]>
Commit | Line | Data |
---|---|---|
70e083d2 TG |
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 | /* | |
22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. | |
23 | * Copyright (c) 2012, Joyent, Inc. All rights reserved. | |
24 | * Copyright (c) 2011, 2015 by Delphix. All rights reserved. | |
25 | * Copyright (c) 2014 by Saso Kiselkov. All rights reserved. | |
26 | * Copyright 2014 Nexenta Systems, Inc. All rights reserved. | |
27 | */ | |
28 | ||
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 | |
62 | * elements of the cache are therefore exactly the same size. So | |
63 | * when adjusting the cache size following a cache miss, its simply | |
64 | * a matter of choosing a single page to evict. In our model, we | |
65 | * have variable sized cache blocks (rangeing from 512 bytes to | |
66 | * 128K bytes). We therefore choose a set of blocks to evict to make | |
67 | * space for a cache miss that approximates as closely as possible | |
68 | * the space used by the new block. | |
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 | |
81 | * adjusting the cache use method 2. We therefore provide two | |
82 | * types of locks: 1) the hash table lock array, and 2) the | |
83 | * arc list locks. | |
84 | * | |
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). | |
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 | |
108 | * protected from simultaneous callbacks from arc_clear_callback() | |
109 | * and arc_do_user_evicts(). | |
110 | * | |
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 | * | |
119 | * Note that the majority of the performance stats are manipulated | |
120 | * with atomic operations. | |
121 | * | |
122 | * The L2ARC uses the l2ad_mtx on each vdev for the following: | |
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> | |
133 | #include <sys/zio_compress.h> | |
134 | #include <sys/zfs_context.h> | |
135 | #include <sys/arc.h> | |
136 | #include <sys/refcount.h> | |
137 | #include <sys/vdev.h> | |
138 | #include <sys/vdev_impl.h> | |
139 | #include <sys/dsl_pool.h> | |
140 | #include <sys/multilist.h> | |
141 | #ifdef _KERNEL | |
142 | #include <sys/vmsystm.h> | |
143 | #include <vm/anon.h> | |
144 | #include <sys/fs/swapnode.h> | |
145 | #include <sys/zpl.h> | |
146 | #include <linux/mm_compat.h> | |
147 | #endif | |
148 | #include <sys/callb.h> | |
149 | #include <sys/kstat.h> | |
150 | #include <sys/dmu_tx.h> | |
151 | #include <zfs_fletcher.h> | |
152 | #include <sys/arc_impl.h> | |
153 | #include <sys/trace_arc.h> | |
154 | ||
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 | ||
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; | |
168 | ||
169 | /* | |
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(). | |
182 | */ | |
183 | int zfs_arc_num_sublists_per_state = 0; | |
184 | ||
185 | /* number of seconds before growing cache again */ | |
186 | static int arc_grow_retry = 5; | |
187 | ||
188 | /* shift of arc_c for calculating overflow limit in arc_get_data_buf */ | |
189 | int zfs_arc_overflow_shift = 8; | |
190 | ||
191 | /* shift of arc_c for calculating both min and max arc_p */ | |
192 | static int arc_p_min_shift = 4; | |
193 | ||
194 | /* log2(fraction of arc to reclaim) */ | |
195 | static int arc_shrink_shift = 7; | |
196 | ||
197 | /* | |
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. | |
205 | */ | |
206 | int arc_no_grow_shift = 5; | |
207 | ||
208 | ||
209 | /* | |
210 | * minimum lifespan of a prefetch block in clock ticks | |
211 | * (initialized in arc_init()) | |
212 | */ | |
213 | static int arc_min_prefetch_lifespan; | |
214 | ||
215 | /* | |
216 | * If this percent of memory is free, don't throttle. | |
217 | */ | |
218 | int arc_lotsfree_percent = 10; | |
219 | ||
220 | static int arc_dead; | |
221 | ||
222 | /* | |
223 | * The arc has filled available memory and has now warmed up. | |
224 | */ | |
225 | static boolean_t arc_warm; | |
226 | ||
227 | /* | |
228 | * These tunables are for performance analysis. | |
229 | */ | |
230 | unsigned long zfs_arc_max = 0; | |
231 | unsigned long zfs_arc_min = 0; | |
232 | unsigned long zfs_arc_meta_limit = 0; | |
233 | unsigned long zfs_arc_meta_min = 0; | |
234 | int zfs_arc_grow_retry = 0; | |
235 | int zfs_arc_shrink_shift = 0; | |
236 | int zfs_arc_p_min_shift = 0; | |
237 | int zfs_disable_dup_eviction = 0; | |
238 | int zfs_arc_average_blocksize = 8 * 1024; /* 8KB */ | |
239 | ||
240 | /* | |
241 | * These tunables are Linux specific | |
242 | */ | |
243 | unsigned long zfs_arc_sys_free = 0; | |
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; | |
250 | int zfs_arc_lotsfree_percent = 10; | |
251 | ||
252 | /* The 6 states: */ | |
253 | static arc_state_t ARC_anon; | |
254 | static arc_state_t ARC_mru; | |
255 | static arc_state_t ARC_mru_ghost; | |
256 | static arc_state_t ARC_mfu; | |
257 | static arc_state_t ARC_mfu_ghost; | |
258 | static arc_state_t ARC_l2c_only; | |
259 | ||
260 | typedef struct arc_stats { | |
261 | kstat_named_t arcstat_hits; | |
262 | kstat_named_t arcstat_misses; | |
263 | kstat_named_t arcstat_demand_data_hits; | |
264 | kstat_named_t arcstat_demand_data_misses; | |
265 | kstat_named_t arcstat_demand_metadata_hits; | |
266 | kstat_named_t arcstat_demand_metadata_misses; | |
267 | kstat_named_t arcstat_prefetch_data_hits; | |
268 | kstat_named_t arcstat_prefetch_data_misses; | |
269 | kstat_named_t arcstat_prefetch_metadata_hits; | |
270 | kstat_named_t arcstat_prefetch_metadata_misses; | |
271 | kstat_named_t arcstat_mru_hits; | |
272 | kstat_named_t arcstat_mru_ghost_hits; | |
273 | kstat_named_t arcstat_mfu_hits; | |
274 | kstat_named_t arcstat_mfu_ghost_hits; | |
275 | kstat_named_t arcstat_deleted; | |
276 | /* | |
277 | * Number of buffers that could not be evicted because the hash lock | |
278 | * was held by another thread. The lock may not necessarily be held | |
279 | * by something using the same buffer, since hash locks are shared | |
280 | * by multiple buffers. | |
281 | */ | |
282 | kstat_named_t arcstat_mutex_miss; | |
283 | /* | |
284 | * Number of buffers skipped because they have I/O in progress, are | |
285 | * indrect prefetch buffers that have not lived long enough, or are | |
286 | * not from the spa we're trying to evict from. | |
287 | */ | |
288 | kstat_named_t arcstat_evict_skip; | |
289 | /* | |
290 | * Number of times arc_evict_state() was unable to evict enough | |
291 | * buffers to reach its target amount. | |
292 | */ | |
293 | kstat_named_t arcstat_evict_not_enough; | |
294 | kstat_named_t arcstat_evict_l2_cached; | |
295 | kstat_named_t arcstat_evict_l2_eligible; | |
296 | kstat_named_t arcstat_evict_l2_ineligible; | |
297 | kstat_named_t arcstat_evict_l2_skip; | |
298 | kstat_named_t arcstat_hash_elements; | |
299 | kstat_named_t arcstat_hash_elements_max; | |
300 | kstat_named_t arcstat_hash_collisions; | |
301 | kstat_named_t arcstat_hash_chains; | |
302 | kstat_named_t arcstat_hash_chain_max; | |
303 | kstat_named_t arcstat_p; | |
304 | kstat_named_t arcstat_c; | |
305 | kstat_named_t arcstat_c_min; | |
306 | kstat_named_t arcstat_c_max; | |
307 | kstat_named_t arcstat_size; | |
308 | /* | |
309 | * Number of bytes consumed by internal ARC structures necessary | |
310 | * for tracking purposes; these structures are not actually | |
311 | * backed by ARC buffers. This includes arc_buf_hdr_t structures | |
312 | * (allocated via arc_buf_hdr_t_full and arc_buf_hdr_t_l2only | |
313 | * caches), and arc_buf_t structures (allocated via arc_buf_t | |
314 | * cache). | |
315 | */ | |
316 | kstat_named_t arcstat_hdr_size; | |
317 | /* | |
318 | * Number of bytes consumed by ARC buffers of type equal to | |
319 | * ARC_BUFC_DATA. This is generally consumed by buffers backing | |
320 | * on disk user data (e.g. plain file contents). | |
321 | */ | |
322 | kstat_named_t arcstat_data_size; | |
323 | /* | |
324 | * Number of bytes consumed by ARC buffers of type equal to | |
325 | * ARC_BUFC_METADATA. This is generally consumed by buffers | |
326 | * backing on disk data that is used for internal ZFS | |
327 | * structures (e.g. ZAP, dnode, indirect blocks, etc). | |
328 | */ | |
329 | kstat_named_t arcstat_metadata_size; | |
330 | /* | |
331 | * Number of bytes consumed by various buffers and structures | |
332 | * not actually backed with ARC buffers. This includes bonus | |
333 | * buffers (allocated directly via zio_buf_* functions), | |
334 | * dmu_buf_impl_t structures (allocated via dmu_buf_impl_t | |
335 | * cache), and dnode_t structures (allocated via dnode_t cache). | |
336 | */ | |
337 | kstat_named_t arcstat_other_size; | |
338 | /* | |
339 | * Total number of bytes consumed by ARC buffers residing in the | |
340 | * arc_anon state. This includes *all* buffers in the arc_anon | |
341 | * state; e.g. data, metadata, evictable, and unevictable buffers | |
342 | * are all included in this value. | |
343 | */ | |
344 | kstat_named_t arcstat_anon_size; | |
345 | /* | |
346 | * Number of bytes consumed by ARC buffers that meet the | |
347 | * following criteria: backing buffers of type ARC_BUFC_DATA, | |
348 | * residing in the arc_anon state, and are eligible for eviction | |
349 | * (e.g. have no outstanding holds on the buffer). | |
350 | */ | |
351 | kstat_named_t arcstat_anon_evictable_data; | |
352 | /* | |
353 | * Number of bytes consumed by ARC buffers that meet the | |
354 | * following criteria: backing buffers of type ARC_BUFC_METADATA, | |
355 | * residing in the arc_anon state, and are eligible for eviction | |
356 | * (e.g. have no outstanding holds on the buffer). | |
357 | */ | |
358 | kstat_named_t arcstat_anon_evictable_metadata; | |
359 | /* | |
360 | * Total number of bytes consumed by ARC buffers residing in the | |
361 | * arc_mru state. This includes *all* buffers in the arc_mru | |
362 | * state; e.g. data, metadata, evictable, and unevictable buffers | |
363 | * are all included in this value. | |
364 | */ | |
365 | kstat_named_t arcstat_mru_size; | |
366 | /* | |
367 | * Number of bytes consumed by ARC buffers that meet the | |
368 | * following criteria: backing buffers of type ARC_BUFC_DATA, | |
369 | * residing in the arc_mru state, and are eligible for eviction | |
370 | * (e.g. have no outstanding holds on the buffer). | |
371 | */ | |
372 | kstat_named_t arcstat_mru_evictable_data; | |
373 | /* | |
374 | * Number of bytes consumed by ARC buffers that meet the | |
375 | * following criteria: backing buffers of type ARC_BUFC_METADATA, | |
376 | * residing in the arc_mru state, and are eligible for eviction | |
377 | * (e.g. have no outstanding holds on the buffer). | |
378 | */ | |
379 | kstat_named_t arcstat_mru_evictable_metadata; | |
380 | /* | |
381 | * Total number of bytes that *would have been* consumed by ARC | |
382 | * buffers in the arc_mru_ghost state. The key thing to note | |
383 | * here, is the fact that this size doesn't actually indicate | |
384 | * RAM consumption. The ghost lists only consist of headers and | |
385 | * don't actually have ARC buffers linked off of these headers. | |
386 | * Thus, *if* the headers had associated ARC buffers, these | |
387 | * buffers *would have* consumed this number of bytes. | |
388 | */ | |
389 | kstat_named_t arcstat_mru_ghost_size; | |
390 | /* | |
391 | * Number of bytes that *would have been* consumed by ARC | |
392 | * buffers that are eligible for eviction, of type | |
393 | * ARC_BUFC_DATA, and linked off the arc_mru_ghost state. | |
394 | */ | |
395 | kstat_named_t arcstat_mru_ghost_evictable_data; | |
396 | /* | |
397 | * Number of bytes that *would have been* consumed by ARC | |
398 | * buffers that are eligible for eviction, of type | |
399 | * ARC_BUFC_METADATA, and linked off the arc_mru_ghost state. | |
400 | */ | |
401 | kstat_named_t arcstat_mru_ghost_evictable_metadata; | |
402 | /* | |
403 | * Total number of bytes consumed by ARC buffers residing in the | |
404 | * arc_mfu state. This includes *all* buffers in the arc_mfu | |
405 | * state; e.g. data, metadata, evictable, and unevictable buffers | |
406 | * are all included in this value. | |
407 | */ | |
408 | kstat_named_t arcstat_mfu_size; | |
409 | /* | |
410 | * Number of bytes consumed by ARC buffers that are eligible for | |
411 | * eviction, of type ARC_BUFC_DATA, and reside in the arc_mfu | |
412 | * state. | |
413 | */ | |
414 | kstat_named_t arcstat_mfu_evictable_data; | |
415 | /* | |
416 | * Number of bytes consumed by ARC buffers that are eligible for | |
417 | * eviction, of type ARC_BUFC_METADATA, and reside in the | |
418 | * arc_mfu state. | |
419 | */ | |
420 | kstat_named_t arcstat_mfu_evictable_metadata; | |
421 | /* | |
422 | * Total number of bytes that *would have been* consumed by ARC | |
423 | * buffers in the arc_mfu_ghost state. See the comment above | |
424 | * arcstat_mru_ghost_size for more details. | |
425 | */ | |
426 | kstat_named_t arcstat_mfu_ghost_size; | |
427 | /* | |
428 | * Number of bytes that *would have been* consumed by ARC | |
429 | * buffers that are eligible for eviction, of type | |
430 | * ARC_BUFC_DATA, and linked off the arc_mfu_ghost state. | |
431 | */ | |
432 | kstat_named_t arcstat_mfu_ghost_evictable_data; | |
433 | /* | |
434 | * Number of bytes that *would have been* consumed by ARC | |
435 | * buffers that are eligible for eviction, of type | |
436 | * ARC_BUFC_METADATA, and linked off the arc_mru_ghost state. | |
437 | */ | |
438 | kstat_named_t arcstat_mfu_ghost_evictable_metadata; | |
439 | kstat_named_t arcstat_l2_hits; | |
440 | kstat_named_t arcstat_l2_misses; | |
441 | kstat_named_t arcstat_l2_feeds; | |
442 | kstat_named_t arcstat_l2_rw_clash; | |
443 | kstat_named_t arcstat_l2_read_bytes; | |
444 | kstat_named_t arcstat_l2_write_bytes; | |
445 | kstat_named_t arcstat_l2_writes_sent; | |
446 | kstat_named_t arcstat_l2_writes_done; | |
447 | kstat_named_t arcstat_l2_writes_error; | |
448 | kstat_named_t arcstat_l2_writes_lock_retry; | |
449 | kstat_named_t arcstat_l2_evict_lock_retry; | |
450 | kstat_named_t arcstat_l2_evict_reading; | |
451 | kstat_named_t arcstat_l2_evict_l1cached; | |
452 | kstat_named_t arcstat_l2_free_on_write; | |
453 | kstat_named_t arcstat_l2_cdata_free_on_write; | |
454 | kstat_named_t arcstat_l2_abort_lowmem; | |
455 | kstat_named_t arcstat_l2_cksum_bad; | |
456 | kstat_named_t arcstat_l2_io_error; | |
457 | kstat_named_t arcstat_l2_size; | |
458 | kstat_named_t arcstat_l2_asize; | |
459 | kstat_named_t arcstat_l2_hdr_size; | |
460 | kstat_named_t arcstat_l2_compress_successes; | |
461 | kstat_named_t arcstat_l2_compress_zeros; | |
462 | kstat_named_t arcstat_l2_compress_failures; | |
463 | kstat_named_t arcstat_memory_throttle_count; | |
464 | kstat_named_t arcstat_duplicate_buffers; | |
465 | kstat_named_t arcstat_duplicate_buffers_size; | |
466 | kstat_named_t arcstat_duplicate_reads; | |
467 | kstat_named_t arcstat_memory_direct_count; | |
468 | kstat_named_t arcstat_memory_indirect_count; | |
469 | kstat_named_t arcstat_no_grow; | |
470 | kstat_named_t arcstat_tempreserve; | |
471 | kstat_named_t arcstat_loaned_bytes; | |
472 | kstat_named_t arcstat_prune; | |
473 | kstat_named_t arcstat_meta_used; | |
474 | kstat_named_t arcstat_meta_limit; | |
475 | kstat_named_t arcstat_meta_max; | |
476 | kstat_named_t arcstat_meta_min; | |
477 | kstat_named_t arcstat_need_free; | |
478 | kstat_named_t arcstat_sys_free; | |
479 | } arc_stats_t; | |
480 | ||
481 | static arc_stats_t arc_stats = { | |
482 | { "hits", KSTAT_DATA_UINT64 }, | |
483 | { "misses", KSTAT_DATA_UINT64 }, | |
484 | { "demand_data_hits", KSTAT_DATA_UINT64 }, | |
485 | { "demand_data_misses", KSTAT_DATA_UINT64 }, | |
486 | { "demand_metadata_hits", KSTAT_DATA_UINT64 }, | |
487 | { "demand_metadata_misses", KSTAT_DATA_UINT64 }, | |
488 | { "prefetch_data_hits", KSTAT_DATA_UINT64 }, | |
489 | { "prefetch_data_misses", KSTAT_DATA_UINT64 }, | |
490 | { "prefetch_metadata_hits", KSTAT_DATA_UINT64 }, | |
491 | { "prefetch_metadata_misses", KSTAT_DATA_UINT64 }, | |
492 | { "mru_hits", KSTAT_DATA_UINT64 }, | |
493 | { "mru_ghost_hits", KSTAT_DATA_UINT64 }, | |
494 | { "mfu_hits", KSTAT_DATA_UINT64 }, | |
495 | { "mfu_ghost_hits", KSTAT_DATA_UINT64 }, | |
496 | { "deleted", KSTAT_DATA_UINT64 }, | |
497 | { "mutex_miss", KSTAT_DATA_UINT64 }, | |
498 | { "evict_skip", KSTAT_DATA_UINT64 }, | |
499 | { "evict_not_enough", KSTAT_DATA_UINT64 }, | |
500 | { "evict_l2_cached", KSTAT_DATA_UINT64 }, | |
501 | { "evict_l2_eligible", KSTAT_DATA_UINT64 }, | |
502 | { "evict_l2_ineligible", KSTAT_DATA_UINT64 }, | |
503 | { "evict_l2_skip", KSTAT_DATA_UINT64 }, | |
504 | { "hash_elements", KSTAT_DATA_UINT64 }, | |
505 | { "hash_elements_max", KSTAT_DATA_UINT64 }, | |
506 | { "hash_collisions", KSTAT_DATA_UINT64 }, | |
507 | { "hash_chains", KSTAT_DATA_UINT64 }, | |
508 | { "hash_chain_max", KSTAT_DATA_UINT64 }, | |
509 | { "p", KSTAT_DATA_UINT64 }, | |
510 | { "c", KSTAT_DATA_UINT64 }, | |
511 | { "c_min", KSTAT_DATA_UINT64 }, | |
512 | { "c_max", KSTAT_DATA_UINT64 }, | |
513 | { "size", KSTAT_DATA_UINT64 }, | |
514 | { "hdr_size", KSTAT_DATA_UINT64 }, | |
515 | { "data_size", KSTAT_DATA_UINT64 }, | |
516 | { "metadata_size", KSTAT_DATA_UINT64 }, | |
517 | { "other_size", KSTAT_DATA_UINT64 }, | |
518 | { "anon_size", KSTAT_DATA_UINT64 }, | |
519 | { "anon_evictable_data", KSTAT_DATA_UINT64 }, | |
520 | { "anon_evictable_metadata", KSTAT_DATA_UINT64 }, | |
521 | { "mru_size", KSTAT_DATA_UINT64 }, | |
522 | { "mru_evictable_data", KSTAT_DATA_UINT64 }, | |
523 | { "mru_evictable_metadata", KSTAT_DATA_UINT64 }, | |
524 | { "mru_ghost_size", KSTAT_DATA_UINT64 }, | |
525 | { "mru_ghost_evictable_data", KSTAT_DATA_UINT64 }, | |
526 | { "mru_ghost_evictable_metadata", KSTAT_DATA_UINT64 }, | |
527 | { "mfu_size", KSTAT_DATA_UINT64 }, | |
528 | { "mfu_evictable_data", KSTAT_DATA_UINT64 }, | |
529 | { "mfu_evictable_metadata", KSTAT_DATA_UINT64 }, | |
530 | { "mfu_ghost_size", KSTAT_DATA_UINT64 }, | |
531 | { "mfu_ghost_evictable_data", KSTAT_DATA_UINT64 }, | |
532 | { "mfu_ghost_evictable_metadata", KSTAT_DATA_UINT64 }, | |
533 | { "l2_hits", KSTAT_DATA_UINT64 }, | |
534 | { "l2_misses", KSTAT_DATA_UINT64 }, | |
535 | { "l2_feeds", KSTAT_DATA_UINT64 }, | |
536 | { "l2_rw_clash", KSTAT_DATA_UINT64 }, | |
537 | { "l2_read_bytes", KSTAT_DATA_UINT64 }, | |
538 | { "l2_write_bytes", KSTAT_DATA_UINT64 }, | |
539 | { "l2_writes_sent", KSTAT_DATA_UINT64 }, | |
540 | { "l2_writes_done", KSTAT_DATA_UINT64 }, | |
541 | { "l2_writes_error", KSTAT_DATA_UINT64 }, | |
542 | { "l2_writes_lock_retry", KSTAT_DATA_UINT64 }, | |
543 | { "l2_evict_lock_retry", KSTAT_DATA_UINT64 }, | |
544 | { "l2_evict_reading", KSTAT_DATA_UINT64 }, | |
545 | { "l2_evict_l1cached", KSTAT_DATA_UINT64 }, | |
546 | { "l2_free_on_write", KSTAT_DATA_UINT64 }, | |
547 | { "l2_cdata_free_on_write", KSTAT_DATA_UINT64 }, | |
548 | { "l2_abort_lowmem", KSTAT_DATA_UINT64 }, | |
549 | { "l2_cksum_bad", KSTAT_DATA_UINT64 }, | |
550 | { "l2_io_error", KSTAT_DATA_UINT64 }, | |
551 | { "l2_size", KSTAT_DATA_UINT64 }, | |
552 | { "l2_asize", KSTAT_DATA_UINT64 }, | |
553 | { "l2_hdr_size", KSTAT_DATA_UINT64 }, | |
554 | { "l2_compress_successes", KSTAT_DATA_UINT64 }, | |
555 | { "l2_compress_zeros", KSTAT_DATA_UINT64 }, | |
556 | { "l2_compress_failures", KSTAT_DATA_UINT64 }, | |
557 | { "memory_throttle_count", KSTAT_DATA_UINT64 }, | |
558 | { "duplicate_buffers", KSTAT_DATA_UINT64 }, | |
559 | { "duplicate_buffers_size", KSTAT_DATA_UINT64 }, | |
560 | { "duplicate_reads", KSTAT_DATA_UINT64 }, | |
561 | { "memory_direct_count", KSTAT_DATA_UINT64 }, | |
562 | { "memory_indirect_count", KSTAT_DATA_UINT64 }, | |
563 | { "arc_no_grow", KSTAT_DATA_UINT64 }, | |
564 | { "arc_tempreserve", KSTAT_DATA_UINT64 }, | |
565 | { "arc_loaned_bytes", KSTAT_DATA_UINT64 }, | |
566 | { "arc_prune", KSTAT_DATA_UINT64 }, | |
567 | { "arc_meta_used", KSTAT_DATA_UINT64 }, | |
568 | { "arc_meta_limit", KSTAT_DATA_UINT64 }, | |
569 | { "arc_meta_max", KSTAT_DATA_UINT64 }, | |
570 | { "arc_meta_min", KSTAT_DATA_UINT64 }, | |
571 | { "arc_need_free", KSTAT_DATA_UINT64 }, | |
572 | { "arc_sys_free", KSTAT_DATA_UINT64 } | |
573 | }; | |
574 | ||
575 | #define ARCSTAT(stat) (arc_stats.stat.value.ui64) | |
576 | ||
577 | #define ARCSTAT_INCR(stat, val) \ | |
578 | atomic_add_64(&arc_stats.stat.value.ui64, (val)) | |
579 | ||
580 | #define ARCSTAT_BUMP(stat) ARCSTAT_INCR(stat, 1) | |
581 | #define ARCSTAT_BUMPDOWN(stat) ARCSTAT_INCR(stat, -1) | |
582 | ||
583 | #define ARCSTAT_MAX(stat, val) { \ | |
584 | uint64_t m; \ | |
585 | while ((val) > (m = arc_stats.stat.value.ui64) && \ | |
586 | (m != atomic_cas_64(&arc_stats.stat.value.ui64, m, (val)))) \ | |
587 | continue; \ | |
588 | } | |
589 | ||
590 | #define ARCSTAT_MAXSTAT(stat) \ | |
591 | ARCSTAT_MAX(stat##_max, arc_stats.stat.value.ui64) | |
592 | ||
593 | /* | |
594 | * We define a macro to allow ARC hits/misses to be easily broken down by | |
595 | * two separate conditions, giving a total of four different subtypes for | |
596 | * each of hits and misses (so eight statistics total). | |
597 | */ | |
598 | #define ARCSTAT_CONDSTAT(cond1, stat1, notstat1, cond2, stat2, notstat2, stat) \ | |
599 | if (cond1) { \ | |
600 | if (cond2) { \ | |
601 | ARCSTAT_BUMP(arcstat_##stat1##_##stat2##_##stat); \ | |
602 | } else { \ | |
603 | ARCSTAT_BUMP(arcstat_##stat1##_##notstat2##_##stat); \ | |
604 | } \ | |
605 | } else { \ | |
606 | if (cond2) { \ | |
607 | ARCSTAT_BUMP(arcstat_##notstat1##_##stat2##_##stat); \ | |
608 | } else { \ | |
609 | ARCSTAT_BUMP(arcstat_##notstat1##_##notstat2##_##stat);\ | |
610 | } \ | |
611 | } | |
612 | ||
613 | kstat_t *arc_ksp; | |
614 | static arc_state_t *arc_anon; | |
615 | static arc_state_t *arc_mru; | |
616 | static arc_state_t *arc_mru_ghost; | |
617 | static arc_state_t *arc_mfu; | |
618 | static arc_state_t *arc_mfu_ghost; | |
619 | static arc_state_t *arc_l2c_only; | |
620 | ||
621 | /* | |
622 | * There are several ARC variables that are critical to export as kstats -- | |
623 | * but we don't want to have to grovel around in the kstat whenever we wish to | |
624 | * manipulate them. For these variables, we therefore define them to be in | |
625 | * terms of the statistic variable. This assures that we are not introducing | |
626 | * the possibility of inconsistency by having shadow copies of the variables, | |
627 | * while still allowing the code to be readable. | |
628 | */ | |
629 | #define arc_size ARCSTAT(arcstat_size) /* actual total arc size */ | |
630 | #define arc_p ARCSTAT(arcstat_p) /* target size of MRU */ | |
631 | #define arc_c ARCSTAT(arcstat_c) /* target size of cache */ | |
632 | #define arc_c_min ARCSTAT(arcstat_c_min) /* min target cache size */ | |
633 | #define arc_c_max ARCSTAT(arcstat_c_max) /* max target cache size */ | |
634 | #define arc_no_grow ARCSTAT(arcstat_no_grow) | |
635 | #define arc_tempreserve ARCSTAT(arcstat_tempreserve) | |
636 | #define arc_loaned_bytes ARCSTAT(arcstat_loaned_bytes) | |
637 | #define arc_meta_limit ARCSTAT(arcstat_meta_limit) /* max size for metadata */ | |
638 | #define arc_meta_min ARCSTAT(arcstat_meta_min) /* min size for metadata */ | |
639 | #define arc_meta_used ARCSTAT(arcstat_meta_used) /* size of metadata */ | |
640 | #define arc_meta_max ARCSTAT(arcstat_meta_max) /* max size of metadata */ | |
641 | #define arc_need_free ARCSTAT(arcstat_need_free) /* bytes to be freed */ | |
642 | #define arc_sys_free ARCSTAT(arcstat_sys_free) /* target system free bytes */ | |
643 | ||
644 | #define L2ARC_IS_VALID_COMPRESS(_c_) \ | |
645 | ((_c_) == ZIO_COMPRESS_LZ4 || (_c_) == ZIO_COMPRESS_EMPTY) | |
646 | ||
647 | static list_t arc_prune_list; | |
648 | static kmutex_t arc_prune_mtx; | |
649 | static taskq_t *arc_prune_taskq; | |
650 | static arc_buf_t *arc_eviction_list; | |
651 | static arc_buf_hdr_t arc_eviction_hdr; | |
652 | ||
653 | #define GHOST_STATE(state) \ | |
654 | ((state) == arc_mru_ghost || (state) == arc_mfu_ghost || \ | |
655 | (state) == arc_l2c_only) | |
656 | ||
657 | #define HDR_IN_HASH_TABLE(hdr) ((hdr)->b_flags & ARC_FLAG_IN_HASH_TABLE) | |
658 | #define HDR_IO_IN_PROGRESS(hdr) ((hdr)->b_flags & ARC_FLAG_IO_IN_PROGRESS) | |
659 | #define HDR_IO_ERROR(hdr) ((hdr)->b_flags & ARC_FLAG_IO_ERROR) | |
660 | #define HDR_PREFETCH(hdr) ((hdr)->b_flags & ARC_FLAG_PREFETCH) | |
661 | #define HDR_FREED_IN_READ(hdr) ((hdr)->b_flags & ARC_FLAG_FREED_IN_READ) | |
662 | #define HDR_BUF_AVAILABLE(hdr) ((hdr)->b_flags & ARC_FLAG_BUF_AVAILABLE) | |
663 | ||
664 | #define HDR_L2CACHE(hdr) ((hdr)->b_flags & ARC_FLAG_L2CACHE) | |
665 | #define HDR_L2COMPRESS(hdr) ((hdr)->b_flags & ARC_FLAG_L2COMPRESS) | |
666 | #define HDR_L2_READING(hdr) \ | |
667 | (((hdr)->b_flags & ARC_FLAG_IO_IN_PROGRESS) && \ | |
668 | ((hdr)->b_flags & ARC_FLAG_HAS_L2HDR)) | |
669 | #define HDR_L2_WRITING(hdr) ((hdr)->b_flags & ARC_FLAG_L2_WRITING) | |
670 | #define HDR_L2_EVICTED(hdr) ((hdr)->b_flags & ARC_FLAG_L2_EVICTED) | |
671 | #define HDR_L2_WRITE_HEAD(hdr) ((hdr)->b_flags & ARC_FLAG_L2_WRITE_HEAD) | |
672 | ||
673 | #define HDR_ISTYPE_METADATA(hdr) \ | |
674 | ((hdr)->b_flags & ARC_FLAG_BUFC_METADATA) | |
675 | #define HDR_ISTYPE_DATA(hdr) (!HDR_ISTYPE_METADATA(hdr)) | |
676 | ||
677 | #define HDR_HAS_L1HDR(hdr) ((hdr)->b_flags & ARC_FLAG_HAS_L1HDR) | |
678 | #define HDR_HAS_L2HDR(hdr) ((hdr)->b_flags & ARC_FLAG_HAS_L2HDR) | |
679 | ||
680 | /* | |
681 | * Other sizes | |
682 | */ | |
683 | ||
684 | #define HDR_FULL_SIZE ((int64_t)sizeof (arc_buf_hdr_t)) | |
685 | #define HDR_L2ONLY_SIZE ((int64_t)offsetof(arc_buf_hdr_t, b_l1hdr)) | |
686 | ||
687 | /* | |
688 | * Hash table routines | |
689 | */ | |
690 | ||
691 | #define HT_LOCK_ALIGN 64 | |
692 | #define HT_LOCK_PAD (P2NPHASE(sizeof (kmutex_t), (HT_LOCK_ALIGN))) | |
693 | ||
694 | struct ht_lock { | |
695 | kmutex_t ht_lock; | |
696 | #ifdef _KERNEL | |
697 | unsigned char pad[HT_LOCK_PAD]; | |
698 | #endif | |
699 | }; | |
700 | ||
701 | #define BUF_LOCKS 8192 | |
702 | typedef struct buf_hash_table { | |
703 | uint64_t ht_mask; | |
704 | arc_buf_hdr_t **ht_table; | |
705 | struct ht_lock ht_locks[BUF_LOCKS]; | |
706 | } buf_hash_table_t; | |
707 | ||
708 | static buf_hash_table_t buf_hash_table; | |
709 | ||
710 | #define BUF_HASH_INDEX(spa, dva, birth) \ | |
711 | (buf_hash(spa, dva, birth) & buf_hash_table.ht_mask) | |
712 | #define BUF_HASH_LOCK_NTRY(idx) (buf_hash_table.ht_locks[idx & (BUF_LOCKS-1)]) | |
713 | #define BUF_HASH_LOCK(idx) (&(BUF_HASH_LOCK_NTRY(idx).ht_lock)) | |
714 | #define HDR_LOCK(hdr) \ | |
715 | (BUF_HASH_LOCK(BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth))) | |
716 | ||
717 | uint64_t zfs_crc64_table[256]; | |
718 | ||
719 | /* | |
720 | * Level 2 ARC | |
721 | */ | |
722 | ||
723 | #define L2ARC_WRITE_SIZE (8 * 1024 * 1024) /* initial write max */ | |
724 | #define L2ARC_HEADROOM 2 /* num of writes */ | |
725 | /* | |
726 | * If we discover during ARC scan any buffers to be compressed, we boost | |
727 | * our headroom for the next scanning cycle by this percentage multiple. | |
728 | */ | |
729 | #define L2ARC_HEADROOM_BOOST 200 | |
730 | #define L2ARC_FEED_SECS 1 /* caching interval secs */ | |
731 | #define L2ARC_FEED_MIN_MS 200 /* min caching interval ms */ | |
732 | ||
733 | /* | |
734 | * Used to distinguish headers that are being process by | |
735 | * l2arc_write_buffers(), but have yet to be assigned to a l2arc disk | |
736 | * address. This can happen when the header is added to the l2arc's list | |
737 | * of buffers to write in the first stage of l2arc_write_buffers(), but | |
738 | * has not yet been written out which happens in the second stage of | |
739 | * l2arc_write_buffers(). | |
740 | */ | |
741 | #define L2ARC_ADDR_UNSET ((uint64_t)(-1)) | |
742 | ||
743 | #define l2arc_writes_sent ARCSTAT(arcstat_l2_writes_sent) | |
744 | #define l2arc_writes_done ARCSTAT(arcstat_l2_writes_done) | |
745 | ||
746 | /* L2ARC Performance Tunables */ | |
747 | unsigned long l2arc_write_max = L2ARC_WRITE_SIZE; /* def max write size */ | |
748 | unsigned long l2arc_write_boost = L2ARC_WRITE_SIZE; /* extra warmup write */ | |
749 | unsigned long l2arc_headroom = L2ARC_HEADROOM; /* # of dev writes */ | |
750 | unsigned long l2arc_headroom_boost = L2ARC_HEADROOM_BOOST; | |
751 | unsigned long l2arc_feed_secs = L2ARC_FEED_SECS; /* interval seconds */ | |
752 | unsigned long l2arc_feed_min_ms = L2ARC_FEED_MIN_MS; /* min interval msecs */ | |
753 | int l2arc_noprefetch = B_TRUE; /* don't cache prefetch bufs */ | |
754 | int l2arc_nocompress = B_FALSE; /* don't compress bufs */ | |
755 | int l2arc_feed_again = B_TRUE; /* turbo warmup */ | |
756 | int l2arc_norw = B_FALSE; /* no reads during writes */ | |
757 | ||
758 | /* | |
759 | * L2ARC Internals | |
760 | */ | |
761 | static list_t L2ARC_dev_list; /* device list */ | |
762 | static list_t *l2arc_dev_list; /* device list pointer */ | |
763 | static kmutex_t l2arc_dev_mtx; /* device list mutex */ | |
764 | static l2arc_dev_t *l2arc_dev_last; /* last device used */ | |
765 | static list_t L2ARC_free_on_write; /* free after write buf list */ | |
766 | static list_t *l2arc_free_on_write; /* free after write list ptr */ | |
767 | static kmutex_t l2arc_free_on_write_mtx; /* mutex for list */ | |
768 | static uint64_t l2arc_ndev; /* number of devices */ | |
769 | ||
770 | typedef struct l2arc_read_callback { | |
771 | arc_buf_t *l2rcb_buf; /* read buffer */ | |
772 | spa_t *l2rcb_spa; /* spa */ | |
773 | blkptr_t l2rcb_bp; /* original blkptr */ | |
774 | zbookmark_phys_t l2rcb_zb; /* original bookmark */ | |
775 | int l2rcb_flags; /* original flags */ | |
776 | enum zio_compress l2rcb_compress; /* applied compress */ | |
777 | } l2arc_read_callback_t; | |
778 | ||
779 | typedef struct l2arc_data_free { | |
780 | /* protected by l2arc_free_on_write_mtx */ | |
781 | void *l2df_data; | |
782 | size_t l2df_size; | |
783 | void (*l2df_func)(void *, size_t); | |
784 | list_node_t l2df_list_node; | |
785 | } l2arc_data_free_t; | |
786 | ||
787 | static kmutex_t l2arc_feed_thr_lock; | |
788 | static kcondvar_t l2arc_feed_thr_cv; | |
789 | static uint8_t l2arc_thread_exit; | |
790 | ||
791 | static void arc_get_data_buf(arc_buf_t *); | |
792 | static void arc_access(arc_buf_hdr_t *, kmutex_t *); | |
793 | static boolean_t arc_is_overflowing(void); | |
794 | static void arc_buf_watch(arc_buf_t *); | |
795 | static void arc_tuning_update(void); | |
796 | ||
797 | static arc_buf_contents_t arc_buf_type(arc_buf_hdr_t *); | |
798 | static uint32_t arc_bufc_to_flags(arc_buf_contents_t); | |
799 | ||
800 | static boolean_t l2arc_write_eligible(uint64_t, arc_buf_hdr_t *); | |
801 | static void l2arc_read_done(zio_t *); | |
802 | ||
803 | static boolean_t l2arc_compress_buf(arc_buf_hdr_t *); | |
804 | static void l2arc_decompress_zio(zio_t *, arc_buf_hdr_t *, enum zio_compress); | |
805 | static void l2arc_release_cdata_buf(arc_buf_hdr_t *); | |
806 | ||
807 | static uint64_t | |
808 | buf_hash(uint64_t spa, const dva_t *dva, uint64_t birth) | |
809 | { | |
810 | uint8_t *vdva = (uint8_t *)dva; | |
811 | uint64_t crc = -1ULL; | |
812 | int i; | |
813 | ||
814 | ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); | |
815 | ||
816 | for (i = 0; i < sizeof (dva_t); i++) | |
817 | crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ vdva[i]) & 0xFF]; | |
818 | ||
819 | crc ^= (spa>>8) ^ birth; | |
820 | ||
821 | return (crc); | |
822 | } | |
823 | ||
824 | #define BUF_EMPTY(buf) \ | |
825 | ((buf)->b_dva.dva_word[0] == 0 && \ | |
826 | (buf)->b_dva.dva_word[1] == 0) | |
827 | ||
828 | #define BUF_EQUAL(spa, dva, birth, buf) \ | |
829 | ((buf)->b_dva.dva_word[0] == (dva)->dva_word[0]) && \ | |
830 | ((buf)->b_dva.dva_word[1] == (dva)->dva_word[1]) && \ | |
831 | ((buf)->b_birth == birth) && ((buf)->b_spa == spa) | |
832 | ||
833 | static void | |
834 | buf_discard_identity(arc_buf_hdr_t *hdr) | |
835 | { | |
836 | hdr->b_dva.dva_word[0] = 0; | |
837 | hdr->b_dva.dva_word[1] = 0; | |
838 | hdr->b_birth = 0; | |
839 | } | |
840 | ||
841 | static arc_buf_hdr_t * | |
