdb ::arc dcmd to | |
4202 | * show up to date information. Since the ::arc command | |
4203 | * does not call the kstat's update function, without | |
4204 | * this call, the command may show stale stats for the | |
4205 | * anon, mru, mru_ghost, mfu, and mfu_ghost lists. Even | |
4206 | * with this change, the data might be up to 1 second | |
4207 | * out of date; but that should suffice. The arc_state_t | |
4208 | * structures can be queried directly if more accurate | |
4209 | * information is needed. | |
4210 | */ | |
4211 | #ifndef __linux__ | |
4212 | if (arc_ksp != NULL) | |
4213 | arc_ksp->ks_update(arc_ksp, KSTAT_READ); | |
4214 | #endif | |
ca67b33a | 4215 | mutex_exit(&arc_reclaim_lock); |
34dc7c2f | 4216 | |
0a252dae GM |
4217 | /* |
4218 | * We call arc_adjust() before (possibly) calling | |
4219 | * arc_kmem_reap_now(), so that we can wake up | |
4220 | * arc_get_data_buf() sooner. | |
4221 | */ | |
4222 | evicted = arc_adjust(); | |
4223 | ||
4224 | int64_t free_memory = arc_available_memory(); | |
ca67b33a | 4225 | if (free_memory < 0) { |
34dc7c2f | 4226 | |
ca67b33a | 4227 | arc_no_grow = B_TRUE; |
b128c09f | 4228 | arc_warm = B_TRUE; |
34dc7c2f | 4229 | |
ca67b33a MA |
4230 | /* |
4231 | * Wait at least zfs_grow_retry (default 5) seconds | |
4232 | * before considering growing. | |
4233 | */ | |
ae6d0c60 | 4234 | growtime = gethrtime() + SEC2NSEC(arc_grow_retry); |
6a8f9b6b | 4235 | |
ca67b33a | 4236 | arc_kmem_reap_now(); |
34dc7c2f | 4237 | |
ca67b33a MA |
4238 | /* |
4239 | * If we are still low on memory, shrink the ARC | |
4240 | * so that we have arc_shrink_min free space. | |
4241 | */ | |
4242 | free_memory = arc_available_memory(); | |
34dc7c2f | 4243 | |
ca67b33a MA |
4244 | to_free = (arc_c >> arc_shrink_shift) - free_memory; |
4245 | if (to_free > 0) { | |
4246 | #ifdef _KERNEL | |
11f552fa | 4247 | to_free = MAX(to_free, arc_need_free); |
ca67b33a MA |
4248 | #endif |
4249 | arc_shrink(to_free); | |
4250 | } | |
4251 | } else if (free_memory < arc_c >> arc_no_grow_shift) { | |
4252 | arc_no_grow = B_TRUE; | |
ae6d0c60 | 4253 | } else if (gethrtime() >= growtime) { |
ca67b33a MA |
4254 | arc_no_grow = B_FALSE; |
4255 | } | |
bce45ec9 | 4256 | |
ca67b33a | 4257 | mutex_enter(&arc_reclaim_lock); |
bce45ec9 | 4258 | |
ca67b33a MA |
4259 | /* |
4260 | * If evicted is zero, we couldn't evict anything via | |
4261 | * arc_adjust(). This could be due to hash lock | |
4262 | * collisions, but more likely due to the majority of | |
4263 | * arc buffers being unevictable. Therefore, even if | |
4264 | * arc_size is above arc_c, another pass is unlikely to | |
4265 | * be helpful and could potentially cause us to enter an | |
4266 | * infinite loop. | |
4267 | */ | |
4268 | if (arc_size <= arc_c || evicted == 0) { | |
4269 | /* | |
4270 | * We're either no longer overflowing, or we | |
4271 | * can't evict anything more, so we should wake | |
11f552fa BB |
4272 | * up any threads before we go to sleep and clear |
4273 | * arc_need_free since nothing more can be done. | |
ca67b33a MA |
4274 | */ |
4275 | cv_broadcast(&arc_reclaim_waiters_cv); | |
11f552fa | 4276 | arc_need_free = 0; |
bce45ec9 | 4277 | |
ca67b33a MA |
4278 | /* |
4279 | * Block until signaled, or after one second (we | |
4280 | * might need to perform arc_kmem_reap_now() | |
4281 | * even if we aren't being signalled) | |
4282 | */ | |
4283 | CALLB_CPR_SAFE_BEGIN(&cpr); | |
a9bb2b68 | 4284 | (void) cv_timedwait_sig_hires(&arc_reclaim_thread_cv, |
ae6d0c60 | 4285 | &arc_reclaim_lock, SEC2NSEC(1), MSEC2NSEC(1), 0); |
ca67b33a MA |
4286 | CALLB_CPR_SAFE_END(&cpr, &arc_reclaim_lock); |
4287 | } | |
ca0bf58d | 4288 | } |
bce45ec9 | 4289 | |
d3c2ae1c | 4290 | arc_reclaim_thread_exit = B_FALSE; |
ca0bf58d PS |
4291 | cv_broadcast(&arc_reclaim_thread_cv); |
4292 | CALLB_CPR_EXIT(&cpr); /* drops arc_reclaim_lock */ | |
4293 | spl_fstrans_unmark(cookie); | |
4294 | thread_exit(); | |
4295 | } | |
4296 | ||
7cb67b45 BB |
4297 | #ifdef _KERNEL |
4298 | /* | |
302f753f BB |
4299 | * Determine the amount of memory eligible for eviction contained in the |
4300 | * ARC. All clean data reported by the ghost lists can always be safely | |
4301 | * evicted. Due to arc_c_min, the same does not hold for all clean data | |
4302 | * contained by the regular mru and mfu lists. | |
4303 | * | |
4304 | * In the case of the regular mru and mfu lists, we need to report as | |
4305 | * much clean data as possible, such that evicting that same reported | |
4306 | * data will not bring arc_size below arc_c_min. Thus, in certain | |
4307 | * circumstances, the total amount of clean data in the mru and mfu | |
4308 | * lists might not actually be evictable. | |
4309 | * | |
4310 | * The following two distinct cases are accounted for: | |
4311 | * | |
4312 | * 1. The sum of the amount of dirty data contained by both the mru and | |
4313 | * mfu lists, plus the ARC's other accounting (e.g. the anon list), | |
4314 | * is greater than or equal to arc_c_min. | |
4315 | * (i.e. amount of dirty data >= arc_c_min) | |
4316 | * | |
4317 | * This is the easy case; all clean data contained by the mru and mfu | |
4318 | * lists is evictable. Evicting all clean data can only drop arc_size | |
4319 | * to the amount of dirty data, which is greater than arc_c_min. | |
4320 | * | |
4321 | * 2. The sum of the amount of dirty data contained by both the mru and | |
4322 | * mfu lists, plus the ARC's other accounting (e.g. the anon list), | |
4323 | * is less than arc_c_min. | |
4324 | * (i.e. arc_c_min > amount of dirty data) | |
4325 | * | |
4326 | * 2.1. arc_size is greater than or equal arc_c_min. | |
4327 | * (i.e. arc_size >= arc_c_min > amount of dirty data) | |
4328 | * | |
4329 | * In this case, not all clean data from the regular mru and mfu | |
4330 | * lists is actually evictable; we must leave enough clean data | |
4331 | * to keep arc_size above arc_c_min. Thus, the maximum amount of | |
4332 | * evictable data from the two lists combined, is exactly the | |
4333 | * difference between arc_size and arc_c_min. | |
4334 | * | |
4335 | * 2.2. arc_size is less than arc_c_min | |
4336 | * (i.e. arc_c_min > arc_size > amount of dirty data) | |
4337 | * | |
4338 | * In this case, none of the data contained in the mru and mfu | |
4339 | * lists is evictable, even if it's clean. Since arc_size is | |
4340 | * already below arc_c_min, evicting any more would only | |
4341 | * increase this negative difference. | |
7cb67b45 | 4342 | */ |
302f753f | 4343 | static uint64_t |
4ea3f864 GM |
4344 | arc_evictable_memory(void) |
4345 | { | |
302f753f | 4346 | uint64_t arc_clean = |
d3c2ae1c GW |
4347 | refcount_count(&arc_mru->arcs_esize[ARC_BUFC_DATA]) + |
4348 | refcount_count(&arc_mru->arcs_esize[ARC_BUFC_METADATA]) + | |
4349 | refcount_count(&arc_mfu->arcs_esize[ARC_BUFC_DATA]) + | |
4350 | refcount_count(&arc_mfu->arcs_esize[ARC_BUFC_METADATA]); | |
302f753f | 4351 | uint64_t ghost_clean = |
d3c2ae1c GW |
4352 | refcount_count(&arc_mru_ghost->arcs_esize[ARC_BUFC_DATA]) + |
4353 | refcount_count(&arc_mru_ghost->arcs_esize[ARC_BUFC_METADATA]) + | |
4354 | refcount_count(&arc_mfu_ghost->arcs_esize[ARC_BUFC_DATA]) + | |
4355 | refcount_count(&arc_mfu_ghost->arcs_esize[ARC_BUFC_METADATA]); | |
302f753f BB |
4356 | uint64_t arc_dirty = MAX((int64_t)arc_size - (int64_t)arc_clean, 0); |
4357 | ||
4358 | if (arc_dirty >= arc_c_min) | |
4359 | return (ghost_clean + arc_clean); | |
4360 | ||
4361 | return (ghost_clean + MAX((int64_t)arc_size - (int64_t)arc_c_min, 0)); | |
4362 | } | |
4363 | ||
ed6e9cc2 TC |
4364 | /* |
4365 | * If sc->nr_to_scan is zero, the caller is requesting a query of the | |
4366 | * number of objects which can potentially be freed. If it is nonzero, | |
4367 | * the request is to free that many objects. | |
4368 | * | |
4369 | * Linux kernels >= 3.12 have the count_objects and scan_objects callbacks | |
4370 | * in struct shrinker and also require the shrinker to return the number | |
4371 | * of objects freed. | |
4372 | * | |
4373 | * Older kernels require the shrinker to return the number of freeable | |
4374 | * objects following the freeing of nr_to_free. | |
4375 | */ | |
4376 | static spl_shrinker_t | |
7e7baeca | 4377 | __arc_shrinker_func(struct shrinker *shrink, struct shrink_control *sc) |
7cb67b45 | 4378 | { |
ed6e9cc2 | 4379 | int64_t pages; |
7cb67b45 | 4380 | |
302f753f BB |
4381 | /* The arc is considered warm once reclaim has occurred */ |
4382 | if (unlikely(arc_warm == B_FALSE)) | |
4383 | arc_warm = B_TRUE; | |
7cb67b45 | 4384 | |
302f753f | 4385 | /* Return the potential number of reclaimable pages */ |
ed6e9cc2 | 4386 | pages = btop((int64_t)arc_evictable_memory()); |
302f753f BB |
4387 | if (sc->nr_to_scan == 0) |
4388 | return (pages); | |
3fd70ee6 BB |
4389 | |
4390 | /* Not allowed to perform filesystem reclaim */ | |
7e7baeca | 4391 | if (!(sc->gfp_mask & __GFP_FS)) |
ed6e9cc2 | 4392 | return (SHRINK_STOP); |
3fd70ee6 | 4393 | |
7cb67b45 | 4394 | /* Reclaim in progress */ |
ca0bf58d | 4395 | if (mutex_tryenter(&arc_reclaim_lock) == 0) |
ed6e9cc2 | 4396 | return (SHRINK_STOP); |
7cb67b45 | 4397 | |
ca0bf58d PS |
4398 | mutex_exit(&arc_reclaim_lock); |
4399 | ||
302f753f BB |
4400 | /* |
4401 | * Evict the requested number of pages by shrinking arc_c the | |
4402 | * requested amount. If there is nothing left to evict just | |
4403 | * reap whatever we can from the various arc slabs. | |
4404 | */ | |
4405 | if (pages > 0) { | |
ca67b33a MA |
4406 | arc_shrink(ptob(sc->nr_to_scan)); |
4407 | arc_kmem_reap_now(); | |
ed6e9cc2 TC |
4408 | #ifdef HAVE_SPLIT_SHRINKER_CALLBACK |
4409 | pages = MAX(pages - btop(arc_evictable_memory()), 0); | |
4410 | #else | |
1e3cb67b | 4411 | pages = btop(arc_evictable_memory()); |
ed6e9cc2 | 4412 | #endif |
302f753f | 4413 | } else { |
ca67b33a | 4414 | arc_kmem_reap_now(); |
ed6e9cc2 | 4415 | pages = SHRINK_STOP; |
302f753f BB |
4416 | } |
4417 | ||
ca0bf58d PS |
4418 | /* |
4419 | * We've reaped what we can, wake up threads. | |
4420 | */ | |
4421 | cv_broadcast(&arc_reclaim_waiters_cv); | |
4422 | ||
302f753f BB |
4423 | /* |
4424 | * When direct reclaim is observed it usually indicates a rapid | |
4425 | * increase in memory pressure. This occurs because the kswapd | |
4426 | * threads were unable to asynchronously keep enough free memory | |
4427 | * available. In this case set arc_no_grow to briefly pause arc | |
4428 | * growth to avoid compounding the memory pressure. | |
4429 | */ | |
7cb67b45 | 4430 | if (current_is_kswapd()) { |
302f753f | 4431 | ARCSTAT_BUMP(arcstat_memory_indirect_count); |
7cb67b45 | 4432 | } else { |
302f753f | 4433 | arc_no_grow = B_TRUE; |
11f552fa | 4434 | arc_need_free = ptob(sc->nr_to_scan); |
302f753f | 4435 | ARCSTAT_BUMP(arcstat_memory_direct_count); |
7cb67b45 BB |
4436 | } |
4437 | ||
1e3cb67b | 4438 | return (pages); |
7cb67b45 | 4439 | } |
7e7baeca | 4440 | SPL_SHRINKER_CALLBACK_WRAPPER(arc_shrinker_func); |
7cb67b45 BB |
4441 | |
4442 | SPL_SHRINKER_DECLARE(arc_shrinker, arc_shrinker_func, DEFAULT_SEEKS); | |
4443 | #endif /* _KERNEL */ | |
4444 | ||
34dc7c2f BB |
4445 | /* |
4446 | * Adapt arc info given the number of bytes we are trying to add and | |
4e33ba4c | 4447 | * the state that we are coming from. This function is only called |
34dc7c2f BB |
4448 | * when we are adding new content to the cache. |
4449 | */ | |
4450 | static void | |
4451 | arc_adapt(int bytes, arc_state_t *state) | |
4452 | { | |
4453 | int mult; | |
728d6ae9 | 4454 | uint64_t arc_p_min = (arc_c >> arc_p_min_shift); |
36da08ef PS |
4455 | int64_t mrug_size = refcount_count(&arc_mru_ghost->arcs_size); |
4456 | int64_t mfug_size = refcount_count(&arc_mfu_ghost->arcs_size); | |
34dc7c2f BB |
4457 | |
4458 | if (state == arc_l2c_only) | |
4459 | return; | |
4460 | ||
4461 | ASSERT(bytes > 0); | |
4462 | /* | |
4463 | * Adapt the target size of the MRU list: | |
4464 | * - if we just hit in the MRU ghost list, then increase | |
4465 | * the target size of the MRU list. | |
4466 | * - if we just hit in the MFU ghost list, then increase | |
4467 | * the target size of the MFU list by decreasing the | |
4468 | * target size of the MRU list. | |
4469 | */ | |
4470 | if (state == arc_mru_ghost) { | |
36da08ef | 4471 | mult = (mrug_size >= mfug_size) ? 1 : (mfug_size / mrug_size); |
62422785 PS |
4472 | if (!zfs_arc_p_dampener_disable) |
4473 | mult = MIN(mult, 10); /* avoid wild arc_p adjustment */ | |
34dc7c2f | 4474 | |
728d6ae9 | 4475 | arc_p = MIN(arc_c - arc_p_min, arc_p + bytes * mult); |
34dc7c2f | 4476 | } else if (state == arc_mfu_ghost) { |
d164b209 BB |
4477 | uint64_t delta; |
4478 | ||
36da08ef | 4479 | mult = (mfug_size >= mrug_size) ? 1 : (mrug_size / mfug_size); |
62422785 PS |
4480 | if (!zfs_arc_p_dampener_disable) |
4481 | mult = MIN(mult, 10); | |
34dc7c2f | 4482 | |
d164b209 | 4483 | delta = MIN(bytes * mult, arc_p); |
728d6ae9 | 4484 | arc_p = MAX(arc_p_min, arc_p - delta); |
34dc7c2f BB |
4485 | } |
4486 | ASSERT((int64_t)arc_p >= 0); | |
4487 | ||
ca67b33a MA |
4488 | if (arc_reclaim_needed()) { |
4489 | cv_signal(&arc_reclaim_thread_cv); | |
4490 | return; | |
4491 | } | |
4492 | ||
34dc7c2f BB |
4493 | if (arc_no_grow) |
4494 | return; | |
4495 | ||
4496 | if (arc_c >= arc_c_max) | |
4497 | return; | |
4498 | ||
4499 | /* | |
4500 | * If we're within (2 * maxblocksize) bytes of the target | |
4501 | * cache size, increment the target cache size | |
4502 | */ | |
935434ef | 4503 | ASSERT3U(arc_c, >=, 2ULL << SPA_MAXBLOCKSHIFT); |
121b3cae | 4504 | if (arc_size >= arc_c - (2ULL << SPA_MAXBLOCKSHIFT)) { |
34dc7c2f BB |
4505 | atomic_add_64(&arc_c, (int64_t)bytes); |
4506 | if (arc_c > arc_c_max) | |
4507 | arc_c = arc_c_max; | |
4508 | else if (state == arc_anon) | |
4509 | atomic_add_64(&arc_p, (int64_t)bytes); | |
4510 | if (arc_p > arc_c) | |
4511 | arc_p = arc_c; | |
4512 | } | |
4513 | ASSERT((int64_t)arc_p >= 0); | |
4514 | } | |
4515 | ||
4516 | /* | |
ca0bf58d PS |
4517 | * Check if arc_size has grown past our upper threshold, determined by |
4518 | * zfs_arc_overflow_shift. | |
34dc7c2f | 4519 | */ |
ca0bf58d PS |
4520 | static boolean_t |
4521 | arc_is_overflowing(void) | |
34dc7c2f | 4522 | { |
ca0bf58d PS |
4523 | /* Always allow at least one block of overflow */ |
4524 | uint64_t overflow = MAX(SPA_MAXBLOCKSIZE, | |
4525 | arc_c >> zfs_arc_overflow_shift); | |
34dc7c2f | 4526 | |
ca0bf58d | 4527 | return (arc_size >= arc_c + overflow); |
34dc7c2f BB |
4528 | } |
4529 | ||
a6255b7f DQ |
4530 | static abd_t * |
4531 | arc_get_data_abd(arc_buf_hdr_t *hdr, uint64_t size, void *tag) | |
4532 | { | |
4533 | arc_buf_contents_t type = arc_buf_type(hdr); | |
4534 | ||
4535 | arc_get_data_impl(hdr, size, tag); | |
4536 | if (type == ARC_BUFC_METADATA) { | |
4537 | return (abd_alloc(size, B_TRUE)); | |
4538 | } else { | |
4539 | ASSERT(type == ARC_BUFC_DATA); | |
4540 | return (abd_alloc(size, B_FALSE)); | |
4541 | } | |
4542 | } | |
4543 | ||
4544 | static void * | |
4545 | arc_get_data_buf(arc_buf_hdr_t *hdr, uint64_t size, void *tag) | |
4546 | { | |
4547 | arc_buf_contents_t type = arc_buf_type(hdr); | |
4548 | ||
4549 | arc_get_data_impl(hdr, size, tag); | |
4550 | if (type == ARC_BUFC_METADATA) { | |
4551 | return (zio_buf_alloc(size)); | |
4552 | } else { | |
4553 | ASSERT(type == ARC_BUFC_DATA); | |
4554 | return (zio_data_buf_alloc(size)); | |
4555 | } | |
4556 | } | |
4557 | ||
34dc7c2f | 4558 | /* |
d3c2ae1c GW |
4559 | * Allocate a block and return it to the caller. If we are hitting the |
4560 | * hard limit for the cache size, we must sleep, waiting for the eviction | |
4561 | * thread to catch up. If we're past the target size but below the hard | |
4562 | * limit, we'll only signal the reclaim thread and continue on. | |
34dc7c2f | 4563 | */ |
a6255b7f DQ |
4564 | static void |
4565 | arc_get_data_impl(arc_buf_hdr_t *hdr, uint64_t size, void *tag) | |
34dc7c2f | 4566 | { |
a6255b7f DQ |
4567 | arc_state_t *state = hdr->b_l1hdr.b_state; |
4568 | arc_buf_contents_t type = arc_buf_type(hdr); | |
34dc7c2f BB |
4569 | |
4570 | arc_adapt(size, state); | |
4571 | ||
4572 | /* | |
ca0bf58d PS |
4573 | * If arc_size is currently overflowing, and has grown past our |
4574 | * upper limit, we must be adding data faster than the evict | |
4575 | * thread can evict. Thus, to ensure we don't compound the | |
4576 | * problem by adding more data and forcing arc_size to grow even | |
4577 | * further past it's target size, we halt and wait for the | |
4578 | * eviction thread to catch up. | |
4579 | * | |
4580 | * It's also possible that the reclaim thread is unable to evict | |
4581 | * enough buffers to get arc_size below the overflow limit (e.g. | |
4582 | * due to buffers being un-evictable, or hash lock collisions). | |
4583 | * In this case, we want to proceed regardless if we're | |
4584 | * overflowing; thus we don't use a while loop here. | |
34dc7c2f | 4585 | */ |
ca0bf58d PS |
4586 | if (arc_is_overflowing()) { |
4587 | mutex_enter(&arc_reclaim_lock); | |
4588 | ||
4589 | /* | |
4590 | * Now that we've acquired the lock, we may no longer be | |
4591 | * over the overflow limit, lets check. | |
4592 | * | |
4593 | * We're ignoring the case of spurious wake ups. If that | |
4594 | * were to happen, it'd let this thread consume an ARC | |
4595 | * buffer before it should have (i.e. before we're under | |
4596 | * the overflow limit and were signalled by the reclaim | |
4597 | * thread). As long as that is a rare occurrence, it | |
4598 | * shouldn't cause any harm. | |
4599 | */ | |
4600 | if (arc_is_overflowing()) { | |
4601 | cv_signal(&arc_reclaim_thread_cv); | |
4602 | cv_wait(&arc_reclaim_waiters_cv, &arc_reclaim_lock); | |
34dc7c2f | 4603 | } |
34dc7c2f | 4604 | |
ca0bf58d | 4605 | mutex_exit(&arc_reclaim_lock); |
34dc7c2f | 4606 | } |
ab26409d | 4607 | |
d3c2ae1c | 4608 | VERIFY3U(hdr->b_type, ==, type); |
da8ccd0e | 4609 | if (type == ARC_BUFC_METADATA) { |
ca0bf58d PS |
4610 | arc_space_consume(size, ARC_SPACE_META); |
4611 | } else { | |
ca0bf58d | 4612 | arc_space_consume(size, ARC_SPACE_DATA); |
da8ccd0e PS |
4613 | } |
4614 | ||
34dc7c2f BB |
4615 | /* |
4616 | * Update the state size. Note that ghost states have a | |
4617 | * "ghost size" and so don't need to be updated. | |
4618 | */ | |
d3c2ae1c | 4619 | if (!GHOST_STATE(state)) { |
34dc7c2f | 4620 | |
d3c2ae1c | 4621 | (void) refcount_add_many(&state->arcs_size, size, tag); |
ca0bf58d PS |
4622 | |
4623 | /* | |
4624 | * If this is reached via arc_read, the link is | |
4625 | * protected by the hash lock. If reached via | |
4626 | * arc_buf_alloc, the header should not be accessed by | |
4627 | * any other thread. And, if reached via arc_read_done, | |
4628 | * the hash lock will protect it if it's found in the | |
4629 | * hash table; otherwise no other thread should be | |
4630 | * trying to [add|remove]_reference it. | |
4631 | */ | |
4632 | if (multilist_link_active(&hdr->b_l1hdr.b_arc_node)) { | |
b9541d6b | 4633 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); |
d3c2ae1c GW |
4634 | (void) refcount_add_many(&state->arcs_esize[type], |
4635 | size, tag); | |
34dc7c2f | 4636 | } |
