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 * Note that the majority of the performance stats are manipulated 112 * with atomic operations. 113 * 114 * The L2ARC uses the l2ad_mtx on each vdev for the following: 115 * 116 * - L2ARC buflist creation 117 * - L2ARC buflist eviction 118 * - L2ARC write completion, which walks L2ARC buflists 119 * - ARC header destruction, as it removes from L2ARC buflists 120 * - ARC header release, as it removes from L2ARC buflists 121 */ 122 123 #include <sys/spa.h> 124 #include <sys/zio.h> 125 #include <sys/zio_compress.h> 126 #include <sys/zfs_context.h> 127 #include <sys/arc.h> 128 #include <sys/refcount.h> 129 #include <sys/vdev.h> 130 #include <sys/vdev_impl.h> 131 #include <sys/dsl_pool.h> 132 #include <sys/multilist.h> 133 #ifdef _KERNEL 134 #include <sys/vmsystm.h> 135 #include <vm/anon.h> 136 #include <sys/fs/swapnode.h> 137 #include <sys/dnlc.h> 138 #endif 139 #include <sys/callb.h> 140 #include <sys/kstat.h> 141 #include <zfs_fletcher.h> 142 143 #ifndef _KERNEL 144 /* set with ZFS_DEBUG=watch, to enable watchpoints on frozen buffers */ 145 boolean_t arc_watch = B_FALSE; 146 int arc_procfd; 147 #endif 148 149 static kmutex_t arc_reclaim_lock; 150 static kcondvar_t arc_reclaim_thread_cv; 151 static boolean_t arc_reclaim_thread_exit; 152 static kcondvar_t arc_reclaim_waiters_cv; 153 154 static kmutex_t arc_user_evicts_lock; 155 static kcondvar_t arc_user_evicts_cv; 156 static boolean_t arc_user_evicts_thread_exit; 157 158 uint_t arc_reduce_dnlc_percent = 3; 159 160 /* 161 * The number of headers to evict in arc_evict_state_impl() before 162 * dropping the sublist lock and evicting from another sublist. A lower 163 * value means we're more likely to evict the "correct" header (i.e. the 164 * oldest header in the arc state), but comes with higher overhead 165 * (i.e. more invocations of arc_evict_state_impl()). 166 */ 167 int zfs_arc_evict_batch_limit = 10; 168 169 /* 170 * The number of sublists used for each of the arc state lists. If this 171 * is not set to a suitable value by the user, it will be configured to 172 * the number of CPUs on the system in arc_init(). 173 */ 174 int zfs_arc_num_sublists_per_state = 0; 175 176 /* number of seconds before growing cache again */ 177 static int arc_grow_retry = 60; 178 179 /* shift of arc_c for calculating overflow limit in arc_get_data_buf */ 180 int zfs_arc_overflow_shift = 8; 181 182 /* shift of arc_c for calculating both min and max arc_p */ 183 static int arc_p_min_shift = 4; 184 185 /* log2(fraction of arc to reclaim) */ 186 static int arc_shrink_shift = 7; 187 188 /* 189 * log2(fraction of ARC which must be free to allow growing). 190 * I.e. If there is less than arc_c >> arc_no_grow_shift free memory, 191 * when reading a new block into the ARC, we will evict an equal-sized block 192 * from the ARC. 193 * 194 * This must be less than arc_shrink_shift, so that when we shrink the ARC, 195 * we will still not allow it to grow. 196 */ 197 int arc_no_grow_shift = 5; 198 199 200 /* 201 * minimum lifespan of a prefetch block in clock ticks 202 * (initialized in arc_init()) 203 */ 204 static int arc_min_prefetch_lifespan; 205 206 /* 207 * If this percent of memory is free, don't throttle. 208 */ 209 int arc_lotsfree_percent = 10; 210 211 static int arc_dead; 212 213 /* 214 * The arc has filled available memory and has now warmed up. 215 */ 216 static boolean_t arc_warm; 217 218 /* 219 * These tunables are for performance analysis. 220 */ 221 uint64_t zfs_arc_max; 222 uint64_t zfs_arc_min; 223 uint64_t zfs_arc_meta_limit = 0; 224 uint64_t zfs_arc_meta_min = 0; 225 int zfs_arc_grow_retry = 0; 226 int zfs_arc_shrink_shift = 0; 227 int zfs_arc_p_min_shift = 0; 228 int zfs_disable_dup_eviction = 0; 229 int zfs_arc_average_blocksize = 8 * 1024; /* 8KB */ 230 231 /* 232 * Note that buffers can be in one of 6 states: 233 * ARC_anon - anonymous (discussed below) 234 * ARC_mru - recently used, currently cached 235 * ARC_mru_ghost - recentely used, no longer in cache 236 * ARC_mfu - frequently used, currently cached 237 * ARC_mfu_ghost - frequently used, no longer in cache 238 * ARC_l2c_only - exists in L2ARC but not other states 239 * When there are no active references to the buffer, they are 240 * are linked onto a list in one of these arc states. These are 241 * the only buffers that can be evicted or deleted. Within each 242 * state there are multiple lists, one for meta-data and one for 243 * non-meta-data. Meta-data (indirect blocks, blocks of dnodes, 244 * etc.) is tracked separately so that it can be managed more 245 * explicitly: favored over data, limited explicitly. 246 * 247 * Anonymous buffers are buffers that are not associated with 248 * a DVA. These are buffers that hold dirty block copies 249 * before they are written to stable storage. By definition, 250 * they are "ref'd" and are considered part of arc_mru 251 * that cannot be freed. Generally, they will aquire a DVA 252 * as they are written and migrate onto the arc_mru list. 253 * 254 * The ARC_l2c_only state is for buffers that are in the second 255 * level ARC but no longer in any of the ARC_m* lists. The second 256 * level ARC itself may also contain buffers that are in any of 257 * the ARC_m* states - meaning that a buffer can exist in two 258 * places. The reason for the ARC_l2c_only state is to keep the 259 * buffer header in the hash table, so that reads that hit the 260 * second level ARC benefit from these fast lookups. 261 */ 262 263 typedef struct arc_state { 264 /* 265 * list of evictable buffers 266 */ 267 multilist_t arcs_list[ARC_BUFC_NUMTYPES]; 268 /* 269 * total amount of evictable data in this state 270 */ 271 uint64_t arcs_lsize[ARC_BUFC_NUMTYPES]; 272 /* 273 * total amount of data in this state; this includes: evictable, 274 * non-evictable, ARC_BUFC_DATA, and ARC_BUFC_METADATA. 275 */ 276 refcount_t arcs_size; 277 } arc_state_t; 278 279 /* The 6 states: */ 280 static arc_state_t ARC_anon; 281 static arc_state_t ARC_mru; 282 static arc_state_t ARC_mru_ghost; 283 static arc_state_t ARC_mfu; 284 static arc_state_t ARC_mfu_ghost; 285 static arc_state_t ARC_l2c_only; 286 287 typedef struct arc_stats { 288 kstat_named_t arcstat_hits; 289 kstat_named_t arcstat_misses; 290 kstat_named_t arcstat_demand_data_hits; 291 kstat_named_t arcstat_demand_data_misses; 292 kstat_named_t arcstat_demand_metadata_hits; 293 kstat_named_t arcstat_demand_metadata_misses; 294 kstat_named_t arcstat_prefetch_data_hits; 295 kstat_named_t arcstat_prefetch_data_misses; 296 kstat_named_t arcstat_prefetch_metadata_hits; 297 kstat_named_t arcstat_prefetch_metadata_misses; 298 kstat_named_t arcstat_mru_hits; 299 kstat_named_t arcstat_mru_ghost_hits; 300 kstat_named_t arcstat_mfu_hits; 301 kstat_named_t arcstat_mfu_ghost_hits; 302 kstat_named_t arcstat_deleted; 303 /* 304 * Number of buffers that could not be evicted because the hash lock 305 * was held by another thread. The lock may not necessarily be held 306 * by something using the same buffer, since hash locks are shared 307 * by multiple buffers. 308 */ 309 kstat_named_t arcstat_mutex_miss; 310 /* 311 * Number of buffers skipped because they have I/O in progress, are 312 * indrect prefetch buffers that have not lived long enough, or are 313 * not from the spa we're trying to evict from. 314 */ 315 kstat_named_t arcstat_evict_skip; 316 /* 317 * Number of times arc_evict_state() was unable to evict enough 318 * buffers to reach it's target amount. 319 */ 320 kstat_named_t arcstat_evict_not_enough; 321 kstat_named_t arcstat_evict_l2_cached; 322 kstat_named_t arcstat_evict_l2_eligible; 323 kstat_named_t arcstat_evict_l2_ineligible; 324 kstat_named_t arcstat_evict_l2_skip; 325 kstat_named_t arcstat_hash_elements; 326 kstat_named_t arcstat_hash_elements_max; 327 kstat_named_t arcstat_hash_collisions; 328 kstat_named_t arcstat_hash_chains; 329 kstat_named_t arcstat_hash_chain_max; 330 kstat_named_t arcstat_p; 331 kstat_named_t arcstat_c; 332 kstat_named_t arcstat_c_min; 333 kstat_named_t arcstat_c_max; 334 kstat_named_t arcstat_size; 335 /* 336 * Number of bytes consumed by internal ARC structures necessary 337 * for tracking purposes; these structures are not actually 338 * backed by ARC buffers. This includes arc_buf_hdr_t structures 339 * (allocated via arc_buf_hdr_t_full and arc_buf_hdr_t_l2only 340 * caches), and arc_buf_t structures (allocated via arc_buf_t 341 * cache). 342 */ 343 kstat_named_t arcstat_hdr_size; 344 /* 345 * Number of bytes consumed by ARC buffers of type equal to 346 * ARC_BUFC_DATA. This is generally consumed by buffers backing 347 * on disk user data (e.g. plain file contents). 348 */ 349 kstat_named_t arcstat_data_size; 350 /* 351 * Number of bytes consumed by ARC buffers of type equal to 352 * ARC_BUFC_METADATA. This is generally consumed by buffers 353 * backing on disk data that is used for internal ZFS 354 * structures (e.g. ZAP, dnode, indirect blocks, etc). 355 */ 356 kstat_named_t arcstat_metadata_size; 357 /* 358 * Number of bytes consumed by various buffers and structures 359 * not actually backed with ARC buffers. This includes bonus 360 * buffers (allocated directly via zio_buf_* functions), 361 * dmu_buf_impl_t structures (allocated via dmu_buf_impl_t 362 * cache), and dnode_t structures (allocated via dnode_t cache). 363 */ 364 kstat_named_t arcstat_other_size; 365 /* 366 * Total number of bytes consumed by ARC buffers residing in the 367 * arc_anon state. This includes *all* buffers in the arc_anon 368 * state; e.g. data, metadata, evictable, and unevictable buffers 369 * are all included in this value. 370 */ 371 kstat_named_t arcstat_anon_size; 372 /* 373 * Number of bytes consumed by ARC buffers that meet the 374 * following criteria: backing buffers of type ARC_BUFC_DATA, 375 * residing in the arc_anon state, and are eligible for eviction 376 * (e.g. have no outstanding holds on the buffer). 377 */ 378 kstat_named_t arcstat_anon_evictable_data; 379 /* 380 * Number of bytes consumed by ARC buffers that meet the 381 * following criteria: backing buffers of type ARC_BUFC_METADATA, 382 * residing in the arc_anon state, and are eligible for eviction 383 * (e.g. have no outstanding holds on the buffer). 384 */ 385 kstat_named_t arcstat_anon_evictable_metadata; 386 /* 387 * Total number of bytes consumed by ARC buffers residing in the 388 * arc_mru state. This includes *all* buffers in the arc_mru 389 * state; e.g. data, metadata, evictable, and unevictable buffers 390 * are all included in this value. 391 */ 392 kstat_named_t arcstat_mru_size; 393 /* 394 * Number of bytes consumed by ARC buffers that meet the 395 * following criteria: backing buffers of type ARC_BUFC_DATA, 396 * residing in the arc_mru state, and are eligible for eviction 397 * (e.g. have no outstanding holds on the buffer). 398 */ 399 kstat_named_t arcstat_mru_evictable_data; 400 /* 401 * Number of bytes consumed by ARC buffers that meet the 402 * following criteria: backing buffers of type ARC_BUFC_METADATA, 403 * residing in the arc_mru state, and are eligible for eviction 404 * (e.g. have no outstanding holds on the buffer). 405 */ 406 kstat_named_t arcstat_mru_evictable_metadata; 407 /* 408 * Total number of bytes that *would have been* consumed by ARC 409 * buffers in the arc_mru_ghost state. The key thing to note 410 * here, is the fact that this size doesn't actually indicate 411 * RAM consumption. The ghost lists only consist of headers and 412 * don't actually have ARC buffers linked off of these headers. 413 * Thus, *if* the headers had associated ARC buffers, these 414 * buffers *would have* consumed this number of bytes. 415 */ 416 kstat_named_t arcstat_mru_ghost_size; 417 /* 418 * Number of bytes that *would have been* consumed by ARC 419 * buffers that are eligible for eviction, of type 420 * ARC_BUFC_DATA, and linked off the arc_mru_ghost state. 421 */ 422 kstat_named_t arcstat_mru_ghost_evictable_data; 423 /* 424 * Number of bytes that *would have been* consumed by ARC 425 * buffers that are eligible for eviction, of type 426 * ARC_BUFC_METADATA, and linked off the arc_mru_ghost state. 427 */ 428 kstat_named_t arcstat_mru_ghost_evictable_metadata; 429 /* 430 * Total number of bytes consumed by ARC buffers residing in the 431 * arc_mfu state. This includes *all* buffers in the arc_mfu 432 * state; e.g. data, metadata, evictable, and unevictable buffers 433 * are all included in this value. 434 */ 435 kstat_named_t arcstat_mfu_size; 436 /* 437 * Number of bytes consumed by ARC buffers that are eligible for 438 * eviction, of type ARC_BUFC_DATA, and reside in the arc_mfu 439 * state. 440 */ 441 kstat_named_t arcstat_mfu_evictable_data; 442 /* 443 * Number of bytes consumed by ARC buffers that are eligible for 444 * eviction, of type ARC_BUFC_METADATA, and reside in the 445 * arc_mfu state. 446 */ 447 kstat_named_t arcstat_mfu_evictable_metadata; 448 /* 449 * Total number of bytes that *would have been* consumed by ARC 450 * buffers in the arc_mfu_ghost state. See the comment above 451 * arcstat_mru_ghost_size for more details. 452 */ 453 kstat_named_t arcstat_mfu_ghost_size; 454 /* 455 * Number of bytes that *would have been* consumed by ARC 456 * buffers that are eligible for eviction, of type 457 * ARC_BUFC_DATA, and linked off the arc_mfu_ghost state. 458 */ 459 kstat_named_t arcstat_mfu_ghost_evictable_data; 460 /* 461 * Number of bytes that *would have been* consumed by ARC 462 * buffers that are eligible for eviction, of type 463 * ARC_BUFC_METADATA, and linked off the arc_mru_ghost state. 464 */ 465 kstat_named_t arcstat_mfu_ghost_evictable_metadata; 466 kstat_named_t arcstat_l2_hits; 467 kstat_named_t arcstat_l2_misses; 468 kstat_named_t arcstat_l2_feeds; 469 kstat_named_t arcstat_l2_rw_clash; 470 kstat_named_t arcstat_l2_read_bytes; 471 kstat_named_t arcstat_l2_write_bytes; 472 kstat_named_t arcstat_l2_writes_sent; 473 kstat_named_t arcstat_l2_writes_done; 474 kstat_named_t arcstat_l2_writes_error; 475 kstat_named_t arcstat_l2_writes_lock_retry; 476 kstat_named_t arcstat_l2_evict_lock_retry; 477 kstat_named_t arcstat_l2_evict_reading; 478 kstat_named_t arcstat_l2_evict_l1cached; 479 kstat_named_t arcstat_l2_free_on_write; 480 kstat_named_t arcstat_l2_cdata_free_on_write; 481 kstat_named_t arcstat_l2_abort_lowmem; 482 kstat_named_t arcstat_l2_cksum_bad; 483 kstat_named_t arcstat_l2_io_error; 484 kstat_named_t arcstat_l2_size; 485 kstat_named_t arcstat_l2_asize; 486 kstat_named_t arcstat_l2_hdr_size; 487 kstat_named_t arcstat_l2_compress_successes; 488 kstat_named_t arcstat_l2_compress_zeros; 489 kstat_named_t arcstat_l2_compress_failures; 490 kstat_named_t arcstat_memory_throttle_count; 491 kstat_named_t arcstat_duplicate_buffers; 492 kstat_named_t arcstat_duplicate_buffers_size; 493 kstat_named_t arcstat_duplicate_reads; 494 kstat_named_t arcstat_meta_used; 495 kstat_named_t arcstat_meta_limit; 496 kstat_named_t arcstat_meta_max; 497 kstat_named_t arcstat_meta_min; 498 } arc_stats_t; 499 500 static arc_stats_t arc_stats = { 501 { "hits", KSTAT_DATA_UINT64 }, 502 { "misses", KSTAT_DATA_UINT64 }, 503 { "demand_data_hits", KSTAT_DATA_UINT64 }, 504 { "demand_data_misses", KSTAT_DATA_UINT64 }, 505 { "demand_metadata_hits", KSTAT_DATA_UINT64 }, 506 { "demand_metadata_misses", KSTAT_DATA_UINT64 }, 507 { "prefetch_data_hits", KSTAT_DATA_UINT64 }, 508 { "prefetch_data_misses", KSTAT_DATA_UINT64 }, 509 { "prefetch_metadata_hits", KSTAT_DATA_UINT64 }, 510 { "prefetch_metadata_misses", KSTAT_DATA_UINT64 }, 511 { "mru_hits", KSTAT_DATA_UINT64 }, 512 { "mru_ghost_hits", KSTAT_DATA_UINT64 }, 513 { "mfu_hits", KSTAT_DATA_UINT64 }, 514 { "mfu_ghost_hits", KSTAT_DATA_UINT64 }, 515 { "deleted", KSTAT_DATA_UINT64 }, 516 { "mutex_miss", KSTAT_DATA_UINT64 }, 517 { "evict_skip", KSTAT_DATA_UINT64 }, 518 { "evict_not_enough", KSTAT_DATA_UINT64 }, 519 { "evict_l2_cached", KSTAT_DATA_UINT64 }, 520 { "evict_l2_eligible", KSTAT_DATA_UINT64 }, 521 { "evict_l2_ineligible", KSTAT_DATA_UINT64 }, 522 { "evict_l2_skip", KSTAT_DATA_UINT64 }, 523 { "hash_elements", KSTAT_DATA_UINT64 }, 524 { "hash_elements_max", KSTAT_DATA_UINT64 }, 525 { "hash_collisions", KSTAT_DATA_UINT64 }, 526 { "hash_chains", KSTAT_DATA_UINT64 }, 527 { "hash_chain_max", KSTAT_DATA_UINT64 }, 528 { "p", KSTAT_DATA_UINT64 }, 529 { "c", KSTAT_DATA_UINT64 }, 530 { "c_min", KSTAT_DATA_UINT64 }, 531 { "c_max", KSTAT_DATA_UINT64 }, 532 { "size", KSTAT_DATA_UINT64 }, 533 { "hdr_size", KSTAT_DATA_UINT64 }, 534 { "data_size", KSTAT_DATA_UINT64 }, 535 { "metadata_size", KSTAT_DATA_UINT64 }, 536 { "other_size", KSTAT_DATA_UINT64 }, 537 { "anon_size", KSTAT_DATA_UINT64 }, 538 { "anon_evictable_data", KSTAT_DATA_UINT64 }, 539 { "anon_evictable_metadata", KSTAT_DATA_UINT64 }, 540 { "mru_size", KSTAT_DATA_UINT64 }, 541 { "mru_evictable_data", KSTAT_DATA_UINT64 }, 542 { "mru_evictable_metadata", KSTAT_DATA_UINT64 }, 543 { "mru_ghost_size", KSTAT_DATA_UINT64 }, 544 { "mru_ghost_evictable_data", KSTAT_DATA_UINT64 }, 545 { "mru_ghost_evictable_metadata", KSTAT_DATA_UINT64 }, 546 { "mfu_size", KSTAT_DATA_UINT64 }, 547 { "mfu_evictable_data", KSTAT_DATA_UINT64 }, 548 { "mfu_evictable_metadata", KSTAT_DATA_UINT64 }, 549 { "mfu_ghost_size", KSTAT_DATA_UINT64 }, 550 { "mfu_ghost_evictable_data", KSTAT_DATA_UINT64 }, 551 { "mfu_ghost_evictable_metadata", KSTAT_DATA_UINT64 }, 552 { "l2_hits", KSTAT_DATA_UINT64 }, 553 { "l2_misses", KSTAT_DATA_UINT64 }, 554 { "l2_feeds", KSTAT_DATA_UINT64 }, 555 { "l2_rw_clash", KSTAT_DATA_UINT64 }, 556 { "l2_read_bytes", KSTAT_DATA_UINT64 }, 557 { "l2_write_bytes", KSTAT_DATA_UINT64 }, 558 { "l2_writes_sent", KSTAT_DATA_UINT64 }, 559 { "l2_writes_done", KSTAT_DATA_UINT64 }, 560 { "l2_writes_error", KSTAT_DATA_UINT64 }, 561 { "l2_writes_lock_retry", KSTAT_DATA_UINT64 }, 562 { "l2_evict_lock_retry", KSTAT_DATA_UINT64 }, 563 { "l2_evict_reading", KSTAT_DATA_UINT64 }, 564 { "l2_evict_l1cached", KSTAT_DATA_UINT64 }, 565 { "l2_free_on_write", KSTAT_DATA_UINT64 }, 566 { "l2_cdata_free_on_write", KSTAT_DATA_UINT64 }, 567 { "l2_abort_lowmem", KSTAT_DATA_UINT64 }, 568 { "l2_cksum_bad", KSTAT_DATA_UINT64 }, 569 { "l2_io_error", KSTAT_DATA_UINT64 }, 570 { "l2_size", KSTAT_DATA_UINT64 }, 571 { "l2_asize", KSTAT_DATA_UINT64 }, 572 { "l2_hdr_size", KSTAT_DATA_UINT64 }, 573 { "l2_compress_successes", KSTAT_DATA_UINT64 }, 574 { "l2_compress_zeros", KSTAT_DATA_UINT64 }, 575 { "l2_compress_failures", KSTAT_DATA_UINT64 }, 576 { "memory_throttle_count", KSTAT_DATA_UINT64 }, 577 { "duplicate_buffers", KSTAT_DATA_UINT64 }, 578 { "duplicate_buffers_size", KSTAT_DATA_UINT64 }, 579 { "duplicate_reads", KSTAT_DATA_UINT64 }, 580 { "arc_meta_used", KSTAT_DATA_UINT64 }, 581 { "arc_meta_limit", KSTAT_DATA_UINT64 }, 582 { "arc_meta_max", KSTAT_DATA_UINT64 }, 583 { "arc_meta_min", KSTAT_DATA_UINT64 } 584 }; 585 586 #define ARCSTAT(stat) (arc_stats.stat.value.ui64) 587 588 #define ARCSTAT_INCR(stat, val) \ 589 atomic_add_64(&arc_stats.stat.value.ui64, (val)) 590 591 #define ARCSTAT_BUMP(stat) ARCSTAT_INCR(stat, 1) 592 #define ARCSTAT_BUMPDOWN(stat) ARCSTAT_INCR(stat, -1) 593 594 #define ARCSTAT_MAX(stat, val) { \ 595 uint64_t m; \ 596 while ((val) > (m = arc_stats.stat.value.ui64) && \ 597 (m != atomic_cas_64(&arc_stats.stat.value.ui64, m, (val)))) \ 598 continue; \ 599 } 600 601 #define ARCSTAT_MAXSTAT(stat) \ 602 ARCSTAT_MAX(stat##_max, arc_stats.stat.value.ui64) 603 604 /* 605 * We define a macro to allow ARC hits/misses to be easily broken down by 606 * two separate conditions, giving a total of four different subtypes for 607 * each of hits and misses (so eight statistics total). 608 */ 609 #define ARCSTAT_CONDSTAT(cond1, stat1, notstat1, cond2, stat2, notstat2, stat) \ 610 if (cond1) { \ 611 if (cond2) { \ 612 ARCSTAT_BUMP(arcstat_##stat1##_##stat2##_##stat); \ 613 } else { \ 614 ARCSTAT_BUMP(arcstat_##stat1##_##notstat2##_##stat); \ 615 } \ 616 } else { \ 617 if (cond2) { \ 618 ARCSTAT_BUMP(arcstat_##notstat1##_##stat2##_##stat); \ 619 } else { \ 620 ARCSTAT_BUMP(arcstat_##notstat1##_##notstat2##_##stat);\ 621 } \ 622 } 623 624 kstat_t *arc_ksp; 625 static arc_state_t *arc_anon; 626 static arc_state_t *arc_mru; 627 static arc_state_t *arc_mru_ghost; 628 static arc_state_t *arc_mfu; 629 static arc_state_t *arc_mfu_ghost; 630 static arc_state_t *arc_l2c_only; 631 632 /* 633 * There are several ARC variables that are critical to export as kstats -- 634 * but we don't want to have to grovel around in the kstat whenever we wish to 635 * manipulate them. For these variables, we therefore define them to be in 636 * terms of the statistic variable. This assures that we are not introducing 637 * the possibility of inconsistency by having shadow copies of the variables, 638 * while still allowing the code to be readable. 639 */ 640 #define arc_size ARCSTAT(arcstat_size) /* actual total arc size */ 641 #define arc_p ARCSTAT(arcstat_p) /* target size of MRU */ 642 #define arc_c ARCSTAT(arcstat_c) /* target size of cache */ 643 #define arc_c_min ARCSTAT(arcstat_c_min) /* min target cache size */ 644 #define arc_c_max ARCSTAT(arcstat_c_max) /* max target cache size */ 645 #define arc_meta_limit ARCSTAT(arcstat_meta_limit) /* max size for metadata */ 646 #define arc_meta_min ARCSTAT(arcstat_meta_min) /* min size for metadata */ 647 #define arc_meta_used ARCSTAT(arcstat_meta_used) /* size of metadata */ 648 #define arc_meta_max ARCSTAT(arcstat_meta_max) /* max size of metadata */ 649 650 #define L2ARC_IS_VALID_COMPRESS(_c_) \ 651 ((_c_) == ZIO_COMPRESS_LZ4 || (_c_) == ZIO_COMPRESS_EMPTY) 652 653 static int arc_no_grow; /* Don't try to grow cache size */ 654 static uint64_t arc_tempreserve; 655 static uint64_t arc_loaned_bytes; 656 657 typedef struct arc_callback arc_callback_t; 658 659 struct arc_callback { 660 void *acb_private; 661 arc_done_func_t *acb_done; 662 arc_buf_t *acb_buf; 663 zio_t *acb_zio_dummy; 664 arc_callback_t *acb_next; 665 }; 666 667 typedef struct arc_write_callback arc_write_callback_t; 668 669 struct arc_write_callback { 670 void *awcb_private; 671 arc_done_func_t *awcb_ready; 672 arc_done_func_t *awcb_physdone; 673 arc_done_func_t *awcb_done; 674 arc_buf_t *awcb_buf; 675 }; 676 677 /* 678 * ARC buffers are separated into multiple structs as a memory saving measure: 679 * - Common fields struct, always defined, and embedded within it: 680 * - L2-only fields, always allocated but undefined when not in L2ARC 681 * - L1-only fields, only allocated when in L1ARC 682 * 683 * Buffer in L1 Buffer only in L2 684 * +------------------------+ +------------------------+ 685 * | arc_buf_hdr_t | | arc_buf_hdr_t | 686 * | | | | 687 * | | | | 688 * | | | | 689 * +------------------------+ +------------------------+ 690 * | l2arc_buf_hdr_t | | l2arc_buf_hdr_t | 691 * | (undefined if L1-only) | | | 692 * +------------------------+ +------------------------+ 693 * | l1arc_buf_hdr_t | 694 * | | 695 * | | 696 * | | 697 * | | 698 * +------------------------+ 699 * 700 * Because it's possible for the L2ARC to become extremely large, we can wind 701 * up eating a lot of memory in L2ARC buffer headers, so the size of a header 702 * is minimized by only allocating the fields necessary for an L1-cached buffer 703 * when a header is actually in the L1 cache. The sub-headers (l1arc_buf_hdr and 704 * l2arc_buf_hdr) are embedded rather than allocated separately to save a couple 705 * words in pointers. arc_hdr_realloc() is used to switch a header between 706 * these two allocation states. 707 */ 708 typedef struct l1arc_buf_hdr { 709 kmutex_t b_freeze_lock; 710 #ifdef ZFS_DEBUG 711 /* 712 * used for debugging wtih kmem_flags - by allocating and freeing 713 * b_thawed when the buffer is thawed, we get a record of the stack 714 * trace that thawed it. 715 */ 716 void *b_thawed; 717 #endif 718 719 arc_buf_t *b_buf; 720 uint32_t b_datacnt; 721 /* for waiting on writes to complete */ 722 kcondvar_t b_cv; 723 724 /* protected by arc state mutex */ 725 arc_state_t *b_state; 726 multilist_node_t b_arc_node; 727 728 /* updated atomically */ 729 clock_t b_arc_access; 730 731 /* self protecting */ 732 refcount_t b_refcnt; 733 734 arc_callback_t *b_acb; 735 /* temporary buffer holder for in-flight compressed data */ 736 void *b_tmp_cdata; 737 } l1arc_buf_hdr_t; 738 739 typedef struct l2arc_dev l2arc_dev_t; 740 741 typedef struct l2arc_buf_hdr { 742 /* protected by arc_buf_hdr mutex */ 743 l2arc_dev_t *b_dev; /* L2ARC device */ 744 uint64_t b_daddr; /* disk address, offset byte */ 745 /* real alloc'd buffer size depending on b_compress applied */ 746 int32_t b_asize; 747 748 list_node_t b_l2node; 749 } l2arc_buf_hdr_t; 750 751 struct arc_buf_hdr { 752 /* protected by hash lock */ 753 dva_t b_dva; 754 uint64_t b_birth; 755 /* 756 * Even though this checksum is only set/verified when a buffer is in 757 * the L1 cache, it needs to be in the set of common fields because it 758 * must be preserved from the time before a buffer is written out to 759 * L2ARC until after it is read back in. 760 */ 761 zio_cksum_t *b_freeze_cksum; 762 763 arc_buf_hdr_t *b_hash_next; 764 arc_flags_t b_flags; 765 766 /* immutable */ 767 int32_t b_size; 768 uint64_t b_spa; 769 770 /* L2ARC fields. Undefined when not in L2ARC. */ 771 l2arc_buf_hdr_t b_l2hdr; 772 /* L1ARC fields. Undefined when in l2arc_only state */ 773 l1arc_buf_hdr_t b_l1hdr; 774 }; 775 776 static arc_buf_t *arc_eviction_list; 777 static arc_buf_hdr_t arc_eviction_hdr; 778 779 #define GHOST_STATE(state) \ 780 ((state) == arc_mru_ghost || (state) == arc_mfu_ghost || \ 781 (state) == arc_l2c_only) 782 783 #define HDR_IN_HASH_TABLE(hdr) ((hdr)->b_flags & ARC_FLAG_IN_HASH_TABLE) 784 #define HDR_IO_IN_PROGRESS(hdr) ((hdr)->b_flags & ARC_FLAG_IO_IN_PROGRESS) 785 #define HDR_IO_ERROR(hdr) ((hdr)->b_flags & ARC_FLAG_IO_ERROR) 786 #define HDR_PREFETCH(hdr) ((hdr)->b_flags & ARC_FLAG_PREFETCH) 787 #define HDR_FREED_IN_READ(hdr) ((hdr)->b_flags & ARC_FLAG_FREED_IN_READ) 788 #define HDR_BUF_AVAILABLE(hdr) ((hdr)->b_flags & ARC_FLAG_BUF_AVAILABLE) 789 790 #define HDR_L2CACHE(hdr) ((hdr)->b_flags & ARC_FLAG_L2CACHE) 791 #define HDR_L2COMPRESS(hdr) ((hdr)->b_flags & ARC_FLAG_L2COMPRESS) 792 #define HDR_L2_READING(hdr) \ 793 (((hdr)->b_flags & ARC_FLAG_IO_IN_PROGRESS) && \ 794 ((hdr)->b_flags & ARC_FLAG_HAS_L2HDR)) 795 #define HDR_L2_WRITING(hdr) ((hdr)->b_flags & ARC_FLAG_L2_WRITING) 796 #define HDR_L2_EVICTED(hdr) ((hdr)->b_flags & ARC_FLAG_L2_EVICTED) 797 #define HDR_L2_WRITE_HEAD(hdr) ((hdr)->b_flags & ARC_FLAG_L2_WRITE_HEAD) 798 799 #define HDR_ISTYPE_METADATA(hdr) \ 800 ((hdr)->b_flags & ARC_FLAG_BUFC_METADATA) 801 #define HDR_ISTYPE_DATA(hdr) (!HDR_ISTYPE_METADATA(hdr)) 802 803 #define HDR_HAS_L1HDR(hdr) ((hdr)->b_flags & ARC_FLAG_HAS_L1HDR) 804 #define HDR_HAS_L2HDR(hdr) ((hdr)->b_flags & ARC_FLAG_HAS_L2HDR) 805 806 /* For storing compression mode in b_flags */ 807 #define HDR_COMPRESS_OFFSET 24 808 #define HDR_COMPRESS_NBITS 7 809 810 #define HDR_GET_COMPRESS(hdr) ((enum zio_compress)BF32_GET(hdr->b_flags, \ 811 HDR_COMPRESS_OFFSET, HDR_COMPRESS_NBITS)) 812 #define HDR_SET_COMPRESS(hdr, cmp) BF32_SET(hdr->b_flags, \ 813 HDR_COMPRESS_OFFSET, HDR_COMPRESS_NBITS, (cmp)) 814 815 /* 816 * Other sizes 817 */ 818 819 #define HDR_FULL_SIZE ((int64_t)sizeof (arc_buf_hdr_t)) 820 #define HDR_L2ONLY_SIZE ((int64_t)offsetof(arc_buf_hdr_t, b_l1hdr)) 821 822 /* 823 * Hash table routines 824 */ 825 826 #define HT_LOCK_PAD 64 827 828 struct ht_lock { 829 kmutex_t ht_lock; 830 #ifdef _KERNEL 831 unsigned char pad[(HT_LOCK_PAD - sizeof (kmutex_t))]; 832 #endif 833 }; 834 835 #define BUF_LOCKS 256 836 typedef struct buf_hash_table { 837 uint64_t ht_mask; 838 arc_buf_hdr_t **ht_table; 839 struct ht_lock ht_locks[BUF_LOCKS]; 840 } buf_hash_table_t; 841 842 static buf_hash_table_t buf_hash_table; 843 844 #define BUF_HASH_INDEX(spa, dva, birth) \ 845 (buf_hash(spa, dva, birth) & buf_hash_table.ht_mask) 846 #define BUF_HASH_LOCK_NTRY(idx) (buf_hash_table.ht_locks[idx & (BUF_LOCKS-1)]) 847 #define BUF_HASH_LOCK(idx) (&(BUF_HASH_LOCK_NTRY(idx).ht_lock)) 848 #define HDR_LOCK(hdr) \ 849 (BUF_HASH_LOCK(BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth))) 850 851 uint64_t zfs_crc64_table[256]; 852 853 /* 854 * Level 2 ARC 855 */ 856 857 #define L2ARC_WRITE_SIZE (8 * 1024 * 1024) /* initial write max */ 858 #define L2ARC_HEADROOM 2 /* num of writes */ 859 /* 860 * If we discover during ARC scan any buffers to be compressed, we boost 861 * our headroom for the next scanning cycle by this percentage multiple. 