842 | buf_hash_find(uint64_t spa, const blkptr_t *bp, kmutex_t **lockp) | |
843 | { | |
844 | const dva_t *dva = BP_IDENTITY(bp); | |
845 | uint64_t birth = BP_PHYSICAL_BIRTH(bp); | |
846 | uint64_t idx = BUF_HASH_INDEX(spa, dva, birth); | |
847 | kmutex_t *hash_lock = BUF_HASH_LOCK(idx); | |
848 | arc_buf_hdr_t *hdr; | |
849 | ||
850 | mutex_enter(hash_lock); | |
851 | for (hdr = buf_hash_table.ht_table[idx]; hdr != NULL; | |
852 | hdr = hdr->b_hash_next) { | |
853 | if (BUF_EQUAL(spa, dva, birth, hdr)) { | |
854 | *lockp = hash_lock; | |
855 | return (hdr); | |
856 | } | |
857 | } | |
858 | mutex_exit(hash_lock); | |
859 | *lockp = NULL; | |
860 | return (NULL); | |
861 | } | |
862 | ||
863 | /* | |
864 | * Insert an entry into the hash table. If there is already an element | |
865 | * equal to elem in the hash table, then the already existing element | |
866 | * will be returned and the new element will not be inserted. | |
867 | * Otherwise returns NULL. | |
868 | * If lockp == NULL, the caller is assumed to already hold the hash lock. | |
869 | */ | |
870 | static arc_buf_hdr_t * | |
871 | buf_hash_insert(arc_buf_hdr_t *hdr, kmutex_t **lockp) | |
872 | { | |
873 | uint64_t idx = BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth); | |
874 | kmutex_t *hash_lock = BUF_HASH_LOCK(idx); | |
875 | arc_buf_hdr_t *fhdr; | |
876 | uint32_t i; | |
877 | ||
878 | ASSERT(!DVA_IS_EMPTY(&hdr->b_dva)); | |
879 | ASSERT(hdr->b_birth != 0); | |
880 | ASSERT(!HDR_IN_HASH_TABLE(hdr)); | |
881 | ||
882 | if (lockp != NULL) { | |
883 | *lockp = hash_lock; | |
884 | mutex_enter(hash_lock); | |
885 | } else { | |
886 | ASSERT(MUTEX_HELD(hash_lock)); | |
887 | } | |
888 | ||
889 | for (fhdr = buf_hash_table.ht_table[idx], i = 0; fhdr != NULL; | |
890 | fhdr = fhdr->b_hash_next, i++) { | |
891 | if (BUF_EQUAL(hdr->b_spa, &hdr->b_dva, hdr->b_birth, fhdr)) | |
892 | return (fhdr); | |
893 | } | |
894 | ||
895 | hdr->b_hash_next = buf_hash_table.ht_table[idx]; | |
896 | buf_hash_table.ht_table[idx] = hdr; | |
897 | hdr->b_flags |= ARC_FLAG_IN_HASH_TABLE; | |
898 | ||
899 | /* collect some hash table performance data */ | |
900 | if (i > 0) { | |
901 | ARCSTAT_BUMP(arcstat_hash_collisions); | |
902 | if (i == 1) | |
903 | ARCSTAT_BUMP(arcstat_hash_chains); | |
904 | ||
905 | ARCSTAT_MAX(arcstat_hash_chain_max, i); | |
906 | } | |
907 | ||
908 | ARCSTAT_BUMP(arcstat_hash_elements); | |
909 | ARCSTAT_MAXSTAT(arcstat_hash_elements); | |
910 | ||
911 | return (NULL); | |
912 | } | |
913 | ||
914 | static void | |
915 | buf_hash_remove(arc_buf_hdr_t *hdr) | |
916 | { | |
917 | arc_buf_hdr_t *fhdr, **hdrp; | |
918 | uint64_t idx = BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth); | |
919 | ||
920 | ASSERT(MUTEX_HELD(BUF_HASH_LOCK(idx))); | |
921 | ASSERT(HDR_IN_HASH_TABLE(hdr)); | |
922 | ||
923 | hdrp = &buf_hash_table.ht_table[idx]; | |
924 | while ((fhdr = *hdrp) != hdr) { | |
925 | ASSERT(fhdr != NULL); | |
926 | hdrp = &fhdr->b_hash_next; | |
927 | } | |
928 | *hdrp = hdr->b_hash_next; | |
929 | hdr->b_hash_next = NULL; | |
930 | hdr->b_flags &= ~ARC_FLAG_IN_HASH_TABLE; | |
931 | ||
932 | /* collect some hash table performance data */ | |
933 | ARCSTAT_BUMPDOWN(arcstat_hash_elements); | |
934 | ||
935 | if (buf_hash_table.ht_table[idx] && | |
936 | buf_hash_table.ht_table[idx]->b_hash_next == NULL) | |
937 | ARCSTAT_BUMPDOWN(arcstat_hash_chains); | |
938 | } | |
939 | ||
940 | /* | |
941 | * Global data structures and functions for the buf kmem cache. | |
942 | */ | |
943 | static kmem_cache_t *hdr_full_cache; | |
944 | static kmem_cache_t *hdr_l2only_cache; | |
945 | static kmem_cache_t *buf_cache; | |
946 | ||
947 | static void | |
948 | buf_fini(void) | |
949 | { | |
950 | int i; | |
951 | ||
952 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
953 | /* | |
954 | * Large allocations which do not require contiguous pages | |
955 | * should be using vmem_free() in the linux kernel\ | |
956 | */ | |
957 | vmem_free(buf_hash_table.ht_table, | |
958 | (buf_hash_table.ht_mask + 1) * sizeof (void *)); | |
959 | #else | |
960 | kmem_free(buf_hash_table.ht_table, | |
961 | (buf_hash_table.ht_mask + 1) * sizeof (void *)); | |
962 | #endif | |
963 | for (i = 0; i < BUF_LOCKS; i++) | |
964 | mutex_destroy(&buf_hash_table.ht_locks[i].ht_lock); | |
965 | kmem_cache_destroy(hdr_full_cache); | |
966 | kmem_cache_destroy(hdr_l2only_cache); | |
967 | kmem_cache_destroy(buf_cache); | |
968 | } | |
969 | ||
970 | /* | |
971 | * Constructor callback - called when the cache is empty | |
972 | * and a new buf is requested. | |
973 | */ | |
974 | /* ARGSUSED */ | |
975 | static int | |
976 | hdr_full_cons(void *vbuf, void *unused, int kmflag) | |
977 | { | |
978 | arc_buf_hdr_t *hdr = vbuf; | |
979 | ||
980 | bzero(hdr, HDR_FULL_SIZE); | |
981 | cv_init(&hdr->b_l1hdr.b_cv, NULL, CV_DEFAULT, NULL); | |
982 | refcount_create(&hdr->b_l1hdr.b_refcnt); | |
983 | mutex_init(&hdr->b_l1hdr.b_freeze_lock, NULL, MUTEX_DEFAULT, NULL); | |
984 | list_link_init(&hdr->b_l1hdr.b_arc_node); | |
985 | list_link_init(&hdr->b_l2hdr.b_l2node); | |
986 | multilist_link_init(&hdr->b_l1hdr.b_arc_node); | |
987 | arc_space_consume(HDR_FULL_SIZE, ARC_SPACE_HDRS); | |
988 | ||
989 | return (0); | |
990 | } | |
991 | ||
992 | /* ARGSUSED */ | |
993 | static int | |
994 | hdr_l2only_cons(void *vbuf, void *unused, int kmflag) | |
995 | { | |
996 | arc_buf_hdr_t *hdr = vbuf; | |
997 | ||
998 | bzero(hdr, HDR_L2ONLY_SIZE); | |
999 | arc_space_consume(HDR_L2ONLY_SIZE, ARC_SPACE_L2HDRS); | |
1000 | ||
1001 | return (0); | |
1002 | } | |
1003 | ||
1004 | /* ARGSUSED */ | |
1005 | static int | |
1006 | buf_cons(void *vbuf, void *unused, int kmflag) | |
1007 | { | |
1008 | arc_buf_t *buf = vbuf; | |
1009 | ||
1010 | bzero(buf, sizeof (arc_buf_t)); | |
1011 | mutex_init(&buf->b_evict_lock, NULL, MUTEX_DEFAULT, NULL); | |
1012 | arc_space_consume(sizeof (arc_buf_t), ARC_SPACE_HDRS); | |
1013 | ||
1014 | return (0); | |
1015 | } | |
1016 | ||
1017 | /* | |
1018 | * Destructor callback - called when a cached buf is | |
1019 | * no longer required. | |
1020 | */ | |
1021 | /* ARGSUSED */ | |
1022 | static void | |
1023 | hdr_full_dest(void *vbuf, void *unused) | |
1024 | { | |
1025 | arc_buf_hdr_t *hdr = vbuf; | |
1026 | ||
1027 | ASSERT(BUF_EMPTY(hdr)); | |
1028 | cv_destroy(&hdr->b_l1hdr.b_cv); | |
1029 | refcount_destroy(&hdr->b_l1hdr.b_refcnt); | |
1030 | mutex_destroy(&hdr->b_l1hdr.b_freeze_lock); | |
1031 | ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); | |
1032 | arc_space_return(HDR_FULL_SIZE, ARC_SPACE_HDRS); | |
1033 | } | |
1034 | ||
1035 | /* ARGSUSED */ | |
1036 | static void | |
1037 | hdr_l2only_dest(void *vbuf, void *unused) | |
1038 | { | |
1039 | ASSERTV(arc_buf_hdr_t *hdr = vbuf); | |
1040 | ||
1041 | ASSERT(BUF_EMPTY(hdr)); | |
1042 | arc_space_return(HDR_L2ONLY_SIZE, ARC_SPACE_L2HDRS); | |
1043 | } | |
1044 | ||
1045 | /* ARGSUSED */ | |
1046 | static void | |
1047 | buf_dest(void *vbuf, void *unused) | |
1048 | { | |
1049 | arc_buf_t *buf = vbuf; | |
1050 | ||
1051 | mutex_destroy(&buf->b_evict_lock); | |
1052 | arc_space_return(sizeof (arc_buf_t), ARC_SPACE_HDRS); | |
1053 | } | |
1054 | ||
1055 | /* | |
1056 | * Reclaim callback -- invoked when memory is low. | |
1057 | */ | |
1058 | /* ARGSUSED */ | |
1059 | static void | |
1060 | hdr_recl(void *unused) | |
1061 | { | |
1062 | dprintf("hdr_recl called\n"); | |
1063 | /* | |
1064 | * umem calls the reclaim func when we destroy the buf cache, | |
1065 | * which is after we do arc_fini(). | |
1066 | */ | |
1067 | if (!arc_dead) | |
1068 | cv_signal(&arc_reclaim_thread_cv); | |
1069 | } | |
1070 | ||
1071 | static void | |
1072 | buf_init(void) | |
1073 | { | |
1074 | uint64_t *ct; | |
1075 | uint64_t hsize = 1ULL << 12; | |
1076 | int i, j; | |
1077 | ||
1078 | /* | |
1079 | * The hash table is big enough to fill all of physical memory | |
1080 | * with an average block size of zfs_arc_average_blocksize (default 8K). | |
1081 | * By default, the table will take up | |
1082 | * totalmem * sizeof(void*) / 8K (1MB per GB with 8-byte pointers). | |
1083 | */ | |
1084 | while (hsize * zfs_arc_average_blocksize < physmem * PAGESIZE) | |
1085 | hsize <<= 1; | |
1086 | retry: | |
1087 | buf_hash_table.ht_mask = hsize - 1; | |
1088 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
1089 | /* | |
1090 | * Large allocations which do not require contiguous pages | |
1091 | * should be using vmem_alloc() in the linux kernel | |
1092 | */ | |
1093 | buf_hash_table.ht_table = | |
1094 | vmem_zalloc(hsize * sizeof (void*), KM_SLEEP); | |
1095 | #else | |
1096 | buf_hash_table.ht_table = | |
1097 | kmem_zalloc(hsize * sizeof (void*), KM_NOSLEEP); | |
1098 | #endif | |
1099 | if (buf_hash_table.ht_table == NULL) { | |
1100 | ASSERT(hsize > (1ULL << 8)); | |
1101 | hsize >>= 1; | |
1102 | goto retry; | |
1103 | } | |
1104 | ||
1105 | hdr_full_cache = kmem_cache_create("arc_buf_hdr_t_full", HDR_FULL_SIZE, | |
1106 | 0, hdr_full_cons, hdr_full_dest, hdr_recl, NULL, NULL, 0); | |
1107 | hdr_l2only_cache = kmem_cache_create("arc_buf_hdr_t_l2only", | |
1108 | HDR_L2ONLY_SIZE, 0, hdr_l2only_cons, hdr_l2only_dest, hdr_recl, | |
1109 | NULL, NULL, 0); | |
1110 | buf_cache = kmem_cache_create("arc_buf_t", sizeof (arc_buf_t), | |
1111 | 0, buf_cons, buf_dest, NULL, NULL, NULL, 0); | |
1112 | ||
1113 | for (i = 0; i < 256; i++) | |
1114 | for (ct = zfs_crc64_table + i, *ct = i, j = 8; j > 0; j--) | |
1115 | *ct = (*ct >> 1) ^ (-(*ct & 1) & ZFS_CRC64_POLY); | |
1116 | ||
1117 | for (i = 0; i < BUF_LOCKS; i++) { | |
1118 | mutex_init(&buf_hash_table.ht_locks[i].ht_lock, | |
1119 | NULL, MUTEX_DEFAULT, NULL); | |
1120 | } | |
1121 | } | |
1122 | ||
1123 | /* | |
1124 | * Transition between the two allocation states for the arc_buf_hdr struct. | |
1125 | * The arc_buf_hdr struct can be allocated with (hdr_full_cache) or without | |
1126 | * (hdr_l2only_cache) the fields necessary for the L1 cache - the smaller | |
1127 | * version is used when a cache buffer is only in the L2ARC in order to reduce | |
1128 | * memory usage. | |
1129 | */ | |
1130 | static arc_buf_hdr_t * | |
1131 | arc_hdr_realloc(arc_buf_hdr_t *hdr, kmem_cache_t *old, kmem_cache_t *new) | |
1132 | { | |
1133 | arc_buf_hdr_t *nhdr; | |
1134 | l2arc_dev_t *dev; | |
1135 | ||
1136 | ASSERT(HDR_HAS_L2HDR(hdr)); | |
1137 | ASSERT((old == hdr_full_cache && new == hdr_l2only_cache) || | |
1138 | (old == hdr_l2only_cache && new == hdr_full_cache)); | |
1139 | ||
1140 | dev = hdr->b_l2hdr.b_dev; | |
1141 | nhdr = kmem_cache_alloc(new, KM_PUSHPAGE); | |
1142 | ||
1143 | ASSERT(MUTEX_HELD(HDR_LOCK(hdr))); | |
1144 | buf_hash_remove(hdr); | |
1145 | ||
1146 | bcopy(hdr, nhdr, HDR_L2ONLY_SIZE); | |
1147 | ||
1148 | if (new == hdr_full_cache) { | |
1149 | nhdr->b_flags |= ARC_FLAG_HAS_L1HDR; | |
1150 | /* | |
1151 | * arc_access and arc_change_state need to be aware that a | |
1152 | * header has just come out of L2ARC, so we set its state to | |
1153 | * l2c_only even though it's about to change. | |
1154 | */ | |
1155 | nhdr->b_l1hdr.b_state = arc_l2c_only; | |
1156 | ||
1157 | /* Verify previous threads set to NULL before freeing */ | |
1158 | ASSERT3P(nhdr->b_l1hdr.b_tmp_cdata, ==, NULL); | |
1159 | } else { | |
1160 | ASSERT(hdr->b_l1hdr.b_buf == NULL); | |
1161 | ASSERT0(hdr->b_l1hdr.b_datacnt); | |
1162 | ||
1163 | /* | |
1164 | * If we've reached here, We must have been called from | |
1165 | * arc_evict_hdr(), as such we should have already been | |
1166 | * removed from any ghost list we were previously on | |
1167 | * (which protects us from racing with arc_evict_state), | |
1168 | * thus no locking is needed during this check. | |
1169 | */ | |
1170 | ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); | |
1171 | ||
1172 | /* | |
1173 | * A buffer must not be moved into the arc_l2c_only | |
1174 | * state if it's not finished being written out to the | |
1175 | * l2arc device. Otherwise, the b_l1hdr.b_tmp_cdata field | |
1176 | * might try to be accessed, even though it was removed. | |
1177 | */ | |
1178 | VERIFY(!HDR_L2_WRITING(hdr)); | |
1179 | VERIFY3P(hdr->b_l1hdr.b_tmp_cdata, ==, NULL); | |
1180 | ||
1181 | nhdr->b_flags &= ~ARC_FLAG_HAS_L1HDR; | |
1182 | } | |
1183 | /* | |
1184 | * The header has been reallocated so we need to re-insert it into any | |
1185 | * lists it was on. | |
1186 | */ | |
1187 | (void) buf_hash_insert(nhdr, NULL); | |
1188 | ||
1189 | ASSERT(list_link_active(&hdr->b_l2hdr.b_l2node)); | |
1190 | ||
1191 | mutex_enter(&dev->l2ad_mtx); | |
1192 | ||
1193 | /* | |
1194 | * We must place the realloc'ed header back into the list at | |
1195 | * the same spot. Otherwise, if it's placed earlier in the list, | |
1196 | * l2arc_write_buffers() could find it during the function's | |
1197 | * write phase, and try to write it out to the l2arc. | |
1198 | */ | |
1199 | list_insert_after(&dev->l2ad_buflist, hdr, nhdr); | |
1200 | list_remove(&dev->l2ad_buflist, hdr); | |
1201 | ||
1202 | mutex_exit(&dev->l2ad_mtx); | |
1203 | ||
1204 | /* | |
1205 | * Since we're using the pointer address as the tag when | |
1206 | * incrementing and decrementing the l2ad_alloc refcount, we | |
1207 | * must remove the old pointer (that we're about to destroy) and | |
1208 | * add the new pointer to the refcount. Otherwise we'd remove | |
1209 | * the wrong pointer address when calling arc_hdr_destroy() later. | |
1210 | */ | |
1211 | ||
1212 | (void) refcount_remove_many(&dev->l2ad_alloc, | |
1213 | hdr->b_l2hdr.b_asize, hdr); | |
1214 | ||
1215 | (void) refcount_add_many(&dev->l2ad_alloc, | |
1216 | nhdr->b_l2hdr.b_asize, nhdr); | |
1217 | ||
1218 | buf_discard_identity(hdr); | |
1219 | hdr->b_freeze_cksum = NULL; | |
1220 | kmem_cache_free(old, hdr); | |
1221 | ||
1222 | return (nhdr); | |
1223 | } | |
1224 | ||
1225 | ||
1226 | #define ARC_MINTIME (hz>>4) /* 62 ms */ | |
1227 | ||
1228 | static void | |
1229 | arc_cksum_verify(arc_buf_t *buf) | |
1230 | { | |
1231 | zio_cksum_t zc; | |
1232 | ||
1233 | if (!(zfs_flags & ZFS_DEBUG_MODIFY)) | |
1234 | return; | |
1235 | ||
1236 | mutex_enter(&buf->b_hdr->b_l1hdr.b_freeze_lock); | |
1237 | if (buf->b_hdr->b_freeze_cksum == NULL || HDR_IO_ERROR(buf->b_hdr)) { | |
1238 | mutex_exit(&buf->b_hdr->b_l1hdr.b_freeze_lock); | |
1239 | return; | |
1240 | } | |
1241 | fletcher_2_native(buf->b_data, buf->b_hdr->b_size, &zc); | |
1242 | if (!ZIO_CHECKSUM_EQUAL(*buf->b_hdr->b_freeze_cksum, zc)) | |
1243 | panic("buffer modified while frozen!"); | |
1244 | mutex_exit(&buf->b_hdr->b_l1hdr.b_freeze_lock); | |
1245 | } | |
1246 | ||
1247 | static int | |
1248 | arc_cksum_equal(arc_buf_t *buf) | |
1249 | { | |
1250 | zio_cksum_t zc; | |
1251 | int equal; | |
1252 | ||
1253 | mutex_enter(&buf->b_hdr->b_l1hdr.b_freeze_lock); | |
1254 | fletcher_2_native(buf->b_data, buf->b_hdr->b_size, &zc); | |
1255 | equal = ZIO_CHECKSUM_EQUAL(*buf->b_hdr->b_freeze_cksum, zc); | |
1256 | mutex_exit(&buf->b_hdr->b_l1hdr.b_freeze_lock); | |
1257 | ||
1258 | return (equal); | |
1259 | } | |
1260 | ||
1261 | static void | |
1262 | arc_cksum_compute(arc_buf_t *buf, boolean_t force) | |
1263 | { | |
1264 | if (!force && !(zfs_flags & ZFS_DEBUG_MODIFY)) | |
1265 | return; | |
1266 | ||
1267 | mutex_enter(&buf->b_hdr->b_l1hdr.b_freeze_lock); | |
1268 | if (buf->b_hdr->b_freeze_cksum != NULL) { | |
1269 | mutex_exit(&buf->b_hdr->b_l1hdr.b_freeze_lock); | |
1270 | return; | |
1271 | } | |
1272 | buf->b_hdr->b_freeze_cksum = kmem_alloc(sizeof (zio_cksum_t), KM_SLEEP); | |
1273 | fletcher_2_native(buf->b_data, buf->b_hdr->b_size, | |
1274 | buf->b_hdr->b_freeze_cksum); | |
1275 | mutex_exit(&buf->b_hdr->b_l1hdr.b_freeze_lock); | |
1276 | arc_buf_watch(buf); | |
1277 | } | |
1278 | ||
1279 | #ifndef _KERNEL | |
1280 | void | |
1281 | arc_buf_sigsegv(int sig, siginfo_t *si, void *unused) | |
1282 | { | |
1283 | panic("Got SIGSEGV at address: 0x%lx\n", (long) si->si_addr); | |
1284 | } | |
1285 | #endif | |
1286 | ||
1287 | /* ARGSUSED */ | |
1288 | static void | |
1289 | arc_buf_unwatch(arc_buf_t *buf) | |
1290 | { | |
1291 | #ifndef _KERNEL | |
1292 | if (arc_watch) { | |
1293 | ASSERT0(mprotect(buf->b_data, buf->b_hdr->b_size, | |
1294 | PROT_READ | PROT_WRITE)); | |
1295 | } | |
1296 | #endif | |
1297 | } | |
1298 | ||
1299 | /* ARGSUSED */ | |
1300 | static void | |
1301 | arc_buf_watch(arc_buf_t *buf) | |
1302 | { | |
1303 | #ifndef _KERNEL | |
1304 | if (arc_watch) | |
1305 | ASSERT0(mprotect(buf->b_data, buf->b_hdr->b_size, PROT_READ)); | |
1306 | #endif | |
1307 | } | |
1308 | ||
1309 | static arc_buf_contents_t | |
1310 | arc_buf_type(arc_buf_hdr_t *hdr) | |
1311 | { | |
1312 | if (HDR_ISTYPE_METADATA(hdr)) { | |
1313 | return (ARC_BUFC_METADATA); | |
1314 | } else { | |
1315 | return (ARC_BUFC_DATA); | |
1316 | } | |
1317 | } | |
1318 | ||
1319 | static uint32_t | |
1320 | arc_bufc_to_flags(arc_buf_contents_t type) | |
1321 | { | |
1322 | switch (type) { | |
1323 | case ARC_BUFC_DATA: | |
1324 | /* metadata field is 0 if buffer contains normal data */ | |
1325 | return (0); | |
1326 | case ARC_BUFC_METADATA: | |
1327 | return (ARC_FLAG_BUFC_METADATA); | |
1328 | default: | |
1329 | break; | |
1330 | } | |
1331 | panic("undefined ARC buffer type!"); | |
1332 | return ((uint32_t)-1); | |
1333 | } | |
1334 | ||
1335 | void | |
1336 | arc_buf_thaw(arc_buf_t *buf) | |
1337 | { | |
1338 | if (zfs_flags & ZFS_DEBUG_MODIFY) { | |
1339 | if (buf->b_hdr->b_l1hdr.b_state != arc_anon) | |
1340 | panic("modifying non-anon buffer!"); | |
1341 | if (HDR_IO_IN_PROGRESS(buf->b_hdr)) | |
1342 | panic("modifying buffer while i/o in progress!"); | |
1343 | arc_cksum_verify(buf); | |
1344 | } | |
1345 | ||
1346 | mutex_enter(&buf->b_hdr->b_l1hdr.b_freeze_lock); | |
1347 | if (buf->b_hdr->b_freeze_cksum != NULL) { | |
1348 | kmem_free(buf->b_hdr->b_freeze_cksum, sizeof (zio_cksum_t)); | |
1349 | buf->b_hdr->b_freeze_cksum = NULL; | |
1350 | } | |
1351 | ||
1352 | mutex_exit(&buf->b_hdr->b_l1hdr.b_freeze_lock); | |
1353 | ||
1354 | arc_buf_unwatch(buf); | |
1355 | } | |
1356 | ||
1357 | void | |
1358 | arc_buf_freeze(arc_buf_t *buf) | |
1359 | { | |
1360 | kmutex_t *hash_lock; | |
1361 | ||
1362 | if (!(zfs_flags & ZFS_DEBUG_MODIFY)) | |
1363 | return; | |
1364 | ||
1365 | hash_lock = HDR_LOCK(buf->b_hdr); | |
1366 | mutex_enter(hash_lock); | |
1367 | ||
1368 | ASSERT(buf->b_hdr->b_freeze_cksum != NULL || | |
1369 | buf->b_hdr->b_l1hdr.b_state == arc_anon); | |
1370 | arc_cksum_compute(buf, B_FALSE); | |
1371 | mutex_exit(hash_lock); | |
1372 | ||
1373 | } | |
1374 | ||
1375 | static void | |
1376 | add_reference(arc_buf_hdr_t *hdr, kmutex_t *hash_lock, void *tag) | |
1377 | { | |
1378 | arc_state_t *state; | |
1379 | ||
1380 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
1381 | ASSERT(MUTEX_HELD(hash_lock)); | |
1382 | ||
1383 | state = hdr->b_l1hdr.b_state; | |
1384 | ||
1385 | if ((refcount_add(&hdr->b_l1hdr.b_refcnt, tag) == 1) && | |
1386 | (state != arc_anon)) { | |
1387 | /* We don't use the L2-only state list. */ | |
1388 | if (state != arc_l2c_only) { | |
1389 | arc_buf_contents_t type = arc_buf_type(hdr); | |
1390 | uint64_t delta = hdr->b_size * hdr->b_l1hdr.b_datacnt; | |
1391 | multilist_t *list = &state->arcs_list[type]; | |
1392 | uint64_t *size = &state->arcs_lsize[type]; | |
1393 | ||
1394 | multilist_remove(list, hdr); | |
1395 | ||
1396 | if (GHOST_STATE(state)) { | |
1397 | ASSERT0(hdr->b_l1hdr.b_datacnt); | |
1398 | ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); | |
1399 | delta = hdr->b_size; | |
1400 | } | |
1401 | ASSERT(delta > 0); | |
1402 | ASSERT3U(*size, >=, delta); | |
1403 | atomic_add_64(size, -delta); | |
1404 | } | |
1405 | /* remove the prefetch flag if we get a reference */ | |
1406 | hdr->b_flags &= ~ARC_FLAG_PREFETCH; | |
1407 | } | |
1408 | } | |
1409 | ||
1410 | static int | |
1411 | remove_reference(arc_buf_hdr_t *hdr, kmutex_t *hash_lock, void *tag) | |
1412 | { | |
1413 | int cnt; | |
1414 | arc_state_t *state = hdr->b_l1hdr.b_state; | |
1415 | ||
1416 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
1417 | ASSERT(state == arc_anon || MUTEX_HELD(hash_lock)); | |
1418 | ASSERT(!GHOST_STATE(state)); | |
1419 | ||
1420 | /* | |
1421 | * arc_l2c_only counts as a ghost state so we don't need to explicitly | |
1422 | * check to prevent usage of the arc_l2c_only list. | |
1423 | */ | |
1424 | if (((cnt = refcount_remove(&hdr->b_l1hdr.b_refcnt, tag)) == 0) && | |
1425 | (state != arc_anon)) { | |
1426 | arc_buf_contents_t type = arc_buf_type(hdr); | |
1427 | multilist_t *list = &state->arcs_list[type]; | |
1428 | uint64_t *size = &state->arcs_lsize[type]; | |
1429 | ||
1430 | multilist_insert(list, hdr); | |
1431 | ||
1432 | ASSERT(hdr->b_l1hdr.b_datacnt > 0); | |
1433 | atomic_add_64(size, hdr->b_size * | |
1434 | hdr->b_l1hdr.b_datacnt); | |
1435 | } | |
1436 | return (cnt); | |
1437 | } | |
1438 | ||
1439 | /* | |
1440 | * Returns detailed information about a specific arc buffer. When the | |
1441 | * state_index argument is set the function will calculate the arc header | |
1442 | * list position for its arc state. Since this requires a linear traversal | |
1443 | * callers are strongly encourage not to do this. However, it can be helpful | |
1444 | * for targeted analysis so the functionality is provided. | |
1445 | */ | |
1446 | void | |
1447 | arc_buf_info(arc_buf_t *ab, arc_buf_info_t *abi, int state_index) | |
1448 | { | |
1449 | arc_buf_hdr_t *hdr = ab->b_hdr; | |
1450 | l1arc_buf_hdr_t *l1hdr = NULL; | |
1451 | l2arc_buf_hdr_t *l2hdr = NULL; | |
1452 | arc_state_t *state = NULL; | |
1453 | ||
1454 | memset(abi, 0, sizeof (arc_buf_info_t)); | |
1455 | ||
1456 | if (hdr == NULL) | |
1457 | return; | |
1458 | ||
1459 | abi->abi_flags = hdr->b_flags; | |
1460 | ||
1461 | if (HDR_HAS_L1HDR(hdr)) { | |
1462 | l1hdr = &hdr->b_l1hdr; | |
1463 | state = l1hdr->b_state; | |
1464 | } | |
1465 | if (HDR_HAS_L2HDR(hdr)) | |
1466 | l2hdr = &hdr->b_l2hdr; | |
1467 | ||
1468 | if (l1hdr) { | |
1469 | abi->abi_datacnt = l1hdr->b_datacnt; | |
1470 | abi->abi_access = l1hdr->b_arc_access; | |
1471 | abi->abi_mru_hits = l1hdr->b_mru_hits; | |
1472 | abi->abi_mru_ghost_hits = l1hdr->b_mru_ghost_hits; | |
1473 | abi->abi_mfu_hits = l1hdr->b_mfu_hits; | |
1474 | abi->abi_mfu_ghost_hits = l1hdr->b_mfu_ghost_hits; | |
1475 | abi->abi_holds = refcount_count(&l1hdr->b_refcnt); | |
1476 | } | |
1477 | ||
1478 | if (l2hdr) { | |
1479 | abi->abi_l2arc_dattr = l2hdr->b_daddr; | |
1480 | abi->abi_l2arc_asize = l2hdr->b_asize; | |
1481 | abi->abi_l2arc_compress = l2hdr->b_compress; | |
1482 | abi->abi_l2arc_hits = l2hdr->b_hits; | |
1483 | } | |
1484 | ||
1485 | abi->abi_state_type = state ? state->arcs_state : ARC_STATE_ANON; | |
1486 | abi->abi_state_contents = arc_buf_type(hdr); | |
1487 | abi->abi_size = hdr->b_size; | |
1488 | } | |
1489 | ||
1490 | /* | |
1491 | * Move the supplied buffer to the indicated state. The hash lock | |
1492 | * for the buffer must be held by the caller. | |
1493 | */ | |
1494 | static void | |
1495 | arc_change_state(arc_state_t *new_state, arc_buf_hdr_t *hdr, | |
1496 | kmutex_t *hash_lock) | |
1497 | { | |
1498 | arc_state_t *old_state; | |
1499 | int64_t refcnt; | |
1500 | uint32_t datacnt; | |
1501 | uint64_t from_delta, to_delta; | |
1502 | arc_buf_contents_t buftype = arc_buf_type(hdr); | |
1503 | ||
1504 | /* | |
1505 | * We almost always have an L1 hdr here, since we call arc_hdr_realloc() | |
1506 | * in arc_read() when bringing a buffer out of the L2ARC. However, the | |
1507 | * L1 hdr doesn't always exist when we change state to arc_anon before | |
1508 | * destroying a header, in which case reallocating to add the L1 hdr is | |
1509 | * pointless. | |
1510 | */ | |
1511 | if (HDR_HAS_L1HDR(hdr)) { | |
1512 | old_state = hdr->b_l1hdr.b_state; | |
1513 | refcnt = refcount_count(&hdr->b_l1hdr.b_refcnt); | |
1514 | datacnt = hdr->b_l1hdr.b_datacnt; | |
1515 | } else { | |
1516 | old_state = arc_l2c_only; | |
1517 | refcnt = 0; | |
1518 | datacnt = 0; | |
1519 | } | |
1520 | ||
1521 | ASSERT(MUTEX_HELD(hash_lock)); | |
1522 | ASSERT3P(new_state, !=, old_state); | |
1523 | ASSERT(refcnt == 0 || datacnt > 0); | |
1524 | ASSERT(!GHOST_STATE(new_state) || datacnt == 0); | |
1525 | ASSERT(old_state != arc_anon || datacnt <= 1); | |
1526 | ||
1527 | from_delta = to_delta = datacnt * hdr->b_size; | |
1528 | ||
1529 | /* | |
1530 | * If this buffer is evictable, transfer it from the | |
1531 | * old state list to the new state list. | |
1532 | */ | |
1533 | if (refcnt == 0) { | |
1534 | if (old_state != arc_anon && old_state != arc_l2c_only) { | |
1535 | uint64_t *size = &old_state->arcs_lsize[buftype]; | |
1536 | ||
1537 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
1538 | multilist_remove(&old_state->arcs_list[buftype], hdr); | |
1539 | ||
1540 | /* | |
1541 | * If prefetching out of the ghost cache, | |
1542 | * we will have a non-zero datacnt. | |
1543 | */ | |
1544 | if (GHOST_STATE(old_state) && datacnt == 0) { | |
1545 | /* ghost elements have a ghost size */ | |
1546 | ASSERT(hdr->b_l1hdr.b_buf == NULL); | |
1547 | from_delta = hdr->b_size; | |
1548 | } | |
1549 | ASSERT3U(*size, >=, from_delta); | |
1550 | atomic_add_64(size, -from_delta); | |
1551 | } | |
1552 | if (new_state != arc_anon && new_state != arc_l2c_only) { | |
1553 | uint64_t *size = &new_state->arcs_lsize[buftype]; | |
1554 | ||
1555 | /* | |
1556 | * An L1 header always exists here, since if we're | |
1557 | * moving to some L1-cached state (i.e. not l2c_only or | |
1558 | * anonymous), we realloc the header to add an L1hdr | |
1559 | * beforehand. | |
1560 | */ | |
1561 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
1562 | multilist_insert(&new_state->arcs_list[buftype], hdr); | |
1563 | ||
1564 | /* ghost elements have a ghost size */ | |
1565 | if (GHOST_STATE(new_state)) { | |
1566 | ASSERT0(datacnt); | |
1567 | ASSERT(hdr->b_l1hdr.b_buf == NULL); | |
1568 | to_delta = hdr->b_size; | |
1569 | } | |
1570 | atomic_add_64(size, to_delta); | |
1571 | } | |
1572 | } | |
1573 | ||
1574 | ASSERT(!BUF_EMPTY(hdr)); | |
1575 | if (new_state == arc_anon && HDR_IN_HASH_TABLE(hdr)) | |
1576 | buf_hash_remove(hdr); | |
1577 | ||
1578 | /* adjust state sizes (ignore arc_l2c_only) */ | |
1579 | ||
1580 | if (to_delta && new_state != arc_l2c_only) { | |
1581 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
1582 | if (GHOST_STATE(new_state)) { | |
1583 | ASSERT0(datacnt); | |
1584 | ||
1585 | /* | |
1586 | * We moving a header to a ghost state, we first | |
1587 | * remove all arc buffers. Thus, we'll have a | |
1588 | * datacnt of zero, and no arc buffer to use for | |
1589 | * the reference. As a result, we use the arc | |
1590 | * header pointer for the reference. | |
1591 | */ | |
1592 | (void) refcount_add_many(&new_state->arcs_size, | |
1593 | hdr->b_size, hdr); | |
1594 | } else { | |
1595 | arc_buf_t *buf; | |
1596 | ASSERT3U(datacnt, !=, 0); | |
1597 | ||
1598 | /* | |
1599 | * Each individual buffer holds a unique reference, | |
1600 | * thus we must remove each of these references one | |
1601 | * at a time. | |
1602 | */ | |
1603 | for (buf = hdr->b_l1hdr.b_buf; buf != NULL; | |
1604 | buf = buf->b_next) { | |
1605 | (void) refcount_add_many(&new_state->arcs_size, | |
1606 | hdr->b_size, buf); | |
1607 | } | |
1608 | } | |
1609 | } | |
1610 | ||
1611 | if (from_delta && old_state != arc_l2c_only) { | |
1612 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
1613 | if (GHOST_STATE(old_state)) { | |
1614 | /* | |
1615 | * When moving a header off of a ghost state, | |
1616 | * there's the possibility for datacnt to be | |
1617 | * non-zero. This is because we first add the | |
1618 | * arc buffer to the header prior to changing | |
1619 | * the header's state. Since we used the header | |
1620 | * for the reference when putting the header on | |
1621 | * the ghost state, we must balance that and use | |
1622 | * the header when removing off the ghost state | |
1623 | * (even though datacnt is non zero). | |
1624 | */ | |
1625 | ||
1626 | IMPLY(datacnt == 0, new_state == arc_anon || | |
1627 | new_state == arc_l2c_only); | |
1628 | ||
1629 | (void) refcount_remove_many(&old_state->arcs_size, | |
1630 | hdr->b_size, hdr); | |
1631 | } else { | |
1632 | arc_buf_t *buf; | |
1633 | ASSERT3U(datacnt, !=, 0); | |
1634 | ||
1635 | /* | |
1636 | * Each individual buffer holds a unique reference, | |
1637 | * thus we must remove each of these references one | |
1638 | * at a time. | |
1639 | */ | |
1640 | for (buf = hdr->b_l1hdr.b_buf; buf != NULL; | |
1641 | buf = buf->b_next) { | |
1642 | (void) refcount_remove_many( | |
1643 | &old_state->arcs_size, hdr->b_size, buf); | |
1644 | } | |
1645 | } | |
1646 | } | |
1647 | ||
1648 | if (HDR_HAS_L1HDR(hdr)) | |
1649 | hdr->b_l1hdr.b_state = new_state; | |
1650 | ||
1651 | /* | |
1652 | * L2 headers should never be on the L2 state list since they don't | |
1653 | * have L1 headers allocated. | |
1654 | */ | |
1655 | ASSERT(multilist_is_empty(&arc_l2c_only->arcs_list[ARC_BUFC_DATA]) && | |
1656 | multilist_is_empty(&arc_l2c_only->arcs_list[ARC_BUFC_METADATA])); | |
1657 | } | |
1658 | ||
1659 | void | |
1660 | arc_space_consume(uint64_t space, arc_space_type_t type) | |
1661 | { | |
1662 | ASSERT(type >= 0 && type < ARC_SPACE_NUMTYPES); | |
1663 | ||
1664 | switch (type) { | |
1665 | default: | |
1666 | break; | |
1667 | case ARC_SPACE_DATA: | |
1668 | ARCSTAT_INCR(arcstat_data_size, space); | |
1669 | break; | |
1670 | case ARC_SPACE_META: | |
1671 | ARCSTAT_INCR(arcstat_metadata_size, space); | |
1672 | break; | |
1673 | case ARC_SPACE_OTHER: | |
1674 | ARCSTAT_INCR(arcstat_other_size, space); | |
1675 | break; | |
1676 | case ARC_SPACE_HDRS: | |
1677 | ARCSTAT_INCR(arcstat_hdr_size, space); | |
1678 | break; | |
1679 | case ARC_SPACE_L2HDRS: | |
1680 | ARCSTAT_INCR(arcstat_l2_hdr_size, space); | |
1681 | break; | |
1682 | } | |
1683 | ||
1684 | if (type != ARC_SPACE_DATA) | |
1685 | ARCSTAT_INCR(arcstat_meta_used, space); | |
1686 | ||
1687 | atomic_add_64(&arc_size, space); | |
1688 | } | |
1689 | ||
1690 | void | |
1691 | arc_space_return(uint64_t space, arc_space_type_t type) | |
1692 | { | |
1693 | ASSERT(type >= 0 && type < ARC_SPACE_NUMTYPES); | |
1694 | ||
1695 | switch (type) { | |
1696 | default: | |
1697 | break; | |
1698 | case ARC_SPACE_DATA: | |
1699 | ARCSTAT_INCR(arcstat_data_size, -space); | |
1700 | break; | |
1701 | case ARC_SPACE_META: | |
1702 | ARCSTAT_INCR(arcstat_metadata_size, -space); | |
1703 | break; | |
1704 | case ARC_SPACE_OTHER: | |
1705 | ARCSTAT_INCR(arcstat_other_size, -space); | |
1706 | break; | |
1707 | case ARC_SPACE_HDRS: | |
1708 | ARCSTAT_INCR(arcstat_hdr_size, -space); | |
1709 | break; | |
1710 | case ARC_SPACE_L2HDRS: | |
1711 | ARCSTAT_INCR(arcstat_l2_hdr_size, -space); | |
1712 | break; | |
1713 | } | |
1714 | ||
1715 | if (type != ARC_SPACE_DATA) { | |
1716 | ASSERT(arc_meta_used >= space); | |
1717 | if (arc_meta_max < arc_meta_used) | |
1718 | arc_meta_max = arc_meta_used; | |
1719 | ARCSTAT_INCR(arcstat_meta_used, -space); | |
1720 | } | |
1721 | ||
1722 | ASSERT(arc_size >= space); | |
1723 | atomic_add_64(&arc_size, -space); | |
1724 | } | |
1725 | ||
1726 | arc_buf_t * | |
1727 | arc_buf_alloc(spa_t *spa, uint64_t size, void *tag, arc_buf_contents_t type) | |
1728 | { | |
1729 | arc_buf_hdr_t *hdr; | |
1730 | arc_buf_t *buf; | |
1731 | ||
1732 | VERIFY3U(size, <=, spa_maxblocksize(spa)); | |
1733 | hdr = kmem_cache_alloc(hdr_full_cache, KM_PUSHPAGE); | |
1734 | ASSERT(BUF_EMPTY(hdr)); | |
1735 | ASSERT3P(hdr->b_freeze_cksum, ==, NULL); | |
1736 | hdr->b_size = size; | |
1737 | hdr->b_spa = spa_load_guid(spa); | |
1738 | hdr->b_l1hdr.b_mru_hits = 0; | |
1739 | hdr->b_l1hdr.b_mru_ghost_hits = 0; | |
1740 | hdr->b_l1hdr.b_mfu_hits = 0; | |
1741 | hdr->b_l1hdr.b_mfu_ghost_hits = 0; | |
1742 | hdr->b_l1hdr.b_l2_hits = 0; | |
1743 | ||
1744 | buf = kmem_cache_alloc(buf_cache, KM_PUSHPAGE); | |
1745 | buf->b_hdr = hdr; | |
1746 | buf->b_data = NULL; | |