d3c2ae1c | 4637 | |
34dc7c2f BB |
4638 | /* |
4639 | * If we are growing the cache, and we are adding anonymous | |
4640 | * data, and we have outgrown arc_p, update arc_p | |
4641 | */ | |
ca0bf58d | 4642 | if (arc_size < arc_c && hdr->b_l1hdr.b_state == arc_anon && |
36da08ef PS |
4643 | (refcount_count(&arc_anon->arcs_size) + |
4644 | refcount_count(&arc_mru->arcs_size) > arc_p)) | |
34dc7c2f BB |
4645 | arc_p = MIN(arc_c, arc_p + size); |
4646 | } | |
a6255b7f DQ |
4647 | } |
4648 | ||
4649 | static void | |
4650 | arc_free_data_abd(arc_buf_hdr_t *hdr, abd_t *abd, uint64_t size, void *tag) | |
4651 | { | |
4652 | arc_free_data_impl(hdr, size, tag); | |
4653 | abd_free(abd); | |
4654 | } | |
4655 | ||
4656 | static void | |
4657 | arc_free_data_buf(arc_buf_hdr_t *hdr, void *buf, uint64_t size, void *tag) | |
4658 | { | |
4659 | arc_buf_contents_t type = arc_buf_type(hdr); | |
4660 | ||
4661 | arc_free_data_impl(hdr, size, tag); | |
4662 | if (type == ARC_BUFC_METADATA) { | |
4663 | zio_buf_free(buf, size); | |
4664 | } else { | |
4665 | ASSERT(type == ARC_BUFC_DATA); | |
4666 | zio_data_buf_free(buf, size); | |
4667 | } | |
d3c2ae1c GW |
4668 | } |
4669 | ||
4670 | /* | |
4671 | * Free the arc data buffer. | |
4672 | */ | |
4673 | static void | |
a6255b7f | 4674 | arc_free_data_impl(arc_buf_hdr_t *hdr, uint64_t size, void *tag) |
d3c2ae1c GW |
4675 | { |
4676 | arc_state_t *state = hdr->b_l1hdr.b_state; | |
4677 | arc_buf_contents_t type = arc_buf_type(hdr); | |
4678 | ||
4679 | /* protected by hash lock, if in the hash table */ | |
4680 | if (multilist_link_active(&hdr->b_l1hdr.b_arc_node)) { | |
4681 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); | |
4682 | ASSERT(state != arc_anon && state != arc_l2c_only); | |
4683 | ||
4684 | (void) refcount_remove_many(&state->arcs_esize[type], | |
4685 | size, tag); | |
4686 | } | |
4687 | (void) refcount_remove_many(&state->arcs_size, size, tag); | |
4688 | ||
4689 | VERIFY3U(hdr->b_type, ==, type); | |
4690 | if (type == ARC_BUFC_METADATA) { | |
d3c2ae1c GW |
4691 | arc_space_return(size, ARC_SPACE_META); |
4692 | } else { | |
4693 | ASSERT(type == ARC_BUFC_DATA); | |
d3c2ae1c GW |
4694 | arc_space_return(size, ARC_SPACE_DATA); |
4695 | } | |
34dc7c2f BB |
4696 | } |
4697 | ||
4698 | /* | |
4699 | * This routine is called whenever a buffer is accessed. | |
4700 | * NOTE: the hash lock is dropped in this function. | |
4701 | */ | |
4702 | static void | |
2a432414 | 4703 | arc_access(arc_buf_hdr_t *hdr, kmutex_t *hash_lock) |
34dc7c2f | 4704 | { |
428870ff BB |
4705 | clock_t now; |
4706 | ||
34dc7c2f | 4707 | ASSERT(MUTEX_HELD(hash_lock)); |
b9541d6b | 4708 | ASSERT(HDR_HAS_L1HDR(hdr)); |
34dc7c2f | 4709 | |
b9541d6b | 4710 | if (hdr->b_l1hdr.b_state == arc_anon) { |
34dc7c2f BB |
4711 | /* |
4712 | * This buffer is not in the cache, and does not | |
4713 | * appear in our "ghost" list. Add the new buffer | |
4714 | * to the MRU state. | |
4715 | */ | |
4716 | ||
b9541d6b CW |
4717 | ASSERT0(hdr->b_l1hdr.b_arc_access); |
4718 | hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); | |
2a432414 GW |
4719 | DTRACE_PROBE1(new_state__mru, arc_buf_hdr_t *, hdr); |
4720 | arc_change_state(arc_mru, hdr, hash_lock); | |
34dc7c2f | 4721 | |
b9541d6b | 4722 | } else if (hdr->b_l1hdr.b_state == arc_mru) { |
428870ff BB |
4723 | now = ddi_get_lbolt(); |
4724 | ||
34dc7c2f BB |
4725 | /* |
4726 | * If this buffer is here because of a prefetch, then either: | |
4727 | * - clear the flag if this is a "referencing" read | |
4728 | * (any subsequent access will bump this into the MFU state). | |
4729 | * or | |
4730 | * - move the buffer to the head of the list if this is | |
4731 | * another prefetch (to make it less likely to be evicted). | |
4732 | */ | |
b9541d6b CW |
4733 | if (HDR_PREFETCH(hdr)) { |
4734 | if (refcount_count(&hdr->b_l1hdr.b_refcnt) == 0) { | |
ca0bf58d PS |
4735 | /* link protected by hash lock */ |
4736 | ASSERT(multilist_link_active( | |
b9541d6b | 4737 | &hdr->b_l1hdr.b_arc_node)); |
34dc7c2f | 4738 | } else { |
d3c2ae1c | 4739 | arc_hdr_clear_flags(hdr, ARC_FLAG_PREFETCH); |
b9541d6b | 4740 | atomic_inc_32(&hdr->b_l1hdr.b_mru_hits); |
34dc7c2f BB |
4741 | ARCSTAT_BUMP(arcstat_mru_hits); |
4742 | } | |
b9541d6b | 4743 | hdr->b_l1hdr.b_arc_access = now; |
34dc7c2f BB |
4744 | return; |
4745 | } | |
4746 | ||
4747 | /* | |
4748 | * This buffer has been "accessed" only once so far, | |
4749 | * but it is still in the cache. Move it to the MFU | |
4750 | * state. | |
4751 | */ | |
b9541d6b CW |
4752 | if (ddi_time_after(now, hdr->b_l1hdr.b_arc_access + |
4753 | ARC_MINTIME)) { | |
34dc7c2f BB |
4754 | /* |
4755 | * More than 125ms have passed since we | |
4756 | * instantiated this buffer. Move it to the | |
4757 | * most frequently used state. | |
4758 | */ | |
b9541d6b | 4759 | hdr->b_l1hdr.b_arc_access = now; |
2a432414 GW |
4760 | DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); |
4761 | arc_change_state(arc_mfu, hdr, hash_lock); | |
34dc7c2f | 4762 | } |
b9541d6b | 4763 | atomic_inc_32(&hdr->b_l1hdr.b_mru_hits); |
34dc7c2f | 4764 | ARCSTAT_BUMP(arcstat_mru_hits); |
b9541d6b | 4765 | } else if (hdr->b_l1hdr.b_state == arc_mru_ghost) { |
34dc7c2f BB |
4766 | arc_state_t *new_state; |
4767 | /* | |
4768 | * This buffer has been "accessed" recently, but | |
4769 | * was evicted from the cache. Move it to the | |
4770 | * MFU state. | |
4771 | */ | |
4772 | ||
b9541d6b | 4773 | if (HDR_PREFETCH(hdr)) { |
34dc7c2f | 4774 | new_state = arc_mru; |
b9541d6b | 4775 | if (refcount_count(&hdr->b_l1hdr.b_refcnt) > 0) |
d3c2ae1c | 4776 | arc_hdr_clear_flags(hdr, ARC_FLAG_PREFETCH); |
2a432414 | 4777 | DTRACE_PROBE1(new_state__mru, arc_buf_hdr_t *, hdr); |
34dc7c2f BB |
4778 | } else { |
4779 | new_state = arc_mfu; | |
2a432414 | 4780 | DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); |
34dc7c2f BB |
4781 | } |
4782 | ||
b9541d6b | 4783 | hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); |
2a432414 | 4784 | arc_change_state(new_state, hdr, hash_lock); |
34dc7c2f | 4785 | |
b9541d6b | 4786 | atomic_inc_32(&hdr->b_l1hdr.b_mru_ghost_hits); |
34dc7c2f | 4787 | ARCSTAT_BUMP(arcstat_mru_ghost_hits); |
b9541d6b | 4788 | } else if (hdr->b_l1hdr.b_state == arc_mfu) { |
34dc7c2f BB |
4789 | /* |
4790 | * This buffer has been accessed more than once and is | |
4791 | * still in the cache. Keep it in the MFU state. | |
4792 | * | |
4793 | * NOTE: an add_reference() that occurred when we did | |
4794 | * the arc_read() will have kicked this off the list. | |
4795 | * If it was a prefetch, we will explicitly move it to | |
4796 | * the head of the list now. | |
4797 | */ | |
b9541d6b CW |
4798 | if ((HDR_PREFETCH(hdr)) != 0) { |
4799 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); | |
ca0bf58d PS |
4800 | /* link protected by hash_lock */ |
4801 | ASSERT(multilist_link_active(&hdr->b_l1hdr.b_arc_node)); | |
34dc7c2f | 4802 | } |
b9541d6b | 4803 | atomic_inc_32(&hdr->b_l1hdr.b_mfu_hits); |
34dc7c2f | 4804 | ARCSTAT_BUMP(arcstat_mfu_hits); |
b9541d6b CW |
4805 | hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); |
4806 | } else if (hdr->b_l1hdr.b_state == arc_mfu_ghost) { | |
34dc7c2f BB |
4807 | arc_state_t *new_state = arc_mfu; |
4808 | /* | |
4809 | * This buffer has been accessed more than once but has | |
4810 | * been evicted from the cache. Move it back to the | |
4811 | * MFU state. | |
4812 | */ | |
4813 | ||
b9541d6b | 4814 | if (HDR_PREFETCH(hdr)) { |
34dc7c2f BB |
4815 | /* |
4816 | * This is a prefetch access... | |
4817 | * move this block back to the MRU state. | |
4818 | */ | |
b9541d6b | 4819 | ASSERT0(refcount_count(&hdr->b_l1hdr.b_refcnt)); |
34dc7c2f BB |
4820 | new_state = arc_mru; |
4821 | } | |
4822 | ||
b9541d6b | 4823 | hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); |
2a432414 GW |
4824 | DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); |
4825 | arc_change_state(new_state, hdr, hash_lock); | |
34dc7c2f | 4826 | |
b9541d6b | 4827 | atomic_inc_32(&hdr->b_l1hdr.b_mfu_ghost_hits); |
34dc7c2f | 4828 | ARCSTAT_BUMP(arcstat_mfu_ghost_hits); |
b9541d6b | 4829 | } else if (hdr->b_l1hdr.b_state == arc_l2c_only) { |
34dc7c2f BB |
4830 | /* |
4831 | * This buffer is on the 2nd Level ARC. | |
4832 | */ | |
4833 | ||
b9541d6b | 4834 | hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); |
2a432414 GW |
4835 | DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); |
4836 | arc_change_state(arc_mfu, hdr, hash_lock); | |
34dc7c2f | 4837 | } else { |
b9541d6b CW |
4838 | cmn_err(CE_PANIC, "invalid arc state 0x%p", |
4839 | hdr->b_l1hdr.b_state); | |
34dc7c2f BB |
4840 | } |
4841 | } | |
4842 | ||
4843 | /* a generic arc_done_func_t which you can use */ | |
4844 | /* ARGSUSED */ | |
4845 | void | |
4846 | arc_bcopy_func(zio_t *zio, arc_buf_t *buf, void *arg) | |
4847 | { | |
428870ff | 4848 | if (zio == NULL || zio->io_error == 0) |
2aa34383 | 4849 | bcopy(buf->b_data, arg, arc_buf_size(buf)); |
d3c2ae1c | 4850 | arc_buf_destroy(buf, arg); |
34dc7c2f BB |
4851 | } |
4852 | ||
4853 | /* a generic arc_done_func_t */ | |
4854 | void | |
4855 | arc_getbuf_func(zio_t *zio, arc_buf_t *buf, void *arg) | |
4856 | { | |
4857 | arc_buf_t **bufp = arg; | |
4858 | if (zio && zio->io_error) { | |
d3c2ae1c | 4859 | arc_buf_destroy(buf, arg); |
34dc7c2f BB |
4860 | *bufp = NULL; |
4861 | } else { | |
4862 | *bufp = buf; | |
428870ff | 4863 | ASSERT(buf->b_data); |
34dc7c2f BB |
4864 | } |
4865 | } | |
4866 | ||
d3c2ae1c GW |
4867 | static void |
4868 | arc_hdr_verify(arc_buf_hdr_t *hdr, blkptr_t *bp) | |
4869 | { | |
4870 | if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) { | |
4871 | ASSERT3U(HDR_GET_PSIZE(hdr), ==, 0); | |
4872 | ASSERT3U(HDR_GET_COMPRESS(hdr), ==, ZIO_COMPRESS_OFF); | |
4873 | } else { | |
4874 | if (HDR_COMPRESSION_ENABLED(hdr)) { | |
4875 | ASSERT3U(HDR_GET_COMPRESS(hdr), ==, | |
4876 | BP_GET_COMPRESS(bp)); | |
4877 | } | |
4878 | ASSERT3U(HDR_GET_LSIZE(hdr), ==, BP_GET_LSIZE(bp)); | |
4879 | ASSERT3U(HDR_GET_PSIZE(hdr), ==, BP_GET_PSIZE(bp)); | |
4880 | } | |
4881 | } | |
4882 | ||
34dc7c2f BB |
4883 | static void |
4884 | arc_read_done(zio_t *zio) | |
4885 | { | |
d3c2ae1c | 4886 | arc_buf_hdr_t *hdr = zio->io_private; |
9b67f605 | 4887 | kmutex_t *hash_lock = NULL; |
524b4217 DK |
4888 | arc_callback_t *callback_list; |
4889 | arc_callback_t *acb; | |
2aa34383 | 4890 | boolean_t freeable = B_FALSE; |
524b4217 | 4891 | boolean_t no_zio_error = (zio->io_error == 0); |
2aa34383 | 4892 | int callback_cnt = 0; |
34dc7c2f BB |
4893 | /* |
4894 | * The hdr was inserted into hash-table and removed from lists | |
4895 | * prior to starting I/O. We should find this header, since | |
4896 | * it's in the hash table, and it should be legit since it's | |
4897 | * not possible to evict it during the I/O. The only possible | |
4898 | * reason for it not to be found is if we were freed during the | |
4899 | * read. | |
4900 | */ | |
9b67f605 MA |
4901 | if (HDR_IN_HASH_TABLE(hdr)) { |
4902 | arc_buf_hdr_t *found; | |
4903 | ||
4904 | ASSERT3U(hdr->b_birth, ==, BP_PHYSICAL_BIRTH(zio->io_bp)); | |
4905 | ASSERT3U(hdr->b_dva.dva_word[0], ==, | |
4906 | BP_IDENTITY(zio->io_bp)->dva_word[0]); | |
4907 | ASSERT3U(hdr->b_dva.dva_word[1], ==, | |
4908 | BP_IDENTITY(zio->io_bp)->dva_word[1]); | |
4909 | ||
d3c2ae1c | 4910 | found = buf_hash_find(hdr->b_spa, zio->io_bp, &hash_lock); |
9b67f605 | 4911 | |
d3c2ae1c | 4912 | ASSERT((found == hdr && |
9b67f605 MA |
4913 | DVA_EQUAL(&hdr->b_dva, BP_IDENTITY(zio->io_bp))) || |
4914 | (found == hdr && HDR_L2_READING(hdr))); | |
d3c2ae1c GW |
4915 | ASSERT3P(hash_lock, !=, NULL); |
4916 | } | |
4917 | ||
524b4217 | 4918 | if (no_zio_error) { |
d3c2ae1c GW |
4919 | /* byteswap if necessary */ |
4920 | if (BP_SHOULD_BYTESWAP(zio->io_bp)) { | |
4921 | if (BP_GET_LEVEL(zio->io_bp) > 0) { | |
4922 | hdr->b_l1hdr.b_byteswap = DMU_BSWAP_UINT64; | |
4923 | } else { | |
4924 | hdr->b_l1hdr.b_byteswap = | |
4925 | DMU_OT_BYTESWAP(BP_GET_TYPE(zio->io_bp)); | |
4926 | } | |
4927 | } else { | |
4928 | hdr->b_l1hdr.b_byteswap = DMU_BSWAP_NUMFUNCS; | |
4929 | } | |
9b67f605 | 4930 | } |
34dc7c2f | 4931 | |
d3c2ae1c | 4932 | arc_hdr_clear_flags(hdr, ARC_FLAG_L2_EVICTED); |
b9541d6b | 4933 | if (l2arc_noprefetch && HDR_PREFETCH(hdr)) |
d3c2ae1c | 4934 | arc_hdr_clear_flags(hdr, ARC_FLAG_L2CACHE); |
34dc7c2f | 4935 | |
b9541d6b | 4936 | callback_list = hdr->b_l1hdr.b_acb; |
d3c2ae1c | 4937 | ASSERT3P(callback_list, !=, NULL); |
34dc7c2f | 4938 | |
524b4217 | 4939 | if (hash_lock && no_zio_error && hdr->b_l1hdr.b_state == arc_anon) { |
428870ff BB |
4940 | /* |
4941 | * Only call arc_access on anonymous buffers. This is because | |
4942 | * if we've issued an I/O for an evicted buffer, we've already | |
4943 | * called arc_access (to prevent any simultaneous readers from | |
4944 | * getting confused). | |
4945 | */ | |
4946 | arc_access(hdr, hash_lock); | |
4947 | } | |
4948 | ||
524b4217 DK |
4949 | /* |
4950 | * If a read request has a callback (i.e. acb_done is not NULL), then we | |
4951 | * make a buf containing the data according to the parameters which were | |
4952 | * passed in. The implementation of arc_buf_alloc_impl() ensures that we | |
4953 | * aren't needlessly decompressing the data multiple times. | |
4954 | */ | |
2aa34383 | 4955 | for (acb = callback_list; acb != NULL; acb = acb->acb_next) { |
524b4217 | 4956 | int error; |
2aa34383 DK |
4957 | if (!acb->acb_done) |
4958 | continue; | |
4959 | ||
524b4217 | 4960 | /* This is a demand read since prefetches don't use callbacks */ |
2aa34383 DK |
4961 | |
4962 | callback_cnt++; | |
524b4217 DK |
4963 | |
4964 | error = arc_buf_alloc_impl(hdr, acb->acb_private, | |
4965 | acb->acb_compressed, no_zio_error, &acb->acb_buf); | |
4966 | if (no_zio_error) { | |
4967 | zio->io_error = error; | |
34dc7c2f BB |
4968 | } |
4969 | } | |
b9541d6b | 4970 | hdr->b_l1hdr.b_acb = NULL; |
d3c2ae1c | 4971 | arc_hdr_clear_flags(hdr, ARC_FLAG_IO_IN_PROGRESS); |
2aa34383 | 4972 | if (callback_cnt == 0) { |
d3c2ae1c GW |
4973 | ASSERT(HDR_PREFETCH(hdr)); |
4974 | ASSERT0(hdr->b_l1hdr.b_bufcnt); | |
a6255b7f | 4975 | ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); |
428870ff | 4976 | } |
34dc7c2f | 4977 | |
b9541d6b CW |
4978 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt) || |
4979 | callback_list != NULL); | |
34dc7c2f | 4980 | |
524b4217 | 4981 | if (no_zio_error) { |
d3c2ae1c GW |
4982 | arc_hdr_verify(hdr, zio->io_bp); |
4983 | } else { | |
4984 | arc_hdr_set_flags(hdr, ARC_FLAG_IO_ERROR); | |
b9541d6b | 4985 | if (hdr->b_l1hdr.b_state != arc_anon) |
34dc7c2f BB |
4986 | arc_change_state(arc_anon, hdr, hash_lock); |
4987 | if (HDR_IN_HASH_TABLE(hdr)) | |
4988 | buf_hash_remove(hdr); | |
b9541d6b | 4989 | freeable = refcount_is_zero(&hdr->b_l1hdr.b_refcnt); |
34dc7c2f BB |
4990 | } |
4991 | ||
4992 | /* | |
4993 | * Broadcast before we drop the hash_lock to avoid the possibility | |
4994 | * that the hdr (and hence the cv) might be freed before we get to | |
4995 | * the cv_broadcast(). | |
4996 | */ | |
b9541d6b | 4997 | cv_broadcast(&hdr->b_l1hdr.b_cv); |
34dc7c2f | 4998 | |
b9541d6b | 4999 | if (hash_lock != NULL) { |
34dc7c2f BB |
5000 | mutex_exit(hash_lock); |
5001 | } else { | |
5002 | /* | |
5003 | * This block was freed while we waited for the read to | |
5004 | * complete. It has been removed from the hash table and | |
5005 | * moved to the anonymous state (so that it won't show up | |
5006 | * in the cache). | |
5007 | */ | |
b9541d6b CW |
5008 | ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon); |
5009 | freeable = refcount_is_zero(&hdr->b_l1hdr.b_refcnt); | |
34dc7c2f BB |
5010 | } |
5011 | ||
5012 | /* execute each callback and free its structure */ | |
5013 | while ((acb = callback_list) != NULL) { | |
5014 | if (acb->acb_done) | |
5015 | acb->acb_done(zio, acb->acb_buf, acb->acb_private); | |
5016 | ||
5017 | if (acb->acb_zio_dummy != NULL) { | |
5018 | acb->acb_zio_dummy->io_error = zio->io_error; | |
5019 | zio_nowait(acb->acb_zio_dummy); | |
5020 | } | |
5021 | ||
5022 | callback_list = acb->acb_next; | |
5023 | kmem_free(acb, sizeof (arc_callback_t)); | |
5024 | } | |
5025 | ||
5026 | if (freeable) | |
5027 | arc_hdr_destroy(hdr); | |
5028 | } | |
5029 | ||
5030 | /* | |
5c839890 | 5031 | * "Read" the block at the specified DVA (in bp) via the |
34dc7c2f BB |
5032 | * cache. If the block is found in the cache, invoke the provided |
5033 | * callback immediately and return. Note that the `zio' parameter | |
5034 | * in the callback will be NULL in this case, since no IO was | |
5035 | * required. If the block is not in the cache pass the read request | |
5036 | * on to the spa with a substitute callback function, so that the | |
5037 | * requested block will be added to the cache. | |
5038 | * | |
5039 | * If a read request arrives for a block that has a read in-progress, | |
5040 | * either wait for the in-progress read to complete (and return the | |
5041 | * results); or, if this is a read with a "done" func, add a record | |
5042 | * to the read to invoke the "done" func when the read completes, | |
5043 | * and return; or just return. | |
5044 | * | |
5045 | * arc_read_done() will invoke all the requested "done" functions | |
5046 | * for readers of this block. | |
5047 | */ | |
5048 | int | |
294f6806 | 5049 | arc_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, arc_done_func_t *done, |
2a432414 GW |
5050 | void *private, zio_priority_t priority, int zio_flags, |
5051 | arc_flags_t *arc_flags, const zbookmark_phys_t *zb) | |
34dc7c2f | 5052 | { |
9b67f605 | 5053 | arc_buf_hdr_t *hdr = NULL; |
9b67f605 | 5054 | kmutex_t *hash_lock = NULL; |
34dc7c2f | 5055 | zio_t *rzio; |
3541dc6d | 5056 | uint64_t guid = spa_load_guid(spa); |
2aa34383 | 5057 | boolean_t compressed_read = (zio_flags & ZIO_FLAG_RAW) != 0; |
1421c891 | 5058 | int rc = 0; |
34dc7c2f | 5059 | |
9b67f605 MA |
5060 | ASSERT(!BP_IS_EMBEDDED(bp) || |
5061 | BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA); | |
5062 | ||
34dc7c2f | 5063 | top: |
9b67f605 MA |
5064 | if (!BP_IS_EMBEDDED(bp)) { |
5065 | /* | |
5066 | * Embedded BP's have no DVA and require no I/O to "read". | |
5067 | * Create an anonymous arc buf to back it. | |
5068 | */ | |
5069 | hdr = buf_hash_find(guid, bp, &hash_lock); | |
5070 | } | |
5071 | ||
a6255b7f | 5072 | if (hdr != NULL && HDR_HAS_L1HDR(hdr) && hdr->b_l1hdr.b_pabd != NULL) { |
d3c2ae1c | 5073 | arc_buf_t *buf = NULL; |
2a432414 | 5074 | *arc_flags |= ARC_FLAG_CACHED; |
34dc7c2f BB |
5075 | |
5076 | if (HDR_IO_IN_PROGRESS(hdr)) { | |
5077 | ||
7f60329a MA |
5078 | if ((hdr->b_flags & ARC_FLAG_PRIO_ASYNC_READ) && |
5079 | priority == ZIO_PRIORITY_SYNC_READ) { | |
5080 | /* | |
5081 | * This sync read must wait for an | |
5082 | * in-progress async read (e.g. a predictive | |
5083 | * prefetch). Async reads are queued | |
5084 | * separately at the vdev_queue layer, so | |
5085 | * this is a form of priority inversion. | |
5086 | * Ideally, we would "inherit" the demand | |
5087 | * i/o's priority by moving the i/o from | |
5088 | * the async queue to the synchronous queue, | |
