]>
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. |
e10b0808 | 23 | * Copyright (c) 2012, Joyent, Inc. All rights reserved. |
cae5b340 | 24 | * Copyright (c) 2011, 2017 by Delphix. All rights reserved. |
e10b0808 | 25 | * Copyright (c) 2014 by Saso Kiselkov. All rights reserved. |
cae5b340 | 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 | |
a08ee875 | 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 | |
a08ee875 | 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 | |
cae5b340 | 80 | * uses method 1, while the internal ARC algorithms for |
a08ee875 | 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 |
cae5b340 | 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 | * | |
cae5b340 | 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 |
cae5b340 | 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 | * | |
e10b0808 | 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 | ||
cae5b340 AX |
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 | |
138 | * caches data in two ways -- in a list of ARC buffers (arc_buf_t) and | |
139 | * also in the arc_buf_hdr_t's private physical data block pointer (b_pabd). | |
140 | * | |
141 | * The L1ARC's data pointer may or may not be uncompressed. The ARC has the | |
142 | * ability to store the physical data (b_pabd) associated with the DVA of the | |
143 | * arc_buf_hdr_t. Since the b_pabd is a copy of the on-disk physical block, | |
144 | * it will match its on-disk compression characteristics. This behavior can be | |
145 | * disabled by setting 'zfs_compressed_arc_enabled' to B_FALSE. When the | |
146 | * compressed ARC functionality is disabled, the b_pabd will point to an | |
147 | * uncompressed version of the on-disk data. | |
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 | |
156 | * "overhead_size" kstat. | |
157 | * | |
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. | |
165 | * | |
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). | |
171 | * | |
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 | |
185 | * | b_pabd +-+ |b_next +---->+-----------+ | |
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 | * +------+ | |
198 | * | |
199 | * When a consumer reads a block, the ARC must first look to see if the | |
200 | * arc_buf_hdr_t is cached. If the hdr is cached then the ARC allocates a new | |
201 | * arc_buf_t and either copies uncompressed data into a new data buffer from an | |
202 | * existing uncompressed arc_buf_t, decompresses the hdr's b_pabd buffer into a | |
203 | * new data buffer, or shares the hdr's b_pabd buffer, depending on whether the | |
204 | * hdr is compressed and the desired compression characteristics of the | |
205 | * arc_buf_t consumer. If the arc_buf_t ends up sharing data with the | |
206 | * arc_buf_hdr_t and both of them are uncompressed then the arc_buf_t must be | |
207 | * the last buffer in the hdr's b_buf list, however a shared compressed buf can | |
208 | * be anywhere in the hdr's list. | |
209 | * | |
210 | * The diagram below shows an example of an uncompressed ARC hdr that is | |
211 | * sharing its data with an arc_buf_t (note that the shared uncompressed buf is | |
212 | * the last element in the buf list): | |
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 +---->+---------+ | |
227 | * | b_pabd +-+ |---------| |b_next +-->NULL | |
228 | * +-----------+ | | | +---------+ | |
229 | * | |b_data +-+ | | | |
230 | * | +---------+ | |b_data +-+ | |
231 | * +->+------+ | +---------+ | | |
232 | * | | | | | |
233 | * uncompressed | | | | | |
234 | * data +------+ | | | |
235 | * ^ +->+------+ | | |
236 | * | uncompressed | | | | |
237 | * | data | | | | |
238 | * | +------+ | | |
239 | * +---------------------------------+ | |
240 | * | |
241 | * Writing to the ARC requires that the ARC first discard the hdr's b_pabd | |
242 | * since the physical block is about to be rewritten. The new data contents | |
243 | * will be contained in the arc_buf_t. As the I/O pipeline performs the write, | |
244 | * it may compress the data before writing it to disk. The ARC will be called | |
245 | * with the transformed data and will bcopy the transformed on-disk block into | |
246 | * a newly allocated b_pabd. Writes are always done into buffers which have | |
247 | * either been loaned (and hence are new and don't have other readers) or | |
248 | * buffers which have been released (and hence have their own hdr, if there | |
249 | * were originally other readers of the buf's original hdr). This ensures that | |
250 | * the ARC only needs to update a single buf and its hdr after a write occurs. | |
251 | * | |
252 | * When the L2ARC is in use, it will also take advantage of the b_pabd. The | |
253 | * L2ARC will always write the contents of b_pabd to the L2ARC. This means | |
254 | * that when compressed ARC is enabled that the L2ARC blocks are identical | |
255 | * to the on-disk block in the main data pool. This provides a significant | |
256 | * advantage since the ARC can leverage the bp's checksum when reading from the | |
257 | * L2ARC to determine if the contents are valid. However, if the compressed | |
258 | * ARC is disabled, then the L2ARC's block must be transformed to look | |
259 | * like the physical block in the main data pool before comparing the | |
260 | * checksum and determining its validity. | |
261 | */ | |
262 | ||
34dc7c2f BB |
263 | #include <sys/spa.h> |
264 | #include <sys/zio.h> | |
cae5b340 | 265 | #include <sys/spa_impl.h> |
c06d4368 | 266 | #include <sys/zio_compress.h> |
cae5b340 | 267 | #include <sys/zio_checksum.h> |
34dc7c2f BB |
268 | #include <sys/zfs_context.h> |
269 | #include <sys/arc.h> | |
e10b0808 | 270 | #include <sys/refcount.h> |
b128c09f | 271 | #include <sys/vdev.h> |
9babb374 | 272 | #include <sys/vdev_impl.h> |
a08ee875 | 273 | #include <sys/dsl_pool.h> |
cae5b340 | 274 | #include <sys/zio_checksum.h> |
e10b0808 | 275 | #include <sys/multilist.h> |
cae5b340 | 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> |
ea04106b | 282 | #include <linux/mm_compat.h> |
42f7b73b | 283 | #include <linux/page_compat.h> |
34dc7c2f BB |
284 | #endif |
285 | #include <sys/callb.h> | |
286 | #include <sys/kstat.h> | |
570827e1 | 287 | #include <sys/dmu_tx.h> |
428870ff | 288 | #include <zfs_fletcher.h> |
ea04106b AX |
289 | #include <sys/arc_impl.h> |
290 | #include <sys/trace_arc.h> | |
34dc7c2f | 291 | |
a08ee875 LG |
292 | #ifndef _KERNEL |
293 | /* set with ZFS_DEBUG=watch, to enable watchpoints on frozen buffers */ | |
294 | boolean_t arc_watch = B_FALSE; | |
295 | #endif | |
296 | ||
e10b0808 AX |
297 | static kmutex_t arc_reclaim_lock; |
298 | static kcondvar_t arc_reclaim_thread_cv; | |
299 | static boolean_t arc_reclaim_thread_exit; | |
300 | static kcondvar_t arc_reclaim_waiters_cv; | |
34dc7c2f | 301 | |
e10b0808 AX |
302 | /* |
303 | * The number of headers to evict in arc_evict_state_impl() before | |
304 | * dropping the sublist lock and evicting from another sublist. A lower | |
305 | * value means we're more likely to evict the "correct" header (i.e. the | |
306 | * oldest header in the arc state), but comes with higher overhead | |
307 | * (i.e. more invocations of arc_evict_state_impl()). | |
308 | */ | |
309 | int zfs_arc_evict_batch_limit = 10; | |
34dc7c2f BB |
310 | |
311 | /* number of seconds before growing cache again */ | |
e10b0808 | 312 | static int arc_grow_retry = 5; |
34dc7c2f | 313 | |
cae5b340 | 314 | /* shift of arc_c for calculating overflow limit in arc_get_data_impl */ |
e10b0808 | 315 | int zfs_arc_overflow_shift = 8; |
ea04106b | 316 | |
e10b0808 AX |
317 | /* shift of arc_c for calculating both min and max arc_p */ |
318 | static int arc_p_min_shift = 4; | |
d164b209 BB |
319 | |
320 | /* log2(fraction of arc to reclaim) */ | |
e10b0808 | 321 | static int arc_shrink_shift = 7; |
d164b209 | 322 | |
cae5b340 AX |
323 | /* percent of pagecache to reclaim arc to */ |
324 | #ifdef _KERNEL | |
325 | static uint_t zfs_arc_pc_percent = 0; | |
326 | #endif | |
327 | ||
34dc7c2f | 328 | /* |
e10b0808 AX |
329 | * log2(fraction of ARC which must be free to allow growing). |
330 | * I.e. If there is less than arc_c >> arc_no_grow_shift free memory, | |
331 | * when reading a new block into the ARC, we will evict an equal-sized block | |
332 | * from the ARC. | |
333 | * | |
334 | * This must be less than arc_shrink_shift, so that when we shrink the ARC, | |
335 | * we will still not allow it to grow. | |
34dc7c2f | 336 | */ |
e10b0808 | 337 | int arc_no_grow_shift = 5; |
c06d4368 | 338 | |
c06d4368 | 339 | |
e10b0808 AX |
340 | /* |
341 | * minimum lifespan of a prefetch block in clock ticks | |
342 | * (initialized in arc_init()) | |
343 | */ | |
344 | static int arc_min_prefetch_lifespan; | |
ea04106b | 345 | |
a08ee875 LG |
346 | /* |
347 | * If this percent of memory is free, don't throttle. | |
348 | */ | |
349 | int arc_lotsfree_percent = 10; | |
350 | ||
34dc7c2f BB |
351 | static int arc_dead; |
352 | ||
b128c09f BB |
353 | /* |
354 | * The arc has filled available memory and has now warmed up. | |
355 | */ | |
356 | static boolean_t arc_warm; | |
357 | ||
cae5b340 AX |
358 | /* |
359 | * log2 fraction of the zio arena to keep free. | |
360 | */ | |
361 | int arc_zio_arena_free_shift = 2; | |
362 | ||
34dc7c2f BB |
363 | /* |
364 | * These tunables are for performance analysis. | |
365 | */ | |
c28b2279 BB |
366 | unsigned long zfs_arc_max = 0; |
367 | unsigned long zfs_arc_min = 0; | |
368 | unsigned long zfs_arc_meta_limit = 0; | |
e10b0808 | 369 | unsigned long zfs_arc_meta_min = 0; |
cae5b340 AX |
370 | unsigned long zfs_arc_dnode_limit = 0; |
371 | unsigned long zfs_arc_dnode_reduce_percent = 10; | |
e10b0808 AX |
372 | int zfs_arc_grow_retry = 0; |
373 | int zfs_arc_shrink_shift = 0; | |
374 | int zfs_arc_p_min_shift = 0; | |
e10b0808 | 375 | int zfs_arc_average_blocksize = 8 * 1024; /* 8KB */ |
34dc7c2f | 376 | |
cae5b340 AX |
377 | int zfs_compressed_arc_enabled = B_TRUE; |
378 | ||
379 | /* | |
380 | * ARC will evict meta buffers that exceed arc_meta_limit. This | |
381 | * tunable make arc_meta_limit adjustable for different workloads. | |
382 | */ | |
383 | unsigned long zfs_arc_meta_limit_percent = 75; | |
384 | ||
385 | /* | |
386 | * Percentage that can be consumed by dnodes of ARC meta buffers. | |
387 | */ | |
388 | unsigned long zfs_arc_dnode_limit_percent = 10; | |
389 | ||
34dc7c2f | 390 | /* |
e10b0808 | 391 | * These tunables are Linux specific |
34dc7c2f | 392 | */ |
e10b0808 AX |
393 | unsigned long zfs_arc_sys_free = 0; |
394 | int zfs_arc_min_prefetch_lifespan = 0; | |
e10b0808 AX |
395 | int zfs_arc_p_dampener_disable = 1; |
396 | int zfs_arc_meta_prune = 10000; | |
397 | int zfs_arc_meta_strategy = ARC_STRATEGY_META_BALANCED; | |
398 | int zfs_arc_meta_adjust_restarts = 4096; | |
399 | int zfs_arc_lotsfree_percent = 10; | |
34dc7c2f BB |
400 | |
401 | /* The 6 states: */ | |
402 | static arc_state_t ARC_anon; | |
403 | static arc_state_t ARC_mru; | |
404 | static arc_state_t ARC_mru_ghost; | |
405 | static arc_state_t ARC_mfu; | |
406 | static arc_state_t ARC_mfu_ghost; | |
407 | static arc_state_t ARC_l2c_only; | |
408 | ||
409 | typedef struct arc_stats { | |
410 | kstat_named_t arcstat_hits; | |
411 | kstat_named_t arcstat_misses; | |
412 | kstat_named_t arcstat_demand_data_hits; | |
413 | kstat_named_t arcstat_demand_data_misses; | |
414 | kstat_named_t arcstat_demand_metadata_hits; | |
415 | kstat_named_t arcstat_demand_metadata_misses; | |
416 | kstat_named_t arcstat_prefetch_data_hits; | |
417 | kstat_named_t arcstat_prefetch_data_misses; | |
418 | kstat_named_t arcstat_prefetch_metadata_hits; | |
419 | kstat_named_t arcstat_prefetch_metadata_misses; | |
420 | kstat_named_t arcstat_mru_hits; | |
421 | kstat_named_t arcstat_mru_ghost_hits; | |
422 | kstat_named_t arcstat_mfu_hits; | |
423 | kstat_named_t arcstat_mfu_ghost_hits; | |
424 | kstat_named_t arcstat_deleted; | |
a08ee875 LG |
425 | /* |
426 | * Number of buffers that could not be evicted because the hash lock | |
427 | * was held by another thread. The lock may not necessarily be held | |
428 | * by something using the same buffer, since hash locks are shared | |
429 | * by multiple buffers. | |
430 | */ | |
34dc7c2f | 431 | kstat_named_t arcstat_mutex_miss; |
047218e2 AX |
432 | /* |
433 | * Number of buffers skipped when updating the access state due to the | |
434 | * header having already been released after acquiring the hash lock. | |
435 | */ | |
436 | kstat_named_t arcstat_access_skip; | |
a08ee875 LG |
437 | /* |
438 | * Number of buffers skipped because they have I/O in progress, are | |
047218e2 | 439 | * indirect prefetch buffers that have not lived long enough, or are |
a08ee875 LG |
440 | * not from the spa we're trying to evict from. |
441 | */ | |
34dc7c2f | 442 | kstat_named_t arcstat_evict_skip; |
e10b0808 AX |
443 | /* |
444 | * Number of times arc_evict_state() was unable to evict enough | |
445 | * buffers to reach its target amount. | |
446 | */ | |
447 | kstat_named_t arcstat_evict_not_enough; | |
428870ff BB |
448 | kstat_named_t arcstat_evict_l2_cached; |
449 | kstat_named_t arcstat_evict_l2_eligible; | |
450 | kstat_named_t arcstat_evict_l2_ineligible; | |
e10b0808 | 451 | kstat_named_t arcstat_evict_l2_skip; |
34dc7c2f BB |
452 | kstat_named_t arcstat_hash_elements; |
453 | kstat_named_t arcstat_hash_elements_max; | |
454 | kstat_named_t arcstat_hash_collisions; | |
455 | kstat_named_t arcstat_hash_chains; | |
456 | kstat_named_t arcstat_hash_chain_max; | |
457 | kstat_named_t arcstat_p; | |
458 | kstat_named_t arcstat_c; | |
459 | kstat_named_t arcstat_c_min; | |
460 | kstat_named_t arcstat_c_max; | |
461 | kstat_named_t arcstat_size; | |
cae5b340 AX |
462 | /* |
463 | * Number of compressed bytes stored in the arc_buf_hdr_t's b_pabd. | |
464 | * Note that the compressed bytes may match the uncompressed bytes | |
465 | * if the block is either not compressed or compressed arc is disabled. | |
466 | */ | |
467 | kstat_named_t arcstat_compressed_size; | |
468 | /* | |
469 | * Uncompressed size of the data stored in b_pabd. If compressed | |
470 | * arc is disabled then this value will be identical to the stat | |
471 | * above. | |
472 | */ | |
473 | kstat_named_t arcstat_uncompressed_size; | |
474 | /* | |
475 | * Number of bytes stored in all the arc_buf_t's. This is classified | |
476 | * as "overhead" since this data is typically short-lived and will | |
477 | * be evicted from the arc when it becomes unreferenced unless the | |
478 | * zfs_keep_uncompressed_metadata or zfs_keep_uncompressed_level | |
479 | * values have been set (see comment in dbuf.c for more information). | |
480 | */ | |
481 | kstat_named_t arcstat_overhead_size; | |
e10b0808 AX |
482 | /* |
483 | * Number of bytes consumed by internal ARC structures necessary | |
484 | * for tracking purposes; these structures are not actually | |
485 | * backed by ARC buffers. This includes arc_buf_hdr_t structures | |
486 | * (allocated via arc_buf_hdr_t_full and arc_buf_hdr_t_l2only | |
487 | * caches), and arc_buf_t structures (allocated via arc_buf_t | |
488 | * cache). | |
489 | */ | |
34dc7c2f | 490 | kstat_named_t arcstat_hdr_size; |
e10b0808 AX |
491 | /* |
492 | * Number of bytes consumed by ARC buffers of type equal to | |
493 | * ARC_BUFC_DATA. This is generally consumed by buffers backing | |
494 | * on disk user data (e.g. plain file contents). | |
495 | */ | |
d164b209 | 496 | kstat_named_t arcstat_data_size; |
e10b0808 AX |
497 | /* |
498 | * Number of bytes consumed by ARC buffers of type equal to | |
499 | * ARC_BUFC_METADATA. This is generally consumed by buffers | |
500 | * backing on disk data that is used for internal ZFS | |
501 | * structures (e.g. ZAP, dnode, indirect blocks, etc). | |
502 | */ | |
503 | kstat_named_t arcstat_metadata_size; | |
504 | /* | |
cae5b340 AX |
505 | * Number of bytes consumed by dmu_buf_impl_t objects. |
506 | */ | |
507 | kstat_named_t arcstat_dbuf_size; | |
508 | /* | |
509 | * Number of bytes consumed by dnode_t objects. | |
510 | */ | |
511 | kstat_named_t arcstat_dnode_size; | |
512 | /* | |
513 | * Number of bytes consumed by bonus buffers. | |
e10b0808 | 514 | */ |
cae5b340 | 515 | kstat_named_t arcstat_bonus_size; |
e10b0808 AX |
516 | /* |
517 | * Total number of bytes consumed by ARC buffers residing in the | |
518 | * arc_anon state. This includes *all* buffers in the arc_anon | |
519 | * state; e.g. data, metadata, evictable, and unevictable buffers | |
520 | * are all included in this value. | |
521 | */ | |
13be560d | 522 | kstat_named_t arcstat_anon_size; |
e10b0808 AX |
523 | /* |
524 | * Number of bytes consumed by ARC buffers that meet the | |
525 | * following criteria: backing buffers of type ARC_BUFC_DATA, | |
526 | * residing in the arc_anon state, and are eligible for eviction | |
527 | * (e.g. have no outstanding holds on the buffer). | |
528 | */ | |
529 | kstat_named_t arcstat_anon_evictable_data; | |
530 | /* | |
531 | * Number of bytes consumed by ARC buffers that meet the | |
532 | * following criteria: backing buffers of type ARC_BUFC_METADATA, | |
533 | * residing in the arc_anon state, and are eligible for eviction | |
534 | * (e.g. have no outstanding holds on the buffer). | |
535 | */ | |
536 | kstat_named_t arcstat_anon_evictable_metadata; | |
537 | /* | |
538 | * Total number of bytes consumed by ARC buffers residing in the | |
539 | * arc_mru state. This includes *all* buffers in the arc_mru | |
540 | * state; e.g. data, metadata, evictable, and unevictable buffers | |
541 | * are all included in this value. | |
542 | */ | |
13be560d | 543 | kstat_named_t arcstat_mru_size; |
e10b0808 AX |
544 | /* |
545 | * Number of bytes consumed by ARC buffers that meet the | |
546 | * following criteria: backing buffers of type ARC_BUFC_DATA, | |
547 | * residing in the arc_mru state, and are eligible for eviction | |
548 | * (e.g. have no outstanding holds on the buffer). | |
549 | */ | |
550 | kstat_named_t arcstat_mru_evictable_data; | |
551 | /* | |
552 | * Number of bytes consumed by ARC buffers that meet the | |
553 | * following criteria: backing buffers of type ARC_BUFC_METADATA, | |
554 | * residing in the arc_mru state, and are eligible for eviction | |
555 | * (e.g. have no outstanding holds on the buffer). | |
556 | */ | |
557 | kstat_named_t arcstat_mru_evictable_metadata; | |
558 | /* | |
559 | * Total number of bytes that *would have been* consumed by ARC | |
560 | * buffers in the arc_mru_ghost state. The key thing to note | |
561 | * here, is the fact that this size doesn't actually indicate | |
562 | * RAM consumption. The ghost lists only consist of headers and | |
563 | * don't actually have ARC buffers linked off of these headers. | |
564 | * Thus, *if* the headers had associated ARC buffers, these | |
565 | * buffers *would have* consumed this number of bytes. | |
566 | */ | |
13be560d | 567 | kstat_named_t arcstat_mru_ghost_size; |
e10b0808 AX |
568 | /* |
569 | * Number of bytes that *would have been* consumed by ARC | |
570 | * buffers that are eligible for eviction, of type | |
571 | * ARC_BUFC_DATA, and linked off the arc_mru_ghost state. | |
572 | */ | |
573 | kstat_named_t arcstat_mru_ghost_evictable_data; | |
574 | /* | |
575 | * Number of bytes that *would have been* consumed by ARC | |
576 | * buffers that are eligible for eviction, of type | |
577 | * ARC_BUFC_METADATA, and linked off the arc_mru_ghost state. | |
578 | */ | |
579 | kstat_named_t arcstat_mru_ghost_evictable_metadata; | |
580 | /* | |
581 | * Total number of bytes consumed by ARC buffers residing in the | |
582 | * arc_mfu state. This includes *all* buffers in the arc_mfu | |
583 | * state; e.g. data, metadata, evictable, and unevictable buffers | |
584 | * are all included in this value. | |
585 | */ | |
13be560d | 586 | kstat_named_t arcstat_mfu_size; |
e10b0808 AX |
587 | /* |
588 | * Number of bytes consumed by ARC buffers that are eligible for | |
589 | * eviction, of type ARC_BUFC_DATA, and reside in the arc_mfu | |
590 | * state. | |
591 | */ | |
592 | kstat_named_t arcstat_mfu_evictable_data; | |
593 | /* | |
594 | * Number of bytes consumed by ARC buffers that are eligible for | |
595 | * eviction, of type ARC_BUFC_METADATA, and reside in the | |
596 | * arc_mfu state. | |
597 | */ | |
598 | kstat_named_t arcstat_mfu_evictable_metadata; | |
599 | /* | |
600 | * Total number of bytes that *would have been* consumed by ARC | |
601 | * buffers in the arc_mfu_ghost state. See the comment above | |
602 | * arcstat_mru_ghost_size for more details. | |
603 | */ | |
13be560d | 604 | kstat_named_t arcstat_mfu_ghost_size; |
e10b0808 AX |
605 | /* |
606 | * Number of bytes that *would have been* consumed by ARC | |
607 | * buffers that are eligible for eviction, of type | |
608 | * ARC_BUFC_DATA, and linked off the arc_mfu_ghost state. | |
609 | */ | |
610 | kstat_named_t arcstat_mfu_ghost_evictable_data; | |
611 | /* | |
612 | * Number of bytes that *would have been* consumed by ARC | |
613 | * buffers that are eligible for eviction, of type | |
614 | * ARC_BUFC_METADATA, and linked off the arc_mru_ghost state. | |
615 | */ | |
616 | kstat_named_t arcstat_mfu_ghost_evictable_metadata; | |
34dc7c2f BB |
617 | kstat_named_t arcstat_l2_hits; |
618 | kstat_named_t arcstat_l2_misses; | |
619 | kstat_named_t arcstat_l2_feeds; | |
620 | kstat_named_t arcstat_l2_rw_clash; | |
d164b209 BB |
621 | kstat_named_t arcstat_l2_read_bytes; |
622 | kstat_named_t arcstat_l2_write_bytes; | |
34dc7c2f BB |
623 | kstat_named_t arcstat_l2_writes_sent; |
624 | kstat_named_t arcstat_l2_writes_done; | |
625 | kstat_named_t arcstat_l2_writes_error; | |
e10b0808 | 626 | kstat_named_t arcstat_l2_writes_lock_retry; |
34dc7c2f BB |
627 | kstat_named_t arcstat_l2_evict_lock_retry; |
628 | kstat_named_t arcstat_l2_evict_reading; | |
e10b0808 | 629 | kstat_named_t arcstat_l2_evict_l1cached; |
34dc7c2f BB |
630 | kstat_named_t arcstat_l2_free_on_write; |
631 | kstat_named_t arcstat_l2_abort_lowmem; | |
632 | kstat_named_t arcstat_l2_cksum_bad; | |
633 | kstat_named_t arcstat_l2_io_error; | |
cae5b340 AX |
634 | kstat_named_t arcstat_l2_lsize; |
635 | kstat_named_t arcstat_l2_psize; | |
34dc7c2f BB |
636 | kstat_named_t arcstat_l2_hdr_size; |
637 | kstat_named_t arcstat_memory_throttle_count; | |
7cb67b45 BB |
638 | kstat_named_t arcstat_memory_direct_count; |
639 | kstat_named_t arcstat_memory_indirect_count; | |
8ec27e97 AX |
640 | kstat_named_t arcstat_memory_all_bytes; |
641 | kstat_named_t arcstat_memory_free_bytes; | |
642 | kstat_named_t arcstat_memory_available_bytes; | |
1834f2d8 BB |
643 | kstat_named_t arcstat_no_grow; |
644 | kstat_named_t arcstat_tempreserve; | |
645 | kstat_named_t arcstat_loaned_bytes; | |
ab26409d | 646 | kstat_named_t arcstat_prune; |
1834f2d8 BB |
647 | kstat_named_t arcstat_meta_used; |
648 | kstat_named_t arcstat_meta_limit; | |
cae5b340 | 649 | kstat_named_t arcstat_dnode_limit; |
1834f2d8 | 650 | kstat_named_t arcstat_meta_max; |
e10b0808 | 651 | kstat_named_t arcstat_meta_min; |
cae5b340 AX |
652 | kstat_named_t arcstat_sync_wait_for_async; |
653 | kstat_named_t arcstat_demand_hit_predictive_prefetch; | |
e10b0808 AX |
654 | kstat_named_t arcstat_need_free; |
655 | kstat_named_t arcstat_sys_free; | |
34dc7c2f BB |
656 | } arc_stats_t; |
657 | ||
658 | static arc_stats_t arc_stats = { | |
659 | { "hits", KSTAT_DATA_UINT64 }, | |
660 | { "misses", KSTAT_DATA_UINT64 }, | |
661 | { "demand_data_hits", KSTAT_DATA_UINT64 }, | |
662 | { "demand_data_misses", KSTAT_DATA_UINT64 }, | |
663 | { "demand_metadata_hits", KSTAT_DATA_UINT64 }, | |
664 | { "demand_metadata_misses", KSTAT_DATA_UINT64 }, | |
665 | { "prefetch_data_hits", KSTAT_DATA_UINT64 }, | |
666 | { "prefetch_data_misses", KSTAT_DATA_UINT64 }, | |
667 | { "prefetch_metadata_hits", KSTAT_DATA_UINT64 }, | |
668 | { "prefetch_metadata_misses", KSTAT_DATA_UINT64 }, | |
669 | { "mru_hits", KSTAT_DATA_UINT64 }, | |
670 | { "mru_ghost_hits", KSTAT_DATA_UINT64 }, | |
671 | { "mfu_hits", KSTAT_DATA_UINT64 }, | |
672 | { "mfu_ghost_hits", KSTAT_DATA_UINT64 }, | |
673 | { "deleted", KSTAT_DATA_UINT64 }, | |
34dc7c2f | 674 | { "mutex_miss", KSTAT_DATA_UINT64 }, |
047218e2 | 675 | { "access_skip", KSTAT_DATA_UINT64 }, |
34dc7c2f | 676 | { "evict_skip", KSTAT_DATA_UINT64 }, |
e10b0808 | 677 | { "evict_not_enough", KSTAT_DATA_UINT64 }, |
428870ff BB |
678 | { "evict_l2_cached", KSTAT_DATA_UINT64 }, |
679 | { "evict_l2_eligible", KSTAT_DATA_UINT64 }, | |
680 | { "evict_l2_ineligible", KSTAT_DATA_UINT64 }, | |
e10b0808 | 681 | { "evict_l2_skip", KSTAT_DATA_UINT64 }, |
34dc7c2f BB |
682 | { "hash_elements", KSTAT_DATA_UINT64 }, |
683 | { "hash_elements_max", KSTAT_DATA_UINT64 }, | |
684 | { "hash_collisions", KSTAT_DATA_UINT64 }, | |
685 | { "hash_chains", KSTAT_DATA_UINT64 }, | |
686 | { "hash_chain_max", KSTAT_DATA_UINT64 }, | |
687 | { "p", KSTAT_DATA_UINT64 }, | |
688 | { "c", KSTAT_DATA_UINT64 }, | |
689 | { "c_min", KSTAT_DATA_UINT64 }, | |
690 | { "c_max", KSTAT_DATA_UINT64 }, | |
691 | { "size", KSTAT_DATA_UINT64 }, | |
cae5b340 AX |
692 | { "compressed_size", KSTAT_DATA_UINT64 }, |
693 | { "uncompressed_size", KSTAT_DATA_UINT64 }, | |
694 | { "overhead_size", KSTAT_DATA_UINT64 }, | |
34dc7c2f | 695 | { "hdr_size", KSTAT_DATA_UINT64 }, |
d164b209 | 696 | { "data_size", KSTAT_DATA_UINT64 }, |
e10b0808 | 697 | { "metadata_size", KSTAT_DATA_UINT64 }, |
cae5b340 AX |
698 | { "dbuf_size", KSTAT_DATA_UINT64 }, |
699 | { "dnode_size", KSTAT_DATA_UINT64 }, | |
700 | { "bonus_size", KSTAT_DATA_UINT64 }, | |
13be560d | 701 | { "anon_size", KSTAT_DATA_UINT64 }, |
e10b0808 AX |
702 | { "anon_evictable_data", KSTAT_DATA_UINT64 }, |
703 | { "anon_evictable_metadata", KSTAT_DATA_UINT64 }, | |
13be560d | 704 | { "mru_size", KSTAT_DATA_UINT64 }, |
e10b0808 AX |
705 | { "mru_evictable_data", KSTAT_DATA_UINT64 }, |
706 | { "mru_evictable_metadata", KSTAT_DATA_UINT64 }, | |
13be560d | 707 | { "mru_ghost_size", KSTAT_DATA_UINT64 }, |
e10b0808 AX |
708 | { "mru_ghost_evictable_data", KSTAT_DATA_UINT64 }, |
709 | { "mru_ghost_evictable_metadata", KSTAT_DATA_UINT64 }, | |
13be560d | 710 | { "mfu_size", KSTAT_DATA_UINT64 }, |
e10b0808 AX |
711 | { "mfu_evictable_data", KSTAT_DATA_UINT64 }, |
712 | { "mfu_evictable_metadata", KSTAT_DATA_UINT64 }, | |
13be560d | 713 | { "mfu_ghost_size", KSTAT_DATA_UINT64 }, |
e10b0808 AX |
714 | { "mfu_ghost_evictable_data", KSTAT_DATA_UINT64 }, |
715 | { "mfu_ghost_evictable_metadata", KSTAT_DATA_UINT64 }, | |
34dc7c2f BB |
716 | { "l2_hits", KSTAT_DATA_UINT64 }, |
717 | { "l2_misses", KSTAT_DATA_UINT64 }, | |
718 | { "l2_feeds", KSTAT_DATA_UINT64 }, | |
719 | { "l2_rw_clash", KSTAT_DATA_UINT64 }, | |
d164b209 BB |
720 | { "l2_read_bytes", KSTAT_DATA_UINT64 }, |
721 | { "l2_write_bytes", KSTAT_DATA_UINT64 }, | |
34dc7c2f BB |
722 | { "l2_writes_sent", KSTAT_DATA_UINT64 }, |
723 | { "l2_writes_done", KSTAT_DATA_UINT64 }, | |
724 | { "l2_writes_error", KSTAT_DATA_UINT64 }, | |
e10b0808 | 725 | { "l2_writes_lock_retry", KSTAT_DATA_UINT64 }, |
34dc7c2f BB |
726 | { "l2_evict_lock_retry", KSTAT_DATA_UINT64 }, |
727 | { "l2_evict_reading", KSTAT_DATA_UINT64 }, | |
e10b0808 | 728 | { "l2_evict_l1cached", KSTAT_DATA_UINT64 }, |
34dc7c2f BB |
729 | { "l2_free_on_write", KSTAT_DATA_UINT64 }, |
730 | { "l2_abort_lowmem", KSTAT_DATA_UINT64 }, | |
731 | { "l2_cksum_bad", KSTAT_DATA_UINT64 }, | |
732 | { "l2_io_error", KSTAT_DATA_UINT64 }, | |
733 | { "l2_size", KSTAT_DATA_UINT64 }, | |
c06d4368 | 734 | { "l2_asize", KSTAT_DATA_UINT64 }, |
34dc7c2f | 735 | { "l2_hdr_size", KSTAT_DATA_UINT64 }, |
1834f2d8 | 736 | { "memory_throttle_count", KSTAT_DATA_UINT64 }, |
7cb67b45 BB |
737 | { "memory_direct_count", KSTAT_DATA_UINT64 }, |
738 | { "memory_indirect_count", KSTAT_DATA_UINT64 }, | |
8ec27e97 AX |
739 | { "memory_all_bytes", KSTAT_DATA_UINT64 }, |
740 | { "memory_free_bytes", KSTAT_DATA_UINT64 }, | |
741 | { "memory_available_bytes", KSTAT_DATA_INT64 }, | |
1834f2d8 BB |
742 | { "arc_no_grow", KSTAT_DATA_UINT64 }, |
743 | { "arc_tempreserve", KSTAT_DATA_UINT64 }, | |
744 | { "arc_loaned_bytes", KSTAT_DATA_UINT64 }, | |
ab26409d | 745 | { "arc_prune", KSTAT_DATA_UINT64 }, |
1834f2d8 BB |
746 | { "arc_meta_used", KSTAT_DATA_UINT64 }, |
747 | { "arc_meta_limit", KSTAT_DATA_UINT64 }, | |
cae5b340 | 748 | { "arc_dnode_limit", KSTAT_DATA_UINT64 }, |
1834f2d8 | 749 | { "arc_meta_max", KSTAT_DATA_UINT64 }, |
e10b0808 | 750 | { "arc_meta_min", KSTAT_DATA_UINT64 }, |
cae5b340 AX |
751 | { "sync_wait_for_async", KSTAT_DATA_UINT64 }, |
752 | { "demand_hit_predictive_prefetch", KSTAT_DATA_UINT64 }, | |
e10b0808 AX |
753 | { "arc_need_free", KSTAT_DATA_UINT64 }, |
754 | { "arc_sys_free", KSTAT_DATA_UINT64 } | |
34dc7c2f BB |
755 | }; |
756 | ||
757 | #define ARCSTAT(stat) (arc_stats.stat.value.ui64) | |
758 | ||
759 | #define ARCSTAT_INCR(stat, val) \ | |
a08ee875 | 760 | atomic_add_64(&arc_stats.stat.value.ui64, (val)) |
34dc7c2f | 761 | |
428870ff | 762 | #define ARCSTAT_BUMP(stat) ARCSTAT_INCR(stat, 1) |
34dc7c2f BB |
763 | #define ARCSTAT_BUMPDOWN(stat) ARCSTAT_INCR(stat, -1) |
764 | ||
765 | #define ARCSTAT_MAX(stat, val) { \ | |
766 | uint64_t m; \ | |
767 | while ((val) > (m = arc_stats.stat.value.ui64) && \ | |
768 | (m != atomic_cas_64(&arc_stats.stat.value.ui64, m, (val)))) \ | |
769 | continue; \ | |
770 | } | |
771 | ||
772 | #define ARCSTAT_MAXSTAT(stat) \ | |
773 | ARCSTAT_MAX(stat##_max, arc_stats.stat.value.ui64) | |
774 | ||
775 | /* | |
776 | * We define a macro to allow ARC hits/misses to be easily broken down by | |
777 | * two separate conditions, giving a total of four different subtypes for | |
778 | * each of hits and misses (so eight statistics total). | |
779 | */ | |
780 | #define ARCSTAT_CONDSTAT(cond1, stat1, notstat1, cond2, stat2, notstat2, stat) \ | |
781 | if (cond1) { \ | |
782 | if (cond2) { \ | |
783 | ARCSTAT_BUMP(arcstat_##stat1##_##stat2##_##stat); \ | |
784 | } else { \ | |
785 | ARCSTAT_BUMP(arcstat_##stat1##_##notstat2##_##stat); \ | |
786 | } \ | |
787 | } else { \ | |
788 | if (cond2) { \ | |
789 | ARCSTAT_BUMP(arcstat_##notstat1##_##stat2##_##stat); \ | |
790 | } else { \ | |
791 | ARCSTAT_BUMP(arcstat_##notstat1##_##notstat2##_##stat);\ | |
792 | } \ | |
793 | } | |
794 | ||
795 | kstat_t *arc_ksp; | |
428870ff | 796 | static arc_state_t *arc_anon; |
34dc7c2f BB |
797 | static arc_state_t *arc_mru; |
798 | static arc_state_t *arc_mru_ghost; | |
799 | static arc_state_t *arc_mfu; | |
800 | static arc_state_t *arc_mfu_ghost; | |
801 | static arc_state_t *arc_l2c_only; | |
802 | ||
803 | /* | |
804 | * There are several ARC variables that are critical to export as kstats -- | |
805 | * but we don't want to have to grovel around in the kstat whenever we wish to | |
806 | * manipulate them. For these variables, we therefore define them to be in | |
807 | * terms of the statistic variable. This assures that we are not introducing | |
808 | * the possibility of inconsistency by having shadow copies of the variables, | |
809 | * while still allowing the code to be readable. | |
810 | */ | |
811 | #define arc_size ARCSTAT(arcstat_size) /* actual total arc size */ | |
812 | #define arc_p ARCSTAT(arcstat_p) /* target size of MRU */ | |
813 | #define arc_c ARCSTAT(arcstat_c) /* target size of cache */ | |
814 | #define arc_c_min ARCSTAT(arcstat_c_min) /* min target cache size */ | |
815 | #define arc_c_max ARCSTAT(arcstat_c_max) /* max target cache size */ | |
cae5b340 | 816 | #define arc_no_grow ARCSTAT(arcstat_no_grow) /* do not grow cache size */ |
1834f2d8 BB |
817 | #define arc_tempreserve ARCSTAT(arcstat_tempreserve) |
818 | #define arc_loaned_bytes ARCSTAT(arcstat_loaned_bytes) | |
a08ee875 | 819 | #define arc_meta_limit ARCSTAT(arcstat_meta_limit) /* max size for metadata */ |
cae5b340 | 820 | #define arc_dnode_limit ARCSTAT(arcstat_dnode_limit) /* max size for dnodes */ |
e10b0808 | 821 | #define arc_meta_min ARCSTAT(arcstat_meta_min) /* min size for metadata */ |
a08ee875 LG |
822 | #define arc_meta_used ARCSTAT(arcstat_meta_used) /* size of metadata */ |
823 | #define arc_meta_max ARCSTAT(arcstat_meta_max) /* max size of metadata */ | |
cae5b340 AX |
824 | #define arc_dbuf_size ARCSTAT(arcstat_dbuf_size) /* dbuf metadata */ |
825 | #define arc_dnode_size ARCSTAT(arcstat_dnode_size) /* dnode metadata */ | |
826 | #define arc_bonus_size ARCSTAT(arcstat_bonus_size) /* bonus buffer metadata */ | |
e10b0808 AX |
827 | #define arc_need_free ARCSTAT(arcstat_need_free) /* bytes to be freed */ |
828 | #define arc_sys_free ARCSTAT(arcstat_sys_free) /* target system free bytes */ | |
34dc7c2f | 829 | |
cae5b340 AX |
830 | /* compressed size of entire arc */ |
831 | #define arc_compressed_size ARCSTAT(arcstat_compressed_size) | |
832 | /* uncompressed size of entire arc */ | |
833 | #define arc_uncompressed_size ARCSTAT(arcstat_uncompressed_size) | |
834 | /* number of bytes in the arc from arc_buf_t's */ | |
835 | #define arc_overhead_size ARCSTAT(arcstat_overhead_size) | |
c06d4368 | 836 | |
ab26409d BB |
837 | static list_t arc_prune_list; |
838 | static kmutex_t arc_prune_mtx; | |
e10b0808 | 839 | static taskq_t *arc_prune_taskq; |
428870ff | 840 | |
34dc7c2f BB |
841 | #define GHOST_STATE(state) \ |
842 | ((state) == arc_mru_ghost || (state) == arc_mfu_ghost || \ | |
843 | (state) == arc_l2c_only) | |
844 | ||
e10b0808 AX |
845 | #define HDR_IN_HASH_TABLE(hdr) ((hdr)->b_flags & ARC_FLAG_IN_HASH_TABLE) |
846 | #define HDR_IO_IN_PROGRESS(hdr) ((hdr)->b_flags & ARC_FLAG_IO_IN_PROGRESS) | |
847 | #define HDR_IO_ERROR(hdr) ((hdr)->b_flags & ARC_FLAG_IO_ERROR) | |
848 | #define HDR_PREFETCH(hdr) ((hdr)->b_flags & ARC_FLAG_PREFETCH) | |
cae5b340 AX |
849 | #define HDR_COMPRESSION_ENABLED(hdr) \ |
850 | ((hdr)->b_flags & ARC_FLAG_COMPRESSED_ARC) | |
e10b0808 AX |
851 | |
852 | #define HDR_L2CACHE(hdr) ((hdr)->b_flags & ARC_FLAG_L2CACHE) | |
e10b0808 | 853 | #define HDR_L2_READING(hdr) \ |
cae5b340 AX |
854 | (((hdr)->b_flags & ARC_FLAG_IO_IN_PROGRESS) && \ |
855 | ((hdr)->b_flags & ARC_FLAG_HAS_L2HDR)) | |
e10b0808 AX |
856 | #define HDR_L2_WRITING(hdr) ((hdr)->b_flags & ARC_FLAG_L2_WRITING) |
857 | #define HDR_L2_EVICTED(hdr) ((hdr)->b_flags & ARC_FLAG_L2_EVICTED) | |
858 | #define HDR_L2_WRITE_HEAD(hdr) ((hdr)->b_flags & ARC_FLAG_L2_WRITE_HEAD) | |
cae5b340 | 859 | #define HDR_SHARED_DATA(hdr) ((hdr)->b_flags & ARC_FLAG_SHARED_DATA) |
e10b0808 AX |
860 | |
861 | #define HDR_ISTYPE_METADATA(hdr) \ | |
cae5b340 | 862 | ((hdr)->b_flags & ARC_FLAG_BUFC_METADATA) |
e10b0808 AX |
863 | #define HDR_ISTYPE_DATA(hdr) (!HDR_ISTYPE_METADATA(hdr)) |
864 | ||
865 | #define HDR_HAS_L1HDR(hdr) ((hdr)->b_flags & ARC_FLAG_HAS_L1HDR) | |
866 | #define HDR_HAS_L2HDR(hdr) ((hdr)->b_flags & ARC_FLAG_HAS_L2HDR) | |
34dc7c2f | 867 | |
cae5b340 AX |
868 | /* For storing compression mode in b_flags */ |
869 | #define HDR_COMPRESS_OFFSET (highbit64(ARC_FLAG_COMPRESS_0) - 1) | |
870 | ||
871 | #define HDR_GET_COMPRESS(hdr) ((enum zio_compress)BF32_GET((hdr)->b_flags, \ | |
872 | HDR_COMPRESS_OFFSET, SPA_COMPRESSBITS)) | |
873 | #define HDR_SET_COMPRESS(hdr, cmp) BF32_SET((hdr)->b_flags, \ | |
874 | HDR_COMPRESS_OFFSET, SPA_COMPRESSBITS, (cmp)); | |
875 | ||
876 | #define ARC_BUF_LAST(buf) ((buf)->b_next == NULL) | |
877 | #define ARC_BUF_SHARED(buf) ((buf)->b_flags & ARC_BUF_FLAG_SHARED) | |
878 | #define ARC_BUF_COMPRESSED(buf) ((buf)->b_flags & ARC_BUF_FLAG_COMPRESSED) | |
879 | ||
34dc7c2f BB |
880 | /* |
881 | * Other sizes | |
882 | */ | |
883 | ||
e10b0808 AX |
884 | #define HDR_FULL_SIZE ((int64_t)sizeof (arc_buf_hdr_t)) |
885 | #define HDR_L2ONLY_SIZE ((int64_t)offsetof(arc_buf_hdr_t, b_l1hdr)) | |
34dc7c2f BB |
886 | |
887 | /* | |
888 | * Hash table routines | |
889 | */ | |
890 | ||
00b46022 BB |
891 | #define HT_LOCK_ALIGN 64 |
892 | #define HT_LOCK_PAD (P2NPHASE(sizeof (kmutex_t), (HT_LOCK_ALIGN))) | |
34dc7c2f BB |
893 | |
894 | struct ht_lock { | |
895 | kmutex_t ht_lock; | |
896 | #ifdef _KERNEL | |
00b46022 | 897 | unsigned char pad[HT_LOCK_PAD]; |
34dc7c2f BB |
898 | #endif |
899 | }; | |
900 | ||
ea04106b | 901 | #define BUF_LOCKS 8192 |
34dc7c2f BB |
902 | typedef struct buf_hash_table { |
903 | uint64_t ht_mask; | |
904 | arc_buf_hdr_t **ht_table; | |
905 | struct ht_lock ht_locks[BUF_LOCKS]; | |
906 | } buf_hash_table_t; | |
907 | ||
908 | static buf_hash_table_t buf_hash_table; | |
909 | ||
910 | #define BUF_HASH_INDEX(spa, dva, birth) \ | |
911 | (buf_hash(spa, dva, birth) & buf_hash_table.ht_mask) | |
912 | #define BUF_HASH_LOCK_NTRY(idx) (buf_hash_table.ht_locks[idx & (BUF_LOCKS-1)]) | |
913 | #define BUF_HASH_LOCK(idx) (&(BUF_HASH_LOCK_NTRY(idx).ht_lock)) | |
428870ff BB |
914 | #define HDR_LOCK(hdr) \ |
915 | (BUF_HASH_LOCK(BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth))) | |
34dc7c2f BB |
916 | |
917 | uint64_t zfs_crc64_table[256]; | |
918 | ||
919 | /* | |
920 | * Level 2 ARC | |
921 | */ | |
922 | ||
923 | #define L2ARC_WRITE_SIZE (8 * 1024 * 1024) /* initial write max */ | |
c06d4368 | 924 | #define L2ARC_HEADROOM 2 /* num of writes */ |
cae5b340 | 925 | |
c06d4368 AX |
926 | /* |
927 | * If we discover during ARC scan any buffers to be compressed, we boost | |
928 | * our headroom for the next scanning cycle by this percentage multiple. | |
929 | */ | |
930 | #define L2ARC_HEADROOM_BOOST 200 | |
d164b209 BB |
931 | #define L2ARC_FEED_SECS 1 /* caching interval secs */ |
932 | #define L2ARC_FEED_MIN_MS 200 /* min caching interval ms */ | |
34dc7c2f | 933 | |
e10b0808 | 934 | /* |
cae5b340 AX |
935 | * We can feed L2ARC from two states of ARC buffers, mru and mfu, |
936 | * and each of the state has two types: data and metadata. | |
e10b0808 | 937 | */ |
cae5b340 | 938 | #define L2ARC_FEED_TYPES 4 |
e10b0808 | 939 | |
34dc7c2f BB |
940 | #define l2arc_writes_sent ARCSTAT(arcstat_l2_writes_sent) |
941 | #define l2arc_writes_done ARCSTAT(arcstat_l2_writes_done) | |
942 | ||
a08ee875 | 943 | /* L2ARC Performance Tunables */ |
abd8610c BB |
944 | unsigned long l2arc_write_max = L2ARC_WRITE_SIZE; /* def max write size */ |
945 | unsigned long l2arc_write_boost = L2ARC_WRITE_SIZE; /* extra warmup write */ | |
946 | unsigned long l2arc_headroom = L2ARC_HEADROOM; /* # of dev writes */ | |
c06d4368 | 947 | unsigned long l2arc_headroom_boost = L2ARC_HEADROOM_BOOST; |
abd8610c BB |
948 | unsigned long l2arc_feed_secs = L2ARC_FEED_SECS; /* interval seconds */ |
949 | unsigned long l2arc_feed_min_ms = L2ARC_FEED_MIN_MS; /* min interval msecs */ | |
950 | int l2arc_noprefetch = B_TRUE; /* don't cache prefetch bufs */ | |
951 | int l2arc_feed_again = B_TRUE; /* turbo warmup */ | |
c06d4368 | 952 | int l2arc_norw = B_FALSE; /* no reads during writes */ |
34dc7c2f BB |
953 | |
954 | /* | |
955 | * L2ARC Internals | |
956 | */ | |
34dc7c2f BB |
957 | static list_t L2ARC_dev_list; /* device list */ |
958 | static list_t *l2arc_dev_list; /* device list pointer */ | |
959 | static kmutex_t l2arc_dev_mtx; /* device list mutex */ | |
960 | static l2arc_dev_t *l2arc_dev_last; /* last device used */ | |
34dc7c2f BB |
961 | static list_t L2ARC_free_on_write; /* free after write buf list */ |
962 | static list_t *l2arc_free_on_write; /* free after write list ptr */ | |
963 | static kmutex_t l2arc_free_on_write_mtx; /* mutex for list */ | |
964 | static uint64_t l2arc_ndev; /* number of devices */ | |
965 | ||
966 | typedef struct l2arc_read_callback { | |
cae5b340 | 967 | arc_buf_hdr_t *l2rcb_hdr; /* read header */ |
c06d4368 | 968 | blkptr_t l2rcb_bp; /* original blkptr */ |
ea04106b | 969 | zbookmark_phys_t l2rcb_zb; /* original bookmark */ |
c06d4368 | 970 | int l2rcb_flags; /* original flags */ |
cae5b340 | 971 | abd_t *l2rcb_abd; /* temporary buffer */ |
34dc7c2f BB |
972 | } l2arc_read_callback_t; |
973 | ||
34dc7c2f BB |
974 | typedef struct l2arc_data_free { |
975 | /* protected by l2arc_free_on_write_mtx */ | |
cae5b340 | 976 | abd_t *l2df_abd; |
34dc7c2f | 977 | size_t l2df_size; |
cae5b340 | 978 | arc_buf_contents_t l2df_type; |
34dc7c2f BB |
979 | list_node_t l2df_list_node; |
980 | } l2arc_data_free_t; | |
981 | ||
982 | static kmutex_t l2arc_feed_thr_lock; | |
983 | static kcondvar_t l2arc_feed_thr_cv; | |
984 | static uint8_t l2arc_thread_exit; | |
985 | ||
cae5b340 AX |
986 | static abd_t *arc_get_data_abd(arc_buf_hdr_t *, uint64_t, void *); |
987 | static void *arc_get_data_buf(arc_buf_hdr_t *, uint64_t, void *); | |
988 | static void arc_get_data_impl(arc_buf_hdr_t *, uint64_t, void *); | |
989 | static void arc_free_data_abd(arc_buf_hdr_t *, abd_t *, uint64_t, void *); | |
990 | static void arc_free_data_buf(arc_buf_hdr_t *, void *, uint64_t, void *); | |
991 | static void arc_free_data_impl(arc_buf_hdr_t *hdr, uint64_t size, void *tag); | |
992 | static void arc_hdr_free_pabd(arc_buf_hdr_t *); | |
993 | static void arc_hdr_alloc_pabd(arc_buf_hdr_t *); | |
e10b0808 AX |
994 | static void arc_access(arc_buf_hdr_t *, kmutex_t *); |
995 | static boolean_t arc_is_overflowing(void); | |
996 | static void arc_buf_watch(arc_buf_t *); | |
997 | static void arc_tuning_update(void); | |
cae5b340 AX |
998 | static void arc_prune_async(int64_t); |
999 | static uint64_t arc_all_memory(void); | |
e10b0808 AX |
1000 | |
1001 | static arc_buf_contents_t arc_buf_type(arc_buf_hdr_t *); | |
1002 | static uint32_t arc_bufc_to_flags(arc_buf_contents_t); | |
cae5b340 AX |
1003 | static inline void arc_hdr_set_flags(arc_buf_hdr_t *hdr, arc_flags_t flags); |