1747 | buf->b_efunc = NULL; | |
1748 | buf->b_private = NULL; | |
1749 | buf->b_next = NULL; | |
1750 | ||
1751 | hdr->b_flags = arc_bufc_to_flags(type); | |
1752 | hdr->b_flags |= ARC_FLAG_HAS_L1HDR; | |
1753 | ||
1754 | hdr->b_l1hdr.b_buf = buf; | |
1755 | hdr->b_l1hdr.b_state = arc_anon; | |
1756 | hdr->b_l1hdr.b_arc_access = 0; | |
1757 | hdr->b_l1hdr.b_datacnt = 1; | |
1758 | hdr->b_l1hdr.b_tmp_cdata = NULL; | |
1759 | ||
1760 | arc_get_data_buf(buf); | |
1761 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); | |
1762 | (void) refcount_add(&hdr->b_l1hdr.b_refcnt, tag); | |
1763 | ||
1764 | return (buf); | |
1765 | } | |
1766 | ||
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 * | |
1776 | arc_loan_buf(spa_t *spa, uint64_t size) | |
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 | ||
1794 | ASSERT(buf->b_data != NULL); | |
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); | |
1798 | ||
1799 | atomic_add_64(&arc_loaned_bytes, -hdr->b_size); | |
1800 | } | |
1801 | ||
1802 | /* Detach an arc_buf from a dbuf (tag) */ | |
1803 | void | |
1804 | arc_loan_inuse_buf(arc_buf_t *buf, void *tag) | |
1805 | { | |
1806 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
1807 | ||
1808 | ASSERT(buf->b_data != NULL); | |
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); | |
1812 | buf->b_efunc = NULL; | |
1813 | buf->b_private = NULL; | |
1814 | ||
1815 | atomic_add_64(&arc_loaned_bytes, hdr->b_size); | |
1816 | } | |
1817 | ||
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 | ||
1825 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
1826 | ASSERT(hdr->b_l1hdr.b_state != arc_anon); | |
1827 | ||
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; | |
1833 | buf->b_next = hdr->b_l1hdr.b_buf; | |
1834 | hdr->b_l1hdr.b_buf = buf; | |
1835 | arc_get_data_buf(buf); | |
1836 | bcopy(from->b_data, buf->b_data, size); | |
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 | */ | |
1844 | if (HDR_ISTYPE_DATA(hdr)) { | |
1845 | ARCSTAT_BUMP(arcstat_duplicate_buffers); | |
1846 | ARCSTAT_INCR(arcstat_duplicate_buffers_size, size); | |
1847 | } | |
1848 | hdr->b_l1hdr.b_datacnt += 1; | |
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 | /* | |
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. | |
1862 | */ | |
1863 | mutex_enter(&buf->b_evict_lock); | |
1864 | if (buf->b_data == NULL) { | |
1865 | mutex_exit(&buf->b_evict_lock); | |
1866 | return; | |
1867 | } | |
1868 | hash_lock = HDR_LOCK(buf->b_hdr); | |
1869 | mutex_enter(hash_lock); | |
1870 | hdr = buf->b_hdr; | |
1871 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
1872 | ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); | |
1873 | mutex_exit(&buf->b_evict_lock); | |
1874 | ||
1875 | ASSERT(hdr->b_l1hdr.b_state == arc_mru || | |
1876 | hdr->b_l1hdr.b_state == arc_mfu); | |
1877 | ||
1878 | add_reference(hdr, hash_lock, tag); | |
1879 | DTRACE_PROBE1(arc__hit, arc_buf_hdr_t *, hdr); | |
1880 | arc_access(hdr, hash_lock); | |
1881 | mutex_exit(hash_lock); | |
1882 | ARCSTAT_BUMP(arcstat_hits); | |
1883 | ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr), | |
1884 | demand, prefetch, !HDR_ISTYPE_METADATA(hdr), | |
1885 | data, metadata, hits); | |
1886 | } | |
1887 | ||
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 | ||
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 | |
1908 | arc_buf_data_free(arc_buf_t *buf, void (*free_func)(void *, size_t)) | |
1909 | { | |
1910 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
1911 | ||
1912 | if (HDR_L2_WRITING(hdr)) { | |
1913 | arc_buf_free_on_write(buf->b_data, hdr->b_size, free_func); | |
1914 | ARCSTAT_BUMP(arcstat_l2_free_on_write); | |
1915 | } else { | |
1916 | free_func(buf->b_data, hdr->b_size); | |
1917 | } | |
1918 | } | |
1919 | ||
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->b_l2hdr.b_compress == 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->b_l2hdr.b_compress == ZIO_COMPRESS_EMPTY) { | |
1962 | ASSERT3P(hdr->b_l1hdr.b_tmp_cdata, ==, NULL); | |
1963 | return; | |
1964 | } | |
1965 | ||
1966 | ASSERT(L2ARC_IS_VALID_COMPRESS(hdr->b_l2hdr.b_compress)); | |
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 | ||
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 | */ | |
1979 | static void | |
1980 | arc_buf_destroy(arc_buf_t *buf, boolean_t remove) | |
1981 | { | |
1982 | arc_buf_t **bufp; | |
1983 | ||
1984 | /* free up data associated with the buf */ | |
1985 | if (buf->b_data != NULL) { | |
1986 | arc_state_t *state = buf->b_hdr->b_l1hdr.b_state; | |
1987 | uint64_t size = buf->b_hdr->b_size; | |
1988 | arc_buf_contents_t type = arc_buf_type(buf->b_hdr); | |
1989 | ||
1990 | arc_cksum_verify(buf); | |
1991 | arc_buf_unwatch(buf); | |
1992 | ||
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); | |
2000 | } | |
2001 | ||
2002 | /* protected by hash lock, if in the hash table */ | |
2003 | if (multilist_link_active(&buf->b_hdr->b_l1hdr.b_arc_node)) { | |
2004 | uint64_t *cnt = &state->arcs_lsize[type]; | |
2005 | ||
2006 | ASSERT(refcount_is_zero( | |
2007 | &buf->b_hdr->b_l1hdr.b_refcnt)); | |
2008 | ASSERT(state != arc_anon && state != arc_l2c_only); | |
2009 | ||
2010 | ASSERT3U(*cnt, >=, size); | |
2011 | atomic_add_64(cnt, -size); | |
2012 | } | |
2013 | ||
2014 | (void) refcount_remove_many(&state->arcs_size, size, buf); | |
2015 | buf->b_data = NULL; | |
2016 | ||
2017 | /* | |
2018 | * If we're destroying a duplicate buffer make sure | |
2019 | * that the appropriate statistics are updated. | |
2020 | */ | |
2021 | if (buf->b_hdr->b_l1hdr.b_datacnt > 1 && | |
2022 | HDR_ISTYPE_DATA(buf->b_hdr)) { | |
2023 | ARCSTAT_BUMPDOWN(arcstat_duplicate_buffers); | |
2024 | ARCSTAT_INCR(arcstat_duplicate_buffers_size, -size); | |
2025 | } | |
2026 | ASSERT(buf->b_hdr->b_l1hdr.b_datacnt > 0); | |
2027 | buf->b_hdr->b_l1hdr.b_datacnt -= 1; | |
2028 | } | |
2029 | ||
2030 | /* only remove the buf if requested */ | |
2031 | if (!remove) | |
2032 | return; | |
2033 | ||
2034 | /* remove the buf from the hdr list */ | |
2035 | for (bufp = &buf->b_hdr->b_l1hdr.b_buf; *bufp != buf; | |
2036 | bufp = &(*bufp)->b_next) | |
2037 | continue; | |
2038 | *bufp = buf->b_next; | |
2039 | buf->b_next = NULL; | |
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 | ||
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 | ||
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 | ||
2099 | static void | |
2100 | arc_hdr_destroy(arc_buf_hdr_t *hdr) | |
2101 | { | |
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 | } | |
2108 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); | |
2109 | ASSERT(!HDR_IN_HASH_TABLE(hdr)); | |
2110 | ||
2111 | if (HDR_HAS_L2HDR(hdr)) { | |
2112 | l2arc_dev_t *dev = hdr->b_l2hdr.b_dev; | |
2113 | boolean_t buflist_held = MUTEX_HELD(&dev->l2ad_mtx); | |
2114 | ||
2115 | if (!buflist_held) | |
2116 | mutex_enter(&dev->l2ad_mtx); | |
2117 | ||
2118 | /* | |
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. | |
2126 | */ | |
2127 | if (HDR_HAS_L2HDR(hdr)) | |
2128 | arc_hdr_l2hdr_destroy(hdr); | |
2129 | ||
2130 | if (!buflist_held) | |
2131 | mutex_exit(&dev->l2ad_mtx); | |
2132 | } | |
2133 | ||
2134 | if (!BUF_EMPTY(hdr)) | |
2135 | buf_discard_identity(hdr); | |
2136 | ||
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 | ||
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) { | |
2147 | mutex_enter(&arc_user_evicts_lock); | |
2148 | mutex_enter(&buf->b_evict_lock); | |
2149 | ASSERT(buf->b_hdr != NULL); | |
2150 | arc_buf_destroy(hdr->b_l1hdr.b_buf, FALSE); | |
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); | |
2156 | cv_signal(&arc_user_evicts_cv); | |
2157 | mutex_exit(&arc_user_evicts_lock); | |
2158 | } else { | |
2159 | arc_buf_destroy(hdr->b_l1hdr.b_buf, TRUE); | |
2160 | } | |
2161 | } | |
2162 | } | |
2163 | ||
2164 | ASSERT3P(hdr->b_hash_next, ==, NULL); | |
2165 | if (HDR_HAS_L1HDR(hdr)) { | |
2166 | ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); | |
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 | } | |
2172 | } | |
2173 | ||
2174 | void | |
2175 | arc_buf_free(arc_buf_t *buf, void *tag) | |
2176 | { | |
2177 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
2178 | int hashed = hdr->b_l1hdr.b_state != arc_anon; | |
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); | |
2187 | hdr = buf->b_hdr; | |
2188 | ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); | |
2189 | ||
2190 | (void) remove_reference(hdr, hash_lock, tag); | |
2191 | if (hdr->b_l1hdr.b_datacnt > 1) { | |
2192 | arc_buf_destroy(buf, TRUE); | |
2193 | } else { | |
2194 | ASSERT(buf == hdr->b_l1hdr.b_buf); | |
2195 | ASSERT(buf->b_efunc == NULL); | |
2196 | hdr->b_flags |= ARC_FLAG_BUF_AVAILABLE; | |
2197 | } | |
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 | */ | |
2206 | mutex_enter(&arc_user_evicts_lock); | |
2207 | (void) remove_reference(hdr, NULL, tag); | |
2208 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); | |
2209 | destroy_hdr = !HDR_IO_IN_PROGRESS(hdr); | |
2210 | mutex_exit(&arc_user_evicts_lock); | |
2211 | if (destroy_hdr) | |
2212 | arc_hdr_destroy(hdr); | |
2213 | } else { | |
2214 | if (remove_reference(hdr, NULL, tag) > 0) | |
2215 | arc_buf_destroy(buf, TRUE); | |
2216 | else | |
2217 | arc_hdr_destroy(hdr); | |
2218 | } | |
2219 | } | |
2220 | ||
2221 | boolean_t | |
2222 | arc_buf_remove_ref(arc_buf_t *buf, void* tag) | |
2223 | { | |
2224 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
2225 | kmutex_t *hash_lock = HDR_LOCK(hdr); | |
2226 | boolean_t no_callback = (buf->b_efunc == NULL); | |
2227 | ||
2228 | if (hdr->b_l1hdr.b_state == arc_anon) { | |
2229 | ASSERT(hdr->b_l1hdr.b_datacnt == 1); | |
2230 | arc_buf_free(buf, tag); | |
2231 | return (no_callback); | |
2232 | } | |
2233 | ||
2234 | mutex_enter(hash_lock); | |
2235 | hdr = buf->b_hdr; | |
2236 | ASSERT(hdr->b_l1hdr.b_datacnt > 0); | |
2237 | ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); | |
2238 | ASSERT(hdr->b_l1hdr.b_state != arc_anon); | |
2239 | ASSERT(buf->b_data != NULL); | |
2240 | ||
2241 | (void) remove_reference(hdr, hash_lock, tag); | |
2242 | if (hdr->b_l1hdr.b_datacnt > 1) { | |
2243 | if (no_callback) | |
2244 | arc_buf_destroy(buf, TRUE); | |
2245 | } else if (no_callback) { | |
2246 | ASSERT(hdr->b_l1hdr.b_buf == buf && buf->b_next == NULL); | |
2247 | ASSERT(buf->b_efunc == NULL); | |
2248 | hdr->b_flags |= ARC_FLAG_BUF_AVAILABLE; | |
2249 | } | |
2250 | ASSERT(no_callback || hdr->b_l1hdr.b_datacnt > 1 || | |
2251 | refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); | |
2252 | mutex_exit(hash_lock); | |
2253 | return (no_callback); | |
2254 | } | |
2255 | ||
2256 | uint64_t | |
2257 | arc_buf_size(arc_buf_t *buf) | |
2258 | { | |
2259 | return (buf->b_hdr->b_size); | |
2260 | } | |
2261 | ||
2262 | /* | |
2263 | * Called from the DMU to determine if the current buffer should be | |
2264 | * evicted. In order to ensure proper locking, the eviction must be initiated | |
2265 | * from the DMU. Return true if the buffer is associated with user data and | |
2266 | * duplicate buffers still exist. | |
2267 | */ | |
2268 | boolean_t | |
2269 | arc_buf_eviction_needed(arc_buf_t *buf) | |
2270 | { | |
2271 | arc_buf_hdr_t *hdr; | |
2272 | boolean_t evict_needed = B_FALSE; | |
2273 | ||
2274 | if (zfs_disable_dup_eviction) | |
2275 | return (B_FALSE); | |
2276 | ||
2277 | mutex_enter(&buf->b_evict_lock); | |
2278 | hdr = buf->b_hdr; | |
2279 | if (hdr == NULL) { | |
2280 | /* | |
2281 | * We are in arc_do_user_evicts(); let that function | |
2282 | * perform the eviction. | |
2283 | */ | |
2284 | ASSERT(buf->b_data == NULL); | |
2285 | mutex_exit(&buf->b_evict_lock); | |
2286 | return (B_FALSE); | |
2287 | } else if (buf->b_data == NULL) { | |
2288 | /* | |
2289 | * We have already been added to the arc eviction list; | |
2290 | * recommend eviction. | |
2291 | */ | |
2292 | ASSERT3P(hdr, ==, &arc_eviction_hdr); | |
2293 | mutex_exit(&buf->b_evict_lock); | |
2294 | return (B_TRUE); | |
2295 | } | |
2296 | ||
2297 | if (hdr->b_l1hdr.b_datacnt > 1 && HDR_ISTYPE_DATA(hdr)) | |
2298 | evict_needed = B_TRUE; | |
2299 | ||
2300 | mutex_exit(&buf->b_evict_lock); | |
2301 | return (evict_needed); | |
2302 | } | |
2303 | ||
2304 | /* | |
2305 | * Evict the arc_buf_hdr that is provided as a parameter. The resultant | |
2306 | * state of the header is dependent on its state prior to entering this | |
2307 | * function. The following transitions are possible: | |
2308 | * | |
2309 | * - arc_mru -> arc_mru_ghost | |
2310 | * - arc_mfu -> arc_mfu_ghost | |
2311 | * - arc_mru_ghost -> arc_l2c_only | |
2312 | * - arc_mru_ghost -> deleted | |
2313 | * - arc_mfu_ghost -> arc_l2c_only | |
2314 | * - arc_mfu_ghost -> deleted | |
2315 | */ | |
2316 | static int64_t | |
2317 | arc_evict_hdr(arc_buf_hdr_t *hdr, kmutex_t *hash_lock) | |
2318 | { | |
2319 | arc_state_t *evicted_state, *state; | |
2320 | int64_t bytes_evicted = 0; | |
2321 | ||
2322 | ASSERT(MUTEX_HELD(hash_lock)); | |
2323 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
2324 | ||
2325 | state = hdr->b_l1hdr.b_state; | |
2326 | if (GHOST_STATE(state)) { | |
2327 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); | |
2328 | ASSERT(hdr->b_l1hdr.b_buf == NULL); | |
2329 | ||
2330 | /* | |
2331 | * l2arc_write_buffers() relies on a header's L1 portion | |
2332 | * (i.e. its b_tmp_cdata field) during its write phase. | |
2333 | * Thus, we cannot push a header onto the arc_l2c_only | |
2334 | * state (removing its L1 piece) until the header is | |
2335 | * done being written to the l2arc. | |
2336 | */ | |
2337 | if (HDR_HAS_L2HDR(hdr) && HDR_L2_WRITING(hdr)) { | |
2338 | ARCSTAT_BUMP(arcstat_evict_l2_skip); | |
2339 | return (bytes_evicted); | |
2340 | } | |
2341 | ||
2342 | ARCSTAT_BUMP(arcstat_deleted); | |
2343 | bytes_evicted += hdr->b_size; | |
2344 | ||
2345 | DTRACE_PROBE1(arc__delete, arc_buf_hdr_t *, hdr); | |
2346 | ||
2347 | if (HDR_HAS_L2HDR(hdr)) { | |
2348 | /* | |
2349 | * This buffer is cached on the 2nd Level ARC; | |
2350 | * don't destroy the header. | |
2351 | */ | |
2352 | arc_change_state(arc_l2c_only, hdr, hash_lock); | |
2353 | /* | |
2354 | * dropping from L1+L2 cached to L2-only, | |
2355 | * realloc to remove the L1 header. | |
2356 | */ | |
2357 | hdr = arc_hdr_realloc(hdr, hdr_full_cache, | |
2358 | hdr_l2only_cache); | |
2359 | } else { | |
2360 | arc_change_state(arc_anon, hdr, hash_lock); | |
2361 | arc_hdr_destroy(hdr); | |
2362 | } | |
2363 | return (bytes_evicted); | |
2364 | } | |
2365 | ||
2366 | ASSERT(state == arc_mru || state == arc_mfu); | |
2367 | evicted_state = (state == arc_mru) ? arc_mru_ghost : arc_mfu_ghost; | |
2368 | ||
2369 | /* prefetch buffers have a minimum lifespan */ | |
2370 | if (HDR_IO_IN_PROGRESS(hdr) || | |
2371 | ((hdr->b_flags & (ARC_FLAG_PREFETCH | ARC_FLAG_INDIRECT)) && | |
2372 | ddi_get_lbolt() - hdr->b_l1hdr.b_arc_access < | |
2373 | arc_min_prefetch_lifespan)) { | |
2374 | ARCSTAT_BUMP(arcstat_evict_skip); | |
2375 | return (bytes_evicted); | |
2376 | } | |
2377 | ||
2378 | ASSERT0(refcount_count(&hdr->b_l1hdr.b_refcnt)); | |
2379 | ASSERT3U(hdr->b_l1hdr.b_datacnt, >, 0); | |
2380 | while (hdr->b_l1hdr.b_buf) { | |
2381 | arc_buf_t *buf = hdr->b_l1hdr.b_buf; | |
2382 | if (!mutex_tryenter(&buf->b_evict_lock)) { | |
2383 | ARCSTAT_BUMP(arcstat_mutex_miss); | |
2384 | break; | |
2385 | } | |
2386 | if (buf->b_data != NULL) | |
2387 | bytes_evicted += hdr->b_size; | |
2388 | if (buf->b_efunc != NULL) { | |
2389 | mutex_enter(&arc_user_evicts_lock); | |
2390 | arc_buf_destroy(buf, FALSE); | |
2391 | hdr->b_l1hdr.b_buf = buf->b_next; | |
2392 | buf->b_hdr = &arc_eviction_hdr; | |
2393 | buf->b_next = arc_eviction_list; | |
2394 | arc_eviction_list = buf; | |
2395 | cv_signal(&arc_user_evicts_cv); | |
2396 | mutex_exit(&arc_user_evicts_lock); | |
2397 | mutex_exit(&buf->b_evict_lock); | |
2398 | } else { | |
2399 | mutex_exit(&buf->b_evict_lock); | |
2400 | arc_buf_destroy(buf, TRUE); | |
2401 | } | |
2402 | } | |
2403 | ||
2404 | if (HDR_HAS_L2HDR(hdr)) { | |
2405 | ARCSTAT_INCR(arcstat_evict_l2_cached, hdr->b_size); | |
2406 | } else { | |
2407 | if (l2arc_write_eligible(hdr->b_spa, hdr)) | |
2408 | ARCSTAT_INCR(arcstat_evict_l2_eligible, hdr->b_size); | |
2409 | else | |
2410 | ARCSTAT_INCR(arcstat_evict_l2_ineligible, hdr->b_size); | |
2411 | } | |
2412 | ||
2413 | if (hdr->b_l1hdr.b_datacnt == 0) { | |
2414 | arc_change_state(evicted_state, hdr, hash_lock); | |
2415 | ASSERT(HDR_IN_HASH_TABLE(hdr)); | |
2416 | hdr->b_flags |= ARC_FLAG_IN_HASH_TABLE; | |
2417 | hdr->b_flags &= ~ARC_FLAG_BUF_AVAILABLE; | |
2418 | DTRACE_PROBE1(arc__evict, arc_buf_hdr_t *, hdr); | |
2419 | } | |
2420 | ||
2421 | return (bytes_evicted); | |
2422 | } | |
2423 | ||
2424 | static uint64_t | |
2425 | arc_evict_state_impl(multilist_t *ml, int idx, arc_buf_hdr_t *marker, | |
2426 | uint64_t spa, int64_t bytes) | |
2427 | { | |
2428 | multilist_sublist_t *mls; | |
2429 | uint64_t bytes_evicted = 0; | |
2430 | arc_buf_hdr_t *hdr; | |
2431 | kmutex_t *hash_lock; | |
2432 | int evict_count = 0; | |
2433 | ||
2434 | ASSERT3P(marker, !=, NULL); | |
2435 | IMPLY(bytes < 0, bytes == ARC_EVICT_ALL); | |
2436 | ||
2437 | mls = multilist_sublist_lock(ml, idx); | |
2438 | ||
2439 | for (hdr = multilist_sublist_prev(mls, marker); hdr != NULL; | |
2440 | hdr = multilist_sublist_prev(mls, marker)) { | |
2441 | if ((bytes != ARC_EVICT_ALL && bytes_evicted >= bytes) || | |
2442 | (evict_count >= zfs_arc_evict_batch_limit)) | |
2443 | break; | |
2444 | ||
2445 | /* | |
2446 | * To keep our iteration location, move the marker | |
2447 | * forward. Since we're not holding hdr's hash lock, we | |
2448 | * must be very careful and not remove 'hdr' from the | |
2449 | * sublist. Otherwise, other consumers might mistake the | |
2450 | * 'hdr' as not being on a sublist when they call the | |
2451 | * multilist_link_active() function (they all rely on | |
2452 | * the hash lock protecting concurrent insertions and | |
2453 | * removals). multilist_sublist_move_forward() was | |
2454 | * specifically implemented to ensure this is the case | |
2455 | * (only 'marker' will be removed and re-inserted). | |
2456 | */ | |
2457 | multilist_sublist_move_forward(mls, marker); | |
2458 | ||
2459 | /* | |
2460 | * The only case where the b_spa field should ever be | |
2461 | * zero, is the marker headers inserted by | |
2462 | * arc_evict_state(). It's possible for multiple threads | |
2463 | * to be calling arc_evict_state() concurrently (e.g. | |
2464 | * dsl_pool_close() and zio_inject_fault()), so we must | |
2465 | * skip any markers we see from these other threads. | |
2466 | */ | |
2467 | if (hdr->b_spa == 0) | |
2468 | continue; | |
2469 | ||
2470 | /* we're only interested in evicting buffers of a certain spa */ | |
2471 | if (spa != 0 && hdr->b_spa != spa) { | |
2472 | ARCSTAT_BUMP(arcstat_evict_skip); | |
2473 | continue; | |
2474 | } | |
2475 | ||
2476 | hash_lock = HDR_LOCK(hdr); | |
2477 | ||
2478 | /* | |
2479 | * We aren't calling this function from any code path | |
2480 | * that would already be holding a hash lock, so we're | |
2481 | * asserting on this assumption to be defensive in case | |
2482 | * this ever changes. Without this check, it would be | |
2483 | * possible to incorrectly increment arcstat_mutex_miss | |
2484 | * below (e.g. if the code changed such that we called | |
2485 | * this function with a hash lock held). | |
2486 | */ | |
2487 | ASSERT(!MUTEX_HELD(hash_lock)); | |
2488 | ||
2489 | if (mutex_tryenter(hash_lock)) { | |
2490 | uint64_t evicted = arc_evict_hdr(hdr, hash_lock); | |
2491 | mutex_exit(hash_lock); | |
2492 | ||
2493 | bytes_evicted += evicted; | |
2494 | ||
2495 | /* | |
2496 | * If evicted is zero, arc_evict_hdr() must have | |
2497 | * decided to skip this header, don't increment | |
2498 | * evict_count in this case. | |
2499 | */ | |
2500 | if (evicted != 0) | |
2501 | evict_count++; | |
2502 | ||
2503 | /* | |
2504 | * If arc_size isn't overflowing, signal any | |
2505 | * threads that might happen to be waiting. | |
2506 | * | |
2507 | * For each header evicted, we wake up a single | |
2508 | * thread. If we used cv_broadcast, we could | |
2509 | * wake up "too many" threads causing arc_size | |
2510 | * to significantly overflow arc_c; since | |
2511 | * arc_get_data_buf() doesn't check for overflow | |
2512 | * when it's woken up (it doesn't because it's | |
2513 | * possible for the ARC to be overflowing while | |
2514 | * full of un-evictable buffers, and the | |
2515 | * function should proceed in this case). | |
2516 | * | |
2517 | * If threads are left sleeping, due to not | |
2518 | * using cv_broadcast, they will be woken up | |
2519 | * just before arc_reclaim_thread() sleeps. | |
2520 | */ | |
2521 | mutex_enter(&arc_reclaim_lock); | |
2522 | if (!arc_is_overflowing()) | |
2523 | cv_signal(&arc_reclaim_waiters_cv); | |
2524 | mutex_exit(&arc_reclaim_lock); | |
2525 | } else { | |
2526 | ARCSTAT_BUMP(arcstat_mutex_miss); | |
2527 | } | |
2528 | } | |
2529 | ||
2530 | multilist_sublist_unlock(mls); | |
2531 | ||
2532 | return (bytes_evicted); | |
2533 | } | |
2534 | ||
2535 | /* | |
2536 | * Evict buffers from the given arc state, until we've removed the | |
2537 | * specified number of bytes. Move the removed buffers to the | |
2538 | * appropriate evict state. | |
2539 | * | |
2540 | * This function makes a "best effort". It skips over any buffers | |
2541 | * it can't get a hash_lock on, and so, may not catch all candidates. | |
2542 | * It may also return without evicting as much space as requested. | |
2543 | * | |
2544 | * If bytes is specified using the special value ARC_EVICT_ALL, this | |
2545 | * will evict all available (i.e. unlocked and evictable) buffers from | |
2546 | * the given arc state; which is used by arc_flush(). | |
2547 | */ | |
2548 | static uint64_t | |
2549 | arc_evict_state(arc_state_t *state, uint64_t spa, int64_t bytes, | |
2550 | arc_buf_contents_t type) | |
2551 | { | |
2552 | uint64_t total_evicted = 0; | |
2553 | multilist_t *ml = &state->arcs_list[type]; | |
2554 | int num_sublists; | |
2555 | arc_buf_hdr_t **markers; | |
2556 | int i; | |
2557 | ||
2558 | IMPLY(bytes < 0, bytes == ARC_EVICT_ALL); | |
2559 | ||
2560 | num_sublists = multilist_get_num_sublists(ml); | |
2561 | ||
2562 | /* | |
2563 | * If we've tried to evict from each sublist, made some | |
2564 | * progress, but still have not hit the target number of bytes | |
2565 | * to evict, we want to keep trying. The markers allow us to | |
2566 | * pick up where we left off for each individual sublist, rather | |
2567 | * than starting from the tail each time. | |
2568 | */ | |
2569 | markers = kmem_zalloc(sizeof (*markers) * num_sublists, KM_SLEEP); | |
2570 | for (i = 0; i < num_sublists; i++) { | |
2571 | multilist_sublist_t *mls; | |
2572 | ||
2573 | markers[i] = kmem_cache_alloc(hdr_full_cache, KM_SLEEP); | |
2574 | ||
2575 | /* | |
2576 | * A b_spa of 0 is used to indicate that this header is | |
2577 | * a marker. This fact is used in arc_adjust_type() and | |
2578 | * arc_evict_state_impl(). | |
2579 | */ | |
2580 | markers[i]->b_spa = 0; | |
2581 | ||
2582 | mls = multilist_sublist_lock(ml, i); | |
2583 | multilist_sublist_insert_tail(mls, markers[i]); | |
2584 | multilist_sublist_unlock(mls); | |
2585 | } | |
2586 | ||
2587 | /* | |
2588 | * While we haven't hit our target number of bytes to evict, or | |
2589 | * we're evicting all available buffers. | |
2590 | */ | |
2591 | while (total_evicted < bytes || bytes == ARC_EVICT_ALL) { | |
2592 | /* | |
2593 | * Start eviction using a randomly selected sublist, | |
2594 | * this is to try and evenly balance eviction across all | |
2595 | * sublists. Always starting at the same sublist | |
2596 | * (e.g. index 0) would cause evictions to favor certain | |
2597 | * sublists over others. | |
2598 | */ | |
2599 | int sublist_idx = multilist_get_random_index(ml); | |
2600 | uint64_t scan_evicted = 0; | |
2601 | ||
2602 | for (i = 0; i < num_sublists; i++) { | |
2603 | uint64_t bytes_remaining; | |
2604 | uint64_t bytes_evicted; | |
2605 | ||
2606 | if (bytes == ARC_EVICT_ALL) | |
2607 | bytes_remaining = ARC_EVICT_ALL; | |
2608 | else if (total_evicted < bytes) | |
2609 | bytes_remaining = bytes - total_evicted; | |
2610 | else | |
2611 | break; | |
2612 | ||
2613 | bytes_evicted = arc_evict_state_impl(ml, sublist_idx, | |
2614 | markers[sublist_idx], spa, bytes_remaining); | |
2615 | ||
2616 | scan_evicted += bytes_evicted; | |
2617 | total_evicted += bytes_evicted; | |
2618 | ||
2619 | /* we've reached the end, wrap to the beginning */ | |
2620 | if (++sublist_idx >= num_sublists) | |
2621 | sublist_idx = 0; | |
2622 | } | |
2623 | ||
2624 | /* | |
2625 | * If we didn't evict anything during this scan, we have | |
2626 | * no reason to believe we'll evict more during another | |
2627 | * scan, so break the loop. | |
2628 | */ | |
2629 | if (scan_evicted == 0) { | |
2630 | /* This isn't possible, let's make that obvious */ | |
2631 | ASSERT3S(bytes, !=, 0); | |
2632 | ||
2633 | /* | |
2634 | * When bytes is ARC_EVICT_ALL, the only way to | |
2635 | * break the loop is when scan_evicted is zero. | |
2636 | * In that case, we actually have evicted enough, | |
2637 | * so we don't want to increment the kstat. | |
2638 | */ | |
2639 | if (bytes != ARC_EVICT_ALL) { | |
2640 | ASSERT3S(total_evicted, <, bytes); | |
2641 | ARCSTAT_BUMP(arcstat_evict_not_enough); | |
2642 | } | |
2643 | ||
2644 | break; | |
2645 | } | |
2646 | } | |
2647 | ||
2648 | for (i = 0; i < num_sublists; i++) { | |
2649 | multilist_sublist_t *mls = multilist_sublist_lock(ml, i); | |
2650 | multilist_sublist_remove(mls, markers[i]); | |
2651 | multilist_sublist_unlock(mls); | |
2652 | ||
2653 | kmem_cache_free(hdr_full_cache, markers[i]); | |
2654 | } | |
2655 | kmem_free(markers, sizeof (*markers) * num_sublists); | |
2656 | ||
2657 | return (total_evicted); | |
2658 | } | |
2659 | ||
2660 | /* | |
2661 | * Flush all "evictable" data of the given type from the arc state | |
2662 | * specified. This will not evict any "active" buffers (i.e. referenced). | |
2663 | * | |
2664 | * When 'retry' is set to FALSE, the function will make a single pass | |
2665 | * over the state and evict any buffers that it can. Since it doesn't | |
2666 | * continually retry the eviction, it might end up leaving some buffers | |
2667 | * in the ARC due to lock misses. | |
2668 | * | |
2669 | * When 'retry' is set to TRUE, the function will continually retry the | |
2670 | * eviction until *all* evictable buffers have been removed from the | |
2671 | * state. As a result, if concurrent insertions into the state are | |
2672 | * allowed (e.g. if the ARC isn't shutting down), this function might | |
2673 | * wind up in an infinite loop, continually trying to evict buffers. | |
2674 | */ | |
2675 | static uint64_t | |
2676 | arc_flush_state(arc_state_t *state, uint64_t spa, arc_buf_contents_t type, | |
2677 | boolean_t retry) | |
2678 | { | |
2679 | uint64_t evicted = 0; | |
2680 | ||
2681 | while (state->arcs_lsize[type] != 0) { | |
2682 | evicted += arc_evict_state(state, spa, ARC_EVICT_ALL, type); | |
2683 | ||
2684 | if (!retry) | |
2685 | break; | |
2686 | } | |
2687 | ||
2688 | return (evicted); | |
2689 | } | |
2690 | ||
2691 | /* | |
2692 | * Helper function for arc_prune_async() it is responsible for safely | |
2693 | * handling the execution of a registered arc_prune_func_t. | |
2694 | */ | |
2695 | static void | |
2696 | arc_prune_task(void *ptr) | |
2697 | { | |
2698 | arc_prune_t *ap = (arc_prune_t *)ptr; | |
2699 | arc_prune_func_t *func = ap->p_pfunc; | |
2700 | ||
2701 | if (func != NULL) | |
2702 | func(ap->p_adjust, ap->p_private); | |
2703 | ||
2704 | refcount_remove(&ap->p_refcnt, func); | |
2705 | } | |
2706 | ||
2707 | /* | |
2708 | * Notify registered consumers they must drop holds on a portion of the ARC | |
2709 | * buffered they reference. This provides a mechanism to ensure the ARC can | |
2710 | * honor the arc_meta_limit and reclaim otherwise pinned ARC buffers. This | |
2711 | * is analogous to dnlc_reduce_cache() but more generic. | |
2712 | * | |
2713 | * This operation is performed asynchronously so it may be safely called | |
2714 | * in the context of the arc_reclaim_thread(). A reference is taken here | |
2715 | * for each registered arc_prune_t and the arc_prune_task() is responsible | |
2716 | * for releasing it once the registered arc_prune_func_t has completed. | |
2717 | */ | |
2718 | static void | |
2719 | arc_prune_async(int64_t adjust) | |
2720 | { | |
2721 | arc_prune_t *ap; | |
2722 | ||
2723 | mutex_enter(&arc_prune_mtx); | |
2724 | for (ap = list_head(&arc_prune_list); ap != NULL; | |
2725 | ap = list_next(&arc_prune_list, ap)) { | |
2726 | ||
2727 | if (refcount_count(&ap->p_refcnt) >= 2) | |
2728 | continue; | |
2729 | ||
2730 | refcount_add(&ap->p_refcnt, ap->p_pfunc); | |
2731 | ap->p_adjust = adjust; | |
2732 | taskq_dispatch(arc_prune_taskq, arc_prune_task, ap, TQ_SLEEP); | |
2733 | ARCSTAT_BUMP(arcstat_prune); | |
2734 | } | |
2735 | mutex_exit(&arc_prune_mtx); | |
2736 | } | |
2737 | ||
2738 | /* | |
2739 | * Evict the specified number of bytes from the state specified, | |
2740 | * restricting eviction to the spa and type given. This function | |
2741 | * prevents us from trying to evict more from a state's list than | |
2742 | * is "evictable", and to skip evicting altogether when passed a | |
2743 | * negative value for "bytes". In contrast, arc_evict_state() will | |
2744 | * evict everything it can, when passed a negative value for "bytes". | |
2745 | */ | |
2746 | static uint64_t | |
2747 | arc_adjust_impl(arc_state_t *state, uint64_t spa, int64_t bytes, | |
2748 | arc_buf_contents_t type) | |
2749 | { | |
2750 | int64_t delta; | |
2751 | ||
2752 | if (bytes > 0 && state->arcs_lsize[type] > 0) { | |
2753 | delta = MIN(state->arcs_lsize[type], bytes); | |
2754 | return (arc_evict_state(state, spa, delta, type)); | |
2755 | } | |
2756 | ||
2757 | return (0); | |
2758 | } | |
2759 | ||
2760 | /* | |
2761 | * The goal of this function is to evict enough meta data buffers from the | |
2762 | * ARC in order to enforce the arc_meta_limit. Achieving this is slightly | |
2763 | * more complicated than it appears because it is common for data buffers | |
2764 | * to have holds on meta data buffers. In addition, dnode meta data buffers | |
2765 | * will be held by the dnodes in the block preventing them from being freed. | |
2766 | * This means we can't simply traverse the ARC and expect to always find | |
2767 | * enough unheld meta data buffer to release. | |
2768 | * | |
2769 | * Therefore, this function has been updated to make alternating passes | |
2770 | * over the ARC releasing data buffers and then newly unheld meta data | |
2771 | * buffers. This ensures forward progress is maintained and arc_meta_used | |
2772 | * will decrease. Normally this is sufficient, but if required the ARC | |
2773 | * will call the registered prune callbacks causing dentry and inodes to | |
2774 | * be dropped from the VFS cache. This will make dnode meta data buffers | |
2775 | * available for reclaim. | |
2776 | */ | |
2777 | static uint64_t | |
2778 | arc_adjust_meta_balanced(void) | |
2779 | { | |
2780 | int64_t adjustmnt, delta, prune = 0; | |
2781 | uint64_t total_evicted = 0; | |
2782 | arc_buf_contents_t type = ARC_BUFC_DATA; | |
2783 | int restarts = MAX(zfs_arc_meta_adjust_restarts, 0); | |
2784 | ||
2785 | restart: | |
2786 | /* | |
2787 | * This slightly differs than the way we evict from the mru in | |
2788 | * arc_adjust because we don't have a "target" value (i.e. no | |
2789 | * "meta" arc_p). As a result, I think we can completely | |
2790 | * cannibalize the metadata in the MRU before we evict the | |
2791 | * metadata from the MFU. I think we probably need to implement a | |
2792 | * "metadata arc_p" value to do this properly. | |
2793 | */ | |
2794 | adjustmnt = arc_meta_used - arc_meta_limit; | |