5089 | * but there is currently no mechanism to do | |
5090 | * so. Track this so that we can evaluate | |
5091 | * the magnitude of this potential performance | |
5092 | * problem. | |
5093 | * | |
5094 | * Note that if the prefetch i/o is already | |
5095 | * active (has been issued to the device), | |
5096 | * the prefetch improved performance, because | |
5097 | * we issued it sooner than we would have | |
5098 | * without the prefetch. | |
5099 | */ | |
5100 | DTRACE_PROBE1(arc__sync__wait__for__async, | |
5101 | arc_buf_hdr_t *, hdr); | |
5102 | ARCSTAT_BUMP(arcstat_sync_wait_for_async); | |
5103 | } | |
5104 | if (hdr->b_flags & ARC_FLAG_PREDICTIVE_PREFETCH) { | |
d3c2ae1c GW |
5105 | arc_hdr_clear_flags(hdr, |
5106 | ARC_FLAG_PREDICTIVE_PREFETCH); | |
7f60329a MA |
5107 | } |
5108 | ||
2a432414 | 5109 | if (*arc_flags & ARC_FLAG_WAIT) { |
b9541d6b | 5110 | cv_wait(&hdr->b_l1hdr.b_cv, hash_lock); |
34dc7c2f BB |
5111 | mutex_exit(hash_lock); |
5112 | goto top; | |
5113 | } | |
2a432414 | 5114 | ASSERT(*arc_flags & ARC_FLAG_NOWAIT); |
34dc7c2f BB |
5115 | |
5116 | if (done) { | |
7f60329a | 5117 | arc_callback_t *acb = NULL; |
34dc7c2f BB |
5118 | |
5119 | acb = kmem_zalloc(sizeof (arc_callback_t), | |
79c76d5b | 5120 | KM_SLEEP); |
34dc7c2f BB |
5121 | acb->acb_done = done; |
5122 | acb->acb_private = private; | |
34dc7c2f BB |
5123 | if (pio != NULL) |
5124 | acb->acb_zio_dummy = zio_null(pio, | |
d164b209 | 5125 | spa, NULL, NULL, NULL, zio_flags); |
34dc7c2f | 5126 | |
d3c2ae1c | 5127 | ASSERT3P(acb->acb_done, !=, NULL); |
b9541d6b CW |
5128 | acb->acb_next = hdr->b_l1hdr.b_acb; |
5129 | hdr->b_l1hdr.b_acb = acb; | |
34dc7c2f | 5130 | mutex_exit(hash_lock); |
1421c891 | 5131 | goto out; |
34dc7c2f BB |
5132 | } |
5133 | mutex_exit(hash_lock); | |
1421c891 | 5134 | goto out; |
34dc7c2f BB |
5135 | } |
5136 | ||
b9541d6b CW |
5137 | ASSERT(hdr->b_l1hdr.b_state == arc_mru || |
5138 | hdr->b_l1hdr.b_state == arc_mfu); | |
34dc7c2f BB |
5139 | |
5140 | if (done) { | |
7f60329a MA |
5141 | if (hdr->b_flags & ARC_FLAG_PREDICTIVE_PREFETCH) { |
5142 | /* | |
5143 | * This is a demand read which does not have to | |
5144 | * wait for i/o because we did a predictive | |
5145 | * prefetch i/o for it, which has completed. | |
5146 | */ | |
5147 | DTRACE_PROBE1( | |
5148 | arc__demand__hit__predictive__prefetch, | |
5149 | arc_buf_hdr_t *, hdr); | |
5150 | ARCSTAT_BUMP( | |
5151 | arcstat_demand_hit_predictive_prefetch); | |
d3c2ae1c GW |
5152 | arc_hdr_clear_flags(hdr, |
5153 | ARC_FLAG_PREDICTIVE_PREFETCH); | |
7f60329a | 5154 | } |
d3c2ae1c GW |
5155 | ASSERT(!BP_IS_EMBEDDED(bp) || !BP_IS_HOLE(bp)); |
5156 | ||
524b4217 DK |
5157 | /* Get a buf with the desired data in it. */ |
5158 | VERIFY0(arc_buf_alloc_impl(hdr, private, | |
5159 | compressed_read, B_TRUE, &buf)); | |
2a432414 | 5160 | } else if (*arc_flags & ARC_FLAG_PREFETCH && |
b9541d6b | 5161 | refcount_count(&hdr->b_l1hdr.b_refcnt) == 0) { |
d3c2ae1c | 5162 | arc_hdr_set_flags(hdr, ARC_FLAG_PREFETCH); |
34dc7c2f BB |
5163 | } |
5164 | DTRACE_PROBE1(arc__hit, arc_buf_hdr_t *, hdr); | |
5165 | arc_access(hdr, hash_lock); | |
2a432414 | 5166 | if (*arc_flags & ARC_FLAG_L2CACHE) |
d3c2ae1c | 5167 | arc_hdr_set_flags(hdr, ARC_FLAG_L2CACHE); |
34dc7c2f BB |
5168 | mutex_exit(hash_lock); |
5169 | ARCSTAT_BUMP(arcstat_hits); | |
b9541d6b CW |
5170 | ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr), |
5171 | demand, prefetch, !HDR_ISTYPE_METADATA(hdr), | |
34dc7c2f BB |
5172 | data, metadata, hits); |
5173 | ||
5174 | if (done) | |
5175 | done(NULL, buf, private); | |
5176 | } else { | |
d3c2ae1c GW |
5177 | uint64_t lsize = BP_GET_LSIZE(bp); |
5178 | uint64_t psize = BP_GET_PSIZE(bp); | |
9b67f605 | 5179 | arc_callback_t *acb; |
b128c09f | 5180 | vdev_t *vd = NULL; |
a117a6d6 | 5181 | uint64_t addr = 0; |
d164b209 | 5182 | boolean_t devw = B_FALSE; |
d3c2ae1c | 5183 | uint64_t size; |
34dc7c2f | 5184 | |
5f6d0b6f BB |
5185 | /* |
5186 | * Gracefully handle a damaged logical block size as a | |
1cdb86cb | 5187 | * checksum error. |
5f6d0b6f | 5188 | */ |
d3c2ae1c | 5189 | if (lsize > spa_maxblocksize(spa)) { |
1cdb86cb | 5190 | rc = SET_ERROR(ECKSUM); |
5f6d0b6f BB |
5191 | goto out; |
5192 | } | |
5193 | ||
34dc7c2f BB |
5194 | if (hdr == NULL) { |
5195 | /* this block is not in the cache */ | |
9b67f605 | 5196 | arc_buf_hdr_t *exists = NULL; |
34dc7c2f | 5197 | arc_buf_contents_t type = BP_GET_BUFC_TYPE(bp); |
d3c2ae1c GW |
5198 | hdr = arc_hdr_alloc(spa_load_guid(spa), psize, lsize, |
5199 | BP_GET_COMPRESS(bp), type); | |
5200 | ||
9b67f605 MA |
5201 | if (!BP_IS_EMBEDDED(bp)) { |
5202 | hdr->b_dva = *BP_IDENTITY(bp); | |
5203 | hdr->b_birth = BP_PHYSICAL_BIRTH(bp); | |
9b67f605 MA |
5204 | exists = buf_hash_insert(hdr, &hash_lock); |
5205 | } | |
5206 | if (exists != NULL) { | |
34dc7c2f BB |
5207 | /* somebody beat us to the hash insert */ |
5208 | mutex_exit(hash_lock); | |
428870ff | 5209 | buf_discard_identity(hdr); |
d3c2ae1c | 5210 | arc_hdr_destroy(hdr); |
34dc7c2f BB |
5211 | goto top; /* restart the IO request */ |
5212 | } | |
34dc7c2f | 5213 | } else { |
b9541d6b CW |
5214 | /* |
5215 | * This block is in the ghost cache. If it was L2-only | |
5216 | * (and thus didn't have an L1 hdr), we realloc the | |
5217 | * header to add an L1 hdr. | |
5218 | */ | |
5219 | if (!HDR_HAS_L1HDR(hdr)) { | |
5220 | hdr = arc_hdr_realloc(hdr, hdr_l2only_cache, | |
5221 | hdr_full_cache); | |
5222 | } | |
5223 | ||
a6255b7f | 5224 | ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); |
b9541d6b | 5225 | ASSERT(GHOST_STATE(hdr->b_l1hdr.b_state)); |
34dc7c2f | 5226 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); |
b9541d6b | 5227 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); |
ca0bf58d | 5228 | ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); |
2aa34383 | 5229 | ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL); |
34dc7c2f | 5230 | |
7f60329a | 5231 | /* |
d3c2ae1c GW |
5232 | * This is a delicate dance that we play here. |
5233 | * This hdr is in the ghost list so we access it | |
5234 | * to move it out of the ghost list before we | |
5235 | * initiate the read. If it's a prefetch then | |
5236 | * it won't have a callback so we'll remove the | |
5237 | * reference that arc_buf_alloc_impl() created. We | |
5238 | * do this after we've called arc_access() to | |
5239 | * avoid hitting an assert in remove_reference(). | |
7f60329a | 5240 | */ |
428870ff | 5241 | arc_access(hdr, hash_lock); |
a6255b7f | 5242 | arc_hdr_alloc_pabd(hdr); |
d3c2ae1c | 5243 | } |
a6255b7f | 5244 | ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); |
d3c2ae1c GW |
5245 | size = arc_hdr_size(hdr); |
5246 | ||
5247 | /* | |
5248 | * If compression is enabled on the hdr, then will do | |
5249 | * RAW I/O and will store the compressed data in the hdr's | |
5250 | * data block. Otherwise, the hdr's data block will contain | |
5251 | * the uncompressed data. | |
5252 | */ | |
5253 | if (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF) { | |
5254 | zio_flags |= ZIO_FLAG_RAW; | |
34dc7c2f BB |
5255 | } |
5256 | ||
d3c2ae1c GW |
5257 | if (*arc_flags & ARC_FLAG_PREFETCH) |
5258 | arc_hdr_set_flags(hdr, ARC_FLAG_PREFETCH); | |
5259 | if (*arc_flags & ARC_FLAG_L2CACHE) | |
5260 | arc_hdr_set_flags(hdr, ARC_FLAG_L2CACHE); | |
5261 | if (BP_GET_LEVEL(bp) > 0) | |
5262 | arc_hdr_set_flags(hdr, ARC_FLAG_INDIRECT); | |
7f60329a | 5263 | if (*arc_flags & ARC_FLAG_PREDICTIVE_PREFETCH) |
d3c2ae1c | 5264 | arc_hdr_set_flags(hdr, ARC_FLAG_PREDICTIVE_PREFETCH); |
b9541d6b | 5265 | ASSERT(!GHOST_STATE(hdr->b_l1hdr.b_state)); |
428870ff | 5266 | |
79c76d5b | 5267 | acb = kmem_zalloc(sizeof (arc_callback_t), KM_SLEEP); |
34dc7c2f BB |
5268 | acb->acb_done = done; |
5269 | acb->acb_private = private; | |
2aa34383 | 5270 | acb->acb_compressed = compressed_read; |
34dc7c2f | 5271 | |
d3c2ae1c | 5272 | ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL); |
b9541d6b | 5273 | hdr->b_l1hdr.b_acb = acb; |
d3c2ae1c | 5274 | arc_hdr_set_flags(hdr, ARC_FLAG_IO_IN_PROGRESS); |
34dc7c2f | 5275 | |
b9541d6b CW |
5276 | if (HDR_HAS_L2HDR(hdr) && |
5277 | (vd = hdr->b_l2hdr.b_dev->l2ad_vdev) != NULL) { | |
5278 | devw = hdr->b_l2hdr.b_dev->l2ad_writing; | |
5279 | addr = hdr->b_l2hdr.b_daddr; | |
b128c09f BB |
5280 | /* |
5281 | * Lock out device removal. | |
5282 | */ | |
5283 | if (vdev_is_dead(vd) || | |
5284 | !spa_config_tryenter(spa, SCL_L2ARC, vd, RW_READER)) | |
5285 | vd = NULL; | |
5286 | } | |
5287 | ||
d3c2ae1c GW |
5288 | if (priority == ZIO_PRIORITY_ASYNC_READ) |
5289 | arc_hdr_set_flags(hdr, ARC_FLAG_PRIO_ASYNC_READ); | |
5290 | else | |
5291 | arc_hdr_clear_flags(hdr, ARC_FLAG_PRIO_ASYNC_READ); | |
5292 | ||
9b67f605 MA |
5293 | if (hash_lock != NULL) |
5294 | mutex_exit(hash_lock); | |
b128c09f | 5295 | |
e49f1e20 WA |
5296 | /* |
5297 | * At this point, we have a level 1 cache miss. Try again in | |
5298 | * L2ARC if possible. | |
5299 | */ | |
d3c2ae1c GW |
5300 | ASSERT3U(HDR_GET_LSIZE(hdr), ==, lsize); |
5301 | ||
428870ff | 5302 | DTRACE_PROBE4(arc__miss, arc_buf_hdr_t *, hdr, blkptr_t *, bp, |
d3c2ae1c | 5303 | uint64_t, lsize, zbookmark_phys_t *, zb); |
34dc7c2f | 5304 | ARCSTAT_BUMP(arcstat_misses); |
b9541d6b CW |
5305 | ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr), |
5306 | demand, prefetch, !HDR_ISTYPE_METADATA(hdr), | |
34dc7c2f BB |
5307 | data, metadata, misses); |
5308 | ||
d164b209 | 5309 | if (vd != NULL && l2arc_ndev != 0 && !(l2arc_norw && devw)) { |
34dc7c2f BB |
5310 | /* |
5311 | * Read from the L2ARC if the following are true: | |
b128c09f BB |
5312 | * 1. The L2ARC vdev was previously cached. |
5313 | * 2. This buffer still has L2ARC metadata. | |
5314 | * 3. This buffer isn't currently writing to the L2ARC. | |
5315 | * 4. The L2ARC entry wasn't evicted, which may | |
5316 | * also have invalidated the vdev. | |
d164b209 | 5317 | * 5. This isn't prefetch and l2arc_noprefetch is set. |
34dc7c2f | 5318 | */ |
b9541d6b | 5319 | if (HDR_HAS_L2HDR(hdr) && |
d164b209 BB |
5320 | !HDR_L2_WRITING(hdr) && !HDR_L2_EVICTED(hdr) && |
5321 | !(l2arc_noprefetch && HDR_PREFETCH(hdr))) { | |
34dc7c2f BB |
5322 | l2arc_read_callback_t *cb; |
5323 | ||
5324 | DTRACE_PROBE1(l2arc__hit, arc_buf_hdr_t *, hdr); | |
5325 | ARCSTAT_BUMP(arcstat_l2_hits); | |
b9541d6b | 5326 | atomic_inc_32(&hdr->b_l2hdr.b_hits); |
34dc7c2f | 5327 | |
34dc7c2f | 5328 | cb = kmem_zalloc(sizeof (l2arc_read_callback_t), |
79c76d5b | 5329 | KM_SLEEP); |
d3c2ae1c | 5330 | cb->l2rcb_hdr = hdr; |
34dc7c2f BB |
5331 | cb->l2rcb_bp = *bp; |
5332 | cb->l2rcb_zb = *zb; | |
b128c09f | 5333 | cb->l2rcb_flags = zio_flags; |
34dc7c2f | 5334 | |
a117a6d6 | 5335 | ASSERT(addr >= VDEV_LABEL_START_SIZE && |
d3c2ae1c | 5336 | addr + lsize < vd->vdev_psize - |
a117a6d6 GW |
5337 | VDEV_LABEL_END_SIZE); |
5338 | ||
34dc7c2f | 5339 | /* |
b128c09f BB |
5340 | * l2arc read. The SCL_L2ARC lock will be |
5341 | * released by l2arc_read_done(). | |
3a17a7a9 SK |
5342 | * Issue a null zio if the underlying buffer |
5343 | * was squashed to zero size by compression. | |
34dc7c2f | 5344 | */ |
d3c2ae1c GW |
5345 | ASSERT3U(HDR_GET_COMPRESS(hdr), !=, |
5346 | ZIO_COMPRESS_EMPTY); | |
5347 | rzio = zio_read_phys(pio, vd, addr, | |
a6255b7f | 5348 | size, hdr->b_l1hdr.b_pabd, |
d3c2ae1c GW |
5349 | ZIO_CHECKSUM_OFF, |
5350 | l2arc_read_done, cb, priority, | |
5351 | zio_flags | ZIO_FLAG_DONT_CACHE | | |
5352 | ZIO_FLAG_CANFAIL | | |
5353 | ZIO_FLAG_DONT_PROPAGATE | | |
5354 | ZIO_FLAG_DONT_RETRY, B_FALSE); | |
5355 | ||
34dc7c2f BB |
5356 | DTRACE_PROBE2(l2arc__read, vdev_t *, vd, |
5357 | zio_t *, rzio); | |
d3c2ae1c | 5358 | ARCSTAT_INCR(arcstat_l2_read_bytes, size); |
34dc7c2f | 5359 | |
2a432414 | 5360 | if (*arc_flags & ARC_FLAG_NOWAIT) { |
b128c09f | 5361 | zio_nowait(rzio); |
1421c891 | 5362 | goto out; |
b128c09f | 5363 | } |
34dc7c2f | 5364 | |
2a432414 | 5365 | ASSERT(*arc_flags & ARC_FLAG_WAIT); |
b128c09f | 5366 | if (zio_wait(rzio) == 0) |
1421c891 | 5367 | goto out; |
b128c09f BB |
5368 | |
5369 | /* l2arc read error; goto zio_read() */ | |
34dc7c2f BB |
5370 | } else { |
5371 | DTRACE_PROBE1(l2arc__miss, | |
5372 | arc_buf_hdr_t *, hdr); | |
5373 | ARCSTAT_BUMP(arcstat_l2_misses); | |
5374 | if (HDR_L2_WRITING(hdr)) | |
5375 | ARCSTAT_BUMP(arcstat_l2_rw_clash); | |
b128c09f | 5376 | spa_config_exit(spa, SCL_L2ARC, vd); |
34dc7c2f | 5377 | } |
d164b209 BB |
5378 | } else { |
5379 | if (vd != NULL) | |
5380 | spa_config_exit(spa, SCL_L2ARC, vd); | |
5381 | if (l2arc_ndev != 0) { | |
5382 | DTRACE_PROBE1(l2arc__miss, | |
5383 | arc_buf_hdr_t *, hdr); | |
5384 | ARCSTAT_BUMP(arcstat_l2_misses); | |
5385 | } | |
34dc7c2f | 5386 | } |
34dc7c2f | 5387 | |
a6255b7f | 5388 | rzio = zio_read(pio, spa, bp, hdr->b_l1hdr.b_pabd, size, |
d3c2ae1c | 5389 | arc_read_done, hdr, priority, zio_flags, zb); |
34dc7c2f | 5390 | |
2a432414 | 5391 | if (*arc_flags & ARC_FLAG_WAIT) { |
1421c891 PS |
5392 | rc = zio_wait(rzio); |
5393 | goto out; | |
5394 | } | |
34dc7c2f | 5395 | |
2a432414 | 5396 | ASSERT(*arc_flags & ARC_FLAG_NOWAIT); |
34dc7c2f BB |
5397 | zio_nowait(rzio); |
5398 | } | |
1421c891 PS |
5399 | |
5400 | out: | |
5401 | spa_read_history_add(spa, zb, *arc_flags); | |
5402 | return (rc); | |
34dc7c2f BB |
5403 | } |
5404 | ||
ab26409d BB |
5405 | arc_prune_t * |
5406 | arc_add_prune_callback(arc_prune_func_t *func, void *private) | |
5407 | { | |
5408 | arc_prune_t *p; | |
5409 | ||
d1d7e268 | 5410 | p = kmem_alloc(sizeof (*p), KM_SLEEP); |
ab26409d BB |
5411 | p->p_pfunc = func; |
5412 | p->p_private = private; | |
5413 | list_link_init(&p->p_node); | |
5414 | refcount_create(&p->p_refcnt); | |
5415 | ||
5416 | mutex_enter(&arc_prune_mtx); | |
5417 | refcount_add(&p->p_refcnt, &arc_prune_list); | |
5418 | list_insert_head(&arc_prune_list, p); | |
5419 | mutex_exit(&arc_prune_mtx); | |
5420 | ||
5421 | return (p); | |
5422 | } | |
5423 | ||
5424 | void | |
5425 | arc_remove_prune_callback(arc_prune_t *p) | |
5426 | { | |
4442f60d | 5427 | boolean_t wait = B_FALSE; |
ab26409d BB |
5428 | mutex_enter(&arc_prune_mtx); |
5429 | list_remove(&arc_prune_list, p); | |
4442f60d CC |
5430 | if (refcount_remove(&p->p_refcnt, &arc_prune_list) > 0) |
5431 | wait = B_TRUE; | |
ab26409d | 5432 | mutex_exit(&arc_prune_mtx); |
4442f60d CC |
5433 | |
5434 | /* wait for arc_prune_task to finish */ | |
5435 | if (wait) | |
5436 | taskq_wait_outstanding(arc_prune_taskq, 0); | |
5437 | ASSERT0(refcount_count(&p->p_refcnt)); | |
5438 | refcount_destroy(&p->p_refcnt); | |
5439 | kmem_free(p, sizeof (*p)); | |
ab26409d BB |
5440 | } |
5441 | ||
df4474f9 MA |
5442 | /* |
5443 | * Notify the arc that a block was freed, and thus will never be used again. | |
5444 | */ | |
5445 | void | |
5446 | arc_freed(spa_t *spa, const blkptr_t *bp) | |
5447 | { | |
5448 | arc_buf_hdr_t *hdr; | |
5449 | kmutex_t *hash_lock; | |
5450 | uint64_t guid = spa_load_guid(spa); | |
5451 | ||
9b67f605 MA |
5452 | ASSERT(!BP_IS_EMBEDDED(bp)); |
5453 | ||
5454 | hdr = buf_hash_find(guid, bp, &hash_lock); | |
df4474f9 MA |
5455 | if (hdr == NULL) |
5456 | return; | |
df4474f9 | 5457 | |
d3c2ae1c GW |
5458 | /* |
5459 | * We might be trying to free a block that is still doing I/O | |
5460 | * (i.e. prefetch) or has a reference (i.e. a dedup-ed, | |
5461 | * dmu_sync-ed block). If this block is being prefetched, then it | |
5462 | * would still have the ARC_FLAG_IO_IN_PROGRESS flag set on the hdr | |
5463 | * until the I/O completes. A block may also have a reference if it is | |
5464 | * part of a dedup-ed, dmu_synced write. The dmu_sync() function would | |
5465 | * have written the new block to its final resting place on disk but | |
5466 | * without the dedup flag set. This would have left the hdr in the MRU | |
5467 | * state and discoverable. When the txg finally syncs it detects that | |
5468 | * the block was overridden in open context and issues an override I/O. | |
5469 | * Since this is a dedup block, the override I/O will determine if the | |
5470 | * block is already in the DDT. If so, then it will replace the io_bp | |
5471 | * with the bp from the DDT and allow the I/O to finish. When the I/O | |
5472 | * reaches the done callback, dbuf_write_override_done, it will | |
5473 | * check to see if the io_bp and io_bp_override are identical. | |
5474 | * If they are not, then it indicates that the bp was replaced with | |
5475 | * the bp in the DDT and the override bp is freed. This allows | |
5476 | * us to arrive here with a reference on a block that is being | |
5477 | * freed. So if we have an I/O in progress, or a reference to | |
5478 | * this hdr, then we don't destroy the hdr. | |
5479 | */ | |
5480 | if (!HDR_HAS_L1HDR(hdr) || (!HDR_IO_IN_PROGRESS(hdr) && | |
5481 | refcount_is_zero(&hdr->b_l1hdr.b_refcnt))) { | |
5482 | arc_change_state(arc_anon, hdr, hash_lock); | |
5483 | arc_hdr_destroy(hdr); | |
df4474f9 | 5484 | mutex_exit(hash_lock); |
bd089c54 | 5485 | } else { |
d3c2ae1c | 5486 | mutex_exit(hash_lock); |
34dc7c2f | 5487 | } |
34dc7c2f | 5488 | |
34dc7c2f BB |
5489 | } |
5490 | ||
5491 | /* | |
e49f1e20 WA |
5492 | * Release this buffer from the cache, making it an anonymous buffer. This |
5493 | * must be done after a read and prior to modifying the buffer contents. | |
34dc7c2f | 5494 | * If the buffer has more than one reference, we must make |
b128c09f | 5495 | * a new hdr for the buffer. |
34dc7c2f BB |
5496 | */ |
5497 | void | |
5498 | arc_release(arc_buf_t *buf, void *tag) | |
5499 | { | |
b9541d6b CW |
5500 | kmutex_t *hash_lock; |
5501 | arc_state_t *state; | |
5502 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
34dc7c2f | 5503 | |
428870ff | 5504 | /* |
ca0bf58d | 5505 | * It would be nice to assert that if its DMU metadata (level > |
428870ff BB |
5506 | * 0 || it's the dnode file), then it must be syncing context. |
5507 | * But we don't know that information at this level. | |
5508 | */ | |
5509 | ||
5510 | mutex_enter(&buf->b_evict_lock); | |
b128c09f | 5511 | |
ca0bf58d PS |
5512 | ASSERT(HDR_HAS_L1HDR(hdr)); |
5513 | ||
b9541d6b CW |
5514 | /* |
5515 | * We don't grab the hash lock prior to this check, because if | |
5516 | * the buffer's header is in the arc_anon state, it won't be | |
5517 | * linked into the hash table. | |
5518 | */ | |
5519 | if (hdr->b_l1hdr.b_state == arc_anon) { | |
5520 | mutex_exit(&buf->b_evict_lock); | |
5521 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); | |
5522 | ASSERT(!HDR_IN_HASH_TABLE(hdr)); | |
5523 | ASSERT(!HDR_HAS_L2HDR(hdr)); | |
d3c2ae1c | 5524 | ASSERT(HDR_EMPTY(hdr)); |
34dc7c2f | 5525 | |
d3c2ae1c | 5526 | ASSERT3U(hdr->b_l1hdr.b_bufcnt, ==, 1); |
b9541d6b CW |
5527 | ASSERT3S(refcount_count(&hdr->b_l1hdr.b_refcnt), ==, 1); |
5528 | ASSERT(!list_link_active(&hdr->b_l1hdr.b_arc_node)); | |
5529 | ||
b9541d6b | 5530 | hdr->b_l1hdr.b_arc_access = 0; |
d3c2ae1c GW |
5531 | |
5532 | /* | |
5533 | * If the buf is being overridden then it may already | |
5534 | * have a hdr that is not empty. | |
5535 | */ | |
5536 | buf_discard_identity(hdr); | |
b9541d6b CW |
5537 | arc_buf_thaw(buf); |
5538 | ||
5539 | return; | |
34dc7c2f BB |
5540 | } |
5541 | ||
b9541d6b CW |