1004 | static inline void arc_hdr_clear_flags(arc_buf_hdr_t *hdr, arc_flags_t flags); | |
e10b0808 AX |
1005 | |
1006 | static boolean_t l2arc_write_eligible(uint64_t, arc_buf_hdr_t *); | |
1007 | static void l2arc_read_done(zio_t *); | |
34dc7c2f BB |
1008 | |
1009 | static uint64_t | |
d164b209 | 1010 | buf_hash(uint64_t spa, const dva_t *dva, uint64_t birth) |
34dc7c2f | 1011 | { |
34dc7c2f BB |
1012 | uint8_t *vdva = (uint8_t *)dva; |
1013 | uint64_t crc = -1ULL; | |
1014 | int i; | |
1015 | ||
1016 | ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); | |
1017 | ||
1018 | for (i = 0; i < sizeof (dva_t); i++) | |
1019 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ vdva[i]) & 0xFF]; | |
1020 | ||
d164b209 | 1021 | crc ^= (spa>>8) ^ birth; |
34dc7c2f BB |
1022 | |
1023 | return (crc); | |
1024 | } | |
1025 | ||
cae5b340 AX |
1026 | #define HDR_EMPTY(hdr) \ |
1027 | ((hdr)->b_dva.dva_word[0] == 0 && \ | |
1028 | (hdr)->b_dva.dva_word[1] == 0) | |
34dc7c2f | 1029 | |
cae5b340 AX |
1030 | #define HDR_EQUAL(spa, dva, birth, hdr) \ |
1031 | ((hdr)->b_dva.dva_word[0] == (dva)->dva_word[0]) && \ | |
1032 | ((hdr)->b_dva.dva_word[1] == (dva)->dva_word[1]) && \ | |
1033 | ((hdr)->b_birth == birth) && ((hdr)->b_spa == spa) | |
34dc7c2f | 1034 | |
428870ff BB |
1035 | static void |
1036 | buf_discard_identity(arc_buf_hdr_t *hdr) | |
1037 | { | |
1038 | hdr->b_dva.dva_word[0] = 0; | |
1039 | hdr->b_dva.dva_word[1] = 0; | |
1040 | hdr->b_birth = 0; | |
428870ff BB |
1041 | } |
1042 | ||
34dc7c2f | 1043 | static arc_buf_hdr_t * |
ea04106b | 1044 | buf_hash_find(uint64_t spa, const blkptr_t *bp, kmutex_t **lockp) |
34dc7c2f | 1045 | { |
ea04106b AX |
1046 | const dva_t *dva = BP_IDENTITY(bp); |
1047 | uint64_t birth = BP_PHYSICAL_BIRTH(bp); | |
34dc7c2f BB |
1048 | uint64_t idx = BUF_HASH_INDEX(spa, dva, birth); |
1049 | kmutex_t *hash_lock = BUF_HASH_LOCK(idx); | |
e10b0808 | 1050 | arc_buf_hdr_t *hdr; |
34dc7c2f BB |
1051 | |
1052 | mutex_enter(hash_lock); | |
e10b0808 AX |
1053 | for (hdr = buf_hash_table.ht_table[idx]; hdr != NULL; |
1054 | hdr = hdr->b_hash_next) { | |
cae5b340 | 1055 | if (HDR_EQUAL(spa, dva, birth, hdr)) { |
34dc7c2f | 1056 | *lockp = hash_lock; |
e10b0808 | 1057 | return (hdr); |
34dc7c2f BB |
1058 | } |
1059 | } | |
1060 | mutex_exit(hash_lock); | |
1061 | *lockp = NULL; | |
1062 | return (NULL); | |
1063 | } | |
1064 | ||
1065 | /* | |
1066 | * Insert an entry into the hash table. If there is already an element | |
1067 | * equal to elem in the hash table, then the already existing element | |
1068 | * will be returned and the new element will not be inserted. | |
1069 | * Otherwise returns NULL. | |
e10b0808 | 1070 | * If lockp == NULL, the caller is assumed to already hold the hash lock. |
34dc7c2f BB |
1071 | */ |
1072 | static arc_buf_hdr_t * | |
e10b0808 | 1073 | buf_hash_insert(arc_buf_hdr_t *hdr, kmutex_t **lockp) |
34dc7c2f | 1074 | { |
e10b0808 | 1075 | uint64_t idx = BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth); |
34dc7c2f | 1076 | kmutex_t *hash_lock = BUF_HASH_LOCK(idx); |
e10b0808 | 1077 | arc_buf_hdr_t *fhdr; |
34dc7c2f BB |
1078 | uint32_t i; |
1079 | ||
e10b0808 AX |
1080 | ASSERT(!DVA_IS_EMPTY(&hdr->b_dva)); |
1081 | ASSERT(hdr->b_birth != 0); | |
1082 | ASSERT(!HDR_IN_HASH_TABLE(hdr)); | |
1083 | ||
1084 | if (lockp != NULL) { | |
1085 | *lockp = hash_lock; | |
1086 | mutex_enter(hash_lock); | |
1087 | } else { | |
1088 | ASSERT(MUTEX_HELD(hash_lock)); | |
1089 | } | |
1090 | ||
1091 | for (fhdr = buf_hash_table.ht_table[idx], i = 0; fhdr != NULL; | |
1092 | fhdr = fhdr->b_hash_next, i++) { | |
cae5b340 | 1093 | if (HDR_EQUAL(hdr->b_spa, &hdr->b_dva, hdr->b_birth, fhdr)) |
e10b0808 | 1094 | return (fhdr); |
34dc7c2f BB |
1095 | } |
1096 | ||
e10b0808 AX |
1097 | hdr->b_hash_next = buf_hash_table.ht_table[idx]; |
1098 | buf_hash_table.ht_table[idx] = hdr; | |
cae5b340 | 1099 | arc_hdr_set_flags(hdr, ARC_FLAG_IN_HASH_TABLE); |
34dc7c2f BB |
1100 | |
1101 | /* collect some hash table performance data */ | |
1102 | if (i > 0) { | |
1103 | ARCSTAT_BUMP(arcstat_hash_collisions); | |
1104 | if (i == 1) | |
1105 | ARCSTAT_BUMP(arcstat_hash_chains); | |
1106 | ||
1107 | ARCSTAT_MAX(arcstat_hash_chain_max, i); | |
1108 | } | |
1109 | ||
1110 | ARCSTAT_BUMP(arcstat_hash_elements); | |
1111 | ARCSTAT_MAXSTAT(arcstat_hash_elements); | |
1112 | ||
1113 | return (NULL); | |
1114 | } | |
1115 | ||
1116 | static void | |
e10b0808 | 1117 | buf_hash_remove(arc_buf_hdr_t *hdr) |
34dc7c2f | 1118 | { |
e10b0808 AX |
1119 | arc_buf_hdr_t *fhdr, **hdrp; |
1120 | uint64_t idx = BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth); | |
34dc7c2f BB |
1121 | |
1122 | ASSERT(MUTEX_HELD(BUF_HASH_LOCK(idx))); | |
e10b0808 | 1123 | ASSERT(HDR_IN_HASH_TABLE(hdr)); |
34dc7c2f | 1124 | |
e10b0808 AX |
1125 | hdrp = &buf_hash_table.ht_table[idx]; |
1126 | while ((fhdr = *hdrp) != hdr) { | |
cae5b340 | 1127 | ASSERT3P(fhdr, !=, NULL); |
e10b0808 | 1128 | hdrp = &fhdr->b_hash_next; |
34dc7c2f | 1129 | } |
e10b0808 AX |
1130 | *hdrp = hdr->b_hash_next; |
1131 | hdr->b_hash_next = NULL; | |
cae5b340 | 1132 | arc_hdr_clear_flags(hdr, ARC_FLAG_IN_HASH_TABLE); |
34dc7c2f BB |
1133 | |
1134 | /* collect some hash table performance data */ | |
1135 | ARCSTAT_BUMPDOWN(arcstat_hash_elements); | |
1136 | ||
1137 | if (buf_hash_table.ht_table[idx] && | |
1138 | buf_hash_table.ht_table[idx]->b_hash_next == NULL) | |
1139 | ARCSTAT_BUMPDOWN(arcstat_hash_chains); | |
1140 | } | |
1141 | ||
1142 | /* | |
1143 | * Global data structures and functions for the buf kmem cache. | |
1144 | */ | |
e10b0808 AX |
1145 | static kmem_cache_t *hdr_full_cache; |
1146 | static kmem_cache_t *hdr_l2only_cache; | |
34dc7c2f BB |
1147 | static kmem_cache_t *buf_cache; |
1148 | ||
1149 | static void | |
1150 | buf_fini(void) | |
1151 | { | |
1152 | int i; | |
1153 | ||
00b46022 | 1154 | #if defined(_KERNEL) && defined(HAVE_SPL) |
a08ee875 LG |
1155 | /* |
1156 | * Large allocations which do not require contiguous pages | |
1157 | * should be using vmem_free() in the linux kernel\ | |
1158 | */ | |
00b46022 BB |
1159 | vmem_free(buf_hash_table.ht_table, |
1160 | (buf_hash_table.ht_mask + 1) * sizeof (void *)); | |
1161 | #else | |
34dc7c2f BB |
1162 | kmem_free(buf_hash_table.ht_table, |
1163 | (buf_hash_table.ht_mask + 1) * sizeof (void *)); | |
00b46022 | 1164 | #endif |
34dc7c2f BB |
1165 | for (i = 0; i < BUF_LOCKS; i++) |
1166 | mutex_destroy(&buf_hash_table.ht_locks[i].ht_lock); | |
e10b0808 AX |
1167 | kmem_cache_destroy(hdr_full_cache); |
1168 | kmem_cache_destroy(hdr_l2only_cache); | |
34dc7c2f BB |
1169 | kmem_cache_destroy(buf_cache); |
1170 | } | |
1171 | ||
1172 | /* | |
1173 | * Constructor callback - called when the cache is empty | |
1174 | * and a new buf is requested. | |
1175 | */ | |
1176 | /* ARGSUSED */ | |
1177 | static int | |
e10b0808 AX |
1178 | hdr_full_cons(void *vbuf, void *unused, int kmflag) |
1179 | { | |
1180 | arc_buf_hdr_t *hdr = vbuf; | |
1181 | ||
1182 | bzero(hdr, HDR_FULL_SIZE); | |
1183 | cv_init(&hdr->b_l1hdr.b_cv, NULL, CV_DEFAULT, NULL); | |
1184 | refcount_create(&hdr->b_l1hdr.b_refcnt); | |
1185 | mutex_init(&hdr->b_l1hdr.b_freeze_lock, NULL, MUTEX_DEFAULT, NULL); | |
1186 | list_link_init(&hdr->b_l1hdr.b_arc_node); | |
1187 | list_link_init(&hdr->b_l2hdr.b_l2node); | |
1188 | multilist_link_init(&hdr->b_l1hdr.b_arc_node); | |
1189 | arc_space_consume(HDR_FULL_SIZE, ARC_SPACE_HDRS); | |
1190 | ||
1191 | return (0); | |
1192 | } | |
1193 | ||
1194 | /* ARGSUSED */ | |
1195 | static int | |
1196 | hdr_l2only_cons(void *vbuf, void *unused, int kmflag) | |
34dc7c2f | 1197 | { |
e10b0808 | 1198 | arc_buf_hdr_t *hdr = vbuf; |
34dc7c2f | 1199 | |
e10b0808 AX |
1200 | bzero(hdr, HDR_L2ONLY_SIZE); |
1201 | arc_space_consume(HDR_L2ONLY_SIZE, ARC_SPACE_L2HDRS); | |
34dc7c2f | 1202 | |
34dc7c2f BB |
1203 | return (0); |
1204 | } | |
1205 | ||
b128c09f BB |
1206 | /* ARGSUSED */ |
1207 | static int | |
1208 | buf_cons(void *vbuf, void *unused, int kmflag) | |
1209 | { | |
1210 | arc_buf_t *buf = vbuf; | |
1211 | ||
1212 | bzero(buf, sizeof (arc_buf_t)); | |
428870ff | 1213 | mutex_init(&buf->b_evict_lock, NULL, MUTEX_DEFAULT, NULL); |
d164b209 BB |
1214 | arc_space_consume(sizeof (arc_buf_t), ARC_SPACE_HDRS); |
1215 | ||
b128c09f BB |
1216 | return (0); |
1217 | } | |
1218 | ||
34dc7c2f BB |
1219 | /* |
1220 | * Destructor callback - called when a cached buf is | |
1221 | * no longer required. | |
1222 | */ | |
1223 | /* ARGSUSED */ | |
1224 | static void | |
e10b0808 AX |
1225 | hdr_full_dest(void *vbuf, void *unused) |
1226 | { | |
1227 | arc_buf_hdr_t *hdr = vbuf; | |
1228 | ||
cae5b340 | 1229 | ASSERT(HDR_EMPTY(hdr)); |
e10b0808 AX |
1230 | cv_destroy(&hdr->b_l1hdr.b_cv); |
1231 | refcount_destroy(&hdr->b_l1hdr.b_refcnt); | |
1232 | mutex_destroy(&hdr->b_l1hdr.b_freeze_lock); | |
1233 | ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); | |
1234 | arc_space_return(HDR_FULL_SIZE, ARC_SPACE_HDRS); | |
1235 | } | |
1236 | ||
1237 | /* ARGSUSED */ | |
1238 | static void | |
1239 | hdr_l2only_dest(void *vbuf, void *unused) | |
34dc7c2f | 1240 | { |
e10b0808 | 1241 | ASSERTV(arc_buf_hdr_t *hdr = vbuf); |
34dc7c2f | 1242 | |
cae5b340 | 1243 | ASSERT(HDR_EMPTY(hdr)); |
e10b0808 | 1244 | arc_space_return(HDR_L2ONLY_SIZE, ARC_SPACE_L2HDRS); |
34dc7c2f BB |
1245 | } |
1246 | ||
b128c09f BB |
1247 | /* ARGSUSED */ |
1248 | static void | |
1249 | buf_dest(void *vbuf, void *unused) | |
1250 | { | |
1251 | arc_buf_t *buf = vbuf; | |
1252 | ||
428870ff | 1253 | mutex_destroy(&buf->b_evict_lock); |
d164b209 | 1254 | arc_space_return(sizeof (arc_buf_t), ARC_SPACE_HDRS); |
b128c09f BB |
1255 | } |
1256 | ||
e10b0808 AX |
1257 | /* |
1258 | * Reclaim callback -- invoked when memory is low. | |
1259 | */ | |
1260 | /* ARGSUSED */ | |
1261 | static void | |
1262 | hdr_recl(void *unused) | |
1263 | { | |
1264 | dprintf("hdr_recl called\n"); | |
1265 | /* | |
1266 | * umem calls the reclaim func when we destroy the buf cache, | |
1267 | * which is after we do arc_fini(). | |
1268 | */ | |
1269 | if (!arc_dead) | |
1270 | cv_signal(&arc_reclaim_thread_cv); | |
1271 | } | |
1272 | ||
34dc7c2f BB |
1273 | static void |
1274 | buf_init(void) | |
1275 | { | |
cae5b340 | 1276 | uint64_t *ct = NULL; |
34dc7c2f BB |
1277 | uint64_t hsize = 1ULL << 12; |
1278 | int i, j; | |
1279 | ||
1280 | /* | |
1281 | * The hash table is big enough to fill all of physical memory | |
ea04106b AX |
1282 | * with an average block size of zfs_arc_average_blocksize (default 8K). |
1283 | * By default, the table will take up | |
1284 | * totalmem * sizeof(void*) / 8K (1MB per GB with 8-byte pointers). | |
34dc7c2f | 1285 | */ |
cae5b340 | 1286 | while (hsize * zfs_arc_average_blocksize < arc_all_memory()) |
34dc7c2f BB |
1287 | hsize <<= 1; |
1288 | retry: | |
1289 | buf_hash_table.ht_mask = hsize - 1; | |
00b46022 | 1290 | #if defined(_KERNEL) && defined(HAVE_SPL) |
a08ee875 LG |
1291 | /* |
1292 | * Large allocations which do not require contiguous pages | |
1293 | * should be using vmem_alloc() in the linux kernel | |
1294 | */ | |
00b46022 BB |
1295 | buf_hash_table.ht_table = |
1296 | vmem_zalloc(hsize * sizeof (void*), KM_SLEEP); | |
1297 | #else | |
34dc7c2f BB |
1298 | buf_hash_table.ht_table = |
1299 | kmem_zalloc(hsize * sizeof (void*), KM_NOSLEEP); | |
00b46022 | 1300 | #endif |
34dc7c2f BB |
1301 | if (buf_hash_table.ht_table == NULL) { |
1302 | ASSERT(hsize > (1ULL << 8)); | |
1303 | hsize >>= 1; | |
1304 | goto retry; | |
1305 | } | |
1306 | ||
e10b0808 AX |
1307 | hdr_full_cache = kmem_cache_create("arc_buf_hdr_t_full", HDR_FULL_SIZE, |
1308 | 0, hdr_full_cons, hdr_full_dest, hdr_recl, NULL, NULL, 0); | |
1309 | hdr_l2only_cache = kmem_cache_create("arc_buf_hdr_t_l2only", | |
1310 | HDR_L2ONLY_SIZE, 0, hdr_l2only_cons, hdr_l2only_dest, hdr_recl, | |
1311 | NULL, NULL, 0); | |
34dc7c2f | 1312 | buf_cache = kmem_cache_create("arc_buf_t", sizeof (arc_buf_t), |
b128c09f | 1313 | 0, buf_cons, buf_dest, NULL, NULL, NULL, 0); |
34dc7c2f BB |
1314 | |
1315 | for (i = 0; i < 256; i++) | |
1316 | for (ct = zfs_crc64_table + i, *ct = i, j = 8; j > 0; j--) | |
1317 | *ct = (*ct >> 1) ^ (-(*ct & 1) & ZFS_CRC64_POLY); | |
1318 | ||
1319 | for (i = 0; i < BUF_LOCKS; i++) { | |
1320 | mutex_init(&buf_hash_table.ht_locks[i].ht_lock, | |
1321 | NULL, MUTEX_DEFAULT, NULL); | |
1322 | } | |
1323 | } | |
1324 | ||
cae5b340 AX |
1325 | #define ARC_MINTIME (hz>>4) /* 62 ms */ |
1326 | ||
e10b0808 | 1327 | /* |
cae5b340 AX |
1328 | * This is the size that the buf occupies in memory. If the buf is compressed, |
1329 | * it will correspond to the compressed size. You should use this method of | |
1330 | * getting the buf size unless you explicitly need the logical size. | |
e10b0808 | 1331 | */ |
cae5b340 AX |
1332 | uint64_t |
1333 | arc_buf_size(arc_buf_t *buf) | |
e10b0808 | 1334 | { |
cae5b340 AX |
1335 | return (ARC_BUF_COMPRESSED(buf) ? |
1336 | HDR_GET_PSIZE(buf->b_hdr) : HDR_GET_LSIZE(buf->b_hdr)); | |
1337 | } | |
e10b0808 | 1338 | |
cae5b340 AX |
1339 | uint64_t |
1340 | arc_buf_lsize(arc_buf_t *buf) | |
1341 | { | |
1342 | return (HDR_GET_LSIZE(buf->b_hdr)); | |
1343 | } | |
e10b0808 | 1344 | |
cae5b340 AX |
1345 | enum zio_compress |
1346 | arc_get_compression(arc_buf_t *buf) | |
1347 | { | |
1348 | return (ARC_BUF_COMPRESSED(buf) ? | |
1349 | HDR_GET_COMPRESS(buf->b_hdr) : ZIO_COMPRESS_OFF); | |
1350 | } | |
e10b0808 | 1351 | |
cae5b340 AX |
1352 | static inline boolean_t |
1353 | arc_buf_is_shared(arc_buf_t *buf) | |
1354 | { | |
1355 | boolean_t shared = (buf->b_data != NULL && | |
1356 | buf->b_hdr->b_l1hdr.b_pabd != NULL && | |
1357 | abd_is_linear(buf->b_hdr->b_l1hdr.b_pabd) && | |
1358 | buf->b_data == abd_to_buf(buf->b_hdr->b_l1hdr.b_pabd)); | |
1359 | IMPLY(shared, HDR_SHARED_DATA(buf->b_hdr)); | |
1360 | IMPLY(shared, ARC_BUF_SHARED(buf)); | |
1361 | IMPLY(shared, ARC_BUF_COMPRESSED(buf) || ARC_BUF_LAST(buf)); | |
e10b0808 AX |
1362 | |
1363 | /* | |
cae5b340 AX |
1364 | * It would be nice to assert arc_can_share() too, but the "hdr isn't |
1365 | * already being shared" requirement prevents us from doing that. | |
e10b0808 AX |
1366 | */ |
1367 | ||
cae5b340 AX |
1368 | return (shared); |
1369 | } | |
e10b0808 | 1370 | |
cae5b340 AX |
1371 | /* |
1372 | * Free the checksum associated with this header. If there is no checksum, this | |
1373 | * is a no-op. | |
1374 | */ | |
1375 | static inline void | |
1376 | arc_cksum_free(arc_buf_hdr_t *hdr) | |
1377 | { | |
1378 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
1379 | mutex_enter(&hdr->b_l1hdr.b_freeze_lock); | |
1380 | if (hdr->b_l1hdr.b_freeze_cksum != NULL) { | |
1381 | kmem_free(hdr->b_l1hdr.b_freeze_cksum, sizeof (zio_cksum_t)); | |
1382 | hdr->b_l1hdr.b_freeze_cksum = NULL; | |
1383 | } | |
1384 | mutex_exit(&hdr->b_l1hdr.b_freeze_lock); | |
e10b0808 AX |
1385 | } |
1386 | ||
cae5b340 AX |
1387 | /* |
1388 | * Return true iff at least one of the bufs on hdr is not compressed. | |
1389 | */ | |
1390 | static boolean_t | |
1391 | arc_hdr_has_uncompressed_buf(arc_buf_hdr_t *hdr) | |
1392 | { | |
1393 | for (arc_buf_t *b = hdr->b_l1hdr.b_buf; b != NULL; b = b->b_next) { | |
1394 | if (!ARC_BUF_COMPRESSED(b)) { | |
1395 | return (B_TRUE); | |
1396 | } | |
1397 | } | |
1398 | return (B_FALSE); | |
1399 | } | |
e10b0808 | 1400 | |
34dc7c2f | 1401 | |
cae5b340 AX |
1402 | /* |
1403 | * If we've turned on the ZFS_DEBUG_MODIFY flag, verify that the buf's data | |
1404 | * matches the checksum that is stored in the hdr. If there is no checksum, | |
1405 | * or if the buf is compressed, this is a no-op. | |
1406 | */ | |
34dc7c2f BB |
1407 | static void |
1408 | arc_cksum_verify(arc_buf_t *buf) | |
1409 | { | |
cae5b340 | 1410 | arc_buf_hdr_t *hdr = buf->b_hdr; |
34dc7c2f BB |
1411 | zio_cksum_t zc; |
1412 | ||
1413 | if (!(zfs_flags & ZFS_DEBUG_MODIFY)) | |
1414 | return; | |
1415 | ||
cae5b340 AX |
1416 | if (ARC_BUF_COMPRESSED(buf)) { |
1417 | ASSERT(hdr->b_l1hdr.b_freeze_cksum == NULL || | |
1418 | arc_hdr_has_uncompressed_buf(hdr)); | |
1419 | return; | |
1420 | } | |
1421 | ||
1422 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
1423 | ||
1424 | mutex_enter(&hdr->b_l1hdr.b_freeze_lock); | |
1425 | if (hdr->b_l1hdr.b_freeze_cksum == NULL || HDR_IO_ERROR(hdr)) { | |
1426 | mutex_exit(&hdr->b_l1hdr.b_freeze_lock); | |
34dc7c2f BB |
1427 | return; |
1428 | } | |
cae5b340 AX |
1429 | |
1430 | fletcher_2_native(buf->b_data, arc_buf_size(buf), NULL, &zc); | |
1431 | if (!ZIO_CHECKSUM_EQUAL(*hdr->b_l1hdr.b_freeze_cksum, zc)) | |
34dc7c2f | 1432 | panic("buffer modified while frozen!"); |
cae5b340 | 1433 | mutex_exit(&hdr->b_l1hdr.b_freeze_lock); |
34dc7c2f BB |
1434 | } |
1435 | ||
cae5b340 AX |
1436 | static boolean_t |
1437 | arc_cksum_is_equal(arc_buf_hdr_t *hdr, zio_t *zio) | |
34dc7c2f | 1438 | { |
cae5b340 AX |
1439 | enum zio_compress compress = BP_GET_COMPRESS(zio->io_bp); |
1440 | boolean_t valid_cksum; | |
34dc7c2f | 1441 | |
cae5b340 AX |
1442 | ASSERT(!BP_IS_EMBEDDED(zio->io_bp)); |
1443 | VERIFY3U(BP_GET_PSIZE(zio->io_bp), ==, HDR_GET_PSIZE(hdr)); | |
1444 | ||
1445 | /* | |
1446 | * We rely on the blkptr's checksum to determine if the block | |
1447 | * is valid or not. When compressed arc is enabled, the l2arc | |
1448 | * writes the block to the l2arc just as it appears in the pool. | |
1449 | * This allows us to use the blkptr's checksum to validate the | |
1450 | * data that we just read off of the l2arc without having to store | |
1451 | * a separate checksum in the arc_buf_hdr_t. However, if compressed | |
1452 | * arc is disabled, then the data written to the l2arc is always | |
1453 | * uncompressed and won't match the block as it exists in the main | |
1454 | * pool. When this is the case, we must first compress it if it is | |
1455 | * compressed on the main pool before we can validate the checksum. | |
1456 | */ | |
1457 | if (!HDR_COMPRESSION_ENABLED(hdr) && compress != ZIO_COMPRESS_OFF) { | |
1458 | uint64_t lsize; | |
1459 | uint64_t csize; | |
1460 | void *cbuf; | |
1461 | ASSERT3U(HDR_GET_COMPRESS(hdr), ==, ZIO_COMPRESS_OFF); | |
1462 | ||
1463 | cbuf = zio_buf_alloc(HDR_GET_PSIZE(hdr)); | |
1464 | lsize = HDR_GET_LSIZE(hdr); | |
1465 | csize = zio_compress_data(compress, zio->io_abd, cbuf, lsize); | |
1466 | ||
1467 | ASSERT3U(csize, <=, HDR_GET_PSIZE(hdr)); | |
1468 | if (csize < HDR_GET_PSIZE(hdr)) { | |
1469 | /* | |
1470 | * Compressed blocks are always a multiple of the | |
1471 | * smallest ashift in the pool. Ideally, we would | |
1472 | * like to round up the csize to the next | |
1473 | * spa_min_ashift but that value may have changed | |
1474 | * since the block was last written. Instead, | |
1475 | * we rely on the fact that the hdr's psize | |
1476 | * was set to the psize of the block when it was | |
1477 | * last written. We set the csize to that value | |
1478 | * and zero out any part that should not contain | |
1479 | * data. | |
1480 | */ | |
1481 | bzero((char *)cbuf + csize, HDR_GET_PSIZE(hdr) - csize); | |
1482 | csize = HDR_GET_PSIZE(hdr); | |
1483 | } | |
1484 | zio_push_transform(zio, cbuf, csize, HDR_GET_PSIZE(hdr), NULL); | |
1485 | } | |
34dc7c2f | 1486 | |
cae5b340 AX |
1487 | /* |
1488 | * Block pointers always store the checksum for the logical data. | |
1489 | * If the block pointer has the gang bit set, then the checksum | |
1490 | * it represents is for the reconstituted data and not for an | |
1491 | * individual gang member. The zio pipeline, however, must be able to | |
1492 | * determine the checksum of each of the gang constituents so it | |
1493 | * treats the checksum comparison differently than what we need | |
1494 | * for l2arc blocks. This prevents us from using the | |
1495 | * zio_checksum_error() interface directly. Instead we must call the | |
1496 | * zio_checksum_error_impl() so that we can ensure the checksum is | |
1497 | * generated using the correct checksum algorithm and accounts for the | |
1498 | * logical I/O size and not just a gang fragment. | |
1499 | */ | |
1500 | valid_cksum = (zio_checksum_error_impl(zio->io_spa, zio->io_bp, | |
1501 | BP_GET_CHECKSUM(zio->io_bp), zio->io_abd, zio->io_size, | |
1502 | zio->io_offset, NULL) == 0); | |
1503 | zio_pop_transforms(zio); | |
1504 | return (valid_cksum); | |
34dc7c2f BB |
1505 | } |
1506 | ||
cae5b340 AX |
1507 | /* |
1508 | * Given a buf full of data, if ZFS_DEBUG_MODIFY is enabled this computes a | |
1509 | * checksum and attaches it to the buf's hdr so that we can ensure that the buf | |
1510 | * isn't modified later on. If buf is compressed or there is already a checksum | |
1511 | * on the hdr, this is a no-op (we only checksum uncompressed bufs). | |
1512 | */ | |
34dc7c2f | 1513 | static void |
cae5b340 | 1514 | arc_cksum_compute(arc_buf_t *buf) |
34dc7c2f | 1515 | { |
cae5b340 AX |
1516 | arc_buf_hdr_t *hdr = buf->b_hdr; |
1517 | ||
1518 | if (!(zfs_flags & ZFS_DEBUG_MODIFY)) | |
34dc7c2f BB |
1519 | return; |
1520 | ||
cae5b340 AX |
1521 | ASSERT(HDR_HAS_L1HDR(hdr)); |
1522 | ||
e10b0808 | 1523 | mutex_enter(&buf->b_hdr->b_l1hdr.b_freeze_lock); |
cae5b340 AX |
1524 | if (hdr->b_l1hdr.b_freeze_cksum != NULL) { |
1525 | ASSERT(arc_hdr_has_uncompressed_buf(hdr)); | |
1526 | mutex_exit(&hdr->b_l1hdr.b_freeze_lock); | |
1527 | return; | |
1528 | } else if (ARC_BUF_COMPRESSED(buf)) { | |
1529 | mutex_exit(&hdr->b_l1hdr.b_freeze_lock); | |
34dc7c2f BB |
1530 | return; |
1531 | } | |
cae5b340 AX |
1532 | |
1533 | ASSERT(!ARC_BUF_COMPRESSED(buf)); | |
1534 | hdr->b_l1hdr.b_freeze_cksum = kmem_alloc(sizeof (zio_cksum_t), | |
1535 | KM_SLEEP); | |
1536 | fletcher_2_native(buf->b_data, arc_buf_size(buf), NULL, | |
1537 | hdr->b_l1hdr.b_freeze_cksum); | |
1538 | mutex_exit(&hdr->b_l1hdr.b_freeze_lock); | |
a08ee875 LG |
1539 | arc_buf_watch(buf); |
1540 | } | |
1541 | ||
1542 | #ifndef _KERNEL | |
1543 | void | |
1544 | arc_buf_sigsegv(int sig, siginfo_t *si, void *unused) | |
1545 | { | |
cae5b340 | 1546 | panic("Got SIGSEGV at address: 0x%lx\n", (long)si->si_addr); |
a08ee875 LG |
1547 | } |
1548 | #endif | |
1549 | ||
1550 | /* ARGSUSED */ | |
1551 | static void | |
1552 | arc_buf_unwatch(arc_buf_t *buf) | |
1553 | { | |
1554 | #ifndef _KERNEL | |
1555 | if (arc_watch) { | |
cae5b340 | 1556 | ASSERT0(mprotect(buf->b_data, arc_buf_size(buf), |
a08ee875 LG |
1557 | PROT_READ | PROT_WRITE)); |
1558 | } | |
1559 | #endif | |
1560 | } | |
1561 | ||
1562 | /* ARGSUSED */ | |
1563 | static void | |
1564 | arc_buf_watch(arc_buf_t *buf) | |
1565 | { | |
1566 | #ifndef _KERNEL | |
1567 | if (arc_watch) | |
cae5b340 AX |
1568 | ASSERT0(mprotect(buf->b_data, arc_buf_size(buf), |
1569 | PROT_READ)); | |
a08ee875 | 1570 | #endif |
34dc7c2f BB |
1571 | } |
1572 | ||
e10b0808 AX |
1573 | static arc_buf_contents_t |
1574 | arc_buf_type(arc_buf_hdr_t *hdr) | |
1575 | { | |
cae5b340 | 1576 | arc_buf_contents_t type; |
e10b0808 | 1577 | if (HDR_ISTYPE_METADATA(hdr)) { |
cae5b340 | 1578 | type = ARC_BUFC_METADATA; |
e10b0808 | 1579 | } else { |
cae5b340 | 1580 | type = ARC_BUFC_DATA; |
e10b0808 | 1581 | } |
cae5b340 AX |
1582 | VERIFY3U(hdr->b_type, ==, type); |
1583 | return (type); | |
1584 | } | |
1585 | ||
1586 | boolean_t | |
1587 | arc_is_metadata(arc_buf_t *buf) | |
1588 | { | |
1589 | return (HDR_ISTYPE_METADATA(buf->b_hdr) != 0); | |
e10b0808 AX |
1590 | } |
1591 | ||
1592 | static uint32_t | |
1593 | arc_bufc_to_flags(arc_buf_contents_t type) | |
1594 | { | |
1595 | switch (type) { | |
1596 | case ARC_BUFC_DATA: | |
1597 | /* metadata field is 0 if buffer contains normal data */ | |
1598 | return (0); | |
1599 | case ARC_BUFC_METADATA: | |
1600 | return (ARC_FLAG_BUFC_METADATA); | |
1601 | default: | |
1602 | break; | |
1603 | } | |
1604 | panic("undefined ARC buffer type!"); | |
1605 | return ((uint32_t)-1); | |
1606 | } | |
1607 | ||
34dc7c2f BB |
1608 | void |
1609 | arc_buf_thaw(arc_buf_t *buf) | |
1610 | { | |
cae5b340 | 1611 | arc_buf_hdr_t *hdr = buf->b_hdr; |
34dc7c2f | 1612 | |
cae5b340 AX |
1613 | ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon); |
1614 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); | |
428870ff | 1615 | |
cae5b340 AX |
1616 | arc_cksum_verify(buf); |
1617 | ||
1618 | /* | |
1619 | * Compressed buffers do not manipulate the b_freeze_cksum or | |
1620 | * allocate b_thawed. | |
1621 | */ | |
1622 | if (ARC_BUF_COMPRESSED(buf)) { | |
1623 | ASSERT(hdr->b_l1hdr.b_freeze_cksum == NULL || | |
1624 | arc_hdr_has_uncompressed_buf(hdr)); | |
1625 | return; | |
1626 | } | |
a08ee875 | 1627 | |
cae5b340 AX |
1628 | ASSERT(HDR_HAS_L1HDR(hdr)); |
1629 | arc_cksum_free(hdr); | |
a08ee875 | 1630 | arc_buf_unwatch(buf); |
34dc7c2f BB |
1631 | } |
1632 | ||
1633 | void | |
1634 | arc_buf_freeze(arc_buf_t *buf) | |
1635 | { | |
cae5b340 | 1636 | arc_buf_hdr_t *hdr = buf->b_hdr; |
428870ff BB |
1637 | kmutex_t *hash_lock; |
1638 | ||
34dc7c2f BB |
1639 | if (!(zfs_flags & ZFS_DEBUG_MODIFY)) |
1640 | return; | |
1641 | ||
cae5b340 AX |
1642 | if (ARC_BUF_COMPRESSED(buf)) { |
1643 | ASSERT(hdr->b_l1hdr.b_freeze_cksum == NULL || | |
1644 | arc_hdr_has_uncompressed_buf(hdr)); | |
1645 | return; | |
1646 | } | |
1647 | ||
1648 | hash_lock = HDR_LOCK(hdr); | |
428870ff BB |
1649 | mutex_enter(hash_lock); |
1650 | ||
cae5b340 AX |
1651 | ASSERT(HDR_HAS_L1HDR(hdr)); |
1652 | ASSERT(hdr->b_l1hdr.b_freeze_cksum != NULL || | |
1653 | hdr->b_l1hdr.b_state == arc_anon); | |
1654 | arc_cksum_compute(buf); | |
428870ff | 1655 | mutex_exit(hash_lock); |
cae5b340 AX |
1656 | } |
1657 | ||
1658 | /* | |
1659 | * The arc_buf_hdr_t's b_flags should never be modified directly. Instead, | |
1660 | * the following functions should be used to ensure that the flags are | |
1661 | * updated in a thread-safe way. When manipulating the flags either | |
1662 | * the hash_lock must be held or the hdr must be undiscoverable. This | |
1663 | * ensures that we're not racing with any other threads when updating | |
1664 | * the flags. | |
1665 | */ | |
1666 | static inline void | |
1667 | arc_hdr_set_flags(arc_buf_hdr_t *hdr, arc_flags_t flags) | |
1668 | { | |
1669 | ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); | |
1670 | hdr->b_flags |= flags; | |
1671 | } | |
a08ee875 | 1672 | |
cae5b340 AX |
1673 | static inline void |
1674 | arc_hdr_clear_flags(arc_buf_hdr_t *hdr, arc_flags_t flags) | |
1675 | { | |
1676 | ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); | |
1677 | hdr->b_flags &= ~flags; | |
34dc7c2f BB |
1678 | } |
1679 | ||
cae5b340 AX |
1680 | /* |
1681 | * Setting the compression bits in the arc_buf_hdr_t's b_flags is | |
1682 | * done in a special way since we have to clear and set bits | |
1683 | * at the same time. Consumers that wish to set the compression bits | |
1684 | * must use this function to ensure that the flags are updated in | |
1685 | * thread-safe manner. | |
1686 | */ | |
34dc7c2f | 1687 | static void |
cae5b340 | 1688 | arc_hdr_set_compress(arc_buf_hdr_t *hdr, enum zio_compress cmp) |
34dc7c2f | 1689 | { |
cae5b340 | 1690 | ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); |
e10b0808 | 1691 | |
cae5b340 AX |
1692 | /* |
1693 | * Holes and embedded blocks will always have a psize = 0 so | |
1694 | * we ignore the compression of the blkptr and set the | |
1695 | * want to uncompress them. Mark them as uncompressed. | |
1696 | */ | |
1697 | if (!zfs_compressed_arc_enabled || HDR_GET_PSIZE(hdr) == 0) { | |
1698 | arc_hdr_clear_flags(hdr, ARC_FLAG_COMPRESSED_ARC); | |
1699 | HDR_SET_COMPRESS(hdr, ZIO_COMPRESS_OFF); | |
1700 | ASSERT(!HDR_COMPRESSION_ENABLED(hdr)); | |
1701 | ASSERT3U(HDR_GET_COMPRESS(hdr), ==, ZIO_COMPRESS_OFF); | |
1702 | } else { | |
1703 | arc_hdr_set_flags(hdr, ARC_FLAG_COMPRESSED_ARC); | |
1704 | HDR_SET_COMPRESS(hdr, cmp); | |
1705 | ASSERT3U(HDR_GET_COMPRESS(hdr), ==, cmp); | |
1706 | ASSERT(HDR_COMPRESSION_ENABLED(hdr)); | |
1707 | } | |
1708 | } | |
34dc7c2f | 1709 | |
cae5b340 AX |
1710 | /* |
1711 | * Looks for another buf on the same hdr which has the data decompressed, copies | |
1712 | * from it, and returns true. If no such buf exists, returns false. | |
1713 | */ | |
1714 | static boolean_t | |
1715 | arc_buf_try_copy_decompressed_data(arc_buf_t *buf) | |
1716 | { | |
1717 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
1718 | boolean_t copied = B_FALSE; | |
e10b0808 | 1719 | |
cae5b340 AX |
1720 | ASSERT(HDR_HAS_L1HDR(hdr)); |
1721 | ASSERT3P(buf->b_data, !=, NULL); | |
1722 | ASSERT(!ARC_BUF_COMPRESSED(buf)); | |
e10b0808 | 1723 | |
cae5b340 AX |
1724 | for (arc_buf_t *from = hdr->b_l1hdr.b_buf; from != NULL; |
1725 | from = from->b_next) { | |
1726 | /* can't use our own data buffer */ | |
1727 | if (from == buf) { | |
1728 | continue; | |
1729 | } | |
e10b0808 | 1730 | |
cae5b340 AX |
1731 | if (!ARC_BUF_COMPRESSED(from)) { |
1732 | bcopy(from->b_data, buf->b_data, arc_buf_size(buf)); | |
1733 | copied = B_TRUE; | |
1734 | break; | |
34dc7c2f | 1735 | } |
34dc7c2f | 1736 | } |
cae5b340 AX |
1737 | |
1738 | /* | |
1739 | * There were no decompressed bufs, so there should not be a | |
1740 | * checksum on the hdr either. | |
1741 | */ | |
1742 | EQUIV(!copied, hdr->b_l1hdr.b_freeze_cksum == NULL); | |
1743 | ||
1744 | return (copied); | |
34dc7c2f BB |
1745 | } |
1746 | ||
cae5b340 AX |
1747 | /* |
1748 | * Given a buf that has a data buffer attached to it, this function will | |
1749 | * efficiently fill the buf with data of the specified compression setting from | |
1750 | * the hdr and update the hdr's b_freeze_cksum if necessary. If the buf and hdr | |
1751 | * are already sharing a data buf, no copy is performed. | |
1752 | * | |
1753 | * If the buf is marked as compressed but uncompressed data was requested, this | |
1754 | * will allocate a new data buffer for the buf, remove that flag, and fill the | |
1755 | * buf with uncompressed data. You can't request a compressed buf on a hdr with | |
1756 | * uncompressed data, and (since we haven't added support for it yet) if you | |
1757 | * want compressed data your buf must already be marked as compressed and have | |
1758 | * the correct-sized data buffer. | |
1759 | */ | |
34dc7c2f | 1760 | static int |
cae5b340 | 1761 | arc_buf_fill(arc_buf_t *buf, boolean_t compressed) |
34dc7c2f | 1762 | { |
cae5b340 AX |
1763 | arc_buf_hdr_t *hdr = buf->b_hdr; |
1764 | boolean_t hdr_compressed = (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF); | |
1765 | dmu_object_byteswap_t bswap = hdr->b_l1hdr.b_byteswap; | |
1766 | ||
1767 | ASSERT3P(buf->b_data, !=, NULL); | |
1768 | IMPLY(compressed, hdr_compressed); | |
1769 | IMPLY(compressed, ARC_BUF_COMPRESSED(buf)); | |
1770 | ||
1771 | if (hdr_compressed == compressed) { | |
1772 | if (!arc_buf_is_shared(buf)) { | |
1773 | abd_copy_to_buf(buf->b_data, hdr->b_l1hdr.b_pabd, | |
1774 | arc_buf_size(buf)); | |
1775 | } | |
1776 | } else { | |
1777 | ASSERT(hdr_compressed); | |
1778 | ASSERT(!compressed); | |
1779 | ASSERT3U(HDR_GET_LSIZE(hdr), !=, HDR_GET_PSIZE(hdr)); | |
1780 | ||
1781 | /* | |
1782 | * If the buf is sharing its data with the hdr, unlink it and | |
1783 | * allocate a new data buffer for the buf. | |
1784 | */ | |
1785 | if (arc_buf_is_shared(buf)) { | |
1786 | ASSERT(ARC_BUF_COMPRESSED(buf)); | |
1787 | ||
1788 | /* We need to give the buf it's own b_data */ | |
1789 | buf->b_flags &= ~ARC_BUF_FLAG_SHARED; | |
1790 | buf->b_data = | |
1791 | arc_get_data_buf(hdr, HDR_GET_LSIZE(hdr), buf); | |
1792 | arc_hdr_clear_flags(hdr, ARC_FLAG_SHARED_DATA); | |
1793 | ||
1794 | /* Previously overhead was 0; just add new overhead */ | |
1795 | ARCSTAT_INCR(arcstat_overhead_size, HDR_GET_LSIZE(hdr)); | |
1796 | } else if (ARC_BUF_COMPRESSED(buf)) { | |
1797 | /* We need to reallocate the buf's b_data */ | |
1798 | arc_free_data_buf(hdr, buf->b_data, HDR_GET_PSIZE(hdr), | |
1799 | buf); | |
1800 | buf->b_data = | |
1801 | arc_get_data_buf(hdr, HDR_GET_LSIZE(hdr), buf); | |
1802 | ||
1803 | /* We increased the size of b_data; update overhead */ | |
1804 | ARCSTAT_INCR(arcstat_overhead_size, | |
1805 | HDR_GET_LSIZE(hdr) - HDR_GET_PSIZE(hdr)); | |
1806 | } | |
1807 | ||
1808 | /* | |
1809 | * Regardless of the buf's previous compression settings, it | |
1810 | * should not be compressed at the end of this function. | |
1811 | */ | |
1812 | buf->b_flags &= ~ARC_BUF_FLAG_COMPRESSED; | |
1813 | ||
1814 | /* | |
1815 | * Try copying the data from another buf which already has a | |
1816 | * decompressed version. If that's not possible, it's time to | |
1817 | * bite the bullet and decompress the data from the hdr. | |
1818 | */ | |
1819 | if (arc_buf_try_copy_decompressed_data(buf)) { | |
1820 | /* Skip byteswapping and checksumming (already done) */ | |
1821 | ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, !=, NULL); | |
1822 | return (0); | |
1823 | } else { | |
1824 | int error = zio_decompress_data(HDR_GET_COMPRESS(hdr), | |
1825 | hdr->b_l1hdr.b_pabd, buf->b_data, | |
1826 | HDR_GET_PSIZE(hdr), HDR_GET_LSIZE(hdr)); | |
1827 | ||
1828 | /* | |
1829 | * Absent hardware errors or software bugs, this should | |
1830 | * be impossible, but log it anyway so we can debug it. | |
1831 | */ | |
1832 | if (error != 0) { | |
1833 | zfs_dbgmsg( | |
1834 | "hdr %p, compress %d, psize %d, lsize %d", | |
1835 | hdr, HDR_GET_COMPRESS(hdr), | |
1836 | HDR_GET_PSIZE(hdr), HDR_GET_LSIZE(hdr)); | |
1837 | return (SET_ERROR(EIO)); | |
1838 | } | |
1839 | } | |
1840 | } | |
1841 | ||
1842 | /* Byteswap the buf's data if necessary */ | |
1843 | if (bswap != DMU_BSWAP_NUMFUNCS) { | |
1844 | ASSERT(!HDR_SHARED_DATA(hdr)); | |
1845 | ASSERT3U(bswap, <, DMU_BSWAP_NUMFUNCS); | |
1846 | dmu_ot_byteswap[bswap].ob_func(buf->b_data, HDR_GET_LSIZE(hdr)); | |
1847 | } | |
1848 | ||
1849 | /* Compute the hdr's checksum if necessary */ | |
1850 | arc_cksum_compute(buf); | |
1851 | ||
1852 | return (0); | |
1853 | } | |
1854 | ||
1855 | int | |
1856 | arc_decompress(arc_buf_t *buf) | |
1857 | { | |
1858 | return (arc_buf_fill(buf, B_FALSE)); | |
1859 | } | |
1860 | ||
1861 | /* | |
1862 | * Return the size of the block, b_pabd, that is stored in the arc_buf_hdr_t. | |
1863 | */ | |
1864 | static uint64_t | |
1865 | arc_hdr_size(arc_buf_hdr_t *hdr) | |
1866 | { | |
1867 | uint64_t size; | |
1868 | ||
1869 | if (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF && | |
1870 | HDR_GET_PSIZE(hdr) > 0) { | |
1871 | size = HDR_GET_PSIZE(hdr); | |
1872 | } else { | |
1873 | ASSERT3U(HDR_GET_LSIZE(hdr), !=, 0); | |
1874 | size = HDR_GET_LSIZE(hdr); | |
1875 | } | |
1876 | return (size); | |
1877 | } | |
1878 | ||
1879 | /* | |
1880 | * Increment the amount of evictable space in the arc_state_t's refcount. | |
1881 | * We account for the space used by the hdr and the arc buf individually | |
1882 | * so that we can add and remove them from the refcount individually. | |
1883 | */ | |
1884 | static void | |
1885 | arc_evictable_space_increment(arc_buf_hdr_t *hdr, arc_state_t *state) | |
1886 | { | |
1887 | arc_buf_contents_t type = arc_buf_type(hdr); | |
1888 | arc_buf_t *buf; | |
1889 | ||
1890 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
1891 | ||
1892 | if (GHOST_STATE(state)) { | |
1893 | ASSERT0(hdr->b_l1hdr.b_bufcnt); | |
1894 | ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); | |
1895 | ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); | |
1896 | (void) refcount_add_many(&state->arcs_esize[type], | |
1897 | HDR_GET_LSIZE(hdr), hdr); | |
1898 | return; | |
1899 | } | |
1900 | ||
1901 | ASSERT(!GHOST_STATE(state)); | |
1902 | if (hdr->b_l1hdr.b_pabd != NULL) { | |
1903 | (void) refcount_add_many(&state->arcs_esize[type], | |
1904 | arc_hdr_size(hdr), hdr); | |
1905 | } | |
1906 | for (buf = hdr->b_l1hdr.b_buf; buf != NULL; buf = buf->b_next) { | |
1907 | if (arc_buf_is_shared(buf)) | |
1908 | continue; | |
1909 | (void) refcount_add_many(&state->arcs_esize[type], | |
1910 | arc_buf_size(buf), buf); | |
1911 | } | |
1912 | } | |
1913 | ||
1914 | /* | |
1915 | * Decrement the amount of evictable space in the arc_state_t's refcount. | |
1916 | * We account for the space used by the hdr and the arc buf individually | |
1917 | * so that we can add and remove them from the refcount individually. | |
1918 | */ | |
1919 | static void | |
1920 | arc_evictable_space_decrement(arc_buf_hdr_t *hdr, arc_state_t *state) | |
1921 | { | |
1922 | arc_buf_contents_t type = arc_buf_type(hdr); | |
1923 | arc_buf_t *buf; | |
1924 | ||
1925 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
1926 | ||
1927 | if (GHOST_STATE(state)) { | |
1928 | ASSERT0(hdr->b_l1hdr.b_bufcnt); | |
1929 | ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); | |
1930 | ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); | |
1931 | (void) refcount_remove_many(&state->arcs_esize[type], | |
1932 | HDR_GET_LSIZE(hdr), hdr); | |
1933 | return; | |
1934 | } | |
1935 | ||
1936 | ASSERT(!GHOST_STATE(state)); | |
1937 | if (hdr->b_l1hdr.b_pabd != NULL) { | |
1938 | (void) refcount_remove_many(&state->arcs_esize[type], | |
1939 | arc_hdr_size(hdr), hdr); | |
1940 | } | |
1941 | for (buf = hdr->b_l1hdr.b_buf; buf != NULL; buf = buf->b_next) { | |
1942 | if (arc_buf_is_shared(buf)) | |
1943 | continue; | |
1944 | (void) refcount_remove_many(&state->arcs_esize[type], | |
1945 | arc_buf_size(buf), buf); | |
1946 | } | |
1947 | } | |
1948 | ||
1949 | /* | |
1950 | * Add a reference to this hdr indicating that someone is actively | |
1951 | * referencing that memory. When the refcount transitions from 0 to 1, | |
1952 | * we remove it from the respective arc_state_t list to indicate that | |
1953 | * it is not evictable. | |
1954 | */ | |
1955 | static void | |
1956 | add_reference(arc_buf_hdr_t *hdr, void *tag) | |
1957 | { | |
1958 | arc_state_t *state; | |
1959 | ||
1960 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
1961 | if (!MUTEX_HELD(HDR_LOCK(hdr))) { | |
1962 | ASSERT(hdr->b_l1hdr.b_state == arc_anon); | |
1963 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); | |
1964 | ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); | |
1965 | } | |
1966 | ||
1967 | state = hdr->b_l1hdr.b_state; | |
1968 | ||
1969 | if ((refcount_add(&hdr->b_l1hdr.b_refcnt, tag) == 1) && | |
1970 | (state != arc_anon)) { | |
1971 | /* We don't use the L2-only state list. */ | |
1972 | if (state != arc_l2c_only) { | |
1973 | multilist_remove(state->arcs_list[arc_buf_type(hdr)], | |
1974 | hdr); | |
1975 | arc_evictable_space_decrement(hdr, state); | |
1976 | } | |
1977 | /* remove the prefetch flag if we get a reference */ | |
1978 | arc_hdr_clear_flags(hdr, ARC_FLAG_PREFETCH); | |
1979 | } | |
1980 | } | |
1981 | ||
1982 | /* | |
1983 | * Remove a reference from this hdr. When the reference transitions from | |
1984 | * 1 to 0 and we're not anonymous, then we add this hdr to the arc_state_t's | |
1985 | * list making it eligible for eviction. | |
1986 | */ | |
1987 | static int | |
1988 | remove_reference(arc_buf_hdr_t *hdr, kmutex_t *hash_lock, void *tag) | |
1989 | { | |
1990 | int cnt; | |
1991 | arc_state_t *state = hdr->b_l1hdr.b_state; | |
34dc7c2f | 1992 | |
e10b0808 | 1993 | ASSERT(HDR_HAS_L1HDR(hdr)); |
34dc7c2f BB |
1994 | ASSERT(state == arc_anon || MUTEX_HELD(hash_lock)); |
1995 | ASSERT(!GHOST_STATE(state)); | |
1996 | ||
e10b0808 AX |
1997 | /* |
1998 | * arc_l2c_only counts as a ghost state so we don't need to explicitly | |
1999 | * check to prevent usage of the arc_l2c_only list. | |
2000 | */ | |
2001 | if (((cnt = refcount_remove(&hdr->b_l1hdr.b_refcnt, tag)) == 0) && | |
34dc7c2f | 2002 | (state != arc_anon)) { |
cae5b340 AX |
2003 | multilist_insert(state->arcs_list[arc_buf_type(hdr)], hdr); |
2004 | ASSERT3U(hdr->b_l1hdr.b_bufcnt, >, 0); | |
2005 | arc_evictable_space_increment(hdr, state); | |
34dc7c2f BB |
2006 | } |
2007 | return (cnt); | |
2008 | } | |
2009 | ||
a08ee875 LG |
2010 | /* |
2011 | * Returns detailed information about a specific arc buffer. When the | |
2012 | * state_index argument is set the function will calculate the arc header | |
2013 | * list position for its arc state. Since this requires a linear traversal | |
2014 | * callers are strongly encourage not to do this. However, it can be helpful | |
2015 | * for targeted analysis so the functionality is provided. | |
2016 | */ | |
2017 | void | |
2018 | arc_buf_info(arc_buf_t *ab, arc_buf_info_t *abi, int state_index) | |
2019 | { | |
2020 | arc_buf_hdr_t *hdr = ab->b_hdr; | |
e10b0808 AX |
2021 | l1arc_buf_hdr_t *l1hdr = NULL; |
2022 | l2arc_buf_hdr_t *l2hdr = NULL; | |
2023 | arc_state_t *state = NULL; | |
2024 | ||
87dac73d AX |
2025 | memset(abi, 0, sizeof (arc_buf_info_t)); |
2026 | ||
2027 | if (hdr == NULL) | |
2028 | return; | |
2029 | ||
2030 | abi->abi_flags = hdr->b_flags; | |
2031 | ||
e10b0808 AX |
2032 | if (HDR_HAS_L1HDR(hdr)) { |
2033 | l1hdr = &hdr->b_l1hdr; | |
2034 | state = l1hdr->b_state; | |
2035 | } | |
2036 | if (HDR_HAS_L2HDR(hdr)) | |
2037 | l2hdr = &hdr->b_l2hdr; | |
a08ee875 | 2038 | |
e10b0808 | 2039 | if (l1hdr) { |
cae5b340 | 2040 | abi->abi_bufcnt = l1hdr->b_bufcnt; |
e10b0808 AX |
2041 | abi->abi_access = l1hdr->b_arc_access; |
2042 | abi->abi_mru_hits = l1hdr->b_mru_hits; | |
2043 | abi->abi_mru_ghost_hits = l1hdr->b_mru_ghost_hits; | |
2044 | abi->abi_mfu_hits = l1hdr->b_mfu_hits; | |
2045 | abi->abi_mfu_ghost_hits = l1hdr->b_mfu_ghost_hits; | |
2046 | abi->abi_holds = refcount_count(&l1hdr->b_refcnt); | |