2795 | ||
2796 | if (adjustmnt > 0 && arc_mru->arcs_lsize[type] > 0) { | |
2797 | delta = MIN(arc_mru->arcs_lsize[type], adjustmnt); | |
2798 | total_evicted += arc_adjust_impl(arc_mru, 0, delta, type); | |
2799 | adjustmnt -= delta; | |
2800 | } | |
2801 | ||
2802 | /* | |
2803 | * We can't afford to recalculate adjustmnt here. If we do, | |
2804 | * new metadata buffers can sneak into the MRU or ANON lists, | |
2805 | * thus penalize the MFU metadata. Although the fudge factor is | |
2806 | * small, it has been empirically shown to be significant for | |
2807 | * certain workloads (e.g. creating many empty directories). As | |
2808 | * such, we use the original calculation for adjustmnt, and | |
2809 | * simply decrement the amount of data evicted from the MRU. | |
2810 | */ | |
2811 | ||
2812 | if (adjustmnt > 0 && arc_mfu->arcs_lsize[type] > 0) { | |
2813 | delta = MIN(arc_mfu->arcs_lsize[type], adjustmnt); | |
2814 | total_evicted += arc_adjust_impl(arc_mfu, 0, delta, type); | |
2815 | } | |
2816 | ||
2817 | adjustmnt = arc_meta_used - arc_meta_limit; | |
2818 | ||
2819 | if (adjustmnt > 0 && arc_mru_ghost->arcs_lsize[type] > 0) { | |
2820 | delta = MIN(adjustmnt, | |
2821 | arc_mru_ghost->arcs_lsize[type]); | |
2822 | total_evicted += arc_adjust_impl(arc_mru_ghost, 0, delta, type); | |
2823 | adjustmnt -= delta; | |
2824 | } | |
2825 | ||
2826 | if (adjustmnt > 0 && arc_mfu_ghost->arcs_lsize[type] > 0) { | |
2827 | delta = MIN(adjustmnt, | |
2828 | arc_mfu_ghost->arcs_lsize[type]); | |
2829 | total_evicted += arc_adjust_impl(arc_mfu_ghost, 0, delta, type); | |
2830 | } | |
2831 | ||
2832 | /* | |
2833 | * If after attempting to make the requested adjustment to the ARC | |
2834 | * the meta limit is still being exceeded then request that the | |
2835 | * higher layers drop some cached objects which have holds on ARC | |
2836 | * meta buffers. Requests to the upper layers will be made with | |
2837 | * increasingly large scan sizes until the ARC is below the limit. | |
2838 | */ | |
2839 | if (arc_meta_used > arc_meta_limit) { | |
2840 | if (type == ARC_BUFC_DATA) { | |
2841 | type = ARC_BUFC_METADATA; | |
2842 | } else { | |
2843 | type = ARC_BUFC_DATA; | |
2844 | ||
2845 | if (zfs_arc_meta_prune) { | |
2846 | prune += zfs_arc_meta_prune; | |
2847 | arc_prune_async(prune); | |
2848 | } | |
2849 | } | |
2850 | ||
2851 | if (restarts > 0) { | |
2852 | restarts--; | |
2853 | goto restart; | |
2854 | } | |
2855 | } | |
2856 | return (total_evicted); | |
2857 | } | |
2858 | ||
2859 | /* | |
2860 | * Evict metadata buffers from the cache, such that arc_meta_used is | |
2861 | * capped by the arc_meta_limit tunable. | |
2862 | */ | |
2863 | static uint64_t | |
2864 | arc_adjust_meta_only(void) | |
2865 | { | |
2866 | uint64_t total_evicted = 0; | |
2867 | int64_t target; | |
2868 | ||
2869 | /* | |
2870 | * If we're over the meta limit, we want to evict enough | |
2871 | * metadata to get back under the meta limit. We don't want to | |
2872 | * evict so much that we drop the MRU below arc_p, though. If | |
2873 | * we're over the meta limit more than we're over arc_p, we | |
2874 | * evict some from the MRU here, and some from the MFU below. | |
2875 | */ | |
2876 | target = MIN((int64_t)(arc_meta_used - arc_meta_limit), | |
2877 | (int64_t)(refcount_count(&arc_anon->arcs_size) + | |
2878 | refcount_count(&arc_mru->arcs_size) - arc_p)); | |
2879 | ||
2880 | total_evicted += arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_METADATA); | |
2881 | ||
2882 | /* | |
2883 | * Similar to the above, we want to evict enough bytes to get us | |
2884 | * below the meta limit, but not so much as to drop us below the | |
2885 | * space alloted to the MFU (which is defined as arc_c - arc_p). | |
2886 | */ | |
2887 | target = MIN((int64_t)(arc_meta_used - arc_meta_limit), | |
2888 | (int64_t)(refcount_count(&arc_mfu->arcs_size) - (arc_c - arc_p))); | |
2889 | ||
2890 | total_evicted += arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_METADATA); | |
2891 | ||
2892 | return (total_evicted); | |
2893 | } | |
2894 | ||
2895 | static uint64_t | |
2896 | arc_adjust_meta(void) | |
2897 | { | |
2898 | if (zfs_arc_meta_strategy == ARC_STRATEGY_META_ONLY) | |
2899 | return (arc_adjust_meta_only()); | |
2900 | else | |
2901 | return (arc_adjust_meta_balanced()); | |
2902 | } | |
2903 | ||
2904 | /* | |
2905 | * Return the type of the oldest buffer in the given arc state | |
2906 | * | |
2907 | * This function will select a random sublist of type ARC_BUFC_DATA and | |
2908 | * a random sublist of type ARC_BUFC_METADATA. The tail of each sublist | |
2909 | * is compared, and the type which contains the "older" buffer will be | |
2910 | * returned. | |
2911 | */ | |
2912 | static arc_buf_contents_t | |
2913 | arc_adjust_type(arc_state_t *state) | |
2914 | { | |
2915 | multilist_t *data_ml = &state->arcs_list[ARC_BUFC_DATA]; | |
2916 | multilist_t *meta_ml = &state->arcs_list[ARC_BUFC_METADATA]; | |
2917 | int data_idx = multilist_get_random_index(data_ml); | |
2918 | int meta_idx = multilist_get_random_index(meta_ml); | |
2919 | multilist_sublist_t *data_mls; | |
2920 | multilist_sublist_t *meta_mls; | |
2921 | arc_buf_contents_t type; | |
2922 | arc_buf_hdr_t *data_hdr; | |
2923 | arc_buf_hdr_t *meta_hdr; | |
2924 | ||
2925 | /* | |
2926 | * We keep the sublist lock until we're finished, to prevent | |
2927 | * the headers from being destroyed via arc_evict_state(). | |
2928 | */ | |
2929 | data_mls = multilist_sublist_lock(data_ml, data_idx); | |
2930 | meta_mls = multilist_sublist_lock(meta_ml, meta_idx); | |
2931 | ||
2932 | /* | |
2933 | * These two loops are to ensure we skip any markers that | |
2934 | * might be at the tail of the lists due to arc_evict_state(). | |
2935 | */ | |
2936 | ||
2937 | for (data_hdr = multilist_sublist_tail(data_mls); data_hdr != NULL; | |
2938 | data_hdr = multilist_sublist_prev(data_mls, data_hdr)) { | |
2939 | if (data_hdr->b_spa != 0) | |
2940 | break; | |
2941 | } | |
2942 | ||
2943 | for (meta_hdr = multilist_sublist_tail(meta_mls); meta_hdr != NULL; | |
2944 | meta_hdr = multilist_sublist_prev(meta_mls, meta_hdr)) { | |
2945 | if (meta_hdr->b_spa != 0) | |
2946 | break; | |
2947 | } | |
2948 | ||
2949 | if (data_hdr == NULL && meta_hdr == NULL) { | |
2950 | type = ARC_BUFC_DATA; | |
2951 | } else if (data_hdr == NULL) { | |
2952 | ASSERT3P(meta_hdr, !=, NULL); | |
2953 | type = ARC_BUFC_METADATA; | |
2954 | } else if (meta_hdr == NULL) { | |
2955 | ASSERT3P(data_hdr, !=, NULL); | |
2956 | type = ARC_BUFC_DATA; | |
2957 | } else { | |
2958 | ASSERT3P(data_hdr, !=, NULL); | |
2959 | ASSERT3P(meta_hdr, !=, NULL); | |
2960 | ||
2961 | /* The headers can't be on the sublist without an L1 header */ | |
2962 | ASSERT(HDR_HAS_L1HDR(data_hdr)); | |
2963 | ASSERT(HDR_HAS_L1HDR(meta_hdr)); | |
2964 | ||
2965 | if (data_hdr->b_l1hdr.b_arc_access < | |
2966 | meta_hdr->b_l1hdr.b_arc_access) { | |
2967 | type = ARC_BUFC_DATA; | |
2968 | } else { | |
2969 | type = ARC_BUFC_METADATA; | |
2970 | } | |
2971 | } | |
2972 | ||
2973 | multilist_sublist_unlock(meta_mls); | |
2974 | multilist_sublist_unlock(data_mls); | |
2975 | ||
2976 | return (type); | |
2977 | } | |
2978 | ||
2979 | /* | |
2980 | * Evict buffers from the cache, such that arc_size is capped by arc_c. | |
2981 | */ | |
2982 | static uint64_t | |
2983 | arc_adjust(void) | |
2984 | { | |
2985 | uint64_t total_evicted = 0; | |
2986 | uint64_t bytes; | |
2987 | int64_t target; | |
2988 | ||
2989 | /* | |
2990 | * If we're over arc_meta_limit, we want to correct that before | |
2991 | * potentially evicting data buffers below. | |
2992 | */ | |
2993 | total_evicted += arc_adjust_meta(); | |
2994 | ||
2995 | /* | |
2996 | * Adjust MRU size | |
2997 | * | |
2998 | * If we're over the target cache size, we want to evict enough | |
2999 | * from the list to get back to our target size. We don't want | |
3000 | * to evict too much from the MRU, such that it drops below | |
3001 | * arc_p. So, if we're over our target cache size more than | |
3002 | * the MRU is over arc_p, we'll evict enough to get back to | |
3003 | * arc_p here, and then evict more from the MFU below. | |
3004 | */ | |
3005 | target = MIN((int64_t)(arc_size - arc_c), | |
3006 | (int64_t)(refcount_count(&arc_anon->arcs_size) + | |
3007 | refcount_count(&arc_mru->arcs_size) + arc_meta_used - arc_p)); | |
3008 | ||
3009 | /* | |
3010 | * If we're below arc_meta_min, always prefer to evict data. | |
3011 | * Otherwise, try to satisfy the requested number of bytes to | |
3012 | * evict from the type which contains older buffers; in an | |
3013 | * effort to keep newer buffers in the cache regardless of their | |
3014 | * type. If we cannot satisfy the number of bytes from this | |
3015 | * type, spill over into the next type. | |
3016 | */ | |
3017 | if (arc_adjust_type(arc_mru) == ARC_BUFC_METADATA && | |
3018 | arc_meta_used > arc_meta_min) { | |
3019 | bytes = arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_METADATA); | |
3020 | total_evicted += bytes; | |
3021 | ||
3022 | /* | |
3023 | * If we couldn't evict our target number of bytes from | |
3024 | * metadata, we try to get the rest from data. | |
3025 | */ | |
3026 | target -= bytes; | |
3027 | ||
3028 | total_evicted += | |
3029 | arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_DATA); | |
3030 | } else { | |
3031 | bytes = arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_DATA); | |
3032 | total_evicted += bytes; | |
3033 | ||
3034 | /* | |
3035 | * If we couldn't evict our target number of bytes from | |
3036 | * data, we try to get the rest from metadata. | |
3037 | */ | |
3038 | target -= bytes; | |
3039 | ||
3040 | total_evicted += | |
3041 | arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_METADATA); | |
3042 | } | |
3043 | ||
3044 | /* | |
3045 | * Adjust MFU size | |
3046 | * | |
3047 | * Now that we've tried to evict enough from the MRU to get its | |
3048 | * size back to arc_p, if we're still above the target cache | |
3049 | * size, we evict the rest from the MFU. | |
3050 | */ | |
3051 | target = arc_size - arc_c; | |
3052 | ||
3053 | if (arc_adjust_type(arc_mfu) == ARC_BUFC_METADATA && | |
3054 | arc_meta_used > arc_meta_min) { | |
3055 | bytes = arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_METADATA); | |
3056 | total_evicted += bytes; | |
3057 | ||
3058 | /* | |
3059 | * If we couldn't evict our target number of bytes from | |
3060 | * metadata, we try to get the rest from data. | |
3061 | */ | |
3062 | target -= bytes; | |
3063 | ||
3064 | total_evicted += | |
3065 | arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_DATA); | |
3066 | } else { | |
3067 | bytes = arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_DATA); | |
3068 | total_evicted += bytes; | |
3069 | ||
3070 | /* | |
3071 | * If we couldn't evict our target number of bytes from | |
3072 | * data, we try to get the rest from data. | |
3073 | */ | |
3074 | target -= bytes; | |
3075 | ||
3076 | total_evicted += | |
3077 | arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_METADATA); | |
3078 | } | |
3079 | ||
3080 | /* | |
3081 | * Adjust ghost lists | |
3082 | * | |
3083 | * In addition to the above, the ARC also defines target values | |
3084 | * for the ghost lists. The sum of the mru list and mru ghost | |
3085 | * list should never exceed the target size of the cache, and | |
3086 | * the sum of the mru list, mfu list, mru ghost list, and mfu | |
3087 | * ghost list should never exceed twice the target size of the | |
3088 | * cache. The following logic enforces these limits on the ghost | |
3089 | * caches, and evicts from them as needed. | |
3090 | */ | |
3091 | target = refcount_count(&arc_mru->arcs_size) + | |
3092 | refcount_count(&arc_mru_ghost->arcs_size) - arc_c; | |
3093 | ||
3094 | bytes = arc_adjust_impl(arc_mru_ghost, 0, target, ARC_BUFC_DATA); | |
3095 | total_evicted += bytes; | |
3096 | ||
3097 | target -= bytes; | |
3098 | ||
3099 | total_evicted += | |
3100 | arc_adjust_impl(arc_mru_ghost, 0, target, ARC_BUFC_METADATA); | |
3101 | ||
3102 | /* | |
3103 | * We assume the sum of the mru list and mfu list is less than | |
3104 | * or equal to arc_c (we enforced this above), which means we | |
3105 | * can use the simpler of the two equations below: | |
3106 | * | |
3107 | * mru + mfu + mru ghost + mfu ghost <= 2 * arc_c | |
3108 | * mru ghost + mfu ghost <= arc_c | |
3109 | */ | |
3110 | target = refcount_count(&arc_mru_ghost->arcs_size) + | |
3111 | refcount_count(&arc_mfu_ghost->arcs_size) - arc_c; | |
3112 | ||
3113 | bytes = arc_adjust_impl(arc_mfu_ghost, 0, target, ARC_BUFC_DATA); | |
3114 | total_evicted += bytes; | |
3115 | ||
3116 | target -= bytes; | |
3117 | ||
3118 | total_evicted += | |
3119 | arc_adjust_impl(arc_mfu_ghost, 0, target, ARC_BUFC_METADATA); | |
3120 | ||
3121 | return (total_evicted); | |
3122 | } | |
3123 | ||
3124 | static void | |
3125 | arc_do_user_evicts(void) | |
3126 | { | |
3127 | mutex_enter(&arc_user_evicts_lock); | |
3128 | while (arc_eviction_list != NULL) { | |
3129 | arc_buf_t *buf = arc_eviction_list; | |
3130 | arc_eviction_list = buf->b_next; | |
3131 | mutex_enter(&buf->b_evict_lock); | |
3132 | buf->b_hdr = NULL; | |
3133 | mutex_exit(&buf->b_evict_lock); | |
3134 | mutex_exit(&arc_user_evicts_lock); | |
3135 | ||
3136 | if (buf->b_efunc != NULL) | |
3137 | VERIFY0(buf->b_efunc(buf->b_private)); | |
3138 | ||
3139 | buf->b_efunc = NULL; | |
3140 | buf->b_private = NULL; | |
3141 | kmem_cache_free(buf_cache, buf); | |
3142 | mutex_enter(&arc_user_evicts_lock); | |
3143 | } | |
3144 | mutex_exit(&arc_user_evicts_lock); | |
3145 | } | |
3146 | ||
3147 | void | |
3148 | arc_flush(spa_t *spa, boolean_t retry) | |
3149 | { | |
3150 | uint64_t guid = 0; | |
3151 | ||
3152 | /* | |
3153 | * If retry is TRUE, a spa must not be specified since we have | |
3154 | * no good way to determine if all of a spa's buffers have been | |
3155 | * evicted from an arc state. | |
3156 | */ | |
3157 | ASSERT(!retry || spa == 0); | |
3158 | ||
3159 | if (spa != NULL) | |
3160 | guid = spa_load_guid(spa); | |
3161 | ||
3162 | (void) arc_flush_state(arc_mru, guid, ARC_BUFC_DATA, retry); | |
3163 | (void) arc_flush_state(arc_mru, guid, ARC_BUFC_METADATA, retry); | |
3164 | ||
3165 | (void) arc_flush_state(arc_mfu, guid, ARC_BUFC_DATA, retry); | |
3166 | (void) arc_flush_state(arc_mfu, guid, ARC_BUFC_METADATA, retry); | |
3167 | ||
3168 | (void) arc_flush_state(arc_mru_ghost, guid, ARC_BUFC_DATA, retry); | |
3169 | (void) arc_flush_state(arc_mru_ghost, guid, ARC_BUFC_METADATA, retry); | |
3170 | ||
3171 | (void) arc_flush_state(arc_mfu_ghost, guid, ARC_BUFC_DATA, retry); | |
3172 | (void) arc_flush_state(arc_mfu_ghost, guid, ARC_BUFC_METADATA, retry); | |
3173 | ||
3174 | arc_do_user_evicts(); | |
3175 | ASSERT(spa || arc_eviction_list == NULL); | |
3176 | } | |
3177 | ||
3178 | void | |
3179 | arc_shrink(int64_t to_free) | |
3180 | { | |
3181 | uint64_t c = arc_c; | |
3182 | ||
3183 | if (c > to_free && c - to_free > arc_c_min) { | |
3184 | arc_c = c - to_free; | |
3185 | atomic_add_64(&arc_p, -(arc_p >> arc_shrink_shift)); | |
3186 | if (arc_c > arc_size) | |
3187 | arc_c = MAX(arc_size, arc_c_min); | |
3188 | if (arc_p > arc_c) | |
3189 | arc_p = (arc_c >> 1); | |
3190 | ASSERT(arc_c >= arc_c_min); | |
3191 | ASSERT((int64_t)arc_p >= 0); | |
3192 | } else { | |
3193 | arc_c = arc_c_min; | |
3194 | } | |
3195 | ||
3196 | if (arc_size > arc_c) | |
3197 | (void) arc_adjust(); | |
3198 | } | |
3199 | ||
3200 | typedef enum free_memory_reason_t { | |
3201 | FMR_UNKNOWN, | |
3202 | FMR_NEEDFREE, | |
3203 | FMR_LOTSFREE, | |
3204 | FMR_SWAPFS_MINFREE, | |
3205 | FMR_PAGES_PP_MAXIMUM, | |
3206 | FMR_HEAP_ARENA, | |
3207 | FMR_ZIO_ARENA, | |
3208 | } free_memory_reason_t; | |
3209 | ||
3210 | int64_t last_free_memory; | |
3211 | free_memory_reason_t last_free_reason; | |
3212 | ||
3213 | #ifdef _KERNEL | |
3214 | /* | |
3215 | * Additional reserve of pages for pp_reserve. | |
3216 | */ | |
3217 | int64_t arc_pages_pp_reserve = 64; | |
3218 | ||
3219 | /* | |
3220 | * Additional reserve of pages for swapfs. | |
3221 | */ | |
3222 | int64_t arc_swapfs_reserve = 64; | |
3223 | #endif /* _KERNEL */ | |
3224 | ||
3225 | /* | |
3226 | * Return the amount of memory that can be consumed before reclaim will be | |
3227 | * needed. Positive if there is sufficient free memory, negative indicates | |
3228 | * the amount of memory that needs to be freed up. | |
3229 | */ | |
3230 | static int64_t | |
3231 | arc_available_memory(void) | |
3232 | { | |
3233 | int64_t lowest = INT64_MAX; | |
3234 | free_memory_reason_t r = FMR_UNKNOWN; | |
3235 | #ifdef _KERNEL | |
3236 | int64_t n; | |
3237 | #ifdef __linux__ | |
3238 | pgcnt_t needfree = btop(arc_need_free); | |
3239 | pgcnt_t lotsfree = btop(arc_sys_free); | |
3240 | pgcnt_t desfree = 0; | |
3241 | #endif | |
3242 | ||
3243 | if (needfree > 0) { | |
3244 | n = PAGESIZE * (-needfree); | |
3245 | if (n < lowest) { | |
3246 | lowest = n; | |
3247 | r = FMR_NEEDFREE; | |
3248 | } | |
3249 | } | |
3250 | ||
3251 | /* | |
3252 | * check that we're out of range of the pageout scanner. It starts to | |
3253 | * schedule paging if freemem is less than lotsfree and needfree. | |
3254 | * lotsfree is the high-water mark for pageout, and needfree is the | |
3255 | * number of needed free pages. We add extra pages here to make sure | |
3256 | * the scanner doesn't start up while we're freeing memory. | |
3257 | */ | |
3258 | n = PAGESIZE * (freemem - lotsfree - needfree - desfree); | |
3259 | if (n < lowest) { | |
3260 | lowest = n; | |
3261 | r = FMR_LOTSFREE; | |
3262 | } | |
3263 | ||
3264 | #ifndef __linux__ | |
3265 | /* | |
3266 | * check to make sure that swapfs has enough space so that anon | |
3267 | * reservations can still succeed. anon_resvmem() checks that the | |
3268 | * availrmem is greater than swapfs_minfree, and the number of reserved | |
3269 | * swap pages. We also add a bit of extra here just to prevent | |
3270 | * circumstances from getting really dire. | |
3271 | */ | |
3272 | n = PAGESIZE * (availrmem - swapfs_minfree - swapfs_reserve - | |
3273 | desfree - arc_swapfs_reserve); | |
3274 | if (n < lowest) { | |
3275 | lowest = n; | |
3276 | r = FMR_SWAPFS_MINFREE; | |
3277 | } | |
3278 | ||
3279 | ||
3280 | /* | |
3281 | * Check that we have enough availrmem that memory locking (e.g., via | |
3282 | * mlock(3C) or memcntl(2)) can still succeed. (pages_pp_maximum | |
3283 | * stores the number of pages that cannot be locked; when availrmem | |
3284 | * drops below pages_pp_maximum, page locking mechanisms such as | |
3285 | * page_pp_lock() will fail.) | |
3286 | */ | |
3287 | n = PAGESIZE * (availrmem - pages_pp_maximum - | |
3288 | arc_pages_pp_reserve); | |
3289 | if (n < lowest) { | |
3290 | lowest = n; | |
3291 | r = FMR_PAGES_PP_MAXIMUM; | |
3292 | } | |
3293 | #endif | |
3294 | ||
3295 | #if defined(__i386) | |
3296 | /* | |
3297 | * If we're on an i386 platform, it's possible that we'll exhaust the | |
3298 | * kernel heap space before we ever run out of available physical | |
3299 | * memory. Most checks of the size of the heap_area compare against | |
3300 | * tune.t_minarmem, which is the minimum available real memory that we | |
3301 | * can have in the system. However, this is generally fixed at 25 pages | |
3302 | * which is so low that it's useless. In this comparison, we seek to | |
3303 | * calculate the total heap-size, and reclaim if more than 3/4ths of the | |
3304 | * heap is allocated. (Or, in the calculation, if less than 1/4th is | |
3305 | * free) | |
3306 | */ | |
3307 | n = vmem_size(heap_arena, VMEM_FREE) - | |
3308 | (vmem_size(heap_arena, VMEM_FREE | VMEM_ALLOC) >> 2); | |
3309 | if (n < lowest) { | |
3310 | lowest = n; | |
3311 | r = FMR_HEAP_ARENA; | |
3312 | } | |
3313 | #endif | |
3314 | ||
3315 | /* | |
3316 | * If zio data pages are being allocated out of a separate heap segment, | |
3317 | * then enforce that the size of available vmem for this arena remains | |
3318 | * above about 1/16th free. | |
3319 | * | |
3320 | * Note: The 1/16th arena free requirement was put in place | |
3321 | * to aggressively evict memory from the arc in order to avoid | |
3322 | * memory fragmentation issues. | |
3323 | */ | |
3324 | if (zio_arena != NULL) { | |
3325 | n = vmem_size(zio_arena, VMEM_FREE) - | |
3326 | (vmem_size(zio_arena, VMEM_ALLOC) >> 4); | |
3327 | if (n < lowest) { | |
3328 | lowest = n; | |
3329 | r = FMR_ZIO_ARENA; | |
3330 | } | |
3331 | } | |
3332 | #else /* _KERNEL */ | |
3333 | /* Every 100 calls, free a small amount */ | |
3334 | if (spa_get_random(100) == 0) | |
3335 | lowest = -1024; | |
3336 | #endif /* _KERNEL */ | |
3337 | ||
3338 | last_free_memory = lowest; | |
3339 | last_free_reason = r; | |
3340 | ||
3341 | return (lowest); | |
3342 | } | |
3343 | ||
3344 | /* | |
3345 | * Determine if the system is under memory pressure and is asking | |
3346 | * to reclaim memory. A return value of TRUE indicates that the system | |
3347 | * is under memory pressure and that the arc should adjust accordingly. | |
3348 | */ | |
3349 | static boolean_t | |
3350 | arc_reclaim_needed(void) | |
3351 | { | |
3352 | return (arc_available_memory() < 0); | |
3353 | } | |
3354 | ||
3355 | static void | |
3356 | arc_kmem_reap_now(void) | |
3357 | { | |
3358 | size_t i; | |
3359 | kmem_cache_t *prev_cache = NULL; | |
3360 | kmem_cache_t *prev_data_cache = NULL; | |
3361 | extern kmem_cache_t *zio_buf_cache[]; | |
3362 | extern kmem_cache_t *zio_data_buf_cache[]; | |
3363 | extern kmem_cache_t *range_seg_cache; | |
3364 | ||
3365 | if ((arc_meta_used >= arc_meta_limit) && zfs_arc_meta_prune) { | |
3366 | /* | |
3367 | * We are exceeding our meta-data cache limit. | |
3368 | * Prune some entries to release holds on meta-data. | |
3369 | */ | |
3370 | arc_prune_async(zfs_arc_meta_prune); | |
3371 | } | |
3372 | ||
3373 | for (i = 0; i < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; i++) { | |
3374 | #ifdef _ILP32 | |
3375 | /* reach upper limit of cache size on 32-bit */ | |
3376 | if (zio_buf_cache[i] == NULL) | |
3377 | break; | |
3378 | #endif | |
3379 | if (zio_buf_cache[i] != prev_cache) { | |
3380 | prev_cache = zio_buf_cache[i]; | |
3381 | kmem_cache_reap_now(zio_buf_cache[i]); | |
3382 | } | |
3383 | if (zio_data_buf_cache[i] != prev_data_cache) { | |
3384 | prev_data_cache = zio_data_buf_cache[i]; | |
3385 | kmem_cache_reap_now(zio_data_buf_cache[i]); | |
3386 | } | |
3387 | } | |
3388 | kmem_cache_reap_now(buf_cache); | |
3389 | kmem_cache_reap_now(hdr_full_cache); | |
3390 | kmem_cache_reap_now(hdr_l2only_cache); | |
3391 | kmem_cache_reap_now(range_seg_cache); | |
3392 | ||
3393 | if (zio_arena != NULL) { | |
3394 | /* | |
3395 | * Ask the vmem arena to reclaim unused memory from its | |
3396 | * quantum caches. | |
3397 | */ | |
3398 | vmem_qcache_reap(zio_arena); | |
3399 | } | |
3400 | } | |
3401 | ||
3402 | /* | |
3403 | * Threads can block in arc_get_data_buf() waiting for this thread to evict | |
3404 | * enough data and signal them to proceed. When this happens, the threads in | |
3405 | * arc_get_data_buf() are sleeping while holding the hash lock for their | |
3406 | * particular arc header. Thus, we must be careful to never sleep on a | |
3407 | * hash lock in this thread. This is to prevent the following deadlock: | |
3408 | * | |
3409 | * - Thread A sleeps on CV in arc_get_data_buf() holding hash lock "L", | |
3410 | * waiting for the reclaim thread to signal it. | |
3411 | * | |
3412 | * - arc_reclaim_thread() tries to acquire hash lock "L" using mutex_enter, | |
3413 | * fails, and goes to sleep forever. | |
3414 | * | |
3415 | * This possible deadlock is avoided by always acquiring a hash lock | |
3416 | * using mutex_tryenter() from arc_reclaim_thread(). | |
3417 | */ | |
3418 | static void | |
3419 | arc_reclaim_thread(void) | |
3420 | { | |
3421 | fstrans_cookie_t cookie = spl_fstrans_mark(); | |
3422 | clock_t growtime = 0; | |
3423 | callb_cpr_t cpr; | |
3424 | ||
3425 | CALLB_CPR_INIT(&cpr, &arc_reclaim_lock, callb_generic_cpr, FTAG); | |
3426 | ||
3427 | mutex_enter(&arc_reclaim_lock); | |
3428 | while (!arc_reclaim_thread_exit) { | |
3429 | int64_t to_free; | |
3430 | int64_t free_memory = arc_available_memory(); | |
3431 | uint64_t evicted = 0; | |
3432 | ||
3433 | arc_tuning_update(); | |
3434 | ||
3435 | mutex_exit(&arc_reclaim_lock); | |
3436 | ||
3437 | if (free_memory < 0) { | |
3438 | ||
3439 | arc_no_grow = B_TRUE; | |
3440 | arc_warm = B_TRUE; | |
3441 | ||
3442 | /* | |
3443 | * Wait at least zfs_grow_retry (default 5) seconds | |
3444 | * before considering growing. | |
3445 | */ | |
3446 | growtime = ddi_get_lbolt() + (arc_grow_retry * hz); | |
3447 | ||
3448 | arc_kmem_reap_now(); | |
3449 | ||
3450 | /* | |
3451 | * If we are still low on memory, shrink the ARC | |
3452 | * so that we have arc_shrink_min free space. | |
3453 | */ | |
3454 | free_memory = arc_available_memory(); | |
3455 | ||
3456 | to_free = (arc_c >> arc_shrink_shift) - free_memory; | |
3457 | if (to_free > 0) { | |
3458 | #ifdef _KERNEL | |
3459 | to_free = MAX(to_free, arc_need_free); | |
3460 | #endif | |
3461 | arc_shrink(to_free); | |
3462 | } | |
3463 | } else if (free_memory < arc_c >> arc_no_grow_shift) { | |
3464 | arc_no_grow = B_TRUE; | |
3465 | } else if (ddi_get_lbolt() >= growtime) { | |
3466 | arc_no_grow = B_FALSE; | |
3467 | } | |
3468 | ||
3469 | evicted = arc_adjust(); | |
3470 | ||
3471 | mutex_enter(&arc_reclaim_lock); | |
3472 | ||
3473 | /* | |
3474 | * If evicted is zero, we couldn't evict anything via | |
3475 | * arc_adjust(). This could be due to hash lock | |
3476 | * collisions, but more likely due to the majority of | |
3477 | * arc buffers being unevictable. Therefore, even if | |
3478 | * arc_size is above arc_c, another pass is unlikely to | |
3479 | * be helpful and could potentially cause us to enter an | |
3480 | * infinite loop. | |
3481 | */ | |
3482 | if (arc_size <= arc_c || evicted == 0) { | |
3483 | /* | |
3484 | * We're either no longer overflowing, or we | |
3485 | * can't evict anything more, so we should wake | |
3486 | * up any threads before we go to sleep and clear | |
3487 | * arc_need_free since nothing more can be done. | |
3488 | */ | |
3489 | cv_broadcast(&arc_reclaim_waiters_cv); | |
3490 | arc_need_free = 0; | |
3491 | ||
3492 | /* | |
3493 | * Block until signaled, or after one second (we | |
3494 | * might need to perform arc_kmem_reap_now() | |
3495 | * even if we aren't being signalled) | |
3496 | */ | |
3497 | CALLB_CPR_SAFE_BEGIN(&cpr); | |
3498 | (void) cv_timedwait_sig(&arc_reclaim_thread_cv, | |
3499 | &arc_reclaim_lock, ddi_get_lbolt() + hz); | |
3500 | CALLB_CPR_SAFE_END(&cpr, &arc_reclaim_lock); | |
3501 | } | |
3502 | } | |
3503 | ||
3504 | arc_reclaim_thread_exit = FALSE; | |
3505 | cv_broadcast(&arc_reclaim_thread_cv); | |
3506 | CALLB_CPR_EXIT(&cpr); /* drops arc_reclaim_lock */ | |
3507 | spl_fstrans_unmark(cookie); | |
3508 | thread_exit(); | |
3509 | } | |
3510 | ||
3511 | static void | |
3512 | arc_user_evicts_thread(void) | |
3513 | { | |
3514 | fstrans_cookie_t cookie = spl_fstrans_mark(); | |
3515 | callb_cpr_t cpr; | |
3516 | ||
3517 | CALLB_CPR_INIT(&cpr, &arc_user_evicts_lock, callb_generic_cpr, FTAG); | |
3518 | ||
3519 | mutex_enter(&arc_user_evicts_lock); | |
3520 | while (!arc_user_evicts_thread_exit) { | |
3521 | mutex_exit(&arc_user_evicts_lock); | |
3522 | ||
3523 | arc_do_user_evicts(); | |
3524 | ||
3525 | /* | |
3526 | * This is necessary in order for the mdb ::arc dcmd to | |
3527 | * show up to date information. Since the ::arc command | |
3528 | * does not call the kstat's update function, without | |
3529 | * this call, the command may show stale stats for the | |
3530 | * anon, mru, mru_ghost, mfu, and mfu_ghost lists. Even | |
3531 | * with this change, the data might be up to 1 second | |
3532 | * out of date; but that should suffice. The arc_state_t | |
3533 | * structures can be queried directly if more accurate | |
3534 | * information is needed. | |
3535 | */ | |
3536 | if (arc_ksp != NULL) | |
3537 | arc_ksp->ks_update(arc_ksp, KSTAT_READ); | |
3538 | ||
3539 | mutex_enter(&arc_user_evicts_lock); | |
3540 | ||
3541 | /* | |
3542 | * Block until signaled, or after one second (we need to | |
3543 | * call the arc's kstat update function regularly). | |
3544 | */ | |
3545 | CALLB_CPR_SAFE_BEGIN(&cpr); | |
3546 | (void) cv_timedwait_sig(&arc_user_evicts_cv, | |
3547 | &arc_user_evicts_lock, ddi_get_lbolt() + hz); | |
3548 | CALLB_CPR_SAFE_END(&cpr, &arc_user_evicts_lock); | |
3549 | } | |
3550 | ||
3551 | arc_user_evicts_thread_exit = FALSE; | |
3552 | cv_broadcast(&arc_user_evicts_cv); | |
3553 | CALLB_CPR_EXIT(&cpr); /* drops arc_user_evicts_lock */ | |
3554 | spl_fstrans_unmark(cookie); | |
3555 | thread_exit(); | |
3556 | } | |
3557 | ||
3558 | #ifdef _KERNEL | |
3559 | /* | |
3560 | * Determine the amount of memory eligible for eviction contained in the | |
3561 | * ARC. All clean data reported by the ghost lists can always be safely | |
3562 | * evicted. Due to arc_c_min, the same does not hold for all clean data | |
3563 | * contained by the regular mru and mfu lists. | |
3564 | * | |
3565 | * In the case of the regular mru and mfu lists, we need to report as | |
3566 | * much clean data as possible, such that evicting that same reported | |
3567 | * data will not bring arc_size below arc_c_min. Thus, in certain | |
3568 | * circumstances, the total amount of clean data in the mru and mfu | |
3569 | * lists might not actually be evictable. | |
3570 | * | |
3571 | * The following two distinct cases are accounted for: | |
3572 | * | |
3573 | * 1. The sum of the amount of dirty data contained by both the mru and | |
3574 | * mfu lists, plus the ARC's other accounting (e.g. the anon list), | |
3575 | * is greater than or equal to arc_c_min. | |
3576 | * (i.e. amount of dirty data >= arc_c_min) | |
3577 | * | |
3578 | * This is the easy case; all clean data contained by the mru and mfu | |
3579 | * lists is evictable. Evicting all clean data can only drop arc_size | |
3580 | * to the amount of dirty data, which is greater than arc_c_min. | |
3581 | * | |
3582 | * 2. The sum of the amount of dirty data contained by both the mru and | |
3583 | * mfu lists, plus the ARC's other accounting (e.g. the anon list), | |
3584 | * is less than arc_c_min. | |
3585 | * (i.e. arc_c_min > amount of dirty data) | |
3586 | * | |
3587 | * 2.1. arc_size is greater than or equal arc_c_min. | |
3588 | * (i.e. arc_size >= arc_c_min > amount of dirty data) | |
3589 | * | |
3590 | * In this case, not all clean data from the regular mru and mfu | |
3591 | * lists is actually evictable; we must leave enough clean data | |
3592 | * to keep arc_size above arc_c_min. Thus, the maximum amount of | |
3593 | * evictable data from the two lists combined, is exactly the | |
3594 | * difference between arc_size and arc_c_min. | |
3595 | * | |
3596 | * 2.2. arc_size is less than arc_c_min | |
3597 | * (i.e. arc_c_min > arc_size > amount of dirty data) | |
3598 | * | |
3599 | * In this case, none of the data contained in the mru and mfu | |
3600 | * lists is evictable, even if it's clean. Since arc_size is | |
3601 | * already below arc_c_min, evicting any more would only | |
3602 | * increase this negative difference. | |
3603 | */ | |
3604 | static uint64_t | |
3605 | arc_evictable_memory(void) { | |
3606 | uint64_t arc_clean = | |
3607 | arc_mru->arcs_lsize[ARC_BUFC_DATA] + | |
3608 | arc_mru->arcs_lsize[ARC_BUFC_METADATA] + | |
3609 | arc_mfu->arcs_lsize[ARC_BUFC_DATA] + | |
3610 | arc_mfu->arcs_lsize[ARC_BUFC_METADATA]; | |
3611 | uint64_t ghost_clean = | |
3612 | arc_mru_ghost->arcs_lsize[ARC_BUFC_DATA] + | |
3613 | arc_mru_ghost->arcs_lsize[ARC_BUFC_METADATA] + | |
3614 | arc_mfu_ghost->arcs_lsize[ARC_BUFC_DATA] + | |
3615 | arc_mfu_ghost->arcs_lsize[ARC_BUFC_METADATA]; | |
3616 | uint64_t arc_dirty = MAX((int64_t)arc_size - (int64_t)arc_clean, 0); | |