862 */ 863 #define L2ARC_HEADROOM_BOOST 200 864 #define L2ARC_FEED_SECS 1 /* caching interval secs */ 865 #define L2ARC_FEED_MIN_MS 200 /* min caching interval ms */ 866 867 /* 868 * Used to distinguish headers that are being process by 869 * l2arc_write_buffers(), but have yet to be assigned to a l2arc disk 870 * address. This can happen when the header is added to the l2arc's list 871 * of buffers to write in the first stage of l2arc_write_buffers(), but 872 * has not yet been written out which happens in the second stage of 873 * l2arc_write_buffers(). 874 */ 875 #define L2ARC_ADDR_UNSET ((uint64_t)(-1)) 876 877 #define l2arc_writes_sent ARCSTAT(arcstat_l2_writes_sent) 878 #define l2arc_writes_done ARCSTAT(arcstat_l2_writes_done) 879 880 /* L2ARC Performance Tunables */ 881 uint64_t l2arc_write_max = L2ARC_WRITE_SIZE; /* default max write size */ 882 uint64_t l2arc_write_boost = L2ARC_WRITE_SIZE; /* extra write during warmup */ 883 uint64_t l2arc_headroom = L2ARC_HEADROOM; /* number of dev writes */ 884 uint64_t l2arc_headroom_boost = L2ARC_HEADROOM_BOOST; 885 uint64_t l2arc_feed_secs = L2ARC_FEED_SECS; /* interval seconds */ 886 uint64_t l2arc_feed_min_ms = L2ARC_FEED_MIN_MS; /* min interval milliseconds */ 887 boolean_t l2arc_noprefetch = B_TRUE; /* don't cache prefetch bufs */ 888 boolean_t l2arc_feed_again = B_TRUE; /* turbo warmup */ 889 boolean_t l2arc_norw = B_TRUE; /* no reads during writes */ 890 891 /* 892 * L2ARC Internals 893 */ 894 struct l2arc_dev { 895 vdev_t *l2ad_vdev; /* vdev */ 896 spa_t *l2ad_spa; /* spa */ 897 uint64_t l2ad_hand; /* next write location */ 898 uint64_t l2ad_start; /* first addr on device */ 899 uint64_t l2ad_end; /* last addr on device */ 900 boolean_t l2ad_first; /* first sweep through */ 901 boolean_t l2ad_writing; /* currently writing */ 902 kmutex_t l2ad_mtx; /* lock for buffer list */ 903 list_t l2ad_buflist; /* buffer list */ 904 list_node_t l2ad_node; /* device list node */ 905 refcount_t l2ad_alloc; /* allocated bytes */ 906 }; 907 908 static list_t L2ARC_dev_list; /* device list */ 909 static list_t *l2arc_dev_list; /* device list pointer */ 910 static kmutex_t l2arc_dev_mtx; /* device list mutex */ 911 static l2arc_dev_t *l2arc_dev_last; /* last device used */ 912 static list_t L2ARC_free_on_write; /* free after write buf list */ 913 static list_t *l2arc_free_on_write; /* free after write list ptr */ 914 static kmutex_t l2arc_free_on_write_mtx; /* mutex for list */ 915 static uint64_t l2arc_ndev; /* number of devices */ 916 917 typedef struct l2arc_read_callback { 918 arc_buf_t *l2rcb_buf; /* read buffer */ 919 spa_t *l2rcb_spa; /* spa */ 920 blkptr_t l2rcb_bp; /* original blkptr */ 921 zbookmark_phys_t l2rcb_zb; /* original bookmark */ 922 int l2rcb_flags; /* original flags */ 923 enum zio_compress l2rcb_compress; /* applied compress */ 924 } l2arc_read_callback_t; 925 926 typedef struct l2arc_write_callback { 927 l2arc_dev_t *l2wcb_dev; /* device info */ 928 arc_buf_hdr_t *l2wcb_head; /* head of write buflist */ 929 } l2arc_write_callback_t; 930 931 typedef struct l2arc_data_free { 932 /* protected by l2arc_free_on_write_mtx */ 933 void *l2df_data; 934 size_t l2df_size; 935 void (*l2df_func)(void *, size_t); 936 list_node_t l2df_list_node; 937 } l2arc_data_free_t; 938 939 static kmutex_t l2arc_feed_thr_lock; 940 static kcondvar_t l2arc_feed_thr_cv; 941 static uint8_t l2arc_thread_exit; 942 943 static void arc_get_data_buf(arc_buf_t *); 944 static void arc_access(arc_buf_hdr_t *, kmutex_t *); 945 static boolean_t arc_is_overflowing(); 946 static void arc_buf_watch(arc_buf_t *); 947 948 static arc_buf_contents_t arc_buf_type(arc_buf_hdr_t *); 949 static uint32_t arc_bufc_to_flags(arc_buf_contents_t); 950 951 static boolean_t l2arc_write_eligible(uint64_t, arc_buf_hdr_t *); 952 static void l2arc_read_done(zio_t *); 953 954 static boolean_t l2arc_compress_buf(arc_buf_hdr_t *); 955 static void l2arc_decompress_zio(zio_t *, arc_buf_hdr_t *, enum zio_compress); 956 static void l2arc_release_cdata_buf(arc_buf_hdr_t *); 957 958 static uint64_t 959 buf_hash(uint64_t spa, const dva_t *dva, uint64_t birth) 960 { 961 uint8_t *vdva = (uint8_t *)dva; 962 uint64_t crc = -1ULL; 963 int i; 964 965 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); 966 967 for (i = 0; i < sizeof (dva_t); i++) 968 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ vdva[i]) & 0xFF]; 969 970 crc ^= (spa>>8) ^ birth; 971 972 return (crc); 973 } 974 975 #define BUF_EMPTY(buf) \ 976 ((buf)->b_dva.dva_word[0] == 0 && \ 977 (buf)->b_dva.dva_word[1] == 0) 978 979 #define BUF_EQUAL(spa, dva, birth, buf) \ 980 ((buf)->b_dva.dva_word[0] == (dva)->dva_word[0]) && \ 981 ((buf)->b_dva.dva_word[1] == (dva)->dva_word[1]) && \ 982 ((buf)->b_birth == birth) && ((buf)->b_spa == spa) 983 984 static void 985 buf_discard_identity(arc_buf_hdr_t *hdr) 986 { 987 hdr->b_dva.dva_word[0] = 0; 988 hdr->b_dva.dva_word[1] = 0; 989 hdr->b_birth = 0; 990 } 991 992 static arc_buf_hdr_t * 993 buf_hash_find(uint64_t spa, const blkptr_t *bp, kmutex_t **lockp) 994 { 995 const dva_t *dva = BP_IDENTITY(bp); 996 uint64_t birth = BP_PHYSICAL_BIRTH(bp); 997 uint64_t idx = BUF_HASH_INDEX(spa, dva, birth); 998 kmutex_t *hash_lock = BUF_HASH_LOCK(idx); 999 arc_buf_hdr_t *hdr; 1000 1001 mutex_enter(hash_lock); 1002 for (hdr = buf_hash_table.ht_table[idx]; hdr != NULL; 1003 hdr = hdr->b_hash_next) { 1004 if (BUF_EQUAL(spa, dva, birth, hdr)) { 1005 *lockp = hash_lock; 1006 return (hdr); 1007 } 1008 } 1009 mutex_exit(hash_lock); 1010 *lockp = NULL; 1011 return (NULL); 1012 } 1013 1014 /* 1015 * Insert an entry into the hash table. If there is already an element 1016 * equal to elem in the hash table, then the already existing element 1017 * will be returned and the new element will not be inserted. 1018 * Otherwise returns NULL. 1019 * If lockp == NULL, the caller is assumed to already hold the hash lock. 1020 */ 1021 static arc_buf_hdr_t * 1022 buf_hash_insert(arc_buf_hdr_t *hdr, kmutex_t **lockp) 1023 { 1024 uint64_t idx = BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth); 1025 kmutex_t *hash_lock = BUF_HASH_LOCK(idx); 1026 arc_buf_hdr_t *fhdr; 1027 uint32_t i; 1028 1029 ASSERT(!DVA_IS_EMPTY(&hdr->b_dva)); 1030 ASSERT(hdr->b_birth != 0); 1031 ASSERT(!HDR_IN_HASH_TABLE(hdr)); 1032 1033 if (lockp != NULL) { 1034 *lockp = hash_lock; 1035 mutex_enter(hash_lock); 1036 } else { 1037 ASSERT(MUTEX_HELD(hash_lock)); 1038 } 1039 1040 for (fhdr = buf_hash_table.ht_table[idx], i = 0; fhdr != NULL; 1041 fhdr = fhdr->b_hash_next, i++) { 1042 if (BUF_EQUAL(hdr->b_spa, &hdr->b_dva, hdr->b_birth, fhdr)) 1043 return (fhdr); 1044 } 1045 1046 hdr->b_hash_next = buf_hash_table.ht_table[idx]; 1047 buf_hash_table.ht_table[idx] = hdr; 1048 hdr->b_flags |= ARC_FLAG_IN_HASH_TABLE; 1049 1050 /* collect some hash table performance data */ 1051 if (i > 0) { 1052 ARCSTAT_BUMP(arcstat_hash_collisions); 1053 if (i == 1) 1054 ARCSTAT_BUMP(arcstat_hash_chains); 1055 1056 ARCSTAT_MAX(arcstat_hash_chain_max, i); 1057 } 1058 1059 ARCSTAT_BUMP(arcstat_hash_elements); 1060 ARCSTAT_MAXSTAT(arcstat_hash_elements); 1061 1062 return (NULL); 1063 } 1064 1065 static void 1066 buf_hash_remove(arc_buf_hdr_t *hdr) 1067 { 1068 arc_buf_hdr_t *fhdr, **hdrp; 1069 uint64_t idx = BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth); 1070 1071 ASSERT(MUTEX_HELD(BUF_HASH_LOCK(idx))); 1072 ASSERT(HDR_IN_HASH_TABLE(hdr)); 1073 1074 hdrp = &buf_hash_table.ht_table[idx]; 1075 while ((fhdr = *hdrp) != hdr) { 1076 ASSERT(fhdr != NULL); 1077 hdrp = &fhdr->b_hash_next; 1078 } 1079 *hdrp = hdr->b_hash_next; 1080 hdr->b_hash_next = NULL; 1081 hdr->b_flags &= ~ARC_FLAG_IN_HASH_TABLE; 1082 1083 /* collect some hash table performance data */ 1084 ARCSTAT_BUMPDOWN(arcstat_hash_elements); 1085 1086 if (buf_hash_table.ht_table[idx] && 1087 buf_hash_table.ht_table[idx]->b_hash_next == NULL) 1088 ARCSTAT_BUMPDOWN(arcstat_hash_chains); 1089 } 1090 1091 /* 1092 * Global data structures and functions for the buf kmem cache. 1093 */ 1094 static kmem_cache_t *hdr_full_cache; 1095 static kmem_cache_t *hdr_l2only_cache; 1096 static kmem_cache_t *buf_cache; 1097 1098 static void 1099 buf_fini(void) 1100 { 1101 int i; 1102 1103 kmem_free(buf_hash_table.ht_table, 1104 (buf_hash_table.ht_mask + 1) * sizeof (void *)); 1105 for (i = 0; i < BUF_LOCKS; i++) 1106 mutex_destroy(&buf_hash_table.ht_locks[i].ht_lock); 1107 kmem_cache_destroy(hdr_full_cache); 1108 kmem_cache_destroy(hdr_l2only_cache); 1109 kmem_cache_destroy(buf_cache); 1110 } 1111 1112 /* 1113 * Constructor callback - called when the cache is empty 1114 * and a new buf is requested. 1115 */ 1116 /* ARGSUSED */ 1117 static int 1118 hdr_full_cons(void *vbuf, void *unused, int kmflag) 1119 { 1120 arc_buf_hdr_t *hdr = vbuf; 1121 1122 bzero(hdr, HDR_FULL_SIZE); 1123 cv_init(&hdr->b_l1hdr.b_cv, NULL, CV_DEFAULT, NULL); 1124 refcount_create(&hdr->b_l1hdr.b_refcnt); 1125 mutex_init(&hdr->b_l1hdr.b_freeze_lock, NULL, MUTEX_DEFAULT, NULL); 1126 multilist_link_init(&hdr->b_l1hdr.b_arc_node); 1127 arc_space_consume(HDR_FULL_SIZE, ARC_SPACE_HDRS); 1128 1129 return (0); 1130 } 1131 1132 /* ARGSUSED */ 1133 static int 1134 hdr_l2only_cons(void *vbuf, void *unused, int kmflag) 1135 { 1136 arc_buf_hdr_t *hdr = vbuf; 1137 1138 bzero(hdr, HDR_L2ONLY_SIZE); 1139 arc_space_consume(HDR_L2ONLY_SIZE, ARC_SPACE_L2HDRS); 1140 1141 return (0); 1142 } 1143 1144 /* ARGSUSED */ 1145 static int 1146 buf_cons(void *vbuf, void *unused, int kmflag) 1147 { 1148 arc_buf_t *buf = vbuf; 1149 1150 bzero(buf, sizeof (arc_buf_t)); 1151 mutex_init(&buf->b_evict_lock, NULL, MUTEX_DEFAULT, NULL); 1152 arc_space_consume(sizeof (arc_buf_t), ARC_SPACE_HDRS); 1153 1154 return (0); 1155 } 1156 1157 /* 1158 * Destructor callback - called when a cached buf is 1159 * no longer required. 1160 */ 1161 /* ARGSUSED */ 1162 static void 1163 hdr_full_dest(void *vbuf, void *unused) 1164 { 1165 arc_buf_hdr_t *hdr = vbuf; 1166 1167 ASSERT(BUF_EMPTY(hdr)); 1168 cv_destroy(&hdr->b_l1hdr.b_cv); 1169 refcount_destroy(&hdr->b_l1hdr.b_refcnt); 1170 mutex_destroy(&hdr->b_l1hdr.b_freeze_lock); 1171 ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); 1172 arc_space_return(HDR_FULL_SIZE, ARC_SPACE_HDRS); 1173 } 1174 1175 /* ARGSUSED */ 1176 static void 1177 hdr_l2only_dest(void *vbuf, void *unused) 1178 { 1179 arc_buf_hdr_t *hdr = vbuf; 1180 1181 ASSERT(BUF_EMPTY(hdr)); 1182 arc_space_return(HDR_L2ONLY_SIZE, ARC_SPACE_L2HDRS); 1183 } 1184 1185 /* ARGSUSED */ 1186 static void 1187 buf_dest(void *vbuf, void *unused) 1188 { 1189 arc_buf_t *buf = vbuf; 1190 1191 mutex_destroy(&buf->b_evict_lock); 1192 arc_space_return(sizeof (arc_buf_t), ARC_SPACE_HDRS); 1193 } 1194 1195 /* 1196 * Reclaim callback -- invoked when memory is low. 1197 */ 1198 /* ARGSUSED */ 1199 static void 1200 hdr_recl(void *unused) 1201 { 1202 dprintf("hdr_recl called\n"); 1203 /* 1204 * umem calls the reclaim func when we destroy the buf cache, 1205 * which is after we do arc_fini(). 1206 */ 1207 if (!arc_dead) 1208 cv_signal(&arc_reclaim_thread_cv); 1209 } 1210 1211 static void 1212 buf_init(void) 1213 { 1214 uint64_t *ct; 1215 uint64_t hsize = 1ULL << 12; 1216 int i, j; 1217 1218 /* 1219 * The hash table is big enough to fill all of physical memory 1220 * with an average block size of zfs_arc_average_blocksize (default 8K). 1221 * By default, the table will take up 1222 * totalmem * sizeof(void*) / 8K (1MB per GB with 8-byte pointers). 1223 */ 1224 while (hsize * zfs_arc_average_blocksize < physmem * PAGESIZE) 1225 hsize <<= 1; 1226 retry: 1227 buf_hash_table.ht_mask = hsize - 1; 1228 buf_hash_table.ht_table = 1229 kmem_zalloc(hsize * sizeof (void*), KM_NOSLEEP); 1230 if (buf_hash_table.ht_table == NULL) { 1231 ASSERT(hsize > (1ULL << 8)); 1232 hsize >>= 1; 1233 goto retry; 1234 } 1235 1236 hdr_full_cache = kmem_cache_create("arc_buf_hdr_t_full", HDR_FULL_SIZE, 1237 0, hdr_full_cons, hdr_full_dest, hdr_recl, NULL, NULL, 0); 1238 hdr_l2only_cache = kmem_cache_create("arc_buf_hdr_t_l2only", 1239 HDR_L2ONLY_SIZE, 0, hdr_l2only_cons, hdr_l2only_dest, hdr_recl, 1240 NULL, NULL, 0); 1241 buf_cache = kmem_cache_create("arc_buf_t", sizeof (arc_buf_t), 1242 0, buf_cons, buf_dest, NULL, NULL, NULL, 0); 1243 1244 for (i = 0; i < 256; i++) 1245 for (ct = zfs_crc64_table + i, *ct = i, j = 8; j > 0; j--) 1246 *ct = (*ct >> 1) ^ (-(*ct & 1) & ZFS_CRC64_POLY); 1247 1248 for (i = 0; i < BUF_LOCKS; i++) { 1249 mutex_init(&buf_hash_table.ht_locks[i].ht_lock, 1250 NULL, MUTEX_DEFAULT, NULL); 1251 } 1252 } 1253 1254 /* 1255 * Transition between the two allocation states for the arc_buf_hdr struct. 1256 * The arc_buf_hdr struct can be allocated with (hdr_full_cache) or without 1257 * (hdr_l2only_cache) the fields necessary for the L1 cache - the smaller 1258 * version is used when a cache buffer is only in the L2ARC in order to reduce 1259 * memory usage. 1260 */ 1261 static arc_buf_hdr_t * 1262 arc_hdr_realloc(arc_buf_hdr_t *hdr, kmem_cache_t *old, kmem_cache_t *new) 1263 { 1264 ASSERT(HDR_HAS_L2HDR(hdr)); 1265 1266 arc_buf_hdr_t *nhdr; 1267 l2arc_dev_t *dev = hdr->b_l2hdr.b_dev; 1268 1269 ASSERT((old == hdr_full_cache && new == hdr_l2only_cache) || 1270 (old == hdr_l2only_cache && new == hdr_full_cache)); 1271 1272 nhdr = kmem_cache_alloc(new, KM_PUSHPAGE); 1273 1274 ASSERT(MUTEX_HELD(HDR_LOCK(hdr))); 1275 buf_hash_remove(hdr); 1276 1277 bcopy(hdr, nhdr, HDR_L2ONLY_SIZE); 1278 1279 if (new == hdr_full_cache) { 1280 nhdr->b_flags |= ARC_FLAG_HAS_L1HDR; 1281 /* 1282 * arc_access and arc_change_state need to be aware that a 1283 * header has just come out of L2ARC, so we set its state to 1284 * l2c_only even though it's about to change. 1285 */ 1286 nhdr->b_l1hdr.b_state = arc_l2c_only; 1287 1288 /* Verify previous threads set to NULL before freeing */ 1289 ASSERT3P(nhdr->b_l1hdr.b_tmp_cdata, ==, NULL); 1290 } else { 1291 ASSERT(hdr->b_l1hdr.b_buf == NULL); 1292 ASSERT0(hdr->b_l1hdr.b_datacnt); 1293 1294 /* 1295 * If we've reached here, We must have been called from 1296 * arc_evict_hdr(), as such we should have already been 1297 * removed from any ghost list we were previously on 1298 * (which protects us from racing with arc_evict_state), 1299 * thus no locking is needed during this check. 1300 */ 1301 ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); 1302 1303 /* 1304 * A buffer must not be moved into the arc_l2c_only 1305 * state if it's not finished being written out to the 1306 * l2arc device. Otherwise, the b_l1hdr.b_tmp_cdata field 1307 * might try to be accessed, even though it was removed. 1308 */ 1309 VERIFY(!HDR_L2_WRITING(hdr)); 1310 VERIFY3P(hdr->b_l1hdr.b_tmp_cdata, ==, NULL); 1311 1312 nhdr->b_flags &= ~ARC_FLAG_HAS_L1HDR; 1313 } 1314 /* 1315 * The header has been reallocated so we need to re-insert it into any 1316 * lists it was on. 1317 */ 1318 (void) buf_hash_insert(nhdr, NULL); 1319 1320 ASSERT(list_link_active(&hdr->b_l2hdr.b_l2node)); 1321 1322 mutex_enter(&dev->l2ad_mtx); 1323 1324 /* 1325 * We must place the realloc'ed header back into the list at 1326 * the same spot. Otherwise, if it's placed earlier in the list, 1327 * l2arc_write_buffers() could find it during the function's 1328 * write phase, and try to write it out to the l2arc. 1329 */ 1330 list_insert_after(&dev->l2ad_buflist, hdr, nhdr); 1331 list_remove(&dev->l2ad_buflist, hdr); 1332 1333 mutex_exit(&dev->l2ad_mtx); 1334 1335 /* 1336 * Since we're using the pointer address as the tag when 1337 * incrementing and decrementing the l2ad_alloc refcount, we 1338 * must remove the old pointer (that we're about to destroy) and 1339 * add the new pointer to the refcount. Otherwise we'd remove 1340 * the wrong pointer address when calling arc_hdr_destroy() later. 1341 */ 1342 1343 (void) refcount_remove_many(&dev->l2ad_alloc, 1344 hdr->b_l2hdr.b_asize, hdr); 1345 1346 (void) refcount_add_many(&dev->l2ad_alloc, 1347 nhdr->b_l2hdr.b_asize, nhdr); 1348 1349 buf_discard_identity(hdr); 1350 hdr->b_freeze_cksum = NULL; 1351 kmem_cache_free(old, hdr); 1352 1353 return (nhdr); 1354 } 1355 1356 1357 #define ARC_MINTIME (hz>>4) /* 62 ms */ 1358 1359 static void 1360 arc_cksum_verify(arc_buf_t *buf) 1361 { 1362 zio_cksum_t zc; 1363 1364 if (!(zfs_flags & ZFS_DEBUG_MODIFY)) 1365 return; 1366 1367 mutex_enter(&buf->b_hdr->b_l1hdr.b_freeze_lock); 1368 if (buf->b_hdr->b_freeze_cksum == NULL || HDR_IO_ERROR(buf->b_hdr)) { 1369 mutex_exit(&buf->b_hdr->b_l1hdr.b_freeze_lock); 1370 return; 1371 } 1372 fletcher_2_native(buf->b_data, buf->b_hdr->b_size, &zc); 1373 if (!ZIO_CHECKSUM_EQUAL(*buf->b_hdr->b_freeze_cksum, zc)) 1374 panic("buffer modified while frozen!"); 1375 mutex_exit(&buf->b_hdr->b_l1hdr.b_freeze_lock); 1376 } 1377 1378 static int 1379 arc_cksum_equal(arc_buf_t *buf) 1380 { 1381 zio_cksum_t zc; 1382 int equal; 1383 1384 mutex_enter(&buf->b_hdr->b_l1hdr.b_freeze_lock); 1385 fletcher_2_native(buf->b_data, buf->b_hdr->b_size, &zc); 1386 equal = ZIO_CHECKSUM_EQUAL(*buf->b_hdr->b_freeze_cksum, zc); 1387 mutex_exit(&buf->b_hdr->b_l1hdr.b_freeze_lock); 1388 1389 return (equal); 1390 } 1391 1392 static void 1393 arc_cksum_compute(arc_buf_t *buf, boolean_t force) 1394 { 1395 if (!force && !(zfs_flags & ZFS_DEBUG_MODIFY)) 1396 return; 1397 1398 mutex_enter(&buf->b_hdr->b_l1hdr.b_freeze_lock); 1399 if (buf->b_hdr->b_freeze_cksum != NULL) { 1400 mutex_exit(&buf->b_hdr->b_l1hdr.b_freeze_lock); 1401 return; 1402 } 1403 buf->b_hdr->b_freeze_cksum = kmem_alloc(sizeof (zio_cksum_t), KM_SLEEP); 1404 fletcher_2_native(buf->b_data, buf->b_hdr->b_size, 1405 buf->b_hdr->b_freeze_cksum); 1406 mutex_exit(&buf->b_hdr->b_l1hdr.b_freeze_lock); 1407 arc_buf_watch(buf); 1408 } 1409 1410 #ifndef _KERNEL 1411 typedef struct procctl { 1412 long cmd; 1413 prwatch_t prwatch; 1414 } procctl_t; 1415 #endif 1416 1417 /* ARGSUSED */ 1418 static void 1419 arc_buf_unwatch(arc_buf_t *buf) 1420 { 1421 #ifndef _KERNEL 1422 if (arc_watch) { 1423 int result; 1424 procctl_t ctl; 1425 ctl.cmd = PCWATCH; 1426 ctl.prwatch.pr_vaddr = (uintptr_t)buf->b_data; 1427 ctl.prwatch.pr_size = 0; 1428 ctl.prwatch.pr_wflags = 0; 1429 result = write(arc_procfd, &ctl, sizeof (ctl)); 1430 ASSERT3U(result, ==, sizeof (ctl)); 1431 } 1432 #endif 1433 } 1434 1435 /* ARGSUSED */ 1436 static void 1437 arc_buf_watch(arc_buf_t *buf) 1438 { 1439 #ifndef _KERNEL 1440 if (arc_watch) { 1441 int result; 1442 procctl_t ctl; 1443 ctl.cmd = PCWATCH; 1444 ctl.prwatch.pr_vaddr = (uintptr_t)buf->b_data; 1445 ctl.prwatch.pr_size = buf->b_hdr->b_size; 1446 ctl.prwatch.pr_wflags = WA_WRITE; 1447 result = write(arc_procfd, &ctl, sizeof (ctl)); 1448 ASSERT3U(result, ==, sizeof (ctl)); 1449 } 1450 #endif 1451 } 1452 1453 static arc_buf_contents_t 1454 arc_buf_type(arc_buf_hdr_t *hdr) 1455 { 1456 if (HDR_ISTYPE_METADATA(hdr)) { 1457 return (ARC_BUFC_METADATA); 1458 } else { 1459 return (ARC_BUFC_DATA); 1460 } 1461 } 1462 1463 static uint32_t 1464 arc_bufc_to_flags(arc_buf_contents_t type) 1465 { 1466 switch (type) { 1467 case ARC_BUFC_DATA: 1468 /* metadata field is 0 if buffer contains normal data */ 1469 return (0); 1470 case ARC_BUFC_METADATA: 1471 return (ARC_FLAG_BUFC_METADATA); 1472 default: 1473 break; 1474 } 1475 panic("undefined ARC buffer type!"); 1476 return ((uint32_t)-1); 1477 } 1478 1479 void 1480 arc_buf_thaw(arc_buf_t *buf) 1481 { 1482 if (zfs_flags & ZFS_DEBUG_MODIFY) { 1483 if (buf->b_hdr->b_l1hdr.b_state != arc_anon) 1484 panic("modifying non-anon buffer!"); 1485 if (HDR_IO_IN_PROGRESS(buf->b_hdr)) 1486 panic("modifying buffer while i/o in progress!"); 1487 arc_cksum_verify(buf); 1488 } 1489 1490 mutex_enter(&buf->b_hdr->b_l1hdr.b_freeze_lock); 1491 if (buf->b_hdr->b_freeze_cksum != NULL) { 1492 kmem_free(buf->b_hdr->b_freeze_cksum, sizeof (zio_cksum_t)); 1493 buf->b_hdr->b_freeze_cksum = NULL; 1494 } 1495 1496 #ifdef ZFS_DEBUG 1497 if (zfs_flags & ZFS_DEBUG_MODIFY) { 1498 if (buf->b_hdr->b_l1hdr.b_thawed != NULL) 1499 kmem_free(buf->b_hdr->b_l1hdr.b_thawed, 1); 1500 buf->b_hdr->b_l1hdr.b_thawed = kmem_alloc(1, KM_SLEEP); 1501 } 1502 #endif 1503 1504 mutex_exit(&buf->b_hdr->b_l1hdr.b_freeze_lock); 1505 1506 arc_buf_unwatch(buf); 1507 } 1508 1509 void 1510 arc_buf_freeze(arc_buf_t *buf) 1511 { 1512 kmutex_t *hash_lock; 1513 1514 if (!(zfs_flags & ZFS_DEBUG_MODIFY)) 1515 return; 1516 1517 hash_lock = HDR_LOCK(buf->b_hdr); 1518 mutex_enter(hash_lock); 1519 1520 ASSERT(buf->b_hdr->b_freeze_cksum != NULL || 1521 buf->b_hdr->b_l1hdr.b_state == arc_anon); 1522 arc_cksum_compute(buf, B_FALSE); 1523 mutex_exit(hash_lock); 1524 1525 } 1526 1527 static void 1528 add_reference(arc_buf_hdr_t *hdr, kmutex_t *hash_lock, void *tag) 1529 { 1530 ASSERT(HDR_HAS_L1HDR(hdr)); 1531 ASSERT(MUTEX_HELD(hash_lock)); 1532 arc_state_t *state = hdr->b_l1hdr.b_state; 1533 1534 if ((refcount_add(&hdr->b_l1hdr.b_refcnt, tag) == 1) && 1535 (state != arc_anon)) { 1536 /* We don't use the L2-only state list. */ 1537 if (state != arc_l2c_only) { 1538 arc_buf_contents_t type = arc_buf_type(hdr); 1539 uint64_t delta = hdr->b_size * hdr->b_l1hdr.b_datacnt; 1540 multilist_t *list = &state->arcs_list[type]; 1541 uint64_t *size = &state->arcs_lsize[type]; 1542 1543 multilist_remove(list, hdr); 1544 1545 if (GHOST_STATE(state)) { 1546 ASSERT0(hdr->b_l1hdr.b_datacnt); 1547 ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); 1548 delta = hdr->b_size; 1549 } 1550 ASSERT(delta > 0); 1551 ASSERT3U(*size, >=, delta); 1552 atomic_add_64(size, -delta); 1553 } 1554 /* remove the prefetch flag if we get a reference */ 1555 hdr->b_flags &= ~ARC_FLAG_PREFETCH; 1556 } 1557 } 1558 1559 static int 1560 remove_reference(arc_buf_hdr_t *hdr, kmutex_t *hash_lock, void *tag) 1561 { 1562 int cnt; 1563 arc_state_t *state = hdr->b_l1hdr.b_state; 1564 1565 ASSERT(HDR_HAS_L1HDR(hdr)); 1566 ASSERT(state == arc_anon || MUTEX_HELD(hash_lock)); 1567 ASSERT(!GHOST_STATE(state)); 1568 1569 /* 1570 * arc_l2c_only counts as a ghost state so we don't need to explicitly 1571 * check to prevent usage of the arc_l2c_only list. 1572 */ 1573 if (((cnt = refcount_remove(&hdr->b_l1hdr.b_refcnt, tag)) == 0) && 1574 (state != arc_anon)) { 1575 arc_buf_contents_t type = arc_buf_type(hdr); 1576 multilist_t *list = &state->arcs_list[type]; 1577 uint64_t *size = &state->arcs_lsize[type]; 1578 1579 multilist_insert(list, hdr); 1580 1581 ASSERT(hdr->b_l1hdr.b_datacnt > 0); 1582 atomic_add_64(size, hdr->b_size * 1583 hdr->b_l1hdr.b_datacnt); 1584 } 1585 return (cnt); 1586 } 1587 1588 /* 1589 * Move the supplied buffer to the indicated state. The hash lock 1590 * for the buffer must be held by the caller. 1591 */ 1592 static void 1593 arc_change_state(arc_state_t *new_state, arc_buf_hdr_t *hdr, 1594 kmutex_t *hash_lock) 1595 { 1596 arc_state_t *old_state; 1597 int64_t refcnt; 1598 uint32_t datacnt; 1599 uint64_t from_delta, to_delta; 1600 arc_buf_contents_t buftype = arc_buf_type(hdr); 1601 1602 /* 1603 * We almost always have an L1 hdr here, since we call arc_hdr_realloc() 1604 * in arc_read() when bringing a buffer out of the L2ARC. However, the 1605 * L1 hdr doesn't always exist when we change state to arc_anon before 1606 * destroying a header, in which case reallocating to add the L1 hdr is 1607 * pointless. 1608 */ 1609 if (HDR_HAS_L1HDR(hdr)) { 1610 old_state = hdr->b_l1hdr.b_state; 1611 refcnt = refcount_count(&hdr->b_l1hdr.b_refcnt); 1612 datacnt = hdr->b_l1hdr.b_datacnt; 1613 } else { 1614 old_state = arc_l2c_only; 1615 refcnt = 0; 1616 datacnt = 0; 1617 } 1618 1619 ASSERT(MUTEX_HELD(hash_lock)); 1620 ASSERT3P(new_state, !