2047 | } | |
2048 | ||
2049 | if (l2hdr) { | |
2050 | abi->abi_l2arc_dattr = l2hdr->b_daddr; | |
e10b0808 AX |
2051 | abi->abi_l2arc_hits = l2hdr->b_hits; |
2052 | } | |
2053 | ||
a08ee875 | 2054 | abi->abi_state_type = state ? state->arcs_state : ARC_STATE_ANON; |
e10b0808 | 2055 | abi->abi_state_contents = arc_buf_type(hdr); |
cae5b340 | 2056 | abi->abi_size = arc_hdr_size(hdr); |
a08ee875 LG |
2057 | } |
2058 | ||
34dc7c2f | 2059 | /* |
e10b0808 | 2060 | * Move the supplied buffer to the indicated state. The hash lock |
34dc7c2f BB |
2061 | * for the buffer must be held by the caller. |
2062 | */ | |
2063 | static void | |
e10b0808 AX |
2064 | arc_change_state(arc_state_t *new_state, arc_buf_hdr_t *hdr, |
2065 | kmutex_t *hash_lock) | |
34dc7c2f | 2066 | { |
e10b0808 AX |
2067 | arc_state_t *old_state; |
2068 | int64_t refcnt; | |
cae5b340 AX |
2069 | uint32_t bufcnt; |
2070 | boolean_t update_old, update_new; | |
e10b0808 AX |
2071 | arc_buf_contents_t buftype = arc_buf_type(hdr); |
2072 | ||
2073 | /* | |
2074 | * We almost always have an L1 hdr here, since we call arc_hdr_realloc() | |
2075 | * in arc_read() when bringing a buffer out of the L2ARC. However, the | |
2076 | * L1 hdr doesn't always exist when we change state to arc_anon before | |
2077 | * destroying a header, in which case reallocating to add the L1 hdr is | |
2078 | * pointless. | |
2079 | */ | |
2080 | if (HDR_HAS_L1HDR(hdr)) { | |
2081 | old_state = hdr->b_l1hdr.b_state; | |
2082 | refcnt = refcount_count(&hdr->b_l1hdr.b_refcnt); | |
cae5b340 AX |
2083 | bufcnt = hdr->b_l1hdr.b_bufcnt; |
2084 | update_old = (bufcnt > 0 || hdr->b_l1hdr.b_pabd != NULL); | |
e10b0808 AX |
2085 | } else { |
2086 | old_state = arc_l2c_only; | |
2087 | refcnt = 0; | |
cae5b340 AX |
2088 | bufcnt = 0; |
2089 | update_old = B_FALSE; | |
e10b0808 | 2090 | } |
cae5b340 | 2091 | update_new = update_old; |
34dc7c2f BB |
2092 | |
2093 | ASSERT(MUTEX_HELD(hash_lock)); | |
a08ee875 | 2094 | ASSERT3P(new_state, !=, old_state); |
cae5b340 AX |
2095 | ASSERT(!GHOST_STATE(new_state) || bufcnt == 0); |
2096 | ASSERT(old_state != arc_anon || bufcnt <= 1); | |
34dc7c2f BB |
2097 | |
2098 | /* | |
2099 | * If this buffer is evictable, transfer it from the | |
2100 | * old state list to the new state list. | |
2101 | */ | |
2102 | if (refcnt == 0) { | |
e10b0808 | 2103 | if (old_state != arc_anon && old_state != arc_l2c_only) { |
e10b0808 | 2104 | ASSERT(HDR_HAS_L1HDR(hdr)); |
cae5b340 | 2105 | multilist_remove(old_state->arcs_list[buftype], hdr); |
34dc7c2f | 2106 | |
cae5b340 AX |
2107 | if (GHOST_STATE(old_state)) { |
2108 | ASSERT0(bufcnt); | |
2109 | ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); | |
2110 | update_old = B_TRUE; | |
34dc7c2f | 2111 | } |
cae5b340 | 2112 | arc_evictable_space_decrement(hdr, old_state); |
34dc7c2f | 2113 | } |
e10b0808 | 2114 | if (new_state != arc_anon && new_state != arc_l2c_only) { |
e10b0808 AX |
2115 | /* |
2116 | * An L1 header always exists here, since if we're | |
2117 | * moving to some L1-cached state (i.e. not l2c_only or | |
2118 | * anonymous), we realloc the header to add an L1hdr | |
2119 | * beforehand. | |
2120 | */ | |
2121 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
cae5b340 | 2122 | multilist_insert(new_state->arcs_list[buftype], hdr); |
34dc7c2f | 2123 | |
34dc7c2f | 2124 | if (GHOST_STATE(new_state)) { |
cae5b340 AX |
2125 | ASSERT0(bufcnt); |
2126 | ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); | |
2127 | update_new = B_TRUE; | |
34dc7c2f | 2128 | } |
cae5b340 | 2129 | arc_evictable_space_increment(hdr, new_state); |
e10b0808 AX |
2130 | } |
2131 | } | |
2132 | ||
cae5b340 | 2133 | ASSERT(!HDR_EMPTY(hdr)); |
e10b0808 AX |
2134 | if (new_state == arc_anon && HDR_IN_HASH_TABLE(hdr)) |
2135 | buf_hash_remove(hdr); | |
2136 | ||
2137 | /* adjust state sizes (ignore arc_l2c_only) */ | |
2138 | ||
cae5b340 | 2139 | if (update_new && new_state != arc_l2c_only) { |
e10b0808 AX |
2140 | ASSERT(HDR_HAS_L1HDR(hdr)); |
2141 | if (GHOST_STATE(new_state)) { | |
cae5b340 | 2142 | ASSERT0(bufcnt); |
e10b0808 AX |
2143 | |
2144 | /* | |
cae5b340 | 2145 | * When moving a header to a ghost state, we first |
e10b0808 | 2146 | * remove all arc buffers. Thus, we'll have a |
cae5b340 | 2147 | * bufcnt of zero, and no arc buffer to use for |
e10b0808 AX |
2148 | * the reference. As a result, we use the arc |
2149 | * header pointer for the reference. | |
2150 | */ | |
2151 | (void) refcount_add_many(&new_state->arcs_size, | |
cae5b340 AX |
2152 | HDR_GET_LSIZE(hdr), hdr); |
2153 | ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); | |
e10b0808 AX |
2154 | } else { |
2155 | arc_buf_t *buf; | |
cae5b340 | 2156 | uint32_t buffers = 0; |
34dc7c2f | 2157 | |
e10b0808 AX |
2158 | /* |
2159 | * Each individual buffer holds a unique reference, | |
2160 | * thus we must remove each of these references one | |
2161 | * at a time. | |
2162 | */ | |
2163 | for (buf = hdr->b_l1hdr.b_buf; buf != NULL; | |
2164 | buf = buf->b_next) { | |
cae5b340 AX |
2165 | ASSERT3U(bufcnt, !=, 0); |
2166 | buffers++; | |
2167 | ||
2168 | /* | |
2169 | * When the arc_buf_t is sharing the data | |
2170 | * block with the hdr, the owner of the | |
2171 | * reference belongs to the hdr. Only | |
2172 | * add to the refcount if the arc_buf_t is | |
2173 | * not shared. | |
2174 | */ | |
2175 | if (arc_buf_is_shared(buf)) | |
2176 | continue; | |
2177 | ||
2178 | (void) refcount_add_many(&new_state->arcs_size, | |
2179 | arc_buf_size(buf), buf); | |
2180 | } | |
2181 | ASSERT3U(bufcnt, ==, buffers); | |
2182 | ||
2183 | if (hdr->b_l1hdr.b_pabd != NULL) { | |
e10b0808 | 2184 | (void) refcount_add_many(&new_state->arcs_size, |
cae5b340 AX |
2185 | arc_hdr_size(hdr), hdr); |
2186 | } else { | |
2187 | ASSERT(GHOST_STATE(old_state)); | |
e10b0808 | 2188 | } |
34dc7c2f BB |
2189 | } |
2190 | } | |
2191 | ||
cae5b340 | 2192 | if (update_old && old_state != arc_l2c_only) { |
e10b0808 AX |
2193 | ASSERT(HDR_HAS_L1HDR(hdr)); |
2194 | if (GHOST_STATE(old_state)) { | |
cae5b340 AX |
2195 | ASSERT0(bufcnt); |
2196 | ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); | |
2197 | ||
e10b0808 AX |
2198 | /* |
2199 | * When moving a header off of a ghost state, | |
cae5b340 AX |
2200 | * the header will not contain any arc buffers. |
2201 | * We use the arc header pointer for the reference | |
2202 | * which is exactly what we did when we put the | |
2203 | * header on the ghost state. | |
e10b0808 AX |
2204 | */ |
2205 | ||
e10b0808 | 2206 | (void) refcount_remove_many(&old_state->arcs_size, |
cae5b340 | 2207 | HDR_GET_LSIZE(hdr), hdr); |
e10b0808 AX |
2208 | } else { |
2209 | arc_buf_t *buf; | |
cae5b340 | 2210 | uint32_t buffers = 0; |
e10b0808 AX |
2211 | |
2212 | /* | |
2213 | * Each individual buffer holds a unique reference, | |
2214 | * thus we must remove each of these references one | |
2215 | * at a time. | |
2216 | */ | |
2217 | for (buf = hdr->b_l1hdr.b_buf; buf != NULL; | |
2218 | buf = buf->b_next) { | |
cae5b340 AX |
2219 | ASSERT3U(bufcnt, !=, 0); |
2220 | buffers++; | |
2221 | ||
2222 | /* | |
2223 | * When the arc_buf_t is sharing the data | |
2224 | * block with the hdr, the owner of the | |
2225 | * reference belongs to the hdr. Only | |
2226 | * add to the refcount if the arc_buf_t is | |
2227 | * not shared. | |
2228 | */ | |
2229 | if (arc_buf_is_shared(buf)) | |
2230 | continue; | |
2231 | ||
e10b0808 | 2232 | (void) refcount_remove_many( |
cae5b340 AX |
2233 | &old_state->arcs_size, arc_buf_size(buf), |
2234 | buf); | |
e10b0808 | 2235 | } |
cae5b340 AX |
2236 | ASSERT3U(bufcnt, ==, buffers); |
2237 | ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); | |
2238 | (void) refcount_remove_many( | |
2239 | &old_state->arcs_size, arc_hdr_size(hdr), hdr); | |
e10b0808 | 2240 | } |
34dc7c2f | 2241 | } |
34dc7c2f | 2242 | |
e10b0808 AX |
2243 | if (HDR_HAS_L1HDR(hdr)) |
2244 | hdr->b_l1hdr.b_state = new_state; | |
2245 | ||
2246 | /* | |
2247 | * L2 headers should never be on the L2 state list since they don't | |
2248 | * have L1 headers allocated. | |
2249 | */ | |
cae5b340 AX |
2250 | ASSERT(multilist_is_empty(arc_l2c_only->arcs_list[ARC_BUFC_DATA]) && |
2251 | multilist_is_empty(arc_l2c_only->arcs_list[ARC_BUFC_METADATA])); | |
34dc7c2f BB |
2252 | } |
2253 | ||
2254 | void | |
d164b209 | 2255 | arc_space_consume(uint64_t space, arc_space_type_t type) |
34dc7c2f | 2256 | { |
d164b209 BB |
2257 | ASSERT(type >= 0 && type < ARC_SPACE_NUMTYPES); |
2258 | ||
2259 | switch (type) { | |
e75c13c3 BB |
2260 | default: |
2261 | break; | |
d164b209 BB |
2262 | case ARC_SPACE_DATA: |
2263 | ARCSTAT_INCR(arcstat_data_size, space); | |
2264 | break; | |
ea04106b | 2265 | case ARC_SPACE_META: |
e10b0808 | 2266 | ARCSTAT_INCR(arcstat_metadata_size, space); |
ea04106b | 2267 | break; |
cae5b340 AX |
2268 | case ARC_SPACE_BONUS: |
2269 | ARCSTAT_INCR(arcstat_bonus_size, space); | |
2270 | break; | |
2271 | case ARC_SPACE_DNODE: | |
2272 | ARCSTAT_INCR(arcstat_dnode_size, space); | |
2273 | break; | |
2274 | case ARC_SPACE_DBUF: | |
2275 | ARCSTAT_INCR(arcstat_dbuf_size, space); | |
d164b209 BB |
2276 | break; |
2277 | case ARC_SPACE_HDRS: | |
2278 | ARCSTAT_INCR(arcstat_hdr_size, space); | |
2279 | break; | |
2280 | case ARC_SPACE_L2HDRS: | |
2281 | ARCSTAT_INCR(arcstat_l2_hdr_size, space); | |
2282 | break; | |
2283 | } | |
2284 | ||
e10b0808 | 2285 | if (type != ARC_SPACE_DATA) |
ea04106b | 2286 | ARCSTAT_INCR(arcstat_meta_used, space); |
ea04106b | 2287 | |
34dc7c2f BB |
2288 | atomic_add_64(&arc_size, space); |
2289 | } | |
2290 | ||
2291 | void | |
d164b209 | 2292 | arc_space_return(uint64_t space, arc_space_type_t type) |
34dc7c2f | 2293 | { |
d164b209 BB |
2294 | ASSERT(type >= 0 && type < ARC_SPACE_NUMTYPES); |
2295 | ||
2296 | switch (type) { | |
e75c13c3 BB |
2297 | default: |
2298 | break; | |
d164b209 BB |
2299 | case ARC_SPACE_DATA: |
2300 | ARCSTAT_INCR(arcstat_data_size, -space); | |
2301 | break; | |
ea04106b | 2302 | case ARC_SPACE_META: |
e10b0808 | 2303 | ARCSTAT_INCR(arcstat_metadata_size, -space); |
ea04106b | 2304 | break; |
cae5b340 AX |
2305 | case ARC_SPACE_BONUS: |
2306 | ARCSTAT_INCR(arcstat_bonus_size, -space); | |
2307 | break; | |
2308 | case ARC_SPACE_DNODE: | |
2309 | ARCSTAT_INCR(arcstat_dnode_size, -space); | |
2310 | break; | |
2311 | case ARC_SPACE_DBUF: | |
2312 | ARCSTAT_INCR(arcstat_dbuf_size, -space); | |
d164b209 BB |
2313 | break; |
2314 | case ARC_SPACE_HDRS: | |
2315 | ARCSTAT_INCR(arcstat_hdr_size, -space); | |
2316 | break; | |
2317 | case ARC_SPACE_L2HDRS: | |
2318 | ARCSTAT_INCR(arcstat_l2_hdr_size, -space); | |
2319 | break; | |
2320 | } | |
2321 | ||
ea04106b AX |
2322 | if (type != ARC_SPACE_DATA) { |
2323 | ASSERT(arc_meta_used >= space); | |
e10b0808 AX |
2324 | if (arc_meta_max < arc_meta_used) |
2325 | arc_meta_max = arc_meta_used; | |
ea04106b AX |
2326 | ARCSTAT_INCR(arcstat_meta_used, -space); |
2327 | } | |
2328 | ||
34dc7c2f BB |
2329 | ASSERT(arc_size >= space); |
2330 | atomic_add_64(&arc_size, -space); | |
2331 | } | |
2332 | ||
cae5b340 AX |
2333 | /* |
2334 | * Given a hdr and a buf, returns whether that buf can share its b_data buffer | |
2335 | * with the hdr's b_pabd. | |
2336 | */ | |
2337 | static boolean_t | |
2338 | arc_can_share(arc_buf_hdr_t *hdr, arc_buf_t *buf) | |
2339 | { | |
2340 | /* | |
2341 | * The criteria for sharing a hdr's data are: | |
2342 | * 1. the hdr's compression matches the buf's compression | |
2343 | * 2. the hdr doesn't need to be byteswapped | |
2344 | * 3. the hdr isn't already being shared | |
2345 | * 4. the buf is either compressed or it is the last buf in the hdr list | |
2346 | * | |
2347 | * Criterion #4 maintains the invariant that shared uncompressed | |
2348 | * bufs must be the final buf in the hdr's b_buf list. Reading this, you | |
2349 | * might ask, "if a compressed buf is allocated first, won't that be the | |
2350 | * last thing in the list?", but in that case it's impossible to create | |
2351 | * a shared uncompressed buf anyway (because the hdr must be compressed | |
2352 | * to have the compressed buf). You might also think that #3 is | |
2353 | * sufficient to make this guarantee, however it's possible | |
2354 | * (specifically in the rare L2ARC write race mentioned in | |
2355 | * arc_buf_alloc_impl()) there will be an existing uncompressed buf that | |
2356 | * is sharable, but wasn't at the time of its allocation. Rather than | |
2357 | * allow a new shared uncompressed buf to be created and then shuffle | |
2358 | * the list around to make it the last element, this simply disallows | |
2359 | * sharing if the new buf isn't the first to be added. | |
2360 | */ | |
2361 | ASSERT3P(buf->b_hdr, ==, hdr); | |
2362 | boolean_t hdr_compressed = HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF; | |
2363 | boolean_t buf_compressed = ARC_BUF_COMPRESSED(buf) != 0; | |
2364 | return (buf_compressed == hdr_compressed && | |
2365 | hdr->b_l1hdr.b_byteswap == DMU_BSWAP_NUMFUNCS && | |
2366 | !HDR_SHARED_DATA(hdr) && | |
2367 | (ARC_BUF_LAST(buf) || ARC_BUF_COMPRESSED(buf))); | |
2368 | } | |
2369 | ||
2370 | /* | |
2371 | * Allocate a buf for this hdr. If you care about the data that's in the hdr, | |
2372 | * or if you want a compressed buffer, pass those flags in. Returns 0 if the | |
2373 | * copy was made successfully, or an error code otherwise. | |
2374 | */ | |
2375 | static int | |
2376 | arc_buf_alloc_impl(arc_buf_hdr_t *hdr, void *tag, boolean_t compressed, | |
2377 | boolean_t fill, arc_buf_t **ret) | |
34dc7c2f | 2378 | { |
34dc7c2f BB |
2379 | arc_buf_t *buf; |
2380 | ||
cae5b340 AX |
2381 | ASSERT(HDR_HAS_L1HDR(hdr)); |
2382 | ASSERT3U(HDR_GET_LSIZE(hdr), >, 0); | |
2383 | VERIFY(hdr->b_type == ARC_BUFC_DATA || | |
2384 | hdr->b_type == ARC_BUFC_METADATA); | |
2385 | ASSERT3P(ret, !=, NULL); | |
2386 | ASSERT3P(*ret, ==, NULL); | |
2387 | ||
e10b0808 AX |
2388 | hdr->b_l1hdr.b_mru_hits = 0; |
2389 | hdr->b_l1hdr.b_mru_ghost_hits = 0; | |
2390 | hdr->b_l1hdr.b_mfu_hits = 0; | |
2391 | hdr->b_l1hdr.b_mfu_ghost_hits = 0; | |
2392 | hdr->b_l1hdr.b_l2_hits = 0; | |
2393 | ||
cae5b340 | 2394 | buf = *ret = kmem_cache_alloc(buf_cache, KM_PUSHPAGE); |
34dc7c2f BB |
2395 | buf->b_hdr = hdr; |
2396 | buf->b_data = NULL; | |
cae5b340 AX |
2397 | buf->b_next = hdr->b_l1hdr.b_buf; |
2398 | buf->b_flags = 0; | |
e10b0808 | 2399 | |
cae5b340 AX |
2400 | add_reference(hdr, tag); |
2401 | ||
2402 | /* | |
2403 | * We're about to change the hdr's b_flags. We must either | |
2404 | * hold the hash_lock or be undiscoverable. | |
2405 | */ | |
2406 | ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); | |
2407 | ||
2408 | /* | |
2409 | * Only honor requests for compressed bufs if the hdr is actually | |
2410 | * compressed. | |
2411 | */ | |
2412 | if (compressed && HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF) | |
2413 | buf->b_flags |= ARC_BUF_FLAG_COMPRESSED; | |
2414 | ||
2415 | /* | |
2416 | * If the hdr's data can be shared then we share the data buffer and | |
2417 | * set the appropriate bit in the hdr's b_flags to indicate the hdr is | |
2418 | * allocate a new buffer to store the buf's data. | |
2419 | * | |
2420 | * There are two additional restrictions here because we're sharing | |
2421 | * hdr -> buf instead of the usual buf -> hdr. First, the hdr can't be | |
2422 | * actively involved in an L2ARC write, because if this buf is used by | |
2423 | * an arc_write() then the hdr's data buffer will be released when the | |
2424 | * write completes, even though the L2ARC write might still be using it. | |
2425 | * Second, the hdr's ABD must be linear so that the buf's user doesn't | |
2426 | * need to be ABD-aware. | |
2427 | */ | |
2428 | boolean_t can_share = arc_can_share(hdr, buf) && !HDR_L2_WRITING(hdr) && | |
2429 | abd_is_linear(hdr->b_l1hdr.b_pabd); | |
2430 | ||
2431 | /* Set up b_data and sharing */ | |
2432 | if (can_share) { | |
2433 | buf->b_data = abd_to_buf(hdr->b_l1hdr.b_pabd); | |
2434 | buf->b_flags |= ARC_BUF_FLAG_SHARED; | |
2435 | arc_hdr_set_flags(hdr, ARC_FLAG_SHARED_DATA); | |
2436 | } else { | |
2437 | buf->b_data = | |
2438 | arc_get_data_buf(hdr, arc_buf_size(buf), buf); | |
2439 | ARCSTAT_INCR(arcstat_overhead_size, arc_buf_size(buf)); | |
2440 | } | |
2441 | VERIFY3P(buf->b_data, !=, NULL); | |
e10b0808 AX |
2442 | |
2443 | hdr->b_l1hdr.b_buf = buf; | |
cae5b340 | 2444 | hdr->b_l1hdr.b_bufcnt += 1; |
e10b0808 | 2445 | |
cae5b340 AX |
2446 | /* |
2447 | * If the user wants the data from the hdr, we need to either copy or | |
2448 | * decompress the data. | |
2449 | */ | |
2450 | if (fill) { | |
2451 | return (arc_buf_fill(buf, ARC_BUF_COMPRESSED(buf) != 0)); | |
2452 | } | |
34dc7c2f | 2453 | |
cae5b340 | 2454 | return (0); |
34dc7c2f BB |
2455 | } |
2456 | ||
9babb374 BB |
2457 | static char *arc_onloan_tag = "onloan"; |
2458 | ||
cae5b340 AX |
2459 | static inline void |
2460 | arc_loaned_bytes_update(int64_t delta) | |
2461 | { | |
2462 | atomic_add_64(&arc_loaned_bytes, delta); | |
2463 | ||
2464 | /* assert that it did not wrap around */ | |
2465 | ASSERT3S(atomic_add_64_nv(&arc_loaned_bytes, 0), >=, 0); | |
2466 | } | |
2467 | ||
9babb374 BB |
2468 | /* |
2469 | * Loan out an anonymous arc buffer. Loaned buffers are not counted as in | |
2470 | * flight data by arc_tempreserve_space() until they are "returned". Loaned | |
2471 | * buffers must be returned to the arc before they can be used by the DMU or | |
2472 | * freed. | |
2473 | */ | |
2474 | arc_buf_t * | |
cae5b340 | 2475 | arc_loan_buf(spa_t *spa, boolean_t is_metadata, int size) |
9babb374 | 2476 | { |
cae5b340 AX |
2477 | arc_buf_t *buf = arc_alloc_buf(spa, arc_onloan_tag, |
2478 | is_metadata ? ARC_BUFC_METADATA : ARC_BUFC_DATA, size); | |
9babb374 | 2479 | |
cae5b340 | 2480 | arc_loaned_bytes_update(size); |
9babb374 | 2481 | |
9babb374 BB |
2482 | return (buf); |
2483 | } | |
2484 | ||
cae5b340 AX |
2485 | arc_buf_t * |
2486 | arc_loan_compressed_buf(spa_t *spa, uint64_t psize, uint64_t lsize, | |
2487 | enum zio_compress compression_type) | |
2488 | { | |
2489 | arc_buf_t *buf = arc_alloc_compressed_buf(spa, arc_onloan_tag, | |
2490 | psize, lsize, compression_type); | |
2491 | ||
2492 | arc_loaned_bytes_update(psize); | |
2493 | ||
2494 | return (buf); | |
2495 | } | |
2496 | ||
2497 | ||
9babb374 BB |
2498 | /* |
2499 | * Return a loaned arc buffer to the arc. | |
2500 | */ | |
2501 | void | |
2502 | arc_return_buf(arc_buf_t *buf, void *tag) | |
2503 | { | |
2504 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
2505 | ||
cae5b340 | 2506 | ASSERT3P(buf->b_data, !=, NULL); |
e10b0808 AX |
2507 | ASSERT(HDR_HAS_L1HDR(hdr)); |
2508 | (void) refcount_add(&hdr->b_l1hdr.b_refcnt, tag); | |
2509 | (void) refcount_remove(&hdr->b_l1hdr.b_refcnt, arc_onloan_tag); | |
9babb374 | 2510 | |
cae5b340 | 2511 | arc_loaned_bytes_update(-arc_buf_size(buf)); |
9babb374 BB |
2512 | } |
2513 | ||
428870ff BB |
2514 | /* Detach an arc_buf from a dbuf (tag) */ |
2515 | void | |
2516 | arc_loan_inuse_buf(arc_buf_t *buf, void *tag) | |
2517 | { | |
e10b0808 | 2518 | arc_buf_hdr_t *hdr = buf->b_hdr; |
428870ff | 2519 | |
cae5b340 | 2520 | ASSERT3P(buf->b_data, !=, NULL); |
e10b0808 AX |
2521 | ASSERT(HDR_HAS_L1HDR(hdr)); |
2522 | (void) refcount_add(&hdr->b_l1hdr.b_refcnt, arc_onloan_tag); | |
2523 | (void) refcount_remove(&hdr->b_l1hdr.b_refcnt, tag); | |
428870ff | 2524 | |
cae5b340 | 2525 | arc_loaned_bytes_update(arc_buf_size(buf)); |
428870ff BB |
2526 | } |
2527 | ||
cae5b340 AX |
2528 | static void |
2529 | l2arc_free_abd_on_write(abd_t *abd, size_t size, arc_buf_contents_t type) | |
34dc7c2f | 2530 | { |
cae5b340 AX |
2531 | l2arc_data_free_t *df = kmem_alloc(sizeof (*df), KM_SLEEP); |
2532 | ||
2533 | df->l2df_abd = abd; | |
2534 | df->l2df_size = size; | |
2535 | df->l2df_type = type; | |
2536 | mutex_enter(&l2arc_free_on_write_mtx); | |
2537 | list_insert_head(l2arc_free_on_write, df); | |
2538 | mutex_exit(&l2arc_free_on_write_mtx); | |
2539 | } | |
2540 | ||
2541 | static void | |
2542 | arc_hdr_free_on_write(arc_buf_hdr_t *hdr) | |
2543 | { | |
2544 | arc_state_t *state = hdr->b_l1hdr.b_state; | |
2545 | arc_buf_contents_t type = arc_buf_type(hdr); | |
2546 | uint64_t size = arc_hdr_size(hdr); | |
2547 | ||
2548 | /* protected by hash lock, if in the hash table */ | |
2549 | if (multilist_link_active(&hdr->b_l1hdr.b_arc_node)) { | |
2550 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); | |
2551 | ASSERT(state != arc_anon && state != arc_l2c_only); | |
2552 | ||
2553 | (void) refcount_remove_many(&state->arcs_esize[type], | |
2554 | size, hdr); | |
2555 | } | |
2556 | (void) refcount_remove_many(&state->arcs_size, size, hdr); | |
2557 | if (type == ARC_BUFC_METADATA) { | |
2558 | arc_space_return(size, ARC_SPACE_META); | |
2559 | } else { | |
2560 | ASSERT(type == ARC_BUFC_DATA); | |
2561 | arc_space_return(size, ARC_SPACE_DATA); | |
2562 | } | |
2563 | ||
2564 | l2arc_free_abd_on_write(hdr->b_l1hdr.b_pabd, size, type); | |
2565 | } | |
2566 | ||
2567 | /* | |
2568 | * Share the arc_buf_t's data with the hdr. Whenever we are sharing the | |
2569 | * data buffer, we transfer the refcount ownership to the hdr and update | |
2570 | * the appropriate kstats. | |
2571 | */ | |
2572 | static void | |
2573 | arc_share_buf(arc_buf_hdr_t *hdr, arc_buf_t *buf) | |
2574 | { | |
2575 | ASSERT(arc_can_share(hdr, buf)); | |
2576 | ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); | |
2577 | ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); | |
2578 | ||
2579 | /* | |
2580 | * Start sharing the data buffer. We transfer the | |
2581 | * refcount ownership to the hdr since it always owns | |
2582 | * the refcount whenever an arc_buf_t is shared. | |
2583 | */ | |
2584 | refcount_transfer_ownership(&hdr->b_l1hdr.b_state->arcs_size, buf, hdr); | |
2585 | hdr->b_l1hdr.b_pabd = abd_get_from_buf(buf->b_data, arc_buf_size(buf)); | |
2586 | abd_take_ownership_of_buf(hdr->b_l1hdr.b_pabd, | |
2587 | HDR_ISTYPE_METADATA(hdr)); | |
2588 | arc_hdr_set_flags(hdr, ARC_FLAG_SHARED_DATA); | |
2589 | buf->b_flags |= ARC_BUF_FLAG_SHARED; | |
2590 | ||
2591 | /* | |
2592 | * Since we've transferred ownership to the hdr we need | |
2593 | * to increment its compressed and uncompressed kstats and | |
2594 | * decrement the overhead size. | |
2595 | */ | |
2596 | ARCSTAT_INCR(arcstat_compressed_size, arc_hdr_size(hdr)); | |
2597 | ARCSTAT_INCR(arcstat_uncompressed_size, HDR_GET_LSIZE(hdr)); | |
2598 | ARCSTAT_INCR(arcstat_overhead_size, -arc_buf_size(buf)); | |
2599 | } | |
2600 | ||
2601 | static void | |
2602 | arc_unshare_buf(arc_buf_hdr_t *hdr, arc_buf_t *buf) | |
2603 | { | |
2604 | ASSERT(arc_buf_is_shared(buf)); | |
2605 | ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); | |
2606 | ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); | |
2607 | ||
2608 | /* | |
2609 | * We are no longer sharing this buffer so we need | |
2610 | * to transfer its ownership to the rightful owner. | |
2611 | */ | |
2612 | refcount_transfer_ownership(&hdr->b_l1hdr.b_state->arcs_size, hdr, buf); | |
2613 | arc_hdr_clear_flags(hdr, ARC_FLAG_SHARED_DATA); | |
2614 | abd_release_ownership_of_buf(hdr->b_l1hdr.b_pabd); | |
2615 | abd_put(hdr->b_l1hdr.b_pabd); | |
2616 | hdr->b_l1hdr.b_pabd = NULL; | |
2617 | buf->b_flags &= ~ARC_BUF_FLAG_SHARED; | |
34dc7c2f | 2618 | |
cae5b340 AX |
2619 | /* |
2620 | * Since the buffer is no longer shared between | |
2621 | * the arc buf and the hdr, count it as overhead. | |
2622 | */ | |
2623 | ARCSTAT_INCR(arcstat_compressed_size, -arc_hdr_size(hdr)); | |
2624 | ARCSTAT_INCR(arcstat_uncompressed_size, -HDR_GET_LSIZE(hdr)); | |
2625 | ARCSTAT_INCR(arcstat_overhead_size, arc_buf_size(buf)); | |
2626 | } | |
2627 | ||
2628 | /* | |
2629 | * Remove an arc_buf_t from the hdr's buf list and return the last | |
2630 | * arc_buf_t on the list. If no buffers remain on the list then return | |
2631 | * NULL. | |
2632 | */ | |
2633 | static arc_buf_t * | |
2634 | arc_buf_remove(arc_buf_hdr_t *hdr, arc_buf_t *buf) | |
2635 | { | |
e10b0808 | 2636 | ASSERT(HDR_HAS_L1HDR(hdr)); |
cae5b340 | 2637 | ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); |
428870ff | 2638 | |
cae5b340 AX |
2639 | arc_buf_t **bufp = &hdr->b_l1hdr.b_buf; |
2640 | arc_buf_t *lastbuf = NULL; | |
1eb5bfa3 GW |
2641 | |
2642 | /* | |
cae5b340 AX |
2643 | * Remove the buf from the hdr list and locate the last |
2644 | * remaining buffer on the list. | |
1eb5bfa3 | 2645 | */ |
cae5b340 AX |
2646 | while (*bufp != NULL) { |
2647 | if (*bufp == buf) | |
2648 | *bufp = buf->b_next; | |
2649 | ||
2650 | /* | |
2651 | * If we've removed a buffer in the middle of | |
2652 | * the list then update the lastbuf and update | |
2653 | * bufp. | |
2654 | */ | |
2655 | if (*bufp != NULL) { | |
2656 | lastbuf = *bufp; | |
2657 | bufp = &(*bufp)->b_next; | |
2658 | } | |
1eb5bfa3 | 2659 | } |
cae5b340 AX |
2660 | buf->b_next = NULL; |
2661 | ASSERT3P(lastbuf, !=, buf); | |
2662 | IMPLY(hdr->b_l1hdr.b_bufcnt > 0, lastbuf != NULL); | |
2663 | IMPLY(hdr->b_l1hdr.b_bufcnt > 0, hdr->b_l1hdr.b_buf != NULL); | |
2664 | IMPLY(lastbuf != NULL, ARC_BUF_LAST(lastbuf)); | |
2665 | ||
2666 | return (lastbuf); | |
34dc7c2f BB |
2667 | } |
2668 | ||
cae5b340 AX |
2669 | /* |
2670 | * Free up buf->b_data and pull the arc_buf_t off of the the arc_buf_hdr_t's | |
2671 | * list and free it. | |
2672 | */ | |
2673 | static void | |
2674 | arc_buf_destroy_impl(arc_buf_t *buf) | |
34dc7c2f | 2675 | { |
cae5b340 | 2676 | arc_buf_hdr_t *hdr = buf->b_hdr; |
34dc7c2f BB |
2677 | |
2678 | /* | |
cae5b340 AX |
2679 | * Free up the data associated with the buf but only if we're not |
2680 | * sharing this with the hdr. If we are sharing it with the hdr, the | |
2681 | * hdr is responsible for doing the free. | |
34dc7c2f | 2682 | */ |
cae5b340 AX |
2683 | if (buf->b_data != NULL) { |
2684 | /* | |
2685 | * We're about to change the hdr's b_flags. We must either | |
2686 | * hold the hash_lock or be undiscoverable. | |
2687 | */ | |
2688 | ASSERT(MUTEX_HELD(HDR_LOCK(hdr)) || HDR_EMPTY(hdr)); | |
2689 | ||
2690 | arc_cksum_verify(buf); | |
2691 | arc_buf_unwatch(buf); | |
2692 | ||
2693 | if (arc_buf_is_shared(buf)) { | |
2694 | arc_hdr_clear_flags(hdr, ARC_FLAG_SHARED_DATA); | |
2695 | } else { | |
2696 | uint64_t size = arc_buf_size(buf); | |
2697 | arc_free_data_buf(hdr, buf->b_data, size, buf); | |
2698 | ARCSTAT_INCR(arcstat_overhead_size, -size); | |
2699 | } | |
2700 | buf->b_data = NULL; | |
2701 | ||
2702 | ASSERT(hdr->b_l1hdr.b_bufcnt > 0); | |
2703 | hdr->b_l1hdr.b_bufcnt -= 1; | |
34dc7c2f | 2704 | } |
34dc7c2f | 2705 | |
cae5b340 | 2706 | arc_buf_t *lastbuf = arc_buf_remove(hdr, buf); |
e10b0808 | 2707 | |
cae5b340 AX |
2708 | if (ARC_BUF_SHARED(buf) && !ARC_BUF_COMPRESSED(buf)) { |
2709 | /* | |
2710 | * If the current arc_buf_t is sharing its data buffer with the | |
2711 | * hdr, then reassign the hdr's b_pabd to share it with the new | |
2712 | * buffer at the end of the list. The shared buffer is always | |
2713 | * the last one on the hdr's buffer list. | |
2714 | * | |
2715 | * There is an equivalent case for compressed bufs, but since | |
2716 | * they aren't guaranteed to be the last buf in the list and | |
2717 | * that is an exceedingly rare case, we just allow that space be | |
2718 | * wasted temporarily. | |
2719 | */ | |
2720 | if (lastbuf != NULL) { | |
2721 | /* Only one buf can be shared at once */ | |
2722 | VERIFY(!arc_buf_is_shared(lastbuf)); | |
2723 | /* hdr is uncompressed so can't have compressed buf */ | |
2724 | VERIFY(!ARC_BUF_COMPRESSED(lastbuf)); | |
2725 | ||
2726 | ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); | |
2727 | arc_hdr_free_pabd(hdr); | |
2728 | ||
2729 | /* | |
2730 | * We must setup a new shared block between the | |
2731 | * last buffer and the hdr. The data would have | |
2732 | * been allocated by the arc buf so we need to transfer | |
2733 | * ownership to the hdr since it's now being shared. | |
2734 | */ | |
2735 | arc_share_buf(hdr, lastbuf); | |
2736 | } | |
2737 | } else if (HDR_SHARED_DATA(hdr)) { | |
2738 | /* | |
2739 | * Uncompressed shared buffers are always at the end | |
2740 | * of the list. Compressed buffers don't have the | |
2741 | * same requirements. This makes it hard to | |
2742 | * simply assert that the lastbuf is shared so | |
2743 | * we rely on the hdr's compression flags to determine | |
2744 | * if we have a compressed, shared buffer. | |
2745 | */ | |
2746 | ASSERT3P(lastbuf, !=, NULL); | |
2747 | ASSERT(arc_buf_is_shared(lastbuf) || | |
2748 | HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF); | |
2749 | } | |
2750 | ||
2751 | /* | |
2752 | * Free the checksum if we're removing the last uncompressed buf from | |
2753 | * this hdr. | |
2754 | */ | |
2755 | if (!arc_hdr_has_uncompressed_buf(hdr)) { | |
2756 | arc_cksum_free(hdr); | |
2757 | } | |
2758 | ||
2759 | /* clean up the buf */ | |
2760 | buf->b_hdr = NULL; | |
2761 | kmem_cache_free(buf_cache, buf); | |
34dc7c2f BB |
2762 | } |
2763 | ||
ea04106b | 2764 | static void |
cae5b340 | 2765 | arc_hdr_alloc_pabd(arc_buf_hdr_t *hdr) |
ea04106b | 2766 | { |
cae5b340 AX |
2767 | ASSERT3U(HDR_GET_LSIZE(hdr), >, 0); |
2768 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
2769 | ASSERT(!HDR_SHARED_DATA(hdr)); | |
ea04106b | 2770 | |
cae5b340 AX |
2771 | ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); |
2772 | hdr->b_l1hdr.b_pabd = arc_get_data_abd(hdr, arc_hdr_size(hdr), hdr); | |
2773 | hdr->b_l1hdr.b_byteswap = DMU_BSWAP_NUMFUNCS; | |
2774 | ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); | |
2775 | ||
2776 | ARCSTAT_INCR(arcstat_compressed_size, arc_hdr_size(hdr)); | |
2777 | ARCSTAT_INCR(arcstat_uncompressed_size, HDR_GET_LSIZE(hdr)); | |
ea04106b AX |
2778 | } |
2779 | ||
34dc7c2f | 2780 | static void |
cae5b340 | 2781 | arc_hdr_free_pabd(arc_buf_hdr_t *hdr) |
34dc7c2f | 2782 | { |
cae5b340 AX |
2783 | ASSERT(HDR_HAS_L1HDR(hdr)); |
2784 | ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); | |
a08ee875 | 2785 | |
cae5b340 AX |
2786 | /* |
2787 | * If the hdr is currently being written to the l2arc then | |
2788 | * we defer freeing the data by adding it to the l2arc_free_on_write | |
2789 | * list. The l2arc will free the data once it's finished | |
2790 | * writing it to the l2arc device. | |
2791 | */ | |
34dc7c2f | 2792 | if (HDR_L2_WRITING(hdr)) { |
cae5b340 | 2793 | arc_hdr_free_on_write(hdr); |
34dc7c2f BB |
2794 | ARCSTAT_BUMP(arcstat_l2_free_on_write); |
2795 | } else { | |
cae5b340 AX |
2796 | arc_free_data_abd(hdr, hdr->b_l1hdr.b_pabd, |
2797 | arc_hdr_size(hdr), hdr); | |
34dc7c2f | 2798 | } |
cae5b340 AX |
2799 | hdr->b_l1hdr.b_pabd = NULL; |
2800 | hdr->b_l1hdr.b_byteswap = DMU_BSWAP_NUMFUNCS; | |
2801 | ||
2802 | ARCSTAT_INCR(arcstat_compressed_size, -arc_hdr_size(hdr)); | |
2803 | ARCSTAT_INCR(arcstat_uncompressed_size, -HDR_GET_LSIZE(hdr)); | |
34dc7c2f BB |
2804 | } |
2805 | ||
cae5b340 AX |
2806 | static arc_buf_hdr_t * |
2807 | arc_hdr_alloc(uint64_t spa, int32_t psize, int32_t lsize, | |
2808 | enum zio_compress compression_type, arc_buf_contents_t type) | |
ea04106b | 2809 | { |
cae5b340 AX |
2810 | arc_buf_hdr_t *hdr; |
2811 | ||
2812 | VERIFY(type == ARC_BUFC_DATA || type == ARC_BUFC_METADATA); | |
2813 | ||
2814 | hdr = kmem_cache_alloc(hdr_full_cache, KM_PUSHPAGE); | |
2815 | ASSERT(HDR_EMPTY(hdr)); | |
2816 | ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL); | |
2817 | HDR_SET_PSIZE(hdr, psize); | |
2818 | HDR_SET_LSIZE(hdr, lsize); | |
2819 | hdr->b_spa = spa; | |
2820 | hdr->b_type = type; | |
2821 | hdr->b_flags = 0; | |
2822 | arc_hdr_set_flags(hdr, arc_bufc_to_flags(type) | ARC_FLAG_HAS_L1HDR); | |
2823 | arc_hdr_set_compress(hdr, compression_type); | |
2824 | ||
2825 | hdr->b_l1hdr.b_state = arc_anon; | |
2826 | hdr->b_l1hdr.b_arc_access = 0; | |
2827 | hdr->b_l1hdr.b_bufcnt = 0; | |
2828 | hdr->b_l1hdr.b_buf = NULL; | |
e10b0808 AX |
2829 | |
2830 | /* | |
cae5b340 AX |
2831 | * Allocate the hdr's buffer. This will contain either |
2832 | * the compressed or uncompressed data depending on the block | |
2833 | * it references and compressed arc enablement. | |
e10b0808 | 2834 | */ |
cae5b340 AX |
2835 | arc_hdr_alloc_pabd(hdr); |
2836 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); | |
2837 | ||
2838 | return (hdr); | |
2839 | } | |
2840 | ||
2841 | /* | |
2842 | * Transition between the two allocation states for the arc_buf_hdr struct. | |
2843 | * The arc_buf_hdr struct can be allocated with (hdr_full_cache) or without | |
2844 | * (hdr_l2only_cache) the fields necessary for the L1 cache - the smaller | |
2845 | * version is used when a cache buffer is only in the L2ARC in order to reduce | |
2846 | * memory usage. | |
2847 | */ | |
2848 | static arc_buf_hdr_t * | |
2849 | arc_hdr_realloc(arc_buf_hdr_t *hdr, kmem_cache_t *old, kmem_cache_t *new) | |
2850 | { | |
2851 | arc_buf_hdr_t *nhdr; | |
2852 | l2arc_dev_t *dev = hdr->b_l2hdr.b_dev; | |
2853 | ||
2854 | ASSERT(HDR_HAS_L2HDR(hdr)); | |
2855 | ASSERT((old == hdr_full_cache && new == hdr_l2only_cache) || | |
2856 | (old == hdr_l2only_cache && new == hdr_full_cache)); | |
2857 | ||
2858 | nhdr = kmem_cache_alloc(new, KM_PUSHPAGE); | |
2859 | ||
2860 | ASSERT(MUTEX_HELD(HDR_LOCK(hdr))); | |
2861 | buf_hash_remove(hdr); | |
2862 | ||
2863 | bcopy(hdr, nhdr, HDR_L2ONLY_SIZE); | |
2864 | ||
2865 | if (new == hdr_full_cache) { | |
2866 | arc_hdr_set_flags(nhdr, ARC_FLAG_HAS_L1HDR); | |
2867 | /* | |
2868 | * arc_access and arc_change_state need to be aware that a | |
2869 | * header has just come out of L2ARC, so we set its state to | |
2870 | * l2c_only even though it's about to change. | |
2871 | */ | |
2872 | nhdr->b_l1hdr.b_state = arc_l2c_only; | |
e10b0808 | 2873 | |
cae5b340 AX |
2874 | /* Verify previous threads set to NULL before freeing */ |
2875 | ASSERT3P(nhdr->b_l1hdr.b_pabd, ==, NULL); | |
2876 | } else { | |
2877 | ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); | |
2878 | ASSERT0(hdr->b_l1hdr.b_bufcnt); | |
2879 | ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL); | |
2880 | ||
2881 | /* | |
2882 | * If we've reached here, We must have been called from | |
2883 | * arc_evict_hdr(), as such we should have already been | |
2884 | * removed from any ghost list we were previously on | |
2885 | * (which protects us from racing with arc_evict_state), | |
2886 | * thus no locking is needed during this check. | |
2887 | */ | |
2888 | ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); | |
2889 | ||
2890 | /* | |
2891 | * A buffer must not be moved into the arc_l2c_only | |
2892 | * state if it's not finished being written out to the | |
2893 | * l2arc device. Otherwise, the b_l1hdr.b_pabd field | |
2894 | * might try to be accessed, even though it was removed. | |
2895 | */ | |
2896 | VERIFY(!HDR_L2_WRITING(hdr)); | |
2897 | VERIFY3P(hdr->b_l1hdr.b_pabd, ==, NULL); | |
2898 | ||
2899 | arc_hdr_clear_flags(nhdr, ARC_FLAG_HAS_L1HDR); | |
2900 | } | |
e10b0808 | 2901 | /* |
cae5b340 AX |
2902 | * The header has been reallocated so we need to re-insert it into any |
2903 | * lists it was on. | |
e10b0808 | 2904 | */ |
cae5b340 AX |
2905 | (void) buf_hash_insert(nhdr, NULL); |
2906 | ||
2907 | ASSERT(list_link_active(&hdr->b_l2hdr.b_l2node)); | |
2908 | ||
2909 | mutex_enter(&dev->l2ad_mtx); | |
ea04106b | 2910 | |
e10b0808 | 2911 | /* |
cae5b340 AX |
2912 | * We must place the realloc'ed header back into the list at |
2913 | * the same spot. Otherwise, if it's placed earlier in the list, | |
2914 | * l2arc_write_buffers() could find it during the function's | |
2915 | * write phase, and try to write it out to the l2arc. | |
e10b0808 | 2916 | */ |
cae5b340 AX |
2917 | list_insert_after(&dev->l2ad_buflist, hdr, nhdr); |
2918 | list_remove(&dev->l2ad_buflist, hdr); | |
2919 | ||
2920 | mutex_exit(&dev->l2ad_mtx); | |
ea04106b | 2921 | |
e10b0808 | 2922 | /* |
cae5b340 AX |
2923 | * Since we're using the pointer address as the tag when |
2924 | * incrementing and decrementing the l2ad_alloc refcount, we | |
2925 | * must remove the old pointer (that we're about to destroy) and | |
2926 | * add the new pointer to the refcount. Otherwise we'd remove | |
2927 | * the wrong pointer address when calling arc_hdr_destroy() later. | |
e10b0808 | 2928 | */ |
e10b0808 | 2929 | |
cae5b340 AX |
2930 | (void) refcount_remove_many(&dev->l2ad_alloc, arc_hdr_size(hdr), hdr); |
2931 | (void) refcount_add_many(&dev->l2ad_alloc, arc_hdr_size(nhdr), nhdr); | |
e10b0808 | 2932 | |
cae5b340 AX |
2933 | buf_discard_identity(hdr); |
2934 | kmem_cache_free(old, hdr); | |
ea04106b | 2935 | |
cae5b340 | 2936 | return (nhdr); |
ea04106b AX |
2937 | } |
2938 | ||
e10b0808 | 2939 | /* |
cae5b340 AX |
2940 | * Allocate a new arc_buf_hdr_t and arc_buf_t and return the buf to the caller. |
2941 | * The buf is returned thawed since we expect the consumer to modify it. | |
e10b0808 | 2942 | */ |
cae5b340 AX |
2943 | arc_buf_t * |
2944 | arc_alloc_buf(spa_t *spa, void *tag, arc_buf_contents_t type, int32_t size) | |
34dc7c2f | 2945 | { |
cae5b340 AX |
2946 | arc_buf_hdr_t *hdr = arc_hdr_alloc(spa_load_guid(spa), size, size, |
2947 | ZIO_COMPRESS_OFF, type); | |
2948 | ASSERT(!MUTEX_HELD(HDR_LOCK(hdr))); | |
428870ff | 2949 | |
cae5b340 AX |
2950 | arc_buf_t *buf = NULL; |
2951 | VERIFY0(arc_buf_alloc_impl(hdr, tag, B_FALSE, B_FALSE, &buf)); | |
2952 | arc_buf_thaw(buf); | |
e10b0808 | 2953 | |
cae5b340 AX |
2954 | return (buf); |
2955 | } | |
34dc7c2f | 2956 | |
cae5b340 AX |
2957 | /* |
2958 | * Allocate a compressed buf in the same manner as arc_alloc_buf. Don't use this | |
2959 | * for bufs containing metadata. | |
2960 | */ | |
2961 | arc_buf_t * | |
2962 | arc_alloc_compressed_buf(spa_t *spa, void *tag, uint64_t psize, uint64_t lsize, | |
2963 | enum zio_compress compression_type) | |
2964 | { | |
2965 | ASSERT3U(lsize, >, 0); | |
2966 | ASSERT3U(lsize, >=, psize); | |
2967 | ASSERT(compression_type > ZIO_COMPRESS_OFF); | |
2968 | ASSERT(compression_type < ZIO_COMPRESS_FUNCTIONS); | |
34dc7c2f | 2969 | |
cae5b340 AX |
2970 | arc_buf_hdr_t *hdr = arc_hdr_alloc(spa_load_guid(spa), psize, lsize, |
2971 | compression_type, ARC_BUFC_DATA); | |
2972 | ASSERT(!MUTEX_HELD(HDR_LOCK(hdr))); | |
e10b0808 | 2973 | |
cae5b340 AX |
2974 | arc_buf_t *buf = NULL; |
2975 | VERIFY0(arc_buf_alloc_impl(hdr, tag, B_TRUE, B_FALSE, &buf)); | |
2976 | arc_buf_thaw(buf); | |
2977 | ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL); | |
1eb5bfa3 | 2978 | |
cae5b340 | 2979 | if (!arc_buf_is_shared(buf)) { |
1eb5bfa3 | 2980 | /* |
cae5b340 AX |
2981 | * To ensure that the hdr has the correct data in it if we call |
2982 | * arc_decompress() on this buf before it's been written to | |
2983 | * disk, it's easiest if we just set up sharing between the | |
2984 | * buf and the hdr. | |
1eb5bfa3 | 2985 | */ |
cae5b340 AX |
2986 | ASSERT(!abd_is_linear(hdr->b_l1hdr.b_pabd)); |
2987 | arc_hdr_free_pabd(hdr); | |
2988 | arc_share_buf(hdr, buf); | |
34dc7c2f BB |
2989 | } |
2990 | ||
cae5b340 | 2991 | return (buf); |
34dc7c2f BB |
2992 | } |
2993 | ||
2994 | static void | |
e10b0808 | 2995 | arc_hdr_l2hdr_destroy(arc_buf_hdr_t *hdr) |
34dc7c2f | 2996 | { |
e10b0808 AX |
2997 | l2arc_buf_hdr_t *l2hdr = &hdr->b_l2hdr; |
2998 | l2arc_dev_t *dev = l2hdr->b_dev; | |
cae5b340 | 2999 | uint64_t psize = arc_hdr_size(hdr); |
e10b0808 AX |
3000 | |
3001 | ASSERT(MUTEX_HELD(&dev->l2ad_mtx)); | |
3002 | ASSERT(HDR_HAS_L2HDR(hdr)); | |
3003 | ||
3004 | list_remove(&dev->l2ad_buflist, hdr); | |
3005 | ||
cae5b340 AX |
3006 | ARCSTAT_INCR(arcstat_l2_psize, -psize); |
3007 | ARCSTAT_INCR(arcstat_l2_lsize, -HDR_GET_LSIZE(hdr)); | |
e10b0808 | 3008 | |
cae5b340 | 3009 | vdev_space_update(dev->l2ad_vdev, -psize, 0, 0); |
e10b0808 | 3010 | |
cae5b340 AX |
3011 | (void) refcount_remove_many(&dev->l2ad_alloc, psize, hdr); |
3012 | arc_hdr_clear_flags(hdr, ARC_FLAG_HAS_L2HDR); | |
e10b0808 AX |
3013 | } |
3014 | ||
3015 | static void | |
3016 | arc_hdr_destroy(arc_buf_hdr_t *hdr) | |
3017 | { | |
3018 | if (HDR_HAS_L1HDR(hdr)) { | |
3019 | ASSERT(hdr->b_l1hdr.b_buf == NULL || | |
cae5b340 | 3020 | hdr->b_l1hdr.b_bufcnt > 0); |
e10b0808 AX |
3021 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); |
3022 | ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon); | |
3023 | } | |
34dc7c2f | 3024 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); |
e10b0808 AX |
3025 | ASSERT(!HDR_IN_HASH_TABLE(hdr)); |
3026 | ||
cae5b340 AX |
3027 | if (!HDR_EMPTY(hdr)) |
3028 | buf_discard_identity(hdr); | |
3029 | ||
e10b0808 AX |
3030 | if (HDR_HAS_L2HDR(hdr)) { |
3031 | l2arc_dev_t *dev = hdr->b_l2hdr.b_dev; | |
3032 | boolean_t buflist_held = MUTEX_HELD(&dev->l2ad_mtx); | |
3033 | ||
3034 | if (!buflist_held) | |
3035 | mutex_enter(&dev->l2ad_mtx); | |
34dc7c2f | 3036 | |
428870ff | 3037 | /* |
e10b0808 AX |
3038 | * Even though we checked this conditional above, we |
3039 | * need to check this again now that we have the | |
3040 | * l2ad_mtx. This is because we could be racing with | |
3041 | * another thread calling l2arc_evict() which might have | |
3042 | * destroyed this header's L2 portion as we were waiting | |
3043 | * to acquire the l2ad_mtx. If that happens, we don't | |
3044 | * want to re-destroy the header's L2 portion. | |
428870ff | 3045 | */ |
e10b0808 AX |
3046 | if (HDR_HAS_L2HDR(hdr)) |