5542 | hash_lock = HDR_LOCK(hdr); |
5543 | mutex_enter(hash_lock); | |
5544 | ||
5545 | /* | |
5546 | * This assignment is only valid as long as the hash_lock is | |
5547 | * held, we must be careful not to reference state or the | |
5548 | * b_state field after dropping the lock. | |
5549 | */ | |
5550 | state = hdr->b_l1hdr.b_state; | |
5551 | ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); | |
5552 | ASSERT3P(state, !=, arc_anon); | |
5553 | ||
5554 | /* this buffer is not on any list */ | |
2aa34383 | 5555 | ASSERT3S(refcount_count(&hdr->b_l1hdr.b_refcnt), >, 0); |
b9541d6b CW |
5556 | |
5557 | if (HDR_HAS_L2HDR(hdr)) { | |
b9541d6b | 5558 | mutex_enter(&hdr->b_l2hdr.b_dev->l2ad_mtx); |
ca0bf58d PS |
5559 | |
5560 | /* | |
d962d5da PS |
5561 | * We have to recheck this conditional again now that |
5562 | * we're holding the l2ad_mtx to prevent a race with | |
5563 | * another thread which might be concurrently calling | |
5564 | * l2arc_evict(). In that case, l2arc_evict() might have | |
5565 | * destroyed the header's L2 portion as we were waiting | |
5566 | * to acquire the l2ad_mtx. | |
ca0bf58d | 5567 | */ |
d962d5da PS |
5568 | if (HDR_HAS_L2HDR(hdr)) |
5569 | arc_hdr_l2hdr_destroy(hdr); | |
ca0bf58d | 5570 | |
b9541d6b | 5571 | mutex_exit(&hdr->b_l2hdr.b_dev->l2ad_mtx); |
b128c09f BB |
5572 | } |
5573 | ||
34dc7c2f BB |
5574 | /* |
5575 | * Do we have more than one buf? | |
5576 | */ | |
d3c2ae1c | 5577 | if (hdr->b_l1hdr.b_bufcnt > 1) { |
34dc7c2f | 5578 | arc_buf_hdr_t *nhdr; |
d164b209 | 5579 | uint64_t spa = hdr->b_spa; |
d3c2ae1c GW |
5580 | uint64_t psize = HDR_GET_PSIZE(hdr); |
5581 | uint64_t lsize = HDR_GET_LSIZE(hdr); | |
5582 | enum zio_compress compress = HDR_GET_COMPRESS(hdr); | |
b9541d6b | 5583 | arc_buf_contents_t type = arc_buf_type(hdr); |
d3c2ae1c GW |
5584 | arc_buf_t *lastbuf = NULL; |
5585 | VERIFY3U(hdr->b_type, ==, type); | |
34dc7c2f | 5586 | |
b9541d6b | 5587 | ASSERT(hdr->b_l1hdr.b_buf != buf || buf->b_next != NULL); |
d3c2ae1c GW |
5588 | (void) remove_reference(hdr, hash_lock, tag); |
5589 | ||
524b4217 | 5590 | if (arc_buf_is_shared(buf) && !ARC_BUF_COMPRESSED(buf)) { |
d3c2ae1c | 5591 | ASSERT3P(hdr->b_l1hdr.b_buf, !=, buf); |
524b4217 DK |
5592 | ASSERT(ARC_BUF_LAST(buf)); |
5593 | } | |
d3c2ae1c | 5594 | |
34dc7c2f | 5595 | /* |
428870ff | 5596 | * Pull the data off of this hdr and attach it to |
d3c2ae1c GW |
5597 | * a new anonymous hdr. Also find the last buffer |
5598 | * in the hdr's buffer list. | |
34dc7c2f | 5599 | */ |
2aa34383 | 5600 | lastbuf = arc_buf_remove(hdr, buf); |
d3c2ae1c | 5601 | ASSERT3P(lastbuf, !=, NULL); |
34dc7c2f | 5602 | |
d3c2ae1c GW |
5603 | /* |
5604 | * If the current arc_buf_t and the hdr are sharing their data | |
524b4217 | 5605 | * buffer, then we must stop sharing that block. |
d3c2ae1c GW |
5606 | */ |
5607 | if (arc_buf_is_shared(buf)) { | |
5608 | ASSERT3P(hdr->b_l1hdr.b_buf, !=, buf); | |
d3c2ae1c GW |
5609 | VERIFY(!arc_buf_is_shared(lastbuf)); |
5610 | ||
5611 | /* | |
5612 | * First, sever the block sharing relationship between | |
5613 | * buf and the arc_buf_hdr_t. Then, setup a new | |
5614 | * block sharing relationship with the last buffer | |
5615 | * on the arc_buf_t list. | |
5616 | */ | |
5617 | arc_unshare_buf(hdr, buf); | |
2aa34383 DK |
5618 | |
5619 | /* | |
a6255b7f | 5620 | * Now we need to recreate the hdr's b_pabd. Since we |
524b4217 | 5621 | * have lastbuf handy, we try to share with it, but if |
a6255b7f | 5622 | * we can't then we allocate a new b_pabd and copy the |
524b4217 | 5623 | * data from buf into it. |
2aa34383 | 5624 | */ |
524b4217 DK |
5625 | if (arc_can_share(hdr, lastbuf)) { |
5626 | arc_share_buf(hdr, lastbuf); | |
5627 | } else { | |
a6255b7f DQ |
5628 | arc_hdr_alloc_pabd(hdr); |
5629 | abd_copy_from_buf(hdr->b_l1hdr.b_pabd, | |
5630 | buf->b_data, psize); | |
2aa34383 | 5631 | } |
d3c2ae1c GW |
5632 | VERIFY3P(lastbuf->b_data, !=, NULL); |
5633 | } else if (HDR_SHARED_DATA(hdr)) { | |
2aa34383 DK |
5634 | /* |
5635 | * Uncompressed shared buffers are always at the end | |
5636 | * of the list. Compressed buffers don't have the | |
5637 | * same requirements. This makes it hard to | |
5638 | * simply assert that the lastbuf is shared so | |
5639 | * we rely on the hdr's compression flags to determine | |
5640 | * if we have a compressed, shared buffer. | |
5641 | */ | |
5642 | ASSERT(arc_buf_is_shared(lastbuf) || | |
5643 | HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF); | |
5644 | ASSERT(!ARC_BUF_SHARED(buf)); | |
d3c2ae1c | 5645 | } |
a6255b7f | 5646 | ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); |
b9541d6b | 5647 | ASSERT3P(state, !=, arc_l2c_only); |
36da08ef | 5648 | |
d3c2ae1c | 5649 | (void) refcount_remove_many(&state->arcs_size, |
2aa34383 | 5650 | arc_buf_size(buf), buf); |
36da08ef | 5651 | |
b9541d6b | 5652 | if (refcount_is_zero(&hdr->b_l1hdr.b_refcnt)) { |
b9541d6b | 5653 | ASSERT3P(state, !=, arc_l2c_only); |
d3c2ae1c | 5654 | (void) refcount_remove_many(&state->arcs_esize[type], |
2aa34383 | 5655 | arc_buf_size(buf), buf); |
34dc7c2f | 5656 | } |
1eb5bfa3 | 5657 | |
d3c2ae1c | 5658 | hdr->b_l1hdr.b_bufcnt -= 1; |
34dc7c2f | 5659 | arc_cksum_verify(buf); |
498877ba | 5660 | arc_buf_unwatch(buf); |
34dc7c2f BB |
5661 | |
5662 | mutex_exit(hash_lock); | |
5663 | ||
d3c2ae1c | 5664 | /* |
a6255b7f | 5665 | * Allocate a new hdr. The new hdr will contain a b_pabd |
d3c2ae1c GW |
5666 | * buffer which will be freed in arc_write(). |
5667 | */ | |
5668 | nhdr = arc_hdr_alloc(spa, psize, lsize, compress, type); | |
5669 | ASSERT3P(nhdr->b_l1hdr.b_buf, ==, NULL); | |
5670 | ASSERT0(nhdr->b_l1hdr.b_bufcnt); | |
5671 | ASSERT0(refcount_count(&nhdr->b_l1hdr.b_refcnt)); | |
5672 | VERIFY3U(nhdr->b_type, ==, type); | |
5673 | ASSERT(!HDR_SHARED_DATA(nhdr)); | |
b9541d6b | 5674 | |
d3c2ae1c GW |
5675 | nhdr->b_l1hdr.b_buf = buf; |
5676 | nhdr->b_l1hdr.b_bufcnt = 1; | |
b9541d6b CW |
5677 | nhdr->b_l1hdr.b_mru_hits = 0; |
5678 | nhdr->b_l1hdr.b_mru_ghost_hits = 0; | |
5679 | nhdr->b_l1hdr.b_mfu_hits = 0; | |
5680 | nhdr->b_l1hdr.b_mfu_ghost_hits = 0; | |
5681 | nhdr->b_l1hdr.b_l2_hits = 0; | |
b9541d6b | 5682 | (void) refcount_add(&nhdr->b_l1hdr.b_refcnt, tag); |
34dc7c2f | 5683 | buf->b_hdr = nhdr; |
d3c2ae1c | 5684 | |
428870ff | 5685 | mutex_exit(&buf->b_evict_lock); |
d3c2ae1c GW |
5686 | (void) refcount_add_many(&arc_anon->arcs_size, |
5687 | HDR_GET_LSIZE(nhdr), buf); | |
34dc7c2f | 5688 | } else { |
428870ff | 5689 | mutex_exit(&buf->b_evict_lock); |
b9541d6b | 5690 | ASSERT(refcount_count(&hdr->b_l1hdr.b_refcnt) == 1); |
ca0bf58d PS |
5691 | /* protected by hash lock, or hdr is on arc_anon */ |
5692 | ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); | |
34dc7c2f | 5693 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); |
b9541d6b CW |
5694 | hdr->b_l1hdr.b_mru_hits = 0; |
5695 | hdr->b_l1hdr.b_mru_ghost_hits = 0; | |
5696 | hdr->b_l1hdr.b_mfu_hits = 0; | |
5697 | hdr->b_l1hdr.b_mfu_ghost_hits = 0; | |
5698 | hdr->b_l1hdr.b_l2_hits = 0; | |
5699 | arc_change_state(arc_anon, hdr, hash_lock); | |
5700 | hdr->b_l1hdr.b_arc_access = 0; | |
5701 | mutex_exit(hash_lock); | |
34dc7c2f | 5702 | |
428870ff | 5703 | buf_discard_identity(hdr); |
34dc7c2f BB |
5704 | arc_buf_thaw(buf); |
5705 | } | |
34dc7c2f BB |
5706 | } |
5707 | ||
5708 | int | |
5709 | arc_released(arc_buf_t *buf) | |
5710 | { | |
b128c09f BB |
5711 | int released; |
5712 | ||
428870ff | 5713 | mutex_enter(&buf->b_evict_lock); |
b9541d6b CW |
5714 | released = (buf->b_data != NULL && |
5715 | buf->b_hdr->b_l1hdr.b_state == arc_anon); | |
428870ff | 5716 | mutex_exit(&buf->b_evict_lock); |
b128c09f | 5717 | return (released); |
34dc7c2f BB |
5718 | } |
5719 | ||
34dc7c2f BB |
5720 | #ifdef ZFS_DEBUG |
5721 | int | |
5722 | arc_referenced(arc_buf_t *buf) | |
5723 | { | |
b128c09f BB |
5724 | int referenced; |
5725 | ||
428870ff | 5726 | mutex_enter(&buf->b_evict_lock); |
b9541d6b | 5727 | referenced = (refcount_count(&buf->b_hdr->b_l1hdr.b_refcnt)); |
428870ff | 5728 | mutex_exit(&buf->b_evict_lock); |
b128c09f | 5729 | return (referenced); |
34dc7c2f BB |
5730 | } |
5731 | #endif | |
5732 | ||
5733 | static void | |
5734 | arc_write_ready(zio_t *zio) | |
5735 | { | |
5736 | arc_write_callback_t *callback = zio->io_private; | |
5737 | arc_buf_t *buf = callback->awcb_buf; | |
5738 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
d3c2ae1c GW |
5739 | uint64_t psize = BP_IS_HOLE(zio->io_bp) ? 0 : BP_GET_PSIZE(zio->io_bp); |
5740 | enum zio_compress compress; | |
a6255b7f | 5741 | fstrans_cookie_t cookie = spl_fstrans_mark(); |
34dc7c2f | 5742 | |
b9541d6b CW |
5743 | ASSERT(HDR_HAS_L1HDR(hdr)); |
5744 | ASSERT(!refcount_is_zero(&buf->b_hdr->b_l1hdr.b_refcnt)); | |
d3c2ae1c | 5745 | ASSERT(hdr->b_l1hdr.b_bufcnt > 0); |
b128c09f | 5746 | |
34dc7c2f | 5747 | /* |
d3c2ae1c GW |
5748 | * If we're reexecuting this zio because the pool suspended, then |
5749 | * cleanup any state that was previously set the first time the | |
2aa34383 | 5750 | * callback was invoked. |
34dc7c2f | 5751 | */ |
d3c2ae1c GW |
5752 | if (zio->io_flags & ZIO_FLAG_REEXECUTED) { |
5753 | arc_cksum_free(hdr); | |
5754 | arc_buf_unwatch(buf); | |
a6255b7f | 5755 | if (hdr->b_l1hdr.b_pabd != NULL) { |
d3c2ae1c | 5756 | if (arc_buf_is_shared(buf)) { |
d3c2ae1c GW |
5757 | arc_unshare_buf(hdr, buf); |
5758 | } else { | |
a6255b7f | 5759 | arc_hdr_free_pabd(hdr); |
d3c2ae1c | 5760 | } |
34dc7c2f | 5761 | } |
34dc7c2f | 5762 | } |
a6255b7f | 5763 | ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); |
d3c2ae1c GW |
5764 | ASSERT(!HDR_SHARED_DATA(hdr)); |
5765 | ASSERT(!arc_buf_is_shared(buf)); | |
5766 | ||
5767 | callback->awcb_ready(zio, buf, callback->awcb_private); | |
5768 | ||
5769 | if (HDR_IO_IN_PROGRESS(hdr)) | |
5770 | ASSERT(zio->io_flags & ZIO_FLAG_REEXECUTED); | |
5771 | ||
5772 | arc_cksum_compute(buf); | |
5773 | arc_hdr_set_flags(hdr, ARC_FLAG_IO_IN_PROGRESS); | |
5774 | ||
5775 | if (BP_IS_HOLE(zio->io_bp) || BP_IS_EMBEDDED(zio->io_bp)) { | |
5776 | compress = ZIO_COMPRESS_OFF; | |
5777 | } else { | |
5778 | ASSERT3U(HDR_GET_LSIZE(hdr), ==, BP_GET_LSIZE(zio->io_bp)); | |
5779 | compress = BP_GET_COMPRESS(zio->io_bp); | |
5780 | } | |
5781 | HDR_SET_PSIZE(hdr, psize); | |
5782 | arc_hdr_set_compress(hdr, compress); | |
5783 | ||
5784 | /* | |
a6255b7f DQ |
5785 | * Fill the hdr with data. If the hdr is compressed, the data we want |
5786 | * is available from the zio, otherwise we can take it from the buf. | |
5787 | * | |
5788 | * We might be able to share the buf's data with the hdr here. However, | |
5789 | * doing so would cause the ARC to be full of linear ABDs if we write a | |
5790 | * lot of shareable data. As a compromise, we check whether scattered | |
5791 | * ABDs are allowed, and assume that if they are then the user wants | |
5792 | * the ARC to be primarily filled with them regardless of the data being | |
5793 | * written. Therefore, if they're allowed then we allocate one and copy | |
5794 | * the data into it; otherwise, we share the data directly if we can. | |
d3c2ae1c | 5795 | */ |
a6255b7f DQ |
5796 | if (zfs_abd_scatter_enabled || !arc_can_share(hdr, buf)) { |
5797 | arc_hdr_alloc_pabd(hdr); | |
5798 | ||
5799 | /* | |
5800 | * Ideally, we would always copy the io_abd into b_pabd, but the | |
5801 | * user may have disabled compressed ARC, thus we must check the | |
5802 | * hdr's compression setting rather than the io_bp's. | |
5803 | */ | |
5804 | if (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF) { | |
5805 | ASSERT3U(BP_GET_COMPRESS(zio->io_bp), !=, | |
5806 | ZIO_COMPRESS_OFF); | |
5807 | ASSERT3U(psize, >, 0); | |
5808 | ||
5809 | abd_copy(hdr->b_l1hdr.b_pabd, zio->io_abd, psize); | |
5810 | } else { | |
5811 | ASSERT3U(zio->io_orig_size, ==, arc_hdr_size(hdr)); | |
5812 | ||
5813 | abd_copy_from_buf(hdr->b_l1hdr.b_pabd, buf->b_data, | |
5814 | arc_buf_size(buf)); | |
5815 | } | |
d3c2ae1c | 5816 | } else { |
a6255b7f | 5817 | ASSERT3P(buf->b_data, ==, abd_to_buf(zio->io_orig_abd)); |
2aa34383 | 5818 | ASSERT3U(zio->io_orig_size, ==, arc_buf_size(buf)); |
d3c2ae1c | 5819 | ASSERT3U(hdr->b_l1hdr.b_bufcnt, ==, 1); |
d3c2ae1c | 5820 | |
d3c2ae1c | 5821 | arc_share_buf(hdr, buf); |
d3c2ae1c | 5822 | } |
a6255b7f | 5823 | |
d3c2ae1c | 5824 | arc_hdr_verify(hdr, zio->io_bp); |
a6255b7f | 5825 | spl_fstrans_unmark(cookie); |
34dc7c2f BB |
5826 | } |
5827 | ||
bc77ba73 PD |
5828 | static void |
5829 | arc_write_children_ready(zio_t *zio) | |
5830 | { | |
5831 | arc_write_callback_t *callback = zio->io_private; | |
5832 | arc_buf_t *buf = callback->awcb_buf; | |
5833 | ||
5834 | callback->awcb_children_ready(zio, buf, callback->awcb_private); | |
5835 | } | |
5836 | ||
e8b96c60 MA |
5837 | /* |
5838 | * The SPA calls this callback for each physical write that happens on behalf | |
5839 | * of a logical write. See the comment in dbuf_write_physdone() for details. | |
5840 | */ | |
5841 | static void | |
5842 | arc_write_physdone(zio_t *zio) | |
5843 | { | |
5844 | arc_write_callback_t *cb = zio->io_private; | |
5845 | if (cb->awcb_physdone != NULL) | |
5846 | cb->awcb_physdone(zio, cb->awcb_buf, cb->awcb_private); | |
5847 | } | |
5848 | ||
34dc7c2f BB |
5849 | static void |
5850 | arc_write_done(zio_t *zio) | |
5851 | { | |
5852 | arc_write_callback_t *callback = zio->io_private; | |
5853 | arc_buf_t *buf = callback->awcb_buf; | |
5854 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
5855 | ||
d3c2ae1c | 5856 | ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL); |
428870ff BB |
5857 | |
5858 | if (zio->io_error == 0) { | |
d3c2ae1c GW |
5859 | arc_hdr_verify(hdr, zio->io_bp); |
5860 | ||
9b67f605 | 5861 | if (BP_IS_HOLE(zio->io_bp) || BP_IS_EMBEDDED(zio->io_bp)) { |
b0bc7a84 MG |
5862 | buf_discard_identity(hdr); |
5863 | } else { | |
5864 | hdr->b_dva = *BP_IDENTITY(zio->io_bp); | |
5865 | hdr->b_birth = BP_PHYSICAL_BIRTH(zio->io_bp); | |
b0bc7a84 | 5866 | } |
428870ff | 5867 | } else { |
d3c2ae1c | 5868 | ASSERT(HDR_EMPTY(hdr)); |
428870ff | 5869 | } |
34dc7c2f | 5870 | |
34dc7c2f | 5871 | /* |
9b67f605 MA |
5872 | * If the block to be written was all-zero or compressed enough to be |
5873 | * embedded in the BP, no write was performed so there will be no | |
5874 | * dva/birth/checksum. The buffer must therefore remain anonymous | |
5875 | * (and uncached). | |
34dc7c2f | 5876 | */ |
d3c2ae1c | 5877 | if (!HDR_EMPTY(hdr)) { |
34dc7c2f BB |
5878 | arc_buf_hdr_t *exists; |
5879 | kmutex_t *hash_lock; | |
5880 | ||
524b4217 | 5881 | ASSERT3U(zio->io_error, ==, 0); |
428870ff | 5882 | |
34dc7c2f BB |
5883 | arc_cksum_verify(buf); |
5884 | ||
5885 | exists = buf_hash_insert(hdr, &hash_lock); | |
b9541d6b | 5886 | if (exists != NULL) { |
34dc7c2f BB |
5887 | /* |
5888 | * This can only happen if we overwrite for | |
5889 | * sync-to-convergence, because we remove | |
5890 | * buffers from the hash table when we arc_free(). | |
5891 | */ | |
428870ff BB |
5892 | if (zio->io_flags & ZIO_FLAG_IO_REWRITE) { |
5893 | if (!BP_EQUAL(&zio->io_bp_orig, zio->io_bp)) | |
5894 | panic("bad overwrite, hdr=%p exists=%p", | |
5895 | (void *)hdr, (void *)exists); | |
b9541d6b CW |
5896 | ASSERT(refcount_is_zero( |
5897 | &exists->b_l1hdr.b_refcnt)); | |
428870ff BB |
5898 | arc_change_state(arc_anon, exists, hash_lock); |
5899 | mutex_exit(hash_lock); | |
5900 | arc_hdr_destroy(exists); | |
5901 | exists = buf_hash_insert(hdr, &hash_lock); | |
5902 | ASSERT3P(exists, ==, NULL); | |
03c6040b GW |
5903 | } else if (zio->io_flags & ZIO_FLAG_NOPWRITE) { |
5904 | /* nopwrite */ | |
5905 | ASSERT(zio->io_prop.zp_nopwrite); | |
5906 | if (!BP_EQUAL(&zio->io_bp_orig, zio->io_bp)) | |
5907 | panic("bad nopwrite, hdr=%p exists=%p", | |
5908 | (void *)hdr, (void *)exists); | |
428870ff BB |
5909 | } else { |
5910 | /* Dedup */ | |
d3c2ae1c | 5911 | ASSERT(hdr->b_l1hdr.b_bufcnt == 1); |
b9541d6b | 5912 | ASSERT(hdr->b_l1hdr.b_state == arc_anon); |
428870ff BB |
5913 | ASSERT(BP_GET_DEDUP(zio->io_bp)); |
5914 | ASSERT(BP_GET_LEVEL(zio->io_bp) == 0); | |
5915 | } | |
34dc7c2f | 5916 | } |
d3c2ae1c | 5917 | arc_hdr_clear_flags(hdr, ARC_FLAG_IO_IN_PROGRESS); |
b128c09f | 5918 | /* if it's not anon, we are doing a scrub */ |
b9541d6b | 5919 | if (exists == NULL && hdr->b_l1hdr.b_state == arc_anon) |
b128c09f | 5920 | arc_access(hdr, hash_lock); |
34dc7c2f | 5921 | mutex_exit(hash_lock); |
34dc7c2f | 5922 | } else { |
d3c2ae1c | 5923 | arc_hdr_clear_flags(hdr, ARC_FLAG_IO_IN_PROGRESS); |
34dc7c2f BB |
5924 | } |
5925 | ||
b9541d6b | 5926 | ASSERT(!refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); |
428870ff | 5927 | callback->awcb_done(zio, buf, callback->awcb_private); |
34dc7c2f | 5928 | |
a6255b7f | 5929 | abd_put(zio->io_abd); |
34dc7c2f BB |
5930 | kmem_free(callback, sizeof (arc_write_callback_t)); |
5931 | } | |
5932 | ||
5933 | zio_t * | |
428870ff | 5934 | arc_write(zio_t *pio, spa_t *spa, uint64_t txg, |
d3c2ae1c | 5935 | blkptr_t *bp, arc_buf_t *buf, boolean_t l2arc, |
bc77ba73 PD |
5936 | const zio_prop_t *zp, arc_done_func_t *ready, |
5937 | arc_done_func_t *children_ready, arc_done_func_t *physdone, | |
e8b96c60 | 5938 | arc_done_func_t *done, void *private, zio_priority_t priority, |
5dbd68a3 | 5939 | int zio_flags, const zbookmark_phys_t *zb) |
34dc7c2f BB |
5940 | { |
5941 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
5942 | arc_write_callback_t *callback; | |
b128c09f | 5943 | zio_t *zio; |
34dc7c2f | 5944 | |
d3c2ae1c GW |
5945 | ASSERT3P(ready, !=, NULL); |
5946 | ASSERT3P(done, !=, NULL); | |
34dc7c2f | 5947 | ASSERT(!HDR_IO_ERROR(hdr)); |
b9541d6b | 5948 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); |
d3c2ae1c GW |
5949 | ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL); |
5950 | ASSERT3U(hdr->b_l1hdr.b_bufcnt, >, 0); | |
b128c09f | 5951 | if (l2arc) |
d3c2ae1c | 5952 | arc_hdr_set_flags(hdr, ARC_FLAG_L2CACHE); |
2aa34383 DK |
5953 | if (ARC_BUF_COMPRESSED(buf)) { |
5954 | ASSERT3U(zp->zp_compress, !=, ZIO_COMPRESS_OFF); | |
5955 | zio_flags |= ZIO_FLAG_RAW; | |
5956 | } | |
79c76d5b | 5957 | callback = kmem_zalloc(sizeof (arc_write_callback_t), KM_SLEEP); |
34dc7c2f | 5958 | callback->awcb_ready = ready; |
bc77ba73 | 5959 | callback->awcb_children_ready = children_ready; |
e8b96c60 | 5960 | callback->awcb_physdone = physdone; |
34dc7c2f BB |
5961 | callback->awcb_done = done; |
5962 | callback->awcb_private = private; | |
5963 | callback->awcb_buf = buf; | |
b128c09f | 5964 | |
d3c2ae1c | 5965 | /* |
a6255b7f | 5966 | * The hdr's b_pabd is now stale, free it now. A new data block |
d3c2ae1c GW |
5967 | * will be allocated when the zio pipeline calls arc_write_ready(). |
5968 | */ | |
a6255b7f | 5969 | if (hdr->b_l1hdr.b_pabd != NULL) { |
d3c2ae1c GW |
5970 | /* |
5971 | * If the buf is currently sharing the data block with | |
5972 | * the hdr then we need to break that relationship here. | |
5973 | * The hdr will remain with a NULL data pointer and the | |
5974 | * buf will take sole ownership of the block. | |
5975 | */ | |
5976 | if (arc_buf_is_shared(buf)) { | |
d3c2ae1c GW |
5977 | arc_unshare_buf(hdr, buf); |
5978 | } else { | |
