]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
1 | /* |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
9 | * or http://www.opensolaris.org/os/licensing. | |
10 | * See the License for the specific language governing permissions | |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | /* | |
428870ff | 22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
36da08ef PS |
23 | * Copyright (c) 2012, Joyent, Inc. All rights reserved. |
24 | * Copyright (c) 2011, 2015 by Delphix. All rights reserved. | |
25 | * Copyright (c) 2014 by Saso Kiselkov. All rights reserved. | |
3bec585e | 26 | * Copyright 2014 Nexenta Systems, Inc. All rights reserved. |
34dc7c2f BB |
27 | */ |
28 | ||
34dc7c2f BB |
29 | /* |
30 | * DVA-based Adjustable Replacement Cache | |
31 | * | |
32 | * While much of the theory of operation used here is | |
33 | * based on the self-tuning, low overhead replacement cache | |
34 | * presented by Megiddo and Modha at FAST 2003, there are some | |
35 | * significant differences: | |
36 | * | |
37 | * 1. The Megiddo and Modha model assumes any page is evictable. | |
38 | * Pages in its cache cannot be "locked" into memory. This makes | |
39 | * the eviction algorithm simple: evict the last page in the list. | |
40 | * This also make the performance characteristics easy to reason | |
41 | * about. Our cache is not so simple. At any given moment, some | |
42 | * subset of the blocks in the cache are un-evictable because we | |
43 | * have handed out a reference to them. Blocks are only evictable | |
44 | * when there are no external references active. This makes | |
45 | * eviction far more problematic: we choose to evict the evictable | |
46 | * blocks that are the "lowest" in the list. | |
47 | * | |
48 | * There are times when it is not possible to evict the requested | |
49 | * space. In these circumstances we are unable to adjust the cache | |
50 | * size. To prevent the cache growing unbounded at these times we | |
51 | * implement a "cache throttle" that slows the flow of new data | |
52 | * into the cache until we can make space available. | |
53 | * | |
54 | * 2. The Megiddo and Modha model assumes a fixed cache size. | |
55 | * Pages are evicted when the cache is full and there is a cache | |
56 | * miss. Our model has a variable sized cache. It grows with | |
57 | * high use, but also tries to react to memory pressure from the | |
58 | * operating system: decreasing its size when system memory is | |
59 | * tight. | |
60 | * | |
61 | * 3. The Megiddo and Modha model assumes a fixed page size. All | |
d3cc8b15 | 62 | * elements of the cache are therefore exactly the same size. So |
34dc7c2f BB |
63 | * when adjusting the cache size following a cache miss, its simply |
64 | * a matter of choosing a single page to evict. In our model, we | |
65 | * have variable sized cache blocks (rangeing from 512 bytes to | |
d3cc8b15 | 66 | * 128K bytes). We therefore choose a set of blocks to evict to make |
34dc7c2f BB |
67 | * space for a cache miss that approximates as closely as possible |
68 | * the space used by the new block. | |
69 | * | |
70 | * See also: "ARC: A Self-Tuning, Low Overhead Replacement Cache" | |
71 | * by N. Megiddo & D. Modha, FAST 2003 | |
72 | */ | |
73 | ||
74 | /* | |
75 | * The locking model: | |
76 | * | |
77 | * A new reference to a cache buffer can be obtained in two | |
78 | * ways: 1) via a hash table lookup using the DVA as a key, | |
79 | * or 2) via one of the ARC lists. The arc_read() interface | |
80 | * uses method 1, while the internal arc algorithms for | |
d3cc8b15 | 81 | * adjusting the cache use method 2. We therefore provide two |
34dc7c2f BB |
82 | * types of locks: 1) the hash table lock array, and 2) the |
83 | * arc list locks. | |
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 | * | |
96 | * Each arc state also has a mutex which is used to protect the | |
97 | * buffer list associated with the state. When attempting to | |
98 | * obtain a hash table lock while holding an arc list lock you | |
99 | * must use: mutex_tryenter() to avoid deadlock. Also note that | |
100 | * the active state mutex must be held before the ghost state mutex. | |
101 | * | |
102 | * Arc buffers may have an associated eviction callback function. | |
103 | * This function will be invoked prior to removing the buffer (e.g. | |
104 | * in arc_do_user_evicts()). Note however that the data associated | |
105 | * with the buffer may be evicted prior to the callback. The callback | |
106 | * must be made with *no locks held* (to prevent deadlock). Additionally, | |
107 | * the users of callbacks must ensure that their private data is | |
bd089c54 | 108 | * protected from simultaneous callbacks from arc_clear_callback() |
34dc7c2f BB |
109 | * and arc_do_user_evicts(). |
110 | * | |
ab26409d BB |
111 | * It as also possible to register a callback which is run when the |
112 | * arc_meta_limit is reached and no buffers can be safely evicted. In | |
113 | * this case the arc user should drop a reference on some arc buffers so | |
114 | * they can be reclaimed and the arc_meta_limit honored. For example, | |
115 | * when using the ZPL each dentry holds a references on a znode. These | |
116 | * dentries must be pruned before the arc buffer holding the znode can | |
117 | * be safely evicted. | |
118 | * | |
34dc7c2f BB |
119 | * Note that the majority of the performance stats are manipulated |
120 | * with atomic operations. | |
121 | * | |
b9541d6b | 122 | * The L2ARC uses the l2ad_mtx on each vdev for the following: |
34dc7c2f BB |
123 | * |
124 | * - L2ARC buflist creation | |
125 | * - L2ARC buflist eviction | |
126 | * - L2ARC write completion, which walks L2ARC buflists | |
127 | * - ARC header destruction, as it removes from L2ARC buflists | |
128 | * - ARC header release, as it removes from L2ARC buflists | |
129 | */ | |
130 | ||
131 | #include <sys/spa.h> | |
132 | #include <sys/zio.h> | |
3a17a7a9 | 133 | #include <sys/zio_compress.h> |
34dc7c2f BB |
134 | #include <sys/zfs_context.h> |
135 | #include <sys/arc.h> | |
36da08ef | 136 | #include <sys/refcount.h> |
b128c09f | 137 | #include <sys/vdev.h> |
9babb374 | 138 | #include <sys/vdev_impl.h> |
e8b96c60 | 139 | #include <sys/dsl_pool.h> |
ca0bf58d | 140 | #include <sys/multilist.h> |
34dc7c2f BB |
141 | #ifdef _KERNEL |
142 | #include <sys/vmsystm.h> | |
143 | #include <vm/anon.h> | |
144 | #include <sys/fs/swapnode.h> | |
ab26409d | 145 | #include <sys/zpl.h> |
aaed7c40 | 146 | #include <linux/mm_compat.h> |
34dc7c2f BB |
147 | #endif |
148 | #include <sys/callb.h> | |
149 | #include <sys/kstat.h> | |
570827e1 | 150 | #include <sys/dmu_tx.h> |
428870ff | 151 | #include <zfs_fletcher.h> |
59ec819a | 152 | #include <sys/arc_impl.h> |
49ee64e5 | 153 | #include <sys/trace_arc.h> |
34dc7c2f | 154 | |
498877ba MA |
155 | #ifndef _KERNEL |
156 | /* set with ZFS_DEBUG=watch, to enable watchpoints on frozen buffers */ | |
157 | boolean_t arc_watch = B_FALSE; | |
158 | #endif | |
159 | ||
ca0bf58d PS |
160 | static kmutex_t arc_reclaim_lock; |
161 | static kcondvar_t arc_reclaim_thread_cv; | |
162 | static boolean_t arc_reclaim_thread_exit; | |
163 | static kcondvar_t arc_reclaim_waiters_cv; | |
164 | ||
165 | static kmutex_t arc_user_evicts_lock; | |
166 | static kcondvar_t arc_user_evicts_cv; | |
167 | static boolean_t arc_user_evicts_thread_exit; | |
34dc7c2f | 168 | |
e8b96c60 | 169 | /* |
ca0bf58d PS |
170 | * The number of headers to evict in arc_evict_state_impl() before |
171 | * dropping the sublist lock and evicting from another sublist. A lower | |
172 | * value means we're more likely to evict the "correct" header (i.e. the | |
173 | * oldest header in the arc state), but comes with higher overhead | |
174 | * (i.e. more invocations of arc_evict_state_impl()). | |
175 | */ | |
176 | int zfs_arc_evict_batch_limit = 10; | |
177 | ||
178 | /* | |
179 | * The number of sublists used for each of the arc state lists. If this | |
180 | * is not set to a suitable value by the user, it will be configured to | |
181 | * the number of CPUs on the system in arc_init(). | |
e8b96c60 | 182 | */ |
ca0bf58d | 183 | int zfs_arc_num_sublists_per_state = 0; |
e8b96c60 | 184 | |
34dc7c2f | 185 | /* number of seconds before growing cache again */ |
ca67b33a | 186 | static int arc_grow_retry = 5; |
34dc7c2f | 187 | |
ca0bf58d | 188 | /* shift of arc_c for calculating overflow limit in arc_get_data_buf */ |
ca67b33a | 189 | int zfs_arc_overflow_shift = 8; |
62422785 | 190 | |
728d6ae9 BB |
191 | /* shift of arc_c for calculating both min and max arc_p */ |
192 | static int arc_p_min_shift = 4; | |
193 | ||
d164b209 | 194 | /* log2(fraction of arc to reclaim) */ |
ca67b33a | 195 | static int arc_shrink_shift = 7; |
d164b209 | 196 | |
34dc7c2f | 197 | /* |
ca67b33a MA |
198 | * log2(fraction of ARC which must be free to allow growing). |
199 | * I.e. If there is less than arc_c >> arc_no_grow_shift free memory, | |
200 | * when reading a new block into the ARC, we will evict an equal-sized block | |
201 | * from the ARC. | |
202 | * | |
203 | * This must be less than arc_shrink_shift, so that when we shrink the ARC, | |
204 | * we will still not allow it to grow. | |
34dc7c2f | 205 | */ |
ca67b33a | 206 | int arc_no_grow_shift = 5; |
bce45ec9 | 207 | |
49ddb315 | 208 | |
ca0bf58d PS |
209 | /* |
210 | * minimum lifespan of a prefetch block in clock ticks | |
211 | * (initialized in arc_init()) | |
212 | */ | |
ca67b33a | 213 | static int arc_min_prefetch_lifespan; |
ca0bf58d | 214 | |
e8b96c60 MA |
215 | /* |
216 | * If this percent of memory is free, don't throttle. | |
217 | */ | |
218 | int arc_lotsfree_percent = 10; | |
219 | ||
34dc7c2f BB |
220 | static int arc_dead; |
221 | ||
b128c09f BB |
222 | /* |
223 | * The arc has filled available memory and has now warmed up. | |
224 | */ | |
225 | static boolean_t arc_warm; | |
226 | ||
34dc7c2f BB |
227 | /* |
228 | * These tunables are for performance analysis. | |
229 | */ | |
c28b2279 BB |
230 | unsigned long zfs_arc_max = 0; |
231 | unsigned long zfs_arc_min = 0; | |
232 | unsigned long zfs_arc_meta_limit = 0; | |
ca0bf58d | 233 | unsigned long zfs_arc_meta_min = 0; |
ca67b33a MA |
234 | int zfs_arc_grow_retry = 0; |
235 | int zfs_arc_shrink_shift = 0; | |
728d6ae9 | 236 | int zfs_arc_p_min_shift = 0; |
ca67b33a MA |
237 | int zfs_disable_dup_eviction = 0; |
238 | int zfs_arc_average_blocksize = 8 * 1024; /* 8KB */ | |
34dc7c2f | 239 | |
bc888666 | 240 | /* |
ca67b33a | 241 | * These tunables are Linux specific |
bc888666 | 242 | */ |
ca67b33a MA |
243 | int zfs_arc_memory_throttle_disable = 1; |
244 | int zfs_arc_min_prefetch_lifespan = 0; | |
245 | int zfs_arc_p_aggressive_disable = 1; | |
246 | int zfs_arc_p_dampener_disable = 1; | |
247 | int zfs_arc_meta_prune = 10000; | |
248 | int zfs_arc_meta_strategy = ARC_STRATEGY_META_BALANCED; | |
249 | int zfs_arc_meta_adjust_restarts = 4096; | |
bc888666 | 250 | |
34dc7c2f BB |
251 | /* The 6 states: */ |
252 | static arc_state_t ARC_anon; | |
253 | static arc_state_t ARC_mru; | |
254 | static arc_state_t ARC_mru_ghost; | |
255 | static arc_state_t ARC_mfu; | |
256 | static arc_state_t ARC_mfu_ghost; | |
257 | static arc_state_t ARC_l2c_only; | |
258 | ||
259 | typedef struct arc_stats { | |
260 | kstat_named_t arcstat_hits; | |
261 | kstat_named_t arcstat_misses; | |
262 | kstat_named_t arcstat_demand_data_hits; | |
263 | kstat_named_t arcstat_demand_data_misses; | |
264 | kstat_named_t arcstat_demand_metadata_hits; | |
265 | kstat_named_t arcstat_demand_metadata_misses; | |
266 | kstat_named_t arcstat_prefetch_data_hits; | |
267 | kstat_named_t arcstat_prefetch_data_misses; | |
268 | kstat_named_t arcstat_prefetch_metadata_hits; | |
269 | kstat_named_t arcstat_prefetch_metadata_misses; | |
270 | kstat_named_t arcstat_mru_hits; | |
271 | kstat_named_t arcstat_mru_ghost_hits; | |
272 | kstat_named_t arcstat_mfu_hits; | |
273 | kstat_named_t arcstat_mfu_ghost_hits; | |
274 | kstat_named_t arcstat_deleted; | |
e49f1e20 WA |
275 | /* |
276 | * Number of buffers that could not be evicted because the hash lock | |
277 | * was held by another thread. The lock may not necessarily be held | |
278 | * by something using the same buffer, since hash locks are shared | |
279 | * by multiple buffers. | |
280 | */ | |
34dc7c2f | 281 | kstat_named_t arcstat_mutex_miss; |
e49f1e20 WA |
282 | /* |
283 | * Number of buffers skipped because they have I/O in progress, are | |
284 | * indrect prefetch buffers that have not lived long enough, or are | |
285 | * not from the spa we're trying to evict from. | |
286 | */ | |
34dc7c2f | 287 | kstat_named_t arcstat_evict_skip; |
ca0bf58d PS |
288 | /* |
289 | * Number of times arc_evict_state() was unable to evict enough | |
290 | * buffers to reach its target amount. | |
291 | */ | |
292 | kstat_named_t arcstat_evict_not_enough; | |
428870ff BB |
293 | kstat_named_t arcstat_evict_l2_cached; |
294 | kstat_named_t arcstat_evict_l2_eligible; | |
295 | kstat_named_t arcstat_evict_l2_ineligible; | |
ca0bf58d | 296 | kstat_named_t arcstat_evict_l2_skip; |
34dc7c2f BB |
297 | kstat_named_t arcstat_hash_elements; |
298 | kstat_named_t arcstat_hash_elements_max; | |
299 | kstat_named_t arcstat_hash_collisions; | |
300 | kstat_named_t arcstat_hash_chains; | |
301 | kstat_named_t arcstat_hash_chain_max; | |
302 | kstat_named_t arcstat_p; | |
303 | kstat_named_t arcstat_c; | |
304 | kstat_named_t arcstat_c_min; | |
305 | kstat_named_t arcstat_c_max; | |
306 | kstat_named_t arcstat_size; | |
500445c0 PS |
307 | /* |
308 | * Number of bytes consumed by internal ARC structures necessary | |
309 | * for tracking purposes; these structures are not actually | |
310 | * backed by ARC buffers. This includes arc_buf_hdr_t structures | |
311 | * (allocated via arc_buf_hdr_t_full and arc_buf_hdr_t_l2only | |
312 | * caches), and arc_buf_t structures (allocated via arc_buf_t | |
313 | * cache). | |
314 | */ | |
34dc7c2f | 315 | kstat_named_t arcstat_hdr_size; |
500445c0 PS |
316 | /* |
317 | * Number of bytes consumed by ARC buffers of type equal to | |
318 | * ARC_BUFC_DATA. This is generally consumed by buffers backing | |
319 | * on disk user data (e.g. plain file contents). | |
320 | */ | |
d164b209 | 321 | kstat_named_t arcstat_data_size; |
500445c0 PS |
322 | /* |
323 | * Number of bytes consumed by ARC buffers of type equal to | |
324 | * ARC_BUFC_METADATA. This is generally consumed by buffers | |
325 | * backing on disk data that is used for internal ZFS | |
326 | * structures (e.g. ZAP, dnode, indirect blocks, etc). | |
327 | */ | |
328 | kstat_named_t arcstat_metadata_size; | |
329 | /* | |
330 | * Number of bytes consumed by various buffers and structures | |
331 | * not actually backed with ARC buffers. This includes bonus | |
332 | * buffers (allocated directly via zio_buf_* functions), | |
333 | * dmu_buf_impl_t structures (allocated via dmu_buf_impl_t | |
334 | * cache), and dnode_t structures (allocated via dnode_t cache). | |
335 | */ | |
d164b209 | 336 | kstat_named_t arcstat_other_size; |
500445c0 PS |
337 | /* |
338 | * Total number of bytes consumed by ARC buffers residing in the | |
339 | * arc_anon state. This includes *all* buffers in the arc_anon | |
340 | * state; e.g. data, metadata, evictable, and unevictable buffers | |
341 | * are all included in this value. | |
342 | */ | |
13be560d | 343 | kstat_named_t arcstat_anon_size; |
500445c0 PS |
344 | /* |
345 | * Number of bytes consumed by ARC buffers that meet the | |
346 | * following criteria: backing buffers of type ARC_BUFC_DATA, | |
347 | * residing in the arc_anon state, and are eligible for eviction | |
348 | * (e.g. have no outstanding holds on the buffer). | |
349 | */ | |
350 | kstat_named_t arcstat_anon_evictable_data; | |
351 | /* | |
352 | * Number of bytes consumed by ARC buffers that meet the | |
353 | * following criteria: backing buffers of type ARC_BUFC_METADATA, | |
354 | * residing in the arc_anon state, and are eligible for eviction | |
355 | * (e.g. have no outstanding holds on the buffer). | |
356 | */ | |
357 | kstat_named_t arcstat_anon_evictable_metadata; | |
358 | /* | |
359 | * Total number of bytes consumed by ARC buffers residing in the | |
360 | * arc_mru state. This includes *all* buffers in the arc_mru | |
361 | * state; e.g. data, metadata, evictable, and unevictable buffers | |
362 | * are all included in this value. | |
363 | */ | |
13be560d | 364 | kstat_named_t arcstat_mru_size; |
500445c0 PS |
365 | /* |
366 | * Number of bytes consumed by ARC buffers that meet the | |
367 | * following criteria: backing buffers of type ARC_BUFC_DATA, | |
368 | * residing in the arc_mru state, and are eligible for eviction | |
369 | * (e.g. have no outstanding holds on the buffer). | |
370 | */ | |
371 | kstat_named_t arcstat_mru_evictable_data; | |
372 | /* | |
373 | * Number of bytes consumed by ARC buffers that meet the | |
374 | * following criteria: backing buffers of type ARC_BUFC_METADATA, | |
375 | * residing in the arc_mru state, and are eligible for eviction | |
376 | * (e.g. have no outstanding holds on the buffer). | |
377 | */ | |
378 | kstat_named_t arcstat_mru_evictable_metadata; | |
379 | /* | |
380 | * Total number of bytes that *would have been* consumed by ARC | |
381 | * buffers in the arc_mru_ghost state. The key thing to note | |
382 | * here, is the fact that this size doesn't actually indicate | |
383 | * RAM consumption. The ghost lists only consist of headers and | |
384 | * don't actually have ARC buffers linked off of these headers. | |
385 | * Thus, *if* the headers had associated ARC buffers, these | |
386 | * buffers *would have* consumed this number of bytes. | |
387 | */ | |
13be560d | 388 | kstat_named_t arcstat_mru_ghost_size; |
500445c0 PS |
389 | /* |
390 | * Number of bytes that *would have been* consumed by ARC | |
391 | * buffers that are eligible for eviction, of type | |
392 | * ARC_BUFC_DATA, and linked off the arc_mru_ghost state. | |
393 | */ | |
394 | kstat_named_t arcstat_mru_ghost_evictable_data; | |
395 | /* | |
396 | * Number of bytes that *would have been* consumed by ARC | |
397 | * buffers that are eligible for eviction, of type | |
398 | * ARC_BUFC_METADATA, and linked off the arc_mru_ghost state. | |
399 | */ | |
400 | kstat_named_t arcstat_mru_ghost_evictable_metadata; | |
401 | /* | |
402 | * Total number of bytes consumed by ARC buffers residing in the | |
403 | * arc_mfu state. This includes *all* buffers in the arc_mfu | |
404 | * state; e.g. data, metadata, evictable, and unevictable buffers | |
405 | * are all included in this value. | |
406 | */ | |
13be560d | 407 | kstat_named_t arcstat_mfu_size; |
500445c0 PS |
408 | /* |
409 | * Number of bytes consumed by ARC buffers that are eligible for | |
410 | * eviction, of type ARC_BUFC_DATA, and reside in the arc_mfu | |
411 | * state. | |
412 | */ | |
413 | kstat_named_t arcstat_mfu_evictable_data; | |
414 | /* | |
415 | * Number of bytes consumed by ARC buffers that are eligible for | |
416 | * eviction, of type ARC_BUFC_METADATA, and reside in the | |
417 | * arc_mfu state. | |
418 | */ | |
419 | kstat_named_t arcstat_mfu_evictable_metadata; | |
420 | /* | |
421 | * Total number of bytes that *would have been* consumed by ARC | |
422 | * buffers in the arc_mfu_ghost state. See the comment above | |
423 | * arcstat_mru_ghost_size for more details. | |
424 | */ | |
13be560d | 425 | kstat_named_t arcstat_mfu_ghost_size; |
500445c0 PS |
426 | /* |
427 | * Number of bytes that *would have been* consumed by ARC | |
428 | * buffers that are eligible for eviction, of type | |
429 | * ARC_BUFC_DATA, and linked off the arc_mfu_ghost state. | |
430 | */ | |
431 | kstat_named_t arcstat_mfu_ghost_evictable_data; | |
432 | /* | |
433 | * Number of bytes that *would have been* consumed by ARC | |
434 | * buffers that are eligible for eviction, of type | |
435 | * ARC_BUFC_METADATA, and linked off the arc_mru_ghost state. | |
436 | */ | |
437 | kstat_named_t arcstat_mfu_ghost_evictable_metadata; | |
34dc7c2f BB |
438 | kstat_named_t arcstat_l2_hits; |
439 | kstat_named_t arcstat_l2_misses; | |
440 | kstat_named_t arcstat_l2_feeds; | |
441 | kstat_named_t arcstat_l2_rw_clash; | |
d164b209 BB |
442 | kstat_named_t arcstat_l2_read_bytes; |
443 | kstat_named_t arcstat_l2_write_bytes; | |
34dc7c2f BB |
444 | kstat_named_t arcstat_l2_writes_sent; |
445 | kstat_named_t arcstat_l2_writes_done; | |
446 | kstat_named_t arcstat_l2_writes_error; | |
ca0bf58d | 447 | kstat_named_t arcstat_l2_writes_lock_retry; |
34dc7c2f BB |
448 | kstat_named_t arcstat_l2_evict_lock_retry; |
449 | kstat_named_t arcstat_l2_evict_reading; | |
b9541d6b | 450 | kstat_named_t arcstat_l2_evict_l1cached; |
34dc7c2f | 451 | kstat_named_t arcstat_l2_free_on_write; |
ca0bf58d | 452 | kstat_named_t arcstat_l2_cdata_free_on_write; |
34dc7c2f BB |
453 | kstat_named_t arcstat_l2_abort_lowmem; |
454 | kstat_named_t arcstat_l2_cksum_bad; | |
455 | kstat_named_t arcstat_l2_io_error; | |
456 | kstat_named_t arcstat_l2_size; | |
3a17a7a9 | 457 | kstat_named_t arcstat_l2_asize; |
34dc7c2f | 458 | kstat_named_t arcstat_l2_hdr_size; |
3a17a7a9 SK |
459 | kstat_named_t arcstat_l2_compress_successes; |
460 | kstat_named_t arcstat_l2_compress_zeros; | |
461 | kstat_named_t arcstat_l2_compress_failures; | |
34dc7c2f | 462 | kstat_named_t arcstat_memory_throttle_count; |
1eb5bfa3 GW |
463 | kstat_named_t arcstat_duplicate_buffers; |
464 | kstat_named_t arcstat_duplicate_buffers_size; | |
465 | kstat_named_t arcstat_duplicate_reads; | |
7cb67b45 BB |
466 | kstat_named_t arcstat_memory_direct_count; |
467 | kstat_named_t arcstat_memory_indirect_count; | |
1834f2d8 BB |
468 | kstat_named_t arcstat_no_grow; |
469 | kstat_named_t arcstat_tempreserve; | |
470 | kstat_named_t arcstat_loaned_bytes; | |
ab26409d | 471 | kstat_named_t arcstat_prune; |
1834f2d8 BB |
472 | kstat_named_t arcstat_meta_used; |
473 | kstat_named_t arcstat_meta_limit; | |
474 | kstat_named_t arcstat_meta_max; | |
ca0bf58d | 475 | kstat_named_t arcstat_meta_min; |
34dc7c2f BB |
476 | } arc_stats_t; |
477 | ||
478 | static arc_stats_t arc_stats = { | |
479 | { "hits", KSTAT_DATA_UINT64 }, | |
480 | { "misses", KSTAT_DATA_UINT64 }, | |
481 | { "demand_data_hits", KSTAT_DATA_UINT64 }, | |
482 | { "demand_data_misses", KSTAT_DATA_UINT64 }, | |
483 | { "demand_metadata_hits", KSTAT_DATA_UINT64 }, | |
484 | { "demand_metadata_misses", KSTAT_DATA_UINT64 }, | |
485 | { "prefetch_data_hits", KSTAT_DATA_UINT64 }, | |
486 | { "prefetch_data_misses", KSTAT_DATA_UINT64 }, | |
487 | { "prefetch_metadata_hits", KSTAT_DATA_UINT64 }, | |
488 | { "prefetch_metadata_misses", KSTAT_DATA_UINT64 }, | |
489 | { "mru_hits", KSTAT_DATA_UINT64 }, | |
490 | { "mru_ghost_hits", KSTAT_DATA_UINT64 }, | |
491 | { "mfu_hits", KSTAT_DATA_UINT64 }, | |
492 | { "mfu_ghost_hits", KSTAT_DATA_UINT64 }, | |
493 | { "deleted", KSTAT_DATA_UINT64 }, | |
34dc7c2f BB |
494 | { "mutex_miss", KSTAT_DATA_UINT64 }, |
495 | { "evict_skip", KSTAT_DATA_UINT64 }, | |
ca0bf58d | 496 | { "evict_not_enough", KSTAT_DATA_UINT64 }, |
428870ff BB |
497 | { "evict_l2_cached", KSTAT_DATA_UINT64 }, |
498 | { "evict_l2_eligible", KSTAT_DATA_UINT64 }, | |
499 | { "evict_l2_ineligible", KSTAT_DATA_UINT64 }, | |
ca0bf58d | 500 | { "evict_l2_skip", KSTAT_DATA_UINT64 }, |
34dc7c2f BB |
501 | { "hash_elements", KSTAT_DATA_UINT64 }, |
502 | { "hash_elements_max", KSTAT_DATA_UINT64 }, | |
503 | { "hash_collisions", KSTAT_DATA_UINT64 }, | |
504 | { "hash_chains", KSTAT_DATA_UINT64 }, | |
505 | { "hash_chain_max", KSTAT_DATA_UINT64 }, | |
506 | { "p", KSTAT_DATA_UINT64 }, | |
507 | { "c", KSTAT_DATA_UINT64 }, | |
508 | { "c_min", KSTAT_DATA_UINT64 }, | |
509 | { "c_max", KSTAT_DATA_UINT64 }, | |
510 | { "size", KSTAT_DATA_UINT64 }, | |
511 | { "hdr_size", KSTAT_DATA_UINT64 }, | |
d164b209 | 512 | { "data_size", KSTAT_DATA_UINT64 }, |
500445c0 | 513 | { "metadata_size", KSTAT_DATA_UINT64 }, |
d164b209 | 514 | { "other_size", KSTAT_DATA_UINT64 }, |
13be560d | 515 | { "anon_size", KSTAT_DATA_UINT64 }, |
500445c0 PS |
516 | { "anon_evictable_data", KSTAT_DATA_UINT64 }, |
517 | { "anon_evictable_metadata", KSTAT_DATA_UINT64 }, | |
13be560d | 518 | { "mru_size", KSTAT_DATA_UINT64 }, |
500445c0 PS |
519 | { "mru_evictable_data", KSTAT_DATA_UINT64 }, |
520 | { "mru_evictable_metadata", KSTAT_DATA_UINT64 }, | |
13be560d | 521 | { "mru_ghost_size", KSTAT_DATA_UINT64 }, |
500445c0 PS |
522 | { "mru_ghost_evictable_data", KSTAT_DATA_UINT64 }, |
523 | { "mru_ghost_evictable_metadata", KSTAT_DATA_UINT64 }, | |
13be560d | 524 | { "mfu_size", KSTAT_DATA_UINT64 }, |
500445c0 PS |
525 | { "mfu_evictable_data", KSTAT_DATA_UINT64 }, |
526 | { "mfu_evictable_metadata", KSTAT_DATA_UINT64 }, | |
13be560d | 527 | { "mfu_ghost_size", KSTAT_DATA_UINT64 }, |
500445c0 PS |
528 | { "mfu_ghost_evictable_data", KSTAT_DATA_UINT64 }, |
529 | { "mfu_ghost_evictable_metadata", KSTAT_DATA_UINT64 }, | |
34dc7c2f BB |
530 | { "l2_hits", KSTAT_DATA_UINT64 }, |
531 | { "l2_misses", KSTAT_DATA_UINT64 }, | |
532 | { "l2_feeds", KSTAT_DATA_UINT64 }, | |
533 | { "l2_rw_clash", KSTAT_DATA_UINT64 }, | |
d164b209 BB |
534 | { "l2_read_bytes", KSTAT_DATA_UINT64 }, |
535 | { "l2_write_bytes", KSTAT_DATA_UINT64 }, | |
34dc7c2f BB |
536 | { "l2_writes_sent", KSTAT_DATA_UINT64 }, |
537 | { "l2_writes_done", KSTAT_DATA_UINT64 }, | |
538 | { "l2_writes_error", KSTAT_DATA_UINT64 }, | |
ca0bf58d | 539 | { "l2_writes_lock_retry", KSTAT_DATA_UINT64 }, |
34dc7c2f BB |
540 | { "l2_evict_lock_retry", KSTAT_DATA_UINT64 }, |
541 | { "l2_evict_reading", KSTAT_DATA_UINT64 }, | |
b9541d6b | 542 | { "l2_evict_l1cached", KSTAT_DATA_UINT64 }, |
34dc7c2f | 543 | { "l2_free_on_write", KSTAT_DATA_UINT64 }, |
ca0bf58d | 544 | { "l2_cdata_free_on_write", KSTAT_DATA_UINT64 }, |
34dc7c2f BB |
545 | { "l2_abort_lowmem", KSTAT_DATA_UINT64 }, |
546 | { "l2_cksum_bad", KSTAT_DATA_UINT64 }, | |
547 | { "l2_io_error", KSTAT_DATA_UINT64 }, | |
548 | { "l2_size", KSTAT_DATA_UINT64 }, | |
3a17a7a9 | 549 | { "l2_asize", KSTAT_DATA_UINT64 }, |
34dc7c2f | 550 | { "l2_hdr_size", KSTAT_DATA_UINT64 }, |
3a17a7a9 SK |
551 | { "l2_compress_successes", KSTAT_DATA_UINT64 }, |
552 | { "l2_compress_zeros", KSTAT_DATA_UINT64 }, | |
553 | { "l2_compress_failures", KSTAT_DATA_UINT64 }, | |
1834f2d8 | 554 | { "memory_throttle_count", KSTAT_DATA_UINT64 }, |
1eb5bfa3 GW |
555 | { "duplicate_buffers", KSTAT_DATA_UINT64 }, |
556 | { "duplicate_buffers_size", KSTAT_DATA_UINT64 }, | |
557 | { "duplicate_reads", KSTAT_DATA_UINT64 }, | |
7cb67b45 BB |
558 | { "memory_direct_count", KSTAT_DATA_UINT64 }, |
559 | { "memory_indirect_count", KSTAT_DATA_UINT64 }, | |
1834f2d8 BB |
560 | { "arc_no_grow", KSTAT_DATA_UINT64 }, |
561 | { "arc_tempreserve", KSTAT_DATA_UINT64 }, | |
562 | { "arc_loaned_bytes", KSTAT_DATA_UINT64 }, | |
ab26409d | 563 | { "arc_prune", KSTAT_DATA_UINT64 }, |
1834f2d8 BB |
564 | { "arc_meta_used", KSTAT_DATA_UINT64 }, |
565 | { "arc_meta_limit", KSTAT_DATA_UINT64 }, | |
566 | { "arc_meta_max", KSTAT_DATA_UINT64 }, | |
500445c0 | 567 | { "arc_meta_min", KSTAT_DATA_UINT64 } |
34dc7c2f BB |
568 | }; |
569 | ||
570 | #define ARCSTAT(stat) (arc_stats.stat.value.ui64) | |
571 | ||
572 | #define ARCSTAT_INCR(stat, val) \ | |
d3cc8b15 | 573 | atomic_add_64(&arc_stats.stat.value.ui64, (val)) |
34dc7c2f | 574 | |
428870ff | 575 | #define ARCSTAT_BUMP(stat) ARCSTAT_INCR(stat, 1) |
34dc7c2f BB |
576 | #define ARCSTAT_BUMPDOWN(stat) ARCSTAT_INCR(stat, -1) |
577 | ||
578 | #define ARCSTAT_MAX(stat, val) { \ | |
579 | uint64_t m; \ | |
580 | while ((val) > (m = arc_stats.stat.value.ui64) && \ | |
581 | (m != atomic_cas_64(&arc_stats.stat.value.ui64, m, (val)))) \ | |
582 | continue; \ | |
583 | } | |
584 | ||
585 | #define ARCSTAT_MAXSTAT(stat) \ | |
586 | ARCSTAT_MAX(stat##_max, arc_stats.stat.value.ui64) | |
587 | ||
588 | /* | |
589 | * We define a macro to allow ARC hits/misses to be easily broken down by | |
590 | * two separate conditions, giving a total of four different subtypes for | |
591 | * each of hits and misses (so eight statistics total). | |
592 | */ | |
593 | #define ARCSTAT_CONDSTAT(cond1, stat1, notstat1, cond2, stat2, notstat2, stat) \ | |
594 | if (cond1) { \ | |
595 | if (cond2) { \ | |
596 | ARCSTAT_BUMP(arcstat_##stat1##_##stat2##_##stat); \ | |
597 | } else { \ | |
598 | ARCSTAT_BUMP(arcstat_##stat1##_##notstat2##_##stat); \ | |
599 | } \ | |
600 | } else { \ | |
601 | if (cond2) { \ | |
602 | ARCSTAT_BUMP(arcstat_##notstat1##_##stat2##_##stat); \ | |
603 | } else { \ | |
604 | ARCSTAT_BUMP(arcstat_##notstat1##_##notstat2##_##stat);\ | |
605 | } \ | |
606 | } | |
607 | ||
608 | kstat_t *arc_ksp; | |
428870ff | 609 | static arc_state_t *arc_anon; |
34dc7c2f BB |
610 | static arc_state_t *arc_mru; |
611 | static arc_state_t *arc_mru_ghost; | |
612 | static arc_state_t *arc_mfu; | |
613 | static arc_state_t *arc_mfu_ghost; | |
614 | static arc_state_t *arc_l2c_only; | |
615 | ||
616 | /* | |
617 | * There are several ARC variables that are critical to export as kstats -- | |
618 | * but we don't want to have to grovel around in the kstat whenever we wish to | |
619 | * manipulate them. For these variables, we therefore define them to be in | |
620 | * terms of the statistic variable. This assures that we are not introducing | |
621 | * the possibility of inconsistency by having shadow copies of the variables, | |
622 | * while still allowing the code to be readable. | |
623 | */ | |
624 | #define arc_size ARCSTAT(arcstat_size) /* actual total arc size */ | |
625 | #define arc_p ARCSTAT(arcstat_p) /* target size of MRU */ | |
626 | #define arc_c ARCSTAT(arcstat_c) /* target size of cache */ | |
627 | #define arc_c_min ARCSTAT(arcstat_c_min) /* min target cache size */ | |
628 | #define arc_c_max ARCSTAT(arcstat_c_max) /* max target cache size */ | |
1834f2d8 BB |
629 | #define arc_no_grow ARCSTAT(arcstat_no_grow) |
630 | #define arc_tempreserve ARCSTAT(arcstat_tempreserve) | |
631 | #define arc_loaned_bytes ARCSTAT(arcstat_loaned_bytes) | |
23c0a133 | 632 | #define arc_meta_limit ARCSTAT(arcstat_meta_limit) /* max size for metadata */ |
ca0bf58d | 633 | #define arc_meta_min ARCSTAT(arcstat_meta_min) /* min size for metadata */ |
23c0a133 GW |
634 | #define arc_meta_used ARCSTAT(arcstat_meta_used) /* size of metadata */ |
635 | #define arc_meta_max ARCSTAT(arcstat_meta_max) /* max size of metadata */ | |
34dc7c2f | 636 | |
3a17a7a9 SK |
637 | #define L2ARC_IS_VALID_COMPRESS(_c_) \ |
638 | ((_c_) == ZIO_COMPRESS_LZ4 || (_c_) == ZIO_COMPRESS_EMPTY) | |
639 | ||
ab26409d BB |
640 | static list_t arc_prune_list; |
641 | static kmutex_t arc_prune_mtx; | |
f6046738 | 642 | static taskq_t *arc_prune_taskq; |
34dc7c2f | 643 | static arc_buf_t *arc_eviction_list; |
34dc7c2f | 644 | static arc_buf_hdr_t arc_eviction_hdr; |
428870ff | 645 | |
34dc7c2f BB |
646 | #define GHOST_STATE(state) \ |
647 | ((state) == arc_mru_ghost || (state) == arc_mfu_ghost || \ | |
648 | (state) == arc_l2c_only) | |
649 | ||
2a432414 GW |
650 | #define HDR_IN_HASH_TABLE(hdr) ((hdr)->b_flags & ARC_FLAG_IN_HASH_TABLE) |
651 | #define HDR_IO_IN_PROGRESS(hdr) ((hdr)->b_flags & ARC_FLAG_IO_IN_PROGRESS) | |
652 | #define HDR_IO_ERROR(hdr) ((hdr)->b_flags & ARC_FLAG_IO_ERROR) | |
653 | #define HDR_PREFETCH(hdr) ((hdr)->b_flags & ARC_FLAG_PREFETCH) | |
654 | #define HDR_FREED_IN_READ(hdr) ((hdr)->b_flags & ARC_FLAG_FREED_IN_READ) | |
655 | #define HDR_BUF_AVAILABLE(hdr) ((hdr)->b_flags & ARC_FLAG_BUF_AVAILABLE) | |
b9541d6b | 656 | |
2a432414 | 657 | #define HDR_L2CACHE(hdr) ((hdr)->b_flags & ARC_FLAG_L2CACHE) |
b9541d6b | 658 | #define HDR_L2COMPRESS(hdr) ((hdr)->b_flags & ARC_FLAG_L2COMPRESS) |
2a432414 | 659 | #define HDR_L2_READING(hdr) \ |
b9541d6b CW |
660 | (((hdr)->b_flags & ARC_FLAG_IO_IN_PROGRESS) && \ |
661 | ((hdr)->b_flags & ARC_FLAG_HAS_L2HDR)) | |
2a432414 GW |
662 | #define HDR_L2_WRITING(hdr) ((hdr)->b_flags & ARC_FLAG_L2_WRITING) |
663 | #define HDR_L2_EVICTED(hdr) ((hdr)->b_flags & ARC_FLAG_L2_EVICTED) | |
664 | #define HDR_L2_WRITE_HEAD(hdr) ((hdr)->b_flags & ARC_FLAG_L2_WRITE_HEAD) | |
34dc7c2f | 665 | |
b9541d6b CW |
666 | #define HDR_ISTYPE_METADATA(hdr) \ |
667 | ((hdr)->b_flags & ARC_FLAG_BUFC_METADATA) | |
668 | #define HDR_ISTYPE_DATA(hdr) (!HDR_ISTYPE_METADATA(hdr)) | |
669 | ||
670 | #define HDR_HAS_L1HDR(hdr) ((hdr)->b_flags & ARC_FLAG_HAS_L1HDR) | |
671 | #define HDR_HAS_L2HDR(hdr) ((hdr)->b_flags & ARC_FLAG_HAS_L2HDR) | |
672 | ||
673 | /* For storing compression mode in b_flags */ | |
674 | #define HDR_COMPRESS_OFFSET 24 | |
675 | #define HDR_COMPRESS_NBITS 7 | |
676 | ||
677 | #define HDR_GET_COMPRESS(hdr) ((enum zio_compress)BF32_GET(hdr->b_flags, \ | |
678 | HDR_COMPRESS_OFFSET, HDR_COMPRESS_NBITS)) | |
679 | #define HDR_SET_COMPRESS(hdr, cmp) BF32_SET(hdr->b_flags, \ | |
680 | HDR_COMPRESS_OFFSET, HDR_COMPRESS_NBITS, (cmp)) | |
681 | ||
34dc7c2f BB |
682 | /* |
683 | * Other sizes | |
684 | */ | |
685 | ||
b9541d6b CW |
686 | #define HDR_FULL_SIZE ((int64_t)sizeof (arc_buf_hdr_t)) |
687 | #define HDR_L2ONLY_SIZE ((int64_t)offsetof(arc_buf_hdr_t, b_l1hdr)) | |
34dc7c2f BB |
688 | |
689 | /* | |
690 | * Hash table routines | |
691 | */ | |
692 | ||
00b46022 BB |
693 | #define HT_LOCK_ALIGN 64 |
694 | #define HT_LOCK_PAD (P2NPHASE(sizeof (kmutex_t), (HT_LOCK_ALIGN))) | |
34dc7c2f BB |
695 | |
696 | struct ht_lock { | |
697 | kmutex_t ht_lock; | |
698 | #ifdef _KERNEL | |
00b46022 | 699 | unsigned char pad[HT_LOCK_PAD]; |
34dc7c2f BB |
700 | #endif |
701 | }; | |
702 | ||
b31d8ea7 | 703 | #define BUF_LOCKS 8192 |
34dc7c2f BB |
704 | typedef struct buf_hash_table { |
705 | uint64_t ht_mask; | |
706 | arc_buf_hdr_t **ht_table; | |
707 | struct ht_lock ht_locks[BUF_LOCKS]; | |
708 | } buf_hash_table_t; | |
709 | ||
710 | static buf_hash_table_t buf_hash_table; | |
711 | ||
712 | #define BUF_HASH_INDEX(spa, dva, birth) \ | |
713 | (buf_hash(spa, dva, birth) & buf_hash_table.ht_mask) | |
714 | #define BUF_HASH_LOCK_NTRY(idx) (buf_hash_table.ht_locks[idx & (BUF_LOCKS-1)]) | |
715 | #define BUF_HASH_LOCK(idx) (&(BUF_HASH_LOCK_NTRY(idx).ht_lock)) | |
428870ff BB |
716 | #define HDR_LOCK(hdr) \ |
717 | (BUF_HASH_LOCK(BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth))) | |
34dc7c2f BB |
718 | |
719 | uint64_t zfs_crc64_table[256]; | |
720 | ||
721 | /* | |
722 | * Level 2 ARC | |
723 | */ | |
724 | ||
725 | #define L2ARC_WRITE_SIZE (8 * 1024 * 1024) /* initial write max */ | |
3a17a7a9 SK |
726 | #define L2ARC_HEADROOM 2 /* num of writes */ |
727 | /* | |
728 | * If we discover during ARC scan any buffers to be compressed, we boost | |
729 | * our headroom for the next scanning cycle by this percentage multiple. | |
730 | */ | |
731 | #define L2ARC_HEADROOM_BOOST 200 | |
d164b209 BB |
732 | #define L2ARC_FEED_SECS 1 /* caching interval secs */ |
733 | #define L2ARC_FEED_MIN_MS 200 /* min caching interval ms */ | |
34dc7c2f | 734 | |
d962d5da PS |
735 | /* |
736 | * Used to distinguish headers that are being process by | |
737 | * l2arc_write_buffers(), but have yet to be assigned to a l2arc disk | |
738 | * address. This can happen when the header is added to the l2arc's list | |
739 | * of buffers to write in the first stage of l2arc_write_buffers(), but | |
740 | * has not yet been written out which happens in the second stage of | |
741 | * l2arc_write_buffers(). | |
742 | */ | |
743 | #define L2ARC_ADDR_UNSET ((uint64_t)(-1)) | |
744 | ||
34dc7c2f BB |
745 | #define l2arc_writes_sent ARCSTAT(arcstat_l2_writes_sent) |
746 | #define l2arc_writes_done ARCSTAT(arcstat_l2_writes_done) | |
747 | ||
d3cc8b15 | 748 | /* L2ARC Performance Tunables */ |
abd8610c BB |
749 | unsigned long l2arc_write_max = L2ARC_WRITE_SIZE; /* def max write size */ |
750 | unsigned long l2arc_write_boost = L2ARC_WRITE_SIZE; /* extra warmup write */ | |
751 | unsigned long l2arc_headroom = L2ARC_HEADROOM; /* # of dev writes */ | |
3a17a7a9 | 752 | unsigned long l2arc_headroom_boost = L2ARC_HEADROOM_BOOST; |
abd8610c BB |
753 | unsigned long l2arc_feed_secs = L2ARC_FEED_SECS; /* interval seconds */ |
754 | unsigned long l2arc_feed_min_ms = L2ARC_FEED_MIN_MS; /* min interval msecs */ | |
755 | int l2arc_noprefetch = B_TRUE; /* don't cache prefetch bufs */ | |
3a17a7a9 | 756 | int l2arc_nocompress = B_FALSE; /* don't compress bufs */ |
abd8610c | 757 | int l2arc_feed_again = B_TRUE; /* turbo warmup */ |
c93504f0 | 758 | int l2arc_norw = B_FALSE; /* no reads during writes */ |
34dc7c2f BB |
759 | |
760 | /* | |
761 | * L2ARC Internals | |
762 | */ | |
34dc7c2f BB |
763 | static list_t L2ARC_dev_list; /* device list */ |
764 | static list_t *l2arc_dev_list; /* device list pointer */ | |
765 | static kmutex_t l2arc_dev_mtx; /* device list mutex */ | |
766 | static l2arc_dev_t *l2arc_dev_last; /* last device used */ | |
34dc7c2f BB |
767 | static list_t L2ARC_free_on_write; /* free after write buf list */ |
768 | static list_t *l2arc_free_on_write; /* free after write list ptr */ | |