3617 | ||
3618 | if (arc_dirty >= arc_c_min) | |
3619 | return (ghost_clean + arc_clean); | |
3620 | ||
3621 | return (ghost_clean + MAX((int64_t)arc_size - (int64_t)arc_c_min, 0)); | |
3622 | } | |
3623 | ||
3624 | /* | |
3625 | * If sc->nr_to_scan is zero, the caller is requesting a query of the | |
3626 | * number of objects which can potentially be freed. If it is nonzero, | |
3627 | * the request is to free that many objects. | |
3628 | * | |
3629 | * Linux kernels >= 3.12 have the count_objects and scan_objects callbacks | |
3630 | * in struct shrinker and also require the shrinker to return the number | |
3631 | * of objects freed. | |
3632 | * | |
3633 | * Older kernels require the shrinker to return the number of freeable | |
3634 | * objects following the freeing of nr_to_free. | |
3635 | */ | |
3636 | static spl_shrinker_t | |
3637 | __arc_shrinker_func(struct shrinker *shrink, struct shrink_control *sc) | |
3638 | { | |
3639 | int64_t pages; | |
3640 | ||
3641 | /* The arc is considered warm once reclaim has occurred */ | |
3642 | if (unlikely(arc_warm == B_FALSE)) | |
3643 | arc_warm = B_TRUE; | |
3644 | ||
3645 | /* Return the potential number of reclaimable pages */ | |
3646 | pages = btop((int64_t)arc_evictable_memory()); | |
3647 | if (sc->nr_to_scan == 0) | |
3648 | return (pages); | |
3649 | ||
3650 | /* Not allowed to perform filesystem reclaim */ | |
3651 | if (!(sc->gfp_mask & __GFP_FS)) | |
3652 | return (SHRINK_STOP); | |
3653 | ||
3654 | /* Reclaim in progress */ | |
3655 | if (mutex_tryenter(&arc_reclaim_lock) == 0) | |
3656 | return (SHRINK_STOP); | |
3657 | ||
3658 | mutex_exit(&arc_reclaim_lock); | |
3659 | ||
3660 | /* | |
3661 | * Evict the requested number of pages by shrinking arc_c the | |
3662 | * requested amount. If there is nothing left to evict just | |
3663 | * reap whatever we can from the various arc slabs. | |
3664 | */ | |
3665 | if (pages > 0) { | |
3666 | arc_shrink(ptob(sc->nr_to_scan)); | |
3667 | arc_kmem_reap_now(); | |
3668 | #ifdef HAVE_SPLIT_SHRINKER_CALLBACK | |
3669 | pages = MAX(pages - btop(arc_evictable_memory()), 0); | |
3670 | #else | |
3671 | pages = btop(arc_evictable_memory()); | |
3672 | #endif | |
3673 | } else { | |
3674 | arc_kmem_reap_now(); | |
3675 | pages = SHRINK_STOP; | |
3676 | } | |
3677 | ||
3678 | /* | |
3679 | * We've reaped what we can, wake up threads. | |
3680 | */ | |
3681 | cv_broadcast(&arc_reclaim_waiters_cv); | |
3682 | ||
3683 | /* | |
3684 | * When direct reclaim is observed it usually indicates a rapid | |
3685 | * increase in memory pressure. This occurs because the kswapd | |
3686 | * threads were unable to asynchronously keep enough free memory | |
3687 | * available. In this case set arc_no_grow to briefly pause arc | |
3688 | * growth to avoid compounding the memory pressure. | |
3689 | */ | |
3690 | if (current_is_kswapd()) { | |
3691 | ARCSTAT_BUMP(arcstat_memory_indirect_count); | |
3692 | } else { | |
3693 | arc_no_grow = B_TRUE; | |
3694 | arc_need_free = ptob(sc->nr_to_scan); | |
3695 | ARCSTAT_BUMP(arcstat_memory_direct_count); | |
3696 | } | |
3697 | ||
3698 | return (pages); | |
3699 | } | |
3700 | SPL_SHRINKER_CALLBACK_WRAPPER(arc_shrinker_func); | |
3701 | ||
3702 | SPL_SHRINKER_DECLARE(arc_shrinker, arc_shrinker_func, DEFAULT_SEEKS); | |
3703 | #endif /* _KERNEL */ | |
3704 | ||
3705 | /* | |
3706 | * Adapt arc info given the number of bytes we are trying to add and | |
3707 | * the state that we are comming from. This function is only called | |
3708 | * when we are adding new content to the cache. | |
3709 | */ | |
3710 | static void | |
3711 | arc_adapt(int bytes, arc_state_t *state) | |
3712 | { | |
3713 | int mult; | |
3714 | uint64_t arc_p_min = (arc_c >> arc_p_min_shift); | |
3715 | int64_t mrug_size = refcount_count(&arc_mru_ghost->arcs_size); | |
3716 | int64_t mfug_size = refcount_count(&arc_mfu_ghost->arcs_size); | |
3717 | ||
3718 | if (state == arc_l2c_only) | |
3719 | return; | |
3720 | ||
3721 | ASSERT(bytes > 0); | |
3722 | /* | |
3723 | * Adapt the target size of the MRU list: | |
3724 | * - if we just hit in the MRU ghost list, then increase | |
3725 | * the target size of the MRU list. | |
3726 | * - if we just hit in the MFU ghost list, then increase | |
3727 | * the target size of the MFU list by decreasing the | |
3728 | * target size of the MRU list. | |
3729 | */ | |
3730 | if (state == arc_mru_ghost) { | |
3731 | mult = (mrug_size >= mfug_size) ? 1 : (mfug_size / mrug_size); | |
3732 | if (!zfs_arc_p_dampener_disable) | |
3733 | mult = MIN(mult, 10); /* avoid wild arc_p adjustment */ | |
3734 | ||
3735 | arc_p = MIN(arc_c - arc_p_min, arc_p + bytes * mult); | |
3736 | } else if (state == arc_mfu_ghost) { | |
3737 | uint64_t delta; | |
3738 | ||
3739 | mult = (mfug_size >= mrug_size) ? 1 : (mrug_size / mfug_size); | |
3740 | if (!zfs_arc_p_dampener_disable) | |
3741 | mult = MIN(mult, 10); | |
3742 | ||
3743 | delta = MIN(bytes * mult, arc_p); | |
3744 | arc_p = MAX(arc_p_min, arc_p - delta); | |
3745 | } | |
3746 | ASSERT((int64_t)arc_p >= 0); | |
3747 | ||
3748 | if (arc_reclaim_needed()) { | |
3749 | cv_signal(&arc_reclaim_thread_cv); | |
3750 | return; | |
3751 | } | |
3752 | ||
3753 | if (arc_no_grow) | |
3754 | return; | |
3755 | ||
3756 | if (arc_c >= arc_c_max) | |
3757 | return; | |
3758 | ||
3759 | /* | |
3760 | * If we're within (2 * maxblocksize) bytes of the target | |
3761 | * cache size, increment the target cache size | |
3762 | */ | |
3763 | ASSERT3U(arc_c, >=, 2ULL << SPA_MAXBLOCKSHIFT); | |
3764 | if (arc_size >= arc_c - (2ULL << SPA_MAXBLOCKSHIFT)) { | |
3765 | atomic_add_64(&arc_c, (int64_t)bytes); | |
3766 | if (arc_c > arc_c_max) | |
3767 | arc_c = arc_c_max; | |
3768 | else if (state == arc_anon) | |
3769 | atomic_add_64(&arc_p, (int64_t)bytes); | |
3770 | if (arc_p > arc_c) | |
3771 | arc_p = arc_c; | |
3772 | } | |
3773 | ASSERT((int64_t)arc_p >= 0); | |
3774 | } | |
3775 | ||
3776 | /* | |
3777 | * Check if arc_size has grown past our upper threshold, determined by | |
3778 | * zfs_arc_overflow_shift. | |
3779 | */ | |
3780 | static boolean_t | |
3781 | arc_is_overflowing(void) | |
3782 | { | |
3783 | /* Always allow at least one block of overflow */ | |
3784 | uint64_t overflow = MAX(SPA_MAXBLOCKSIZE, | |
3785 | arc_c >> zfs_arc_overflow_shift); | |
3786 | ||
3787 | return (arc_size >= arc_c + overflow); | |
3788 | } | |
3789 | ||
3790 | /* | |
3791 | * The buffer, supplied as the first argument, needs a data block. If we | |
3792 | * are hitting the hard limit for the cache size, we must sleep, waiting | |
3793 | * for the eviction thread to catch up. If we're past the target size | |
3794 | * but below the hard limit, we'll only signal the reclaim thread and | |
3795 | * continue on. | |
3796 | */ | |
3797 | static void | |
3798 | arc_get_data_buf(arc_buf_t *buf) | |
3799 | { | |
3800 | arc_state_t *state = buf->b_hdr->b_l1hdr.b_state; | |
3801 | uint64_t size = buf->b_hdr->b_size; | |
3802 | arc_buf_contents_t type = arc_buf_type(buf->b_hdr); | |
3803 | ||
3804 | arc_adapt(size, state); | |
3805 | ||
3806 | /* | |
3807 | * If arc_size is currently overflowing, and has grown past our | |
3808 | * upper limit, we must be adding data faster than the evict | |
3809 | * thread can evict. Thus, to ensure we don't compound the | |
3810 | * problem by adding more data and forcing arc_size to grow even | |
3811 | * further past it's target size, we halt and wait for the | |
3812 | * eviction thread to catch up. | |
3813 | * | |
3814 | * It's also possible that the reclaim thread is unable to evict | |
3815 | * enough buffers to get arc_size below the overflow limit (e.g. | |
3816 | * due to buffers being un-evictable, or hash lock collisions). | |
3817 | * In this case, we want to proceed regardless if we're | |
3818 | * overflowing; thus we don't use a while loop here. | |
3819 | */ | |
3820 | if (arc_is_overflowing()) { | |
3821 | mutex_enter(&arc_reclaim_lock); | |
3822 | ||
3823 | /* | |
3824 | * Now that we've acquired the lock, we may no longer be | |
3825 | * over the overflow limit, lets check. | |
3826 | * | |
3827 | * We're ignoring the case of spurious wake ups. If that | |
3828 | * were to happen, it'd let this thread consume an ARC | |
3829 | * buffer before it should have (i.e. before we're under | |
3830 | * the overflow limit and were signalled by the reclaim | |
3831 | * thread). As long as that is a rare occurrence, it | |
3832 | * shouldn't cause any harm. | |
3833 | */ | |
3834 | if (arc_is_overflowing()) { | |
3835 | cv_signal(&arc_reclaim_thread_cv); | |
3836 | cv_wait(&arc_reclaim_waiters_cv, &arc_reclaim_lock); | |
3837 | } | |
3838 | ||
3839 | mutex_exit(&arc_reclaim_lock); | |
3840 | } | |
3841 | ||
3842 | if (type == ARC_BUFC_METADATA) { | |
3843 | buf->b_data = zio_buf_alloc(size); | |
3844 | arc_space_consume(size, ARC_SPACE_META); | |
3845 | } else { | |
3846 | ASSERT(type == ARC_BUFC_DATA); | |
3847 | buf->b_data = zio_data_buf_alloc(size); | |
3848 | arc_space_consume(size, ARC_SPACE_DATA); | |
3849 | } | |
3850 | ||
3851 | /* | |
3852 | * Update the state size. Note that ghost states have a | |
3853 | * "ghost size" and so don't need to be updated. | |
3854 | */ | |
3855 | if (!GHOST_STATE(buf->b_hdr->b_l1hdr.b_state)) { | |
3856 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
3857 | arc_state_t *state = hdr->b_l1hdr.b_state; | |
3858 | ||
3859 | (void) refcount_add_many(&state->arcs_size, size, buf); | |
3860 | ||
3861 | /* | |
3862 | * If this is reached via arc_read, the link is | |
3863 | * protected by the hash lock. If reached via | |
3864 | * arc_buf_alloc, the header should not be accessed by | |
3865 | * any other thread. And, if reached via arc_read_done, | |
3866 | * the hash lock will protect it if it's found in the | |
3867 | * hash table; otherwise no other thread should be | |
3868 | * trying to [add|remove]_reference it. | |
3869 | */ | |
3870 | if (multilist_link_active(&hdr->b_l1hdr.b_arc_node)) { | |
3871 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); | |
3872 | atomic_add_64(&hdr->b_l1hdr.b_state->arcs_lsize[type], | |
3873 | size); | |
3874 | } | |
3875 | /* | |
3876 | * If we are growing the cache, and we are adding anonymous | |
3877 | * data, and we have outgrown arc_p, update arc_p | |
3878 | */ | |
3879 | if (arc_size < arc_c && hdr->b_l1hdr.b_state == arc_anon && | |
3880 | (refcount_count(&arc_anon->arcs_size) + | |
3881 | refcount_count(&arc_mru->arcs_size) > arc_p)) | |
3882 | arc_p = MIN(arc_c, arc_p + size); | |
3883 | } | |
3884 | } | |
3885 | ||
3886 | /* | |
3887 | * This routine is called whenever a buffer is accessed. | |
3888 | * NOTE: the hash lock is dropped in this function. | |
3889 | */ | |
3890 | static void | |
3891 | arc_access(arc_buf_hdr_t *hdr, kmutex_t *hash_lock) | |
3892 | { | |
3893 | clock_t now; | |
3894 | ||
3895 | ASSERT(MUTEX_HELD(hash_lock)); | |
3896 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
3897 | ||
3898 | if (hdr->b_l1hdr.b_state == arc_anon) { | |
3899 | /* | |
3900 | * This buffer is not in the cache, and does not | |
3901 | * appear in our "ghost" list. Add the new buffer | |
3902 | * to the MRU state. | |
3903 | */ | |
3904 | ||
3905 | ASSERT0(hdr->b_l1hdr.b_arc_access); | |
3906 | hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); | |
3907 | DTRACE_PROBE1(new_state__mru, arc_buf_hdr_t *, hdr); | |
3908 | arc_change_state(arc_mru, hdr, hash_lock); | |
3909 | ||
3910 | } else if (hdr->b_l1hdr.b_state == arc_mru) { | |
3911 | now = ddi_get_lbolt(); | |
3912 | ||
3913 | /* | |
3914 | * If this buffer is here because of a prefetch, then either: | |
3915 | * - clear the flag if this is a "referencing" read | |
3916 | * (any subsequent access will bump this into the MFU state). | |
3917 | * or | |
3918 | * - move the buffer to the head of the list if this is | |
3919 | * another prefetch (to make it less likely to be evicted). | |
3920 | */ | |
3921 | if (HDR_PREFETCH(hdr)) { | |
3922 | if (refcount_count(&hdr->b_l1hdr.b_refcnt) == 0) { | |
3923 | /* link protected by hash lock */ | |
3924 | ASSERT(multilist_link_active( | |
3925 | &hdr->b_l1hdr.b_arc_node)); | |
3926 | } else { | |
3927 | hdr->b_flags &= ~ARC_FLAG_PREFETCH; | |
3928 | atomic_inc_32(&hdr->b_l1hdr.b_mru_hits); | |
3929 | ARCSTAT_BUMP(arcstat_mru_hits); | |
3930 | } | |
3931 | hdr->b_l1hdr.b_arc_access = now; | |
3932 | return; | |
3933 | } | |
3934 | ||
3935 | /* | |
3936 | * This buffer has been "accessed" only once so far, | |
3937 | * but it is still in the cache. Move it to the MFU | |
3938 | * state. | |
3939 | */ | |
3940 | if (ddi_time_after(now, hdr->b_l1hdr.b_arc_access + | |
3941 | ARC_MINTIME)) { | |
3942 | /* | |
3943 | * More than 125ms have passed since we | |
3944 | * instantiated this buffer. Move it to the | |
3945 | * most frequently used state. | |
3946 | */ | |
3947 | hdr->b_l1hdr.b_arc_access = now; | |
3948 | DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); | |
3949 | arc_change_state(arc_mfu, hdr, hash_lock); | |
3950 | } | |
3951 | atomic_inc_32(&hdr->b_l1hdr.b_mru_hits); | |
3952 | ARCSTAT_BUMP(arcstat_mru_hits); | |
3953 | } else if (hdr->b_l1hdr.b_state == arc_mru_ghost) { | |
3954 | arc_state_t *new_state; | |
3955 | /* | |
3956 | * This buffer has been "accessed" recently, but | |
3957 | * was evicted from the cache. Move it to the | |
3958 | * MFU state. | |
3959 | */ | |
3960 | ||
3961 | if (HDR_PREFETCH(hdr)) { | |
3962 | new_state = arc_mru; | |
3963 | if (refcount_count(&hdr->b_l1hdr.b_refcnt) > 0) | |
3964 | hdr->b_flags &= ~ARC_FLAG_PREFETCH; | |
3965 | DTRACE_PROBE1(new_state__mru, arc_buf_hdr_t *, hdr); | |
3966 | } else { | |
3967 | new_state = arc_mfu; | |
3968 | DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); | |
3969 | } | |
3970 | ||
3971 | hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); | |
3972 | arc_change_state(new_state, hdr, hash_lock); | |
3973 | ||
3974 | atomic_inc_32(&hdr->b_l1hdr.b_mru_ghost_hits); | |
3975 | ARCSTAT_BUMP(arcstat_mru_ghost_hits); | |
3976 | } else if (hdr->b_l1hdr.b_state == arc_mfu) { | |
3977 | /* | |
3978 | * This buffer has been accessed more than once and is | |
3979 | * still in the cache. Keep it in the MFU state. | |
3980 | * | |
3981 | * NOTE: an add_reference() that occurred when we did | |
3982 | * the arc_read() will have kicked this off the list. | |
3983 | * If it was a prefetch, we will explicitly move it to | |
3984 | * the head of the list now. | |
3985 | */ | |
3986 | if ((HDR_PREFETCH(hdr)) != 0) { | |
3987 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); | |
3988 | /* link protected by hash_lock */ | |
3989 | ASSERT(multilist_link_active(&hdr->b_l1hdr.b_arc_node)); | |
3990 | } | |
3991 | atomic_inc_32(&hdr->b_l1hdr.b_mfu_hits); | |
3992 | ARCSTAT_BUMP(arcstat_mfu_hits); | |
3993 | hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); | |
3994 | } else if (hdr->b_l1hdr.b_state == arc_mfu_ghost) { | |
3995 | arc_state_t *new_state = arc_mfu; | |
3996 | /* | |
3997 | * This buffer has been accessed more than once but has | |
3998 | * been evicted from the cache. Move it back to the | |
3999 | * MFU state. | |
4000 | */ | |
4001 | ||
4002 | if (HDR_PREFETCH(hdr)) { | |
4003 | /* | |
4004 | * This is a prefetch access... | |
4005 | * move this block back to the MRU state. | |
4006 | */ | |
4007 | ASSERT0(refcount_count(&hdr->b_l1hdr.b_refcnt)); | |
4008 | new_state = arc_mru; | |
4009 | } | |
4010 | ||
4011 | hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); | |
4012 | DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); | |
4013 | arc_change_state(new_state, hdr, hash_lock); | |
4014 | ||
4015 | atomic_inc_32(&hdr->b_l1hdr.b_mfu_ghost_hits); | |
4016 | ARCSTAT_BUMP(arcstat_mfu_ghost_hits); | |
4017 | } else if (hdr->b_l1hdr.b_state == arc_l2c_only) { | |
4018 | /* | |
4019 | * This buffer is on the 2nd Level ARC. | |
4020 | */ | |
4021 | ||
4022 | hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); | |
4023 | DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); | |
4024 | arc_change_state(arc_mfu, hdr, hash_lock); | |
4025 | } else { | |
4026 | cmn_err(CE_PANIC, "invalid arc state 0x%p", | |
4027 | hdr->b_l1hdr.b_state); | |
4028 | } | |
4029 | } | |
4030 | ||
4031 | /* a generic arc_done_func_t which you can use */ | |
4032 | /* ARGSUSED */ | |
4033 | void | |
4034 | arc_bcopy_func(zio_t *zio, arc_buf_t *buf, void *arg) | |
4035 | { | |
4036 | if (zio == NULL || zio->io_error == 0) | |
4037 | bcopy(buf->b_data, arg, buf->b_hdr->b_size); | |
4038 | VERIFY(arc_buf_remove_ref(buf, arg)); | |
4039 | } | |
4040 | ||
4041 | /* a generic arc_done_func_t */ | |
4042 | void | |
4043 | arc_getbuf_func(zio_t *zio, arc_buf_t *buf, void *arg) | |
4044 | { | |
4045 | arc_buf_t **bufp = arg; | |
4046 | if (zio && zio->io_error) { | |
4047 | VERIFY(arc_buf_remove_ref(buf, arg)); | |
4048 | *bufp = NULL; | |
4049 | } else { | |
4050 | *bufp = buf; | |
4051 | ASSERT(buf->b_data); | |
4052 | } | |
4053 | } | |
4054 | ||
4055 | static void | |
4056 | arc_read_done(zio_t *zio) | |
4057 | { | |
4058 | arc_buf_hdr_t *hdr; | |
4059 | arc_buf_t *buf; | |
4060 | arc_buf_t *abuf; /* buffer we're assigning to callback */ | |
4061 | kmutex_t *hash_lock = NULL; | |
4062 | arc_callback_t *callback_list, *acb; | |
4063 | int freeable = FALSE; | |
4064 | ||
4065 | buf = zio->io_private; | |
4066 | hdr = buf->b_hdr; | |
4067 | ||
4068 | /* | |
4069 | * The hdr was inserted into hash-table and removed from lists | |
4070 | * prior to starting I/O. We should find this header, since | |
4071 | * it's in the hash table, and it should be legit since it's | |
4072 | * not possible to evict it during the I/O. The only possible | |
4073 | * reason for it not to be found is if we were freed during the | |
4074 | * read. | |
4075 | */ | |
4076 | if (HDR_IN_HASH_TABLE(hdr)) { | |
4077 | arc_buf_hdr_t *found; | |
4078 | ||
4079 | ASSERT3U(hdr->b_birth, ==, BP_PHYSICAL_BIRTH(zio->io_bp)); | |
4080 | ASSERT3U(hdr->b_dva.dva_word[0], ==, | |
4081 | BP_IDENTITY(zio->io_bp)->dva_word[0]); | |
4082 | ASSERT3U(hdr->b_dva.dva_word[1], ==, | |
4083 | BP_IDENTITY(zio->io_bp)->dva_word[1]); | |
4084 | ||
4085 | found = buf_hash_find(hdr->b_spa, zio->io_bp, | |
4086 | &hash_lock); | |
4087 | ||
4088 | ASSERT((found == NULL && HDR_FREED_IN_READ(hdr) && | |
4089 | hash_lock == NULL) || | |
4090 | (found == hdr && | |
4091 | DVA_EQUAL(&hdr->b_dva, BP_IDENTITY(zio->io_bp))) || | |
4092 | (found == hdr && HDR_L2_READING(hdr))); | |
4093 | } | |
4094 | ||
4095 | hdr->b_flags &= ~ARC_FLAG_L2_EVICTED; | |
4096 | if (l2arc_noprefetch && HDR_PREFETCH(hdr)) | |
4097 | hdr->b_flags &= ~ARC_FLAG_L2CACHE; | |
4098 | ||
4099 | /* byteswap if necessary */ | |
4100 | callback_list = hdr->b_l1hdr.b_acb; | |
4101 | ASSERT(callback_list != NULL); | |
4102 | if (BP_SHOULD_BYTESWAP(zio->io_bp) && zio->io_error == 0) { | |
4103 | dmu_object_byteswap_t bswap = | |
4104 | DMU_OT_BYTESWAP(BP_GET_TYPE(zio->io_bp)); | |
4105 | if (BP_GET_LEVEL(zio->io_bp) > 0) | |
4106 | byteswap_uint64_array(buf->b_data, hdr->b_size); | |
4107 | else | |
4108 | dmu_ot_byteswap[bswap].ob_func(buf->b_data, hdr->b_size); | |
4109 | } | |
4110 | ||
4111 | arc_cksum_compute(buf, B_FALSE); | |
4112 | arc_buf_watch(buf); | |
4113 | ||
4114 | if (hash_lock && zio->io_error == 0 && | |
4115 | hdr->b_l1hdr.b_state == arc_anon) { | |
4116 | /* | |
4117 | * Only call arc_access on anonymous buffers. This is because | |
4118 | * if we've issued an I/O for an evicted buffer, we've already | |
4119 | * called arc_access (to prevent any simultaneous readers from | |
4120 | * getting confused). | |
4121 | */ | |
4122 | arc_access(hdr, hash_lock); | |
4123 | } | |
4124 | ||
4125 | /* create copies of the data buffer for the callers */ | |
4126 | abuf = buf; | |
4127 | for (acb = callback_list; acb; acb = acb->acb_next) { | |
4128 | if (acb->acb_done) { | |
4129 | if (abuf == NULL) { | |
4130 | ARCSTAT_BUMP(arcstat_duplicate_reads); | |
4131 | abuf = arc_buf_clone(buf); | |
4132 | } | |
4133 | acb->acb_buf = abuf; | |
4134 | abuf = NULL; | |
4135 | } | |
4136 | } | |
4137 | hdr->b_l1hdr.b_acb = NULL; | |
4138 | hdr->b_flags &= ~ARC_FLAG_IO_IN_PROGRESS; | |
4139 | ASSERT(!HDR_BUF_AVAILABLE(hdr)); | |
4140 | if (abuf == buf) { | |
4141 | ASSERT(buf->b_efunc == NULL); | |
4142 | ASSERT(hdr->b_l1hdr.b_datacnt == 1); | |
4143 | hdr->b_flags |= ARC_FLAG_BUF_AVAILABLE; | |
4144 | } | |
4145 | ||
4146 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt) || | |
4147 | callback_list != NULL); | |
4148 | ||
4149 | if (zio->io_error != 0) { | |
4150 | hdr->b_flags |= ARC_FLAG_IO_ERROR; | |
4151 | if (hdr->b_l1hdr.b_state != arc_anon) | |
4152 | arc_change_state(arc_anon, hdr, hash_lock); | |
4153 | if (HDR_IN_HASH_TABLE(hdr)) | |
4154 | buf_hash_remove(hdr); | |
4155 | freeable = refcount_is_zero(&hdr->b_l1hdr.b_refcnt); | |
4156 | } | |
4157 | ||
4158 | /* | |
4159 | * Broadcast before we drop the hash_lock to avoid the possibility | |
4160 | * that the hdr (and hence the cv) might be freed before we get to | |
4161 | * the cv_broadcast(). | |
4162 | */ | |
4163 | cv_broadcast(&hdr->b_l1hdr.b_cv); | |
4164 | ||
4165 | if (hash_lock != NULL) { | |
4166 | mutex_exit(hash_lock); | |
4167 | } else { | |
4168 | /* | |
4169 | * This block was freed while we waited for the read to | |
4170 | * complete. It has been removed from the hash table and | |
4171 | * moved to the anonymous state (so that it won't show up | |
4172 | * in the cache). | |
4173 | */ | |
4174 | ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon); | |
4175 | freeable = refcount_is_zero(&hdr->b_l1hdr.b_refcnt); | |
4176 | } | |
4177 | ||
4178 | /* execute each callback and free its structure */ | |
4179 | while ((acb = callback_list) != NULL) { | |
4180 | if (acb->acb_done) | |
4181 | acb->acb_done(zio, acb->acb_buf, acb->acb_private); | |
4182 | ||
4183 | if (acb->acb_zio_dummy != NULL) { | |
4184 | acb->acb_zio_dummy->io_error = zio->io_error; | |
4185 | zio_nowait(acb->acb_zio_dummy); | |
4186 | } | |
4187 | ||
4188 | callback_list = acb->acb_next; | |
4189 | kmem_free(acb, sizeof (arc_callback_t)); | |
4190 | } | |
4191 | ||
4192 | if (freeable) | |
4193 | arc_hdr_destroy(hdr); | |
4194 | } | |
4195 | ||
4196 | /* | |
4197 | * "Read" the block at the specified DVA (in bp) via the | |
4198 | * cache. If the block is found in the cache, invoke the provided | |
4199 | * callback immediately and return. Note that the `zio' parameter | |
4200 | * in the callback will be NULL in this case, since no IO was | |
4201 | * required. If the block is not in the cache pass the read request | |
4202 | * on to the spa with a substitute callback function, so that the | |
4203 | * requested block will be added to the cache. | |
4204 | * | |
4205 | * If a read request arrives for a block that has a read in-progress, | |
4206 | * either wait for the in-progress read to complete (and return the | |
4207 | * results); or, if this is a read with a "done" func, add a record | |
4208 | * to the read to invoke the "done" func when the read completes, | |
4209 | * and return; or just return. | |
4210 | * | |
4211 | * arc_read_done() will invoke all the requested "done" functions | |
4212 | * for readers of this block. | |
4213 | */ | |
4214 | int | |
4215 | arc_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, arc_done_func_t *done, | |
4216 | void *private, zio_priority_t priority, int zio_flags, | |
4217 | arc_flags_t *arc_flags, const zbookmark_phys_t *zb) | |
4218 | { | |
4219 | arc_buf_hdr_t *hdr = NULL; | |
4220 | arc_buf_t *buf = NULL; | |
4221 | kmutex_t *hash_lock = NULL; | |
4222 | zio_t *rzio; | |
4223 | uint64_t guid = spa_load_guid(spa); | |
4224 | int rc = 0; | |
4225 | ||
4226 | ASSERT(!BP_IS_EMBEDDED(bp) || | |
4227 | BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA); | |
4228 | ||
4229 | top: | |
4230 | if (!BP_IS_EMBEDDED(bp)) { | |
4231 | /* | |
4232 | * Embedded BP's have no DVA and require no I/O to "read". | |
4233 | * Create an anonymous arc buf to back it. | |
4234 | */ | |
4235 | hdr = buf_hash_find(guid, bp, &hash_lock); | |
4236 | } | |
4237 | ||
4238 | if (hdr != NULL && HDR_HAS_L1HDR(hdr) && hdr->b_l1hdr.b_datacnt > 0) { | |
4239 | ||
4240 | *arc_flags |= ARC_FLAG_CACHED; | |
4241 | ||
4242 | if (HDR_IO_IN_PROGRESS(hdr)) { | |
4243 | ||
4244 | if (*arc_flags & ARC_FLAG_WAIT) { | |
4245 | cv_wait(&hdr->b_l1hdr.b_cv, hash_lock); | |
4246 | mutex_exit(hash_lock); | |
4247 | goto top; | |
4248 | } | |
4249 | ASSERT(*arc_flags & ARC_FLAG_NOWAIT); | |
4250 | ||
4251 | if (done) { | |
4252 | arc_callback_t *acb = NULL; | |
4253 | ||
4254 | acb = kmem_zalloc(sizeof (arc_callback_t), | |
4255 | KM_SLEEP); | |
4256 | acb->acb_done = done; | |
4257 | acb->acb_private = private; | |
4258 | if (pio != NULL) | |
4259 | acb->acb_zio_dummy = zio_null(pio, | |
4260 | spa, NULL, NULL, NULL, zio_flags); | |
4261 | ||
4262 | ASSERT(acb->acb_done != NULL); | |
4263 | acb->acb_next = hdr->b_l1hdr.b_acb; | |
4264 | hdr->b_l1hdr.b_acb = acb; | |
4265 | add_reference(hdr, hash_lock, private); | |
4266 | mutex_exit(hash_lock); | |
4267 | goto out; | |
4268 | } | |
4269 | mutex_exit(hash_lock); | |
4270 | goto out; | |
4271 | } | |
4272 | ||
4273 | ASSERT(hdr->b_l1hdr.b_state == arc_mru || | |
4274 | hdr->b_l1hdr.b_state == arc_mfu); | |
4275 | ||
4276 | if (done) { | |
4277 | add_reference(hdr, hash_lock, private); | |
4278 | /* | |
4279 | * If this block is already in use, create a new | |
4280 | * copy of the data so that we will be guaranteed | |
4281 | * that arc_release() will always succeed. | |
4282 | */ | |
4283 | buf = hdr->b_l1hdr.b_buf; | |
4284 | ASSERT(buf); | |
4285 | ASSERT(buf->b_data); | |
4286 | if (HDR_BUF_AVAILABLE(hdr)) { | |
4287 | ASSERT(buf->b_efunc == NULL); | |
4288 | hdr->b_flags &= ~ARC_FLAG_BUF_AVAILABLE; | |
4289 | } else { | |
4290 | buf = arc_buf_clone(buf); | |
4291 | } | |
4292 | ||
4293 | } else if (*arc_flags & ARC_FLAG_PREFETCH && | |
4294 | refcount_count(&hdr->b_l1hdr.b_refcnt) == 0) { | |
4295 | hdr->b_flags |= ARC_FLAG_PREFETCH; | |
4296 | } | |
4297 | DTRACE_PROBE1(arc__hit, arc_buf_hdr_t *, hdr); | |
4298 | arc_access(hdr, hash_lock); | |
4299 | if (*arc_flags & ARC_FLAG_L2CACHE) | |
4300 | hdr->b_flags |= ARC_FLAG_L2CACHE; | |
4301 | if (*arc_flags & ARC_FLAG_L2COMPRESS) | |
4302 | hdr->b_flags |= ARC_FLAG_L2COMPRESS; | |
4303 | mutex_exit(hash_lock); | |
4304 | ARCSTAT_BUMP(arcstat_hits); | |
4305 | ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr), | |
4306 | demand, prefetch, !HDR_ISTYPE_METADATA(hdr), | |
4307 | data, metadata, hits); | |
4308 | ||
4309 | if (done) | |
4310 | done(NULL, buf, private); | |
4311 | } else { | |
4312 | uint64_t size = BP_GET_LSIZE(bp); | |
4313 | arc_callback_t *acb; | |
4314 | vdev_t *vd = NULL; | |
4315 | uint64_t addr = 0; | |
4316 | boolean_t devw = B_FALSE; | |
4317 | enum zio_compress b_compress = ZIO_COMPRESS_OFF; | |
4318 | int32_t b_asize = 0; | |
4319 | ||
4320 | /* | |
4321 | * Gracefully handle a damaged logical block size as a | |
4322 | * checksum error. | |
4323 | */ | |
4324 | if (size > spa_maxblocksize(spa)) { | |
4325 | ASSERT3P(buf, ==, NULL); | |
4326 | rc = SET_ERROR(ECKSUM); | |
4327 | goto out; | |
4328 | } | |
4329 | ||
4330 | if (hdr == NULL) { | |
4331 | /* this block is not in the cache */ | |
4332 | arc_buf_hdr_t *exists = NULL; | |
4333 | arc_buf_contents_t type = BP_GET_BUFC_TYPE(bp); | |
4334 | buf = arc_buf_alloc(spa, size, private, type); | |
4335 | hdr = buf->b_hdr; | |
4336 | if (!BP_IS_EMBEDDED(bp)) { | |
4337 | hdr->b_dva = *BP_IDENTITY(bp); | |
4338 | hdr->b_birth = BP_PHYSICAL_BIRTH(bp); | |
4339 | exists = buf_hash_insert(hdr, &hash_lock); | |
4340 | } | |
4341 | if (exists != NULL) { | |
4342 | /* somebody beat us to the hash insert */ | |
4343 | mutex_exit(hash_lock); | |
4344 | buf_discard_identity(hdr); | |
4345 | (void) arc_buf_remove_ref(buf, private); | |
4346 | goto top; /* restart the IO request */ | |
4347 | } | |
4348 | ||
4349 | /* if this is a prefetch, we don't have a reference */ | |
4350 | if (*arc_flags & ARC_FLAG_PREFETCH) { | |
4351 | (void) remove_reference(hdr, hash_lock, | |
4352 | private); | |
4353 | hdr->b_flags |= ARC_FLAG_PREFETCH; | |
4354 | } | |
4355 | if (*arc_flags & ARC_FLAG_L2CACHE) | |
4356 | hdr->b_flags |= ARC_FLAG_L2CACHE; | |
4357 | if (*arc_flags & ARC_FLAG_L2COMPRESS) | |
4358 | hdr->b_flags |= ARC_FLAG_L2COMPRESS; | |
4359 | if (BP_GET_LEVEL(bp) > 0) | |
4360 | hdr->b_flags |= ARC_FLAG_INDIRECT; | |
4361 | } else { | |
4362 | /* | |
4363 | * This block is in the ghost cache. If it was L2-only | |
4364 | * (and thus didn't have an L1 hdr), we realloc the | |
4365 | * header to add an L1 hdr. | |
4366 | */ | |
4367 | if (!HDR_HAS_L1HDR(hdr)) { | |
4368 | hdr = arc_hdr_realloc(hdr, hdr_l2only_cache, | |
4369 | hdr_full_cache); | |
4370 | } | |
4371 | ||
4372 | ASSERT(GHOST_STATE(hdr->b_l1hdr.b_state)); | |
4373 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); | |
4374 | ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); | |
4375 | ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); | |
4376 | ||
4377 | /* if this is a prefetch, we don't have a reference */ | |
4378 | if (*arc_flags & ARC_FLAG_PREFETCH) | |
4379 | hdr->b_flags |= ARC_FLAG_PREFETCH; | |
4380 | else | |
4381 | add_reference(hdr, hash_lock, private); | |
4382 | if (*arc_flags & ARC_FLAG_L2CACHE) | |
4383 | hdr->b_flags |= ARC_FLAG_L2CACHE; | |
4384 | if (*arc_flags & ARC_FLAG_L2COMPRESS) | |
4385 | hdr->b_flags |= ARC_FLAG_L2COMPRESS; | |
4386 | buf = kmem_cache_alloc(buf_cache, KM_PUSHPAGE); | |
4387 | buf->b_hdr = hdr; | |
4388 | buf->b_data = NULL; | |
4389 | buf->b_efunc = NULL; | |
4390 | buf->b_private = NULL; | |
4391 | buf->b_next = NULL; | |
4392 | hdr->b_l1hdr.b_buf = buf; | |
4393 | ASSERT0(hdr->b_l1hdr.b_datacnt); | |
4394 | hdr->b_l1hdr.b_datacnt = 1; | |
4395 | arc_get_data_buf(buf); | |
4396 | arc_access(hdr, hash_lock); | |
4397 | } | |
4398 | ||
4399 | ASSERT(!GHOST_STATE(hdr->b_l1hdr.b_state)); | |
4400 | ||
4401 | acb = kmem_zalloc(sizeof (arc_callback_t), KM_SLEEP); | |
4402 | acb->acb_done = done; | |
4403 | acb->acb_private = private; | |
4404 | ||
4405 | ASSERT(hdr->b_l1hdr.b_acb == NULL); | |
4406 | hdr->b_l1hdr.b_acb = acb; | |
4407 | hdr->b_flags |= ARC_FLAG_IO_IN_PROGRESS; | |
4408 | ||
4409 | if (HDR_HAS_L2HDR(hdr) && | |
4410 | (vd = hdr->b_l2hdr.b_dev->l2ad_vdev) != NULL) { | |
4411 | devw = hdr->b_l2hdr.b_dev->l2ad_writing; | |
4412 | addr = hdr->b_l2hdr.b_daddr; | |
4413 | b_compress = hdr->b_l2hdr.b_compress; | |
4414 | b_asize = hdr->b_l2hdr.b_asize; | |
4415 | /* | |
4416 | * Lock out device removal. | |
4417 | */ | |
4418 | if (vdev_is_dead(vd) || | |
4419 | !spa_config_tryenter(spa, SCL_L2ARC, vd, RW_READER)) | |
4420 | vd = NULL; | |
4421 | } | |
4422 | ||
4423 | if (hash_lock != NULL) | |
4424 | mutex_exit(hash_lock); | |
4425 | ||
4426 | /* | |
4427 | * At this point, we have a level 1 cache miss. Try again in | |
4428 | * L2ARC if possible. | |
4429 | */ | |
4430 | ASSERT3U(hdr->b_size, ==, size); | |
4431 | DTRACE_PROBE4(arc__miss, arc_buf_hdr_t *, hdr, blkptr_t *, bp, | |
4432 | uint64_t, size, zbookmark_phys_t *, zb); | |
4433 | ARCSTAT_BUMP(arcstat_misses); | |
4434 | ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr), | |
4435 | demand, prefetch, !HDR_ISTYPE_METADATA(hdr), | |
4436 | data, metadata, misses); | |
4437 | ||
4438 | if (vd != NULL && l2arc_ndev != 0 && !(l2arc_norw && devw)) { | |
4439 | /* | |
4440 | * Read from the L2ARC if the following are true: | |
4441 | * 1. The L2ARC vdev was previously cached. | |
4442 | * 2. This buffer still has L2ARC metadata. | |
4443 | * 3. This buffer isn't currently writing to the L2ARC. | |
4444 | * 4. The L2ARC entry wasn't evicted, which may | |
4445 | * also have invalidated the vdev. | |
4446 | * 5. This isn't prefetch and l2arc_noprefetch is set. | |
4447 | */ | |
4448 | if (HDR_HAS_L2HDR(hdr) && | |
4449 | !HDR_L2_WRITING(hdr) && !HDR_L2_EVICTED(hdr) && | |