=, old_state); 1621 ASSERT(refcnt == 0 || datacnt > 0); 1622 ASSERT(!GHOST_STATE(new_state) || datacnt == 0); 1623 ASSERT(old_state != arc_anon || datacnt <= 1); 1624 1625 from_delta = to_delta = datacnt * hdr->b_size; 1626 1627 /* 1628 * If this buffer is evictable, transfer it from the 1629 * old state list to the new state list. 1630 */ 1631 if (refcnt == 0) { 1632 if (old_state != arc_anon && old_state != arc_l2c_only) { 1633 uint64_t *size = &old_state->arcs_lsize[buftype]; 1634 1635 ASSERT(HDR_HAS_L1HDR(hdr)); 1636 multilist_remove(&old_state->arcs_list[buftype], hdr); 1637 1638 /* 1639 * If prefetching out of the ghost cache, 1640 * we will have a non-zero datacnt. 1641 */ 1642 if (GHOST_STATE(old_state) && datacnt == 0) { 1643 /* ghost elements have a ghost size */ 1644 ASSERT(hdr->b_l1hdr.b_buf == NULL); 1645 from_delta = hdr->b_size; 1646 } 1647 ASSERT3U(*size, >=, from_delta); 1648 atomic_add_64(size, -from_delta); 1649 } 1650 if (new_state != arc_anon && new_state != arc_l2c_only) { 1651 uint64_t *size = &new_state->arcs_lsize[buftype]; 1652 1653 /* 1654 * An L1 header always exists here, since if we're 1655 * moving to some L1-cached state (i.e. not l2c_only or 1656 * anonymous), we realloc the header to add an L1hdr 1657 * beforehand. 1658 */ 1659 ASSERT(HDR_HAS_L1HDR(hdr)); 1660 multilist_insert(&new_state->arcs_list[buftype], hdr); 1661 1662 /* ghost elements have a ghost size */ 1663 if (GHOST_STATE(new_state)) { 1664 ASSERT0(datacnt); 1665 ASSERT(hdr->b_l1hdr.b_buf == NULL); 1666 to_delta = hdr->b_size; 1667 } 1668 atomic_add_64(size, to_delta); 1669 } 1670 } 1671 1672 ASSERT(!BUF_EMPTY(hdr)); 1673 if (new_state == arc_anon && HDR_IN_HASH_TABLE(hdr)) 1674 buf_hash_remove(hdr); 1675 1676 /* adjust state sizes (ignore arc_l2c_only) */ 1677 1678 if (to_delta && new_state != arc_l2c_only) { 1679 ASSERT(HDR_HAS_L1HDR(hdr)); 1680 if (GHOST_STATE(new_state)) { 1681 ASSERT0(datacnt); 1682 1683 /* 1684 * We moving a header to a ghost state, we first 1685 * remove all arc buffers. Thus, we'll have a 1686 * datacnt of zero, and no arc buffer to use for 1687 * the reference. As a result, we use the arc 1688 * header pointer for the reference. 1689 */ 1690 (void) refcount_add_many(&new_state->arcs_size, 1691 hdr->b_size, hdr); 1692 } else { 1693 ASSERT3U(datacnt, !=, 0); 1694 1695 /* 1696 * Each individual buffer holds a unique reference, 1697 * thus we must remove each of these references one 1698 * at a time. 1699 */ 1700 for (arc_buf_t *buf = hdr->b_l1hdr.b_buf; buf != NULL; 1701 buf = buf->b_next) { 1702 (void) refcount_add_many(&new_state->arcs_size, 1703 hdr->b_size, buf); 1704 } 1705 } 1706 } 1707 1708 if (from_delta && old_state != arc_l2c_only) { 1709 ASSERT(HDR_HAS_L1HDR(hdr)); 1710 if (GHOST_STATE(old_state)) { 1711 /* 1712 * When moving a header off of a ghost state, 1713 * there's the possibility for datacnt to be 1714 * non-zero. This is because we first add the 1715 * arc buffer to the header prior to changing 1716 * the header's state. Since we used the header 1717 * for the reference when putting the header on 1718 * the ghost state, we must balance that and use 1719 * the header when removing off the ghost state 1720 * (even though datacnt is non zero). 1721 */ 1722 1723 IMPLY(datacnt == 0, new_state == arc_anon || 1724 new_state == arc_l2c_only); 1725 1726 (void) refcount_remove_many(&old_state->arcs_size, 1727 hdr->b_size, hdr); 1728 } else { 1729 ASSERT3P(datacnt, !=, 0); 1730 1731 /* 1732 * Each individual buffer holds a unique reference, 1733 * thus we must remove each of these references one 1734 * at a time. 1735 */ 1736 for (arc_buf_t *buf = hdr->b_l1hdr.b_buf; buf != NULL; 1737 buf = buf->b_next) { 1738 (void) refcount_remove_many( 1739 &old_state->arcs_size, hdr->b_size, buf); 1740 } 1741 } 1742 } 1743 1744 if (HDR_HAS_L1HDR(hdr)) 1745 hdr->b_l1hdr.b_state = new_state; 1746 1747 /* 1748 * L2 headers should never be on the L2 state list since they don't 1749 * have L1 headers allocated. 1750 */ 1751 ASSERT(multilist_is_empty(&arc_l2c_only->arcs_list[ARC_BUFC_DATA]) && 1752 multilist_is_empty(&arc_l2c_only->arcs_list[ARC_BUFC_METADATA])); 1753 } 1754 1755 void 1756 arc_space_consume(uint64_t space, arc_space_type_t type) 1757 { 1758 ASSERT(type >= 0 && type < ARC_SPACE_NUMTYPES); 1759 1760 switch (type) { 1761 case ARC_SPACE_DATA: 1762 ARCSTAT_INCR(arcstat_data_size, space); 1763 break; 1764 case ARC_SPACE_META: 1765 ARCSTAT_INCR(arcstat_metadata_size, space); 1766 break; 1767 case ARC_SPACE_OTHER: 1768 ARCSTAT_INCR(arcstat_other_size, space); 1769 break; 1770 case ARC_SPACE_HDRS: 1771 ARCSTAT_INCR(arcstat_hdr_size, space); 1772 break; 1773 case ARC_SPACE_L2HDRS: 1774 ARCSTAT_INCR(arcstat_l2_hdr_size, space); 1775 break; 1776 } 1777 1778 if (type != ARC_SPACE_DATA) 1779 ARCSTAT_INCR(arcstat_meta_used, space); 1780 1781 atomic_add_64(&arc_size, space); 1782 } 1783 1784 void 1785 arc_space_return(uint64_t space, arc_space_type_t type) 1786 { 1787 ASSERT(type >= 0 && type < ARC_SPACE_NUMTYPES); 1788 1789 switch (type) { 1790 case ARC_SPACE_DATA: 1791 ARCSTAT_INCR(arcstat_data_size, -space); 1792 break; 1793 case ARC_SPACE_META: 1794 ARCSTAT_INCR(arcstat_metadata_size, -space); 1795 break; 1796 case ARC_SPACE_OTHER: 1797 ARCSTAT_INCR(arcstat_other_size, -space); 1798 break; 1799 case ARC_SPACE_HDRS: 1800 ARCSTAT_INCR(arcstat_hdr_size, -space); 1801 break; 1802 case ARC_SPACE_L2HDRS: 1803 ARCSTAT_INCR(arcstat_l2_hdr_size, -space); 1804 break; 1805 } 1806 1807 if (type != ARC_SPACE_DATA) { 1808 ASSERT(arc_meta_used >= space); 1809 if (arc_meta_max < arc_meta_used) 1810 arc_meta_max = arc_meta_used; 1811 ARCSTAT_INCR(arcstat_meta_used, -space); 1812 } 1813 1814 ASSERT(arc_size >= space); 1815 atomic_add_64(&arc_size, -space); 1816 } 1817 1818 arc_buf_t * 1819 arc_buf_alloc(spa_t *spa, int32_t size, void *tag, arc_buf_contents_t type) 1820 { 1821 arc_buf_hdr_t *hdr; 1822 arc_buf_t *buf; 1823 1824 ASSERT3U(size, >, 0); 1825 hdr = kmem_cache_alloc(hdr_full_cache, KM_PUSHPAGE); 1826 ASSERT(BUF_EMPTY(hdr)); 1827 ASSERT3P(hdr->b_freeze_cksum, ==, NULL); 1828 hdr->b_size = size; 1829 hdr->b_spa = spa_load_guid(spa); 1830 1831 buf = kmem_cache_alloc(buf_cache, KM_PUSHPAGE); 1832 buf->b_hdr = hdr; 1833 buf->b_data = NULL; 1834 buf->b_efunc = NULL; 1835 buf->b_private = NULL; 1836 buf->b_next = NULL; 1837 1838 hdr->b_flags = arc_bufc_to_flags(type); 1839 hdr->b_flags |= ARC_FLAG_HAS_L1HDR; 1840 1841 hdr->b_l1hdr.b_buf = buf; 1842 hdr->b_l1hdr.b_state = arc_anon; 1843 hdr->b_l1hdr.b_arc_access = 0; 1844 hdr->b_l1hdr.b_datacnt = 1; 1845 hdr->b_l1hdr.b_tmp_cdata = NULL; 1846 1847 arc_get_data_buf(buf); 1848 ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); 1849 (void) refcount_add(&hdr->b_l1hdr.b_refcnt, tag); 1850 1851 return (buf); 1852 } 1853 1854 static char *arc_onloan_tag = "onloan"; 1855 1856 /* 1857 * Loan out an anonymous arc buffer. Loaned buffers are not counted as in 1858 * flight data by arc_tempreserve_space() until they are "returned". Loaned 1859 * buffers must be returned to the arc before they can be used by the DMU or 1860 * freed. 1861 */ 1862 arc_buf_t * 1863 arc_loan_buf(spa_t *spa, int size) 1864 { 1865 arc_buf_t *buf; 1866 1867 buf = arc_buf_alloc(spa, size, arc_onloan_tag, ARC_BUFC_DATA); 1868 1869 atomic_add_64(&arc_loaned_bytes, size); 1870 return (buf); 1871 } 1872 1873 /* 1874 * Return a loaned arc buffer to the arc. 1875 */ 1876 void 1877 arc_return_buf(arc_buf_t *buf, void *tag) 1878 { 1879 arc_buf_hdr_t *hdr = buf->b_hdr; 1880 1881 ASSERT(buf->b_data != NULL); 1882 ASSERT(HDR_HAS_L1HDR(hdr)); 1883 (void) refcount_add(&hdr->b_l1hdr.b_refcnt, tag); 1884 (void) refcount_remove(&hdr->b_l1hdr.b_refcnt, arc_onloan_tag); 1885 1886 atomic_add_64(&arc_loaned_bytes, -hdr->b_size); 1887 } 1888 1889 /* Detach an arc_buf from a dbuf (tag) */ 1890 void 1891 arc_loan_inuse_buf(arc_buf_t *buf, void *tag) 1892 { 1893 arc_buf_hdr_t *hdr = buf->b_hdr; 1894 1895 ASSERT(buf->b_data != NULL); 1896 ASSERT(HDR_HAS_L1HDR(hdr)); 1897 (void) refcount_add(&hdr->b_l1hdr.b_refcnt, arc_onloan_tag); 1898 (void) refcount_remove(&hdr->b_l1hdr.b_refcnt, tag); 1899 buf->b_efunc = NULL; 1900 buf->b_private = NULL; 1901 1902 atomic_add_64(&arc_loaned_bytes, hdr->b_size); 1903 } 1904 1905 static arc_buf_t * 1906 arc_buf_clone(arc_buf_t *from) 1907 { 1908 arc_buf_t *buf; 1909 arc_buf_hdr_t *hdr = from->b_hdr; 1910 uint64_t size = hdr->b_size; 1911 1912 ASSERT(HDR_HAS_L1HDR(hdr)); 1913 ASSERT(hdr->b_l1hdr.b_state != arc_anon); 1914 1915 buf = kmem_cache_alloc(buf_cache, KM_PUSHPAGE); 1916 buf->b_hdr = hdr; 1917 buf->b_data = NULL; 1918 buf->b_efunc = NULL; 1919 buf->b_private = NULL; 1920 buf->b_next = hdr->b_l1hdr.b_buf; 1921 hdr->b_l1hdr.b_buf = buf; 1922 arc_get_data_buf(buf); 1923 bcopy(from->b_data, buf->b_data, size); 1924 1925 /* 1926 * This buffer already exists in the arc so create a duplicate 1927 * copy for the caller. If the buffer is associated with user data 1928 * then track the size and number of duplicates. These stats will be 1929 * updated as duplicate buffers are created and destroyed. 1930 */ 1931 if (HDR_ISTYPE_DATA(hdr)) { 1932 ARCSTAT_BUMP(arcstat_duplicate_buffers); 1933 ARCSTAT_INCR(arcstat_duplicate_buffers_size, size); 1934 } 1935 hdr->b_l1hdr.b_datacnt += 1; 1936 return (buf); 1937 } 1938 1939 void 1940 arc_buf_add_ref(arc_buf_t *buf, void* tag) 1941 { 1942 arc_buf_hdr_t *hdr; 1943 kmutex_t *hash_lock; 1944 1945 /* 1946 * Check to see if this buffer is evicted. Callers 1947 * must verify b_data != NULL to know if the add_ref 1948 * was successful. 1949 */ 1950 mutex_enter(&buf->b_evict_lock); 1951 if (buf->b_data == NULL) { 1952 mutex_exit(&buf->b_evict_lock); 1953 return; 1954 } 1955 hash_lock = HDR_LOCK(buf->b_hdr); 1956 mutex_enter(hash_lock); 1957 hdr = buf->b_hdr; 1958 ASSERT(HDR_HAS_L1HDR(hdr)); 1959 ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); 1960 mutex_exit(&buf->b_evict_lock); 1961 1962 ASSERT(hdr->b_l1hdr.b_state == arc_mru || 1963 hdr->b_l1hdr.b_state == arc_mfu); 1964 1965 add_reference(hdr, hash_lock, tag); 1966 DTRACE_PROBE1(arc__hit, arc_buf_hdr_t *, hdr); 1967 arc_access(hdr, hash_lock); 1968 mutex_exit(hash_lock); 1969 ARCSTAT_BUMP(arcstat_hits); 1970 ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr), 1971 demand, prefetch, !HDR_ISTYPE_METADATA(hdr), 1972 data, metadata, hits); 1973 } 1974 1975 static void 1976 arc_buf_free_on_write(void *data, size_t size, 1977 void (*free_func)(void *, size_t)) 1978 { 1979 l2arc_data_free_t *df; 1980 1981 df = kmem_alloc(sizeof (*df), KM_SLEEP); 1982 df->l2df_data = data; 1983 df->l2df_size = size; 1984 df->l2df_func = free_func; 1985 mutex_enter(&l2arc_free_on_write_mtx); 1986 list_insert_head(l2arc_free_on_write, df); 1987 mutex_exit(&l2arc_free_on_write_mtx); 1988 } 1989 1990 /* 1991 * Free the arc data buffer. If it is an l2arc write in progress, 1992 * the buffer is placed on l2arc_free_on_write to be freed later. 1993 */ 1994 static void 1995 arc_buf_data_free(arc_buf_t *buf, void (*free_func)(void *, size_t)) 1996 { 1997 arc_buf_hdr_t *hdr = buf->b_hdr; 1998 1999 if (HDR_L2_WRITING(hdr)) { 2000 arc_buf_free_on_write(buf->b_data, hdr->b_size, free_func); 2001 ARCSTAT_BUMP(arcstat_l2_free_on_write); 2002 } else { 2003 free_func(buf->b_data, hdr->b_size); 2004 } 2005 } 2006 2007 static void 2008 arc_buf_l2_cdata_free(arc_buf_hdr_t *hdr) 2009 { 2010 ASSERT(HDR_HAS_L2HDR(hdr)); 2011 ASSERT(MUTEX_HELD(&hdr->b_l2hdr.b_dev->l2ad_mtx)); 2012 2013 /* 2014 * The b_tmp_cdata field is linked off of the b_l1hdr, so if 2015 * that doesn't exist, the header is in the arc_l2c_only state, 2016 * and there isn't anything to free (it's already been freed). 2017 */ 2018 if (!HDR_HAS_L1HDR(hdr)) 2019 return; 2020 2021 /* 2022 * The header isn't being written to the l2arc device, thus it 2023 * shouldn't have a b_tmp_cdata to free. 2024 */ 2025 if (!HDR_L2_WRITING(hdr)) { 2026 ASSERT3P(hdr->b_l1hdr.b_tmp_cdata, ==, NULL); 2027 return; 2028 } 2029 2030 /* 2031 * The header does not have compression enabled. This can be due 2032 * to the buffer not being compressible, or because we're 2033 * freeing the buffer before the second phase of 2034 * l2arc_write_buffer() has started (which does the compression 2035 * step). In either case, b_tmp_cdata does not point to a 2036 * separately compressed buffer, so there's nothing to free (it 2037 * points to the same buffer as the arc_buf_t's b_data field). 2038 */ 2039 if (HDR_GET_COMPRESS(hdr) == ZIO_COMPRESS_OFF) { 2040 hdr->b_l1hdr.b_tmp_cdata = NULL; 2041 return; 2042 } 2043 2044 /* 2045 * There's nothing to free since the buffer was all zero's and 2046 * compressed to a zero length buffer. 2047 */ 2048 if (HDR_GET_COMPRESS(hdr) == ZIO_COMPRESS_EMPTY) { 2049 ASSERT3P(hdr->b_l1hdr.b_tmp_cdata, ==, NULL); 2050 return; 2051 } 2052 2053 ASSERT(L2ARC_IS_VALID_COMPRESS(HDR_GET_COMPRESS(hdr))); 2054 2055 arc_buf_free_on_write(hdr->b_l1hdr.b_tmp_cdata, 2056 hdr->b_size, zio_data_buf_free); 2057 2058 ARCSTAT_BUMP(arcstat_l2_cdata_free_on_write); 2059 hdr->b_l1hdr.b_tmp_cdata = NULL; 2060 } 2061 2062 /* 2063 * Free up buf->b_data and if 'remove' is set, then pull the 2064 * arc_buf_t off of the the arc_buf_hdr_t's list and free it. 2065 */ 2066 static void 2067 arc_buf_destroy(arc_buf_t *buf, boolean_t remove) 2068 { 2069 arc_buf_t **bufp; 2070 2071 /* free up data associated with the buf */ 2072 if (buf->b_data != NULL) { 2073 arc_state_t *state = buf->b_hdr->b_l1hdr.b_state; 2074 uint64_t size = buf->b_hdr->b_size; 2075 arc_buf_contents_t type = arc_buf_type(buf->b_hdr); 2076 2077 arc_cksum_verify(buf); 2078 arc_buf_unwatch(buf); 2079 2080 if (type == ARC_BUFC_METADATA) { 2081 arc_buf_data_free(buf, zio_buf_free); 2082 arc_space_return(size, ARC_SPACE_META); 2083 } else { 2084 ASSERT(type == ARC_BUFC_DATA); 2085 arc_buf_data_free(buf, zio_data_buf_free); 2086 arc_space_return(size, ARC_SPACE_DATA); 2087 } 2088 2089 /* protected by hash lock, if in the hash table */ 2090 if (multilist_link_active(&buf->b_hdr->b_l1hdr.b_arc_node)) { 2091 uint64_t *cnt = &state->arcs_lsize[type]; 2092 2093 ASSERT(refcount_is_zero( 2094 &buf->b_hdr->b_l1hdr.b_refcnt)); 2095 ASSERT(state != arc_anon && state != arc_l2c_only); 2096 2097 ASSERT3U(*cnt, >=, size); 2098 atomic_add_64(cnt, -size); 2099 } 2100 2101 (void) refcount_remove_many(&state->arcs_size, size, buf); 2102 buf->b_data = NULL; 2103 2104 /* 2105 * If we're destroying a duplicate buffer make sure 2106 * that the appropriate statistics are updated. 2107 */ 2108 if (buf->b_hdr->b_l1hdr.b_datacnt > 1 && 2109 HDR_ISTYPE_DATA(buf->b_hdr)) { 2110 ARCSTAT_BUMPDOWN(arcstat_duplicate_buffers); 2111 ARCSTAT_INCR(arcstat_duplicate_buffers_size, -size); 2112 } 2113 ASSERT(buf->b_hdr->b_l1hdr.b_datacnt > 0); 2114 buf->b_hdr->b_l1hdr.b_datacnt -= 1; 2115 } 2116 2117 /* only remove the buf if requested */ 2118 if (!remove) 2119 return; 2120 2121 /* remove the buf from the hdr list */ 2122 for (bufp = &buf->b_hdr->b_l1hdr.b_buf; *bufp != buf; 2123 bufp = &(*bufp)->b_next) 2124 continue; 2125 *bufp = buf->b_next; 2126 buf->b_next = NULL; 2127 2128 ASSERT(buf->b_efunc == NULL); 2129 2130 /* clean up the buf */ 2131 buf->b_hdr = NULL; 2132 kmem_cache_free(buf_cache, buf); 2133 } 2134 2135 static void 2136 arc_hdr_l2hdr_destroy(arc_buf_hdr_t *hdr) 2137 { 2138 l2arc_buf_hdr_t *l2hdr = &hdr->b_l2hdr; 2139 l2arc_dev_t *dev = l2hdr->b_dev; 2140 2141 ASSERT(MUTEX_HELD(&dev->l2ad_mtx)); 2142 ASSERT(HDR_HAS_L2HDR(hdr)); 2143 2144 list_remove(&dev->l2ad_buflist, hdr); 2145 2146 /* 2147 * We don't want to leak the b_tmp_cdata buffer that was 2148 * allocated in l2arc_write_buffers() 2149 */ 2150 arc_buf_l2_cdata_free(hdr); 2151 2152 /* 2153 * If the l2hdr's b_daddr is equal to L2ARC_ADDR_UNSET, then 2154 * this header is being processed by l2arc_write_buffers() (i.e. 2155 * it's in the first stage of l2arc_write_buffers()). 2156 * Re-affirming that truth here, just to serve as a reminder. If 2157 * b_daddr does not equal L2ARC_ADDR_UNSET, then the header may or 2158 * may not have its HDR_L2_WRITING flag set. (the write may have 2159 * completed, in which case HDR_L2_WRITING will be false and the 2160 * b_daddr field will point to the address of the buffer on disk). 2161 */ 2162 IMPLY(l2hdr->b_daddr == L2ARC_ADDR_UNSET, HDR_L2_WRITING(hdr)); 2163 2164 /* 2165 * If b_daddr is equal to L2ARC_ADDR_UNSET, we're racing with 2166 * l2arc_write_buffers(). Since we've just removed this header 2167 * from the l2arc buffer list, this header will never reach the 2168 * second stage of l2arc_write_buffers(), which increments the 2169 * accounting stats for this header. Thus, we must be careful 2170 * not to decrement them for this header either. 2171 */ 2172 if (l2hdr->b_daddr != L2ARC_ADDR_UNSET) { 2173 ARCSTAT_INCR(arcstat_l2_asize, -l2hdr->b_asize); 2174 ARCSTAT_INCR(arcstat_l2_size, -hdr->b_size); 2175 2176 vdev_space_update(dev->l2ad_vdev, 2177 -l2hdr->b_asize, 0, 0); 2178 2179 (void) refcount_remove_many(&dev->l2ad_alloc, 2180 l2hdr->b_asize, hdr); 2181 } 2182 2183 hdr->b_flags &= ~ARC_FLAG_HAS_L2HDR; 2184 } 2185 2186 static void 2187 arc_hdr_destroy(arc_buf_hdr_t *hdr) 2188 { 2189 if (HDR_HAS_L1HDR(hdr)) { 2190 ASSERT(hdr->b_l1hdr.b_buf == NULL || 2191 hdr->b_l1hdr.b_datacnt > 0); 2192 ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); 2193 ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon); 2194 } 2195 ASSERT(!HDR_IO_IN_PROGRESS(hdr)); 2196 ASSERT(!HDR_IN_HASH_TABLE(hdr)); 2197 2198 if (HDR_HAS_L2HDR(hdr)) { 2199 l2arc_dev_t *dev = hdr->b_l2hdr.b_dev; 2200 boolean_t buflist_held = MUTEX_HELD(&dev->l2ad_mtx); 2201 2202 if (!buflist_held) 2203 mutex_enter(&dev->l2ad_mtx); 2204 2205 /* 2206 * Even though we checked this conditional above, we 2207 * need to check this again now that we have the 2208 * l2ad_mtx. This is because we could be racing with 2209 * another thread calling l2arc_evict() which might have 2210 * destroyed this header's L2 portion as we were waiting 2211 * to acquire the l2ad_mtx. If that happens, we don't 2212 * want to re-destroy the header's L2 portion. 2213 */ 2214 if (HDR_HAS_L2HDR(hdr)) 2215 arc_hdr_l2hdr_destroy(hdr); 2216 2217 if (!buflist_held) 2218 mutex_exit(&dev->l2ad_mtx); 2219 } 2220 2221 if (!BUF_EMPTY(hdr)) 2222 buf_discard_identity(hdr); 2223 2224 if (hdr->b_freeze_cksum != NULL) { 2225 kmem_free(hdr->b_freeze_cksum, sizeof (zio_cksum_t)); 2226 hdr->b_freeze_cksum = NULL; 2227 } 2228 2229 if (HDR_HAS_L1HDR(hdr)) { 2230 while (hdr->b_l1hdr.b_buf) { 2231 arc_buf_t *buf = hdr->b_l1hdr.b_buf; 2232 2233 if (buf->b_efunc != NULL) { 2234 mutex_enter(&arc_user_evicts_lock); 2235 mutex_enter(&buf->b_evict_lock); 2236 ASSERT(buf->b_hdr != NULL); 2237 arc_buf_destroy(hdr->b_l1hdr.b_buf, FALSE); 2238 hdr->b_l1hdr.b_buf = buf->b_next; 2239 buf->b_hdr = &arc_eviction_hdr; 2240 buf->b_next = arc_eviction_list; 2241 arc_eviction_list = buf; 2242 mutex_exit(&buf->b_evict_lock); 2243 cv_signal(&arc_user_evicts_cv); 2244 mutex_exit(&arc_user_evicts_lock); 2245 } else { 2246 arc_buf_destroy(hdr->b_l1hdr.b_buf, TRUE); 2247 } 2248 } 2249 #ifdef ZFS_DEBUG 2250 if (hdr->b_l1hdr.b_thawed != NULL) { 2251 kmem_free(hdr->b_l1hdr.b_thawed, 1); 2252 hdr->b_l1hdr.b_thawed = NULL; 2253 } 2254 #endif 2255 } 2256 2257 ASSERT3P(hdr->b_hash_next, ==, NULL); 2258 if (HDR_HAS_L1HDR(hdr)) { 2259 ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); 2260 ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL); 2261 kmem_cache_free(hdr_full_cache, hdr); 2262 } else { 2263 kmem_cache_free(hdr_l2only_cache, hdr); 2264 } 2265 } 2266 2267 void 2268 arc_buf_free(arc_buf_t *buf, void *tag) 2269 { 2270 arc_buf_hdr_t *hdr = buf->b_hdr; 2271 int hashed = hdr->b_l1hdr.b_state != arc_anon; 2272 2273 ASSERT(buf->b_efunc == NULL); 2274 ASSERT(buf->b_data != NULL); 2275 2276 if (hashed) { 2277 kmutex_t *hash_lock = HDR_LOCK(hdr); 2278 2279 mutex_enter(hash_lock); 2280 hdr = buf->b_hdr; 2281 ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); 2282 2283 (void) remove_reference(hdr, hash_lock, tag); 2284 if (hdr->b_l1hdr.b_datacnt > 1) { 2285 arc_buf_destroy(buf, TRUE); 2286 } else { 2287 ASSERT(buf == hdr->b_l1hdr.b_buf); 2288 ASSERT(buf->b_efunc == NULL); 2289 hdr->b_flags |= ARC_FLAG_BUF_AVAILABLE; 2290 } 2291 mutex_exit(hash_lock); 2292 } else if (HDR_IO_IN_PROGRESS(hdr)) { 2293 int destroy_hdr; 2294 /* 2295 * We are in the middle of an async write. Don't destroy 2296 * this buffer unless the write completes before we finish 2297 * decrementing the reference count. 2298 */ 2299 mutex_enter(&arc_user_evicts_lock); 2300 (void) remove_reference(hdr, NULL, tag); 2301 ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); 2302 destroy_hdr = !HDR_IO_IN_PROGRESS(hdr); 2303 mutex_exit(&arc_user_evicts_lock); 2304 if (destroy_hdr) 2305 arc_hdr_destroy(hdr); 2306 } else { 2307 if (remove_reference(hdr, NULL, tag) > 0) 2308 arc_buf_destroy(buf, TRUE); 2309 else 2310 arc_hdr_destroy(hdr); 2311 } 2312 } 2313 2314 boolean_t 2315 arc_buf_remove_ref(arc_buf_t *buf, void* tag) 2316 { 2317 arc_buf_hdr_t *hdr = buf->b_hdr; 2318 kmutex_t *hash_lock = HDR_LOCK(hdr); 2319 boolean_t no_callback = (buf->b_efunc == NULL); 2320 2321 if (hdr->b_l1hdr.b_state == arc_anon) { 2322 ASSERT(hdr->b_l1hdr.b_datacnt == 1); 2323 arc_buf_free(buf, tag); 2324 return (no_callback); 2325 } 2326 2327 mutex_enter(hash_lock); 2328 hdr = buf->b_hdr; 2329 ASSERT(hdr->b_l1hdr.b_datacnt > 0); 2330 ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); 2331 ASSERT(hdr->b_l1hdr.b_state != arc_anon); 2332 ASSERT(buf->b_data != NULL); 2333 2334 (void) remove_reference(hdr, hash_lock, tag); 2335 if (hdr->b_l1hdr.b_datacnt > 1) { 2336 if (no_callback) 2337 arc_buf_destroy(buf, TRUE); 2338 } else if (no_callback) { 2339 ASSERT(hdr->b_l1hdr.b_buf == buf && buf->b_next == NULL); 2340 ASSERT(buf->b_efunc == NULL); 2341 hdr->b_flags |= ARC_FLAG_BUF_AVAILABLE; 2342 } 2343 ASSERT(no_callback || hdr->b_l1hdr.b_datacnt > 1 || 2344 refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); 2345 mutex_exit(hash_lock); 2346 return (no_callback); 2347 } 2348 2349 int32_t 2350 arc_buf_size(arc_buf_t *buf) 2351 { 2352 return (buf->b_hdr->b_size); 2353 } 2354 2355 /* 2356 * Called from the DMU to determine if the current buffer should be 2357 * evicted. In order to ensure proper locking, the eviction must be initiated 2358 * from the DMU. Return true if the buffer is associated with user data and 2359 * duplicate buffers still exist. 2360 */ 2361 boolean_t 2362 arc_buf_eviction_needed(arc_buf_t *buf) 2363 { 2364 arc_buf_hdr_t *hdr; 2365 boolean_t evict_needed = B_FALSE; 2366 2367 if (zfs_disable_dup_eviction) 2368 return (B_FALSE); 2369 2370 mutex_enter(&buf->b_evict_lock); 2371 hdr = buf->b_hdr; 2372 if (hdr == NULL) { 2373 /* 2374 * We are in arc_do_user_evicts(); let that function 2375 * perform the eviction. 2376 */ 2377 ASSERT(buf->b_data == NULL); 2378 mutex_exit(&buf->b_evict_lock); 2379 return (B_FALSE); 2380 } else if (buf->b_data == NULL) { 2381 /* 2382 * We have already been added to the arc eviction list; 2383 * recommend eviction. 2384 */ 2385 ASSERT3P(hdr, ==, &arc_eviction_hdr); 2386 mutex_exit(&buf->b_evict_lock); 2387 return (B_TRUE); 2388 } 2389 2390 if (hdr->b_l1hdr.b_datacnt > 1 && HDR_ISTYPE_DATA(hdr)) 2391 evict_needed = B_TRUE; 2392 2393 mutex_exit(&buf->b_evict_lock); 2394 return (evict_needed); 2395 } 2396 2397 /* 2398 * Evict the arc_buf_hdr that is provided as a parameter. The resultant 2399 * state of the header is dependent on it's state prior to entering this 2400 * function. The following transitions are possible: 2401 * 2402 * - arc_mru -> arc_mru_ghost 2403 * - arc_mfu -> arc_mfu_ghost 2404 * - arc_mru_ghost -> arc_l2c_only 2405 * - arc_mru_ghost -> deleted 2406 * - arc_mfu_ghost -> arc_l2c_only 2407 * - arc_mfu_ghost -> deleted 2408 */ 2409 static int64_t 2410 arc_evict_hdr(arc_buf_hdr_t *hdr, kmutex_t *hash_lock) 2411 { 2412 arc_state_t *evicted_state, *state; 2413 int64_t bytes_evicted = 0; 2414 2415 ASSERT(MUTEX_HELD(hash_lock)); 2416 ASSERT(HDR_HAS_L1HDR(hdr)); 2417 2418 state = hdr->b_l1hdr.b_state; 2419 if (GHOST_STATE(state)) { 2420 ASSERT(!HDR_IO_IN_PROGRESS(hdr)); 2421 ASSERT(hdr->b_l1hdr.b_buf == NULL); 2422 2423 /* 2424 * l2arc_write_buffers() relies on a header's L1 portion 2425 * (i.e. it's b_tmp_cdata field) during it's write phase. 2426 * Thus, we cannot push a header onto the arc_l2c_only 2427 * state (removing it's L1 piece) until the header is 2428 * done being written to the l2arc. 2429 */ 2430 if (HDR_HAS_L2HDR(hdr) && HDR_L2_WRITING(hdr)) { 2431 ARCSTAT_BUMP(arcstat_evict_l2_skip); 2432 return (bytes_evicted); 2433 } 2434 2435 ARCSTAT_BUMP(arcstat_deleted); 2436 bytes_evicted += hdr->b_size; 2437 2438 DTRACE_PROBE1(arc__delete, arc_buf_hdr_t *, hdr); 2439 2440 if (HDR_HAS_L2HDR(hdr)) { 2441 /* 2442 * This buffer is cached on the 2nd Level ARC; 2443 * don't destroy the header. 2444 */ 2445 arc_change_state(arc_l2c_only, hdr, hash_lock); 2446 /* 2447 * dropping from L1+L2 cached to L2-only, 2448 * realloc to remove the L1 header. 2449 */ 2450 hdr = arc_hdr_realloc(hdr, hdr_full_cache, 2451 hdr_l2only_cache); 2452 } else { 2453 arc_change_state(arc_anon, hdr, hash_lock); 2454 arc_hdr_destroy(hdr); 2455 } 2456 return (bytes_evicted); 2457 } 2458 2459 ASSERT(state == arc_mru || state == arc_mfu); 2460 evicted_state = (state == arc_mru) ? arc_mru_ghost : arc_mfu_ghost; 2461 2462 /* prefetch buffers have a minimum lifespan */ 2463 if (HDR_IO_IN_PROGRESS(hdr) || 2464 ((hdr->b_flags & (ARC_FLAG_PREFETCH | ARC_FLAG_INDIRECT)) && 2465 ddi_get_lbolt() - hdr->b_l1hdr.b_arc_access < 2466 arc_min_prefetch_lifespan)) { 2467 ARCSTAT_BUMP(arcstat_evict_skip); 2468 return (bytes_evicted); 2469 } 2470 2471 ASSERT0(refcount_count(&hdr->b_l1hdr.b_refcnt)); 2472 ASSERT3U(hdr->b_l1hdr.b_datacnt, >, 0); 2473 while (hdr->b_l1hdr.b_buf) { 2474 arc_buf_t *buf = hdr->b_l1hdr.b_buf; 2475 if (!mutex_tryenter(&buf->b_evict_lock)) { 2476 ARCSTAT_BUMP(arcstat_mutex_miss); 2477 break; 2478 } 2479 if (buf->b_data != NULL) 2480 bytes_evicted += hdr->b_size; 2481 if (buf->b_efunc != NULL) { 2482 mutex_enter(&arc_user_evicts_lock); 2483 arc_buf_destroy(buf, FALSE); 2484 hdr->b_l1hdr.b_buf = buf->b_next; 2485 buf->b_hdr = &arc_eviction_hdr; 2486 buf->b_next = arc_eviction_list; 2487 arc_eviction_list = buf; 2488 cv_signal(&arc_user_evicts_cv); 2489 mutex_exit(&arc_user_evicts_lock); 2490 mutex_exit(&buf->b_evict_lock); 2491 } else { 2492 mutex_exit(&buf->b_evict_lock); 2493 arc_buf_destroy(buf, TRUE); 2494 } 2495 } 2496 2497 if (HDR_HAS_L2HDR(hdr)) { 2498 ARCSTAT_INCR(arcstat_evict_l2_cached, hdr->b_size); 2499 } else { 2500 if (l2arc_write_eligible(hdr->b_spa, hdr)) 2501 ARCSTAT_INCR(arcstat_evict_l2_eligible, hdr->b_size); 2502 else 2503 ARCSTAT_INCR(arcstat_evict_l2_ineligible, hdr->b_size); 2504 } 2505 2506 if (hdr->b_l1hdr.b_datacnt == 0) { 2507 arc_change_state(evicted_state, hdr, hash_lock); 2508 ASSERT(HDR_IN_HASH_TABLE(hdr)); 2509 hdr->b_flags |= ARC_FLAG_IN_HASH_TABLE; 2510 hdr->b_flags &= ~ARC_FLAG_BUF_AVAILABLE; 2511 DTRACE_PROBE1(arc__evict, arc_buf_hdr_t *, hdr); 2512 } 2513 2514 return (bytes_evicted); 2515 } 2516 2517 static uint64_t 2518 arc_evict_state_impl(multilist_t *ml, int idx, arc_buf_hdr_t *marker, 2519 uint64_t spa, int64_t bytes) 2520 { 2521 multilist_sublist_t *mls; 2522 uint64_t bytes_evicted = 0; 2523 arc_buf_hdr_t *hdr; 2524 kmutex_t *hash_lock; 2525 int evict_count = 0; 2526 2527 ASSERT3P(marker, !