769 | static kmutex_t l2arc_free_on_write_mtx; /* mutex for list */ | |
770 | static uint64_t l2arc_ndev; /* number of devices */ | |
771 | ||
772 | typedef struct l2arc_read_callback { | |
3a17a7a9 SK |
773 | arc_buf_t *l2rcb_buf; /* read buffer */ |
774 | spa_t *l2rcb_spa; /* spa */ | |
775 | blkptr_t l2rcb_bp; /* original blkptr */ | |
5dbd68a3 | 776 | zbookmark_phys_t l2rcb_zb; /* original bookmark */ |
3a17a7a9 SK |
777 | int l2rcb_flags; /* original flags */ |
778 | enum zio_compress l2rcb_compress; /* applied compress */ | |
34dc7c2f BB |
779 | } l2arc_read_callback_t; |
780 | ||
34dc7c2f BB |
781 | typedef struct l2arc_data_free { |
782 | /* protected by l2arc_free_on_write_mtx */ | |
783 | void *l2df_data; | |
784 | size_t l2df_size; | |
785 | void (*l2df_func)(void *, size_t); | |
786 | list_node_t l2df_list_node; | |
787 | } l2arc_data_free_t; | |
788 | ||
789 | static kmutex_t l2arc_feed_thr_lock; | |
790 | static kcondvar_t l2arc_feed_thr_cv; | |
791 | static uint8_t l2arc_thread_exit; | |
792 | ||
2a432414 GW |
793 | static void arc_get_data_buf(arc_buf_t *); |
794 | static void arc_access(arc_buf_hdr_t *, kmutex_t *); | |
ca0bf58d | 795 | static boolean_t arc_is_overflowing(void); |
2a432414 | 796 | static void arc_buf_watch(arc_buf_t *); |
ca67b33a | 797 | static void arc_tuning_update(void); |
2a432414 | 798 | |
b9541d6b CW |
799 | static arc_buf_contents_t arc_buf_type(arc_buf_hdr_t *); |
800 | static uint32_t arc_bufc_to_flags(arc_buf_contents_t); | |
801 | ||
2a432414 GW |
802 | static boolean_t l2arc_write_eligible(uint64_t, arc_buf_hdr_t *); |
803 | static void l2arc_read_done(zio_t *); | |
34dc7c2f | 804 | |
b9541d6b | 805 | static boolean_t l2arc_compress_buf(arc_buf_hdr_t *); |
2a432414 GW |
806 | static void l2arc_decompress_zio(zio_t *, arc_buf_hdr_t *, enum zio_compress); |
807 | static void l2arc_release_cdata_buf(arc_buf_hdr_t *); | |
3a17a7a9 | 808 | |
34dc7c2f | 809 | static uint64_t |
d164b209 | 810 | buf_hash(uint64_t spa, const dva_t *dva, uint64_t birth) |
34dc7c2f | 811 | { |
34dc7c2f BB |
812 | uint8_t *vdva = (uint8_t *)dva; |
813 | uint64_t crc = -1ULL; | |
814 | int i; | |
815 | ||
816 | ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); | |
817 | ||
818 | for (i = 0; i < sizeof (dva_t); i++) | |
819 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ vdva[i]) & 0xFF]; | |
820 | ||
d164b209 | 821 | crc ^= (spa>>8) ^ birth; |
34dc7c2f BB |
822 | |
823 | return (crc); | |
824 | } | |
825 | ||
826 | #define BUF_EMPTY(buf) \ | |
827 | ((buf)->b_dva.dva_word[0] == 0 && \ | |
b9541d6b | 828 | (buf)->b_dva.dva_word[1] == 0) |
34dc7c2f BB |
829 | |
830 | #define BUF_EQUAL(spa, dva, birth, buf) \ | |
831 | ((buf)->b_dva.dva_word[0] == (dva)->dva_word[0]) && \ | |
832 | ((buf)->b_dva.dva_word[1] == (dva)->dva_word[1]) && \ | |
833 | ((buf)->b_birth == birth) && ((buf)->b_spa == spa) | |
834 | ||
428870ff BB |
835 | static void |
836 | buf_discard_identity(arc_buf_hdr_t *hdr) | |
837 | { | |
838 | hdr->b_dva.dva_word[0] = 0; | |
839 | hdr->b_dva.dva_word[1] = 0; | |
840 | hdr->b_birth = 0; | |
428870ff BB |
841 | } |
842 | ||
34dc7c2f | 843 | static arc_buf_hdr_t * |
9b67f605 | 844 | buf_hash_find(uint64_t spa, const blkptr_t *bp, kmutex_t **lockp) |
34dc7c2f | 845 | { |
9b67f605 MA |
846 | const dva_t *dva = BP_IDENTITY(bp); |
847 | uint64_t birth = BP_PHYSICAL_BIRTH(bp); | |
34dc7c2f BB |
848 | uint64_t idx = BUF_HASH_INDEX(spa, dva, birth); |
849 | kmutex_t *hash_lock = BUF_HASH_LOCK(idx); | |
2a432414 | 850 | arc_buf_hdr_t *hdr; |
34dc7c2f BB |
851 | |
852 | mutex_enter(hash_lock); | |
2a432414 GW |
853 | for (hdr = buf_hash_table.ht_table[idx]; hdr != NULL; |
854 | hdr = hdr->b_hash_next) { | |
855 | if (BUF_EQUAL(spa, dva, birth, hdr)) { | |
34dc7c2f | 856 | *lockp = hash_lock; |
2a432414 | 857 | return (hdr); |
34dc7c2f BB |
858 | } |
859 | } | |
860 | mutex_exit(hash_lock); | |
861 | *lockp = NULL; | |
862 | return (NULL); | |
863 | } | |
864 | ||
865 | /* | |
866 | * Insert an entry into the hash table. If there is already an element | |
867 | * equal to elem in the hash table, then the already existing element | |
868 | * will be returned and the new element will not be inserted. | |
869 | * Otherwise returns NULL. | |
b9541d6b | 870 | * If lockp == NULL, the caller is assumed to already hold the hash lock. |
34dc7c2f BB |
871 | */ |
872 | static arc_buf_hdr_t * | |
2a432414 | 873 | buf_hash_insert(arc_buf_hdr_t *hdr, kmutex_t **lockp) |
34dc7c2f | 874 | { |
2a432414 | 875 | uint64_t idx = BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth); |
34dc7c2f | 876 | kmutex_t *hash_lock = BUF_HASH_LOCK(idx); |
2a432414 | 877 | arc_buf_hdr_t *fhdr; |
34dc7c2f BB |
878 | uint32_t i; |
879 | ||
2a432414 GW |
880 | ASSERT(!DVA_IS_EMPTY(&hdr->b_dva)); |
881 | ASSERT(hdr->b_birth != 0); | |
882 | ASSERT(!HDR_IN_HASH_TABLE(hdr)); | |
b9541d6b CW |
883 | |
884 | if (lockp != NULL) { | |
885 | *lockp = hash_lock; | |
886 | mutex_enter(hash_lock); | |
887 | } else { | |
888 | ASSERT(MUTEX_HELD(hash_lock)); | |
889 | } | |
890 | ||
2a432414 GW |
891 | for (fhdr = buf_hash_table.ht_table[idx], i = 0; fhdr != NULL; |
892 | fhdr = fhdr->b_hash_next, i++) { | |
893 | if (BUF_EQUAL(hdr->b_spa, &hdr->b_dva, hdr->b_birth, fhdr)) | |
894 | return (fhdr); | |
34dc7c2f BB |
895 | } |
896 | ||
2a432414 GW |
897 | hdr->b_hash_next = buf_hash_table.ht_table[idx]; |
898 | buf_hash_table.ht_table[idx] = hdr; | |
899 | hdr->b_flags |= ARC_FLAG_IN_HASH_TABLE; | |
34dc7c2f BB |
900 | |
901 | /* collect some hash table performance data */ | |
902 | if (i > 0) { | |
903 | ARCSTAT_BUMP(arcstat_hash_collisions); | |
904 | if (i == 1) | |
905 | ARCSTAT_BUMP(arcstat_hash_chains); | |
906 | ||
907 | ARCSTAT_MAX(arcstat_hash_chain_max, i); | |
908 | } | |
909 | ||
910 | ARCSTAT_BUMP(arcstat_hash_elements); | |
911 | ARCSTAT_MAXSTAT(arcstat_hash_elements); | |
912 | ||
913 | return (NULL); | |
914 | } | |
915 | ||
916 | static void | |
2a432414 | 917 | buf_hash_remove(arc_buf_hdr_t *hdr) |
34dc7c2f | 918 | { |
2a432414 GW |
919 | arc_buf_hdr_t *fhdr, **hdrp; |
920 | uint64_t idx = BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth); | |
34dc7c2f BB |
921 | |
922 | ASSERT(MUTEX_HELD(BUF_HASH_LOCK(idx))); | |
2a432414 | 923 | ASSERT(HDR_IN_HASH_TABLE(hdr)); |
34dc7c2f | 924 | |
2a432414 GW |
925 | hdrp = &buf_hash_table.ht_table[idx]; |
926 | while ((fhdr = *hdrp) != hdr) { | |
927 | ASSERT(fhdr != NULL); | |
928 | hdrp = &fhdr->b_hash_next; | |
34dc7c2f | 929 | } |
2a432414 GW |
930 | *hdrp = hdr->b_hash_next; |
931 | hdr->b_hash_next = NULL; | |
932 | hdr->b_flags &= ~ARC_FLAG_IN_HASH_TABLE; | |
34dc7c2f BB |
933 | |
934 | /* collect some hash table performance data */ | |
935 | ARCSTAT_BUMPDOWN(arcstat_hash_elements); | |
936 | ||
937 | if (buf_hash_table.ht_table[idx] && | |
938 | buf_hash_table.ht_table[idx]->b_hash_next == NULL) | |
939 | ARCSTAT_BUMPDOWN(arcstat_hash_chains); | |
940 | } | |
941 | ||
942 | /* | |
943 | * Global data structures and functions for the buf kmem cache. | |
944 | */ | |
b9541d6b CW |
945 | static kmem_cache_t *hdr_full_cache; |
946 | static kmem_cache_t *hdr_l2only_cache; | |
34dc7c2f BB |
947 | static kmem_cache_t *buf_cache; |
948 | ||
949 | static void | |
950 | buf_fini(void) | |
951 | { | |
952 | int i; | |
953 | ||
00b46022 | 954 | #if defined(_KERNEL) && defined(HAVE_SPL) |
d1d7e268 MK |
955 | /* |
956 | * Large allocations which do not require contiguous pages | |
957 | * should be using vmem_free() in the linux kernel\ | |
958 | */ | |
00b46022 BB |
959 | vmem_free(buf_hash_table.ht_table, |
960 | (buf_hash_table.ht_mask + 1) * sizeof (void *)); | |
961 | #else | |
34dc7c2f BB |
962 | kmem_free(buf_hash_table.ht_table, |
963 | (buf_hash_table.ht_mask + 1) * sizeof (void *)); | |
00b46022 | 964 | #endif |
34dc7c2f BB |
965 | for (i = 0; i < BUF_LOCKS; i++) |
966 | mutex_destroy(&buf_hash_table.ht_locks[i].ht_lock); | |
b9541d6b CW |
967 | kmem_cache_destroy(hdr_full_cache); |
968 | kmem_cache_destroy(hdr_l2only_cache); | |
34dc7c2f BB |
969 | kmem_cache_destroy(buf_cache); |
970 | } | |
971 | ||
972 | /* | |
973 | * Constructor callback - called when the cache is empty | |
974 | * and a new buf is requested. | |
975 | */ | |
976 | /* ARGSUSED */ | |
977 | static int | |
b9541d6b CW |
978 | hdr_full_cons(void *vbuf, void *unused, int kmflag) |
979 | { | |
980 | arc_buf_hdr_t *hdr = vbuf; | |
981 | ||
982 | bzero(hdr, HDR_FULL_SIZE); | |
983 | cv_init(&hdr->b_l1hdr.b_cv, NULL, CV_DEFAULT, NULL); | |
984 | refcount_create(&hdr->b_l1hdr.b_refcnt); | |
985 | mutex_init(&hdr->b_l1hdr.b_freeze_lock, NULL, MUTEX_DEFAULT, NULL); | |
986 | list_link_init(&hdr->b_l1hdr.b_arc_node); | |
987 | list_link_init(&hdr->b_l2hdr.b_l2node); | |
ca0bf58d | 988 | multilist_link_init(&hdr->b_l1hdr.b_arc_node); |
b9541d6b CW |
989 | arc_space_consume(HDR_FULL_SIZE, ARC_SPACE_HDRS); |
990 | ||
991 | return (0); | |
992 | } | |
993 | ||
994 | /* ARGSUSED */ | |
995 | static int | |
996 | hdr_l2only_cons(void *vbuf, void *unused, int kmflag) | |
34dc7c2f | 997 | { |
2a432414 GW |
998 | arc_buf_hdr_t *hdr = vbuf; |
999 | ||
b9541d6b CW |
1000 | bzero(hdr, HDR_L2ONLY_SIZE); |
1001 | arc_space_consume(HDR_L2ONLY_SIZE, ARC_SPACE_L2HDRS); | |
34dc7c2f | 1002 | |
34dc7c2f BB |
1003 | return (0); |
1004 | } | |
1005 | ||
b128c09f BB |
1006 | /* ARGSUSED */ |
1007 | static int | |
1008 | buf_cons(void *vbuf, void *unused, int kmflag) | |
1009 | { | |
1010 | arc_buf_t *buf = vbuf; | |
1011 | ||
1012 | bzero(buf, sizeof (arc_buf_t)); | |
428870ff | 1013 | mutex_init(&buf->b_evict_lock, NULL, MUTEX_DEFAULT, NULL); |
d164b209 BB |
1014 | arc_space_consume(sizeof (arc_buf_t), ARC_SPACE_HDRS); |
1015 | ||
b128c09f BB |
1016 | return (0); |
1017 | } | |
1018 | ||
34dc7c2f BB |
1019 | /* |
1020 | * Destructor callback - called when a cached buf is | |
1021 | * no longer required. | |
1022 | */ | |
1023 | /* ARGSUSED */ | |
1024 | static void | |
b9541d6b | 1025 | hdr_full_dest(void *vbuf, void *unused) |
34dc7c2f | 1026 | { |
2a432414 | 1027 | arc_buf_hdr_t *hdr = vbuf; |
34dc7c2f | 1028 | |
2a432414 | 1029 | ASSERT(BUF_EMPTY(hdr)); |
b9541d6b CW |
1030 | cv_destroy(&hdr->b_l1hdr.b_cv); |
1031 | refcount_destroy(&hdr->b_l1hdr.b_refcnt); | |
1032 | mutex_destroy(&hdr->b_l1hdr.b_freeze_lock); | |
ca0bf58d | 1033 | ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); |
b9541d6b CW |
1034 | arc_space_return(HDR_FULL_SIZE, ARC_SPACE_HDRS); |
1035 | } | |
1036 | ||
1037 | /* ARGSUSED */ | |
1038 | static void | |
1039 | hdr_l2only_dest(void *vbuf, void *unused) | |
1040 | { | |
1041 | ASSERTV(arc_buf_hdr_t *hdr = vbuf); | |
1042 | ||
1043 | ASSERT(BUF_EMPTY(hdr)); | |
1044 | arc_space_return(HDR_L2ONLY_SIZE, ARC_SPACE_L2HDRS); | |
34dc7c2f BB |
1045 | } |
1046 | ||
b128c09f BB |
1047 | /* ARGSUSED */ |
1048 | static void | |
1049 | buf_dest(void *vbuf, void *unused) | |
1050 | { | |
1051 | arc_buf_t *buf = vbuf; | |
1052 | ||
428870ff | 1053 | mutex_destroy(&buf->b_evict_lock); |
d164b209 | 1054 | arc_space_return(sizeof (arc_buf_t), ARC_SPACE_HDRS); |
b128c09f BB |
1055 | } |
1056 | ||
8c8af9d8 BB |
1057 | /* |
1058 | * Reclaim callback -- invoked when memory is low. | |
1059 | */ | |
1060 | /* ARGSUSED */ | |
1061 | static void | |
1062 | hdr_recl(void *unused) | |
1063 | { | |
1064 | dprintf("hdr_recl called\n"); | |
1065 | /* | |
1066 | * umem calls the reclaim func when we destroy the buf cache, | |
1067 | * which is after we do arc_fini(). | |
1068 | */ | |
1069 | if (!arc_dead) | |
1070 | cv_signal(&arc_reclaim_thread_cv); | |
1071 | } | |
1072 | ||
34dc7c2f BB |
1073 | static void |
1074 | buf_init(void) | |
1075 | { | |
1076 | uint64_t *ct; | |
1077 | uint64_t hsize = 1ULL << 12; | |
1078 | int i, j; | |
1079 | ||
1080 | /* | |
1081 | * The hash table is big enough to fill all of physical memory | |
49ddb315 MA |
1082 | * with an average block size of zfs_arc_average_blocksize (default 8K). |
1083 | * By default, the table will take up | |
1084 | * totalmem * sizeof(void*) / 8K (1MB per GB with 8-byte pointers). | |
34dc7c2f | 1085 | */ |
49ddb315 | 1086 | while (hsize * zfs_arc_average_blocksize < physmem * PAGESIZE) |
34dc7c2f BB |
1087 | hsize <<= 1; |
1088 | retry: | |
1089 | buf_hash_table.ht_mask = hsize - 1; | |
00b46022 | 1090 | #if defined(_KERNEL) && defined(HAVE_SPL) |
d1d7e268 MK |
1091 | /* |
1092 | * Large allocations which do not require contiguous pages | |
1093 | * should be using vmem_alloc() in the linux kernel | |
1094 | */ | |
00b46022 BB |
1095 | buf_hash_table.ht_table = |
1096 | vmem_zalloc(hsize * sizeof (void*), KM_SLEEP); | |
1097 | #else | |
34dc7c2f BB |
1098 | buf_hash_table.ht_table = |
1099 | kmem_zalloc(hsize * sizeof (void*), KM_NOSLEEP); | |
00b46022 | 1100 | #endif |
34dc7c2f BB |
1101 | if (buf_hash_table.ht_table == NULL) { |
1102 | ASSERT(hsize > (1ULL << 8)); | |
1103 | hsize >>= 1; | |
1104 | goto retry; | |
1105 | } | |
1106 | ||
b9541d6b | 1107 | hdr_full_cache = kmem_cache_create("arc_buf_hdr_t_full", HDR_FULL_SIZE, |
8c8af9d8 | 1108 | 0, hdr_full_cons, hdr_full_dest, hdr_recl, NULL, NULL, 0); |
b9541d6b | 1109 | hdr_l2only_cache = kmem_cache_create("arc_buf_hdr_t_l2only", |
8c8af9d8 | 1110 | HDR_L2ONLY_SIZE, 0, hdr_l2only_cons, hdr_l2only_dest, hdr_recl, |
b9541d6b | 1111 | NULL, NULL, 0); |
34dc7c2f | 1112 | buf_cache = kmem_cache_create("arc_buf_t", sizeof (arc_buf_t), |
b128c09f | 1113 | 0, buf_cons, buf_dest, NULL, NULL, NULL, 0); |
34dc7c2f BB |
1114 | |
1115 | for (i = 0; i < 256; i++) | |
1116 | for (ct = zfs_crc64_table + i, *ct = i, j = 8; j > 0; j--) | |
1117 | *ct = (*ct >> 1) ^ (-(*ct & 1) & ZFS_CRC64_POLY); | |
1118 | ||
1119 | for (i = 0; i < BUF_LOCKS; i++) { | |
1120 | mutex_init(&buf_hash_table.ht_locks[i].ht_lock, | |
40d06e3c | 1121 | NULL, MUTEX_DEFAULT, NULL); |
34dc7c2f BB |
1122 | } |
1123 | } | |
1124 | ||
b9541d6b CW |
1125 | /* |
1126 | * Transition between the two allocation states for the arc_buf_hdr struct. | |
1127 | * The arc_buf_hdr struct can be allocated with (hdr_full_cache) or without | |
1128 | * (hdr_l2only_cache) the fields necessary for the L1 cache - the smaller | |
1129 | * version is used when a cache buffer is only in the L2ARC in order to reduce | |
1130 | * memory usage. | |
1131 | */ | |
1132 | static arc_buf_hdr_t * | |
1133 | arc_hdr_realloc(arc_buf_hdr_t *hdr, kmem_cache_t *old, kmem_cache_t *new) | |
1134 | { | |
1135 | arc_buf_hdr_t *nhdr; | |
1136 | l2arc_dev_t *dev; | |
1137 | ||
1138 | ASSERT(HDR_HAS_L2HDR(hdr)); | |
1139 | ASSERT((old == hdr_full_cache && new == hdr_l2only_cache) || | |
1140 | (old == hdr_l2only_cache && new == hdr_full_cache)); | |
1141 | ||
1142 | dev = hdr->b_l2hdr.b_dev; | |
1143 | nhdr = kmem_cache_alloc(new, KM_PUSHPAGE); | |
1144 | ||
1145 | ASSERT(MUTEX_HELD(HDR_LOCK(hdr))); | |
1146 | buf_hash_remove(hdr); | |
1147 | ||
1148 | bcopy(hdr, nhdr, HDR_L2ONLY_SIZE); | |
d962d5da | 1149 | |
b9541d6b CW |
1150 | if (new == hdr_full_cache) { |
1151 | nhdr->b_flags |= ARC_FLAG_HAS_L1HDR; | |
1152 | /* | |
1153 | * arc_access and arc_change_state need to be aware that a | |
1154 | * header has just come out of L2ARC, so we set its state to | |
1155 | * l2c_only even though it's about to change. | |
1156 | */ | |
1157 | nhdr->b_l1hdr.b_state = arc_l2c_only; | |
ca0bf58d PS |
1158 | |
1159 | /* Verify previous threads set to NULL before freeing */ | |
1160 | ASSERT3P(nhdr->b_l1hdr.b_tmp_cdata, ==, NULL); | |
b9541d6b CW |
1161 | } else { |
1162 | ASSERT(hdr->b_l1hdr.b_buf == NULL); | |
1163 | ASSERT0(hdr->b_l1hdr.b_datacnt); | |
ca0bf58d PS |
1164 | |
1165 | /* | |
1166 | * If we've reached here, We must have been called from | |
1167 | * arc_evict_hdr(), as such we should have already been | |
1168 | * removed from any ghost list we were previously on | |
1169 | * (which protects us from racing with arc_evict_state), | |
1170 | * thus no locking is needed during this check. | |
1171 | */ | |
1172 | ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); | |
1173 | ||
b9541d6b | 1174 | /* |
ca0bf58d PS |
1175 | * A buffer must not be moved into the arc_l2c_only |
1176 | * state if it's not finished being written out to the | |
1177 | * l2arc device. Otherwise, the b_l1hdr.b_tmp_cdata field | |
1178 | * might try to be accessed, even though it was removed. | |
b9541d6b | 1179 | */ |
ca0bf58d PS |
1180 | VERIFY(!HDR_L2_WRITING(hdr)); |
1181 | VERIFY3P(hdr->b_l1hdr.b_tmp_cdata, ==, NULL); | |
1182 | ||
b9541d6b CW |
1183 | nhdr->b_flags &= ~ARC_FLAG_HAS_L1HDR; |
1184 | } | |
1185 | /* | |
1186 | * The header has been reallocated so we need to re-insert it into any | |
1187 | * lists it was on. | |
1188 | */ | |
1189 | (void) buf_hash_insert(nhdr, NULL); | |
1190 | ||
1191 | ASSERT(list_link_active(&hdr->b_l2hdr.b_l2node)); | |
1192 | ||
1193 | mutex_enter(&dev->l2ad_mtx); | |
1194 | ||
1195 | /* | |
1196 | * We must place the realloc'ed header back into the list at | |
1197 | * the same spot. Otherwise, if it's placed earlier in the list, | |
1198 | * l2arc_write_buffers() could find it during the function's | |
1199 | * write phase, and try to write it out to the l2arc. | |
1200 | */ | |
1201 | list_insert_after(&dev->l2ad_buflist, hdr, nhdr); | |
1202 | list_remove(&dev->l2ad_buflist, hdr); | |
1203 | ||
1204 | mutex_exit(&dev->l2ad_mtx); | |
1205 | ||
d962d5da PS |
1206 | /* |
1207 | * Since we're using the pointer address as the tag when | |
1208 | * incrementing and decrementing the l2ad_alloc refcount, we | |
1209 | * must remove the old pointer (that we're about to destroy) and | |
1210 | * add the new pointer to the refcount. Otherwise we'd remove | |
1211 | * the wrong pointer address when calling arc_hdr_destroy() later. | |
1212 | */ | |
1213 | ||
1214 | (void) refcount_remove_many(&dev->l2ad_alloc, | |
1215 | hdr->b_l2hdr.b_asize, hdr); | |
1216 | ||
1217 | (void) refcount_add_many(&dev->l2ad_alloc, | |
1218 | nhdr->b_l2hdr.b_asize, nhdr); | |
1219 | ||
b9541d6b CW |
1220 | buf_discard_identity(hdr); |
1221 | hdr->b_freeze_cksum = NULL; | |
1222 | kmem_cache_free(old, hdr); | |
1223 | ||
1224 | return (nhdr); | |
1225 | } | |
1226 | ||
1227 | ||
34dc7c2f BB |
1228 | #define ARC_MINTIME (hz>>4) /* 62 ms */ |
1229 | ||
1230 | static void | |
1231 | arc_cksum_verify(arc_buf_t *buf) | |
1232 | { | |
1233 | zio_cksum_t zc; | |
1234 | ||
1235 | if (!(zfs_flags & ZFS_DEBUG_MODIFY)) | |
1236 | return; | |
1237 | ||
b9541d6b CW |
1238 | mutex_enter(&buf->b_hdr->b_l1hdr.b_freeze_lock); |
1239 | if (buf->b_hdr->b_freeze_cksum == NULL || HDR_IO_ERROR(buf->b_hdr)) { | |
1240 | mutex_exit(&buf->b_hdr->b_l1hdr.b_freeze_lock); | |
34dc7c2f BB |
1241 | return; |
1242 | } | |
1243 | fletcher_2_native(buf->b_data, buf->b_hdr->b_size, &zc); | |
1244 | if (!ZIO_CHECKSUM_EQUAL(*buf->b_hdr->b_freeze_cksum, zc)) | |
1245 | panic("buffer modified while frozen!"); | |
b9541d6b | 1246 | mutex_exit(&buf->b_hdr->b_l1hdr.b_freeze_lock); |
34dc7c2f BB |
1247 | } |
1248 | ||
1249 | static int | |
1250 | arc_cksum_equal(arc_buf_t *buf) | |
1251 | { | |
1252 | zio_cksum_t zc; | |
1253 | int equal; | |
1254 | ||
b9541d6b | 1255 | mutex_enter(&buf->b_hdr->b_l1hdr.b_freeze_lock); |
34dc7c2f BB |
1256 | fletcher_2_native(buf->b_data, buf->b_hdr->b_size, &zc); |
1257 | equal = ZIO_CHECKSUM_EQUAL(*buf->b_hdr->b_freeze_cksum, zc); | |
b9541d6b | 1258 | mutex_exit(&buf->b_hdr->b_l1hdr.b_freeze_lock); |
34dc7c2f BB |
1259 | |
1260 | return (equal); | |
1261 | } | |
1262 | ||
1263 | static void | |
1264 | arc_cksum_compute(arc_buf_t *buf, boolean_t force) | |
1265 | { | |
1266 | if (!force && !(zfs_flags & ZFS_DEBUG_MODIFY)) | |
1267 | return; | |
1268 | ||
b9541d6b | 1269 | mutex_enter(&buf->b_hdr->b_l1hdr.b_freeze_lock); |
34dc7c2f | 1270 | if (buf->b_hdr->b_freeze_cksum != NULL) { |
b9541d6b | 1271 | mutex_exit(&buf->b_hdr->b_l1hdr.b_freeze_lock); |
34dc7c2f BB |
1272 | return; |
1273 | } | |
409dc1a5 | 1274 | buf->b_hdr->b_freeze_cksum = kmem_alloc(sizeof (zio_cksum_t), |
79c76d5b | 1275 | KM_SLEEP); |
34dc7c2f BB |
1276 | fletcher_2_native(buf->b_data, buf->b_hdr->b_size, |
1277 | buf->b_hdr->b_freeze_cksum); | |
b9541d6b | 1278 | mutex_exit(&buf->b_hdr->b_l1hdr.b_freeze_lock); |
498877ba MA |
1279 | arc_buf_watch(buf); |
1280 | } | |
1281 | ||
1282 | #ifndef _KERNEL | |
1283 | void | |
1284 | arc_buf_sigsegv(int sig, siginfo_t *si, void *unused) | |
1285 | { | |
1286 | panic("Got SIGSEGV at address: 0x%lx\n", (long) si->si_addr); | |
1287 | } | |
1288 | #endif | |
1289 | ||
1290 | /* ARGSUSED */ | |
1291 | static void | |
1292 | arc_buf_unwatch(arc_buf_t *buf) | |
1293 | { | |
1294 | #ifndef _KERNEL | |
1295 | if (arc_watch) { | |
1296 | ASSERT0(mprotect(buf->b_data, buf->b_hdr->b_size, | |
1297 | PROT_READ | PROT_WRITE)); | |
1298 | } | |
1299 | #endif | |
1300 | } | |
1301 | ||
1302 | /* ARGSUSED */ | |
1303 | static void | |
1304 | arc_buf_watch(arc_buf_t *buf) | |
1305 | { | |
1306 | #ifndef _KERNEL | |
1307 | if (arc_watch) | |
1308 | ASSERT0(mprotect(buf->b_data, buf->b_hdr->b_size, PROT_READ)); | |
1309 | #endif | |
34dc7c2f BB |
1310 | } |
1311 | ||
b9541d6b CW |
1312 | static arc_buf_contents_t |
1313 | arc_buf_type(arc_buf_hdr_t *hdr) | |
1314 | { | |
1315 | if (HDR_ISTYPE_METADATA(hdr)) { | |
1316 | return (ARC_BUFC_METADATA); | |
1317 | } else { | |
1318 | return (ARC_BUFC_DATA); | |
1319 | } | |
1320 | } | |
1321 | ||
1322 | static uint32_t | |
1323 | arc_bufc_to_flags(arc_buf_contents_t type) | |
1324 | { | |
1325 | switch (type) { | |
1326 | case ARC_BUFC_DATA: | |
1327 | /* metadata field is 0 if buffer contains normal data */ | |
1328 | return (0); | |
1329 | case ARC_BUFC_METADATA: | |
1330 | return (ARC_FLAG_BUFC_METADATA); | |
1331 | default: | |
1332 | break; | |
1333 | } | |
1334 | panic("undefined ARC buffer type!"); | |
1335 | return ((uint32_t)-1); | |
1336 | } | |
1337 | ||
34dc7c2f BB |
1338 | void |
1339 | arc_buf_thaw(arc_buf_t *buf) | |
1340 | { | |
1341 | if (zfs_flags & ZFS_DEBUG_MODIFY) { | |
b9541d6b | 1342 | if (buf->b_hdr->b_l1hdr.b_state != arc_anon) |
34dc7c2f | 1343 | panic("modifying non-anon buffer!"); |
b9541d6b | 1344 | if (HDR_IO_IN_PROGRESS(buf->b_hdr)) |
34dc7c2f BB |
1345 | panic("modifying buffer while i/o in progress!"); |
1346 | arc_cksum_verify(buf); | |
1347 | } | |
1348 | ||
b9541d6b | 1349 | mutex_enter(&buf->b_hdr->b_l1hdr.b_freeze_lock); |
34dc7c2f BB |
1350 | if (buf->b_hdr->b_freeze_cksum != NULL) { |
1351 | kmem_free(buf->b_hdr->b_freeze_cksum, sizeof (zio_cksum_t)); | |
1352 | buf->b_hdr->b_freeze_cksum = NULL; | |
1353 | } | |
428870ff | 1354 | |
b9541d6b | 1355 | mutex_exit(&buf->b_hdr->b_l1hdr.b_freeze_lock); |
498877ba MA |
1356 | |
1357 | arc_buf_unwatch(buf); | |
34dc7c2f BB |
1358 | } |
1359 | ||
1360 | void | |
1361 | arc_buf_freeze(arc_buf_t *buf) | |
1362 | { | |
428870ff BB |
1363 | kmutex_t *hash_lock; |
1364 | ||
34dc7c2f BB |
1365 | if (!(zfs_flags & ZFS_DEBUG_MODIFY)) |
1366 | return; | |
1367 | ||
428870ff BB |
1368 | hash_lock = HDR_LOCK(buf->b_hdr); |
1369 | mutex_enter(hash_lock); | |
1370 | ||
34dc7c2f | 1371 | ASSERT(buf->b_hdr->b_freeze_cksum != NULL || |
b9541d6b | 1372 | buf->b_hdr->b_l1hdr.b_state == arc_anon); |
34dc7c2f | 1373 | arc_cksum_compute(buf, B_FALSE); |
428870ff | 1374 | mutex_exit(hash_lock); |
498877ba | 1375 | |
34dc7c2f BB |
1376 | } |
1377 | ||
1378 | static void | |
2a432414 | 1379 | add_reference(arc_buf_hdr_t *hdr, kmutex_t *hash_lock, void *tag) |
34dc7c2f | 1380 | { |
b9541d6b CW |
1381 | arc_state_t *state; |
1382 | ||
1383 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
34dc7c2f BB |
1384 | ASSERT(MUTEX_HELD(hash_lock)); |
1385 | ||
b9541d6b CW |
1386 | state = hdr->b_l1hdr.b_state; |
1387 | ||
1388 | if ((refcount_add(&hdr->b_l1hdr.b_refcnt, tag) == 1) && | |
1389 | (state != arc_anon)) { | |
1390 | /* We don't use the L2-only state list. */ | |
1391 | if (state != arc_l2c_only) { | |
ca0bf58d | 1392 | arc_buf_contents_t type = arc_buf_type(hdr); |
b9541d6b | 1393 | uint64_t delta = hdr->b_size * hdr->b_l1hdr.b_datacnt; |
ca0bf58d PS |
1394 | multilist_t *list = &state->arcs_list[type]; |
1395 | uint64_t *size = &state->arcs_lsize[type]; | |
1396 | ||
1397 | multilist_remove(list, hdr); | |
b9541d6b | 1398 | |
b9541d6b CW |
1399 | if (GHOST_STATE(state)) { |
1400 | ASSERT0(hdr->b_l1hdr.b_datacnt); | |
1401 | ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); | |
1402 | delta = hdr->b_size; | |
1403 | } | |
1404 | ASSERT(delta > 0); | |
1405 | ASSERT3U(*size, >=, delta); | |
1406 | atomic_add_64(size, -delta); | |
34dc7c2f | 1407 | } |
b128c09f | 1408 | /* remove the prefetch flag if we get a reference */ |
b9541d6b | 1409 | hdr->b_flags &= ~ARC_FLAG_PREFETCH; |
34dc7c2f BB |
1410 | } |
1411 | } | |
1412 | ||
1413 | static int | |
2a432414 | 1414 | remove_reference(arc_buf_hdr_t *hdr, kmutex_t *hash_lock, void *tag) |
34dc7c2f BB |
1415 | { |
1416 | int cnt; | |
b9541d6b | 1417 | arc_state_t *state = hdr->b_l1hdr.b_state; |
34dc7c2f | 1418 | |
b9541d6b | 1419 | ASSERT(HDR_HAS_L1HDR(hdr)); |
34dc7c2f BB |
1420 | ASSERT(state == arc_anon || MUTEX_HELD(hash_lock)); |
1421 | ASSERT(!GHOST_STATE(state)); | |
1422 | ||
b9541d6b CW |
1423 | /* |
1424 | * arc_l2c_only counts as a ghost state so we don't need to explicitly | |
1425 | * check to prevent usage of the arc_l2c_only list. | |
1426 | */ | |
1427 | if (((cnt = refcount_remove(&hdr->b_l1hdr.b_refcnt, tag)) == 0) && | |
34dc7c2f | 1428 | (state != arc_anon)) { |
ca0bf58d PS |
1429 | arc_buf_contents_t type = arc_buf_type(hdr); |
1430 | multilist_t *list = &state->arcs_list[type]; | |
1431 | uint64_t *size = &state->arcs_lsize[type]; | |
1432 | ||
1433 | multilist_insert(list, hdr); | |
34dc7c2f | 1434 | |
b9541d6b CW |
1435 | ASSERT(hdr->b_l1hdr.b_datacnt > 0); |
1436 | atomic_add_64(size, hdr->b_size * | |
1437 | hdr->b_l1hdr.b_datacnt); | |
34dc7c2f BB |
1438 | } |
1439 | return (cnt); | |
1440 | } | |
1441 | ||
e0b0ca98 BB |
1442 | /* |
1443 | * Returns detailed information about a specific arc buffer. When the | |
1444 | * state_index argument is set the function will calculate the arc header | |
1445 | * list position for its arc state. Since this requires a linear traversal | |
1446 | * callers are strongly encourage not to do this. However, it can be helpful | |
1447 | * for targeted analysis so the functionality is provided. | |
1448 | */ | |
1449 | void | |
1450 | arc_buf_info(arc_buf_t *ab, arc_buf_info_t *abi, int state_index) | |
1451 | { | |
1452 | arc_buf_hdr_t *hdr = ab->b_hdr; | |
b9541d6b CW |
1453 | l1arc_buf_hdr_t *l1hdr = NULL; |
1454 | l2arc_buf_hdr_t *l2hdr = NULL; | |
1455 | arc_state_t *state = NULL; | |
1456 | ||
1457 | if (HDR_HAS_L1HDR(hdr)) { | |
1458 | l1hdr = &hdr->b_l1hdr; | |
1459 | state = l1hdr->b_state; | |
1460 | } | |
1461 | if (HDR_HAS_L2HDR(hdr)) | |
1462 | l2hdr = &hdr->b_l2hdr; | |
e0b0ca98 | 1463 | |
d1d7e268 | 1464 | memset(abi, 0, sizeof (arc_buf_info_t)); |
e0b0ca98 | 1465 | abi->abi_flags = hdr->b_flags; |
b9541d6b CW |
1466 | |
1467 | if (l1hdr) { | |
1468 | abi->abi_datacnt = l1hdr->b_datacnt; | |
1469 | abi->abi_access = l1hdr->b_arc_access; | |
1470 | abi->abi_mru_hits = l1hdr->b_mru_hits; | |
1471 | abi->abi_mru_ghost_hits = l1hdr->b_mru_ghost_hits; | |
1472 | abi->abi_mfu_hits = l1hdr->b_mfu_hits; | |
1473 | abi->abi_mfu_ghost_hits = l1hdr->b_mfu_ghost_hits; | |
1474 | abi->abi_holds = refcount_count(&l1hdr->b_refcnt); | |
1475 | } | |
1476 | ||
1477 | if (l2hdr) { | |
1478 | abi->abi_l2arc_dattr = l2hdr->b_daddr; | |
1479 | abi->abi_l2arc_asize = l2hdr->b_asize; | |
1480 | abi->abi_l2arc_compress = HDR_GET_COMPRESS(hdr); | |
1481 | abi->abi_l2arc_hits = l2hdr->b_hits; | |
1482 | } | |
1483 | ||
e0b0ca98 | 1484 | abi->abi_state_type = state ? state->arcs_state : ARC_STATE_ANON; |
b9541d6b | 1485 | abi->abi_state_contents = arc_buf_type(hdr); |
e0b0ca98 | 1486 | abi->abi_size = hdr->b_size; |
e0b0ca98 BB |
1487 | } |
1488 | ||
34dc7c2f | 1489 | /* |
ca0bf58d | 1490 | * Move the supplied buffer to the indicated state. The hash lock |
34dc7c2f BB |
1491 | * for the buffer must be held by the caller. |
1492 | */ | |
1493 | static void | |
2a432414 GW |
1494 | arc_change_state(arc_state_t *new_state, arc_buf_hdr_t *hdr, |
1495 | kmutex_t *hash_lock) | |
34dc7c2f | 1496 | { |
b9541d6b CW |
1497 | arc_state_t *old_state; |
1498 | int64_t refcnt; | |
1499 | uint32_t datacnt; | |
34dc7c2f | 1500 | uint64_t from_delta, to_delta; |
b9541d6b CW |
1501 | arc_buf_contents_t buftype = arc_buf_type(hdr); |
1502 | ||
1503 | /* | |
1504 | * We almost always have an L1 hdr here, since we call arc_hdr_realloc() | |
1505 | * in arc_read() when bringing a buffer out of the L2ARC. However, the | |
1506 | * L1 hdr doesn't always exist when we change state to arc_anon before | |
1507 | * destroying a header, in which case reallocating to add the L1 hdr is | |
1508 | * pointless. | |
1509 | */ | |
1510 | if (HDR_HAS_L1HDR(hdr)) { | |
1511 | old_state = hdr->b_l1hdr.b_state; | |
1512 | refcnt = refcount_count(&hdr->b_l1hdr.b_refcnt); | |
1513 | datacnt = hdr->b_l1hdr.b_datacnt; | |
1514 | } else { | |
1515 | old_state = arc_l2c_only; | |
1516 | refcnt = 0; | |
1517 | datacnt = 0; | |
1518 | } | |
34dc7c2f BB |
1519 | |
1520 | ASSERT(MUTEX_HELD(hash_lock)); | |
e8b96c60 | 1521 | ASSERT3P(new_state, !=, old_state); |
b9541d6b CW |
1522 | ASSERT(refcnt == 0 || datacnt > 0); |
1523 | ASSERT(!GHOST_STATE(new_state) || datacnt == 0); | |
1524 | ASSERT(old_state != arc_anon || datacnt <= 1); | |
34dc7c2f | 1525 | |
b9541d6b | 1526 | from_delta = to_delta = datacnt * hdr->b_size; |
34dc7c2f BB |
1527 | |
1528 | /* | |
1529 | * If this buffer is evictable, transfer it from the | |
1530 | * old state list to the new state list. | |
1531 | */ | |
1532 | if (refcnt == 0) { | |
b9541d6b | 1533 | if (old_state != arc_anon && old_state != arc_l2c_only) { |
b9541d6b | 1534 | uint64_t *size = &old_state->arcs_lsize[buftype]; |
34dc7c2f | 1535 | |
b9541d6b | 1536 | ASSERT(HDR_HAS_L1HDR(hdr)); |
ca0bf58d | 1537 | multilist_remove(&old_state->arcs_list[buftype], hdr); |
34dc7c2f BB |
1538 | |
1539 | /* | |
1540 | * If prefetching out of the ghost cache, | |
428870ff | 1541 | * we will have a non-zero datacnt. |
34dc7c2f | 1542 | */ |
b9541d6b | 1543 | if (GHOST_STATE(old_state) && datacnt == 0) { |
34dc7c2f | 1544 | /* ghost elements have a ghost size */ |
b9541d6b | 1545 | ASSERT(hdr->b_l1hdr.b_buf == NULL); |
2a432414 | 1546 | from_delta = hdr->b_size; |
34dc7c2f BB |
1547 | } |
1548 | ASSERT3U(*size, >=, from_delta); | |
1549 | atomic_add_64(size, -from_delta); | |
34dc7c2f | 1550 | } |
b9541d6b | 1551 | if (new_state != arc_anon && new_state != arc_l2c_only) { |
b9541d6b | 1552 | uint64_t *size = &new_state->arcs_lsize[buftype]; |
34dc7c2f | 1553 | |
b9541d6b CW |
1554 | /* |
1555 | * An L1 header always exists here, since if we're | |
1556 | * moving to some L1-cached state (i.e. not l2c_only or | |
1557 | * anonymous), we realloc the header to add an L1hdr | |
1558 | * beforehand. | |
1559 | */ | |
1560 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
ca0bf58d | 1561 | multilist_insert(&new_state->arcs_list[buftype], hdr); |
34dc7c2f BB |
1562 | |
1563 | /* ghost elements have a ghost size */ | |
1564 | if (GHOST_STATE(new_state)) { | |
b9541d6b CW |
1565 | ASSERT0(datacnt); |
1566 | ASSERT(hdr->b_l1hdr.b_buf == NULL); | |
2a432414 | 1567 | to_delta = hdr->b_size; |
34dc7c2f BB |
1568 | } |
1569 | atomic_add_64(size, to_delta); | |
34dc7c2f BB |
1570 | } |
1571 | } | |
1572 | ||
2a432414 GW |
1573 | ASSERT(!BUF_EMPTY(hdr)); |
1574 | if (new_state == arc_anon && HDR_IN_HASH_TABLE(hdr)) | |
1575 | buf_hash_remove(hdr); | |
34dc7c2f | 1576 | |
b9541d6b | 1577 | /* adjust state sizes (ignore arc_l2c_only) */ |
36da08ef PS |
1578 | |
1579 | if (to_delta && new_state != arc_l2c_only) { | |
1580 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
1581 | if (GHOST_STATE(new_state)) { | |
1582 | ASSERT0(datacnt); | |
1583 | ||
1584 | /* | |
1585 | * We moving a header to a ghost state, we first | |
1586 | * remove all arc buffers. Thus, we'll have a | |
1587 | * datacnt of zero, and no arc buffer to use for | |
1588 | * the reference. As a result, we use the arc | |
1589 | * header pointer for the reference. | |
1590 | */ | |
1591 | (void) refcount_add_many(&new_state->arcs_size, | |
1592 | hdr->b_size, hdr); | |
1593 | } else { | |
1594 | arc_buf_t *buf; | |
1595 | ASSERT3U(datacnt, !=, 0); | |
1596 | ||
1597 | /* | |
1598 | * Each individual buffer holds a unique reference, | |
1599 | * thus we must remove each of these references one | |
1600 | * at a time. | |
1601 | */ | |
1602 | for (buf = hdr->b_l1hdr.b_buf; buf != NULL; | |
1603 | buf = buf->b_next) { | |
1604 | (void) refcount_add_many(&new_state->arcs_size, | |
1605 | hdr->b_size, buf); | |
1606 | } | |
1607 | } | |
1608 | } | |
1609 | ||
b9541d6b | 1610 | if (from_delta && old_state != arc_l2c_only) { |
36da08ef PS |
1611 | ASSERT(HDR_HAS_L1HDR(hdr)); |
1612 | if (GHOST_STATE(old_state)) { | |
1613 | /* | |
1614 | * When moving a header off of a ghost state, | |
1615 | * there's the possibility for datacnt to be | |
1616 | * non-zero. This is because we first add the | |
1617 | * arc buffer to the header prior to changing | |
1618 | * the header's state. Since we used the header | |
1619 | * for the reference when putting the header on | |
1620 | * the ghost state, we must balance that and use | |
1621 | * the header when removing off the ghost state | |
1622 | * (even though datacnt is non zero). | |
1623 | */ | |
1624 | ||
1625 | IMPLY(datacnt == 0, new_state == arc_anon || | |
1626 | new_state == arc_l2c_only); | |
1627 | ||
1628 | (void) refcount_remove_many(&old_state->arcs_size, | |
1629 | hdr->b_size, hdr); | |
1630 | } else { | |
1631 | arc_buf_t *buf; | |
1632 | ASSERT3U(datacnt, !=, 0); | |
1633 | ||
1634 | /* | |
1635 | * Each individual buffer holds a unique reference, | |
1636 | * thus we must remove each of these references one | |
1637 | * at a time. | |
1638 | */ | |
1639 | for (buf = hdr->b_l1hdr.b_buf; buf != NULL; | |
1640 | buf = buf->b_next) { | |
1641 | (void) refcount_remove_many( | |
1642 | &old_state->arcs_size, hdr->b_size, buf); | |
1643 | } | |
1644 | } | |
34dc7c2f | 1645 | } |
36da08ef | 1646 | |
b9541d6b CW |
1647 | if (HDR_HAS_L1HDR(hdr)) |
1648 | hdr->b_l1hdr.b_state = new_state; | |
34dc7c2f | 1649 | |
b9541d6b CW |
1650 | /* |
1651 | * L2 headers should never be on the L2 state list since they don't | |
1652 | * have L1 headers allocated. | |
1653 | */ | |
ca0bf58d PS |
1654 | ASSERT(multilist_is_empty(&arc_l2c_only->arcs_list[ARC_BUFC_DATA]) && |
1655 | multilist_is_empty(&arc_l2c_only->arcs_list[ARC_BUFC_METADATA])); | |
34dc7c2f BB |
1656 | } |
1657 | ||
1658 | void | |
d164b209 | 1659 | arc_space_consume(uint64_t space, arc_space_type_t type) |
34dc7c2f | 1660 | { |
d164b209 BB |
1661 | ASSERT(type >= 0 && type < ARC_SPACE_NUMTYPES); |
1662 | ||
1663 | switch (type) { | |
e75c13c3 BB |
1664 | default: |
1665 | break; | |
d164b209 BB |
1666 | case ARC_SPACE_DATA: |
1667 | ARCSTAT_INCR(arcstat_data_size, space); | |
1668 | break; | |
cc7f677c | 1669 | case ARC_SPACE_META: |
500445c0 | 1670 | ARCSTAT_INCR(arcstat_metadata_size, space); |
cc7f677c | 1671 | break; |
d164b209 BB |
1672 | case ARC_SPACE_OTHER: |
1673 | ARCSTAT_INCR(arcstat_other_size, space); | |
1674 | break; | |
1675 | case ARC_SPACE_HDRS: | |
1676 | ARCSTAT_INCR(arcstat_hdr_size, space); | |
1677 | break; | |
1678 | case ARC_SPACE_L2HDRS: | |
1679 | ARCSTAT_INCR(arcstat_l2_hdr_size, space); | |
1680 | break; | |
1681 | } | |
1682 | ||
500445c0 | 1683 | if (type != ARC_SPACE_DATA) |
cc7f677c PS |
1684 | ARCSTAT_INCR(arcstat_meta_used, space); |
1685 | ||
34dc7c2f BB |
1686 | atomic_add_64(&arc_size, space); |
1687 | } | |
1688 | ||
1689 | void | |
d164b209 | 1690 | arc_space_return(uint64_t space, arc_space_type_t type) |
34dc7c2f | 1691 | { |
d164b209 BB |
1692 | ASSERT(type >= 0 && type < ARC_SPACE_NUMTYPES); |
1693 | ||
1694 | switch (type) { | |
e75c13c3 BB |
1695 | default: |
1696 | break; | |
d164b209 BB |
1697 | case ARC_SPACE_DATA: |
1698 | ARCSTAT_INCR(arcstat_data_size, -space); | |
1699 | break; | |
cc7f677c | 1700 | case ARC_SPACE_META: |
500445c0 | 1701 | ARCSTAT_INCR(arcstat_metadata_size, -space); |
cc7f677c | 1702 | break; |
d164b209 BB |
1703 | case ARC_SPACE_OTHER: |
1704 | ARCSTAT_INCR(arcstat_other_size, -space); | |
1705 | break; | |
1706 | case ARC_SPACE_HDRS: | |
1707 | ARCSTAT_INCR(arcstat_hdr_size, -space); | |
1708 | break; | |
1709 | case ARC_SPACE_L2HDRS: | |
1710 | ARCSTAT_INCR(arcstat_l2_hdr_size, -space); | |
1711 | break; | |
1712 | } | |
1713 | ||
cc7f677c PS |
1714 | if (type != ARC_SPACE_DATA) { |
1715 | ASSERT(arc_meta_used >= space); | |
500445c0 PS |
1716 | if (arc_meta_max < arc_meta_used) |
1717 | arc_meta_max = arc_meta_used; | |
cc7f677c PS |
1718 | ARCSTAT_INCR(arcstat_meta_used, -space); |
1719 | } | |
1720 | ||
34dc7c2f BB |
1721 | ASSERT(arc_size >= space); |
1722 | atomic_add_64(&arc_size, -space); | |
1723 | } | |
1724 | ||
34dc7c2f | 1725 | arc_buf_t * |