4450 | !(l2arc_noprefetch && HDR_PREFETCH(hdr))) { | |
4451 | l2arc_read_callback_t *cb; | |
4452 | ||
4453 | DTRACE_PROBE1(l2arc__hit, arc_buf_hdr_t *, hdr); | |
4454 | ARCSTAT_BUMP(arcstat_l2_hits); | |
4455 | atomic_inc_32(&hdr->b_l2hdr.b_hits); | |
4456 | ||
4457 | cb = kmem_zalloc(sizeof (l2arc_read_callback_t), | |
4458 | KM_SLEEP); | |
4459 | cb->l2rcb_buf = buf; | |
4460 | cb->l2rcb_spa = spa; | |
4461 | cb->l2rcb_bp = *bp; | |
4462 | cb->l2rcb_zb = *zb; | |
4463 | cb->l2rcb_flags = zio_flags; | |
4464 | cb->l2rcb_compress = b_compress; | |
4465 | ||
4466 | ASSERT(addr >= VDEV_LABEL_START_SIZE && | |
4467 | addr + size < vd->vdev_psize - | |
4468 | VDEV_LABEL_END_SIZE); | |
4469 | ||
4470 | /* | |
4471 | * l2arc read. The SCL_L2ARC lock will be | |
4472 | * released by l2arc_read_done(). | |
4473 | * Issue a null zio if the underlying buffer | |
4474 | * was squashed to zero size by compression. | |
4475 | */ | |
4476 | if (b_compress == ZIO_COMPRESS_EMPTY) { | |
4477 | rzio = zio_null(pio, spa, vd, | |
4478 | l2arc_read_done, cb, | |
4479 | zio_flags | ZIO_FLAG_DONT_CACHE | | |
4480 | ZIO_FLAG_CANFAIL | | |
4481 | ZIO_FLAG_DONT_PROPAGATE | | |
4482 | ZIO_FLAG_DONT_RETRY); | |
4483 | } else { | |
4484 | rzio = zio_read_phys(pio, vd, addr, | |
4485 | b_asize, buf->b_data, | |
4486 | ZIO_CHECKSUM_OFF, | |
4487 | l2arc_read_done, cb, priority, | |
4488 | zio_flags | ZIO_FLAG_DONT_CACHE | | |
4489 | ZIO_FLAG_CANFAIL | | |
4490 | ZIO_FLAG_DONT_PROPAGATE | | |
4491 | ZIO_FLAG_DONT_RETRY, B_FALSE); | |
4492 | } | |
4493 | DTRACE_PROBE2(l2arc__read, vdev_t *, vd, | |
4494 | zio_t *, rzio); | |
4495 | ARCSTAT_INCR(arcstat_l2_read_bytes, b_asize); | |
4496 | ||
4497 | if (*arc_flags & ARC_FLAG_NOWAIT) { | |
4498 | zio_nowait(rzio); | |
4499 | goto out; | |
4500 | } | |
4501 | ||
4502 | ASSERT(*arc_flags & ARC_FLAG_WAIT); | |
4503 | if (zio_wait(rzio) == 0) | |
4504 | goto out; | |
4505 | ||
4506 | /* l2arc read error; goto zio_read() */ | |
4507 | } else { | |
4508 | DTRACE_PROBE1(l2arc__miss, | |
4509 | arc_buf_hdr_t *, hdr); | |
4510 | ARCSTAT_BUMP(arcstat_l2_misses); | |
4511 | if (HDR_L2_WRITING(hdr)) | |
4512 | ARCSTAT_BUMP(arcstat_l2_rw_clash); | |
4513 | spa_config_exit(spa, SCL_L2ARC, vd); | |
4514 | } | |
4515 | } else { | |
4516 | if (vd != NULL) | |
4517 | spa_config_exit(spa, SCL_L2ARC, vd); | |
4518 | if (l2arc_ndev != 0) { | |
4519 | DTRACE_PROBE1(l2arc__miss, | |
4520 | arc_buf_hdr_t *, hdr); | |
4521 | ARCSTAT_BUMP(arcstat_l2_misses); | |
4522 | } | |
4523 | } | |
4524 | ||
4525 | rzio = zio_read(pio, spa, bp, buf->b_data, size, | |
4526 | arc_read_done, buf, priority, zio_flags, zb); | |
4527 | ||
4528 | if (*arc_flags & ARC_FLAG_WAIT) { | |
4529 | rc = zio_wait(rzio); | |
4530 | goto out; | |
4531 | } | |
4532 | ||
4533 | ASSERT(*arc_flags & ARC_FLAG_NOWAIT); | |
4534 | zio_nowait(rzio); | |
4535 | } | |
4536 | ||
4537 | out: | |
4538 | spa_read_history_add(spa, zb, *arc_flags); | |
4539 | return (rc); | |
4540 | } | |
4541 | ||
4542 | arc_prune_t * | |
4543 | arc_add_prune_callback(arc_prune_func_t *func, void *private) | |
4544 | { | |
4545 | arc_prune_t *p; | |
4546 | ||
4547 | p = kmem_alloc(sizeof (*p), KM_SLEEP); | |
4548 | p->p_pfunc = func; | |
4549 | p->p_private = private; | |
4550 | list_link_init(&p->p_node); | |
4551 | refcount_create(&p->p_refcnt); | |
4552 | ||
4553 | mutex_enter(&arc_prune_mtx); | |
4554 | refcount_add(&p->p_refcnt, &arc_prune_list); | |
4555 | list_insert_head(&arc_prune_list, p); | |
4556 | mutex_exit(&arc_prune_mtx); | |
4557 | ||
4558 | return (p); | |
4559 | } | |
4560 | ||
4561 | void | |
4562 | arc_remove_prune_callback(arc_prune_t *p) | |
4563 | { | |
4564 | boolean_t wait = B_FALSE; | |
4565 | mutex_enter(&arc_prune_mtx); | |
4566 | list_remove(&arc_prune_list, p); | |
4567 | if (refcount_remove(&p->p_refcnt, &arc_prune_list) > 0) | |
4568 | wait = B_TRUE; | |
4569 | mutex_exit(&arc_prune_mtx); | |
4570 | ||
4571 | /* wait for arc_prune_task to finish */ | |
4572 | if (wait) | |
4573 | taskq_wait_outstanding(arc_prune_taskq, 0); | |
4574 | ASSERT0(refcount_count(&p->p_refcnt)); | |
4575 | refcount_destroy(&p->p_refcnt); | |
4576 | kmem_free(p, sizeof (*p)); | |
4577 | } | |
4578 | ||
4579 | void | |
4580 | arc_set_callback(arc_buf_t *buf, arc_evict_func_t *func, void *private) | |
4581 | { | |
4582 | ASSERT(buf->b_hdr != NULL); | |
4583 | ASSERT(buf->b_hdr->b_l1hdr.b_state != arc_anon); | |
4584 | ASSERT(!refcount_is_zero(&buf->b_hdr->b_l1hdr.b_refcnt) || | |
4585 | func == NULL); | |
4586 | ASSERT(buf->b_efunc == NULL); | |
4587 | ASSERT(!HDR_BUF_AVAILABLE(buf->b_hdr)); | |
4588 | ||
4589 | buf->b_efunc = func; | |
4590 | buf->b_private = private; | |
4591 | } | |
4592 | ||
4593 | /* | |
4594 | * Notify the arc that a block was freed, and thus will never be used again. | |
4595 | */ | |
4596 | void | |
4597 | arc_freed(spa_t *spa, const blkptr_t *bp) | |
4598 | { | |
4599 | arc_buf_hdr_t *hdr; | |
4600 | kmutex_t *hash_lock; | |
4601 | uint64_t guid = spa_load_guid(spa); | |
4602 | ||
4603 | ASSERT(!BP_IS_EMBEDDED(bp)); | |
4604 | ||
4605 | hdr = buf_hash_find(guid, bp, &hash_lock); | |
4606 | if (hdr == NULL) | |
4607 | return; | |
4608 | if (HDR_BUF_AVAILABLE(hdr)) { | |
4609 | arc_buf_t *buf = hdr->b_l1hdr.b_buf; | |
4610 | add_reference(hdr, hash_lock, FTAG); | |
4611 | hdr->b_flags &= ~ARC_FLAG_BUF_AVAILABLE; | |
4612 | mutex_exit(hash_lock); | |
4613 | ||
4614 | arc_release(buf, FTAG); | |
4615 | (void) arc_buf_remove_ref(buf, FTAG); | |
4616 | } else { | |
4617 | mutex_exit(hash_lock); | |
4618 | } | |
4619 | ||
4620 | } | |
4621 | ||
4622 | /* | |
4623 | * Clear the user eviction callback set by arc_set_callback(), first calling | |
4624 | * it if it exists. Because the presence of a callback keeps an arc_buf cached | |
4625 | * clearing the callback may result in the arc_buf being destroyed. However, | |
4626 | * it will not result in the *last* arc_buf being destroyed, hence the data | |
4627 | * will remain cached in the ARC. We make a copy of the arc buffer here so | |
4628 | * that we can process the callback without holding any locks. | |
4629 | * | |
4630 | * It's possible that the callback is already in the process of being cleared | |
4631 | * by another thread. In this case we can not clear the callback. | |
4632 | * | |
4633 | * Returns B_TRUE if the callback was successfully called and cleared. | |
4634 | */ | |
4635 | boolean_t | |
4636 | arc_clear_callback(arc_buf_t *buf) | |
4637 | { | |
4638 | arc_buf_hdr_t *hdr; | |
4639 | kmutex_t *hash_lock; | |
4640 | arc_evict_func_t *efunc = buf->b_efunc; | |
4641 | void *private = buf->b_private; | |
4642 | ||
4643 | mutex_enter(&buf->b_evict_lock); | |
4644 | hdr = buf->b_hdr; | |
4645 | if (hdr == NULL) { | |
4646 | /* | |
4647 | * We are in arc_do_user_evicts(). | |
4648 | */ | |
4649 | ASSERT(buf->b_data == NULL); | |
4650 | mutex_exit(&buf->b_evict_lock); | |
4651 | return (B_FALSE); | |
4652 | } else if (buf->b_data == NULL) { | |
4653 | /* | |
4654 | * We are on the eviction list; process this buffer now | |
4655 | * but let arc_do_user_evicts() do the reaping. | |
4656 | */ | |
4657 | buf->b_efunc = NULL; | |
4658 | mutex_exit(&buf->b_evict_lock); | |
4659 | VERIFY0(efunc(private)); | |
4660 | return (B_TRUE); | |
4661 | } | |
4662 | hash_lock = HDR_LOCK(hdr); | |
4663 | mutex_enter(hash_lock); | |
4664 | hdr = buf->b_hdr; | |
4665 | ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); | |
4666 | ||
4667 | ASSERT3U(refcount_count(&hdr->b_l1hdr.b_refcnt), <, | |
4668 | hdr->b_l1hdr.b_datacnt); | |
4669 | ASSERT(hdr->b_l1hdr.b_state == arc_mru || | |
4670 | hdr->b_l1hdr.b_state == arc_mfu); | |
4671 | ||
4672 | buf->b_efunc = NULL; | |
4673 | buf->b_private = NULL; | |
4674 | ||
4675 | if (hdr->b_l1hdr.b_datacnt > 1) { | |
4676 | mutex_exit(&buf->b_evict_lock); | |
4677 | arc_buf_destroy(buf, TRUE); | |
4678 | } else { | |
4679 | ASSERT(buf == hdr->b_l1hdr.b_buf); | |
4680 | hdr->b_flags |= ARC_FLAG_BUF_AVAILABLE; | |
4681 | mutex_exit(&buf->b_evict_lock); | |
4682 | } | |
4683 | ||
4684 | mutex_exit(hash_lock); | |
4685 | VERIFY0(efunc(private)); | |
4686 | return (B_TRUE); | |
4687 | } | |
4688 | ||
4689 | /* | |
4690 | * Release this buffer from the cache, making it an anonymous buffer. This | |
4691 | * must be done after a read and prior to modifying the buffer contents. | |
4692 | * If the buffer has more than one reference, we must make | |
4693 | * a new hdr for the buffer. | |
4694 | */ | |
4695 | void | |
4696 | arc_release(arc_buf_t *buf, void *tag) | |
4697 | { | |
4698 | kmutex_t *hash_lock; | |
4699 | arc_state_t *state; | |
4700 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
4701 | ||
4702 | /* | |
4703 | * It would be nice to assert that if its DMU metadata (level > | |
4704 | * 0 || it's the dnode file), then it must be syncing context. | |
4705 | * But we don't know that information at this level. | |
4706 | */ | |
4707 | ||
4708 | mutex_enter(&buf->b_evict_lock); | |
4709 | ||
4710 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
4711 | ||
4712 | /* | |
4713 | * We don't grab the hash lock prior to this check, because if | |
4714 | * the buffer's header is in the arc_anon state, it won't be | |
4715 | * linked into the hash table. | |
4716 | */ | |
4717 | if (hdr->b_l1hdr.b_state == arc_anon) { | |
4718 | mutex_exit(&buf->b_evict_lock); | |
4719 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); | |
4720 | ASSERT(!HDR_IN_HASH_TABLE(hdr)); | |
4721 | ASSERT(!HDR_HAS_L2HDR(hdr)); | |
4722 | ASSERT(BUF_EMPTY(hdr)); | |
4723 | ||
4724 | ASSERT3U(hdr->b_l1hdr.b_datacnt, ==, 1); | |
4725 | ASSERT3S(refcount_count(&hdr->b_l1hdr.b_refcnt), ==, 1); | |
4726 | ASSERT(!list_link_active(&hdr->b_l1hdr.b_arc_node)); | |
4727 | ||
4728 | ASSERT3P(buf->b_efunc, ==, NULL); | |
4729 | ASSERT3P(buf->b_private, ==, NULL); | |
4730 | ||
4731 | hdr->b_l1hdr.b_arc_access = 0; | |
4732 | arc_buf_thaw(buf); | |
4733 | ||
4734 | return; | |
4735 | } | |
4736 | ||
4737 | hash_lock = HDR_LOCK(hdr); | |
4738 | mutex_enter(hash_lock); | |
4739 | ||
4740 | /* | |
4741 | * This assignment is only valid as long as the hash_lock is | |
4742 | * held, we must be careful not to reference state or the | |
4743 | * b_state field after dropping the lock. | |
4744 | */ | |
4745 | state = hdr->b_l1hdr.b_state; | |
4746 | ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); | |
4747 | ASSERT3P(state, !=, arc_anon); | |
4748 | ||
4749 | /* this buffer is not on any list */ | |
4750 | ASSERT(refcount_count(&hdr->b_l1hdr.b_refcnt) > 0); | |
4751 | ||
4752 | if (HDR_HAS_L2HDR(hdr)) { | |
4753 | mutex_enter(&hdr->b_l2hdr.b_dev->l2ad_mtx); | |
4754 | ||
4755 | /* | |
4756 | * We have to recheck this conditional again now that | |
4757 | * we're holding the l2ad_mtx to prevent a race with | |
4758 | * another thread which might be concurrently calling | |
4759 | * l2arc_evict(). In that case, l2arc_evict() might have | |
4760 | * destroyed the header's L2 portion as we were waiting | |
4761 | * to acquire the l2ad_mtx. | |
4762 | */ | |
4763 | if (HDR_HAS_L2HDR(hdr)) | |
4764 | arc_hdr_l2hdr_destroy(hdr); | |
4765 | ||
4766 | mutex_exit(&hdr->b_l2hdr.b_dev->l2ad_mtx); | |
4767 | } | |
4768 | ||
4769 | /* | |
4770 | * Do we have more than one buf? | |
4771 | */ | |
4772 | if (hdr->b_l1hdr.b_datacnt > 1) { | |
4773 | arc_buf_hdr_t *nhdr; | |
4774 | arc_buf_t **bufp; | |
4775 | uint64_t blksz = hdr->b_size; | |
4776 | uint64_t spa = hdr->b_spa; | |
4777 | arc_buf_contents_t type = arc_buf_type(hdr); | |
4778 | uint32_t flags = hdr->b_flags; | |
4779 | ||
4780 | ASSERT(hdr->b_l1hdr.b_buf != buf || buf->b_next != NULL); | |
4781 | /* | |
4782 | * Pull the data off of this hdr and attach it to | |
4783 | * a new anonymous hdr. | |
4784 | */ | |
4785 | (void) remove_reference(hdr, hash_lock, tag); | |
4786 | bufp = &hdr->b_l1hdr.b_buf; | |
4787 | while (*bufp != buf) | |
4788 | bufp = &(*bufp)->b_next; | |
4789 | *bufp = buf->b_next; | |
4790 | buf->b_next = NULL; | |
4791 | ||
4792 | ASSERT3P(state, !=, arc_l2c_only); | |
4793 | ||
4794 | (void) refcount_remove_many( | |
4795 | &state->arcs_size, hdr->b_size, buf); | |
4796 | ||
4797 | if (refcount_is_zero(&hdr->b_l1hdr.b_refcnt)) { | |
4798 | uint64_t *size; | |
4799 | ||
4800 | ASSERT3P(state, !=, arc_l2c_only); | |
4801 | size = &state->arcs_lsize[type]; | |
4802 | ASSERT3U(*size, >=, hdr->b_size); | |
4803 | atomic_add_64(size, -hdr->b_size); | |
4804 | } | |
4805 | ||
4806 | /* | |
4807 | * We're releasing a duplicate user data buffer, update | |
4808 | * our statistics accordingly. | |
4809 | */ | |
4810 | if (HDR_ISTYPE_DATA(hdr)) { | |
4811 | ARCSTAT_BUMPDOWN(arcstat_duplicate_buffers); | |
4812 | ARCSTAT_INCR(arcstat_duplicate_buffers_size, | |
4813 | -hdr->b_size); | |
4814 | } | |
4815 | hdr->b_l1hdr.b_datacnt -= 1; | |
4816 | arc_cksum_verify(buf); | |
4817 | arc_buf_unwatch(buf); | |
4818 | ||
4819 | mutex_exit(hash_lock); | |
4820 | ||
4821 | nhdr = kmem_cache_alloc(hdr_full_cache, KM_PUSHPAGE); | |
4822 | nhdr->b_size = blksz; | |
4823 | nhdr->b_spa = spa; | |
4824 | ||
4825 | nhdr->b_l1hdr.b_mru_hits = 0; | |
4826 | nhdr->b_l1hdr.b_mru_ghost_hits = 0; | |
4827 | nhdr->b_l1hdr.b_mfu_hits = 0; | |
4828 | nhdr->b_l1hdr.b_mfu_ghost_hits = 0; | |
4829 | nhdr->b_l1hdr.b_l2_hits = 0; | |
4830 | nhdr->b_flags = flags & ARC_FLAG_L2_WRITING; | |
4831 | nhdr->b_flags |= arc_bufc_to_flags(type); | |
4832 | nhdr->b_flags |= ARC_FLAG_HAS_L1HDR; | |
4833 | ||
4834 | nhdr->b_l1hdr.b_buf = buf; | |
4835 | nhdr->b_l1hdr.b_datacnt = 1; | |
4836 | nhdr->b_l1hdr.b_state = arc_anon; | |
4837 | nhdr->b_l1hdr.b_arc_access = 0; | |
4838 | nhdr->b_l1hdr.b_tmp_cdata = NULL; | |
4839 | nhdr->b_freeze_cksum = NULL; | |
4840 | ||
4841 | (void) refcount_add(&nhdr->b_l1hdr.b_refcnt, tag); | |
4842 | buf->b_hdr = nhdr; | |
4843 | mutex_exit(&buf->b_evict_lock); | |
4844 | (void) refcount_add_many(&arc_anon->arcs_size, blksz, buf); | |
4845 | } else { | |
4846 | mutex_exit(&buf->b_evict_lock); | |
4847 | ASSERT(refcount_count(&hdr->b_l1hdr.b_refcnt) == 1); | |
4848 | /* protected by hash lock, or hdr is on arc_anon */ | |
4849 | ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); | |
4850 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); | |
4851 | hdr->b_l1hdr.b_mru_hits = 0; | |
4852 | hdr->b_l1hdr.b_mru_ghost_hits = 0; | |
4853 | hdr->b_l1hdr.b_mfu_hits = 0; | |
4854 | hdr->b_l1hdr.b_mfu_ghost_hits = 0; | |
4855 | hdr->b_l1hdr.b_l2_hits = 0; | |
4856 | arc_change_state(arc_anon, hdr, hash_lock); | |
4857 | hdr->b_l1hdr.b_arc_access = 0; | |
4858 | mutex_exit(hash_lock); | |
4859 | ||
4860 | buf_discard_identity(hdr); | |
4861 | arc_buf_thaw(buf); | |
4862 | } | |
4863 | buf->b_efunc = NULL; | |
4864 | buf->b_private = NULL; | |
4865 | } | |
4866 | ||
4867 | int | |
4868 | arc_released(arc_buf_t *buf) | |
4869 | { | |
4870 | int released; | |
4871 | ||
4872 | mutex_enter(&buf->b_evict_lock); | |
4873 | released = (buf->b_data != NULL && | |
4874 | buf->b_hdr->b_l1hdr.b_state == arc_anon); | |
4875 | mutex_exit(&buf->b_evict_lock); | |
4876 | return (released); | |
4877 | } | |
4878 | ||
4879 | #ifdef ZFS_DEBUG | |
4880 | int | |
4881 | arc_referenced(arc_buf_t *buf) | |
4882 | { | |
4883 | int referenced; | |
4884 | ||
4885 | mutex_enter(&buf->b_evict_lock); | |
4886 | referenced = (refcount_count(&buf->b_hdr->b_l1hdr.b_refcnt)); | |
4887 | mutex_exit(&buf->b_evict_lock); | |
4888 | return (referenced); | |
4889 | } | |
4890 | #endif | |
4891 | ||
4892 | static void | |
4893 | arc_write_ready(zio_t *zio) | |
4894 | { | |
4895 | arc_write_callback_t *callback = zio->io_private; | |
4896 | arc_buf_t *buf = callback->awcb_buf; | |
4897 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
4898 | ||
4899 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
4900 | ASSERT(!refcount_is_zero(&buf->b_hdr->b_l1hdr.b_refcnt)); | |
4901 | ASSERT(hdr->b_l1hdr.b_datacnt > 0); | |
4902 | callback->awcb_ready(zio, buf, callback->awcb_private); | |
4903 | ||
4904 | /* | |
4905 | * If the IO is already in progress, then this is a re-write | |
4906 | * attempt, so we need to thaw and re-compute the cksum. | |
4907 | * It is the responsibility of the callback to handle the | |
4908 | * accounting for any re-write attempt. | |
4909 | */ | |
4910 | if (HDR_IO_IN_PROGRESS(hdr)) { | |
4911 | mutex_enter(&hdr->b_l1hdr.b_freeze_lock); | |
4912 | if (hdr->b_freeze_cksum != NULL) { | |
4913 | kmem_free(hdr->b_freeze_cksum, sizeof (zio_cksum_t)); | |
4914 | hdr->b_freeze_cksum = NULL; | |
4915 | } | |
4916 | mutex_exit(&hdr->b_l1hdr.b_freeze_lock); | |
4917 | } | |
4918 | arc_cksum_compute(buf, B_FALSE); | |
4919 | hdr->b_flags |= ARC_FLAG_IO_IN_PROGRESS; | |
4920 | } | |
4921 | ||
4922 | /* | |
4923 | * The SPA calls this callback for each physical write that happens on behalf | |
4924 | * of a logical write. See the comment in dbuf_write_physdone() for details. | |
4925 | */ | |
4926 | static void | |
4927 | arc_write_physdone(zio_t *zio) | |
4928 | { | |
4929 | arc_write_callback_t *cb = zio->io_private; | |
4930 | if (cb->awcb_physdone != NULL) | |
4931 | cb->awcb_physdone(zio, cb->awcb_buf, cb->awcb_private); | |
4932 | } | |
4933 | ||
4934 | static void | |
4935 | arc_write_done(zio_t *zio) | |
4936 | { | |
4937 | arc_write_callback_t *callback = zio->io_private; | |
4938 | arc_buf_t *buf = callback->awcb_buf; | |
4939 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
4940 | ||
4941 | ASSERT(hdr->b_l1hdr.b_acb == NULL); | |
4942 | ||
4943 | if (zio->io_error == 0) { | |
4944 | if (BP_IS_HOLE(zio->io_bp) || BP_IS_EMBEDDED(zio->io_bp)) { | |
4945 | buf_discard_identity(hdr); | |
4946 | } else { | |
4947 | hdr->b_dva = *BP_IDENTITY(zio->io_bp); | |
4948 | hdr->b_birth = BP_PHYSICAL_BIRTH(zio->io_bp); | |
4949 | } | |
4950 | } else { | |
4951 | ASSERT(BUF_EMPTY(hdr)); | |
4952 | } | |
4953 | ||
4954 | /* | |
4955 | * If the block to be written was all-zero or compressed enough to be | |
4956 | * embedded in the BP, no write was performed so there will be no | |
4957 | * dva/birth/checksum. The buffer must therefore remain anonymous | |
4958 | * (and uncached). | |
4959 | */ | |
4960 | if (!BUF_EMPTY(hdr)) { | |
4961 | arc_buf_hdr_t *exists; | |
4962 | kmutex_t *hash_lock; | |
4963 | ||
4964 | ASSERT(zio->io_error == 0); | |
4965 | ||
4966 | arc_cksum_verify(buf); | |
4967 | ||
4968 | exists = buf_hash_insert(hdr, &hash_lock); | |
4969 | if (exists != NULL) { | |
4970 | /* | |
4971 | * This can only happen if we overwrite for | |
4972 | * sync-to-convergence, because we remove | |
4973 | * buffers from the hash table when we arc_free(). | |
4974 | */ | |
4975 | if (zio->io_flags & ZIO_FLAG_IO_REWRITE) { | |
4976 | if (!BP_EQUAL(&zio->io_bp_orig, zio->io_bp)) | |
4977 | panic("bad overwrite, hdr=%p exists=%p", | |
4978 | (void *)hdr, (void *)exists); | |
4979 | ASSERT(refcount_is_zero( | |
4980 | &exists->b_l1hdr.b_refcnt)); | |
4981 | arc_change_state(arc_anon, exists, hash_lock); | |
4982 | mutex_exit(hash_lock); | |
4983 | arc_hdr_destroy(exists); | |
4984 | exists = buf_hash_insert(hdr, &hash_lock); | |
4985 | ASSERT3P(exists, ==, NULL); | |
4986 | } else if (zio->io_flags & ZIO_FLAG_NOPWRITE) { | |
4987 | /* nopwrite */ | |
4988 | ASSERT(zio->io_prop.zp_nopwrite); | |
4989 | if (!BP_EQUAL(&zio->io_bp_orig, zio->io_bp)) | |
4990 | panic("bad nopwrite, hdr=%p exists=%p", | |
4991 | (void *)hdr, (void *)exists); | |
4992 | } else { | |
4993 | /* Dedup */ | |
4994 | ASSERT(hdr->b_l1hdr.b_datacnt == 1); | |
4995 | ASSERT(hdr->b_l1hdr.b_state == arc_anon); | |
4996 | ASSERT(BP_GET_DEDUP(zio->io_bp)); | |
4997 | ASSERT(BP_GET_LEVEL(zio->io_bp) == 0); | |
4998 | } | |
4999 | } | |
5000 | hdr->b_flags &= ~ARC_FLAG_IO_IN_PROGRESS; | |
5001 | /* if it's not anon, we are doing a scrub */ | |
5002 | if (exists == NULL && hdr->b_l1hdr.b_state == arc_anon) | |
5003 | arc_access(hdr, hash_lock); | |
5004 | mutex_exit(hash_lock); | |
5005 | } else { | |
5006 | hdr->b_flags &= ~ARC_FLAG_IO_IN_PROGRESS; | |
5007 | } | |
5008 | ||
5009 | ASSERT(!refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); | |
5010 | callback->awcb_done(zio, buf, callback->awcb_private); | |
5011 | ||
5012 | kmem_free(callback, sizeof (arc_write_callback_t)); | |
5013 | } | |
5014 | ||
5015 | zio_t * | |
5016 | arc_write(zio_t *pio, spa_t *spa, uint64_t txg, | |
5017 | blkptr_t *bp, arc_buf_t *buf, boolean_t l2arc, boolean_t l2arc_compress, | |
5018 | const zio_prop_t *zp, arc_done_func_t *ready, arc_done_func_t *physdone, | |
5019 | arc_done_func_t *done, void *private, zio_priority_t priority, | |
5020 | int zio_flags, const zbookmark_phys_t *zb) | |
5021 | { | |
5022 | arc_buf_hdr_t *hdr = buf->b_hdr; | |
5023 | arc_write_callback_t *callback; | |
5024 | zio_t *zio; | |
5025 | ||
5026 | ASSERT(ready != NULL); | |
5027 | ASSERT(done != NULL); | |
5028 | ASSERT(!HDR_IO_ERROR(hdr)); | |
5029 | ASSERT(!HDR_IO_IN_PROGRESS(hdr)); | |
5030 | ASSERT(hdr->b_l1hdr.b_acb == NULL); | |
5031 | ASSERT(hdr->b_l1hdr.b_datacnt > 0); | |
5032 | if (l2arc) | |
5033 | hdr->b_flags |= ARC_FLAG_L2CACHE; | |
5034 | if (l2arc_compress) | |
5035 | hdr->b_flags |= ARC_FLAG_L2COMPRESS; | |
5036 | callback = kmem_zalloc(sizeof (arc_write_callback_t), KM_SLEEP); | |
5037 | callback->awcb_ready = ready; | |
5038 | callback->awcb_physdone = physdone; | |
5039 | callback->awcb_done = done; | |
5040 | callback->awcb_private = private; | |
5041 | callback->awcb_buf = buf; | |
5042 | ||
5043 | zio = zio_write(pio, spa, txg, bp, buf->b_data, hdr->b_size, zp, | |
5044 | arc_write_ready, arc_write_physdone, arc_write_done, callback, | |
5045 | priority, zio_flags, zb); | |
5046 | ||
5047 | return (zio); | |
5048 | } | |
5049 | ||
5050 | static int | |
5051 | arc_memory_throttle(uint64_t reserve, uint64_t txg) | |
5052 | { | |
5053 | #ifdef _KERNEL | |
5054 | uint64_t available_memory = ptob(freemem); | |
5055 | static uint64_t page_load = 0; | |
5056 | static uint64_t last_txg = 0; | |
5057 | #ifdef __linux__ | |
5058 | pgcnt_t minfree = btop(arc_sys_free / 4); | |
5059 | #endif | |
5060 | ||
5061 | if (freemem > physmem * arc_lotsfree_percent / 100) | |
5062 | return (0); | |
5063 | ||
5064 | if (txg > last_txg) { | |
5065 | last_txg = txg; | |
5066 | page_load = 0; | |
5067 | } | |
5068 | ||
5069 | /* | |
5070 | * If we are in pageout, we know that memory is already tight, | |
5071 | * the arc is already going to be evicting, so we just want to | |
5072 | * continue to let page writes occur as quickly as possible. | |
5073 | */ | |
5074 | if (current_is_kswapd()) { | |
5075 | if (page_load > MAX(ptob(minfree), available_memory) / 4) { | |
5076 | DMU_TX_STAT_BUMP(dmu_tx_memory_reclaim); | |
5077 | return (SET_ERROR(ERESTART)); | |
5078 | } | |
5079 | /* Note: reserve is inflated, so we deflate */ | |
5080 | page_load += reserve / 8; | |
5081 | return (0); | |
5082 | } else if (page_load > 0 && arc_reclaim_needed()) { | |
5083 | /* memory is low, delay before restarting */ | |
5084 | ARCSTAT_INCR(arcstat_memory_throttle_count, 1); | |
5085 | DMU_TX_STAT_BUMP(dmu_tx_memory_reclaim); | |
5086 | return (SET_ERROR(EAGAIN)); | |
5087 | } | |
5088 | page_load = 0; | |
5089 | #endif | |
5090 | return (0); | |
5091 | } | |
5092 | ||
5093 | void | |
5094 | arc_tempreserve_clear(uint64_t reserve) | |
5095 | { | |
5096 | atomic_add_64(&arc_tempreserve, -reserve); | |
5097 | ASSERT((int64_t)arc_tempreserve >= 0); | |
5098 | } | |
5099 | ||
5100 | int | |
5101 | arc_tempreserve_space(uint64_t reserve, uint64_t txg) | |
5102 | { | |
5103 | int error; | |
5104 | uint64_t anon_size; | |
5105 | ||
5106 | if (!arc_no_grow && | |
5107 | reserve > arc_c/4 && | |
5108 | reserve * 4 > (2ULL << SPA_MAXBLOCKSHIFT)) | |
5109 | arc_c = MIN(arc_c_max, reserve * 4); | |
5110 | ||
5111 | /* | |
5112 | * Throttle when the calculated memory footprint for the TXG | |
5113 | * exceeds the target ARC size. | |
5114 | */ | |
5115 | if (reserve > arc_c) { | |
5116 | DMU_TX_STAT_BUMP(dmu_tx_memory_reserve); | |
5117 | return (SET_ERROR(ERESTART)); | |
5118 | } | |
5119 | ||
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 | */ | |
5125 | anon_size = MAX((int64_t)(refcount_count(&arc_anon->arcs_size) - | |
5126 | arc_loaned_bytes), 0); | |
5127 | ||
5128 | /* | |
5129 | * Writes will, almost always, require additional memory allocations | |
5130 | * in order to compress/encrypt/etc the data. We therefore need to | |
5131 | * make sure that there is sufficient available memory for this. | |
5132 | */ | |
5133 | error = arc_memory_throttle(reserve, txg); | |
5134 | if (error != 0) | |
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 | */ | |
5144 | ||
5145 | if (reserve + arc_tempreserve + anon_size > arc_c / 2 && | |
5146 | anon_size > arc_c / 4) { | |
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); | |
5153 | DMU_TX_STAT_BUMP(dmu_tx_dirty_throttle); | |
5154 | return (SET_ERROR(ERESTART)); | |
5155 | } | |
5156 | atomic_add_64(&arc_tempreserve, reserve); | |
5157 | return (0); | |
5158 | } | |
5159 | ||
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 | { | |
5164 | size->value.ui64 = refcount_count(&state->arcs_size); | |
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) { | |
5175 | return (EACCES); | |
5176 | } else { | |
5177 | arc_kstat_update_state(arc_anon, | |
5178 | &as->arcstat_anon_size, | |
5179 | &as->arcstat_anon_evictable_data, | |
5180 | &as->arcstat_anon_evictable_metadata); | |
5181 | arc_kstat_update_state(arc_mru, | |
5182 | &as->arcstat_mru_size, | |
5183 | &as->arcstat_mru_evictable_data, | |
5184 | &as->arcstat_mru_evictable_metadata); | |
5185 | arc_kstat_update_state(arc_mru_ghost, | |
5186 | &as->arcstat_mru_ghost_size, | |
5187 | &as->arcstat_mru_ghost_evictable_data, | |
5188 | &as->arcstat_mru_ghost_evictable_metadata); | |
5189 | arc_kstat_update_state(arc_mfu, | |
5190 | &as->arcstat_mfu_size, | |
5191 | &as->arcstat_mfu_evictable_data, | |
5192 | &as->arcstat_mfu_evictable_metadata); | |
5193 | arc_kstat_update_state(arc_mfu_ghost, | |
5194 | &as->arcstat_mfu_ghost_size, | |
5195 | &as->arcstat_mfu_ghost_evictable_data, | |
5196 | &as->arcstat_mfu_ghost_evictable_metadata); | |
5197 | } | |
5198 | ||
5199 | return (0); | |
5200 | } | |
5201 | ||
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 | ||
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, (3 * arc_c_max) / 4); | |
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 | ||
5287 | /* Valid range: 1 - N */ | |
5288 | if (zfs_arc_p_min_shift) | |
5289 | arc_p_min_shift = zfs_arc_p_min_shift; | |
5290 | ||
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 | /* Valid range: 0 - 100 */ | |
5296 | if ((zfs_arc_lotsfree_percent >= 0) && | |
5297 | (zfs_arc_lotsfree_percent <= 100)) | |
5298 | arc_lotsfree_percent = zfs_arc_lotsfree_percent; | |
5299 | ||
5300 | /* Valid range: 0 - <all physical memory> */ | |
5301 | if ((zfs_arc_sys_free) && (zfs_arc_sys_free != arc_sys_free)) | |
5302 | arc_sys_free = MIN(MAX(zfs_arc_sys_free, 0), ptob(physmem)); | |
5303 | ||
5304 | } | |
5305 | ||
5306 | void | |
5307 | arc_init(void) | |
5308 | { | |
5309 | /* | |
5310 | * allmem is "all memory that we could possibly use". | |
5311 | */ | |
5312 | #ifdef _KERNEL | |
5313 | uint64_t allmem = ptob(physmem); | |
5314 | #else | |
5315 | uint64_t allmem = (physmem * PAGESIZE) / 2; | |
5316 | #endif | |
5317 | ||
5318 | mutex_init(&arc_reclaim_lock, NULL, MUTEX_DEFAULT, NULL); | |
5319 | cv_init(&arc_reclaim_thread_cv, NULL, CV_DEFAULT, NULL); | |
5320 | cv_init(&arc_reclaim_waiters_cv, NULL, CV_DEFAULT, NULL); | |
5321 | ||
5322 | mutex_init(&arc_user_evicts_lock, NULL, MUTEX_DEFAULT, NULL); | |
5323 | cv_init(&arc_user_evicts_cv, NULL, CV_DEFAULT, NULL); | |
5324 | ||
5325 | /* Convert seconds to clock ticks */ | |
5326 | arc_min_prefetch_lifespan = 1 * hz; | |
5327 | ||
5328 | /* Start out with 1/8 of all memory */ | |
5329 | arc_c = allmem / 8; | |
5330 | ||
5331 | #ifdef _KERNEL | |
5332 | /* | |
5333 | * On architectures where the physical memory can be larger | |
5334 | * than the addressable space (intel in 32-bit mode), we may | |
5335 | * need to limit the cache to 1/8 of VM size. | |
5336 | */ | |
5337 | arc_c = MIN(arc_c, vmem_size(heap_arena, VMEM_ALLOC | VMEM_FREE) / 8); | |
5338 | ||
5339 | /* | |
5340 | * Register a shrinker to support synchronous (direct) memory | |
5341 | * reclaim from the arc. This is done to prevent kswapd from | |
5342 | * swapping out pages when it is preferable to shrink the arc. | |
5343 | */ | |
5344 | spl_register_shrinker(&arc_shrinker); | |
5345 | ||
5346 | /* Set to 1/64 of all memory or a minimum of 512K */ | |
5347 | arc_sys_free = MAX(ptob(physmem / 64), (512 * 1024)); | |
5348 | arc_need_free = 0; | |
5349 | #endif | |
5350 | ||
5351 | /* Set min cache to allow safe operation of arc_adapt() */ | |
5352 | arc_c_min = 2ULL << SPA_MAXBLOCKSHIFT; | |
5353 | /* Set max to 1/2 of all memory */ | |
5354 | arc_c_max = allmem / 2; | |
5355 | ||
5356 | arc_c = arc_c_max; | |
5357 | arc_p = (arc_c >> 1); | |
5358 | ||
5359 | /* Set min to 1/2 of arc_c_min */ | |
5360 | arc_meta_min = 1ULL << SPA_MAXBLOCKSHIFT; | |
5361 | /* Initialize maximum observed usage to zero */ | |
5362 | arc_meta_max = 0; | |
5363 | /* Set limit to 3/4 of arc_c_max with a floor of arc_meta_min */ | |
5364 | arc_meta_limit = MAX((3 * arc_c_max) / 4, arc_meta_min); | |
5365 | ||
5366 | /* Apply user specified tunings */ | |
5367 | arc_tuning_update(); | |
5368 | ||
5369 | if (zfs_arc_num_sublists_per_state < 1) | |
5370 | zfs_arc_num_sublists_per_state = MAX(boot_ncpus, 1); | |
5371 | ||
5372 | /* if kmem_flags are set, lets try to use less memory */ | |
5373 | if (kmem_debugging()) | |
5374 | arc_c = arc_c / 2; | |
5375 | if (arc_c < arc_c_min) | |
5376 | arc_c = arc_c_min; | |
5377 | ||
5378 | arc_anon = &ARC_anon; | |
5379 | arc_mru = &ARC_mru; | |
5380 | arc_mru_ghost = &ARC_mru_ghost; | |
5381 | arc_mfu = &ARC_mfu; | |
5382 | arc_mfu_ghost = &ARC_mfu_ghost; | |
5383 | arc_l2c_only = &ARC_l2c_only; | |
5384 | arc_size = 0; | |
5385 | ||
5386 | multilist_create(&arc_mru->arcs_list[ARC_BUFC_METADATA], | |
5387 | sizeof (arc_buf_hdr_t), | |
5388 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
5389 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
5390 | multilist_create(&arc_mru->arcs_list[ARC_BUFC_DATA], | |
5391 | sizeof (arc_buf_hdr_t), | |
5392 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
5393 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
5394 | multilist_create(&arc_mru_ghost->arcs_list[ARC_BUFC_METADATA], | |
5395 | sizeof (arc_buf_hdr_t), | |
5396 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
5397 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
5398 | multilist_create(&arc_mru_ghost->arcs_list[ARC_BUFC_DATA], | |
5399 | sizeof (arc_buf_hdr_t), | |
5400 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
5401 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
5402 | multilist_create(&arc_mfu->arcs_list[ARC_BUFC_METADATA], | |
5403 | sizeof (arc_buf_hdr_t), | |
5404 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
5405 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
5406 | multilist_create(&arc_mfu->arcs_list[ARC_BUFC_DATA], | |
5407 | sizeof (arc_buf_hdr_t), | |
5408 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
5409 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
5410 | multilist_create(&arc_mfu_ghost->arcs_list[ARC_BUFC_METADATA], | |
5411 | sizeof (arc_buf_hdr_t), | |
5412 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
5413 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
5414 | multilist_create(&arc_mfu_ghost->arcs_list[ARC_BUFC_DATA], | |
5415 | sizeof (arc_buf_hdr_t), | |
5416 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
5417 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
5418 | multilist_create(&arc_l2c_only->arcs_list[ARC_BUFC_METADATA], | |
5419 | sizeof (arc_buf_hdr_t), | |
5420 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