3047 | arc_hdr_l2hdr_destroy(hdr); | |
428870ff BB |
3048 | |
3049 | if (!buflist_held) | |
e10b0808 | 3050 | mutex_exit(&dev->l2ad_mtx); |
34dc7c2f BB |
3051 | } |
3052 | ||
cae5b340 AX |
3053 | if (HDR_HAS_L1HDR(hdr)) { |
3054 | arc_cksum_free(hdr); | |
34dc7c2f | 3055 | |
cae5b340 AX |
3056 | while (hdr->b_l1hdr.b_buf != NULL) |
3057 | arc_buf_destroy_impl(hdr->b_l1hdr.b_buf); | |
34dc7c2f | 3058 | |
cae5b340 AX |
3059 | if (hdr->b_l1hdr.b_pabd != NULL) |
3060 | arc_hdr_free_pabd(hdr); | |
e10b0808 AX |
3061 | } |
3062 | ||
34dc7c2f | 3063 | ASSERT3P(hdr->b_hash_next, ==, NULL); |
e10b0808 AX |
3064 | if (HDR_HAS_L1HDR(hdr)) { |
3065 | ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); | |
3066 | ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL); | |
3067 | kmem_cache_free(hdr_full_cache, hdr); | |
3068 | } else { | |
3069 | kmem_cache_free(hdr_l2only_cache, hdr); | |
3070 | } | |
34dc7c2f BB |
3071 | } |
3072 | ||
3073 | void | |
cae5b340 | 3074 | arc_buf_destroy(arc_buf_t *buf, void* tag) |
34dc7c2f BB |
3075 | { |
3076 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
e10b0808 | 3077 | kmutex_t *hash_lock = HDR_LOCK(hdr); |
34dc7c2f | 3078 | |
e10b0808 | 3079 | if (hdr->b_l1hdr.b_state == arc_anon) { |
cae5b340 AX |
3080 | ASSERT3U(hdr->b_l1hdr.b_bufcnt, ==, 1); |
3081 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); | |
3082 | VERIFY0(remove_reference(hdr, NULL, tag)); | |
3083 | arc_hdr_destroy(hdr); | |
3084 | return; | |
34dc7c2f BB |
3085 | } |
3086 | ||
3087 | mutex_enter(hash_lock); | |
cae5b340 AX |
3088 | ASSERT3P(hdr, ==, buf->b_hdr); |
3089 | ASSERT(hdr->b_l1hdr.b_bufcnt > 0); | |
428870ff | 3090 | ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); |
cae5b340 AX |
3091 | ASSERT3P(hdr->b_l1hdr.b_state, !=, arc_anon); |
3092 | ASSERT3P(buf->b_data, !=, NULL); | |
34dc7c2f BB |
3093 | |
3094 | (void) remove_reference(hdr, hash_lock, tag); | |
cae5b340 | 3095 | arc_buf_destroy_impl(buf); |
34dc7c2f | 3096 | mutex_exit(hash_lock); |
1eb5bfa3 GW |
3097 | } |
3098 | ||
34dc7c2f | 3099 | /* |
e10b0808 AX |
3100 | * Evict the arc_buf_hdr that is provided as a parameter. The resultant |
3101 | * state of the header is dependent on its state prior to entering this | |
3102 | * function. The following transitions are possible: | |
34dc7c2f | 3103 | * |
e10b0808 AX |
3104 | * - arc_mru -> arc_mru_ghost |
3105 | * - arc_mfu -> arc_mfu_ghost | |
3106 | * - arc_mru_ghost -> arc_l2c_only | |
3107 | * - arc_mru_ghost -> deleted | |
3108 | * - arc_mfu_ghost -> arc_l2c_only | |
3109 | * - arc_mfu_ghost -> deleted | |
34dc7c2f | 3110 | */ |
e10b0808 AX |
3111 | static int64_t |
3112 | arc_evict_hdr(arc_buf_hdr_t *hdr, kmutex_t *hash_lock) | |
34dc7c2f | 3113 | { |
e10b0808 AX |
3114 | arc_state_t *evicted_state, *state; |
3115 | int64_t bytes_evicted = 0; | |
34dc7c2f | 3116 | |
e10b0808 AX |
3117 | ASSERT(MUTEX_HELD(hash_lock)); |
3118 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
a08ee875 | 3119 | |
e10b0808 AX |
3120 | state = hdr->b_l1hdr.b_state; |
3121 | if (GHOST_STATE(state)) { | |
3122 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); | |
cae5b340 | 3123 | ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); |
a08ee875 LG |
3124 | |
3125 | /* | |
e10b0808 | 3126 | * l2arc_write_buffers() relies on a header's L1 portion |
cae5b340 | 3127 | * (i.e. its b_pabd field) during it's write phase. |
e10b0808 AX |
3128 | * Thus, we cannot push a header onto the arc_l2c_only |
3129 | * state (removing its L1 piece) until the header is | |
3130 | * done being written to the l2arc. | |
a08ee875 | 3131 | */ |
e10b0808 AX |
3132 | if (HDR_HAS_L2HDR(hdr) && HDR_L2_WRITING(hdr)) { |
3133 | ARCSTAT_BUMP(arcstat_evict_l2_skip); | |
3134 | return (bytes_evicted); | |
a08ee875 LG |
3135 | } |
3136 | ||
e10b0808 | 3137 | ARCSTAT_BUMP(arcstat_deleted); |
cae5b340 | 3138 | bytes_evicted += HDR_GET_LSIZE(hdr); |
428870ff | 3139 | |
e10b0808 | 3140 | DTRACE_PROBE1(arc__delete, arc_buf_hdr_t *, hdr); |
428870ff | 3141 | |
e10b0808 | 3142 | if (HDR_HAS_L2HDR(hdr)) { |
cae5b340 | 3143 | ASSERT(hdr->b_l1hdr.b_pabd == NULL); |
e10b0808 AX |
3144 | /* |
3145 | * This buffer is cached on the 2nd Level ARC; | |
3146 | * don't destroy the header. | |
3147 | */ | |
3148 | arc_change_state(arc_l2c_only, hdr, hash_lock); | |
3149 | /* | |
3150 | * dropping from L1+L2 cached to L2-only, | |
3151 | * realloc to remove the L1 header. | |
3152 | */ | |
3153 | hdr = arc_hdr_realloc(hdr, hdr_full_cache, | |
3154 | hdr_l2only_cache); | |
34dc7c2f | 3155 | } else { |
e10b0808 AX |
3156 | arc_change_state(arc_anon, hdr, hash_lock); |
3157 | arc_hdr_destroy(hdr); | |
34dc7c2f | 3158 | } |
e10b0808 | 3159 | return (bytes_evicted); |
34dc7c2f BB |
3160 | } |
3161 | ||
e10b0808 AX |
3162 | ASSERT(state == arc_mru || state == arc_mfu); |
3163 | evicted_state = (state == arc_mru) ? arc_mru_ghost : arc_mfu_ghost; | |
34dc7c2f | 3164 | |
e10b0808 AX |
3165 | /* prefetch buffers have a minimum lifespan */ |
3166 | if (HDR_IO_IN_PROGRESS(hdr) || | |
3167 | ((hdr->b_flags & (ARC_FLAG_PREFETCH | ARC_FLAG_INDIRECT)) && | |
3168 | ddi_get_lbolt() - hdr->b_l1hdr.b_arc_access < | |
3169 | arc_min_prefetch_lifespan)) { | |
3170 | ARCSTAT_BUMP(arcstat_evict_skip); | |
3171 | return (bytes_evicted); | |
ea04106b AX |
3172 | } |
3173 | ||
e10b0808 | 3174 | ASSERT0(refcount_count(&hdr->b_l1hdr.b_refcnt)); |
e10b0808 AX |
3175 | while (hdr->b_l1hdr.b_buf) { |
3176 | arc_buf_t *buf = hdr->b_l1hdr.b_buf; | |
3177 | if (!mutex_tryenter(&buf->b_evict_lock)) { | |
3178 | ARCSTAT_BUMP(arcstat_mutex_miss); | |
3179 | break; | |
3180 | } | |
3181 | if (buf->b_data != NULL) | |
cae5b340 AX |
3182 | bytes_evicted += HDR_GET_LSIZE(hdr); |
3183 | mutex_exit(&buf->b_evict_lock); | |
3184 | arc_buf_destroy_impl(buf); | |
e10b0808 | 3185 | } |
34dc7c2f | 3186 | |
e10b0808 | 3187 | if (HDR_HAS_L2HDR(hdr)) { |
cae5b340 | 3188 | ARCSTAT_INCR(arcstat_evict_l2_cached, HDR_GET_LSIZE(hdr)); |
e10b0808 | 3189 | } else { |
cae5b340 AX |
3190 | if (l2arc_write_eligible(hdr->b_spa, hdr)) { |
3191 | ARCSTAT_INCR(arcstat_evict_l2_eligible, | |
3192 | HDR_GET_LSIZE(hdr)); | |
3193 | } else { | |
3194 | ARCSTAT_INCR(arcstat_evict_l2_ineligible, | |
3195 | HDR_GET_LSIZE(hdr)); | |
3196 | } | |
e10b0808 | 3197 | } |
34dc7c2f | 3198 | |
cae5b340 AX |
3199 | if (hdr->b_l1hdr.b_bufcnt == 0) { |
3200 | arc_cksum_free(hdr); | |
3201 | ||
3202 | bytes_evicted += arc_hdr_size(hdr); | |
3203 | ||
3204 | /* | |
3205 | * If this hdr is being evicted and has a compressed | |
3206 | * buffer then we discard it here before we change states. | |
3207 | * This ensures that the accounting is updated correctly | |
3208 | * in arc_free_data_impl(). | |
3209 | */ | |
3210 | arc_hdr_free_pabd(hdr); | |
3211 | ||
e10b0808 AX |
3212 | arc_change_state(evicted_state, hdr, hash_lock); |
3213 | ASSERT(HDR_IN_HASH_TABLE(hdr)); | |
cae5b340 | 3214 | arc_hdr_set_flags(hdr, ARC_FLAG_IN_HASH_TABLE); |
e10b0808 AX |
3215 | DTRACE_PROBE1(arc__evict, arc_buf_hdr_t *, hdr); |
3216 | } | |
34dc7c2f | 3217 | |
e10b0808 | 3218 | return (bytes_evicted); |
34dc7c2f BB |
3219 | } |
3220 | ||
e10b0808 AX |
3221 | static uint64_t |
3222 | arc_evict_state_impl(multilist_t *ml, int idx, arc_buf_hdr_t *marker, | |
3223 | uint64_t spa, int64_t bytes) | |
34dc7c2f | 3224 | { |
e10b0808 AX |
3225 | multilist_sublist_t *mls; |
3226 | uint64_t bytes_evicted = 0; | |
3227 | arc_buf_hdr_t *hdr; | |
34dc7c2f | 3228 | kmutex_t *hash_lock; |
e10b0808 | 3229 | int evict_count = 0; |
34dc7c2f | 3230 | |
e10b0808 AX |
3231 | ASSERT3P(marker, !=, NULL); |
3232 | IMPLY(bytes < 0, bytes == ARC_EVICT_ALL); | |
572e2857 | 3233 | |
e10b0808 | 3234 | mls = multilist_sublist_lock(ml, idx); |
572e2857 | 3235 | |
e10b0808 AX |
3236 | for (hdr = multilist_sublist_prev(mls, marker); hdr != NULL; |
3237 | hdr = multilist_sublist_prev(mls, marker)) { | |
3238 | if ((bytes != ARC_EVICT_ALL && bytes_evicted >= bytes) || | |
3239 | (evict_count >= zfs_arc_evict_batch_limit)) | |
3240 | break; | |
a08ee875 LG |
3241 | |
3242 | /* | |
e10b0808 AX |
3243 | * To keep our iteration location, move the marker |
3244 | * forward. Since we're not holding hdr's hash lock, we | |
3245 | * must be very careful and not remove 'hdr' from the | |
3246 | * sublist. Otherwise, other consumers might mistake the | |
3247 | * 'hdr' as not being on a sublist when they call the | |
3248 | * multilist_link_active() function (they all rely on | |
3249 | * the hash lock protecting concurrent insertions and | |
3250 | * removals). multilist_sublist_move_forward() was | |
3251 | * specifically implemented to ensure this is the case | |
3252 | * (only 'marker' will be removed and re-inserted). | |
a08ee875 | 3253 | */ |
e10b0808 AX |
3254 | multilist_sublist_move_forward(mls, marker); |
3255 | ||
3256 | /* | |
3257 | * The only case where the b_spa field should ever be | |
3258 | * zero, is the marker headers inserted by | |
3259 | * arc_evict_state(). It's possible for multiple threads | |
3260 | * to be calling arc_evict_state() concurrently (e.g. | |
3261 | * dsl_pool_close() and zio_inject_fault()), so we must | |
3262 | * skip any markers we see from these other threads. | |
3263 | */ | |
3264 | if (hdr->b_spa == 0) | |
3265 | continue; | |
3266 | ||
3267 | /* we're only interested in evicting buffers of a certain spa */ | |
3268 | if (spa != 0 && hdr->b_spa != spa) { | |
3269 | ARCSTAT_BUMP(arcstat_evict_skip); | |
a08ee875 LG |
3270 | continue; |
3271 | } | |
e10b0808 AX |
3272 | |
3273 | hash_lock = HDR_LOCK(hdr); | |
3274 | ||
3275 | /* | |
3276 | * We aren't calling this function from any code path | |
3277 | * that would already be holding a hash lock, so we're | |
3278 | * asserting on this assumption to be defensive in case | |
3279 | * this ever changes. Without this check, it would be | |
3280 | * possible to incorrectly increment arcstat_mutex_miss | |
3281 | * below (e.g. if the code changed such that we called | |
3282 | * this function with a hash lock held). | |
3283 | */ | |
3284 | ASSERT(!MUTEX_HELD(hash_lock)); | |
3285 | ||
34dc7c2f | 3286 | if (mutex_tryenter(hash_lock)) { |
e10b0808 AX |
3287 | uint64_t evicted = arc_evict_hdr(hdr, hash_lock); |
3288 | mutex_exit(hash_lock); | |
34dc7c2f | 3289 | |
e10b0808 | 3290 | bytes_evicted += evicted; |
34dc7c2f | 3291 | |
572e2857 | 3292 | /* |
e10b0808 AX |
3293 | * If evicted is zero, arc_evict_hdr() must have |
3294 | * decided to skip this header, don't increment | |
3295 | * evict_count in this case. | |
572e2857 | 3296 | */ |
e10b0808 AX |
3297 | if (evicted != 0) |
3298 | evict_count++; | |
3299 | ||
3300 | /* | |
3301 | * If arc_size isn't overflowing, signal any | |
3302 | * threads that might happen to be waiting. | |
3303 | * | |
3304 | * For each header evicted, we wake up a single | |
3305 | * thread. If we used cv_broadcast, we could | |
3306 | * wake up "too many" threads causing arc_size | |
3307 | * to significantly overflow arc_c; since | |
cae5b340 | 3308 | * arc_get_data_impl() doesn't check for overflow |
e10b0808 AX |
3309 | * when it's woken up (it doesn't because it's |
3310 | * possible for the ARC to be overflowing while | |
3311 | * full of un-evictable buffers, and the | |
3312 | * function should proceed in this case). | |
3313 | * | |
3314 | * If threads are left sleeping, due to not | |
3315 | * using cv_broadcast, they will be woken up | |
3316 | * just before arc_reclaim_thread() sleeps. | |
3317 | */ | |
3318 | mutex_enter(&arc_reclaim_lock); | |
3319 | if (!arc_is_overflowing()) | |
3320 | cv_signal(&arc_reclaim_waiters_cv); | |
3321 | mutex_exit(&arc_reclaim_lock); | |
a08ee875 | 3322 | } else { |
e10b0808 | 3323 | ARCSTAT_BUMP(arcstat_mutex_miss); |
a08ee875 | 3324 | } |
34dc7c2f | 3325 | } |
34dc7c2f | 3326 | |
e10b0808 | 3327 | multilist_sublist_unlock(mls); |
34dc7c2f | 3328 | |
e10b0808 | 3329 | return (bytes_evicted); |
34dc7c2f BB |
3330 | } |
3331 | ||
e10b0808 AX |
3332 | /* |
3333 | * Evict buffers from the given arc state, until we've removed the | |
3334 | * specified number of bytes. Move the removed buffers to the | |
3335 | * appropriate evict state. | |
3336 | * | |
3337 | * This function makes a "best effort". It skips over any buffers | |
3338 | * it can't get a hash_lock on, and so, may not catch all candidates. | |
3339 | * It may also return without evicting as much space as requested. | |
3340 | * | |
3341 | * If bytes is specified using the special value ARC_EVICT_ALL, this | |
3342 | * will evict all available (i.e. unlocked and evictable) buffers from | |
3343 | * the given arc state; which is used by arc_flush(). | |
3344 | */ | |
3345 | static uint64_t | |
3346 | arc_evict_state(arc_state_t *state, uint64_t spa, int64_t bytes, | |
3347 | arc_buf_contents_t type) | |
34dc7c2f | 3348 | { |
e10b0808 | 3349 | uint64_t total_evicted = 0; |
cae5b340 | 3350 | multilist_t *ml = state->arcs_list[type]; |
e10b0808 AX |
3351 | int num_sublists; |
3352 | arc_buf_hdr_t **markers; | |
3353 | int i; | |
3354 | ||
3355 | IMPLY(bytes < 0, bytes == ARC_EVICT_ALL); | |
3356 | ||
3357 | num_sublists = multilist_get_num_sublists(ml); | |
d164b209 BB |
3358 | |
3359 | /* | |
e10b0808 AX |
3360 | * If we've tried to evict from each sublist, made some |
3361 | * progress, but still have not hit the target number of bytes | |
3362 | * to evict, we want to keep trying. The markers allow us to | |
3363 | * pick up where we left off for each individual sublist, rather | |
3364 | * than starting from the tail each time. | |
d164b209 | 3365 | */ |
e10b0808 AX |
3366 | markers = kmem_zalloc(sizeof (*markers) * num_sublists, KM_SLEEP); |
3367 | for (i = 0; i < num_sublists; i++) { | |
3368 | multilist_sublist_t *mls; | |
34dc7c2f | 3369 | |
e10b0808 AX |
3370 | markers[i] = kmem_cache_alloc(hdr_full_cache, KM_SLEEP); |
3371 | ||
3372 | /* | |
3373 | * A b_spa of 0 is used to indicate that this header is | |
3374 | * a marker. This fact is used in arc_adjust_type() and | |
3375 | * arc_evict_state_impl(). | |
3376 | */ | |
3377 | markers[i]->b_spa = 0; | |
34dc7c2f | 3378 | |
e10b0808 AX |
3379 | mls = multilist_sublist_lock(ml, i); |
3380 | multilist_sublist_insert_tail(mls, markers[i]); | |
3381 | multilist_sublist_unlock(mls); | |
34dc7c2f BB |
3382 | } |
3383 | ||
d164b209 | 3384 | /* |
e10b0808 AX |
3385 | * While we haven't hit our target number of bytes to evict, or |
3386 | * we're evicting all available buffers. | |
d164b209 | 3387 | */ |
e10b0808 | 3388 | while (total_evicted < bytes || bytes == ARC_EVICT_ALL) { |
cae5b340 AX |
3389 | int sublist_idx = multilist_get_random_index(ml); |
3390 | uint64_t scan_evicted = 0; | |
3391 | ||
3392 | /* | |
3393 | * Try to reduce pinned dnodes with a floor of arc_dnode_limit. | |
3394 | * Request that 10% of the LRUs be scanned by the superblock | |
3395 | * shrinker. | |
3396 | */ | |
3397 | if (type == ARC_BUFC_DATA && arc_dnode_size > arc_dnode_limit) | |
3398 | arc_prune_async((arc_dnode_size - arc_dnode_limit) / | |
3399 | sizeof (dnode_t) / zfs_arc_dnode_reduce_percent); | |
3400 | ||
e10b0808 AX |
3401 | /* |
3402 | * Start eviction using a randomly selected sublist, | |
3403 | * this is to try and evenly balance eviction across all | |
3404 | * sublists. Always starting at the same sublist | |
3405 | * (e.g. index 0) would cause evictions to favor certain | |
3406 | * sublists over others. | |
3407 | */ | |
e10b0808 AX |
3408 | for (i = 0; i < num_sublists; i++) { |
3409 | uint64_t bytes_remaining; | |
3410 | uint64_t bytes_evicted; | |
d164b209 | 3411 | |
e10b0808 AX |
3412 | if (bytes == ARC_EVICT_ALL) |
3413 | bytes_remaining = ARC_EVICT_ALL; | |
3414 | else if (total_evicted < bytes) | |
3415 | bytes_remaining = bytes - total_evicted; | |
3416 | else | |
3417 | break; | |
34dc7c2f | 3418 | |
e10b0808 AX |
3419 | bytes_evicted = arc_evict_state_impl(ml, sublist_idx, |
3420 | markers[sublist_idx], spa, bytes_remaining); | |
3421 | ||
3422 | scan_evicted += bytes_evicted; | |
3423 | total_evicted += bytes_evicted; | |
34dc7c2f | 3424 | |
e10b0808 AX |
3425 | /* we've reached the end, wrap to the beginning */ |
3426 | if (++sublist_idx >= num_sublists) | |
3427 | sublist_idx = 0; | |
3428 | } | |
3429 | ||
3430 | /* | |
3431 | * If we didn't evict anything during this scan, we have | |
3432 | * no reason to believe we'll evict more during another | |
3433 | * scan, so break the loop. | |
3434 | */ | |
3435 | if (scan_evicted == 0) { | |
3436 | /* This isn't possible, let's make that obvious */ | |
3437 | ASSERT3S(bytes, !=, 0); | |
3438 | ||
3439 | /* | |
3440 | * When bytes is ARC_EVICT_ALL, the only way to | |
3441 | * break the loop is when scan_evicted is zero. | |
3442 | * In that case, we actually have evicted enough, | |
3443 | * so we don't want to increment the kstat. | |
3444 | */ | |
3445 | if (bytes != ARC_EVICT_ALL) { | |
3446 | ASSERT3S(total_evicted, <, bytes); | |
3447 | ARCSTAT_BUMP(arcstat_evict_not_enough); | |
3448 | } | |
d164b209 | 3449 | |
e10b0808 AX |
3450 | break; |
3451 | } | |
d164b209 | 3452 | } |
34dc7c2f | 3453 | |
e10b0808 AX |
3454 | for (i = 0; i < num_sublists; i++) { |
3455 | multilist_sublist_t *mls = multilist_sublist_lock(ml, i); | |
3456 | multilist_sublist_remove(mls, markers[i]); | |
3457 | multilist_sublist_unlock(mls); | |
34dc7c2f | 3458 | |
e10b0808 | 3459 | kmem_cache_free(hdr_full_cache, markers[i]); |
34dc7c2f | 3460 | } |
e10b0808 AX |
3461 | kmem_free(markers, sizeof (*markers) * num_sublists); |
3462 | ||
3463 | return (total_evicted); | |
34dc7c2f BB |
3464 | } |
3465 | ||
ab26409d | 3466 | /* |
e10b0808 AX |
3467 | * Flush all "evictable" data of the given type from the arc state |
3468 | * specified. This will not evict any "active" buffers (i.e. referenced). | |
3469 | * | |
cae5b340 | 3470 | * When 'retry' is set to B_FALSE, the function will make a single pass |
e10b0808 AX |
3471 | * over the state and evict any buffers that it can. Since it doesn't |
3472 | * continually retry the eviction, it might end up leaving some buffers | |
3473 | * in the ARC due to lock misses. | |
3474 | * | |
cae5b340 | 3475 | * When 'retry' is set to B_TRUE, the function will continually retry the |
e10b0808 AX |
3476 | * eviction until *all* evictable buffers have been removed from the |
3477 | * state. As a result, if concurrent insertions into the state are | |
3478 | * allowed (e.g. if the ARC isn't shutting down), this function might | |
3479 | * wind up in an infinite loop, continually trying to evict buffers. | |
ab26409d | 3480 | */ |
e10b0808 AX |
3481 | static uint64_t |
3482 | arc_flush_state(arc_state_t *state, uint64_t spa, arc_buf_contents_t type, | |
3483 | boolean_t retry) | |
ab26409d | 3484 | { |
e10b0808 | 3485 | uint64_t evicted = 0; |
ab26409d | 3486 | |
cae5b340 | 3487 | while (refcount_count(&state->arcs_esize[type]) != 0) { |
e10b0808 | 3488 | evicted += arc_evict_state(state, spa, ARC_EVICT_ALL, type); |
ab26409d | 3489 | |
e10b0808 AX |
3490 | if (!retry) |
3491 | break; | |
3492 | } | |
ab26409d | 3493 | |
e10b0808 AX |
3494 | return (evicted); |
3495 | } | |
ab26409d | 3496 | |
e10b0808 AX |
3497 | /* |
3498 | * Helper function for arc_prune_async() it is responsible for safely | |
3499 | * handling the execution of a registered arc_prune_func_t. | |
3500 | */ | |
3501 | static void | |
3502 | arc_prune_task(void *ptr) | |
3503 | { | |
3504 | arc_prune_t *ap = (arc_prune_t *)ptr; | |
3505 | arc_prune_func_t *func = ap->p_pfunc; | |
ab26409d | 3506 | |
e10b0808 AX |
3507 | if (func != NULL) |
3508 | func(ap->p_adjust, ap->p_private); | |
ab26409d | 3509 | |
87dac73d | 3510 | refcount_remove(&ap->p_refcnt, func); |
ab26409d BB |
3511 | } |
3512 | ||
e10b0808 AX |
3513 | /* |
3514 | * Notify registered consumers they must drop holds on a portion of the ARC | |
3515 | * buffered they reference. This provides a mechanism to ensure the ARC can | |
3516 | * honor the arc_meta_limit and reclaim otherwise pinned ARC buffers. This | |
3517 | * is analogous to dnlc_reduce_cache() but more generic. | |
3518 | * | |
3519 | * This operation is performed asynchronously so it may be safely called | |
3520 | * in the context of the arc_reclaim_thread(). A reference is taken here | |
3521 | * for each registered arc_prune_t and the arc_prune_task() is responsible | |
3522 | * for releasing it once the registered arc_prune_func_t has completed. | |
3523 | */ | |
34dc7c2f | 3524 | static void |
e10b0808 | 3525 | arc_prune_async(int64_t adjust) |
34dc7c2f | 3526 | { |
e10b0808 | 3527 | arc_prune_t *ap; |
34dc7c2f | 3528 | |
e10b0808 AX |
3529 | mutex_enter(&arc_prune_mtx); |
3530 | for (ap = list_head(&arc_prune_list); ap != NULL; | |
3531 | ap = list_next(&arc_prune_list, ap)) { | |
34dc7c2f | 3532 | |
e10b0808 AX |
3533 | if (refcount_count(&ap->p_refcnt) >= 2) |
3534 | continue; | |
3535 | ||
3536 | refcount_add(&ap->p_refcnt, ap->p_pfunc); | |
3537 | ap->p_adjust = adjust; | |
cae5b340 AX |
3538 | if (taskq_dispatch(arc_prune_taskq, arc_prune_task, |
3539 | ap, TQ_SLEEP) == TASKQID_INVALID) { | |
3540 | refcount_remove(&ap->p_refcnt, ap->p_pfunc); | |
3541 | continue; | |
3542 | } | |
e10b0808 | 3543 | ARCSTAT_BUMP(arcstat_prune); |
34dc7c2f | 3544 | } |
e10b0808 AX |
3545 | mutex_exit(&arc_prune_mtx); |
3546 | } | |
3547 | ||
3548 | /* | |
3549 | * Evict the specified number of bytes from the state specified, | |
3550 | * restricting eviction to the spa and type given. This function | |
3551 | * prevents us from trying to evict more from a state's list than | |
3552 | * is "evictable", and to skip evicting altogether when passed a | |
3553 | * negative value for "bytes". In contrast, arc_evict_state() will | |
3554 | * evict everything it can, when passed a negative value for "bytes". | |
3555 | */ | |
3556 | static uint64_t | |
3557 | arc_adjust_impl(arc_state_t *state, uint64_t spa, int64_t bytes, | |
3558 | arc_buf_contents_t type) | |
3559 | { | |
3560 | int64_t delta; | |
3561 | ||
cae5b340 AX |
3562 | if (bytes > 0 && refcount_count(&state->arcs_esize[type]) > 0) { |
3563 | delta = MIN(refcount_count(&state->arcs_esize[type]), bytes); | |
e10b0808 AX |
3564 | return (arc_evict_state(state, spa, delta, type)); |
3565 | } | |
3566 | ||
3567 | return (0); | |
34dc7c2f BB |
3568 | } |
3569 | ||
ab26409d | 3570 | /* |
ea04106b AX |
3571 | * The goal of this function is to evict enough meta data buffers from the |
3572 | * ARC in order to enforce the arc_meta_limit. Achieving this is slightly | |
3573 | * more complicated than it appears because it is common for data buffers | |
3574 | * to have holds on meta data buffers. In addition, dnode meta data buffers | |
3575 | * will be held by the dnodes in the block preventing them from being freed. | |
3576 | * This means we can't simply traverse the ARC and expect to always find | |
3577 | * enough unheld meta data buffer to release. | |
3578 | * | |
3579 | * Therefore, this function has been updated to make alternating passes | |
3580 | * over the ARC releasing data buffers and then newly unheld meta data | |
3581 | * buffers. This ensures forward progress is maintained and arc_meta_used | |
3582 | * will decrease. Normally this is sufficient, but if required the ARC | |
3583 | * will call the registered prune callbacks causing dentry and inodes to | |
3584 | * be dropped from the VFS cache. This will make dnode meta data buffers | |
3585 | * available for reclaim. | |
ab26409d | 3586 | */ |
e10b0808 AX |
3587 | static uint64_t |
3588 | arc_adjust_meta_balanced(void) | |
ab26409d | 3589 | { |
cae5b340 | 3590 | int64_t delta, prune = 0, adjustmnt; |
e10b0808 | 3591 | uint64_t total_evicted = 0; |
ea04106b | 3592 | arc_buf_contents_t type = ARC_BUFC_DATA; |
e10b0808 | 3593 | int restarts = MAX(zfs_arc_meta_adjust_restarts, 0); |
ea04106b AX |
3594 | |
3595 | restart: | |
3596 | /* | |
3597 | * This slightly differs than the way we evict from the mru in | |
3598 | * arc_adjust because we don't have a "target" value (i.e. no | |
3599 | * "meta" arc_p). As a result, I think we can completely | |
3600 | * cannibalize the metadata in the MRU before we evict the | |
3601 | * metadata from the MFU. I think we probably need to implement a | |
3602 | * "metadata arc_p" value to do this properly. | |
3603 | */ | |
3604 | adjustmnt = arc_meta_used - arc_meta_limit; | |
3605 | ||
cae5b340 AX |
3606 | if (adjustmnt > 0 && refcount_count(&arc_mru->arcs_esize[type]) > 0) { |
3607 | delta = MIN(refcount_count(&arc_mru->arcs_esize[type]), | |
3608 | adjustmnt); | |
e10b0808 | 3609 | total_evicted += arc_adjust_impl(arc_mru, 0, delta, type); |
ea04106b AX |
3610 | adjustmnt -= delta; |
3611 | } | |
ab26409d | 3612 | |
ea04106b AX |
3613 | /* |
3614 | * We can't afford to recalculate adjustmnt here. If we do, | |
3615 | * new metadata buffers can sneak into the MRU or ANON lists, | |
3616 | * thus penalize the MFU metadata. Although the fudge factor is | |
3617 | * small, it has been empirically shown to be significant for | |
3618 | * certain workloads (e.g. creating many empty directories). As | |
3619 | * such, we use the original calculation for adjustmnt, and | |
3620 | * simply decrement the amount of data evicted from the MRU. | |
3621 | */ | |
3622 | ||
cae5b340 AX |
3623 | if (adjustmnt > 0 && refcount_count(&arc_mfu->arcs_esize[type]) > 0) { |
3624 | delta = MIN(refcount_count(&arc_mfu->arcs_esize[type]), | |
3625 | adjustmnt); | |
e10b0808 | 3626 | total_evicted += arc_adjust_impl(arc_mfu, 0, delta, type); |
ea04106b AX |
3627 | } |
3628 | ||
3629 | adjustmnt = arc_meta_used - arc_meta_limit; | |
3630 | ||
cae5b340 AX |
3631 | if (adjustmnt > 0 && |
3632 | refcount_count(&arc_mru_ghost->arcs_esize[type]) > 0) { | |
ea04106b | 3633 | delta = MIN(adjustmnt, |
cae5b340 | 3634 | refcount_count(&arc_mru_ghost->arcs_esize[type])); |
e10b0808 | 3635 | total_evicted += arc_adjust_impl(arc_mru_ghost, 0, delta, type); |
ea04106b | 3636 | adjustmnt -= delta; |
ab26409d BB |
3637 | } |
3638 | ||
cae5b340 AX |
3639 | if (adjustmnt > 0 && |
3640 | refcount_count(&arc_mfu_ghost->arcs_esize[type]) > 0) { | |
ea04106b | 3641 | delta = MIN(adjustmnt, |
cae5b340 | 3642 | refcount_count(&arc_mfu_ghost->arcs_esize[type])); |
e10b0808 | 3643 | total_evicted += arc_adjust_impl(arc_mfu_ghost, 0, delta, type); |
ab26409d BB |
3644 | } |
3645 | ||
ea04106b AX |
3646 | /* |
3647 | * If after attempting to make the requested adjustment to the ARC | |
3648 | * the meta limit is still being exceeded then request that the | |
3649 | * higher layers drop some cached objects which have holds on ARC | |
3650 | * meta buffers. Requests to the upper layers will be made with | |
3651 | * increasingly large scan sizes until the ARC is below the limit. | |
3652 | */ | |
3653 | if (arc_meta_used > arc_meta_limit) { | |
3654 | if (type == ARC_BUFC_DATA) { | |
3655 | type = ARC_BUFC_METADATA; | |
3656 | } else { | |
3657 | type = ARC_BUFC_DATA; | |
3658 | ||
3659 | if (zfs_arc_meta_prune) { | |
3660 | prune += zfs_arc_meta_prune; | |
e10b0808 | 3661 | arc_prune_async(prune); |
ea04106b AX |
3662 | } |
3663 | } | |
3664 | ||
3665 | if (restarts > 0) { | |
3666 | restarts--; | |
3667 | goto restart; | |
3668 | } | |
3669 | } | |
e10b0808 | 3670 | return (total_evicted); |
ab26409d BB |
3671 | } |
3672 | ||
34dc7c2f | 3673 | /* |
e10b0808 AX |
3674 | * Evict metadata buffers from the cache, such that arc_meta_used is |
3675 | * capped by the arc_meta_limit tunable. | |
34dc7c2f | 3676 | */ |
e10b0808 AX |
3677 | static uint64_t |
3678 | arc_adjust_meta_only(void) | |
34dc7c2f | 3679 | { |
e10b0808 AX |
3680 | uint64_t total_evicted = 0; |
3681 | int64_t target; | |
d164b209 | 3682 | |
e10b0808 AX |
3683 | /* |
3684 | * If we're over the meta limit, we want to evict enough | |
3685 | * metadata to get back under the meta limit. We don't want to | |
3686 | * evict so much that we drop the MRU below arc_p, though. If | |
3687 | * we're over the meta limit more than we're over arc_p, we | |
3688 | * evict some from the MRU here, and some from the MFU below. | |
3689 | */ | |
3690 | target = MIN((int64_t)(arc_meta_used - arc_meta_limit), | |
3691 | (int64_t)(refcount_count(&arc_anon->arcs_size) + | |
3692 | refcount_count(&arc_mru->arcs_size) - arc_p)); | |
3693 | ||
3694 | total_evicted += arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_METADATA); | |
3695 | ||
3696 | /* | |
3697 | * Similar to the above, we want to evict enough bytes to get us | |
3698 | * below the meta limit, but not so much as to drop us below the | |
cae5b340 | 3699 | * space allotted to the MFU (which is defined as arc_c - arc_p). |
e10b0808 AX |
3700 | */ |
3701 | target = MIN((int64_t)(arc_meta_used - arc_meta_limit), | |
3702 | (int64_t)(refcount_count(&arc_mfu->arcs_size) - (arc_c - arc_p))); | |
3703 | ||
3704 | total_evicted += arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_METADATA); | |
3705 | ||
3706 | return (total_evicted); | |
3707 | } | |
3708 | ||
3709 | static uint64_t | |
3710 | arc_adjust_meta(void) | |
3711 | { | |
3712 | if (zfs_arc_meta_strategy == ARC_STRATEGY_META_ONLY) | |
3713 | return (arc_adjust_meta_only()); | |
3714 | else | |
3715 | return (arc_adjust_meta_balanced()); | |
3716 | } | |
3717 | ||
3718 | /* | |
3719 | * Return the type of the oldest buffer in the given arc state | |
3720 | * | |
3721 | * This function will select a random sublist of type ARC_BUFC_DATA and | |
3722 | * a random sublist of type ARC_BUFC_METADATA. The tail of each sublist | |
3723 | * is compared, and the type which contains the "older" buffer will be | |
3724 | * returned. | |
3725 | */ | |
3726 | static arc_buf_contents_t | |
3727 | arc_adjust_type(arc_state_t *state) | |
3728 | { | |
cae5b340 AX |
3729 | multilist_t *data_ml = state->arcs_list[ARC_BUFC_DATA]; |
3730 | multilist_t *meta_ml = state->arcs_list[ARC_BUFC_METADATA]; | |
e10b0808 AX |
3731 | int data_idx = multilist_get_random_index(data_ml); |
3732 | int meta_idx = multilist_get_random_index(meta_ml); | |
3733 | multilist_sublist_t *data_mls; | |
3734 | multilist_sublist_t *meta_mls; | |
3735 | arc_buf_contents_t type; | |
3736 | arc_buf_hdr_t *data_hdr; | |
3737 | arc_buf_hdr_t *meta_hdr; | |
3738 | ||
3739 | /* | |
3740 | * We keep the sublist lock until we're finished, to prevent | |
3741 | * the headers from being destroyed via arc_evict_state(). | |
3742 | */ | |
3743 | data_mls = multilist_sublist_lock(data_ml, data_idx); | |
3744 | meta_mls = multilist_sublist_lock(meta_ml, meta_idx); | |
3745 | ||
3746 | /* | |
3747 | * These two loops are to ensure we skip any markers that | |
3748 | * might be at the tail of the lists due to arc_evict_state(). | |
3749 | */ | |
d164b209 | 3750 | |
e10b0808 AX |
3751 | for (data_hdr = multilist_sublist_tail(data_mls); data_hdr != NULL; |
3752 | data_hdr = multilist_sublist_prev(data_mls, data_hdr)) { | |
3753 | if (data_hdr->b_spa != 0) | |
34dc7c2f BB |
3754 | break; |
3755 | } | |
e10b0808 AX |
3756 | |
3757 | for (meta_hdr = multilist_sublist_tail(meta_mls); meta_hdr != NULL; | |
3758 | meta_hdr = multilist_sublist_prev(meta_mls, meta_hdr)) { | |
3759 | if (meta_hdr->b_spa != 0) | |
34dc7c2f BB |
3760 | break; |
3761 | } | |
e10b0808 AX |
3762 | |
3763 | if (data_hdr == NULL && meta_hdr == NULL) { | |
3764 | type = ARC_BUFC_DATA; | |
3765 | } else if (data_hdr == NULL) { | |
3766 | ASSERT3P(meta_hdr, !=, NULL); | |
3767 | type = ARC_BUFC_METADATA; | |
3768 | } else if (meta_hdr == NULL) { | |
3769 | ASSERT3P(data_hdr, !=, NULL); | |
3770 | type = ARC_BUFC_DATA; | |
3771 | } else { | |
3772 | ASSERT3P(data_hdr, !=, NULL); | |
3773 | ASSERT3P(meta_hdr, !=, NULL); | |
3774 | ||
3775 | /* The headers can't be on the sublist without an L1 header */ | |
3776 | ASSERT(HDR_HAS_L1HDR(data_hdr)); | |
3777 | ASSERT(HDR_HAS_L1HDR(meta_hdr)); | |
3778 | ||
3779 | if (data_hdr->b_l1hdr.b_arc_access < | |
3780 | meta_hdr->b_l1hdr.b_arc_access) { | |
3781 | type = ARC_BUFC_DATA; | |
3782 | } else { | |
3783 | type = ARC_BUFC_METADATA; | |
3784 | } | |
34dc7c2f | 3785 | } |
e10b0808 AX |
3786 | |
3787 | multilist_sublist_unlock(meta_mls); | |
3788 | multilist_sublist_unlock(data_mls); | |
3789 | ||
3790 | return (type); | |
3791 | } | |
3792 | ||
3793 | /* | |
3794 | * Evict buffers from the cache, such that arc_size is capped by arc_c. | |
3795 | */ | |
3796 | static uint64_t | |
3797 | arc_adjust(void) | |
3798 | { | |
3799 | uint64_t total_evicted = 0; | |
3800 | uint64_t bytes; | |
3801 | int64_t target; | |
3802 | ||
3803 | /* | |
3804 | * If we're over arc_meta_limit, we want to correct that before | |
3805 | * potentially evicting data buffers below. | |
3806 | */ | |
3807 | total_evicted += arc_adjust_meta(); | |
3808 | ||
3809 | /* | |
3810 | * Adjust MRU size | |
3811 | * | |
3812 | * If we're over the target cache size, we want to evict enough | |
3813 | * from the list to get back to our target size. We don't want | |
3814 | * to evict too much from the MRU, such that it drops below | |
3815 | * arc_p. So, if we're over our target cache size more than | |
3816 | * the MRU is over arc_p, we'll evict enough to get back to | |
3817 | * arc_p here, and then evict more from the MFU below. | |
3818 | */ | |
3819 | target = MIN((int64_t)(arc_size - arc_c), | |
3820 | (int64_t)(refcount_count(&arc_anon->arcs_size) + | |
3821 | refcount_count(&arc_mru->arcs_size) + arc_meta_used - arc_p)); | |
3822 | ||
3823 | /* | |
3824 | * If we're below arc_meta_min, always prefer to evict data. | |
3825 | * Otherwise, try to satisfy the requested number of bytes to | |
3826 | * evict from the type which contains older buffers; in an | |
3827 | * effort to keep newer buffers in the cache regardless of their | |
3828 | * type. If we cannot satisfy the number of bytes from this | |
3829 | * type, spill over into the next type. | |
3830 | */ | |
3831 | if (arc_adjust_type(arc_mru) == ARC_BUFC_METADATA && | |
3832 | arc_meta_used > arc_meta_min) { | |
3833 | bytes = arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_METADATA); | |
3834 | total_evicted += bytes; | |
3835 | ||
3836 | /* | |
3837 | * If we couldn't evict our target number of bytes from | |
3838 | * metadata, we try to get the rest from data. | |
3839 | */ | |
3840 | target -= bytes; | |
3841 | ||
3842 | total_evicted += | |
3843 | arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_DATA); | |
3844 | } else { | |
3845 | bytes = arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_DATA); | |
3846 | total_evicted += bytes; | |
3847 | ||
3848 | /* | |
3849 | * If we couldn't evict our target number of bytes from | |
3850 | * data, we try to get the rest from metadata. | |
3851 | */ | |
3852 | target -= bytes; | |
3853 | ||
3854 | total_evicted += | |
3855 | arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_METADATA); | |
34dc7c2f BB |
3856 | } |
3857 | ||
e10b0808 AX |
3858 | /* |
3859 | * Adjust MFU size | |
3860 | * | |
3861 | * Now that we've tried to evict enough from the MRU to get its | |
3862 | * size back to arc_p, if we're still above the target cache | |
3863 | * size, we evict the rest from the MFU. | |
3864 | */ | |
3865 | target = arc_size - arc_c; | |
3866 | ||
3867 | if (arc_adjust_type(arc_mfu) == ARC_BUFC_METADATA && | |
3868 | arc_meta_used > arc_meta_min) { | |
3869 | bytes = arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_METADATA); | |
3870 | total_evicted += bytes; | |
3871 | ||
3872 | /* | |
3873 | * If we couldn't evict our target number of bytes from | |
3874 | * metadata, we try to get the rest from data. | |
3875 | */ | |
3876 | target -= bytes; | |
3877 | ||
3878 | total_evicted += | |
3879 | arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_DATA); | |
3880 | } else { | |
3881 | bytes = arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_DATA); | |
3882 | total_evicted += bytes; | |
3883 | ||
3884 | /* | |
3885 | * If we couldn't evict our target number of bytes from | |
3886 | * data, we try to get the rest from data. | |
3887 | */ | |
3888 | target -= bytes; | |
3889 | ||
3890 | total_evicted += | |
3891 | arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_METADATA); | |
3892 | } | |
3893 | ||
3894 | /* | |
3895 | * Adjust ghost lists | |
3896 | * | |
3897 | * In addition to the above, the ARC also defines target values | |
3898 | * for the ghost lists. The sum of the mru list and mru ghost | |
3899 | * list should never exceed the target size of the cache, and | |
3900 | * the sum of the mru list, mfu list, mru ghost list, and mfu | |
3901 | * ghost list should never exceed twice the target size of the | |
3902 | * cache. The following logic enforces these limits on the ghost | |
3903 | * caches, and evicts from them as needed. | |
3904 | */ | |
3905 | target = refcount_count(&arc_mru->arcs_size) + | |
3906 | refcount_count(&arc_mru_ghost->arcs_size) - arc_c; | |
3907 | ||
3908 | bytes = arc_adjust_impl(arc_mru_ghost, 0, target, ARC_BUFC_DATA); | |
3909 | total_evicted += bytes; | |
3910 | ||
3911 | target -= bytes; | |
3912 | ||
3913 | total_evicted += | |
3914 | arc_adjust_impl(arc_mru_ghost, 0, target, ARC_BUFC_METADATA); | |
3915 | ||
3916 | /* | |
3917 | * We assume the sum of the mru list and mfu list is less than | |
3918 | * or equal to arc_c (we enforced this above), which means we | |
3919 | * can use the simpler of the two equations below: | |
3920 | * | |
3921 | * mru + mfu + mru ghost + mfu ghost <= 2 * arc_c | |
3922 | * mru ghost + mfu ghost <= arc_c | |
3923 | */ | |
3924 | target = refcount_count(&arc_mru_ghost->arcs_size) + | |
3925 | refcount_count(&arc_mfu_ghost->arcs_size) - arc_c; | |
3926 | ||
3927 | bytes = arc_adjust_impl(arc_mfu_ghost, 0, target, ARC_BUFC_DATA); | |
3928 | total_evicted += bytes; | |
3929 | ||
3930 | target -= bytes; | |
3931 | ||
3932 | total_evicted += | |
3933 | arc_adjust_impl(arc_mfu_ghost, 0, target, ARC_BUFC_METADATA); | |
3934 | ||
3935 | return (total_evicted); | |
3936 | } | |
3937 | ||
e10b0808 AX |
3938 | void |
3939 | arc_flush(spa_t *spa, boolean_t retry) | |
3940 | { | |
3941 | uint64_t guid = 0; | |
3942 | ||
3943 | /* | |
cae5b340 | 3944 | * If retry is B_TRUE, a spa must not be specified since we have |
e10b0808 AX |
3945 | * no good way to determine if all of a spa's buffers have been |
3946 | * evicted from an arc state. | |
3947 | */ | |
3948 | ASSERT(!retry || spa == 0); | |
3949 | ||
3950 | if (spa != NULL) | |
3951 | guid = spa_load_guid(spa); | |
3952 | ||
3953 | (void) arc_flush_state(arc_mru, guid, ARC_BUFC_DATA, retry); | |
3954 | (void) arc_flush_state(arc_mru, guid, ARC_BUFC_METADATA, retry); | |
3955 | ||
3956 | (void) arc_flush_state(arc_mfu, guid, ARC_BUFC_DATA, retry); | |
3957 | (void) arc_flush_state(arc_mfu, guid, ARC_BUFC_METADATA, retry); | |
3958 | ||
3959 | (void) arc_flush_state(arc_mru_ghost, guid, ARC_BUFC_DATA, retry); | |
3960 | (void) arc_flush_state(arc_mru_ghost, guid, ARC_BUFC_METADATA, retry); | |
3961 | ||
3962 | (void) arc_flush_state(arc_mfu_ghost, guid, ARC_BUFC_DATA, retry); | |
3963 | (void) arc_flush_state(arc_mfu_ghost, guid, ARC_BUFC_METADATA, retry); | |
34dc7c2f BB |
3964 | } |
3965 | ||
34dc7c2f | 3966 | void |
e10b0808 | 3967 | arc_shrink(int64_t to_free) |
34dc7c2f | 3968 | { |
4d815aed | 3969 | uint64_t c = arc_c; |
34dc7c2f | 3970 | |
4d815aed AX |
3971 | if (c > to_free && c - to_free > arc_c_min) { |
3972 | arc_c = c - to_free; | |
e10b0808 | 3973 | atomic_add_64(&arc_p, -(arc_p >> arc_shrink_shift)); |
34dc7c2f BB |
3974 | if (arc_c > arc_size) |
3975 | arc_c = MAX(arc_size, arc_c_min); | |
3976 | if (arc_p > arc_c) | |
3977 | arc_p = (arc_c >> 1); | |
3978 | ASSERT(arc_c >= arc_c_min); | |
3979 | ASSERT((int64_t)arc_p >= 0); | |
4d815aed AX |
3980 | } else { |
3981 | arc_c = arc_c_min; | |
34dc7c2f BB |
3982 | } |
3983 | ||
3984 | if (arc_size > arc_c) | |
e10b0808 AX |
3985 | (void) arc_adjust(); |
3986 | } | |
3987 | ||
cae5b340 AX |
3988 | /* |
3989 | * Return maximum amount of memory that we could possibly use. Reduced | |
3990 | * to half of all memory in user space which is primarily used for testing. | |
3991 | */ | |
3992 | static uint64_t | |
3993 | arc_all_memory(void) | |
3994 | { | |
3995 | #ifdef _KERNEL | |
8ec27e97 AX |
3996 | #ifdef CONFIG_HIGHMEM |
3997 | return (ptob(totalram_pages - totalhigh_pages)); | |
3998 | #else | |
3999 | return (ptob(totalram_pages)); | |
4000 | #endif /* CONFIG_HIGHMEM */ | |
cae5b340 AX |
4001 | #else |
4002 | return (ptob(physmem) / 2); | |
8ec27e97 | 4003 | #endif /* _KERNEL */ |
cae5b340 AX |
4004 | } |
4005 | ||
8ec27e97 AX |
4006 | /* |
4007 | * Return the amount of memory that is considered free. In user space | |
4008 | * which is primarily used for testing we pretend that free memory ranges | |
4009 | * from 0-20% of all memory. | |
4010 | */ | |
cae5b340 AX |
4011 | static uint64_t |
4012 | arc_free_memory(void) | |
4013 | { | |
8ec27e97 AX |
4014 | #ifdef _KERNEL |
4015 | #ifdef CONFIG_HIGHMEM | |
4016 | struct sysinfo si; | |
4017 | si_meminfo(&si); | |
4018 | return (ptob(si.freeram - si.freehigh)); | |
4019 | #else | |
8ec27e97 | 4020 | return (ptob(nr_free_pages() + |
42f7b73b AX |
4021 | nr_inactive_file_pages() + |
4022 | nr_inactive_anon_pages() + | |
4023 | nr_slab_reclaimable_pages())); | |
4024 | ||
8ec27e97 AX |
4025 | #endif /* CONFIG_HIGHMEM */ |
4026 | #else | |
4027 | return (spa_get_random(arc_all_memory() * 20 / 100)); | |
4028 | #endif /* _KERNEL */ | |
cae5b340 | 4029 | } |
cae5b340 | 4030 | |
e10b0808 AX |