a6255b7f | 5979 | arc_hdr_free_pabd(hdr); |
d3c2ae1c GW |
5980 | } |
5981 | VERIFY3P(buf->b_data, !=, NULL); | |
5982 | arc_hdr_set_compress(hdr, ZIO_COMPRESS_OFF); | |
5983 | } | |
5984 | ASSERT(!arc_buf_is_shared(buf)); | |
a6255b7f | 5985 | ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); |
d3c2ae1c | 5986 | |
a6255b7f DQ |
5987 | zio = zio_write(pio, spa, txg, bp, |
5988 | abd_get_from_buf(buf->b_data, HDR_GET_LSIZE(hdr)), | |
2aa34383 | 5989 | HDR_GET_LSIZE(hdr), arc_buf_size(buf), zp, |
bc77ba73 PD |
5990 | arc_write_ready, |
5991 | (children_ready != NULL) ? arc_write_children_ready : NULL, | |
5992 | arc_write_physdone, arc_write_done, callback, | |
e8b96c60 | 5993 | priority, zio_flags, zb); |
34dc7c2f BB |
5994 | |
5995 | return (zio); | |
5996 | } | |
5997 | ||
34dc7c2f | 5998 | static int |
e8b96c60 | 5999 | arc_memory_throttle(uint64_t reserve, uint64_t txg) |
34dc7c2f BB |
6000 | { |
6001 | #ifdef _KERNEL | |
7e8bddd0 BB |
6002 | uint64_t available_memory = ptob(freemem); |
6003 | static uint64_t page_load = 0; | |
6004 | static uint64_t last_txg = 0; | |
6005 | #ifdef __linux__ | |
6006 | pgcnt_t minfree = btop(arc_sys_free / 4); | |
6007 | #endif | |
0c5493d4 | 6008 | |
9edb3695 BB |
6009 | #if defined(__i386) |
6010 | available_memory = | |
6011 | MIN(available_memory, vmem_size(heap_arena, VMEM_FREE)); | |
6012 | #endif | |
6013 | ||
6014 | if (available_memory > arc_all_memory() * arc_lotsfree_percent / 100) | |
ca67b33a MA |
6015 | return (0); |
6016 | ||
7e8bddd0 BB |
6017 | if (txg > last_txg) { |
6018 | last_txg = txg; | |
6019 | page_load = 0; | |
6020 | } | |
7e8bddd0 BB |
6021 | /* |
6022 | * If we are in pageout, we know that memory is already tight, | |
6023 | * the arc is already going to be evicting, so we just want to | |
6024 | * continue to let page writes occur as quickly as possible. | |
6025 | */ | |
6026 | if (current_is_kswapd()) { | |
6027 | if (page_load > MAX(ptob(minfree), available_memory) / 4) { | |
6028 | DMU_TX_STAT_BUMP(dmu_tx_memory_reclaim); | |
6029 | return (SET_ERROR(ERESTART)); | |
6030 | } | |
6031 | /* Note: reserve is inflated, so we deflate */ | |
6032 | page_load += reserve / 8; | |
6033 | return (0); | |
6034 | } else if (page_load > 0 && arc_reclaim_needed()) { | |
ca67b33a | 6035 | /* memory is low, delay before restarting */ |
34dc7c2f | 6036 | ARCSTAT_INCR(arcstat_memory_throttle_count, 1); |
570827e1 | 6037 | DMU_TX_STAT_BUMP(dmu_tx_memory_reclaim); |
2e528b49 | 6038 | return (SET_ERROR(EAGAIN)); |
34dc7c2f | 6039 | } |
7e8bddd0 | 6040 | page_load = 0; |
34dc7c2f BB |
6041 | #endif |
6042 | return (0); | |
6043 | } | |
6044 | ||
6045 | void | |
6046 | arc_tempreserve_clear(uint64_t reserve) | |
6047 | { | |
6048 | atomic_add_64(&arc_tempreserve, -reserve); | |
6049 | ASSERT((int64_t)arc_tempreserve >= 0); | |
6050 | } | |
6051 | ||
6052 | int | |
6053 | arc_tempreserve_space(uint64_t reserve, uint64_t txg) | |
6054 | { | |
6055 | int error; | |
9babb374 | 6056 | uint64_t anon_size; |
34dc7c2f | 6057 | |
1b8951b3 TC |
6058 | if (!arc_no_grow && |
6059 | reserve > arc_c/4 && | |
6060 | reserve * 4 > (2ULL << SPA_MAXBLOCKSHIFT)) | |
34dc7c2f | 6061 | arc_c = MIN(arc_c_max, reserve * 4); |
12f9a6a3 BB |
6062 | |
6063 | /* | |
6064 | * Throttle when the calculated memory footprint for the TXG | |
6065 | * exceeds the target ARC size. | |
6066 | */ | |
570827e1 BB |
6067 | if (reserve > arc_c) { |
6068 | DMU_TX_STAT_BUMP(dmu_tx_memory_reserve); | |
12f9a6a3 | 6069 | return (SET_ERROR(ERESTART)); |
570827e1 | 6070 | } |
34dc7c2f | 6071 | |
9babb374 BB |
6072 | /* |
6073 | * Don't count loaned bufs as in flight dirty data to prevent long | |
6074 | * network delays from blocking transactions that are ready to be | |
6075 | * assigned to a txg. | |
6076 | */ | |
36da08ef PS |
6077 | anon_size = MAX((int64_t)(refcount_count(&arc_anon->arcs_size) - |
6078 | arc_loaned_bytes), 0); | |
9babb374 | 6079 | |
34dc7c2f BB |
6080 | /* |
6081 | * Writes will, almost always, require additional memory allocations | |
d3cc8b15 | 6082 | * in order to compress/encrypt/etc the data. We therefore need to |
34dc7c2f BB |
6083 | * make sure that there is sufficient available memory for this. |
6084 | */ | |
e8b96c60 MA |
6085 | error = arc_memory_throttle(reserve, txg); |
6086 | if (error != 0) | |
34dc7c2f BB |
6087 | return (error); |
6088 | ||
6089 | /* | |
6090 | * Throttle writes when the amount of dirty data in the cache | |
6091 | * gets too large. We try to keep the cache less than half full | |
6092 | * of dirty blocks so that our sync times don't grow too large. | |
6093 | * Note: if two requests come in concurrently, we might let them | |
6094 | * both succeed, when one of them should fail. Not a huge deal. | |
6095 | */ | |
9babb374 BB |
6096 | |
6097 | if (reserve + arc_tempreserve + anon_size > arc_c / 2 && | |
6098 | anon_size > arc_c / 4) { | |
d3c2ae1c GW |
6099 | uint64_t meta_esize = |
6100 | refcount_count(&arc_anon->arcs_esize[ARC_BUFC_METADATA]); | |
6101 | uint64_t data_esize = | |
6102 | refcount_count(&arc_anon->arcs_esize[ARC_BUFC_DATA]); | |
34dc7c2f BB |
6103 | dprintf("failing, arc_tempreserve=%lluK anon_meta=%lluK " |
6104 | "anon_data=%lluK tempreserve=%lluK arc_c=%lluK\n", | |
d3c2ae1c GW |
6105 | arc_tempreserve >> 10, meta_esize >> 10, |
6106 | data_esize >> 10, reserve >> 10, arc_c >> 10); | |
570827e1 | 6107 | DMU_TX_STAT_BUMP(dmu_tx_dirty_throttle); |
2e528b49 | 6108 | return (SET_ERROR(ERESTART)); |
34dc7c2f BB |
6109 | } |
6110 | atomic_add_64(&arc_tempreserve, reserve); | |
6111 | return (0); | |
6112 | } | |
6113 | ||
13be560d BB |
6114 | static void |
6115 | arc_kstat_update_state(arc_state_t *state, kstat_named_t *size, | |
6116 | kstat_named_t *evict_data, kstat_named_t *evict_metadata) | |
6117 | { | |
36da08ef | 6118 | size->value.ui64 = refcount_count(&state->arcs_size); |
d3c2ae1c GW |
6119 | evict_data->value.ui64 = |
6120 | refcount_count(&state->arcs_esize[ARC_BUFC_DATA]); | |
6121 | evict_metadata->value.ui64 = | |
6122 | refcount_count(&state->arcs_esize[ARC_BUFC_METADATA]); | |
13be560d BB |
6123 | } |
6124 | ||
6125 | static int | |
6126 | arc_kstat_update(kstat_t *ksp, int rw) | |
6127 | { | |
6128 | arc_stats_t *as = ksp->ks_data; | |
6129 | ||
6130 | if (rw == KSTAT_WRITE) { | |
500445c0 | 6131 | return (EACCES); |
13be560d BB |
6132 | } else { |
6133 | arc_kstat_update_state(arc_anon, | |
6134 | &as->arcstat_anon_size, | |
500445c0 PS |
6135 | &as->arcstat_anon_evictable_data, |
6136 | &as->arcstat_anon_evictable_metadata); | |
13be560d BB |
6137 | arc_kstat_update_state(arc_mru, |
6138 | &as->arcstat_mru_size, | |
500445c0 PS |
6139 | &as->arcstat_mru_evictable_data, |
6140 | &as->arcstat_mru_evictable_metadata); | |
13be560d BB |
6141 | arc_kstat_update_state(arc_mru_ghost, |
6142 | &as->arcstat_mru_ghost_size, | |
500445c0 PS |
6143 | &as->arcstat_mru_ghost_evictable_data, |
6144 | &as->arcstat_mru_ghost_evictable_metadata); | |
13be560d BB |
6145 | arc_kstat_update_state(arc_mfu, |
6146 | &as->arcstat_mfu_size, | |
500445c0 PS |
6147 | &as->arcstat_mfu_evictable_data, |
6148 | &as->arcstat_mfu_evictable_metadata); | |
fc41c640 | 6149 | arc_kstat_update_state(arc_mfu_ghost, |
13be560d | 6150 | &as->arcstat_mfu_ghost_size, |
500445c0 PS |
6151 | &as->arcstat_mfu_ghost_evictable_data, |
6152 | &as->arcstat_mfu_ghost_evictable_metadata); | |
13be560d BB |
6153 | } |
6154 | ||
6155 | return (0); | |
6156 | } | |
6157 | ||
ca0bf58d PS |
6158 | /* |
6159 | * This function *must* return indices evenly distributed between all | |
6160 | * sublists of the multilist. This is needed due to how the ARC eviction | |
6161 | * code is laid out; arc_evict_state() assumes ARC buffers are evenly | |
6162 | * distributed between all sublists and uses this assumption when | |
6163 | * deciding which sublist to evict from and how much to evict from it. | |
6164 | */ | |
6165 | unsigned int | |
6166 | arc_state_multilist_index_func(multilist_t *ml, void *obj) | |
6167 | { | |
6168 | arc_buf_hdr_t *hdr = obj; | |
6169 | ||
6170 | /* | |
6171 | * We rely on b_dva to generate evenly distributed index | |
6172 | * numbers using buf_hash below. So, as an added precaution, | |
6173 | * let's make sure we never add empty buffers to the arc lists. | |
6174 | */ | |
d3c2ae1c | 6175 | ASSERT(!HDR_EMPTY(hdr)); |
ca0bf58d PS |
6176 | |
6177 | /* | |
6178 | * The assumption here, is the hash value for a given | |
6179 | * arc_buf_hdr_t will remain constant throughout its lifetime | |
6180 | * (i.e. its b_spa, b_dva, and b_birth fields don't change). | |
6181 | * Thus, we don't need to store the header's sublist index | |
6182 | * on insertion, as this index can be recalculated on removal. | |
6183 | * | |
6184 | * Also, the low order bits of the hash value are thought to be | |
6185 | * distributed evenly. Otherwise, in the case that the multilist | |
6186 | * has a power of two number of sublists, each sublists' usage | |
6187 | * would not be evenly distributed. | |
6188 | */ | |
6189 | return (buf_hash(hdr->b_spa, &hdr->b_dva, hdr->b_birth) % | |
6190 | multilist_get_num_sublists(ml)); | |
6191 | } | |
6192 | ||
ca67b33a MA |
6193 | /* |
6194 | * Called during module initialization and periodically thereafter to | |
6195 | * apply reasonable changes to the exposed performance tunings. Non-zero | |
6196 | * zfs_* values which differ from the currently set values will be applied. | |
6197 | */ | |
6198 | static void | |
6199 | arc_tuning_update(void) | |
6200 | { | |
9edb3695 BB |
6201 | uint64_t percent, allmem = arc_all_memory(); |
6202 | ||
ca67b33a MA |
6203 | /* Valid range: 64M - <all physical memory> */ |
6204 | if ((zfs_arc_max) && (zfs_arc_max != arc_c_max) && | |
9edb3695 | 6205 | (zfs_arc_max > 64 << 20) && (zfs_arc_max < allmem) && |
ca67b33a MA |
6206 | (zfs_arc_max > arc_c_min)) { |
6207 | arc_c_max = zfs_arc_max; | |
6208 | arc_c = arc_c_max; | |
6209 | arc_p = (arc_c >> 1); | |
9907cc1c G |
6210 | /* Valid range of arc_meta_limit: arc_meta_min - arc_c_max */ |
6211 | percent = MIN(zfs_arc_meta_limit_percent, 100); | |
6212 | arc_meta_limit = MAX(arc_meta_min, (percent * arc_c_max) / 100); | |
6213 | percent = MIN(zfs_arc_dnode_limit_percent, 100); | |
6214 | arc_dnode_limit = (percent * arc_meta_limit) / 100; | |
ca67b33a MA |
6215 | } |
6216 | ||
6217 | /* Valid range: 32M - <arc_c_max> */ | |
6218 | if ((zfs_arc_min) && (zfs_arc_min != arc_c_min) && | |
6219 | (zfs_arc_min >= 2ULL << SPA_MAXBLOCKSHIFT) && | |
6220 | (zfs_arc_min <= arc_c_max)) { | |
6221 | arc_c_min = zfs_arc_min; | |
6222 | arc_c = MAX(arc_c, arc_c_min); | |
6223 | } | |
6224 | ||
6225 | /* Valid range: 16M - <arc_c_max> */ | |
6226 | if ((zfs_arc_meta_min) && (zfs_arc_meta_min != arc_meta_min) && | |
6227 | (zfs_arc_meta_min >= 1ULL << SPA_MAXBLOCKSHIFT) && | |
6228 | (zfs_arc_meta_min <= arc_c_max)) { | |
6229 | arc_meta_min = zfs_arc_meta_min; | |
6230 | arc_meta_limit = MAX(arc_meta_limit, arc_meta_min); | |
25458cbe | 6231 | arc_dnode_limit = arc_meta_limit / 10; |
ca67b33a MA |
6232 | } |
6233 | ||
6234 | /* Valid range: <arc_meta_min> - <arc_c_max> */ | |
6235 | if ((zfs_arc_meta_limit) && (zfs_arc_meta_limit != arc_meta_limit) && | |
6236 | (zfs_arc_meta_limit >= zfs_arc_meta_min) && | |
6237 | (zfs_arc_meta_limit <= arc_c_max)) | |
6238 | arc_meta_limit = zfs_arc_meta_limit; | |
6239 | ||
25458cbe TC |
6240 | /* Valid range: <arc_meta_min> - <arc_c_max> */ |
6241 | if ((zfs_arc_dnode_limit) && (zfs_arc_dnode_limit != arc_dnode_limit) && | |
6242 | (zfs_arc_dnode_limit >= zfs_arc_meta_min) && | |
6243 | (zfs_arc_dnode_limit <= arc_c_max)) | |
6244 | arc_dnode_limit = zfs_arc_dnode_limit; | |
6245 | ||
ca67b33a MA |
6246 | /* Valid range: 1 - N */ |
6247 | if (zfs_arc_grow_retry) | |
6248 | arc_grow_retry = zfs_arc_grow_retry; | |
6249 | ||
6250 | /* Valid range: 1 - N */ | |
6251 | if (zfs_arc_shrink_shift) { | |
6252 | arc_shrink_shift = zfs_arc_shrink_shift; | |
6253 | arc_no_grow_shift = MIN(arc_no_grow_shift, arc_shrink_shift -1); | |
6254 | } | |
6255 | ||
728d6ae9 BB |
6256 | /* Valid range: 1 - N */ |
6257 | if (zfs_arc_p_min_shift) | |
6258 | arc_p_min_shift = zfs_arc_p_min_shift; | |
6259 | ||
ca67b33a MA |
6260 | /* Valid range: 1 - N ticks */ |
6261 | if (zfs_arc_min_prefetch_lifespan) | |
6262 | arc_min_prefetch_lifespan = zfs_arc_min_prefetch_lifespan; | |
11f552fa | 6263 | |
7e8bddd0 BB |
6264 | /* Valid range: 0 - 100 */ |
6265 | if ((zfs_arc_lotsfree_percent >= 0) && | |
6266 | (zfs_arc_lotsfree_percent <= 100)) | |
6267 | arc_lotsfree_percent = zfs_arc_lotsfree_percent; | |
6268 | ||
11f552fa BB |
6269 | /* Valid range: 0 - <all physical memory> */ |
6270 | if ((zfs_arc_sys_free) && (zfs_arc_sys_free != arc_sys_free)) | |
9edb3695 | 6271 | arc_sys_free = MIN(MAX(zfs_arc_sys_free, 0), allmem); |
7e8bddd0 | 6272 | |
ca67b33a MA |
6273 | } |
6274 | ||
d3c2ae1c GW |
6275 | static void |
6276 | arc_state_init(void) | |
6277 | { | |
6278 | arc_anon = &ARC_anon; | |
6279 | arc_mru = &ARC_mru; | |
6280 | arc_mru_ghost = &ARC_mru_ghost; | |
6281 | arc_mfu = &ARC_mfu; | |
6282 | arc_mfu_ghost = &ARC_mfu_ghost; | |
6283 | arc_l2c_only = &ARC_l2c_only; | |
6284 | ||
6285 | multilist_create(&arc_mru->arcs_list[ARC_BUFC_METADATA], | |
6286 | sizeof (arc_buf_hdr_t), | |
6287 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
6288 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
6289 | multilist_create(&arc_mru->arcs_list[ARC_BUFC_DATA], | |
6290 | sizeof (arc_buf_hdr_t), | |
6291 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
6292 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
6293 | multilist_create(&arc_mru_ghost->arcs_list[ARC_BUFC_METADATA], | |
6294 | sizeof (arc_buf_hdr_t), | |
6295 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
6296 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
6297 | multilist_create(&arc_mru_ghost->arcs_list[ARC_BUFC_DATA], | |
6298 | sizeof (arc_buf_hdr_t), | |
6299 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
6300 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
6301 | multilist_create(&arc_mfu->arcs_list[ARC_BUFC_METADATA], | |
6302 | sizeof (arc_buf_hdr_t), | |
6303 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
6304 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
6305 | multilist_create(&arc_mfu->arcs_list[ARC_BUFC_DATA], | |
6306 | sizeof (arc_buf_hdr_t), | |
6307 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
6308 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
6309 | multilist_create(&arc_mfu_ghost->arcs_list[ARC_BUFC_METADATA], | |
6310 | sizeof (arc_buf_hdr_t), | |
6311 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
6312 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
6313 | multilist_create(&arc_mfu_ghost->arcs_list[ARC_BUFC_DATA], | |
6314 | sizeof (arc_buf_hdr_t), | |
6315 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
6316 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
6317 | multilist_create(&arc_l2c_only->arcs_list[ARC_BUFC_METADATA], | |
6318 | sizeof (arc_buf_hdr_t), | |
6319 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
6320 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
6321 | multilist_create(&arc_l2c_only->arcs_list[ARC_BUFC_DATA], | |
6322 | sizeof (arc_buf_hdr_t), | |
6323 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
6324 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
6325 | ||
6326 | refcount_create(&arc_anon->arcs_esize[ARC_BUFC_METADATA]); | |
6327 | refcount_create(&arc_anon->arcs_esize[ARC_BUFC_DATA]); | |
6328 | refcount_create(&arc_mru->arcs_esize[ARC_BUFC_METADATA]); | |
6329 | refcount_create(&arc_mru->arcs_esize[ARC_BUFC_DATA]); | |
6330 | refcount_create(&arc_mru_ghost->arcs_esize[ARC_BUFC_METADATA]); | |
6331 | refcount_create(&arc_mru_ghost->arcs_esize[ARC_BUFC_DATA]); | |
6332 | refcount_create(&arc_mfu->arcs_esize[ARC_BUFC_METADATA]); | |
6333 | refcount_create(&arc_mfu->arcs_esize[ARC_BUFC_DATA]); | |
6334 | refcount_create(&arc_mfu_ghost->arcs_esize[ARC_BUFC_METADATA]); | |
6335 | refcount_create(&arc_mfu_ghost->arcs_esize[ARC_BUFC_DATA]); | |
6336 | refcount_create(&arc_l2c_only->arcs_esize[ARC_BUFC_METADATA]); | |
6337 | refcount_create(&arc_l2c_only->arcs_esize[ARC_BUFC_DATA]); | |
6338 | ||
6339 | refcount_create(&arc_anon->arcs_size); | |
6340 | refcount_create(&arc_mru->arcs_size); | |
6341 | refcount_create(&arc_mru_ghost->arcs_size); | |
6342 | refcount_create(&arc_mfu->arcs_size); | |
6343 | refcount_create(&arc_mfu_ghost->arcs_size); | |
6344 | refcount_create(&arc_l2c_only->arcs_size); | |
6345 | ||
6346 | arc_anon->arcs_state = ARC_STATE_ANON; | |
6347 | arc_mru->arcs_state = ARC_STATE_MRU; | |
6348 | arc_mru_ghost->arcs_state = ARC_STATE_MRU_GHOST; | |
6349 | arc_mfu->arcs_state = ARC_STATE_MFU; | |
6350 | arc_mfu_ghost->arcs_state = ARC_STATE_MFU_GHOST; | |
6351 | arc_l2c_only->arcs_state = ARC_STATE_L2C_ONLY; | |
6352 | } | |
6353 | ||
6354 | static void | |
6355 | arc_state_fini(void) | |
6356 | { | |
6357 | refcount_destroy(&arc_anon->arcs_esize[ARC_BUFC_METADATA]); | |
6358 | refcount_destroy(&arc_anon->arcs_esize[ARC_BUFC_DATA]); | |
6359 | refcount_destroy(&arc_mru->arcs_esize[ARC_BUFC_METADATA]); | |
6360 | refcount_destroy(&arc_mru->arcs_esize[ARC_BUFC_DATA]); | |
6361 | refcount_destroy(&arc_mru_ghost->arcs_esize[ARC_BUFC_METADATA]); | |
6362 | refcount_destroy(&arc_mru_ghost->arcs_esize[ARC_BUFC_DATA]); | |
6363 | refcount_destroy(&arc_mfu->arcs_esize[ARC_BUFC_METADATA]); | |
6364 | refcount_destroy(&arc_mfu->arcs_esize[ARC_BUFC_DATA]); | |
6365 | refcount_destroy(&arc_mfu_ghost->arcs_esize[ARC_BUFC_METADATA]); | |
6366 | refcount_destroy(&arc_mfu_ghost->arcs_esize[ARC_BUFC_DATA]); | |
6367 | refcount_destroy(&arc_l2c_only->arcs_esize[ARC_BUFC_METADATA]); | |
6368 | refcount_destroy(&arc_l2c_only->arcs_esize[ARC_BUFC_DATA]); | |
6369 | ||
6370 | refcount_destroy(&arc_anon->arcs_size); | |
6371 | refcount_destroy(&arc_mru->arcs_size); | |
6372 | refcount_destroy(&arc_mru_ghost->arcs_size); | |
6373 | refcount_destroy(&arc_mfu->arcs_size); | |
6374 | refcount_destroy(&arc_mfu_ghost->arcs_size); | |
6375 | refcount_destroy(&arc_l2c_only->arcs_size); | |
6376 | ||
6377 | multilist_destroy(&arc_mru->arcs_list[ARC_BUFC_METADATA]); | |
6378 | multilist_destroy(&arc_mru_ghost->arcs_list[ARC_BUFC_METADATA]); | |
6379 | multilist_destroy(&arc_mfu->arcs_list[ARC_BUFC_METADATA]); | |
6380 | multilist_destroy(&arc_mfu_ghost->arcs_list[ARC_BUFC_METADATA]); | |
6381 | multilist_destroy(&arc_mru->arcs_list[ARC_BUFC_DATA]); | |
6382 | multilist_destroy(&arc_mru_ghost->arcs_list[ARC_BUFC_DATA]); | |
6383 | multilist_destroy(&arc_mfu->arcs_list[ARC_BUFC_DATA]); | |
6384 | multilist_destroy(&arc_mfu_ghost->arcs_list[ARC_BUFC_DATA]); | |
75710332 BB |
6385 | multilist_destroy(&arc_l2c_only->arcs_list[ARC_BUFC_METADATA]); |
6386 | multilist_destroy(&arc_l2c_only->arcs_list[ARC_BUFC_DATA]); | |
d3c2ae1c GW |
6387 | } |
6388 | ||
6389 | uint64_t | |
6390 | arc_max_bytes(void) | |
6391 | { | |
6392 | return (arc_c_max); | |
6393 | } | |
6394 | ||
34dc7c2f BB |
6395 | void |
6396 | arc_init(void) | |
6397 | { | |
9edb3695 | 6398 | uint64_t percent, allmem = arc_all_memory(); |