=, NULL); 2528 IMPLY(bytes < 0, bytes == ARC_EVICT_ALL); 2529 2530 mls = multilist_sublist_lock(ml, idx); 2531 2532 for (hdr = multilist_sublist_prev(mls, marker); hdr != NULL; 2533 hdr = multilist_sublist_prev(mls, marker)) { 2534 if ((bytes != ARC_EVICT_ALL && bytes_evicted >= bytes) || 2535 (evict_count >= zfs_arc_evict_batch_limit)) 2536 break; 2537 2538 /* 2539 * To keep our iteration location, move the marker 2540 * forward. Since we're not holding hdr's hash lock, we 2541 * must be very careful and not remove 'hdr' from the 2542 * sublist. Otherwise, other consumers might mistake the 2543 * 'hdr' as not being on a sublist when they call the 2544 * multilist_link_active() function (they all rely on 2545 * the hash lock protecting concurrent insertions and 2546 * removals). multilist_sublist_move_forward() was 2547 * specifically implemented to ensure this is the case 2548 * (only 'marker' will be removed and re-inserted). 2549 */ 2550 multilist_sublist_move_forward(mls, marker); 2551 2552 /* 2553 * The only case where the b_spa field should ever be 2554 * zero, is the marker headers inserted by 2555 * arc_evict_state(). It's possible for multiple threads 2556 * to be calling arc_evict_state() concurrently (e.g. 2557 * dsl_pool_close() and zio_inject_fault()), so we must 2558 * skip any markers we see from these other threads. 2559 */ 2560 if (hdr->b_spa == 0) 2561 continue; 2562 2563 /* we're only interested in evicting buffers of a certain spa */ 2564 if (spa != 0 && hdr->b_spa != spa) { 2565 ARCSTAT_BUMP(arcstat_evict_skip); 2566 continue; 2567 } 2568 2569 hash_lock = HDR_LOCK(hdr); 2570 2571 /* 2572 * We aren't calling this function from any code path 2573 * that would already be holding a hash lock, so we're 2574 * asserting on this assumption to be defensive in case 2575 * this ever changes. Without this check, it would be 2576 * possible to incorrectly increment arcstat_mutex_miss 2577 * below (e.g. if the code changed such that we called 2578 * this function with a hash lock held). 2579 */ 2580 ASSERT(!MUTEX_HELD(hash_lock)); 2581 2582 if (mutex_tryenter(hash_lock)) { 2583 uint64_t evicted = arc_evict_hdr(hdr, hash_lock); 2584 mutex_exit(hash_lock); 2585 2586 bytes_evicted += evicted; 2587 2588 /* 2589 * If evicted is zero, arc_evict_hdr() must have 2590 * decided to skip this header, don't increment 2591 * evict_count in this case. 2592 */ 2593 if (evicted != 0) 2594 evict_count++; 2595 2596 /* 2597 * If arc_size isn't overflowing, signal any 2598 * threads that might happen to be waiting. 2599 * 2600 * For each header evicted, we wake up a single 2601 * thread. If we used cv_broadcast, we could 2602 * wake up "too many" threads causing arc_size 2603 * to significantly overflow arc_c; since 2604 * arc_get_data_buf() doesn't check for overflow 2605 * when it's woken up (it doesn't because it's 2606 * possible for the ARC to be overflowing while 2607 * full of un-evictable buffers, and the 2608 * function should proceed in this case). 2609 * 2610 * If threads are left sleeping, due to not 2611 * using cv_broadcast, they will be woken up 2612 * just before arc_reclaim_thread() sleeps. 2613 */ 2614 mutex_enter(&arc_reclaim_lock); 2615 if (!arc_is_overflowing()) 2616 cv_signal(&arc_reclaim_waiters_cv); 2617 mutex_exit(&arc_reclaim_lock); 2618 } else { 2619 ARCSTAT_BUMP(arcstat_mutex_miss); 2620 } 2621 } 2622 2623 multilist_sublist_unlock(mls); 2624 2625 return (bytes_evicted); 2626 } 2627 2628 /* 2629 * Evict buffers from the given arc state, until we've removed the 2630 * specified number of bytes. Move the removed buffers to the 2631 * appropriate evict state. 2632 * 2633 * This function makes a "best effort". It skips over any buffers 2634 * it can't get a hash_lock on, and so, may not catch all candidates. 2635 * It may also return without evicting as much space as requested. 2636 * 2637 * If bytes is specified using the special value ARC_EVICT_ALL, this 2638 * will evict all available (i.e. unlocked and evictable) buffers from 2639 * the given arc state; which is used by arc_flush(). 2640 */ 2641 static uint64_t 2642 arc_evict_state(arc_state_t *state, uint64_t spa, int64_t bytes, 2643 arc_buf_contents_t type) 2644 { 2645 uint64_t total_evicted = 0; 2646 multilist_t *ml = &state->arcs_list[type]; 2647 int num_sublists; 2648 arc_buf_hdr_t **markers; 2649 2650 IMPLY(bytes < 0, bytes == ARC_EVICT_ALL); 2651 2652 num_sublists = multilist_get_num_sublists(ml); 2653 2654 /* 2655 * If we've tried to evict from each sublist, made some 2656 * progress, but still have not hit the target number of bytes 2657 * to evict, we want to keep trying. The markers allow us to 2658 * pick up where we left off for each individual sublist, rather 2659 * than starting from the tail each time. 2660 */ 2661 markers = kmem_zalloc(sizeof (*markers) * num_sublists, KM_SLEEP); 2662 for (int i = 0; i < num_sublists; i++) { 2663 markers[i] = kmem_cache_alloc(hdr_full_cache, KM_SLEEP); 2664 2665 /* 2666 * A b_spa of 0 is used to indicate that this header is 2667 * a marker. This fact is used in arc_adjust_type() and 2668 * arc_evict_state_impl(). 2669 */ 2670 markers[i]->b_spa = 0; 2671 2672 multilist_sublist_t *mls = multilist_sublist_lock(ml, i); 2673 multilist_sublist_insert_tail(mls, markers[i]); 2674 multilist_sublist_unlock(mls); 2675 } 2676 2677 /* 2678 * While we haven't hit our target number of bytes to evict, or 2679 * we're evicting all available buffers. 2680 */ 2681 while (total_evicted < bytes || bytes == ARC_EVICT_ALL) { 2682 /* 2683 * Start eviction using a randomly selected sublist, 2684 * this is to try and evenly balance eviction across all 2685 * sublists. Always starting at the same sublist 2686 * (e.g. index 0) would cause evictions to favor certain 2687 * sublists over others. 2688 */ 2689 int sublist_idx = multilist_get_random_index(ml); 2690 uint64_t scan_evicted = 0; 2691 2692 for (int i = 0; i < num_sublists; i++) { 2693 uint64_t bytes_remaining; 2694 uint64_t bytes_evicted; 2695 2696 if (bytes == ARC_EVICT_ALL) 2697 bytes_remaining = ARC_EVICT_ALL; 2698 else if (total_evicted < bytes) 2699 bytes_remaining = bytes - total_evicted; 2700 else 2701 break; 2702 2703 bytes_evicted = arc_evict_state_impl(ml, sublist_idx, 2704 markers[sublist_idx], spa, bytes_remaining); 2705 2706 scan_evicted += bytes_evicted; 2707 total_evicted += bytes_evicted; 2708 2709 /* we've reached the end, wrap to the beginning */ 2710 if (++sublist_idx >= num_sublists) 2711 sublist_idx = 0; 2712 } 2713 2714 /* 2715 * If we didn't evict anything during this scan, we have 2716 * no reason to believe we'll evict more during another 2717 * scan, so break the loop. 2718 */ 2719 if (scan_evicted == 0) { 2720 /* This isn't possible, let's make that obvious */ 2721 ASSERT3S(bytes, !=, 0); 2722 2723 /* 2724 * When bytes is ARC_EVICT_ALL, the only way to 2725 * break the loop is when scan_evicted is zero. 2726 * In that case, we actually have evicted enough, 2727 * so we don't want to increment the kstat. 2728 */ 2729 if (bytes != ARC_EVICT_ALL) { 2730 ASSERT3S(total_evicted, <, bytes); 2731 ARCSTAT_BUMP(arcstat_evict_not_enough); 2732 } 2733 2734 break; 2735 } 2736 } 2737 2738 for (int i = 0; i < num_sublists; i++) { 2739 multilist_sublist_t *mls = multilist_sublist_lock(ml, i); 2740 multilist_sublist_remove(mls, markers[i]); 2741 multilist_sublist_unlock(mls); 2742 2743 kmem_cache_free(hdr_full_cache, markers[i]); 2744 } 2745 kmem_free(markers, sizeof (*markers) * num_sublists); 2746 2747 return (total_evicted); 2748 } 2749 2750 /* 2751 * Flush all "evictable" data of the given type from the arc state 2752 * specified. This will not evict any "active" buffers (i.e. referenced). 2753 * 2754 * When 'retry' is set to FALSE, the function will make a single pass 2755 * over the state and evict any buffers that it can. Since it doesn't 2756 * continually retry the eviction, it might end up leaving some buffers 2757 * in the ARC due to lock misses. 2758 * 2759 * When 'retry' is set to TRUE, the function will continually retry the 2760 * eviction until *all* evictable buffers have been removed from the 2761 * state. As a result, if concurrent insertions into the state are 2762 * allowed (e.g. if the ARC isn't shutting down), this function might 2763 * wind up in an infinite loop, continually trying to evict buffers. 2764 */ 2765 static uint64_t 2766 arc_flush_state(arc_state_t *state, uint64_t spa, arc_buf_contents_t type, 2767 boolean_t retry) 2768 { 2769 uint64_t evicted = 0; 2770 2771 while (state->arcs_lsize[type] != 0) { 2772 evicted += arc_evict_state(state, spa, ARC_EVICT_ALL, type); 2773 2774 if (!retry) 2775 break; 2776 } 2777 2778 return (evicted); 2779 } 2780 2781 /* 2782 * Evict the specified number of bytes from the state specified, 2783 * restricting eviction to the spa and type given. This function 2784 * prevents us from trying to evict more from a state's list than 2785 * is "evictable", and to skip evicting altogether when passed a 2786 * negative value for "bytes". In contrast, arc_evict_state() will 2787 * evict everything it can, when passed a negative value for "bytes". 2788 */ 2789 static uint64_t 2790 arc_adjust_impl(arc_state_t *state, uint64_t spa, int64_t bytes, 2791 arc_buf_contents_t type) 2792 { 2793 int64_t delta; 2794 2795 if (bytes > 0 && state->arcs_lsize[type] > 0) { 2796 delta = MIN(state->arcs_lsize[type], bytes); 2797 return (arc_evict_state(state, spa, delta, type)); 2798 } 2799 2800 return (0); 2801 } 2802 2803 /* 2804 * Evict metadata buffers from the cache, such that arc_meta_used is 2805 * capped by the arc_meta_limit tunable. 2806 */ 2807 static uint64_t 2808 arc_adjust_meta(void) 2809 { 2810 uint64_t total_evicted = 0; 2811 int64_t target; 2812 2813 /* 2814 * If we're over the meta limit, we want to evict enough 2815 * metadata to get back under the meta limit. We don't want to 2816 * evict so much that we drop the MRU below arc_p, though. If 2817 * we're over the meta limit more than we're over arc_p, we 2818 * evict some from the MRU here, and some from the MFU below. 2819 */ 2820 target = MIN((int64_t)(arc_meta_used - arc_meta_limit), 2821 (int64_t)(refcount_count(&arc_anon->arcs_size) + 2822 refcount_count(&arc_mru->arcs_size) - arc_p)); 2823 2824 total_evicted += arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_METADATA); 2825 2826 /* 2827 * Similar to the above, we want to evict enough bytes to get us 2828 * below the meta limit, but not so much as to drop us below the 2829 * space alloted to the MFU (which is defined as arc_c - arc_p). 2830 */ 2831 target = MIN((int64_t)(arc_meta_used - arc_meta_limit), 2832 (int64_t)(refcount_count(&arc_mfu->arcs_size) - (arc_c - arc_p))); 2833 2834 total_evicted += arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_METADATA); 2835 2836 return (total_evicted); 2837 } 2838 2839 /* 2840 * Return the type of the oldest buffer in the given arc state 2841 * 2842 * This function will select a random sublist of type ARC_BUFC_DATA and 2843 * a random sublist of type ARC_BUFC_METADATA. The tail of each sublist 2844 * is compared, and the type which contains the "older" buffer will be 2845 * returned. 2846 */ 2847 static arc_buf_contents_t 2848 arc_adjust_type(arc_state_t *state) 2849 { 2850 multilist_t *data_ml = &state->arcs_list[ARC_BUFC_DATA]; 2851 multilist_t *meta_ml = &state->arcs_list[ARC_BUFC_METADATA]; 2852 int data_idx = multilist_get_random_index(data_ml); 2853 int meta_idx = multilist_get_random_index(meta_ml); 2854 multilist_sublist_t *data_mls; 2855 multilist_sublist_t *meta_mls; 2856 arc_buf_contents_t type; 2857 arc_buf_hdr_t *data_hdr; 2858 arc_buf_hdr_t *meta_hdr; 2859 2860 /* 2861 * We keep the sublist lock until we're finished, to prevent 2862 * the headers from being destroyed via arc_evict_state(). 2863 */ 2864 data_mls = multilist_sublist_lock(data_ml, data_idx); 2865 meta_mls = multilist_sublist_lock(meta_ml, meta_idx); 2866 2867 /* 2868 * These two loops are to ensure we skip any markers that 2869 * might be at the tail of the lists due to arc_evict_state(). 2870 */ 2871 2872 for (data_hdr = multilist_sublist_tail(data_mls); data_hdr != NULL; 2873 data_hdr = multilist_sublist_prev(data_mls, data_hdr)) { 2874 if (data_hdr->b_spa != 0) 2875 break; 2876 } 2877 2878 for (meta_hdr = multilist_sublist_tail(meta_mls); meta_hdr != NULL; 2879 meta_hdr = multilist_sublist_prev(meta_mls, meta_hdr)) { 2880 if (meta_hdr->b_spa != 0) 2881 break; 2882 } 2883 2884 if (data_hdr == NULL && meta_hdr == NULL) { 2885 type = ARC_BUFC_DATA; 2886 } else if (data_hdr == NULL) { 2887 ASSERT3P(meta_hdr, !=, NULL); 2888 type = ARC_BUFC_METADATA; 2889 } else if (meta_hdr == NULL) { 2890 ASSERT3P(data_hdr, !=, NULL); 2891 type = ARC_BUFC_DATA; 2892 } else { 2893 ASSERT3P(data_hdr, !=, NULL); 2894 ASSERT3P(meta_hdr, !=, NULL); 2895 2896 /* The headers can't be on the sublist without an L1 header */ 2897 ASSERT(HDR_HAS_L1HDR(data_hdr)); 2898 ASSERT(HDR_HAS_L1HDR(meta_hdr)); 2899 2900 if (data_hdr->b_l1hdr.b_arc_access < 2901 meta_hdr->b_l1hdr.b_arc_access) { 2902 type = ARC_BUFC_DATA; 2903 } else { 2904 type = ARC_BUFC_METADATA; 2905 } 2906 } 2907 2908 multilist_sublist_unlock(meta_mls); 2909 multilist_sublist_unlock(data_mls); 2910 2911 return (type); 2912 } 2913 2914 /* 2915 * Evict buffers from the cache, such that arc_size is capped by arc_c. 2916 */ 2917 static uint64_t 2918 arc_adjust(void) 2919 { 2920 uint64_t total_evicted = 0; 2921 uint64_t bytes; 2922 int64_t target; 2923 2924 /* 2925 * If we're over arc_meta_limit, we want to correct that before 2926 * potentially evicting data buffers below. 2927 */ 2928 total_evicted += arc_adjust_meta(); 2929 2930 /* 2931 * Adjust MRU size 2932 * 2933 * If we're over the target cache size, we want to evict enough 2934 * from the list to get back to our target size. We don't want 2935 * to evict too much from the MRU, such that it drops below 2936 * arc_p. So, if we're over our target cache size more than 2937 * the MRU is over arc_p, we'll evict enough to get back to 2938 * arc_p here, and then evict more from the MFU below. 2939 */ 2940 target = MIN((int64_t)(arc_size - arc_c), 2941 (int64_t)(refcount_count(&arc_anon->arcs_size) + 2942 refcount_count(&arc_mru->arcs_size) + arc_meta_used - arc_p)); 2943 2944 /* 2945 * If we're below arc_meta_min, always prefer to evict data. 2946 * Otherwise, try to satisfy the requested number of bytes to 2947 * evict from the type which contains older buffers; in an 2948 * effort to keep newer buffers in the cache regardless of their 2949 * type. If we cannot satisfy the number of bytes from this 2950 * type, spill over into the next type. 2951 */ 2952 if (arc_adjust_type(arc_mru) == ARC_BUFC_METADATA && 2953 arc_meta_used > arc_meta_min) { 2954 bytes = arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_METADATA); 2955 total_evicted += bytes; 2956 2957 /* 2958 * If we couldn't evict our target number of bytes from 2959 * metadata, we try to get the rest from data. 2960 */ 2961 target -= bytes; 2962 2963 total_evicted += 2964 arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_DATA); 2965 } else { 2966 bytes = arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_DATA); 2967 total_evicted += bytes; 2968 2969 /* 2970 * If we couldn't evict our target number of bytes from 2971 * data, we try to get the rest from metadata. 2972 */ 2973 target -= bytes; 2974 2975 total_evicted += 2976 arc_adjust_impl(arc_mru, 0, target, ARC_BUFC_METADATA); 2977 } 2978 2979 /* 2980 * Adjust MFU size 2981 * 2982 * Now that we've tried to evict enough from the MRU to get its 2983 * size back to arc_p, if we're still above the target cache 2984 * size, we evict the rest from the MFU. 2985 */ 2986 target = arc_size - arc_c; 2987 2988 if (arc_adjust_type(arc_mru) == ARC_BUFC_METADATA && 2989 arc_meta_used > arc_meta_min) { 2990 bytes = arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_METADATA); 2991 total_evicted += bytes; 2992 2993 /* 2994 * If we couldn't evict our target number of bytes from 2995 * metadata, we try to get the rest from data. 2996 */ 2997 target -= bytes; 2998 2999 total_evicted += 3000 arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_DATA); 3001 } else { 3002 bytes = arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_DATA); 3003 total_evicted += bytes; 3004 3005 /* 3006 * If we couldn't evict our target number of bytes from 3007 * data, we try to get the rest from data. 3008 */ 3009 target -= bytes; 3010 3011 total_evicted += 3012 arc_adjust_impl(arc_mfu, 0, target, ARC_BUFC_METADATA); 3013 } 3014 3015 /* 3016 * Adjust ghost lists 3017 * 3018 * In addition to the above, the ARC also defines target values 3019 * for the ghost lists. The sum of the mru list and mru ghost 3020 * list should never exceed the target size of the cache, and 3021 * the sum of the mru list, mfu list, mru ghost list, and mfu 3022 * ghost list should never exceed twice the target size of the 3023 * cache. The following logic enforces these limits on the ghost 3024 * caches, and evicts from them as needed. 3025 */ 3026 target = refcount_count(&arc_mru->arcs_size) + 3027 refcount_count(&arc_mru_ghost->arcs_size) - arc_c; 3028 3029 bytes = arc_adjust_impl(arc_mru_ghost, 0, target, ARC_BUFC_DATA); 3030 total_evicted += bytes; 3031 3032 target -= bytes; 3033 3034 total_evicted += 3035 arc_adjust_impl(arc_mru_ghost, 0, target, ARC_BUFC_METADATA); 3036 3037 /* 3038 * We assume the sum of the mru list and mfu list is less than 3039 * or equal to arc_c (we enforced this above), which means we 3040 * can use the simpler of the two equations below: 3041 * 3042 * mru + mfu + mru ghost + mfu ghost <= 2 * arc_c 3043 * mru ghost + mfu ghost <= arc_c 3044 */ 3045 target = refcount_count(&arc_mru_ghost->arcs_size) + 3046 refcount_count(&arc_mfu_ghost->arcs_size) - arc_c; 3047 3048 bytes = arc_adjust_impl(arc_mfu_ghost, 0, target, ARC_BUFC_DATA); 3049 total_evicted += bytes; 3050 3051 target -= bytes; 3052 3053 total_evicted += 3054 arc_adjust_impl(arc_mfu_ghost, 0, target, ARC_BUFC_METADATA); 3055 3056 return (total_evicted); 3057 } 3058 3059 static void 3060 arc_do_user_evicts(void) 3061 { 3062 mutex_enter(&arc_user_evicts_lock); 3063 while (arc_eviction_list != NULL) { 3064 arc_buf_t *buf = arc_eviction_list; 3065 arc_eviction_list = buf->b_next; 3066 mutex_enter(&buf->b_evict_lock); 3067 buf->b_hdr = NULL; 3068 mutex_exit(&buf->b_evict_lock); 3069 mutex_exit(&arc_user_evicts_lock); 3070 3071 if (buf->b_efunc != NULL) 3072 VERIFY0(buf->b_efunc(buf->b_private)); 3073 3074 buf->b_efunc = NULL; 3075 buf->b_private = NULL; 3076 kmem_cache_free(buf_cache, buf); 3077 mutex_enter(&arc_user_evicts_lock); 3078 } 3079 mutex_exit(&arc_user_evicts_lock); 3080 } 3081 3082 void 3083 arc_flush(spa_t *spa, boolean_t retry) 3084 { 3085 uint64_t guid = 0; 3086 3087 /* 3088 * If retry is TRUE, a spa must not be specified since we have 3089 * no good way to determine if all of a spa's buffers have been 3090 * evicted from an arc state. 3091 */ 3092 ASSERT(!retry || spa == 0); 3093 3094 if (spa != NULL) 3095 guid = spa_load_guid(spa); 3096 3097 (void) arc_flush_state(arc_mru, guid, ARC_BUFC_DATA, retry); 3098 (void) arc_flush_state(arc_mru, guid, ARC_BUFC_METADATA, retry); 3099 3100 (void) arc_flush_state(arc_mfu, guid, ARC_BUFC_DATA, retry); 3101 (void) arc_flush_state(arc_mfu, guid, ARC_BUFC_METADATA, retry); 3102 3103 (void) arc_flush_state(arc_mru_ghost, guid, ARC_BUFC_DATA, retry); 3104 (void) arc_flush_state(arc_mru_ghost, guid, ARC_BUFC_METADATA, retry); 3105 3106 (void) arc_flush_state(arc_mfu_ghost, guid, ARC_BUFC_DATA, retry); 3107 (void) arc_flush_state(arc_mfu_ghost, guid, ARC_BUFC_METADATA, retry); 3108 3109 arc_do_user_evicts(); 3110 ASSERT(spa || arc_eviction_list == NULL); 3111 } 3112 3113 void 3114 arc_shrink(int64_t to_free) 3115 { 3116 if (arc_c > arc_c_min) { 3117 3118 if (arc_c > arc_c_min + to_free) 3119 atomic_add_64(&arc_c, -to_free); 3120 else 3121 arc_c = arc_c_min; 3122 3123 atomic_add_64(&arc_p, -(arc_p >> arc_shrink_shift)); 3124 if (arc_c > arc_size) 3125 arc_c = MAX(arc_size, arc_c_min); 3126 if (arc_p > arc_c) 3127 arc_p = (arc_c >> 1); 3128 ASSERT(arc_c >= arc_c_min); 3129 ASSERT((int64_t)arc_p >= 0); 3130 } 3131 3132 if (arc_size > arc_c) 3133 (void) arc_adjust(); 3134 } 3135 3136 typedef enum free_memory_reason_t { 3137 FMR_UNKNOWN, 3138 FMR_NEEDFREE, 3139 FMR_LOTSFREE, 3140 FMR_SWAPFS_MINFREE, 3141 FMR_PAGES_PP_MAXIMUM, 3142 FMR_HEAP_ARENA, 3143 FMR_ZIO_ARENA, 3144 } free_memory_reason_t; 3145 3146 int64_t last_free_memory; 3147 free_memory_reason_t last_free_reason; 3148 3149 /* 3150 * Additional reserve of pages for pp_reserve. 3151 */ 3152 int64_t arc_pages_pp_reserve = 64; 3153 3154 /* 3155 * Additional reserve of pages for swapfs. 3156 */ 3157 int64_t arc_swapfs_reserve = 64; 3158 3159 /* 3160 * Return the amount of memory that can be consumed before reclaim will be 3161 * needed. Positive if there is sufficient free memory, negative indicates 3162 * the amount of memory that needs to be freed up. 3163 */ 3164 static int64_t 3165 arc_available_memory(void) 3166 { 3167 int64_t lowest = INT64_MAX; 3168 int64_t n; 3169 free_memory_reason_t r = FMR_UNKNOWN; 3170 3171 #ifdef _KERNEL 3172 if (needfree > 0) { 3173 n = PAGESIZE * (-needfree); 3174 if (n < lowest) { 3175 lowest = n; 3176 r = FMR_NEEDFREE; 3177 } 3178 } 3179 3180 /* 3181 * check that we're out of range of the pageout scanner. It starts to 3182 * schedule paging if freemem is less than lotsfree and needfree. 3183 * lotsfree is the high-water mark for pageout, and needfree is the 3184 * number of needed free pages. We add extra pages here to make sure 3185 * the scanner doesn't start up while we're freeing memory. 3186 */ 3187 n = PAGESIZE * (freemem - lotsfree - needfree - desfree); 3188 if (n < lowest) { 3189 lowest = n; 3190 r = FMR_LOTSFREE; 3191 } 3192 3193 /* 3194 * check to make sure that swapfs has enough space so that anon 3195 * reservations can still succeed. anon_resvmem() checks that the 3196 * availrmem is greater than swapfs_minfree, and the number of reserved 3197 * swap pages. We also add a bit of extra here just to prevent 3198 * circumstances from getting really dire. 3199 */ 3200 n = PAGESIZE * (availrmem - swapfs_minfree - swapfs_reserve - 3201 desfree - arc_swapfs_reserve); 3202 if (n < lowest) { 3203 lowest = n; 3204 r = FMR_SWAPFS_MINFREE; 3205 } 3206 3207 3208 /* 3209 * Check that we have enough availrmem that memory locking (e.g., via 3210 * mlock(3C) or memcntl(2)) can still succeed. (pages_pp_maximum 3211 * stores the number of pages that cannot be locked; when availrmem 3212 * drops below pages_pp_maximum, page locking mechanisms such as 3213 * page_pp_lock() will fail.) 3214 */ 3215 n = PAGESIZE * (availrmem - pages_pp_maximum - 3216 arc_pages_pp_reserve); 3217 if (n < lowest) { 3218 lowest = n; 3219 r = FMR_PAGES_PP_MAXIMUM; 3220 } 3221 3222 #if defined(__i386) 3223 /* 3224 * If we're on an i386 platform, it's possible that we'll exhaust the 3225 * kernel heap space before we ever run out of available physical 3226 * memory. Most checks of the size of the heap_area compare against 3227 * tune.t_minarmem, which is the minimum available real memory that we 3228 * can have in the system. However, this is generally fixed at 25 pages 3229 * which is so low that it's useless. In this comparison, we seek to 3230 * calculate the total heap-size, and reclaim if more than 3/4ths of the 3231 * heap is allocated. (Or, in the calculation, if less than 1/4th is 3232 * free) 3233 */ 3234 n = vmem_size(heap_arena, VMEM_FREE) - 3235 (vmem_size(heap_arena, VMEM_FREE | VMEM_ALLOC) >> 2); 3236 if (n < lowest) { 3237 lowest = n; 3238 r = FMR_HEAP_ARENA; 3239 } 3240 #endif 3241 3242 /* 3243 * If zio data pages are being allocated out of a separate heap segment, 3244 * then enforce that the size of available vmem for this arena remains 3245 * above about 1/16th free. 3246 * 3247 * Note: The 1/16th arena free requirement was put in place 3248 * to aggressively evict memory from the arc in order to avoid 3249 * memory fragmentation issues. 3250 */ 3251 if (zio_arena != NULL) { 3252 n = vmem_size(zio_arena, VMEM_FREE) - 3253 (vmem_size(zio_arena, VMEM_ALLOC) >> 4); 3254 if (n < lowest) { 3255 lowest = n; 3256 r = FMR_ZIO_ARENA; 3257 } 3258 } 3259 #else 3260 /* Every 100 calls, free a small amount */ 3261 if (spa_get_random(100) == 0) 3262 lowest = -1024; 3263 #endif 3264 3265 last_free_memory = lowest; 3266 last_free_reason = r; 3267 3268 return (lowest); 3269 } 3270 3271 3272 /* 3273 * Determine if the system is under memory pressure and is asking 3274 * to reclaim memory. A return value of TRUE indicates that the system 3275 * is under memory pressure and that the arc should adjust accordingly. 3276 */ 3277 static boolean_t 3278 arc_reclaim_needed(void) 3279 { 3280 return (arc_available_memory() < 0); 3281 } 3282 3283 static void 3284 arc_kmem_reap_now(void) 3285 { 3286 size_t i; 3287 kmem_cache_t *prev_cache = NULL; 3288 kmem_cache_t *prev_data_cache = NULL; 3289 extern kmem_cache_t *zio_buf_cache[]; 3290 extern kmem_cache_t *zio_data_buf_cache[]; 3291 extern kmem_cache_t *range_seg_cache; 3292 3293 #ifdef _KERNEL 3294 if (arc_meta_used >= arc_meta_limit) { 3295 /* 3296 * We are exceeding our meta-data cache limit. 