5421 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
5422 | multilist_create(&arc_l2c_only->arcs_list[ARC_BUFC_DATA], | |
5423 | sizeof (arc_buf_hdr_t), | |
5424 | offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), | |
5425 | zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); | |
5426 | ||
5427 | arc_anon->arcs_state = ARC_STATE_ANON; | |
5428 | arc_mru->arcs_state = ARC_STATE_MRU; | |
5429 | arc_mru_ghost->arcs_state = ARC_STATE_MRU_GHOST; | |
5430 | arc_mfu->arcs_state = ARC_STATE_MFU; | |
5431 | arc_mfu_ghost->arcs_state = ARC_STATE_MFU_GHOST; | |
5432 | arc_l2c_only->arcs_state = ARC_STATE_L2C_ONLY; | |
5433 | ||
5434 | refcount_create(&arc_anon->arcs_size); | |
5435 | refcount_create(&arc_mru->arcs_size); | |
5436 | refcount_create(&arc_mru_ghost->arcs_size); | |
5437 | refcount_create(&arc_mfu->arcs_size); | |
5438 | refcount_create(&arc_mfu_ghost->arcs_size); | |
5439 | refcount_create(&arc_l2c_only->arcs_size); | |
5440 | ||
5441 | buf_init(); | |
5442 | ||
5443 | arc_reclaim_thread_exit = FALSE; | |
5444 | arc_user_evicts_thread_exit = FALSE; | |
5445 | list_create(&arc_prune_list, sizeof (arc_prune_t), | |
5446 | offsetof(arc_prune_t, p_node)); | |
5447 | arc_eviction_list = NULL; | |
5448 | mutex_init(&arc_prune_mtx, NULL, MUTEX_DEFAULT, NULL); | |
5449 | bzero(&arc_eviction_hdr, sizeof (arc_buf_hdr_t)); | |
5450 | ||
5451 | arc_prune_taskq = taskq_create("arc_prune", max_ncpus, defclsyspri, | |
5452 | max_ncpus, INT_MAX, TASKQ_PREPOPULATE | TASKQ_DYNAMIC); | |
5453 | ||
5454 | arc_ksp = kstat_create("zfs", 0, "arcstats", "misc", KSTAT_TYPE_NAMED, | |
5455 | sizeof (arc_stats) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL); | |
5456 | ||
5457 | if (arc_ksp != NULL) { | |
5458 | arc_ksp->ks_data = &arc_stats; | |
5459 | arc_ksp->ks_update = arc_kstat_update; | |
5460 | kstat_install(arc_ksp); | |
5461 | } | |
5462 | ||
5463 | (void) thread_create(NULL, 0, arc_reclaim_thread, NULL, 0, &p0, | |
5464 | TS_RUN, defclsyspri); | |
5465 | ||
5466 | (void) thread_create(NULL, 0, arc_user_evicts_thread, NULL, 0, &p0, | |
5467 | TS_RUN, defclsyspri); | |
5468 | ||
5469 | arc_dead = FALSE; | |
5470 | arc_warm = B_FALSE; | |
5471 | ||
5472 | /* | |
5473 | * Calculate maximum amount of dirty data per pool. | |
5474 | * | |
5475 | * If it has been set by a module parameter, take that. | |
5476 | * Otherwise, use a percentage of physical memory defined by | |
5477 | * zfs_dirty_data_max_percent (default 10%) with a cap at | |
9784fa9e | 5478 | * zfs_dirty_data_max_max (default 4G or 25% of physical memory). |
70e083d2 TG |
5479 | */ |
5480 | if (zfs_dirty_data_max_max == 0) | |
9784fa9e CIK |
5481 | zfs_dirty_data_max_max = MIN(4ULL * 1024 * 1024 * 1024, |
5482 | (uint64_t)physmem * PAGESIZE * | |
5483 | zfs_dirty_data_max_max_percent / 100); | |
70e083d2 TG |
5484 | |
5485 | if (zfs_dirty_data_max == 0) { | |
5486 | zfs_dirty_data_max = (uint64_t)physmem * PAGESIZE * | |
5487 | zfs_dirty_data_max_percent / 100; | |
5488 | zfs_dirty_data_max = MIN(zfs_dirty_data_max, | |
5489 | zfs_dirty_data_max_max); | |
5490 | } | |
5491 | } | |
5492 | ||
5493 | void | |
5494 | arc_fini(void) | |
5495 | { | |
5496 | arc_prune_t *p; | |
5497 | ||
5498 | #ifdef _KERNEL | |
5499 | spl_unregister_shrinker(&arc_shrinker); | |
5500 | #endif /* _KERNEL */ | |
5501 | ||
5502 | mutex_enter(&arc_reclaim_lock); | |
5503 | arc_reclaim_thread_exit = TRUE; | |
5504 | /* | |
5505 | * The reclaim thread will set arc_reclaim_thread_exit back to | |
5506 | * FALSE when it is finished exiting; we're waiting for that. | |
5507 | */ | |
5508 | while (arc_reclaim_thread_exit) { | |
5509 | cv_signal(&arc_reclaim_thread_cv); | |
5510 | cv_wait(&arc_reclaim_thread_cv, &arc_reclaim_lock); | |
5511 | } | |
5512 | mutex_exit(&arc_reclaim_lock); | |
5513 | ||
5514 | mutex_enter(&arc_user_evicts_lock); | |
5515 | arc_user_evicts_thread_exit = TRUE; | |
5516 | /* | |
5517 | * The user evicts thread will set arc_user_evicts_thread_exit | |
5518 | * to FALSE when it is finished exiting; we're waiting for that. | |
5519 | */ | |
5520 | while (arc_user_evicts_thread_exit) { | |
5521 | cv_signal(&arc_user_evicts_cv); | |
5522 | cv_wait(&arc_user_evicts_cv, &arc_user_evicts_lock); | |
5523 | } | |
5524 | mutex_exit(&arc_user_evicts_lock); | |
5525 | ||
5526 | /* Use TRUE to ensure *all* buffers are evicted */ | |
5527 | arc_flush(NULL, TRUE); | |
5528 | ||
5529 | arc_dead = TRUE; | |
5530 | ||
5531 | if (arc_ksp != NULL) { | |
5532 | kstat_delete(arc_ksp); | |
5533 | arc_ksp = NULL; | |
5534 | } | |
5535 | ||
5536 | taskq_wait(arc_prune_taskq); | |
5537 | taskq_destroy(arc_prune_taskq); | |
5538 | ||
5539 | mutex_enter(&arc_prune_mtx); | |
5540 | while ((p = list_head(&arc_prune_list)) != NULL) { | |
5541 | list_remove(&arc_prune_list, p); | |
5542 | refcount_remove(&p->p_refcnt, &arc_prune_list); | |
5543 | refcount_destroy(&p->p_refcnt); | |
5544 | kmem_free(p, sizeof (*p)); | |
5545 | } | |
5546 | mutex_exit(&arc_prune_mtx); | |
5547 | ||
5548 | list_destroy(&arc_prune_list); | |
5549 | mutex_destroy(&arc_prune_mtx); | |
5550 | mutex_destroy(&arc_reclaim_lock); | |
5551 | cv_destroy(&arc_reclaim_thread_cv); | |
5552 | cv_destroy(&arc_reclaim_waiters_cv); | |
5553 | ||
5554 | mutex_destroy(&arc_user_evicts_lock); | |
5555 | cv_destroy(&arc_user_evicts_cv); | |
5556 | ||
5557 | refcount_destroy(&arc_anon->arcs_size); | |
5558 | refcount_destroy(&arc_mru->arcs_size); | |
5559 | refcount_destroy(&arc_mru_ghost->arcs_size); | |
5560 | refcount_destroy(&arc_mfu->arcs_size); | |
5561 | refcount_destroy(&arc_mfu_ghost->arcs_size); | |
5562 | refcount_destroy(&arc_l2c_only->arcs_size); | |
5563 | ||
5564 | multilist_destroy(&arc_mru->arcs_list[ARC_BUFC_METADATA]); | |
5565 | multilist_destroy(&arc_mru_ghost->arcs_list[ARC_BUFC_METADATA]); | |
5566 | multilist_destroy(&arc_mfu->arcs_list[ARC_BUFC_METADATA]); | |
5567 | multilist_destroy(&arc_mfu_ghost->arcs_list[ARC_BUFC_METADATA]); | |
5568 | multilist_destroy(&arc_mru->arcs_list[ARC_BUFC_DATA]); | |
5569 | multilist_destroy(&arc_mru_ghost->arcs_list[ARC_BUFC_DATA]); | |
5570 | multilist_destroy(&arc_mfu->arcs_list[ARC_BUFC_DATA]); | |
5571 | multilist_destroy(&arc_mfu_ghost->arcs_list[ARC_BUFC_DATA]); | |
5572 | multilist_destroy(&arc_l2c_only->arcs_list[ARC_BUFC_METADATA]); | |
5573 | multilist_destroy(&arc_l2c_only->arcs_list[ARC_BUFC_DATA]); | |
5574 | ||
5575 | buf_fini(); | |
5576 | ||
5577 | ASSERT0(arc_loaned_bytes); | |
5578 | } | |
5579 | ||
5580 | /* | |
5581 | * Level 2 ARC | |
5582 | * | |
5583 | * The level 2 ARC (L2ARC) is a cache layer in-between main memory and disk. | |
5584 | * It uses dedicated storage devices to hold cached data, which are populated | |
5585 | * using large infrequent writes. The main role of this cache is to boost | |
5586 | * the performance of random read workloads. The intended L2ARC devices | |
5587 | * include short-stroked disks, solid state disks, and other media with | |
5588 | * substantially faster read latency than disk. | |
5589 | * | |
5590 | * +-----------------------+ | |
5591 | * | ARC | | |
5592 | * +-----------------------+ | |
5593 | * | ^ ^ | |
5594 | * | | | | |
5595 | * l2arc_feed_thread() arc_read() | |
5596 | * | | | | |
5597 | * | l2arc read | | |
5598 | * V | | | |
5599 | * +---------------+ | | |
5600 | * | L2ARC | | | |
5601 | * +---------------+ | | |
5602 | * | ^ | | |
5603 | * l2arc_write() | | | |
5604 | * | | | | |
5605 | * V | | | |
5606 | * +-------+ +-------+ | |
5607 | * | vdev | | vdev | | |
5608 | * | cache | | cache | | |
5609 | * +-------+ +-------+ | |
5610 | * +=========+ .-----. | |
5611 | * : L2ARC : |-_____-| | |
5612 | * : devices : | Disks | | |
5613 | * +=========+ `-_____-' | |
5614 | * | |
5615 | * Read requests are satisfied from the following sources, in order: | |
5616 | * | |
5617 | * 1) ARC | |
5618 | * 2) vdev cache of L2ARC devices | |
5619 | * 3) L2ARC devices | |
5620 | * 4) vdev cache of disks | |
5621 | * 5) disks | |
5622 | * | |
5623 | * Some L2ARC device types exhibit extremely slow write performance. | |
5624 | * To accommodate for this there are some significant differences between | |
5625 | * the L2ARC and traditional cache design: | |
5626 | * | |
5627 | * 1. There is no eviction path from the ARC to the L2ARC. Evictions from | |
5628 | * the ARC behave as usual, freeing buffers and placing headers on ghost | |
5629 | * lists. The ARC does not send buffers to the L2ARC during eviction as | |
5630 | * this would add inflated write latencies for all ARC memory pressure. | |
5631 | * | |
5632 | * 2. The L2ARC attempts to cache data from the ARC before it is evicted. | |
5633 | * It does this by periodically scanning buffers from the eviction-end of | |
5634 | * the MFU and MRU ARC lists, copying them to the L2ARC devices if they are | |
5635 | * not already there. It scans until a headroom of buffers is satisfied, | |
5636 | * which itself is a buffer for ARC eviction. If a compressible buffer is | |
5637 | * found during scanning and selected for writing to an L2ARC device, we | |
5638 | * temporarily boost scanning headroom during the next scan cycle to make | |
5639 | * sure we adapt to compression effects (which might significantly reduce | |
5640 | * the data volume we write to L2ARC). The thread that does this is | |
5641 | * l2arc_feed_thread(), illustrated below; example sizes are included to | |
5642 | * provide a better sense of ratio than this diagram: | |
5643 | * | |
5644 | * head --> tail | |
5645 | * +---------------------+----------+ | |
5646 | * ARC_mfu |:::::#:::::::::::::::|o#o###o###|-->. # already on L2ARC | |
5647 | * +---------------------+----------+ | o L2ARC eligible | |
5648 | * ARC_mru |:#:::::::::::::::::::|#o#ooo####|-->| : ARC buffer | |
5649 | * +---------------------+----------+ | | |
5650 | * 15.9 Gbytes ^ 32 Mbytes | | |
5651 | * headroom | | |
5652 | * l2arc_feed_thread() | |
5653 | * | | |
5654 | * l2arc write hand <--[oooo]--' | |
5655 | * | 8 Mbyte | |
5656 | * | write max | |
5657 | * V | |
5658 | * +==============================+ | |
5659 | * L2ARC dev |####|#|###|###| |####| ... | | |
5660 | * +==============================+ | |
5661 | * 32 Gbytes | |
5662 | * | |
5663 | * 3. If an ARC buffer is copied to the L2ARC but then hit instead of | |
5664 | * evicted, then the L2ARC has cached a buffer much sooner than it probably | |
5665 | * needed to, potentially wasting L2ARC device bandwidth and storage. It is | |
5666 | * safe to say that this is an uncommon case, since buffers at the end of | |
5667 | * the ARC lists have moved there due to inactivity. | |
5668 | * | |
5669 | * 4. If the ARC evicts faster than the L2ARC can maintain a headroom, | |
5670 | * then the L2ARC simply misses copying some buffers. This serves as a | |
5671 | * pressure valve to prevent heavy read workloads from both stalling the ARC | |
5672 | * with waits and clogging the L2ARC with writes. This also helps prevent | |
5673 | * the potential for the L2ARC to churn if it attempts to cache content too | |
5674 | * quickly, such as during backups of the entire pool. | |
5675 | * | |
5676 | * 5. After system boot and before the ARC has filled main memory, there are | |
5677 | * no evictions from the ARC and so the tails of the ARC_mfu and ARC_mru | |
5678 | * lists can remain mostly static. Instead of searching from tail of these | |
5679 | * lists as pictured, the l2arc_feed_thread() will search from the list heads | |
5680 | * for eligible buffers, greatly increasing its chance of finding them. | |
5681 | * | |
5682 | * The L2ARC device write speed is also boosted during this time so that | |
5683 | * the L2ARC warms up faster. Since there have been no ARC evictions yet, | |
5684 | * there are no L2ARC reads, and no fear of degrading read performance | |
5685 | * through increased writes. | |
5686 | * | |
5687 | * 6. Writes to the L2ARC devices are grouped and sent in-sequence, so that | |
5688 | * the vdev queue can aggregate them into larger and fewer writes. Each | |
5689 | * device is written to in a rotor fashion, sweeping writes through | |
5690 | * available space then repeating. | |
5691 | * | |
5692 | * 7. The L2ARC does not store dirty content. It never needs to flush | |
5693 | * write buffers back to disk based storage. | |
5694 | * | |
5695 | * 8. If an ARC buffer is written (and dirtied) which also exists in the | |
5696 | * L2ARC, the now stale L2ARC buffer is immediately dropped. | |
5697 | * | |
5698 | * The performance of the L2ARC can be tweaked by a number of tunables, which | |
5699 | * may be necessary for different workloads: | |
5700 | * | |
5701 | * l2arc_write_max max write bytes per interval | |
5702 | * l2arc_write_boost extra write bytes during device warmup | |
5703 | * l2arc_noprefetch skip caching prefetched buffers | |
5704 | * l2arc_nocompress skip compressing buffers | |
5705 | * l2arc_headroom number of max device writes to precache | |
5706 | * l2arc_headroom_boost when we find compressed buffers during ARC | |
5707 | * scanning, we multiply headroom by this | |
5708 | * percentage factor for the next scan cycle, | |
5709 | * since more compressed buffers are likely to | |
5710 | * be present | |
5711 | * l2arc_feed_secs seconds between L2ARC writing | |
5712 | * | |
5713 | * Tunables may be removed or added as future performance improvements are | |
5714 | * integrated, and also may become zpool properties. | |
5715 | * | |
5716 | * There are three key functions that control how the L2ARC warms up: | |
5717 | * | |
5718 | * l2arc_write_eligible() check if a buffer is eligible to cache | |
5719 | * l2arc_write_size() calculate how much to write | |
5720 | * l2arc_write_interval() calculate sleep delay between writes | |
5721 | * | |
5722 | * These three functions determine what to write, how much, and how quickly | |
5723 | * to send writes. | |
5724 | */ | |
5725 | ||
5726 | static boolean_t | |
5727 | l2arc_write_eligible(uint64_t spa_guid, arc_buf_hdr_t *hdr) | |
5728 | { | |
5729 | /* | |
5730 | * A buffer is *not* eligible for the L2ARC if it: | |
5731 | * 1. belongs to a different spa. | |
5732 | * 2. is already cached on the L2ARC. | |
5733 | * 3. has an I/O in progress (it may be an incomplete read). | |
5734 | * 4. is flagged not eligible (zfs property). | |
5735 | */ | |
5736 | if (hdr->b_spa != spa_guid || HDR_HAS_L2HDR(hdr) || | |
5737 | HDR_IO_IN_PROGRESS(hdr) || !HDR_L2CACHE(hdr)) | |
5738 | return (B_FALSE); | |
5739 | ||
5740 | return (B_TRUE); | |
5741 | } | |
5742 | ||
5743 | static uint64_t | |
5744 | l2arc_write_size(void) | |
5745 | { | |
5746 | uint64_t size; | |
5747 | ||
5748 | /* | |
5749 | * Make sure our globals have meaningful values in case the user | |
5750 | * altered them. | |
5751 | */ | |
5752 | size = l2arc_write_max; | |
5753 | if (size == 0) { | |
5754 | cmn_err(CE_NOTE, "Bad value for l2arc_write_max, value must " | |
5755 | "be greater than zero, resetting it to the default (%d)", | |
5756 | L2ARC_WRITE_SIZE); | |
5757 | size = l2arc_write_max = L2ARC_WRITE_SIZE; | |
5758 | } | |
5759 | ||
5760 | if (arc_warm == B_FALSE) | |
5761 | size += l2arc_write_boost; | |
5762 | ||
5763 | return (size); | |
5764 | ||
5765 | } | |
5766 | ||
5767 | static clock_t | |
5768 | l2arc_write_interval(clock_t began, uint64_t wanted, uint64_t wrote) | |
5769 | { | |
5770 | clock_t interval, next, now; | |
5771 | ||
5772 | /* | |
5773 | * If the ARC lists are busy, increase our write rate; if the | |
5774 | * lists are stale, idle back. This is achieved by checking | |
5775 | * how much we previously wrote - if it was more than half of | |
5776 | * what we wanted, schedule the next write much sooner. | |
5777 | */ | |
5778 | if (l2arc_feed_again && wrote > (wanted / 2)) | |
5779 | interval = (hz * l2arc_feed_min_ms) / 1000; | |
5780 | else | |
5781 | interval = hz * l2arc_feed_secs; | |
5782 | ||
5783 | now = ddi_get_lbolt(); | |
5784 | next = MAX(now, MIN(now + interval, began + interval)); | |
5785 | ||
5786 | return (next); | |
5787 | } | |
5788 | ||
5789 | /* | |
5790 | * Cycle through L2ARC devices. This is how L2ARC load balances. | |
5791 | * If a device is returned, this also returns holding the spa config lock. | |
5792 | */ | |
5793 | static l2arc_dev_t * | |
5794 | l2arc_dev_get_next(void) | |
5795 | { | |
5796 | l2arc_dev_t *first, *next = NULL; | |
5797 | ||
5798 | /* | |
5799 | * Lock out the removal of spas (spa_namespace_lock), then removal | |
5800 | * of cache devices (l2arc_dev_mtx). Once a device has been selected, | |
5801 | * both locks will be dropped and a spa config lock held instead. | |
5802 | */ | |
5803 | mutex_enter(&spa_namespace_lock); | |
5804 | mutex_enter(&l2arc_dev_mtx); | |
5805 | ||
5806 | /* if there are no vdevs, there is nothing to do */ | |
5807 | if (l2arc_ndev == 0) | |
5808 | goto out; | |
5809 | ||
5810 | first = NULL; | |
5811 | next = l2arc_dev_last; | |
5812 | do { | |
5813 | /* loop around the list looking for a non-faulted vdev */ | |
5814 | if (next == NULL) { | |
5815 | next = list_head(l2arc_dev_list); | |
5816 | } else { | |
5817 | next = list_next(l2arc_dev_list, next); | |
5818 | if (next == NULL) | |
5819 | next = list_head(l2arc_dev_list); | |
5820 | } | |
5821 | ||
5822 | /* if we have come back to the start, bail out */ | |
5823 | if (first == NULL) | |
5824 | first = next; | |
5825 | else if (next == first) | |
5826 | break; | |
5827 | ||
5828 | } while (vdev_is_dead(next->l2ad_vdev)); | |
5829 | ||
5830 | /* if we were unable to find any usable vdevs, return NULL */ | |
5831 | if (vdev_is_dead(next->l2ad_vdev)) | |
5832 | next = NULL; | |
5833 | ||
5834 | l2arc_dev_last = next; | |
5835 | ||
5836 | out: | |
5837 | mutex_exit(&l2arc_dev_mtx); | |
5838 | ||
5839 | /* | |
5840 | * Grab the config lock to prevent the 'next' device from being | |
5841 | * removed while we are writing to it. | |
5842 | */ | |
5843 | if (next != NULL) | |
5844 | spa_config_enter(next->l2ad_spa, SCL_L2ARC, next, RW_READER); | |
5845 | mutex_exit(&spa_namespace_lock); | |
5846 | ||
5847 | return (next); | |
5848 | } | |
5849 | ||
5850 | /* | |
5851 | * Free buffers that were tagged for destruction. | |
5852 | */ | |
5853 | static void | |
5854 | l2arc_do_free_on_write(void) | |
5855 | { | |
5856 | list_t *buflist; | |
5857 | l2arc_data_free_t *df, *df_prev; | |
5858 | ||
5859 | mutex_enter(&l2arc_free_on_write_mtx); | |
5860 | buflist = l2arc_free_on_write; | |
5861 | ||
5862 | for (df = list_tail(buflist); df; df = df_prev) { | |
5863 | df_prev = list_prev(buflist, df); | |
5864 | ASSERT(df->l2df_data != NULL); | |
5865 | ASSERT(df->l2df_func != NULL); | |
5866 | df->l2df_func(df->l2df_data, df->l2df_size); | |
5867 | list_remove(buflist, df); | |
5868 | kmem_free(df, sizeof (l2arc_data_free_t)); | |
5869 | } | |
5870 | ||
5871 | mutex_exit(&l2arc_free_on_write_mtx); | |
5872 | } | |
5873 | ||
5874 | /* | |
5875 | * A write to a cache device has completed. Update all headers to allow | |
5876 | * reads from these buffers to begin. | |
5877 | */ | |
5878 | static void | |
5879 | l2arc_write_done(zio_t *zio) | |
5880 | { | |
5881 | l2arc_write_callback_t *cb; | |
5882 | l2arc_dev_t *dev; | |
5883 | list_t *buflist; | |
5884 | arc_buf_hdr_t *head, *hdr, *hdr_prev; | |
5885 | kmutex_t *hash_lock; | |
5886 | int64_t bytes_dropped = 0; | |
5887 | ||
5888 | cb = zio->io_private; | |
5889 | ASSERT(cb != NULL); | |
5890 | dev = cb->l2wcb_dev; | |
5891 | ASSERT(dev != NULL); | |
5892 | head = cb->l2wcb_head; | |
5893 | ASSERT(head != NULL); | |
5894 | buflist = &dev->l2ad_buflist; | |
5895 | ASSERT(buflist != NULL); | |
5896 | DTRACE_PROBE2(l2arc__iodone, zio_t *, zio, | |
5897 | l2arc_write_callback_t *, cb); | |
5898 | ||
5899 | if (zio->io_error != 0) | |
5900 | ARCSTAT_BUMP(arcstat_l2_writes_error); | |
5901 | ||
5902 | /* | |
5903 | * All writes completed, or an error was hit. | |
5904 | */ | |
5905 | top: | |
5906 | mutex_enter(&dev->l2ad_mtx); | |
5907 | for (hdr = list_prev(buflist, head); hdr; hdr = hdr_prev) { | |
5908 | hdr_prev = list_prev(buflist, hdr); | |
5909 | ||
5910 | hash_lock = HDR_LOCK(hdr); | |
5911 | ||
5912 | /* | |
5913 | * We cannot use mutex_enter or else we can deadlock | |
5914 | * with l2arc_write_buffers (due to swapping the order | |
5915 | * the hash lock and l2ad_mtx are taken). | |
5916 | */ | |
5917 | if (!mutex_tryenter(hash_lock)) { | |
5918 | /* | |
5919 | * Missed the hash lock. We must retry so we | |
5920 | * don't leave the ARC_FLAG_L2_WRITING bit set. | |
5921 | */ | |
5922 | ARCSTAT_BUMP(arcstat_l2_writes_lock_retry); | |
5923 | ||
5924 | /* | |
5925 | * We don't want to rescan the headers we've | |
5926 | * already marked as having been written out, so | |
5927 | * we reinsert the head node so we can pick up | |
5928 | * where we left off. | |
5929 | */ | |
5930 | list_remove(buflist, head); | |
5931 | list_insert_after(buflist, hdr, head); | |
5932 | ||
5933 | mutex_exit(&dev->l2ad_mtx); | |
5934 | ||
5935 | /* | |
5936 | * We wait for the hash lock to become available | |
5937 | * to try and prevent busy waiting, and increase | |
5938 | * the chance we'll be able to acquire the lock | |
5939 | * the next time around. | |
5940 | */ | |
5941 | mutex_enter(hash_lock); | |
5942 | mutex_exit(hash_lock); | |
5943 | goto top; | |
5944 | } | |
5945 | ||
5946 | /* | |
5947 | * We could not have been moved into the arc_l2c_only | |
5948 | * state while in-flight due to our ARC_FLAG_L2_WRITING | |
5949 | * bit being set. Let's just ensure that's being enforced. | |
5950 | */ | |
5951 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
5952 | ||
5953 | /* | |
5954 | * We may have allocated a buffer for L2ARC compression, | |
5955 | * we must release it to avoid leaking this data. | |
5956 | */ | |
5957 | l2arc_release_cdata_buf(hdr); | |
5958 | ||
5959 | if (zio->io_error != 0) { | |
5960 | /* | |
5961 | * Error - drop L2ARC entry. | |
5962 | */ | |
5963 | list_remove(buflist, hdr); | |
5964 | hdr->b_flags &= ~ARC_FLAG_HAS_L2HDR; | |
5965 | ||
5966 | ARCSTAT_INCR(arcstat_l2_asize, -hdr->b_l2hdr.b_asize); | |
5967 | ARCSTAT_INCR(arcstat_l2_size, -hdr->b_size); | |
5968 | ||
5969 | bytes_dropped += hdr->b_l2hdr.b_asize; | |
5970 | (void) refcount_remove_many(&dev->l2ad_alloc, | |
5971 | hdr->b_l2hdr.b_asize, hdr); | |
5972 | } | |
5973 | ||
5974 | /* | |
5975 | * Allow ARC to begin reads and ghost list evictions to | |
5976 | * this L2ARC entry. | |
5977 | */ | |
5978 | hdr->b_flags &= ~ARC_FLAG_L2_WRITING; | |
5979 | ||
5980 | mutex_exit(hash_lock); | |
5981 | } | |
5982 | ||
5983 | atomic_inc_64(&l2arc_writes_done); | |
5984 | list_remove(buflist, head); | |
5985 | ASSERT(!HDR_HAS_L1HDR(head)); | |
5986 | kmem_cache_free(hdr_l2only_cache, head); | |
5987 | mutex_exit(&dev->l2ad_mtx); | |
5988 | ||
5989 | vdev_space_update(dev->l2ad_vdev, -bytes_dropped, 0, 0); | |
5990 | ||
5991 | l2arc_do_free_on_write(); | |
5992 | ||
5993 | kmem_free(cb, sizeof (l2arc_write_callback_t)); | |
5994 | } | |
5995 | ||
5996 | /* | |
5997 | * A read to a cache device completed. Validate buffer contents before | |
5998 | * handing over to the regular ARC routines. | |
5999 | */ | |
6000 | static void | |
6001 | l2arc_read_done(zio_t *zio) | |
6002 | { | |
6003 | l2arc_read_callback_t *cb; | |
6004 | arc_buf_hdr_t *hdr; | |
6005 | arc_buf_t *buf; | |
6006 | kmutex_t *hash_lock; | |
6007 | int equal; | |
6008 | ||
6009 | ASSERT(zio->io_vd != NULL); | |
6010 | ASSERT(zio->io_flags & ZIO_FLAG_DONT_PROPAGATE); | |
6011 | ||
6012 | spa_config_exit(zio->io_spa, SCL_L2ARC, zio->io_vd); | |
6013 | ||
6014 | cb = zio->io_private; | |
6015 | ASSERT(cb != NULL); | |
6016 | buf = cb->l2rcb_buf; | |
6017 | ASSERT(buf != NULL); | |
6018 | ||
6019 | hash_lock = HDR_LOCK(buf->b_hdr); | |
6020 | mutex_enter(hash_lock); | |
6021 | hdr = buf->b_hdr; | |
6022 | ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); | |
6023 | ||
6024 | /* | |
6025 | * If the buffer was compressed, decompress it first. | |
6026 | */ | |
6027 | if (cb->l2rcb_compress != ZIO_COMPRESS_OFF) | |
6028 | l2arc_decompress_zio(zio, hdr, cb->l2rcb_compress); | |
6029 | ASSERT(zio->io_data != NULL); | |
6030 | ASSERT3U(zio->io_size, ==, hdr->b_size); | |
6031 | ASSERT3U(BP_GET_LSIZE(&cb->l2rcb_bp), ==, hdr->b_size); | |
6032 | ||
6033 | /* | |
6034 | * Check this survived the L2ARC journey. | |
6035 | */ | |
6036 | equal = arc_cksum_equal(buf); | |
6037 | if (equal && zio->io_error == 0 && !HDR_L2_EVICTED(hdr)) { | |
6038 | mutex_exit(hash_lock); | |
6039 | zio->io_private = buf; | |
6040 | zio->io_bp_copy = cb->l2rcb_bp; /* XXX fix in L2ARC 2.0 */ | |
6041 | zio->io_bp = &zio->io_bp_copy; /* XXX fix in L2ARC 2.0 */ | |
6042 | arc_read_done(zio); | |
6043 | } else { | |
6044 | mutex_exit(hash_lock); | |
6045 | /* | |
6046 | * Buffer didn't survive caching. Increment stats and | |
6047 | * reissue to the original storage device. | |
6048 | */ | |
6049 | if (zio->io_error != 0) { | |
6050 | ARCSTAT_BUMP(arcstat_l2_io_error); | |
6051 | } else { | |
6052 | zio->io_error = SET_ERROR(EIO); | |
6053 | } | |
6054 | if (!equal) | |
6055 | ARCSTAT_BUMP(arcstat_l2_cksum_bad); | |
6056 | ||
6057 | /* | |
6058 | * If there's no waiter, issue an async i/o to the primary | |
6059 | * storage now. If there *is* a waiter, the caller must | |
6060 | * issue the i/o in a context where it's OK to block. | |
6061 | */ | |
6062 | if (zio->io_waiter == NULL) { | |
6063 | zio_t *pio = zio_unique_parent(zio); | |
6064 | ||
6065 | ASSERT(!pio || pio->io_child_type == ZIO_CHILD_LOGICAL); | |
6066 | ||
6067 | zio_nowait(zio_read(pio, cb->l2rcb_spa, &cb->l2rcb_bp, | |
6068 | buf->b_data, hdr->b_size, arc_read_done, buf, | |
6069 | zio->io_priority, cb->l2rcb_flags, &cb->l2rcb_zb)); | |
6070 | } | |
6071 | } | |
6072 | ||
6073 | kmem_free(cb, sizeof (l2arc_read_callback_t)); | |
6074 | } | |
6075 | ||
6076 | /* | |
6077 | * This is the list priority from which the L2ARC will search for pages to | |
6078 | * cache. This is used within loops (0..3) to cycle through lists in the | |
6079 | * desired order. This order can have a significant effect on cache | |
6080 | * performance. | |
6081 | * | |
6082 | * Currently the metadata lists are hit first, MFU then MRU, followed by | |
6083 | * the data lists. This function returns a locked list, and also returns | |
6084 | * the lock pointer. | |
6085 | */ | |
6086 | static multilist_sublist_t * | |
6087 | l2arc_sublist_lock(int list_num) | |
6088 | { | |
6089 | multilist_t *ml = NULL; | |
6090 | unsigned int idx; | |
6091 | ||
6092 | ASSERT(list_num >= 0 && list_num <= 3); | |
6093 | ||
6094 | switch (list_num) { | |
6095 | case 0: | |
6096 | ml = &arc_mfu->arcs_list[ARC_BUFC_METADATA]; | |
6097 | break; | |
6098 | case 1: | |
6099 | ml = &arc_mru->arcs_list[ARC_BUFC_METADATA]; | |
6100 | break; | |
6101 | case 2: | |
6102 | ml = &arc_mfu->arcs_list[ARC_BUFC_DATA]; | |
6103 | break; | |
6104 | case 3: | |
6105 | ml = &arc_mru->arcs_list[ARC_BUFC_DATA]; | |
6106 | break; | |
6107 | } | |
6108 | ||
6109 | /* | |
6110 | * Return a randomly-selected sublist. This is acceptable | |
6111 | * because the caller feeds only a little bit of data for each | |
6112 | * call (8MB). Subsequent calls will result in different | |
6113 | * sublists being selected. | |
6114 | */ | |
6115 | idx = multilist_get_random_index(ml); | |
6116 | return (multilist_sublist_lock(ml, idx)); | |
6117 | } | |
6118 | ||
6119 | /* | |
6120 | * Evict buffers from the device write hand to the distance specified in | |
6121 | * bytes. This distance may span populated buffers, it may span nothing. | |
6122 | * This is clearing a region on the L2ARC device ready for writing. | |
6123 | * If the 'all' boolean is set, every buffer is evicted. | |
6124 | */ | |
6125 | static void | |
6126 | l2arc_evict(l2arc_dev_t *dev, uint64_t distance, boolean_t all) | |
6127 | { | |
6128 | list_t *buflist; | |
6129 | arc_buf_hdr_t *hdr, *hdr_prev; | |
6130 | kmutex_t *hash_lock; | |
6131 | uint64_t taddr; | |
6132 | ||
6133 | buflist = &dev->l2ad_buflist; | |
6134 | ||
6135 | if (!all && dev->l2ad_first) { | |
6136 | /* | |
6137 | * This is the first sweep through the device. There is | |
6138 | * nothing to evict. | |
6139 | */ | |
6140 | return; | |
6141 | } | |
6142 | ||
6143 | if (dev->l2ad_hand >= (dev->l2ad_end - (2 * distance))) { | |
6144 | /* | |
6145 | * When nearing the end of the device, evict to the end | |
6146 | * before the device write hand jumps to the start. | |
6147 | */ | |
6148 | taddr = dev->l2ad_end; | |
6149 | } else { | |
6150 | taddr = dev->l2ad_hand + distance; | |
6151 | } | |
6152 | DTRACE_PROBE4(l2arc__evict, l2arc_dev_t *, dev, list_t *, buflist, | |
6153 | uint64_t, taddr, boolean_t, all); | |
6154 | ||
6155 | top: | |
6156 | mutex_enter(&dev->l2ad_mtx); | |
6157 | for (hdr = list_tail(buflist); hdr; hdr = hdr_prev) { | |
6158 | hdr_prev = list_prev(buflist, hdr); | |
6159 | ||
6160 | hash_lock = HDR_LOCK(hdr); | |
6161 | ||
6162 | /* | |
6163 | * We cannot use mutex_enter or else we can deadlock | |
6164 | * with l2arc_write_buffers (due to swapping the order | |
6165 | * the hash lock and l2ad_mtx are taken). | |
6166 | */ | |
6167 | if (!mutex_tryenter(hash_lock)) { | |
6168 | /* | |
6169 | * Missed the hash lock. Retry. | |
6170 | */ | |
6171 | ARCSTAT_BUMP(arcstat_l2_evict_lock_retry); | |
6172 | mutex_exit(&dev->l2ad_mtx); | |
6173 | mutex_enter(hash_lock); | |
6174 | mutex_exit(hash_lock); | |
6175 | goto top; | |
6176 | } | |
6177 | ||
6178 | if (HDR_L2_WRITE_HEAD(hdr)) { | |
6179 | /* | |
6180 | * We hit a write head node. Leave it for | |
6181 | * l2arc_write_done(). | |
6182 | */ | |
6183 | list_remove(buflist, hdr); | |
6184 | mutex_exit(hash_lock); | |
6185 | continue; | |
6186 | } | |
6187 | ||
6188 | if (!all && HDR_HAS_L2HDR(hdr) && | |
6189 | (hdr->b_l2hdr.b_daddr > taddr || | |
6190 | hdr->b_l2hdr.b_daddr < dev->l2ad_hand)) { | |
6191 | /* | |
6192 | * We've evicted to the target address, | |
6193 | * or the end of the device. | |
6194 | */ | |
6195 | mutex_exit(hash_lock); | |
6196 | break; | |
6197 | } | |
6198 | ||
6199 | ASSERT(HDR_HAS_L2HDR(hdr)); | |
6200 | if (!HDR_HAS_L1HDR(hdr)) { | |
6201 | ASSERT(!HDR_L2_READING(hdr)); | |
6202 | /* | |
6203 | * This doesn't exist in the ARC. Destroy. | |
6204 | * arc_hdr_destroy() will call list_remove() | |
6205 | * and decrement arcstat_l2_size. | |
6206 | */ | |
6207 | arc_change_state(arc_anon, hdr, hash_lock); | |
6208 | arc_hdr_destroy(hdr); | |
6209 | } else { | |
6210 | ASSERT(hdr->b_l1hdr.b_state != arc_l2c_only); | |
6211 | ARCSTAT_BUMP(arcstat_l2_evict_l1cached); | |
6212 | /* | |
6213 | * Invalidate issued or about to be issued | |
6214 | * reads, since we may be about to write | |
6215 | * over this location. | |
6216 | */ | |
6217 | if (HDR_L2_READING(hdr)) { | |
6218 | ARCSTAT_BUMP(arcstat_l2_evict_reading); | |
6219 | hdr->b_flags |= ARC_FLAG_L2_EVICTED; | |
6220 | } | |
6221 | ||
6222 | /* Ensure this header has finished being written */ | |
6223 | ASSERT(!HDR_L2_WRITING(hdr)); | |
6224 | ASSERT3P(hdr->b_l1hdr.b_tmp_cdata, ==, NULL); | |
6225 | ||
6226 | arc_hdr_l2hdr_destroy(hdr); | |
6227 | } | |
6228 | mutex_exit(hash_lock); | |
6229 | } | |
6230 | mutex_exit(&dev->l2ad_mtx); | |
6231 | } | |
6232 | ||
6233 | /* | |
6234 | * Find and write ARC buffers to the L2ARC device. | |
6235 | * | |
6236 | * An ARC_FLAG_L2_WRITING flag is set so that the L2ARC buffers are not valid | |
6237 | * for reading until they have completed writing. | |
6238 | * The headroom_boost is an in-out parameter used to maintain headroom boost | |
6239 | * state between calls to this function. | |
6240 | * | |
6241 | * Returns the number of bytes actually written (which may be smaller than | |
6242 | * the delta by which the device hand has changed due to alignment). | |
6243 | */ | |
6244 | static uint64_t | |
6245 | l2arc_write_buffers(spa_t *spa, l2arc_dev_t *dev, uint64_t target_sz, | |
6246 | boolean_t *headroom_boost) | |
6247 | { | |
6248 | arc_buf_hdr_t *hdr, *hdr_prev, *head; | |