5f6d0b6f | 1726 | arc_buf_alloc(spa_t *spa, uint64_t size, void *tag, arc_buf_contents_t type) |
34dc7c2f BB |
1727 | { |
1728 | arc_buf_hdr_t *hdr; | |
1729 | arc_buf_t *buf; | |
1730 | ||
f1512ee6 | 1731 | VERIFY3U(size, <=, spa_maxblocksize(spa)); |
b9541d6b | 1732 | hdr = kmem_cache_alloc(hdr_full_cache, KM_PUSHPAGE); |
34dc7c2f | 1733 | ASSERT(BUF_EMPTY(hdr)); |
b9541d6b | 1734 | ASSERT3P(hdr->b_freeze_cksum, ==, NULL); |
34dc7c2f | 1735 | hdr->b_size = size; |
3541dc6d | 1736 | hdr->b_spa = spa_load_guid(spa); |
b9541d6b CW |
1737 | hdr->b_l1hdr.b_mru_hits = 0; |
1738 | hdr->b_l1hdr.b_mru_ghost_hits = 0; | |
1739 | hdr->b_l1hdr.b_mfu_hits = 0; | |
1740 | hdr->b_l1hdr.b_mfu_ghost_hits = 0; | |
1741 | hdr->b_l1hdr.b_l2_hits = 0; | |
1742 | ||
34dc7c2f BB |
1743 | buf = kmem_cache_alloc(buf_cache, KM_PUSHPAGE); |
1744 | buf->b_hdr = hdr; | |
1745 | buf->b_data = NULL; | |
1746 | buf->b_efunc = NULL; | |
1747 | buf->b_private = NULL; | |
1748 | buf->b_next = NULL; | |
b9541d6b CW |
1749 | |
1750 | hdr->b_flags = arc_bufc_to_flags(type); | |
1751 | hdr->b_flags |= ARC_FLAG_HAS_L1HDR; | |
1752 | ||
1753 | hdr->b_l1hdr.b_buf = buf; | |
1754 | hdr->b_l1hdr.b_state = arc_anon; | |
1755 | hdr->b_l1hdr.b_arc_access = 0; | |
1756 | hdr->b_l1hdr.b_datacnt = 1; | |
ca0bf58d | 1757 | hdr->b_l1hdr.b_tmp_cdata = NULL; |
b9541d6b | 1758 | |
34dc7c2f | 1759 | arc_get_data_buf(buf); |
b9541d6b CW |
1760 | |
1761 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); | |
1762 | (void) refcount_add(&hdr->b_l1hdr.b_refcnt, tag); | |
34dc7c2f BB |
1763 | |
1764 | return (buf); | |
1765 | } | |
1766 | ||
9babb374 BB |
1767 | static char *arc_onloan_tag = "onloan"; |
1768 | ||
1769 | /* | |
1770 | * Loan out an anonymous arc buffer. Loaned buffers are not counted as in | |
1771 | * flight data by arc_tempreserve_space() until they are "returned". Loaned | |
1772 | * buffers must be returned to the arc before they can be used by the DMU or | |
1773 | * freed. | |
1774 | */ | |
1775 | arc_buf_t * | |
5f6d0b6f | 1776 | arc_loan_buf(spa_t *spa, uint64_t size) |
9babb374 BB |
1777 | { |
1778 | arc_buf_t *buf; | |
1779 | ||
1780 | buf = arc_buf_alloc(spa, size, arc_onloan_tag, ARC_BUFC_DATA); | |
1781 | ||
1782 | atomic_add_64(&arc_loaned_bytes, size); | |
1783 | return (buf); | |
1784 | } | |
1785 | ||
1786 | /* | |
1787 | * Return a loaned arc buffer to the arc. | |
1788 | */ | |
1789 | void | |
1790 | arc_return_buf(arc_buf_t *buf, void *tag) | |
1791 | { | |
1792 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
1793 | ||
9babb374 | 1794 | ASSERT(buf->b_data != NULL); |
b9541d6b CW |
1795 | ASSERT(HDR_HAS_L1HDR(hdr)); |
1796 | (void) refcount_add(&hdr->b_l1hdr.b_refcnt, tag); | |
1797 | (void) refcount_remove(&hdr->b_l1hdr.b_refcnt, arc_onloan_tag); | |
9babb374 BB |
1798 | |
1799 | atomic_add_64(&arc_loaned_bytes, -hdr->b_size); | |
1800 | } | |
1801 | ||
428870ff BB |
1802 | /* Detach an arc_buf from a dbuf (tag) */ |
1803 | void | |
1804 | arc_loan_inuse_buf(arc_buf_t *buf, void *tag) | |
1805 | { | |
b9541d6b | 1806 | arc_buf_hdr_t *hdr = buf->b_hdr; |
428870ff BB |
1807 | |
1808 | ASSERT(buf->b_data != NULL); | |
b9541d6b CW |
1809 | ASSERT(HDR_HAS_L1HDR(hdr)); |
1810 | (void) refcount_add(&hdr->b_l1hdr.b_refcnt, arc_onloan_tag); | |
1811 | (void) refcount_remove(&hdr->b_l1hdr.b_refcnt, tag); | |
428870ff BB |
1812 | buf->b_efunc = NULL; |
1813 | buf->b_private = NULL; | |
1814 | ||
1815 | atomic_add_64(&arc_loaned_bytes, hdr->b_size); | |
1816 | } | |
1817 | ||
34dc7c2f BB |
1818 | static arc_buf_t * |
1819 | arc_buf_clone(arc_buf_t *from) | |
1820 | { | |
1821 | arc_buf_t *buf; | |
1822 | arc_buf_hdr_t *hdr = from->b_hdr; | |
1823 | uint64_t size = hdr->b_size; | |
1824 | ||
b9541d6b CW |
1825 | ASSERT(HDR_HAS_L1HDR(hdr)); |
1826 | ASSERT(hdr->b_l1hdr.b_state != arc_anon); | |
428870ff | 1827 | |
34dc7c2f BB |
1828 | buf = kmem_cache_alloc(buf_cache, KM_PUSHPAGE); |
1829 | buf->b_hdr = hdr; | |
1830 | buf->b_data = NULL; | |
1831 | buf->b_efunc = NULL; | |
1832 | buf->b_private = NULL; | |
b9541d6b CW |
1833 | buf->b_next = hdr->b_l1hdr.b_buf; |
1834 | hdr->b_l1hdr.b_buf = buf; | |
34dc7c2f BB |
1835 | arc_get_data_buf(buf); |
1836 | bcopy(from->b_data, buf->b_data, size); | |
1eb5bfa3 GW |
1837 | |
1838 | /* | |
1839 | * This buffer already exists in the arc so create a duplicate | |
1840 | * copy for the caller. If the buffer is associated with user data | |
1841 | * then track the size and number of duplicates. These stats will be | |
1842 | * updated as duplicate buffers are created and destroyed. | |
1843 | */ | |
b9541d6b | 1844 | if (HDR_ISTYPE_DATA(hdr)) { |
1eb5bfa3 GW |
1845 | ARCSTAT_BUMP(arcstat_duplicate_buffers); |
1846 | ARCSTAT_INCR(arcstat_duplicate_buffers_size, size); | |
1847 | } | |
b9541d6b | 1848 | hdr->b_l1hdr.b_datacnt += 1; |
34dc7c2f BB |
1849 | return (buf); |
1850 | } | |
1851 | ||
1852 | void | |
1853 | arc_buf_add_ref(arc_buf_t *buf, void* tag) | |
1854 | { | |
1855 | arc_buf_hdr_t *hdr; | |
1856 | kmutex_t *hash_lock; | |
1857 | ||
1858 | /* | |
b128c09f BB |
1859 | * Check to see if this buffer is evicted. Callers |
1860 | * must verify b_data != NULL to know if the add_ref | |
1861 | * was successful. | |
34dc7c2f | 1862 | */ |
428870ff | 1863 | mutex_enter(&buf->b_evict_lock); |
b128c09f | 1864 | if (buf->b_data == NULL) { |
428870ff | 1865 | mutex_exit(&buf->b_evict_lock); |
34dc7c2f BB |
1866 | return; |
1867 | } | |
428870ff | 1868 | hash_lock = HDR_LOCK(buf->b_hdr); |
34dc7c2f | 1869 | mutex_enter(hash_lock); |
428870ff | 1870 | hdr = buf->b_hdr; |
b9541d6b | 1871 | ASSERT(HDR_HAS_L1HDR(hdr)); |
428870ff BB |
1872 | ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); |
1873 | mutex_exit(&buf->b_evict_lock); | |
34dc7c2f | 1874 | |
b9541d6b CW |
1875 | ASSERT(hdr->b_l1hdr.b_state == arc_mru || |
1876 | hdr->b_l1hdr.b_state == arc_mfu); | |
1877 | ||
34dc7c2f | 1878 | add_reference(hdr, hash_lock, tag); |
d164b209 | 1879 | DTRACE_PROBE1(arc__hit, arc_buf_hdr_t *, hdr); |
34dc7c2f BB |
1880 | arc_access(hdr, hash_lock); |
1881 | mutex_exit(hash_lock); | |
1882 | ARCSTAT_BUMP(arcstat_hits); | |
b9541d6b CW |
1883 | ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr), |
1884 | demand, prefetch, !HDR_ISTYPE_METADATA(hdr), | |
34dc7c2f BB |
1885 | data, metadata, hits); |
1886 | } | |
1887 | ||
ca0bf58d PS |
1888 | static void |
1889 | arc_buf_free_on_write(void *data, size_t size, | |
1890 | void (*free_func)(void *, size_t)) | |
1891 | { | |
1892 | l2arc_data_free_t *df; | |
1893 | ||
1894 | df = kmem_alloc(sizeof (*df), KM_SLEEP); | |
1895 | df->l2df_data = data; | |
1896 | df->l2df_size = size; | |
1897 | df->l2df_func = free_func; | |
1898 | mutex_enter(&l2arc_free_on_write_mtx); | |
1899 | list_insert_head(l2arc_free_on_write, df); | |
1900 | mutex_exit(&l2arc_free_on_write_mtx); | |
1901 | } | |
1902 | ||
34dc7c2f BB |
1903 | /* |
1904 | * Free the arc data buffer. If it is an l2arc write in progress, | |
1905 | * the buffer is placed on l2arc_free_on_write to be freed later. | |
1906 | */ | |
1907 | static void | |
498877ba | 1908 | arc_buf_data_free(arc_buf_t *buf, void (*free_func)(void *, size_t)) |
34dc7c2f | 1909 | { |
498877ba MA |
1910 | arc_buf_hdr_t *hdr = buf->b_hdr; |
1911 | ||
34dc7c2f | 1912 | if (HDR_L2_WRITING(hdr)) { |
ca0bf58d | 1913 | arc_buf_free_on_write(buf->b_data, hdr->b_size, free_func); |
34dc7c2f BB |
1914 | ARCSTAT_BUMP(arcstat_l2_free_on_write); |
1915 | } else { | |
498877ba | 1916 | free_func(buf->b_data, hdr->b_size); |
34dc7c2f BB |
1917 | } |
1918 | } | |
1919 | ||
ca0bf58d PS |
1920 | static void |
1921 | arc_buf_l2_cdata_free(arc_buf_hdr_t *hdr) | |
1922 | { | |
1923 | ASSERT(HDR_HAS_L2HDR(hdr)); | |
1924 | ASSERT(MUTEX_HELD(&hdr->b_l2hdr.b_dev->l2ad_mtx)); | |
1925 | ||
1926 | /* | |
1927 | * The b_tmp_cdata field is linked off of the b_l1hdr, so if | |
1928 | * that doesn't exist, the header is in the arc_l2c_only state, | |
1929 | * and there isn't anything to free (it's already been freed). | |
1930 | */ | |
1931 | if (!HDR_HAS_L1HDR(hdr)) | |
1932 | return; | |
1933 | ||
1934 | /* | |
1935 | * The header isn't being written to the l2arc device, thus it | |
1936 | * shouldn't have a b_tmp_cdata to free. | |
1937 | */ | |
1938 | if (!HDR_L2_WRITING(hdr)) { | |
1939 | ASSERT3P(hdr->b_l1hdr.b_tmp_cdata, ==, NULL); | |
1940 | return; | |
1941 | } | |
1942 | ||
1943 | /* | |
1944 | * The header does not have compression enabled. This can be due | |
1945 | * to the buffer not being compressible, or because we're | |
1946 | * freeing the buffer before the second phase of | |
1947 | * l2arc_write_buffer() has started (which does the compression | |
1948 | * step). In either case, b_tmp_cdata does not point to a | |
1949 | * separately compressed buffer, so there's nothing to free (it | |
1950 | * points to the same buffer as the arc_buf_t's b_data field). | |
1951 | */ | |
1952 | if (HDR_GET_COMPRESS(hdr) == ZIO_COMPRESS_OFF) { | |
1953 | hdr->b_l1hdr.b_tmp_cdata = NULL; | |
1954 | return; | |
1955 | } | |
1956 | ||
1957 | /* | |
1958 | * There's nothing to free since the buffer was all zero's and | |
1959 | * compressed to a zero length buffer. | |
1960 | */ | |
1961 | if (HDR_GET_COMPRESS(hdr) == ZIO_COMPRESS_EMPTY) { | |
1962 | ASSERT3P(hdr->b_l1hdr.b_tmp_cdata, ==, NULL); | |
1963 | return; | |
1964 | } | |
1965 | ||
1966 | ASSERT(L2ARC_IS_VALID_COMPRESS(HDR_GET_COMPRESS(hdr))); | |
1967 | ||
1968 | arc_buf_free_on_write(hdr->b_l1hdr.b_tmp_cdata, | |
1969 | hdr->b_size, zio_data_buf_free); | |
1970 | ||
1971 | ARCSTAT_BUMP(arcstat_l2_cdata_free_on_write); | |
1972 | hdr->b_l1hdr.b_tmp_cdata = NULL; | |
1973 | } | |
1974 | ||
bd089c54 MA |
1975 | /* |
1976 | * Free up buf->b_data and if 'remove' is set, then pull the | |
1977 | * arc_buf_t off of the the arc_buf_hdr_t's list and free it. | |
1978 | */ | |
34dc7c2f | 1979 | static void |
ca0bf58d | 1980 | arc_buf_destroy(arc_buf_t *buf, boolean_t remove) |
34dc7c2f BB |
1981 | { |
1982 | arc_buf_t **bufp; | |
1983 | ||
1984 | /* free up data associated with the buf */ | |
b9541d6b CW |
1985 | if (buf->b_data != NULL) { |
1986 | arc_state_t *state = buf->b_hdr->b_l1hdr.b_state; | |
34dc7c2f | 1987 | uint64_t size = buf->b_hdr->b_size; |
b9541d6b | 1988 | arc_buf_contents_t type = arc_buf_type(buf->b_hdr); |
34dc7c2f BB |
1989 | |
1990 | arc_cksum_verify(buf); | |
498877ba | 1991 | arc_buf_unwatch(buf); |
428870ff | 1992 | |
ca0bf58d PS |
1993 | if (type == ARC_BUFC_METADATA) { |
1994 | arc_buf_data_free(buf, zio_buf_free); | |
1995 | arc_space_return(size, ARC_SPACE_META); | |
1996 | } else { | |
1997 | ASSERT(type == ARC_BUFC_DATA); | |
1998 | arc_buf_data_free(buf, zio_data_buf_free); | |
1999 | arc_space_return(size, ARC_SPACE_DATA); | |
34dc7c2f | 2000 | } |
ca0bf58d PS |
2001 | |
2002 | /* protected by hash lock, if in the hash table */ | |
2003 | if (multilist_link_active(&buf->b_hdr->b_l1hdr.b_arc_node)) { | |
34dc7c2f BB |
2004 | uint64_t *cnt = &state->arcs_lsize[type]; |
2005 | ||
b9541d6b CW |
2006 | ASSERT(refcount_is_zero( |
2007 | &buf->b_hdr->b_l1hdr.b_refcnt)); | |
2008 | ASSERT(state != arc_anon && state != arc_l2c_only); | |
34dc7c2f BB |
2009 | |
2010 | ASSERT3U(*cnt, >=, size); | |
2011 | atomic_add_64(cnt, -size); | |
2012 | } | |
36da08ef PS |
2013 | |
2014 | (void) refcount_remove_many(&state->arcs_size, size, buf); | |
34dc7c2f | 2015 | buf->b_data = NULL; |
1eb5bfa3 GW |
2016 | |
2017 | /* | |
2018 | * If we're destroying a duplicate buffer make sure | |
2019 | * that the appropriate statistics are updated. | |
2020 | */ | |
b9541d6b CW |
2021 | if (buf->b_hdr->b_l1hdr.b_datacnt > 1 && |
2022 | HDR_ISTYPE_DATA(buf->b_hdr)) { | |
1eb5bfa3 GW |
2023 | ARCSTAT_BUMPDOWN(arcstat_duplicate_buffers); |
2024 | ARCSTAT_INCR(arcstat_duplicate_buffers_size, -size); | |
2025 | } | |
b9541d6b CW |
2026 | ASSERT(buf->b_hdr->b_l1hdr.b_datacnt > 0); |
2027 | buf->b_hdr->b_l1hdr.b_datacnt -= 1; | |
34dc7c2f BB |
2028 | } |
2029 | ||
2030 | /* only remove the buf if requested */ | |
bd089c54 | 2031 | if (!remove) |
34dc7c2f BB |
2032 | return; |
2033 | ||
2034 | /* remove the buf from the hdr list */ | |
b9541d6b CW |
2035 | for (bufp = &buf->b_hdr->b_l1hdr.b_buf; *bufp != buf; |
2036 | bufp = &(*bufp)->b_next) | |
34dc7c2f BB |
2037 | continue; |
2038 | *bufp = buf->b_next; | |
428870ff | 2039 | buf->b_next = NULL; |
34dc7c2f BB |
2040 | |
2041 | ASSERT(buf->b_efunc == NULL); | |
2042 | ||
2043 | /* clean up the buf */ | |
2044 | buf->b_hdr = NULL; | |
2045 | kmem_cache_free(buf_cache, buf); | |
2046 | } | |
2047 | ||
d962d5da PS |
2048 | static void |
2049 | arc_hdr_l2hdr_destroy(arc_buf_hdr_t *hdr) | |
2050 | { | |
2051 | l2arc_buf_hdr_t *l2hdr = &hdr->b_l2hdr; | |
2052 | l2arc_dev_t *dev = l2hdr->b_dev; | |
2053 | ||
2054 | ASSERT(MUTEX_HELD(&dev->l2ad_mtx)); | |
2055 | ASSERT(HDR_HAS_L2HDR(hdr)); | |
2056 | ||
2057 | list_remove(&dev->l2ad_buflist, hdr); | |
2058 | ||
d962d5da PS |
2059 | /* |
2060 | * We don't want to leak the b_tmp_cdata buffer that was | |
2061 | * allocated in l2arc_write_buffers() | |
2062 | */ | |
2063 | arc_buf_l2_cdata_free(hdr); | |
2064 | ||
2065 | /* | |
2066 | * If the l2hdr's b_daddr is equal to L2ARC_ADDR_UNSET, then | |
2067 | * this header is being processed by l2arc_write_buffers() (i.e. | |
2068 | * it's in the first stage of l2arc_write_buffers()). | |
2069 | * Re-affirming that truth here, just to serve as a reminder. If | |
2070 | * b_daddr does not equal L2ARC_ADDR_UNSET, then the header may or | |
2071 | * may not have its HDR_L2_WRITING flag set. (the write may have | |
2072 | * completed, in which case HDR_L2_WRITING will be false and the | |
2073 | * b_daddr field will point to the address of the buffer on disk). | |
2074 | */ | |
2075 | IMPLY(l2hdr->b_daddr == L2ARC_ADDR_UNSET, HDR_L2_WRITING(hdr)); | |
2076 | ||
2077 | /* | |
2078 | * If b_daddr is equal to L2ARC_ADDR_UNSET, we're racing with | |
2079 | * l2arc_write_buffers(). Since we've just removed this header | |
2080 | * from the l2arc buffer list, this header will never reach the | |
2081 | * second stage of l2arc_write_buffers(), which increments the | |
2082 | * accounting stats for this header. Thus, we must be careful | |
2083 | * not to decrement them for this header either. | |
2084 | */ | |
2085 | if (l2hdr->b_daddr != L2ARC_ADDR_UNSET) { | |
2086 | ARCSTAT_INCR(arcstat_l2_asize, -l2hdr->b_asize); | |
2087 | ARCSTAT_INCR(arcstat_l2_size, -hdr->b_size); | |
2088 | ||
2089 | vdev_space_update(dev->l2ad_vdev, | |
2090 | -l2hdr->b_asize, 0, 0); | |
2091 | ||
2092 | (void) refcount_remove_many(&dev->l2ad_alloc, | |
2093 | l2hdr->b_asize, hdr); | |
2094 | } | |
2095 | ||
2096 | hdr->b_flags &= ~ARC_FLAG_HAS_L2HDR; | |
2097 | } | |
2098 | ||
34dc7c2f BB |
2099 | static void |
2100 | arc_hdr_destroy(arc_buf_hdr_t *hdr) | |
2101 | { | |
b9541d6b CW |
2102 | if (HDR_HAS_L1HDR(hdr)) { |
2103 | ASSERT(hdr->b_l1hdr.b_buf == NULL || | |
2104 | hdr->b_l1hdr.b_datacnt > 0); | |
2105 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); | |
2106 | ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon); | |
2107 | } | |
34dc7c2f | 2108 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); |
b9541d6b CW |
2109 | ASSERT(!HDR_IN_HASH_TABLE(hdr)); |
2110 | ||
2111 | if (HDR_HAS_L2HDR(hdr)) { | |
d962d5da PS |
2112 | l2arc_dev_t *dev = hdr->b_l2hdr.b_dev; |
2113 | boolean_t buflist_held = MUTEX_HELD(&dev->l2ad_mtx); | |
428870ff | 2114 | |
d962d5da PS |
2115 | if (!buflist_held) |
2116 | mutex_enter(&dev->l2ad_mtx); | |
b9541d6b | 2117 | |
ca0bf58d | 2118 | /* |
d962d5da PS |
2119 | * Even though we checked this conditional above, we |
2120 | * need to check this again now that we have the | |
2121 | * l2ad_mtx. This is because we could be racing with | |
2122 | * another thread calling l2arc_evict() which might have | |
2123 | * destroyed this header's L2 portion as we were waiting | |
2124 | * to acquire the l2ad_mtx. If that happens, we don't | |
2125 | * want to re-destroy the header's L2 portion. | |
ca0bf58d | 2126 | */ |
d962d5da PS |
2127 | if (HDR_HAS_L2HDR(hdr)) |
2128 | arc_hdr_l2hdr_destroy(hdr); | |
428870ff BB |
2129 | |
2130 | if (!buflist_held) | |
d962d5da | 2131 | mutex_exit(&dev->l2ad_mtx); |
34dc7c2f BB |
2132 | } |
2133 | ||
b9541d6b | 2134 | if (!BUF_EMPTY(hdr)) |
428870ff | 2135 | buf_discard_identity(hdr); |
b9541d6b | 2136 | |
34dc7c2f BB |
2137 | if (hdr->b_freeze_cksum != NULL) { |
2138 | kmem_free(hdr->b_freeze_cksum, sizeof (zio_cksum_t)); | |
2139 | hdr->b_freeze_cksum = NULL; | |
2140 | } | |
2141 | ||
b9541d6b CW |
2142 | if (HDR_HAS_L1HDR(hdr)) { |
2143 | while (hdr->b_l1hdr.b_buf) { | |
2144 | arc_buf_t *buf = hdr->b_l1hdr.b_buf; | |
2145 | ||
2146 | if (buf->b_efunc != NULL) { | |
ca0bf58d | 2147 | mutex_enter(&arc_user_evicts_lock); |
b9541d6b CW |
2148 | mutex_enter(&buf->b_evict_lock); |
2149 | ASSERT(buf->b_hdr != NULL); | |
ca0bf58d | 2150 | arc_buf_destroy(hdr->b_l1hdr.b_buf, FALSE); |
b9541d6b CW |
2151 | hdr->b_l1hdr.b_buf = buf->b_next; |
2152 | buf->b_hdr = &arc_eviction_hdr; | |
2153 | buf->b_next = arc_eviction_list; | |
2154 | arc_eviction_list = buf; | |
2155 | mutex_exit(&buf->b_evict_lock); | |
ca0bf58d PS |
2156 | cv_signal(&arc_user_evicts_cv); |
2157 | mutex_exit(&arc_user_evicts_lock); | |
b9541d6b | 2158 | } else { |
ca0bf58d | 2159 | arc_buf_destroy(hdr->b_l1hdr.b_buf, TRUE); |
b9541d6b CW |
2160 | } |
2161 | } | |
2162 | } | |
2163 | ||
34dc7c2f | 2164 | ASSERT3P(hdr->b_hash_next, ==, NULL); |
b9541d6b | 2165 | if (HDR_HAS_L1HDR(hdr)) { |
ca0bf58d | 2166 | ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); |
b9541d6b CW |
2167 | ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL); |
2168 | kmem_cache_free(hdr_full_cache, hdr); | |
2169 | } else { | |
2170 | kmem_cache_free(hdr_l2only_cache, hdr); | |
2171 | } | |
34dc7c2f BB |
2172 | } |
2173 | ||
2174 | void | |
2175 | arc_buf_free(arc_buf_t *buf, void *tag) | |
2176 | { | |
2177 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
b9541d6b | 2178 | int hashed = hdr->b_l1hdr.b_state != arc_anon; |
34dc7c2f BB |
2179 | |
2180 | ASSERT(buf->b_efunc == NULL); | |
2181 | ASSERT(buf->b_data != NULL); | |
2182 | ||
2183 | if (hashed) { | |
2184 | kmutex_t *hash_lock = HDR_LOCK(hdr); | |
2185 | ||
2186 | mutex_enter(hash_lock); | |
428870ff BB |
2187 | hdr = buf->b_hdr; |
2188 | ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); | |
2189 | ||
34dc7c2f | 2190 | (void) remove_reference(hdr, hash_lock, tag); |
b9541d6b | 2191 | if (hdr->b_l1hdr.b_datacnt > 1) { |
ca0bf58d | 2192 | arc_buf_destroy(buf, TRUE); |
428870ff | 2193 | } else { |
b9541d6b | 2194 | ASSERT(buf == hdr->b_l1hdr.b_buf); |
428870ff | 2195 | ASSERT(buf->b_efunc == NULL); |
2a432414 | 2196 | hdr->b_flags |= ARC_FLAG_BUF_AVAILABLE; |
428870ff | 2197 | } |
34dc7c2f BB |
2198 | mutex_exit(hash_lock); |
2199 | } else if (HDR_IO_IN_PROGRESS(hdr)) { | |
2200 | int destroy_hdr; | |
2201 | /* | |
2202 | * We are in the middle of an async write. Don't destroy | |
2203 | * this buffer unless the write completes before we finish | |
2204 | * decrementing the reference count. | |
2205 | */ | |
ca0bf58d | 2206 | mutex_enter(&arc_user_evicts_lock); |
34dc7c2f | 2207 | (void) remove_reference(hdr, NULL, tag); |
b9541d6b | 2208 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); |
34dc7c2f | 2209 | destroy_hdr = !HDR_IO_IN_PROGRESS(hdr); |
ca0bf58d | 2210 | mutex_exit(&arc_user_evicts_lock); |
34dc7c2f BB |
2211 | if (destroy_hdr) |
2212 | arc_hdr_destroy(hdr); | |
2213 | } else { | |
428870ff | 2214 | if (remove_reference(hdr, NULL, tag) > 0) |
ca0bf58d | 2215 | arc_buf_destroy(buf, TRUE); |
428870ff | 2216 | else |
34dc7c2f | 2217 | arc_hdr_destroy(hdr); |
34dc7c2f BB |
2218 | } |
2219 | } | |
2220 | ||
13fe0198 | 2221 | boolean_t |
34dc7c2f BB |
2222 | arc_buf_remove_ref(arc_buf_t *buf, void* tag) |
2223 | { | |
2224 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
b4f7f105 | 2225 | kmutex_t *hash_lock = NULL; |
13fe0198 | 2226 | boolean_t no_callback = (buf->b_efunc == NULL); |
34dc7c2f | 2227 | |
b9541d6b CW |
2228 | if (hdr->b_l1hdr.b_state == arc_anon) { |
2229 | ASSERT(hdr->b_l1hdr.b_datacnt == 1); | |
34dc7c2f BB |
2230 | arc_buf_free(buf, tag); |
2231 | return (no_callback); | |
2232 | } | |
2233 | ||
b4f7f105 | 2234 | hash_lock = HDR_LOCK(hdr); |
34dc7c2f | 2235 | mutex_enter(hash_lock); |
428870ff | 2236 | hdr = buf->b_hdr; |
b9541d6b | 2237 | ASSERT(hdr->b_l1hdr.b_datacnt > 0); |
428870ff | 2238 | ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); |
b9541d6b | 2239 | ASSERT(hdr->b_l1hdr.b_state != arc_anon); |
34dc7c2f BB |
2240 | ASSERT(buf->b_data != NULL); |
2241 | ||
2242 | (void) remove_reference(hdr, hash_lock, tag); | |
b9541d6b | 2243 | if (hdr->b_l1hdr.b_datacnt > 1) { |
34dc7c2f | 2244 | if (no_callback) |
ca0bf58d | 2245 | arc_buf_destroy(buf, TRUE); |
34dc7c2f | 2246 | } else if (no_callback) { |
b9541d6b | 2247 | ASSERT(hdr->b_l1hdr.b_buf == buf && buf->b_next == NULL); |
428870ff | 2248 | ASSERT(buf->b_efunc == NULL); |
2a432414 | 2249 | hdr->b_flags |= ARC_FLAG_BUF_AVAILABLE; |
34dc7c2f | 2250 | } |
b9541d6b CW |
2251 | ASSERT(no_callback || hdr->b_l1hdr.b_datacnt > 1 || |
2252 | refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); | |
34dc7c2f BB |
2253 | mutex_exit(hash_lock); |
2254 | return (no_callback); | |
2255 | } | |
2256 | ||
5f6d0b6f | 2257 | uint64_t |
34dc7c2f BB |
2258 | arc_buf_size(arc_buf_t *buf) |
2259 | { | |
2260 | return (buf->b_hdr->b_size); | |
2261 | } | |
2262 | ||
1eb5bfa3 GW |
2263 | /* |
2264 | * Called from the DMU to determine if the current buffer should be | |
2265 | * evicted. In order to ensure proper locking, the eviction must be initiated | |
2266 | * from the DMU. Return true if the buffer is associated with user data and | |
2267 | * duplicate buffers still exist. | |
2268 | */ | |
2269 | boolean_t | |
2270 | arc_buf_eviction_needed(arc_buf_t *buf) | |
2271 | { | |
2272 | arc_buf_hdr_t *hdr; | |
2273 | boolean_t evict_needed = B_FALSE; | |
2274 | ||
2275 | if (zfs_disable_dup_eviction) | |
2276 | return (B_FALSE); | |
2277 | ||
2278 | mutex_enter(&buf->b_evict_lock); | |
2279 | hdr = buf->b_hdr; | |
2280 | if (hdr == NULL) { | |
2281 | /* | |
2282 | * We are in arc_do_user_evicts(); let that function | |
2283 | * perform the eviction. | |
2284 | */ | |
2285 | ASSERT(buf->b_data == NULL); | |
2286 | mutex_exit(&buf->b_evict_lock); | |
2287 | return (B_FALSE); | |
2288 | } else if (buf->b_data == NULL) { | |
2289 | /* | |
2290 | * We have already been added to the arc eviction list; | |
2291 | * recommend eviction. | |
2292 | */ | |
2293 | ASSERT3P(hdr, ==, &arc_eviction_hdr); | |
2294 | mutex_exit(&buf->b_evict_lock); | |
2295 | return (B_TRUE); | |
2296 | } | |
2297 | ||
b9541d6b | 2298 | if (hdr->b_l1hdr.b_datacnt > 1 && HDR_ISTYPE_DATA(hdr)) |
1eb5bfa3 GW |
2299 | evict_needed = B_TRUE; |
2300 | ||
2301 | mutex_exit(&buf->b_evict_lock); | |
2302 | return (evict_needed); | |
2303 | } | |
2304 | ||
34dc7c2f | 2305 | /* |
ca0bf58d PS |
2306 | * Evict the arc_buf_hdr that is provided as a parameter. The resultant |
2307 | * state of the header is dependent on its state prior to entering this | |
2308 | * function. The following transitions are possible: | |
34dc7c2f | 2309 | * |
ca0bf58d PS |
2310 | * - arc_mru -> arc_mru_ghost |
2311 | * - arc_mfu -> arc_mfu_ghost | |
2312 | * - arc_mru_ghost -> arc_l2c_only | |
2313 | * - arc_mru_ghost -> deleted | |
2314 | * - arc_mfu_ghost -> arc_l2c_only | |
2315 | * - arc_mfu_ghost -> deleted | |
34dc7c2f | 2316 | */ |
ca0bf58d PS |
2317 | static int64_t |
2318 | arc_evict_hdr(arc_buf_hdr_t *hdr, kmutex_t *hash_lock) | |
34dc7c2f | 2319 | { |
ca0bf58d PS |
2320 | arc_state_t *evicted_state, *state; |
2321 | int64_t bytes_evicted = 0; | |
34dc7c2f | 2322 | |
ca0bf58d PS |
2323 | ASSERT(MUTEX_HELD(hash_lock)); |
2324 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
e8b96c60 | 2325 | |
ca0bf58d PS |
2326 | state = hdr->b_l1hdr.b_state; |
2327 | if (GHOST_STATE(state)) { | |
2328 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); | |
2329 | ASSERT(hdr->b_l1hdr.b_buf == NULL); | |
e8b96c60 MA |
2330 | |
2331 | /* | |
ca0bf58d PS |
2332 | * l2arc_write_buffers() relies on a header's L1 portion |
2333 | * (i.e. its b_tmp_cdata field) during its write phase. | |
2334 | * Thus, we cannot push a header onto the arc_l2c_only | |
2335 | * state (removing its L1 piece) until the header is | |
2336 | * done being written to the l2arc. | |
e8b96c60 | 2337 | */ |
ca0bf58d PS |
2338 | if (HDR_HAS_L2HDR(hdr) && HDR_L2_WRITING(hdr)) { |
2339 | ARCSTAT_BUMP(arcstat_evict_l2_skip); | |
2340 | return (bytes_evicted); | |
e8b96c60 MA |
2341 | } |
2342 | ||
ca0bf58d PS |
2343 | ARCSTAT_BUMP(arcstat_deleted); |
2344 | bytes_evicted += hdr->b_size; | |
428870ff | 2345 | |
ca0bf58d | 2346 | DTRACE_PROBE1(arc__delete, arc_buf_hdr_t *, hdr); |
428870ff | 2347 | |
ca0bf58d PS |
2348 | if (HDR_HAS_L2HDR(hdr)) { |
2349 | /* | |
2350 | * This buffer is cached on the 2nd Level ARC; | |
2351 | * don't destroy the header. | |
2352 | */ | |
2353 | arc_change_state(arc_l2c_only, hdr, hash_lock); | |
2354 | /* | |
2355 | * dropping from L1+L2 cached to L2-only, | |
2356 | * realloc to remove the L1 header. | |
2357 | */ | |
2358 | hdr = arc_hdr_realloc(hdr, hdr_full_cache, | |
2359 | hdr_l2only_cache); | |
34dc7c2f | 2360 | } else { |
ca0bf58d PS |
2361 | arc_change_state(arc_anon, hdr, hash_lock); |
2362 | arc_hdr_destroy(hdr); | |
34dc7c2f | 2363 | } |
ca0bf58d | 2364 | return (bytes_evicted); |
34dc7c2f BB |
2365 | } |
2366 | ||
ca0bf58d PS |
2367 | ASSERT(state == arc_mru || state == arc_mfu); |
2368 | evicted_state = (state == arc_mru) ? arc_mru_ghost : arc_mfu_ghost; | |
34dc7c2f | 2369 | |
ca0bf58d PS |
2370 | /* prefetch buffers have a minimum lifespan */ |
2371 | if (HDR_IO_IN_PROGRESS(hdr) || | |
2372 | ((hdr->b_flags & (ARC_FLAG_PREFETCH | ARC_FLAG_INDIRECT)) && | |
2373 | ddi_get_lbolt() - hdr->b_l1hdr.b_arc_access < | |
2374 | arc_min_prefetch_lifespan)) { | |
2375 | ARCSTAT_BUMP(arcstat_evict_skip); | |
2376 | return (bytes_evicted); | |
da8ccd0e PS |
2377 | } |
2378 | ||
ca0bf58d PS |
2379 | ASSERT0(refcount_count(&hdr->b_l1hdr.b_refcnt)); |
2380 | ASSERT3U(hdr->b_l1hdr.b_datacnt, >, 0); | |
2381 | while (hdr->b_l1hdr.b_buf) { | |
2382 | arc_buf_t *buf = hdr->b_l1hdr.b_buf; | |
2383 | if (!mutex_tryenter(&buf->b_evict_lock)) { | |
2384 | ARCSTAT_BUMP(arcstat_mutex_miss); | |
2385 | break; | |
2386 | } | |
2387 | if (buf->b_data != NULL) | |
2388 | bytes_evicted += hdr->b_size; | |
2389 | if (buf->b_efunc != NULL) { | |
2390 | mutex_enter(&arc_user_evicts_lock); | |
2391 | arc_buf_destroy(buf, FALSE); | |
2392 | hdr->b_l1hdr.b_buf = buf->b_next; | |
2393 | buf->b_hdr = &arc_eviction_hdr; | |
2394 | buf->b_next = arc_eviction_list; | |
2395 | arc_eviction_list = buf; | |
2396 | cv_signal(&arc_user_evicts_cv); | |
2397 | mutex_exit(&arc_user_evicts_lock); | |
2398 | mutex_exit(&buf->b_evict_lock); | |
2399 | } else { | |
2400 | mutex_exit(&buf->b_evict_lock); | |
2401 | arc_buf_destroy(buf, TRUE); | |
2402 | } | |
2403 | } | |
34dc7c2f | 2404 | |
ca0bf58d PS |
2405 | if (HDR_HAS_L2HDR(hdr)) { |
2406 | ARCSTAT_INCR(arcstat_evict_l2_cached, hdr->b_size); | |
2407 | } else { | |
2408 | if (l2arc_write_eligible(hdr->b_spa, hdr)) | |
2409 | ARCSTAT_INCR(arcstat_evict_l2_eligible, hdr->b_size); | |
2410 | else | |
2411 | ARCSTAT_INCR(arcstat_evict_l2_ineligible, hdr->b_size); | |
2412 | } | |
34dc7c2f | 2413 | |
ca0bf58d PS |
2414 | if (hdr->b_l1hdr.b_datacnt == 0) { |
2415 | arc_change_state(evicted_state, hdr, hash_lock); | |
2416 | ASSERT(HDR_IN_HASH_TABLE(hdr)); | |
2417 | hdr->b_flags |= ARC_FLAG_IN_HASH_TABLE; | |
2418 | hdr->b_flags &= ~ARC_FLAG_BUF_AVAILABLE; | |
2419 | DTRACE_PROBE1(arc__evict, arc_buf_hdr_t *, hdr); | |
2420 | } | |
34dc7c2f | 2421 | |
ca0bf58d | 2422 | return (bytes_evicted); |
34dc7c2f BB |
2423 | } |
2424 | ||
ca0bf58d PS |
2425 | static uint64_t |
2426 | arc_evict_state_impl(multilist_t *ml, int idx, arc_buf_hdr_t *marker, | |
2427 | uint64_t spa, int64_t bytes) | |
34dc7c2f | 2428 | { |
ca0bf58d PS |
2429 | multilist_sublist_t *mls; |
2430 | uint64_t bytes_evicted = 0; | |
2431 | arc_buf_hdr_t *hdr; | |
34dc7c2f | 2432 | kmutex_t *hash_lock; |
ca0bf58d | 2433 | int evict_count = 0; |
34dc7c2f | 2434 | |
ca0bf58d PS |
2435 | ASSERT3P(marker, !=, NULL); |
2436 | ASSERTV(if (bytes < 0) ASSERT(bytes == ARC_EVICT_ALL)); | |
2437 | ||
2438 | mls = multilist_sublist_lock(ml, idx); | |
572e2857 | 2439 | |
ca0bf58d PS |
2440 | for (hdr = multilist_sublist_prev(mls, marker); hdr != NULL; |
2441 | hdr = multilist_sublist_prev(mls, marker)) { | |
2442 | if ((bytes != ARC_EVICT_ALL && bytes_evicted >= bytes) || | |
2443 | (evict_count >= zfs_arc_evict_batch_limit)) | |
2444 | break; | |
2445 | ||
2446 | /* | |
2447 | * To keep our iteration location, move the marker | |
2448 | * forward. Since we're not holding hdr's hash lock, we | |
2449 | * must be very careful and not remove 'hdr' from the | |
2450 | * sublist. Otherwise, other consumers might mistake the | |
2451 | * 'hdr' as not being on a sublist when they call the | |
2452 | * multilist_link_active() function (they all rely on | |
2453 | * the hash lock protecting concurrent insertions and | |
2454 | * removals). multilist_sublist_move_forward() was | |
2455 | * specifically implemented to ensure this is the case | |
2456 | * (only 'marker' will be removed and re-inserted). | |
2457 | */ | |
2458 | multilist_sublist_move_forward(mls, marker); | |
2459 | ||
2460 | /* | |
2461 | * The only case where the b_spa field should ever be | |
2462 | * zero, is the marker headers inserted by | |
2463 | * arc_evict_state(). It's possible for multiple threads | |
2464 | * to be calling arc_evict_state() concurrently (e.g. | |
2465 | * dsl_pool_close() and zio_inject_fault()), so we must | |
2466 | * skip any markers we see from these other threads. | |
2467 | */ | |
2a432414 | 2468 | if (hdr->b_spa == 0) |
572e2857 BB |
2469 | continue; |
2470 | ||
ca0bf58d PS |
2471 | /* we're only interested in evicting buffers of a certain spa */ |
2472 | if (spa != 0 && hdr->b_spa != spa) { | |
2473 | ARCSTAT_BUMP(arcstat_evict_skip); | |
428870ff | 2474 | continue; |
ca0bf58d PS |
2475 | } |
2476 | ||
2477 | hash_lock = HDR_LOCK(hdr); | |
e8b96c60 MA |
2478 | |
2479 | /* | |
ca0bf58d PS |
2480 | * We aren't calling this function from any code path |
2481 | * that would already be holding a hash lock, so we're | |
2482 | * asserting on this assumption to be defensive in case | |
2483 | * this ever changes. Without this check, it would be | |
2484 | * possible to incorrectly increment arcstat_mutex_miss | |
2485 | * below (e.g. if the code changed such that we called | |
2486 | * this function with a hash lock held). | |
e8b96c60 | 2487 | */ |
ca0bf58d PS |
2488 | ASSERT(!MUTEX_HELD(hash_lock)); |
2489 | ||
34dc7c2f | 2490 | if (mutex_tryenter(hash_lock)) { |
ca0bf58d PS |
2491 | uint64_t evicted = arc_evict_hdr(hdr, hash_lock); |
2492 | mutex_exit(hash_lock); | |
34dc7c2f | 2493 | |
ca0bf58d | 2494 | bytes_evicted += evicted; |
34dc7c2f | 2495 | |
572e2857 | 2496 | /* |
ca0bf58d PS |
2497 | * If evicted is zero, arc_evict_hdr() must have |
2498 | * decided to skip this header, don't increment | |
2499 | * evict_count in this case. | |
572e2857 | 2500 | */ |
ca0bf58d PS |
2501 | if (evicted != 0) |
2502 | evict_count++; | |
2503 | ||
2504 | /* | |
2505 | * If arc_size isn't overflowing, signal any | |
2506 | * threads that might happen to be waiting. | |
2507 | * | |
2508 | * For each header evicted, we wake up a single | |
2509 | * thread. If we used cv_broadcast, we could | |
2510 | * wake up "too many" threads causing arc_size | |
2511 | * to significantly overflow arc_c; since | |
2512 | * arc_get_data_buf() doesn't check for overflow | |
2513 | * when it's woken up (it doesn't because it's | |
2514 | * possible for the ARC to be overflowing while | |
2515 | * full of un-evictable buffers, and the | |
2516 | * function should proceed in this case). | |
2517 | * | |
2518 | * If threads are left sleeping, due to not | |
2519 | * using cv_broadcast, they will be woken up | |
2520 | * just before arc_reclaim_thread() sleeps. | |
2521 | */ | |
2522 | mutex_enter(&arc_reclaim_lock); | |
2523 | if (!arc_is_overflowing()) | |
2524 | cv_signal(&arc_reclaim_waiters_cv); | |
2525 | mutex_exit(&arc_reclaim_lock); | |
e8b96c60 | 2526 | } else { |
ca0bf58d | 2527 | ARCSTAT_BUMP(arcstat_mutex_miss); |
e8b96c60 | 2528 | } |
34dc7c2f | 2529 | } |
34dc7c2f | 2530 | |
ca0bf58d | 2531 | multilist_sublist_unlock(mls); |
34dc7c2f | 2532 | |
ca0bf58d | 2533 | return (bytes_evicted); |
34dc7c2f BB |
2534 | } |
2535 | ||
ca0bf58d PS |
2536 | /* |
2537 | * Evict buffers from the given arc state, until we've removed the | |
2538 | * specified number of bytes. Move the removed buffers to the | |
2539 | * appropriate evict state. | |
2540 | * | |
2541 | * This function makes a "best effort". It skips over any buffers | |
2542 | * it can't get a hash_lock on, and so, may not catch all candidates. | |
2543 | * It may also return without evicting as much space as requested. | |
2544 | * | |
2545 | * If bytes is specified using the special value ARC_EVICT_ALL, this | |
2546 | * will evict all available (i.e. unlocked and evictable) buffers from | |
2547 | * the given arc state; which is used by arc_flush(). | |
2548 | */ | |
2549 | static uint64_t | |
2550 | arc_evict_state(arc_state_t *state, uint64_t spa, int64_t bytes, | |
2551 | arc_buf_contents_t type) | |
34dc7c2f | 2552 | { |
ca0bf58d PS |
2553 | uint64_t total_evicted = 0; |
2554 | multilist_t *ml = &state->arcs_list[type]; | |
2555 | int num_sublists; | |
2556 | arc_buf_hdr_t **markers; | |
2557 | int i; | |
2558 | ||
2559 | ASSERTV(if (bytes < 0) ASSERT(bytes == ARC_EVICT_ALL)); | |
2560 | ||
2561 | num_sublists = multilist_get_num_sublists(ml); | |
d164b209 BB |
2562 | |
2563 | /* | |
ca0bf58d PS |
2564 | * If we've tried to evict from each sublist, made some |
2565 | * progress, but still have not hit the target number of bytes | |
2566 | * to evict, we want to keep trying. The markers allow us to | |
2567 | * pick up where we left off for each individual sublist, rather | |
2568 | * than starting from the tail each time. | |
d164b209 | 2569 | */ |
ca0bf58d PS |
2570 | markers = kmem_zalloc(sizeof (*markers) * num_sublists, KM_SLEEP); |
2571 | for (i = 0; i < num_sublists; i++) { | |
2572 | multilist_sublist_t *mls; | |
34dc7c2f | 2573 | |
ca0bf58d PS |
2574 | markers[i] = kmem_cache_alloc(hdr_full_cache, KM_SLEEP); |
2575 | ||
2576 | /* | |
2577 | * A b_spa of 0 is used to indicate that this header is | |
2578 | * a marker. This fact is used in arc_adjust_type() and | |
2579 | * arc_evict_state_impl(). | |
2580 | */ | |
2581 | markers[i]->b_spa = 0; | |
34dc7c2f | 2582 | |
ca0bf58d PS |
2583 | mls = multilist_sublist_lock(ml, i); |
2584 | multilist_sublist_insert_tail(mls, markers[i]); | |
2585 | multilist_sublist_unlock(mls); | |
34dc7c2f BB |
2586 | } |
2587 | ||
d164b209 | 2588 | /* |
ca0bf58d PS |
2589 | * While we haven't hit our target number of bytes to evict, or |
2590 | * we're evicting all available buffers. | |
d164b209 | 2591 | */ |
ca0bf58d PS |
2592 | while (total_evicted < bytes || bytes == ARC_EVICT_ALL) { |
2593 | /* | |
2594 | * Start eviction using a randomly selected sublist, | |
2595 | * this is to try and evenly balance eviction across all | |
2596 | * sublists. Always starting at the same sublist | |
2597 | * (e.g. index 0) would cause evictions to favor certain | |
2598 | * sublists over others. | |
2599 | */ | |
2600 | int sublist_idx = multilist_get_random_index(ml); | |
2601 | uint64_t scan_evicted = 0; | |
34dc7c2f | 2602 | |
ca0bf58d PS |
2603 | for (i = 0; i < num_sublists; i++) { |
2604 | uint64_t bytes_remaining; | |
2605 | uint64_t bytes_evicted; | |
d164b209 | 2606 | |
ca0bf58d PS |
2607 | if (bytes == ARC_EVICT_ALL) |
2608 | bytes_remaining = ARC_EVICT_ALL; | |
2609 | else if (total_evicted < bytes) | |
2610 | bytes_remaining = bytes - total_evicted; | |
2611 | else | |
2612 | break; | |
34dc7c2f | 2613 | |
ca0bf58d PS |
2614 | bytes_evicted = arc_evict_state_impl(ml, sublist_idx, |
2615 | markers[sublist_idx], spa, bytes_remaining); | |
2616 | ||
2617 | scan_evicted += bytes_evicted; | |
2618 | total_evicted += bytes_evicted; | |
2619 | ||
2620 | /* we've reached the end, wrap to the beginning */ | |
2621 | if (++sublist_idx >= num_sublists) | |
2622 | sublist_idx = 0; | |
2623 | } | |
2624 | ||
2625 | /* | |
2626 | * If we didn't evict anything during this scan, we have | |
2627 | * no reason to believe we'll evict more during another | |
2628 | * scan, so break the loop. | |
2629 | */ | |
2630 | if (scan_evicted == 0) { | |
2631 | /* This isn't possible, let's make that obvious */ | |
2632 | ASSERT3S(bytes, !=, 0); | |
34dc7c2f | 2633 | |
ca0bf58d PS |
2634 | /* |