ca67b33a | 6399 | |
ca0bf58d PS |
6400 | mutex_init(&arc_reclaim_lock, NULL, MUTEX_DEFAULT, NULL); |
6401 | cv_init(&arc_reclaim_thread_cv, NULL, CV_DEFAULT, NULL); | |
6402 | cv_init(&arc_reclaim_waiters_cv, NULL, CV_DEFAULT, NULL); | |
6403 | ||
34dc7c2f | 6404 | /* Convert seconds to clock ticks */ |
ca67b33a | 6405 | arc_min_prefetch_lifespan = 1 * hz; |
34dc7c2f | 6406 | |
34dc7c2f | 6407 | #ifdef _KERNEL |
7cb67b45 BB |
6408 | /* |
6409 | * Register a shrinker to support synchronous (direct) memory | |
6410 | * reclaim from the arc. This is done to prevent kswapd from | |
6411 | * swapping out pages when it is preferable to shrink the arc. | |
6412 | */ | |
6413 | spl_register_shrinker(&arc_shrinker); | |
11f552fa BB |
6414 | |
6415 | /* Set to 1/64 of all memory or a minimum of 512K */ | |
9edb3695 | 6416 | arc_sys_free = MAX(allmem / 64, (512 * 1024)); |
11f552fa | 6417 | arc_need_free = 0; |
34dc7c2f BB |
6418 | #endif |
6419 | ||
0a1f8cd9 TC |
6420 | /* Set max to 1/2 of all memory */ |
6421 | arc_c_max = allmem / 2; | |
6422 | ||
ab5cbbd1 BB |
6423 | /* |
6424 | * In userland, there's only the memory pressure that we artificially | |
6425 | * create (see arc_available_memory()). Don't let arc_c get too | |
6426 | * small, because it can cause transactions to be larger than | |
6427 | * arc_c, causing arc_tempreserve_space() to fail. | |
6428 | */ | |
6429 | #ifndef _KERNEL | |
0a1f8cd9 | 6430 | arc_c_min = MAX(arc_c_max / 2, 2ULL << SPA_MAXBLOCKSHIFT); |
ab5cbbd1 | 6431 | #else |
121b3cae | 6432 | arc_c_min = 2ULL << SPA_MAXBLOCKSHIFT; |
ab5cbbd1 BB |
6433 | #endif |
6434 | ||
34dc7c2f BB |
6435 | arc_c = arc_c_max; |
6436 | arc_p = (arc_c >> 1); | |
d3c2ae1c | 6437 | arc_size = 0; |
34dc7c2f | 6438 | |
ca67b33a MA |
6439 | /* Set min to 1/2 of arc_c_min */ |
6440 | arc_meta_min = 1ULL << SPA_MAXBLOCKSHIFT; | |
6441 | /* Initialize maximum observed usage to zero */ | |
1834f2d8 | 6442 | arc_meta_max = 0; |
9907cc1c G |
6443 | /* |
6444 | * Set arc_meta_limit to a percent of arc_c_max with a floor of | |
6445 | * arc_meta_min, and a ceiling of arc_c_max. | |
6446 | */ | |
6447 | percent = MIN(zfs_arc_meta_limit_percent, 100); | |
6448 | arc_meta_limit = MAX(arc_meta_min, (percent * arc_c_max) / 100); | |
6449 | percent = MIN(zfs_arc_dnode_limit_percent, 100); | |
6450 | arc_dnode_limit = (percent * arc_meta_limit) / 100; | |
34dc7c2f | 6451 | |
ca67b33a MA |
6452 | /* Apply user specified tunings */ |
6453 | arc_tuning_update(); | |
c52fca13 | 6454 | |
ca0bf58d | 6455 | if (zfs_arc_num_sublists_per_state < 1) |
ca67b33a | 6456 | zfs_arc_num_sublists_per_state = MAX(boot_ncpus, 1); |
ca0bf58d | 6457 | |
34dc7c2f BB |
6458 | /* if kmem_flags are set, lets try to use less memory */ |
6459 | if (kmem_debugging()) | |
6460 | arc_c = arc_c / 2; | |
6461 | if (arc_c < arc_c_min) | |
6462 | arc_c = arc_c_min; | |
6463 | ||
d3c2ae1c | 6464 | arc_state_init(); |
34dc7c2f BB |
6465 | buf_init(); |
6466 | ||
ab26409d BB |
6467 | list_create(&arc_prune_list, sizeof (arc_prune_t), |
6468 | offsetof(arc_prune_t, p_node)); | |
ab26409d | 6469 | mutex_init(&arc_prune_mtx, NULL, MUTEX_DEFAULT, NULL); |
34dc7c2f | 6470 | |
1229323d | 6471 | arc_prune_taskq = taskq_create("arc_prune", max_ncpus, defclsyspri, |
aa9af22c | 6472 | max_ncpus, INT_MAX, TASKQ_PREPOPULATE | TASKQ_DYNAMIC); |
f6046738 | 6473 | |
d3c2ae1c GW |
6474 | arc_reclaim_thread_exit = B_FALSE; |
6475 | ||
34dc7c2f BB |
6476 | arc_ksp = kstat_create("zfs", 0, "arcstats", "misc", KSTAT_TYPE_NAMED, |
6477 | sizeof (arc_stats) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL); | |
6478 | ||
6479 | if (arc_ksp != NULL) { | |
6480 | arc_ksp->ks_data = &arc_stats; | |
13be560d | 6481 | arc_ksp->ks_update = arc_kstat_update; |
34dc7c2f BB |
6482 | kstat_install(arc_ksp); |
6483 | } | |
6484 | ||
ca67b33a | 6485 | (void) thread_create(NULL, 0, arc_reclaim_thread, NULL, 0, &p0, |
1229323d | 6486 | TS_RUN, defclsyspri); |
34dc7c2f | 6487 | |
d3c2ae1c | 6488 | arc_dead = B_FALSE; |
b128c09f | 6489 | arc_warm = B_FALSE; |
34dc7c2f | 6490 | |
e8b96c60 MA |
6491 | /* |
6492 | * Calculate maximum amount of dirty data per pool. | |
6493 | * | |
6494 | * If it has been set by a module parameter, take that. | |
6495 | * Otherwise, use a percentage of physical memory defined by | |
6496 | * zfs_dirty_data_max_percent (default 10%) with a cap at | |
6497 | * zfs_dirty_data_max_max (default 25% of physical memory). | |
6498 | */ | |
6499 | if (zfs_dirty_data_max_max == 0) | |
9edb3695 | 6500 | zfs_dirty_data_max_max = allmem * |
e8b96c60 MA |
6501 | zfs_dirty_data_max_max_percent / 100; |
6502 | ||
6503 | if (zfs_dirty_data_max == 0) { | |
9edb3695 | 6504 | zfs_dirty_data_max = allmem * |
e8b96c60 MA |
6505 | zfs_dirty_data_max_percent / 100; |
6506 | zfs_dirty_data_max = MIN(zfs_dirty_data_max, | |
6507 | zfs_dirty_data_max_max); | |
6508 | } | |
34dc7c2f BB |
6509 | } |
6510 | ||
6511 | void | |
6512 | arc_fini(void) | |
6513 | { | |
ab26409d BB |
6514 | arc_prune_t *p; |
6515 | ||
7cb67b45 BB |
6516 | #ifdef _KERNEL |
6517 | spl_unregister_shrinker(&arc_shrinker); | |
6518 | #endif /* _KERNEL */ | |
6519 | ||
ca0bf58d | 6520 | mutex_enter(&arc_reclaim_lock); |
d3c2ae1c | 6521 | arc_reclaim_thread_exit = B_TRUE; |
ca0bf58d PS |
6522 | /* |
6523 | * The reclaim thread will set arc_reclaim_thread_exit back to | |
d3c2ae1c | 6524 | * B_FALSE when it is finished exiting; we're waiting for that. |
ca0bf58d PS |
6525 | */ |
6526 | while (arc_reclaim_thread_exit) { | |
6527 | cv_signal(&arc_reclaim_thread_cv); | |
6528 | cv_wait(&arc_reclaim_thread_cv, &arc_reclaim_lock); | |
6529 | } | |
6530 | mutex_exit(&arc_reclaim_lock); | |
6531 | ||
d3c2ae1c GW |
6532 | /* Use B_TRUE to ensure *all* buffers are evicted */ |
6533 | arc_flush(NULL, B_TRUE); | |
34dc7c2f | 6534 | |
d3c2ae1c | 6535 | arc_dead = B_TRUE; |
34dc7c2f BB |
6536 | |
6537 | if (arc_ksp != NULL) { | |
6538 | kstat_delete(arc_ksp); | |
6539 | arc_ksp = NULL; | |
6540 | } | |
6541 | ||
f6046738 BB |
6542 | taskq_wait(arc_prune_taskq); |
6543 | taskq_destroy(arc_prune_taskq); | |
6544 | ||
ab26409d BB |
6545 | mutex_enter(&arc_prune_mtx); |
6546 | while ((p = list_head(&arc_prune_list)) != NULL) { | |
6547 | list_remove(&arc_prune_list, p); | |
6548 | refcount_remove(&p->p_refcnt, &arc_prune_list); | |
6549 | refcount_destroy(&p->p_refcnt); | |
6550 | kmem_free(p, sizeof (*p)); | |
6551 | } | |
6552 | mutex_exit(&arc_prune_mtx); | |
6553 | ||
6554 | list_destroy(&arc_prune_list); | |
6555 | mutex_destroy(&arc_prune_mtx); | |
ca0bf58d PS |
6556 | mutex_destroy(&arc_reclaim_lock); |
6557 | cv_destroy(&arc_reclaim_thread_cv); | |
6558 | cv_destroy(&arc_reclaim_waiters_cv); | |
6559 | ||
d3c2ae1c | 6560 | arc_state_fini(); |
34dc7c2f | 6561 | buf_fini(); |
9babb374 | 6562 | |
b9541d6b | 6563 | ASSERT0(arc_loaned_bytes); |
34dc7c2f BB |
6564 | } |
6565 | ||
6566 | /* | |
6567 | * Level 2 ARC | |
6568 | * | |
6569 | * The level 2 ARC (L2ARC) is a cache layer in-between main memory and disk. | |
6570 | * It uses dedicated storage devices to hold cached data, which are populated | |
6571 | * using large infrequent writes. The main role of this cache is to boost | |
6572 | * the performance of random read workloads. The intended L2ARC devices | |
6573 | * include short-stroked disks, solid state disks, and other media with | |
6574 | * substantially faster read latency than disk. | |
6575 | * | |
6576 | * +-----------------------+ | |
6577 | * | ARC | | |
6578 | * +-----------------------+ | |
6579 | * | ^ ^ | |
6580 | * | | | | |
6581 | * l2arc_feed_thread() arc_read() | |
6582 | * | | | | |
6583 | * | l2arc read | | |
6584 | * V | | | |
6585 | * +---------------+ | | |
6586 | * | L2ARC | | | |
6587 | * +---------------+ | | |
6588 | * | ^ | | |
6589 | * l2arc_write() | | | |
6590 | * | | | | |
6591 | * V | | | |
6592 | * +-------+ +-------+ | |
6593 | * | vdev | | vdev | | |
6594 | * | cache | | cache | | |
6595 | * +-------+ +-------+ | |
6596 | * +=========+ .-----. | |
6597 | * : L2ARC : |-_____-| | |
6598 | * : devices : | Disks | | |
6599 | * +=========+ `-_____-' | |
6600 | * | |
6601 | * Read requests are satisfied from the following sources, in order: | |
6602 | * | |
6603 | * 1) ARC | |
6604 | * 2) vdev cache of L2ARC devices | |
6605 | * 3) L2ARC devices | |
6606 | * 4) vdev cache of disks | |
6607 | * 5) disks | |
6608 | * | |
6609 | * Some L2ARC device types exhibit extremely slow write performance. | |
6610 | * To accommodate for this there are some significant differences between | |
6611 | * the L2ARC and traditional cache design: | |
6612 | * | |
6613 | * 1. There is no eviction path from the ARC to the L2ARC. Evictions from | |
6614 | * the ARC behave as usual, freeing buffers and placing headers on ghost | |
6615 | * lists. The ARC does not send buffers to the L2ARC during eviction as | |
6616 | * this would add inflated write latencies for all ARC memory pressure. | |
6617 | * | |
6618 | * 2. The L2ARC attempts to cache data from the ARC before it is evicted. | |
6619 | * It does this by periodically scanning buffers from the eviction-end of | |
6620 | * the MFU and MRU ARC lists, copying them to the L2ARC devices if they are | |
3a17a7a9 SK |
6621 | * not already there. It scans until a headroom of buffers is satisfied, |
6622 | * which itself is a buffer for ARC eviction. If a compressible buffer is | |
6623 | * found during scanning and selected for writing to an L2ARC device, we | |
6624 | * temporarily boost scanning headroom during the next scan cycle to make | |
6625 | * sure we adapt to compression effects (which might significantly reduce | |
6626 | * the data volume we write to L2ARC). The thread that does this is | |
34dc7c2f BB |
6627 | * l2arc_feed_thread(), illustrated below; example sizes are included to |
6628 | * provide a better sense of ratio than this diagram: | |
6629 | * | |
6630 | * head --> tail | |
6631 | * +---------------------+----------+ | |
6632 | * ARC_mfu |:::::#:::::::::::::::|o#o###o###|-->. # already on L2ARC | |
6633 | * +---------------------+----------+ | o L2ARC eligible | |
6634 | * ARC_mru |:#:::::::::::::::::::|#o#ooo####|-->| : ARC buffer | |
6635 | * +---------------------+----------+ | | |
6636 | * 15.9 Gbytes ^ 32 Mbytes | | |
6637 | * headroom | | |
6638 | * l2arc_feed_thread() | |
6639 | * | | |
6640 | * l2arc write hand <--[oooo]--' | |
6641 | * | 8 Mbyte | |
6642 | * | write max | |
6643 | * V | |
6644 | * +==============================+ | |
6645 | * L2ARC dev |####|#|###|###| |####| ... | | |
6646 | * +==============================+ | |
6647 | * 32 Gbytes | |
6648 | * | |
6649 | * 3. If an ARC buffer is copied to the L2ARC but then hit instead of | |
6650 | * evicted, then the L2ARC has cached a buffer much sooner than it probably | |
6651 | * needed to, potentially wasting L2ARC device bandwidth and storage. It is | |
6652 | * safe to say that this is an uncommon case, since buffers at the end of | |
6653 | * the ARC lists have moved there due to inactivity. | |
6654 | * | |
6655 | * 4. If the ARC evicts faster than the L2ARC can maintain a headroom, | |
6656 | * then the L2ARC simply misses copying some buffers. This serves as a | |
6657 | * pressure valve to prevent heavy read workloads from both stalling the ARC | |
6658 | * with waits and clogging the L2ARC with writes. This also helps prevent | |
6659 | * the potential for the L2ARC to churn if it attempts to cache content too | |
6660 | * quickly, such as during backups of the entire pool. | |
6661 | * | |
b128c09f BB |
6662 | * 5. After system boot and before the ARC has filled main memory, there are |
6663 | * no evictions from the ARC and so the tails of the ARC_mfu and ARC_mru | |
6664 | * lists can remain mostly static. Instead of searching from tail of these | |
6665 | * lists as pictured, the l2arc_feed_thread() will search from the list heads | |
6666 | * for eligible buffers, greatly increasing its chance of finding them. | |
6667 | * | |
6668 | * The L2ARC device write speed is also boosted during this time so that | |
6669 | * the L2ARC warms up faster. Since there have been no ARC evictions yet, | |
6670 | * there are no L2ARC reads, and no fear of degrading read performance | |
6671 | * through increased writes. | |
6672 | * | |
6673 | * 6. Writes to the L2ARC devices are grouped and sent in-sequence, so that | |
34dc7c2f BB |
6674 | * the vdev queue can aggregate them into larger and fewer writes. Each |
6675 | * device is written to in a rotor fashion, sweeping writes through | |
6676 | * available space then repeating. | |
6677 | * | |
b128c09f | 6678 | * 7. The L2ARC does not store dirty content. It never needs to flush |
34dc7c2f BB |
6679 | * write buffers back to disk based storage. |
6680 | * | |
b128c09f | 6681 | * 8. If an ARC buffer is written (and dirtied) which also exists in the |
34dc7c2f BB |
6682 | * L2ARC, the now stale L2ARC buffer is immediately dropped. |
6683 | * | |
6684 | * The performance of the L2ARC can be tweaked by a number of tunables, which | |
6685 | * may be necessary for different workloads: | |
6686 | * | |
6687 | * l2arc_write_max max write bytes per interval | |
b128c09f | 6688 | * l2arc_write_boost extra write bytes during device warmup |
34dc7c2f BB |
6689 | * l2arc_noprefetch skip caching prefetched buffers |
6690 | * l2arc_headroom number of max device writes to precache | |
3a17a7a9 SK |
6691 | * l2arc_headroom_boost when we find compressed buffers during ARC |
6692 | * scanning, we multiply headroom by this | |
6693 | * percentage factor for the next scan cycle, | |
6694 | * since more compressed buffers are likely to | |
6695 | * be present | |
34dc7c2f BB |
6696 | * l2arc_feed_secs seconds between L2ARC writing |
6697 | * | |
6698 | * Tunables may be removed or added as future performance improvements are | |
6699 | * integrated, and also may become zpool properties. | |
d164b209 BB |
6700 | * |
6701 | * There are three key functions that control how the L2ARC warms up: | |
6702 | * | |
6703 | * l2arc_write_eligible() check if a buffer is eligible to cache | |
6704 | * l2arc_write_size() calculate how much to write | |
6705 | * l2arc_write_interval() calculate sleep delay between writes | |
6706 | * | |
6707 | * These three functions determine what to write, how much, and how quickly | |
6708 | * to send writes. | |
34dc7c2f BB |
6709 | */ |
6710 | ||
d164b209 | 6711 | static boolean_t |
2a432414 | 6712 | l2arc_write_eligible(uint64_t spa_guid, arc_buf_hdr_t *hdr) |
d164b209 BB |
6713 | { |
6714 | /* | |
6715 | * A buffer is *not* eligible for the L2ARC if it: | |
6716 | * 1. belongs to a different spa. | |
428870ff BB |
6717 | * 2. is already cached on the L2ARC. |
6718 | * 3. has an I/O in progress (it may be an incomplete read). | |
6719 | * 4. is flagged not eligible (zfs property). | |
d164b209 | 6720 | */ |
b9541d6b | 6721 | if (hdr->b_spa != spa_guid || HDR_HAS_L2HDR(hdr) || |
2a432414 | 6722 | HDR_IO_IN_PROGRESS(hdr) || !HDR_L2CACHE(hdr)) |
d164b209 BB |
6723 | return (B_FALSE); |
6724 | ||
6725 | return (B_TRUE); | |
6726 | } | |
6727 | ||
6728 | static uint64_t | |
3a17a7a9 | 6729 | l2arc_write_size(void) |
d164b209 BB |
6730 | { |
6731 | uint64_t size; | |
6732 | ||
3a17a7a9 SK |
6733 | /* |
6734 | * Make sure our globals have meaningful values in case the user | |
6735 | * altered them. | |
6736 | */ | |
6737 | size = l2arc_write_max; | |
6738 | if (size == 0) { | |
6739 | cmn_err(CE_NOTE, "Bad value for l2arc_write_max, value must " | |
6740 | "be greater than zero, resetting it to the default (%d)", | |
6741 | L2ARC_WRITE_SIZE); | |
6742 | size = l2arc_write_max = L2ARC_WRITE_SIZE; | |
6743 | } | |
d164b209 BB |
6744 | |
6745 | if (arc_warm == B_FALSE) | |
3a17a7a9 | 6746 | size += l2arc_write_boost; |
d164b209 BB |
6747 | |
6748 | return (size); | |
6749 | ||
6750 | } | |
6751 | ||
6752 | static clock_t | |
6753 | l2arc_write_interval(clock_t began, uint64_t wanted, uint64_t wrote) | |
6754 | { | |
428870ff | 6755 | clock_t interval, next, now; |
d164b209 BB |
6756 | |
6757 | /* | |
6758 | * If the ARC lists are busy, increase our write rate; if the | |
6759 | * lists are stale, idle back. This is achieved by checking | |
6760 | * how much we previously wrote - if it was more than half of | |
6761 | * what we wanted, schedule the next write much sooner. | |
6762 | */ | |
6763 | if (l2arc_feed_again && wrote > (wanted / 2)) | |
6764 | interval = (hz * l2arc_feed_min_ms) / 1000; | |
6765 | else | |
6766 | interval = hz * l2arc_feed_secs; | |
6767 | ||
428870ff BB |
6768 | now = ddi_get_lbolt(); |
6769 | next = MAX(now, MIN(now + interval, began + interval)); | |
d164b209 BB |
6770 | |
6771 | return (next); | |
6772 | } | |
6773 | ||
34dc7c2f BB |
6774 | /* |
6775 | * Cycle through L2ARC devices. This is how L2ARC load balances. | |
b128c09f | 6776 | * If a device is returned, this also returns holding the spa config lock. |
34dc7c2f BB |
6777 | */ |
6778 | static l2arc_dev_t * | |
6779 | l2arc_dev_get_next(void) | |
6780 | { | |
b128c09f | 6781 | l2arc_dev_t *first, *next = NULL; |
34dc7c2f | 6782 | |
b128c09f BB |
6783 | /* |
6784 | * Lock out the removal of spas (spa_namespace_lock), then removal | |
6785 | * of cache devices (l2arc_dev_mtx). Once a device has been selected, | |
6786 | * both locks will be dropped and a spa config lock held instead. | |
6787 | */ | |
6788 | mutex_enter(&spa_namespace_lock); | |
6789 | mutex_enter(&l2arc_dev_mtx); | |
6790 | ||
6791 | /* if there are no vdevs, there is nothing to do */ | |
6792 | if (l2arc_ndev == 0) | |
6793 | goto out; | |
6794 | ||
6795 | first = NULL; | |
6796 | next = l2arc_dev_last; | |
6797 | do { | |
6798 | /* loop around the list looking for a non-faulted vdev */ | |
6799 | if (next == NULL) { | |
34dc7c2f | 6800 | next = list_head(l2arc_dev_list); |
b128c09f BB |
6801 | } else { |
6802 | next = list_next(l2arc_dev_list, next); | |
6803 | if (next == NULL) | |
6804 | next = list_head(l2arc_dev_list); | |
6805 | } | |
6806 | ||
6807 | /* if we have come back to the start, bail out */ | |
6808 | if (first == NULL) | |
6809 | first = next; | |
6810 | else if (next == first) | |
6811 | break; | |
6812 | ||
6813 | } while (vdev_is_dead(next->l2ad_vdev)); | |
6814 | ||
6815 | /* if we were unable to find any usable vdevs, return NULL */ | |
6816 | if (vdev_is_dead(next->l2ad_vdev)) | |
6817 | next = NULL; | |
34dc7c2f BB |
6818 | |
6819 | l2arc_dev_last = next; | |
6820 | ||
b128c09f BB |
6821 | out: |
6822 | mutex_exit(&l2arc_dev_mtx); | |
6823 | ||
6824 | /* | |
6825 | * Grab the config lock to prevent the 'next' device from being | |
6826 | * removed while we are writing to it. | |
6827 | */ | |
6828 | if (next != NULL) | |
6829 | spa_config_enter(next->l2ad_spa, SCL_L2ARC, next, RW_READER); | |
6830 | mutex_exit(&spa_namespace_lock); | |
6831 | ||
34dc7c2f BB |
6832 | return (next); |
6833 | } | |
6834 | ||
b128c09f BB |
6835 | /* |
6836 | * Free buffers that were tagged for destruction. | |
6837 | */ | |
6838 | static void | |
0bc8fd78 | 6839 | l2arc_do_free_on_write(void) |
b128c09f BB |
6840 | { |
6841 | list_t *buflist; | |
6842 | l2arc_data_free_t *df, *df_prev; | |