4031 | typedef enum free_memory_reason_t { |
4032 | FMR_UNKNOWN, | |
4033 | FMR_NEEDFREE, | |
4034 | FMR_LOTSFREE, | |
4035 | FMR_SWAPFS_MINFREE, | |
4036 | FMR_PAGES_PP_MAXIMUM, | |
4037 | FMR_HEAP_ARENA, | |
4038 | FMR_ZIO_ARENA, | |
4039 | } free_memory_reason_t; | |
4040 | ||
4041 | int64_t last_free_memory; | |
4042 | free_memory_reason_t last_free_reason; | |
4043 | ||
4044 | #ifdef _KERNEL | |
4045 | /* | |
4046 | * Additional reserve of pages for pp_reserve. | |
4047 | */ | |
4048 | int64_t arc_pages_pp_reserve = 64; | |
4049 | ||
4050 | /* | |
4051 | * Additional reserve of pages for swapfs. | |
4052 | */ | |
4053 | int64_t arc_swapfs_reserve = 64; | |
4054 | #endif /* _KERNEL */ | |
4055 | ||
4056 | /* | |
4057 | * Return the amount of memory that can be consumed before reclaim will be | |
4058 | * needed. Positive if there is sufficient free memory, negative indicates | |
4059 | * the amount of memory that needs to be freed up. | |
4060 | */ | |
4061 | static int64_t | |
4062 | arc_available_memory(void) | |
4063 | { | |
4064 | int64_t lowest = INT64_MAX; | |
4065 | free_memory_reason_t r = FMR_UNKNOWN; | |
4066 | #ifdef _KERNEL | |
4067 | int64_t n; | |
4068 | #ifdef __linux__ | |
8ec27e97 AX |
4069 | #ifdef freemem |
4070 | #undef freemem | |
4071 | #endif | |
e10b0808 AX |
4072 | pgcnt_t needfree = btop(arc_need_free); |
4073 | pgcnt_t lotsfree = btop(arc_sys_free); | |
4074 | pgcnt_t desfree = 0; | |
8ec27e97 | 4075 | pgcnt_t freemem = btop(arc_free_memory()); |
cae5b340 AX |
4076 | #endif |
4077 | ||
e10b0808 AX |
4078 | if (needfree > 0) { |
4079 | n = PAGESIZE * (-needfree); | |
4080 | if (n < lowest) { | |
4081 | lowest = n; | |
4082 | r = FMR_NEEDFREE; | |
4083 | } | |
4084 | } | |
4085 | ||
4086 | /* | |
4087 | * check that we're out of range of the pageout scanner. It starts to | |
4088 | * schedule paging if freemem is less than lotsfree and needfree. | |
4089 | * lotsfree is the high-water mark for pageout, and needfree is the | |
4090 | * number of needed free pages. We add extra pages here to make sure | |
4091 | * the scanner doesn't start up while we're freeing memory. | |
4092 | */ | |
8ec27e97 | 4093 | n = PAGESIZE * (freemem - lotsfree - needfree - desfree); |
e10b0808 AX |
4094 | if (n < lowest) { |
4095 | lowest = n; | |
4096 | r = FMR_LOTSFREE; | |
4097 | } | |
4098 | ||
4099 | #ifndef __linux__ | |
4100 | /* | |
4101 | * check to make sure that swapfs has enough space so that anon | |
4102 | * reservations can still succeed. anon_resvmem() checks that the | |
4103 | * availrmem is greater than swapfs_minfree, and the number of reserved | |
4104 | * swap pages. We also add a bit of extra here just to prevent | |
4105 | * circumstances from getting really dire. | |
4106 | */ | |
4107 | n = PAGESIZE * (availrmem - swapfs_minfree - swapfs_reserve - | |
4108 | desfree - arc_swapfs_reserve); | |
4109 | if (n < lowest) { | |
4110 | lowest = n; | |
4111 | r = FMR_SWAPFS_MINFREE; | |
4112 | } | |
4113 | ||
e10b0808 AX |
4114 | /* |
4115 | * Check that we have enough availrmem that memory locking (e.g., via | |
4116 | * mlock(3C) or memcntl(2)) can still succeed. (pages_pp_maximum | |
4117 | * stores the number of pages that cannot be locked; when availrmem | |
4118 | * drops below pages_pp_maximum, page locking mechanisms such as | |
4119 | * page_pp_lock() will fail.) | |
4120 | */ | |
4121 | n = PAGESIZE * (availrmem - pages_pp_maximum - | |
4122 | arc_pages_pp_reserve); | |
4123 | if (n < lowest) { | |
4124 | lowest = n; | |
4125 | r = FMR_PAGES_PP_MAXIMUM; | |
4126 | } | |
4127 | #endif | |
4128 | ||
8ec27e97 | 4129 | #if defined(_ILP32) |
e10b0808 | 4130 | /* |
8ec27e97 | 4131 | * If we're on a 32-bit platform, it's possible that we'll exhaust the |
e10b0808 AX |
4132 | * kernel heap space before we ever run out of available physical |
4133 | * memory. Most checks of the size of the heap_area compare against | |
4134 | * tune.t_minarmem, which is the minimum available real memory that we | |
4135 | * can have in the system. However, this is generally fixed at 25 pages | |
4136 | * which is so low that it's useless. In this comparison, we seek to | |
4137 | * calculate the total heap-size, and reclaim if more than 3/4ths of the | |
4138 | * heap is allocated. (Or, in the calculation, if less than 1/4th is | |
4139 | * free) | |
4140 | */ | |
4141 | n = vmem_size(heap_arena, VMEM_FREE) - | |
4142 | (vmem_size(heap_arena, VMEM_FREE | VMEM_ALLOC) >> 2); | |
4143 | if (n < lowest) { | |
4144 | lowest = n; | |
4145 | r = FMR_HEAP_ARENA; | |
4146 | } | |
4147 | #endif | |
4148 | ||
4149 | /* | |
4150 | * If zio data pages are being allocated out of a separate heap segment, | |
4151 | * then enforce that the size of available vmem for this arena remains | |
cae5b340 | 4152 | * above about 1/4th (1/(2^arc_zio_arena_free_shift)) free. |
e10b0808 | 4153 | * |
cae5b340 AX |
4154 | * Note that reducing the arc_zio_arena_free_shift keeps more virtual |
4155 | * memory (in the zio_arena) free, which can avoid memory | |
4156 | * fragmentation issues. | |
e10b0808 AX |
4157 | */ |
4158 | if (zio_arena != NULL) { | |
cae5b340 AX |
4159 | n = (int64_t)vmem_size(zio_arena, VMEM_FREE) - |
4160 | (vmem_size(zio_arena, VMEM_ALLOC) >> | |
4161 | arc_zio_arena_free_shift); | |
e10b0808 AX |
4162 | if (n < lowest) { |
4163 | lowest = n; | |
4164 | r = FMR_ZIO_ARENA; | |
4165 | } | |
4166 | } | |
4167 | #else /* _KERNEL */ | |
4168 | /* Every 100 calls, free a small amount */ | |
4169 | if (spa_get_random(100) == 0) | |
4170 | lowest = -1024; | |
4171 | #endif /* _KERNEL */ | |
4172 | ||
4173 | last_free_memory = lowest; | |
4174 | last_free_reason = r; | |
4175 | ||
4176 | return (lowest); | |
4177 | } | |
4178 | ||
4179 | /* | |
4180 | * Determine if the system is under memory pressure and is asking | |
cae5b340 | 4181 | * to reclaim memory. A return value of B_TRUE indicates that the system |
e10b0808 AX |
4182 | * is under memory pressure and that the arc should adjust accordingly. |
4183 | */ | |
4184 | static boolean_t | |
4185 | arc_reclaim_needed(void) | |
4186 | { | |
4187 | return (arc_available_memory() < 0); | |
34dc7c2f BB |
4188 | } |
4189 | ||
34dc7c2f | 4190 | static void |
e10b0808 | 4191 | arc_kmem_reap_now(void) |
34dc7c2f BB |
4192 | { |
4193 | size_t i; | |
4194 | kmem_cache_t *prev_cache = NULL; | |
4195 | kmem_cache_t *prev_data_cache = NULL; | |
4196 | extern kmem_cache_t *zio_buf_cache[]; | |
4197 | extern kmem_cache_t *zio_data_buf_cache[]; | |
e10b0808 | 4198 | extern kmem_cache_t *range_seg_cache; |
34dc7c2f | 4199 | |
8ec27e97 | 4200 | #ifdef _KERNEL |
e10b0808 AX |
4201 | if ((arc_meta_used >= arc_meta_limit) && zfs_arc_meta_prune) { |
4202 | /* | |
4203 | * We are exceeding our meta-data cache limit. | |
4204 | * Prune some entries to release holds on meta-data. | |
4205 | */ | |
4206 | arc_prune_async(zfs_arc_meta_prune); | |
4207 | } | |
8ec27e97 AX |
4208 | #if defined(_ILP32) |
4209 | /* | |
4210 | * Reclaim unused memory from all kmem caches. | |
4211 | */ | |
4212 | kmem_reap(); | |
4213 | #endif | |
4214 | #endif | |
34dc7c2f BB |
4215 | |
4216 | for (i = 0; i < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; i++) { | |
8ec27e97 | 4217 | #if defined(_ILP32) |
94a40997 AX |
4218 | /* reach upper limit of cache size on 32-bit */ |
4219 | if (zio_buf_cache[i] == NULL) | |
4220 | break; | |
4221 | #endif | |
34dc7c2f BB |
4222 | if (zio_buf_cache[i] != prev_cache) { |
4223 | prev_cache = zio_buf_cache[i]; | |
4224 | kmem_cache_reap_now(zio_buf_cache[i]); | |
4225 | } | |
4226 | if (zio_data_buf_cache[i] != prev_data_cache) { | |
4227 | prev_data_cache = zio_data_buf_cache[i]; | |
4228 | kmem_cache_reap_now(zio_data_buf_cache[i]); | |
4229 | } | |
4230 | } | |
4231 | kmem_cache_reap_now(buf_cache); | |
e10b0808 AX |
4232 | kmem_cache_reap_now(hdr_full_cache); |
4233 | kmem_cache_reap_now(hdr_l2only_cache); | |
4234 | kmem_cache_reap_now(range_seg_cache); | |
4235 | ||
4236 | if (zio_arena != NULL) { | |
4237 | /* | |
4238 | * Ask the vmem arena to reclaim unused memory from its | |
4239 | * quantum caches. | |
4240 | */ | |
4241 | vmem_qcache_reap(zio_arena); | |
4242 | } | |
34dc7c2f BB |
4243 | } |
4244 | ||
302f753f | 4245 | /* |
cae5b340 | 4246 | * Threads can block in arc_get_data_impl() waiting for this thread to evict |
e10b0808 | 4247 | * enough data and signal them to proceed. When this happens, the threads in |
cae5b340 | 4248 | * arc_get_data_impl() are sleeping while holding the hash lock for their |
e10b0808 AX |
4249 | * particular arc header. Thus, we must be careful to never sleep on a |
4250 | * hash lock in this thread. This is to prevent the following deadlock: | |
4251 | * | |
cae5b340 | 4252 | * - Thread A sleeps on CV in arc_get_data_impl() holding hash lock "L", |
e10b0808 AX |
4253 | * waiting for the reclaim thread to signal it. |
4254 | * | |
4255 | * - arc_reclaim_thread() tries to acquire hash lock "L" using mutex_enter, | |
4256 | * fails, and goes to sleep forever. | |
4257 | * | |
4258 | * This possible deadlock is avoided by always acquiring a hash lock | |
4259 | * using mutex_tryenter() from arc_reclaim_thread(). | |
302f753f | 4260 | */ |
34dc7c2f | 4261 | static void |
e10b0808 | 4262 | arc_reclaim_thread(void) |
34dc7c2f | 4263 | { |
e10b0808 | 4264 | fstrans_cookie_t cookie = spl_fstrans_mark(); |
cae5b340 | 4265 | hrtime_t growtime = 0; |
34dc7c2f BB |
4266 | callb_cpr_t cpr; |
4267 | ||
e10b0808 | 4268 | CALLB_CPR_INIT(&cpr, &arc_reclaim_lock, callb_generic_cpr, FTAG); |
34dc7c2f | 4269 | |
e10b0808 AX |
4270 | mutex_enter(&arc_reclaim_lock); |
4271 | while (!arc_reclaim_thread_exit) { | |
4272 | int64_t to_free; | |
e10b0808 | 4273 | uint64_t evicted = 0; |
cae5b340 | 4274 | uint64_t need_free = arc_need_free; |
e10b0808 | 4275 | arc_tuning_update(); |
34dc7c2f | 4276 | |
cae5b340 AX |
4277 | /* |
4278 | * This is necessary in order for the mdb ::arc dcmd to | |
4279 | * show up to date information. Since the ::arc command | |
4280 | * does not call the kstat's update function, without | |
4281 | * this call, the command may show stale stats for the | |
4282 | * anon, mru, mru_ghost, mfu, and mfu_ghost lists. Even | |
4283 | * with this change, the data might be up to 1 second | |
4284 | * out of date; but that should suffice. The arc_state_t | |
4285 | * structures can be queried directly if more accurate | |
4286 | * information is needed. | |
4287 | */ | |
4288 | #ifndef __linux__ | |
4289 | if (arc_ksp != NULL) | |
4290 | arc_ksp->ks_update(arc_ksp, KSTAT_READ); | |
4291 | #endif | |
e10b0808 | 4292 | mutex_exit(&arc_reclaim_lock); |
34dc7c2f | 4293 | |
cae5b340 AX |
4294 | /* |
4295 | * We call arc_adjust() before (possibly) calling | |
4296 | * arc_kmem_reap_now(), so that we can wake up | |
4297 | * arc_get_data_buf() sooner. | |
4298 | */ | |
4299 | evicted = arc_adjust(); | |
4300 | ||
4301 | int64_t free_memory = arc_available_memory(); | |
e10b0808 | 4302 | if (free_memory < 0) { |
34dc7c2f | 4303 | |
e10b0808 | 4304 | arc_no_grow = B_TRUE; |
b128c09f | 4305 | arc_warm = B_TRUE; |
e10b0808 AX |
4306 | |
4307 | /* | |
4308 | * Wait at least zfs_grow_retry (default 5) seconds | |
4309 | * before considering growing. | |
4310 | */ | |
cae5b340 | 4311 | growtime = gethrtime() + SEC2NSEC(arc_grow_retry); |
e10b0808 AX |
4312 | |
4313 | arc_kmem_reap_now(); | |
4314 | ||
4315 | /* | |
4316 | * If we are still low on memory, shrink the ARC | |
4317 | * so that we have arc_shrink_min free space. | |
4318 | */ | |
4319 | free_memory = arc_available_memory(); | |
4320 | ||
4321 | to_free = (arc_c >> arc_shrink_shift) - free_memory; | |
4322 | if (to_free > 0) { | |
4323 | #ifdef _KERNEL | |
cae5b340 | 4324 | to_free = MAX(to_free, need_free); |
e10b0808 AX |
4325 | #endif |
4326 | arc_shrink(to_free); | |
4327 | } | |
4328 | } else if (free_memory < arc_c >> arc_no_grow_shift) { | |
4329 | arc_no_grow = B_TRUE; | |
cae5b340 | 4330 | } else if (gethrtime() >= growtime) { |
e10b0808 | 4331 | arc_no_grow = B_FALSE; |
302f753f | 4332 | } |
34dc7c2f | 4333 | |
e10b0808 | 4334 | mutex_enter(&arc_reclaim_lock); |
6a8f9b6b | 4335 | |
e10b0808 AX |
4336 | /* |
4337 | * If evicted is zero, we couldn't evict anything via | |
4338 | * arc_adjust(). This could be due to hash lock | |
4339 | * collisions, but more likely due to the majority of | |
4340 | * arc buffers being unevictable. Therefore, even if | |
4341 | * arc_size is above arc_c, another pass is unlikely to | |
4342 | * be helpful and could potentially cause us to enter an | |
4343 | * infinite loop. | |
4344 | */ | |
4345 | if (arc_size <= arc_c || evicted == 0) { | |
4346 | /* | |
4347 | * We're either no longer overflowing, or we | |
4348 | * can't evict anything more, so we should wake | |
cae5b340 AX |
4349 | * up any threads before we go to sleep and remove |
4350 | * the bytes we were working on from arc_need_free | |
4351 | * since nothing more will be done here. | |
e10b0808 AX |
4352 | */ |
4353 | cv_broadcast(&arc_reclaim_waiters_cv); | |
cae5b340 | 4354 | ARCSTAT_INCR(arcstat_need_free, -need_free); |
34dc7c2f | 4355 | |
e10b0808 AX |
4356 | /* |
4357 | * Block until signaled, or after one second (we | |
4358 | * might need to perform arc_kmem_reap_now() | |
4359 | * even if we aren't being signalled) | |
4360 | */ | |
4361 | CALLB_CPR_SAFE_BEGIN(&cpr); | |
cae5b340 AX |
4362 | (void) cv_timedwait_sig_hires(&arc_reclaim_thread_cv, |
4363 | &arc_reclaim_lock, SEC2NSEC(1), MSEC2NSEC(1), 0); | |
e10b0808 AX |
4364 | CALLB_CPR_SAFE_END(&cpr, &arc_reclaim_lock); |
4365 | } | |
4366 | } | |
34dc7c2f | 4367 | |
cae5b340 | 4368 | arc_reclaim_thread_exit = B_FALSE; |
e10b0808 AX |
4369 | cv_broadcast(&arc_reclaim_thread_cv); |
4370 | CALLB_CPR_EXIT(&cpr); /* drops arc_reclaim_lock */ | |
4371 | spl_fstrans_unmark(cookie); | |
4372 | thread_exit(); | |
4373 | } | |
c06d4368 | 4374 | |
7cb67b45 BB |
4375 | #ifdef _KERNEL |
4376 | /* | |
302f753f BB |
4377 | * Determine the amount of memory eligible for eviction contained in the |
4378 | * ARC. All clean data reported by the ghost lists can always be safely | |
4379 | * evicted. Due to arc_c_min, the same does not hold for all clean data | |
4380 | * contained by the regular mru and mfu lists. | |
4381 | * | |
4382 | * In the case of the regular mru and mfu lists, we need to report as | |
4383 | * much clean data as possible, such that evicting that same reported | |
4384 | * data will not bring arc_size below arc_c_min. Thus, in certain | |
4385 | * circumstances, the total amount of clean data in the mru and mfu | |
4386 | * lists might not actually be evictable. | |
4387 | * | |
4388 | * The following two distinct cases are accounted for: | |
4389 | * | |
4390 | * 1. The sum of the amount of dirty data contained by both the mru and | |
4391 | * mfu lists, plus the ARC's other accounting (e.g. the anon list), | |
4392 | * is greater than or equal to arc_c_min. | |
4393 | * (i.e. amount of dirty data >= arc_c_min) | |
4394 | * | |
4395 | * This is the easy case; all clean data contained by the mru and mfu | |
4396 | * lists is evictable. Evicting all clean data can only drop arc_size | |
4397 | * to the amount of dirty data, which is greater than arc_c_min. | |
4398 | * | |
4399 | * 2. The sum of the amount of dirty data contained by both the mru and | |
4400 | * mfu lists, plus the ARC's other accounting (e.g. the anon list), | |
4401 | * is less than arc_c_min. | |
4402 | * (i.e. arc_c_min > amount of dirty data) | |
4403 | * | |
4404 | * 2.1. arc_size is greater than or equal arc_c_min. | |
4405 | * (i.e. arc_size >= arc_c_min > amount of dirty data) | |
4406 | * | |
4407 | * In this case, not all clean data from the regular mru and mfu | |
4408 | * lists is actually evictable; we must leave enough clean data | |
4409 | * to keep arc_size above arc_c_min. Thus, the maximum amount of | |
4410 | * evictable data from the two lists combined, is exactly the | |
4411 | * difference between arc_size and arc_c_min. | |
4412 | * | |
4413 | * 2.2. arc_size is less than arc_c_min | |
4414 | * (i.e. arc_c_min > arc_size > amount of dirty data) | |
4415 | * | |
4416 | * In this case, none of the data contained in the mru and mfu | |
4417 | * lists is evictable, even if it's clean. Since arc_size is | |
4418 | * already below arc_c_min, evicting any more would only | |
4419 | * increase this negative difference. | |
7cb67b45 | 4420 | */ |
302f753f | 4421 | static uint64_t |
cae5b340 AX |
4422 | arc_evictable_memory(void) |
4423 | { | |
302f753f | 4424 | uint64_t arc_clean = |
cae5b340 AX |
4425 | refcount_count(&arc_mru->arcs_esize[ARC_BUFC_DATA]) + |
4426 | refcount_count(&arc_mru->arcs_esize[ARC_BUFC_METADATA]) + | |
4427 | refcount_count(&arc_mfu->arcs_esize[ARC_BUFC_DATA]) + | |
4428 | refcount_count(&arc_mfu->arcs_esize[ARC_BUFC_METADATA]); | |
302f753f BB |
4429 | uint64_t arc_dirty = MAX((int64_t)arc_size - (int64_t)arc_clean, 0); |
4430 | ||
cae5b340 AX |
4431 | /* |
4432 | * Scale reported evictable memory in proportion to page cache, cap | |
4433 | * at specified min/max. | |
4434 | */ | |
42f7b73b | 4435 | uint64_t min = (ptob(nr_file_pages()) / 100) * zfs_arc_pc_percent; |
cae5b340 | 4436 | min = MAX(arc_c_min, MIN(arc_c_max, min)); |
302f753f | 4437 | |
cae5b340 AX |
4438 | if (arc_dirty >= min) |
4439 | return (arc_clean); | |
4440 | ||
4441 | return (MAX((int64_t)arc_size - (int64_t)min, 0)); | |
302f753f BB |
4442 | } |
4443 | ||
ea04106b AX |
4444 | /* |
4445 | * If sc->nr_to_scan is zero, the caller is requesting a query of the | |
4446 | * number of objects which can potentially be freed. If it is nonzero, | |
4447 | * the request is to free that many objects. | |
4448 | * | |
4449 | * Linux kernels >= 3.12 have the count_objects and scan_objects callbacks | |
4450 | * in struct shrinker and also require the shrinker to return the number | |
4451 | * of objects freed. | |
4452 | * | |
4453 | * Older kernels require the shrinker to return the number of freeable | |
4454 | * objects following the freeing of nr_to_free. | |
4455 | */ | |
4456 | static spl_shrinker_t | |
7e7baeca | 4457 | __arc_shrinker_func(struct shrinker *shrink, struct shrink_control *sc) |
7cb67b45 | 4458 | { |
ea04106b | 4459 | int64_t pages; |
7cb67b45 | 4460 | |
302f753f BB |
4461 | /* The arc is considered warm once reclaim has occurred */ |
4462 | if (unlikely(arc_warm == B_FALSE)) | |
4463 | arc_warm = B_TRUE; | |
7cb67b45 | 4464 | |
302f753f | 4465 | /* Return the potential number of reclaimable pages */ |
ea04106b | 4466 | pages = btop((int64_t)arc_evictable_memory()); |
302f753f BB |
4467 | if (sc->nr_to_scan == 0) |
4468 | return (pages); | |
3fd70ee6 BB |
4469 | |
4470 | /* Not allowed to perform filesystem reclaim */ | |
7e7baeca | 4471 | if (!(sc->gfp_mask & __GFP_FS)) |
ea04106b | 4472 | return (SHRINK_STOP); |
3fd70ee6 | 4473 | |
7cb67b45 | 4474 | /* Reclaim in progress */ |
cae5b340 AX |
4475 | if (mutex_tryenter(&arc_reclaim_lock) == 0) { |
4476 | ARCSTAT_INCR(arcstat_need_free, ptob(sc->nr_to_scan)); | |
4477 | return (0); | |
4478 | } | |
7cb67b45 | 4479 | |
e10b0808 AX |
4480 | mutex_exit(&arc_reclaim_lock); |
4481 | ||
302f753f BB |
4482 | /* |
4483 | * Evict the requested number of pages by shrinking arc_c the | |
cae5b340 | 4484 | * requested amount. |
302f753f BB |
4485 | */ |
4486 | if (pages > 0) { | |
e10b0808 | 4487 | arc_shrink(ptob(sc->nr_to_scan)); |
cae5b340 AX |
4488 | if (current_is_kswapd()) |
4489 | arc_kmem_reap_now(); | |
ea04106b | 4490 | #ifdef HAVE_SPLIT_SHRINKER_CALLBACK |
cae5b340 AX |
4491 | pages = MAX((int64_t)pages - |
4492 | (int64_t)btop(arc_evictable_memory()), 0); | |
ea04106b AX |
4493 | #else |
4494 | pages = btop(arc_evictable_memory()); | |
4495 | #endif | |
cae5b340 AX |
4496 | /* |
4497 | * We've shrunk what we can, wake up threads. | |
4498 | */ | |
4499 | cv_broadcast(&arc_reclaim_waiters_cv); | |
4500 | } else | |
ea04106b | 4501 | pages = SHRINK_STOP; |
e10b0808 | 4502 | |
302f753f BB |
4503 | /* |
4504 | * When direct reclaim is observed it usually indicates a rapid | |
4505 | * increase in memory pressure. This occurs because the kswapd | |
4506 | * threads were unable to asynchronously keep enough free memory | |
4507 | * available. In this case set arc_no_grow to briefly pause arc | |
4508 | * growth to avoid compounding the memory pressure. | |
4509 | */ | |
7cb67b45 | 4510 | if (current_is_kswapd()) { |
302f753f | 4511 | ARCSTAT_BUMP(arcstat_memory_indirect_count); |
7cb67b45 | 4512 | } else { |
302f753f | 4513 | arc_no_grow = B_TRUE; |
cae5b340 | 4514 | arc_kmem_reap_now(); |
302f753f | 4515 | ARCSTAT_BUMP(arcstat_memory_direct_count); |
7cb67b45 BB |
4516 | } |
4517 | ||
ea04106b | 4518 | return (pages); |
7cb67b45 | 4519 | } |
7e7baeca | 4520 | SPL_SHRINKER_CALLBACK_WRAPPER(arc_shrinker_func); |
7cb67b45 BB |
4521 | |
4522 | SPL_SHRINKER_DECLARE(arc_shrinker, arc_shrinker_func, DEFAULT_SEEKS); | |
4523 | #endif /* _KERNEL */ | |
4524 | ||
34dc7c2f BB |
4525 | /* |
4526 | * Adapt arc info given the number of bytes we are trying to add and | |
cae5b340 | 4527 | * the state that we are coming from. This function is only called |
34dc7c2f BB |
4528 | * when we are adding new content to the cache. |
4529 | */ | |
4530 | static void | |
4531 | arc_adapt(int bytes, arc_state_t *state) | |
4532 | { | |
4533 | int mult; | |
e10b0808 AX |
4534 | uint64_t arc_p_min = (arc_c >> arc_p_min_shift); |
4535 | int64_t mrug_size = refcount_count(&arc_mru_ghost->arcs_size); | |
4536 | int64_t mfug_size = refcount_count(&arc_mfu_ghost->arcs_size); | |
34dc7c2f BB |
4537 | |
4538 | if (state == arc_l2c_only) | |
4539 | return; | |
4540 | ||
4541 | ASSERT(bytes > 0); | |
4542 | /* | |
4543 | * Adapt the target size of the MRU list: | |
4544 | * - if we just hit in the MRU ghost list, then increase | |
4545 | * the target size of the MRU list. | |
4546 | * - if we just hit in the MFU ghost list, then increase | |
4547 | * the target size of the MFU list by decreasing the | |
4548 | * target size of the MRU list. | |
4549 | */ | |
4550 | if (state == arc_mru_ghost) { | |
e10b0808 | 4551 | mult = (mrug_size >= mfug_size) ? 1 : (mfug_size / mrug_size); |
ea04106b AX |
4552 | if (!zfs_arc_p_dampener_disable) |
4553 | mult = MIN(mult, 10); /* avoid wild arc_p adjustment */ | |
4554 | ||
e10b0808 | 4555 | arc_p = MIN(arc_c - arc_p_min, arc_p + bytes * mult); |
34dc7c2f | 4556 | } else if (state == arc_mfu_ghost) { |
d164b209 BB |
4557 | uint64_t delta; |
4558 | ||
e10b0808 | 4559 | mult = (mfug_size >= mrug_size) ? 1 : (mrug_size / mfug_size); |
ea04106b AX |
4560 | if (!zfs_arc_p_dampener_disable) |
4561 | mult = MIN(mult, 10); | |
34dc7c2f | 4562 | |
d164b209 | 4563 | delta = MIN(bytes * mult, arc_p); |
e10b0808 | 4564 | arc_p = MAX(arc_p_min, arc_p - delta); |
34dc7c2f BB |
4565 | } |
4566 | ASSERT((int64_t)arc_p >= 0); | |
4567 | ||
e10b0808 AX |
4568 | if (arc_reclaim_needed()) { |
4569 | cv_signal(&arc_reclaim_thread_cv); | |
4570 | return; | |
4571 | } | |
4572 | ||
34dc7c2f BB |
4573 | if (arc_no_grow) |
4574 | return; | |
4575 | ||
4576 | if (arc_c >= arc_c_max) | |
4577 | return; | |
4578 | ||
4579 | /* | |
4580 | * If we're within (2 * maxblocksize) bytes of the target | |
4581 | * cache size, increment the target cache size | |
4582 | */ | |
4d815aed | 4583 | ASSERT3U(arc_c, >=, 2ULL << SPA_MAXBLOCKSHIFT); |
e10b0808 | 4584 | if (arc_size >= arc_c - (2ULL << SPA_MAXBLOCKSHIFT)) { |
34dc7c2f BB |
4585 | atomic_add_64(&arc_c, (int64_t)bytes); |
4586 | if (arc_c > arc_c_max) | |
4587 | arc_c = arc_c_max; | |
4588 | else if (state == arc_anon) | |
4589 | atomic_add_64(&arc_p, (int64_t)bytes); | |
4590 | if (arc_p > arc_c) | |
4591 | arc_p = arc_c; | |
4592 | } | |
4593 | ASSERT((int64_t)arc_p >= 0); | |
4594 | } | |
4595 | ||
4596 | /* | |
e10b0808 AX |
4597 | * Check if arc_size has grown past our upper threshold, determined by |
4598 | * zfs_arc_overflow_shift. | |
34dc7c2f | 4599 | */ |
e10b0808 AX |
4600 | static boolean_t |
4601 | arc_is_overflowing(void) | |
34dc7c2f | 4602 | { |
e10b0808 AX |
4603 | /* Always allow at least one block of overflow */ |
4604 | uint64_t overflow = MAX(SPA_MAXBLOCKSIZE, | |
4605 | arc_c >> zfs_arc_overflow_shift); | |
34dc7c2f | 4606 | |
e10b0808 | 4607 | return (arc_size >= arc_c + overflow); |
34dc7c2f BB |
4608 | } |
4609 | ||
cae5b340 AX |
4610 | static abd_t * |
4611 | arc_get_data_abd(arc_buf_hdr_t *hdr, uint64_t size, void *tag) | |
4612 | { | |
4613 | arc_buf_contents_t type = arc_buf_type(hdr); | |
4614 | ||
4615 | arc_get_data_impl(hdr, size, tag); | |
4616 | if (type == ARC_BUFC_METADATA) { | |
4617 | return (abd_alloc(size, B_TRUE)); | |
4618 | } else { | |
4619 | ASSERT(type == ARC_BUFC_DATA); | |
4620 | return (abd_alloc(size, B_FALSE)); | |
4621 | } | |
4622 | } | |
4623 | ||
4624 | static void * | |
4625 | arc_get_data_buf(arc_buf_hdr_t *hdr, uint64_t size, void *tag) | |
4626 | { | |
4627 | arc_buf_contents_t type = arc_buf_type(hdr); | |
4628 | ||
4629 | arc_get_data_impl(hdr, size, tag); | |
4630 | if (type == ARC_BUFC_METADATA) { | |
4631 | return (zio_buf_alloc(size)); | |
4632 | } else { | |
4633 | ASSERT(type == ARC_BUFC_DATA); | |
4634 | return (zio_data_buf_alloc(size)); | |
4635 | } | |
4636 | } | |
4637 | ||
34dc7c2f | 4638 | /* |
cae5b340 AX |
4639 | * Allocate a block and return it to the caller. If we are hitting the |
4640 | * hard limit for the cache size, we must sleep, waiting for the eviction | |
4641 | * thread to catch up. If we're past the target size but below the hard | |
4642 | * limit, we'll only signal the reclaim thread and continue on. | |
34dc7c2f BB |
4643 | */ |
4644 | static void | |
cae5b340 | 4645 | arc_get_data_impl(arc_buf_hdr_t *hdr, uint64_t size, void *tag) |
34dc7c2f | 4646 | { |
cae5b340 AX |
4647 | arc_state_t *state = hdr->b_l1hdr.b_state; |
4648 | arc_buf_contents_t type = arc_buf_type(hdr); | |
34dc7c2f BB |
4649 | |
4650 | arc_adapt(size, state); | |
4651 | ||
4652 | /* | |
e10b0808 AX |
4653 | * If arc_size is currently overflowing, and has grown past our |
4654 | * upper limit, we must be adding data faster than the evict | |
4655 | * thread can evict. Thus, to ensure we don't compound the | |
4656 | * problem by adding more data and forcing arc_size to grow even | |
4657 | * further past it's target size, we halt and wait for the | |
4658 | * eviction thread to catch up. | |
4659 | * | |
4660 | * It's also possible that the reclaim thread is unable to evict | |
4661 | * enough buffers to get arc_size below the overflow limit (e.g. | |
4662 | * due to buffers being un-evictable, or hash lock collisions). | |
4663 | * In this case, we want to proceed regardless if we're | |
4664 | * overflowing; thus we don't use a while loop here. | |
34dc7c2f | 4665 | */ |
e10b0808 AX |
4666 | if (arc_is_overflowing()) { |
4667 | mutex_enter(&arc_reclaim_lock); | |
4668 | ||
4669 | /* | |
4670 | * Now that we've acquired the lock, we may no longer be | |
4671 | * over the overflow limit, lets check. | |
4672 | * | |
4673 | * We're ignoring the case of spurious wake ups. If that | |
4674 | * were to happen, it'd let this thread consume an ARC | |
4675 | * buffer before it should have (i.e. before we're under | |
4676 | * the overflow limit and were signalled by the reclaim | |
4677 | * thread). As long as that is a rare occurrence, it | |
4678 | * shouldn't cause any harm. | |
4679 | */ | |
4680 | if (arc_is_overflowing()) { | |
4681 | cv_signal(&arc_reclaim_thread_cv); | |
4682 | cv_wait(&arc_reclaim_waiters_cv, &arc_reclaim_lock); | |
34dc7c2f | 4683 | } |
34dc7c2f | 4684 | |
e10b0808 | 4685 | mutex_exit(&arc_reclaim_lock); |
34dc7c2f | 4686 | } |
ab26409d | 4687 | |
cae5b340 | 4688 | VERIFY3U(hdr->b_type, ==, type); |
ea04106b | 4689 | if (type == ARC_BUFC_METADATA) { |
e10b0808 AX |
4690 | arc_space_consume(size, ARC_SPACE_META); |
4691 | } else { | |
e10b0808 | 4692 | arc_space_consume(size, ARC_SPACE_DATA); |
ea04106b AX |
4693 | } |
4694 | ||
34dc7c2f BB |
4695 | /* |
4696 | * Update the state size. Note that ghost states have a | |
4697 | * "ghost size" and so don't need to be updated. | |
4698 | */ | |
cae5b340 | 4699 | if (!GHOST_STATE(state)) { |
34dc7c2f | 4700 | |
cae5b340 | 4701 | (void) refcount_add_many(&state->arcs_size, size, tag); |
e10b0808 AX |
4702 | |
4703 | /* | |
4704 | * If this is reached via arc_read, the link is | |
4705 | * protected by the hash lock. If reached via | |
4706 | * arc_buf_alloc, the header should not be accessed by | |
4707 | * any other thread. And, if reached via arc_read_done, | |
4708 | * the hash lock will protect it if it's found in the | |
4709 | * hash table; otherwise no other thread should be | |
4710 | * trying to [add|remove]_reference it. | |
4711 | */ | |
4712 | if (multilist_link_active(&hdr->b_l1hdr.b_arc_node)) { | |
4713 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); | |
cae5b340 AX |
4714 | (void) refcount_add_many(&state->arcs_esize[type], |
4715 | size, tag); | |
34dc7c2f | 4716 | } |
cae5b340 | 4717 | |
34dc7c2f BB |
4718 | /* |
4719 | * If we are growing the cache, and we are adding anonymous | |
4720 | * data, and we have outgrown arc_p, update arc_p | |
4721 | */ | |
e10b0808 AX |
4722 | if (arc_size < arc_c && hdr->b_l1hdr.b_state == arc_anon && |
4723 | (refcount_count(&arc_anon->arcs_size) + | |
4724 | refcount_count(&arc_mru->arcs_size) > arc_p)) | |
34dc7c2f BB |
4725 | arc_p = MIN(arc_c, arc_p + size); |
4726 | } | |
4727 | } | |
4728 | ||
cae5b340 AX |
4729 | static void |
4730 | arc_free_data_abd(arc_buf_hdr_t *hdr, abd_t *abd, uint64_t size, void *tag) | |
4731 | { | |
4732 | arc_free_data_impl(hdr, size, tag); | |
4733 | abd_free(abd); | |
4734 | } | |
4735 | ||
4736 | static void | |
4737 | arc_free_data_buf(arc_buf_hdr_t *hdr, void *buf, uint64_t size, void *tag) | |
4738 | { | |
4739 | arc_buf_contents_t type = arc_buf_type(hdr); | |
4740 | ||
4741 | arc_free_data_impl(hdr, size, tag); | |
4742 | if (type == ARC_BUFC_METADATA) { | |
4743 | zio_buf_free(buf, size); | |
4744 | } else { | |
4745 | ASSERT(type == ARC_BUFC_DATA); | |
4746 | zio_data_buf_free(buf, size); | |
4747 | } | |
4748 | } | |
4749 | ||
4750 | /* | |
4751 | * Free the arc data buffer. | |
4752 | */ | |
4753 | static void | |
4754 | arc_free_data_impl(arc_buf_hdr_t *hdr, uint64_t size, void *tag) | |
4755 | { | |
4756 | arc_state_t *state = hdr->b_l1hdr.b_state; | |
4757 | arc_buf_contents_t type = arc_buf_type(hdr); | |
4758 | ||
4759 | /* protected by hash lock, if in the hash table */ | |
4760 | if (multilist_link_active(&hdr->b_l1hdr.b_arc_node)) { | |
4761 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); | |
4762 | ASSERT(state != arc_anon && state != arc_l2c_only); | |
4763 | ||
4764 | (void) refcount_remove_many(&state->arcs_esize[type], | |
4765 | size, tag); | |
4766 | } | |
4767 | (void) refcount_remove_many(&state->arcs_size, size, tag); | |
4768 | ||
4769 | VERIFY3U(hdr->b_type, ==, type); | |
4770 | if (type == ARC_BUFC_METADATA) { | |
4771 | arc_space_return(size, ARC_SPACE_META); | |
4772 | } else { | |
4773 | ASSERT(type == ARC_BUFC_DATA); | |
4774 | arc_space_return(size, ARC_SPACE_DATA); | |
4775 | } | |
4776 | } | |
4777 | ||
34dc7c2f BB |
4778 | /* |
4779 | * This routine is called whenever a buffer is accessed. | |
4780 | * NOTE: the hash lock is dropped in this function. | |
4781 | */ | |
4782 | static void | |
e10b0808 | 4783 | arc_access(arc_buf_hdr_t *hdr, kmutex_t *hash_lock) |
34dc7c2f | 4784 | { |
428870ff BB |
4785 | clock_t now; |
4786 | ||
34dc7c2f | 4787 | ASSERT(MUTEX_HELD(hash_lock)); |
e10b0808 | 4788 | ASSERT(HDR_HAS_L1HDR(hdr)); |
34dc7c2f | 4789 | |
e10b0808 | 4790 | if (hdr->b_l1hdr.b_state == arc_anon) { |
34dc7c2f BB |
4791 | /* |
4792 | * This buffer is not in the cache, and does not | |
4793 | * appear in our "ghost" list. Add the new buffer | |
4794 | * to the MRU state. | |
4795 | */ | |
4796 | ||
e10b0808 AX |
4797 | ASSERT0(hdr->b_l1hdr.b_arc_access); |
4798 | hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); | |
4799 | DTRACE_PROBE1(new_state__mru, arc_buf_hdr_t *, hdr); | |
4800 | arc_change_state(arc_mru, hdr, hash_lock); | |
34dc7c2f | 4801 | |
e10b0808 | 4802 | } else if (hdr->b_l1hdr.b_state == arc_mru) { |
428870ff BB |
4803 | now = ddi_get_lbolt(); |
4804 | ||
34dc7c2f BB |
4805 | /* |
4806 | * If this buffer is here because of a prefetch, then either: | |
4807 | * - clear the flag if this is a "referencing" read | |
4808 | * (any subsequent access will bump this into the MFU state). | |
4809 | * or | |
4810 | * - move the buffer to the head of the list if this is | |
4811 | * another prefetch (to make it less likely to be evicted). | |
4812 | */ | |
e10b0808 AX |
4813 | if (HDR_PREFETCH(hdr)) { |
4814 | if (refcount_count(&hdr->b_l1hdr.b_refcnt) == 0) { | |
4815 | /* link protected by hash lock */ | |
4816 | ASSERT(multilist_link_active( | |
4817 | &hdr->b_l1hdr.b_arc_node)); | |
34dc7c2f | 4818 | } else { |
cae5b340 | 4819 | arc_hdr_clear_flags(hdr, ARC_FLAG_PREFETCH); |
e10b0808 | 4820 | atomic_inc_32(&hdr->b_l1hdr.b_mru_hits); |
34dc7c2f BB |
4821 | ARCSTAT_BUMP(arcstat_mru_hits); |
4822 | } | |
e10b0808 | 4823 | hdr->b_l1hdr.b_arc_access = now; |
34dc7c2f BB |
4824 | return; |
4825 | } | |
4826 | ||
4827 | /* | |
4828 | * This buffer has been "accessed" only once so far, | |
4829 | * but it is still in the cache. Move it to the MFU | |
4830 | * state. | |
4831 | */ | |
e10b0808 AX |
4832 | if (ddi_time_after(now, hdr->b_l1hdr.b_arc_access + |
4833 | ARC_MINTIME)) { | |
34dc7c2f BB |
4834 | /* |
4835 | * More than 125ms have passed since we | |
4836 | * instantiated this buffer. Move it to the | |
4837 | * most frequently used state. | |
4838 | */ | |
e10b0808 AX |
4839 | hdr->b_l1hdr.b_arc_access = now; |
4840 | DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); | |
4841 | arc_change_state(arc_mfu, hdr, hash_lock); | |
34dc7c2f | 4842 | } |
e10b0808 | 4843 | atomic_inc_32(&hdr->b_l1hdr.b_mru_hits); |
34dc7c2f | 4844 | ARCSTAT_BUMP(arcstat_mru_hits); |
e10b0808 | 4845 | } else if (hdr->b_l1hdr.b_state == arc_mru_ghost) { |
34dc7c2f BB |
4846 | arc_state_t *new_state; |
4847 | /* | |
4848 | * This buffer has been "accessed" recently, but | |
4849 | * was evicted from the cache. Move it to the | |
4850 | * MFU state. | |
4851 | */ | |
4852 | ||
e10b0808 | 4853 | if (HDR_PREFETCH(hdr)) { |
34dc7c2f | 4854 | new_state = arc_mru; |
e10b0808 | 4855 | if (refcount_count(&hdr->b_l1hdr.b_refcnt) > 0) |
cae5b340 | 4856 | arc_hdr_clear_flags(hdr, ARC_FLAG_PREFETCH); |
e10b0808 | 4857 | DTRACE_PROBE1(new_state__mru, arc_buf_hdr_t *, hdr); |
34dc7c2f BB |
4858 | } else { |
4859 | new_state = arc_mfu; | |
e10b0808 | 4860 | DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); |
34dc7c2f BB |
4861 | } |
4862 | ||
e10b0808 AX |
4863 | hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); |
4864 | arc_change_state(new_state, hdr, hash_lock); | |
34dc7c2f | 4865 | |
e10b0808 | 4866 | atomic_inc_32(&hdr->b_l1hdr.b_mru_ghost_hits); |
34dc7c2f | 4867 | ARCSTAT_BUMP(arcstat_mru_ghost_hits); |
e10b0808 | 4868 | } else if (hdr->b_l1hdr.b_state == arc_mfu) { |
34dc7c2f BB |
4869 | /* |
4870 | * This buffer has been accessed more than once and is | |
4871 | * still in the cache. Keep it in the MFU state. | |
4872 | * | |
4873 | * NOTE: an add_reference() that occurred when we did | |
4874 | * the arc_read() will have kicked this off the list. | |
4875 | * If it was a prefetch, we will explicitly move it to | |
4876 | * the head of the list now. | |
4877 | */ | |
e10b0808 AX |
4878 | if ((HDR_PREFETCH(hdr)) != 0) { |
4879 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); | |
4880 | /* link protected by hash_lock */ | |
4881 | ASSERT(multilist_link_active(&hdr->b_l1hdr.b_arc_node)); | |
34dc7c2f | 4882 | } |
e10b0808 | 4883 | atomic_inc_32(&hdr->b_l1hdr.b_mfu_hits); |
34dc7c2f | 4884 | ARCSTAT_BUMP(arcstat_mfu_hits); |
e10b0808 AX |
4885 | hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); |
4886 | } else if (hdr->b_l1hdr.b_state == arc_mfu_ghost) { | |
34dc7c2f BB |
4887 | arc_state_t *new_state = arc_mfu; |
4888 | /* | |
4889 | * This buffer has been accessed more than once but has | |
4890 | * been evicted from the cache. Move it back to the | |
4891 | * MFU state. | |
4892 | */ | |
4893 | ||
e10b0808 | 4894 | if (HDR_PREFETCH(hdr)) { |
34dc7c2f BB |
4895 | /* |
4896 | * This is a prefetch access... | |
4897 | * move this block back to the MRU state. | |
4898 | */ | |
e10b0808 | 4899 | ASSERT0(refcount_count(&hdr->b_l1hdr.b_refcnt)); |
34dc7c2f BB |
4900 | new_state = arc_mru; |
4901 | } | |
4902 | ||
e10b0808 AX |
4903 | hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); |
4904 | DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); | |
4905 | arc_change_state(new_state, hdr, hash_lock); | |
34dc7c2f | 4906 | |
e10b0808 | 4907 | atomic_inc_32(&hdr->b_l1hdr.b_mfu_ghost_hits); |
34dc7c2f | 4908 | ARCSTAT_BUMP(arcstat_mfu_ghost_hits); |
e10b0808 | 4909 | } else if (hdr->b_l1hdr.b_state == arc_l2c_only) { |
34dc7c2f BB |
4910 | /* |
4911 | * This buffer is on the 2nd Level ARC. | |
4912 | */ | |
4913 | ||
e10b0808 AX |
4914 | hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); |
4915 | DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); | |
4916 | arc_change_state(arc_mfu, hdr, hash_lock); | |
34dc7c2f | 4917 | } else { |
e10b0808 AX |
4918 | cmn_err(CE_PANIC, "invalid arc state 0x%p", |
4919 | hdr->b_l1hdr.b_state); | |
34dc7c2f BB |
4920 | } |
4921 | } | |
4922 | ||
047218e2 AX |
4923 | /* |
4924 | * This routine is called by dbuf_hold() to update the arc_access() state | |
4925 | * which otherwise would be skipped for entries in the dbuf cache. | |
4926 | */ | |
4927 | void | |
4928 | arc_buf_access(arc_buf_t *buf) | |
4929 | { | |
4930 | mutex_enter(&buf->b_evict_lock); | |
4931 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
4932 | ||
4933 | /* | |
4934 | * Avoid taking the hash_lock when possible as an optimization. | |
4935 | * The header must be checked again under the hash_lock in order | |
4936 | * to handle the case where it is concurrently being released. | |
4937 | */ | |
4938 | if (hdr->b_l1hdr.b_state == arc_anon || HDR_EMPTY(hdr)) { | |
4939 | mutex_exit(&buf->b_evict_lock); | |
4940 | return; | |
4941 | } | |
4942 | ||
4943 | kmutex_t *hash_lock = HDR_LOCK(hdr); | |
4944 | mutex_enter(hash_lock); | |
4945 | ||
4946 | if (hdr->b_l1hdr.b_state == arc_anon || HDR_EMPTY(hdr)) { | |
4947 | mutex_exit(hash_lock); | |
4948 | mutex_exit(&buf->b_evict_lock); | |
4949 | ARCSTAT_BUMP(arcstat_access_skip); | |
4950 | return; | |
4951 | } | |
4952 | ||
4953 | mutex_exit(&buf->b_evict_lock); | |
4954 | ||
4955 | ASSERT(hdr->b_l1hdr.b_state == arc_mru || | |
4956 | hdr->b_l1hdr.b_state == arc_mfu); | |
4957 | ||
4958 | DTRACE_PROBE1(arc__hit, arc_buf_hdr_t *, hdr); | |
4959 | arc_access(hdr, hash_lock); | |
4960 | mutex_exit(hash_lock); | |
4961 | ||
4962 | ARCSTAT_BUMP(arcstat_hits); | |
4963 | ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr), demand, prefetch, | |
4964 | !HDR_ISTYPE_METADATA(hdr), data, metadata, hits); | |
4965 | } | |
4966 | ||
4967 | /* a generic arc_read_done_func_t which you can use */ | |
34dc7c2f BB |
4968 | /* ARGSUSED */ |
4969 | void | |
4970 | arc_bcopy_func(zio_t *zio, arc_buf_t *buf, void *arg) | |
4971 | { | |
428870ff | 4972 | if (zio == NULL || zio->io_error == 0) |
cae5b340 AX |
4973 | bcopy(buf->b_data, arg, arc_buf_size(buf)); |
4974 | arc_buf_destroy(buf, arg); | |
34dc7c2f BB |
4975 | } |
4976 | ||
4977 | /* a generic arc_done_func_t */ | |
4978 | void | |
4979 | arc_getbuf_func(zio_t *zio, arc_buf_t *buf, void *arg) | |
4980 | { | |
4981 | arc_buf_t **bufp = arg; | |
4982 | if (zio && zio->io_error) { | |
cae5b340 | 4983 | arc_buf_destroy(buf, arg); |
34dc7c2f BB |
4984 | *bufp = NULL; |
4985 | } else { | |
4986 | *bufp = buf; | |
428870ff | 4987 | ASSERT(buf->b_data); |
34dc7c2f BB |
4988 | } |
4989 | } | |
4990 | ||
cae5b340 AX |
4991 | static void |
4992 | arc_hdr_verify(arc_buf_hdr_t *hdr, blkptr_t *bp) | |
4993 | { | |
4994 | if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) { | |
4995 | ASSERT3U(HDR_GET_PSIZE(hdr), ==, 0); | |
4996 | ASSERT3U(HDR_GET_COMPRESS(hdr), ==, ZIO_COMPRESS_OFF); | |
4997 | } else { | |
4998 | if (HDR_COMPRESSION_ENABLED(hdr)) { | |
4999 | ASSERT3U(HDR_GET_COMPRESS(hdr), ==, | |
5000 | BP_GET_COMPRESS(bp)); | |
5001 | } | |
5002 | ASSERT3U(HDR_GET_LSIZE(hdr), ==, BP_GET_LSIZE(bp)); | |
5003 | ASSERT3U(HDR_GET_PSIZE(hdr), ==, BP_GET_PSIZE(bp)); | |
5004 | } | |
5005 | } | |
5006 | ||
34dc7c2f BB |
5007 | static void |
5008 | arc_read_done(zio_t *zio) | |
5009 | { | |
cae5b340 | 5010 | arc_buf_hdr_t *hdr = zio->io_private; |
ea04106b | 5011 | kmutex_t *hash_lock = NULL; |
cae5b340 AX |
5012 | arc_callback_t *callback_list; |
5013 | arc_callback_t *acb; | |
5014 | boolean_t freeable = B_FALSE; | |
5015 | boolean_t no_zio_error = (zio->io_error == 0); | |
34dc7c2f BB |
5016 | |
5017 | /* | |
5018 | * The hdr was inserted into hash-table and removed from lists | |
5019 | * prior to starting I/O. We should find this header, since | |
5020 | * it's in the hash table, and it should be legit since it's | |
5021 | * not possible to evict it during the I/O. The only possible | |
5022 | * reason for it not to be found is if we were freed during the | |
5023 | * read. | |
5024 | */ | |
ea04106b AX |
5025 | if (HDR_IN_HASH_TABLE(hdr)) { |