2635 | * When bytes is ARC_EVICT_ALL, the only way to | |
2636 | * break the loop is when scan_evicted is zero. | |
2637 | * In that case, we actually have evicted enough, | |
2638 | * so we don't want to increment the kstat. | |
2639 | */ | |
2640 | if (bytes != ARC_EVICT_ALL) { | |
2641 | ASSERT3S(total_evicted, <, bytes); | |
2642 | ARCSTAT_BUMP(arcstat_evict_not_enough); | |
2643 | } | |
d164b209 | 2644 | |
ca0bf58d PS |
2645 | break; |
2646 | } | |
d164b209 | 2647 | } |
34dc7c2f | 2648 | |
ca0bf58d PS |
2649 | for (i = 0; i < num_sublists; i++) { |
2650 | multilist_sublist_t *mls = multilist_sublist_lock(ml, i); | |
2651 | multilist_sublist_remove(mls, markers[i]); | |
2652 | multilist_sublist_unlock(mls); | |
34dc7c2f | 2653 | |
ca0bf58d | 2654 | kmem_cache_free(hdr_full_cache, markers[i]); |
34dc7c2f | 2655 | } |
ca0bf58d PS |
2656 | kmem_free(markers, sizeof (*markers) * num_sublists); |
2657 | ||
2658 | return (total_evicted); | |
2659 | } | |
2660 | ||
2661 | /* | |
2662 | * Flush all "evictable" data of the given type from the arc state | |
2663 | * specified. This will not evict any "active" buffers (i.e. referenced). | |
2664 | * | |
2665 | * When 'retry' is set to FALSE, the function will make a single pass | |
2666 | * over the state and evict any buffers that it can. Since it doesn't | |
2667 | * continually retry the eviction, it might end up leaving some buffers | |
2668 | * in the ARC due to lock misses. | |
2669 | * | |
2670 | * When 'retry' is set to TRUE, the function will continually retry the | |
2671 | * eviction until *all* evictable buffers have been removed from the | |
2672 | * state. As a result, if concurrent insertions into the state are | |
2673 | * allowed (e.g. if the ARC isn't shutting down), this function might | |
2674 | * wind up in an infinite loop, continually trying to evict buffers. | |
2675 | */ | |
2676 | static uint64_t | |
2677 | arc_flush_state(arc_state_t *state, uint64_t spa, arc_buf_contents_t type, | |
2678 | boolean_t retry) | |
2679 | { | |
2680 | uint64_t evicted = 0; | |
2681 | ||
2682 | while (state->arcs_lsize[type] != 0) { | |
2683 | evicted += arc_evict_state(state, spa, ARC_EVICT_ALL, type); | |
2684 | ||
2685 | if (!retry) | |
2686 | break; | |
2687 | } | |
2688 | ||
2689 | return (evicted); | |
34dc7c2f BB |
2690 | } |
2691 | ||
ab26409d | 2692 | /* |
f6046738 BB |
2693 | * Helper function for arc_prune() it is responsible for safely handling |
2694 | * the execution of a registered arc_prune_func_t. | |
ab26409d BB |
2695 | */ |
2696 | static void | |
f6046738 | 2697 | arc_prune_task(void *ptr) |
ab26409d | 2698 | { |
f6046738 BB |
2699 | arc_prune_t *ap = (arc_prune_t *)ptr; |
2700 | arc_prune_func_t *func = ap->p_pfunc; | |
ab26409d | 2701 | |
f6046738 BB |
2702 | if (func != NULL) |
2703 | func(ap->p_adjust, ap->p_private); | |
ab26409d | 2704 | |
f6046738 BB |
2705 | /* Callback unregistered concurrently with execution */ |
2706 | if (refcount_remove(&ap->p_refcnt, func) == 0) { | |
2707 | ASSERT(!list_link_active(&ap->p_node)); | |
2708 | refcount_destroy(&ap->p_refcnt); | |
2709 | kmem_free(ap, sizeof (*ap)); | |
2710 | } | |
2711 | } | |
ab26409d | 2712 | |
f6046738 BB |
2713 | /* |
2714 | * Notify registered consumers they must drop holds on a portion of the ARC | |
2715 | * buffered they reference. This provides a mechanism to ensure the ARC can | |
2716 | * honor the arc_meta_limit and reclaim otherwise pinned ARC buffers. This | |
2717 | * is analogous to dnlc_reduce_cache() but more generic. | |
2718 | * | |
2719 | * This operation is performed asyncronously so it may be safely called | |
ca67b33a | 2720 | * in the context of the arc_reclaim_thread(). A reference is taken here |
f6046738 BB |
2721 | * for each registered arc_prune_t and the arc_prune_task() is responsible |
2722 | * for releasing it once the registered arc_prune_func_t has completed. | |
2723 | */ | |
2724 | static void | |
2725 | arc_prune_async(int64_t adjust) | |
2726 | { | |
2727 | arc_prune_t *ap; | |
ab26409d | 2728 | |
f6046738 BB |
2729 | mutex_enter(&arc_prune_mtx); |
2730 | for (ap = list_head(&arc_prune_list); ap != NULL; | |
2731 | ap = list_next(&arc_prune_list, ap)) { | |
ab26409d | 2732 | |
f6046738 BB |
2733 | if (refcount_count(&ap->p_refcnt) >= 2) |
2734 | continue; | |
ab26409d | 2735 | |
f6046738 BB |
2736 | refcount_add(&ap->p_refcnt, ap->p_pfunc); |
2737 | ap->p_adjust = adjust; | |
2738 | taskq_dispatch(arc_prune_taskq, arc_prune_task, ap, TQ_SLEEP); | |
2739 | ARCSTAT_BUMP(arcstat_prune); | |
ab26409d | 2740 | } |
ab26409d BB |
2741 | mutex_exit(&arc_prune_mtx); |
2742 | } | |
2743 | ||
f6046738 BB |
2744 | static void |
2745 | arc_prune(int64_t adjust) | |
2746 | { | |
2747 | arc_prune_async(adjust); | |
2748 | taskq_wait_outstanding(arc_prune_taskq, 0); | |
2749 | } | |
2750 | ||
ca0bf58d PS |
2751 | /* |
2752 | * Evict the specified number of bytes from the state specified, | |
2753 | * restricting eviction to the spa and type given. This function | |
2754 | * prevents us from trying to evict more from a state's list than | |
2755 | * is "evictable", and to skip evicting altogether when passed a | |
2756 | * negative value for "bytes". In contrast, arc_evict_state() will | |
2757 | * evict everything it can, when passed a negative value for "bytes". | |
2758 | */ | |
2759 | static uint64_t | |
2760 | arc_adjust_impl(arc_state_t *state, uint64_t spa, int64_t bytes, | |
2761 | arc_buf_contents_t type) | |
2762 | { | |
2763 | int64_t delta; | |
2764 | ||
2765 | if (bytes > 0 && state->arcs_lsize[type] > 0) { | |
2766 | delta = MIN(state->arcs_lsize[type], bytes); | |
2767 | return (arc_evict_state(state, spa, delta, type)); | |
2768 | } | |
2769 | ||
2770 | return (0); | |
2771 | } | |
2772 | ||
2773 | /* | |
2774 | * The goal of this function is to evict enough meta data buffers from the | |
2775 | * ARC in order to enforce the arc_meta_limit. Achieving this is slightly | |
2776 | * more complicated than it appears because it is common for data buffers | |
2777 | * to have holds on meta data buffers. In addition, dnode meta data buffers | |
2778 | * will be held by the dnodes in the block preventing them from being freed. | |
2779 | * This means we can't simply traverse the ARC and expect to always find | |
2780 | * enough unheld meta data buffer to release. | |
2781 | * | |
2782 | * Therefore, this function has been updated to make alternating passes | |
2783 | * over the ARC releasing data buffers and then newly unheld meta data | |
2784 | * buffers. This ensures forward progress is maintained and arc_meta_used | |
2785 | * will decrease. Normally this is sufficient, but if required the ARC | |
2786 | * will call the registered prune callbacks causing dentry and inodes to | |
2787 | * be dropped from the VFS cache. This will make dnode meta data buffers | |
2788 | * available for reclaim. | |
2789 | */ | |
2790 | static uint64_t | |
f6046738 | 2791 | arc_adjust_meta_balanced(void) |
ca0bf58d PS |
2792 | { |
2793 | int64_t adjustmnt, delta, prune = 0; | |
2794 | uint64_t total_evicted = 0; | |
2795 | arc_buf_contents_t type = ARC_BUFC_DATA; | |
ca67b33a | 2796 | int restarts = MAX(zfs_arc_meta_adjust_restarts, 0); |
ca0bf58d PS |
2797 | |
2798 | restart: | |
2799 | /* | |
2800 | * This slightly differs than the way we evict from the mru in | |
2801 | * arc_adjust because we don't have a "target" value (i.e. no | |
2802 | * "meta" arc_p). As a result, I think we can completely | |
2803 | * cannibalize the metadata in the MRU before we evict the | |
2804 | * metadata from the MFU. I think we probably need to implement a | |
2805 | * "metadata arc_p" value to do this properly. | |
2806 | */ | |
2807 | adjustmnt = arc_meta_used - arc_meta_limit; | |
2808 | ||
2809 | if (adjustmnt > 0 && arc_mru->arcs_lsize[type] > 0) { | |
2810 | delta = MIN(arc_mru->arcs_lsize[type], adjustmnt); | |
2811 | total_evicted += arc_adjust_impl(arc_mru, 0, delta, type); | |
2812 | adjustmnt -= delta; | |
2813 | } | |
2814 | ||
2815 | /* | |
2816 | * We can't afford to recalculate adjustmnt here. If we do, | |
2817 | * new metadata buffers can sneak into the MRU or ANON lists, | |
2818 | * thus penalize the MFU metadata. Although the fudge factor is | |
2819 | * small, it has been empirically shown to be significant for | |
2820 | * certain workloads (e.g. creating many empty directories). As | |
2821 | * such, we use the original calculation for adjustmnt, and | |
2822 | * simply decrement the amount of data evicted from the MRU. | |
2823 | */ | |
2824 | ||
2825 | if (adjustmnt > 0 && arc_mfu->arcs_lsize[type] > 0) { | |
2826 | delta = MIN(arc_mfu->arcs_lsize[type], adjustmnt); | |
2827 | total_evicted += arc_adjust_impl(arc_mfu, 0, delta, type); | |
2828 | } | |
2829 | ||
2830 | adjustmnt = arc_meta_used - arc_meta_limit; | |
2831 | ||
2832 | if (adjustmnt > 0 && arc_mru_ghost->arcs_lsize[type] > 0) { | |
2833 | delta = MIN(adjustmnt, | |
2834 | arc_mru_ghost->arcs_lsize[type]); | |
2835 | total_evicted += arc_adjust_impl(arc_mru_ghost, 0, delta, type); | |
2836 | adjustmnt -= delta; | |
2837 | } | |
2838 | ||
2839 | if (adjustmnt > 0 && arc_mfu_ghost->arcs_lsize[type] > 0) { | |
2840 | delta = MIN(adjustmnt, | |
2841 | arc_mfu_ghost->arcs_lsize[type]); | |
2842 | total_evicted += arc_adjust_impl(arc_mfu_ghost, 0, delta, type); | |
2843 | } | |
2844 | ||
2845 | /* | |
2846 | * If after attempting to make the requested adjustment to the ARC | |
2847 | * the meta limit is still being exceeded then request that the | |
2848 | * higher layers drop some cached objects which have holds on ARC | |
2849 | * meta buffers. Requests to the upper layers will be made with | |
2850 | * increasingly large scan sizes until the ARC is below the limit. | |
2851 | */ | |
2852 | if (arc_meta_used > arc_meta_limit) { | |
2853 | if (type == ARC_BUFC_DATA) { | |
2854 | type = ARC_BUFC_METADATA; | |
2855 | } else { | |
2856 | type = ARC_BUFC_DATA; | |
2857 | ||
2858 | if (zfs_arc_meta_prune) { | |
2859 | prune += zfs_arc_meta_prune; | |
f6046738 | 2860 | arc_prune_async(prune); |
ca0bf58d PS |
2861 | } |
2862 | } | |
2863 | ||
2864 | if (restarts > 0) { | |
2865 | restarts--; | |
2866 | goto restart; | |
2867 | } | |
2868 | } | |
2869 | return (total_evicted); | |
2870 | } | |
2871 | ||
f6046738 BB |
2872 | /* |
2873 | * Evict metadata buffers from the cache, such that arc_meta_used is | |
2874 | * capped by the arc_meta_limit tunable. | |
2875 | */ | |
2876 | static uint64_t | |
2877 | arc_adjust_meta_only(void) | |
2878 | { | |
2879 | uint64_t total_evicted = 0; | |
2880 | int64_t target; | |
2881 | ||
2882 | /* | |
2883 | * If we're over the meta limit, we want to evict enough | |
2884 | * metadata to get back under the meta limit. We don't want to | |
2885 | * evict so much that we drop the MRU below arc_p, though. If | |
2886 | * we're over the meta limit more than we're over arc_p, we | |
2887 | * evict some from the MRU here, and some from the MFU below. | |
2888 | */ | |
2889 | target = MIN((int64_t)(arc_meta_used - arc_meta_limit), | |
36da08ef PS |
2890 | (int64_t)(refcount_count(&arc_anon->arcs_size) + |
2891 | refcount_count(&arc_mru->arcs_size) - arc_p)); | |
f6046738 BB |
2892 | |
2893 | total_evicted += arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_METADATA); | |
2894 | ||
2895 | /* | |
2896 | * Similar to the above, we want to evict enough bytes to get us | |
2897 | * below the meta limit, but not so much as to drop us below the | |
2898 | * space alloted to the MFU (which is defined as arc_c - arc_p). | |
2899 | */ | |
2900 | target = MIN((int64_t)(arc_meta_used - arc_meta_limit), | |
36da08ef | 2901 | (int64_t)(refcount_count(&arc_mfu->arcs_size) - (arc_c - arc_p))); |
f6046738 BB |
2902 | |
2903 | total_evicted += arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_METADATA); | |
2904 | ||
2905 | return (total_evicted); | |
2906 | } | |
2907 | ||
2908 | static uint64_t | |
2909 | arc_adjust_meta(void) | |
2910 | { | |
2911 | if (zfs_arc_meta_strategy == ARC_STRATEGY_META_ONLY) | |
2912 | return (arc_adjust_meta_only()); | |
2913 | else | |
2914 | return (arc_adjust_meta_balanced()); | |
2915 | } | |
2916 | ||
ca0bf58d PS |
2917 | /* |
2918 | * Return the type of the oldest buffer in the given arc state | |
2919 | * | |
2920 | * This function will select a random sublist of type ARC_BUFC_DATA and | |
2921 | * a random sublist of type ARC_BUFC_METADATA. The tail of each sublist | |
2922 | * is compared, and the type which contains the "older" buffer will be | |
2923 | * returned. | |
2924 | */ | |
2925 | static arc_buf_contents_t | |
2926 | arc_adjust_type(arc_state_t *state) | |
2927 | { | |
2928 | multilist_t *data_ml = &state->arcs_list[ARC_BUFC_DATA]; | |
2929 | multilist_t *meta_ml = &state->arcs_list[ARC_BUFC_METADATA]; | |
2930 | int data_idx = multilist_get_random_index(data_ml); | |
2931 | int meta_idx = multilist_get_random_index(meta_ml); | |
2932 | multilist_sublist_t *data_mls; | |
2933 | multilist_sublist_t *meta_mls; | |
2934 | arc_buf_contents_t type; | |
2935 | arc_buf_hdr_t *data_hdr; | |
2936 | arc_buf_hdr_t *meta_hdr; | |
2937 | ||
2938 | /* | |
2939 | * We keep the sublist lock until we're finished, to prevent | |
2940 | * the headers from being destroyed via arc_evict_state(). | |
2941 | */ | |
2942 | data_mls = multilist_sublist_lock(data_ml, data_idx); | |
2943 | meta_mls = multilist_sublist_lock(meta_ml, meta_idx); | |
2944 | ||
2945 | /* | |
2946 | * These two loops are to ensure we skip any markers that | |
2947 | * might be at the tail of the lists due to arc_evict_state(). | |
2948 | */ | |
2949 | ||
2950 | for (data_hdr = multilist_sublist_tail(data_mls); data_hdr != NULL; | |
2951 | data_hdr = multilist_sublist_prev(data_mls, data_hdr)) { | |
2952 | if (data_hdr->b_spa != 0) | |
2953 | break; | |
2954 | } | |
2955 | ||
2956 | for (meta_hdr = multilist_sublist_tail(meta_mls); meta_hdr != NULL; | |
2957 | meta_hdr = multilist_sublist_prev(meta_mls, meta_hdr)) { | |
2958 | if (meta_hdr->b_spa != 0) | |
2959 | break; | |
2960 | } | |
2961 | ||
2962 | if (data_hdr == NULL && meta_hdr == NULL) { | |
2963 | type = ARC_BUFC_DATA; | |
2964 | } else if (data_hdr == NULL) { | |
2965 | ASSERT3P(meta_hdr, !=, NULL); | |
2966 | type = ARC_BUFC_METADATA; | |
2967 | } else if (meta_hdr == NULL) { | |
2968 | ASSERT3P(data_hdr, !=, NULL); | |
2969 | type = ARC_BUFC_DATA; | |
2970 | } else { | |
2971 | ASSERT3P(data_hdr, !=, NULL); | |
2972 | ASSERT3P(meta_hdr, !=, NULL); | |
2973 | ||
2974 | /* The headers can't be on the sublist without an L1 header */ | |
2975 | ASSERT(HDR_HAS_L1HDR(data_hdr)); | |
2976 | ASSERT(HDR_HAS_L1HDR(meta_hdr)); | |
2977 | ||
2978 | if (data_hdr->b_l1hdr.b_arc_access < | |
2979 | meta_hdr->b_l1hdr.b_arc_access) { | |
2980 | type = ARC_BUFC_DATA; | |
2981 | } else { | |
2982 | type = ARC_BUFC_METADATA; | |
2983 | } | |
2984 | } | |
2985 | ||
2986 | multilist_sublist_unlock(meta_mls); | |
2987 | multilist_sublist_unlock(data_mls); | |
2988 | ||
2989 | return (type); | |
2990 | } | |
2991 | ||
2992 | /* | |
2993 | * Evict buffers from the cache, such that arc_size is capped by arc_c. | |
2994 | */ | |
2995 | static uint64_t | |
2996 | arc_adjust(void) | |
2997 | { | |
2998 | uint64_t total_evicted = 0; | |
2999 | uint64_t bytes; | |
3000 | int64_t target; | |
3001 | ||
3002 | /* | |
3003 | * If we're over arc_meta_limit, we want to correct that before | |
3004 | * potentially evicting data buffers below. | |
3005 | */ | |
3006 | total_evicted += arc_adjust_meta(); | |
3007 | ||
3008 | /* | |
3009 | * Adjust MRU size | |
3010 | * | |
3011 | * If we're over the target cache size, we want to evict enough | |
3012 | * from the list to get back to our target size. We don't want | |
3013 | * to evict too much from the MRU, such that it drops below | |
3014 | * arc_p. So, if we're over our target cache size more than | |
3015 | * the MRU is over arc_p, we'll evict enough to get back to | |
3016 | * arc_p here, and then evict more from the MFU below. | |
3017 | */ | |
3018 | target = MIN((int64_t)(arc_size - arc_c), | |
36da08ef PS |
3019 | (int64_t)(refcount_count(&arc_anon->arcs_size) + |
3020 | refcount_count(&arc_mru->arcs_size) + arc_meta_used - arc_p)); | |
ca0bf58d PS |
3021 | |
3022 | /* | |
3023 | * If we're below arc_meta_min, always prefer to evict data. | |
3024 | * Otherwise, try to satisfy the requested number of bytes to | |
3025 | * evict from the type which contains older buffers; in an | |
3026 | * effort to keep newer buffers in the cache regardless of their | |
3027 | * type. If we cannot satisfy the number of bytes from this | |
3028 | * type, spill over into the next type. | |
3029 | */ | |
3030 | if (arc_adjust_type(arc_mru) == ARC_BUFC_METADATA && | |
3031 | arc_meta_used > arc_meta_min) { | |
3032 | bytes = arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_METADATA); | |
3033 | total_evicted += bytes; | |
3034 | ||
3035 | /* | |
3036 | * If we couldn't evict our target number of bytes from | |
3037 | * metadata, we try to get the rest from data. | |
3038 | */ | |
3039 | target -= bytes; | |
3040 | ||
3041 | total_evicted += | |
3042 | arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_DATA); | |
3043 | } else { | |
3044 | bytes = arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_DATA); | |
3045 | total_evicted += bytes; | |
3046 | ||
3047 | /* | |
3048 | * If we couldn't evict our target number of bytes from | |
3049 | * data, we try to get the rest from metadata. | |
3050 | */ | |
3051 | target -= bytes; | |
3052 | ||
3053 | total_evicted += | |
3054 | arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_METADATA); | |
3055 | } | |
3056 | ||
3057 | /* | |
3058 | * Adjust MFU size | |
3059 | * | |
3060 | * Now that we've tried to evict enough from the MRU to get its | |
3061 | * size back to arc_p, if we're still above the target cache | |
3062 | * size, we evict the rest from the MFU. | |
3063 | */ | |
3064 | target = arc_size - arc_c; | |
3065 | ||
a7b10a93 | 3066 | if (arc_adjust_type(arc_mfu) == ARC_BUFC_METADATA && |
ca0bf58d PS |
3067 | arc_meta_used > arc_meta_min) { |
3068 | bytes = arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_METADATA); | |
3069 | total_evicted += bytes; | |
3070 | ||
3071 | /* | |
3072 | * If we couldn't evict our target number of bytes from | |
3073 | * metadata, we try to get the rest from data. | |
3074 | */ | |
3075 | target -= bytes; | |
3076 | ||
3077 | total_evicted += | |
3078 | arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_DATA); | |
3079 | } else { | |
3080 | bytes = arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_DATA); | |
3081 | total_evicted += bytes; | |
3082 | ||
3083 | /* | |
3084 | * If we couldn't evict our target number of bytes from | |
3085 | * data, we try to get the rest from data. | |
3086 | */ | |
3087 | target -= bytes; | |
3088 | ||
3089 | total_evicted += | |
3090 | arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_METADATA); | |
3091 | } | |
3092 | ||
3093 | /* | |
3094 | * Adjust ghost lists | |
3095 | * | |
3096 | * In addition to the above, the ARC also defines target values | |
3097 | * for the ghost lists. The sum of the mru list and mru ghost | |
3098 | * list should never exceed the target size of the cache, and | |
3099 | * the sum of the mru list, mfu list, mru ghost list, and mfu | |
3100 | * ghost list should never exceed twice the target size of the | |
3101 | * cache. The following logic enforces these limits on the ghost | |
3102 | * caches, and evicts from them as needed. | |
3103 | */ | |
36da08ef PS |
3104 | target = refcount_count(&arc_mru->arcs_size) + |
3105 | refcount_count(&arc_mru_ghost->arcs_size) - arc_c; | |
ca0bf58d PS |
3106 | |
3107 | bytes = arc_adjust_impl(arc_mru_ghost, 0, target, ARC_BUFC_DATA); | |
3108 | total_evicted += bytes; | |
3109 | ||
3110 | target -= bytes; | |
3111 | ||
3112 | total_evicted += | |
3113 | arc_adjust_impl(arc_mru_ghost, 0, target, ARC_BUFC_METADATA); | |
3114 | ||
3115 | /* | |
3116 | * We assume the sum of the mru list and mfu list is less than | |
3117 | * or equal to arc_c (we enforced this above), which means we | |
3118 | * can use the simpler of the two equations below: | |
3119 | * | |
3120 | * mru + mfu + mru ghost + mfu ghost <= 2 * arc_c | |
3121 | * mru ghost + mfu ghost <= arc_c | |
3122 | */ | |
36da08ef PS |
3123 | target = refcount_count(&arc_mru_ghost->arcs_size) + |
3124 | refcount_count(&arc_mfu_ghost->arcs_size) - arc_c; | |
ca0bf58d PS |
3125 | |
3126 | bytes = arc_adjust_impl(arc_mfu_ghost, 0, target, ARC_BUFC_DATA); | |
3127 | total_evicted += bytes; | |
3128 | ||
3129 | target -= bytes; | |
3130 | ||
3131 | total_evicted += | |
3132 | arc_adjust_impl(arc_mfu_ghost, 0, target, ARC_BUFC_METADATA); | |
3133 | ||
3134 | return (total_evicted); | |
3135 | } | |
3136 | ||
34dc7c2f BB |
3137 | static void |
3138 | arc_do_user_evicts(void) | |
3139 | { | |
ca0bf58d | 3140 | mutex_enter(&arc_user_evicts_lock); |
34dc7c2f BB |
3141 | while (arc_eviction_list != NULL) { |
3142 | arc_buf_t *buf = arc_eviction_list; | |
3143 | arc_eviction_list = buf->b_next; | |
428870ff | 3144 | mutex_enter(&buf->b_evict_lock); |
34dc7c2f | 3145 | buf->b_hdr = NULL; |
428870ff | 3146 | mutex_exit(&buf->b_evict_lock); |
ca0bf58d | 3147 | mutex_exit(&arc_user_evicts_lock); |
34dc7c2f BB |
3148 | |
3149 | if (buf->b_efunc != NULL) | |
bd089c54 | 3150 | VERIFY0(buf->b_efunc(buf->b_private)); |
34dc7c2f BB |
3151 | |
3152 | buf->b_efunc = NULL; | |
3153 | buf->b_private = NULL; | |
3154 | kmem_cache_free(buf_cache, buf); | |
ca0bf58d | 3155 | mutex_enter(&arc_user_evicts_lock); |
34dc7c2f | 3156 | } |
ca0bf58d | 3157 | mutex_exit(&arc_user_evicts_lock); |
34dc7c2f BB |
3158 | } |
3159 | ||
ca0bf58d PS |
3160 | void |
3161 | arc_flush(spa_t *spa, boolean_t retry) | |
ab26409d | 3162 | { |
ca0bf58d | 3163 | uint64_t guid = 0; |
94520ca4 | 3164 | |
bc888666 | 3165 | /* |
ca0bf58d PS |
3166 | * If retry is TRUE, a spa must not be specified since we have |
3167 | * no good way to determine if all of a spa's buffers have been | |
3168 | * evicted from an arc state. | |
bc888666 | 3169 | */ |
ca0bf58d | 3170 | ASSERT(!retry || spa == 0); |
d164b209 | 3171 | |
b9541d6b | 3172 | if (spa != NULL) |
3541dc6d | 3173 | guid = spa_load_guid(spa); |
d164b209 | 3174 | |
ca0bf58d PS |
3175 | (void) arc_flush_state(arc_mru, guid, ARC_BUFC_DATA, retry); |
3176 | (void) arc_flush_state(arc_mru, guid, ARC_BUFC_METADATA, retry); | |
3177 | ||
3178 | (void) arc_flush_state(arc_mfu, guid, ARC_BUFC_DATA, retry); | |
3179 | (void) arc_flush_state(arc_mfu, guid, ARC_BUFC_METADATA, retry); | |
3180 | ||
3181 | (void) arc_flush_state(arc_mru_ghost, guid, ARC_BUFC_DATA, retry); | |
3182 | (void) arc_flush_state(arc_mru_ghost, guid, ARC_BUFC_METADATA, retry); | |
34dc7c2f | 3183 | |
ca0bf58d PS |
3184 | (void) arc_flush_state(arc_mfu_ghost, guid, ARC_BUFC_DATA, retry); |
3185 | (void) arc_flush_state(arc_mfu_ghost, guid, ARC_BUFC_METADATA, retry); | |
34dc7c2f | 3186 | |
34dc7c2f | 3187 | arc_do_user_evicts(); |
34dc7c2f BB |
3188 | ASSERT(spa || arc_eviction_list == NULL); |
3189 | } | |
3190 | ||
34dc7c2f | 3191 | void |
ca67b33a | 3192 | arc_shrink(int64_t to_free) |
34dc7c2f BB |
3193 | { |
3194 | if (arc_c > arc_c_min) { | |
302f753f | 3195 | |
34dc7c2f BB |
3196 | if (arc_c > arc_c_min + to_free) |
3197 | atomic_add_64(&arc_c, -to_free); | |
3198 | else | |
3199 | arc_c = arc_c_min; | |
3200 | ||
ca67b33a | 3201 | atomic_add_64(&arc_p, -(arc_p >> arc_shrink_shift)); |
34dc7c2f BB |
3202 | if (arc_c > arc_size) |
3203 | arc_c = MAX(arc_size, arc_c_min); | |
3204 | if (arc_p > arc_c) | |
3205 | arc_p = (arc_c >> 1); | |
3206 | ASSERT(arc_c >= arc_c_min); | |
3207 | ASSERT((int64_t)arc_p >= 0); | |
3208 | } | |
3209 | ||
3210 | if (arc_size > arc_c) | |
ca0bf58d | 3211 | (void) arc_adjust(); |
34dc7c2f BB |
3212 | } |
3213 | ||
ca67b33a MA |
3214 | typedef enum free_memory_reason_t { |
3215 | FMR_UNKNOWN, | |
3216 | FMR_NEEDFREE, | |
3217 | FMR_LOTSFREE, | |
3218 | FMR_SWAPFS_MINFREE, | |
3219 | FMR_PAGES_PP_MAXIMUM, | |
3220 | FMR_HEAP_ARENA, | |
3221 | FMR_ZIO_ARENA, | |
3222 | } free_memory_reason_t; | |
3223 | ||
3224 | int64_t last_free_memory; | |
3225 | free_memory_reason_t last_free_reason; | |
3226 | ||
3227 | #ifdef _KERNEL | |
3228 | #ifdef __linux__ | |
3229 | /* | |
3230 | * expiration time for arc_no_grow set by direct memory reclaim. | |
3231 | */ | |
3232 | static clock_t arc_grow_time = 0; | |
3233 | #else | |
3234 | /* | |
3235 | * Additional reserve of pages for pp_reserve. | |
3236 | */ | |
3237 | int64_t arc_pages_pp_reserve = 64; | |
3238 | ||
3239 | /* | |
3240 | * Additional reserve of pages for swapfs. | |
3241 | */ | |
3242 | int64_t arc_swapfs_reserve = 64; | |
3243 | #endif | |
3244 | #endif /* _KERNEL */ | |
3245 | ||
3246 | /* | |
3247 | * Return the amount of memory that can be consumed before reclaim will be | |
3248 | * needed. Positive if there is sufficient free memory, negative indicates | |
3249 | * the amount of memory that needs to be freed up. | |
3250 | */ | |
3251 | static int64_t | |
3252 | arc_available_memory(void) | |
3253 | { | |
3254 | int64_t lowest = INT64_MAX; | |
3255 | free_memory_reason_t r = FMR_UNKNOWN; | |
3256 | ||
3257 | #ifdef _KERNEL | |
3258 | #ifdef __linux__ | |
3259 | /* | |
3260 | * Under Linux we are not allowed to directly interrogate the global | |
3261 | * memory state. Instead rely on observing that direct reclaim has | |
3262 | * recently occurred therefore the system must be low on memory. The | |
3263 | * exact values returned are not critical but should be small. | |
3264 | */ | |
3265 | if (ddi_time_after_eq(ddi_get_lbolt(), arc_grow_time)) | |
3266 | lowest = PAGE_SIZE; | |
3267 | else | |
3268 | lowest = -PAGE_SIZE; | |
3269 | #else | |
3270 | int64_t n; | |
3271 | ||
3272 | /* | |
3273 | * Platforms like illumos have greater visibility in to the memory | |
3274 | * subsystem and can return a more detailed analysis of memory. | |
3275 | */ | |
3276 | if (needfree > 0) { | |
3277 | n = PAGESIZE * (-needfree); | |
3278 | if (n < lowest) { | |
3279 | lowest = n; | |
3280 | r = FMR_NEEDFREE; | |
3281 | } | |
3282 | } | |
3283 | ||
3284 | /* | |
3285 | * check that we're out of range of the pageout scanner. It starts to | |
3286 | * schedule paging if freemem is less than lotsfree and needfree. | |
3287 | * lotsfree is the high-water mark for pageout, and needfree is the | |
3288 | * number of needed free pages. We add extra pages here to make sure | |
3289 | * the scanner doesn't start up while we're freeing memory. | |
3290 | */ | |
3291 | n = PAGESIZE * (freemem - lotsfree - needfree - desfree); | |
3292 | if (n < lowest) { | |
3293 | lowest = n; | |
3294 | r = FMR_LOTSFREE; | |
3295 | } | |
3296 | ||
3297 | /* | |
3298 | * check to make sure that swapfs has enough space so that anon | |
3299 | * reservations can still succeed. anon_resvmem() checks that the | |
3300 | * availrmem is greater than swapfs_minfree, and the number of reserved | |
3301 | * swap pages. We also add a bit of extra here just to prevent | |
3302 | * circumstances from getting really dire. | |
3303 | */ | |
3304 | n = PAGESIZE * (availrmem - swapfs_minfree - swapfs_reserve - | |
3305 | desfree - arc_swapfs_reserve); | |
3306 | if (n < lowest) { | |
3307 | lowest = n; | |
3308 | r = FMR_SWAPFS_MINFREE; | |
3309 | } | |
3310 | ||
3311 | ||
3312 | /* | |
3313 | * Check that we have enough availrmem that memory locking (e.g., via | |
3314 | * mlock(3C) or memcntl(2)) can still succeed. (pages_pp_maximum | |
3315 | * stores the number of pages that cannot be locked; when availrmem | |
3316 | * drops below pages_pp_maximum, page locking mechanisms such as | |
3317 | * page_pp_lock() will fail.) | |
3318 | */ | |
3319 | n = PAGESIZE * (availrmem - pages_pp_maximum - | |
3320 | arc_pages_pp_reserve); | |
3321 | if (n < lowest) { | |
3322 | lowest = n; | |
3323 | r = FMR_PAGES_PP_MAXIMUM; | |
3324 | } | |
3325 | ||
3326 | #if defined(__i386) | |
3327 | /* | |
3328 | * If we're on an i386 platform, it's possible that we'll exhaust the | |
3329 | * kernel heap space before we ever run out of available physical | |
3330 | * memory. Most checks of the size of the heap_area compare against | |
3331 | * tune.t_minarmem, which is the minimum available real memory that we | |
3332 | * can have in the system. However, this is generally fixed at 25 pages | |
3333 | * which is so low that it's useless. In this comparison, we seek to | |
3334 | * calculate the total heap-size, and reclaim if more than 3/4ths of the | |
3335 | * heap is allocated. (Or, in the calculation, if less than 1/4th is | |
3336 | * free) | |
3337 | */ | |
3338 | n = vmem_size(heap_arena, VMEM_FREE) - | |
3339 | (vmem_size(heap_arena, VMEM_FREE | VMEM_ALLOC) >> 2); | |
3340 | if (n < lowest) { | |
3341 | lowest = n; | |
3342 | r = FMR_HEAP_ARENA; | |
3343 | } | |
3344 | #endif | |
3345 | ||
3346 | /* | |
3347 | * If zio data pages are being allocated out of a separate heap segment, | |
3348 | * then enforce that the size of available vmem for this arena remains | |
3349 | * above about 1/16th free. | |
3350 | * | |
3351 | * Note: The 1/16th arena free requirement was put in place | |
3352 | * to aggressively evict memory from the arc in order to avoid | |
3353 | * memory fragmentation issues. | |
3354 | */ | |
3355 | if (zio_arena != NULL) { | |
3356 | n = vmem_size(zio_arena, VMEM_FREE) - | |
3357 | (vmem_size(zio_arena, VMEM_ALLOC) >> 4); | |
3358 | if (n < lowest) { | |
3359 | lowest = n; | |
3360 | r = FMR_ZIO_ARENA; | |
3361 | } | |
3362 | } | |
3363 | #endif /* __linux__ */ | |
3364 | #else | |
3365 | /* Every 100 calls, free a small amount */ | |
3366 | if (spa_get_random(100) == 0) | |
3367 | lowest = -1024; | |
3368 | #endif | |
3369 | ||
3370 | last_free_memory = lowest; | |
3371 | last_free_reason = r; | |
3372 | ||
3373 | return (lowest); | |
3374 | } | |
3375 | ||
3376 | /* | |
3377 | * Determine if the system is under memory pressure and is asking | |
3378 | * to reclaim memory. A return value of TRUE indicates that the system | |
3379 | * is under memory pressure and that the arc should adjust accordingly. | |
3380 | */ | |
3381 | static boolean_t | |
3382 | arc_reclaim_needed(void) | |
3383 | { | |
3384 | return (arc_available_memory() < 0); | |
3385 | } | |
3386 | ||
34dc7c2f | 3387 | static void |
ca67b33a | 3388 | arc_kmem_reap_now(void) |
34dc7c2f BB |
3389 | { |
3390 | size_t i; | |
3391 | kmem_cache_t *prev_cache = NULL; | |
3392 | kmem_cache_t *prev_data_cache = NULL; | |
3393 | extern kmem_cache_t *zio_buf_cache[]; | |
3394 | extern kmem_cache_t *zio_data_buf_cache[]; | |
669dedb3 | 3395 | extern kmem_cache_t *range_seg_cache; |
34dc7c2f | 3396 | |
f6046738 BB |
3397 | if ((arc_meta_used >= arc_meta_limit) && zfs_arc_meta_prune) { |
3398 | /* | |
3399 | * We are exceeding our meta-data cache limit. | |
3400 | * Prune some entries to release holds on meta-data. | |
3401 | */ | |
3402 | arc_prune(zfs_arc_meta_prune); | |
3403 | } | |
3404 | ||
34dc7c2f BB |
3405 | for (i = 0; i < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; i++) { |
3406 | if (zio_buf_cache[i] != prev_cache) { | |
3407 | prev_cache = zio_buf_cache[i]; | |
3408 | kmem_cache_reap_now(zio_buf_cache[i]); | |
3409 | } | |
3410 | if (zio_data_buf_cache[i] != prev_data_cache) { | |
3411 | prev_data_cache = zio_data_buf_cache[i]; | |
3412 | kmem_cache_reap_now(zio_data_buf_cache[i]); | |
3413 | } | |
3414 | } | |
ca0bf58d | 3415 | kmem_cache_reap_now(buf_cache); |
b9541d6b CW |
3416 | kmem_cache_reap_now(hdr_full_cache); |
3417 | kmem_cache_reap_now(hdr_l2only_cache); | |
669dedb3 | 3418 | kmem_cache_reap_now(range_seg_cache); |
ca67b33a MA |
3419 | |
3420 | if (zio_arena != NULL) { | |
3421 | /* | |
3422 | * Ask the vmem arena to reclaim unused memory from its | |
3423 | * quantum caches. | |
3424 | */ | |
3425 | vmem_qcache_reap(zio_arena); | |
3426 | } | |
34dc7c2f BB |
3427 | } |
3428 | ||
302f753f | 3429 | /* |
ca0bf58d PS |
3430 | * Threads can block in arc_get_data_buf() waiting for this thread to evict |
3431 | * enough data and signal them to proceed. When this happens, the threads in | |
3432 | * arc_get_data_buf() are sleeping while holding the hash lock for their | |
3433 | * particular arc header. Thus, we must be careful to never sleep on a | |
3434 | * hash lock in this thread. This is to prevent the following deadlock: | |
3435 | * | |
3436 | * - Thread A sleeps on CV in arc_get_data_buf() holding hash lock "L", | |
3437 | * waiting for the reclaim thread to signal it. | |
3438 | * | |
3439 | * - arc_reclaim_thread() tries to acquire hash lock "L" using mutex_enter, | |
3440 | * fails, and goes to sleep forever. | |
3441 | * | |
3442 | * This possible deadlock is avoided by always acquiring a hash lock | |
3443 | * using mutex_tryenter() from arc_reclaim_thread(). | |
302f753f | 3444 | */ |
34dc7c2f | 3445 | static void |
ca67b33a | 3446 | arc_reclaim_thread(void) |
34dc7c2f | 3447 | { |
ca67b33a MA |
3448 | fstrans_cookie_t cookie = spl_fstrans_mark(); |
3449 | clock_t growtime = 0; | |
34dc7c2f BB |
3450 | callb_cpr_t cpr; |
3451 | ||
ca0bf58d | 3452 | CALLB_CPR_INIT(&cpr, &arc_reclaim_lock, callb_generic_cpr, FTAG); |
34dc7c2f | 3453 | |
ca0bf58d | 3454 | mutex_enter(&arc_reclaim_lock); |
ca67b33a MA |
3455 | while (!arc_reclaim_thread_exit) { |
3456 | int64_t to_free; | |
3457 | int64_t free_memory = arc_available_memory(); | |
3458 | uint64_t evicted = 0; | |
302f753f | 3459 | |
ca67b33a | 3460 | arc_tuning_update(); |
34dc7c2f | 3461 | |
ca67b33a | 3462 | mutex_exit(&arc_reclaim_lock); |
34dc7c2f | 3463 | |
ca67b33a | 3464 | if (free_memory < 0) { |
34dc7c2f | 3465 | |
ca67b33a | 3466 | arc_no_grow = B_TRUE; |
b128c09f | 3467 | arc_warm = B_TRUE; |
34dc7c2f | 3468 | |
ca67b33a MA |
3469 | /* |
3470 | * Wait at least zfs_grow_retry (default 5) seconds | |
3471 | * before considering growing. | |
3472 | */ | |
3473 | growtime = ddi_get_lbolt() + (arc_grow_retry * hz); | |
6a8f9b6b | 3474 | |
ca67b33a | 3475 | arc_kmem_reap_now(); |
34dc7c2f | 3476 | |
ca67b33a MA |
3477 | /* |
3478 | * If we are still low on memory, shrink the ARC | |
3479 | * so that we have arc_shrink_min free space. | |
3480 | */ | |
3481 | free_memory = arc_available_memory(); | |
34dc7c2f | 3482 | |
ca67b33a MA |
3483 | to_free = (arc_c >> arc_shrink_shift) - free_memory; |
3484 | if (to_free > 0) { | |
3485 | #ifdef _KERNEL | |
3486 | to_free = MAX(to_free, ptob(needfree)); | |
3487 | #endif | |
3488 | arc_shrink(to_free); | |
3489 | } | |
3490 | } else if (free_memory < arc_c >> arc_no_grow_shift) { | |
3491 | arc_no_grow = B_TRUE; | |
3492 | } else if (ddi_get_lbolt() >= growtime) { | |
3493 | arc_no_grow = B_FALSE; | |
3494 | } | |
bce45ec9 | 3495 | |
ca67b33a | 3496 | evicted = arc_adjust(); |
bce45ec9 | 3497 | |
ca67b33a | 3498 | mutex_enter(&arc_reclaim_lock); |
bce45ec9 | 3499 | |
ca67b33a MA |
3500 | /* |
3501 | * If evicted is zero, we couldn't evict anything via | |
3502 | * arc_adjust(). This could be due to hash lock | |
3503 | * collisions, but more likely due to the majority of | |
3504 | * arc buffers being unevictable. Therefore, even if | |
3505 | * arc_size is above arc_c, another pass is unlikely to | |
3506 | * be helpful and could potentially cause us to enter an | |
3507 | * infinite loop. | |
3508 | */ | |
3509 | if (arc_size <= arc_c || evicted == 0) { | |
3510 | /* | |
3511 | * We're either no longer overflowing, or we | |
3512 | * can't evict anything more, so we should wake | |
3513 | * up any threads before we go to sleep. | |
3514 | */ | |
3515 | cv_broadcast(&arc_reclaim_waiters_cv); | |
bce45ec9 | 3516 | |
ca67b33a MA |
3517 | /* |
3518 | * Block until signaled, or after one second (we | |
3519 | * might need to perform arc_kmem_reap_now() | |
3520 | * even if we aren't being signalled) | |
3521 | */ | |
3522 | CALLB_CPR_SAFE_BEGIN(&cpr); | |
3523 | (void) cv_timedwait_sig(&arc_reclaim_thread_cv, | |
3524 | &arc_reclaim_lock, ddi_get_lbolt() + hz); | |
3525 | CALLB_CPR_SAFE_END(&cpr, &arc_reclaim_lock); | |
3526 | } | |
ca0bf58d | 3527 | } |
bce45ec9 | 3528 | |
ca67b33a | 3529 | arc_reclaim_thread_exit = FALSE; |
ca0bf58d PS |
3530 | cv_broadcast(&arc_reclaim_thread_cv); |
3531 | CALLB_CPR_EXIT(&cpr); /* drops arc_reclaim_lock */ | |
3532 | spl_fstrans_unmark(cookie); | |
3533 | thread_exit(); | |
3534 | } | |
3535 | ||
3536 | static void | |
3537 | arc_user_evicts_thread(void) | |
3538 | { | |
ca67b33a | 3539 | fstrans_cookie_t cookie = spl_fstrans_mark(); |
ca0bf58d | 3540 | callb_cpr_t cpr; |
bce45ec9 | 3541 | |
ca0bf58d | 3542 | CALLB_CPR_INIT(&cpr, &arc_user_evicts_lock, callb_generic_cpr, FTAG); |
bce45ec9 | 3543 | |
ca0bf58d PS |
3544 | mutex_enter(&arc_user_evicts_lock); |
3545 | while (!arc_user_evicts_thread_exit) { | |
3546 | mutex_exit(&arc_user_evicts_lock); | |
3547 | ||
3548 | arc_do_user_evicts(); | |
3549 | ||
3550 | /* | |
3551 | * This is necessary in order for the mdb ::arc dcmd to | |