6843 | ||
6844 | mutex_enter(&l2arc_free_on_write_mtx); | |
6845 | buflist = l2arc_free_on_write; | |
6846 | ||
6847 | for (df = list_tail(buflist); df; df = df_prev) { | |
6848 | df_prev = list_prev(buflist, df); | |
a6255b7f DQ |
6849 | ASSERT3P(df->l2df_abd, !=, NULL); |
6850 | abd_free(df->l2df_abd); | |
b128c09f BB |
6851 | list_remove(buflist, df); |
6852 | kmem_free(df, sizeof (l2arc_data_free_t)); | |
6853 | } | |
6854 | ||
6855 | mutex_exit(&l2arc_free_on_write_mtx); | |
6856 | } | |
6857 | ||
34dc7c2f BB |
6858 | /* |
6859 | * A write to a cache device has completed. Update all headers to allow | |
6860 | * reads from these buffers to begin. | |
6861 | */ | |
6862 | static void | |
6863 | l2arc_write_done(zio_t *zio) | |
6864 | { | |
6865 | l2arc_write_callback_t *cb; | |
6866 | l2arc_dev_t *dev; | |
6867 | list_t *buflist; | |
2a432414 | 6868 | arc_buf_hdr_t *head, *hdr, *hdr_prev; |
34dc7c2f | 6869 | kmutex_t *hash_lock; |
3bec585e | 6870 | int64_t bytes_dropped = 0; |
34dc7c2f BB |
6871 | |
6872 | cb = zio->io_private; | |
d3c2ae1c | 6873 | ASSERT3P(cb, !=, NULL); |
34dc7c2f | 6874 | dev = cb->l2wcb_dev; |
d3c2ae1c | 6875 | ASSERT3P(dev, !=, NULL); |
34dc7c2f | 6876 | head = cb->l2wcb_head; |
d3c2ae1c | 6877 | ASSERT3P(head, !=, NULL); |
b9541d6b | 6878 | buflist = &dev->l2ad_buflist; |
d3c2ae1c | 6879 | ASSERT3P(buflist, !=, NULL); |
34dc7c2f BB |
6880 | DTRACE_PROBE2(l2arc__iodone, zio_t *, zio, |
6881 | l2arc_write_callback_t *, cb); | |
6882 | ||
6883 | if (zio->io_error != 0) | |
6884 | ARCSTAT_BUMP(arcstat_l2_writes_error); | |
6885 | ||
34dc7c2f BB |
6886 | /* |
6887 | * All writes completed, or an error was hit. | |
6888 | */ | |
ca0bf58d PS |
6889 | top: |
6890 | mutex_enter(&dev->l2ad_mtx); | |
2a432414 GW |
6891 | for (hdr = list_prev(buflist, head); hdr; hdr = hdr_prev) { |
6892 | hdr_prev = list_prev(buflist, hdr); | |
34dc7c2f | 6893 | |
2a432414 | 6894 | hash_lock = HDR_LOCK(hdr); |
ca0bf58d PS |
6895 | |
6896 | /* | |
6897 | * We cannot use mutex_enter or else we can deadlock | |
6898 | * with l2arc_write_buffers (due to swapping the order | |
6899 | * the hash lock and l2ad_mtx are taken). | |
6900 | */ | |
34dc7c2f BB |
6901 | if (!mutex_tryenter(hash_lock)) { |
6902 | /* | |
ca0bf58d PS |
6903 | * Missed the hash lock. We must retry so we |
6904 | * don't leave the ARC_FLAG_L2_WRITING bit set. | |
34dc7c2f | 6905 | */ |
ca0bf58d PS |
6906 | ARCSTAT_BUMP(arcstat_l2_writes_lock_retry); |
6907 | ||
6908 | /* | |
6909 | * We don't want to rescan the headers we've | |
6910 | * already marked as having been written out, so | |
6911 | * we reinsert the head node so we can pick up | |
6912 | * where we left off. | |
6913 | */ | |
6914 | list_remove(buflist, head); | |
6915 | list_insert_after(buflist, hdr, head); | |
6916 | ||
6917 | mutex_exit(&dev->l2ad_mtx); | |
6918 | ||
6919 | /* | |
6920 | * We wait for the hash lock to become available | |
6921 | * to try and prevent busy waiting, and increase | |
6922 | * the chance we'll be able to acquire the lock | |
6923 | * the next time around. | |
6924 | */ | |
6925 | mutex_enter(hash_lock); | |
6926 | mutex_exit(hash_lock); | |
6927 | goto top; | |
34dc7c2f BB |
6928 | } |
6929 | ||
b9541d6b | 6930 | /* |
ca0bf58d PS |
6931 | * We could not have been moved into the arc_l2c_only |
6932 | * state while in-flight due to our ARC_FLAG_L2_WRITING | |
6933 | * bit being set. Let's just ensure that's being enforced. | |
6934 | */ | |
6935 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
6936 | ||
8a09d5fd BB |
6937 | /* |
6938 | * Skipped - drop L2ARC entry and mark the header as no | |
6939 | * longer L2 eligibile. | |
6940 | */ | |
d3c2ae1c | 6941 | if (zio->io_error != 0) { |
34dc7c2f | 6942 | /* |
b128c09f | 6943 | * Error - drop L2ARC entry. |
34dc7c2f | 6944 | */ |
2a432414 | 6945 | list_remove(buflist, hdr); |
d3c2ae1c | 6946 | arc_hdr_clear_flags(hdr, ARC_FLAG_HAS_L2HDR); |
b9541d6b | 6947 | |
d3c2ae1c GW |
6948 | ARCSTAT_INCR(arcstat_l2_asize, -arc_hdr_size(hdr)); |
6949 | ARCSTAT_INCR(arcstat_l2_size, -HDR_GET_LSIZE(hdr)); | |
d962d5da | 6950 | |
d3c2ae1c | 6951 | bytes_dropped += arc_hdr_size(hdr); |
d962d5da | 6952 | (void) refcount_remove_many(&dev->l2ad_alloc, |
d3c2ae1c | 6953 | arc_hdr_size(hdr), hdr); |
34dc7c2f BB |
6954 | } |
6955 | ||
6956 | /* | |
ca0bf58d PS |
6957 | * Allow ARC to begin reads and ghost list evictions to |
6958 | * this L2ARC entry. | |
34dc7c2f | 6959 | */ |
d3c2ae1c | 6960 | arc_hdr_clear_flags(hdr, ARC_FLAG_L2_WRITING); |
34dc7c2f BB |
6961 | |
6962 | mutex_exit(hash_lock); | |
6963 | } | |
6964 | ||
6965 | atomic_inc_64(&l2arc_writes_done); | |
6966 | list_remove(buflist, head); | |
b9541d6b CW |
6967 | ASSERT(!HDR_HAS_L1HDR(head)); |
6968 | kmem_cache_free(hdr_l2only_cache, head); | |
6969 | mutex_exit(&dev->l2ad_mtx); | |
34dc7c2f | 6970 | |
3bec585e SK |
6971 | vdev_space_update(dev->l2ad_vdev, -bytes_dropped, 0, 0); |
6972 | ||
b128c09f | 6973 | l2arc_do_free_on_write(); |
34dc7c2f BB |
6974 | |
6975 | kmem_free(cb, sizeof (l2arc_write_callback_t)); | |
6976 | } | |
6977 | ||
6978 | /* | |
6979 | * A read to a cache device completed. Validate buffer contents before | |
6980 | * handing over to the regular ARC routines. | |
6981 | */ | |
6982 | static void | |
6983 | l2arc_read_done(zio_t *zio) | |
6984 | { | |
6985 | l2arc_read_callback_t *cb; | |
6986 | arc_buf_hdr_t *hdr; | |
34dc7c2f | 6987 | kmutex_t *hash_lock; |
d3c2ae1c | 6988 | boolean_t valid_cksum; |
b128c09f | 6989 | |
d3c2ae1c | 6990 | ASSERT3P(zio->io_vd, !=, NULL); |
b128c09f BB |
6991 | ASSERT(zio->io_flags & ZIO_FLAG_DONT_PROPAGATE); |
6992 | ||
6993 | spa_config_exit(zio->io_spa, SCL_L2ARC, zio->io_vd); | |
34dc7c2f BB |
6994 | |
6995 | cb = zio->io_private; | |
d3c2ae1c GW |
6996 | ASSERT3P(cb, !=, NULL); |
6997 | hdr = cb->l2rcb_hdr; | |
6998 | ASSERT3P(hdr, !=, NULL); | |
34dc7c2f | 6999 | |
d3c2ae1c | 7000 | hash_lock = HDR_LOCK(hdr); |
34dc7c2f | 7001 | mutex_enter(hash_lock); |
428870ff | 7002 | ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); |
34dc7c2f | 7003 | |
a6255b7f | 7004 | ASSERT3P(zio->io_abd, !=, NULL); |
3a17a7a9 | 7005 | |
34dc7c2f BB |
7006 | /* |
7007 | * Check this survived the L2ARC journey. | |
7008 | */ | |
a6255b7f | 7009 | ASSERT3P(zio->io_abd, ==, hdr->b_l1hdr.b_pabd); |
d3c2ae1c GW |
7010 | zio->io_bp_copy = cb->l2rcb_bp; /* XXX fix in L2ARC 2.0 */ |
7011 | zio->io_bp = &zio->io_bp_copy; /* XXX fix in L2ARC 2.0 */ | |
7012 | ||
7013 | valid_cksum = arc_cksum_is_equal(hdr, zio); | |
7014 | if (valid_cksum && zio->io_error == 0 && !HDR_L2_EVICTED(hdr)) { | |
34dc7c2f | 7015 | mutex_exit(hash_lock); |
d3c2ae1c | 7016 | zio->io_private = hdr; |
34dc7c2f BB |
7017 | arc_read_done(zio); |
7018 | } else { | |
7019 | mutex_exit(hash_lock); | |
7020 | /* | |
7021 | * Buffer didn't survive caching. Increment stats and | |
7022 | * reissue to the original storage device. | |
7023 | */ | |
b128c09f | 7024 | if (zio->io_error != 0) { |
34dc7c2f | 7025 | ARCSTAT_BUMP(arcstat_l2_io_error); |
b128c09f | 7026 | } else { |
2e528b49 | 7027 | zio->io_error = SET_ERROR(EIO); |
b128c09f | 7028 | } |
d3c2ae1c | 7029 | if (!valid_cksum) |
34dc7c2f BB |
7030 | ARCSTAT_BUMP(arcstat_l2_cksum_bad); |
7031 | ||
34dc7c2f | 7032 | /* |
b128c09f BB |
7033 | * If there's no waiter, issue an async i/o to the primary |
7034 | * storage now. If there *is* a waiter, the caller must | |
7035 | * issue the i/o in a context where it's OK to block. | |
34dc7c2f | 7036 | */ |
d164b209 BB |
7037 | if (zio->io_waiter == NULL) { |
7038 | zio_t *pio = zio_unique_parent(zio); | |
7039 | ||
7040 | ASSERT(!pio || pio->io_child_type == ZIO_CHILD_LOGICAL); | |
7041 | ||
d3c2ae1c | 7042 | zio_nowait(zio_read(pio, zio->io_spa, zio->io_bp, |
a6255b7f | 7043 | hdr->b_l1hdr.b_pabd, zio->io_size, arc_read_done, |
d3c2ae1c GW |
7044 | hdr, zio->io_priority, cb->l2rcb_flags, |
7045 | &cb->l2rcb_zb)); | |
d164b209 | 7046 | } |
34dc7c2f BB |
7047 | } |
7048 | ||
7049 | kmem_free(cb, sizeof (l2arc_read_callback_t)); | |
7050 | } | |
7051 | ||
7052 | /* | |
7053 | * This is the list priority from which the L2ARC will search for pages to | |
7054 | * cache. This is used within loops (0..3) to cycle through lists in the | |
7055 | * desired order. This order can have a significant effect on cache | |
7056 | * performance. | |
7057 | * | |
7058 | * Currently the metadata lists are hit first, MFU then MRU, followed by | |
7059 | * the data lists. This function returns a locked list, and also returns | |
7060 | * the lock pointer. | |
7061 | */ | |
ca0bf58d PS |
7062 | static multilist_sublist_t * |
7063 | l2arc_sublist_lock(int list_num) | |
34dc7c2f | 7064 | { |
ca0bf58d PS |
7065 | multilist_t *ml = NULL; |
7066 | unsigned int idx; | |
34dc7c2f | 7067 | |
4aafab91 | 7068 | ASSERT(list_num >= 0 && list_num < L2ARC_FEED_TYPES); |
34dc7c2f BB |
7069 | |
7070 | switch (list_num) { | |
7071 | case 0: | |
ca0bf58d | 7072 | ml = &arc_mfu->arcs_list[ARC_BUFC_METADATA]; |
34dc7c2f BB |
7073 | break; |
7074 | case 1: | |
ca0bf58d | 7075 | ml = &arc_mru->arcs_list[ARC_BUFC_METADATA]; |
34dc7c2f BB |
7076 | break; |
7077 | case 2: | |
ca0bf58d | 7078 | ml = &arc_mfu->arcs_list[ARC_BUFC_DATA]; |
34dc7c2f BB |
7079 | break; |
7080 | case 3: | |
ca0bf58d | 7081 | ml = &arc_mru->arcs_list[ARC_BUFC_DATA]; |
34dc7c2f | 7082 | break; |
4aafab91 G |
7083 | default: |
7084 | return (NULL); | |
34dc7c2f BB |
7085 | } |
7086 | ||
ca0bf58d PS |
7087 | /* |
7088 | * Return a randomly-selected sublist. This is acceptable | |
7089 | * because the caller feeds only a little bit of data for each | |
7090 | * call (8MB). Subsequent calls will result in different | |
7091 | * sublists being selected. | |
7092 | */ | |
7093 | idx = multilist_get_random_index(ml); | |
7094 | return (multilist_sublist_lock(ml, idx)); | |
34dc7c2f BB |
7095 | } |
7096 | ||
7097 | /* | |
7098 | * Evict buffers from the device write hand to the distance specified in | |
7099 | * bytes. This distance may span populated buffers, it may span nothing. | |
7100 | * This is clearing a region on the L2ARC device ready for writing. | |
7101 | * If the 'all' boolean is set, every buffer is evicted. | |
7102 | */ | |
7103 | static void | |
7104 | l2arc_evict(l2arc_dev_t *dev, uint64_t distance, boolean_t all) | |
7105 | { | |
7106 | list_t *buflist; | |
2a432414 | 7107 | arc_buf_hdr_t *hdr, *hdr_prev; |
34dc7c2f BB |
7108 | kmutex_t *hash_lock; |
7109 | uint64_t taddr; | |
7110 | ||
b9541d6b | 7111 | buflist = &dev->l2ad_buflist; |
34dc7c2f BB |
7112 | |
7113 | if (!all && dev->l2ad_first) { | |
7114 | /* | |
7115 | * This is the first sweep through the device. There is | |
7116 | * nothing to evict. | |
7117 | */ | |
7118 | return; | |
7119 | } | |
7120 | ||
b128c09f | 7121 | if (dev->l2ad_hand >= (dev->l2ad_end - (2 * distance))) { |
34dc7c2f BB |
7122 | /* |
7123 | * When nearing the end of the device, evict to the end | |
7124 | * before the device write hand jumps to the start. | |
7125 | */ | |
7126 | taddr = dev->l2ad_end; | |
7127 | } else { | |
7128 | taddr = dev->l2ad_hand + distance; | |
7129 | } | |
7130 | DTRACE_PROBE4(l2arc__evict, l2arc_dev_t *, dev, list_t *, buflist, | |
7131 | uint64_t, taddr, boolean_t, all); | |
7132 | ||
7133 | top: | |
b9541d6b | 7134 | mutex_enter(&dev->l2ad_mtx); |
2a432414 GW |
7135 | for (hdr = list_tail(buflist); hdr; hdr = hdr_prev) { |
7136 | hdr_prev = list_prev(buflist, hdr); | |
34dc7c2f | 7137 | |
2a432414 | 7138 | hash_lock = HDR_LOCK(hdr); |
ca0bf58d PS |
7139 | |
7140 | /* | |
7141 | * We cannot use mutex_enter or else we can deadlock | |
7142 | * with l2arc_write_buffers (due to swapping the order | |
7143 | * the hash lock and l2ad_mtx are taken). | |
7144 | */ | |
34dc7c2f BB |
7145 | if (!mutex_tryenter(hash_lock)) { |
7146 | /* | |
7147 | * Missed the hash lock. Retry. | |
7148 | */ | |
7149 | ARCSTAT_BUMP(arcstat_l2_evict_lock_retry); | |
b9541d6b | 7150 | mutex_exit(&dev->l2ad_mtx); |
34dc7c2f BB |
7151 | mutex_enter(hash_lock); |
7152 | mutex_exit(hash_lock); | |
7153 | goto top; | |
7154 | } | |
7155 | ||
2a432414 | 7156 | if (HDR_L2_WRITE_HEAD(hdr)) { |
34dc7c2f BB |
7157 | /* |
7158 | * We hit a write head node. Leave it for | |
7159 | * l2arc_write_done(). | |
7160 | */ | |
2a432414 | 7161 | list_remove(buflist, hdr); |
34dc7c2f BB |
7162 | mutex_exit(hash_lock); |
7163 | continue; | |
7164 | } | |
7165 | ||
b9541d6b CW |
7166 | if (!all && HDR_HAS_L2HDR(hdr) && |
7167 | (hdr->b_l2hdr.b_daddr > taddr || | |
7168 | hdr->b_l2hdr.b_daddr < dev->l2ad_hand)) { | |
34dc7c2f BB |
7169 | /* |
7170 | * We've evicted to the target address, | |
7171 | * or the end of the device. | |
7172 | */ | |
7173 | mutex_exit(hash_lock); | |
7174 | break; | |
7175 | } | |
7176 | ||
b9541d6b CW |
7177 | ASSERT(HDR_HAS_L2HDR(hdr)); |
7178 | if (!HDR_HAS_L1HDR(hdr)) { | |
2a432414 | 7179 | ASSERT(!HDR_L2_READING(hdr)); |
34dc7c2f BB |
7180 | /* |
7181 | * This doesn't exist in the ARC. Destroy. | |
7182 | * arc_hdr_destroy() will call list_remove() | |
7183 | * and decrement arcstat_l2_size. | |
7184 | */ | |
2a432414 GW |
7185 | arc_change_state(arc_anon, hdr, hash_lock); |
7186 | arc_hdr_destroy(hdr); | |
34dc7c2f | 7187 | } else { |
b9541d6b CW |
7188 | ASSERT(hdr->b_l1hdr.b_state != arc_l2c_only); |
7189 | ARCSTAT_BUMP(arcstat_l2_evict_l1cached); | |
b128c09f BB |
7190 | /* |
7191 | * Invalidate issued or about to be issued | |
7192 | * reads, since we may be about to write | |
7193 | * over this location. | |
7194 | */ | |
2a432414 | 7195 | if (HDR_L2_READING(hdr)) { |
b128c09f | 7196 | ARCSTAT_BUMP(arcstat_l2_evict_reading); |
d3c2ae1c | 7197 | arc_hdr_set_flags(hdr, ARC_FLAG_L2_EVICTED); |
b128c09f BB |
7198 | } |
7199 | ||
ca0bf58d PS |
7200 | /* Ensure this header has finished being written */ |
7201 | ASSERT(!HDR_L2_WRITING(hdr)); | |
d962d5da PS |
7202 | |
7203 | arc_hdr_l2hdr_destroy(hdr); | |
34dc7c2f BB |
7204 | } |
7205 | mutex_exit(hash_lock); | |
7206 | } | |
b9541d6b | 7207 | mutex_exit(&dev->l2ad_mtx); |
34dc7c2f BB |
7208 | } |
7209 | ||
7210 | /* | |
7211 | * Find and write ARC buffers to the L2ARC device. | |
7212 | * | |
2a432414 | 7213 | * An ARC_FLAG_L2_WRITING flag is set so that the L2ARC buffers are not valid |
34dc7c2f | 7214 | * for reading until they have completed writing. |
3a17a7a9 SK |
7215 | * The headroom_boost is an in-out parameter used to maintain headroom boost |
7216 | * state between calls to this function. | |
7217 | * | |
7218 | * Returns the number of bytes actually written (which may be smaller than | |
7219 | * the delta by which the device hand has changed due to alignment). | |
34dc7c2f | 7220 | */ |
d164b209 | 7221 | static uint64_t |
d3c2ae1c | 7222 | l2arc_write_buffers(spa_t *spa, l2arc_dev_t *dev, uint64_t target_sz) |
34dc7c2f | 7223 | { |
2a432414 | 7224 | arc_buf_hdr_t *hdr, *hdr_prev, *head; |
d3c2ae1c | 7225 | uint64_t write_asize, write_psize, write_sz, headroom; |
3a17a7a9 | 7226 | boolean_t full; |
34dc7c2f BB |
7227 | l2arc_write_callback_t *cb; |
7228 | zio_t *pio, *wzio; | |
3541dc6d | 7229 | uint64_t guid = spa_load_guid(spa); |
d6320ddb | 7230 | int try; |
34dc7c2f | 7231 | |
d3c2ae1c | 7232 | ASSERT3P(dev->l2ad_vdev, !=, NULL); |
3a17a7a9 | 7233 | |
34dc7c2f | 7234 | pio = NULL; |
d3c2ae1c | 7235 | write_sz = write_asize = write_psize = 0; |
34dc7c2f | 7236 | full = B_FALSE; |
b9541d6b | 7237 | head = kmem_cache_alloc(hdr_l2only_cache, KM_PUSHPAGE); |
d3c2ae1c | 7238 | arc_hdr_set_flags(head, ARC_FLAG_L2_WRITE_HEAD | ARC_FLAG_HAS_L2HDR); |
3a17a7a9 | 7239 | |
34dc7c2f BB |
7240 | /* |
7241 | * Copy buffers for L2ARC writing. | |
7242 | */ | |
4aafab91 | 7243 | for (try = 0; try < L2ARC_FEED_TYPES; try++) { |
ca0bf58d | 7244 | multilist_sublist_t *mls = l2arc_sublist_lock(try); |
3a17a7a9 SK |
7245 | uint64_t passed_sz = 0; |
7246 | ||
4aafab91 G |
7247 | VERIFY3P(mls, !=, NULL); |
7248 | ||
b128c09f BB |
7249 | /* |
7250 | * L2ARC fast warmup. | |
7251 | * | |
7252 | * Until the ARC is warm and starts to evict, read from the | |
7253 | * head of the ARC lists rather than the tail. | |
7254 | */ | |
b128c09f | 7255 | if (arc_warm == B_FALSE) |
ca0bf58d | 7256 | hdr = multilist_sublist_head(mls); |
b128c09f | 7257 | else |
ca0bf58d | 7258 | hdr = multilist_sublist_tail(mls); |
b128c09f | 7259 | |
3a17a7a9 | 7260 | headroom = target_sz * l2arc_headroom; |
d3c2ae1c | 7261 | if (zfs_compressed_arc_enabled) |
3a17a7a9 SK |
7262 | headroom = (headroom * l2arc_headroom_boost) / 100; |
7263 | ||
2a432414 | 7264 | for (; hdr; hdr = hdr_prev) { |
3a17a7a9 | 7265 | kmutex_t *hash_lock; |
d3c2ae1c | 7266 | uint64_t asize, size; |
a6255b7f | 7267 | abd_t *to_write; |
3a17a7a9 | 7268 | |
b128c09f | 7269 | if (arc_warm == B_FALSE) |
ca0bf58d | 7270 | hdr_prev = multilist_sublist_next(mls, hdr); |
b128c09f | 7271 | else |
ca0bf58d | 7272 | hdr_prev = multilist_sublist_prev(mls, hdr); |
34dc7c2f | 7273 | |
2a432414 | 7274 | hash_lock = HDR_LOCK(hdr); |
3a17a7a9 | 7275 | if (!mutex_tryenter(hash_lock)) { |
34dc7c2f BB |
7276 | /* |
7277 | * Skip this buffer rather than waiting. | |
7278 | */ | |
7279 | continue; | |
7280 | } | |
7281 | ||
d3c2ae1c | 7282 | passed_sz += HDR_GET_LSIZE(hdr); |
34dc7c2f BB |
7283 | if (passed_sz > headroom) { |
7284 | /* | |
7285 | * Searched too far. | |
7286 | */ | |
7287 | mutex_exit(hash_lock); | |
7288 | break; | |
7289 | } | |
7290 | ||
2a432414 | 7291 | if (!l2arc_write_eligible(guid, hdr)) { |
34dc7c2f BB |
7292 | mutex_exit(hash_lock); |
7293 | continue; | |
7294 | } | |
7295 | ||
d3c2ae1c | 7296 | if ((write_asize + HDR_GET_LSIZE(hdr)) > target_sz) { |
34dc7c2f BB |
7297 | full = B_TRUE; |
7298 | mutex_exit(hash_lock); | |
7299 | break; | |
7300 | } | |
7301 | ||
34dc7c2f BB |
7302 | if (pio == NULL) { |
7303 | /* | |
7304 | * Insert a dummy header on the buflist so | |
7305 | * l2arc_write_done() can find where the | |
7306 | * write buffers begin without searching. | |
7307 | */ | |
ca0bf58d | 7308 | mutex_enter(&dev->l2ad_mtx); |
b9541d6b | 7309 | list_insert_head(&dev->l2ad_buflist, head); |
ca0bf58d | 7310 | mutex_exit(&dev->l2ad_mtx); |
34dc7c2f | 7311 | |
96c080cb BB |
7312 | cb = kmem_alloc( |
7313 | sizeof (l2arc_write_callback_t), KM_SLEEP); | |
34dc7c2f BB |
7314 | cb->l2wcb_dev = dev; |
7315 | cb->l2wcb_head = head; | |
7316 | pio = zio_root(spa, l2arc_write_done, cb, | |
7317 | ZIO_FLAG_CANFAIL); | |
7318 | } | |
7319 | ||
b9541d6b | 7320 | hdr->b_l2hdr.b_dev = dev; |
b9541d6b | 7321 | hdr->b_l2hdr.b_hits = 0; |
3a17a7a9 | 7322 | |
d3c2ae1c GW |
7323 | hdr->b_l2hdr.b_daddr = dev->l2ad_hand; |
7324 | arc_hdr_set_flags(hdr, | |
7325 | ARC_FLAG_L2_WRITING | ARC_FLAG_HAS_L2HDR); | |
3a17a7a9 | 7326 | |
ca0bf58d | 7327 | mutex_enter(&dev->l2ad_mtx); |
b9541d6b | 7328 | list_insert_head(&dev->l2ad_buflist, hdr); |
ca0bf58d | 7329 | mutex_exit(&dev->l2ad_mtx); |
34dc7c2f BB |
7330 | |
7331 | /* | |
d3c2ae1c GW |
7332 | * We rely on the L1 portion of the header below, so |
7333 | * it's invalid for this header to have been evicted out | |
7334 | * of the ghost cache, prior to being written out. The | |
7335 | * ARC_FLAG_L2_WRITING bit ensures this won't happen. | |