3297 * Purge some DNLC entries to release holds on meta-data. 3298 */ 3299 dnlc_reduce_cache((void *)(uintptr_t)arc_reduce_dnlc_percent); 3300 } 3301 #if defined(__i386) 3302 /* 3303 * Reclaim unused memory from all kmem caches. 3304 */ 3305 kmem_reap(); 3306 #endif 3307 #endif 3308 3309 for (i = 0; i < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; i++) { 3310 if (zio_buf_cache[i] != prev_cache) { 3311 prev_cache = zio_buf_cache[i]; 3312 kmem_cache_reap_now(zio_buf_cache[i]); 3313 } 3314 if (zio_data_buf_cache[i] != prev_data_cache) { 3315 prev_data_cache = zio_data_buf_cache[i]; 3316 kmem_cache_reap_now(zio_data_buf_cache[i]); 3317 } 3318 } 3319 kmem_cache_reap_now(buf_cache); 3320 kmem_cache_reap_now(hdr_full_cache); 3321 kmem_cache_reap_now(hdr_l2only_cache); 3322 kmem_cache_reap_now(range_seg_cache); 3323 3324 if (zio_arena != NULL) { 3325 /* 3326 * Ask the vmem arena to reclaim unused memory from its 3327 * quantum caches. 3328 */ 3329 vmem_qcache_reap(zio_arena); 3330 } 3331 } 3332 3333 /* 3334 * Threads can block in arc_get_data_buf() waiting for this thread to evict 3335 * enough data and signal them to proceed. When this happens, the threads in 3336 * arc_get_data_buf() are sleeping while holding the hash lock for their 3337 * particular arc header. Thus, we must be careful to never sleep on a 3338 * hash lock in this thread. This is to prevent the following deadlock: 3339 * 3340 * - Thread A sleeps on CV in arc_get_data_buf() holding hash lock "L", 3341 * waiting for the reclaim thread to signal it. 3342 * 3343 * - arc_reclaim_thread() tries to acquire hash lock "L" using mutex_enter, 3344 * fails, and goes to sleep forever. 3345 * 3346 * This possible deadlock is avoided by always acquiring a hash lock 3347 * using mutex_tryenter() from arc_reclaim_thread(). 3348 */ 3349 static void 3350 arc_reclaim_thread(void) 3351 { 3352 clock_t growtime = 0; 3353 callb_cpr_t cpr; 3354 3355 CALLB_CPR_INIT(&cpr, &arc_reclaim_lock, callb_generic_cpr, FTAG); 3356 3357 mutex_enter(&arc_reclaim_lock); 3358 while (!arc_reclaim_thread_exit) { 3359 int64_t free_memory = arc_available_memory(); 3360 uint64_t evicted = 0; 3361 3362 mutex_exit(&arc_reclaim_lock); 3363 3364 if (free_memory < 0) { 3365 3366 arc_no_grow = B_TRUE; 3367 arc_warm = B_TRUE; 3368 3369 /* 3370 * Wait at least zfs_grow_retry (default 60) seconds 3371 * before considering growing. 3372 */ 3373 growtime = ddi_get_lbolt() + (arc_grow_retry * hz); 3374 3375 arc_kmem_reap_now(); 3376 3377 /* 3378 * If we are still low on memory, shrink the ARC 3379 * so that we have arc_shrink_min free space. 3380 */ 3381 free_memory = arc_available_memory(); 3382 3383 int64_t to_free = 3384 (arc_c >> arc_shrink_shift) - free_memory; 3385 if (to_free > 0) { 3386 #ifdef _KERNEL 3387 to_free = MAX(to_free, ptob(needfree)); 3388 #endif 3389 arc_shrink(to_free); 3390 } 3391 } else if (free_memory < arc_c >> arc_no_grow_shift) { 3392 arc_no_grow = B_TRUE; 3393 } else if (ddi_get_lbolt() >= growtime) { 3394 arc_no_grow = B_FALSE; 3395 } 3396 3397 evicted = arc_adjust(); 3398 3399 mutex_enter(&arc_reclaim_lock); 3400 3401 /* 3402 * If evicted is zero, we couldn't evict anything via 3403 * arc_adjust(). This could be due to hash lock 3404 * collisions, but more likely due to the majority of 3405 * arc buffers being unevictable. Therefore, even if 3406 * arc_size is above arc_c, another pass is unlikely to 3407 * be helpful and could potentially cause us to enter an 3408 * infinite loop. 3409 */ 3410 if (arc_size <= arc_c || evicted == 0) { 3411 /* 3412 * We're either no longer overflowing, or we 3413 * can't evict anything more, so we should wake 3414 * up any threads before we go to sleep. 3415 */ 3416 cv_broadcast(&arc_reclaim_waiters_cv); 3417 3418 /* 3419 * Block until signaled, or after one second (we 3420 * might need to perform arc_kmem_reap_now() 3421 * even if we aren't being signalled) 3422 */ 3423 CALLB_CPR_SAFE_BEGIN(&cpr); 3424 (void) cv_timedwait(&arc_reclaim_thread_cv, 3425 &arc_reclaim_lock, ddi_get_lbolt() + hz); 3426 CALLB_CPR_SAFE_END(&cpr, &arc_reclaim_lock); 3427 } 3428 } 3429 3430 arc_reclaim_thread_exit = FALSE; 3431 cv_broadcast(&arc_reclaim_thread_cv); 3432 CALLB_CPR_EXIT(&cpr); /* drops arc_reclaim_lock */ 3433 thread_exit(); 3434 } 3435 3436 static void 3437 arc_user_evicts_thread(void) 3438 { 3439 callb_cpr_t cpr; 3440 3441 CALLB_CPR_INIT(&cpr, &arc_user_evicts_lock, callb_generic_cpr, FTAG); 3442 3443 mutex_enter(&arc_user_evicts_lock); 3444 while (!arc_user_evicts_thread_exit) { 3445 mutex_exit(&arc_user_evicts_lock); 3446 3447 arc_do_user_evicts(); 3448 3449 /* 3450 * This is necessary in order for the mdb ::arc dcmd to 3451 * show up to date information. Since the ::arc command 3452 * does not call the kstat's update function, without 3453 * this call, the command may show stale stats for the 3454 * anon, mru, mru_ghost, mfu, and mfu_ghost lists. Even 3455 * with this change, the data might be up to 1 second 3456 * out of date; but that should suffice. The arc_state_t 3457 * structures can be queried directly if more accurate 3458 * information is needed. 3459 */ 3460 if (arc_ksp != NULL) 3461 arc_ksp->ks_update(arc_ksp, KSTAT_READ); 3462 3463 mutex_enter(&arc_user_evicts_lock); 3464 3465 /* 3466 * Block until signaled, or after one second (we need to 3467 * call the arc's kstat update function regularly). 3468 */ 3469 CALLB_CPR_SAFE_BEGIN(&cpr); 3470 (void) cv_timedwait(&arc_user_evicts_cv, 3471 &arc_user_evicts_lock, ddi_get_lbolt() + hz); 3472 CALLB_CPR_SAFE_END(&cpr, &arc_user_evicts_lock); 3473 } 3474 3475 arc_user_evicts_thread_exit = FALSE; 3476 cv_broadcast(&arc_user_evicts_cv); 3477 CALLB_CPR_EXIT(&cpr); /* drops arc_user_evicts_lock */ 3478 thread_exit(); 3479 } 3480 3481 /* 3482 * Adapt arc info given the number of bytes we are trying to add and 3483 * the state that we are comming from. This function is only called 3484 * when we are adding new content to the cache. 3485 */ 3486 static void 3487 arc_adapt(int bytes, arc_state_t *state) 3488 { 3489 int mult; 3490 uint64_t arc_p_min = (arc_c >> arc_p_min_shift); 3491 int64_t mrug_size = refcount_count(&arc_mru_ghost->arcs_size); 3492 int64_t mfug_size = refcount_count(&arc_mfu_ghost->arcs_size); 3493 3494 if (state == arc_l2c_only) 3495 return; 3496 3497 ASSERT(bytes > 0); 3498 /* 3499 * Adapt the target size of the MRU list: 3500 * - if we just hit in the MRU ghost list, then increase 3501 * the target size of the MRU list. 3502 * - if we just hit in the MFU ghost list, then increase 3503 * the target size of the MFU list by decreasing the 3504 * target size of the MRU list. 3505 */ 3506 if (state == arc_mru_ghost) { 3507 mult = (mrug_size >= mfug_size) ? 1 : (mfug_size / mrug_size); 3508 mult = MIN(mult, 10); /* avoid wild arc_p adjustment */ 3509 3510 arc_p = MIN(arc_c - arc_p_min, arc_p + bytes * mult); 3511 } else if (state == arc_mfu_ghost) { 3512 uint64_t delta; 3513 3514 mult = (mfug_size >= mrug_size) ? 1 : (mrug_size / mfug_size); 3515 mult = MIN(mult, 10); 3516 3517 delta = MIN(bytes * mult, arc_p); 3518 arc_p = MAX(arc_p_min, arc_p - delta); 3519 } 3520 ASSERT((int64_t)arc_p >= 0); 3521 3522 if (arc_reclaim_needed()) { 3523 cv_signal(&arc_reclaim_thread_cv); 3524 return; 3525 } 3526 3527 if (arc_no_grow) 3528 return; 3529 3530 if (arc_c >= arc_c_max) 3531 return; 3532 3533 /* 3534 * If we're within (2 * maxblocksize) bytes of the target 3535 * cache size, increment the target cache size 3536 */ 3537 if (arc_size > arc_c - (2ULL << SPA_MAXBLOCKSHIFT)) { 3538 atomic_add_64(&arc_c, (int64_t)bytes); 3539 if (arc_c > arc_c_max) 3540 arc_c = arc_c_max; 3541 else if (state == arc_anon) 3542 atomic_add_64(&arc_p, (int64_t)bytes); 3543 if (arc_p > arc_c) 3544 arc_p = arc_c; 3545 } 3546 ASSERT((int64_t)arc_p >= 0); 3547 } 3548 3549 /* 3550 * Check if arc_size has grown past our upper threshold, determined by 3551 * zfs_arc_overflow_shift. 3552 */ 3553 static boolean_t 3554 arc_is_overflowing(void) 3555 { 3556 /* Always allow at least one block of overflow */ 3557 uint64_t overflow = MAX(SPA_MAXBLOCKSIZE, 3558 arc_c >> zfs_arc_overflow_shift); 3559 3560 return (arc_size >= arc_c + overflow); 3561 } 3562 3563 /* 3564 * The buffer, supplied as the first argument, needs a data block. If we 3565 * are hitting the hard limit for the cache size, we must sleep, waiting 3566 * for the eviction thread to catch up. If we're past the target size 3567 * but below the hard limit, we'll only signal the reclaim thread and 3568 * continue on. 3569 */ 3570 static void 3571 arc_get_data_buf(arc_buf_t *buf) 3572 { 3573 arc_state_t *state = buf->b_hdr->b_l1hdr.b_state; 3574 uint64_t size = buf->b_hdr->b_size; 3575 arc_buf_contents_t type = arc_buf_type(buf->b_hdr); 3576 3577 arc_adapt(size, state); 3578 3579 /* 3580 * If arc_size is currently overflowing, and has grown past our 3581 * upper limit, we must be adding data faster than the evict 3582 * thread can evict. Thus, to ensure we don't compound the 3583 * problem by adding more data and forcing arc_size to grow even 3584 * further past it's target size, we halt and wait for the 3585 * eviction thread to catch up. 3586 * 3587 * It's also possible that the reclaim thread is unable to evict 3588 * enough buffers to get arc_size below the overflow limit (e.g. 3589 * due to buffers being un-evictable, or hash lock collisions). 3590 * In this case, we want to proceed regardless if we're 3591 * overflowing; thus we don't use a while loop here. 3592 */ 3593 if (arc_is_overflowing()) { 3594 mutex_enter(&arc_reclaim_lock); 3595 3596 /* 3597 * Now that we've acquired the lock, we may no longer be 3598 * over the overflow limit, lets check. 3599 * 3600 * We're ignoring the case of spurious wake ups. If that 3601 * were to happen, it'd let this thread consume an ARC 3602 * buffer before it should have (i.e. before we're under 3603 * the overflow limit and were signalled by the reclaim 3604 * thread). As long as that is a rare occurrence, it 3605 * shouldn't cause any harm. 3606 */ 3607 if (arc_is_overflowing()) { 3608 cv_signal(&arc_reclaim_thread_cv); 3609 cv_wait(&arc_reclaim_waiters_cv, &arc_reclaim_lock); 3610 } 3611 3612 mutex_exit(&arc_reclaim_lock); 3613 } 3614 3615 if (type == ARC_BUFC_METADATA) { 3616 buf->b_data = zio_buf_alloc(size); 3617 arc_space_consume(size, ARC_SPACE_META); 3618 } else { 3619 ASSERT(type == ARC_BUFC_DATA); 3620 buf->b_data = zio_data_buf_alloc(size); 3621 arc_space_consume(size, ARC_SPACE_DATA); 3622 } 3623 3624 /* 3625 * Update the state size. Note that ghost states have a 3626 * "ghost size" and so don't need to be updated. 3627 */ 3628 if (!GHOST_STATE(buf->b_hdr->b_l1hdr.b_state)) { 3629 arc_buf_hdr_t *hdr = buf->b_hdr; 3630 arc_state_t *state = hdr->b_l1hdr.b_state; 3631 3632 (void) refcount_add_many(&state->arcs_size, size, buf); 3633 3634 /* 3635 * If this is reached via arc_read, the link is 3636 * protected by the hash lock. If reached via 3637 * arc_buf_alloc, the header should not be accessed by 3638 * any other thread. And, if reached via arc_read_done, 3639 * the hash lock will protect it if it's found in the 3640 * hash table; otherwise no other thread should be 3641 * trying to [add|remove]_reference it. 3642 */ 3643 if (multilist_link_active(&hdr->b_l1hdr.b_arc_node)) { 3644 ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); 3645 atomic_add_64(&hdr->b_l1hdr.b_state->arcs_lsize[type], 3646 size); 3647 } 3648 /* 3649 * If we are growing the cache, and we are adding anonymous 3650 * data, and we have outgrown arc_p, update arc_p 3651 */ 3652 if (arc_size < arc_c && hdr->b_l1hdr.b_state == arc_anon && 3653 (refcount_count(&arc_anon->arcs_size) + 3654 refcount_count(&arc_mru->arcs_size) > arc_p)) 3655 arc_p = MIN(arc_c, arc_p + size); 3656 } 3657 } 3658 3659 /* 3660 * This routine is called whenever a buffer is accessed. 3661 * NOTE: the hash lock is dropped in this function. 3662 */ 3663 static void 3664 arc_access(arc_buf_hdr_t *hdr, kmutex_t *hash_lock) 3665 { 3666 clock_t now; 3667 3668 ASSERT(MUTEX_HELD(hash_lock)); 3669 ASSERT(HDR_HAS_L1HDR(hdr)); 3670 3671 if (hdr->b_l1hdr.b_state == arc_anon) { 3672 /* 3673 * This buffer is not in the cache, and does not 3674 * appear in our "ghost" list. Add the new buffer 3675 * to the MRU state. 3676 */ 3677 3678 ASSERT0(hdr->b_l1hdr.b_arc_access); 3679 hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); 3680 DTRACE_PROBE1(new_state__mru, arc_buf_hdr_t *, hdr); 3681 arc_change_state(arc_mru, hdr, hash_lock); 3682 3683 } else if (hdr->b_l1hdr.b_state == arc_mru) { 3684 now = ddi_get_lbolt(); 3685 3686 /* 3687 * If this buffer is here because of a prefetch, then either: 3688 * - clear the flag if this is a "referencing" read 3689 * (any subsequent access will bump this into the MFU state). 3690 * or 3691 * - move the buffer to the head of the list if this is 3692 * another prefetch (to make it less likely to be evicted). 3693 */ 3694 if (HDR_PREFETCH(hdr)) { 3695 if (refcount_count(&hdr->b_l1hdr.b_refcnt) == 0) { 3696 /* link protected by hash lock */ 3697 ASSERT(multilist_link_active( 3698 &hdr->b_l1hdr.b_arc_node)); 3699 } else { 3700 hdr->b_flags &= ~ARC_FLAG_PREFETCH; 3701 ARCSTAT_BUMP(arcstat_mru_hits); 3702 } 3703 hdr->b_l1hdr.b_arc_access = now; 3704 return; 3705 } 3706 3707 /* 3708 * This buffer has been "accessed" only once so far, 3709 * but it is still in the cache. Move it to the MFU 3710 * state. 3711 */ 3712 if (now > hdr->b_l1hdr.b_arc_access + ARC_MINTIME) { 3713 /* 3714 * More than 125ms have passed since we 3715 * instantiated this buffer. Move it to the 3716 * most frequently used state. 3717 */ 3718 hdr->b_l1hdr.b_arc_access = now; 3719 DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); 3720 arc_change_state(arc_mfu, hdr, hash_lock); 3721 } 3722 ARCSTAT_BUMP(arcstat_mru_hits); 3723 } else if (hdr->b_l1hdr.b_state == arc_mru_ghost) { 3724 arc_state_t *new_state; 3725 /* 3726 * This buffer has been "accessed" recently, but 3727 * was evicted from the cache. Move it to the 3728 * MFU state. 3729 */ 3730 3731 if (HDR_PREFETCH(hdr)) { 3732 new_state = arc_mru; 3733 if (refcount_count(&hdr->b_l1hdr.b_refcnt) > 0) 3734 hdr->b_flags &= ~ARC_FLAG_PREFETCH; 3735 DTRACE_PROBE1(new_state__mru, arc_buf_hdr_t *, hdr); 3736 } else { 3737 new_state = arc_mfu; 3738 DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); 3739 } 3740 3741 hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); 3742 arc_change_state(new_state, hdr, hash_lock); 3743 3744 ARCSTAT_BUMP(arcstat_mru_ghost_hits); 3745 } else if (hdr->b_l1hdr.b_state == arc_mfu) { 3746 /* 3747 * This buffer has been accessed more than once and is 3748 * still in the cache. Keep it in the MFU state. 3749 * 3750 * NOTE: an add_reference() that occurred when we did 3751 * the arc_read() will have kicked this off the list. 3752 * If it was a prefetch, we will explicitly move it to 3753 * the head of the list now. 3754 */ 3755 if ((HDR_PREFETCH(hdr)) != 0) { 3756 ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); 3757 /* link protected by hash_lock */ 3758 ASSERT(multilist_link_active(&hdr->b_l1hdr.b_arc_node)); 3759 } 3760 ARCSTAT_BUMP(arcstat_mfu_hits); 3761 hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); 3762 } else if (hdr->b_l1hdr.b_state == arc_mfu_ghost) { 3763 arc_state_t *new_state = arc_mfu; 3764 /* 3765 * This buffer has been accessed more than once but has 3766 * been evicted from the cache. Move it back to the 3767 * MFU state. 3768 */ 3769 3770 if (HDR_PREFETCH(hdr)) { 3771 /* 3772 * This is a prefetch access... 3773 * move this block back to the MRU state. 3774 */ 3775 ASSERT0(refcount_count(&hdr->b_l1hdr.b_refcnt)); 3776 new_state = arc_mru; 3777 } 3778 3779 hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); 3780 DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); 3781 arc_change_state(new_state, hdr, hash_lock); 3782 3783 ARCSTAT_BUMP(arcstat_mfu_ghost_hits); 3784 } else if (hdr->b_l1hdr.b_state == arc_l2c_only) { 3785 /* 3786 * This buffer is on the 2nd Level ARC. 3787 */ 3788 3789 hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); 3790 DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); 3791 arc_change_state(arc_mfu, hdr, hash_lock); 3792 } else { 3793 ASSERT(!"invalid arc state"); 3794 } 3795 } 3796 3797 /* a generic arc_done_func_t which you can use */ 3798 /* ARGSUSED */ 3799 void 3800 arc_bcopy_func(zio_t *zio, arc_buf_t *buf, void *arg) 3801 { 3802 if (zio == NULL || zio->io_error == 0) 3803 bcopy(buf->b_data, arg, buf->b_hdr->b_size); 3804 VERIFY(arc_buf_remove_ref(buf, arg)); 3805 } 3806 3807 /* a generic arc_done_func_t */ 3808 void 3809 arc_getbuf_func(zio_t *zio, arc_buf_t *buf, void *arg) 3810 { 3811 arc_buf_t **bufp = arg; 3812 if (zio && zio->io_error) { 3813 VERIFY(arc_buf_remove_ref(buf, arg)); 3814 *bufp = NULL; 3815 } else { 3816 *bufp = buf; 3817 ASSERT(buf->b_data); 3818 } 3819 } 3820 3821 static void 3822 arc_read_done(zio_t *zio) 3823 { 3824 arc_buf_hdr_t *hdr; 3825 arc_buf_t *buf; 3826 arc_buf_t *abuf; /* buffer we're assigning to callback */ 3827 kmutex_t *hash_lock = NULL; 3828 arc_callback_t *callback_list, *acb; 3829 int freeable = FALSE; 3830 3831 buf = zio->io_private; 3832 hdr = buf->b_hdr; 3833 3834 /* 3835 * The hdr was inserted into hash-table and removed from lists 3836 * prior to starting I/O. We should find this header, since 3837 * it's in the hash table, and it should be legit since it's 3838 * not possible to evict it during the I/O. The only possible 3839 * reason for it not to be found is if we were freed during the 3840 * read. 3841 */ 3842 if (HDR_IN_HASH_TABLE(hdr)) { 3843 ASSERT3U(hdr->b_birth, ==, BP_PHYSICAL_BIRTH(zio->io_bp)); 3844 ASSERT3U(hdr->b_dva.dva_word[0], ==, 3845 BP_IDENTITY(zio->io_bp)->dva_word[0]); 3846 ASSERT3U(hdr->b_dva.dva_word[1], ==, 3847 BP_IDENTITY(zio->io_bp)->dva_word[1]); 3848 3849 arc_buf_hdr_t *found = buf_hash_find(hdr->b_spa, zio->io_bp, 3850 &hash_lock); 3851 3852 ASSERT((found == NULL && HDR_FREED_IN_READ(hdr) && 3853 hash_lock == NULL) || 3854 (found == hdr && 3855 DVA_EQUAL(&hdr->b_dva, BP_IDENTITY(zio->io_bp))) || 3856 (found == hdr && HDR_L2_READING(hdr))); 3857 } 3858 3859 hdr->b_flags &= ~ARC_FLAG_L2_EVICTED; 3860 if (l2arc_noprefetch && HDR_PREFETCH(hdr)) 3861 hdr->b_flags &= ~ARC_FLAG_L2CACHE; 3862 3863 /* byteswap if necessary */ 3864 callback_list = hdr->b_l1hdr.b_acb; 3865 ASSERT(callback_list != NULL); 3866 if (BP_SHOULD_BYTESWAP(zio->io_bp) && zio->io_error == 0) { 3867 dmu_object_byteswap_t bswap = 3868 DMU_OT_BYTESWAP(BP_GET_TYPE(zio->io_bp)); 3869 arc_byteswap_func_t *func = BP_GET_LEVEL(zio->io_bp) > 0 ? 3870 byteswap_uint64_array : 3871 dmu_ot_byteswap[bswap].ob_func; 3872 func(buf->b_data, hdr->b_size); 3873 } 3874 3875 arc_cksum_compute(buf, B_FALSE); 3876 arc_buf_watch(buf); 3877 3878 if (hash_lock && zio->io_error == 0 && 3879 hdr->b_l1hdr.b_state == arc_anon) { 3880 /* 3881 * Only call arc_access on anonymous buffers. This is because 3882 * if we've issued an I/O for an evicted buffer, we've already 3883 * called arc_access (to prevent any simultaneous readers from 3884 * getting confused). 3885 */ 3886 arc_access(hdr, hash_lock); 3887 } 3888 3889 /* create copies of the data buffer for the callers */ 3890 abuf = buf; 3891 for (acb = callback_list; acb; acb = acb->acb_next) { 3892 if (acb->acb_done) { 3893 if (abuf == NULL) { 3894 ARCSTAT_BUMP(arcstat_duplicate_reads); 3895 abuf = arc_buf_clone(buf); 3896 } 3897 acb->acb_buf = abuf; 3898 abuf = NULL; 3899 } 3900 } 3901 hdr->b_l1hdr.b_acb = NULL; 3902 hdr->b_flags &= ~ARC_FLAG_IO_IN_PROGRESS; 3903 ASSERT(!HDR_BUF_AVAILABLE(hdr)); 3904 if (abuf == buf) { 3905 ASSERT(buf->b_efunc == NULL); 3906 ASSERT(hdr->b_l1hdr.b_datacnt == 1); 3907 hdr->b_flags |= ARC_FLAG_BUF_AVAILABLE; 3908 } 3909 3910 ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt) || 3911 callback_list != NULL); 3912 3913 if (zio->io_error != 0) { 3914 hdr->b_flags |= ARC_FLAG_IO_ERROR; 3915 if (hdr->b_l1hdr.b_state != arc_anon) 3916 arc_change_state(arc_anon, hdr, hash_lock); 3917 if (HDR_IN_HASH_TABLE(hdr)) 3918 buf_hash_remove(hdr); 3919 freeable = refcount_is_zero(&hdr->b_l1hdr.b_refcnt); 3920 } 3921 3922 /* 3923 * Broadcast before we drop the hash_lock to avoid the possibility 3924 * that the hdr (and hence the cv) might be freed before we get to 3925 * the cv_broadcast(). 3926 */ 3927 cv_broadcast(&hdr->b_l1hdr.b_cv); 3928 3929 if (hash_lock != NULL) { 3930 mutex_exit(hash_lock); 3931 } else { 3932 /* 3933 * This block was freed while we waited for the read to 3934 * complete. It has been removed from the hash table and 3935 * moved to the anonymous state (so that it won't show up 3936 * in the cache). 3937 */ 3938 ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon); 3939 freeable = refcount_is_zero(&hdr->b_l1hdr.b_refcnt); 3940 } 3941 3942 /* execute each callback and free its structure */ 3943 while ((acb = callback_list) != NULL) { 3944 if (acb->acb_done) 3945 acb->acb_done(zio, acb->acb_buf, acb->acb_private); 3946 3947 if (acb->acb_zio_dummy != NULL) { 3948 acb->acb_zio_dummy->io_error = zio->io_error; 3949 zio_nowait(acb->acb_zio_dummy); 3950 } 3951 3952 callback_list = acb->acb_next; 3953 kmem_free(acb, sizeof (arc_callback_t)); 3954 } 3955 3956 if (freeable) 3957 arc_hdr_destroy(hdr); 3958 } 3959 3960 /* 3961 * "Read" the block at the specified DVA (in bp) via the 3962 * cache. If the block is found in the cache, invoke the provided 3963 * callback immediately and return. Note that the `zio' parameter 3964 * in the callback will be NULL in this case, since no IO was 3965 * required. If the block is not in the cache pass the read request 3966 * on to the spa with a substitute callback function, so that the 3967 * requested block will be added to the cache. 3968 * 3969 * If a read request arrives for a block that has a read in-progress, 3970 * either wait for the in-progress read to complete (and return the 3971 * results); or, if this is a read with a "done" func, add a record 3972 * to the read to invoke the "done" func when the read completes, 3973 * and return; or just return. 3974 * 3975 * arc_read_done() will invoke all the requested "done" functions 3976 * for readers of this block. 3977 */ 3978 int 3979 arc_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, arc_done_func_t *done, 3980 void *private, zio_priority_t priority, int zio_flags, 3981 arc_flags_t *arc_flags, const zbookmark_phys_t *zb) 3982 { 3983 arc_buf_hdr_t *hdr = NULL; 3984 arc_buf_t *buf = NULL; 3985 kmutex_t *hash_lock = NULL; 3986 zio_t *rzio; 3987 uint64_t guid = spa_load_guid(spa); 3988 3989 ASSERT(!BP_IS_EMBEDDED(bp) || 3990 BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA); 3991 3992 top: 3993 if (!BP_IS_EMBEDDED(bp)) { 3994 /* 3995 * Embedded BP's have no DVA and require no I/O to "read". 3996 * Create an anonymous arc buf to back it. 3997 */ 3998 hdr = buf_hash_find(guid, bp, &hash_lock); 3999 } 4000 4001 if (hdr != NULL && HDR_HAS_L1HDR(hdr) && hdr->b_l1hdr.b_datacnt > 0) { 4002 4003 *arc_flags |= ARC_FLAG_CACHED; 4004 4005 if (HDR_IO_IN_PROGRESS(hdr)) { 4006 4007 if (*arc_flags & ARC_FLAG_WAIT) { 4008 cv_wait(&hdr->b_l1hdr.b_cv, hash_lock); 4009 mutex_exit(hash_lock); 4010 goto top; 4011 } 4012 ASSERT(*arc_flags & ARC_FLAG_NOWAIT); 4013 4014 if (done) { 4015 arc_callback_t *acb = NULL; 4016 4017 acb = kmem_zalloc(sizeof (arc_callback_t), 4018 KM_SLEEP); 4019 acb->acb_done = done; 4020 acb->acb_private = private; 4021 if (pio != NULL) 4022 acb->acb_zio_dummy = zio_null(pio, 4023 spa, NULL, NULL, NULL, zio_flags); 4024 4025 ASSERT(acb->acb_done != NULL); 4026 acb->acb_next = hdr->b_l1hdr.b_acb; 4027 hdr->b_l1hdr.b_acb = acb; 4028 add_reference(hdr, hash_lock, private); 4029 mutex_exit(hash_lock); 4030 return (0); 4031 } 4032 mutex_exit(hash_lock); 4033 return (0); 4034 } 4035 4036 ASSERT(hdr->b_l1hdr.b_state == arc_mru || 4037 hdr->b_l1hdr.b_state == arc_mfu); 4038 4039 if (done) { 4040 add_reference(hdr, hash_lock, private); 4041 /* 4042 * If this block is already in use, create a new 4043 * copy of the data so that we will be guaranteed 4044 * that arc_release() will always succeed. 