5026 | arc_buf_hdr_t *found; | |
5027 | ||
5028 | ASSERT3U(hdr->b_birth, ==, BP_PHYSICAL_BIRTH(zio->io_bp)); | |
5029 | ASSERT3U(hdr->b_dva.dva_word[0], ==, | |
5030 | BP_IDENTITY(zio->io_bp)->dva_word[0]); | |
5031 | ASSERT3U(hdr->b_dva.dva_word[1], ==, | |
5032 | BP_IDENTITY(zio->io_bp)->dva_word[1]); | |
5033 | ||
cae5b340 | 5034 | found = buf_hash_find(hdr->b_spa, zio->io_bp, &hash_lock); |
ea04106b | 5035 | |
cae5b340 | 5036 | ASSERT((found == hdr && |
ea04106b AX |
5037 | DVA_EQUAL(&hdr->b_dva, BP_IDENTITY(zio->io_bp))) || |
5038 | (found == hdr && HDR_L2_READING(hdr))); | |
cae5b340 AX |
5039 | ASSERT3P(hash_lock, !=, NULL); |
5040 | } | |
5041 | ||
5042 | if (no_zio_error) { | |
5043 | /* byteswap if necessary */ | |
5044 | if (BP_SHOULD_BYTESWAP(zio->io_bp)) { | |
5045 | if (BP_GET_LEVEL(zio->io_bp) > 0) { | |
5046 | hdr->b_l1hdr.b_byteswap = DMU_BSWAP_UINT64; | |
5047 | } else { | |
5048 | hdr->b_l1hdr.b_byteswap = | |
5049 | DMU_OT_BYTESWAP(BP_GET_TYPE(zio->io_bp)); | |
5050 | } | |
5051 | } else { | |
5052 | hdr->b_l1hdr.b_byteswap = DMU_BSWAP_NUMFUNCS; | |
5053 | } | |
ea04106b | 5054 | } |
34dc7c2f | 5055 | |
cae5b340 | 5056 | arc_hdr_clear_flags(hdr, ARC_FLAG_L2_EVICTED); |
e10b0808 | 5057 | if (l2arc_noprefetch && HDR_PREFETCH(hdr)) |
cae5b340 | 5058 | arc_hdr_clear_flags(hdr, ARC_FLAG_L2CACHE); |
34dc7c2f | 5059 | |
e10b0808 | 5060 | callback_list = hdr->b_l1hdr.b_acb; |
cae5b340 | 5061 | ASSERT3P(callback_list, !=, NULL); |
34dc7c2f | 5062 | |
cae5b340 | 5063 | if (hash_lock && no_zio_error && hdr->b_l1hdr.b_state == arc_anon) { |
428870ff BB |
5064 | /* |
5065 | * Only call arc_access on anonymous buffers. This is because | |
5066 | * if we've issued an I/O for an evicted buffer, we've already | |
5067 | * called arc_access (to prevent any simultaneous readers from | |
5068 | * getting confused). | |
5069 | */ | |
5070 | arc_access(hdr, hash_lock); | |
5071 | } | |
5072 | ||
cae5b340 AX |
5073 | /* |
5074 | * If a read request has a callback (i.e. acb_done is not NULL), then we | |
5075 | * make a buf containing the data according to the parameters which were | |
5076 | * passed in. The implementation of arc_buf_alloc_impl() ensures that we | |
5077 | * aren't needlessly decompressing the data multiple times. | |
5078 | */ | |
5079 | int callback_cnt = 0; | |
5080 | for (acb = callback_list; acb != NULL; acb = acb->acb_next) { | |
5081 | if (!acb->acb_done) | |
5082 | continue; | |
5083 | ||
5084 | /* This is a demand read since prefetches don't use callbacks */ | |
5085 | callback_cnt++; | |
5086 | ||
5087 | int error = arc_buf_alloc_impl(hdr, acb->acb_private, | |
5088 | acb->acb_compressed, no_zio_error, &acb->acb_buf); | |
5089 | if (no_zio_error) { | |
5090 | zio->io_error = error; | |
34dc7c2f BB |
5091 | } |
5092 | } | |
e10b0808 | 5093 | hdr->b_l1hdr.b_acb = NULL; |
cae5b340 AX |
5094 | arc_hdr_clear_flags(hdr, ARC_FLAG_IO_IN_PROGRESS); |
5095 | if (callback_cnt == 0) { | |
5096 | ASSERT(HDR_PREFETCH(hdr)); | |
5097 | ASSERT0(hdr->b_l1hdr.b_bufcnt); | |
5098 | ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); | |
428870ff | 5099 | } |
34dc7c2f | 5100 | |
e10b0808 AX |
5101 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt) || |
5102 | callback_list != NULL); | |
34dc7c2f | 5103 | |
cae5b340 AX |
5104 | if (no_zio_error) { |
5105 | arc_hdr_verify(hdr, zio->io_bp); | |
5106 | } else { | |
5107 | arc_hdr_set_flags(hdr, ARC_FLAG_IO_ERROR); | |
e10b0808 | 5108 | if (hdr->b_l1hdr.b_state != arc_anon) |
34dc7c2f BB |
5109 | arc_change_state(arc_anon, hdr, hash_lock); |
5110 | if (HDR_IN_HASH_TABLE(hdr)) | |
5111 | buf_hash_remove(hdr); | |
e10b0808 | 5112 | freeable = refcount_is_zero(&hdr->b_l1hdr.b_refcnt); |
34dc7c2f BB |
5113 | } |
5114 | ||
5115 | /* | |
5116 | * Broadcast before we drop the hash_lock to avoid the possibility | |
5117 | * that the hdr (and hence the cv) might be freed before we get to | |
5118 | * the cv_broadcast(). | |
5119 | */ | |
e10b0808 | 5120 | cv_broadcast(&hdr->b_l1hdr.b_cv); |
34dc7c2f | 5121 | |
e10b0808 | 5122 | if (hash_lock != NULL) { |
34dc7c2f BB |
5123 | mutex_exit(hash_lock); |
5124 | } else { | |
5125 | /* | |
5126 | * This block was freed while we waited for the read to | |
5127 | * complete. It has been removed from the hash table and | |
5128 | * moved to the anonymous state (so that it won't show up | |
5129 | * in the cache). | |
5130 | */ | |
e10b0808 AX |
5131 | ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon); |
5132 | freeable = refcount_is_zero(&hdr->b_l1hdr.b_refcnt); | |
34dc7c2f BB |
5133 | } |
5134 | ||
5135 | /* execute each callback and free its structure */ | |
5136 | while ((acb = callback_list) != NULL) { | |
5137 | if (acb->acb_done) | |
5138 | acb->acb_done(zio, acb->acb_buf, acb->acb_private); | |
5139 | ||
5140 | if (acb->acb_zio_dummy != NULL) { | |
5141 | acb->acb_zio_dummy->io_error = zio->io_error; | |
5142 | zio_nowait(acb->acb_zio_dummy); | |
5143 | } | |
5144 | ||
5145 | callback_list = acb->acb_next; | |
5146 | kmem_free(acb, sizeof (arc_callback_t)); | |
5147 | } | |
5148 | ||
5149 | if (freeable) | |
5150 | arc_hdr_destroy(hdr); | |
5151 | } | |
5152 | ||
5153 | /* | |
5c839890 | 5154 | * "Read" the block at the specified DVA (in bp) via the |
34dc7c2f BB |
5155 | * cache. If the block is found in the cache, invoke the provided |
5156 | * callback immediately and return. Note that the `zio' parameter | |
5157 | * in the callback will be NULL in this case, since no IO was | |
5158 | * required. If the block is not in the cache pass the read request | |
5159 | * on to the spa with a substitute callback function, so that the | |
5160 | * requested block will be added to the cache. | |
5161 | * | |
5162 | * If a read request arrives for a block that has a read in-progress, | |
5163 | * either wait for the in-progress read to complete (and return the | |
5164 | * results); or, if this is a read with a "done" func, add a record | |
5165 | * to the read to invoke the "done" func when the read completes, | |
5166 | * and return; or just return. | |
5167 | * | |
5168 | * arc_read_done() will invoke all the requested "done" functions | |
5169 | * for readers of this block. | |
5170 | */ | |
5171 | int | |
c06d4368 | 5172 | arc_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, arc_done_func_t *done, |
e10b0808 AX |
5173 | void *private, zio_priority_t priority, int zio_flags, |
5174 | arc_flags_t *arc_flags, const zbookmark_phys_t *zb) | |
34dc7c2f | 5175 | { |
ea04106b | 5176 | arc_buf_hdr_t *hdr = NULL; |
ea04106b | 5177 | kmutex_t *hash_lock = NULL; |
34dc7c2f | 5178 | zio_t *rzio; |
3541dc6d | 5179 | uint64_t guid = spa_load_guid(spa); |
cae5b340 | 5180 | boolean_t compressed_read = (zio_flags & ZIO_FLAG_RAW) != 0; |
a08ee875 | 5181 | int rc = 0; |
34dc7c2f | 5182 | |
ea04106b AX |
5183 | ASSERT(!BP_IS_EMBEDDED(bp) || |
5184 | BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA); | |
5185 | ||
34dc7c2f | 5186 | top: |
ea04106b AX |
5187 | if (!BP_IS_EMBEDDED(bp)) { |
5188 | /* | |
5189 | * Embedded BP's have no DVA and require no I/O to "read". | |
5190 | * Create an anonymous arc buf to back it. | |
5191 | */ | |
5192 | hdr = buf_hash_find(guid, bp, &hash_lock); | |
5193 | } | |
5194 | ||
cae5b340 AX |
5195 | if (hdr != NULL && HDR_HAS_L1HDR(hdr) && hdr->b_l1hdr.b_pabd != NULL) { |
5196 | arc_buf_t *buf = NULL; | |
e10b0808 | 5197 | *arc_flags |= ARC_FLAG_CACHED; |
34dc7c2f BB |
5198 | |
5199 | if (HDR_IO_IN_PROGRESS(hdr)) { | |
5200 | ||
cae5b340 AX |
5201 | if ((hdr->b_flags & ARC_FLAG_PRIO_ASYNC_READ) && |
5202 | priority == ZIO_PRIORITY_SYNC_READ) { | |
5203 | /* | |
5204 | * This sync read must wait for an | |
5205 | * in-progress async read (e.g. a predictive | |
5206 | * prefetch). Async reads are queued | |
5207 | * separately at the vdev_queue layer, so | |
5208 | * this is a form of priority inversion. | |
5209 | * Ideally, we would "inherit" the demand | |
5210 | * i/o's priority by moving the i/o from | |
5211 | * the async queue to the synchronous queue, | |
5212 | * but there is currently no mechanism to do | |
5213 | * so. Track this so that we can evaluate | |
5214 | * the magnitude of this potential performance | |
5215 | * problem. | |
5216 | * | |
5217 | * Note that if the prefetch i/o is already | |
5218 | * active (has been issued to the device), | |
5219 | * the prefetch improved performance, because | |
5220 | * we issued it sooner than we would have | |
5221 | * without the prefetch. | |
5222 | */ | |
5223 | DTRACE_PROBE1(arc__sync__wait__for__async, | |
5224 | arc_buf_hdr_t *, hdr); | |
5225 | ARCSTAT_BUMP(arcstat_sync_wait_for_async); | |
5226 | } | |
5227 | if (hdr->b_flags & ARC_FLAG_PREDICTIVE_PREFETCH) { | |
5228 | arc_hdr_clear_flags(hdr, | |
5229 | ARC_FLAG_PREDICTIVE_PREFETCH); | |
5230 | } | |
5231 | ||
e10b0808 AX |
5232 | if (*arc_flags & ARC_FLAG_WAIT) { |
5233 | cv_wait(&hdr->b_l1hdr.b_cv, hash_lock); | |
34dc7c2f BB |
5234 | mutex_exit(hash_lock); |
5235 | goto top; | |
5236 | } | |
e10b0808 | 5237 | ASSERT(*arc_flags & ARC_FLAG_NOWAIT); |
34dc7c2f BB |
5238 | |
5239 | if (done) { | |
cae5b340 | 5240 | arc_callback_t *acb = NULL; |
34dc7c2f BB |
5241 | |
5242 | acb = kmem_zalloc(sizeof (arc_callback_t), | |
ea04106b | 5243 | KM_SLEEP); |
34dc7c2f BB |
5244 | acb->acb_done = done; |
5245 | acb->acb_private = private; | |
cae5b340 | 5246 | acb->acb_compressed = compressed_read; |
34dc7c2f BB |
5247 | if (pio != NULL) |
5248 | acb->acb_zio_dummy = zio_null(pio, | |
d164b209 | 5249 | spa, NULL, NULL, NULL, zio_flags); |
34dc7c2f | 5250 | |
cae5b340 | 5251 | ASSERT3P(acb->acb_done, !=, NULL); |
e10b0808 AX |
5252 | acb->acb_next = hdr->b_l1hdr.b_acb; |
5253 | hdr->b_l1hdr.b_acb = acb; | |
34dc7c2f | 5254 | mutex_exit(hash_lock); |
a08ee875 | 5255 | goto out; |
34dc7c2f BB |
5256 | } |
5257 | mutex_exit(hash_lock); | |
a08ee875 | 5258 | goto out; |
34dc7c2f BB |
5259 | } |
5260 | ||
e10b0808 AX |
5261 | ASSERT(hdr->b_l1hdr.b_state == arc_mru || |
5262 | hdr->b_l1hdr.b_state == arc_mfu); | |
34dc7c2f BB |
5263 | |
5264 | if (done) { | |
cae5b340 AX |
5265 | if (hdr->b_flags & ARC_FLAG_PREDICTIVE_PREFETCH) { |
5266 | /* | |
5267 | * This is a demand read which does not have to | |
5268 | * wait for i/o because we did a predictive | |
5269 | * prefetch i/o for it, which has completed. | |
5270 | */ | |
5271 | DTRACE_PROBE1( | |
5272 | arc__demand__hit__predictive__prefetch, | |
5273 | arc_buf_hdr_t *, hdr); | |
5274 | ARCSTAT_BUMP( | |
5275 | arcstat_demand_hit_predictive_prefetch); | |
5276 | arc_hdr_clear_flags(hdr, | |
5277 | ARC_FLAG_PREDICTIVE_PREFETCH); | |
34dc7c2f | 5278 | } |
cae5b340 | 5279 | ASSERT(!BP_IS_EMBEDDED(bp) || !BP_IS_HOLE(bp)); |
428870ff | 5280 | |
cae5b340 AX |
5281 | /* Get a buf with the desired data in it. */ |
5282 | VERIFY0(arc_buf_alloc_impl(hdr, private, | |
5283 | compressed_read, B_TRUE, &buf)); | |
e10b0808 AX |
5284 | } else if (*arc_flags & ARC_FLAG_PREFETCH && |
5285 | refcount_count(&hdr->b_l1hdr.b_refcnt) == 0) { | |
cae5b340 | 5286 | arc_hdr_set_flags(hdr, ARC_FLAG_PREFETCH); |
34dc7c2f BB |
5287 | } |
5288 | DTRACE_PROBE1(arc__hit, arc_buf_hdr_t *, hdr); | |
5289 | arc_access(hdr, hash_lock); | |
e10b0808 | 5290 | if (*arc_flags & ARC_FLAG_L2CACHE) |
cae5b340 | 5291 | arc_hdr_set_flags(hdr, ARC_FLAG_L2CACHE); |
34dc7c2f BB |
5292 | mutex_exit(hash_lock); |
5293 | ARCSTAT_BUMP(arcstat_hits); | |
e10b0808 AX |
5294 | ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr), |
5295 | demand, prefetch, !HDR_ISTYPE_METADATA(hdr), | |
34dc7c2f BB |
5296 | data, metadata, hits); |
5297 | ||
5298 | if (done) | |
5299 | done(NULL, buf, private); | |
5300 | } else { | |
cae5b340 AX |
5301 | uint64_t lsize = BP_GET_LSIZE(bp); |
5302 | uint64_t psize = BP_GET_PSIZE(bp); | |
ea04106b | 5303 | arc_callback_t *acb; |
b128c09f | 5304 | vdev_t *vd = NULL; |
a08ee875 | 5305 | uint64_t addr = 0; |
d164b209 | 5306 | boolean_t devw = B_FALSE; |
cae5b340 | 5307 | uint64_t size; |
ea04106b AX |
5308 | |
5309 | /* | |
5310 | * Gracefully handle a damaged logical block size as a | |
87dac73d | 5311 | * checksum error. |
ea04106b | 5312 | */ |
cae5b340 | 5313 | if (lsize > spa_maxblocksize(spa)) { |
87dac73d | 5314 | rc = SET_ERROR(ECKSUM); |
ea04106b AX |
5315 | goto out; |
5316 | } | |
34dc7c2f BB |
5317 | |
5318 | if (hdr == NULL) { | |
5319 | /* this block is not in the cache */ | |
ea04106b | 5320 | arc_buf_hdr_t *exists = NULL; |
34dc7c2f | 5321 | arc_buf_contents_t type = BP_GET_BUFC_TYPE(bp); |
cae5b340 AX |
5322 | hdr = arc_hdr_alloc(spa_load_guid(spa), psize, lsize, |
5323 | BP_GET_COMPRESS(bp), type); | |
5324 | ||
ea04106b AX |
5325 | if (!BP_IS_EMBEDDED(bp)) { |
5326 | hdr->b_dva = *BP_IDENTITY(bp); | |
5327 | hdr->b_birth = BP_PHYSICAL_BIRTH(bp); | |
ea04106b AX |
5328 | exists = buf_hash_insert(hdr, &hash_lock); |
5329 | } | |
5330 | if (exists != NULL) { | |
34dc7c2f BB |
5331 | /* somebody beat us to the hash insert */ |
5332 | mutex_exit(hash_lock); | |
428870ff | 5333 | buf_discard_identity(hdr); |
cae5b340 | 5334 | arc_hdr_destroy(hdr); |
34dc7c2f BB |
5335 | goto top; /* restart the IO request */ |
5336 | } | |
34dc7c2f | 5337 | } else { |
e10b0808 AX |
5338 | /* |
5339 | * This block is in the ghost cache. If it was L2-only | |
5340 | * (and thus didn't have an L1 hdr), we realloc the | |
5341 | * header to add an L1 hdr. | |
5342 | */ | |
5343 | if (!HDR_HAS_L1HDR(hdr)) { | |
5344 | hdr = arc_hdr_realloc(hdr, hdr_l2only_cache, | |
5345 | hdr_full_cache); | |
5346 | } | |
5347 | ||
cae5b340 | 5348 | ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); |
e10b0808 | 5349 | ASSERT(GHOST_STATE(hdr->b_l1hdr.b_state)); |
34dc7c2f | 5350 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); |
e10b0808 AX |
5351 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); |
5352 | ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); | |
cae5b340 | 5353 | ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL); |
34dc7c2f | 5354 | |
cae5b340 AX |
5355 | /* |
5356 | * This is a delicate dance that we play here. | |
5357 | * This hdr is in the ghost list so we access it | |
5358 | * to move it out of the ghost list before we | |
5359 | * initiate the read. If it's a prefetch then | |
5360 | * it won't have a callback so we'll remove the | |
5361 | * reference that arc_buf_alloc_impl() created. We | |
5362 | * do this after we've called arc_access() to | |
5363 | * avoid hitting an assert in remove_reference(). | |
5364 | */ | |
428870ff | 5365 | arc_access(hdr, hash_lock); |
cae5b340 | 5366 | arc_hdr_alloc_pabd(hdr); |
34dc7c2f | 5367 | } |
cae5b340 AX |
5368 | ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); |
5369 | size = arc_hdr_size(hdr); | |
34dc7c2f | 5370 | |
cae5b340 AX |
5371 | /* |
5372 | * If compression is enabled on the hdr, then will do | |
5373 | * RAW I/O and will store the compressed data in the hdr's | |
5374 | * data block. Otherwise, the hdr's data block will contain | |
5375 | * the uncompressed data. | |
5376 | */ | |
5377 | if (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF) { | |
5378 | zio_flags |= ZIO_FLAG_RAW; | |
5379 | } | |
5380 | ||
5381 | if (*arc_flags & ARC_FLAG_PREFETCH) | |
5382 | arc_hdr_set_flags(hdr, ARC_FLAG_PREFETCH); | |
5383 | if (*arc_flags & ARC_FLAG_L2CACHE) | |
5384 | arc_hdr_set_flags(hdr, ARC_FLAG_L2CACHE); | |
5385 | if (BP_GET_LEVEL(bp) > 0) | |
5386 | arc_hdr_set_flags(hdr, ARC_FLAG_INDIRECT); | |
5387 | if (*arc_flags & ARC_FLAG_PREDICTIVE_PREFETCH) | |
5388 | arc_hdr_set_flags(hdr, ARC_FLAG_PREDICTIVE_PREFETCH); | |
e10b0808 | 5389 | ASSERT(!GHOST_STATE(hdr->b_l1hdr.b_state)); |
428870ff | 5390 | |
ea04106b | 5391 | acb = kmem_zalloc(sizeof (arc_callback_t), KM_SLEEP); |
34dc7c2f BB |
5392 | acb->acb_done = done; |
5393 | acb->acb_private = private; | |
cae5b340 | 5394 | acb->acb_compressed = compressed_read; |
34dc7c2f | 5395 | |
cae5b340 | 5396 | ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL); |
e10b0808 | 5397 | hdr->b_l1hdr.b_acb = acb; |
cae5b340 | 5398 | arc_hdr_set_flags(hdr, ARC_FLAG_IO_IN_PROGRESS); |
34dc7c2f | 5399 | |
e10b0808 AX |
5400 | if (HDR_HAS_L2HDR(hdr) && |
5401 | (vd = hdr->b_l2hdr.b_dev->l2ad_vdev) != NULL) { | |
5402 | devw = hdr->b_l2hdr.b_dev->l2ad_writing; | |
5403 | addr = hdr->b_l2hdr.b_daddr; | |
b128c09f BB |
5404 | /* |
5405 | * Lock out device removal. | |
5406 | */ | |
5407 | if (vdev_is_dead(vd) || | |
5408 | !spa_config_tryenter(spa, SCL_L2ARC, vd, RW_READER)) | |
5409 | vd = NULL; | |
5410 | } | |
5411 | ||
cae5b340 AX |
5412 | if (priority == ZIO_PRIORITY_ASYNC_READ) |
5413 | arc_hdr_set_flags(hdr, ARC_FLAG_PRIO_ASYNC_READ); | |
5414 | else | |
5415 | arc_hdr_clear_flags(hdr, ARC_FLAG_PRIO_ASYNC_READ); | |
5416 | ||
ea04106b AX |
5417 | if (hash_lock != NULL) |
5418 | mutex_exit(hash_lock); | |
b128c09f | 5419 | |
a08ee875 LG |
5420 | /* |
5421 | * At this point, we have a level 1 cache miss. Try again in | |
5422 | * L2ARC if possible. | |
5423 | */ | |
cae5b340 AX |
5424 | ASSERT3U(HDR_GET_LSIZE(hdr), ==, lsize); |
5425 | ||
428870ff | 5426 | DTRACE_PROBE4(arc__miss, arc_buf_hdr_t *, hdr, blkptr_t *, bp, |
cae5b340 | 5427 | uint64_t, lsize, zbookmark_phys_t *, zb); |
34dc7c2f | 5428 | ARCSTAT_BUMP(arcstat_misses); |
e10b0808 AX |
5429 | ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr), |
5430 | demand, prefetch, !HDR_ISTYPE_METADATA(hdr), | |
34dc7c2f BB |
5431 | data, metadata, misses); |
5432 | ||
d164b209 | 5433 | if (vd != NULL && l2arc_ndev != 0 && !(l2arc_norw && devw)) { |
34dc7c2f BB |
5434 | /* |
5435 | * Read from the L2ARC if the following are true: | |
b128c09f BB |
5436 | * 1. The L2ARC vdev was previously cached. |
5437 | * 2. This buffer still has L2ARC metadata. | |
5438 | * 3. This buffer isn't currently writing to the L2ARC. | |
5439 | * 4. The L2ARC entry wasn't evicted, which may | |
5440 | * also have invalidated the vdev. | |
d164b209 | 5441 | * 5. This isn't prefetch and l2arc_noprefetch is set. |
34dc7c2f | 5442 | */ |
e10b0808 | 5443 | if (HDR_HAS_L2HDR(hdr) && |
d164b209 BB |
5444 | !HDR_L2_WRITING(hdr) && !HDR_L2_EVICTED(hdr) && |
5445 | !(l2arc_noprefetch && HDR_PREFETCH(hdr))) { | |
34dc7c2f | 5446 | l2arc_read_callback_t *cb; |
cae5b340 AX |
5447 | abd_t *abd; |
5448 | uint64_t asize; | |
34dc7c2f BB |
5449 | |
5450 | DTRACE_PROBE1(l2arc__hit, arc_buf_hdr_t *, hdr); | |
5451 | ARCSTAT_BUMP(arcstat_l2_hits); | |
e10b0808 | 5452 | atomic_inc_32(&hdr->b_l2hdr.b_hits); |
34dc7c2f | 5453 | |
34dc7c2f | 5454 | cb = kmem_zalloc(sizeof (l2arc_read_callback_t), |
ea04106b | 5455 | KM_SLEEP); |
cae5b340 | 5456 | cb->l2rcb_hdr = hdr; |
34dc7c2f BB |
5457 | cb->l2rcb_bp = *bp; |
5458 | cb->l2rcb_zb = *zb; | |
b128c09f | 5459 | cb->l2rcb_flags = zio_flags; |
cae5b340 AX |
5460 | |
5461 | asize = vdev_psize_to_asize(vd, size); | |
5462 | if (asize != size) { | |
5463 | abd = abd_alloc_for_io(asize, | |
5464 | HDR_ISTYPE_METADATA(hdr)); | |
5465 | cb->l2rcb_abd = abd; | |
5466 | } else { | |
5467 | abd = hdr->b_l1hdr.b_pabd; | |
5468 | } | |
34dc7c2f | 5469 | |
a08ee875 | 5470 | ASSERT(addr >= VDEV_LABEL_START_SIZE && |
cae5b340 | 5471 | addr + asize <= vd->vdev_psize - |
a08ee875 LG |
5472 | VDEV_LABEL_END_SIZE); |
5473 | ||
34dc7c2f | 5474 | /* |
b128c09f BB |
5475 | * l2arc read. The SCL_L2ARC lock will be |
5476 | * released by l2arc_read_done(). | |
c06d4368 AX |
5477 | * Issue a null zio if the underlying buffer |
5478 | * was squashed to zero size by compression. | |
34dc7c2f | 5479 | */ |
cae5b340 AX |
5480 | ASSERT3U(HDR_GET_COMPRESS(hdr), !=, |
5481 | ZIO_COMPRESS_EMPTY); | |
5482 | rzio = zio_read_phys(pio, vd, addr, | |
5483 | asize, abd, | |
5484 | ZIO_CHECKSUM_OFF, | |
5485 | l2arc_read_done, cb, priority, | |
5486 | zio_flags | ZIO_FLAG_DONT_CACHE | | |
5487 | ZIO_FLAG_CANFAIL | | |
5488 | ZIO_FLAG_DONT_PROPAGATE | | |
5489 | ZIO_FLAG_DONT_RETRY, B_FALSE); | |
5490 | ||
34dc7c2f BB |
5491 | DTRACE_PROBE2(l2arc__read, vdev_t *, vd, |
5492 | zio_t *, rzio); | |
cae5b340 | 5493 | ARCSTAT_INCR(arcstat_l2_read_bytes, size); |
34dc7c2f | 5494 | |
e10b0808 | 5495 | if (*arc_flags & ARC_FLAG_NOWAIT) { |
b128c09f | 5496 | zio_nowait(rzio); |
a08ee875 | 5497 | goto out; |
b128c09f | 5498 | } |
34dc7c2f | 5499 | |
e10b0808 | 5500 | ASSERT(*arc_flags & ARC_FLAG_WAIT); |
b128c09f | 5501 | if (zio_wait(rzio) == 0) |
a08ee875 | 5502 | goto out; |
b128c09f BB |
5503 | |
5504 | /* l2arc read error; goto zio_read() */ | |
34dc7c2f BB |
5505 | } else { |
5506 | DTRACE_PROBE1(l2arc__miss, | |
5507 | arc_buf_hdr_t *, hdr); | |
5508 | ARCSTAT_BUMP(arcstat_l2_misses); | |
5509 | if (HDR_L2_WRITING(hdr)) | |
5510 | ARCSTAT_BUMP(arcstat_l2_rw_clash); | |
b128c09f | 5511 | spa_config_exit(spa, SCL_L2ARC, vd); |
34dc7c2f | 5512 | } |
d164b209 BB |
5513 | } else { |
5514 | if (vd != NULL) | |
5515 | spa_config_exit(spa, SCL_L2ARC, vd); | |
5516 | if (l2arc_ndev != 0) { | |
5517 | DTRACE_PROBE1(l2arc__miss, | |
5518 | arc_buf_hdr_t *, hdr); | |
5519 | ARCSTAT_BUMP(arcstat_l2_misses); | |
5520 | } | |
34dc7c2f | 5521 | } |
34dc7c2f | 5522 | |
cae5b340 AX |
5523 | rzio = zio_read(pio, spa, bp, hdr->b_l1hdr.b_pabd, size, |
5524 | arc_read_done, hdr, priority, zio_flags, zb); | |
34dc7c2f | 5525 | |
e10b0808 | 5526 | if (*arc_flags & ARC_FLAG_WAIT) { |
a08ee875 LG |
5527 | rc = zio_wait(rzio); |
5528 | goto out; | |
5529 | } | |
34dc7c2f | 5530 | |
e10b0808 | 5531 | ASSERT(*arc_flags & ARC_FLAG_NOWAIT); |
34dc7c2f BB |
5532 | zio_nowait(rzio); |
5533 | } | |
a08ee875 LG |
5534 | |
5535 | out: | |
5536 | spa_read_history_add(spa, zb, *arc_flags); | |
5537 | return (rc); | |
34dc7c2f BB |
5538 | } |
5539 | ||
ab26409d BB |
5540 | arc_prune_t * |
5541 | arc_add_prune_callback(arc_prune_func_t *func, void *private) | |
5542 | { | |
5543 | arc_prune_t *p; | |
5544 | ||
a08ee875 | 5545 | p = kmem_alloc(sizeof (*p), KM_SLEEP); |
ab26409d BB |
5546 | p->p_pfunc = func; |
5547 | p->p_private = private; | |
5548 | list_link_init(&p->p_node); | |
5549 | refcount_create(&p->p_refcnt); | |
5550 | ||
5551 | mutex_enter(&arc_prune_mtx); | |
5552 | refcount_add(&p->p_refcnt, &arc_prune_list); | |
5553 | list_insert_head(&arc_prune_list, p); | |
5554 | mutex_exit(&arc_prune_mtx); | |
5555 | ||
5556 | return (p); | |
5557 | } | |
5558 | ||
5559 | void | |
5560 | arc_remove_prune_callback(arc_prune_t *p) | |
5561 | { | |
87dac73d | 5562 | boolean_t wait = B_FALSE; |
ab26409d BB |
5563 | mutex_enter(&arc_prune_mtx); |
5564 | list_remove(&arc_prune_list, p); | |
87dac73d AX |
5565 | if (refcount_remove(&p->p_refcnt, &arc_prune_list) > 0) |
5566 | wait = B_TRUE; | |
ab26409d | 5567 | mutex_exit(&arc_prune_mtx); |
87dac73d AX |
5568 | |
5569 | /* wait for arc_prune_task to finish */ | |
5570 | if (wait) | |
5571 | taskq_wait_outstanding(arc_prune_taskq, 0); | |
5572 | ASSERT0(refcount_count(&p->p_refcnt)); | |
5573 | refcount_destroy(&p->p_refcnt); | |
5574 | kmem_free(p, sizeof (*p)); | |
ab26409d BB |
5575 | } |
5576 | ||
c06d4368 AX |
5577 | /* |
5578 | * Notify the arc that a block was freed, and thus will never be used again. | |
5579 | */ | |
5580 | void | |
5581 | arc_freed(spa_t *spa, const blkptr_t *bp) | |
5582 | { | |
5583 | arc_buf_hdr_t *hdr; | |
5584 | kmutex_t *hash_lock; | |
5585 | uint64_t guid = spa_load_guid(spa); | |
5586 | ||
ea04106b AX |
5587 | ASSERT(!BP_IS_EMBEDDED(bp)); |
5588 | ||
5589 | hdr = buf_hash_find(guid, bp, &hash_lock); | |
c06d4368 AX |
5590 | if (hdr == NULL) |
5591 | return; | |
c06d4368 | 5592 | |
cae5b340 AX |
5593 | /* |
5594 | * We might be trying to free a block that is still doing I/O | |
5595 | * (i.e. prefetch) or has a reference (i.e. a dedup-ed, | |
5596 | * dmu_sync-ed block). If this block is being prefetched, then it | |
5597 | * would still have the ARC_FLAG_IO_IN_PROGRESS flag set on the hdr | |
5598 | * until the I/O completes. A block may also have a reference if it is | |
5599 | * part of a dedup-ed, dmu_synced write. The dmu_sync() function would | |
5600 | * have written the new block to its final resting place on disk but | |
5601 | * without the dedup flag set. This would have left the hdr in the MRU | |
5602 | * state and discoverable. When the txg finally syncs it detects that | |
5603 | * the block was overridden in open context and issues an override I/O. | |
5604 | * Since this is a dedup block, the override I/O will determine if the | |
5605 | * block is already in the DDT. If so, then it will replace the io_bp | |
5606 | * with the bp from the DDT and allow the I/O to finish. When the I/O | |
5607 | * reaches the done callback, dbuf_write_override_done, it will | |
5608 | * check to see if the io_bp and io_bp_override are identical. | |
5609 | * If they are not, then it indicates that the bp was replaced with | |
5610 | * the bp in the DDT and the override bp is freed. This allows | |
5611 | * us to arrive here with a reference on a block that is being | |
5612 | * freed. So if we have an I/O in progress, or a reference to | |
5613 | * this hdr, then we don't destroy the hdr. | |
5614 | */ | |
5615 | if (!HDR_HAS_L1HDR(hdr) || (!HDR_IO_IN_PROGRESS(hdr) && | |
5616 | refcount_is_zero(&hdr->b_l1hdr.b_refcnt))) { | |
5617 | arc_change_state(arc_anon, hdr, hash_lock); | |
5618 | arc_hdr_destroy(hdr); | |
c06d4368 | 5619 | mutex_exit(hash_lock); |
ea04106b | 5620 | } else { |
cae5b340 | 5621 | mutex_exit(hash_lock); |
34dc7c2f | 5622 | } |
34dc7c2f | 5623 | |
34dc7c2f BB |
5624 | } |
5625 | ||
5626 | /* | |
a08ee875 LG |
5627 | * Release this buffer from the cache, making it an anonymous buffer. This |
5628 | * must be done after a read and prior to modifying the buffer contents. | |
34dc7c2f | 5629 | * If the buffer has more than one reference, we must make |
b128c09f | 5630 | * a new hdr for the buffer. |
34dc7c2f BB |
5631 | */ |
5632 | void | |
5633 | arc_release(arc_buf_t *buf, void *tag) | |
5634 | { | |
e10b0808 AX |
5635 | kmutex_t *hash_lock; |
5636 | arc_state_t *state; | |
5637 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
34dc7c2f | 5638 | |
428870ff | 5639 | /* |
e10b0808 | 5640 | * It would be nice to assert that if its DMU metadata (level > |
428870ff BB |
5641 | * 0 || it's the dnode file), then it must be syncing context. |
5642 | * But we don't know that information at this level. | |
5643 | */ | |
5644 | ||
5645 | mutex_enter(&buf->b_evict_lock); | |
b128c09f | 5646 | |
e10b0808 | 5647 | ASSERT(HDR_HAS_L1HDR(hdr)); |
34dc7c2f | 5648 | |
e10b0808 AX |
5649 | /* |
5650 | * We don't grab the hash lock prior to this check, because if | |
5651 | * the buffer's header is in the arc_anon state, it won't be | |
5652 | * linked into the hash table. | |
5653 | */ | |
5654 | if (hdr->b_l1hdr.b_state == arc_anon) { | |
5655 | mutex_exit(&buf->b_evict_lock); | |
5656 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); | |
5657 | ASSERT(!HDR_IN_HASH_TABLE(hdr)); | |
5658 | ASSERT(!HDR_HAS_L2HDR(hdr)); | |
cae5b340 | 5659 | ASSERT(HDR_EMPTY(hdr)); |
e10b0808 | 5660 | |
cae5b340 | 5661 | ASSERT3U(hdr->b_l1hdr.b_bufcnt, ==, 1); |
e10b0808 AX |
5662 | ASSERT3S(refcount_count(&hdr->b_l1hdr.b_refcnt), ==, 1); |
5663 | ASSERT(!list_link_active(&hdr->b_l1hdr.b_arc_node)); | |
5664 | ||
e10b0808 | 5665 | hdr->b_l1hdr.b_arc_access = 0; |
cae5b340 AX |
5666 | |
5667 | /* | |
5668 | * If the buf is being overridden then it may already | |
5669 | * have a hdr that is not empty. | |
5670 | */ | |
5671 | buf_discard_identity(hdr); | |
e10b0808 AX |
5672 | arc_buf_thaw(buf); |
5673 | ||
5674 | return; | |
34dc7c2f BB |
5675 | } |
5676 | ||
e10b0808 AX |
5677 | hash_lock = HDR_LOCK(hdr); |
5678 | mutex_enter(hash_lock); | |
5679 | ||
5680 | /* | |
5681 | * This assignment is only valid as long as the hash_lock is | |
5682 | * held, we must be careful not to reference state or the | |
5683 | * b_state field after dropping the lock. | |
5684 | */ | |
5685 | state = hdr->b_l1hdr.b_state; | |
5686 | ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); | |
5687 | ASSERT3P(state, !=, arc_anon); | |
5688 | ||
5689 | /* this buffer is not on any list */ | |
cae5b340 | 5690 | ASSERT3S(refcount_count(&hdr->b_l1hdr.b_refcnt), >, 0); |
e10b0808 AX |
5691 | |
5692 | if (HDR_HAS_L2HDR(hdr)) { | |
5693 | mutex_enter(&hdr->b_l2hdr.b_dev->l2ad_mtx); | |
5694 | ||
5695 | /* | |
5696 | * We have to recheck this conditional again now that | |
5697 | * we're holding the l2ad_mtx to prevent a race with | |
5698 | * another thread which might be concurrently calling | |
5699 | * l2arc_evict(). In that case, l2arc_evict() might have | |
5700 | * destroyed the header's L2 portion as we were waiting | |
5701 | * to acquire the l2ad_mtx. | |
5702 | */ | |
5703 | if (HDR_HAS_L2HDR(hdr)) | |
5704 | arc_hdr_l2hdr_destroy(hdr); | |
5705 | ||
5706 | mutex_exit(&hdr->b_l2hdr.b_dev->l2ad_mtx); | |
b128c09f BB |
5707 | } |
5708 | ||
34dc7c2f BB |
5709 | /* |
5710 | * Do we have more than one buf? | |
5711 | */ | |
cae5b340 | 5712 | if (hdr->b_l1hdr.b_bufcnt > 1) { |
34dc7c2f | 5713 | arc_buf_hdr_t *nhdr; |
d164b209 | 5714 | uint64_t spa = hdr->b_spa; |
cae5b340 AX |
5715 | uint64_t psize = HDR_GET_PSIZE(hdr); |
5716 | uint64_t lsize = HDR_GET_LSIZE(hdr); | |
5717 | enum zio_compress compress = HDR_GET_COMPRESS(hdr); | |
e10b0808 | 5718 | arc_buf_contents_t type = arc_buf_type(hdr); |
cae5b340 | 5719 | VERIFY3U(hdr->b_type, ==, type); |
34dc7c2f | 5720 | |
e10b0808 | 5721 | ASSERT(hdr->b_l1hdr.b_buf != buf || buf->b_next != NULL); |
cae5b340 AX |
5722 | (void) remove_reference(hdr, hash_lock, tag); |
5723 | ||
5724 | if (arc_buf_is_shared(buf) && !ARC_BUF_COMPRESSED(buf)) { | |
5725 | ASSERT3P(hdr->b_l1hdr.b_buf, !=, buf); | |
5726 | ASSERT(ARC_BUF_LAST(buf)); | |
5727 | } | |
5728 | ||
34dc7c2f | 5729 | /* |
428870ff | 5730 | * Pull the data off of this hdr and attach it to |
cae5b340 AX |
5731 | * a new anonymous hdr. Also find the last buffer |
5732 | * in the hdr's buffer list. | |
34dc7c2f | 5733 | */ |
cae5b340 AX |
5734 | arc_buf_t *lastbuf = arc_buf_remove(hdr, buf); |
5735 | ASSERT3P(lastbuf, !=, NULL); | |
5736 | ||
5737 | /* | |
5738 | * If the current arc_buf_t and the hdr are sharing their data | |
5739 | * buffer, then we must stop sharing that block. | |
5740 | */ | |
5741 | if (arc_buf_is_shared(buf)) { | |
5742 | ASSERT3P(hdr->b_l1hdr.b_buf, !=, buf); | |
5743 | VERIFY(!arc_buf_is_shared(lastbuf)); | |
5744 | ||
5745 | /* | |
5746 | * First, sever the block sharing relationship between | |
5747 | * buf and the arc_buf_hdr_t. | |
5748 | */ | |
5749 | arc_unshare_buf(hdr, buf); | |
34dc7c2f | 5750 | |
cae5b340 AX |
5751 | /* |
5752 | * Now we need to recreate the hdr's b_pabd. Since we | |
5753 | * have lastbuf handy, we try to share with it, but if | |
5754 | * we can't then we allocate a new b_pabd and copy the | |
5755 | * data from buf into it. | |
5756 | */ | |
5757 | if (arc_can_share(hdr, lastbuf)) { | |
5758 | arc_share_buf(hdr, lastbuf); | |
5759 | } else { | |
5760 | arc_hdr_alloc_pabd(hdr); | |
5761 | abd_copy_from_buf(hdr->b_l1hdr.b_pabd, | |
5762 | buf->b_data, psize); | |
5763 | } | |
5764 | VERIFY3P(lastbuf->b_data, !=, NULL); | |
5765 | } else if (HDR_SHARED_DATA(hdr)) { | |
5766 | /* | |
5767 | * Uncompressed shared buffers are always at the end | |
5768 | * of the list. Compressed buffers don't have the | |
5769 | * same requirements. This makes it hard to | |
5770 | * simply assert that the lastbuf is shared so | |
5771 | * we rely on the hdr's compression flags to determine | |
5772 | * if we have a compressed, shared buffer. | |
5773 | */ | |
5774 | ASSERT(arc_buf_is_shared(lastbuf) || | |
5775 | HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF); | |
5776 | ASSERT(!ARC_BUF_SHARED(buf)); | |
5777 | } | |
5778 | ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); | |
e10b0808 AX |
5779 | ASSERT3P(state, !=, arc_l2c_only); |
5780 | ||
cae5b340 AX |
5781 | (void) refcount_remove_many(&state->arcs_size, |
5782 | arc_buf_size(buf), buf); | |
e10b0808 AX |
5783 | |
5784 | if (refcount_is_zero(&hdr->b_l1hdr.b_refcnt)) { | |
e10b0808 | 5785 | ASSERT3P(state, !=, arc_l2c_only); |
cae5b340 AX |
5786 | (void) refcount_remove_many(&state->arcs_esize[type], |
5787 | arc_buf_size(buf), buf); | |
34dc7c2f | 5788 | } |
1eb5bfa3 | 5789 | |
cae5b340 | 5790 | hdr->b_l1hdr.b_bufcnt -= 1; |
34dc7c2f | 5791 | arc_cksum_verify(buf); |
a08ee875 | 5792 | arc_buf_unwatch(buf); |
34dc7c2f | 5793 | |
cae5b340 AX |
5794 | /* if this is the last uncompressed buf free the checksum */ |
5795 | if (!arc_hdr_has_uncompressed_buf(hdr)) | |
5796 | arc_cksum_free(hdr); | |
5797 | ||
34dc7c2f BB |
5798 | mutex_exit(hash_lock); |
5799 | ||
cae5b340 AX |
5800 | /* |
5801 | * Allocate a new hdr. The new hdr will contain a b_pabd | |
5802 | * buffer which will be freed in arc_write(). | |
5803 | */ | |
5804 | nhdr = arc_hdr_alloc(spa, psize, lsize, compress, type); | |
5805 | ASSERT3P(nhdr->b_l1hdr.b_buf, ==, NULL); | |
5806 | ASSERT0(nhdr->b_l1hdr.b_bufcnt); | |
5807 | ASSERT0(refcount_count(&nhdr->b_l1hdr.b_refcnt)); | |
5808 | VERIFY3U(nhdr->b_type, ==, type); | |
5809 | ASSERT(!HDR_SHARED_DATA(nhdr)); | |
e10b0808 | 5810 | |
cae5b340 AX |
5811 | nhdr->b_l1hdr.b_buf = buf; |
5812 | nhdr->b_l1hdr.b_bufcnt = 1; | |
e10b0808 AX |
5813 | nhdr->b_l1hdr.b_mru_hits = 0; |
5814 | nhdr->b_l1hdr.b_mru_ghost_hits = 0; | |
5815 | nhdr->b_l1hdr.b_mfu_hits = 0; | |
5816 | nhdr->b_l1hdr.b_mfu_ghost_hits = 0; | |
5817 | nhdr->b_l1hdr.b_l2_hits = 0; | |
e10b0808 | 5818 | (void) refcount_add(&nhdr->b_l1hdr.b_refcnt, tag); |
34dc7c2f | 5819 | buf->b_hdr = nhdr; |
cae5b340 | 5820 | |
428870ff | 5821 | mutex_exit(&buf->b_evict_lock); |
cae5b340 AX |
5822 | (void) refcount_add_many(&arc_anon->arcs_size, |
5823 | HDR_GET_LSIZE(nhdr), buf); | |
34dc7c2f | 5824 | } else { |
428870ff | 5825 | mutex_exit(&buf->b_evict_lock); |
e10b0808 AX |
5826 | ASSERT(refcount_count(&hdr->b_l1hdr.b_refcnt) == 1); |
5827 | /* protected by hash lock, or hdr is on arc_anon */ | |
5828 | ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); | |
34dc7c2f | 5829 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); |
e10b0808 AX |
5830 | hdr->b_l1hdr.b_mru_hits = 0; |
5831 | hdr->b_l1hdr.b_mru_ghost_hits = 0; | |
5832 | hdr->b_l1hdr.b_mfu_hits = 0; | |
5833 | hdr->b_l1hdr.b_mfu_ghost_hits = 0; | |
5834 | hdr->b_l1hdr.b_l2_hits = 0; | |
5835 | arc_change_state(arc_anon, hdr, hash_lock); | |
5836 | hdr->b_l1hdr.b_arc_access = 0; | |
5837 | mutex_exit(hash_lock); | |
34dc7c2f | 5838 | |
428870ff | 5839 | buf_discard_identity(hdr); |
34dc7c2f BB |
5840 | arc_buf_thaw(buf); |
5841 | } | |
34dc7c2f BB |
5842 | } |
5843 | ||
5844 | int | |
5845 | arc_released(arc_buf_t *buf) | |
5846 | { | |
b128c09f BB |
5847 | int released; |
5848 | ||
428870ff | 5849 | mutex_enter(&buf->b_evict_lock); |
e10b0808 AX |
5850 | released = (buf->b_data != NULL && |
5851 | buf->b_hdr->b_l1hdr.b_state == arc_anon); | |
428870ff | 5852 | mutex_exit(&buf->b_evict_lock); |
b128c09f | 5853 | return (released); |
34dc7c2f BB |
5854 | } |
5855 | ||
34dc7c2f BB |
5856 | #ifdef ZFS_DEBUG |
5857 | int | |
5858 | arc_referenced(arc_buf_t *buf) | |
5859 | { | |
b128c09f BB |
5860 | int referenced; |
5861 | ||
428870ff | 5862 | mutex_enter(&buf->b_evict_lock); |
e10b0808 | 5863 | referenced = (refcount_count(&buf->b_hdr->b_l1hdr.b_refcnt)); |
428870ff | 5864 | mutex_exit(&buf->b_evict_lock); |
b128c09f | 5865 | return (referenced); |
34dc7c2f BB |
5866 | } |
5867 | #endif | |
5868 | ||
5869 | static void | |
5870 | arc_write_ready(zio_t *zio) | |
5871 | { | |
5872 | arc_write_callback_t *callback = zio->io_private; | |
5873 | arc_buf_t *buf = callback->awcb_buf; | |
5874 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
cae5b340 AX |
5875 | uint64_t psize = BP_IS_HOLE(zio->io_bp) ? 0 : BP_GET_PSIZE(zio->io_bp); |
5876 | enum zio_compress compress; | |
5877 | fstrans_cookie_t cookie = spl_fstrans_mark(); | |
34dc7c2f | 5878 | |
e10b0808 AX |
5879 | ASSERT(HDR_HAS_L1HDR(hdr)); |
5880 | ASSERT(!refcount_is_zero(&buf->b_hdr->b_l1hdr.b_refcnt)); | |
cae5b340 AX |
5881 | ASSERT(hdr->b_l1hdr.b_bufcnt > 0); |
5882 | ||
5883 | /* | |
5884 | * If we're reexecuting this zio because the pool suspended, then | |
5885 | * cleanup any state that was previously set the first time the | |
5886 | * callback was invoked. | |
5887 | */ | |
5888 | if (zio->io_flags & ZIO_FLAG_REEXECUTED) { | |
5889 | arc_cksum_free(hdr); | |
5890 | arc_buf_unwatch(buf); | |
5891 | if (hdr->b_l1hdr.b_pabd != NULL) { | |
5892 | if (arc_buf_is_shared(buf)) { | |
5893 | arc_unshare_buf(hdr, buf); | |
5894 | } else { | |
5895 | arc_hdr_free_pabd(hdr); | |
5896 | } | |
5897 | } | |
5898 | } | |
5899 | ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); | |
5900 | ASSERT(!HDR_SHARED_DATA(hdr)); | |
5901 | ASSERT(!arc_buf_is_shared(buf)); | |
5902 | ||
b128c09f BB |
5903 | callback->awcb_ready(zio, buf, callback->awcb_private); |
5904 | ||
cae5b340 AX |
5905 | if (HDR_IO_IN_PROGRESS(hdr)) |
5906 | ASSERT(zio->io_flags & ZIO_FLAG_REEXECUTED); | |
5907 | ||
5908 | arc_cksum_compute(buf); | |
5909 | arc_hdr_set_flags(hdr, ARC_FLAG_IO_IN_PROGRESS); | |
5910 | ||
5911 | if (BP_IS_HOLE(zio->io_bp) || BP_IS_EMBEDDED(zio->io_bp)) { | |
5912 | compress = ZIO_COMPRESS_OFF; | |
5913 | } else { | |
5914 | ASSERT3U(HDR_GET_LSIZE(hdr), ==, BP_GET_LSIZE(zio->io_bp)); | |
5915 | compress = BP_GET_COMPRESS(zio->io_bp); | |
5916 | } | |
5917 | HDR_SET_PSIZE(hdr, psize); | |
5918 | arc_hdr_set_compress(hdr, compress); | |
5919 | ||
34dc7c2f | 5920 | /* |
cae5b340 AX |
5921 | * Fill the hdr with data. If the hdr is compressed, the data we want |
5922 | * is available from the zio, otherwise we can take it from the buf. | |
5923 | * | |
5924 | * We might be able to share the buf's data with the hdr here. However, | |
5925 | * doing so would cause the ARC to be full of linear ABDs if we write a | |
5926 | * lot of shareable data. As a compromise, we check whether scattered | |
5927 | * ABDs are allowed, and assume that if they are then the user wants | |
5928 | * the ARC to be primarily filled with them regardless of the data being | |
5929 | * written. Therefore, if they're allowed then we allocate one and copy | |
5930 | * the data into it; otherwise, we share the data directly if we can. | |
34dc7c2f | 5931 | */ |
cae5b340 AX |
5932 | if (zfs_abd_scatter_enabled || !arc_can_share(hdr, buf)) { |
5933 | arc_hdr_alloc_pabd(hdr); | |
5934 | ||
5935 | /* | |
5936 | * Ideally, we would always copy the io_abd into b_pabd, but the | |
5937 | * user may have disabled compressed ARC, thus we must check the | |
5938 | * hdr's compression setting rather than the io_bp's. | |
5939 | */ | |
5940 | if (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF) { | |
5941 | ASSERT3U(BP_GET_COMPRESS(zio->io_bp), !=, | |
5942 | ZIO_COMPRESS_OFF); | |
5943 | ASSERT3U(psize, >, 0); | |
5944 | ||
5945 | abd_copy(hdr->b_l1hdr.b_pabd, zio->io_abd, psize); | |
5946 | } else { | |
5947 | ASSERT3U(zio->io_orig_size, ==, arc_hdr_size(hdr)); | |
5948 | ||
5949 | abd_copy_from_buf(hdr->b_l1hdr.b_pabd, buf->b_data, | |
5950 | arc_buf_size(buf)); | |
34dc7c2f | 5951 | } |
cae5b340 AX |
5952 | } else { |
5953 | ASSERT3P(buf->b_data, ==, abd_to_buf(zio->io_orig_abd)); | |
5954 | ASSERT3U(zio->io_orig_size, ==, arc_buf_size(buf)); | |
5955 | ASSERT3U(hdr->b_l1hdr.b_bufcnt, ==, 1); | |
5956 | ||
5957 | arc_share_buf(hdr, buf); | |
34dc7c2f | 5958 | } |
cae5b340 AX |
5959 | |
5960 | arc_hdr_verify(hdr, zio->io_bp); | |
5961 | spl_fstrans_unmark(cookie); | |
5962 | } | |
5963 | ||
5964 | static void | |
5965 | arc_write_children_ready(zio_t *zio) | |
5966 | { | |
5967 | arc_write_callback_t *callback = zio->io_private; | |
5968 | arc_buf_t *buf = callback->awcb_buf; | |
5969 | ||
5970 | callback->awcb_children_ready(zio, buf, callback->awcb_private); | |
34dc7c2f BB |
5971 | } |
5972 | ||
a08ee875 LG |
5973 | /* |
5974 | * The SPA calls this callback for each physical write that happens on behalf | |
5975 | * of a logical write. See the comment in dbuf_write_physdone() for details. | |
5976 | */ | |
5977 | static void | |
5978 | arc_write_physdone(zio_t *zio) | |
5979 | { | |
5980 | arc_write_callback_t *cb = zio->io_private; | |
5981 | if (cb->awcb_physdone != NULL) | |
5982 | cb->awcb_physdone(zio, cb->awcb_buf, cb->awcb_private); | |
5983 | } | |
5984 | ||
34dc7c2f BB |
5985 | static void |
5986 | arc_write_done(zio_t *zio) | |
5987 | { | |
5988 | arc_write_callback_t *callback = zio->io_private; | |
5989 | arc_buf_t *buf = callback->awcb_buf; | |
5990 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
5991 | ||
cae5b340 | 5992 | ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL); |
428870ff BB |
5993 | |
5994 | if (zio->io_error == 0) { | |
cae5b340 AX |
5995 | arc_hdr_verify(hdr, zio->io_bp); |
5996 | ||
ea04106b AX |
5997 | if (BP_IS_HOLE(zio->io_bp) || BP_IS_EMBEDDED(zio->io_bp)) { |
5998 | buf_discard_identity(hdr); | |
5999 | } else { | |
6000 | hdr->b_dva = *BP_IDENTITY(zio->io_bp); | |
6001 | hdr->b_birth = BP_PHYSICAL_BIRTH(zio->io_bp); | |
ea04106b | 6002 | } |
428870ff | 6003 | } else { |
cae5b340 | 6004 | ASSERT(HDR_EMPTY(hdr)); |
428870ff | 6005 | } |
34dc7c2f | 6006 | |
34dc7c2f | 6007 | /* |
ea04106b AX |
6008 | * If the block to be written was all-zero or compressed enough to be |
6009 | * embedded in the BP, no write was performed so there will be no | |
6010 | * dva/birth/checksum. The buffer must therefore remain anonymous | |
6011 | * (and uncached). | |
34dc7c2f | 6012 | */ |