34dc7c2f | 7336 | */ |
d3c2ae1c | 7337 | ASSERT(HDR_HAS_L1HDR(hdr)); |
3a17a7a9 | 7338 | |
d3c2ae1c | 7339 | ASSERT3U(HDR_GET_PSIZE(hdr), >, 0); |
a6255b7f | 7340 | ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); |
d3c2ae1c GW |
7341 | ASSERT3U(arc_hdr_size(hdr), >, 0); |
7342 | size = arc_hdr_size(hdr); | |
d962d5da | 7343 | |
d3c2ae1c | 7344 | (void) refcount_add_many(&dev->l2ad_alloc, size, hdr); |
3a17a7a9 | 7345 | |
8a09d5fd | 7346 | /* |
d3c2ae1c GW |
7347 | * Normally the L2ARC can use the hdr's data, but if |
7348 | * we're sharing data between the hdr and one of its | |
7349 | * bufs, L2ARC needs its own copy of the data so that | |
7350 | * the ZIO below can't race with the buf consumer. To | |
7351 | * ensure that this copy will be available for the | |
7352 | * lifetime of the ZIO and be cleaned up afterwards, we | |
7353 | * add it to the l2arc_free_on_write queue. | |
8a09d5fd | 7354 | */ |
d3c2ae1c | 7355 | if (!HDR_SHARED_DATA(hdr)) { |
a6255b7f | 7356 | to_write = hdr->b_l1hdr.b_pabd; |
d3c2ae1c | 7357 | } else { |
a6255b7f DQ |
7358 | to_write = abd_alloc_for_io(size, |
7359 | HDR_ISTYPE_METADATA(hdr)); | |
7360 | abd_copy(to_write, hdr->b_l1hdr.b_pabd, size); | |
7361 | l2arc_free_abd_on_write(to_write, size, | |
7362 | arc_buf_type(hdr)); | |
d3c2ae1c | 7363 | } |
34dc7c2f | 7364 | wzio = zio_write_phys(pio, dev->l2ad_vdev, |
d3c2ae1c GW |
7365 | hdr->b_l2hdr.b_daddr, size, to_write, |
7366 | ZIO_CHECKSUM_OFF, NULL, hdr, | |
7367 | ZIO_PRIORITY_ASYNC_WRITE, | |
34dc7c2f BB |
7368 | ZIO_FLAG_CANFAIL, B_FALSE); |
7369 | ||
d3c2ae1c | 7370 | write_sz += HDR_GET_LSIZE(hdr); |
34dc7c2f BB |
7371 | DTRACE_PROBE2(l2arc__write, vdev_t *, dev->l2ad_vdev, |
7372 | zio_t *, wzio); | |
d962d5da | 7373 | |
d3c2ae1c | 7374 | write_asize += size; |
b128c09f BB |
7375 | /* |
7376 | * Keep the clock hand suitably device-aligned. | |
7377 | */ | |
d3c2ae1c GW |
7378 | asize = vdev_psize_to_asize(dev->l2ad_vdev, size); |
7379 | write_psize += asize; | |
7380 | dev->l2ad_hand += asize; | |
7381 | ||
7382 | mutex_exit(hash_lock); | |
7383 | ||
7384 | (void) zio_nowait(wzio); | |
34dc7c2f | 7385 | } |
d3c2ae1c GW |
7386 | |
7387 | multilist_sublist_unlock(mls); | |
7388 | ||
7389 | if (full == B_TRUE) | |
7390 | break; | |
34dc7c2f | 7391 | } |
34dc7c2f | 7392 | |
d3c2ae1c GW |
7393 | /* No buffers selected for writing? */ |
7394 | if (pio == NULL) { | |
7395 | ASSERT0(write_sz); | |
7396 | ASSERT(!HDR_HAS_L1HDR(head)); | |
7397 | kmem_cache_free(hdr_l2only_cache, head); | |
7398 | return (0); | |
7399 | } | |
34dc7c2f | 7400 | |
3a17a7a9 | 7401 | ASSERT3U(write_asize, <=, target_sz); |
34dc7c2f | 7402 | ARCSTAT_BUMP(arcstat_l2_writes_sent); |
3a17a7a9 | 7403 | ARCSTAT_INCR(arcstat_l2_write_bytes, write_asize); |
34dc7c2f | 7404 | ARCSTAT_INCR(arcstat_l2_size, write_sz); |
d3c2ae1c GW |
7405 | ARCSTAT_INCR(arcstat_l2_asize, write_asize); |
7406 | vdev_space_update(dev->l2ad_vdev, write_asize, 0, 0); | |
34dc7c2f BB |
7407 | |
7408 | /* | |
7409 | * Bump device hand to the device start if it is approaching the end. | |
7410 | * l2arc_evict() will already have evicted ahead for this case. | |
7411 | */ | |
b128c09f | 7412 | if (dev->l2ad_hand >= (dev->l2ad_end - target_sz)) { |
34dc7c2f | 7413 | dev->l2ad_hand = dev->l2ad_start; |
34dc7c2f BB |
7414 | dev->l2ad_first = B_FALSE; |
7415 | } | |
7416 | ||
d164b209 | 7417 | dev->l2ad_writing = B_TRUE; |
34dc7c2f | 7418 | (void) zio_wait(pio); |
d164b209 BB |
7419 | dev->l2ad_writing = B_FALSE; |
7420 | ||
3a17a7a9 SK |
7421 | return (write_asize); |
7422 | } | |
7423 | ||
34dc7c2f BB |
7424 | /* |
7425 | * This thread feeds the L2ARC at regular intervals. This is the beating | |
7426 | * heart of the L2ARC. | |
7427 | */ | |
7428 | static void | |
7429 | l2arc_feed_thread(void) | |
7430 | { | |
7431 | callb_cpr_t cpr; | |
7432 | l2arc_dev_t *dev; | |
7433 | spa_t *spa; | |
d164b209 | 7434 | uint64_t size, wrote; |
428870ff | 7435 | clock_t begin, next = ddi_get_lbolt(); |
40d06e3c | 7436 | fstrans_cookie_t cookie; |
34dc7c2f BB |
7437 | |
7438 | CALLB_CPR_INIT(&cpr, &l2arc_feed_thr_lock, callb_generic_cpr, FTAG); | |
7439 | ||
7440 | mutex_enter(&l2arc_feed_thr_lock); | |
7441 | ||
40d06e3c | 7442 | cookie = spl_fstrans_mark(); |
34dc7c2f | 7443 | while (l2arc_thread_exit == 0) { |
34dc7c2f | 7444 | CALLB_CPR_SAFE_BEGIN(&cpr); |
b64ccd6c | 7445 | (void) cv_timedwait_sig(&l2arc_feed_thr_cv, |
5b63b3eb | 7446 | &l2arc_feed_thr_lock, next); |
34dc7c2f | 7447 | CALLB_CPR_SAFE_END(&cpr, &l2arc_feed_thr_lock); |
428870ff | 7448 | next = ddi_get_lbolt() + hz; |
34dc7c2f BB |
7449 | |
7450 | /* | |
b128c09f | 7451 | * Quick check for L2ARC devices. |
34dc7c2f BB |
7452 | */ |
7453 | mutex_enter(&l2arc_dev_mtx); | |
7454 | if (l2arc_ndev == 0) { | |
7455 | mutex_exit(&l2arc_dev_mtx); | |
7456 | continue; | |
7457 | } | |
b128c09f | 7458 | mutex_exit(&l2arc_dev_mtx); |
428870ff | 7459 | begin = ddi_get_lbolt(); |
34dc7c2f BB |
7460 | |
7461 | /* | |
b128c09f BB |
7462 | * This selects the next l2arc device to write to, and in |
7463 | * doing so the next spa to feed from: dev->l2ad_spa. This | |
7464 | * will return NULL if there are now no l2arc devices or if | |
7465 | * they are all faulted. | |
7466 | * | |
7467 | * If a device is returned, its spa's config lock is also | |
7468 | * held to prevent device removal. l2arc_dev_get_next() | |
7469 | * will grab and release l2arc_dev_mtx. | |
34dc7c2f | 7470 | */ |
b128c09f | 7471 | if ((dev = l2arc_dev_get_next()) == NULL) |
34dc7c2f | 7472 | continue; |
b128c09f BB |
7473 | |
7474 | spa = dev->l2ad_spa; | |
d3c2ae1c | 7475 | ASSERT3P(spa, !=, NULL); |
34dc7c2f | 7476 | |
572e2857 BB |
7477 | /* |
7478 | * If the pool is read-only then force the feed thread to | |
7479 | * sleep a little longer. | |
7480 | */ | |
7481 | if (!spa_writeable(spa)) { | |
7482 | next = ddi_get_lbolt() + 5 * l2arc_feed_secs * hz; | |
7483 | spa_config_exit(spa, SCL_L2ARC, dev); | |
7484 | continue; | |
7485 | } | |
7486 | ||
34dc7c2f | 7487 | /* |
b128c09f | 7488 | * Avoid contributing to memory pressure. |
34dc7c2f | 7489 | */ |
ca67b33a | 7490 | if (arc_reclaim_needed()) { |
b128c09f BB |
7491 | ARCSTAT_BUMP(arcstat_l2_abort_lowmem); |
7492 | spa_config_exit(spa, SCL_L2ARC, dev); | |
34dc7c2f BB |
7493 | continue; |
7494 | } | |
b128c09f | 7495 | |
34dc7c2f BB |
7496 | ARCSTAT_BUMP(arcstat_l2_feeds); |
7497 | ||
3a17a7a9 | 7498 | size = l2arc_write_size(); |
b128c09f | 7499 | |
34dc7c2f BB |
7500 | /* |
7501 | * Evict L2ARC buffers that will be overwritten. | |
7502 | */ | |
b128c09f | 7503 | l2arc_evict(dev, size, B_FALSE); |
34dc7c2f BB |
7504 | |
7505 | /* | |
7506 | * Write ARC buffers. | |
7507 | */ | |
d3c2ae1c | 7508 | wrote = l2arc_write_buffers(spa, dev, size); |
d164b209 BB |
7509 | |
7510 | /* | |
7511 | * Calculate interval between writes. | |
7512 | */ | |
7513 | next = l2arc_write_interval(begin, size, wrote); | |
b128c09f | 7514 | spa_config_exit(spa, SCL_L2ARC, dev); |
34dc7c2f | 7515 | } |
40d06e3c | 7516 | spl_fstrans_unmark(cookie); |
34dc7c2f BB |
7517 | |
7518 | l2arc_thread_exit = 0; | |
7519 | cv_broadcast(&l2arc_feed_thr_cv); | |
7520 | CALLB_CPR_EXIT(&cpr); /* drops l2arc_feed_thr_lock */ | |
7521 | thread_exit(); | |
7522 | } | |
7523 | ||
b128c09f BB |
7524 | boolean_t |
7525 | l2arc_vdev_present(vdev_t *vd) | |
7526 | { | |
7527 | l2arc_dev_t *dev; | |
7528 | ||
7529 | mutex_enter(&l2arc_dev_mtx); | |
7530 | for (dev = list_head(l2arc_dev_list); dev != NULL; | |
7531 | dev = list_next(l2arc_dev_list, dev)) { | |
7532 | if (dev->l2ad_vdev == vd) | |
7533 | break; | |
7534 | } | |
7535 | mutex_exit(&l2arc_dev_mtx); | |
7536 | ||
7537 | return (dev != NULL); | |
7538 | } | |
7539 | ||
34dc7c2f BB |
7540 | /* |
7541 | * Add a vdev for use by the L2ARC. By this point the spa has already | |
7542 | * validated the vdev and opened it. | |
7543 | */ | |
7544 | void | |
9babb374 | 7545 | l2arc_add_vdev(spa_t *spa, vdev_t *vd) |
34dc7c2f BB |
7546 | { |
7547 | l2arc_dev_t *adddev; | |
7548 | ||
b128c09f BB |
7549 | ASSERT(!l2arc_vdev_present(vd)); |
7550 | ||
34dc7c2f BB |
7551 | /* |
7552 | * Create a new l2arc device entry. | |
7553 | */ | |
7554 | adddev = kmem_zalloc(sizeof (l2arc_dev_t), KM_SLEEP); | |
7555 | adddev->l2ad_spa = spa; | |
7556 | adddev->l2ad_vdev = vd; | |
9babb374 BB |
7557 | adddev->l2ad_start = VDEV_LABEL_START_SIZE; |
7558 | adddev->l2ad_end = VDEV_LABEL_START_SIZE + vdev_get_min_asize(vd); | |
34dc7c2f | 7559 | adddev->l2ad_hand = adddev->l2ad_start; |
34dc7c2f | 7560 | adddev->l2ad_first = B_TRUE; |
d164b209 | 7561 | adddev->l2ad_writing = B_FALSE; |
98f72a53 | 7562 | list_link_init(&adddev->l2ad_node); |
34dc7c2f | 7563 | |
b9541d6b | 7564 | mutex_init(&adddev->l2ad_mtx, NULL, MUTEX_DEFAULT, NULL); |
34dc7c2f BB |
7565 | /* |
7566 | * This is a list of all ARC buffers that are still valid on the | |
7567 | * device. | |
7568 | */ | |
b9541d6b CW |
7569 | list_create(&adddev->l2ad_buflist, sizeof (arc_buf_hdr_t), |
7570 | offsetof(arc_buf_hdr_t, b_l2hdr.b_l2node)); | |
34dc7c2f | 7571 | |
428870ff | 7572 | vdev_space_update(vd, 0, 0, adddev->l2ad_end - adddev->l2ad_hand); |
d962d5da | 7573 | refcount_create(&adddev->l2ad_alloc); |
34dc7c2f BB |
7574 | |
7575 | /* | |
7576 | * Add device to global list | |
7577 | */ | |
7578 | mutex_enter(&l2arc_dev_mtx); | |
7579 | list_insert_head(l2arc_dev_list, adddev); | |
7580 | atomic_inc_64(&l2arc_ndev); | |
7581 | mutex_exit(&l2arc_dev_mtx); | |
7582 | } | |
7583 | ||
7584 | /* | |
7585 | * Remove a vdev from the L2ARC. | |
7586 | */ | |
7587 | void | |
7588 | l2arc_remove_vdev(vdev_t *vd) | |
7589 | { | |
7590 | l2arc_dev_t *dev, *nextdev, *remdev = NULL; | |
7591 | ||
34dc7c2f BB |
7592 | /* |
7593 | * Find the device by vdev | |
7594 | */ | |
7595 | mutex_enter(&l2arc_dev_mtx); | |
7596 | for (dev = list_head(l2arc_dev_list); dev; dev = nextdev) { | |
7597 | nextdev = list_next(l2arc_dev_list, dev); | |
7598 | if (vd == dev->l2ad_vdev) { | |
7599 | remdev = dev; | |
7600 | break; | |
7601 | } | |
7602 | } | |
d3c2ae1c | 7603 | ASSERT3P(remdev, !=, NULL); |
34dc7c2f BB |
7604 | |
7605 | /* | |
7606 | * Remove device from global list | |
7607 | */ | |
7608 | list_remove(l2arc_dev_list, remdev); | |
7609 | l2arc_dev_last = NULL; /* may have been invalidated */ | |
b128c09f BB |
7610 | atomic_dec_64(&l2arc_ndev); |
7611 | mutex_exit(&l2arc_dev_mtx); | |
34dc7c2f BB |
7612 | |
7613 | /* | |
7614 | * Clear all buflists and ARC references. L2ARC device flush. | |
7615 | */ | |
7616 | l2arc_evict(remdev, 0, B_TRUE); | |
b9541d6b CW |
7617 | list_destroy(&remdev->l2ad_buflist); |
7618 | mutex_destroy(&remdev->l2ad_mtx); | |
d962d5da | 7619 | refcount_destroy(&remdev->l2ad_alloc); |
34dc7c2f | 7620 | kmem_free(remdev, sizeof (l2arc_dev_t)); |
34dc7c2f BB |
7621 | } |
7622 | ||
7623 | void | |
b128c09f | 7624 | l2arc_init(void) |
34dc7c2f BB |
7625 | { |
7626 | l2arc_thread_exit = 0; | |
7627 | l2arc_ndev = 0; | |
7628 | l2arc_writes_sent = 0; | |
7629 | l2arc_writes_done = 0; | |
7630 | ||
7631 | mutex_init(&l2arc_feed_thr_lock, NULL, MUTEX_DEFAULT, NULL); | |
7632 | cv_init(&l2arc_feed_thr_cv, NULL, CV_DEFAULT, NULL); | |
7633 | mutex_init(&l2arc_dev_mtx, NULL, MUTEX_DEFAULT, NULL); | |
34dc7c2f BB |
7634 | mutex_init(&l2arc_free_on_write_mtx, NULL, MUTEX_DEFAULT, NULL); |
7635 | ||
7636 | l2arc_dev_list = &L2ARC_dev_list; | |
7637 | l2arc_free_on_write = &L2ARC_free_on_write; | |
7638 | list_create(l2arc_dev_list, sizeof (l2arc_dev_t), | |
7639 | offsetof(l2arc_dev_t, l2ad_node)); | |
7640 | list_create(l2arc_free_on_write, sizeof (l2arc_data_free_t), | |
7641 | offsetof(l2arc_data_free_t, l2df_list_node)); | |
34dc7c2f BB |
7642 | } |
7643 | ||
7644 | void | |
b128c09f | 7645 | l2arc_fini(void) |
34dc7c2f | 7646 | { |
b128c09f BB |
7647 | /* |
7648 | * This is called from dmu_fini(), which is called from spa_fini(); | |
7649 | * Because of this, we can assume that all l2arc devices have | |
7650 | * already been removed when the pools themselves were removed. | |
7651 | */ | |
7652 | ||
7653 | l2arc_do_free_on_write(); | |
34dc7c2f BB |
7654 | |
7655 | mutex_destroy(&l2arc_feed_thr_lock); | |
7656 | cv_destroy(&l2arc_feed_thr_cv); | |
7657 | mutex_destroy(&l2arc_dev_mtx); | |
34dc7c2f BB |
7658 | mutex_destroy(&l2arc_free_on_write_mtx); |
7659 | ||
7660 | list_destroy(l2arc_dev_list); | |
7661 | list_destroy(l2arc_free_on_write); | |
7662 | } | |
b128c09f BB |
7663 | |
7664 | void | |
7665 | l2arc_start(void) | |
7666 | { | |
fb5f0bc8 | 7667 | if (!(spa_mode_global & FWRITE)) |
b128c09f BB |
7668 | return; |
7669 | ||
7670 | (void) thread_create(NULL, 0, l2arc_feed_thread, NULL, 0, &p0, | |
1229323d | 7671 | TS_RUN, defclsyspri); |
b128c09f BB |
7672 | } |
7673 | ||
7674 | void | |
7675 | l2arc_stop(void) | |
7676 | { | |
fb5f0bc8 | 7677 | if (!(spa_mode_global & FWRITE)) |
b128c09f BB |
7678 | return; |
7679 | ||
7680 | mutex_enter(&l2arc_feed_thr_lock); | |
7681 | cv_signal(&l2arc_feed_thr_cv); /* kick thread out of startup */ | |
7682 | l2arc_thread_exit = 1; | |
7683 | while (l2arc_thread_exit != 0) | |
7684 | cv_wait(&l2arc_feed_thr_cv, &l2arc_feed_thr_lock); | |
7685 | mutex_exit(&l2arc_feed_thr_lock); | |
7686 | } | |
c28b2279 BB |
7687 | |
7688 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
0f699108 AZ |
7689 | EXPORT_SYMBOL(arc_buf_size); |
7690 | EXPORT_SYMBOL(arc_write); | |
c28b2279 | 7691 | EXPORT_SYMBOL(arc_read); |
e0b0ca98 | 7692 | EXPORT_SYMBOL(arc_buf_info); |
c28b2279 | 7693 | EXPORT_SYMBOL(arc_getbuf_func); |
ab26409d BB |
7694 | EXPORT_SYMBOL(arc_add_prune_callback); |
7695 | EXPORT_SYMBOL(arc_remove_prune_callback); | |
c28b2279 | 7696 | |
02730c33 | 7697 | /* BEGIN CSTYLED */ |
bce45ec9 | 7698 | module_param(zfs_arc_min, ulong, 0644); |
c409e464 | 7699 | MODULE_PARM_DESC(zfs_arc_min, "Min arc size"); |
c28b2279 | 7700 | |
bce45ec9 | 7701 | module_param(zfs_arc_max, ulong, 0644); |
c409e464 | 7702 | MODULE_PARM_DESC(zfs_arc_max, "Max arc size"); |
c28b2279 | 7703 | |
bce45ec9 | 7704 | module_param(zfs_arc_meta_limit, ulong, 0644); |
c28b2279 | 7705 | MODULE_PARM_DESC(zfs_arc_meta_limit, "Meta limit for arc size"); |
6a8f9b6b | 7706 | |
9907cc1c G |
7707 | module_param(zfs_arc_meta_limit_percent, ulong, 0644); |
7708 | MODULE_PARM_DESC(zfs_arc_meta_limit_percent, | |
7709 | "Percent of arc size for arc meta limit"); | |
7710 | ||
ca0bf58d PS |
7711 | module_param(zfs_arc_meta_min, ulong, 0644); |
7712 | MODULE_PARM_DESC(zfs_arc_meta_min, "Min arc metadata"); | |
7713 | ||
bce45ec9 | 7714 | module_param(zfs_arc_meta_prune, int, 0644); |
2cbb06b5 | 7715 | MODULE_PARM_DESC(zfs_arc_meta_prune, "Meta objects to scan for prune"); |
c409e464 | 7716 | |
ca67b33a | 7717 | module_param(zfs_arc_meta_adjust_restarts, int, 0644); |
bc888666 BB |
7718 | MODULE_PARM_DESC(zfs_arc_meta_adjust_restarts, |
7719 | "Limit number of restarts in arc_adjust_meta"); | |
7720 | ||
f6046738 BB |
7721 | module_param(zfs_arc_meta_strategy, int, 0644); |
7722 | MODULE_PARM_DESC(zfs_arc_meta_strategy, "Meta reclaim strategy"); | |
7723 | ||
bce45ec9 | 7724 | module_param(zfs_arc_grow_retry, int, 0644); |
c409e464 BB |
7725 | MODULE_PARM_DESC(zfs_arc_grow_retry, "Seconds before growing arc size"); |
7726 | ||
89c8cac4 PS |
7727 | module_param(zfs_arc_p_aggressive_disable, int, 0644); |
7728 | MODULE_PARM_DESC(zfs_arc_p_aggressive_disable, "disable aggressive arc_p grow"); | |
7729 | ||
62422785 PS |
7730 | module_param(zfs_arc_p_dampener_disable, int, 0644); |
7731 | MODULE_PARM_DESC(zfs_arc_p_dampener_disable, "disable arc_p adapt dampener"); | |
7732 | ||
bce45ec9 | 7733 | module_param(zfs_arc_shrink_shift, int, 0644); |
c409e464 BB |
7734 | MODULE_PARM_DESC(zfs_arc_shrink_shift, "log2(fraction of arc to reclaim)"); |
7735 | ||
728d6ae9 BB |
7736 | module_param(zfs_arc_p_min_shift, int, 0644); |
7737 | MODULE_PARM_DESC(zfs_arc_p_min_shift, "arc_c shift to calc min/max arc_p"); | |
7738 | ||
49ddb315 MA |
7739 | module_param(zfs_arc_average_blocksize, int, 0444); |
7740 | MODULE_PARM_DESC(zfs_arc_average_blocksize, "Target average block size"); | |
7741 | ||
d3c2ae1c | 7742 | module_param(zfs_compressed_arc_enabled, int, 0644); |
544596c5 | 7743 | MODULE_PARM_DESC(zfs_compressed_arc_enabled, "Disable compressed arc buffers"); |
d3c2ae1c | 7744 | |
bce45ec9 BB |
7745 | module_param(zfs_arc_min_prefetch_lifespan, int, 0644); |
7746 | MODULE_PARM_DESC(zfs_arc_min_prefetch_lifespan, "Min life of prefetch block"); | |
7747 | ||
ca0bf58d PS |
7748 | module_param(zfs_arc_num_sublists_per_state, int, 0644); |
7749 | MODULE_PARM_DESC(zfs_arc_num_sublists_per_state, | |
7750 | "Number of sublists used in each of the ARC state lists"); | |
7751 | ||
bce45ec9 | 7752 | module_param(l2arc_write_max, ulong, 0644); |
abd8610c BB |
7753 | MODULE_PARM_DESC(l2arc_write_max, "Max write bytes per interval"); |
7754 | ||
bce45ec9 | 7755 | module_param(l2arc_write_boost, ulong, 0644); |
abd8610c BB |
7756 | MODULE_PARM_DESC(l2arc_write_boost, "Extra write bytes during device warmup"); |
7757 | ||
bce45ec9 | 7758 | module_param(l2arc_headroom, ulong, 0644); |
abd8610c BB |
7759 | MODULE_PARM_DESC(l2arc_headroom, "Number of max device writes to precache"); |
7760 | ||
3a17a7a9 SK |
7761 | module_param(l2arc_headroom_boost, ulong, 0644); |
7762 | MODULE_PARM_DESC(l2arc_headroom_boost, "Compressed l2arc_headroom multiplier"); | |
7763 | ||
bce45ec9 | 7764 | module_param(l2arc_feed_secs, ulong, 0644); |
abd8610c BB |
7765 | MODULE_PARM_DESC(l2arc_feed_secs, "Seconds between L2ARC writing"); |
7766 | ||
bce45ec9 | 7767 | module_param(l2arc_feed_min_ms, ulong, 0644); |
abd8610c BB |
7768 | MODULE_PARM_DESC(l2arc_feed_min_ms, "Min feed interval in milliseconds"); |
7769 | ||
bce45ec9 | 7770 | module_param(l2arc_noprefetch, int, 0644); |
abd8610c BB |
7771 | MODULE_PARM_DESC(l2arc_noprefetch, "Skip caching prefetched buffers"); |
7772 | ||
bce45ec9 | 7773 | module_param(l2arc_feed_again, int, 0644); |
abd8610c BB |
7774 | MODULE_PARM_DESC(l2arc_feed_again, "Turbo L2ARC warmup"); |
7775 | ||
bce45ec9 | 7776 | module_param(l2arc_norw, int, 0644); |
abd8610c BB |
7777 | MODULE_PARM_DESC(l2arc_norw, "No reads during writes"); |
7778 | ||
7e8bddd0 BB |
7779 | module_param(zfs_arc_lotsfree_percent, int, 0644); |
7780 | MODULE_PARM_DESC(zfs_arc_lotsfree_percent, | |
7781 | "System free memory I/O throttle in bytes"); | |
7782 | ||
11f552fa BB |
7783 | module_param(zfs_arc_sys_free, ulong, 0644); |
7784 | MODULE_PARM_DESC(zfs_arc_sys_free, "System free memory target size in bytes"); | |
7785 | ||
25458cbe TC |
7786 | module_param(zfs_arc_dnode_limit, ulong, 0644); |
7787 | MODULE_PARM_DESC(zfs_arc_dnode_limit, "Minimum bytes of dnodes in arc"); | |
7788 | ||
9907cc1c G |
7789 | module_param(zfs_arc_dnode_limit_percent, ulong, 0644); |
7790 | MODULE_PARM_DESC(zfs_arc_dnode_limit_percent, | |
7791 | "Percent of ARC meta buffers for dnodes"); | |
7792 | ||
25458cbe TC |
7793 | module_param(zfs_arc_dnode_reduce_percent, ulong, 0644); |
7794 | MODULE_PARM_DESC(zfs_arc_dnode_reduce_percent, | |
7795 | "Percentage of excess dnodes to try to unpin"); | |
02730c33 | 7796 | /* END CSTYLED */ |
c28b2279 | 7797 | #endif |