4045 */ 4046 buf = hdr->b_l1hdr.b_buf; 4047 ASSERT(buf); 4048 ASSERT(buf->b_data); 4049 if (HDR_BUF_AVAILABLE(hdr)) { 4050 ASSERT(buf->b_efunc == NULL); 4051 hdr->b_flags &= ~ARC_FLAG_BUF_AVAILABLE; 4052 } else { 4053 buf = arc_buf_clone(buf); 4054 } 4055 4056 } else if (*arc_flags & ARC_FLAG_PREFETCH && 4057 refcount_count(&hdr->b_l1hdr.b_refcnt) == 0) { 4058 hdr->b_flags |= ARC_FLAG_PREFETCH; 4059 } 4060 DTRACE_PROBE1(arc__hit, arc_buf_hdr_t *, hdr); 4061 arc_access(hdr, hash_lock); 4062 if (*arc_flags & ARC_FLAG_L2CACHE) 4063 hdr->b_flags |= ARC_FLAG_L2CACHE; 4064 if (*arc_flags & ARC_FLAG_L2COMPRESS) 4065 hdr->b_flags |= ARC_FLAG_L2COMPRESS; 4066 mutex_exit(hash_lock); 4067 ARCSTAT_BUMP(arcstat_hits); 4068 ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr), 4069 demand, prefetch, !HDR_ISTYPE_METADATA(hdr), 4070 data, metadata, hits); 4071 4072 if (done) 4073 done(NULL, buf, private); 4074 } else { 4075 uint64_t size = BP_GET_LSIZE(bp); 4076 arc_callback_t *acb; 4077 vdev_t *vd = NULL; 4078 uint64_t addr = 0; 4079 boolean_t devw = B_FALSE; 4080 enum zio_compress b_compress = ZIO_COMPRESS_OFF; 4081 int32_t b_asize = 0; 4082 4083 if (hdr == NULL) { 4084 /* this block is not in the cache */ 4085 arc_buf_hdr_t *exists = NULL; 4086 arc_buf_contents_t type = BP_GET_BUFC_TYPE(bp); 4087 buf = arc_buf_alloc(spa, size, private, type); 4088 hdr = buf->b_hdr; 4089 if (!BP_IS_EMBEDDED(bp)) { 4090 hdr->b_dva = *BP_IDENTITY(bp); 4091 hdr->b_birth = BP_PHYSICAL_BIRTH(bp); 4092 exists = buf_hash_insert(hdr, &hash_lock); 4093 } 4094 if (exists != NULL) { 4095 /* somebody beat us to the hash insert */ 4096 mutex_exit(hash_lock); 4097 buf_discard_identity(hdr); 4098 (void) arc_buf_remove_ref(buf, private); 4099 goto top; /* restart the IO request */ 4100 } 4101 4102 /* if this is a prefetch, we don't have a reference */ 4103 if (*arc_flags & ARC_FLAG_PREFETCH) { 4104 (void) remove_reference(hdr, hash_lock, 4105 private); 4106 hdr->b_flags |= ARC_FLAG_PREFETCH; 4107 } 4108 if (*arc_flags & ARC_FLAG_L2CACHE) 4109 hdr->b_flags |= ARC_FLAG_L2CACHE; 4110 if (*arc_flags & ARC_FLAG_L2COMPRESS) 4111 hdr->b_flags |= ARC_FLAG_L2COMPRESS; 4112 if (BP_GET_LEVEL(bp) > 0) 4113 hdr->b_flags |= ARC_FLAG_INDIRECT; 4114 } else { 4115 /* 4116 * This block is in the ghost cache. If it was L2-only 4117 * (and thus didn't have an L1 hdr), we realloc the 4118 * header to add an L1 hdr. 4119 */ 4120 if (!HDR_HAS_L1HDR(hdr)) { 4121 hdr = arc_hdr_realloc(hdr, hdr_l2only_cache, 4122 hdr_full_cache); 4123 } 4124 4125 ASSERT(GHOST_STATE(hdr->b_l1hdr.b_state)); 4126 ASSERT(!HDR_IO_IN_PROGRESS(hdr)); 4127 ASSERT(refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); 4128 ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); 4129 4130 /* if this is a prefetch, we don't have a reference */ 4131 if (*arc_flags & ARC_FLAG_PREFETCH) 4132 hdr->b_flags |= ARC_FLAG_PREFETCH; 4133 else 4134 add_reference(hdr, hash_lock, private); 4135 if (*arc_flags & ARC_FLAG_L2CACHE) 4136 hdr->b_flags |= ARC_FLAG_L2CACHE; 4137 if (*arc_flags & ARC_FLAG_L2COMPRESS) 4138 hdr->b_flags |= ARC_FLAG_L2COMPRESS; 4139 buf = kmem_cache_alloc(buf_cache, KM_PUSHPAGE); 4140 buf->b_hdr = hdr; 4141 buf->b_data = NULL; 4142 buf->b_efunc = NULL; 4143 buf->b_private = NULL; 4144 buf->b_next = NULL; 4145 hdr->b_l1hdr.b_buf = buf; 4146 ASSERT0(hdr->b_l1hdr.b_datacnt); 4147 hdr->b_l1hdr.b_datacnt = 1; 4148 arc_get_data_buf(buf); 4149 arc_access(hdr, hash_lock); 4150 } 4151 4152 ASSERT(!GHOST_STATE(hdr->b_l1hdr.b_state)); 4153 4154 acb = kmem_zalloc(sizeof (arc_callback_t), KM_SLEEP); 4155 acb->acb_done = done; 4156 acb->acb_private = private; 4157 4158 ASSERT(hdr->b_l1hdr.b_acb == NULL); 4159 hdr->b_l1hdr.b_acb = acb; 4160 hdr->b_flags |= ARC_FLAG_IO_IN_PROGRESS; 4161 4162 if (HDR_HAS_L2HDR(hdr) && 4163 (vd = hdr->b_l2hdr.b_dev->l2ad_vdev) != NULL) { 4164 devw = hdr->b_l2hdr.b_dev->l2ad_writing; 4165 addr = hdr->b_l2hdr.b_daddr; 4166 b_compress = HDR_GET_COMPRESS(hdr); 4167 b_asize = hdr->b_l2hdr.b_asize; 4168 /* 4169 * Lock out device removal. 4170 */ 4171 if (vdev_is_dead(vd) || 4172 !spa_config_tryenter(spa, SCL_L2ARC, vd, RW_READER)) 4173 vd = NULL; 4174 } 4175 4176 if (hash_lock != NULL) 4177 mutex_exit(hash_lock); 4178 4179 /* 4180 * At this point, we have a level 1 cache miss. Try again in 4181 * L2ARC if possible. 4182 */ 4183 ASSERT3U(hdr->b_size, ==, size); 4184 DTRACE_PROBE4(arc__miss, arc_buf_hdr_t *, hdr, blkptr_t *, bp, 4185 uint64_t, size, zbookmark_phys_t *, zb); 4186 ARCSTAT_BUMP(arcstat_misses); 4187 ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr), 4188 demand, prefetch, !HDR_ISTYPE_METADATA(hdr), 4189 data, metadata, misses); 4190 4191 if (vd != NULL && l2arc_ndev != 0 && !(l2arc_norw && devw)) { 4192 /* 4193 * Read from the L2ARC if the following are true: 4194 * 1. The L2ARC vdev was previously cached. 4195 * 2. This buffer still has L2ARC metadata. 4196 * 3. This buffer isn't currently writing to the L2ARC. 4197 * 4. The L2ARC entry wasn't evicted, which may 4198 * also have invalidated the vdev. 4199 * 5. This isn't prefetch and l2arc_noprefetch is set. 4200 */ 4201 if (HDR_HAS_L2HDR(hdr) && 4202 !HDR_L2_WRITING(hdr) && !HDR_L2_EVICTED(hdr) && 4203 !(l2arc_noprefetch && HDR_PREFETCH(hdr))) { 4204 l2arc_read_callback_t *cb; 4205 4206 DTRACE_PROBE1(l2arc__hit, arc_buf_hdr_t *, hdr); 4207 ARCSTAT_BUMP(arcstat_l2_hits); 4208 4209 cb = kmem_zalloc(sizeof (l2arc_read_callback_t), 4210 KM_SLEEP); 4211 cb->l2rcb_buf = buf; 4212 cb->l2rcb_spa = spa; 4213 cb->l2rcb_bp = *bp; 4214 cb->l2rcb_zb = *zb; 4215 cb->l2rcb_flags = zio_flags; 4216 cb->l2rcb_compress = b_compress; 4217 4218 ASSERT(addr >= VDEV_LABEL_START_SIZE && 4219 addr + size < vd->vdev_psize - 4220 VDEV_LABEL_END_SIZE); 4221 4222 /* 4223 * l2arc read. The SCL_L2ARC lock will be 4224 * released by l2arc_read_done(). 4225 * Issue a null zio if the underlying buffer 4226 * was squashed to zero size by compression. 4227 */ 4228 if (b_compress == ZIO_COMPRESS_EMPTY) { 4229 rzio = zio_null(pio, spa, vd, 4230 l2arc_read_done, cb, 4231 zio_flags | ZIO_FLAG_DONT_CACHE | 4232 ZIO_FLAG_CANFAIL | 4233 ZIO_FLAG_DONT_PROPAGATE | 4234 ZIO_FLAG_DONT_RETRY); 4235 } else { 4236 rzio = zio_read_phys(pio, vd, addr, 4237 b_asize, buf->b_data, 4238 ZIO_CHECKSUM_OFF, 4239 l2arc_read_done, cb, priority, 4240 zio_flags | ZIO_FLAG_DONT_CACHE | 4241 ZIO_FLAG_CANFAIL | 4242 ZIO_FLAG_DONT_PROPAGATE | 4243 ZIO_FLAG_DONT_RETRY, B_FALSE); 4244 } 4245 DTRACE_PROBE2(l2arc__read, vdev_t *, vd, 4246 zio_t *, rzio); 4247 ARCSTAT_INCR(arcstat_l2_read_bytes, b_asize); 4248 4249 if (*arc_flags & ARC_FLAG_NOWAIT) { 4250 zio_nowait(rzio); 4251 return (0); 4252 } 4253 4254 ASSERT(*arc_flags & ARC_FLAG_WAIT); 4255 if (zio_wait(rzio) == 0) 4256 return (0); 4257 4258 /* l2arc read error; goto zio_read() */ 4259 } else { 4260 DTRACE_PROBE1(l2arc__miss, 4261 arc_buf_hdr_t *, hdr); 4262 ARCSTAT_BUMP(arcstat_l2_misses); 4263 if (HDR_L2_WRITING(hdr)) 4264 ARCSTAT_BUMP(arcstat_l2_rw_clash); 4265 spa_config_exit(spa, SCL_L2ARC, vd); 4266 } 4267 } else { 4268 if (vd != NULL) 4269 spa_config_exit(spa, SCL_L2ARC, vd); 4270 if (l2arc_ndev != 0) { 4271 DTRACE_PROBE1(l2arc__miss, 4272 arc_buf_hdr_t *, hdr); 4273 ARCSTAT_BUMP(arcstat_l2_misses); 4274 } 4275 } 4276 4277 rzio = zio_read(pio, spa, bp, buf->b_data, size, 4278 arc_read_done, buf, priority, zio_flags, zb); 4279 4280 if (*arc_flags & ARC_FLAG_WAIT) 4281 return (zio_wait(rzio)); 4282 4283 ASSERT(*arc_flags & ARC_FLAG_NOWAIT); 4284 zio_nowait(rzio); 4285 } 4286 return (0); 4287 } 4288 4289 void 4290 arc_set_callback(arc_buf_t *buf, arc_evict_func_t *func, void *private) 4291 { 4292 ASSERT(buf->b_hdr != NULL); 4293 ASSERT(buf->b_hdr->b_l1hdr.b_state != arc_anon); 4294 ASSERT(!refcount_is_zero(&buf->b_hdr->b_l1hdr.b_refcnt) || 4295 func == NULL); 4296 ASSERT(buf->b_efunc == NULL); 4297 ASSERT(!HDR_BUF_AVAILABLE(buf->b_hdr)); 4298 4299 buf->b_efunc = func; 4300 buf->b_private = private; 4301 } 4302 4303 /* 4304 * Notify the arc that a block was freed, and thus will never be used again. 4305 */ 4306 void 4307 arc_freed(spa_t *spa, const blkptr_t *bp) 4308 { 4309 arc_buf_hdr_t *hdr; 4310 kmutex_t *hash_lock; 4311 uint64_t guid = spa_load_guid(spa); 4312 4313 ASSERT(!BP_IS_EMBEDDED(bp)); 4314 4315 hdr = buf_hash_find(guid, bp, &hash_lock); 4316 if (hdr == NULL) 4317 return; 4318 if (HDR_BUF_AVAILABLE(hdr)) { 4319 arc_buf_t *buf = hdr->b_l1hdr.b_buf; 4320 add_reference(hdr, hash_lock, FTAG); 4321 hdr->b_flags &= ~ARC_FLAG_BUF_AVAILABLE; 4322 mutex_exit(hash_lock); 4323 4324 arc_release(buf, FTAG); 4325 (void) arc_buf_remove_ref(buf, FTAG); 4326 } else { 4327 mutex_exit(hash_lock); 4328 } 4329 4330 } 4331 4332 /* 4333 * Clear the user eviction callback set by arc_set_callback(), first calling 4334 * it if it exists. Because the presence of a callback keeps an arc_buf cached 4335 * clearing the callback may result in the arc_buf being destroyed. However, 4336 * it will not result in the *last* arc_buf being destroyed, hence the data 4337 * will remain cached in the ARC. We make a copy of the arc buffer here so 4338 * that we can process the callback without holding any locks. 4339 * 4340 * It's possible that the callback is already in the process of being cleared 4341 * by another thread. In this case we can not clear the callback. 4342 * 4343 * Returns B_TRUE if the callback was successfully called and cleared. 4344 */ 4345 boolean_t 4346 arc_clear_callback(arc_buf_t *buf) 4347 { 4348 arc_buf_hdr_t *hdr; 4349 kmutex_t *hash_lock; 4350 arc_evict_func_t *efunc = buf->b_efunc; 4351 void *private = buf->b_private; 4352 4353 mutex_enter(&buf->b_evict_lock); 4354 hdr = buf->b_hdr; 4355 if (hdr == NULL) { 4356 /* 4357 * We are in arc_do_user_evicts(). 4358 */ 4359 ASSERT(buf->b_data == NULL); 4360 mutex_exit(&buf->b_evict_lock); 4361 return (B_FALSE); 4362 } else if (buf->b_data == NULL) { 4363 /* 4364 * We are on the eviction list; process this buffer now 4365 * but let arc_do_user_evicts() do the reaping. 4366 */ 4367 buf->b_efunc = NULL; 4368 mutex_exit(&buf->b_evict_lock); 4369 VERIFY0(efunc(private)); 4370 return (B_TRUE); 4371 } 4372 hash_lock = HDR_LOCK(hdr); 4373 mutex_enter(hash_lock); 4374 hdr = buf->b_hdr; 4375 ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); 4376 4377 ASSERT3U(refcount_count(&hdr->b_l1hdr.b_refcnt), <, 4378 hdr->b_l1hdr.b_datacnt); 4379 ASSERT(hdr->b_l1hdr.b_state == arc_mru || 4380 hdr->b_l1hdr.b_state == arc_mfu); 4381 4382 buf->b_efunc = NULL; 4383 buf->b_private = NULL; 4384 4385 if (hdr->b_l1hdr.b_datacnt > 1) { 4386 mutex_exit(&buf->b_evict_lock); 4387 arc_buf_destroy(buf, TRUE); 4388 } else { 4389 ASSERT(buf == hdr->b_l1hdr.b_buf); 4390 hdr->b_flags |= ARC_FLAG_BUF_AVAILABLE; 4391 mutex_exit(&buf->b_evict_lock); 4392 } 4393 4394 mutex_exit(hash_lock); 4395 VERIFY0(efunc(private)); 4396 return (B_TRUE); 4397 } 4398 4399 /* 4400 * Release this buffer from the cache, making it an anonymous buffer. This 4401 * must be done after a read and prior to modifying the buffer contents. 4402 * If the buffer has more than one reference, we must make 4403 * a new hdr for the buffer. 4404 */ 4405 void 4406 arc_release(arc_buf_t *buf, void *tag) 4407 { 4408 arc_buf_hdr_t *hdr = buf->b_hdr; 4409 4410 /* 4411 * It would be nice to assert that if it's DMU metadata (level > 4412 * 0 || it's the dnode file), then it must be syncing context. 4413 * But we don't know that information at this level. 4414 */ 4415 4416 mutex_enter(&buf->b_evict_lock); 4417 4418 ASSERT(HDR_HAS_L1HDR(hdr)); 4419 4420 /* 4421 * We don't grab the hash lock prior to this check, because if 4422 * the buffer's header is in the arc_anon state, it won't be 4423 * linked into the hash table. 4424 */ 4425 if (hdr->b_l1hdr.b_state == arc_anon) { 4426 mutex_exit(&buf->b_evict_lock); 4427 ASSERT(!HDR_IO_IN_PROGRESS(hdr)); 4428 ASSERT(!HDR_IN_HASH_TABLE(hdr)); 4429 ASSERT(!HDR_HAS_L2HDR(hdr)); 4430 ASSERT(BUF_EMPTY(hdr)); 4431 4432 ASSERT3U(hdr->b_l1hdr.b_datacnt, ==, 1); 4433 ASSERT3S(refcount_count(&hdr->b_l1hdr.b_refcnt), ==, 1); 4434 ASSERT(!list_link_active(&hdr->b_l1hdr.b_arc_node)); 4435 4436 ASSERT3P(buf->b_efunc, ==, NULL); 4437 ASSERT3P(buf->b_private, ==, NULL); 4438 4439 hdr->b_l1hdr.b_arc_access = 0; 4440 arc_buf_thaw(buf); 4441 4442 return; 4443 } 4444 4445 kmutex_t *hash_lock = HDR_LOCK(hdr); 4446 mutex_enter(hash_lock); 4447 4448 /* 4449 * This assignment is only valid as long as the hash_lock is 4450 * held, we must be careful not to reference state or the 4451 * b_state field after dropping the lock. 4452 */ 4453 arc_state_t *state = hdr->b_l1hdr.b_state; 4454 ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); 4455 ASSERT3P(state, !=, arc_anon); 4456 4457 /* this buffer is not on any list */ 4458 ASSERT(refcount_count(&hdr->b_l1hdr.b_refcnt) > 0); 4459 4460 if (HDR_HAS_L2HDR(hdr)) { 4461 mutex_enter(&hdr->b_l2hdr.b_dev->l2ad_mtx); 4462 4463 /* 4464 * We have to recheck this conditional again now that 4465 * we're holding the l2ad_mtx to prevent a race with 4466 * another thread which might be concurrently calling 4467 * l2arc_evict(). In that case, l2arc_evict() might have 4468 * destroyed the header's L2 portion as we were waiting 4469 * to acquire the l2ad_mtx. 4470 */ 4471 if (HDR_HAS_L2HDR(hdr)) 4472 arc_hdr_l2hdr_destroy(hdr); 4473 4474 mutex_exit(&hdr->b_l2hdr.b_dev->l2ad_mtx); 4475 } 4476 4477 /* 4478 * Do we have more than one buf? 4479 */ 4480 if (hdr->b_l1hdr.b_datacnt > 1) { 4481 arc_buf_hdr_t *nhdr; 4482 arc_buf_t **bufp; 4483 uint64_t blksz = hdr->b_size; 4484 uint64_t spa = hdr->b_spa; 4485 arc_buf_contents_t type = arc_buf_type(hdr); 4486 uint32_t flags = hdr->b_flags; 4487 4488 ASSERT(hdr->b_l1hdr.b_buf != buf || buf->b_next != NULL); 4489 /* 4490 * Pull the data off of this hdr and attach it to 4491 * a new anonymous hdr. 4492 */ 4493 (void) remove_reference(hdr, hash_lock, tag); 4494 bufp = &hdr->b_l1hdr.b_buf; 4495 while (*bufp != buf) 4496 bufp = &(*bufp)->b_next; 4497 *bufp = buf->b_next; 4498 buf->b_next = NULL; 4499 4500 ASSERT3P(state, !=, arc_l2c_only); 4501 4502 (void) refcount_remove_many( 4503 &state->arcs_size, hdr->b_size, buf); 4504 4505 if (refcount_is_zero(&hdr->b_l1hdr.b_refcnt)) { 4506 ASSERT3P(state, !=, arc_l2c_only); 4507 uint64_t *size = &state->arcs_lsize[type]; 4508 ASSERT3U(*size, >=, hdr->b_size); 4509 atomic_add_64(size, -hdr->b_size); 4510 } 4511 4512 /* 4513 * We're releasing a duplicate user data buffer, update 4514 * our statistics accordingly. 4515 */ 4516 if (HDR_ISTYPE_DATA(hdr)) { 4517 ARCSTAT_BUMPDOWN(arcstat_duplicate_buffers); 4518 ARCSTAT_INCR(arcstat_duplicate_buffers_size, 4519 -hdr->b_size); 4520 } 4521 hdr->b_l1hdr.b_datacnt -= 1; 4522 arc_cksum_verify(buf); 4523 arc_buf_unwatch(buf); 4524 4525 mutex_exit(hash_lock); 4526 4527 nhdr = kmem_cache_alloc(hdr_full_cache, KM_PUSHPAGE); 4528 nhdr->b_size = blksz; 4529 nhdr->b_spa = spa; 4530 4531 nhdr->b_flags = flags & ARC_FLAG_L2_WRITING; 4532 nhdr->b_flags |= arc_bufc_to_flags(type); 4533 nhdr->b_flags |= ARC_FLAG_HAS_L1HDR; 4534 4535 nhdr->b_l1hdr.b_buf = buf; 4536 nhdr->b_l1hdr.b_datacnt = 1; 4537 nhdr->b_l1hdr.b_state = arc_anon; 4538 nhdr->b_l1hdr.b_arc_access = 0; 4539 nhdr->b_l1hdr.b_tmp_cdata = NULL; 4540 nhdr->b_freeze_cksum = NULL; 4541 4542 (void) refcount_add(&nhdr->b_l1hdr.b_refcnt, tag); 4543 buf->b_hdr = nhdr; 4544 mutex_exit(&buf->b_evict_lock); 4545 (void) refcount_add_many(&arc_anon->arcs_size, blksz, buf); 4546 } else { 4547 mutex_exit(&buf->b_evict_lock); 4548 ASSERT(refcount_count(&hdr->b_l1hdr.b_refcnt) == 1); 4549 /* protected by hash lock, or hdr is on arc_anon */ 4550 ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); 4551 ASSERT(!HDR_IO_IN_PROGRESS(hdr)); 4552 arc_change_state(arc_anon, hdr, hash_lock); 4553 hdr->b_l1hdr.b_arc_access = 0; 4554 mutex_exit(hash_lock); 4555 4556 buf_discard_identity(hdr); 4557 arc_buf_thaw(buf); 4558 } 4559 buf->b_efunc = NULL; 4560 buf->b_private = NULL; 4561 } 4562 4563 int 4564 arc_released(arc_buf_t *buf) 4565 { 4566 int released; 4567 4568 mutex_enter(&buf->b_evict_lock); 4569 released = (buf->b_data != NULL && 4570 buf->b_hdr->b_l1hdr.b_state == arc_anon); 4571 mutex_exit(&buf->b_evict_lock); 4572 return (released); 4573 } 4574 4575 #ifdef ZFS_DEBUG 4576 int 4577 arc_referenced(arc_buf_t *buf) 4578 { 4579 int referenced; 4580 4581 mutex_enter(&buf->b_evict_lock); 4582 referenced = (refcount_count(&buf->b_hdr->b_l1hdr.b_refcnt)); 4583 mutex_exit(&buf->b_evict_lock); 4584 return (referenced); 4585 } 4586 #endif 4587 4588 static void 4589 arc_write_ready(zio_t *zio) 4590 { 4591 arc_write_callback_t *callback = zio->io_private; 4592 arc_buf_t *buf = callback->awcb_buf; 4593 arc_buf_hdr_t *hdr = buf->b_hdr; 4594 4595 ASSERT(HDR_HAS_L1HDR(hdr)); 4596 ASSERT(!refcount_is_zero(&buf->b_hdr->b_l1hdr.b_refcnt)); 4597 ASSERT(hdr->b_l1hdr.b_datacnt > 0); 4598 callback->awcb_ready(zio, buf, callback->awcb_private); 4599 4600 /* 4601 * If the IO is already in progress, then this is a re-write 4602 * attempt, so we need to thaw and re-compute the cksum. 4603 * It is the responsibility of the callback to handle the 4604 * accounting for any re-write attempt. 4605 */ 4606 if (HDR_IO_IN_PROGRESS(hdr)) { 4607 mutex_enter(&hdr->b_l1hdr.b_freeze_lock); 4608 if (hdr->b_freeze_cksum != NULL) { 4609 kmem_free(hdr->b_freeze_cksum, sizeof (zio_cksum_t)); 4610 hdr->b_freeze_cksum = NULL; 4611 } 4612 mutex_exit(&hdr->b_l1hdr.b_freeze_lock); 4613 } 4614 arc_cksum_compute(buf, B_FALSE); 4615 hdr->b_flags |= ARC_FLAG_IO_IN_PROGRESS; 4616 } 4617 4618 /* 4619 * The SPA calls this callback for each physical write that happens on behalf 4620 * of a logical write. See the comment in dbuf_write_physdone() for details. 4621 */ 4622 static void 4623 arc_write_physdone(zio_t *zio) 4624 { 4625 arc_write_callback_t *cb = zio->io_private; 4626 if (cb->awcb_physdone != NULL) 4627 cb->awcb_physdone(zio, cb->awcb_buf, cb->awcb_private); 4628 } 4629 4630 static void 4631 arc_write_done(zio_t *zio) 4632 { 4633 arc_write_callback_t *callback = zio->io_private; 4634 arc_buf_t *buf = callback->awcb_buf; 4635 arc_buf_hdr_t *hdr = buf->b_hdr; 4636 4637 ASSERT(hdr->b_l1hdr.b_acb == NULL); 4638 4639 if (zio->io_error == 0) { 4640 if (BP_IS_HOLE(zio->io_bp) || BP_IS_EMBEDDED(zio->io_bp)) { 4641 buf_discard_identity(hdr); 4642 } else { 4643 hdr->b_dva = *BP_IDENTITY(zio->io_bp); 4644 hdr->b_birth = BP_PHYSICAL_BIRTH(zio->io_bp); 4645 } 4646 } else { 4647 ASSERT(BUF_EMPTY(hdr)); 4648 } 4649 4650 /* 4651 * If the block to be written was all-zero or compressed enough to be 4652 * embedded in the BP, no write was performed so there will be no 4653 * dva/birth/checksum. The buffer must therefore remain anonymous 4654 * (and uncached). 4655 */ 4656 if (!BUF_EMPTY(hdr)) { 4657 arc_buf_hdr_t *exists; 4658 kmutex_t *hash_lock; 4659 4660 ASSERT(zio->io_error == 0); 4661 4662 arc_cksum_verify(buf); 4663 4664 exists = buf_hash_insert(hdr, &hash_lock); 4665 if (exists != NULL) { 4666 /* 4667 * This can only happen if we overwrite for 4668 * sync-to-convergence, because we remove 4669 * buffers from the hash table when we arc_free(). 4670 */ 4671 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) { 4672 if (!BP_EQUAL(&zio->io_bp_orig, zio->io_bp)) 4673 panic("bad overwrite, hdr=%p exists=%p", 4674 (void *)hdr, (void *)exists); 4675 ASSERT(refcount_is_zero( 4676 &exists->b_l1hdr.b_refcnt)); 4677 arc_change_state(arc_anon, exists, hash_lock); 4678 mutex_exit(hash_lock); 4679 arc_hdr_destroy(exists); 4680 exists = buf_hash_insert(hdr, &hash_lock); 4681 ASSERT3P(exists, ==, NULL); 4682 } else if (zio->io_flags & ZIO_FLAG_NOPWRITE) { 4683 /* nopwrite */ 4684 ASSERT(zio->io_prop.zp_nopwrite); 4685 if (!BP_EQUAL(&zio->io_bp_orig, zio->io_bp)) 4686 panic("bad nopwrite, hdr=%p exists=%p", 4687 (void *)hdr, (void *)exists); 4688 } else { 4689 /* Dedup */ 4690 ASSERT(hdr->b_l1hdr.b_datacnt == 1); 4691 ASSERT(hdr->b_l1hdr.b_state == arc_anon); 4692 ASSERT(BP_GET_DEDUP(zio->io_bp)); 4693 ASSERT(BP_GET_LEVEL(zio->io_bp) == 0); 4694 } 4695 } 4696 hdr->b_flags &= ~ARC_FLAG_IO_IN_PROGRESS; 4697 /* if it's not anon, we are doing a scrub */ 4698 if (exists == NULL && hdr->b_l1hdr.b_state == arc_anon) 4699 arc_access(hdr, hash_lock); 4700 mutex_exit(hash_lock); 4701 } else { 4702 hdr->b_flags &= ~ARC_FLAG_IO_IN_PROGRESS; 4703 } 4704 4705 ASSERT(!refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); 4706 callback->awcb_done(zio, buf, callback->awcb_private); 4707 4708 kmem_free(callback, sizeof (arc_write_callback_t)); 4709 } 4710 4711 zio_t * 4712 arc_write(zio_t *pio, spa_t *spa, uint64_t txg, 4713 blkptr_t *bp, arc_buf_t *buf, boolean_t l2arc, boolean_t l2arc_compress, 4714 const zio_prop_t *zp, arc_done_func_t *ready, arc_done_func_t *physdone, 4715 arc_done_func_t *done, void *private, zio_priority_t priority, 4716 int zio_flags, const zbookmark_phys_t *zb) 4717 { 4718 arc_buf_hdr_t *hdr = buf->b_hdr; 4719 arc_write_callback_t *callback; 4720 zio_t *zio; 4721 4722 ASSERT(ready != NULL); 4723 ASSERT(done != NULL); 4724 ASSERT(!HDR_IO_ERROR(hdr)); 4725 ASSERT(!HDR_IO_IN_PROGRESS(hdr)); 4726 ASSERT(hdr->b_l1hdr.b_acb == NULL); 4727 ASSERT(hdr->b_l1hdr.b_datacnt > 0); 4728 if (l2arc) 4729 hdr->b_flags |= ARC_FLAG_L2CACHE; 4730 if (l2arc_compress) 4731 hdr->b_flags |= ARC_FLAG_L2COMPRESS; 4732 callback = kmem_zalloc(sizeof (arc_write_callback_t), KM_SLEEP); 4733 callback->awcb_ready = ready; 4734 callback->awcb_physdone = physdone; 4735 callback->awcb_done = done; 4736 callback->awcb_private = private; 4737 callback->awcb_buf = buf; 4738 4739 zio = zio_write(pio, spa, txg, bp, buf->b_data, hdr->b_size, zp, 4740 arc_write_ready, arc_write_physdone, arc_write_done, callback, 4741 priority, zio_flags, zb); 4742 4743 return (zio); 4744 } 4745 4746 static int 4747 arc_memory_throttle(uint64_t reserve, uint64_t txg) 4748 { 4749 #ifdef _KERNEL 4750 uint64_t available_memory = ptob(freemem); 4751 static uint64_t page_load = 0; 4752 static uint64_t last_txg = 0; 4753 4754 #if defined(__i386) 4755 available_memory = 4756 MIN(available_memory, vmem_size(heap_arena, VMEM_FREE)); 4757 #endif 4758 4759 if (freemem > physmem * arc_lotsfree_percent / 100) 4760 return (0); 4761 4762 if (txg > last_txg) { 4763 last_txg = txg; 4764 page_load = 0; 4765 } 4766 /* 4767 * If we are in pageout, we know that memory is already tight, 4768 * the arc is already going to be evicting, so we just want to 4769 * continue to let page writes occur as quickly as possible. 4770 */ 4771 if (curproc == proc_pageout) { 4772 if (page_load > MAX(ptob(minfree), available_memory) / 4) 4773 return (SET_ERROR(ERESTART)); 4774 /* Note: reserve is inflated, so we deflate */ 4775 page_load += reserve / 8; 4776 return (0); 4777 } else if (page_load > 0 && arc_reclaim_needed()) { 4778 /* memory is low, delay before restarting */ 4779 ARCSTAT_INCR(arcstat_memory_throttle_count, 1); 4780 return (SET_ERROR(EAGAIN)); 4781 } 4782 page_load = 0; 4783 #endif 4784 return (0); 4785 } 4786 4787 void 4788 arc_tempreserve_clear(uint64_t reserve) 4789 { 4790 atomic_add_64(&arc_tempreserve, -reserve); 4791 ASSERT((int64_t)arc_tempreserve >= 0); 4792 } 4793 4794 int 4795 arc_tempreserve_space(uint64_t reserve, uint64_t txg) 4796 { 4797 int error; 4798 uint64_t anon_size; 4799 4800 if (reserve > arc_c/4 && !arc_no_grow) 4801 arc_c = MIN(arc_c_max, reserve * 4); 4802 if (reserve > arc_c) 4803 return (SET_ERROR(ENOMEM)); 4804 4805 /* 4806 * Don't count loaned bufs as in flight dirty data to prevent long 4807 * network delays from blocking transactions that are ready to be 4808 * assigned to a txg. 4809 */ 4810 anon_size = MAX((int64_t)(refcount_count(&arc_anon->arcs_size) - 4811 arc_loaned_bytes), 0); 4812 4813 /* 4814 * Writes will, almost always, require additional memory allocations 4815 * in order to compress/encrypt/etc the data. We therefore need to 4816 * make sure that there is sufficient available memory for this. 4817 */ 4818 error = arc_memory_throttle(reserve, txg); 4819 if (error != 0) 4820 return (error); 4821 4822 /* 4823 * Throttle writes when the amount of dirty data in the cache 4824 * gets too large. We try to keep the cache less than half full 4825 * of dirty blocks so that our sync times don't grow too large. 4826 * Note: if two requests come in concurrently, we might let them 4827 * both succeed, when one of them should fail. Not a huge deal. 4828 */ 4829 4830 if (reserve + arc_tempreserve + anon_size > arc_c / 2 && 4831 anon_size > arc_c / 4) { 4832 dprintf("failing, arc_tempreserve=%lluK anon_meta=%lluK " 4833 "anon_data=%lluK tempreserve=%lluK arc_c=%lluK\n", 4834 arc_tempreserve>>10, 4835 arc_anon->arcs_lsize[ARC_BUFC_METADATA]>>10, 4836 arc_anon->arcs_lsize[ARC_BUFC_DATA]>>10, 4837 reserve>>10, arc_c>>10); 4838 return (SET_ERROR(ERESTART)); 4839 } 4840 atomic_add_64(&arc_tempreserve, reserve); 4841 return (0); 4842 } 4843 4844 static void 4845 arc_kstat_update_state(arc_state_t *state, kstat_named_t *size, 4846 kstat_named_t *evict_data, kstat_named_t *evict_metadata) 4847 { 4848 size->value.ui64 = refcount_count(&state->arcs_size); 4849 evict_data->value.ui64 = state->arcs_lsize[ARC_BUFC_DATA]; 4850 evict_metadata->value.ui64 = state->arcs_lsize[ARC_BUFC_METADATA]; 4851 } 4852 4853 static int 4854 arc_kstat_update(kstat_t *ksp, int rw) 4855 { 4856 arc_stats_t *as = ksp->ks_data; 4857 4858 if (rw == KSTAT_WRITE) { 4859 return (EACCES); 4860 } else { 4861 arc_kstat_update_state(arc_anon, 4862 &as->arcstat_anon_size, 4863 &as->arcstat_anon_evictable_data, 4864 &as->arcstat_anon_evictable_metadata); 4865 arc_kstat_update_state(arc_mru, 4866 &as->arcstat_mru_size, 4867 &as->arcstat_mru_evictable_data, 4868 &as->arcstat_mru_evictable_metadata); 4869 arc_kstat_update_state(arc_mru_ghost, 4870 &as->arcstat_mru_ghost_size, 4871 &as->arcstat_mru_ghost_evictable_data, 4872 &as->arcstat_mru_ghost_evictable_metadata); 4873 arc_kstat_update_state(arc_mfu, 4874 &as->arcstat_mfu_size, 4875 &as->arcstat_mfu_evictable_data, 4876 &as->arcstat_mfu_evictable_metadata); 4877 arc_kstat_update_state(arc_mfu_ghost, 4878 &as->arcstat_mfu_ghost_size, 4879 &as->arcstat_mfu_ghost_evictable_data, 4880 &as->arcstat_mfu_ghost_evictable_metadata); 4881 } 4882 4883 return (0); 4884 } 4885 4886 /* 4887 * This function *must* return indices evenly distributed between all 4888 * sublists of the multilist. This is needed due to how the ARC eviction 4889 * code is laid out; arc_evict_state() assumes ARC buffers are evenly 4890 * distributed between all sublists and uses this assumption when 4891 * deciding which sublist to evict from and how much to evict from it. 4892 */ 4893 unsigned int 4894 arc_state_multilist_index_func(multilist_t *ml, void *obj) 4895 { 4896 arc_buf_hdr_t *hdr = obj; 4897 4898 /* 4899 * We rely on b_dva to generate evenly distributed index 4900 * numbers using buf_hash below. So, as an added precaution, 4901 * let's make sure we never add empty buffers to the arc lists. 4902 */ 4903 ASSERT(!BUF_EMPTY(hdr)); 4904 4905 /* 4906 * The assumption here, is the hash value for a given 4907 * arc_buf_hdr_t will remain constant throughout it's lifetime 4908 * (i.e. it's b_spa, b_dva, and b_birth fields don't change). 4909 * Thus, we don't need to store the header's sublist index 4910 * on insertion, as this index can be recalculated on removal. 4911 * 4912 * Also, the low order bits of the hash value are thought to be 4913 * distributed evenly. Otherwise, in the case that the multilist 4914 * has a power of two number of sublists, each sublists' usage 4915 * would not be evenly distributed. 4916 */ 4917 return (buf_hash(hdr->b_spa, &hdr->b_dva, hdr->b_birth) % 4918 multilist_get_num_sublists(ml)); 4919 } 4920 4921 void 4922 arc_init(void) 4923 { 4924 /* 4925 * allmem is "all memory that we could possibly use". 