cae5b340 | 6013 | if (!HDR_EMPTY(hdr)) { |
34dc7c2f BB |
6014 | arc_buf_hdr_t *exists; |
6015 | kmutex_t *hash_lock; | |
6016 | ||
cae5b340 | 6017 | ASSERT3U(zio->io_error, ==, 0); |
428870ff | 6018 | |
34dc7c2f BB |
6019 | arc_cksum_verify(buf); |
6020 | ||
6021 | exists = buf_hash_insert(hdr, &hash_lock); | |
e10b0808 | 6022 | if (exists != NULL) { |
34dc7c2f BB |
6023 | /* |
6024 | * This can only happen if we overwrite for | |
6025 | * sync-to-convergence, because we remove | |
6026 | * buffers from the hash table when we arc_free(). | |
6027 | */ | |
428870ff BB |
6028 | if (zio->io_flags & ZIO_FLAG_IO_REWRITE) { |
6029 | if (!BP_EQUAL(&zio->io_bp_orig, zio->io_bp)) | |
6030 | panic("bad overwrite, hdr=%p exists=%p", | |
6031 | (void *)hdr, (void *)exists); | |
e10b0808 AX |
6032 | ASSERT(refcount_is_zero( |
6033 | &exists->b_l1hdr.b_refcnt)); | |
428870ff BB |
6034 | arc_change_state(arc_anon, exists, hash_lock); |
6035 | mutex_exit(hash_lock); | |
6036 | arc_hdr_destroy(exists); | |
6037 | exists = buf_hash_insert(hdr, &hash_lock); | |
6038 | ASSERT3P(exists, ==, NULL); | |
a08ee875 LG |
6039 | } else if (zio->io_flags & ZIO_FLAG_NOPWRITE) { |
6040 | /* nopwrite */ | |
6041 | ASSERT(zio->io_prop.zp_nopwrite); | |
6042 | if (!BP_EQUAL(&zio->io_bp_orig, zio->io_bp)) | |
6043 | panic("bad nopwrite, hdr=%p exists=%p", | |
6044 | (void *)hdr, (void *)exists); | |
428870ff BB |
6045 | } else { |
6046 | /* Dedup */ | |
cae5b340 | 6047 | ASSERT(hdr->b_l1hdr.b_bufcnt == 1); |
e10b0808 | 6048 | ASSERT(hdr->b_l1hdr.b_state == arc_anon); |
428870ff BB |
6049 | ASSERT(BP_GET_DEDUP(zio->io_bp)); |
6050 | ASSERT(BP_GET_LEVEL(zio->io_bp) == 0); | |
6051 | } | |
34dc7c2f | 6052 | } |
cae5b340 | 6053 | arc_hdr_clear_flags(hdr, ARC_FLAG_IO_IN_PROGRESS); |
b128c09f | 6054 | /* if it's not anon, we are doing a scrub */ |
e10b0808 | 6055 | if (exists == NULL && hdr->b_l1hdr.b_state == arc_anon) |
b128c09f | 6056 | arc_access(hdr, hash_lock); |
34dc7c2f | 6057 | mutex_exit(hash_lock); |
34dc7c2f | 6058 | } else { |
cae5b340 | 6059 | arc_hdr_clear_flags(hdr, ARC_FLAG_IO_IN_PROGRESS); |
34dc7c2f BB |
6060 | } |
6061 | ||
e10b0808 | 6062 | ASSERT(!refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); |
428870ff | 6063 | callback->awcb_done(zio, buf, callback->awcb_private); |
34dc7c2f | 6064 | |
cae5b340 | 6065 | abd_put(zio->io_abd); |
34dc7c2f BB |
6066 | kmem_free(callback, sizeof (arc_write_callback_t)); |
6067 | } | |
6068 | ||
6069 | zio_t * | |
428870ff | 6070 | arc_write(zio_t *pio, spa_t *spa, uint64_t txg, |
cae5b340 AX |
6071 | blkptr_t *bp, arc_buf_t *buf, boolean_t l2arc, |
6072 | const zio_prop_t *zp, arc_done_func_t *ready, | |
6073 | arc_done_func_t *children_ready, arc_done_func_t *physdone, | |
a08ee875 | 6074 | arc_done_func_t *done, void *private, zio_priority_t priority, |
ea04106b | 6075 | int zio_flags, const zbookmark_phys_t *zb) |
34dc7c2f BB |
6076 | { |
6077 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
6078 | arc_write_callback_t *callback; | |
b128c09f | 6079 | zio_t *zio; |
cae5b340 | 6080 | zio_prop_t localprop = *zp; |
34dc7c2f | 6081 | |
cae5b340 AX |
6082 | ASSERT3P(ready, !=, NULL); |
6083 | ASSERT3P(done, !=, NULL); | |
34dc7c2f | 6084 | ASSERT(!HDR_IO_ERROR(hdr)); |
e10b0808 | 6085 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); |
cae5b340 AX |
6086 | ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL); |
6087 | ASSERT3U(hdr->b_l1hdr.b_bufcnt, >, 0); | |
b128c09f | 6088 | if (l2arc) |
cae5b340 AX |
6089 | arc_hdr_set_flags(hdr, ARC_FLAG_L2CACHE); |
6090 | if (ARC_BUF_COMPRESSED(buf)) { | |
6091 | /* | |
6092 | * We're writing a pre-compressed buffer. Make the | |
6093 | * compression algorithm requested by the zio_prop_t match | |
6094 | * the pre-compressed buffer's compression algorithm. | |
6095 | */ | |
6096 | localprop.zp_compress = HDR_GET_COMPRESS(hdr); | |
6097 | ||
6098 | ASSERT3U(HDR_GET_LSIZE(hdr), !=, arc_buf_size(buf)); | |
6099 | zio_flags |= ZIO_FLAG_RAW; | |
6100 | } | |
ea04106b | 6101 | callback = kmem_zalloc(sizeof (arc_write_callback_t), KM_SLEEP); |
34dc7c2f | 6102 | callback->awcb_ready = ready; |
cae5b340 | 6103 | callback->awcb_children_ready = children_ready; |
a08ee875 | 6104 | callback->awcb_physdone = physdone; |
34dc7c2f BB |
6105 | callback->awcb_done = done; |
6106 | callback->awcb_private = private; | |
6107 | callback->awcb_buf = buf; | |
b128c09f | 6108 | |
cae5b340 AX |
6109 | /* |
6110 | * The hdr's b_pabd is now stale, free it now. A new data block | |
6111 | * will be allocated when the zio pipeline calls arc_write_ready(). | |
6112 | */ | |
6113 | if (hdr->b_l1hdr.b_pabd != NULL) { | |
6114 | /* | |
6115 | * If the buf is currently sharing the data block with | |
6116 | * the hdr then we need to break that relationship here. | |
6117 | * The hdr will remain with a NULL data pointer and the | |
6118 | * buf will take sole ownership of the block. | |
6119 | */ | |
6120 | if (arc_buf_is_shared(buf)) { | |
6121 | arc_unshare_buf(hdr, buf); | |
6122 | } else { | |
6123 | arc_hdr_free_pabd(hdr); | |
6124 | } | |
6125 | VERIFY3P(buf->b_data, !=, NULL); | |
6126 | arc_hdr_set_compress(hdr, ZIO_COMPRESS_OFF); | |
6127 | } | |
41d74433 AX |
6128 | |
6129 | if (!(zio_flags & ZIO_FLAG_RAW)) | |
6130 | arc_hdr_set_compress(hdr, ZIO_COMPRESS_OFF); | |
6131 | ||
cae5b340 AX |
6132 | ASSERT(!arc_buf_is_shared(buf)); |
6133 | ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); | |
6134 | ||
6135 | zio = zio_write(pio, spa, txg, bp, | |
6136 | abd_get_from_buf(buf->b_data, HDR_GET_LSIZE(hdr)), | |
6137 | HDR_GET_LSIZE(hdr), arc_buf_size(buf), &localprop, arc_write_ready, | |
6138 | (children_ready != NULL) ? arc_write_children_ready : NULL, | |
6139 | arc_write_physdone, arc_write_done, callback, | |
a08ee875 | 6140 | priority, zio_flags, zb); |
34dc7c2f BB |
6141 | |
6142 | return (zio); | |
6143 | } | |
6144 | ||
34dc7c2f | 6145 | static int |
a08ee875 | 6146 | arc_memory_throttle(uint64_t reserve, uint64_t txg) |
34dc7c2f BB |
6147 | { |
6148 | #ifdef _KERNEL | |
8ec27e97 | 6149 | uint64_t available_memory = arc_free_memory(); |
e10b0808 AX |
6150 | static uint64_t page_load = 0; |
6151 | static uint64_t last_txg = 0; | |
e10b0808 | 6152 | |
8ec27e97 | 6153 | #if defined(_ILP32) |
cae5b340 AX |
6154 | available_memory = |
6155 | MIN(available_memory, vmem_size(heap_arena, VMEM_FREE)); | |
6156 | #endif | |
6157 | ||
6158 | if (available_memory > arc_all_memory() * arc_lotsfree_percent / 100) | |
0c5493d4 BB |
6159 | return (0); |
6160 | ||
e10b0808 AX |
6161 | if (txg > last_txg) { |
6162 | last_txg = txg; | |
6163 | page_load = 0; | |
6164 | } | |
e10b0808 AX |
6165 | /* |
6166 | * If we are in pageout, we know that memory is already tight, | |
6167 | * the arc is already going to be evicting, so we just want to | |
6168 | * continue to let page writes occur as quickly as possible. | |
6169 | */ | |
6170 | if (current_is_kswapd()) { | |
8ec27e97 | 6171 | if (page_load > MAX(arc_sys_free / 4, available_memory) / 4) { |
e10b0808 AX |
6172 | DMU_TX_STAT_BUMP(dmu_tx_memory_reclaim); |
6173 | return (SET_ERROR(ERESTART)); | |
6174 | } | |
6175 | /* Note: reserve is inflated, so we deflate */ | |
6176 | page_load += reserve / 8; | |
6177 | return (0); | |
6178 | } else if (page_load > 0 && arc_reclaim_needed()) { | |
6179 | /* memory is low, delay before restarting */ | |
34dc7c2f | 6180 | ARCSTAT_INCR(arcstat_memory_throttle_count, 1); |
570827e1 | 6181 | DMU_TX_STAT_BUMP(dmu_tx_memory_reclaim); |
a08ee875 | 6182 | return (SET_ERROR(EAGAIN)); |
34dc7c2f | 6183 | } |
e10b0808 | 6184 | page_load = 0; |
34dc7c2f BB |
6185 | #endif |
6186 | return (0); | |
6187 | } | |
6188 | ||
6189 | void | |
6190 | arc_tempreserve_clear(uint64_t reserve) | |
6191 | { | |
6192 | atomic_add_64(&arc_tempreserve, -reserve); | |
6193 | ASSERT((int64_t)arc_tempreserve >= 0); | |
6194 | } | |
6195 | ||
6196 | int | |
6197 | arc_tempreserve_space(uint64_t reserve, uint64_t txg) | |
6198 | { | |
6199 | int error; | |
9babb374 | 6200 | uint64_t anon_size; |
34dc7c2f | 6201 | |
4d815aed AX |
6202 | if (!arc_no_grow && |
6203 | reserve > arc_c/4 && | |
6204 | reserve * 4 > (2ULL << SPA_MAXBLOCKSHIFT)) | |
34dc7c2f | 6205 | arc_c = MIN(arc_c_max, reserve * 4); |
ea04106b AX |
6206 | |
6207 | /* | |
6208 | * Throttle when the calculated memory footprint for the TXG | |
6209 | * exceeds the target ARC size. | |
6210 | */ | |
570827e1 BB |
6211 | if (reserve > arc_c) { |
6212 | DMU_TX_STAT_BUMP(dmu_tx_memory_reserve); | |
ea04106b | 6213 | return (SET_ERROR(ERESTART)); |
570827e1 | 6214 | } |
34dc7c2f | 6215 | |
9babb374 BB |
6216 | /* |
6217 | * Don't count loaned bufs as in flight dirty data to prevent long | |
6218 | * network delays from blocking transactions that are ready to be | |
6219 | * assigned to a txg. | |
6220 | */ | |
cae5b340 AX |
6221 | |
6222 | /* assert that it has not wrapped around */ | |
6223 | ASSERT3S(atomic_add_64_nv(&arc_loaned_bytes, 0), >=, 0); | |
6224 | ||
e10b0808 AX |
6225 | anon_size = MAX((int64_t)(refcount_count(&arc_anon->arcs_size) - |
6226 | arc_loaned_bytes), 0); | |
9babb374 | 6227 | |
34dc7c2f BB |
6228 | /* |
6229 | * Writes will, almost always, require additional memory allocations | |
a08ee875 | 6230 | * in order to compress/encrypt/etc the data. We therefore need to |
34dc7c2f BB |
6231 | * make sure that there is sufficient available memory for this. |
6232 | */ | |
a08ee875 LG |
6233 | error = arc_memory_throttle(reserve, txg); |
6234 | if (error != 0) | |
34dc7c2f BB |
6235 | return (error); |
6236 | ||
6237 | /* | |
6238 | * Throttle writes when the amount of dirty data in the cache | |
6239 | * gets too large. We try to keep the cache less than half full | |
6240 | * of dirty blocks so that our sync times don't grow too large. | |
6241 | * Note: if two requests come in concurrently, we might let them | |
6242 | * both succeed, when one of them should fail. Not a huge deal. | |
6243 | */ | |
9babb374 BB |
6244 | |
6245 | if (reserve + arc_tempreserve + anon_size > arc_c / 2 && | |
6246 | anon_size > arc_c / 4) { | |
cae5b340 AX |
6247 | uint64_t meta_esize = |
6248 | refcount_count(&arc_anon->arcs_esize[ARC_BUFC_METADATA]); | |
6249 | uint64_t data_esize = | |
6250 | refcount_count(&arc_anon->arcs_esize[ARC_BUFC_DATA]); | |
34dc7c2f BB |
6251 | dprintf("failing, arc_tempreserve=%lluK anon_meta=%lluK " |
6252 | "anon_data=%lluK tempreserve=%lluK arc_c=%lluK\n", | |
cae5b340 AX |
6253 | arc_tempreserve >> 10, meta_esize >> 10, |
6254 | data_esize >> 10, reserve >> 10, arc_c >> 10); | |
570827e1 | 6255 | DMU_TX_STAT_BUMP(dmu_tx_dirty_throttle); |
a08ee875 | 6256 | return (SET_ERROR(ERESTART)); |
34dc7c2f BB |
6257 | } |
6258 | atomic_add_64(&arc_tempreserve, reserve); | |
6259 | return (0); | |
6260 | } | |
6261 | ||
13be560d BB |
6262 | static void |
6263 | arc_kstat_update_state(arc_state_t *state, kstat_named_t *size, | |
6264 | kstat_named_t *evict_data, kstat_named_t *evict_metadata) | |
6265 | { | |
e10b0808 | 6266 | size->value.ui64 = refcount_count(&state->arcs_size); |
cae5b340 AX |
6267 | evict_data->value.ui64 = |
6268 | refcount_count(&state->arcs_esize[ARC_BUFC_DATA]); | |
6269 | evict_metadata->value.ui64 = | |
6270 | refcount_count(&state->arcs_esize[ARC_BUFC_METADATA]); | |
13be560d BB |
6271 | } |
6272 | ||
6273 | static int | |
6274 | arc_kstat_update(kstat_t *ksp, int rw) | |
6275 | { | |
6276 | arc_stats_t *as = ksp->ks_data; | |
6277 | ||
6278 | if (rw == KSTAT_WRITE) { | |
e10b0808 | 6279 | return (EACCES); |
13be560d BB |
6280 | } else { |
6281 | arc_kstat_update_state(arc_anon, | |
6282 | &as->arcstat_anon_size, | |
e10b0808 AX |
6283 | &as->arcstat_anon_evictable_data, |
6284 | &as->arcstat_anon_evictable_metadata); | |
13be560d BB |
6285 | arc_kstat_update_state(arc_mru, |
6286 | &as->arcstat_mru_size, | |
e10b0808 AX |
6287 | &as->arcstat_mru_evictable_data, |
6288 | &as->arcstat_mru_evictable_metadata); | |
13be560d BB |
6289 | arc_kstat_update_state(arc_mru_ghost, |
6290 | &as->arcstat_mru_ghost_size, | |
e10b0808 AX |
6291 | &as->arcstat_mru_ghost_evictable_data, |
6292 | &as->arcstat_mru_ghost_evictable_metadata); | |
13be560d BB |
6293 | arc_kstat_update_state(arc_mfu, |
6294 | &as->arcstat_mfu_size, | |
e10b0808 AX |
6295 | &as->arcstat_mfu_evictable_data, |
6296 | &as->arcstat_mfu_evictable_metadata); | |
fc41c640 | 6297 | arc_kstat_update_state(arc_mfu_ghost, |
13be560d | 6298 | &as->arcstat_mfu_ghost_size, |
e10b0808 AX |
6299 | &as->arcstat_mfu_ghost_evictable_data, |
6300 | &as->arcstat_mfu_ghost_evictable_metadata); | |
8ec27e97 AX |
6301 | |
6302 | as->arcstat_memory_all_bytes.value.ui64 = | |
6303 | arc_all_memory(); | |
6304 | as->arcstat_memory_free_bytes.value.ui64 = | |
6305 | arc_free_memory(); | |
6306 | as->arcstat_memory_available_bytes.value.i64 = | |
6307 | arc_available_memory(); | |
13be560d BB |
6308 | } |
6309 | ||
6310 | return (0); | |
6311 | } | |
6312 | ||
e10b0808 AX |
6313 | /* |
6314 | * This function *must* return indices evenly distributed between all | |
6315 | * sublists of the multilist. This is needed due to how the ARC eviction | |
6316 | * code is laid out; arc_evict_state() assumes ARC buffers are evenly | |
6317 | * distributed between all sublists and uses this assumption when | |
6318 | * deciding which sublist to evict from and how much to evict from it. | |
6319 | */ | |
6320 | unsigned int | |
6321 | arc_state_multilist_index_func(multilist_t *ml, void *obj) | |
6322 | { | |
6323 | arc_buf_hdr_t *hdr = obj; | |
6324 | ||
6325 | /* | |
6326 | * We rely on b_dva to generate evenly distributed index | |
6327 | * numbers using buf_hash below. So, as an added precaution, | |
6328 | * let's make sure we never add empty buffers to the arc lists. | |
6329 | */ | |
cae5b340 | 6330 | ASSERT(!HDR_EMPTY(hdr)); |
e10b0808 AX |
6331 | |
6332 | /* | |
6333 | * The assumption here, is the hash value for a given | |
6334 | * arc_buf_hdr_t will remain constant throughout its lifetime | |
6335 | * (i.e. its b_spa, b_dva, and b_birth fields don't change). | |
6336 | * Thus, we don't need to store the header's sublist index | |
6337 | * on insertion, as this index can be recalculated on removal. | |
6338 | * | |
6339 | * Also, the low order bits of the hash value are thought to be | |
6340 | * distributed evenly. Otherwise, in the case that the multilist | |
6341 | * has a power of two number of sublists, each sublists' usage | |
6342 | * would not be evenly distributed. | |
6343 | */ | |
6344 | return (buf_hash(hdr->b_spa, &hdr->b_dva, hdr->b_birth) % | |
6345 | multilist_get_num_sublists(ml)); | |
6346 | } | |
6347 | ||
6348 | /* | |
6349 | * Called during module initialization and periodically thereafter to | |
6350 | * apply reasonable changes to the exposed performance tunings. Non-zero | |
6351 | * zfs_* values which differ from the currently set values will be applied. | |
6352 | */ | |
6353 | static void | |
6354 | arc_tuning_update(void) | |
6355 | { | |
cae5b340 AX |
6356 | uint64_t allmem = arc_all_memory(); |
6357 | unsigned long limit; | |
6358 | ||
e10b0808 AX |
6359 | /* Valid range: 64M - <all physical memory> */ |
6360 | if ((zfs_arc_max) && (zfs_arc_max != arc_c_max) && | |
42f7b73b | 6361 | (zfs_arc_max >= 64 << 20) && (zfs_arc_max < allmem) && |
e10b0808 AX |
6362 | (zfs_arc_max > arc_c_min)) { |
6363 | arc_c_max = zfs_arc_max; | |
6364 | arc_c = arc_c_max; | |
6365 | arc_p = (arc_c >> 1); | |
cae5b340 AX |
6366 | if (arc_meta_limit > arc_c_max) |
6367 | arc_meta_limit = arc_c_max; | |
6368 | if (arc_dnode_limit > arc_meta_limit) | |
6369 | arc_dnode_limit = arc_meta_limit; | |
e10b0808 AX |
6370 | } |
6371 | ||
6372 | /* Valid range: 32M - <arc_c_max> */ | |
6373 | if ((zfs_arc_min) && (zfs_arc_min != arc_c_min) && | |
6374 | (zfs_arc_min >= 2ULL << SPA_MAXBLOCKSHIFT) && | |
6375 | (zfs_arc_min <= arc_c_max)) { | |
6376 | arc_c_min = zfs_arc_min; | |
6377 | arc_c = MAX(arc_c, arc_c_min); | |
6378 | } | |
6379 | ||
6380 | /* Valid range: 16M - <arc_c_max> */ | |
6381 | if ((zfs_arc_meta_min) && (zfs_arc_meta_min != arc_meta_min) && | |
6382 | (zfs_arc_meta_min >= 1ULL << SPA_MAXBLOCKSHIFT) && | |
6383 | (zfs_arc_meta_min <= arc_c_max)) { | |
6384 | arc_meta_min = zfs_arc_meta_min; | |
cae5b340 AX |
6385 | if (arc_meta_limit < arc_meta_min) |
6386 | arc_meta_limit = arc_meta_min; | |
6387 | if (arc_dnode_limit < arc_meta_min) | |
6388 | arc_dnode_limit = arc_meta_min; | |
e10b0808 AX |
6389 | } |
6390 | ||
6391 | /* Valid range: <arc_meta_min> - <arc_c_max> */ | |
cae5b340 AX |
6392 | limit = zfs_arc_meta_limit ? zfs_arc_meta_limit : |
6393 | MIN(zfs_arc_meta_limit_percent, 100) * arc_c_max / 100; | |
6394 | if ((limit != arc_meta_limit) && | |
6395 | (limit >= arc_meta_min) && | |
6396 | (limit <= arc_c_max)) | |
6397 | arc_meta_limit = limit; | |
6398 | ||
6399 | /* Valid range: <arc_meta_min> - <arc_meta_limit> */ | |
6400 | limit = zfs_arc_dnode_limit ? zfs_arc_dnode_limit : | |
6401 | MIN(zfs_arc_dnode_limit_percent, 100) * arc_meta_limit / 100; | |
6402 | if ((limit != arc_dnode_limit) && | |
6403 | (limit >= arc_meta_min) && | |
6404 | (limit <= arc_meta_limit)) | |
6405 | arc_dnode_limit = limit; | |
e10b0808 AX |
6406 | |
6407 | /* Valid range: 1 - N */ | |
6408 | if (zfs_arc_grow_retry) | |
6409 | arc_grow_retry = zfs_arc_grow_retry; | |
6410 | ||
cae5b340 AX |
6411 | /* Valid range: 1 - N */ |
6412 | if (zfs_arc_shrink_shift) { | |
6413 | arc_shrink_shift = zfs_arc_shrink_shift; | |
6414 | arc_no_grow_shift = MIN(arc_no_grow_shift, arc_shrink_shift -1); | |
6415 | } | |
6416 | ||
6417 | /* Valid range: 1 - N */ | |
6418 | if (zfs_arc_p_min_shift) | |
6419 | arc_p_min_shift = zfs_arc_p_min_shift; | |
6420 | ||
6421 | /* Valid range: 1 - N ticks */ | |
6422 | if (zfs_arc_min_prefetch_lifespan) | |
6423 | arc_min_prefetch_lifespan = zfs_arc_min_prefetch_lifespan; | |
6424 | ||
6425 | /* Valid range: 0 - 100 */ | |
6426 | if ((zfs_arc_lotsfree_percent >= 0) && | |
6427 | (zfs_arc_lotsfree_percent <= 100)) | |
6428 | arc_lotsfree_percent = zfs_arc_lotsfree_percent; | |
6429 | ||
6430 | /* Valid range: 0 - <all physical memory> */ | |
6431 | if ((zfs_arc_sys_free) && (zfs_arc_sys_free != arc_sys_free)) | |
6432 | arc_sys_free = MIN(MAX(zfs_arc_sys_free, 0), allmem); | |
6433 | ||
6434 | } | |
6435 | ||
6436 | static void | |
6437 | arc_state_init(void) | |
6438 | { | |
6439 | arc_anon = &ARC_anon; | |
6440 | arc_mru = &ARC_mru; | |
6441 | arc_mru_ghost = &ARC_mru_ghost; | |
6442 | arc_mfu = &ARC_mfu; | |
6443 | arc_mfu_ghost = &ARC_mfu_ghost; | |
6444 | arc_l2c_only = &ARC_l2c_only; | |
6445 | ||
6446 | arc_mru->arcs_list[ARC_BUFC_METADATA] = | |
6447 | multilist_create(sizeof (arc_buf_hdr_t), | |
6448 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
6449 | arc_state_multilist_index_func); | |
6450 | arc_mru->arcs_list[ARC_BUFC_DATA] = | |
6451 | multilist_create(sizeof (arc_buf_hdr_t), | |
6452 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
6453 | arc_state_multilist_index_func); | |
6454 | arc_mru_ghost->arcs_list[ARC_BUFC_METADATA] = | |
6455 | multilist_create(sizeof (arc_buf_hdr_t), | |
6456 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
6457 | arc_state_multilist_index_func); | |
6458 | arc_mru_ghost->arcs_list[ARC_BUFC_DATA] = | |
6459 | multilist_create(sizeof (arc_buf_hdr_t), | |
6460 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
6461 | arc_state_multilist_index_func); | |
6462 | arc_mfu->arcs_list[ARC_BUFC_METADATA] = | |
6463 | multilist_create(sizeof (arc_buf_hdr_t), | |
6464 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
6465 | arc_state_multilist_index_func); | |
6466 | arc_mfu->arcs_list[ARC_BUFC_DATA] = | |
6467 | multilist_create(sizeof (arc_buf_hdr_t), | |
6468 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
6469 | arc_state_multilist_index_func); | |
6470 | arc_mfu_ghost->arcs_list[ARC_BUFC_METADATA] = | |
6471 | multilist_create(sizeof (arc_buf_hdr_t), | |
6472 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
6473 | arc_state_multilist_index_func); | |
6474 | arc_mfu_ghost->arcs_list[ARC_BUFC_DATA] = | |
6475 | multilist_create(sizeof (arc_buf_hdr_t), | |
6476 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
6477 | arc_state_multilist_index_func); | |
6478 | arc_l2c_only->arcs_list[ARC_BUFC_METADATA] = | |
6479 | multilist_create(sizeof (arc_buf_hdr_t), | |
6480 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
6481 | arc_state_multilist_index_func); | |
6482 | arc_l2c_only->arcs_list[ARC_BUFC_DATA] = | |
6483 | multilist_create(sizeof (arc_buf_hdr_t), | |
6484 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
6485 | arc_state_multilist_index_func); | |
6486 | ||
6487 | refcount_create(&arc_anon->arcs_esize[ARC_BUFC_METADATA]); | |
6488 | refcount_create(&arc_anon->arcs_esize[ARC_BUFC_DATA]); | |
6489 | refcount_create(&arc_mru->arcs_esize[ARC_BUFC_METADATA]); | |
6490 | refcount_create(&arc_mru->arcs_esize[ARC_BUFC_DATA]); | |
6491 | refcount_create(&arc_mru_ghost->arcs_esize[ARC_BUFC_METADATA]); | |
6492 | refcount_create(&arc_mru_ghost->arcs_esize[ARC_BUFC_DATA]); | |
6493 | refcount_create(&arc_mfu->arcs_esize[ARC_BUFC_METADATA]); | |
6494 | refcount_create(&arc_mfu->arcs_esize[ARC_BUFC_DATA]); | |
6495 | refcount_create(&arc_mfu_ghost->arcs_esize[ARC_BUFC_METADATA]); | |
6496 | refcount_create(&arc_mfu_ghost->arcs_esize[ARC_BUFC_DATA]); | |
6497 | refcount_create(&arc_l2c_only->arcs_esize[ARC_BUFC_METADATA]); | |
6498 | refcount_create(&arc_l2c_only->arcs_esize[ARC_BUFC_DATA]); | |
6499 | ||
6500 | refcount_create(&arc_anon->arcs_size); | |
6501 | refcount_create(&arc_mru->arcs_size); | |
6502 | refcount_create(&arc_mru_ghost->arcs_size); | |
6503 | refcount_create(&arc_mfu->arcs_size); | |
6504 | refcount_create(&arc_mfu_ghost->arcs_size); | |
6505 | refcount_create(&arc_l2c_only->arcs_size); | |
e10b0808 | 6506 | |
cae5b340 AX |
6507 | arc_anon->arcs_state = ARC_STATE_ANON; |
6508 | arc_mru->arcs_state = ARC_STATE_MRU; | |
6509 | arc_mru_ghost->arcs_state = ARC_STATE_MRU_GHOST; | |
6510 | arc_mfu->arcs_state = ARC_STATE_MFU; | |
6511 | arc_mfu_ghost->arcs_state = ARC_STATE_MFU_GHOST; | |
6512 | arc_l2c_only->arcs_state = ARC_STATE_L2C_ONLY; | |
6513 | } | |
e10b0808 | 6514 | |
cae5b340 AX |
6515 | static void |
6516 | arc_state_fini(void) | |
6517 | { | |
6518 | refcount_destroy(&arc_anon->arcs_esize[ARC_BUFC_METADATA]); | |
6519 | refcount_destroy(&arc_anon->arcs_esize[ARC_BUFC_DATA]); | |
6520 | refcount_destroy(&arc_mru->arcs_esize[ARC_BUFC_METADATA]); | |
6521 | refcount_destroy(&arc_mru->arcs_esize[ARC_BUFC_DATA]); | |
6522 | refcount_destroy(&arc_mru_ghost->arcs_esize[ARC_BUFC_METADATA]); | |
6523 | refcount_destroy(&arc_mru_ghost->arcs_esize[ARC_BUFC_DATA]); | |
6524 | refcount_destroy(&arc_mfu->arcs_esize[ARC_BUFC_METADATA]); | |
6525 | refcount_destroy(&arc_mfu->arcs_esize[ARC_BUFC_DATA]); | |
6526 | refcount_destroy(&arc_mfu_ghost->arcs_esize[ARC_BUFC_METADATA]); | |
6527 | refcount_destroy(&arc_mfu_ghost->arcs_esize[ARC_BUFC_DATA]); | |
6528 | refcount_destroy(&arc_l2c_only->arcs_esize[ARC_BUFC_METADATA]); | |
6529 | refcount_destroy(&arc_l2c_only->arcs_esize[ARC_BUFC_DATA]); | |
e10b0808 | 6530 | |
cae5b340 AX |
6531 | refcount_destroy(&arc_anon->arcs_size); |
6532 | refcount_destroy(&arc_mru->arcs_size); | |
6533 | refcount_destroy(&arc_mru_ghost->arcs_size); | |
6534 | refcount_destroy(&arc_mfu->arcs_size); | |
6535 | refcount_destroy(&arc_mfu_ghost->arcs_size); | |
6536 | refcount_destroy(&arc_l2c_only->arcs_size); | |
e10b0808 | 6537 | |
cae5b340 AX |
6538 | multilist_destroy(arc_mru->arcs_list[ARC_BUFC_METADATA]); |
6539 | multilist_destroy(arc_mru_ghost->arcs_list[ARC_BUFC_METADATA]); | |
6540 | multilist_destroy(arc_mfu->arcs_list[ARC_BUFC_METADATA]); | |
6541 | multilist_destroy(arc_mfu_ghost->arcs_list[ARC_BUFC_METADATA]); | |
6542 | multilist_destroy(arc_mru->arcs_list[ARC_BUFC_DATA]); | |
6543 | multilist_destroy(arc_mru_ghost->arcs_list[ARC_BUFC_DATA]); | |
6544 | multilist_destroy(arc_mfu->arcs_list[ARC_BUFC_DATA]); | |
6545 | multilist_destroy(arc_mfu_ghost->arcs_list[ARC_BUFC_DATA]); | |
6546 | multilist_destroy(arc_l2c_only->arcs_list[ARC_BUFC_METADATA]); | |
6547 | multilist_destroy(arc_l2c_only->arcs_list[ARC_BUFC_DATA]); | |
6548 | } | |
e10b0808 | 6549 | |
cae5b340 | 6550 | uint64_t |
8ec27e97 | 6551 | arc_target_bytes(void) |
cae5b340 | 6552 | { |
8ec27e97 | 6553 | return (arc_c); |
e10b0808 AX |
6554 | } |
6555 | ||
34dc7c2f BB |
6556 | void |
6557 | arc_init(void) | |
6558 | { | |
cae5b340 | 6559 | uint64_t percent, allmem = arc_all_memory(); |
e10b0808 AX |
6560 | |
6561 | mutex_init(&arc_reclaim_lock, NULL, MUTEX_DEFAULT, NULL); | |
6562 | cv_init(&arc_reclaim_thread_cv, NULL, CV_DEFAULT, NULL); | |
6563 | cv_init(&arc_reclaim_waiters_cv, NULL, CV_DEFAULT, NULL); | |
6564 | ||
34dc7c2f | 6565 | /* Convert seconds to clock ticks */ |
e10b0808 | 6566 | arc_min_prefetch_lifespan = 1 * hz; |
34dc7c2f | 6567 | |
34dc7c2f | 6568 | #ifdef _KERNEL |
7cb67b45 BB |
6569 | /* |
6570 | * Register a shrinker to support synchronous (direct) memory | |
6571 | * reclaim from the arc. This is done to prevent kswapd from | |
6572 | * swapping out pages when it is preferable to shrink the arc. | |
6573 | */ | |
6574 | spl_register_shrinker(&arc_shrinker); | |
34dc7c2f | 6575 | |
e10b0808 | 6576 | /* Set to 1/64 of all memory or a minimum of 512K */ |
cae5b340 | 6577 | arc_sys_free = MAX(allmem / 64, (512 * 1024)); |
e10b0808 AX |
6578 | arc_need_free = 0; |
6579 | #endif | |
34dc7c2f | 6580 | |
e10b0808 AX |
6581 | /* Set max to 1/2 of all memory */ |
6582 | arc_c_max = allmem / 2; | |
34dc7c2f | 6583 | |
cae5b340 AX |
6584 | #ifdef _KERNEL |
6585 | /* Set min cache to 1/32 of all memory, or 32MB, whichever is more */ | |
6586 | arc_c_min = MAX(allmem / 32, 2ULL << SPA_MAXBLOCKSHIFT); | |
6587 | #else | |
6588 | /* | |
6589 | * In userland, there's only the memory pressure that we artificially | |
6590 | * create (see arc_available_memory()). Don't let arc_c get too | |
6591 | * small, because it can cause transactions to be larger than | |
6592 | * arc_c, causing arc_tempreserve_space() to fail. | |
6593 | */ | |
6594 | arc_c_min = MAX(arc_c_max / 2, 2ULL << SPA_MAXBLOCKSHIFT); | |
6595 | #endif | |
6596 | ||
34dc7c2f BB |
6597 | arc_c = arc_c_max; |
6598 | arc_p = (arc_c >> 1); | |
cae5b340 | 6599 | arc_size = 0; |
34dc7c2f | 6600 | |
e10b0808 AX |
6601 | /* Set min to 1/2 of arc_c_min */ |
6602 | arc_meta_min = 1ULL << SPA_MAXBLOCKSHIFT; | |
6603 | /* Initialize maximum observed usage to zero */ | |
1834f2d8 | 6604 | arc_meta_max = 0; |
cae5b340 AX |
6605 | /* |
6606 | * Set arc_meta_limit to a percent of arc_c_max with a floor of | |
6607 | * arc_meta_min, and a ceiling of arc_c_max. | |
6608 | */ | |
6609 | percent = MIN(zfs_arc_meta_limit_percent, 100); | |
6610 | arc_meta_limit = MAX(arc_meta_min, (percent * arc_c_max) / 100); | |
6611 | percent = MIN(zfs_arc_dnode_limit_percent, 100); | |
6612 | arc_dnode_limit = (percent * arc_meta_limit) / 100; | |
34dc7c2f | 6613 | |
e10b0808 AX |
6614 | /* Apply user specified tunings */ |
6615 | arc_tuning_update(); | |
6616 | ||
34dc7c2f BB |
6617 | /* if kmem_flags are set, lets try to use less memory */ |
6618 | if (kmem_debugging()) | |
6619 | arc_c = arc_c / 2; | |
6620 | if (arc_c < arc_c_min) | |
6621 | arc_c = arc_c_min; | |
6622 | ||
cae5b340 | 6623 | arc_state_init(); |
34dc7c2f BB |
6624 | buf_init(); |
6625 | ||
ab26409d BB |
6626 | list_create(&arc_prune_list, sizeof (arc_prune_t), |
6627 | offsetof(arc_prune_t, p_node)); | |
ab26409d | 6628 | mutex_init(&arc_prune_mtx, NULL, MUTEX_DEFAULT, NULL); |
34dc7c2f | 6629 | |
e10b0808 AX |
6630 | arc_prune_taskq = taskq_create("arc_prune", max_ncpus, defclsyspri, |
6631 | max_ncpus, INT_MAX, TASKQ_PREPOPULATE | TASKQ_DYNAMIC); | |
6632 | ||
cae5b340 AX |
6633 | arc_reclaim_thread_exit = B_FALSE; |
6634 | ||
34dc7c2f BB |
6635 | arc_ksp = kstat_create("zfs", 0, "arcstats", "misc", KSTAT_TYPE_NAMED, |
6636 | sizeof (arc_stats) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL); | |
6637 | ||
6638 | if (arc_ksp != NULL) { | |
6639 | arc_ksp->ks_data = &arc_stats; | |
13be560d | 6640 | arc_ksp->ks_update = arc_kstat_update; |
34dc7c2f BB |
6641 | kstat_install(arc_ksp); |
6642 | } | |
6643 | ||
e10b0808 AX |
6644 | (void) thread_create(NULL, 0, arc_reclaim_thread, NULL, 0, &p0, |
6645 | TS_RUN, defclsyspri); | |
6646 | ||
cae5b340 | 6647 | arc_dead = B_FALSE; |
b128c09f | 6648 | arc_warm = B_FALSE; |
34dc7c2f | 6649 | |
a08ee875 LG |
6650 | /* |
6651 | * Calculate maximum amount of dirty data per pool. | |
6652 | * | |
6653 | * If it has been set by a module parameter, take that. | |
6654 | * Otherwise, use a percentage of physical memory defined by | |
6655 | * zfs_dirty_data_max_percent (default 10%) with a cap at | |
22929307 | 6656 | * zfs_dirty_data_max_max (default 4G or 25% of physical memory). |
a08ee875 LG |
6657 | */ |
6658 | if (zfs_dirty_data_max_max == 0) | |
22929307 | 6659 | zfs_dirty_data_max_max = MIN(4ULL * 1024 * 1024 * 1024, |
cae5b340 | 6660 | allmem * zfs_dirty_data_max_max_percent / 100); |
a08ee875 LG |
6661 | |
6662 | if (zfs_dirty_data_max == 0) { | |
cae5b340 | 6663 | zfs_dirty_data_max = allmem * |
a08ee875 LG |
6664 | zfs_dirty_data_max_percent / 100; |
6665 | zfs_dirty_data_max = MIN(zfs_dirty_data_max, | |
6666 | zfs_dirty_data_max_max); | |
6667 | } | |
34dc7c2f BB |
6668 | } |
6669 | ||
6670 | void | |
6671 | arc_fini(void) | |
6672 | { | |
ab26409d BB |
6673 | arc_prune_t *p; |
6674 | ||
7cb67b45 BB |
6675 | #ifdef _KERNEL |
6676 | spl_unregister_shrinker(&arc_shrinker); | |
6677 | #endif /* _KERNEL */ | |
6678 | ||
e10b0808 | 6679 | mutex_enter(&arc_reclaim_lock); |
cae5b340 | 6680 | arc_reclaim_thread_exit = B_TRUE; |
e10b0808 AX |
6681 | /* |
6682 | * The reclaim thread will set arc_reclaim_thread_exit back to | |
cae5b340 | 6683 | * B_FALSE when it is finished exiting; we're waiting for that. |
e10b0808 AX |
6684 | */ |
6685 | while (arc_reclaim_thread_exit) { | |
6686 | cv_signal(&arc_reclaim_thread_cv); | |
6687 | cv_wait(&arc_reclaim_thread_cv, &arc_reclaim_lock); | |
6688 | } | |
6689 | mutex_exit(&arc_reclaim_lock); | |
6690 | ||
cae5b340 AX |
6691 | /* Use B_TRUE to ensure *all* buffers are evicted */ |
6692 | arc_flush(NULL, B_TRUE); | |
34dc7c2f | 6693 | |
cae5b340 | 6694 | arc_dead = B_TRUE; |
34dc7c2f BB |
6695 | |
6696 | if (arc_ksp != NULL) { | |
6697 | kstat_delete(arc_ksp); | |
6698 | arc_ksp = NULL; | |
6699 | } | |
6700 | ||
e10b0808 AX |
6701 | taskq_wait(arc_prune_taskq); |
6702 | taskq_destroy(arc_prune_taskq); | |
6703 | ||
ab26409d BB |
6704 | mutex_enter(&arc_prune_mtx); |
6705 | while ((p = list_head(&arc_prune_list)) != NULL) { | |
6706 | list_remove(&arc_prune_list, p); | |
6707 | refcount_remove(&p->p_refcnt, &arc_prune_list); | |
6708 | refcount_destroy(&p->p_refcnt); | |
6709 | kmem_free(p, sizeof (*p)); | |
6710 | } | |
6711 | mutex_exit(&arc_prune_mtx); | |
6712 | ||
6713 | list_destroy(&arc_prune_list); | |
6714 | mutex_destroy(&arc_prune_mtx); | |
e10b0808 AX |
6715 | mutex_destroy(&arc_reclaim_lock); |
6716 | cv_destroy(&arc_reclaim_thread_cv); | |
6717 | cv_destroy(&arc_reclaim_waiters_cv); | |
6718 | ||
cae5b340 | 6719 | arc_state_fini(); |
34dc7c2f | 6720 | buf_fini(); |
9babb374 | 6721 | |
e10b0808 | 6722 | ASSERT0(arc_loaned_bytes); |
34dc7c2f BB |
6723 | } |
6724 | ||
6725 | /* | |
6726 | * Level 2 ARC | |
6727 | * | |
6728 | * The level 2 ARC (L2ARC) is a cache layer in-between main memory and disk. | |
6729 | * It uses dedicated storage devices to hold cached data, which are populated | |
6730 | * using large infrequent writes. The main role of this cache is to boost | |
6731 | * the performance of random read workloads. The intended L2ARC devices | |
6732 | * include short-stroked disks, solid state disks, and other media with | |
6733 | * substantially faster read latency than disk. | |
6734 | * | |
6735 | * +-----------------------+ | |
6736 | * | ARC | | |
6737 | * +-----------------------+ | |
6738 | * | ^ ^ | |
6739 | * | | | | |
6740 | * l2arc_feed_thread() arc_read() | |
6741 | * | | | | |
6742 | * | l2arc read | | |
6743 | * V | | | |
6744 | * +---------------+ | | |
6745 | * | L2ARC | | | |
6746 | * +---------------+ | | |
6747 | * | ^ | | |
6748 | * l2arc_write() | | | |
6749 | * | | | | |
6750 | * V | | | |
6751 | * +-------+ +-------+ | |
6752 | * | vdev | | vdev | | |
6753 | * | cache | | cache | | |
6754 | * +-------+ +-------+ | |
6755 | * +=========+ .-----. | |
6756 | * : L2ARC : |-_____-| | |
6757 | * : devices : | Disks | | |
6758 | * +=========+ `-_____-' | |
6759 | * | |
6760 | * Read requests are satisfied from the following sources, in order: | |
6761 | * | |
6762 | * 1) ARC | |
6763 | * 2) vdev cache of L2ARC devices | |
6764 | * 3) L2ARC devices | |
6765 | * 4) vdev cache of disks | |
6766 | * 5) disks | |
6767 | * | |
6768 | * Some L2ARC device types exhibit extremely slow write performance. | |
6769 | * To accommodate for this there are some significant differences between | |
6770 | * the L2ARC and traditional cache design: | |
6771 | * | |
6772 | * 1. There is no eviction path from the ARC to the L2ARC. Evictions from | |
6773 | * the ARC behave as usual, freeing buffers and placing headers on ghost | |
6774 | * lists. The ARC does not send buffers to the L2ARC during eviction as | |
6775 | * this would add inflated write latencies for all ARC memory pressure. | |
6776 | * | |
6777 | * 2. The L2ARC attempts to cache data from the ARC before it is evicted. | |
6778 | * It does this by periodically scanning buffers from the eviction-end of | |
6779 | * the MFU and MRU ARC lists, copying them to the L2ARC devices if they are | |
c06d4368 AX |
6780 | * not already there. It scans until a headroom of buffers is satisfied, |
6781 | * which itself is a buffer for ARC eviction. If a compressible buffer is | |
6782 | * found during scanning and selected for writing to an L2ARC device, we | |
6783 | * temporarily boost scanning headroom during the next scan cycle to make | |
6784 | * sure we adapt to compression effects (which might significantly reduce | |
6785 | * the data volume we write to L2ARC). The thread that does this is | |
34dc7c2f BB |
6786 | * l2arc_feed_thread(), illustrated below; example sizes are included to |
6787 | * provide a better sense of ratio than this diagram: | |
6788 | * | |
6789 | * head --> tail | |
6790 | * +---------------------+----------+ | |
6791 | * ARC_mfu |:::::#:::::::::::::::|o#o###o###|-->. # already on L2ARC | |
6792 | * +---------------------+----------+ | o L2ARC eligible | |
6793 | * ARC_mru |:#:::::::::::::::::::|#o#ooo####|-->| : ARC buffer | |
6794 | * +---------------------+----------+ | | |
6795 | * 15.9 Gbytes ^ 32 Mbytes | | |
6796 | * headroom | | |
6797 | * l2arc_feed_thread() | |
6798 | * | | |
6799 | * l2arc write hand <--[oooo]--' | |
6800 | * | 8 Mbyte | |
6801 | * | write max | |
6802 | * V | |
6803 | * +==============================+ | |
6804 | * L2ARC dev |####|#|###|###| |####| ... | | |
6805 | * +==============================+ | |
6806 | * 32 Gbytes | |
6807 | * | |
6808 | * 3. If an ARC buffer is copied to the L2ARC but then hit instead of | |
6809 | * evicted, then the L2ARC has cached a buffer much sooner than it probably | |
6810 | * needed to, potentially wasting L2ARC device bandwidth and storage. It is | |
6811 | * safe to say that this is an uncommon case, since buffers at the end of | |
6812 | * the ARC lists have moved there due to inactivity. | |
6813 | * | |
6814 | * 4. If the ARC evicts faster than the L2ARC can maintain a headroom, | |
6815 | * then the L2ARC simply misses copying some buffers. This serves as a | |
6816 | * pressure valve to prevent heavy read workloads from both stalling the ARC | |
6817 | * with waits and clogging the L2ARC with writes. This also helps prevent | |
6818 | * the potential for the L2ARC to churn if it attempts to cache content too | |
6819 | * quickly, such as during backups of the entire pool. | |
6820 | * | |
b128c09f BB |
6821 | * 5. After system boot and before the ARC has filled main memory, there are |
6822 | * no evictions from the ARC and so the tails of the ARC_mfu and ARC_mru | |
6823 | * lists can remain mostly static. Instead of searching from tail of these | |
6824 | * lists as pictured, the l2arc_feed_thread() will search from the list heads | |
6825 | * for eligible buffers, greatly increasing its chance of finding them. | |
6826 | * | |
6827 | * The L2ARC device write speed is also boosted during this time so that | |
6828 | * the L2ARC warms up faster. Since there have been no ARC evictions yet, | |
6829 | * there are no L2ARC reads, and no fear of degrading read performance | |
6830 | * through increased writes. | |
6831 | * | |
6832 | * 6. Writes to the L2ARC devices are grouped and sent in-sequence, so that | |
34dc7c2f BB |
6833 | * the vdev queue can aggregate them into larger and fewer writes. Each |
6834 | * device is written to in a rotor fashion, sweeping writes through | |
6835 | * available space then repeating. | |
6836 | * | |
b128c09f | 6837 | * 7. The L2ARC does not store dirty content. It never needs to flush |
34dc7c2f BB |
6838 | * write buffers back to disk based storage. |
6839 | * | |
b128c09f | 6840 | * 8. If an ARC buffer is written (and dirtied) which also exists in the |
34dc7c2f BB |
6841 | * L2ARC, the now stale L2ARC buffer is immediately dropped. |
6842 | * | |
6843 | * The performance of the L2ARC can be tweaked by a number of tunables, which | |
6844 | * may be necessary for different workloads: | |
6845 | * | |
6846 | * l2arc_write_max max write bytes per interval | |
b128c09f | 6847 | * l2arc_write_boost extra write bytes during device warmup |
34dc7c2f BB |
6848 | * l2arc_noprefetch skip caching prefetched buffers |
6849 | * l2arc_headroom number of max device writes to precache | |
c06d4368 AX |
6850 | * l2arc_headroom_boost when we find compressed buffers during ARC |
6851 | * scanning, we multiply headroom by this | |
6852 | * percentage factor for the next scan cycle, | |
6853 | * since more compressed buffers are likely to | |
6854 | * be present | |
34dc7c2f BB |
6855 | * l2arc_feed_secs seconds between L2ARC writing |
6856 | * | |
6857 | * Tunables may be removed or added as future performance improvements are | |
6858 | * integrated, and also may become zpool properties. | |
d164b209 BB |
6859 | * |
6860 | * There are three key functions that control how the L2ARC warms up: | |
6861 | * | |
6862 | * l2arc_write_eligible() check if a buffer is eligible to cache | |
6863 | * l2arc_write_size() calculate how much to write | |
6864 | * l2arc_write_interval() calculate sleep delay between writes | |
6865 | * | |
6866 | * These three functions determine what to write, how much, and how quickly | |
6867 | * to send writes. | |
34dc7c2f BB |
6868 | */ |
6869 | ||
d164b209 | 6870 | static boolean_t |
e10b0808 | 6871 | l2arc_write_eligible(uint64_t spa_guid, arc_buf_hdr_t *hdr) |
d164b209 BB |
6872 | { |
6873 | /* | |
6874 | * A buffer is *not* eligible for the L2ARC if it: | |
6875 | * 1. belongs to a different spa. | |
428870ff BB |
6876 | * 2. is already cached on the L2ARC. |
6877 | * 3. has an I/O in progress (it may be an incomplete read). | |
6878 | * 4. is flagged not eligible (zfs property). | |
d164b209 | 6879 | */ |
e10b0808 AX |
6880 | if (hdr->b_spa != spa_guid || HDR_HAS_L2HDR(hdr) || |
6881 | HDR_IO_IN_PROGRESS(hdr) || !HDR_L2CACHE(hdr)) | |
d164b209 BB |
6882 | return (B_FALSE); |
6883 | ||
6884 | return (B_TRUE); | |
6885 | } | |
6886 | ||
6887 | static uint64_t | |
c06d4368 | 6888 | l2arc_write_size(void) |
d164b209 BB |
6889 | { |
6890 | uint64_t size; | |
6891 | ||
c06d4368 AX |
6892 | /* |
6893 | * Make sure our globals have meaningful values in case the user | |
6894 | * altered them. | |
6895 | */ | |
6896 | size = l2arc_write_max; | |
6897 | if (size == 0) { | |
6898 | cmn_err(CE_NOTE, "Bad value for l2arc_write_max, value must " | |
6899 | "be greater than zero, resetting it to the default (%d)", | |
6900 | L2ARC_WRITE_SIZE); | |
6901 | size = l2arc_write_max = L2ARC_WRITE_SIZE; | |
6902 | } | |
d164b209 BB |
6903 | |
6904 | if (arc_warm == B_FALSE) | |
c06d4368 | 6905 | size += l2arc_write_boost; |
d164b209 BB |
6906 | |
6907 | return (size); | |
6908 | ||
6909 | } | |
6910 | ||
6911 | static clock_t | |
6912 | l2arc_write_interval(clock_t began, uint64_t wanted, uint64_t wrote) | |
6913 | { | |
428870ff | 6914 | clock_t interval, next, now; |
d164b209 BB |
6915 | |
6916 | /* | |
6917 | * If the ARC lists are busy, increase our write rate; if the | |
6918 | * lists are stale, idle back. This is achieved by checking | |
6919 | * how much we previously wrote - if it was more than half of | |
6920 | * what we wanted, schedule the next write much sooner. | |
6921 | */ | |
6922 | if (l2arc_feed_again && wrote > (wanted / 2)) | |
6923 | interval = (hz * l2arc_feed_min_ms) / 1000; | |
6924 | else | |
6925 | interval = hz * l2arc_feed_secs; | |
6926 | ||
428870ff BB |
6927 | now = ddi_get_lbolt(); |
6928 | next = MAX(now, MIN(now + interval, began + interval)); | |
d164b209 BB |
6929 | |
6930 | return (next); | |
6931 | } | |
6932 | ||
34dc7c2f BB |
6933 | /* |
6934 | * Cycle through L2ARC devices. This is how L2ARC load balances. | |
b128c09f | 6935 | * If a device is returned, this also returns holding the spa config lock. |