3552 | * show up to date information. Since the ::arc command | |
3553 | * does not call the kstat's update function, without | |
3554 | * this call, the command may show stale stats for the | |
3555 | * anon, mru, mru_ghost, mfu, and mfu_ghost lists. Even | |
3556 | * with this change, the data might be up to 1 second | |
3557 | * out of date; but that should suffice. The arc_state_t | |
3558 | * structures can be queried directly if more accurate | |
3559 | * information is needed. | |
3560 | */ | |
3561 | if (arc_ksp != NULL) | |
3562 | arc_ksp->ks_update(arc_ksp, KSTAT_READ); | |
3563 | ||
3564 | mutex_enter(&arc_user_evicts_lock); | |
3565 | ||
3566 | /* | |
3567 | * Block until signaled, or after one second (we need to | |
3568 | * call the arc's kstat update function regularly). | |
3569 | */ | |
3570 | CALLB_CPR_SAFE_BEGIN(&cpr); | |
b64ccd6c | 3571 | (void) cv_timedwait_sig(&arc_user_evicts_cv, |
ca0bf58d PS |
3572 | &arc_user_evicts_lock, ddi_get_lbolt() + hz); |
3573 | CALLB_CPR_SAFE_END(&cpr, &arc_user_evicts_lock); | |
34dc7c2f BB |
3574 | } |
3575 | ||
ca0bf58d PS |
3576 | arc_user_evicts_thread_exit = FALSE; |
3577 | cv_broadcast(&arc_user_evicts_cv); | |
3578 | CALLB_CPR_EXIT(&cpr); /* drops arc_user_evicts_lock */ | |
40d06e3c | 3579 | spl_fstrans_unmark(cookie); |
34dc7c2f BB |
3580 | thread_exit(); |
3581 | } | |
3582 | ||
7cb67b45 BB |
3583 | #ifdef _KERNEL |
3584 | /* | |
302f753f BB |
3585 | * Determine the amount of memory eligible for eviction contained in the |
3586 | * ARC. All clean data reported by the ghost lists can always be safely | |
3587 | * evicted. Due to arc_c_min, the same does not hold for all clean data | |
3588 | * contained by the regular mru and mfu lists. | |
3589 | * | |
3590 | * In the case of the regular mru and mfu lists, we need to report as | |
3591 | * much clean data as possible, such that evicting that same reported | |
3592 | * data will not bring arc_size below arc_c_min. Thus, in certain | |
3593 | * circumstances, the total amount of clean data in the mru and mfu | |
3594 | * lists might not actually be evictable. | |
3595 | * | |
3596 | * The following two distinct cases are accounted for: | |
3597 | * | |
3598 | * 1. The sum of the amount of dirty data contained by both the mru and | |
3599 | * mfu lists, plus the ARC's other accounting (e.g. the anon list), | |
3600 | * is greater than or equal to arc_c_min. | |
3601 | * (i.e. amount of dirty data >= arc_c_min) | |
3602 | * | |
3603 | * This is the easy case; all clean data contained by the mru and mfu | |
3604 | * lists is evictable. Evicting all clean data can only drop arc_size | |
3605 | * to the amount of dirty data, which is greater than arc_c_min. | |
3606 | * | |
3607 | * 2. The sum of the amount of dirty data contained by both the mru and | |
3608 | * mfu lists, plus the ARC's other accounting (e.g. the anon list), | |
3609 | * is less than arc_c_min. | |
3610 | * (i.e. arc_c_min > amount of dirty data) | |
3611 | * | |
3612 | * 2.1. arc_size is greater than or equal arc_c_min. | |
3613 | * (i.e. arc_size >= arc_c_min > amount of dirty data) | |
3614 | * | |
3615 | * In this case, not all clean data from the regular mru and mfu | |
3616 | * lists is actually evictable; we must leave enough clean data | |
3617 | * to keep arc_size above arc_c_min. Thus, the maximum amount of | |
3618 | * evictable data from the two lists combined, is exactly the | |
3619 | * difference between arc_size and arc_c_min. | |
3620 | * | |
3621 | * 2.2. arc_size is less than arc_c_min | |
3622 | * (i.e. arc_c_min > arc_size > amount of dirty data) | |
3623 | * | |
3624 | * In this case, none of the data contained in the mru and mfu | |
3625 | * lists is evictable, even if it's clean. Since arc_size is | |
3626 | * already below arc_c_min, evicting any more would only | |
3627 | * increase this negative difference. | |
7cb67b45 | 3628 | */ |
302f753f BB |
3629 | static uint64_t |
3630 | arc_evictable_memory(void) { | |
3631 | uint64_t arc_clean = | |
3632 | arc_mru->arcs_lsize[ARC_BUFC_DATA] + | |
3633 | arc_mru->arcs_lsize[ARC_BUFC_METADATA] + | |
3634 | arc_mfu->arcs_lsize[ARC_BUFC_DATA] + | |
3635 | arc_mfu->arcs_lsize[ARC_BUFC_METADATA]; | |
3636 | uint64_t ghost_clean = | |
3637 | arc_mru_ghost->arcs_lsize[ARC_BUFC_DATA] + | |
3638 | arc_mru_ghost->arcs_lsize[ARC_BUFC_METADATA] + | |
3639 | arc_mfu_ghost->arcs_lsize[ARC_BUFC_DATA] + | |
3640 | arc_mfu_ghost->arcs_lsize[ARC_BUFC_METADATA]; | |
3641 | uint64_t arc_dirty = MAX((int64_t)arc_size - (int64_t)arc_clean, 0); | |
3642 | ||
3643 | if (arc_dirty >= arc_c_min) | |
3644 | return (ghost_clean + arc_clean); | |
3645 | ||
3646 | return (ghost_clean + MAX((int64_t)arc_size - (int64_t)arc_c_min, 0)); | |
3647 | } | |
3648 | ||
ed6e9cc2 TC |
3649 | /* |
3650 | * If sc->nr_to_scan is zero, the caller is requesting a query of the | |
3651 | * number of objects which can potentially be freed. If it is nonzero, | |
3652 | * the request is to free that many objects. | |
3653 | * | |
3654 | * Linux kernels >= 3.12 have the count_objects and scan_objects callbacks | |
3655 | * in struct shrinker and also require the shrinker to return the number | |
3656 | * of objects freed. | |
3657 | * | |
3658 | * Older kernels require the shrinker to return the number of freeable | |
3659 | * objects following the freeing of nr_to_free. | |
3660 | */ | |
3661 | static spl_shrinker_t | |
7e7baeca | 3662 | __arc_shrinker_func(struct shrinker *shrink, struct shrink_control *sc) |
7cb67b45 | 3663 | { |
ed6e9cc2 | 3664 | int64_t pages; |
7cb67b45 | 3665 | |
302f753f BB |
3666 | /* The arc is considered warm once reclaim has occurred */ |
3667 | if (unlikely(arc_warm == B_FALSE)) | |
3668 | arc_warm = B_TRUE; | |
7cb67b45 | 3669 | |
302f753f | 3670 | /* Return the potential number of reclaimable pages */ |
ed6e9cc2 | 3671 | pages = btop((int64_t)arc_evictable_memory()); |
302f753f BB |
3672 | if (sc->nr_to_scan == 0) |
3673 | return (pages); | |
3fd70ee6 BB |
3674 | |
3675 | /* Not allowed to perform filesystem reclaim */ | |
7e7baeca | 3676 | if (!(sc->gfp_mask & __GFP_FS)) |
ed6e9cc2 | 3677 | return (SHRINK_STOP); |
3fd70ee6 | 3678 | |
7cb67b45 | 3679 | /* Reclaim in progress */ |
ca0bf58d | 3680 | if (mutex_tryenter(&arc_reclaim_lock) == 0) |
ed6e9cc2 | 3681 | return (SHRINK_STOP); |
7cb67b45 | 3682 | |
ca0bf58d PS |
3683 | mutex_exit(&arc_reclaim_lock); |
3684 | ||
302f753f BB |
3685 | /* |
3686 | * Evict the requested number of pages by shrinking arc_c the | |
3687 | * requested amount. If there is nothing left to evict just | |
3688 | * reap whatever we can from the various arc slabs. | |
3689 | */ | |
3690 | if (pages > 0) { | |
ca67b33a MA |
3691 | arc_shrink(ptob(sc->nr_to_scan)); |
3692 | arc_kmem_reap_now(); | |
ed6e9cc2 TC |
3693 | #ifdef HAVE_SPLIT_SHRINKER_CALLBACK |
3694 | pages = MAX(pages - btop(arc_evictable_memory()), 0); | |
3695 | #else | |
1e3cb67b | 3696 | pages = btop(arc_evictable_memory()); |
ed6e9cc2 | 3697 | #endif |
302f753f | 3698 | } else { |
ca67b33a | 3699 | arc_kmem_reap_now(); |
ed6e9cc2 | 3700 | pages = SHRINK_STOP; |
302f753f BB |
3701 | } |
3702 | ||
ca0bf58d PS |
3703 | /* |
3704 | * We've reaped what we can, wake up threads. | |
3705 | */ | |
3706 | cv_broadcast(&arc_reclaim_waiters_cv); | |
3707 | ||
302f753f BB |
3708 | /* |
3709 | * When direct reclaim is observed it usually indicates a rapid | |
3710 | * increase in memory pressure. This occurs because the kswapd | |
3711 | * threads were unable to asynchronously keep enough free memory | |
3712 | * available. In this case set arc_no_grow to briefly pause arc | |
3713 | * growth to avoid compounding the memory pressure. | |
3714 | */ | |
7cb67b45 | 3715 | if (current_is_kswapd()) { |
302f753f | 3716 | ARCSTAT_BUMP(arcstat_memory_indirect_count); |
7cb67b45 | 3717 | } else { |
302f753f | 3718 | arc_no_grow = B_TRUE; |
bce45ec9 | 3719 | arc_grow_time = ddi_get_lbolt() + (zfs_arc_grow_retry * hz); |
302f753f | 3720 | ARCSTAT_BUMP(arcstat_memory_direct_count); |
7cb67b45 BB |
3721 | } |
3722 | ||
1e3cb67b | 3723 | return (pages); |
7cb67b45 | 3724 | } |
7e7baeca | 3725 | SPL_SHRINKER_CALLBACK_WRAPPER(arc_shrinker_func); |
7cb67b45 BB |
3726 | |
3727 | SPL_SHRINKER_DECLARE(arc_shrinker, arc_shrinker_func, DEFAULT_SEEKS); | |
3728 | #endif /* _KERNEL */ | |
3729 | ||
34dc7c2f BB |
3730 | /* |
3731 | * Adapt arc info given the number of bytes we are trying to add and | |
3732 | * the state that we are comming from. This function is only called | |
3733 | * when we are adding new content to the cache. | |
3734 | */ | |
3735 | static void | |
3736 | arc_adapt(int bytes, arc_state_t *state) | |
3737 | { | |
3738 | int mult; | |
728d6ae9 | 3739 | uint64_t arc_p_min = (arc_c >> arc_p_min_shift); |
36da08ef PS |
3740 | int64_t mrug_size = refcount_count(&arc_mru_ghost->arcs_size); |
3741 | int64_t mfug_size = refcount_count(&arc_mfu_ghost->arcs_size); | |
34dc7c2f BB |
3742 | |
3743 | if (state == arc_l2c_only) | |
3744 | return; | |
3745 | ||
3746 | ASSERT(bytes > 0); | |
3747 | /* | |
3748 | * Adapt the target size of the MRU list: | |
3749 | * - if we just hit in the MRU ghost list, then increase | |
3750 | * the target size of the MRU list. | |
3751 | * - if we just hit in the MFU ghost list, then increase | |
3752 | * the target size of the MFU list by decreasing the | |
3753 | * target size of the MRU list. | |
3754 | */ | |
3755 | if (state == arc_mru_ghost) { | |
36da08ef | 3756 | mult = (mrug_size >= mfug_size) ? 1 : (mfug_size / mrug_size); |
62422785 PS |
3757 | if (!zfs_arc_p_dampener_disable) |
3758 | mult = MIN(mult, 10); /* avoid wild arc_p adjustment */ | |
34dc7c2f | 3759 | |
728d6ae9 | 3760 | arc_p = MIN(arc_c - arc_p_min, arc_p + bytes * mult); |
34dc7c2f | 3761 | } else if (state == arc_mfu_ghost) { |
d164b209 BB |
3762 | uint64_t delta; |
3763 | ||
36da08ef | 3764 | mult = (mfug_size >= mrug_size) ? 1 : (mrug_size / mfug_size); |
62422785 PS |
3765 | if (!zfs_arc_p_dampener_disable) |
3766 | mult = MIN(mult, 10); | |
34dc7c2f | 3767 | |
d164b209 | 3768 | delta = MIN(bytes * mult, arc_p); |
728d6ae9 | 3769 | arc_p = MAX(arc_p_min, arc_p - delta); |
34dc7c2f BB |
3770 | } |
3771 | ASSERT((int64_t)arc_p >= 0); | |
3772 | ||
ca67b33a MA |
3773 | if (arc_reclaim_needed()) { |
3774 | cv_signal(&arc_reclaim_thread_cv); | |
3775 | return; | |
3776 | } | |
3777 | ||
34dc7c2f BB |
3778 | if (arc_no_grow) |
3779 | return; | |
3780 | ||
3781 | if (arc_c >= arc_c_max) | |
3782 | return; | |
3783 | ||
3784 | /* | |
3785 | * If we're within (2 * maxblocksize) bytes of the target | |
3786 | * cache size, increment the target cache size | |
3787 | */ | |
121b3cae TC |
3788 | VERIFY3U(arc_c, >=, 2ULL << SPA_MAXBLOCKSHIFT); |
3789 | if (arc_size >= arc_c - (2ULL << SPA_MAXBLOCKSHIFT)) { | |
34dc7c2f BB |
3790 | atomic_add_64(&arc_c, (int64_t)bytes); |
3791 | if (arc_c > arc_c_max) | |
3792 | arc_c = arc_c_max; | |
3793 | else if (state == arc_anon) | |
3794 | atomic_add_64(&arc_p, (int64_t)bytes); | |
3795 | if (arc_p > arc_c) | |
3796 | arc_p = arc_c; | |
3797 | } | |
3798 | ASSERT((int64_t)arc_p >= 0); | |
3799 | } | |
3800 | ||
3801 | /* | |
ca0bf58d PS |
3802 | * Check if arc_size has grown past our upper threshold, determined by |
3803 | * zfs_arc_overflow_shift. | |
34dc7c2f | 3804 | */ |
ca0bf58d PS |
3805 | static boolean_t |
3806 | arc_is_overflowing(void) | |
34dc7c2f | 3807 | { |
ca0bf58d PS |
3808 | /* Always allow at least one block of overflow */ |
3809 | uint64_t overflow = MAX(SPA_MAXBLOCKSIZE, | |
3810 | arc_c >> zfs_arc_overflow_shift); | |
34dc7c2f | 3811 | |
ca0bf58d | 3812 | return (arc_size >= arc_c + overflow); |
34dc7c2f BB |
3813 | } |
3814 | ||
3815 | /* | |
ca0bf58d PS |
3816 | * The buffer, supplied as the first argument, needs a data block. If we |
3817 | * are hitting the hard limit for the cache size, we must sleep, waiting | |
3818 | * for the eviction thread to catch up. If we're past the target size | |
3819 | * but below the hard limit, we'll only signal the reclaim thread and | |
3820 | * continue on. | |
34dc7c2f BB |
3821 | */ |
3822 | static void | |
3823 | arc_get_data_buf(arc_buf_t *buf) | |
3824 | { | |
b9541d6b | 3825 | arc_state_t *state = buf->b_hdr->b_l1hdr.b_state; |
34dc7c2f | 3826 | uint64_t size = buf->b_hdr->b_size; |
b9541d6b | 3827 | arc_buf_contents_t type = arc_buf_type(buf->b_hdr); |
34dc7c2f BB |
3828 | |
3829 | arc_adapt(size, state); | |
3830 | ||
3831 | /* | |
ca0bf58d PS |
3832 | * If arc_size is currently overflowing, and has grown past our |
3833 | * upper limit, we must be adding data faster than the evict | |
3834 | * thread can evict. Thus, to ensure we don't compound the | |
3835 | * problem by adding more data and forcing arc_size to grow even | |
3836 | * further past it's target size, we halt and wait for the | |
3837 | * eviction thread to catch up. | |
3838 | * | |
3839 | * It's also possible that the reclaim thread is unable to evict | |
3840 | * enough buffers to get arc_size below the overflow limit (e.g. | |
3841 | * due to buffers being un-evictable, or hash lock collisions). | |
3842 | * In this case, we want to proceed regardless if we're | |
3843 | * overflowing; thus we don't use a while loop here. | |
34dc7c2f | 3844 | */ |
ca0bf58d PS |
3845 | if (arc_is_overflowing()) { |
3846 | mutex_enter(&arc_reclaim_lock); | |
3847 | ||
3848 | /* | |
3849 | * Now that we've acquired the lock, we may no longer be | |
3850 | * over the overflow limit, lets check. | |
3851 | * | |
3852 | * We're ignoring the case of spurious wake ups. If that | |
3853 | * were to happen, it'd let this thread consume an ARC | |
3854 | * buffer before it should have (i.e. before we're under | |
3855 | * the overflow limit and were signalled by the reclaim | |
3856 | * thread). As long as that is a rare occurrence, it | |
3857 | * shouldn't cause any harm. | |
3858 | */ | |
3859 | if (arc_is_overflowing()) { | |
3860 | cv_signal(&arc_reclaim_thread_cv); | |
3861 | cv_wait(&arc_reclaim_waiters_cv, &arc_reclaim_lock); | |
34dc7c2f | 3862 | } |
34dc7c2f | 3863 | |
ca0bf58d | 3864 | mutex_exit(&arc_reclaim_lock); |
34dc7c2f | 3865 | } |
ab26409d | 3866 | |
da8ccd0e | 3867 | if (type == ARC_BUFC_METADATA) { |
ca0bf58d PS |
3868 | buf->b_data = zio_buf_alloc(size); |
3869 | arc_space_consume(size, ARC_SPACE_META); | |
3870 | } else { | |
3871 | ASSERT(type == ARC_BUFC_DATA); | |
3872 | buf->b_data = zio_data_buf_alloc(size); | |
3873 | arc_space_consume(size, ARC_SPACE_DATA); | |
da8ccd0e PS |
3874 | } |
3875 | ||
34dc7c2f BB |
3876 | /* |
3877 | * Update the state size. Note that ghost states have a | |
3878 | * "ghost size" and so don't need to be updated. | |
3879 | */ | |
b9541d6b | 3880 | if (!GHOST_STATE(buf->b_hdr->b_l1hdr.b_state)) { |
34dc7c2f | 3881 | arc_buf_hdr_t *hdr = buf->b_hdr; |
36da08ef | 3882 | arc_state_t *state = hdr->b_l1hdr.b_state; |
34dc7c2f | 3883 | |
36da08ef | 3884 | (void) refcount_add_many(&state->arcs_size, size, buf); |
ca0bf58d PS |
3885 | |
3886 | /* | |
3887 | * If this is reached via arc_read, the link is | |
3888 | * protected by the hash lock. If reached via | |
3889 | * arc_buf_alloc, the header should not be accessed by | |
3890 | * any other thread. And, if reached via arc_read_done, | |
3891 | * the hash lock will protect it if it's found in the | |
3892 | * hash table; otherwise no other thread should be | |
3893 | * trying to [add|remove]_reference it. | |
3894 | */ | |
3895 | if (multilist_link_active(&hdr->b_l1hdr.b_arc_node)) { | |
b9541d6b CW |
3896 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); |
3897 | atomic_add_64(&hdr->b_l1hdr.b_state->arcs_lsize[type], | |
3898 | size); | |
34dc7c2f BB |
3899 | } |
3900 | /* | |
3901 | * If we are growing the cache, and we are adding anonymous | |
3902 | * data, and we have outgrown arc_p, update arc_p | |
3903 | */ | |
ca0bf58d | 3904 | if (arc_size < arc_c && hdr->b_l1hdr.b_state == arc_anon && |
36da08ef PS |
3905 | (refcount_count(&arc_anon->arcs_size) + |
3906 | refcount_count(&arc_mru->arcs_size) > arc_p)) | |
34dc7c2f BB |
3907 | arc_p = MIN(arc_c, arc_p + size); |
3908 | } | |
3909 | } | |
3910 | ||
3911 | /* | |
3912 | * This routine is called whenever a buffer is accessed. | |
3913 | * NOTE: the hash lock is dropped in this function. | |
3914 | */ | |
3915 | static void | |
2a432414 | 3916 | arc_access(arc_buf_hdr_t *hdr, kmutex_t *hash_lock) |
34dc7c2f | 3917 | { |
428870ff BB |
3918 | clock_t now; |
3919 | ||
34dc7c2f | 3920 | ASSERT(MUTEX_HELD(hash_lock)); |
b9541d6b | 3921 | ASSERT(HDR_HAS_L1HDR(hdr)); |
34dc7c2f | 3922 | |
b9541d6b | 3923 | if (hdr->b_l1hdr.b_state == arc_anon) { |
34dc7c2f BB |
3924 | /* |
3925 | * This buffer is not in the cache, and does not | |
3926 | * appear in our "ghost" list. Add the new buffer | |
3927 | * to the MRU state. | |
3928 | */ | |
3929 | ||
b9541d6b CW |
3930 | ASSERT0(hdr->b_l1hdr.b_arc_access); |
3931 | hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); | |
2a432414 GW |
3932 | DTRACE_PROBE1(new_state__mru, arc_buf_hdr_t *, hdr); |
3933 | arc_change_state(arc_mru, hdr, hash_lock); | |
34dc7c2f | 3934 | |
b9541d6b | 3935 | } else if (hdr->b_l1hdr.b_state == arc_mru) { |
428870ff BB |
3936 | now = ddi_get_lbolt(); |
3937 | ||
34dc7c2f BB |
3938 | /* |
3939 | * If this buffer is here because of a prefetch, then either: | |
3940 | * - clear the flag if this is a "referencing" read | |
3941 | * (any subsequent access will bump this into the MFU state). | |
3942 | * or | |
3943 | * - move the buffer to the head of the list if this is | |
3944 | * another prefetch (to make it less likely to be evicted). | |
3945 | */ | |
b9541d6b CW |
3946 | if (HDR_PREFETCH(hdr)) { |
3947 | if (refcount_count(&hdr->b_l1hdr.b_refcnt) == 0) { | |
ca0bf58d PS |
3948 | /* link protected by hash lock */ |
3949 | ASSERT(multilist_link_active( | |
b9541d6b | 3950 | &hdr->b_l1hdr.b_arc_node)); |
34dc7c2f | 3951 | } else { |
2a432414 | 3952 | hdr->b_flags &= ~ARC_FLAG_PREFETCH; |
b9541d6b | 3953 | atomic_inc_32(&hdr->b_l1hdr.b_mru_hits); |
34dc7c2f BB |
3954 | ARCSTAT_BUMP(arcstat_mru_hits); |
3955 | } | |
b9541d6b | 3956 | hdr->b_l1hdr.b_arc_access = now; |
34dc7c2f BB |
3957 | return; |
3958 | } | |
3959 | ||
3960 | /* | |
3961 | * This buffer has been "accessed" only once so far, | |
3962 | * but it is still in the cache. Move it to the MFU | |
3963 | * state. | |
3964 | */ | |
b9541d6b CW |
3965 | if (ddi_time_after(now, hdr->b_l1hdr.b_arc_access + |
3966 | ARC_MINTIME)) { | |
34dc7c2f BB |
3967 | /* |
3968 | * More than 125ms have passed since we | |
3969 | * instantiated this buffer. Move it to the | |
3970 | * most frequently used state. | |
3971 | */ | |
b9541d6b | 3972 | hdr->b_l1hdr.b_arc_access = now; |
2a432414 GW |
3973 | DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); |
3974 | arc_change_state(arc_mfu, hdr, hash_lock); | |
34dc7c2f | 3975 | } |
b9541d6b | 3976 | atomic_inc_32(&hdr->b_l1hdr.b_mru_hits); |
34dc7c2f | 3977 | ARCSTAT_BUMP(arcstat_mru_hits); |
b9541d6b | 3978 | } else if (hdr->b_l1hdr.b_state == arc_mru_ghost) { |
34dc7c2f BB |
3979 | arc_state_t *new_state; |
3980 | /* | |
3981 | * This buffer has been "accessed" recently, but | |
3982 | * was evicted from the cache. Move it to the | |
3983 | * MFU state. | |
3984 | */ | |
3985 | ||
b9541d6b | 3986 | if (HDR_PREFETCH(hdr)) { |
34dc7c2f | 3987 | new_state = arc_mru; |
b9541d6b | 3988 | if (refcount_count(&hdr->b_l1hdr.b_refcnt) > 0) |
2a432414 GW |
3989 | hdr->b_flags &= ~ARC_FLAG_PREFETCH; |
3990 | DTRACE_PROBE1(new_state__mru, arc_buf_hdr_t *, hdr); | |
34dc7c2f BB |
3991 | } else { |
3992 | new_state = arc_mfu; | |
2a432414 | 3993 | DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); |
34dc7c2f BB |
3994 | } |
3995 | ||
b9541d6b | 3996 | hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); |
2a432414 | 3997 | arc_change_state(new_state, hdr, hash_lock); |
34dc7c2f | 3998 | |
b9541d6b | 3999 | atomic_inc_32(&hdr->b_l1hdr.b_mru_ghost_hits); |
34dc7c2f | 4000 | ARCSTAT_BUMP(arcstat_mru_ghost_hits); |
b9541d6b | 4001 | } else if (hdr->b_l1hdr.b_state == arc_mfu) { |
34dc7c2f BB |
4002 | /* |
4003 | * This buffer has been accessed more than once and is | |
4004 | * still in the cache. Keep it in the MFU state. | |
4005 | * | |
4006 | * NOTE: an add_reference() that occurred when we did | |
4007 | * the arc_read() will have kicked this off the list. | |
4008 | * If it was a prefetch, we will explicitly move it to | |
4009 | * the head of the list now. | |
4010 | */ | |
b9541d6b CW |
4011 | if ((HDR_PREFETCH(hdr)) != 0) { |
4012 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); | |
ca0bf58d PS |
4013 | /* link protected by hash_lock */ |
4014 | ASSERT(multilist_link_active(&hdr->b_l1hdr.b_arc_node)); | |
34dc7c2f | 4015 | } |
b9541d6b | 4016 | atomic_inc_32(&hdr->b_l1hdr.b_mfu_hits); |
34dc7c2f | 4017 | ARCSTAT_BUMP(arcstat_mfu_hits); |
b9541d6b CW |
4018 | hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); |
4019 | } else if (hdr->b_l1hdr.b_state == arc_mfu_ghost) { | |
34dc7c2f BB |
4020 | arc_state_t *new_state = arc_mfu; |
4021 | /* | |
4022 | * This buffer has been accessed more than once but has | |
4023 | * been evicted from the cache. Move it back to the | |
4024 | * MFU state. | |
4025 | */ | |
4026 | ||
b9541d6b | 4027 | if (HDR_PREFETCH(hdr)) { |
34dc7c2f BB |
4028 | /* |
4029 | * This is a prefetch access... | |
4030 | * move this block back to the MRU state. | |
4031 | */ | |
b9541d6b | 4032 | ASSERT0(refcount_count(&hdr->b_l1hdr.b_refcnt)); |
34dc7c2f BB |
4033 | new_state = arc_mru; |
4034 | } | |
4035 | ||
b9541d6b | 4036 | hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); |
2a432414 GW |
4037 | DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); |
4038 | arc_change_state(new_state, hdr, hash_lock); | |
34dc7c2f | 4039 | |
b9541d6b | 4040 | atomic_inc_32(&hdr->b_l1hdr.b_mfu_ghost_hits); |
34dc7c2f | 4041 | ARCSTAT_BUMP(arcstat_mfu_ghost_hits); |
b9541d6b | 4042 | } else if (hdr->b_l1hdr.b_state == arc_l2c_only) { |
34dc7c2f BB |
4043 | /* |
4044 | * This buffer is on the 2nd Level ARC. | |
4045 | */ | |
4046 | ||
b9541d6b | 4047 | hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); |
2a432414 GW |
4048 | DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); |
4049 | arc_change_state(arc_mfu, hdr, hash_lock); | |
34dc7c2f | 4050 | } else { |
b9541d6b CW |
4051 | cmn_err(CE_PANIC, "invalid arc state 0x%p", |
4052 | hdr->b_l1hdr.b_state); | |
34dc7c2f BB |
4053 | } |
4054 | } | |
4055 | ||
4056 | /* a generic arc_done_func_t which you can use */ | |
4057 | /* ARGSUSED */ | |
4058 | void | |
4059 | arc_bcopy_func(zio_t *zio, arc_buf_t *buf, void *arg) | |
4060 | { | |
428870ff BB |
4061 | if (zio == NULL || zio->io_error == 0) |
4062 | bcopy(buf->b_data, arg, buf->b_hdr->b_size); | |
13fe0198 | 4063 | VERIFY(arc_buf_remove_ref(buf, arg)); |
34dc7c2f BB |
4064 | } |
4065 | ||
4066 | /* a generic arc_done_func_t */ | |
4067 | void | |
4068 | arc_getbuf_func(zio_t *zio, arc_buf_t *buf, void *arg) | |
4069 | { | |
4070 | arc_buf_t **bufp = arg; | |
4071 | if (zio && zio->io_error) { | |
13fe0198 | 4072 | VERIFY(arc_buf_remove_ref(buf, arg)); |
34dc7c2f BB |
4073 | *bufp = NULL; |
4074 | } else { | |
4075 | *bufp = buf; | |
428870ff | 4076 | ASSERT(buf->b_data); |
34dc7c2f BB |
4077 | } |
4078 | } | |
4079 | ||
4080 | static void | |
4081 | arc_read_done(zio_t *zio) | |
4082 | { | |
9b67f605 | 4083 | arc_buf_hdr_t *hdr; |
34dc7c2f BB |
4084 | arc_buf_t *buf; |
4085 | arc_buf_t *abuf; /* buffer we're assigning to callback */ | |
9b67f605 | 4086 | kmutex_t *hash_lock = NULL; |
34dc7c2f BB |
4087 | arc_callback_t *callback_list, *acb; |
4088 | int freeable = FALSE; | |
4089 | ||
4090 | buf = zio->io_private; | |
4091 | hdr = buf->b_hdr; | |
4092 | ||
4093 | /* | |
4094 | * The hdr was inserted into hash-table and removed from lists | |
4095 | * prior to starting I/O. We should find this header, since | |
4096 | * it's in the hash table, and it should be legit since it's | |
4097 | * not possible to evict it during the I/O. The only possible | |
4098 | * reason for it not to be found is if we were freed during the | |
4099 | * read. | |
4100 | */ | |
9b67f605 MA |
4101 | if (HDR_IN_HASH_TABLE(hdr)) { |
4102 | arc_buf_hdr_t *found; | |
4103 | ||
4104 | ASSERT3U(hdr->b_birth, ==, BP_PHYSICAL_BIRTH(zio->io_bp)); | |
4105 | ASSERT3U(hdr->b_dva.dva_word[0], ==, | |
4106 | BP_IDENTITY(zio->io_bp)->dva_word[0]); | |
4107 | ASSERT3U(hdr->b_dva.dva_word[1], ==, | |
4108 | BP_IDENTITY(zio->io_bp)->dva_word[1]); | |
4109 | ||
4110 | found = buf_hash_find(hdr->b_spa, zio->io_bp, | |
4111 | &hash_lock); | |
4112 | ||
4113 | ASSERT((found == NULL && HDR_FREED_IN_READ(hdr) && | |
4114 | hash_lock == NULL) || | |
4115 | (found == hdr && | |
4116 | DVA_EQUAL(&hdr->b_dva, BP_IDENTITY(zio->io_bp))) || | |
4117 | (found == hdr && HDR_L2_READING(hdr))); | |
4118 | } | |
34dc7c2f | 4119 | |
2a432414 | 4120 | hdr->b_flags &= ~ARC_FLAG_L2_EVICTED; |
b9541d6b | 4121 | if (l2arc_noprefetch && HDR_PREFETCH(hdr)) |
2a432414 | 4122 | hdr->b_flags &= ~ARC_FLAG_L2CACHE; |
34dc7c2f BB |
4123 | |
4124 | /* byteswap if necessary */ | |
b9541d6b | 4125 | callback_list = hdr->b_l1hdr.b_acb; |
34dc7c2f | 4126 | ASSERT(callback_list != NULL); |
428870ff | 4127 | if (BP_SHOULD_BYTESWAP(zio->io_bp) && zio->io_error == 0) { |
9ae529ec CS |
4128 | dmu_object_byteswap_t bswap = |
4129 | DMU_OT_BYTESWAP(BP_GET_TYPE(zio->io_bp)); | |
b01615d5 RY |
4130 | if (BP_GET_LEVEL(zio->io_bp) > 0) |
4131 | byteswap_uint64_array(buf->b_data, hdr->b_size); | |
4132 | else | |
4133 | dmu_ot_byteswap[bswap].ob_func(buf->b_data, hdr->b_size); | |
b128c09f | 4134 | } |
34dc7c2f BB |
4135 | |
4136 | arc_cksum_compute(buf, B_FALSE); | |
498877ba | 4137 | arc_buf_watch(buf); |
34dc7c2f | 4138 | |
b9541d6b CW |
4139 | if (hash_lock && zio->io_error == 0 && |
4140 | hdr->b_l1hdr.b_state == arc_anon) { | |
428870ff BB |
4141 | /* |
4142 | * Only call arc_access on anonymous buffers. This is because | |
4143 | * if we've issued an I/O for an evicted buffer, we've already | |
4144 | * called arc_access (to prevent any simultaneous readers from | |
4145 | * getting confused). | |
4146 | */ | |
4147 | arc_access(hdr, hash_lock); | |
4148 | } | |
4149 | ||
34dc7c2f BB |
4150 | /* create copies of the data buffer for the callers */ |
4151 | abuf = buf; | |
4152 | for (acb = callback_list; acb; acb = acb->acb_next) { | |
4153 | if (acb->acb_done) { | |
1eb5bfa3 GW |
4154 | if (abuf == NULL) { |
4155 | ARCSTAT_BUMP(arcstat_duplicate_reads); | |
34dc7c2f | 4156 | abuf = arc_buf_clone(buf); |
1eb5bfa3 | 4157 | } |
34dc7c2f BB |
4158 | acb->acb_buf = abuf; |
4159 | abuf = NULL; | |
4160 | } | |
4161 | } | |
b9541d6b | 4162 | hdr->b_l1hdr.b_acb = NULL; |
2a432414 | 4163 | hdr->b_flags &= ~ARC_FLAG_IO_IN_PROGRESS; |
34dc7c2f | 4164 | ASSERT(!HDR_BUF_AVAILABLE(hdr)); |
428870ff BB |
4165 | if (abuf == buf) { |
4166 | ASSERT(buf->b_efunc == NULL); | |
b9541d6b | 4167 | ASSERT(hdr->b_l1hdr.b_datacnt == 1); |
2a432414 | 4168 | hdr->b_flags |= ARC_FLAG_BUF_AVAILABLE; |
428870ff | 4169 | } |
34dc7c2f | 4170 | |
b9541d6b CW |
4171 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt) || |
4172 | callback_list != NULL); | |
34dc7c2f BB |
4173 | |
4174 | if (zio->io_error != 0) { | |
2a432414 | 4175 | hdr->b_flags |= ARC_FLAG_IO_ERROR; |
b9541d6b | 4176 | if (hdr->b_l1hdr.b_state != arc_anon) |
34dc7c2f BB |
4177 | arc_change_state(arc_anon, hdr, hash_lock); |
4178 | if (HDR_IN_HASH_TABLE(hdr)) | |
4179 | buf_hash_remove(hdr); | |
b9541d6b | 4180 | freeable = refcount_is_zero(&hdr->b_l1hdr.b_refcnt); |
34dc7c2f BB |
4181 | } |
4182 | ||
4183 | /* | |
4184 | * Broadcast before we drop the hash_lock to avoid the possibility | |
4185 | * that the hdr (and hence the cv) might be freed before we get to | |
4186 | * the cv_broadcast(). | |
4187 | */ | |
b9541d6b | 4188 | cv_broadcast(&hdr->b_l1hdr.b_cv); |
34dc7c2f | 4189 | |
b9541d6b | 4190 | if (hash_lock != NULL) { |
34dc7c2f BB |
4191 | mutex_exit(hash_lock); |
4192 | } else { | |
4193 | /* | |
4194 | * This block was freed while we waited for the read to | |
4195 | * complete. It has been removed from the hash table and | |
4196 | * moved to the anonymous state (so that it won't show up | |
4197 | * in the cache). | |
4198 | */ | |
b9541d6b CW |
4199 | ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon); |
4200 | freeable = refcount_is_zero(&hdr->b_l1hdr.b_refcnt); | |
34dc7c2f BB |
4201 | } |
4202 | ||
4203 | /* execute each callback and free its structure */ | |
4204 | while ((acb = callback_list) != NULL) { | |
4205 | if (acb->acb_done) | |
4206 | acb->acb_done(zio, acb->acb_buf, acb->acb_private); | |
4207 | ||
4208 | if (acb->acb_zio_dummy != NULL) { | |
4209 | acb->acb_zio_dummy->io_error = zio->io_error; | |
4210 | zio_nowait(acb->acb_zio_dummy); | |
4211 | } | |
4212 | ||
4213 | callback_list = acb->acb_next; | |
4214 | kmem_free(acb, sizeof (arc_callback_t)); | |
4215 | } | |
4216 | ||
4217 | if (freeable) | |
4218 | arc_hdr_destroy(hdr); | |
4219 | } | |
4220 | ||
4221 | /* | |
5c839890 | 4222 | * "Read" the block at the specified DVA (in bp) via the |
34dc7c2f BB |
4223 | * cache. If the block is found in the cache, invoke the provided |
4224 | * callback immediately and return. Note that the `zio' parameter | |
4225 | * in the callback will be NULL in this case, since no IO was | |
4226 | * required. If the block is not in the cache pass the read request | |
4227 | * on to the spa with a substitute callback function, so that the | |
4228 | * requested block will be added to the cache. | |
4229 | * | |
4230 | * If a read request arrives for a block that has a read in-progress, | |
4231 | * either wait for the in-progress read to complete (and return the | |
4232 | * results); or, if this is a read with a "done" func, add a record | |
4233 | * to the read to invoke the "done" func when the read completes, | |
4234 | * and return; or just return. | |
4235 | * | |
4236 | * arc_read_done() will invoke all the requested "done" functions | |
4237 | * for readers of this block. | |
4238 | */ | |
4239 | int | |
294f6806 | 4240 | arc_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, arc_done_func_t *done, |
2a432414 GW |
4241 | void *private, zio_priority_t priority, int zio_flags, |
4242 | arc_flags_t *arc_flags, const zbookmark_phys_t *zb) | |
34dc7c2f | 4243 | { |
9b67f605 | 4244 | arc_buf_hdr_t *hdr = NULL; |
d4ed6673 | 4245 | arc_buf_t *buf = NULL; |
9b67f605 | 4246 | kmutex_t *hash_lock = NULL; |
34dc7c2f | 4247 | zio_t *rzio; |
3541dc6d | 4248 | uint64_t guid = spa_load_guid(spa); |
1421c891 | 4249 | int rc = 0; |
34dc7c2f | 4250 | |
9b67f605 MA |
4251 | ASSERT(!BP_IS_EMBEDDED(bp) || |
4252 | BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA); | |
4253 | ||
34dc7c2f | 4254 | top: |
9b67f605 MA |
4255 | if (!BP_IS_EMBEDDED(bp)) { |
4256 | /* | |
4257 | * Embedded BP's have no DVA and require no I/O to "read". | |
4258 | * Create an anonymous arc buf to back it. | |
4259 | */ | |
4260 | hdr = buf_hash_find(guid, bp, &hash_lock); | |
4261 | } | |
4262 | ||
b9541d6b | 4263 | if (hdr != NULL && HDR_HAS_L1HDR(hdr) && hdr->b_l1hdr.b_datacnt > 0) { |
34dc7c2f | 4264 | |
2a432414 | 4265 | *arc_flags |= ARC_FLAG_CACHED; |
34dc7c2f BB |
4266 | |
4267 | if (HDR_IO_IN_PROGRESS(hdr)) { | |
4268 | ||
2a432414 | 4269 | if (*arc_flags & ARC_FLAG_WAIT) { |
b9541d6b | 4270 | cv_wait(&hdr->b_l1hdr.b_cv, hash_lock); |
34dc7c2f BB |
4271 | mutex_exit(hash_lock); |
4272 | goto top; | |
4273 | } | |
2a432414 | 4274 | ASSERT(*arc_flags & ARC_FLAG_NOWAIT); |
34dc7c2f BB |
4275 | |
4276 | if (done) { | |
4277 | arc_callback_t *acb = NULL; | |
4278 | ||
4279 | acb = kmem_zalloc(sizeof (arc_callback_t), | |
79c76d5b | 4280 | KM_SLEEP); |
34dc7c2f BB |
4281 | acb->acb_done = done; |
4282 | acb->acb_private = private; | |
34dc7c2f BB |
4283 | if (pio != NULL) |
4284 | acb->acb_zio_dummy = zio_null(pio, | |
d164b209 | 4285 | spa, NULL, NULL, NULL, zio_flags); |
34dc7c2f BB |
4286 | |
4287 | ASSERT(acb->acb_done != NULL); | |
b9541d6b CW |
4288 | acb->acb_next = hdr->b_l1hdr.b_acb; |
4289 | hdr->b_l1hdr.b_acb = acb; | |
34dc7c2f BB |
4290 | add_reference(hdr, hash_lock, private); |
4291 | mutex_exit(hash_lock); | |
1421c891 | 4292 | goto out; |
34dc7c2f BB |
4293 | } |
4294 | mutex_exit(hash_lock); | |
1421c891 | 4295 | goto out; |
34dc7c2f BB |
4296 | } |
4297 | ||
b9541d6b CW |
4298 | ASSERT(hdr->b_l1hdr.b_state == arc_mru || |
4299 | hdr->b_l1hdr.b_state == arc_mfu); | |
34dc7c2f BB |
4300 | |
4301 | if (done) { | |
4302 | add_reference(hdr, hash_lock, private); | |
4303 | /* | |
4304 | * If this block is already in use, create a new | |
4305 | * copy of the data so that we will be guaranteed | |
4306 | * that arc_release() will always succeed. | |
4307 | */ | |
b9541d6b | 4308 | buf = hdr->b_l1hdr.b_buf; |
34dc7c2f BB |
4309 | ASSERT(buf); |
4310 | ASSERT(buf->b_data); | |
4311 | if (HDR_BUF_AVAILABLE(hdr)) { | |
4312 | ASSERT(buf->b_efunc == NULL); | |
2a432414 | 4313 | hdr->b_flags &= ~ARC_FLAG_BUF_AVAILABLE; |
34dc7c2f BB |
4314 | } else { |
4315 | buf = arc_buf_clone(buf); | |
4316 | } | |
428870ff | 4317 | |
2a432414 | 4318 | } else if (*arc_flags & ARC_FLAG_PREFETCH && |
b9541d6b | 4319 | refcount_count(&hdr->b_l1hdr.b_refcnt) == 0) { |
2a432414 | 4320 | hdr->b_flags |= ARC_FLAG_PREFETCH; |
34dc7c2f BB |
4321 | } |
4322 | DTRACE_PROBE1(arc__hit, arc_buf_hdr_t *, hdr); | |
4323 | arc_access(hdr, hash_lock); | |
2a432414 GW |
4324 | if (*arc_flags & ARC_FLAG_L2CACHE) |
4325 | hdr->b_flags |= ARC_FLAG_L2CACHE; | |
4326 | if (*arc_flags & ARC_FLAG_L2COMPRESS) | |
4327 | hdr->b_flags |= ARC_FLAG_L2COMPRESS; | |
34dc7c2f BB |
4328 | mutex_exit(hash_lock); |
4329 | ARCSTAT_BUMP(arcstat_hits); | |
b9541d6b CW |
4330 | ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr), |
4331 | demand, prefetch, !HDR_ISTYPE_METADATA(hdr), | |
34dc7c2f BB |
4332 | data, metadata, hits); |
4333 | ||
4334 | if (done) | |
4335 | done(NULL, buf, private); | |
4336 | } else { | |
4337 | uint64_t size = BP_GET_LSIZE(bp); | |
9b67f605 | 4338 | arc_callback_t *acb; |
b128c09f | 4339 | vdev_t *vd = NULL; |
a117a6d6 | 4340 | uint64_t addr = 0; |
d164b209 | 4341 | boolean_t devw = B_FALSE; |
0ed212dc | 4342 | enum zio_compress b_compress = ZIO_COMPRESS_OFF; |
b9541d6b | 4343 | int32_t b_asize = 0; |
34dc7c2f | 4344 | |
5f6d0b6f BB |
4345 | /* |
4346 | * Gracefully handle a damaged logical block size as a | |
4347 | * checksum error by passing a dummy zio to the done callback. | |
4348 | */ | |
f1512ee6 | 4349 | if (size > spa_maxblocksize(spa)) { |
5f6d0b6f BB |
4350 | if (done) { |
4351 | rzio = zio_null(pio, spa, NULL, | |
4352 | NULL, NULL, zio_flags); | |
4353 | rzio->io_error = ECKSUM; | |
4354 | done(rzio, buf, private); | |
4355 | zio_nowait(rzio); | |
4356 | } | |
4357 | rc = ECKSUM; | |
4358 | goto out; | |
4359 | } | |
4360 | ||
34dc7c2f BB |
4361 | if (hdr == NULL) { |
4362 | /* this block is not in the cache */ | |
9b67f605 | 4363 | arc_buf_hdr_t *exists = NULL; |
34dc7c2f BB |
4364 | arc_buf_contents_t type = BP_GET_BUFC_TYPE(bp); |
4365 | buf = arc_buf_alloc(spa, size, private, type); | |
4366 | hdr = buf->b_hdr; | |
9b67f605 MA |
4367 | if (!BP_IS_EMBEDDED(bp)) { |
4368 | hdr->b_dva = *BP_IDENTITY(bp); | |
4369 | hdr->b_birth = BP_PHYSICAL_BIRTH(bp); | |
9b67f605 MA |
4370 | exists = buf_hash_insert(hdr, &hash_lock); |
4371 | } | |
4372 | if (exists != NULL) { | |
34dc7c2f BB |
4373 | /* somebody beat us to the hash insert */ |
4374 | mutex_exit(hash_lock); | |
428870ff | 4375 | buf_discard_identity(hdr); |
34dc7c2f BB |
4376 | (void) arc_buf_remove_ref(buf, private); |
4377 | goto top; /* restart the IO request */ | |
4378 | } | |
2a432414 | 4379 | |
34dc7c2f | 4380 | /* if this is a prefetch, we don't have a reference */ |
2a432414 | 4381 | if (*arc_flags & ARC_FLAG_PREFETCH) { |
34dc7c2f BB |
4382 | (void) remove_reference(hdr, hash_lock, |
4383 | private); | |
2a432414 | 4384 | hdr->b_flags |= ARC_FLAG_PREFETCH; |
34dc7c2f | 4385 | } |
2a432414 GW |
4386 | if (*arc_flags & ARC_FLAG_L2CACHE) |
4387 | hdr->b_flags |= ARC_FLAG_L2CACHE; | |
4388 | if (*arc_flags & ARC_FLAG_L2COMPRESS) | |
4389 | hdr->b_flags |= ARC_FLAG_L2COMPRESS; | |
34dc7c2f | 4390 | if (BP_GET_LEVEL(bp) > 0) |
2a432414 | 4391 | hdr->b_flags |= ARC_FLAG_INDIRECT; |
34dc7c2f | 4392 | } else { |
b9541d6b CW |
4393 | /* |
4394 | * This block is in the ghost cache. If it was L2-only | |
4395 | * (and thus didn't have an L1 hdr), we realloc the | |
4396 | * header to add an L1 hdr. | |
4397 | */ | |
4398 | if (!HDR_HAS_L1HDR(hdr)) { | |
4399 | hdr = arc_hdr_realloc(hdr, hdr_l2only_cache, | |
4400 | hdr_full_cache); | |
4401 | } | |
4402 | ||
4403 | ASSERT(GHOST_STATE(hdr->b_l1hdr.b_state)); | |
34dc7c2f | 4404 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); |
b9541d6b | 4405 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); |
ca0bf58d | 4406 | ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); |