4926 */ 4927 #ifdef _KERNEL 4928 uint64_t allmem = ptob(physmem - swapfs_minfree); 4929 #else 4930 uint64_t allmem = (physmem * PAGESIZE) / 2; 4931 #endif 4932 4933 mutex_init(&arc_reclaim_lock, NULL, MUTEX_DEFAULT, NULL); 4934 cv_init(&arc_reclaim_thread_cv, NULL, CV_DEFAULT, NULL); 4935 cv_init(&arc_reclaim_waiters_cv, NULL, CV_DEFAULT, NULL); 4936 4937 mutex_init(&arc_user_evicts_lock, NULL, MUTEX_DEFAULT, NULL); 4938 cv_init(&arc_user_evicts_cv, NULL, CV_DEFAULT, NULL); 4939 4940 /* Convert seconds to clock ticks */ 4941 arc_min_prefetch_lifespan = 1 * hz; 4942 4943 /* Start out with 1/8 of all memory */ 4944 arc_c = allmem / 8; 4945 4946 #ifdef _KERNEL 4947 /* 4948 * On architectures where the physical memory can be larger 4949 * than the addressable space (intel in 32-bit mode), we may 4950 * need to limit the cache to 1/8 of VM size. 4951 */ 4952 arc_c = MIN(arc_c, vmem_size(heap_arena, VMEM_ALLOC | VMEM_FREE) / 8); 4953 #endif 4954 4955 /* set min cache to 1/32 of all memory, or 64MB, whichever is more */ 4956 arc_c_min = MAX(allmem / 32, 64 << 20); 4957 /* set max to 3/4 of all memory, or all but 1GB, whichever is more */ 4958 if (allmem >= 1 << 30) 4959 arc_c_max = allmem - (1 << 30); 4960 else 4961 arc_c_max = arc_c_min; 4962 arc_c_max = MAX(allmem * 3 / 4, arc_c_max); 4963 4964 /* 4965 * Allow the tunables to override our calculations if they are 4966 * reasonable (ie. over 64MB) 4967 */ 4968 if (zfs_arc_max > 64 << 20 && zfs_arc_max < allmem) 4969 arc_c_max = zfs_arc_max; 4970 if (zfs_arc_min > 64 << 20 && zfs_arc_min <= arc_c_max) 4971 arc_c_min = zfs_arc_min; 4972 4973 arc_c = arc_c_max; 4974 arc_p = (arc_c >> 1); 4975 4976 /* limit meta-data to 1/4 of the arc capacity */ 4977 arc_meta_limit = arc_c_max / 4; 4978 4979 /* Allow the tunable to override if it is reasonable */ 4980 if (zfs_arc_meta_limit > 0 && zfs_arc_meta_limit <= arc_c_max) 4981 arc_meta_limit = zfs_arc_meta_limit; 4982 4983 if (arc_c_min < arc_meta_limit / 2 && zfs_arc_min == 0) 4984 arc_c_min = arc_meta_limit / 2; 4985 4986 if (zfs_arc_meta_min > 0) { 4987 arc_meta_min = zfs_arc_meta_min; 4988 } else { 4989 arc_meta_min = arc_c_min / 2; 4990 } 4991 4992 if (zfs_arc_grow_retry > 0) 4993 arc_grow_retry = zfs_arc_grow_retry; 4994 4995 if (zfs_arc_shrink_shift > 0) 4996 arc_shrink_shift = zfs_arc_shrink_shift; 4997 4998 /* 4999 * Ensure that arc_no_grow_shift is less than arc_shrink_shift. 5000 */ 5001 if (arc_no_grow_shift >= arc_shrink_shift) 5002 arc_no_grow_shift = arc_shrink_shift - 1; 5003 5004 if (zfs_arc_p_min_shift > 0) 5005 arc_p_min_shift = zfs_arc_p_min_shift; 5006 5007 if (zfs_arc_num_sublists_per_state < 1) 5008 zfs_arc_num_sublists_per_state = MAX(boot_ncpus, 1); 5009 5010 /* if kmem_flags are set, lets try to use less memory */ 5011 if (kmem_debugging()) 5012 arc_c = arc_c / 2; 5013 if (arc_c < arc_c_min) 5014 arc_c = arc_c_min; 5015 5016 arc_anon = &ARC_anon; 5017 arc_mru = &ARC_mru; 5018 arc_mru_ghost = &ARC_mru_ghost; 5019 arc_mfu = &ARC_mfu; 5020 arc_mfu_ghost = &ARC_mfu_ghost; 5021 arc_l2c_only = &ARC_l2c_only; 5022 arc_size = 0; 5023 5024 multilist_create(&arc_mru->arcs_list[ARC_BUFC_METADATA], 5025 sizeof (arc_buf_hdr_t), 5026 offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), 5027 zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); 5028 multilist_create(&arc_mru->arcs_list[ARC_BUFC_DATA], 5029 sizeof (arc_buf_hdr_t), 5030 offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), 5031 zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); 5032 multilist_create(&arc_mru_ghost->arcs_list[ARC_BUFC_METADATA], 5033 sizeof (arc_buf_hdr_t), 5034 offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), 5035 zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); 5036 multilist_create(&arc_mru_ghost->arcs_list[ARC_BUFC_DATA], 5037 sizeof (arc_buf_hdr_t), 5038 offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), 5039 zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); 5040 multilist_create(&arc_mfu->arcs_list[ARC_BUFC_METADATA], 5041 sizeof (arc_buf_hdr_t), 5042 offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), 5043 zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); 5044 multilist_create(&arc_mfu->arcs_list[ARC_BUFC_DATA], 5045 sizeof (arc_buf_hdr_t), 5046 offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), 5047 zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); 5048 multilist_create(&arc_mfu_ghost->arcs_list[ARC_BUFC_METADATA], 5049 sizeof (arc_buf_hdr_t), 5050 offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), 5051 zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); 5052 multilist_create(&arc_mfu_ghost->arcs_list[ARC_BUFC_DATA], 5053 sizeof (arc_buf_hdr_t), 5054 offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), 5055 zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); 5056 multilist_create(&arc_l2c_only->arcs_list[ARC_BUFC_METADATA], 5057 sizeof (arc_buf_hdr_t), 5058 offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), 5059 zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); 5060 multilist_create(&arc_l2c_only->arcs_list[ARC_BUFC_DATA], 5061 sizeof (arc_buf_hdr_t), 5062 offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), 5063 zfs_arc_num_sublists_per_state, arc_state_multilist_index_func); 5064 5065 refcount_create(&arc_anon->arcs_size); 5066 refcount_create(&arc_mru->arcs_size); 5067 refcount_create(&arc_mru_ghost->arcs_size); 5068 refcount_create(&arc_mfu->arcs_size); 5069 refcount_create(&arc_mfu_ghost->arcs_size); 5070 refcount_create(&arc_l2c_only->arcs_size); 5071 5072 buf_init(); 5073 5074 arc_reclaim_thread_exit = FALSE; 5075 arc_user_evicts_thread_exit = FALSE; 5076 arc_eviction_list = NULL; 5077 bzero(&arc_eviction_hdr, sizeof (arc_buf_hdr_t)); 5078 5079 arc_ksp = kstat_create("zfs", 0, "arcstats", "misc", KSTAT_TYPE_NAMED, 5080 sizeof (arc_stats) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL); 5081 5082 if (arc_ksp != NULL) { 5083 arc_ksp->ks_data = &arc_stats; 5084 arc_ksp->ks_update = arc_kstat_update; 5085 kstat_install(arc_ksp); 5086 } 5087 5088 (void) thread_create(NULL, 0, arc_reclaim_thread, NULL, 0, &p0, 5089 TS_RUN, minclsyspri); 5090 5091 (void) thread_create(NULL, 0, arc_user_evicts_thread, NULL, 0, &p0, 5092 TS_RUN, minclsyspri); 5093 5094 arc_dead = FALSE; 5095 arc_warm = B_FALSE; 5096 5097 /* 5098 * Calculate maximum amount of dirty data per pool. 5099 * 5100 * If it has been set by /etc/system, take that. 5101 * Otherwise, use a percentage of physical memory defined by 5102 * zfs_dirty_data_max_percent (default 10%) with a cap at 5103 * zfs_dirty_data_max_max (default 4GB). 5104 */ 5105 if (zfs_dirty_data_max == 0) { 5106 zfs_dirty_data_max = physmem * PAGESIZE * 5107 zfs_dirty_data_max_percent / 100; 5108 zfs_dirty_data_max = MIN(zfs_dirty_data_max, 5109 zfs_dirty_data_max_max); 5110 } 5111 } 5112 5113 void 5114 arc_fini(void) 5115 { 5116 mutex_enter(&arc_reclaim_lock); 5117 arc_reclaim_thread_exit = TRUE; 5118 /* 5119 * The reclaim thread will set arc_reclaim_thread_exit back to 5120 * FALSE when it is finished exiting; we're waiting for that. 5121 */ 5122 while (arc_reclaim_thread_exit) { 5123 cv_signal(&arc_reclaim_thread_cv); 5124 cv_wait(&arc_reclaim_thread_cv, &arc_reclaim_lock); 5125 } 5126 mutex_exit(&arc_reclaim_lock); 5127 5128 mutex_enter(&arc_user_evicts_lock); 5129 arc_user_evicts_thread_exit = TRUE; 5130 /* 5131 * The user evicts thread will set arc_user_evicts_thread_exit 5132 * to FALSE when it is finished exiting; we're waiting for that. 5133 */ 5134 while (arc_user_evicts_thread_exit) { 5135 cv_signal(&arc_user_evicts_cv); 5136 cv_wait(&arc_user_evicts_cv, &arc_user_evicts_lock); 5137 } 5138 mutex_exit(&arc_user_evicts_lock); 5139 5140 /* Use TRUE to ensure *all* buffers are evicted */ 5141 arc_flush(NULL, TRUE); 5142 5143 arc_dead = TRUE; 5144 5145 if (arc_ksp != NULL) { 5146 kstat_delete(arc_ksp); 5147 arc_ksp = NULL; 5148 } 5149 5150 mutex_destroy(&arc_reclaim_lock); 5151 cv_destroy(&arc_reclaim_thread_cv); 5152 cv_destroy(&arc_reclaim_waiters_cv); 5153 5154 mutex_destroy(&arc_user_evicts_lock); 5155 cv_destroy(&arc_user_evicts_cv); 5156 5157 refcount_destroy(&arc_anon->arcs_size); 5158 refcount_destroy(&arc_mru->arcs_size); 5159 refcount_destroy(&arc_mru_ghost->arcs_size); 5160 refcount_destroy(&arc_mfu->arcs_size); 5161 refcount_destroy(&arc_mfu_ghost->arcs_size); 5162 refcount_destroy(&arc_l2c_only->arcs_size); 5163 5164 multilist_destroy(&arc_mru->arcs_list[ARC_BUFC_METADATA]); 5165 multilist_destroy(&arc_mru_ghost->arcs_list[ARC_BUFC_METADATA]); 5166 multilist_destroy(&arc_mfu->arcs_list[ARC_BUFC_METADATA]); 5167 multilist_destroy(&arc_mfu_ghost->arcs_list[ARC_BUFC_METADATA]); 5168 multilist_destroy(&arc_mru->arcs_list[ARC_BUFC_DATA]); 5169 multilist_destroy(&arc_mru_ghost->arcs_list[ARC_BUFC_DATA]); 5170 multilist_destroy(&arc_mfu->arcs_list[ARC_BUFC_DATA]); 5171 multilist_destroy(&arc_mfu_ghost->arcs_list[ARC_BUFC_DATA]); 5172 5173 buf_fini(); 5174 5175 ASSERT0(arc_loaned_bytes); 5176 } 5177 5178 /* 5179 * Level 2 ARC 5180 * 5181 * The level 2 ARC (L2ARC) is a cache layer in-between main memory and disk. 5182 * It uses dedicated storage devices to hold cached data, which are populated 5183 * using large infrequent writes. The main role of this cache is to boost 5184 * the performance of random read workloads. The intended L2ARC devices 5185 * include short-stroked disks, solid state disks, and other media with 5186 * substantially faster read latency than disk. 5187 * 5188 * +-----------------------+ 5189 * | ARC | 5190 * +-----------------------+ 5191 * | ^ ^ 5192 * | | | 5193 * l2arc_feed_thread() arc_read() 5194 * | | | 5195 * | l2arc read | 5196 * V | | 5197 * +---------------+ | 5198 * | L2ARC | | 5199 * +---------------+ | 5200 * | ^ | 5201 * l2arc_write() | | 5202 * | | | 5203 * V | | 5204 * +-------+ +-------+ 5205 * | vdev | | vdev | 5206 * | cache | | cache | 5207 * +-------+ +-------+ 5208 * +=========+ .-----. 5209 * : L2ARC : |-_____-| 5210 * : devices : | Disks | 5211 * +=========+ `-_____-' 5212 * 5213 * Read requests are satisfied from the following sources, in order: 5214 * 5215 * 1) ARC 5216 * 2) vdev cache of L2ARC devices 5217 * 3) L2ARC devices 5218 * 4) vdev cache of disks 5219 * 5) disks 5220 * 5221 * Some L2ARC device types exhibit extremely slow write performance. 5222 * To accommodate for this there are some significant differences between 5223 * the L2ARC and traditional cache design: 5224 * 5225 * 1. There is no eviction path from the ARC to the L2ARC. Evictions from 5226 * the ARC behave as usual, freeing buffers and placing headers on ghost 5227 * lists. The ARC does not send buffers to the L2ARC during eviction as 5228 * this would add inflated write latencies for all ARC memory pressure. 5229 * 5230 * 2. The L2ARC attempts to cache data from the ARC before it is evicted. 5231 * It does this by periodically scanning buffers from the eviction-end of 5232 * the MFU and MRU ARC lists, copying them to the L2ARC devices if they are 5233 * not already there. It scans until a headroom of buffers is satisfied, 5234 * which itself is a buffer for ARC eviction. If a compressible buffer is 5235 * found during scanning and selected for writing to an L2ARC device, we 5236 * temporarily boost scanning headroom during the next scan cycle to make 5237 * sure we adapt to compression effects (which might significantly reduce 5238 * the data volume we write to L2ARC). The thread that does this is 5239 * l2arc_feed_thread(), illustrated below; example sizes are included to 5240 * provide a better sense of ratio than this diagram: 5241 * 5242 * head --> tail 5243 * +---------------------+----------+ 5244 * ARC_mfu |:::::#:::::::::::::::|o#o###o###|-->. # already on L2ARC 5245 * +---------------------+----------+ | o L2ARC eligible 5246 * ARC_mru |:#:::::::::::::::::::|#o#ooo####|-->| : ARC buffer 5247 * +---------------------+----------+ | 5248 * 15.9 Gbytes ^ 32 Mbytes | 5249 * headroom | 5250 * l2arc_feed_thread() 5251 * | 5252 * l2arc write hand <--[oooo]--' 5253 * | 8 Mbyte 5254 * | write max 5255 * V 5256 * +==============================+ 5257 * L2ARC dev |####|#|###|###| |####| ... | 5258 * +==============================+ 5259 * 32 Gbytes 5260 * 5261 * 3. If an ARC buffer is copied to the L2ARC but then hit instead of 5262 * evicted, then the L2ARC has cached a buffer much sooner than it probably 5263 * needed to, potentially wasting L2ARC device bandwidth and storage. It is 5264 * safe to say that this is an uncommon case, since buffers at the end of 5265 * the ARC lists have moved there due to inactivity. 5266 * 5267 * 4. If the ARC evicts faster than the L2ARC can maintain a headroom, 5268 * then the L2ARC simply misses copying some buffers. This serves as a 5269 * pressure valve to prevent heavy read workloads from both stalling the ARC 5270 * with waits and clogging the L2ARC with writes. This also helps prevent 5271 * the potential for the L2ARC to churn if it attempts to cache content too 5272 * quickly, such as during backups of the entire pool. 5273 * 5274 * 5. After system boot and before the ARC has filled main memory, there are 5275 * no evictions from the ARC and so the tails of the ARC_mfu and ARC_mru 5276 * lists can remain mostly static. Instead of searching from tail of these 5277 * lists as pictured, the l2arc_feed_thread() will search from the list heads 5278 * for eligible buffers, greatly increasing its chance of finding them. 5279 * 5280 * The L2ARC device write speed is also boosted during this time so that 5281 * the L2ARC warms up faster. Since there have been no ARC evictions yet, 5282 * there are no L2ARC reads, and no fear of degrading read performance 5283 * through increased writes. 5284 * 5285 * 6. Writes to the L2ARC devices are grouped and sent in-sequence, so that 5286 * the vdev queue can aggregate them into larger and fewer writes. Each 5287 * device is written to in a rotor fashion, sweeping writes through 5288 * available space then repeating. 5289 * 5290 * 7. The L2ARC does not store dirty content. It never needs to flush 5291 * write buffers back to disk based storage. 5292 * 5293 * 8. If an ARC buffer is written (and dirtied) which also exists in the 5294 * L2ARC, the now stale L2ARC buffer is immediately dropped. 5295 * 5296 * The performance of the L2ARC can be tweaked by a number of tunables, which 5297 * may be necessary for different workloads: 5298 * 5299 * l2arc_write_max max write bytes per interval 5300 * l2arc_write_boost extra write bytes during device warmup 5301 * l2arc_noprefetch skip caching prefetched buffers 5302 * l2arc_headroom number of max device writes to precache 5303 * l2arc_headroom_boost when we find compressed buffers during ARC 5304 * scanning, we multiply headroom by this 5305 * percentage factor for the next scan cycle, 5306 * since more compressed buffers are likely to 5307 * be present 5308 * l2arc_feed_secs seconds between L2ARC writing 5309 * 5310 * Tunables may be removed or added as future performance improvements are 5311 * integrated, and also may become zpool properties. 5312 * 5313 * There are three key functions that control how the L2ARC warms up: 5314 * 5315 * l2arc_write_eligible() check if a buffer is eligible to cache 5316 * l2arc_write_size() calculate how much to write 5317 * l2arc_write_interval() calculate sleep delay between writes 5318 * 5319 * These three functions determine what to write, how much, and how quickly 5320 * to send writes. 5321 */ 5322 5323 static boolean_t 5324 l2arc_write_eligible(uint64_t spa_guid, arc_buf_hdr_t *hdr) 5325 { 5326 /* 5327 * A buffer is *not* eligible for the L2ARC if it: 5328 * 1. belongs to a different spa. 5329 * 2. is already cached on the L2ARC. 5330 * 3. has an I/O in progress (it may be an incomplete read). 5331 * 4. is flagged not eligible (zfs property). 5332 */ 5333 if (hdr->b_spa != spa_guid || HDR_HAS_L2HDR(hdr) || 5334 HDR_IO_IN_PROGRESS(hdr) || !HDR_L2CACHE(hdr)) 5335 return (B_FALSE); 5336 5337 return (B_TRUE); 5338 } 5339 5340 static uint64_t 5341 l2arc_write_size(void) 5342 { 5343 uint64_t size; 5344 5345 /* 5346 * Make sure our globals have meaningful values in case the user 5347 * altered them. 5348 */ 5349 size = l2arc_write_max; 5350 if (size == 0) { 5351 cmn_err(CE_NOTE, "Bad value for l2arc_write_max, value must " 5352 "be greater than zero, resetting it to the default (%d)", 5353 L2ARC_WRITE_SIZE); 5354 size = l2arc_write_max = L2ARC_WRITE_SIZE; 5355 } 5356 5357 if (arc_warm == B_FALSE) 5358 size += l2arc_write_boost; 5359 5360 return (size); 5361 5362 } 5363 5364 static clock_t 5365 l2arc_write_interval(clock_t began, uint64_t wanted, uint64_t wrote) 5366 { 5367 clock_t interval, next, now; 5368 5369 /* 5370 * If the ARC lists are busy, increase our write rate; if the 5371 * lists are stale, idle back. This is achieved by checking 5372 * how much we previously wrote - if it was more than half of 5373 * what we wanted, schedule the next write much sooner. 5374 */ 5375 if (l2arc_feed_again && wrote > (wanted / 2)) 5376 interval = (hz * l2arc_feed_min_ms) / 1000; 5377 else 5378 interval = hz * l2arc_feed_secs; 5379 5380 now = ddi_get_lbolt(); 5381 next = MAX(now, MIN(now + interval, began + interval)); 5382 5383 return (next); 5384 } 5385 5386 /* 5387 * Cycle through L2ARC devices. This is how L2ARC load balances. 5388 * If a device is returned, this also returns holding the spa config lock. 5389 */ 5390 static l2arc_dev_t * 5391 l2arc_dev_get_next(void) 5392 { 5393 l2arc_dev_t *first, *next = NULL; 5394 5395 /* 5396 * Lock out the removal of spas (spa_namespace_lock), then removal 5397 * of cache devices (l2arc_dev_mtx). Once a device has been selected, 5398 * both locks will be dropped and a spa config lock held instead. 5399 */ 5400 mutex_enter(&spa_namespace_lock); 5401 mutex_enter(&l2arc_dev_mtx); 5402 5403 /* if there are no vdevs, there is nothing to do */ 5404 if (l2arc_ndev == 0) 5405 goto out; 5406 5407 first = NULL; 5408 next = l2arc_dev_last; 5409 do { 5410 /* loop around the list looking for a non-faulted vdev */ 5411 if (next == NULL) { 5412 next = list_head(l2arc_dev_list); 5413 } else { 5414 next = list_next(l2arc_dev_list, next); 5415 if (next == NULL) 5416 next = list_head(l2arc_dev_list); 5417 } 5418 5419 /* if we have come back to the start, bail out */ 5420 if (first == NULL) 5421 first = next; 5422 else if (next == first) 5423 break; 5424 5425 } while (vdev_is_dead(next->l2ad_vdev)); 5426 5427 /* if we were unable to find any usable vdevs, return NULL */ 5428 if (vdev_is_dead(next->l2ad_vdev)) 5429 next = NULL; 5430 5431 l2arc_dev_last = next; 5432 5433 out: 5434 mutex_exit(&l2arc_dev_mtx); 5435 5436 /* 5437 * Grab the config lock to prevent the 'next' device from being 5438 * removed while we are writing to it. 5439 */ 5440 if (next != NULL) 5441 spa_config_enter(next->l2ad_spa, SCL_L2ARC, next, RW_READER); 5442 mutex_exit(&spa_namespace_lock); 5443 5444 return (next); 5445 } 5446 5447 /* 5448 * Free buffers that were tagged for destruction. 5449 */ 5450 static void 5451 l2arc_do_free_on_write() 5452 { 5453 list_t *buflist; 5454 l2arc_data_free_t *df, *df_prev; 5455 5456 mutex_enter(&l2arc_free_on_write_mtx); 5457 buflist = l2arc_free_on_write; 5458 5459 for (df = list_tail(buflist); df; df = df_prev) { 5460 df_prev = list_prev(buflist, df); 5461 ASSERT(df->l2df_data != NULL); 5462 ASSERT(df->l2df_func != NULL); 5463 df->l2df_func(df->l2df_data, df->l2df_size); 5464 list_remove(buflist, df); 5465 kmem_free(df, sizeof (l2arc_data_free_t)); 5466 } 5467 5468 mutex_exit(&l2arc_free_on_write_mtx); 5469 } 5470 5471 /* 5472 * A write to a cache device has completed. Update all headers to allow 5473 * reads from these buffers to begin. 5474 */ 5475 static void 5476 l2arc_write_done(zio_t *zio) 5477 { 5478 l2arc_write_callback_t *cb; 5479 l2arc_dev_t *dev; 5480 list_t *buflist; 5481 arc_buf_hdr_t *head, *hdr, *hdr_prev; 5482 kmutex_t *hash_lock; 5483 int64_t bytes_dropped = 0; 5484 5485 cb = zio->io_private; 5486 ASSERT(cb != NULL); 5487 dev = cb->l2wcb_dev; 5488 ASSERT(dev != NULL); 5489 head = cb->l2wcb_head; 5490 ASSERT(head != NULL); 5491 buflist = &dev->l2ad_buflist; 5492 ASSERT(buflist != NULL); 5493 DTRACE_PROBE2(l2arc__iodone, zio_t *, zio, 5494 l2arc_write_callback_t *, cb); 5495 5496 if (zio->io_error != 0) 5497 ARCSTAT_BUMP(arcstat_l2_writes_error); 5498 5499 /* 5500 * All writes completed, or an error was hit. 5501 */ 5502 top: 5503 mutex_enter(&dev->l2ad_mtx); 5504 for (hdr = list_prev(buflist, head); hdr; hdr = hdr_prev) { 5505 hdr_prev = list_prev(buflist, hdr); 5506 5507 hash_lock = HDR_LOCK(hdr); 5508 5509 /* 5510 * We cannot use mutex_enter or else we can deadlock 5511 * with l2arc_write_buffers (due to swapping the order 5512 * the hash lock and l2ad_mtx are taken). 5513 */ 5514 if (!mutex_tryenter(hash_lock)) { 5515 /* 5516 * Missed the hash lock. We must retry so we 5517 * don't leave the ARC_FLAG_L2_WRITING bit set. 5518 */ 5519 ARCSTAT_BUMP(arcstat_l2_writes_lock_retry); 5520 5521 /* 5522 * We don't want to rescan the headers we've 5523 * already marked as having been written out, so 5524 * we reinsert the head node so we can pick up 5525 * where we left off. 5526 */ 5527 list_remove(buflist, head); 5528 list_insert_after(buflist, hdr, head); 5529 5530 mutex_exit(&dev->l2ad_mtx); 5531 5532 /* 5533 * We wait for the hash lock to become available 5534 * to try and prevent busy waiting, and increase 5535 * the chance we'll be able to acquire the lock 5536 * the next time around. 5537 */ 5538 mutex_enter(hash_lock); 5539 mutex_exit(hash_lock); 5540 goto top; 5541 } 5542 5543 /* 5544 * We could not have been moved into the arc_l2c_only 5545 * state while in-flight due to our ARC_FLAG_L2_WRITING 5546 * bit being set. Let's just ensure that's being enforced. 5547 */ 5548 ASSERT(HDR_HAS_L1HDR(hdr)); 5549 5550 /* 5551 * We may have allocated a buffer for L2ARC compression, 5552 * we must release it to avoid leaking this data. 5553 */ 5554 l2arc_release_cdata_buf(hdr); 5555 5556 if (zio->io_error != 0) { 5557 /* 5558 * Error - drop L2ARC entry. 5559 */ 5560 list_remove(buflist, hdr); 5561 hdr->b_flags &= ~ARC_FLAG_HAS_L2HDR; 5562 5563 ARCSTAT_INCR(arcstat_l2_asize, -hdr->b_l2hdr.b_asize); 5564 ARCSTAT_INCR(arcstat_l2_size, -hdr->b_size); 5565 5566 bytes_dropped += hdr->b_l2hdr.b_asize; 5567 (void) refcount_remove_many(&dev->l2ad_alloc, 5568 hdr->b_l2hdr.b_asize, hdr); 5569 } 5570 5571 /* 5572 * Allow ARC to begin reads and ghost list evictions to 5573 * this L2ARC entry. 5574 */ 5575 hdr->b_flags &= ~ARC_FLAG_L2_WRITING; 5576 5577 mutex_exit(hash_lock); 5578 } 5579 5580 atomic_inc_64(&l2arc_writes_done); 5581 list_remove(buflist, head); 5582 ASSERT(!HDR_HAS_L1HDR(head)); 5583 kmem_cache_free(hdr_l2only_cache, head); 5584 mutex_exit(&dev->l2ad_mtx); 5585 5586 vdev_space_update(dev->l2ad_vdev, -bytes_dropped, 0, 0); 5587 5588 l2arc_do_free_on_write(); 5589 5590 kmem_free(cb, sizeof (l2arc_write_callback_t)); 5591 } 5592 5593 /* 5594 * A read to a cache device completed. Validate buffer contents before 5595 * handing over to the regular ARC routines. 5596 */ 5597 static void 5598 l2arc_read_done(zio_t *zio) 5599 { 5600 l2arc_read_callback_t *cb; 5601 arc_buf_hdr_t *hdr; 5602 arc_buf_t *buf; 5603 kmutex_t *hash_lock; 5604 int equal; 5605 5606 ASSERT(zio->io_vd != NULL); 5607 ASSERT(zio->io_flags & ZIO_FLAG_DONT_PROPAGATE); 5608 5609 spa_config_exit(zio->io_spa, SCL_L2ARC, zio->io_vd); 5610 5611 cb = zio->io_private; 5612 ASSERT(cb != NULL); 5613 buf = cb->l2rcb_buf; 5614 ASSERT(buf != NULL); 5615 5616 hash_lock = HDR_LOCK(buf->b_hdr); 5617 mutex_enter(hash_lock); 5618 hdr = buf->b_hdr; 5619 ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); 5620 5621 /* 5622 * If the buffer was compressed, decompress it first. 5623 */ 5624 if (cb->l2rcb_compress != ZIO_COMPRESS_OFF) 5625 l2arc_decompress_zio(zio, hdr, cb->l2rcb_compress); 5626 ASSERT(zio->io_data != NULL); 5627 5628 /* 5629 * Check this survived the L2ARC journey. 5630 */ 5631 equal = arc_cksum_equal(buf); 5632 if (equal && zio->io_error == 0 && !HDR_L2_EVICTED(hdr)) { 5633 mutex_exit(hash_lock); 5634 zio->io_private = buf; 5635 zio->io_bp_copy = cb->l2rcb_bp; /* XXX fix in L2ARC 2.0 */ 5636 zio->io_bp = &zio->io_bp_copy; /* XXX fix in L2ARC 2.0 */ 5637 arc_read_done(zio); 5638 } else { 5639 mutex_exit(hash_lock); 5640 /* 5641 * Buffer didn't survive caching. Increment stats and 5642 * reissue to the original storage device. 5643 */ 5644 if (zio->io_error != 0) { 5645 ARCSTAT_BUMP(arcstat_l2_io_error); 5646 } else { 5647 zio->io_error = SET_ERROR(EIO); 5648 } 5649 if (!equal) 5650 ARCSTAT_BUMP(arcstat_l2_cksum_bad); 5651 5652 /* 5653 * If there's no waiter, issue an async i/o to the primary 5654 * storage now. If there *is* a waiter, the caller must 5655 * issue the i/o in a context where it's OK to block. 5656 */ 5657 if (zio->io_waiter == NULL) { 5658 zio_t *pio = zio_unique_parent(zio); 5659 5660 ASSERT(!pio || pio->io_child_type == ZIO_CHILD_LOGICAL); 5661 5662 zio_nowait(zio_read(pio, cb->l2rcb_spa, &cb->l2rcb_bp, 5663 buf->b_data, zio->io_size, arc_read_done, buf, 5664 zio->io_priority, cb->l2rcb_flags, &cb->l2rcb_zb)); 5665 } 5666 } 5667 5668 kmem_free(cb, sizeof (l2arc_read_callback_t)); 5669 } 5670 5671 /* 5672 * This is the list priority from which the L2ARC will search for pages to 5673 * cache. This is used within loops (0..3) to cycle through lists in the 5674 * desired order. This order can have a significant effect on cache 5675 * performance. 5676 * 5677 * Currently the metadata lists are hit first, MFU then MRU, followed by 5678 * the data lists. This function returns a locked list, and also returns 5679 * the lock pointer. 5680 */ 5681 static multilist_sublist_t * 5682 l2arc_sublist_lock(int list_num) 5683 { 5684 multilist_t *ml = NULL; 5685 unsigned int idx; 5686 5687 ASSERT(list_num >= 0 && list_num <= 3); 5688 5689 switch (list_num) { 5690 case 0: 5691 ml = &arc_mfu->arcs_list[ARC_BUFC_METADATA]; 5692 break; 5693 case 1: 5694 ml = &arc_mru->arcs_list[ARC_BUFC_METADATA]; 5695 break; 5696 case 2: 5697 ml = &arc_mfu->arcs_list[ARC_BUFC_DATA]; 5698 break; 5699 case 3: 5700 ml = &arc_mru->arcs_list[ARC_BUFC_DATA]; 5701 break; 5702 } 5703 5704 /* 5705 * Return a randomly-selected sublist. This is acceptable 5706 * because the caller feeds only a little bit of data for each 5707 * call (8MB). Subsequent calls will result in different 5708 * sublists being selected. 