6249 | uint64_t write_asize, write_sz, headroom, buf_compress_minsz, | |
6250 | stats_size; | |
6251 | void *buf_data; | |
6252 | boolean_t full; | |
6253 | l2arc_write_callback_t *cb; | |
6254 | zio_t *pio, *wzio; | |
6255 | uint64_t guid = spa_load_guid(spa); | |
6256 | int try; | |
6257 | const boolean_t do_headroom_boost = *headroom_boost; | |
6258 | ||
6259 | ASSERT(dev->l2ad_vdev != NULL); | |
6260 | ||
6261 | /* Lower the flag now, we might want to raise it again later. */ | |
6262 | *headroom_boost = B_FALSE; | |
6263 | ||
6264 | pio = NULL; | |
6265 | write_sz = write_asize = 0; | |
6266 | full = B_FALSE; | |
6267 | head = kmem_cache_alloc(hdr_l2only_cache, KM_PUSHPAGE); | |
6268 | head->b_flags |= ARC_FLAG_L2_WRITE_HEAD; | |
6269 | head->b_flags |= ARC_FLAG_HAS_L2HDR; | |
6270 | ||
6271 | /* | |
6272 | * We will want to try to compress buffers that are at least 2x the | |
6273 | * device sector size. | |
6274 | */ | |
6275 | buf_compress_minsz = 2 << dev->l2ad_vdev->vdev_ashift; | |
6276 | ||
6277 | /* | |
6278 | * Copy buffers for L2ARC writing. | |
6279 | */ | |
6280 | for (try = 0; try <= 3; try++) { | |
6281 | multilist_sublist_t *mls = l2arc_sublist_lock(try); | |
6282 | uint64_t passed_sz = 0; | |
6283 | ||
6284 | /* | |
6285 | * L2ARC fast warmup. | |
6286 | * | |
6287 | * Until the ARC is warm and starts to evict, read from the | |
6288 | * head of the ARC lists rather than the tail. | |
6289 | */ | |
6290 | if (arc_warm == B_FALSE) | |
6291 | hdr = multilist_sublist_head(mls); | |
6292 | else | |
6293 | hdr = multilist_sublist_tail(mls); | |
6294 | ||
6295 | headroom = target_sz * l2arc_headroom; | |
6296 | if (do_headroom_boost) | |
6297 | headroom = (headroom * l2arc_headroom_boost) / 100; | |
6298 | ||
6299 | for (; hdr; hdr = hdr_prev) { | |
6300 | kmutex_t *hash_lock; | |
6301 | uint64_t buf_sz; | |
6302 | uint64_t buf_a_sz; | |
6303 | ||
6304 | if (arc_warm == B_FALSE) | |
6305 | hdr_prev = multilist_sublist_next(mls, hdr); | |
6306 | else | |
6307 | hdr_prev = multilist_sublist_prev(mls, hdr); | |
6308 | ||
6309 | hash_lock = HDR_LOCK(hdr); | |
6310 | if (!mutex_tryenter(hash_lock)) { | |
6311 | /* | |
6312 | * Skip this buffer rather than waiting. | |
6313 | */ | |
6314 | continue; | |
6315 | } | |
6316 | ||
6317 | passed_sz += hdr->b_size; | |
6318 | if (passed_sz > headroom) { | |
6319 | /* | |
6320 | * Searched too far. | |
6321 | */ | |
6322 | mutex_exit(hash_lock); | |
6323 | break; | |
6324 | } | |
6325 | ||
6326 | if (!l2arc_write_eligible(guid, hdr)) { | |
6327 | mutex_exit(hash_lock); | |
6328 | continue; | |
6329 | } | |
6330 | ||
6331 | /* | |
6332 | * Assume that the buffer is not going to be compressed | |
6333 | * and could take more space on disk because of a larger | |
6334 | * disk block size. | |
6335 | */ | |
6336 | buf_sz = hdr->b_size; | |
6337 | buf_a_sz = vdev_psize_to_asize(dev->l2ad_vdev, buf_sz); | |
6338 | ||
6339 | if ((write_asize + buf_a_sz) > target_sz) { | |
6340 | full = B_TRUE; | |
6341 | mutex_exit(hash_lock); | |
6342 | break; | |
6343 | } | |
6344 | ||
6345 | if (pio == NULL) { | |
6346 | /* | |
6347 | * Insert a dummy header on the buflist so | |
6348 | * l2arc_write_done() can find where the | |
6349 | * write buffers begin without searching. | |
6350 | */ | |
6351 | mutex_enter(&dev->l2ad_mtx); | |
6352 | list_insert_head(&dev->l2ad_buflist, head); | |
6353 | mutex_exit(&dev->l2ad_mtx); | |
6354 | ||
6355 | cb = kmem_alloc( | |
6356 | sizeof (l2arc_write_callback_t), KM_SLEEP); | |
6357 | cb->l2wcb_dev = dev; | |
6358 | cb->l2wcb_head = head; | |
6359 | pio = zio_root(spa, l2arc_write_done, cb, | |
6360 | ZIO_FLAG_CANFAIL); | |
6361 | } | |
6362 | ||
6363 | /* | |
6364 | * Create and add a new L2ARC header. | |
6365 | */ | |
6366 | hdr->b_l2hdr.b_dev = dev; | |
6367 | hdr->b_flags |= ARC_FLAG_L2_WRITING; | |
6368 | /* | |
6369 | * Temporarily stash the data buffer in b_tmp_cdata. | |
6370 | * The subsequent write step will pick it up from | |
6371 | * there. This is because can't access b_l1hdr.b_buf | |
6372 | * without holding the hash_lock, which we in turn | |
6373 | * can't access without holding the ARC list locks | |
6374 | * (which we want to avoid during compression/writing) | |
6375 | */ | |
6376 | hdr->b_l2hdr.b_compress = ZIO_COMPRESS_OFF; | |
6377 | hdr->b_l2hdr.b_asize = hdr->b_size; | |
6378 | hdr->b_l2hdr.b_hits = 0; | |
6379 | hdr->b_l1hdr.b_tmp_cdata = hdr->b_l1hdr.b_buf->b_data; | |
6380 | ||
6381 | /* | |
6382 | * Explicitly set the b_daddr field to a known | |
6383 | * value which means "invalid address". This | |
6384 | * enables us to differentiate which stage of | |
6385 | * l2arc_write_buffers() the particular header | |
6386 | * is in (e.g. this loop, or the one below). | |
6387 | * ARC_FLAG_L2_WRITING is not enough to make | |
6388 | * this distinction, and we need to know in | |
6389 | * order to do proper l2arc vdev accounting in | |
6390 | * arc_release() and arc_hdr_destroy(). | |
6391 | * | |
6392 | * Note, we can't use a new flag to distinguish | |
6393 | * the two stages because we don't hold the | |
6394 | * header's hash_lock below, in the second stage | |
6395 | * of this function. Thus, we can't simply | |
6396 | * change the b_flags field to denote that the | |
6397 | * IO has been sent. We can change the b_daddr | |
6398 | * field of the L2 portion, though, since we'll | |
6399 | * be holding the l2ad_mtx; which is why we're | |
6400 | * using it to denote the header's state change. | |
6401 | */ | |
6402 | hdr->b_l2hdr.b_daddr = L2ARC_ADDR_UNSET; | |
6403 | hdr->b_flags |= ARC_FLAG_HAS_L2HDR; | |
6404 | ||
6405 | mutex_enter(&dev->l2ad_mtx); | |
6406 | list_insert_head(&dev->l2ad_buflist, hdr); | |
6407 | mutex_exit(&dev->l2ad_mtx); | |
6408 | ||
6409 | /* | |
6410 | * Compute and store the buffer cksum before | |
6411 | * writing. On debug the cksum is verified first. | |
6412 | */ | |
6413 | arc_cksum_verify(hdr->b_l1hdr.b_buf); | |
6414 | arc_cksum_compute(hdr->b_l1hdr.b_buf, B_TRUE); | |
6415 | ||
6416 | mutex_exit(hash_lock); | |
6417 | ||
6418 | write_sz += buf_sz; | |
6419 | write_asize += buf_a_sz; | |
6420 | } | |
6421 | ||
6422 | multilist_sublist_unlock(mls); | |
6423 | ||
6424 | if (full == B_TRUE) | |
6425 | break; | |
6426 | } | |
6427 | ||
6428 | /* No buffers selected for writing? */ | |
6429 | if (pio == NULL) { | |
6430 | ASSERT0(write_sz); | |
6431 | ASSERT(!HDR_HAS_L1HDR(head)); | |
6432 | kmem_cache_free(hdr_l2only_cache, head); | |
6433 | return (0); | |
6434 | } | |
6435 | ||
6436 | mutex_enter(&dev->l2ad_mtx); | |
6437 | ||
6438 | /* | |
6439 | * Note that elsewhere in this file arcstat_l2_asize | |
6440 | * and the used space on l2ad_vdev are updated using b_asize, | |
6441 | * which is not necessarily rounded up to the device block size. | |
6442 | * Too keep accounting consistent we do the same here as well: | |
6443 | * stats_size accumulates the sum of b_asize of the written buffers, | |
6444 | * while write_asize accumulates the sum of b_asize rounded up | |
6445 | * to the device block size. | |
6446 | * The latter sum is used only to validate the corectness of the code. | |
6447 | */ | |
6448 | stats_size = 0; | |
6449 | write_asize = 0; | |
6450 | ||
6451 | /* | |
6452 | * Now start writing the buffers. We're starting at the write head | |
6453 | * and work backwards, retracing the course of the buffer selector | |
6454 | * loop above. | |
6455 | */ | |
6456 | for (hdr = list_prev(&dev->l2ad_buflist, head); hdr; | |
6457 | hdr = list_prev(&dev->l2ad_buflist, hdr)) { | |
6458 | uint64_t buf_sz; | |
6459 | ||
6460 | /* | |
6461 | * We rely on the L1 portion of the header below, so | |
6462 | * it's invalid for this header to have been evicted out | |
6463 | * of the ghost cache, prior to being written out. The | |
6464 | * ARC_FLAG_L2_WRITING bit ensures this won't happen. | |
6465 | */ | |
6466 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
6467 | ||
6468 | /* | |
6469 | * We shouldn't need to lock the buffer here, since we flagged | |
6470 | * it as ARC_FLAG_L2_WRITING in the previous step, but we must | |
6471 | * take care to only access its L2 cache parameters. In | |
6472 | * particular, hdr->l1hdr.b_buf may be invalid by now due to | |
6473 | * ARC eviction. | |
6474 | */ | |
6475 | hdr->b_l2hdr.b_daddr = dev->l2ad_hand; | |
6476 | ||
6477 | if ((!l2arc_nocompress && HDR_L2COMPRESS(hdr)) && | |
6478 | hdr->b_l2hdr.b_asize >= buf_compress_minsz) { | |
6479 | if (l2arc_compress_buf(hdr)) { | |
6480 | /* | |
6481 | * If compression succeeded, enable headroom | |
6482 | * boost on the next scan cycle. | |
6483 | */ | |
6484 | *headroom_boost = B_TRUE; | |
6485 | } | |
6486 | } | |
6487 | ||
6488 | /* | |
6489 | * Pick up the buffer data we had previously stashed away | |
6490 | * (and now potentially also compressed). | |
6491 | */ | |
6492 | buf_data = hdr->b_l1hdr.b_tmp_cdata; | |
6493 | buf_sz = hdr->b_l2hdr.b_asize; | |
6494 | ||
6495 | /* | |
6496 | * We need to do this regardless if buf_sz is zero or | |
6497 | * not, otherwise, when this l2hdr is evicted we'll | |
6498 | * remove a reference that was never added. | |
6499 | */ | |
6500 | (void) refcount_add_many(&dev->l2ad_alloc, buf_sz, hdr); | |
6501 | ||
6502 | /* Compression may have squashed the buffer to zero length. */ | |
6503 | if (buf_sz != 0) { | |
6504 | uint64_t buf_a_sz; | |
6505 | ||
6506 | wzio = zio_write_phys(pio, dev->l2ad_vdev, | |
6507 | dev->l2ad_hand, buf_sz, buf_data, ZIO_CHECKSUM_OFF, | |
6508 | NULL, NULL, ZIO_PRIORITY_ASYNC_WRITE, | |
6509 | ZIO_FLAG_CANFAIL, B_FALSE); | |
6510 | ||
6511 | DTRACE_PROBE2(l2arc__write, vdev_t *, dev->l2ad_vdev, | |
6512 | zio_t *, wzio); | |
6513 | (void) zio_nowait(wzio); | |
6514 | ||
6515 | stats_size += buf_sz; | |
6516 | ||
6517 | /* | |
6518 | * Keep the clock hand suitably device-aligned. | |
6519 | */ | |
6520 | buf_a_sz = vdev_psize_to_asize(dev->l2ad_vdev, buf_sz); | |
6521 | write_asize += buf_a_sz; | |
6522 | dev->l2ad_hand += buf_a_sz; | |
6523 | } | |
6524 | } | |
6525 | ||
6526 | mutex_exit(&dev->l2ad_mtx); | |
6527 | ||
6528 | ASSERT3U(write_asize, <=, target_sz); | |
6529 | ARCSTAT_BUMP(arcstat_l2_writes_sent); | |
6530 | ARCSTAT_INCR(arcstat_l2_write_bytes, write_asize); | |
6531 | ARCSTAT_INCR(arcstat_l2_size, write_sz); | |
6532 | ARCSTAT_INCR(arcstat_l2_asize, stats_size); | |
6533 | vdev_space_update(dev->l2ad_vdev, stats_size, 0, 0); | |
6534 | ||
6535 | /* | |
6536 | * Bump device hand to the device start if it is approaching the end. | |
6537 | * l2arc_evict() will already have evicted ahead for this case. | |
6538 | */ | |
6539 | if (dev->l2ad_hand >= (dev->l2ad_end - target_sz)) { | |
6540 | dev->l2ad_hand = dev->l2ad_start; | |
6541 | dev->l2ad_first = B_FALSE; | |
6542 | } | |
6543 | ||
6544 | dev->l2ad_writing = B_TRUE; | |
6545 | (void) zio_wait(pio); | |
6546 | dev->l2ad_writing = B_FALSE; | |
6547 | ||
6548 | return (write_asize); | |
6549 | } | |
6550 | ||
6551 | /* | |
6552 | * Compresses an L2ARC buffer. | |
6553 | * The data to be compressed must be prefilled in l1hdr.b_tmp_cdata and its | |
6554 | * size in l2hdr->b_asize. This routine tries to compress the data and | |
6555 | * depending on the compression result there are three possible outcomes: | |
6556 | * *) The buffer was incompressible. The original l2hdr contents were left | |
6557 | * untouched and are ready for writing to an L2 device. | |
6558 | * *) The buffer was all-zeros, so there is no need to write it to an L2 | |
6559 | * device. To indicate this situation b_tmp_cdata is NULL'ed, b_asize is | |
6560 | * set to zero and b_compress is set to ZIO_COMPRESS_EMPTY. | |
6561 | * *) Compression succeeded and b_tmp_cdata was replaced with a temporary | |
6562 | * data buffer which holds the compressed data to be written, and b_asize | |
6563 | * tells us how much data there is. b_compress is set to the appropriate | |
6564 | * compression algorithm. Once writing is done, invoke | |
6565 | * l2arc_release_cdata_buf on this l2hdr to free this temporary buffer. | |
6566 | * | |
6567 | * Returns B_TRUE if compression succeeded, or B_FALSE if it didn't (the | |
6568 | * buffer was incompressible). | |
6569 | */ | |
6570 | static boolean_t | |
6571 | l2arc_compress_buf(arc_buf_hdr_t *hdr) | |
6572 | { | |
6573 | void *cdata; | |
6574 | size_t csize, len, rounded; | |
6575 | l2arc_buf_hdr_t *l2hdr; | |
6576 | ||
6577 | ASSERT(HDR_HAS_L2HDR(hdr)); | |
6578 | ||
6579 | l2hdr = &hdr->b_l2hdr; | |
6580 | ||
6581 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
6582 | ASSERT3U(l2hdr->b_compress, ==, ZIO_COMPRESS_OFF); | |
6583 | ASSERT(hdr->b_l1hdr.b_tmp_cdata != NULL); | |
6584 | ||
6585 | len = l2hdr->b_asize; | |
6586 | cdata = zio_data_buf_alloc(len); | |
6587 | ASSERT3P(cdata, !=, NULL); | |
6588 | csize = zio_compress_data(ZIO_COMPRESS_LZ4, hdr->b_l1hdr.b_tmp_cdata, | |
6589 | cdata, l2hdr->b_asize); | |
6590 | ||
6591 | rounded = P2ROUNDUP(csize, (size_t)SPA_MINBLOCKSIZE); | |
6592 | if (rounded > csize) { | |
6593 | bzero((char *)cdata + csize, rounded - csize); | |
6594 | csize = rounded; | |
6595 | } | |
6596 | ||
6597 | if (csize == 0) { | |
6598 | /* zero block, indicate that there's nothing to write */ | |
6599 | zio_data_buf_free(cdata, len); | |
6600 | l2hdr->b_compress = ZIO_COMPRESS_EMPTY; | |
6601 | l2hdr->b_asize = 0; | |
6602 | hdr->b_l1hdr.b_tmp_cdata = NULL; | |
6603 | ARCSTAT_BUMP(arcstat_l2_compress_zeros); | |
6604 | return (B_TRUE); | |
6605 | } else if (csize > 0 && csize < len) { | |
6606 | /* | |
6607 | * Compression succeeded, we'll keep the cdata around for | |
6608 | * writing and release it afterwards. | |
6609 | */ | |
6610 | l2hdr->b_compress = ZIO_COMPRESS_LZ4; | |
6611 | l2hdr->b_asize = csize; | |
6612 | hdr->b_l1hdr.b_tmp_cdata = cdata; | |
6613 | ARCSTAT_BUMP(arcstat_l2_compress_successes); | |
6614 | return (B_TRUE); | |
6615 | } else { | |
6616 | /* | |
6617 | * Compression failed, release the compressed buffer. | |
6618 | * l2hdr will be left unmodified. | |
6619 | */ | |
6620 | zio_data_buf_free(cdata, len); | |
6621 | ARCSTAT_BUMP(arcstat_l2_compress_failures); | |
6622 | return (B_FALSE); | |
6623 | } | |
6624 | } | |
6625 | ||
6626 | /* | |
6627 | * Decompresses a zio read back from an l2arc device. On success, the | |
6628 | * underlying zio's io_data buffer is overwritten by the uncompressed | |
6629 | * version. On decompression error (corrupt compressed stream), the | |
6630 | * zio->io_error value is set to signal an I/O error. | |
6631 | * | |
6632 | * Please note that the compressed data stream is not checksummed, so | |
6633 | * if the underlying device is experiencing data corruption, we may feed | |
6634 | * corrupt data to the decompressor, so the decompressor needs to be | |
6635 | * able to handle this situation (LZ4 does). | |
6636 | */ | |
6637 | static void | |
6638 | l2arc_decompress_zio(zio_t *zio, arc_buf_hdr_t *hdr, enum zio_compress c) | |
6639 | { | |
6640 | uint64_t csize; | |
6641 | void *cdata; | |
6642 | ||
6643 | ASSERT(L2ARC_IS_VALID_COMPRESS(c)); | |
6644 | ||
6645 | if (zio->io_error != 0) { | |
6646 | /* | |
6647 | * An io error has occured, just restore the original io | |
6648 | * size in preparation for a main pool read. | |
6649 | */ | |
6650 | zio->io_orig_size = zio->io_size = hdr->b_size; | |
6651 | return; | |
6652 | } | |
6653 | ||
6654 | if (c == ZIO_COMPRESS_EMPTY) { | |
6655 | /* | |
6656 | * An empty buffer results in a null zio, which means we | |
6657 | * need to fill its io_data after we're done restoring the | |
6658 | * buffer's contents. | |
6659 | */ | |
6660 | ASSERT(hdr->b_l1hdr.b_buf != NULL); | |
6661 | bzero(hdr->b_l1hdr.b_buf->b_data, hdr->b_size); | |
6662 | zio->io_data = zio->io_orig_data = hdr->b_l1hdr.b_buf->b_data; | |
6663 | } else { | |
6664 | ASSERT(zio->io_data != NULL); | |
6665 | /* | |
6666 | * We copy the compressed data from the start of the arc buffer | |
6667 | * (the zio_read will have pulled in only what we need, the | |
6668 | * rest is garbage which we will overwrite at decompression) | |
6669 | * and then decompress back to the ARC data buffer. This way we | |
6670 | * can minimize copying by simply decompressing back over the | |
6671 | * original compressed data (rather than decompressing to an | |
6672 | * aux buffer and then copying back the uncompressed buffer, | |
6673 | * which is likely to be much larger). | |
6674 | */ | |
6675 | csize = zio->io_size; | |
6676 | cdata = zio_data_buf_alloc(csize); | |
6677 | bcopy(zio->io_data, cdata, csize); | |
6678 | if (zio_decompress_data(c, cdata, zio->io_data, csize, | |
6679 | hdr->b_size) != 0) | |
6680 | zio->io_error = EIO; | |
6681 | zio_data_buf_free(cdata, csize); | |
6682 | } | |
6683 | ||
6684 | /* Restore the expected uncompressed IO size. */ | |
6685 | zio->io_orig_size = zio->io_size = hdr->b_size; | |
6686 | } | |
6687 | ||
6688 | /* | |
6689 | * Releases the temporary b_tmp_cdata buffer in an l2arc header structure. | |
6690 | * This buffer serves as a temporary holder of compressed data while | |
6691 | * the buffer entry is being written to an l2arc device. Once that is | |
6692 | * done, we can dispose of it. | |
6693 | */ | |
6694 | static void | |
6695 | l2arc_release_cdata_buf(arc_buf_hdr_t *hdr) | |
6696 | { | |
6697 | enum zio_compress comp; | |
6698 | ||
6699 | ASSERT(HDR_HAS_L1HDR(hdr)); | |
6700 | ASSERT(HDR_HAS_L2HDR(hdr)); | |
6701 | comp = hdr->b_l2hdr.b_compress; | |
6702 | ASSERT(comp == ZIO_COMPRESS_OFF || L2ARC_IS_VALID_COMPRESS(comp)); | |
6703 | ||
6704 | if (comp == ZIO_COMPRESS_OFF) { | |
6705 | /* | |
6706 | * In this case, b_tmp_cdata points to the same buffer | |
6707 | * as the arc_buf_t's b_data field. We don't want to | |
6708 | * free it, since the arc_buf_t will handle that. | |
6709 | */ | |
6710 | hdr->b_l1hdr.b_tmp_cdata = NULL; | |
6711 | } else if (comp == ZIO_COMPRESS_EMPTY) { | |
6712 | /* | |
6713 | * In this case, b_tmp_cdata was compressed to an empty | |
6714 | * buffer, thus there's nothing to free and b_tmp_cdata | |
6715 | * should have been set to NULL in l2arc_write_buffers(). | |
6716 | */ | |
6717 | ASSERT3P(hdr->b_l1hdr.b_tmp_cdata, ==, NULL); | |
6718 | } else { | |
6719 | /* | |
6720 | * If the data was compressed, then we've allocated a | |
6721 | * temporary buffer for it, so now we need to release it. | |
6722 | */ | |
6723 | ASSERT(hdr->b_l1hdr.b_tmp_cdata != NULL); | |
6724 | zio_data_buf_free(hdr->b_l1hdr.b_tmp_cdata, | |
6725 | hdr->b_size); | |
6726 | hdr->b_l1hdr.b_tmp_cdata = NULL; | |
6727 | } | |
6728 | ||
6729 | } | |
6730 | ||
6731 | /* | |
6732 | * This thread feeds the L2ARC at regular intervals. This is the beating | |
6733 | * heart of the L2ARC. | |
6734 | */ | |
6735 | static void | |
6736 | l2arc_feed_thread(void) | |
6737 | { | |
6738 | callb_cpr_t cpr; | |
6739 | l2arc_dev_t *dev; | |
6740 | spa_t *spa; | |
6741 | uint64_t size, wrote; | |
6742 | clock_t begin, next = ddi_get_lbolt(); | |
6743 | boolean_t headroom_boost = B_FALSE; | |
6744 | fstrans_cookie_t cookie; | |
6745 | ||
6746 | CALLB_CPR_INIT(&cpr, &l2arc_feed_thr_lock, callb_generic_cpr, FTAG); | |
6747 | ||
6748 | mutex_enter(&l2arc_feed_thr_lock); | |
6749 | ||
6750 | cookie = spl_fstrans_mark(); | |
6751 | while (l2arc_thread_exit == 0) { | |
6752 | CALLB_CPR_SAFE_BEGIN(&cpr); | |
6753 | (void) cv_timedwait_sig(&l2arc_feed_thr_cv, | |
6754 | &l2arc_feed_thr_lock, next); | |
6755 | CALLB_CPR_SAFE_END(&cpr, &l2arc_feed_thr_lock); | |
6756 | next = ddi_get_lbolt() + hz; | |
6757 | ||
6758 | /* | |
6759 | * Quick check for L2ARC devices. | |
6760 | */ | |
6761 | mutex_enter(&l2arc_dev_mtx); | |
6762 | if (l2arc_ndev == 0) { | |
6763 | mutex_exit(&l2arc_dev_mtx); | |
6764 | continue; | |
6765 | } | |
6766 | mutex_exit(&l2arc_dev_mtx); | |
6767 | begin = ddi_get_lbolt(); | |
6768 | ||
6769 | /* | |
6770 | * This selects the next l2arc device to write to, and in | |
6771 | * doing so the next spa to feed from: dev->l2ad_spa. This | |
6772 | * will return NULL if there are now no l2arc devices or if | |
6773 | * they are all faulted. | |
6774 | * | |
6775 | * If a device is returned, its spa's config lock is also | |
6776 | * held to prevent device removal. l2arc_dev_get_next() | |
6777 | * will grab and release l2arc_dev_mtx. | |
6778 | */ | |
6779 | if ((dev = l2arc_dev_get_next()) == NULL) | |
6780 | continue; | |
6781 | ||
6782 | spa = dev->l2ad_spa; | |
6783 | ASSERT(spa != NULL); | |
6784 | ||
6785 | /* | |
6786 | * If the pool is read-only then force the feed thread to | |
6787 | * sleep a little longer. | |
6788 | */ | |
6789 | if (!spa_writeable(spa)) { | |
6790 | next = ddi_get_lbolt() + 5 * l2arc_feed_secs * hz; | |
6791 | spa_config_exit(spa, SCL_L2ARC, dev); | |
6792 | continue; | |
6793 | } | |
6794 | ||
6795 | /* | |
6796 | * Avoid contributing to memory pressure. | |
6797 | */ | |
6798 | if (arc_reclaim_needed()) { | |
6799 | ARCSTAT_BUMP(arcstat_l2_abort_lowmem); | |
6800 | spa_config_exit(spa, SCL_L2ARC, dev); | |
6801 | continue; | |
6802 | } | |
6803 | ||
6804 | ARCSTAT_BUMP(arcstat_l2_feeds); | |
6805 | ||
6806 | size = l2arc_write_size(); | |
6807 | ||
6808 | /* | |
6809 | * Evict L2ARC buffers that will be overwritten. | |
6810 | */ | |
6811 | l2arc_evict(dev, size, B_FALSE); | |
6812 | ||
6813 | /* | |
6814 | * Write ARC buffers. | |
6815 | */ | |
6816 | wrote = l2arc_write_buffers(spa, dev, size, &headroom_boost); | |
6817 | ||
6818 | /* | |
6819 | * Calculate interval between writes. | |
6820 | */ | |
6821 | next = l2arc_write_interval(begin, size, wrote); | |
6822 | spa_config_exit(spa, SCL_L2ARC, dev); | |
6823 | } | |
6824 | spl_fstrans_unmark(cookie); | |
6825 | ||
6826 | l2arc_thread_exit = 0; | |
6827 | cv_broadcast(&l2arc_feed_thr_cv); | |
6828 | CALLB_CPR_EXIT(&cpr); /* drops l2arc_feed_thr_lock */ | |
6829 | thread_exit(); | |
6830 | } | |
6831 | ||
6832 | boolean_t | |
6833 | l2arc_vdev_present(vdev_t *vd) | |
6834 | { | |
6835 | l2arc_dev_t *dev; | |
6836 | ||
6837 | mutex_enter(&l2arc_dev_mtx); | |
6838 | for (dev = list_head(l2arc_dev_list); dev != NULL; | |
6839 | dev = list_next(l2arc_dev_list, dev)) { | |
6840 | if (dev->l2ad_vdev == vd) | |
6841 | break; | |
6842 | } | |
6843 | mutex_exit(&l2arc_dev_mtx); | |
6844 | ||
6845 | return (dev != NULL); | |
6846 | } | |
6847 | ||
6848 | /* | |
6849 | * Add a vdev for use by the L2ARC. By this point the spa has already | |
6850 | * validated the vdev and opened it. | |
6851 | */ | |
6852 | void | |
6853 | l2arc_add_vdev(spa_t *spa, vdev_t *vd) | |
6854 | { | |
6855 | l2arc_dev_t *adddev; | |
6856 | ||
6857 | ASSERT(!l2arc_vdev_present(vd)); | |
6858 | ||
6859 | /* | |
6860 | * Create a new l2arc device entry. | |
6861 | */ | |
6862 | adddev = kmem_zalloc(sizeof (l2arc_dev_t), KM_SLEEP); | |
6863 | adddev->l2ad_spa = spa; | |
6864 | adddev->l2ad_vdev = vd; | |
6865 | adddev->l2ad_start = VDEV_LABEL_START_SIZE; | |
6866 | adddev->l2ad_end = VDEV_LABEL_START_SIZE + vdev_get_min_asize(vd); | |
6867 | adddev->l2ad_hand = adddev->l2ad_start; | |
6868 | adddev->l2ad_first = B_TRUE; | |
6869 | adddev->l2ad_writing = B_FALSE; | |
6870 | list_link_init(&adddev->l2ad_node); | |
6871 | ||
6872 | mutex_init(&adddev->l2ad_mtx, NULL, MUTEX_DEFAULT, NULL); | |
6873 | /* | |
6874 | * This is a list of all ARC buffers that are still valid on the | |
6875 | * device. | |
6876 | */ | |
6877 | list_create(&adddev->l2ad_buflist, sizeof (arc_buf_hdr_t), | |
6878 | offsetof(arc_buf_hdr_t, b_l2hdr.b_l2node)); | |
6879 | ||
6880 | vdev_space_update(vd, 0, 0, adddev->l2ad_end - adddev->l2ad_hand); | |
6881 | refcount_create(&adddev->l2ad_alloc); | |
6882 | ||
6883 | /* | |
6884 | * Add device to global list | |
6885 | */ | |
6886 | mutex_enter(&l2arc_dev_mtx); | |
6887 | list_insert_head(l2arc_dev_list, adddev); | |
6888 | atomic_inc_64(&l2arc_ndev); | |
6889 | mutex_exit(&l2arc_dev_mtx); | |
6890 | } | |
6891 | ||
6892 | /* | |
6893 | * Remove a vdev from the L2ARC. | |
6894 | */ | |
6895 | void | |
6896 | l2arc_remove_vdev(vdev_t *vd) | |
6897 | { | |
6898 | l2arc_dev_t *dev, *nextdev, *remdev = NULL; | |
6899 | ||
6900 | /* | |
6901 | * Find the device by vdev | |
6902 | */ | |
6903 | mutex_enter(&l2arc_dev_mtx); | |
6904 | for (dev = list_head(l2arc_dev_list); dev; dev = nextdev) { | |
6905 | nextdev = list_next(l2arc_dev_list, dev); | |
6906 | if (vd == dev->l2ad_vdev) { | |
6907 | remdev = dev; | |
6908 | break; | |
6909 | } | |
6910 | } | |
6911 | ASSERT(remdev != NULL); | |
6912 | ||
6913 | /* | |
6914 | * Remove device from global list | |
6915 | */ | |
6916 | list_remove(l2arc_dev_list, remdev); | |
6917 | l2arc_dev_last = NULL; /* may have been invalidated */ | |
6918 | atomic_dec_64(&l2arc_ndev); | |
6919 | mutex_exit(&l2arc_dev_mtx); | |
6920 | ||
6921 | /* | |
6922 | * Clear all buflists and ARC references. L2ARC device flush. | |
6923 | */ | |
6924 | l2arc_evict(remdev, 0, B_TRUE); | |
6925 | list_destroy(&remdev->l2ad_buflist); | |
6926 | mutex_destroy(&remdev->l2ad_mtx); | |
6927 | refcount_destroy(&remdev->l2ad_alloc); | |
6928 | kmem_free(remdev, sizeof (l2arc_dev_t)); | |
6929 | } | |
6930 | ||
6931 | void | |
6932 | l2arc_init(void) | |
6933 | { | |
6934 | l2arc_thread_exit = 0; | |
6935 | l2arc_ndev = 0; | |
6936 | l2arc_writes_sent = 0; | |
6937 | l2arc_writes_done = 0; | |
6938 | ||
6939 | mutex_init(&l2arc_feed_thr_lock, NULL, MUTEX_DEFAULT, NULL); | |
6940 | cv_init(&l2arc_feed_thr_cv, NULL, CV_DEFAULT, NULL); | |
6941 | mutex_init(&l2arc_dev_mtx, NULL, MUTEX_DEFAULT, NULL); | |
6942 | mutex_init(&l2arc_free_on_write_mtx, NULL, MUTEX_DEFAULT, NULL); | |
6943 | ||
6944 | l2arc_dev_list = &L2ARC_dev_list; | |
6945 | l2arc_free_on_write = &L2ARC_free_on_write; | |
6946 | list_create(l2arc_dev_list, sizeof (l2arc_dev_t), | |
6947 | offsetof(l2arc_dev_t, l2ad_node)); | |
6948 | list_create(l2arc_free_on_write, sizeof (l2arc_data_free_t), | |
6949 | offsetof(l2arc_data_free_t, l2df_list_node)); | |
6950 | } | |
6951 | ||
6952 | void | |
6953 | l2arc_fini(void) | |
6954 | { | |
6955 | /* | |
6956 | * This is called from dmu_fini(), which is called from spa_fini(); | |
6957 | * Because of this, we can assume that all l2arc devices have | |
6958 | * already been removed when the pools themselves were removed. | |
6959 | */ | |
6960 | ||
6961 | l2arc_do_free_on_write(); | |
6962 | ||
6963 | mutex_destroy(&l2arc_feed_thr_lock); | |
6964 | cv_destroy(&l2arc_feed_thr_cv); | |
6965 | mutex_destroy(&l2arc_dev_mtx); | |
6966 | mutex_destroy(&l2arc_free_on_write_mtx); | |
6967 | ||
6968 | list_destroy(l2arc_dev_list); | |
6969 | list_destroy(l2arc_free_on_write); | |
6970 | } | |
6971 | ||
6972 | void | |
6973 | l2arc_start(void) | |
6974 | { | |
6975 | if (!(spa_mode_global & FWRITE)) | |
6976 | return; | |
6977 | ||
6978 | (void) thread_create(NULL, 0, l2arc_feed_thread, NULL, 0, &p0, | |
6979 | TS_RUN, defclsyspri); | |
6980 | } | |
6981 | ||
6982 | void | |
6983 | l2arc_stop(void) | |
6984 | { | |
6985 | if (!(spa_mode_global & FWRITE)) | |
6986 | return; | |
6987 | ||
6988 | mutex_enter(&l2arc_feed_thr_lock); | |
6989 | cv_signal(&l2arc_feed_thr_cv); /* kick thread out of startup */ | |
6990 | l2arc_thread_exit = 1; | |
6991 | while (l2arc_thread_exit != 0) | |
6992 | cv_wait(&l2arc_feed_thr_cv, &l2arc_feed_thr_lock); | |
6993 | mutex_exit(&l2arc_feed_thr_lock); | |
6994 | } | |
6995 | ||
6996 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
6997 | EXPORT_SYMBOL(arc_buf_size); | |
6998 | EXPORT_SYMBOL(arc_write); | |
6999 | EXPORT_SYMBOL(arc_read); | |
7000 | EXPORT_SYMBOL(arc_buf_remove_ref); | |
7001 | EXPORT_SYMBOL(arc_buf_info); | |
7002 | EXPORT_SYMBOL(arc_getbuf_func); | |
7003 | EXPORT_SYMBOL(arc_add_prune_callback); | |
7004 | EXPORT_SYMBOL(arc_remove_prune_callback); | |
7005 | ||
7006 | module_param(zfs_arc_min, ulong, 0644); | |
7007 | MODULE_PARM_DESC(zfs_arc_min, "Min arc size"); | |
7008 | ||
7009 | module_param(zfs_arc_max, ulong, 0644); | |
7010 | MODULE_PARM_DESC(zfs_arc_max, "Max arc size"); | |
7011 | ||
7012 | module_param(zfs_arc_meta_limit, ulong, 0644); | |
7013 | MODULE_PARM_DESC(zfs_arc_meta_limit, "Meta limit for arc size"); | |
7014 | ||
7015 | module_param(zfs_arc_meta_min, ulong, 0644); | |
7016 | MODULE_PARM_DESC(zfs_arc_meta_min, "Min arc metadata"); | |
7017 | ||
7018 | module_param(zfs_arc_meta_prune, int, 0644); | |
7019 | MODULE_PARM_DESC(zfs_arc_meta_prune, "Meta objects to scan for prune"); | |
7020 | ||
7021 | module_param(zfs_arc_meta_adjust_restarts, int, 0644); | |
7022 | MODULE_PARM_DESC(zfs_arc_meta_adjust_restarts, | |
7023 | "Limit number of restarts in arc_adjust_meta"); | |
7024 | ||
7025 | module_param(zfs_arc_meta_strategy, int, 0644); | |
7026 | MODULE_PARM_DESC(zfs_arc_meta_strategy, "Meta reclaim strategy"); | |
7027 | ||
7028 | module_param(zfs_arc_grow_retry, int, 0644); | |
7029 | MODULE_PARM_DESC(zfs_arc_grow_retry, "Seconds before growing arc size"); | |
7030 | ||
7031 | module_param(zfs_arc_p_aggressive_disable, int, 0644); | |
7032 | MODULE_PARM_DESC(zfs_arc_p_aggressive_disable, "disable aggressive arc_p grow"); | |
7033 | ||
7034 | module_param(zfs_arc_p_dampener_disable, int, 0644); | |
7035 | MODULE_PARM_DESC(zfs_arc_p_dampener_disable, "disable arc_p adapt dampener"); | |
7036 | ||
7037 | module_param(zfs_arc_shrink_shift, int, 0644); | |
7038 | MODULE_PARM_DESC(zfs_arc_shrink_shift, "log2(fraction of arc to reclaim)"); | |
7039 | ||
7040 | module_param(zfs_arc_p_min_shift, int, 0644); | |
7041 | MODULE_PARM_DESC(zfs_arc_p_min_shift, "arc_c shift to calc min/max arc_p"); | |
7042 | ||
7043 | module_param(zfs_disable_dup_eviction, int, 0644); | |
7044 | MODULE_PARM_DESC(zfs_disable_dup_eviction, "disable duplicate buffer eviction"); | |
7045 | ||
7046 | module_param(zfs_arc_average_blocksize, int, 0444); | |
7047 | MODULE_PARM_DESC(zfs_arc_average_blocksize, "Target average block size"); | |
7048 | ||
7049 | module_param(zfs_arc_min_prefetch_lifespan, int, 0644); | |
7050 | MODULE_PARM_DESC(zfs_arc_min_prefetch_lifespan, "Min life of prefetch block"); | |
7051 | ||
7052 | module_param(zfs_arc_num_sublists_per_state, int, 0644); | |
7053 | MODULE_PARM_DESC(zfs_arc_num_sublists_per_state, | |
7054 | "Number of sublists used in each of the ARC state lists"); | |
7055 | ||
7056 | module_param(l2arc_write_max, ulong, 0644); | |
7057 | MODULE_PARM_DESC(l2arc_write_max, "Max write bytes per interval"); | |
7058 | ||
7059 | module_param(l2arc_write_boost, ulong, 0644); | |
7060 | MODULE_PARM_DESC(l2arc_write_boost, "Extra write bytes during device warmup"); | |
7061 | ||
7062 | module_param(l2arc_headroom, ulong, 0644); | |
7063 | MODULE_PARM_DESC(l2arc_headroom, "Number of max device writes to precache"); | |
7064 | ||
7065 | module_param(l2arc_headroom_boost, ulong, 0644); | |
7066 | MODULE_PARM_DESC(l2arc_headroom_boost, "Compressed l2arc_headroom multiplier"); | |
7067 | ||
7068 | module_param(l2arc_feed_secs, ulong, 0644); | |
7069 | MODULE_PARM_DESC(l2arc_feed_secs, "Seconds between L2ARC writing"); | |
7070 | ||
7071 | module_param(l2arc_feed_min_ms, ulong, 0644); | |
7072 | MODULE_PARM_DESC(l2arc_feed_min_ms, "Min feed interval in milliseconds"); | |
7073 | ||
7074 | module_param(l2arc_noprefetch, int, 0644); | |
7075 | MODULE_PARM_DESC(l2arc_noprefetch, "Skip caching prefetched buffers"); | |
7076 | ||
7077 | module_param(l2arc_nocompress, int, 0644); | |
7078 | MODULE_PARM_DESC(l2arc_nocompress, "Skip compressing L2ARC buffers"); | |
7079 | ||
7080 | module_param(l2arc_feed_again, int, 0644); | |
7081 | MODULE_PARM_DESC(l2arc_feed_again, "Turbo L2ARC warmup"); | |
7082 | ||
7083 | module_param(l2arc_norw, int, 0644); | |
7084 | MODULE_PARM_DESC(l2arc_norw, "No reads during writes"); | |
7085 | ||
7086 | module_param(zfs_arc_lotsfree_percent, int, 0644); | |
7087 | MODULE_PARM_DESC(zfs_arc_lotsfree_percent, | |
7088 | "System free memory I/O throttle in bytes"); | |
7089 | ||
7090 | module_param(zfs_arc_sys_free, ulong, 0644); | |
7091 | MODULE_PARM_DESC(zfs_arc_sys_free, "System free memory target size in bytes"); | |
7092 | ||
7093 | #endif |