34dc7c2f BB |
4407 | |
4408 | /* if this is a prefetch, we don't have a reference */ | |
2a432414 GW |
4409 | if (*arc_flags & ARC_FLAG_PREFETCH) |
4410 | hdr->b_flags |= ARC_FLAG_PREFETCH; | |
34dc7c2f BB |
4411 | else |
4412 | add_reference(hdr, hash_lock, private); | |
2a432414 GW |
4413 | if (*arc_flags & ARC_FLAG_L2CACHE) |
4414 | hdr->b_flags |= ARC_FLAG_L2CACHE; | |
4415 | if (*arc_flags & ARC_FLAG_L2COMPRESS) | |
4416 | hdr->b_flags |= ARC_FLAG_L2COMPRESS; | |
34dc7c2f BB |
4417 | buf = kmem_cache_alloc(buf_cache, KM_PUSHPAGE); |
4418 | buf->b_hdr = hdr; | |
4419 | buf->b_data = NULL; | |
4420 | buf->b_efunc = NULL; | |
4421 | buf->b_private = NULL; | |
4422 | buf->b_next = NULL; | |
b9541d6b CW |
4423 | hdr->b_l1hdr.b_buf = buf; |
4424 | ASSERT0(hdr->b_l1hdr.b_datacnt); | |
4425 | hdr->b_l1hdr.b_datacnt = 1; | |
428870ff BB |
4426 | arc_get_data_buf(buf); |
4427 | arc_access(hdr, hash_lock); | |
34dc7c2f BB |
4428 | } |
4429 | ||
b9541d6b | 4430 | ASSERT(!GHOST_STATE(hdr->b_l1hdr.b_state)); |
428870ff | 4431 | |
79c76d5b | 4432 | acb = kmem_zalloc(sizeof (arc_callback_t), KM_SLEEP); |
34dc7c2f BB |
4433 | acb->acb_done = done; |
4434 | acb->acb_private = private; | |
34dc7c2f | 4435 | |
b9541d6b CW |
4436 | ASSERT(hdr->b_l1hdr.b_acb == NULL); |
4437 | hdr->b_l1hdr.b_acb = acb; | |
2a432414 | 4438 | hdr->b_flags |= ARC_FLAG_IO_IN_PROGRESS; |
34dc7c2f | 4439 | |
b9541d6b CW |
4440 | if (HDR_HAS_L2HDR(hdr) && |
4441 | (vd = hdr->b_l2hdr.b_dev->l2ad_vdev) != NULL) { | |
4442 | devw = hdr->b_l2hdr.b_dev->l2ad_writing; | |
4443 | addr = hdr->b_l2hdr.b_daddr; | |
4444 | b_compress = HDR_GET_COMPRESS(hdr); | |
4445 | b_asize = hdr->b_l2hdr.b_asize; | |
b128c09f BB |
4446 | /* |
4447 | * Lock out device removal. | |
4448 | */ | |
4449 | if (vdev_is_dead(vd) || | |
4450 | !spa_config_tryenter(spa, SCL_L2ARC, vd, RW_READER)) | |
4451 | vd = NULL; | |
4452 | } | |
4453 | ||
9b67f605 MA |
4454 | if (hash_lock != NULL) |
4455 | mutex_exit(hash_lock); | |
b128c09f | 4456 | |
e49f1e20 WA |
4457 | /* |
4458 | * At this point, we have a level 1 cache miss. Try again in | |
4459 | * L2ARC if possible. | |
4460 | */ | |
34dc7c2f | 4461 | ASSERT3U(hdr->b_size, ==, size); |
428870ff | 4462 | DTRACE_PROBE4(arc__miss, arc_buf_hdr_t *, hdr, blkptr_t *, bp, |
5dbd68a3 | 4463 | uint64_t, size, zbookmark_phys_t *, zb); |
34dc7c2f | 4464 | ARCSTAT_BUMP(arcstat_misses); |
b9541d6b CW |
4465 | ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr), |
4466 | demand, prefetch, !HDR_ISTYPE_METADATA(hdr), | |
34dc7c2f BB |
4467 | data, metadata, misses); |
4468 | ||
d164b209 | 4469 | if (vd != NULL && l2arc_ndev != 0 && !(l2arc_norw && devw)) { |
34dc7c2f BB |
4470 | /* |
4471 | * Read from the L2ARC if the following are true: | |
b128c09f BB |
4472 | * 1. The L2ARC vdev was previously cached. |
4473 | * 2. This buffer still has L2ARC metadata. | |
4474 | * 3. This buffer isn't currently writing to the L2ARC. | |
4475 | * 4. The L2ARC entry wasn't evicted, which may | |
4476 | * also have invalidated the vdev. | |
d164b209 | 4477 | * 5. This isn't prefetch and l2arc_noprefetch is set. |
34dc7c2f | 4478 | */ |
b9541d6b | 4479 | if (HDR_HAS_L2HDR(hdr) && |
d164b209 BB |
4480 | !HDR_L2_WRITING(hdr) && !HDR_L2_EVICTED(hdr) && |
4481 | !(l2arc_noprefetch && HDR_PREFETCH(hdr))) { | |
34dc7c2f BB |
4482 | l2arc_read_callback_t *cb; |
4483 | ||
4484 | DTRACE_PROBE1(l2arc__hit, arc_buf_hdr_t *, hdr); | |
4485 | ARCSTAT_BUMP(arcstat_l2_hits); | |
b9541d6b | 4486 | atomic_inc_32(&hdr->b_l2hdr.b_hits); |
34dc7c2f | 4487 | |
34dc7c2f | 4488 | cb = kmem_zalloc(sizeof (l2arc_read_callback_t), |
79c76d5b | 4489 | KM_SLEEP); |
34dc7c2f BB |
4490 | cb->l2rcb_buf = buf; |
4491 | cb->l2rcb_spa = spa; | |
4492 | cb->l2rcb_bp = *bp; | |
4493 | cb->l2rcb_zb = *zb; | |
b128c09f | 4494 | cb->l2rcb_flags = zio_flags; |
0ed212dc | 4495 | cb->l2rcb_compress = b_compress; |
34dc7c2f | 4496 | |
a117a6d6 GW |
4497 | ASSERT(addr >= VDEV_LABEL_START_SIZE && |
4498 | addr + size < vd->vdev_psize - | |
4499 | VDEV_LABEL_END_SIZE); | |
4500 | ||
34dc7c2f | 4501 | /* |
b128c09f BB |
4502 | * l2arc read. The SCL_L2ARC lock will be |
4503 | * released by l2arc_read_done(). | |
3a17a7a9 SK |
4504 | * Issue a null zio if the underlying buffer |
4505 | * was squashed to zero size by compression. | |
34dc7c2f | 4506 | */ |
0ed212dc | 4507 | if (b_compress == ZIO_COMPRESS_EMPTY) { |
3a17a7a9 SK |
4508 | rzio = zio_null(pio, spa, vd, |
4509 | l2arc_read_done, cb, | |
4510 | zio_flags | ZIO_FLAG_DONT_CACHE | | |
4511 | ZIO_FLAG_CANFAIL | | |
4512 | ZIO_FLAG_DONT_PROPAGATE | | |
4513 | ZIO_FLAG_DONT_RETRY); | |
4514 | } else { | |
4515 | rzio = zio_read_phys(pio, vd, addr, | |
0ed212dc BP |
4516 | b_asize, buf->b_data, |
4517 | ZIO_CHECKSUM_OFF, | |
3a17a7a9 SK |
4518 | l2arc_read_done, cb, priority, |
4519 | zio_flags | ZIO_FLAG_DONT_CACHE | | |
4520 | ZIO_FLAG_CANFAIL | | |
4521 | ZIO_FLAG_DONT_PROPAGATE | | |
4522 | ZIO_FLAG_DONT_RETRY, B_FALSE); | |
4523 | } | |
34dc7c2f BB |
4524 | DTRACE_PROBE2(l2arc__read, vdev_t *, vd, |
4525 | zio_t *, rzio); | |
0ed212dc | 4526 | ARCSTAT_INCR(arcstat_l2_read_bytes, b_asize); |
34dc7c2f | 4527 | |
2a432414 | 4528 | if (*arc_flags & ARC_FLAG_NOWAIT) { |
b128c09f | 4529 | zio_nowait(rzio); |
1421c891 | 4530 | goto out; |
b128c09f | 4531 | } |
34dc7c2f | 4532 | |
2a432414 | 4533 | ASSERT(*arc_flags & ARC_FLAG_WAIT); |
b128c09f | 4534 | if (zio_wait(rzio) == 0) |
1421c891 | 4535 | goto out; |
b128c09f BB |
4536 | |
4537 | /* l2arc read error; goto zio_read() */ | |
34dc7c2f BB |
4538 | } else { |
4539 | DTRACE_PROBE1(l2arc__miss, | |
4540 | arc_buf_hdr_t *, hdr); | |
4541 | ARCSTAT_BUMP(arcstat_l2_misses); | |
4542 | if (HDR_L2_WRITING(hdr)) | |
4543 | ARCSTAT_BUMP(arcstat_l2_rw_clash); | |
b128c09f | 4544 | spa_config_exit(spa, SCL_L2ARC, vd); |
34dc7c2f | 4545 | } |
d164b209 BB |
4546 | } else { |
4547 | if (vd != NULL) | |
4548 | spa_config_exit(spa, SCL_L2ARC, vd); | |
4549 | if (l2arc_ndev != 0) { | |
4550 | DTRACE_PROBE1(l2arc__miss, | |
4551 | arc_buf_hdr_t *, hdr); | |
4552 | ARCSTAT_BUMP(arcstat_l2_misses); | |
4553 | } | |
34dc7c2f | 4554 | } |
34dc7c2f BB |
4555 | |
4556 | rzio = zio_read(pio, spa, bp, buf->b_data, size, | |
b128c09f | 4557 | arc_read_done, buf, priority, zio_flags, zb); |
34dc7c2f | 4558 | |
2a432414 | 4559 | if (*arc_flags & ARC_FLAG_WAIT) { |
1421c891 PS |
4560 | rc = zio_wait(rzio); |
4561 | goto out; | |
4562 | } | |
34dc7c2f | 4563 | |
2a432414 | 4564 | ASSERT(*arc_flags & ARC_FLAG_NOWAIT); |
34dc7c2f BB |
4565 | zio_nowait(rzio); |
4566 | } | |
1421c891 PS |
4567 | |
4568 | out: | |
4569 | spa_read_history_add(spa, zb, *arc_flags); | |
4570 | return (rc); | |
34dc7c2f BB |
4571 | } |
4572 | ||
ab26409d BB |
4573 | arc_prune_t * |
4574 | arc_add_prune_callback(arc_prune_func_t *func, void *private) | |
4575 | { | |
4576 | arc_prune_t *p; | |
4577 | ||
d1d7e268 | 4578 | p = kmem_alloc(sizeof (*p), KM_SLEEP); |
ab26409d BB |
4579 | p->p_pfunc = func; |
4580 | p->p_private = private; | |
4581 | list_link_init(&p->p_node); | |
4582 | refcount_create(&p->p_refcnt); | |
4583 | ||
4584 | mutex_enter(&arc_prune_mtx); | |
4585 | refcount_add(&p->p_refcnt, &arc_prune_list); | |
4586 | list_insert_head(&arc_prune_list, p); | |
4587 | mutex_exit(&arc_prune_mtx); | |
4588 | ||
4589 | return (p); | |
4590 | } | |
4591 | ||
4592 | void | |
4593 | arc_remove_prune_callback(arc_prune_t *p) | |
4594 | { | |
4595 | mutex_enter(&arc_prune_mtx); | |
4596 | list_remove(&arc_prune_list, p); | |
4597 | if (refcount_remove(&p->p_refcnt, &arc_prune_list) == 0) { | |
4598 | refcount_destroy(&p->p_refcnt); | |
4599 | kmem_free(p, sizeof (*p)); | |
4600 | } | |
4601 | mutex_exit(&arc_prune_mtx); | |
4602 | } | |
4603 | ||
34dc7c2f BB |
4604 | void |
4605 | arc_set_callback(arc_buf_t *buf, arc_evict_func_t *func, void *private) | |
4606 | { | |
4607 | ASSERT(buf->b_hdr != NULL); | |
b9541d6b CW |
4608 | ASSERT(buf->b_hdr->b_l1hdr.b_state != arc_anon); |
4609 | ASSERT(!refcount_is_zero(&buf->b_hdr->b_l1hdr.b_refcnt) || | |
4610 | func == NULL); | |
428870ff BB |
4611 | ASSERT(buf->b_efunc == NULL); |
4612 | ASSERT(!HDR_BUF_AVAILABLE(buf->b_hdr)); | |
4613 | ||
34dc7c2f BB |
4614 | buf->b_efunc = func; |
4615 | buf->b_private = private; | |
4616 | } | |
4617 | ||
df4474f9 MA |
4618 | /* |
4619 | * Notify the arc that a block was freed, and thus will never be used again. | |
4620 | */ | |
4621 | void | |
4622 | arc_freed(spa_t *spa, const blkptr_t *bp) | |
4623 | { | |
4624 | arc_buf_hdr_t *hdr; | |
4625 | kmutex_t *hash_lock; | |
4626 | uint64_t guid = spa_load_guid(spa); | |
4627 | ||
9b67f605 MA |
4628 | ASSERT(!BP_IS_EMBEDDED(bp)); |
4629 | ||
4630 | hdr = buf_hash_find(guid, bp, &hash_lock); | |
df4474f9 MA |
4631 | if (hdr == NULL) |
4632 | return; | |
4633 | if (HDR_BUF_AVAILABLE(hdr)) { | |
b9541d6b | 4634 | arc_buf_t *buf = hdr->b_l1hdr.b_buf; |
df4474f9 | 4635 | add_reference(hdr, hash_lock, FTAG); |
2a432414 | 4636 | hdr->b_flags &= ~ARC_FLAG_BUF_AVAILABLE; |
df4474f9 MA |
4637 | mutex_exit(hash_lock); |
4638 | ||
4639 | arc_release(buf, FTAG); | |
4640 | (void) arc_buf_remove_ref(buf, FTAG); | |
4641 | } else { | |
4642 | mutex_exit(hash_lock); | |
4643 | } | |
4644 | ||
4645 | } | |
4646 | ||
34dc7c2f | 4647 | /* |
bd089c54 MA |
4648 | * Clear the user eviction callback set by arc_set_callback(), first calling |
4649 | * it if it exists. Because the presence of a callback keeps an arc_buf cached | |
4650 | * clearing the callback may result in the arc_buf being destroyed. However, | |
4651 | * it will not result in the *last* arc_buf being destroyed, hence the data | |
4652 | * will remain cached in the ARC. We make a copy of the arc buffer here so | |
4653 | * that we can process the callback without holding any locks. | |
4654 | * | |
4655 | * It's possible that the callback is already in the process of being cleared | |
4656 | * by another thread. In this case we can not clear the callback. | |
4657 | * | |
4658 | * Returns B_TRUE if the callback was successfully called and cleared. | |
34dc7c2f | 4659 | */ |
bd089c54 MA |
4660 | boolean_t |
4661 | arc_clear_callback(arc_buf_t *buf) | |
34dc7c2f BB |
4662 | { |
4663 | arc_buf_hdr_t *hdr; | |
4664 | kmutex_t *hash_lock; | |
bd089c54 MA |
4665 | arc_evict_func_t *efunc = buf->b_efunc; |
4666 | void *private = buf->b_private; | |
34dc7c2f | 4667 | |
428870ff | 4668 | mutex_enter(&buf->b_evict_lock); |
34dc7c2f BB |
4669 | hdr = buf->b_hdr; |
4670 | if (hdr == NULL) { | |
4671 | /* | |
4672 | * We are in arc_do_user_evicts(). | |
4673 | */ | |
4674 | ASSERT(buf->b_data == NULL); | |
428870ff | 4675 | mutex_exit(&buf->b_evict_lock); |
bd089c54 | 4676 | return (B_FALSE); |
b128c09f | 4677 | } else if (buf->b_data == NULL) { |
34dc7c2f | 4678 | /* |
b128c09f BB |
4679 | * We are on the eviction list; process this buffer now |
4680 | * but let arc_do_user_evicts() do the reaping. | |
34dc7c2f | 4681 | */ |
b128c09f | 4682 | buf->b_efunc = NULL; |
428870ff | 4683 | mutex_exit(&buf->b_evict_lock); |
bd089c54 MA |
4684 | VERIFY0(efunc(private)); |
4685 | return (B_TRUE); | |
34dc7c2f | 4686 | } |
b128c09f BB |
4687 | hash_lock = HDR_LOCK(hdr); |
4688 | mutex_enter(hash_lock); | |
428870ff BB |
4689 | hdr = buf->b_hdr; |
4690 | ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); | |
34dc7c2f | 4691 | |
b9541d6b CW |
4692 | ASSERT3U(refcount_count(&hdr->b_l1hdr.b_refcnt), <, |
4693 | hdr->b_l1hdr.b_datacnt); | |
4694 | ASSERT(hdr->b_l1hdr.b_state == arc_mru || | |
4695 | hdr->b_l1hdr.b_state == arc_mfu); | |
34dc7c2f | 4696 | |
bd089c54 MA |
4697 | buf->b_efunc = NULL; |
4698 | buf->b_private = NULL; | |
34dc7c2f | 4699 | |
b9541d6b | 4700 | if (hdr->b_l1hdr.b_datacnt > 1) { |
bd089c54 | 4701 | mutex_exit(&buf->b_evict_lock); |
ca0bf58d | 4702 | arc_buf_destroy(buf, TRUE); |
bd089c54 | 4703 | } else { |
b9541d6b | 4704 | ASSERT(buf == hdr->b_l1hdr.b_buf); |
2a432414 | 4705 | hdr->b_flags |= ARC_FLAG_BUF_AVAILABLE; |
bd089c54 | 4706 | mutex_exit(&buf->b_evict_lock); |
34dc7c2f | 4707 | } |
34dc7c2f | 4708 | |
bd089c54 MA |
4709 | mutex_exit(hash_lock); |
4710 | VERIFY0(efunc(private)); | |
4711 | return (B_TRUE); | |
34dc7c2f BB |
4712 | } |
4713 | ||
4714 | /* | |
e49f1e20 WA |
4715 | * Release this buffer from the cache, making it an anonymous buffer. This |
4716 | * must be done after a read and prior to modifying the buffer contents. | |
34dc7c2f | 4717 | * If the buffer has more than one reference, we must make |
b128c09f | 4718 | * a new hdr for the buffer. |
34dc7c2f BB |
4719 | */ |
4720 | void | |
4721 | arc_release(arc_buf_t *buf, void *tag) | |
4722 | { | |
b9541d6b CW |
4723 | kmutex_t *hash_lock; |
4724 | arc_state_t *state; | |
4725 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
34dc7c2f | 4726 | |
428870ff | 4727 | /* |
ca0bf58d | 4728 | * It would be nice to assert that if its DMU metadata (level > |
428870ff BB |
4729 | * 0 || it's the dnode file), then it must be syncing context. |
4730 | * But we don't know that information at this level. | |
4731 | */ | |
4732 | ||
4733 | mutex_enter(&buf->b_evict_lock); | |
b128c09f | 4734 | |
ca0bf58d PS |
4735 | ASSERT(HDR_HAS_L1HDR(hdr)); |
4736 | ||
b9541d6b CW |
4737 | /* |
4738 | * We don't grab the hash lock prior to this check, because if | |
4739 | * the buffer's header is in the arc_anon state, it won't be | |
4740 | * linked into the hash table. | |
4741 | */ | |
4742 | if (hdr->b_l1hdr.b_state == arc_anon) { | |
4743 | mutex_exit(&buf->b_evict_lock); | |
4744 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); | |
4745 | ASSERT(!HDR_IN_HASH_TABLE(hdr)); | |
4746 | ASSERT(!HDR_HAS_L2HDR(hdr)); | |
4747 | ASSERT(BUF_EMPTY(hdr)); | |
34dc7c2f | 4748 | |
b9541d6b CW |
4749 | ASSERT3U(hdr->b_l1hdr.b_datacnt, ==, 1); |
4750 | ASSERT3S(refcount_count(&hdr->b_l1hdr.b_refcnt), ==, 1); | |
4751 | ASSERT(!list_link_active(&hdr->b_l1hdr.b_arc_node)); | |
4752 | ||
4753 | ASSERT3P(buf->b_efunc, ==, NULL); | |
4754 | ASSERT3P(buf->b_private, ==, NULL); | |
4755 | ||
4756 | hdr->b_l1hdr.b_arc_access = 0; | |
4757 | arc_buf_thaw(buf); | |
4758 | ||
4759 | return; | |
34dc7c2f BB |
4760 | } |
4761 | ||
b9541d6b CW |
4762 | hash_lock = HDR_LOCK(hdr); |
4763 | mutex_enter(hash_lock); | |
4764 | ||
4765 | /* | |
4766 | * This assignment is only valid as long as the hash_lock is | |
4767 | * held, we must be careful not to reference state or the | |
4768 | * b_state field after dropping the lock. | |
4769 | */ | |
4770 | state = hdr->b_l1hdr.b_state; | |
4771 | ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); | |
4772 | ASSERT3P(state, !=, arc_anon); | |
4773 | ||
4774 | /* this buffer is not on any list */ | |
4775 | ASSERT(refcount_count(&hdr->b_l1hdr.b_refcnt) > 0); | |
4776 | ||
4777 | if (HDR_HAS_L2HDR(hdr)) { | |
b9541d6b | 4778 | mutex_enter(&hdr->b_l2hdr.b_dev->l2ad_mtx); |
ca0bf58d PS |
4779 | |
4780 | /* | |
d962d5da PS |
4781 | * We have to recheck this conditional again now that |
4782 | * we're holding the l2ad_mtx to prevent a race with | |
4783 | * another thread which might be concurrently calling | |
4784 | * l2arc_evict(). In that case, l2arc_evict() might have | |
4785 | * destroyed the header's L2 portion as we were waiting | |
4786 | * to acquire the l2ad_mtx. | |
ca0bf58d | 4787 | */ |
d962d5da PS |
4788 | if (HDR_HAS_L2HDR(hdr)) |
4789 | arc_hdr_l2hdr_destroy(hdr); | |
ca0bf58d | 4790 | |
b9541d6b | 4791 | mutex_exit(&hdr->b_l2hdr.b_dev->l2ad_mtx); |
b128c09f BB |
4792 | } |
4793 | ||
34dc7c2f BB |
4794 | /* |
4795 | * Do we have more than one buf? | |
4796 | */ | |
b9541d6b | 4797 | if (hdr->b_l1hdr.b_datacnt > 1) { |
34dc7c2f BB |
4798 | arc_buf_hdr_t *nhdr; |
4799 | arc_buf_t **bufp; | |
4800 | uint64_t blksz = hdr->b_size; | |
d164b209 | 4801 | uint64_t spa = hdr->b_spa; |
b9541d6b | 4802 | arc_buf_contents_t type = arc_buf_type(hdr); |
34dc7c2f BB |
4803 | uint32_t flags = hdr->b_flags; |
4804 | ||
b9541d6b | 4805 | ASSERT(hdr->b_l1hdr.b_buf != buf || buf->b_next != NULL); |
34dc7c2f | 4806 | /* |
428870ff BB |
4807 | * Pull the data off of this hdr and attach it to |
4808 | * a new anonymous hdr. | |
34dc7c2f BB |
4809 | */ |
4810 | (void) remove_reference(hdr, hash_lock, tag); | |
b9541d6b | 4811 | bufp = &hdr->b_l1hdr.b_buf; |
34dc7c2f BB |
4812 | while (*bufp != buf) |
4813 | bufp = &(*bufp)->b_next; | |
428870ff | 4814 | *bufp = buf->b_next; |
34dc7c2f BB |
4815 | buf->b_next = NULL; |
4816 | ||
b9541d6b | 4817 | ASSERT3P(state, !=, arc_l2c_only); |
36da08ef PS |
4818 | |
4819 | (void) refcount_remove_many( | |
4820 | &state->arcs_size, hdr->b_size, buf); | |
4821 | ||
b9541d6b CW |
4822 | if (refcount_is_zero(&hdr->b_l1hdr.b_refcnt)) { |
4823 | uint64_t *size; | |
4824 | ||
4825 | ASSERT3P(state, !=, arc_l2c_only); | |
4826 | size = &state->arcs_lsize[type]; | |
34dc7c2f BB |
4827 | ASSERT3U(*size, >=, hdr->b_size); |
4828 | atomic_add_64(size, -hdr->b_size); | |
4829 | } | |
1eb5bfa3 GW |
4830 | |
4831 | /* | |
4832 | * We're releasing a duplicate user data buffer, update | |
4833 | * our statistics accordingly. | |
4834 | */ | |
b9541d6b | 4835 | if (HDR_ISTYPE_DATA(hdr)) { |
1eb5bfa3 GW |
4836 | ARCSTAT_BUMPDOWN(arcstat_duplicate_buffers); |
4837 | ARCSTAT_INCR(arcstat_duplicate_buffers_size, | |
4838 | -hdr->b_size); | |
4839 | } | |
b9541d6b | 4840 | hdr->b_l1hdr.b_datacnt -= 1; |
34dc7c2f | 4841 | arc_cksum_verify(buf); |
498877ba | 4842 | arc_buf_unwatch(buf); |
34dc7c2f BB |
4843 | |
4844 | mutex_exit(hash_lock); | |
4845 | ||
b9541d6b | 4846 | nhdr = kmem_cache_alloc(hdr_full_cache, KM_PUSHPAGE); |
34dc7c2f BB |
4847 | nhdr->b_size = blksz; |
4848 | nhdr->b_spa = spa; | |
b9541d6b CW |
4849 | |
4850 | nhdr->b_l1hdr.b_mru_hits = 0; | |
4851 | nhdr->b_l1hdr.b_mru_ghost_hits = 0; | |
4852 | nhdr->b_l1hdr.b_mfu_hits = 0; | |
4853 | nhdr->b_l1hdr.b_mfu_ghost_hits = 0; | |
4854 | nhdr->b_l1hdr.b_l2_hits = 0; | |
2a432414 | 4855 | nhdr->b_flags = flags & ARC_FLAG_L2_WRITING; |
b9541d6b CW |
4856 | nhdr->b_flags |= arc_bufc_to_flags(type); |
4857 | nhdr->b_flags |= ARC_FLAG_HAS_L1HDR; | |
4858 | ||
4859 | nhdr->b_l1hdr.b_buf = buf; | |
4860 | nhdr->b_l1hdr.b_datacnt = 1; | |
4861 | nhdr->b_l1hdr.b_state = arc_anon; | |
4862 | nhdr->b_l1hdr.b_arc_access = 0; | |
ca0bf58d | 4863 | nhdr->b_l1hdr.b_tmp_cdata = NULL; |
34dc7c2f | 4864 | nhdr->b_freeze_cksum = NULL; |
b9541d6b CW |
4865 | |
4866 | (void) refcount_add(&nhdr->b_l1hdr.b_refcnt, tag); | |
34dc7c2f | 4867 | buf->b_hdr = nhdr; |
428870ff | 4868 | mutex_exit(&buf->b_evict_lock); |
36da08ef | 4869 | (void) refcount_add_many(&arc_anon->arcs_size, blksz, buf); |
34dc7c2f | 4870 | } else { |
428870ff | 4871 | mutex_exit(&buf->b_evict_lock); |
b9541d6b | 4872 | ASSERT(refcount_count(&hdr->b_l1hdr.b_refcnt) == 1); |
ca0bf58d PS |
4873 | /* protected by hash lock, or hdr is on arc_anon */ |
4874 | ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); | |
34dc7c2f | 4875 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); |
b9541d6b CW |
4876 | hdr->b_l1hdr.b_mru_hits = 0; |
4877 | hdr->b_l1hdr.b_mru_ghost_hits = 0; | |
4878 | hdr->b_l1hdr.b_mfu_hits = 0; | |
4879 | hdr->b_l1hdr.b_mfu_ghost_hits = 0; | |
4880 | hdr->b_l1hdr.b_l2_hits = 0; | |
4881 | arc_change_state(arc_anon, hdr, hash_lock); | |
4882 | hdr->b_l1hdr.b_arc_access = 0; | |
4883 | mutex_exit(hash_lock); | |
34dc7c2f | 4884 | |
428870ff | 4885 | buf_discard_identity(hdr); |
34dc7c2f BB |
4886 | arc_buf_thaw(buf); |
4887 | } | |
4888 | buf->b_efunc = NULL; | |
4889 | buf->b_private = NULL; | |
34dc7c2f BB |
4890 | } |
4891 | ||
4892 | int | |
4893 | arc_released(arc_buf_t *buf) | |
4894 | { | |
b128c09f BB |
4895 | int released; |
4896 | ||
428870ff | 4897 | mutex_enter(&buf->b_evict_lock); |
b9541d6b CW |
4898 | released = (buf->b_data != NULL && |
4899 | buf->b_hdr->b_l1hdr.b_state == arc_anon); | |
428870ff | 4900 | mutex_exit(&buf->b_evict_lock); |
b128c09f | 4901 | return (released); |
34dc7c2f BB |
4902 | } |
4903 | ||
34dc7c2f BB |
4904 | #ifdef ZFS_DEBUG |
4905 | int | |
4906 | arc_referenced(arc_buf_t *buf) | |
4907 | { | |
b128c09f BB |
4908 | int referenced; |
4909 | ||
428870ff | 4910 | mutex_enter(&buf->b_evict_lock); |
b9541d6b | 4911 | referenced = (refcount_count(&buf->b_hdr->b_l1hdr.b_refcnt)); |
428870ff | 4912 | mutex_exit(&buf->b_evict_lock); |
b128c09f | 4913 | return (referenced); |
34dc7c2f BB |
4914 | } |
4915 | #endif | |
4916 | ||
4917 | static void | |
4918 | arc_write_ready(zio_t *zio) | |
4919 | { | |
4920 | arc_write_callback_t *callback = zio->io_private; | |
4921 | arc_buf_t *buf = callback->awcb_buf; | |
4922 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
4923 | ||
b9541d6b CW |
4924 | ASSERT(HDR_HAS_L1HDR(hdr)); |
4925 | ASSERT(!refcount_is_zero(&buf->b_hdr->b_l1hdr.b_refcnt)); | |
4926 | ASSERT(hdr->b_l1hdr.b_datacnt > 0); | |
b128c09f BB |
4927 | callback->awcb_ready(zio, buf, callback->awcb_private); |
4928 | ||
34dc7c2f BB |
4929 | /* |
4930 | * If the IO is already in progress, then this is a re-write | |
b128c09f BB |
4931 | * attempt, so we need to thaw and re-compute the cksum. |
4932 | * It is the responsibility of the callback to handle the | |
4933 | * accounting for any re-write attempt. | |
34dc7c2f BB |
4934 | */ |
4935 | if (HDR_IO_IN_PROGRESS(hdr)) { | |
b9541d6b | 4936 | mutex_enter(&hdr->b_l1hdr.b_freeze_lock); |
34dc7c2f BB |
4937 | if (hdr->b_freeze_cksum != NULL) { |
4938 | kmem_free(hdr->b_freeze_cksum, sizeof (zio_cksum_t)); | |
4939 | hdr->b_freeze_cksum = NULL; | |
4940 | } | |
b9541d6b | 4941 | mutex_exit(&hdr->b_l1hdr.b_freeze_lock); |
34dc7c2f BB |
4942 | } |
4943 | arc_cksum_compute(buf, B_FALSE); | |
2a432414 | 4944 | hdr->b_flags |= ARC_FLAG_IO_IN_PROGRESS; |
34dc7c2f BB |
4945 | } |
4946 | ||
e8b96c60 MA |
4947 | /* |
4948 | * The SPA calls this callback for each physical write that happens on behalf | |
4949 | * of a logical write. See the comment in dbuf_write_physdone() for details. | |
4950 | */ | |
4951 | static void | |
4952 | arc_write_physdone(zio_t *zio) | |
4953 | { | |
4954 | arc_write_callback_t *cb = zio->io_private; | |
4955 | if (cb->awcb_physdone != NULL) | |
4956 | cb->awcb_physdone(zio, cb->awcb_buf, cb->awcb_private); | |
4957 | } | |
4958 | ||
34dc7c2f BB |
4959 | static void |
4960 | arc_write_done(zio_t *zio) | |
4961 | { | |
4962 | arc_write_callback_t *callback = zio->io_private; | |
4963 | arc_buf_t *buf = callback->awcb_buf; | |
4964 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
4965 | ||
b9541d6b | 4966 | ASSERT(hdr->b_l1hdr.b_acb == NULL); |
428870ff BB |
4967 | |
4968 | if (zio->io_error == 0) { | |
9b67f605 | 4969 | if (BP_IS_HOLE(zio->io_bp) || BP_IS_EMBEDDED(zio->io_bp)) { |
b0bc7a84 MG |
4970 | buf_discard_identity(hdr); |
4971 | } else { | |
4972 | hdr->b_dva = *BP_IDENTITY(zio->io_bp); | |
4973 | hdr->b_birth = BP_PHYSICAL_BIRTH(zio->io_bp); | |
b0bc7a84 | 4974 | } |
428870ff BB |
4975 | } else { |
4976 | ASSERT(BUF_EMPTY(hdr)); | |
4977 | } | |
34dc7c2f | 4978 | |
34dc7c2f | 4979 | /* |
9b67f605 MA |
4980 | * If the block to be written was all-zero or compressed enough to be |
4981 | * embedded in the BP, no write was performed so there will be no | |
4982 | * dva/birth/checksum. The buffer must therefore remain anonymous | |
4983 | * (and uncached). | |
34dc7c2f BB |
4984 | */ |
4985 | if (!BUF_EMPTY(hdr)) { | |
4986 | arc_buf_hdr_t *exists; | |
4987 | kmutex_t *hash_lock; | |
4988 | ||
428870ff BB |
4989 | ASSERT(zio->io_error == 0); |
4990 | ||
34dc7c2f BB |
4991 | arc_cksum_verify(buf); |
4992 | ||
4993 | exists = buf_hash_insert(hdr, &hash_lock); | |
b9541d6b | 4994 | if (exists != NULL) { |
34dc7c2f BB |
4995 | /* |
4996 | * This can only happen if we overwrite for | |
4997 | * sync-to-convergence, because we remove | |
4998 | * buffers from the hash table when we arc_free(). | |
4999 | */ | |
428870ff BB |
5000 | if (zio->io_flags & ZIO_FLAG_IO_REWRITE) { |
5001 | if (!BP_EQUAL(&zio->io_bp_orig, zio->io_bp)) | |
5002 | panic("bad overwrite, hdr=%p exists=%p", | |
5003 | (void *)hdr, (void *)exists); | |
b9541d6b CW |
5004 | ASSERT(refcount_is_zero( |
5005 | &exists->b_l1hdr.b_refcnt)); | |
428870ff BB |
5006 | arc_change_state(arc_anon, exists, hash_lock); |
5007 | mutex_exit(hash_lock); | |
5008 | arc_hdr_destroy(exists); | |
5009 | exists = buf_hash_insert(hdr, &hash_lock); | |
5010 | ASSERT3P(exists, ==, NULL); | |
03c6040b GW |
5011 | } else if (zio->io_flags & ZIO_FLAG_NOPWRITE) { |
5012 | /* nopwrite */ | |
5013 | ASSERT(zio->io_prop.zp_nopwrite); | |
5014 | if (!BP_EQUAL(&zio->io_bp_orig, zio->io_bp)) | |
5015 | panic("bad nopwrite, hdr=%p exists=%p", | |
5016 | (void *)hdr, (void *)exists); | |
428870ff BB |
5017 | } else { |
5018 | /* Dedup */ | |
b9541d6b CW |
5019 | ASSERT(hdr->b_l1hdr.b_datacnt == 1); |
5020 | ASSERT(hdr->b_l1hdr.b_state == arc_anon); | |
428870ff BB |
5021 | ASSERT(BP_GET_DEDUP(zio->io_bp)); |
5022 | ASSERT(BP_GET_LEVEL(zio->io_bp) == 0); | |
5023 | } | |
34dc7c2f | 5024 | } |
2a432414 | 5025 | hdr->b_flags &= ~ARC_FLAG_IO_IN_PROGRESS; |
b128c09f | 5026 | /* if it's not anon, we are doing a scrub */ |
b9541d6b | 5027 | if (exists == NULL && hdr->b_l1hdr.b_state == arc_anon) |
b128c09f | 5028 | arc_access(hdr, hash_lock); |
34dc7c2f | 5029 | mutex_exit(hash_lock); |
34dc7c2f | 5030 | } else { |
2a432414 | 5031 | hdr->b_flags &= ~ARC_FLAG_IO_IN_PROGRESS; |
34dc7c2f BB |
5032 | } |
5033 | ||
b9541d6b | 5034 | ASSERT(!refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); |
428870ff | 5035 | callback->awcb_done(zio, buf, callback->awcb_private); |
34dc7c2f BB |
5036 | |
5037 | kmem_free(callback, sizeof (arc_write_callback_t)); | |
5038 | } | |
5039 | ||
5040 | zio_t * | |
428870ff | 5041 | arc_write(zio_t *pio, spa_t *spa, uint64_t txg, |
3a17a7a9 | 5042 | blkptr_t *bp, arc_buf_t *buf, boolean_t l2arc, boolean_t l2arc_compress, |
e8b96c60 MA |
5043 | const zio_prop_t *zp, arc_done_func_t *ready, arc_done_func_t *physdone, |
5044 | arc_done_func_t *done, void *private, zio_priority_t priority, | |
5dbd68a3 | 5045 | int zio_flags, const zbookmark_phys_t *zb) |
34dc7c2f BB |
5046 | { |
5047 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
5048 | arc_write_callback_t *callback; | |
b128c09f | 5049 | zio_t *zio; |
34dc7c2f | 5050 | |
b128c09f | 5051 | ASSERT(ready != NULL); |
428870ff | 5052 | ASSERT(done != NULL); |
34dc7c2f | 5053 | ASSERT(!HDR_IO_ERROR(hdr)); |
b9541d6b CW |
5054 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); |
5055 | ASSERT(hdr->b_l1hdr.b_acb == NULL); | |
5056 | ASSERT(hdr->b_l1hdr.b_datacnt > 0); | |
b128c09f | 5057 | if (l2arc) |
2a432414 | 5058 | hdr->b_flags |= ARC_FLAG_L2CACHE; |
3a17a7a9 | 5059 | if (l2arc_compress) |
2a432414 | 5060 | hdr->b_flags |= ARC_FLAG_L2COMPRESS; |
79c76d5b | 5061 | callback = kmem_zalloc(sizeof (arc_write_callback_t), KM_SLEEP); |
34dc7c2f | 5062 | callback->awcb_ready = ready; |
e8b96c60 | 5063 | callback->awcb_physdone = physdone; |
34dc7c2f BB |
5064 | callback->awcb_done = done; |
5065 | callback->awcb_private = private; | |
5066 | callback->awcb_buf = buf; | |
b128c09f | 5067 | |
428870ff | 5068 | zio = zio_write(pio, spa, txg, bp, buf->b_data, hdr->b_size, zp, |
e8b96c60 MA |
5069 | arc_write_ready, arc_write_physdone, arc_write_done, callback, |
5070 | priority, zio_flags, zb); | |
34dc7c2f BB |
5071 | |
5072 | return (zio); | |
5073 | } | |
5074 | ||
34dc7c2f | 5075 | static int |
e8b96c60 | 5076 | arc_memory_throttle(uint64_t reserve, uint64_t txg) |
34dc7c2f BB |
5077 | { |
5078 | #ifdef _KERNEL | |
0c5493d4 BB |
5079 | if (zfs_arc_memory_throttle_disable) |
5080 | return (0); | |
5081 | ||
ca67b33a MA |
5082 | if (freemem > physmem * arc_lotsfree_percent / 100) |
5083 | return (0); | |
5084 | ||
5085 | if (arc_reclaim_needed()) { | |
5086 | /* memory is low, delay before restarting */ | |
34dc7c2f | 5087 | ARCSTAT_INCR(arcstat_memory_throttle_count, 1); |
570827e1 | 5088 | DMU_TX_STAT_BUMP(dmu_tx_memory_reclaim); |
2e528b49 | 5089 | return (SET_ERROR(EAGAIN)); |
34dc7c2f | 5090 | } |
34dc7c2f BB |
5091 | #endif |
5092 | return (0); | |
5093 | } | |
5094 | ||
5095 | void | |
5096 | arc_tempreserve_clear(uint64_t reserve) | |
5097 | { | |
5098 | atomic_add_64(&arc_tempreserve, -reserve); | |
5099 | ASSERT((int64_t)arc_tempreserve >= 0); | |
5100 | } | |
5101 | ||
5102 | int | |
5103 | arc_tempreserve_space(uint64_t reserve, uint64_t txg) | |
5104 | { | |
5105 | int error; | |
9babb374 | 5106 | uint64_t anon_size; |
34dc7c2f | 5107 | |
34dc7c2f BB |
5108 | if (reserve > arc_c/4 && !arc_no_grow) |
5109 | arc_c = MIN(arc_c_max, reserve * 4); | |
12f9a6a3 BB |
5110 | |
5111 | /* | |
5112 | * Throttle when the calculated memory footprint for the TXG | |
5113 | * exceeds the target ARC size. | |
5114 | */ | |
570827e1 BB |
5115 | if (reserve > arc_c) { |
5116 | DMU_TX_STAT_BUMP(dmu_tx_memory_reserve); | |
12f9a6a3 | 5117 | return (SET_ERROR(ERESTART)); |
570827e1 | 5118 | } |
34dc7c2f | 5119 | |
9babb374 BB |
5120 | /* |
5121 | * Don't count loaned bufs as in flight dirty data to prevent long | |
5122 | * network delays from blocking transactions that are ready to be | |
5123 | * assigned to a txg. | |
5124 | */ | |
36da08ef PS |
5125 | anon_size = MAX((int64_t)(refcount_count(&arc_anon->arcs_size) - |
5126 | arc_loaned_bytes), 0); | |
9babb374 | 5127 | |
34dc7c2f BB |
5128 | /* |
5129 | * Writes will, almost always, require additional memory allocations | |
d3cc8b15 | 5130 | * in order to compress/encrypt/etc the data. We therefore need to |
34dc7c2f BB |
5131 | * make sure that there is sufficient available memory for this. |
5132 | */ | |
e8b96c60 MA |
5133 | error = arc_memory_throttle(reserve, txg); |
5134 | if (error != 0) | |
34dc7c2f BB |
5135 | return (error); |
5136 | ||
5137 | /* | |
5138 | * Throttle writes when the amount of dirty data in the cache | |
5139 | * gets too large. We try to keep the cache less than half full | |
5140 | * of dirty blocks so that our sync times don't grow too large. | |
5141 | * Note: if two requests come in concurrently, we might let them | |
5142 | * both succeed, when one of them should fail. Not a huge deal. | |
5143 | */ | |
9babb374 BB |
5144 | |
5145 | if (reserve + arc_tempreserve + anon_size > arc_c / 2 && | |
5146 | anon_size > arc_c / 4) { | |
34dc7c2f BB |
5147 | dprintf("failing, arc_tempreserve=%lluK anon_meta=%lluK " |
5148 | "anon_data=%lluK tempreserve=%lluK arc_c=%lluK\n", | |
5149 | arc_tempreserve>>10, | |
5150 | arc_anon->arcs_lsize[ARC_BUFC_METADATA]>>10, | |
5151 | arc_anon->arcs_lsize[ARC_BUFC_DATA]>>10, | |
5152 | reserve>>10, arc_c>>10); | |
570827e1 | 5153 | DMU_TX_STAT_BUMP(dmu_tx_dirty_throttle); |
2e528b49 | 5154 | return (SET_ERROR(ERESTART)); |
34dc7c2f BB |
5155 | } |
5156 | atomic_add_64(&arc_tempreserve, reserve); | |
5157 | return (0); | |
5158 | } | |
5159 | ||
13be560d BB |
5160 | static void |
5161 | arc_kstat_update_state(arc_state_t *state, kstat_named_t *size, | |
5162 | kstat_named_t *evict_data, kstat_named_t *evict_metadata) | |
5163 | { | |
36da08ef | 5164 | size->value.ui64 = refcount_count(&state->arcs_size); |
13be560d BB |
5165 | evict_data->value.ui64 = state->arcs_lsize[ARC_BUFC_DATA]; |
5166 | evict_metadata->value.ui64 = state->arcs_lsize[ARC_BUFC_METADATA]; | |
5167 | } | |
5168 | ||
5169 | static int | |
5170 | arc_kstat_update(kstat_t *ksp, int rw) | |
5171 | { | |
5172 | arc_stats_t *as = ksp->ks_data; | |
5173 | ||
5174 | if (rw == KSTAT_WRITE) { | |
500445c0 | 5175 | return (EACCES); |
13be560d BB |
5176 | } else { |
5177 | arc_kstat_update_state(arc_anon, | |
5178 | &as->arcstat_anon_size, | |
500445c0 PS |
5179 | &as->arcstat_anon_evictable_data, |
5180 | &as->arcstat_anon_evictable_metadata); | |
13be560d BB |
5181 | arc_kstat_update_state(arc_mru, |
5182 | &as->arcstat_mru_size, | |
500445c0 PS |
5183 | &as->arcstat_mru_evictable_data, |
5184 | &as->arcstat_mru_evictable_metadata); | |
13be560d BB |
5185 | arc_kstat_update_state(arc_mru_ghost, |
5186 | &as->arcstat_mru_ghost_size, | |
500445c0 PS |
5187 | &as->arcstat_mru_ghost_evictable_data, |
5188 | &as->arcstat_mru_ghost_evictable_metadata); | |
13be560d BB |
5189 | arc_kstat_update_state(arc_mfu, |
5190 | &as->arcstat_mfu_size, | |
500445c0 PS |
5191 | &as->arcstat_mfu_evictable_data, |
5192 | &as->arcstat_mfu_evictable_metadata); | |
fc41c640 | 5193 | arc_kstat_update_state(arc_mfu_ghost, |
13be560d | 5194 | &as->arcstat_mfu_ghost_size, |
500445c0 PS |
5195 | &as->arcstat_mfu_ghost_evictable_data, |
5196 | &as->arcstat_mfu_ghost_evictable_metadata); | |
13be560d BB |
5197 | } |
5198 | ||
5199 | return (0); | |
5200 | } | |
5201 | ||
ca0bf58d PS |
5202 | /* |
5203 | * This function *must* return indices evenly distributed between all | |
5204 | * sublists of the multilist. This is needed due to how the ARC eviction | |
5205 | * code is laid out; arc_evict_state() assumes ARC buffers are evenly | |
5206 | * distributed between all sublists and uses this assumption when | |
5207 | * deciding which sublist to evict from and how much to evict from it. | |
5208 | */ | |
5209 | unsigned int | |
5210 | arc_state_multilist_index_func(multilist_t *ml, void *obj) | |
5211 | { | |
5212 | arc_buf_hdr_t *hdr = obj; | |
5213 | ||
5214 | /* | |
5215 | * We rely on b_dva to generate evenly distributed index | |
5216 | * numbers using buf_hash below. So, as an added precaution, | |
5217 | * let's make sure we never add empty buffers to the arc lists. | |
5218 | */ | |
5219 | ASSERT(!BUF_EMPTY(hdr)); | |
5220 | ||
5221 | /* | |
5222 | * The assumption here, is the hash value for a given | |
5223 | * arc_buf_hdr_t will remain constant throughout its lifetime | |
5224 | * (i.e. its b_spa, b_dva, and b_birth fields don't change). | |
5225 | * Thus, we don't need to store the header's sublist index | |
5226 | * on insertion, as this index can be recalculated on removal. | |
5227 | * | |
5228 | * Also, the low order bits of the hash value are thought to be | |
5229 | * distributed evenly. Otherwise, in the case that the multilist | |
5230 | * has a power of two number of sublists, each sublists' usage | |
5231 | * would not be evenly distributed. | |
5232 | */ | |
5233 | return (buf_hash(hdr->b_spa, &hdr->b_dva, hdr->b_birth) % | |
5234 | multilist_get_num_sublists(ml)); | |
5235 | } | |
5236 | ||
ca67b33a MA |
5237 | /* |
5238 | * Called during module initialization and periodically thereafter to | |
5239 | * apply reasonable changes to the exposed performance tunings. Non-zero | |
5240 | * zfs_* values which differ from the currently set values will be applied. | |
5241 | */ | |
5242 | static void | |
5243 | arc_tuning_update(void) | |
5244 | { | |
5245 | /* Valid range: 64M - <all physical memory> */ | |
5246 | if ((zfs_arc_max) && (zfs_arc_max != arc_c_max) && | |
5247 | (zfs_arc_max > 64 << 20) && (zfs_arc_max < ptob(physmem)) && | |
5248 | (zfs_arc_max > arc_c_min)) { | |
5249 | arc_c_max = zfs_arc_max; | |
5250 | arc_c = arc_c_max; | |
5251 | arc_p = (arc_c >> 1); | |
5252 | arc_meta_limit = MIN(arc_meta_limit, arc_c_max); | |
5253 | } | |
5254 | ||
5255 | /* Valid range: 32M - <arc_c_max> */ | |
5256 | if ((zfs_arc_min) && (zfs_arc_min != arc_c_min) && | |
5257 | (zfs_arc_min >= 2ULL << SPA_MAXBLOCKSHIFT) && | |
5258 | (zfs_arc_min <= arc_c_max)) { | |
5259 | arc_c_min = zfs_arc_min; | |
5260 | arc_c = MAX(arc_c, arc_c_min); | |
5261 | } | |
5262 | ||
5263 | /* Valid range: 16M - <arc_c_max> */ | |
5264 | if ((zfs_arc_meta_min) && (zfs_arc_meta_min != arc_meta_min) && | |
5265 | (zfs_arc_meta_min >= 1ULL << SPA_MAXBLOCKSHIFT) && | |
5266 | (zfs_arc_meta_min <= arc_c_max)) { | |
5267 | arc_meta_min = zfs_arc_meta_min; | |
5268 | arc_meta_limit = MAX(arc_meta_limit, arc_meta_min); | |
5269 | } | |
5270 | ||
5271 | /* Valid range: <arc_meta_min> - <arc_c_max> */ | |
5272 | if ((zfs_arc_meta_limit) && (zfs_arc_meta_limit != arc_meta_limit) && | |
5273 | (zfs_arc_meta_limit >= zfs_arc_meta_min) && | |
5274 | (zfs_arc_meta_limit <= arc_c_max)) | |
5275 | arc_meta_limit = zfs_arc_meta_limit; | |
5276 | ||
5277 | /* Valid range: 1 - N */ | |
5278 | if (zfs_arc_grow_retry) | |
5279 | arc_grow_retry = zfs_arc_grow_retry; | |
5280 | ||
5281 | /* Valid range: 1 - N */ | |
5282 | if (zfs_arc_shrink_shift) { | |
5283 | arc_shrink_shift = zfs_arc_shrink_shift; | |
5284 | arc_no_grow_shift = MIN(arc_no_grow_shift, arc_shrink_shift -1); | |
5285 | } | |
5286 | ||
728d6ae9 BB |
5287 | /* Valid range: 1 - N */ |
5288 | if (zfs_arc_p_min_shift) | |
5289 | arc_p_min_shift = zfs_arc_p_min_shift; | |
5290 | ||
ca67b33a MA |
5291 | /* Valid range: 1 - N ticks */ |
5292 | if (zfs_arc_min_prefetch_lifespan) | |
5293 | arc_min_prefetch_lifespan = zfs_arc_min_prefetch_lifespan; | |
5294 | } | |
5295 | ||
34dc7c2f BB |
5296 | void |
5297 | arc_init(void) | |
5298 | { | |
ca67b33a MA |
5299 | /* |
5300 | * allmem is "all memory that we could possibly use". | |
5301 | */ | |
5302 | #ifdef _KERNEL | |
5303 | uint64_t allmem = ptob(physmem); | |
5304 | #else | |
5305 | uint64_t allmem = (physmem * PAGESIZE) / 2; | |