5709 */ 5710 idx = multilist_get_random_index(ml); 5711 return (multilist_sublist_lock(ml, idx)); 5712 } 5713 5714 /* 5715 * Evict buffers from the device write hand to the distance specified in 5716 * bytes. This distance may span populated buffers, it may span nothing. 5717 * This is clearing a region on the L2ARC device ready for writing. 5718 * If the 'all' boolean is set, every buffer is evicted. 5719 */ 5720 static void 5721 l2arc_evict(l2arc_dev_t *dev, uint64_t distance, boolean_t all) 5722 { 5723 list_t *buflist; 5724 arc_buf_hdr_t *hdr, *hdr_prev; 5725 kmutex_t *hash_lock; 5726 uint64_t taddr; 5727 5728 buflist = &dev->l2ad_buflist; 5729 5730 if (!all && dev->l2ad_first) { 5731 /* 5732 * This is the first sweep through the device. There is 5733 * nothing to evict. 5734 */ 5735 return; 5736 } 5737 5738 if (dev->l2ad_hand >= (dev->l2ad_end - (2 * distance))) { 5739 /* 5740 * When nearing the end of the device, evict to the end 5741 * before the device write hand jumps to the start. 5742 */ 5743 taddr = dev->l2ad_end; 5744 } else { 5745 taddr = dev->l2ad_hand + distance; 5746 } 5747 DTRACE_PROBE4(l2arc__evict, l2arc_dev_t *, dev, list_t *, buflist, 5748 uint64_t, taddr, boolean_t, all); 5749 5750 top: 5751 mutex_enter(&dev->l2ad_mtx); 5752 for (hdr = list_tail(buflist); hdr; hdr = hdr_prev) { 5753 hdr_prev = list_prev(buflist, hdr); 5754 5755 hash_lock = HDR_LOCK(hdr); 5756 5757 /* 5758 * We cannot use mutex_enter or else we can deadlock 5759 * with l2arc_write_buffers (due to swapping the order 5760 * the hash lock and l2ad_mtx are taken). 5761 */ 5762 if (!mutex_tryenter(hash_lock)) { 5763 /* 5764 * Missed the hash lock. Retry. 5765 */ 5766 ARCSTAT_BUMP(arcstat_l2_evict_lock_retry); 5767 mutex_exit(&dev->l2ad_mtx); 5768 mutex_enter(hash_lock); 5769 mutex_exit(hash_lock); 5770 goto top; 5771 } 5772 5773 if (HDR_L2_WRITE_HEAD(hdr)) { 5774 /* 5775 * We hit a write head node. Leave it for 5776 * l2arc_write_done(). 5777 */ 5778 list_remove(buflist, hdr); 5779 mutex_exit(hash_lock); 5780 continue; 5781 } 5782 5783 if (!all && HDR_HAS_L2HDR(hdr) && 5784 (hdr->b_l2hdr.b_daddr > taddr || 5785 hdr->b_l2hdr.b_daddr < dev->l2ad_hand)) { 5786 /* 5787 * We've evicted to the target address, 5788 * or the end of the device. 5789 */ 5790 mutex_exit(hash_lock); 5791 break; 5792 } 5793 5794 ASSERT(HDR_HAS_L2HDR(hdr)); 5795 if (!HDR_HAS_L1HDR(hdr)) { 5796 ASSERT(!HDR_L2_READING(hdr)); 5797 /* 5798 * This doesn't exist in the ARC. Destroy. 5799 * arc_hdr_destroy() will call list_remove() 5800 * and decrement arcstat_l2_size. 5801 */ 5802 arc_change_state(arc_anon, hdr, hash_lock); 5803 arc_hdr_destroy(hdr); 5804 } else { 5805 ASSERT(hdr->b_l1hdr.b_state != arc_l2c_only); 5806 ARCSTAT_BUMP(arcstat_l2_evict_l1cached); 5807 /* 5808 * Invalidate issued or about to be issued 5809 * reads, since we may be about to write 5810 * over this location. 5811 */ 5812 if (HDR_L2_READING(hdr)) { 5813 ARCSTAT_BUMP(arcstat_l2_evict_reading); 5814 hdr->b_flags |= ARC_FLAG_L2_EVICTED; 5815 } 5816 5817 /* Ensure this header has finished being written */ 5818 ASSERT(!HDR_L2_WRITING(hdr)); 5819 ASSERT3P(hdr->b_l1hdr.b_tmp_cdata, ==, NULL); 5820 5821 arc_hdr_l2hdr_destroy(hdr); 5822 } 5823 mutex_exit(hash_lock); 5824 } 5825 mutex_exit(&dev->l2ad_mtx); 5826 } 5827 5828 /* 5829 * Find and write ARC buffers to the L2ARC device. 5830 * 5831 * An ARC_FLAG_L2_WRITING flag is set so that the L2ARC buffers are not valid 5832 * for reading until they have completed writing. 5833 * The headroom_boost is an in-out parameter used to maintain headroom boost 5834 * state between calls to this function. 5835 * 5836 * Returns the number of bytes actually written (which may be smaller than 5837 * the delta by which the device hand has changed due to alignment). 5838 */ 5839 static uint64_t 5840 l2arc_write_buffers(spa_t *spa, l2arc_dev_t *dev, uint64_t target_sz, 5841 boolean_t *headroom_boost) 5842 { 5843 arc_buf_hdr_t *hdr, *hdr_prev, *head; 5844 uint64_t write_asize, write_psize, write_sz, headroom, 5845 buf_compress_minsz; 5846 void *buf_data; 5847 boolean_t full; 5848 l2arc_write_callback_t *cb; 5849 zio_t *pio, *wzio; 5850 uint64_t guid = spa_load_guid(spa); 5851 const boolean_t do_headroom_boost = *headroom_boost; 5852 5853 ASSERT(dev->l2ad_vdev != NULL); 5854 5855 /* Lower the flag now, we might want to raise it again later. */ 5856 *headroom_boost = B_FALSE; 5857 5858 pio = NULL; 5859 write_sz = write_asize = write_psize = 0; 5860 full = B_FALSE; 5861 head = kmem_cache_alloc(hdr_l2only_cache, KM_PUSHPAGE); 5862 head->b_flags |= ARC_FLAG_L2_WRITE_HEAD; 5863 head->b_flags |= ARC_FLAG_HAS_L2HDR; 5864 5865 /* 5866 * We will want to try to compress buffers that are at least 2x the 5867 * device sector size. 5868 */ 5869 buf_compress_minsz = 2 << dev->l2ad_vdev->vdev_ashift; 5870 5871 /* 5872 * Copy buffers for L2ARC writing. 5873 */ 5874 for (int try = 0; try <= 3; try++) { 5875 multilist_sublist_t *mls = l2arc_sublist_lock(try); 5876 uint64_t passed_sz = 0; 5877 5878 /* 5879 * L2ARC fast warmup. 5880 * 5881 * Until the ARC is warm and starts to evict, read from the 5882 * head of the ARC lists rather than the tail. 5883 */ 5884 if (arc_warm == B_FALSE) 5885 hdr = multilist_sublist_head(mls); 5886 else 5887 hdr = multilist_sublist_tail(mls); 5888 5889 headroom = target_sz * l2arc_headroom; 5890 if (do_headroom_boost) 5891 headroom = (headroom * l2arc_headroom_boost) / 100; 5892 5893 for (; hdr; hdr = hdr_prev) { 5894 kmutex_t *hash_lock; 5895 uint64_t buf_sz; 5896 5897 if (arc_warm == B_FALSE) 5898 hdr_prev = multilist_sublist_next(mls, hdr); 5899 else 5900 hdr_prev = multilist_sublist_prev(mls, hdr); 5901 5902 hash_lock = HDR_LOCK(hdr); 5903 if (!mutex_tryenter(hash_lock)) { 5904 /* 5905 * Skip this buffer rather than waiting. 5906 */ 5907 continue; 5908 } 5909 5910 passed_sz += hdr->b_size; 5911 if (passed_sz > headroom) { 5912 /* 5913 * Searched too far. 5914 */ 5915 mutex_exit(hash_lock); 5916 break; 5917 } 5918 5919 if (!l2arc_write_eligible(guid, hdr)) { 5920 mutex_exit(hash_lock); 5921 continue; 5922 } 5923 5924 if ((write_sz + hdr->b_size) > target_sz) { 5925 full = B_TRUE; 5926 mutex_exit(hash_lock); 5927 break; 5928 } 5929 5930 if (pio == NULL) { 5931 /* 5932 * Insert a dummy header on the buflist so 5933 * l2arc_write_done() can find where the 5934 * write buffers begin without searching. 5935 */ 5936 mutex_enter(&dev->l2ad_mtx); 5937 list_insert_head(&dev->l2ad_buflist, head); 5938 mutex_exit(&dev->l2ad_mtx); 5939 5940 cb = kmem_alloc( 5941 sizeof (l2arc_write_callback_t), KM_SLEEP); 5942 cb->l2wcb_dev = dev; 5943 cb->l2wcb_head = head; 5944 pio = zio_root(spa, l2arc_write_done, cb, 5945 ZIO_FLAG_CANFAIL); 5946 } 5947 5948 /* 5949 * Create and add a new L2ARC header. 5950 */ 5951 hdr->b_l2hdr.b_dev = dev; 5952 hdr->b_flags |= ARC_FLAG_L2_WRITING; 5953 /* 5954 * Temporarily stash the data buffer in b_tmp_cdata. 5955 * The subsequent write step will pick it up from 5956 * there. This is because can't access b_l1hdr.b_buf 5957 * without holding the hash_lock, which we in turn 5958 * can't access without holding the ARC list locks 5959 * (which we want to avoid during compression/writing). 5960 */ 5961 HDR_SET_COMPRESS(hdr, ZIO_COMPRESS_OFF); 5962 hdr->b_l2hdr.b_asize = hdr->b_size; 5963 hdr->b_l1hdr.b_tmp_cdata = hdr->b_l1hdr.b_buf->b_data; 5964 5965 /* 5966 * Explicitly set the b_daddr field to a known 5967 * value which means "invalid address". This 5968 * enables us to differentiate which stage of 5969 * l2arc_write_buffers() the particular header 5970 * is in (e.g. this loop, or the one below). 5971 * ARC_FLAG_L2_WRITING is not enough to make 5972 * this distinction, and we need to know in 5973 * order to do proper l2arc vdev accounting in 5974 * arc_release() and arc_hdr_destroy(). 5975 * 5976 * Note, we can't use a new flag to distinguish 5977 * the two stages because we don't hold the 5978 * header's hash_lock below, in the second stage 5979 * of this function. Thus, we can't simply 5980 * change the b_flags field to denote that the 5981 * IO has been sent. We can change the b_daddr 5982 * field of the L2 portion, though, since we'll 5983 * be holding the l2ad_mtx; which is why we're 5984 * using it to denote the header's state change. 5985 */ 5986 hdr->b_l2hdr.b_daddr = L2ARC_ADDR_UNSET; 5987 5988 buf_sz = hdr->b_size; 5989 hdr->b_flags |= ARC_FLAG_HAS_L2HDR; 5990 5991 mutex_enter(&dev->l2ad_mtx); 5992 list_insert_head(&dev->l2ad_buflist, hdr); 5993 mutex_exit(&dev->l2ad_mtx); 5994 5995 /* 5996 * Compute and store the buffer cksum before 5997 * writing. On debug the cksum is verified first. 5998 */ 5999 arc_cksum_verify(hdr->b_l1hdr.b_buf); 6000 arc_cksum_compute(hdr->b_l1hdr.b_buf, B_TRUE); 6001 6002 mutex_exit(hash_lock); 6003 6004 write_sz += buf_sz; 6005 } 6006 6007 multilist_sublist_unlock(mls); 6008 6009 if (full == B_TRUE) 6010 break; 6011 } 6012 6013 /* No buffers selected for writing? */ 6014 if (pio == NULL) { 6015 ASSERT0(write_sz); 6016 ASSERT(!HDR_HAS_L1HDR(head)); 6017 kmem_cache_free(hdr_l2only_cache, head); 6018 return (0); 6019 } 6020 6021 mutex_enter(&dev->l2ad_mtx); 6022 6023 /* 6024 * Now start writing the buffers. We're starting at the write head 6025 * and work backwards, retracing the course of the buffer selector 6026 * loop above. 6027 */ 6028 for (hdr = list_prev(&dev->l2ad_buflist, head); hdr; 6029 hdr = list_prev(&dev->l2ad_buflist, hdr)) { 6030 uint64_t buf_sz; 6031 6032 /* 6033 * We rely on the L1 portion of the header below, so 6034 * it's invalid for this header to have been evicted out 6035 * of the ghost cache, prior to being written out. The 6036 * ARC_FLAG_L2_WRITING bit ensures this won't happen. 6037 */ 6038 ASSERT(HDR_HAS_L1HDR(hdr)); 6039 6040 /* 6041 * We shouldn't need to lock the buffer here, since we flagged 6042 * it as ARC_FLAG_L2_WRITING in the previous step, but we must 6043 * take care to only access its L2 cache parameters. In 6044 * particular, hdr->l1hdr.b_buf may be invalid by now due to 6045 * ARC eviction. 6046 */ 6047 hdr->b_l2hdr.b_daddr = dev->l2ad_hand; 6048 6049 if ((HDR_L2COMPRESS(hdr)) && 6050 hdr->b_l2hdr.b_asize >= buf_compress_minsz) { 6051 if (l2arc_compress_buf(hdr)) { 6052 /* 6053 * If compression succeeded, enable headroom 6054 * boost on the next scan cycle. 6055 */ 6056 *headroom_boost = B_TRUE; 6057 } 6058 } 6059 6060 /* 6061 * Pick up the buffer data we had previously stashed away 6062 * (and now potentially also compressed). 6063 */ 6064 buf_data = hdr->b_l1hdr.b_tmp_cdata; 6065 buf_sz = hdr->b_l2hdr.b_asize; 6066 6067 /* 6068 * We need to do this regardless if buf_sz is zero or 6069 * not, otherwise, when this l2hdr is evicted we'll 6070 * remove a reference that was never added. 6071 */ 6072 (void) refcount_add_many(&dev->l2ad_alloc, buf_sz, hdr); 6073 6074 /* Compression may have squashed the buffer to zero length. */ 6075 if (buf_sz != 0) { 6076 uint64_t buf_p_sz; 6077 6078 wzio = zio_write_phys(pio, dev->l2ad_vdev, 6079 dev->l2ad_hand, buf_sz, buf_data, ZIO_CHECKSUM_OFF, 6080 NULL, NULL, ZIO_PRIORITY_ASYNC_WRITE, 6081 ZIO_FLAG_CANFAIL, B_FALSE); 6082 6083 DTRACE_PROBE2(l2arc__write, vdev_t *, dev->l2ad_vdev, 6084 zio_t *, wzio); 6085 (void) zio_nowait(wzio); 6086 6087 write_asize += buf_sz; 6088 6089 /* 6090 * Keep the clock hand suitably device-aligned. 6091 */ 6092 buf_p_sz = vdev_psize_to_asize(dev->l2ad_vdev, buf_sz); 6093 write_psize += buf_p_sz; 6094 dev->l2ad_hand += buf_p_sz; 6095 } 6096 } 6097 6098 mutex_exit(&dev->l2ad_mtx); 6099 6100 ASSERT3U(write_asize, <=, target_sz); 6101 ARCSTAT_BUMP(arcstat_l2_writes_sent); 6102 ARCSTAT_INCR(arcstat_l2_write_bytes, write_asize); 6103 ARCSTAT_INCR(arcstat_l2_size, write_sz); 6104 ARCSTAT_INCR(arcstat_l2_asize, write_asize); 6105 vdev_space_update(dev->l2ad_vdev, write_asize, 0, 0); 6106 6107 /* 6108 * Bump device hand to the device start if it is approaching the end. 6109 * l2arc_evict() will already have evicted ahead for this case. 6110 */ 6111 if (dev->l2ad_hand >= (dev->l2ad_end - target_sz)) { 6112 dev->l2ad_hand = dev->l2ad_start; 6113 dev->l2ad_first = B_FALSE; 6114 } 6115 6116 dev->l2ad_writing = B_TRUE; 6117 (void) zio_wait(pio); 6118 dev->l2ad_writing = B_FALSE; 6119 6120 return (write_asize); 6121 } 6122 6123 /* 6124 * Compresses an L2ARC buffer. 6125 * The data to be compressed must be prefilled in l1hdr.b_tmp_cdata and its 6126 * size in l2hdr->b_asize. This routine tries to compress the data and 6127 * depending on the compression result there are three possible outcomes: 6128 * *) The buffer was incompressible. The original l2hdr contents were left 6129 * untouched and are ready for writing to an L2 device. 6130 * *) The buffer was all-zeros, so there is no need to write it to an L2 6131 * device. To indicate this situation b_tmp_cdata is NULL'ed, b_asize is 6132 * set to zero and b_compress is set to ZIO_COMPRESS_EMPTY. 6133 * *) Compression succeeded and b_tmp_cdata was replaced with a temporary 6134 * data buffer which holds the compressed data to be written, and b_asize 6135 * tells us how much data there is. b_compress is set to the appropriate 6136 * compression algorithm. Once writing is done, invoke 6137 * l2arc_release_cdata_buf on this l2hdr to free this temporary buffer. 6138 * 6139 * Returns B_TRUE if compression succeeded, or B_FALSE if it didn't (the 6140 * buffer was incompressible). 6141 */ 6142 static boolean_t 6143 l2arc_compress_buf(arc_buf_hdr_t *hdr) 6144 { 6145 void *cdata; 6146 size_t csize, len, rounded; 6147 ASSERT(HDR_HAS_L2HDR(hdr)); 6148 l2arc_buf_hdr_t *l2hdr = &hdr->b_l2hdr; 6149 6150 ASSERT(HDR_HAS_L1HDR(hdr)); 6151 ASSERT(HDR_GET_COMPRESS(hdr) == ZIO_COMPRESS_OFF); 6152 ASSERT(hdr->b_l1hdr.b_tmp_cdata != NULL); 6153 6154 len = l2hdr->b_asize; 6155 cdata = zio_data_buf_alloc(len); 6156 ASSERT3P(cdata, !=, NULL); 6157 csize = zio_compress_data(ZIO_COMPRESS_LZ4, hdr->b_l1hdr.b_tmp_cdata, 6158 cdata, l2hdr->b_asize); 6159 6160 rounded = P2ROUNDUP(csize, (size_t)SPA_MINBLOCKSIZE); 6161 if (rounded > csize) { 6162 bzero((char *)cdata + csize, rounded - csize); 6163 csize = rounded; 6164 } 6165 6166 if (csize == 0) { 6167 /* zero block, indicate that there's nothing to write */ 6168 zio_data_buf_free(cdata, len); 6169 HDR_SET_COMPRESS(hdr, ZIO_COMPRESS_EMPTY); 6170 l2hdr->b_asize = 0; 6171 hdr->b_l1hdr.b_tmp_cdata = NULL; 6172 ARCSTAT_BUMP(arcstat_l2_compress_zeros); 6173 return (B_TRUE); 6174 } else if (csize > 0 && csize < len) { 6175 /* 6176 * Compression succeeded, we'll keep the cdata around for 6177 * writing and release it afterwards. 6178 */ 6179 HDR_SET_COMPRESS(hdr, ZIO_COMPRESS_LZ4); 6180 l2hdr->b_asize = csize; 6181 hdr->b_l1hdr.b_tmp_cdata = cdata; 6182 ARCSTAT_BUMP(arcstat_l2_compress_successes); 6183 return (B_TRUE); 6184 } else { 6185 /* 6186 * Compression failed, release the compressed buffer. 6187 * l2hdr will be left unmodified. 6188 */ 6189 zio_data_buf_free(cdata, len); 6190 ARCSTAT_BUMP(arcstat_l2_compress_failures); 6191 return (B_FALSE); 6192 } 6193 } 6194 6195 /* 6196 * Decompresses a zio read back from an l2arc device. On success, the 6197 * underlying zio's io_data buffer is overwritten by the uncompressed 6198 * version. On decompression error (corrupt compressed stream), the 6199 * zio->io_error value is set to signal an I/O error. 6200 * 6201 * Please note that the compressed data stream is not checksummed, so 6202 * if the underlying device is experiencing data corruption, we may feed 6203 * corrupt data to the decompressor, so the decompressor needs to be 6204 * able to handle this situation (LZ4 does). 6205 */ 6206 static void 6207 l2arc_decompress_zio(zio_t *zio, arc_buf_hdr_t *hdr, enum zio_compress c) 6208 { 6209 ASSERT(L2ARC_IS_VALID_COMPRESS(c)); 6210 6211 if (zio->io_error != 0) { 6212 /* 6213 * An io error has occured, just restore the original io 6214 * size in preparation for a main pool read. 6215 */ 6216 zio->io_orig_size = zio->io_size = hdr->b_size; 6217 return; 6218 } 6219 6220 if (c == ZIO_COMPRESS_EMPTY) { 6221 /* 6222 * An empty buffer results in a null zio, which means we 6223 * need to fill its io_data after we're done restoring the 6224 * buffer's contents. 6225 */ 6226 ASSERT(hdr->b_l1hdr.b_buf != NULL); 6227 bzero(hdr->b_l1hdr.b_buf->b_data, hdr->b_size); 6228 zio->io_data = zio->io_orig_data = hdr->b_l1hdr.b_buf->b_data; 6229 } else { 6230 ASSERT(zio->io_data != NULL); 6231 /* 6232 * We copy the compressed data from the start of the arc buffer 6233 * (the zio_read will have pulled in only what we need, the 6234 * rest is garbage which we will overwrite at decompression) 6235 * and then decompress back to the ARC data buffer. This way we 6236 * can minimize copying by simply decompressing back over the 6237 * original compressed data (rather than decompressing to an 6238 * aux buffer and then copying back the uncompressed buffer, 6239 * which is likely to be much larger). 6240 */ 6241 uint64_t csize; 6242 void *cdata; 6243 6244 csize = zio->io_size; 6245 cdata = zio_data_buf_alloc(csize); 6246 bcopy(zio->io_data, cdata, csize); 6247 if (zio_decompress_data(c, cdata, zio->io_data, csize, 6248 hdr->b_size) != 0) 6249 zio->io_error = EIO; 6250 zio_data_buf_free(cdata, csize); 6251 } 6252 6253 /* Restore the expected uncompressed IO size. */ 6254 zio->io_orig_size = zio->io_size = hdr->b_size; 6255 } 6256 6257 /* 6258 * Releases the temporary b_tmp_cdata buffer in an l2arc header structure. 6259 * This buffer serves as a temporary holder of compressed data while 6260 * the buffer entry is being written to an l2arc device. Once that is 6261 * done, we can dispose of it. 6262 */ 6263 static void 6264 l2arc_release_cdata_buf(arc_buf_hdr_t *hdr) 6265 { 6266 enum zio_compress comp = HDR_GET_COMPRESS(hdr); 6267 6268 ASSERT(HDR_HAS_L1HDR(hdr)); 6269 ASSERT(comp == ZIO_COMPRESS_OFF || L2ARC_IS_VALID_COMPRESS(comp)); 6270 6271 if (comp == ZIO_COMPRESS_OFF) { 6272 /* 6273 * In this case, b_tmp_cdata points to the same buffer 6274 * as the arc_buf_t's b_data field. We don't want to 6275 * free it, since the arc_buf_t will handle that. 6276 */ 6277 hdr->b_l1hdr.b_tmp_cdata = NULL; 6278 } else if (comp == ZIO_COMPRESS_EMPTY) { 6279 /* 6280 * In this case, b_tmp_cdata was compressed to an empty 6281 * buffer, thus there's nothing to free and b_tmp_cdata 6282 * should have been set to NULL in l2arc_write_buffers(). 6283 */ 6284 ASSERT3P(hdr->b_l1hdr.b_tmp_cdata, ==, NULL); 6285 } else { 6286 /* 6287 * If the data was compressed, then we've allocated a 6288 * temporary buffer for it, so now we need to release it. 6289 */ 6290 ASSERT(hdr->b_l1hdr.b_tmp_cdata != NULL); 6291 zio_data_buf_free(hdr->b_l1hdr.b_tmp_cdata, 6292 hdr->b_size); 6293 hdr->b_l1hdr.b_tmp_cdata = NULL; 6294 } 6295 6296 } 6297 6298 /* 6299 * This thread feeds the L2ARC at regular intervals. This is the beating 6300 * heart of the L2ARC. 6301 */ 6302 static void 6303 l2arc_feed_thread(void) 6304 { 6305 callb_cpr_t cpr; 6306 l2arc_dev_t *dev; 6307 spa_t *spa; 6308 uint64_t size, wrote; 6309 clock_t begin, next = ddi_get_lbolt(); 6310 boolean_t headroom_boost = B_FALSE; 6311 6312 CALLB_CPR_INIT(&cpr, &l2arc_feed_thr_lock, callb_generic_cpr, FTAG); 6313 6314 mutex_enter(&l2arc_feed_thr_lock); 6315 6316 while (l2arc_thread_exit == 0) { 6317 CALLB_CPR_SAFE_BEGIN(&cpr); 6318 (void) cv_timedwait(&l2arc_feed_thr_cv, &l2arc_feed_thr_lock, 6319 next); 6320 CALLB_CPR_SAFE_END(&cpr, &l2arc_feed_thr_lock); 6321 next = ddi_get_lbolt() + hz; 6322 6323 /* 6324 * Quick check for L2ARC devices. 6325 */ 6326 mutex_enter(&l2arc_dev_mtx); 6327 if (l2arc_ndev == 0) { 6328 mutex_exit(&l2arc_dev_mtx); 6329 continue; 6330 } 6331 mutex_exit(&l2arc_dev_mtx); 6332 begin = ddi_get_lbolt(); 6333 6334 /* 6335 * This selects the next l2arc device to write to, and in 6336 * doing so the next spa to feed from: dev->l2ad_spa. This 6337 * will return NULL if there are now no l2arc devices or if 6338 * they are all faulted. 6339 * 6340 * If a device is returned, its spa's config lock is also 6341 * held to prevent device removal. l2arc_dev_get_next() 6342 * will grab and release l2arc_dev_mtx. 6343 */ 6344 if ((dev = l2arc_dev_get_next()) == NULL) 6345 continue; 6346 6347 spa = dev->l2ad_spa; 6348 ASSERT(spa != NULL); 6349 6350 /* 6351 * If the pool is read-only then force the feed thread to 6352 * sleep a little longer. 6353 */ 6354 if (!spa_writeable(spa)) { 6355 next = ddi_get_lbolt() + 5 * l2arc_feed_secs * hz; 6356 spa_config_exit(spa, SCL_L2ARC, dev); 6357 continue; 6358 } 6359 6360 /* 6361 * Avoid contributing to memory pressure. 6362 */ 6363 if (arc_reclaim_needed()) { 6364 ARCSTAT_BUMP(arcstat_l2_abort_lowmem); 6365 spa_config_exit(spa, SCL_L2ARC, dev); 6366 continue; 6367 } 6368 6369 ARCSTAT_BUMP(arcstat_l2_feeds); 6370 6371 size = l2arc_write_size(); 6372 6373 /* 6374 * Evict L2ARC buffers that will be overwritten. 6375 */ 6376 l2arc_evict(dev, size, B_FALSE); 6377 6378 /* 6379 * Write ARC buffers. 6380 */ 6381 wrote = l2arc_write_buffers(spa, dev, size, &headroom_boost); 6382 6383 /* 6384 * Calculate interval between writes. 6385 */ 6386 next = l2arc_write_interval(begin, size, wrote); 6387 spa_config_exit(spa, SCL_L2ARC, dev); 6388 } 6389 6390 l2arc_thread_exit = 0; 6391 cv_broadcast(&l2arc_feed_thr_cv); 6392 CALLB_CPR_EXIT(&cpr); /* drops l2arc_feed_thr_lock */ 6393 thread_exit(); 6394 } 6395 6396 boolean_t 6397 l2arc_vdev_present(vdev_t *vd) 6398 { 6399 l2arc_dev_t *dev; 6400 6401 mutex_enter(&l2arc_dev_mtx); 6402 for (dev = list_head(l2arc_dev_list); dev != NULL; 6403 dev = list_next(l2arc_dev_list, dev)) { 6404 if (dev->l2ad_vdev == vd) 6405 break; 6406 } 6407 mutex_exit(&l2arc_dev_mtx); 6408 6409 return (dev != NULL); 6410 } 6411 6412 /* 6413 * Add a vdev for use by the L2ARC. By this point the spa has already 6414 * validated the vdev and opened it. 6415 */ 6416 void 6417 l2arc_add_vdev(spa_t *spa, vdev_t *vd) 6418 { 6419 l2arc_dev_t *adddev; 6420 6421 ASSERT(!l2arc_vdev_present(vd)); 6422 6423 /* 6424 * Create a new l2arc device entry. 6425 */ 6426 adddev = kmem_zalloc(sizeof (l2arc_dev_t), KM_SLEEP); 6427 adddev->l2ad_spa = spa; 6428 adddev->l2ad_vdev = vd; 6429 adddev->l2ad_start = VDEV_LABEL_START_SIZE; 6430 adddev->l2ad_end = VDEV_LABEL_START_SIZE + vdev_get_min_asize(vd); 6431 adddev->l2ad_hand = adddev->l2ad_start; 6432 adddev->l2ad_first = B_TRUE; 6433 adddev->l2ad_writing = B_FALSE; 6434 6435 mutex_init(&adddev->l2ad_mtx, NULL, MUTEX_DEFAULT, NULL); 6436 /* 6437 * This is a list of all ARC buffers that are still valid on the 6438 * device. 6439 */ 6440 list_create(&adddev->l2ad_buflist, sizeof (arc_buf_hdr_t), 6441 offsetof(arc_buf_hdr_t, b_l2hdr.b_l2node)); 6442 6443 vdev_space_update(vd, 0, 0, adddev->l2ad_end - adddev->l2ad_hand); 6444 refcount_create(&adddev->l2ad_alloc); 6445 6446 /* 6447 * Add device to global list 6448 */ 6449 mutex_enter(&l2arc_dev_mtx); 6450 list_insert_head(l2arc_dev_list, adddev); 6451 atomic_inc_64(&l2arc_ndev); 6452 mutex_exit(&l2arc_dev_mtx); 6453 } 6454 6455 /* 6456 * Remove a vdev from the L2ARC. 6457 */ 6458 void 6459 l2arc_remove_vdev(vdev_t *vd) 6460 { 6461 l2arc_dev_t *dev, *nextdev, *remdev = NULL; 6462 6463 /* 6464 * Find the device by vdev 6465 */ 6466 mutex_enter(&l2arc_dev_mtx); 6467 for (dev = list_head(l2arc_dev_list); dev; dev = nextdev) { 6468 nextdev = list_next(l2arc_dev_list, dev); 6469 if (vd == dev->l2ad_vdev) { 6470 remdev = dev; 6471 break; 6472 } 6473 } 6474 ASSERT(remdev != NULL); 6475 6476 /* 6477 * Remove device from global list 6478 */ 6479 list_remove(l2arc_dev_list, remdev); 6480 l2arc_dev_last = NULL; /* may have been invalidated */ 6481 atomic_dec_64(&l2arc_ndev); 6482 mutex_exit(&l2arc_dev_mtx); 6483 6484 /* 6485 * Clear all buflists and ARC references. L2ARC device flush. 6486 */ 6487 l2arc_evict(remdev, 0, B_TRUE); 6488 list_destroy(&remdev->l2ad_buflist); 6489 mutex_destroy(&remdev->l2ad_mtx); 6490 refcount_destroy(&remdev->l2ad_alloc); 6491 kmem_free(remdev, sizeof (l2arc_dev_t)); 6492 } 6493 6494 void 6495 l2arc_init(void) 6496 { 6497 l2arc_thread_exit = 0; 6498 l2arc_ndev = 0; 6499 l2arc_writes_sent = 0; 6500 l2arc_writes_done = 0; 6501 6502 mutex_init(&l2arc_feed_thr_lock, NULL, MUTEX_DEFAULT, NULL); 6503 cv_init(&l2arc_feed_thr_cv, NULL, CV_DEFAULT, NULL); 6504 mutex_init(&l2arc_dev_mtx, NULL, MUTEX_DEFAULT, NULL); 6505 mutex_init(&l2arc_free_on_write_mtx, NULL, MUTEX_DEFAULT, NULL); 6506 6507 l2arc_dev_list = &L2ARC_dev_list; 6508 l2arc_free_on_write = &L2ARC_free_on_write; 6509 list_create(l2arc_dev_list, sizeof (l2arc_dev_t), 6510 offsetof(l2arc_dev_t, l2ad_node)); 6511 list_create(l2arc_free_on_write, sizeof (l2arc_data_free_t), 6512 offsetof(l2arc_data_free_t, l2df_list_node)); 6513 } 6514 6515 void 6516 l2arc_fini(void) 6517 { 6518 /* 6519 * This is called from dmu_fini(), which is called from spa_fini(); 6520 * Because of this, we can assume that all l2arc devices have 6521 * already been removed when the pools themselves were removed. 6522 */ 6523 6524 l2arc_do_free_on_write(); 6525 6526 mutex_destroy(&l2arc_feed_thr_lock); 6527 cv_destroy(&l2arc_feed_thr_cv); 6528 mutex_destroy(&l2arc_dev_mtx); 6529 mutex_destroy(&l2arc_free_on_write_mtx); 6530 6531 list_destroy(l2arc_dev_list); 6532 list_destroy(l2arc_free_on_write); 6533 } 6534 6535 void 6536 l2arc_start(void) 6537 { 6538 if (!(spa_mode_global & FWRITE)) 6539 return; 6540 6541 (void) thread_create(NULL, 0, l2arc_feed_thread, NULL, 0, &p0, 6542 TS_RUN, minclsyspri); 6543 } 6544 6545 void 6546 l2arc_stop(void) 6547 { 6548 if (!(spa_mode_global & FWRITE)) 6549 return; 6550 6551 mutex_enter(&l2arc_feed_thr_lock); 6552 cv_signal(&l2arc_feed_thr_cv); /* kick thread out of startup */ 6553 l2arc_thread_exit = 1; 6554 while (l2arc_thread_exit != 0) 6555 cv_wait(&l2arc_feed_thr_cv, &l2arc_feed_thr_lock); 6556 mutex_exit(&l2arc_feed_thr_lock); 6557 } 6558