34dc7c2f BB |
6936 | */ |
6937 | static l2arc_dev_t * | |
6938 | l2arc_dev_get_next(void) | |
6939 | { | |
b128c09f | 6940 | l2arc_dev_t *first, *next = NULL; |
34dc7c2f | 6941 | |
b128c09f BB |
6942 | /* |
6943 | * Lock out the removal of spas (spa_namespace_lock), then removal | |
6944 | * of cache devices (l2arc_dev_mtx). Once a device has been selected, | |
6945 | * both locks will be dropped and a spa config lock held instead. | |
6946 | */ | |
6947 | mutex_enter(&spa_namespace_lock); | |
6948 | mutex_enter(&l2arc_dev_mtx); | |
6949 | ||
6950 | /* if there are no vdevs, there is nothing to do */ | |
6951 | if (l2arc_ndev == 0) | |
6952 | goto out; | |
6953 | ||
6954 | first = NULL; | |
6955 | next = l2arc_dev_last; | |
6956 | do { | |
6957 | /* loop around the list looking for a non-faulted vdev */ | |
6958 | if (next == NULL) { | |
34dc7c2f | 6959 | next = list_head(l2arc_dev_list); |
b128c09f BB |
6960 | } else { |
6961 | next = list_next(l2arc_dev_list, next); | |
6962 | if (next == NULL) | |
6963 | next = list_head(l2arc_dev_list); | |
6964 | } | |
6965 | ||
6966 | /* if we have come back to the start, bail out */ | |
6967 | if (first == NULL) | |
6968 | first = next; | |
6969 | else if (next == first) | |
6970 | break; | |
6971 | ||
6972 | } while (vdev_is_dead(next->l2ad_vdev)); | |
6973 | ||
6974 | /* if we were unable to find any usable vdevs, return NULL */ | |
6975 | if (vdev_is_dead(next->l2ad_vdev)) | |
6976 | next = NULL; | |
34dc7c2f BB |
6977 | |
6978 | l2arc_dev_last = next; | |
6979 | ||
b128c09f BB |
6980 | out: |
6981 | mutex_exit(&l2arc_dev_mtx); | |
6982 | ||
6983 | /* | |
6984 | * Grab the config lock to prevent the 'next' device from being | |
6985 | * removed while we are writing to it. | |
6986 | */ | |
6987 | if (next != NULL) | |
6988 | spa_config_enter(next->l2ad_spa, SCL_L2ARC, next, RW_READER); | |
6989 | mutex_exit(&spa_namespace_lock); | |
6990 | ||
34dc7c2f BB |
6991 | return (next); |
6992 | } | |
6993 | ||
b128c09f BB |
6994 | /* |
6995 | * Free buffers that were tagged for destruction. | |
6996 | */ | |
6997 | static void | |
0bc8fd78 | 6998 | l2arc_do_free_on_write(void) |
b128c09f BB |
6999 | { |
7000 | list_t *buflist; | |
7001 | l2arc_data_free_t *df, *df_prev; | |
7002 | ||
7003 | mutex_enter(&l2arc_free_on_write_mtx); | |
7004 | buflist = l2arc_free_on_write; | |
7005 | ||
7006 | for (df = list_tail(buflist); df; df = df_prev) { | |
7007 | df_prev = list_prev(buflist, df); | |
cae5b340 AX |
7008 | ASSERT3P(df->l2df_abd, !=, NULL); |
7009 | abd_free(df->l2df_abd); | |
b128c09f BB |
7010 | list_remove(buflist, df); |
7011 | kmem_free(df, sizeof (l2arc_data_free_t)); | |
7012 | } | |
7013 | ||
7014 | mutex_exit(&l2arc_free_on_write_mtx); | |
7015 | } | |
7016 | ||
34dc7c2f BB |
7017 | /* |
7018 | * A write to a cache device has completed. Update all headers to allow | |
7019 | * reads from these buffers to begin. | |
7020 | */ | |
7021 | static void | |
7022 | l2arc_write_done(zio_t *zio) | |
7023 | { | |
7024 | l2arc_write_callback_t *cb; | |
7025 | l2arc_dev_t *dev; | |
7026 | list_t *buflist; | |
e10b0808 | 7027 | arc_buf_hdr_t *head, *hdr, *hdr_prev; |
34dc7c2f | 7028 | kmutex_t *hash_lock; |
ea04106b | 7029 | int64_t bytes_dropped = 0; |
34dc7c2f BB |
7030 | |
7031 | cb = zio->io_private; | |
cae5b340 | 7032 | ASSERT3P(cb, !=, NULL); |
34dc7c2f | 7033 | dev = cb->l2wcb_dev; |
cae5b340 | 7034 | ASSERT3P(dev, !=, NULL); |
34dc7c2f | 7035 | head = cb->l2wcb_head; |
cae5b340 | 7036 | ASSERT3P(head, !=, NULL); |
e10b0808 | 7037 | buflist = &dev->l2ad_buflist; |
cae5b340 | 7038 | ASSERT3P(buflist, !=, NULL); |
34dc7c2f BB |
7039 | DTRACE_PROBE2(l2arc__iodone, zio_t *, zio, |
7040 | l2arc_write_callback_t *, cb); | |
7041 | ||
7042 | if (zio->io_error != 0) | |
7043 | ARCSTAT_BUMP(arcstat_l2_writes_error); | |
7044 | ||
34dc7c2f BB |
7045 | /* |
7046 | * All writes completed, or an error was hit. | |
7047 | */ | |
e10b0808 AX |
7048 | top: |
7049 | mutex_enter(&dev->l2ad_mtx); | |
7050 | for (hdr = list_prev(buflist, head); hdr; hdr = hdr_prev) { | |
7051 | hdr_prev = list_prev(buflist, hdr); | |
7052 | ||
7053 | hash_lock = HDR_LOCK(hdr); | |
a08ee875 LG |
7054 | |
7055 | /* | |
e10b0808 AX |
7056 | * We cannot use mutex_enter or else we can deadlock |
7057 | * with l2arc_write_buffers (due to swapping the order | |
7058 | * the hash lock and l2ad_mtx are taken). | |
a08ee875 | 7059 | */ |
34dc7c2f BB |
7060 | if (!mutex_tryenter(hash_lock)) { |
7061 | /* | |
e10b0808 AX |
7062 | * Missed the hash lock. We must retry so we |
7063 | * don't leave the ARC_FLAG_L2_WRITING bit set. | |
34dc7c2f | 7064 | */ |
e10b0808 AX |
7065 | ARCSTAT_BUMP(arcstat_l2_writes_lock_retry); |
7066 | ||
7067 | /* | |
7068 | * We don't want to rescan the headers we've | |
7069 | * already marked as having been written out, so | |
7070 | * we reinsert the head node so we can pick up | |
7071 | * where we left off. | |
7072 | */ | |
7073 | list_remove(buflist, head); | |
7074 | list_insert_after(buflist, hdr, head); | |
7075 | ||
7076 | mutex_exit(&dev->l2ad_mtx); | |
7077 | ||
7078 | /* | |
7079 | * We wait for the hash lock to become available | |
7080 | * to try and prevent busy waiting, and increase | |
7081 | * the chance we'll be able to acquire the lock | |
7082 | * the next time around. | |
7083 | */ | |
7084 | mutex_enter(hash_lock); | |
7085 | mutex_exit(hash_lock); | |
7086 | goto top; | |
34dc7c2f BB |
7087 | } |
7088 | ||
e10b0808 AX |
7089 | /* |
7090 | * We could not have been moved into the arc_l2c_only | |
7091 | * state while in-flight due to our ARC_FLAG_L2_WRITING | |
7092 | * bit being set. Let's just ensure that's being enforced. | |
7093 | */ | |
7094 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
7095 | ||
7096 | /* | |
cae5b340 AX |
7097 | * Skipped - drop L2ARC entry and mark the header as no |
7098 | * longer L2 eligibile. | |
e10b0808 | 7099 | */ |
34dc7c2f BB |
7100 | if (zio->io_error != 0) { |
7101 | /* | |
b128c09f | 7102 | * Error - drop L2ARC entry. |
34dc7c2f | 7103 | */ |
e10b0808 | 7104 | list_remove(buflist, hdr); |
cae5b340 | 7105 | arc_hdr_clear_flags(hdr, ARC_FLAG_HAS_L2HDR); |
e10b0808 | 7106 | |
cae5b340 AX |
7107 | ARCSTAT_INCR(arcstat_l2_psize, -arc_hdr_size(hdr)); |
7108 | ARCSTAT_INCR(arcstat_l2_lsize, -HDR_GET_LSIZE(hdr)); | |
e10b0808 | 7109 | |
cae5b340 | 7110 | bytes_dropped += arc_hdr_size(hdr); |
e10b0808 | 7111 | (void) refcount_remove_many(&dev->l2ad_alloc, |
cae5b340 | 7112 | arc_hdr_size(hdr), hdr); |
34dc7c2f BB |
7113 | } |
7114 | ||
7115 | /* | |
e10b0808 AX |
7116 | * Allow ARC to begin reads and ghost list evictions to |
7117 | * this L2ARC entry. | |
34dc7c2f | 7118 | */ |
cae5b340 | 7119 | arc_hdr_clear_flags(hdr, ARC_FLAG_L2_WRITING); |
34dc7c2f BB |
7120 | |
7121 | mutex_exit(hash_lock); | |
7122 | } | |
7123 | ||
7124 | atomic_inc_64(&l2arc_writes_done); | |
7125 | list_remove(buflist, head); | |
e10b0808 AX |
7126 | ASSERT(!HDR_HAS_L1HDR(head)); |
7127 | kmem_cache_free(hdr_l2only_cache, head); | |
7128 | mutex_exit(&dev->l2ad_mtx); | |
34dc7c2f | 7129 | |
ea04106b AX |
7130 | vdev_space_update(dev->l2ad_vdev, -bytes_dropped, 0, 0); |
7131 | ||
b128c09f | 7132 | l2arc_do_free_on_write(); |
34dc7c2f BB |
7133 | |
7134 | kmem_free(cb, sizeof (l2arc_write_callback_t)); | |
7135 | } | |
7136 | ||
7137 | /* | |
7138 | * A read to a cache device completed. Validate buffer contents before | |
7139 | * handing over to the regular ARC routines. | |
7140 | */ | |
7141 | static void | |
7142 | l2arc_read_done(zio_t *zio) | |
7143 | { | |
7144 | l2arc_read_callback_t *cb; | |
7145 | arc_buf_hdr_t *hdr; | |
34dc7c2f | 7146 | kmutex_t *hash_lock; |
cae5b340 | 7147 | boolean_t valid_cksum; |
b128c09f | 7148 | |
cae5b340 | 7149 | ASSERT3P(zio->io_vd, !=, NULL); |
b128c09f BB |
7150 | ASSERT(zio->io_flags & ZIO_FLAG_DONT_PROPAGATE); |
7151 | ||
7152 | spa_config_exit(zio->io_spa, SCL_L2ARC, zio->io_vd); | |
34dc7c2f BB |
7153 | |
7154 | cb = zio->io_private; | |
cae5b340 AX |
7155 | ASSERT3P(cb, !=, NULL); |
7156 | hdr = cb->l2rcb_hdr; | |
7157 | ASSERT3P(hdr, !=, NULL); | |
34dc7c2f | 7158 | |
cae5b340 | 7159 | hash_lock = HDR_LOCK(hdr); |
34dc7c2f | 7160 | mutex_enter(hash_lock); |
428870ff | 7161 | ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); |
34dc7c2f | 7162 | |
c06d4368 | 7163 | /* |
cae5b340 AX |
7164 | * If the data was read into a temporary buffer, |
7165 | * move it and free the buffer. | |
c06d4368 | 7166 | */ |
cae5b340 AX |
7167 | if (cb->l2rcb_abd != NULL) { |
7168 | ASSERT3U(arc_hdr_size(hdr), <, zio->io_size); | |
7169 | if (zio->io_error == 0) { | |
7170 | abd_copy(hdr->b_l1hdr.b_pabd, cb->l2rcb_abd, | |
7171 | arc_hdr_size(hdr)); | |
7172 | } | |
7173 | ||
7174 | /* | |
7175 | * The following must be done regardless of whether | |
7176 | * there was an error: | |
7177 | * - free the temporary buffer | |
7178 | * - point zio to the real ARC buffer | |
7179 | * - set zio size accordingly | |
7180 | * These are required because zio is either re-used for | |
7181 | * an I/O of the block in the case of the error | |
7182 | * or the zio is passed to arc_read_done() and it | |
7183 | * needs real data. | |
7184 | */ | |
7185 | abd_free(cb->l2rcb_abd); | |
7186 | zio->io_size = zio->io_orig_size = arc_hdr_size(hdr); | |
7187 | zio->io_abd = zio->io_orig_abd = hdr->b_l1hdr.b_pabd; | |
7188 | } | |
7189 | ||
7190 | ASSERT3P(zio->io_abd, !=, NULL); | |
c06d4368 | 7191 | |
34dc7c2f BB |
7192 | /* |
7193 | * Check this survived the L2ARC journey. | |
7194 | */ | |
cae5b340 AX |
7195 | ASSERT3P(zio->io_abd, ==, hdr->b_l1hdr.b_pabd); |
7196 | zio->io_bp_copy = cb->l2rcb_bp; /* XXX fix in L2ARC 2.0 */ | |
7197 | zio->io_bp = &zio->io_bp_copy; /* XXX fix in L2ARC 2.0 */ | |
7198 | ||
7199 | valid_cksum = arc_cksum_is_equal(hdr, zio); | |
7200 | if (valid_cksum && zio->io_error == 0 && !HDR_L2_EVICTED(hdr)) { | |
34dc7c2f | 7201 | mutex_exit(hash_lock); |
cae5b340 | 7202 | zio->io_private = hdr; |
34dc7c2f BB |
7203 | arc_read_done(zio); |
7204 | } else { | |
7205 | mutex_exit(hash_lock); | |
7206 | /* | |
7207 | * Buffer didn't survive caching. Increment stats and | |
7208 | * reissue to the original storage device. | |
7209 | */ | |
b128c09f | 7210 | if (zio->io_error != 0) { |
34dc7c2f | 7211 | ARCSTAT_BUMP(arcstat_l2_io_error); |
b128c09f | 7212 | } else { |
a08ee875 | 7213 | zio->io_error = SET_ERROR(EIO); |
b128c09f | 7214 | } |
cae5b340 | 7215 | if (!valid_cksum) |
34dc7c2f BB |
7216 | ARCSTAT_BUMP(arcstat_l2_cksum_bad); |
7217 | ||
34dc7c2f | 7218 | /* |
b128c09f BB |
7219 | * If there's no waiter, issue an async i/o to the primary |
7220 | * storage now. If there *is* a waiter, the caller must | |
7221 | * issue the i/o in a context where it's OK to block. | |
34dc7c2f | 7222 | */ |
d164b209 BB |
7223 | if (zio->io_waiter == NULL) { |
7224 | zio_t *pio = zio_unique_parent(zio); | |
7225 | ||
7226 | ASSERT(!pio || pio->io_child_type == ZIO_CHILD_LOGICAL); | |
7227 | ||
cae5b340 AX |
7228 | zio_nowait(zio_read(pio, zio->io_spa, zio->io_bp, |
7229 | hdr->b_l1hdr.b_pabd, zio->io_size, arc_read_done, | |
7230 | hdr, zio->io_priority, cb->l2rcb_flags, | |
7231 | &cb->l2rcb_zb)); | |
d164b209 | 7232 | } |
34dc7c2f BB |
7233 | } |
7234 | ||
7235 | kmem_free(cb, sizeof (l2arc_read_callback_t)); | |
7236 | } | |
7237 | ||
7238 | /* | |
7239 | * This is the list priority from which the L2ARC will search for pages to | |
7240 | * cache. This is used within loops (0..3) to cycle through lists in the | |
7241 | * desired order. This order can have a significant effect on cache | |
7242 | * performance. | |
7243 | * | |
7244 | * Currently the metadata lists are hit first, MFU then MRU, followed by | |
7245 | * the data lists. This function returns a locked list, and also returns | |
7246 | * the lock pointer. | |
7247 | */ | |
e10b0808 AX |
7248 | static multilist_sublist_t * |
7249 | l2arc_sublist_lock(int list_num) | |
34dc7c2f | 7250 | { |
e10b0808 AX |
7251 | multilist_t *ml = NULL; |
7252 | unsigned int idx; | |
34dc7c2f | 7253 | |
cae5b340 | 7254 | ASSERT(list_num >= 0 && list_num < L2ARC_FEED_TYPES); |
34dc7c2f BB |
7255 | |
7256 | switch (list_num) { | |
7257 | case 0: | |
cae5b340 | 7258 | ml = arc_mfu->arcs_list[ARC_BUFC_METADATA]; |
34dc7c2f BB |
7259 | break; |
7260 | case 1: | |
cae5b340 | 7261 | ml = arc_mru->arcs_list[ARC_BUFC_METADATA]; |
34dc7c2f BB |
7262 | break; |
7263 | case 2: | |
cae5b340 | 7264 | ml = arc_mfu->arcs_list[ARC_BUFC_DATA]; |
34dc7c2f BB |
7265 | break; |
7266 | case 3: | |
cae5b340 | 7267 | ml = arc_mru->arcs_list[ARC_BUFC_DATA]; |
34dc7c2f | 7268 | break; |
cae5b340 AX |
7269 | default: |
7270 | return (NULL); | |
34dc7c2f BB |
7271 | } |
7272 | ||
e10b0808 AX |
7273 | /* |
7274 | * Return a randomly-selected sublist. This is acceptable | |
7275 | * because the caller feeds only a little bit of data for each | |
7276 | * call (8MB). Subsequent calls will result in different | |
7277 | * sublists being selected. | |
7278 | */ | |
7279 | idx = multilist_get_random_index(ml); | |
7280 | return (multilist_sublist_lock(ml, idx)); | |
34dc7c2f BB |
7281 | } |
7282 | ||
7283 | /* | |
7284 | * Evict buffers from the device write hand to the distance specified in | |
7285 | * bytes. This distance may span populated buffers, it may span nothing. | |
7286 | * This is clearing a region on the L2ARC device ready for writing. | |
7287 | * If the 'all' boolean is set, every buffer is evicted. | |
7288 | */ | |
7289 | static void | |
7290 | l2arc_evict(l2arc_dev_t *dev, uint64_t distance, boolean_t all) | |
7291 | { | |
7292 | list_t *buflist; | |
e10b0808 | 7293 | arc_buf_hdr_t *hdr, *hdr_prev; |
34dc7c2f BB |
7294 | kmutex_t *hash_lock; |
7295 | uint64_t taddr; | |
7296 | ||
e10b0808 | 7297 | buflist = &dev->l2ad_buflist; |
34dc7c2f BB |
7298 | |
7299 | if (!all && dev->l2ad_first) { | |
7300 | /* | |
7301 | * This is the first sweep through the device. There is | |
7302 | * nothing to evict. | |
7303 | */ | |
7304 | return; | |
7305 | } | |
7306 | ||
b128c09f | 7307 | if (dev->l2ad_hand >= (dev->l2ad_end - (2 * distance))) { |
34dc7c2f BB |
7308 | /* |
7309 | * When nearing the end of the device, evict to the end | |
7310 | * before the device write hand jumps to the start. | |
7311 | */ | |
7312 | taddr = dev->l2ad_end; | |
7313 | } else { | |
7314 | taddr = dev->l2ad_hand + distance; | |
7315 | } | |
7316 | DTRACE_PROBE4(l2arc__evict, l2arc_dev_t *, dev, list_t *, buflist, | |
7317 | uint64_t, taddr, boolean_t, all); | |
7318 | ||
7319 | top: | |
e10b0808 AX |
7320 | mutex_enter(&dev->l2ad_mtx); |
7321 | for (hdr = list_tail(buflist); hdr; hdr = hdr_prev) { | |
7322 | hdr_prev = list_prev(buflist, hdr); | |
34dc7c2f | 7323 | |
e10b0808 AX |
7324 | hash_lock = HDR_LOCK(hdr); |
7325 | ||
7326 | /* | |
7327 | * We cannot use mutex_enter or else we can deadlock | |
7328 | * with l2arc_write_buffers (due to swapping the order | |
7329 | * the hash lock and l2ad_mtx are taken). | |
7330 | */ | |
34dc7c2f BB |
7331 | if (!mutex_tryenter(hash_lock)) { |
7332 | /* | |
7333 | * Missed the hash lock. Retry. | |
7334 | */ | |
7335 | ARCSTAT_BUMP(arcstat_l2_evict_lock_retry); | |
e10b0808 | 7336 | mutex_exit(&dev->l2ad_mtx); |
34dc7c2f BB |
7337 | mutex_enter(hash_lock); |
7338 | mutex_exit(hash_lock); | |
7339 | goto top; | |
7340 | } | |
7341 | ||
e10b0808 | 7342 | if (HDR_L2_WRITE_HEAD(hdr)) { |
34dc7c2f BB |
7343 | /* |
7344 | * We hit a write head node. Leave it for | |
7345 | * l2arc_write_done(). | |
7346 | */ | |
e10b0808 | 7347 | list_remove(buflist, hdr); |
34dc7c2f BB |
7348 | mutex_exit(hash_lock); |
7349 | continue; | |
7350 | } | |
7351 | ||
e10b0808 AX |
7352 | if (!all && HDR_HAS_L2HDR(hdr) && |
7353 | (hdr->b_l2hdr.b_daddr > taddr || | |
7354 | hdr->b_l2hdr.b_daddr < dev->l2ad_hand)) { | |
34dc7c2f BB |
7355 | /* |
7356 | * We've evicted to the target address, | |
7357 | * or the end of the device. | |
7358 | */ | |
7359 | mutex_exit(hash_lock); | |
7360 | break; | |
7361 | } | |
7362 | ||
e10b0808 AX |
7363 | ASSERT(HDR_HAS_L2HDR(hdr)); |
7364 | if (!HDR_HAS_L1HDR(hdr)) { | |
7365 | ASSERT(!HDR_L2_READING(hdr)); | |
34dc7c2f BB |
7366 | /* |
7367 | * This doesn't exist in the ARC. Destroy. | |
7368 | * arc_hdr_destroy() will call list_remove() | |
cae5b340 | 7369 | * and decrement arcstat_l2_lsize. |
34dc7c2f | 7370 | */ |
e10b0808 AX |
7371 | arc_change_state(arc_anon, hdr, hash_lock); |
7372 | arc_hdr_destroy(hdr); | |
34dc7c2f | 7373 | } else { |
e10b0808 AX |
7374 | ASSERT(hdr->b_l1hdr.b_state != arc_l2c_only); |
7375 | ARCSTAT_BUMP(arcstat_l2_evict_l1cached); | |
b128c09f BB |
7376 | /* |
7377 | * Invalidate issued or about to be issued | |
7378 | * reads, since we may be about to write | |
7379 | * over this location. | |
7380 | */ | |
e10b0808 | 7381 | if (HDR_L2_READING(hdr)) { |
b128c09f | 7382 | ARCSTAT_BUMP(arcstat_l2_evict_reading); |
cae5b340 | 7383 | arc_hdr_set_flags(hdr, ARC_FLAG_L2_EVICTED); |
b128c09f BB |
7384 | } |
7385 | ||
e10b0808 AX |
7386 | /* Ensure this header has finished being written */ |
7387 | ASSERT(!HDR_L2_WRITING(hdr)); | |
34dc7c2f | 7388 | |
e10b0808 | 7389 | arc_hdr_l2hdr_destroy(hdr); |
34dc7c2f BB |
7390 | } |
7391 | mutex_exit(hash_lock); | |
7392 | } | |
e10b0808 | 7393 | mutex_exit(&dev->l2ad_mtx); |
34dc7c2f BB |
7394 | } |
7395 | ||
7396 | /* | |
7397 | * Find and write ARC buffers to the L2ARC device. | |
7398 | * | |
e10b0808 | 7399 | * An ARC_FLAG_L2_WRITING flag is set so that the L2ARC buffers are not valid |
34dc7c2f | 7400 | * for reading until they have completed writing. |
c06d4368 AX |
7401 | * The headroom_boost is an in-out parameter used to maintain headroom boost |
7402 | * state between calls to this function. | |
7403 | * | |
7404 | * Returns the number of bytes actually written (which may be smaller than | |
7405 | * the delta by which the device hand has changed due to alignment). | |
34dc7c2f | 7406 | */ |
d164b209 | 7407 | static uint64_t |
cae5b340 | 7408 | l2arc_write_buffers(spa_t *spa, l2arc_dev_t *dev, uint64_t target_sz) |
34dc7c2f | 7409 | { |
e10b0808 | 7410 | arc_buf_hdr_t *hdr, *hdr_prev, *head; |
cae5b340 | 7411 | uint64_t write_asize, write_psize, write_lsize, headroom; |
c06d4368 | 7412 | boolean_t full; |
34dc7c2f BB |
7413 | l2arc_write_callback_t *cb; |
7414 | zio_t *pio, *wzio; | |
3541dc6d | 7415 | uint64_t guid = spa_load_guid(spa); |
d6320ddb | 7416 | int try; |
34dc7c2f | 7417 | |
cae5b340 | 7418 | ASSERT3P(dev->l2ad_vdev, !=, NULL); |
c06d4368 | 7419 | |
34dc7c2f | 7420 | pio = NULL; |
cae5b340 | 7421 | write_lsize = write_asize = write_psize = 0; |
34dc7c2f | 7422 | full = B_FALSE; |
e10b0808 | 7423 | head = kmem_cache_alloc(hdr_l2only_cache, KM_PUSHPAGE); |
cae5b340 | 7424 | arc_hdr_set_flags(head, ARC_FLAG_L2_WRITE_HEAD | ARC_FLAG_HAS_L2HDR); |
c06d4368 | 7425 | |
34dc7c2f BB |
7426 | /* |
7427 | * Copy buffers for L2ARC writing. | |
7428 | */ | |
cae5b340 | 7429 | for (try = 0; try < L2ARC_FEED_TYPES; try++) { |
e10b0808 | 7430 | multilist_sublist_t *mls = l2arc_sublist_lock(try); |
c06d4368 AX |
7431 | uint64_t passed_sz = 0; |
7432 | ||
cae5b340 AX |
7433 | VERIFY3P(mls, !=, NULL); |
7434 | ||
b128c09f BB |
7435 | /* |
7436 | * L2ARC fast warmup. | |
7437 | * | |
7438 | * Until the ARC is warm and starts to evict, read from the | |
7439 | * head of the ARC lists rather than the tail. | |
7440 | */ | |
b128c09f | 7441 | if (arc_warm == B_FALSE) |
e10b0808 | 7442 | hdr = multilist_sublist_head(mls); |
b128c09f | 7443 | else |
e10b0808 | 7444 | hdr = multilist_sublist_tail(mls); |
b128c09f | 7445 | |
c06d4368 | 7446 | headroom = target_sz * l2arc_headroom; |
cae5b340 | 7447 | if (zfs_compressed_arc_enabled) |
c06d4368 AX |
7448 | headroom = (headroom * l2arc_headroom_boost) / 100; |
7449 | ||
e10b0808 | 7450 | for (; hdr; hdr = hdr_prev) { |
c06d4368 | 7451 | kmutex_t *hash_lock; |
c06d4368 | 7452 | |
b128c09f | 7453 | if (arc_warm == B_FALSE) |
e10b0808 | 7454 | hdr_prev = multilist_sublist_next(mls, hdr); |
b128c09f | 7455 | else |
e10b0808 | 7456 | hdr_prev = multilist_sublist_prev(mls, hdr); |
34dc7c2f | 7457 | |
e10b0808 | 7458 | hash_lock = HDR_LOCK(hdr); |
c06d4368 | 7459 | if (!mutex_tryenter(hash_lock)) { |
34dc7c2f BB |
7460 | /* |
7461 | * Skip this buffer rather than waiting. | |
7462 | */ | |
7463 | continue; | |
7464 | } | |
7465 | ||
cae5b340 | 7466 | passed_sz += HDR_GET_LSIZE(hdr); |
34dc7c2f BB |
7467 | if (passed_sz > headroom) { |
7468 | /* | |
7469 | * Searched too far. | |
7470 | */ | |
7471 | mutex_exit(hash_lock); | |
7472 | break; | |
7473 | } | |
7474 | ||
e10b0808 | 7475 | if (!l2arc_write_eligible(guid, hdr)) { |
34dc7c2f BB |
7476 | mutex_exit(hash_lock); |
7477 | continue; | |
7478 | } | |
7479 | ||
e10b0808 | 7480 | /* |
cae5b340 AX |
7481 | * We rely on the L1 portion of the header below, so |
7482 | * it's invalid for this header to have been evicted out | |
7483 | * of the ghost cache, prior to being written out. The | |
7484 | * ARC_FLAG_L2_WRITING bit ensures this won't happen. | |
e10b0808 | 7485 | */ |
cae5b340 AX |
7486 | ASSERT(HDR_HAS_L1HDR(hdr)); |
7487 | ||
7488 | ASSERT3U(HDR_GET_PSIZE(hdr), >, 0); | |
7489 | ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); | |
7490 | ASSERT3U(arc_hdr_size(hdr), >, 0); | |
7491 | uint64_t psize = arc_hdr_size(hdr); | |
7492 | uint64_t asize = vdev_psize_to_asize(dev->l2ad_vdev, | |
7493 | psize); | |
e10b0808 | 7494 | |
cae5b340 | 7495 | if ((write_asize + asize) > target_sz) { |
34dc7c2f BB |
7496 | full = B_TRUE; |
7497 | mutex_exit(hash_lock); | |
7498 | break; | |
7499 | } | |
7500 | ||
34dc7c2f BB |
7501 | if (pio == NULL) { |
7502 | /* | |
7503 | * Insert a dummy header on the buflist so | |
7504 | * l2arc_write_done() can find where the | |
7505 | * write buffers begin without searching. | |
7506 | */ | |
e10b0808 AX |
7507 | mutex_enter(&dev->l2ad_mtx); |
7508 | list_insert_head(&dev->l2ad_buflist, head); | |
7509 | mutex_exit(&dev->l2ad_mtx); | |
34dc7c2f | 7510 | |
e10b0808 AX |
7511 | cb = kmem_alloc( |
7512 | sizeof (l2arc_write_callback_t), KM_SLEEP); | |
34dc7c2f BB |
7513 | cb->l2wcb_dev = dev; |
7514 | cb->l2wcb_head = head; | |
7515 | pio = zio_root(spa, l2arc_write_done, cb, | |
7516 | ZIO_FLAG_CANFAIL); | |
7517 | } | |
7518 | ||
e10b0808 | 7519 | hdr->b_l2hdr.b_dev = dev; |
e10b0808 | 7520 | hdr->b_l2hdr.b_hits = 0; |
c06d4368 | 7521 | |
cae5b340 AX |
7522 | hdr->b_l2hdr.b_daddr = dev->l2ad_hand; |
7523 | arc_hdr_set_flags(hdr, | |
7524 | ARC_FLAG_L2_WRITING | ARC_FLAG_HAS_L2HDR); | |
c06d4368 | 7525 | |
e10b0808 AX |
7526 | mutex_enter(&dev->l2ad_mtx); |
7527 | list_insert_head(&dev->l2ad_buflist, hdr); | |
7528 | mutex_exit(&dev->l2ad_mtx); | |
34dc7c2f | 7529 | |
cae5b340 AX |
7530 | (void) refcount_add_many(&dev->l2ad_alloc, psize, hdr); |
7531 | ||
34dc7c2f | 7532 | /* |
cae5b340 AX |
7533 | * Normally the L2ARC can use the hdr's data, but if |
7534 | * we're sharing data between the hdr and one of its | |
7535 | * bufs, L2ARC needs its own copy of the data so that | |
7536 | * the ZIO below can't race with the buf consumer. | |
7537 | * Another case where we need to create a copy of the | |
7538 | * data is when the buffer size is not device-aligned | |
7539 | * and we need to pad the block to make it such. | |
7540 | * That also keeps the clock hand suitably aligned. | |
7541 | * | |
7542 | * To ensure that the copy will be available for the | |
7543 | * lifetime of the ZIO and be cleaned up afterwards, we | |
7544 | * add it to the l2arc_free_on_write queue. | |
34dc7c2f | 7545 | */ |
cae5b340 AX |
7546 | abd_t *to_write; |
7547 | if (!HDR_SHARED_DATA(hdr) && psize == asize) { | |
7548 | to_write = hdr->b_l1hdr.b_pabd; | |
7549 | } else { | |
7550 | to_write = abd_alloc_for_io(asize, | |
7551 | HDR_ISTYPE_METADATA(hdr)); | |
7552 | abd_copy(to_write, hdr->b_l1hdr.b_pabd, psize); | |
7553 | if (asize != psize) { | |
7554 | abd_zero_off(to_write, psize, | |
7555 | asize - psize); | |
7556 | } | |
7557 | l2arc_free_abd_on_write(to_write, asize, | |
7558 | arc_buf_type(hdr)); | |
7559 | } | |
7560 | wzio = zio_write_phys(pio, dev->l2ad_vdev, | |
7561 | hdr->b_l2hdr.b_daddr, asize, to_write, | |
7562 | ZIO_CHECKSUM_OFF, NULL, hdr, | |
7563 | ZIO_PRIORITY_ASYNC_WRITE, | |
7564 | ZIO_FLAG_CANFAIL, B_FALSE); | |
7565 | ||
7566 | write_lsize += HDR_GET_LSIZE(hdr); | |
7567 | DTRACE_PROBE2(l2arc__write, vdev_t *, dev->l2ad_vdev, | |
7568 | zio_t *, wzio); | |
7569 | ||
7570 | write_psize += psize; | |
7571 | write_asize += asize; | |
7572 | dev->l2ad_hand += asize; | |
34dc7c2f BB |
7573 | |
7574 | mutex_exit(hash_lock); | |
7575 | ||
cae5b340 | 7576 | (void) zio_nowait(wzio); |
c06d4368 AX |
7577 | } |
7578 | ||
e10b0808 | 7579 | multilist_sublist_unlock(mls); |
c06d4368 AX |
7580 | |
7581 | if (full == B_TRUE) | |
7582 | break; | |
7583 | } | |
7584 | ||
7585 | /* No buffers selected for writing? */ | |
7586 | if (pio == NULL) { | |
cae5b340 | 7587 | ASSERT0(write_lsize); |
e10b0808 AX |
7588 | ASSERT(!HDR_HAS_L1HDR(head)); |
7589 | kmem_cache_free(hdr_l2only_cache, head); | |
c06d4368 AX |
7590 | return (0); |
7591 | } | |
7592 | ||
c06d4368 | 7593 | ASSERT3U(write_asize, <=, target_sz); |
34dc7c2f | 7594 | ARCSTAT_BUMP(arcstat_l2_writes_sent); |
cae5b340 AX |
7595 | ARCSTAT_INCR(arcstat_l2_write_bytes, write_psize); |
7596 | ARCSTAT_INCR(arcstat_l2_lsize, write_lsize); | |
7597 | ARCSTAT_INCR(arcstat_l2_psize, write_psize); | |
7598 | vdev_space_update(dev->l2ad_vdev, write_psize, 0, 0); | |
34dc7c2f BB |
7599 | |
7600 | /* | |
7601 | * Bump device hand to the device start if it is approaching the end. | |
7602 | * l2arc_evict() will already have evicted ahead for this case. | |
7603 | */ | |
b128c09f | 7604 | if (dev->l2ad_hand >= (dev->l2ad_end - target_sz)) { |
34dc7c2f | 7605 | dev->l2ad_hand = dev->l2ad_start; |
34dc7c2f BB |
7606 | dev->l2ad_first = B_FALSE; |
7607 | } | |
7608 | ||
d164b209 | 7609 | dev->l2ad_writing = B_TRUE; |
34dc7c2f | 7610 | (void) zio_wait(pio); |
d164b209 BB |
7611 | dev->l2ad_writing = B_FALSE; |
7612 | ||
c06d4368 AX |
7613 | return (write_asize); |
7614 | } | |
7615 | ||
34dc7c2f BB |
7616 | /* |
7617 | * This thread feeds the L2ARC at regular intervals. This is the beating | |
7618 | * heart of the L2ARC. | |
7619 | */ | |
7620 | static void | |
7621 | l2arc_feed_thread(void) | |
7622 | { | |
7623 | callb_cpr_t cpr; | |
7624 | l2arc_dev_t *dev; | |
7625 | spa_t *spa; | |
d164b209 | 7626 | uint64_t size, wrote; |
428870ff | 7627 | clock_t begin, next = ddi_get_lbolt(); |
ea04106b | 7628 | fstrans_cookie_t cookie; |
34dc7c2f BB |
7629 | |
7630 | CALLB_CPR_INIT(&cpr, &l2arc_feed_thr_lock, callb_generic_cpr, FTAG); | |
7631 | ||
7632 | mutex_enter(&l2arc_feed_thr_lock); | |
7633 | ||
ea04106b | 7634 | cookie = spl_fstrans_mark(); |
34dc7c2f | 7635 | while (l2arc_thread_exit == 0) { |
34dc7c2f | 7636 | CALLB_CPR_SAFE_BEGIN(&cpr); |
e10b0808 | 7637 | (void) cv_timedwait_sig(&l2arc_feed_thr_cv, |
5b63b3eb | 7638 | &l2arc_feed_thr_lock, next); |
34dc7c2f | 7639 | CALLB_CPR_SAFE_END(&cpr, &l2arc_feed_thr_lock); |
428870ff | 7640 | next = ddi_get_lbolt() + hz; |
34dc7c2f BB |
7641 | |
7642 | /* | |
b128c09f | 7643 | * Quick check for L2ARC devices. |
34dc7c2f BB |
7644 | */ |
7645 | mutex_enter(&l2arc_dev_mtx); | |
7646 | if (l2arc_ndev == 0) { | |
7647 | mutex_exit(&l2arc_dev_mtx); | |
7648 | continue; | |
7649 | } | |
b128c09f | 7650 | mutex_exit(&l2arc_dev_mtx); |
428870ff | 7651 | begin = ddi_get_lbolt(); |
34dc7c2f BB |
7652 | |
7653 | /* | |
b128c09f BB |
7654 | * This selects the next l2arc device to write to, and in |
7655 | * doing so the next spa to feed from: dev->l2ad_spa. This | |
7656 | * will return NULL if there are now no l2arc devices or if | |
7657 | * they are all faulted. | |
7658 | * | |
7659 | * If a device is returned, its spa's config lock is also | |
7660 | * held to prevent device removal. l2arc_dev_get_next() | |
7661 | * will grab and release l2arc_dev_mtx. | |
34dc7c2f | 7662 | */ |
b128c09f | 7663 | if ((dev = l2arc_dev_get_next()) == NULL) |
34dc7c2f | 7664 | continue; |
b128c09f BB |
7665 | |
7666 | spa = dev->l2ad_spa; | |
cae5b340 | 7667 | ASSERT3P(spa, !=, NULL); |
34dc7c2f | 7668 | |
572e2857 BB |
7669 | /* |
7670 | * If the pool is read-only then force the feed thread to | |
7671 | * sleep a little longer. | |
7672 | */ | |
7673 | if (!spa_writeable(spa)) { | |
7674 | next = ddi_get_lbolt() + 5 * l2arc_feed_secs * hz; | |
7675 | spa_config_exit(spa, SCL_L2ARC, dev); | |
7676 | continue; | |
7677 | } | |
7678 | ||
34dc7c2f | 7679 | /* |
b128c09f | 7680 | * Avoid contributing to memory pressure. |
34dc7c2f | 7681 | */ |
e10b0808 | 7682 | if (arc_reclaim_needed()) { |
b128c09f BB |
7683 | ARCSTAT_BUMP(arcstat_l2_abort_lowmem); |
7684 | spa_config_exit(spa, SCL_L2ARC, dev); | |
34dc7c2f BB |
7685 | continue; |
7686 | } | |
b128c09f | 7687 | |
34dc7c2f BB |
7688 | ARCSTAT_BUMP(arcstat_l2_feeds); |
7689 | ||
c06d4368 | 7690 | size = l2arc_write_size(); |
b128c09f | 7691 | |
34dc7c2f BB |
7692 | /* |
7693 | * Evict L2ARC buffers that will be overwritten. | |
7694 | */ | |
b128c09f | 7695 | l2arc_evict(dev, size, B_FALSE); |
34dc7c2f BB |
7696 | |
7697 | /* | |
7698 | * Write ARC buffers. | |
7699 | */ | |
cae5b340 | 7700 | wrote = l2arc_write_buffers(spa, dev, size); |
d164b209 BB |
7701 | |
7702 | /* | |
7703 | * Calculate interval between writes. | |
7704 | */ | |
7705 | next = l2arc_write_interval(begin, size, wrote); | |
b128c09f | 7706 | spa_config_exit(spa, SCL_L2ARC, dev); |
34dc7c2f | 7707 | } |
ea04106b | 7708 | spl_fstrans_unmark(cookie); |
34dc7c2f BB |
7709 | |
7710 | l2arc_thread_exit = 0; | |
7711 | cv_broadcast(&l2arc_feed_thr_cv); | |
7712 | CALLB_CPR_EXIT(&cpr); /* drops l2arc_feed_thr_lock */ | |
7713 | thread_exit(); | |
7714 | } | |
7715 | ||
b128c09f BB |
7716 | boolean_t |
7717 | l2arc_vdev_present(vdev_t *vd) | |
7718 | { | |
7719 | l2arc_dev_t *dev; | |
7720 | ||
7721 | mutex_enter(&l2arc_dev_mtx); | |
7722 | for (dev = list_head(l2arc_dev_list); dev != NULL; | |
7723 | dev = list_next(l2arc_dev_list, dev)) { | |
7724 | if (dev->l2ad_vdev == vd) | |
7725 | break; | |
7726 | } | |
7727 | mutex_exit(&l2arc_dev_mtx); | |
7728 | ||
7729 | return (dev != NULL); | |
7730 | } | |
7731 | ||
34dc7c2f BB |
7732 | /* |
7733 | * Add a vdev for use by the L2ARC. By this point the spa has already | |
7734 | * validated the vdev and opened it. | |
7735 | */ | |
7736 | void | |
9babb374 | 7737 | l2arc_add_vdev(spa_t *spa, vdev_t *vd) |
34dc7c2f BB |
7738 | { |
7739 | l2arc_dev_t *adddev; | |
7740 | ||
b128c09f BB |
7741 | ASSERT(!l2arc_vdev_present(vd)); |
7742 | ||
34dc7c2f BB |
7743 | /* |
7744 | * Create a new l2arc device entry. | |
7745 | */ | |
7746 | adddev = kmem_zalloc(sizeof (l2arc_dev_t), KM_SLEEP); | |
7747 | adddev->l2ad_spa = spa; | |
7748 | adddev->l2ad_vdev = vd; | |
9babb374 BB |
7749 | adddev->l2ad_start = VDEV_LABEL_START_SIZE; |
7750 | adddev->l2ad_end = VDEV_LABEL_START_SIZE + vdev_get_min_asize(vd); | |
34dc7c2f | 7751 | adddev->l2ad_hand = adddev->l2ad_start; |
34dc7c2f | 7752 | adddev->l2ad_first = B_TRUE; |
d164b209 | 7753 | adddev->l2ad_writing = B_FALSE; |
98f72a53 | 7754 | list_link_init(&adddev->l2ad_node); |
34dc7c2f | 7755 | |
e10b0808 | 7756 | mutex_init(&adddev->l2ad_mtx, NULL, MUTEX_DEFAULT, NULL); |
34dc7c2f BB |
7757 | /* |
7758 | * This is a list of all ARC buffers that are still valid on the | |
7759 | * device. | |
7760 | */ | |
e10b0808 AX |
7761 | list_create(&adddev->l2ad_buflist, sizeof (arc_buf_hdr_t), |
7762 | offsetof(arc_buf_hdr_t, b_l2hdr.b_l2node)); | |
34dc7c2f | 7763 | |
428870ff | 7764 | vdev_space_update(vd, 0, 0, adddev->l2ad_end - adddev->l2ad_hand); |
e10b0808 | 7765 | refcount_create(&adddev->l2ad_alloc); |
34dc7c2f BB |
7766 | |
7767 | /* | |
7768 | * Add device to global list | |
7769 | */ | |
7770 | mutex_enter(&l2arc_dev_mtx); | |
7771 | list_insert_head(l2arc_dev_list, adddev); | |
7772 | atomic_inc_64(&l2arc_ndev); | |
7773 | mutex_exit(&l2arc_dev_mtx); | |
7774 | } | |
7775 | ||
7776 | /* | |
7777 | * Remove a vdev from the L2ARC. | |
7778 | */ | |
7779 | void | |
7780 | l2arc_remove_vdev(vdev_t *vd) | |
7781 | { | |
7782 | l2arc_dev_t *dev, *nextdev, *remdev = NULL; | |
7783 | ||
34dc7c2f BB |
7784 | /* |
7785 | * Find the device by vdev | |
7786 | */ | |
7787 | mutex_enter(&l2arc_dev_mtx); | |
7788 | for (dev = list_head(l2arc_dev_list); dev; dev = nextdev) { | |
7789 | nextdev = list_next(l2arc_dev_list, dev); | |
7790 | if (vd == dev->l2ad_vdev) { | |
7791 | remdev = dev; | |
7792 | break; | |
7793 | } | |
7794 | } | |
cae5b340 | 7795 | ASSERT3P(remdev, !=, NULL); |
34dc7c2f BB |
7796 | |
7797 | /* | |
7798 | * Remove device from global list | |
7799 | */ | |
7800 | list_remove(l2arc_dev_list, remdev); | |
7801 | l2arc_dev_last = NULL; /* may have been invalidated */ | |
b128c09f BB |
7802 | atomic_dec_64(&l2arc_ndev); |
7803 | mutex_exit(&l2arc_dev_mtx); | |
34dc7c2f BB |
7804 | |
7805 | /* | |
7806 | * Clear all buflists and ARC references. L2ARC device flush. | |
7807 | */ | |
7808 | l2arc_evict(remdev, 0, B_TRUE); | |
e10b0808 AX |
7809 | list_destroy(&remdev->l2ad_buflist); |
7810 | mutex_destroy(&remdev->l2ad_mtx); | |
7811 | refcount_destroy(&remdev->l2ad_alloc); | |
34dc7c2f | 7812 | kmem_free(remdev, sizeof (l2arc_dev_t)); |
34dc7c2f BB |
7813 | } |
7814 | ||
7815 | void | |
b128c09f | 7816 | l2arc_init(void) |
34dc7c2f BB |
7817 | { |
7818 | l2arc_thread_exit = 0; | |
7819 | l2arc_ndev = 0; | |
7820 | l2arc_writes_sent = 0; | |
7821 | l2arc_writes_done = 0; | |
7822 | ||
7823 | mutex_init(&l2arc_feed_thr_lock, NULL, MUTEX_DEFAULT, NULL); | |
7824 | cv_init(&l2arc_feed_thr_cv, NULL, CV_DEFAULT, NULL); | |
7825 | mutex_init(&l2arc_dev_mtx, NULL, MUTEX_DEFAULT, NULL); | |
34dc7c2f BB |
7826 | mutex_init(&l2arc_free_on_write_mtx, NULL, MUTEX_DEFAULT, NULL); |
7827 | ||
7828 | l2arc_dev_list = &L2ARC_dev_list; | |
7829 | l2arc_free_on_write = &L2ARC_free_on_write; | |
7830 | list_create(l2arc_dev_list, sizeof (l2arc_dev_t), | |
7831 | offsetof(l2arc_dev_t, l2ad_node)); | |
7832 | list_create(l2arc_free_on_write, sizeof (l2arc_data_free_t), | |
7833 | offsetof(l2arc_data_free_t, l2df_list_node)); | |
34dc7c2f BB |
7834 | } |
7835 | ||
7836 | void | |
b128c09f | 7837 | l2arc_fini(void) |
34dc7c2f | 7838 | { |
b128c09f BB |
7839 | /* |
7840 | * This is called from dmu_fini(), which is called from spa_fini(); | |
7841 | * Because of this, we can assume that all l2arc devices have | |
7842 | * already been removed when the pools themselves were removed. | |
7843 | */ | |
7844 | ||
7845 | l2arc_do_free_on_write(); | |
34dc7c2f BB |
7846 | |
7847 | mutex_destroy(&l2arc_feed_thr_lock); | |
7848 | cv_destroy(&l2arc_feed_thr_cv); | |
7849 | mutex_destroy(&l2arc_dev_mtx); | |
34dc7c2f BB |
7850 | mutex_destroy(&l2arc_free_on_write_mtx); |
7851 | ||
7852 | list_destroy(l2arc_dev_list); | |
7853 | list_destroy(l2arc_free_on_write); | |
7854 | } | |
b128c09f BB |
7855 | |
7856 | void | |
7857 | l2arc_start(void) | |
7858 | { | |
fb5f0bc8 | 7859 | if (!(spa_mode_global & FWRITE)) |
b128c09f BB |
7860 | return; |
7861 | ||
7862 | (void) thread_create(NULL, 0, l2arc_feed_thread, NULL, 0, &p0, | |
e10b0808 | 7863 | TS_RUN, defclsyspri); |
b128c09f BB |
7864 | } |
7865 | ||
7866 | void | |
7867 | l2arc_stop(void) | |
7868 | { | |
fb5f0bc8 | 7869 | if (!(spa_mode_global & FWRITE)) |
b128c09f BB |
7870 | return; |
7871 | ||
7872 | mutex_enter(&l2arc_feed_thr_lock); | |
7873 | cv_signal(&l2arc_feed_thr_cv); /* kick thread out of startup */ | |
7874 | l2arc_thread_exit = 1; | |
7875 | while (l2arc_thread_exit != 0) | |
7876 | cv_wait(&l2arc_feed_thr_cv, &l2arc_feed_thr_lock); | |
7877 | mutex_exit(&l2arc_feed_thr_lock); | |
7878 | } | |
c28b2279 BB |
7879 | |
7880 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
ea04106b AX |
7881 | EXPORT_SYMBOL(arc_buf_size); |
7882 | EXPORT_SYMBOL(arc_write); | |
c28b2279 | 7883 | EXPORT_SYMBOL(arc_read); |
a08ee875 | 7884 | EXPORT_SYMBOL(arc_buf_info); |
c28b2279 | 7885 | EXPORT_SYMBOL(arc_getbuf_func); |
ab26409d BB |
7886 | EXPORT_SYMBOL(arc_add_prune_callback); |
7887 | EXPORT_SYMBOL(arc_remove_prune_callback); | |
c28b2279 | 7888 | |
cae5b340 | 7889 | /* BEGIN CSTYLED */ |
c06d4368 | 7890 | module_param(zfs_arc_min, ulong, 0644); |
c409e464 | 7891 | MODULE_PARM_DESC(zfs_arc_min, "Min arc size"); |
c28b2279 | 7892 | |
c06d4368 | 7893 | module_param(zfs_arc_max, ulong, 0644); |
c409e464 | 7894 | MODULE_PARM_DESC(zfs_arc_max, "Max arc size"); |
c28b2279 | 7895 | |
c06d4368 | 7896 | module_param(zfs_arc_meta_limit, ulong, 0644); |
c28b2279 | 7897 | MODULE_PARM_DESC(zfs_arc_meta_limit, "Meta limit for arc size"); |
6a8f9b6b | 7898 | |
cae5b340 AX |
7899 | module_param(zfs_arc_meta_limit_percent, ulong, 0644); |
7900 | MODULE_PARM_DESC(zfs_arc_meta_limit_percent, | |
7901 | "Percent of arc size for arc meta limit"); | |
7902 | ||
e10b0808 AX |
7903 | module_param(zfs_arc_meta_min, ulong, 0644); |
7904 | MODULE_PARM_DESC(zfs_arc_meta_min, "Min arc metadata"); | |
7905 | ||
c06d4368 | 7906 | module_param(zfs_arc_meta_prune, int, 0644); |
ea04106b AX |
7907 | MODULE_PARM_DESC(zfs_arc_meta_prune, "Meta objects to scan for prune"); |
7908 | ||
e10b0808 | 7909 | module_param(zfs_arc_meta_adjust_restarts, int, 0644); |
ea04106b AX |
7910 | MODULE_PARM_DESC(zfs_arc_meta_adjust_restarts, |
7911 | "Limit number of restarts in arc_adjust_meta"); | |
c409e464 | 7912 | |
e10b0808 AX |
7913 | module_param(zfs_arc_meta_strategy, int, 0644); |
7914 | MODULE_PARM_DESC(zfs_arc_meta_strategy, "Meta reclaim strategy"); | |
7915 | ||
c06d4368 | 7916 | module_param(zfs_arc_grow_retry, int, 0644); |
c409e464 BB |
7917 | MODULE_PARM_DESC(zfs_arc_grow_retry, "Seconds before growing arc size"); |
7918 | ||
ea04106b AX |
7919 | module_param(zfs_arc_p_dampener_disable, int, 0644); |
7920 | MODULE_PARM_DESC(zfs_arc_p_dampener_disable, "disable arc_p adapt dampener"); | |
7921 | ||
c06d4368 | 7922 | module_param(zfs_arc_shrink_shift, int, 0644); |
c409e464 BB |
7923 | MODULE_PARM_DESC(zfs_arc_shrink_shift, "log2(fraction of arc to reclaim)"); |
7924 | ||
cae5b340 AX |
7925 | module_param(zfs_arc_pc_percent, uint, 0644); |
7926 | MODULE_PARM_DESC(zfs_arc_pc_percent, | |
7927 | "Percent of pagecache to reclaim arc to"); | |
7928 | ||
e10b0808 AX |
7929 | module_param(zfs_arc_p_min_shift, int, 0644); |
7930 | MODULE_PARM_DESC(zfs_arc_p_min_shift, "arc_c shift to calc min/max arc_p"); | |
7931 | ||
ea04106b AX |
7932 | module_param(zfs_arc_average_blocksize, int, 0444); |
7933 | MODULE_PARM_DESC(zfs_arc_average_blocksize, "Target average block size"); | |
7934 | ||
cae5b340 AX |
7935 | module_param(zfs_compressed_arc_enabled, int, 0644); |
7936 | MODULE_PARM_DESC(zfs_compressed_arc_enabled, "Disable compressed arc buffers"); | |
7937 | ||
c06d4368 AX |
7938 | module_param(zfs_arc_min_prefetch_lifespan, int, 0644); |
7939 | MODULE_PARM_DESC(zfs_arc_min_prefetch_lifespan, "Min life of prefetch block"); | |
7940 | ||
7941 | module_param(l2arc_write_max, ulong, 0644); | |
abd8610c BB |
7942 | MODULE_PARM_DESC(l2arc_write_max, "Max write bytes per interval"); |
7943 | ||
c06d4368 | 7944 | module_param(l2arc_write_boost, ulong, 0644); |
abd8610c BB |
7945 | MODULE_PARM_DESC(l2arc_write_boost, "Extra write bytes during device warmup"); |
7946 | ||
c06d4368 | 7947 | module_param(l2arc_headroom, ulong, 0644); |
abd8610c BB |
7948 | MODULE_PARM_DESC(l2arc_headroom, "Number of max device writes to precache"); |
7949 | ||
c06d4368 AX |
7950 | module_param(l2arc_headroom_boost, ulong, 0644); |
7951 | MODULE_PARM_DESC(l2arc_headroom_boost, "Compressed l2arc_headroom multiplier"); | |
7952 | ||
7953 | module_param(l2arc_feed_secs, ulong, 0644); | |
abd8610c BB |
7954 | MODULE_PARM_DESC(l2arc_feed_secs, "Seconds between L2ARC writing"); |
7955 | ||
c06d4368 | 7956 | module_param(l2arc_feed_min_ms, ulong, 0644); |
abd8610c BB |
7957 | MODULE_PARM_DESC(l2arc_feed_min_ms, "Min feed interval in milliseconds"); |
7958 | ||
c06d4368 | 7959 | module_param(l2arc_noprefetch, int, 0644); |
abd8610c BB |
7960 | MODULE_PARM_DESC(l2arc_noprefetch, "Skip caching prefetched buffers"); |
7961 | ||
c06d4368 | 7962 | module_param(l2arc_feed_again, int, 0644); |
abd8610c BB |
7963 | MODULE_PARM_DESC(l2arc_feed_again, "Turbo L2ARC warmup"); |
7964 | ||
c06d4368 | 7965 | module_param(l2arc_norw, int, 0644); |
abd8610c BB |
7966 | MODULE_PARM_DESC(l2arc_norw, "No reads during writes"); |
7967 | ||
e10b0808 AX |
7968 | module_param(zfs_arc_lotsfree_percent, int, 0644); |
7969 | MODULE_PARM_DESC(zfs_arc_lotsfree_percent, | |
7970 | "System free memory I/O throttle in bytes"); | |
7971 | ||
7972 | module_param(zfs_arc_sys_free, ulong, 0644); | |
7973 | MODULE_PARM_DESC(zfs_arc_sys_free, "System free memory target size in bytes"); | |
7974 | ||
cae5b340 AX |
7975 | module_param(zfs_arc_dnode_limit, ulong, 0644); |
7976 | MODULE_PARM_DESC(zfs_arc_dnode_limit, "Minimum bytes of dnodes in arc"); | |
7977 | ||
7978 | module_param(zfs_arc_dnode_limit_percent, ulong, 0644); | |
7979 | MODULE_PARM_DESC(zfs_arc_dnode_limit_percent, | |
7980 | "Percent of ARC meta buffers for dnodes"); | |
7981 | ||
7982 | module_param(zfs_arc_dnode_reduce_percent, ulong, 0644); | |
7983 | MODULE_PARM_DESC(zfs_arc_dnode_reduce_percent, | |
7984 | "Percentage of excess dnodes to try to unpin"); | |
7985 | /* END CSTYLED */ | |
c28b2279 | 7986 | #endif |