5306 | #endif | |
5307 | ||
ca0bf58d PS |
5308 | mutex_init(&arc_reclaim_lock, NULL, MUTEX_DEFAULT, NULL); |
5309 | cv_init(&arc_reclaim_thread_cv, NULL, CV_DEFAULT, NULL); | |
5310 | cv_init(&arc_reclaim_waiters_cv, NULL, CV_DEFAULT, NULL); | |
5311 | ||
5312 | mutex_init(&arc_user_evicts_lock, NULL, MUTEX_DEFAULT, NULL); | |
5313 | cv_init(&arc_user_evicts_cv, NULL, CV_DEFAULT, NULL); | |
34dc7c2f BB |
5314 | |
5315 | /* Convert seconds to clock ticks */ | |
ca67b33a | 5316 | arc_min_prefetch_lifespan = 1 * hz; |
34dc7c2f BB |
5317 | |
5318 | /* Start out with 1/8 of all memory */ | |
ca67b33a | 5319 | arc_c = allmem / 8; |
34dc7c2f BB |
5320 | |
5321 | #ifdef _KERNEL | |
5322 | /* | |
5323 | * On architectures where the physical memory can be larger | |
5324 | * than the addressable space (intel in 32-bit mode), we may | |
5325 | * need to limit the cache to 1/8 of VM size. | |
5326 | */ | |
5327 | arc_c = MIN(arc_c, vmem_size(heap_arena, VMEM_ALLOC | VMEM_FREE) / 8); | |
ca67b33a | 5328 | |
7cb67b45 BB |
5329 | /* |
5330 | * Register a shrinker to support synchronous (direct) memory | |
5331 | * reclaim from the arc. This is done to prevent kswapd from | |
5332 | * swapping out pages when it is preferable to shrink the arc. | |
5333 | */ | |
5334 | spl_register_shrinker(&arc_shrinker); | |
34dc7c2f BB |
5335 | #endif |
5336 | ||
ca67b33a | 5337 | /* Set min cache to allow safe operation of arc_adapt() */ |
121b3cae | 5338 | arc_c_min = 2ULL << SPA_MAXBLOCKSHIFT; |
ca67b33a MA |
5339 | /* Set max to 1/2 of all memory */ |
5340 | arc_c_max = allmem / 2; | |
34dc7c2f BB |
5341 | |
5342 | arc_c = arc_c_max; | |
5343 | arc_p = (arc_c >> 1); | |
5344 | ||
ca67b33a MA |
5345 | /* Set min to 1/2 of arc_c_min */ |
5346 | arc_meta_min = 1ULL << SPA_MAXBLOCKSHIFT; | |
5347 | /* Initialize maximum observed usage to zero */ | |
1834f2d8 | 5348 | arc_meta_max = 0; |
ca67b33a MA |
5349 | /* Set limit to 3/4 of arc_c_max with a floor of arc_meta_min */ |
5350 | arc_meta_limit = MAX((3 * arc_c_max) / 4, arc_meta_min); | |
34dc7c2f | 5351 | |
ca67b33a MA |
5352 | /* Apply user specified tunings */ |
5353 | arc_tuning_update(); | |
c52fca13 | 5354 | |
ca0bf58d | 5355 | if (zfs_arc_num_sublists_per_state < 1) |
ca67b33a | 5356 | zfs_arc_num_sublists_per_state = MAX(boot_ncpus, 1); |
ca0bf58d | 5357 | |
34dc7c2f BB |
5358 | /* if kmem_flags are set, lets try to use less memory */ |
5359 | if (kmem_debugging()) | |
5360 | arc_c = arc_c / 2; | |
5361 | if (arc_c < arc_c_min) | |
5362 | arc_c = arc_c_min; | |
5363 | ||
5364 | arc_anon = &ARC_anon; | |
5365 | arc_mru = &ARC_mru; | |
5366 | arc_mru_ghost = &ARC_mru_ghost; | |
5367 | arc_mfu = &ARC_mfu; | |
5368 | arc_mfu_ghost = &ARC_mfu_ghost; | |
5369 | arc_l2c_only = &ARC_l2c_only; | |
5370 | arc_size = 0; | |
5371 | ||
ca0bf58d | 5372 | multilist_create(&arc_mru->arcs_list[ARC_BUFC_METADATA], |
b9541d6b | 5373 | sizeof (arc_buf_hdr_t), |
ca0bf58d PS |
5374 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), |
5375 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
5376 | multilist_create(&arc_mru->arcs_list[ARC_BUFC_DATA], | |
b9541d6b | 5377 | sizeof (arc_buf_hdr_t), |
ca0bf58d PS |
5378 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), |
5379 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
5380 | multilist_create(&arc_mru_ghost->arcs_list[ARC_BUFC_METADATA], | |
b9541d6b | 5381 | sizeof (arc_buf_hdr_t), |
ca0bf58d PS |
5382 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), |
5383 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
5384 | multilist_create(&arc_mru_ghost->arcs_list[ARC_BUFC_DATA], | |
b9541d6b | 5385 | sizeof (arc_buf_hdr_t), |
ca0bf58d PS |
5386 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), |
5387 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
5388 | multilist_create(&arc_mfu->arcs_list[ARC_BUFC_METADATA], | |
b9541d6b | 5389 | sizeof (arc_buf_hdr_t), |
ca0bf58d PS |
5390 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), |
5391 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
5392 | multilist_create(&arc_mfu->arcs_list[ARC_BUFC_DATA], | |
b9541d6b | 5393 | sizeof (arc_buf_hdr_t), |
ca0bf58d PS |
5394 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), |
5395 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
5396 | multilist_create(&arc_mfu_ghost->arcs_list[ARC_BUFC_METADATA], | |
b9541d6b | 5397 | sizeof (arc_buf_hdr_t), |
ca0bf58d PS |
5398 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), |
5399 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
5400 | multilist_create(&arc_mfu_ghost->arcs_list[ARC_BUFC_DATA], | |
b9541d6b | 5401 | sizeof (arc_buf_hdr_t), |
ca0bf58d PS |
5402 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), |
5403 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
5404 | multilist_create(&arc_l2c_only->arcs_list[ARC_BUFC_METADATA], | |
b9541d6b | 5405 | sizeof (arc_buf_hdr_t), |
ca0bf58d PS |
5406 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), |
5407 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
5408 | multilist_create(&arc_l2c_only->arcs_list[ARC_BUFC_DATA], | |
b9541d6b | 5409 | sizeof (arc_buf_hdr_t), |
ca0bf58d PS |
5410 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), |
5411 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
34dc7c2f | 5412 | |
e0b0ca98 BB |
5413 | arc_anon->arcs_state = ARC_STATE_ANON; |
5414 | arc_mru->arcs_state = ARC_STATE_MRU; | |
5415 | arc_mru_ghost->arcs_state = ARC_STATE_MRU_GHOST; | |
5416 | arc_mfu->arcs_state = ARC_STATE_MFU; | |
5417 | arc_mfu_ghost->arcs_state = ARC_STATE_MFU_GHOST; | |
5418 | arc_l2c_only->arcs_state = ARC_STATE_L2C_ONLY; | |
5419 | ||
36da08ef PS |
5420 | refcount_create(&arc_anon->arcs_size); |
5421 | refcount_create(&arc_mru->arcs_size); | |
5422 | refcount_create(&arc_mru_ghost->arcs_size); | |
5423 | refcount_create(&arc_mfu->arcs_size); | |
5424 | refcount_create(&arc_mfu_ghost->arcs_size); | |
5425 | refcount_create(&arc_l2c_only->arcs_size); | |
5426 | ||
34dc7c2f BB |
5427 | buf_init(); |
5428 | ||
ca0bf58d PS |
5429 | arc_reclaim_thread_exit = FALSE; |
5430 | arc_user_evicts_thread_exit = FALSE; | |
ab26409d BB |
5431 | list_create(&arc_prune_list, sizeof (arc_prune_t), |
5432 | offsetof(arc_prune_t, p_node)); | |
34dc7c2f | 5433 | arc_eviction_list = NULL; |
ab26409d | 5434 | mutex_init(&arc_prune_mtx, NULL, MUTEX_DEFAULT, NULL); |
34dc7c2f BB |
5435 | bzero(&arc_eviction_hdr, sizeof (arc_buf_hdr_t)); |
5436 | ||
f6046738 | 5437 | arc_prune_taskq = taskq_create("arc_prune", max_ncpus, minclsyspri, |
aa9af22c | 5438 | max_ncpus, INT_MAX, TASKQ_PREPOPULATE | TASKQ_DYNAMIC); |
f6046738 | 5439 | |
34dc7c2f BB |
5440 | arc_ksp = kstat_create("zfs", 0, "arcstats", "misc", KSTAT_TYPE_NAMED, |
5441 | sizeof (arc_stats) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL); | |
5442 | ||
5443 | if (arc_ksp != NULL) { | |
5444 | arc_ksp->ks_data = &arc_stats; | |
13be560d | 5445 | arc_ksp->ks_update = arc_kstat_update; |
34dc7c2f BB |
5446 | kstat_install(arc_ksp); |
5447 | } | |
5448 | ||
ca67b33a | 5449 | (void) thread_create(NULL, 0, arc_reclaim_thread, NULL, 0, &p0, |
34dc7c2f BB |
5450 | TS_RUN, minclsyspri); |
5451 | ||
ca0bf58d PS |
5452 | (void) thread_create(NULL, 0, arc_user_evicts_thread, NULL, 0, &p0, |
5453 | TS_RUN, minclsyspri); | |
5454 | ||
34dc7c2f | 5455 | arc_dead = FALSE; |
b128c09f | 5456 | arc_warm = B_FALSE; |
34dc7c2f | 5457 | |
e8b96c60 MA |
5458 | /* |
5459 | * Calculate maximum amount of dirty data per pool. | |
5460 | * | |
5461 | * If it has been set by a module parameter, take that. | |
5462 | * Otherwise, use a percentage of physical memory defined by | |
5463 | * zfs_dirty_data_max_percent (default 10%) with a cap at | |
5464 | * zfs_dirty_data_max_max (default 25% of physical memory). | |
5465 | */ | |
5466 | if (zfs_dirty_data_max_max == 0) | |
5467 | zfs_dirty_data_max_max = physmem * PAGESIZE * | |
5468 | zfs_dirty_data_max_max_percent / 100; | |
5469 | ||
5470 | if (zfs_dirty_data_max == 0) { | |
5471 | zfs_dirty_data_max = physmem * PAGESIZE * | |
5472 | zfs_dirty_data_max_percent / 100; | |
5473 | zfs_dirty_data_max = MIN(zfs_dirty_data_max, | |
5474 | zfs_dirty_data_max_max); | |
5475 | } | |
34dc7c2f BB |
5476 | } |
5477 | ||
5478 | void | |
5479 | arc_fini(void) | |
5480 | { | |
ab26409d BB |
5481 | arc_prune_t *p; |
5482 | ||
7cb67b45 BB |
5483 | #ifdef _KERNEL |
5484 | spl_unregister_shrinker(&arc_shrinker); | |
5485 | #endif /* _KERNEL */ | |
5486 | ||
ca0bf58d PS |
5487 | mutex_enter(&arc_reclaim_lock); |
5488 | arc_reclaim_thread_exit = TRUE; | |
5489 | /* | |
5490 | * The reclaim thread will set arc_reclaim_thread_exit back to | |
5491 | * FALSE when it is finished exiting; we're waiting for that. | |
5492 | */ | |
5493 | while (arc_reclaim_thread_exit) { | |
5494 | cv_signal(&arc_reclaim_thread_cv); | |
5495 | cv_wait(&arc_reclaim_thread_cv, &arc_reclaim_lock); | |
5496 | } | |
5497 | mutex_exit(&arc_reclaim_lock); | |
5498 | ||
5499 | mutex_enter(&arc_user_evicts_lock); | |
5500 | arc_user_evicts_thread_exit = TRUE; | |
5501 | /* | |
5502 | * The user evicts thread will set arc_user_evicts_thread_exit | |
5503 | * to FALSE when it is finished exiting; we're waiting for that. | |
5504 | */ | |
5505 | while (arc_user_evicts_thread_exit) { | |
5506 | cv_signal(&arc_user_evicts_cv); | |
5507 | cv_wait(&arc_user_evicts_cv, &arc_user_evicts_lock); | |
5508 | } | |
5509 | mutex_exit(&arc_user_evicts_lock); | |
34dc7c2f | 5510 | |
ca0bf58d PS |
5511 | /* Use TRUE to ensure *all* buffers are evicted */ |
5512 | arc_flush(NULL, TRUE); | |
34dc7c2f BB |
5513 | |
5514 | arc_dead = TRUE; | |
5515 | ||
5516 | if (arc_ksp != NULL) { | |
5517 | kstat_delete(arc_ksp); | |
5518 | arc_ksp = NULL; | |
5519 | } | |
5520 | ||
f6046738 BB |
5521 | taskq_wait(arc_prune_taskq); |
5522 | taskq_destroy(arc_prune_taskq); | |
5523 | ||
ab26409d BB |
5524 | mutex_enter(&arc_prune_mtx); |
5525 | while ((p = list_head(&arc_prune_list)) != NULL) { | |
5526 | list_remove(&arc_prune_list, p); | |
5527 | refcount_remove(&p->p_refcnt, &arc_prune_list); | |
5528 | refcount_destroy(&p->p_refcnt); | |
5529 | kmem_free(p, sizeof (*p)); | |
5530 | } | |
5531 | mutex_exit(&arc_prune_mtx); | |
5532 | ||
5533 | list_destroy(&arc_prune_list); | |
5534 | mutex_destroy(&arc_prune_mtx); | |
ca0bf58d PS |
5535 | mutex_destroy(&arc_reclaim_lock); |
5536 | cv_destroy(&arc_reclaim_thread_cv); | |
5537 | cv_destroy(&arc_reclaim_waiters_cv); | |
5538 | ||
5539 | mutex_destroy(&arc_user_evicts_lock); | |
5540 | cv_destroy(&arc_user_evicts_cv); | |
5541 | ||
36da08ef PS |
5542 | refcount_destroy(&arc_anon->arcs_size); |
5543 | refcount_destroy(&arc_mru->arcs_size); | |
5544 | refcount_destroy(&arc_mru_ghost->arcs_size); | |
5545 | refcount_destroy(&arc_mfu->arcs_size); | |
5546 | refcount_destroy(&arc_mfu_ghost->arcs_size); | |
5547 | refcount_destroy(&arc_l2c_only->arcs_size); | |
5548 | ||
ca0bf58d PS |
5549 | multilist_destroy(&arc_mru->arcs_list[ARC_BUFC_METADATA]); |
5550 | multilist_destroy(&arc_mru_ghost->arcs_list[ARC_BUFC_METADATA]); | |
5551 | multilist_destroy(&arc_mfu->arcs_list[ARC_BUFC_METADATA]); | |
5552 | multilist_destroy(&arc_mfu_ghost->arcs_list[ARC_BUFC_METADATA]); | |
5553 | multilist_destroy(&arc_mru->arcs_list[ARC_BUFC_DATA]); | |
5554 | multilist_destroy(&arc_mru_ghost->arcs_list[ARC_BUFC_DATA]); | |
5555 | multilist_destroy(&arc_mfu->arcs_list[ARC_BUFC_DATA]); | |
5556 | multilist_destroy(&arc_mfu_ghost->arcs_list[ARC_BUFC_DATA]); | |
5557 | multilist_destroy(&arc_l2c_only->arcs_list[ARC_BUFC_METADATA]); | |
5558 | multilist_destroy(&arc_l2c_only->arcs_list[ARC_BUFC_DATA]); | |
34dc7c2f BB |
5559 | |
5560 | buf_fini(); | |
9babb374 | 5561 | |
b9541d6b | 5562 | ASSERT0(arc_loaned_bytes); |
34dc7c2f BB |
5563 | } |
5564 | ||
5565 | /* | |
5566 | * Level 2 ARC | |
5567 | * | |
5568 | * The level 2 ARC (L2ARC) is a cache layer in-between main memory and disk. | |
5569 | * It uses dedicated storage devices to hold cached data, which are populated | |
5570 | * using large infrequent writes. The main role of this cache is to boost | |
5571 | * the performance of random read workloads. The intended L2ARC devices | |
5572 | * include short-stroked disks, solid state disks, and other media with | |
5573 | * substantially faster read latency than disk. | |
5574 | * | |
5575 | * +-----------------------+ | |
5576 | * | ARC | | |
5577 | * +-----------------------+ | |
5578 | * | ^ ^ | |
5579 | * | | | | |
5580 | * l2arc_feed_thread() arc_read() | |
5581 | * | | | | |
5582 | * | l2arc read | | |
5583 | * V | | | |
5584 | * +---------------+ | | |
5585 | * | L2ARC | | | |
5586 | * +---------------+ | | |
5587 | * | ^ | | |
5588 | * l2arc_write() | | | |
5589 | * | | | | |
5590 | * V | | | |
5591 | * +-------+ +-------+ | |
5592 | * | vdev | | vdev | | |
5593 | * | cache | | cache | | |
5594 | * +-------+ +-------+ | |
5595 | * +=========+ .-----. | |
5596 | * : L2ARC : |-_____-| | |
5597 | * : devices : | Disks | | |
5598 | * +=========+ `-_____-' | |
5599 | * | |
5600 | * Read requests are satisfied from the following sources, in order: | |
5601 | * | |
5602 | * 1) ARC | |
5603 | * 2) vdev cache of L2ARC devices | |
5604 | * 3) L2ARC devices | |
5605 | * 4) vdev cache of disks | |
5606 | * 5) disks | |
5607 | * | |
5608 | * Some L2ARC device types exhibit extremely slow write performance. | |
5609 | * To accommodate for this there are some significant differences between | |
5610 | * the L2ARC and traditional cache design: | |
5611 | * | |
5612 | * 1. There is no eviction path from the ARC to the L2ARC. Evictions from | |
5613 | * the ARC behave as usual, freeing buffers and placing headers on ghost | |
5614 | * lists. The ARC does not send buffers to the L2ARC during eviction as | |
5615 | * this would add inflated write latencies for all ARC memory pressure. | |
5616 | * | |
5617 | * 2. The L2ARC attempts to cache data from the ARC before it is evicted. | |
5618 | * It does this by periodically scanning buffers from the eviction-end of | |
5619 | * the MFU and MRU ARC lists, copying them to the L2ARC devices if they are | |
3a17a7a9 SK |
5620 | * not already there. It scans until a headroom of buffers is satisfied, |
5621 | * which itself is a buffer for ARC eviction. If a compressible buffer is | |
5622 | * found during scanning and selected for writing to an L2ARC device, we | |
5623 | * temporarily boost scanning headroom during the next scan cycle to make | |
5624 | * sure we adapt to compression effects (which might significantly reduce | |
5625 | * the data volume we write to L2ARC). The thread that does this is | |
34dc7c2f BB |
5626 | * l2arc_feed_thread(), illustrated below; example sizes are included to |
5627 | * provide a better sense of ratio than this diagram: | |
5628 | * | |
5629 | * head --> tail | |
5630 | * +---------------------+----------+ | |
5631 | * ARC_mfu |:::::#:::::::::::::::|o#o###o###|-->. # already on L2ARC | |
5632 | * +---------------------+----------+ | o L2ARC eligible | |
5633 | * ARC_mru |:#:::::::::::::::::::|#o#ooo####|-->| : ARC buffer | |
5634 | * +---------------------+----------+ | | |
5635 | * 15.9 Gbytes ^ 32 Mbytes | | |
5636 | * headroom | | |
5637 | * l2arc_feed_thread() | |
5638 | * | | |
5639 | * l2arc write hand <--[oooo]--' | |
5640 | * | 8 Mbyte | |
5641 | * | write max | |
5642 | * V | |
5643 | * +==============================+ | |
5644 | * L2ARC dev |####|#|###|###| |####| ... | | |
5645 | * +==============================+ | |
5646 | * 32 Gbytes | |
5647 | * | |
5648 | * 3. If an ARC buffer is copied to the L2ARC but then hit instead of | |
5649 | * evicted, then the L2ARC has cached a buffer much sooner than it probably | |
5650 | * needed to, potentially wasting L2ARC device bandwidth and storage. It is | |
5651 | * safe to say that this is an uncommon case, since buffers at the end of | |
5652 | * the ARC lists have moved there due to inactivity. | |
5653 | * | |
5654 | * 4. If the ARC evicts faster than the L2ARC can maintain a headroom, | |
5655 | * then the L2ARC simply misses copying some buffers. This serves as a | |
5656 | * pressure valve to prevent heavy read workloads from both stalling the ARC | |
5657 | * with waits and clogging the L2ARC with writes. This also helps prevent | |
5658 | * the potential for the L2ARC to churn if it attempts to cache content too | |
5659 | * quickly, such as during backups of the entire pool. | |
5660 | * | |
b128c09f BB |
5661 | * 5. After system boot and before the ARC has filled main memory, there are |
5662 | * no evictions from the ARC and so the tails of the ARC_mfu and ARC_mru | |
5663 | * lists can remain mostly static. Instead of searching from tail of these | |
5664 | * lists as pictured, the l2arc_feed_thread() will search from the list heads | |
5665 | * for eligible buffers, greatly increasing its chance of finding them. | |
5666 | * | |
5667 | * The L2ARC device write speed is also boosted during this time so that | |
5668 | * the L2ARC warms up faster. Since there have been no ARC evictions yet, | |
5669 | * there are no L2ARC reads, and no fear of degrading read performance | |
5670 | * through increased writes. | |
5671 | * | |
5672 | * 6. Writes to the L2ARC devices are grouped and sent in-sequence, so that | |
34dc7c2f BB |
5673 | * the vdev queue can aggregate them into larger and fewer writes. Each |
5674 | * device is written to in a rotor fashion, sweeping writes through | |
5675 | * available space then repeating. | |
5676 | * | |
b128c09f | 5677 | * 7. The L2ARC does not store dirty content. It never needs to flush |
34dc7c2f BB |
5678 | * write buffers back to disk based storage. |
5679 | * | |
b128c09f | 5680 | * 8. If an ARC buffer is written (and dirtied) which also exists in the |
34dc7c2f BB |
5681 | * L2ARC, the now stale L2ARC buffer is immediately dropped. |
5682 | * | |
5683 | * The performance of the L2ARC can be tweaked by a number of tunables, which | |
5684 | * may be necessary for different workloads: | |
5685 | * | |
5686 | * l2arc_write_max max write bytes per interval | |
b128c09f | 5687 | * l2arc_write_boost extra write bytes during device warmup |
34dc7c2f | 5688 | * l2arc_noprefetch skip caching prefetched buffers |
3a17a7a9 | 5689 | * l2arc_nocompress skip compressing buffers |
34dc7c2f | 5690 | * l2arc_headroom number of max device writes to precache |
3a17a7a9 SK |
5691 | * l2arc_headroom_boost when we find compressed buffers during ARC |
5692 | * scanning, we multiply headroom by this | |
5693 | * percentage factor for the next scan cycle, | |
5694 | * since more compressed buffers are likely to | |
5695 | * be present | |
34dc7c2f BB |
5696 | * l2arc_feed_secs seconds between L2ARC writing |
5697 | * | |
5698 | * Tunables may be removed or added as future performance improvements are | |
5699 | * integrated, and also may become zpool properties. | |
d164b209 BB |
5700 | * |
5701 | * There are three key functions that control how the L2ARC warms up: | |
5702 | * | |
5703 | * l2arc_write_eligible() check if a buffer is eligible to cache | |
5704 | * l2arc_write_size() calculate how much to write | |
5705 | * l2arc_write_interval() calculate sleep delay between writes | |
5706 | * | |
5707 | * These three functions determine what to write, how much, and how quickly | |
5708 | * to send writes. | |
34dc7c2f BB |
5709 | */ |
5710 | ||
d164b209 | 5711 | static boolean_t |
2a432414 | 5712 | l2arc_write_eligible(uint64_t spa_guid, arc_buf_hdr_t *hdr) |
d164b209 BB |
5713 | { |
5714 | /* | |
5715 | * A buffer is *not* eligible for the L2ARC if it: | |
5716 | * 1. belongs to a different spa. | |
428870ff BB |
5717 | * 2. is already cached on the L2ARC. |
5718 | * 3. has an I/O in progress (it may be an incomplete read). | |
5719 | * 4. is flagged not eligible (zfs property). | |
d164b209 | 5720 | */ |
b9541d6b | 5721 | if (hdr->b_spa != spa_guid || HDR_HAS_L2HDR(hdr) || |
2a432414 | 5722 | HDR_IO_IN_PROGRESS(hdr) || !HDR_L2CACHE(hdr)) |
d164b209 BB |
5723 | return (B_FALSE); |
5724 | ||
5725 | return (B_TRUE); | |
5726 | } | |
5727 | ||
5728 | static uint64_t | |
3a17a7a9 | 5729 | l2arc_write_size(void) |
d164b209 BB |
5730 | { |
5731 | uint64_t size; | |
5732 | ||
3a17a7a9 SK |
5733 | /* |
5734 | * Make sure our globals have meaningful values in case the user | |
5735 | * altered them. | |
5736 | */ | |
5737 | size = l2arc_write_max; | |
5738 | if (size == 0) { | |
5739 | cmn_err(CE_NOTE, "Bad value for l2arc_write_max, value must " | |
5740 | "be greater than zero, resetting it to the default (%d)", | |
5741 | L2ARC_WRITE_SIZE); | |
5742 | size = l2arc_write_max = L2ARC_WRITE_SIZE; | |
5743 | } | |
d164b209 BB |
5744 | |
5745 | if (arc_warm == B_FALSE) | |
3a17a7a9 | 5746 | size += l2arc_write_boost; |
d164b209 BB |
5747 | |
5748 | return (size); | |
5749 | ||
5750 | } | |
5751 | ||
5752 | static clock_t | |
5753 | l2arc_write_interval(clock_t began, uint64_t wanted, uint64_t wrote) | |
5754 | { | |
428870ff | 5755 | clock_t interval, next, now; |
d164b209 BB |
5756 | |
5757 | /* | |
5758 | * If the ARC lists are busy, increase our write rate; if the | |
5759 | * lists are stale, idle back. This is achieved by checking | |
5760 | * how much we previously wrote - if it was more than half of | |
5761 | * what we wanted, schedule the next write much sooner. | |
5762 | */ | |
5763 | if (l2arc_feed_again && wrote > (wanted / 2)) | |
5764 | interval = (hz * l2arc_feed_min_ms) / 1000; | |
5765 | else | |
5766 | interval = hz * l2arc_feed_secs; | |
5767 | ||
428870ff BB |
5768 | now = ddi_get_lbolt(); |
5769 | next = MAX(now, MIN(now + interval, began + interval)); | |
d164b209 BB |
5770 | |
5771 | return (next); | |
5772 | } | |
5773 | ||
34dc7c2f BB |
5774 | /* |
5775 | * Cycle through L2ARC devices. This is how L2ARC load balances. | |
b128c09f | 5776 | * If a device is returned, this also returns holding the spa config lock. |
34dc7c2f BB |
5777 | */ |
5778 | static l2arc_dev_t * | |
5779 | l2arc_dev_get_next(void) | |
5780 | { | |
b128c09f | 5781 | l2arc_dev_t *first, *next = NULL; |
34dc7c2f | 5782 | |
b128c09f BB |
5783 | /* |
5784 | * Lock out the removal of spas (spa_namespace_lock), then removal | |
5785 | * of cache devices (l2arc_dev_mtx). Once a device has been selected, | |
5786 | * both locks will be dropped and a spa config lock held instead. | |
5787 | */ | |
5788 | mutex_enter(&spa_namespace_lock); | |
5789 | mutex_enter(&l2arc_dev_mtx); | |
5790 | ||
5791 | /* if there are no vdevs, there is nothing to do */ | |
5792 | if (l2arc_ndev == 0) | |
5793 | goto out; | |
5794 | ||
5795 | first = NULL; | |
5796 | next = l2arc_dev_last; | |
5797 | do { | |
5798 | /* loop around the list looking for a non-faulted vdev */ | |
5799 | if (next == NULL) { | |
34dc7c2f | 5800 | next = list_head(l2arc_dev_list); |
b128c09f BB |
5801 | } else { |
5802 | next = list_next(l2arc_dev_list, next); | |
5803 | if (next == NULL) | |
5804 | next = list_head(l2arc_dev_list); | |
5805 | } | |
5806 | ||
5807 | /* if we have come back to the start, bail out */ | |
5808 | if (first == NULL) | |
5809 | first = next; | |
5810 | else if (next == first) | |
5811 | break; | |
5812 | ||
5813 | } while (vdev_is_dead(next->l2ad_vdev)); | |
5814 | ||
5815 | /* if we were unable to find any usable vdevs, return NULL */ | |
5816 | if (vdev_is_dead(next->l2ad_vdev)) | |
5817 | next = NULL; | |
34dc7c2f BB |
5818 | |
5819 | l2arc_dev_last = next; | |
5820 | ||
b128c09f BB |
5821 | out: |
5822 | mutex_exit(&l2arc_dev_mtx); | |
5823 | ||
5824 | /* | |
5825 | * Grab the config lock to prevent the 'next' device from being | |
5826 | * removed while we are writing to it. | |
5827 | */ | |
5828 | if (next != NULL) | |
5829 | spa_config_enter(next->l2ad_spa, SCL_L2ARC, next, RW_READER); | |
5830 | mutex_exit(&spa_namespace_lock); | |
5831 | ||
34dc7c2f BB |
5832 | return (next); |
5833 | } | |
5834 | ||
b128c09f BB |
5835 | /* |
5836 | * Free buffers that were tagged for destruction. | |
5837 | */ | |
5838 | static void | |
0bc8fd78 | 5839 | l2arc_do_free_on_write(void) |
b128c09f BB |
5840 | { |
5841 | list_t *buflist; | |
5842 | l2arc_data_free_t *df, *df_prev; | |
5843 | ||
5844 | mutex_enter(&l2arc_free_on_write_mtx); | |
5845 | buflist = l2arc_free_on_write; | |
5846 | ||
5847 | for (df = list_tail(buflist); df; df = df_prev) { | |
5848 | df_prev = list_prev(buflist, df); | |
5849 | ASSERT(df->l2df_data != NULL); | |
5850 | ASSERT(df->l2df_func != NULL); | |
5851 | df->l2df_func(df->l2df_data, df->l2df_size); | |
5852 | list_remove(buflist, df); | |
5853 | kmem_free(df, sizeof (l2arc_data_free_t)); | |
5854 | } | |
5855 | ||
5856 | mutex_exit(&l2arc_free_on_write_mtx); | |
5857 | } | |
5858 | ||
34dc7c2f BB |
5859 | /* |
5860 | * A write to a cache device has completed. Update all headers to allow | |
5861 | * reads from these buffers to begin. | |
5862 | */ | |
5863 | static void | |
5864 | l2arc_write_done(zio_t *zio) | |
5865 | { | |
5866 | l2arc_write_callback_t *cb; | |
5867 | l2arc_dev_t *dev; | |
5868 | list_t *buflist; | |
2a432414 | 5869 | arc_buf_hdr_t *head, *hdr, *hdr_prev; |
34dc7c2f | 5870 | kmutex_t *hash_lock; |
3bec585e | 5871 | int64_t bytes_dropped = 0; |
34dc7c2f BB |
5872 | |
5873 | cb = zio->io_private; | |
5874 | ASSERT(cb != NULL); | |
5875 | dev = cb->l2wcb_dev; | |
5876 | ASSERT(dev != NULL); | |
5877 | head = cb->l2wcb_head; | |
5878 | ASSERT(head != NULL); | |
b9541d6b | 5879 | buflist = &dev->l2ad_buflist; |
34dc7c2f BB |
5880 | ASSERT(buflist != NULL); |
5881 | DTRACE_PROBE2(l2arc__iodone, zio_t *, zio, | |
5882 | l2arc_write_callback_t *, cb); | |
5883 | ||
5884 | if (zio->io_error != 0) | |
5885 | ARCSTAT_BUMP(arcstat_l2_writes_error); | |
5886 | ||
34dc7c2f BB |
5887 | /* |
5888 | * All writes completed, or an error was hit. | |
5889 | */ | |
ca0bf58d PS |
5890 | top: |
5891 | mutex_enter(&dev->l2ad_mtx); | |
2a432414 GW |
5892 | for (hdr = list_prev(buflist, head); hdr; hdr = hdr_prev) { |
5893 | hdr_prev = list_prev(buflist, hdr); | |
34dc7c2f | 5894 | |
2a432414 | 5895 | hash_lock = HDR_LOCK(hdr); |
ca0bf58d PS |
5896 | |
5897 | /* | |
5898 | * We cannot use mutex_enter or else we can deadlock | |
5899 | * with l2arc_write_buffers (due to swapping the order | |
5900 | * the hash lock and l2ad_mtx are taken). | |
5901 | */ | |
34dc7c2f BB |
5902 | if (!mutex_tryenter(hash_lock)) { |
5903 | /* | |
ca0bf58d PS |
5904 | * Missed the hash lock. We must retry so we |
5905 | * don't leave the ARC_FLAG_L2_WRITING bit set. | |
34dc7c2f | 5906 | */ |
ca0bf58d PS |
5907 | ARCSTAT_BUMP(arcstat_l2_writes_lock_retry); |
5908 | ||
5909 | /* | |
5910 | * We don't want to rescan the headers we've | |
5911 | * already marked as having been written out, so | |
5912 | * we reinsert the head node so we can pick up | |
5913 | * where we left off. | |
5914 | */ | |
5915 | list_remove(buflist, head); | |
5916 | list_insert_after(buflist, hdr, head); | |
5917 | ||
5918 | mutex_exit(&dev->l2ad_mtx); | |
5919 | ||
5920 | /* | |
5921 | * We wait for the hash lock to become available | |
5922 | * to try and prevent busy waiting, and increase | |
5923 | * the chance we'll be able to acquire the lock | |
5924 | * the next time around. | |
5925 | */ | |
5926 | mutex_enter(hash_lock); | |
5927 | mutex_exit(hash_lock); | |
5928 | goto top; | |
34dc7c2f BB |
5929 | } |
5930 | ||
b9541d6b | 5931 | /* |
ca0bf58d PS |
5932 | * We could not have been moved into the arc_l2c_only |
5933 | * state while in-flight due to our ARC_FLAG_L2_WRITING | |
5934 | * bit being set. Let's just ensure that's being enforced. | |
5935 | */ | |
5936 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
5937 | ||
5938 | /* | |
5939 | * We may have allocated a buffer for L2ARC compression, | |
5940 | * we must release it to avoid leaking this data. | |
b9541d6b | 5941 | */ |
ca0bf58d | 5942 | l2arc_release_cdata_buf(hdr); |
b9541d6b | 5943 | |
34dc7c2f BB |
5944 | if (zio->io_error != 0) { |
5945 | /* | |
b128c09f | 5946 | * Error - drop L2ARC entry. |
34dc7c2f | 5947 | */ |
2a432414 | 5948 | list_remove(buflist, hdr); |
b9541d6b CW |
5949 | hdr->b_flags &= ~ARC_FLAG_HAS_L2HDR; |
5950 | ||
5951 | ARCSTAT_INCR(arcstat_l2_asize, -hdr->b_l2hdr.b_asize); | |
2a432414 | 5952 | ARCSTAT_INCR(arcstat_l2_size, -hdr->b_size); |
d962d5da PS |
5953 | |
5954 | bytes_dropped += hdr->b_l2hdr.b_asize; | |
5955 | (void) refcount_remove_many(&dev->l2ad_alloc, | |
5956 | hdr->b_l2hdr.b_asize, hdr); | |
34dc7c2f BB |
5957 | } |
5958 | ||
5959 | /* | |
ca0bf58d PS |
5960 | * Allow ARC to begin reads and ghost list evictions to |
5961 | * this L2ARC entry. | |
34dc7c2f | 5962 | */ |
2a432414 | 5963 | hdr->b_flags &= ~ARC_FLAG_L2_WRITING; |
34dc7c2f BB |
5964 | |
5965 | mutex_exit(hash_lock); | |
5966 | } | |
5967 | ||
5968 | atomic_inc_64(&l2arc_writes_done); | |
5969 | list_remove(buflist, head); | |
b9541d6b CW |
5970 | ASSERT(!HDR_HAS_L1HDR(head)); |
5971 | kmem_cache_free(hdr_l2only_cache, head); | |
5972 | mutex_exit(&dev->l2ad_mtx); | |
34dc7c2f | 5973 | |
3bec585e SK |
5974 | vdev_space_update(dev->l2ad_vdev, -bytes_dropped, 0, 0); |
5975 | ||
b128c09f | 5976 | l2arc_do_free_on_write(); |
34dc7c2f BB |
5977 | |
5978 | kmem_free(cb, sizeof (l2arc_write_callback_t)); | |
5979 | } | |
5980 | ||
5981 | /* | |
5982 | * A read to a cache device completed. Validate buffer contents before | |
5983 | * handing over to the regular ARC routines. | |
5984 | */ | |
5985 | static void | |
5986 | l2arc_read_done(zio_t *zio) | |
5987 | { | |
5988 | l2arc_read_callback_t *cb; | |
5989 | arc_buf_hdr_t *hdr; | |
5990 | arc_buf_t *buf; | |
34dc7c2f | 5991 | kmutex_t *hash_lock; |
b128c09f BB |
5992 | int equal; |
5993 | ||
5994 | ASSERT(zio->io_vd != NULL); | |
5995 | ASSERT(zio->io_flags & ZIO_FLAG_DONT_PROPAGATE); | |
5996 | ||
5997 | spa_config_exit(zio->io_spa, SCL_L2ARC, zio->io_vd); | |
34dc7c2f BB |
5998 | |
5999 | cb = zio->io_private; | |
6000 | ASSERT(cb != NULL); | |
6001 | buf = cb->l2rcb_buf; | |
6002 | ASSERT(buf != NULL); | |
34dc7c2f | 6003 | |
428870ff | 6004 | hash_lock = HDR_LOCK(buf->b_hdr); |
34dc7c2f | 6005 | mutex_enter(hash_lock); |
428870ff BB |
6006 | hdr = buf->b_hdr; |
6007 | ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); | |
34dc7c2f | 6008 | |
3a17a7a9 SK |
6009 | /* |
6010 | * If the buffer was compressed, decompress it first. | |
6011 | */ | |
6012 | if (cb->l2rcb_compress != ZIO_COMPRESS_OFF) | |
6013 | l2arc_decompress_zio(zio, hdr, cb->l2rcb_compress); | |
6014 | ASSERT(zio->io_data != NULL); | |
6015 | ||
34dc7c2f BB |
6016 | /* |
6017 | * Check this survived the L2ARC journey. | |
6018 | */ | |
6019 | equal = arc_cksum_equal(buf); | |
6020 | if (equal && zio->io_error == 0 && !HDR_L2_EVICTED(hdr)) { | |
6021 | mutex_exit(hash_lock); | |
6022 | zio->io_private = buf; | |
b128c09f BB |
6023 | zio->io_bp_copy = cb->l2rcb_bp; /* XXX fix in L2ARC 2.0 */ |
6024 | zio->io_bp = &zio->io_bp_copy; /* XXX fix in L2ARC 2.0 */ | |
34dc7c2f BB |
6025 | arc_read_done(zio); |
6026 | } else { | |
6027 | mutex_exit(hash_lock); | |
6028 | /* | |
6029 | * Buffer didn't survive caching. Increment stats and | |
6030 | * reissue to the original storage device. | |
6031 | */ | |
b128c09f | 6032 | if (zio->io_error != 0) { |
34dc7c2f | 6033 | ARCSTAT_BUMP(arcstat_l2_io_error); |
b128c09f | 6034 | } else { |
2e528b49 | 6035 | zio->io_error = SET_ERROR(EIO); |
b128c09f | 6036 | } |
34dc7c2f BB |
6037 | if (!equal) |
6038 | ARCSTAT_BUMP(arcstat_l2_cksum_bad); | |
6039 | ||
34dc7c2f | 6040 | /* |
b128c09f BB |
6041 | * If there's no waiter, issue an async i/o to the primary |
6042 | * storage now. If there *is* a waiter, the caller must | |
6043 | * issue the i/o in a context where it's OK to block. | |
34dc7c2f | 6044 | */ |
d164b209 BB |
6045 | if (zio->io_waiter == NULL) { |
6046 | zio_t *pio = zio_unique_parent(zio); | |
6047 | ||
6048 | ASSERT(!pio || pio->io_child_type == ZIO_CHILD_LOGICAL); | |
6049 | ||
6050 | zio_nowait(zio_read(pio, cb->l2rcb_spa, &cb->l2rcb_bp, | |
b128c09f BB |
6051 | buf->b_data, zio->io_size, arc_read_done, buf, |
6052 | zio->io_priority, cb->l2rcb_flags, &cb->l2rcb_zb)); | |
d164b209 | 6053 | } |
34dc7c2f BB |
6054 | } |
6055 | ||
6056 | kmem_free(cb, sizeof (l2arc_read_callback_t)); | |
6057 | } | |
6058 | ||
6059 | /* | |
6060 | * This is the list priority from which the L2ARC will search for pages to | |
6061 | * cache. This is used within loops (0..3) to cycle through lists in the | |
6062 | * desired order. This order can have a significant effect on cache | |
6063 | * performance. | |
6064 | * | |
6065 | * Currently the metadata lists are hit first, MFU then MRU, followed by | |
6066 | * the data lists. This function returns a locked list, and also returns | |
6067 | * the lock pointer. | |
6068 | */ | |
ca0bf58d PS |
6069 | static multilist_sublist_t * |
6070 | l2arc_sublist_lock(int list_num) | |
34dc7c2f | 6071 | { |
ca0bf58d PS |
6072 | multilist_t *ml = NULL; |
6073 | unsigned int idx; | |
34dc7c2f BB |
6074 | |
6075 | ASSERT(list_num >= 0 && list_num <= 3); | |
6076 | ||
6077 | switch (list_num) { | |
6078 | case 0: | |
ca0bf58d | 6079 | ml = &arc_mfu->arcs_list[ARC_BUFC_METADATA]; |
34dc7c2f BB |
6080 | break; |
6081 | case 1: | |
ca0bf58d | 6082 | ml = &arc_mru->arcs_list[ARC_BUFC_METADATA]; |
34dc7c2f BB |
6083 | break; |
6084 | case 2: | |
ca0bf58d | 6085 | ml = &arc_mfu->arcs_list[ARC_BUFC_DATA]; |
34dc7c2f BB |
6086 | break; |
6087 | case 3: | |
ca0bf58d | 6088 | ml = &arc_mru->arcs_list[ARC_BUFC_DATA]; |
34dc7c2f BB |
6089 | break; |
6090 | } | |
6091 | ||
ca0bf58d PS |
6092 | /* |
6093 | * Return a randomly-selected sublist. This is acceptable | |
6094 | * because the caller feeds only a little bit of data for each | |
6095 | * call (8MB). Subsequent calls will result in different | |
6096 | * sublists being selected. | |
6097 | */ | |
6098 | idx = multilist_get_random_index(ml); | |
6099 | return (multilist_sublist_lock(ml, idx)); | |
34dc7c2f BB |
6100 | } |
6101 | ||
6102 | /* | |
6103 | * Evict buffers from the device write hand to the distance specified in | |
6104 | * bytes. This distance may span populated buffers, it may span nothing. | |
6105 | * This is clearing a region on the L2ARC device ready for writing. | |
6106 | * If the 'all' boolean is set, every buffer is evicted. | |
6107 | */ | |
6108 | static void | |
6109 | l2arc_evict(l2arc_dev_t *dev, uint64_t distance, boolean_t all) | |
6110 | { | |
6111 | list_t *buflist; | |
2a432414 | 6112 | arc_buf_hdr_t *hdr, *hdr_prev; |
34dc7c2f BB |
6113 | kmutex_t *hash_lock; |
6114 | uint64_t taddr; | |
6115 | ||
b9541d6b | 6116 | buflist = &dev->l2ad_buflist; |
34dc7c2f BB |
6117 | |
6118 | if (!all && dev->l2ad_first) { | |
6119 | /* | |
6120 | * This is the first sweep through the device. There is | |
6121 | * nothing to evict. | |
6122 | */ | |
6123 | return; | |
6124 | } | |
6125 | ||
b128c09f | 6126 | if (dev->l2ad_hand >= (dev->l2ad_end - (2 * distance))) { |
34dc7c2f BB |
6127 | /* |
6128 | * When nearing the end of the device, evict to the end | |
6129 | * before the device write hand jumps to the start. | |
6130 | */ | |
6131 | taddr = dev->l2ad_end; | |
6132 | } else { | |
6133 | taddr = dev->l2ad_hand + distance; | |
6134 | } | |
6135 | DTRACE_PROBE4(l2arc__evict, l2arc_dev_t *, dev, list_t *, buflist, | |
6136 | uint64_t, taddr, boolean_t, all); | |
6137 | ||
6138 | top: | |
b9541d6b | 6139 | mutex_enter(&dev->l2ad_mtx); |
2a432414 GW |
6140 | for (hdr = list_tail(buflist); hdr; hdr = hdr_prev) { |
6141 | hdr_prev = list_prev(buflist, hdr); | |
34dc7c2f | 6142 | |
2a432414 | 6143 | hash_lock = HDR_LOCK(hdr); |
ca0bf58d PS |
6144 | |
6145 | /* | |
6146 | * We cannot use mutex_enter or else we can deadlock | |
6147 | * with l2arc_write_buffers (due to swapping the order | |
6148 | * the hash lock and l2ad_mtx are taken). | |
6149 | */ | |
34dc7c2f BB |
6150 | if (!mutex_tryenter(hash_lock)) { |
6151 | /* | |
6152 | * Missed the hash lock. Retry. | |
6153 | */ | |
6154 | ARCSTAT_BUMP(arcstat_l2_evict_lock_retry); | |
b9541d6b | 6155 | mutex_exit(&dev->l2ad_mtx); |
34dc7c2f BB |
6156 | mutex_enter(hash_lock); |
6157 | mutex_exit(hash_lock); | |
6158 | goto top; | |
6159 | } | |
6160 | ||
2a432414 | 6161 | if (HDR_L2_WRITE_HEAD(hdr)) { |
34dc7c2f BB |
6162 | /* |
6163 | * We hit a write head node. Leave it for | |
6164 | * l2arc_write_done(). | |
6165 | */ | |
2a432414 | 6166 | list_remove(buflist, hdr); |
34dc7c2f BB |
6167 | mutex_exit(hash_lock); |
6168 | continue; | |
6169 | } | |
6170 | ||
b9541d6b CW |
6171 | if (!all && HDR_HAS_L2HDR(hdr) && |
6172 | (hdr->b_l2hdr.b_daddr > taddr || | |
6173 | hdr->b_l2hdr.b_daddr < dev->l2ad_hand)) { | |
34dc7c2f BB |
6174 | /* |
6175 | * We've evicted to the target address, | |
6176 | * or the end of the device. | |
6177 | */ | |
6178 | mutex_exit(hash_lock); | |
6179 | break; | |
6180 | } | |
6181 | ||
b9541d6b CW |
6182 | ASSERT(HDR_HAS_L2HDR(hdr)); |
6183 | if (!HDR_HAS_L1HDR(hdr)) { | |
2a432414 | 6184 | ASSERT(!HDR_L2_READING(hdr)); |
34dc7c2f BB |
6185 | /* |
6186 | * This doesn't exist in the ARC. Destroy. | |
6187 | * arc_hdr_destroy() will call list_remove() | |
6188 | * and decrement arcstat_l2_size. | |
6189 | */ | |
2a432414 GW |
6190 | arc_change_state(arc_anon, hdr, hash_lock); |
6191 | arc_hdr_destroy(hdr); | |
34dc7c2f | 6192 | } else { |
b9541d6b CW |
6193 | ASSERT(hdr->b_l1hdr.b_state != arc_l2c_only); |
6194 | ARCSTAT_BUMP(arcstat_l2_evict_l1cached); | |
b128c09f BB |
6195 | /* |
6196 | * Invalidate issued or about to be issued | |
6197 | * reads, since we may be about to write | |
6198 | * over this location. | |
6199 | */ | |
2a432414 | 6200 | if (HDR_L2_READING(hdr)) { |
b128c09f | 6201 | ARCSTAT_BUMP(arcstat_l2_evict_reading); |
2a432414 | 6202 | hdr->b_flags |= ARC_FLAG_L2_EVICTED; |
b128c09f BB |
6203 | } |
6204 | ||
ca0bf58d PS |
6205 | /* Ensure this header has finished being written */ |
6206 | ASSERT(!HDR_L2_WRITING(hdr)); | |
6207 | ASSERT3P(hdr->b_l1hdr.b_tmp_cdata, ==, NULL); | |
d962d5da PS |
6208 | |
6209 | arc_hdr_l2hdr_destroy(hdr); | |
34dc7c2f BB |
6210 | } |
6211 | mutex_exit(hash_lock); | |
6212 | } | |
b9541d6b | 6213 | mutex_exit(&dev->l2ad_mtx); |
34dc7c2f BB |
6214 | } |
6215 | ||
6216 | /* | |
6217 | * Find and write ARC buffers to the L2ARC device. | |
6218 | * | |
2a432414 | 6219 | * An ARC_FLAG_L2_WRITING flag is set so that the L2ARC buffers are not valid |
34dc7c2f | 6220 | * for reading until they have completed writing. |
3a17a7a9 SK |
6221 | * The headroom_boost is an in-out parameter used to maintain headroom boost |
6222 | * state between calls to this function. | |
6223 | * | |
6224 | * Returns the number of bytes actually written (which may be smaller than | |
6225 | * the delta by which the device hand has changed due to alignment). | |
34dc7c2f | 6226 | */ |
d164b209 | 6227 | static uint64_t |
3a17a7a9 SK |
6228 | l2arc_write_buffers(spa_t *spa, l2arc_dev_t *dev, uint64_t target_sz, |
6229 | boolean_t *headroom_boost) | |
34dc7c2f | 6230 | { |
2a432414 | 6231 | arc_buf_hdr_t *hdr, *hdr_prev, *head; |
ef56b078 AG |
6232 | uint64_t write_asize, write_sz, headroom, buf_compress_minsz, |
6233 | stats_size; | |
34dc7c2f | 6234 | void *buf_data; |
3a17a7a9 | 6235 | boolean_t full; |
34dc7c2f BB |
6236 | l2arc_write_callback_t *cb; |
6237 | zio_t *pio, *wzio; | |
3541dc6d | 6238 | uint64_t guid = spa_load_guid(spa); |
d6320ddb | 6239 | int try; |
3a17a7a9 | 6240 | const boolean_t do_headroom_boost = *headroom_boost; |
34dc7c2f | 6241 | |
34dc7c2f BB |
6242 | ASSERT(dev->l2ad_vdev != NULL); |
6243 | ||
3a17a7a9 SK |
6244 | /* Lower the flag now, we might want to raise it again later. */ |
6245 | *headroom_boost = B_FALSE; | |
6246 | ||
34dc7c2f | 6247 | pio = NULL; |
ef56b078 | 6248 | write_sz = write_asize = 0; |
34dc7c2f | 6249 | full = B_FALSE; |
b9541d6b | 6250 | head = kmem_cache_alloc(hdr_l2only_cache, KM_PUSHPAGE); |
2a432414 | 6251 | head->b_flags |= ARC_FLAG_L2_WRITE_HEAD; |
b9541d6b | 6252 | head->b_flags |= ARC_FLAG_HAS_L2HDR; |
34dc7c2f | 6253 | |
3a17a7a9 SK |
6254 | /* |
6255 | * We will want to try to compress buffers that are at least 2x the | |
6256 | * device sector size. | |
6257 | */ | |
6258 | buf_compress_minsz = 2 << dev->l2ad_vdev->vdev_ashift; | |
6259 | ||
34dc7c2f BB |
6260 | /* |
6261 | * Copy buffers for L2ARC writing. | |
6262 | */ | |
d6320ddb | 6263 | for (try = 0; try <= 3; try++) { |
ca0bf58d | 6264 | multilist_sublist_t *mls = l2arc_sublist_lock(try); |
3a17a7a9 SK |
6265 | uint64_t passed_sz = 0; |
6266 | ||
b128c09f BB |
6267 | /* |
6268 | * L2ARC fast warmup. | |
6269 | * | |
6270 | * Until the ARC is warm and starts to evict, read from the | |
6271 | * head of the ARC lists rather than the tail. | |
6272 | */ | |
b128c09f | 6273 | if (arc_warm == B_FALSE) |
ca0bf58d | 6274 | hdr = multilist_sublist_head(mls); |
b128c09f | 6275 | else |
ca0bf58d | 6276 | hdr = multilist_sublist_tail(mls); |
b128c09f | 6277 | |
3a17a7a9 SK |
6278 | headroom = target_sz * l2arc_headroom; |
6279 | if (do_headroom_boost) | |
6280 | headroom = (headroom * l2arc_headroom_boost) / 100; | |
6281 | ||
2a432414 | 6282 | for (; hdr; hdr = hdr_prev) { |
3a17a7a9 SK |
6283 | kmutex_t *hash_lock; |
6284 | uint64_t buf_sz; | |
ef56b078 | 6285 | uint64_t buf_a_sz; |
3a17a7a9 | 6286 | |
b128c09f | 6287 | if (arc_warm == B_FALSE) |
ca0bf58d | 6288 | hdr_prev = multilist_sublist_next(mls, hdr); |
b128c09f | 6289 | else |
ca0bf58d | 6290 | hdr_prev = multilist_sublist_prev(mls, hdr); |
34dc7c2f | 6291 | |
2a432414 | 6292 | hash_lock = HDR_LOCK(hdr); |
3a17a7a9 | 6293 | if (!mutex_tryenter(hash_lock)) { |
34dc7c2f BB |
6294 | /* |
6295 | * Skip this buffer rather than waiting. | |
6296 | */ | |
6297 | continue; | |
6298 | } | |
6299 | ||
2a432414 | 6300 | passed_sz += hdr->b_size; |
34dc7c2f BB |
6301 | if (passed_sz > headroom) { |
6302 | /* | |
6303 | * Searched too far. | |
6304 | */ | |
6305 | mutex_exit(hash_lock); | |
6306 | break; | |
6307 | } | |
6308 | ||
2a432414 | 6309 | if (!l2arc_write_eligible(guid, hdr)) { |
34dc7c2f BB |
6310 | mutex_exit(hash_lock); |
6311 | continue; | |
6312 | } | |
6313 | ||
ef56b078 AG |
6314 | /* |
6315 | * Assume that the buffer is not going to be compressed | |
6316 | * and could take more space on disk because of a larger | |
6317 | * disk block size. | |
6318 | */ | |
6319 | buf_sz = hdr->b_size; | |
6320 | buf_a_sz = vdev_psize_to_asize(dev->l2ad_vdev, buf_sz); | |
6321 | ||
6322 | if ((write_asize + buf_a_sz) > target_sz) { | |
34dc7c2f BB |
6323 | full = B_TRUE; |
6324 | mutex_exit(hash_lock); | |
6325 | break; | |
6326 | } | |
6327 | ||
34dc7c2f BB |
6328 | if (pio == NULL) { |
6329 | /* | |
6330 | * Insert a dummy header on the buflist so | |
6331 | * l2arc_write_done() can find where the | |
6332 | * write buffers begin without searching. | |
6333 | */ | |
ca0bf58d | 6334 | mutex_enter(&dev->l2ad_mtx); |
b9541d6b | 6335 | list_insert_head(&dev->l2ad_buflist, head); |
ca0bf58d | 6336 | mutex_exit(&dev->l2ad_mtx); |
34dc7c2f | 6337 | |
409dc1a5 | 6338 | cb = kmem_alloc(sizeof (l2arc_write_callback_t), |
79c76d5b | 6339 | KM_SLEEP); |
34dc7c2f BB |
6340 | cb->l2wcb_dev = dev; |
6341 | cb->l2wcb_head = head; | |
6342 | pio = zio_root(spa, l2arc_write_done, cb, | |
6343 | ZIO_FLAG_CANFAIL); | |
6344 | } | |
6345 | ||
6346 | /* | |
6347 | * Create and add a new L2ARC header. | |
6348 | */ | |
b9541d6b | 6349 | hdr->b_l2hdr.b_dev = dev; |
2a432414 | 6350 | hdr->b_flags |= ARC_FLAG_L2_WRITING; |
3a17a7a9 SK |
6351 | /* |
6352 | * Temporarily stash the data buffer in b_tmp_cdata. | |
6353 | * The subsequent write step will pick it up from | |
b9541d6b | 6354 | * there. This is because can't access b_l1hdr.b_buf |
3a17a7a9 SK |
6355 | * without holding the hash_lock, which we in turn |
6356 | * can't access without holding the ARC list locks | |
6357 | * (which we want to avoid during compression/writing) | |
6358 | */ | |
b9541d6b CW |
6359 | HDR_SET_COMPRESS(hdr, ZIO_COMPRESS_OFF); |
6360 | hdr->b_l2hdr.b_asize = hdr->b_size; | |
6361 | hdr->b_l2hdr.b_hits = 0; | |
6362 | hdr->b_l1hdr.b_tmp_cdata = hdr->b_l1hdr.b_buf->b_data; | |
3a17a7a9 | 6363 | |
d962d5da PS |
6364 | /* |
6365 | * Explicitly set the b_daddr field to a known | |
6366 | * value which means "invalid address". This | |
6367 | * enables us to differentiate which stage of | |
6368 | * l2arc_write_buffers() the particular header | |
6369 | * is in (e.g. this loop, or the one below). | |
6370 | * ARC_FLAG_L2_WRITING is not enough to make | |
6371 | * this distinction, and we need to know in | |
6372 | * order to do proper l2arc vdev accounting in | |
6373 | * arc_release() and arc_hdr_destroy(). | |
6374 | * | |
6375 | * Note, we can't use a new flag to distinguish | |
6376 | * the two stages because we don't hold the | |
6377 | * header's hash_lock below, in the second stage | |
6378 | * of this function. Thus, we can't simply | |
6379 | * change the b_flags field to denote that the | |
6380 | * IO has been sent. We can change the b_daddr | |
6381 | * field of the L2 portion, though, since we'll | |
6382 | * be holding the l2ad_mtx; which is why we're | |
6383 | * using it to denote the header's state change. | |
6384 | */ | |
6385 | hdr->b_l2hdr.b_daddr = L2ARC_ADDR_UNSET; | |
b9541d6b | 6386 | hdr->b_flags |= ARC_FLAG_HAS_L2HDR; |
3a17a7a9 | 6387 | |
ca0bf58d | 6388 | mutex_enter(&dev->l2ad_mtx); |
b9541d6b | 6389 | list_insert_head(&dev->l2ad_buflist, hdr); |
ca0bf58d | 6390 | mutex_exit(&dev->l2ad_mtx); |
34dc7c2f BB |
6391 | |
6392 | /* | |
6393 | * Compute and store the buffer cksum before | |
6394 | * writing. On debug the cksum is verified first. | |
6395 | */ | |
b9541d6b CW |
6396 | arc_cksum_verify(hdr->b_l1hdr.b_buf); |
6397 | arc_cksum_compute(hdr->b_l1hdr.b_buf, B_TRUE); | |
34dc7c2f BB |
6398 | |
6399 | mutex_exit(hash_lock); | |
6400 | ||
3a17a7a9 | 6401 | write_sz += buf_sz; |
ef56b078 | 6402 | write_asize += buf_a_sz; |
3a17a7a9 SK |
6403 | } |
6404 | ||
ca0bf58d | 6405 | multilist_sublist_unlock(mls); |
3a17a7a9 SK |
6406 | |
6407 | if (full == B_TRUE) | |
6408 | break; | |
6409 | } | |
6410 | ||
6411 | /* No buffers selected for writing? */ | |
6412 | if (pio == NULL) { | |
6413 | ASSERT0(write_sz); | |
b9541d6b CW |
6414 | ASSERT(!HDR_HAS_L1HDR(head)); |
6415 | kmem_cache_free(hdr_l2only_cache, head); | |
3a17a7a9 SK |
6416 | return (0); |
6417 | } | |
6418 | ||
ca0bf58d PS |
6419 | mutex_enter(&dev->l2ad_mtx); |
6420 | ||
ef56b078 AG |
6421 | /* |
6422 | * Note that elsewhere in this file arcstat_l2_asize | |
6423 | * and the used space on l2ad_vdev are updated using b_asize, | |
6424 | * which is not necessarily rounded up to the device block size. | |
6425 | * Too keep accounting consistent we do the same here as well: | |
6426 | * stats_size accumulates the sum of b_asize of the written buffers, | |
6427 | * while write_asize accumulates the sum of b_asize rounded up | |
6428 | * to the device block size. | |
6429 | * The latter sum is used only to validate the corectness of the code. | |
6430 | */ | |
6431 | stats_size = 0; | |
6432 | write_asize = 0; | |
6433 | ||
3a17a7a9 SK |
6434 | /* |
6435 | * Now start writing the buffers. We're starting at the write head | |
6436 | * and work backwards, retracing the course of the buffer selector | |
6437 | * loop above. | |
6438 | */ | |
b9541d6b CW |
6439 | for (hdr = list_prev(&dev->l2ad_buflist, head); hdr; |
6440 | hdr = list_prev(&dev->l2ad_buflist, hdr)) { | |
3a17a7a9 SK |
6441 | uint64_t buf_sz; |
6442 | ||
ca0bf58d PS |
6443 | /* |
6444 | * We rely on the L1 portion of the header below, so | |
6445 | * it's invalid for this header to have been evicted out | |
6446 | * of the ghost cache, prior to being written out. The | |
6447 | * ARC_FLAG_L2_WRITING bit ensures this won't happen. | |
6448 | */ | |
6449 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
6450 | ||
3a17a7a9 SK |
6451 | /* |
6452 | * We shouldn't need to lock the buffer here, since we flagged | |
2a432414 GW |
6453 | * it as ARC_FLAG_L2_WRITING in the previous step, but we must |
6454 | * take care to only access its L2 cache parameters. In | |
b9541d6b | 6455 | * particular, hdr->l1hdr.b_buf may be invalid by now due to |
2a432414 | 6456 | * ARC eviction. |
3a17a7a9 | 6457 | */ |
b9541d6b | 6458 | hdr->b_l2hdr.b_daddr = dev->l2ad_hand; |
3a17a7a9 | 6459 | |
b9541d6b CW |
6460 | if ((!l2arc_nocompress && HDR_L2COMPRESS(hdr)) && |
6461 | hdr->b_l2hdr.b_asize >= buf_compress_minsz) { | |
6462 | if (l2arc_compress_buf(hdr)) { | |
3a17a7a9 SK |
6463 | /* |
6464 | * If compression succeeded, enable headroom | |
6465 | * boost on the next scan cycle. | |
6466 | */ | |
6467 | *headroom_boost = B_TRUE; | |
6468 | } | |
6469 | } | |
6470 | ||
6471 | /* | |
6472 | * Pick up the buffer data we had previously stashed away | |
6473 | * (and now potentially also compressed). | |
6474 | */ | |
b9541d6b CW |
6475 | buf_data = hdr->b_l1hdr.b_tmp_cdata; |
6476 | buf_sz = hdr->b_l2hdr.b_asize; | |
3a17a7a9 | 6477 | |
d962d5da PS |
6478 | /* |
6479 | * We need to do this regardless if buf_sz is zero or | |
6480 | * not, otherwise, when this l2hdr is evicted we'll | |
6481 | * remove a reference that was never added. | |
6482 | */ | |
6483 | (void) refcount_add_many(&dev->l2ad_alloc, buf_sz, hdr); | |
6484 | ||
3a17a7a9 SK |
6485 | /* Compression may have squashed the buffer to zero length. */ |
6486 | if (buf_sz != 0) { | |
ef56b078 | 6487 | uint64_t buf_a_sz; |
3a17a7a9 | 6488 | |
34dc7c2f BB |
6489 | wzio = zio_write_phys(pio, dev->l2ad_vdev, |
6490 | dev->l2ad_hand, buf_sz, buf_data, ZIO_CHECKSUM_OFF, | |
6491 | NULL, NULL, ZIO_PRIORITY_ASYNC_WRITE, | |
6492 | ZIO_FLAG_CANFAIL, B_FALSE); | |
6493 | ||
6494 | DTRACE_PROBE2(l2arc__write, vdev_t *, dev->l2ad_vdev, | |
6495 | zio_t *, wzio); | |
6496 | (void) zio_nowait(wzio); | |
6497 | ||
ef56b078 | 6498 | stats_size += buf_sz; |
d962d5da | 6499 | |
b128c09f BB |
6500 | /* |
6501 | * Keep the clock hand suitably device-aligned. | |
6502 | */ | |
ef56b078 AG |
6503 | buf_a_sz = vdev_psize_to_asize(dev->l2ad_vdev, buf_sz); |
6504 | write_asize += buf_a_sz; | |
6505 | dev->l2ad_hand += buf_a_sz; | |
34dc7c2f | 6506 | } |
34dc7c2f | 6507 | } |
34dc7c2f | 6508 | |
b9541d6b | 6509 | mutex_exit(&dev->l2ad_mtx); |
34dc7c2f | 6510 | |
3a17a7a9 | 6511 | ASSERT3U(write_asize, <=, target_sz); |
34dc7c2f | 6512 | ARCSTAT_BUMP(arcstat_l2_writes_sent); |
3a17a7a9 | 6513 | ARCSTAT_INCR(arcstat_l2_write_bytes, write_asize); |
34dc7c2f | 6514 | ARCSTAT_INCR(arcstat_l2_size, write_sz); |
ef56b078 AG |
6515 | ARCSTAT_INCR(arcstat_l2_asize, stats_size); |
6516 | vdev_space_update(dev->l2ad_vdev, stats_size, 0, 0); | |
34dc7c2f BB |
6517 | |
6518 | /* | |
6519 | * Bump device hand to the device start if it is approaching the end. | |
6520 | * l2arc_evict() will already have evicted ahead for this case. | |
6521 | */ | |
b128c09f | 6522 | if (dev->l2ad_hand >= (dev->l2ad_end - target_sz)) { |
34dc7c2f | 6523 | dev->l2ad_hand = dev->l2ad_start; |
34dc7c2f BB |
6524 | dev->l2ad_first = B_FALSE; |
6525 | } | |
6526 | ||
d164b209 | 6527 | dev->l2ad_writing = B_TRUE; |
34dc7c2f | 6528 | (void) zio_wait(pio); |
d164b209 BB |
6529 | dev->l2ad_writing = B_FALSE; |
6530 | ||
3a17a7a9 SK |
6531 | return (write_asize); |
6532 | } | |
6533 | ||
6534 | /* | |
6535 | * Compresses an L2ARC buffer. | |
b9541d6b | 6536 | * The data to be compressed must be prefilled in l1hdr.b_tmp_cdata and its |
3a17a7a9 SK |
6537 | * size in l2hdr->b_asize. This routine tries to compress the data and |
6538 | * depending on the compression result there are three possible outcomes: | |
6539 | * *) The buffer was incompressible. The original l2hdr contents were left | |
6540 | * untouched and are ready for writing to an L2 device. | |
6541 | * *) The buffer was all-zeros, so there is no need to write it to an L2 | |
6542 | * device. To indicate this situation b_tmp_cdata is NULL'ed, b_asize is | |
6543 | * set to zero and b_compress is set to ZIO_COMPRESS_EMPTY. | |
6544 | * *) Compression succeeded and b_tmp_cdata was replaced with a temporary | |
6545 | * data buffer which holds the compressed data to be written, and b_asize | |
6546 | * tells us how much data there is. b_compress is set to the appropriate | |
6547 | * compression algorithm. Once writing is done, invoke | |
6548 | * l2arc_release_cdata_buf on this l2hdr to free this temporary buffer. | |
6549 | * | |
6550 | * Returns B_TRUE if compression succeeded, or B_FALSE if it didn't (the | |
6551 | * buffer was incompressible). | |
6552 | */ | |
6553 | static boolean_t | |
b9541d6b | 6554 | l2arc_compress_buf(arc_buf_hdr_t *hdr) |
3a17a7a9 SK |
6555 | { |
6556 | void *cdata; | |
9b67f605 | 6557 | size_t csize, len, rounded; |
b9541d6b | 6558 | l2arc_buf_hdr_t *l2hdr; |
3a17a7a9 | 6559 | |
b9541d6b CW |
6560 | ASSERT(HDR_HAS_L2HDR(hdr)); |
6561 | ||
6562 | l2hdr = &hdr->b_l2hdr; | |
6563 | ||
6564 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
6565 | ASSERT(HDR_GET_COMPRESS(hdr) == ZIO_COMPRESS_OFF); | |
6566 | ASSERT(hdr->b_l1hdr.b_tmp_cdata != NULL); | |
3a17a7a9 SK |
6567 | |
6568 | len = l2hdr->b_asize; | |
6569 | cdata = zio_data_buf_alloc(len); | |
b9541d6b CW |
6570 | ASSERT3P(cdata, !=, NULL); |
6571 | csize = zio_compress_data(ZIO_COMPRESS_LZ4, hdr->b_l1hdr.b_tmp_cdata, | |
3a17a7a9 SK |
6572 | cdata, l2hdr->b_asize); |
6573 | ||
9b67f605 MA |
6574 | rounded = P2ROUNDUP(csize, (size_t)SPA_MINBLOCKSIZE); |
6575 | if (rounded > csize) { | |
6576 | bzero((char *)cdata + csize, rounded - csize); | |
6577 | csize = rounded; | |
6578 | } | |
6579 | ||
3a17a7a9 SK |
6580 | if (csize == 0) { |
6581 | /* zero block, indicate that there's nothing to write */ | |
6582 | zio_data_buf_free(cdata, len); | |
b9541d6b | 6583 | HDR_SET_COMPRESS(hdr, ZIO_COMPRESS_EMPTY); |
3a17a7a9 | 6584 | l2hdr->b_asize = 0; |
b9541d6b | 6585 | hdr->b_l1hdr.b_tmp_cdata = NULL; |
3a17a7a9 SK |
6586 | ARCSTAT_BUMP(arcstat_l2_compress_zeros); |
6587 | return (B_TRUE); | |
6588 | } else if (csize > 0 && csize < len) { | |
6589 | /* | |
6590 | * Compression succeeded, we'll keep the cdata around for | |
6591 | * writing and release it afterwards. | |
6592 | */ | |
b9541d6b | 6593 | HDR_SET_COMPRESS(hdr, ZIO_COMPRESS_LZ4); |
3a17a7a9 | 6594 | l2hdr->b_asize = csize; |
b9541d6b | 6595 | hdr->b_l1hdr.b_tmp_cdata = cdata; |
3a17a7a9 SK |
6596 | ARCSTAT_BUMP(arcstat_l2_compress_successes); |
6597 | return (B_TRUE); | |
6598 | } else { | |
6599 | /* | |
6600 | * Compression failed, release the compressed buffer. | |
6601 | * l2hdr will be left unmodified. | |
6602 | */ | |
6603 | zio_data_buf_free(cdata, len); | |
6604 | ARCSTAT_BUMP(arcstat_l2_compress_failures); | |
6605 | return (B_FALSE); | |
6606 | } | |
6607 | } | |
6608 | ||
6609 | /* | |
6610 | * Decompresses a zio read back from an l2arc device. On success, the | |
6611 | * underlying zio's io_data buffer is overwritten by the uncompressed | |
6612 | * version. On decompression error (corrupt compressed stream), the | |
6613 | * zio->io_error value is set to signal an I/O error. | |
6614 | * | |
6615 | * Please note that the compressed data stream is not checksummed, so | |
6616 | * if the underlying device is experiencing data corruption, we may feed | |
6617 | * corrupt data to the decompressor, so the decompressor needs to be | |
6618 | * able to handle this situation (LZ4 does). | |
6619 | */ | |
6620 | static void | |
6621 | l2arc_decompress_zio(zio_t *zio, arc_buf_hdr_t *hdr, enum zio_compress c) | |
6622 | { | |
6623 | uint64_t csize; | |
6624 | void *cdata; | |
6625 | ||
6626 | ASSERT(L2ARC_IS_VALID_COMPRESS(c)); | |
6627 | ||
6628 | if (zio->io_error != 0) { | |
6629 | /* | |
6630 | * An io error has occured, just restore the original io | |
6631 | * size in preparation for a main pool read. | |
6632 | */ | |
6633 | zio->io_orig_size = zio->io_size = hdr->b_size; | |
6634 | return; | |
6635 | } | |
6636 | ||
6637 | if (c == ZIO_COMPRESS_EMPTY) { | |
6638 | /* | |
6639 | * An empty buffer results in a null zio, which means we | |
6640 | * need to fill its io_data after we're done restoring the | |
6641 | * buffer's contents. | |
6642 | */ | |
b9541d6b CW |
6643 | ASSERT(hdr->b_l1hdr.b_buf != NULL); |
6644 | bzero(hdr->b_l1hdr.b_buf->b_data, hdr->b_size); | |
6645 | zio->io_data = zio->io_orig_data = hdr->b_l1hdr.b_buf->b_data; | |
3a17a7a9 SK |
6646 | } else { |
6647 | ASSERT(zio->io_data != NULL); | |
6648 | /* | |
6649 | * We copy the compressed data from the start of the arc buffer | |
6650 | * (the zio_read will have pulled in only what we need, the | |
6651 | * rest is garbage which we will overwrite at decompression) | |
6652 | * and then decompress back to the ARC data buffer. This way we | |
6653 | * can minimize copying by simply decompressing back over the | |
6654 | * original compressed data (rather than decompressing to an | |
6655 | * aux buffer and then copying back the uncompressed buffer, | |
6656 | * which is likely to be much larger). | |
6657 | */ | |
6658 | csize = zio->io_size; | |
6659 | cdata = zio_data_buf_alloc(csize); | |
6660 | bcopy(zio->io_data, cdata, csize); | |
6661 | if (zio_decompress_data(c, cdata, zio->io_data, csize, | |
6662 | hdr->b_size) != 0) | |
2e528b49 | 6663 | zio->io_error = SET_ERROR(EIO); |
3a17a7a9 SK |
6664 | zio_data_buf_free(cdata, csize); |
6665 | } | |
6666 | ||
6667 | /* Restore the expected uncompressed IO size. */ | |
6668 | zio->io_orig_size = zio->io_size = hdr->b_size; | |
6669 | } | |
6670 | ||
6671 | /* | |
6672 | * Releases the temporary b_tmp_cdata buffer in an l2arc header structure. | |
6673 | * This buffer serves as a temporary holder of compressed data while | |
6674 | * the buffer entry is being written to an l2arc device. Once that is | |
6675 | * done, we can dispose of it. | |
6676 | */ | |
6677 | static void | |
2a432414 | 6678 | l2arc_release_cdata_buf(arc_buf_hdr_t *hdr) |
3a17a7a9 | 6679 | { |
ca0bf58d PS |
6680 | enum zio_compress comp = HDR_GET_COMPRESS(hdr); |
6681 | ||
b9541d6b | 6682 | ASSERT(HDR_HAS_L1HDR(hdr)); |
ca0bf58d PS |
6683 | ASSERT(comp == ZIO_COMPRESS_OFF || L2ARC_IS_VALID_COMPRESS(comp)); |
6684 | ||
6685 | if (comp == ZIO_COMPRESS_OFF) { | |
6686 | /* | |
6687 | * In this case, b_tmp_cdata points to the same buffer | |
6688 | * as the arc_buf_t's b_data field. We don't want to | |
6689 | * free it, since the arc_buf_t will handle that. | |
6690 | */ | |
6691 | hdr->b_l1hdr.b_tmp_cdata = NULL; | |
6692 | } else if (comp == ZIO_COMPRESS_EMPTY) { | |
6693 | /* | |
6694 | * In this case, b_tmp_cdata was compressed to an empty | |
6695 | * buffer, thus there's nothing to free and b_tmp_cdata | |
6696 | * should have been set to NULL in l2arc_write_buffers(). | |
6697 | */ | |
6698 | ASSERT3P(hdr->b_l1hdr.b_tmp_cdata, ==, NULL); | |
6699 | } else { | |
3a17a7a9 SK |
6700 | /* |
6701 | * If the data was compressed, then we've allocated a | |
6702 | * temporary buffer for it, so now we need to release it. | |
6703 | */ | |
b9541d6b CW |
6704 | ASSERT(hdr->b_l1hdr.b_tmp_cdata != NULL); |
6705 | zio_data_buf_free(hdr->b_l1hdr.b_tmp_cdata, | |
6706 | hdr->b_size); | |
ca0bf58d | 6707 | hdr->b_l1hdr.b_tmp_cdata = NULL; |
3a17a7a9 | 6708 | } |
ca0bf58d | 6709 | |
34dc7c2f BB |
6710 | } |
6711 | ||
6712 | /* | |
6713 | * This thread feeds the L2ARC at regular intervals. This is the beating | |
6714 | * heart of the L2ARC. | |
6715 | */ | |
6716 | static void | |
6717 | l2arc_feed_thread(void) | |
6718 | { | |
6719 | callb_cpr_t cpr; | |
6720 | l2arc_dev_t *dev; | |
6721 | spa_t *spa; | |
d164b209 | 6722 | uint64_t size, wrote; |
428870ff | 6723 | clock_t begin, next = ddi_get_lbolt(); |
3a17a7a9 | 6724 | boolean_t headroom_boost = B_FALSE; |
40d06e3c | 6725 | fstrans_cookie_t cookie; |
34dc7c2f BB |
6726 | |
6727 | CALLB_CPR_INIT(&cpr, &l2arc_feed_thr_lock, callb_generic_cpr, FTAG); | |
6728 | ||
6729 | mutex_enter(&l2arc_feed_thr_lock); | |
6730 | ||
40d06e3c | 6731 | cookie = spl_fstrans_mark(); |
34dc7c2f | 6732 | while (l2arc_thread_exit == 0) { |
34dc7c2f | 6733 | CALLB_CPR_SAFE_BEGIN(&cpr); |
b64ccd6c | 6734 | (void) cv_timedwait_sig(&l2arc_feed_thr_cv, |
5b63b3eb | 6735 | &l2arc_feed_thr_lock, next); |
34dc7c2f | 6736 | CALLB_CPR_SAFE_END(&cpr, &l2arc_feed_thr_lock); |
428870ff | 6737 | next = ddi_get_lbolt() + hz; |
34dc7c2f BB |
6738 | |
6739 | /* | |
b128c09f | 6740 | * Quick check for L2ARC devices. |
34dc7c2f BB |
6741 | */ |
6742 | mutex_enter(&l2arc_dev_mtx); | |
6743 | if (l2arc_ndev == 0) { | |
6744 | mutex_exit(&l2arc_dev_mtx); | |
6745 | continue; | |
6746 | } | |
b128c09f | 6747 | mutex_exit(&l2arc_dev_mtx); |
428870ff | 6748 | begin = ddi_get_lbolt(); |
34dc7c2f BB |
6749 | |
6750 | /* | |
b128c09f BB |
6751 | * This selects the next l2arc device to write to, and in |
6752 | * doing so the next spa to feed from: dev->l2ad_spa. This | |
6753 | * will return NULL if there are now no l2arc devices or if | |
6754 | * they are all faulted. | |
6755 | * | |
6756 | * If a device is returned, its spa's config lock is also | |
6757 | * held to prevent device removal. l2arc_dev_get_next() | |
6758 | * will grab and release l2arc_dev_mtx. | |
34dc7c2f | 6759 | */ |
b128c09f | 6760 | if ((dev = l2arc_dev_get_next()) == NULL) |
34dc7c2f | 6761 | continue; |
b128c09f BB |
6762 | |
6763 | spa = dev->l2ad_spa; | |
6764 | ASSERT(spa != NULL); | |
34dc7c2f | 6765 | |
572e2857 BB |
6766 | /* |
6767 | * If the pool is read-only then force the feed thread to | |
6768 | * sleep a little longer. | |
6769 | */ | |
6770 | if (!spa_writeable(spa)) { | |
6771 | next = ddi_get_lbolt() + 5 * l2arc_feed_secs * hz; | |
6772 | spa_config_exit(spa, SCL_L2ARC, dev); | |
6773 | continue; | |
6774 | } | |
6775 | ||
34dc7c2f | 6776 | /* |
b128c09f | 6777 | * Avoid contributing to memory pressure. |
34dc7c2f | 6778 | */ |
ca67b33a | 6779 | if (arc_reclaim_needed()) { |
b128c09f BB |
6780 | ARCSTAT_BUMP(arcstat_l2_abort_lowmem); |
6781 | spa_config_exit(spa, SCL_L2ARC, dev); | |
34dc7c2f BB |
6782 | continue; |
6783 | } | |
b128c09f | 6784 | |
34dc7c2f BB |
6785 | ARCSTAT_BUMP(arcstat_l2_feeds); |
6786 | ||
3a17a7a9 | 6787 | size = l2arc_write_size(); |
b128c09f | 6788 | |
34dc7c2f BB |
6789 | /* |
6790 | * Evict L2ARC buffers that will be overwritten. | |
6791 | */ | |
b128c09f | 6792 | l2arc_evict(dev, size, B_FALSE); |
34dc7c2f BB |
6793 | |
6794 | /* | |
6795 | * Write ARC buffers. | |
6796 | */ | |
3a17a7a9 | 6797 | wrote = l2arc_write_buffers(spa, dev, size, &headroom_boost); |
d164b209 BB |
6798 | |
6799 | /* | |
6800 | * Calculate interval between writes. | |
6801 | */ | |
6802 | next = l2arc_write_interval(begin, size, wrote); | |
b128c09f | 6803 | spa_config_exit(spa, SCL_L2ARC, dev); |
34dc7c2f | 6804 | } |
40d06e3c | 6805 | spl_fstrans_unmark(cookie); |
34dc7c2f BB |
6806 | |
6807 | l2arc_thread_exit = 0; | |
6808 | cv_broadcast(&l2arc_feed_thr_cv); | |
6809 | CALLB_CPR_EXIT(&cpr); /* drops l2arc_feed_thr_lock */ | |
6810 | thread_exit(); | |
6811 | } | |
6812 | ||
b128c09f BB |
6813 | boolean_t |
6814 | l2arc_vdev_present(vdev_t *vd) | |
6815 | { | |
6816 | l2arc_dev_t *dev; | |
6817 | ||
6818 | mutex_enter(&l2arc_dev_mtx); | |
6819 | for (dev = list_head(l2arc_dev_list); dev != NULL; | |
6820 | dev = list_next(l2arc_dev_list, dev)) { | |
6821 | if (dev->l2ad_vdev == vd) | |
6822 | break; | |
6823 | } | |
6824 | mutex_exit(&l2arc_dev_mtx); | |
6825 | ||
6826 | return (dev != NULL); | |
6827 | } | |
6828 | ||
34dc7c2f BB |
6829 | /* |
6830 | * Add a vdev for use by the L2ARC. By this point the spa has already | |
6831 | * validated the vdev and opened it. | |
6832 | */ | |
6833 | void | |
9babb374 | 6834 | l2arc_add_vdev(spa_t *spa, vdev_t *vd) |
34dc7c2f BB |
6835 | { |
6836 | l2arc_dev_t *adddev; | |
6837 | ||
b128c09f BB |
6838 | ASSERT(!l2arc_vdev_present(vd)); |
6839 | ||
34dc7c2f BB |
6840 | /* |
6841 | * Create a new l2arc device entry. | |
6842 | */ | |
6843 | adddev = kmem_zalloc(sizeof (l2arc_dev_t), KM_SLEEP); | |
6844 | adddev->l2ad_spa = spa; | |
6845 | adddev->l2ad_vdev = vd; | |
9babb374 BB |
6846 | adddev->l2ad_start = VDEV_LABEL_START_SIZE; |
6847 | adddev->l2ad_end = VDEV_LABEL_START_SIZE + vdev_get_min_asize(vd); | |
34dc7c2f | 6848 | adddev->l2ad_hand = adddev->l2ad_start; |
34dc7c2f | 6849 | adddev->l2ad_first = B_TRUE; |
d164b209 | 6850 | adddev->l2ad_writing = B_FALSE; |
98f72a53 | 6851 | list_link_init(&adddev->l2ad_node); |
34dc7c2f | 6852 | |
b9541d6b | 6853 | mutex_init(&adddev->l2ad_mtx, NULL, MUTEX_DEFAULT, NULL); |
34dc7c2f BB |
6854 | /* |
6855 | * This is a list of all ARC buffers that are still valid on the | |
6856 | * device. | |
6857 | */ | |
b9541d6b CW |
6858 | list_create(&adddev->l2ad_buflist, sizeof (arc_buf_hdr_t), |
6859 | offsetof(arc_buf_hdr_t, b_l2hdr.b_l2node)); | |
34dc7c2f | 6860 | |
428870ff | 6861 | vdev_space_update(vd, 0, 0, adddev->l2ad_end - adddev->l2ad_hand); |
d962d5da | 6862 | refcount_create(&adddev->l2ad_alloc); |
34dc7c2f BB |
6863 | |
6864 | /* | |
6865 | * Add device to global list | |
6866 | */ | |
6867 | mutex_enter(&l2arc_dev_mtx); | |
6868 | list_insert_head(l2arc_dev_list, adddev); | |
6869 | atomic_inc_64(&l2arc_ndev); | |
6870 | mutex_exit(&l2arc_dev_mtx); | |
6871 | } | |
6872 | ||
6873 | /* | |
6874 | * Remove a vdev from the L2ARC. | |
6875 | */ | |
6876 | void | |
6877 | l2arc_remove_vdev(vdev_t *vd) | |
6878 | { | |
6879 | l2arc_dev_t *dev, *nextdev, *remdev = NULL; | |
6880 | ||
34dc7c2f BB |
6881 | /* |
6882 | * Find the device by vdev | |
6883 | */ | |
6884 | mutex_enter(&l2arc_dev_mtx); | |
6885 | for (dev = list_head(l2arc_dev_list); dev; dev = nextdev) { | |
6886 | nextdev = list_next(l2arc_dev_list, dev); | |
6887 | if (vd == dev->l2ad_vdev) { | |
6888 | remdev = dev; | |
6889 | break; | |
6890 | } | |
6891 | } | |
6892 | ASSERT(remdev != NULL); | |
6893 | ||
6894 | /* | |
6895 | * Remove device from global list | |
6896 | */ | |
6897 | list_remove(l2arc_dev_list, remdev); | |
6898 | l2arc_dev_last = NULL; /* may have been invalidated */ | |
b128c09f BB |
6899 | atomic_dec_64(&l2arc_ndev); |
6900 | mutex_exit(&l2arc_dev_mtx); | |
34dc7c2f BB |
6901 | |
6902 | /* | |
6903 | * Clear all buflists and ARC references. L2ARC device flush. | |
6904 | */ | |
6905 | l2arc_evict(remdev, 0, B_TRUE); | |
b9541d6b CW |
6906 | list_destroy(&remdev->l2ad_buflist); |
6907 | mutex_destroy(&remdev->l2ad_mtx); | |
d962d5da | 6908 | refcount_destroy(&remdev->l2ad_alloc); |
34dc7c2f | 6909 | kmem_free(remdev, sizeof (l2arc_dev_t)); |
34dc7c2f BB |
6910 | } |
6911 | ||
6912 | void | |
b128c09f | 6913 | l2arc_init(void) |
34dc7c2f BB |
6914 | { |
6915 | l2arc_thread_exit = 0; | |
6916 | l2arc_ndev = 0; | |
6917 | l2arc_writes_sent = 0; | |
6918 | l2arc_writes_done = 0; | |
6919 | ||
6920 | mutex_init(&l2arc_feed_thr_lock, NULL, MUTEX_DEFAULT, NULL); | |
6921 | cv_init(&l2arc_feed_thr_cv, NULL, CV_DEFAULT, NULL); | |
6922 | mutex_init(&l2arc_dev_mtx, NULL, MUTEX_DEFAULT, NULL); | |
34dc7c2f BB |
6923 | mutex_init(&l2arc_free_on_write_mtx, NULL, MUTEX_DEFAULT, NULL); |
6924 | ||
6925 | l2arc_dev_list = &L2ARC_dev_list; | |
6926 | l2arc_free_on_write = &L2ARC_free_on_write; | |
6927 | list_create(l2arc_dev_list, sizeof (l2arc_dev_t), | |
6928 | offsetof(l2arc_dev_t, l2ad_node)); | |
6929 | list_create(l2arc_free_on_write, sizeof (l2arc_data_free_t), | |
6930 | offsetof(l2arc_data_free_t, l2df_list_node)); | |
34dc7c2f BB |
6931 | } |
6932 | ||
6933 | void | |
b128c09f | 6934 | l2arc_fini(void) |
34dc7c2f | 6935 | { |
b128c09f BB |
6936 | /* |
6937 | * This is called from dmu_fini(), which is called from spa_fini(); | |
6938 | * Because of this, we can assume that all l2arc devices have | |
6939 | * already been removed when the pools themselves were removed. | |
6940 | */ | |
6941 | ||
6942 | l2arc_do_free_on_write(); | |
34dc7c2f BB |
6943 | |
6944 | mutex_destroy(&l2arc_feed_thr_lock); | |
6945 | cv_destroy(&l2arc_feed_thr_cv); | |
6946 | mutex_destroy(&l2arc_dev_mtx); | |
34dc7c2f BB |
6947 | mutex_destroy(&l2arc_free_on_write_mtx); |
6948 | ||
6949 | list_destroy(l2arc_dev_list); | |
6950 | list_destroy(l2arc_free_on_write); | |
6951 | } | |
b128c09f BB |
6952 | |
6953 | void | |
6954 | l2arc_start(void) | |
6955 | { | |
fb5f0bc8 | 6956 | if (!(spa_mode_global & FWRITE)) |
b128c09f BB |
6957 | return; |
6958 | ||
6959 | (void) thread_create(NULL, 0, l2arc_feed_thread, NULL, 0, &p0, | |
6960 | TS_RUN, minclsyspri); | |
6961 | } | |
6962 | ||
6963 | void | |
6964 | l2arc_stop(void) | |
6965 | { | |
fb5f0bc8 | 6966 | if (!(spa_mode_global & FWRITE)) |
b128c09f BB |
6967 | return; |
6968 | ||
6969 | mutex_enter(&l2arc_feed_thr_lock); | |
6970 | cv_signal(&l2arc_feed_thr_cv); /* kick thread out of startup */ | |
6971 | l2arc_thread_exit = 1; | |
6972 | while (l2arc_thread_exit != 0) | |
6973 | cv_wait(&l2arc_feed_thr_cv, &l2arc_feed_thr_lock); | |
6974 | mutex_exit(&l2arc_feed_thr_lock); | |
6975 | } | |
c28b2279 BB |
6976 | |
6977 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
0f699108 AZ |
6978 | EXPORT_SYMBOL(arc_buf_size); |
6979 | EXPORT_SYMBOL(arc_write); | |
c28b2279 BB |
6980 | EXPORT_SYMBOL(arc_read); |
6981 | EXPORT_SYMBOL(arc_buf_remove_ref); | |
e0b0ca98 | 6982 | EXPORT_SYMBOL(arc_buf_info); |
c28b2279 | 6983 | EXPORT_SYMBOL(arc_getbuf_func); |
ab26409d BB |
6984 | EXPORT_SYMBOL(arc_add_prune_callback); |
6985 | EXPORT_SYMBOL(arc_remove_prune_callback); | |
c28b2279 | 6986 | |
bce45ec9 | 6987 | module_param(zfs_arc_min, ulong, 0644); |
c409e464 | 6988 | MODULE_PARM_DESC(zfs_arc_min, "Min arc size"); |
c28b2279 | 6989 | |
bce45ec9 | 6990 | module_param(zfs_arc_max, ulong, 0644); |
c409e464 | 6991 | MODULE_PARM_DESC(zfs_arc_max, "Max arc size"); |
c28b2279 | 6992 | |
bce45ec9 | 6993 | module_param(zfs_arc_meta_limit, ulong, 0644); |
c28b2279 | 6994 | MODULE_PARM_DESC(zfs_arc_meta_limit, "Meta limit for arc size"); |
6a8f9b6b | 6995 | |
ca0bf58d PS |
6996 | module_param(zfs_arc_meta_min, ulong, 0644); |
6997 | MODULE_PARM_DESC(zfs_arc_meta_min, "Min arc metadata"); | |
6998 | ||
bce45ec9 | 6999 | module_param(zfs_arc_meta_prune, int, 0644); |
2cbb06b5 | 7000 | MODULE_PARM_DESC(zfs_arc_meta_prune, "Meta objects to scan for prune"); |
c409e464 | 7001 | |
ca67b33a | 7002 | module_param(zfs_arc_meta_adjust_restarts, int, 0644); |
bc888666 BB |
7003 | MODULE_PARM_DESC(zfs_arc_meta_adjust_restarts, |
7004 | "Limit number of restarts in arc_adjust_meta"); | |
7005 | ||
f6046738 BB |
7006 | module_param(zfs_arc_meta_strategy, int, 0644); |
7007 | MODULE_PARM_DESC(zfs_arc_meta_strategy, "Meta reclaim strategy"); | |
7008 | ||
bce45ec9 | 7009 | module_param(zfs_arc_grow_retry, int, 0644); |
c409e464 BB |
7010 | MODULE_PARM_DESC(zfs_arc_grow_retry, "Seconds before growing arc size"); |
7011 | ||
89c8cac4 PS |
7012 | module_param(zfs_arc_p_aggressive_disable, int, 0644); |
7013 | MODULE_PARM_DESC(zfs_arc_p_aggressive_disable, "disable aggressive arc_p grow"); | |
7014 | ||
62422785 PS |
7015 | module_param(zfs_arc_p_dampener_disable, int, 0644); |
7016 | MODULE_PARM_DESC(zfs_arc_p_dampener_disable, "disable arc_p adapt dampener"); | |
7017 | ||
bce45ec9 | 7018 | module_param(zfs_arc_shrink_shift, int, 0644); |
c409e464 BB |
7019 | MODULE_PARM_DESC(zfs_arc_shrink_shift, "log2(fraction of arc to reclaim)"); |
7020 | ||
728d6ae9 BB |
7021 | module_param(zfs_arc_p_min_shift, int, 0644); |
7022 | MODULE_PARM_DESC(zfs_arc_p_min_shift, "arc_c shift to calc min/max arc_p"); | |
7023 | ||
1f7c30df BB |
7024 | module_param(zfs_disable_dup_eviction, int, 0644); |
7025 | MODULE_PARM_DESC(zfs_disable_dup_eviction, "disable duplicate buffer eviction"); | |
7026 | ||
49ddb315 MA |
7027 | module_param(zfs_arc_average_blocksize, int, 0444); |
7028 | MODULE_PARM_DESC(zfs_arc_average_blocksize, "Target average block size"); | |
7029 | ||
0c5493d4 BB |
7030 | module_param(zfs_arc_memory_throttle_disable, int, 0644); |
7031 | MODULE_PARM_DESC(zfs_arc_memory_throttle_disable, "disable memory throttle"); | |
7032 | ||
bce45ec9 BB |
7033 | module_param(zfs_arc_min_prefetch_lifespan, int, 0644); |
7034 | MODULE_PARM_DESC(zfs_arc_min_prefetch_lifespan, "Min life of prefetch block"); | |
7035 | ||
ca0bf58d PS |
7036 | module_param(zfs_arc_num_sublists_per_state, int, 0644); |
7037 | MODULE_PARM_DESC(zfs_arc_num_sublists_per_state, | |
7038 | "Number of sublists used in each of the ARC state lists"); | |
7039 | ||
bce45ec9 | 7040 | module_param(l2arc_write_max, ulong, 0644); |
abd8610c BB |
7041 | MODULE_PARM_DESC(l2arc_write_max, "Max write bytes per interval"); |
7042 | ||
bce45ec9 | 7043 | module_param(l2arc_write_boost, ulong, 0644); |
abd8610c BB |
7044 | MODULE_PARM_DESC(l2arc_write_boost, "Extra write bytes during device warmup"); |
7045 | ||
bce45ec9 | 7046 | module_param(l2arc_headroom, ulong, 0644); |
abd8610c BB |
7047 | MODULE_PARM_DESC(l2arc_headroom, "Number of max device writes to precache"); |
7048 | ||
3a17a7a9 SK |
7049 | module_param(l2arc_headroom_boost, ulong, 0644); |
7050 | MODULE_PARM_DESC(l2arc_headroom_boost, "Compressed l2arc_headroom multiplier"); | |
7051 | ||
bce45ec9 | 7052 | module_param(l2arc_feed_secs, ulong, 0644); |
abd8610c BB |
7053 | MODULE_PARM_DESC(l2arc_feed_secs, "Seconds between L2ARC writing"); |
7054 | ||
bce45ec9 | 7055 | module_param(l2arc_feed_min_ms, ulong, 0644); |
abd8610c BB |
7056 | MODULE_PARM_DESC(l2arc_feed_min_ms, "Min feed interval in milliseconds"); |
7057 | ||
bce45ec9 | 7058 | module_param(l2arc_noprefetch, int, 0644); |
abd8610c BB |
7059 | MODULE_PARM_DESC(l2arc_noprefetch, "Skip caching prefetched buffers"); |
7060 | ||
3a17a7a9 SK |
7061 | module_param(l2arc_nocompress, int, 0644); |
7062 | MODULE_PARM_DESC(l2arc_nocompress, "Skip compressing L2ARC buffers"); | |
7063 | ||
bce45ec9 | 7064 | module_param(l2arc_feed_again, int, 0644); |
abd8610c BB |
7065 | MODULE_PARM_DESC(l2arc_feed_again, "Turbo L2ARC warmup"); |
7066 | ||
bce45ec9 | 7067 | module_param(l2arc_norw, int, 0644); |
abd8610c BB |
7068 | MODULE_PARM_DESC(l2arc_norw, "No reads during writes"); |
7069 | ||
c28b2279 | 7070 | #endif |