1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _BCACHE_H 3 #define _BCACHE_H 4 5 /* 6 * SOME HIGH LEVEL CODE DOCUMENTATION: 7 * 8 * Bcache mostly works with cache sets, cache devices, and backing devices. 9 * 10 * Support for multiple cache devices hasn't quite been finished off yet, but 11 * it's about 95% plumbed through. A cache set and its cache devices is sort of 12 * like a md raid array and its component devices. Most of the code doesn't care 13 * about individual cache devices, the main abstraction is the cache set. 14 * 15 * Multiple cache devices is intended to give us the ability to mirror dirty 16 * cached data and metadata, without mirroring clean cached data. 17 * 18 * Backing devices are different, in that they have a lifetime independent of a 19 * cache set. When you register a newly formatted backing device it'll come up 20 * in passthrough mode, and then you can attach and detach a backing device from 21 * a cache set at runtime - while it's mounted and in use. Detaching implicitly 22 * invalidates any cached data for that backing device. 23 * 24 * A cache set can have multiple (many) backing devices attached to it. 25 * 26 * There's also flash only volumes - this is the reason for the distinction 27 * between struct cached_dev and struct bcache_device. A flash only volume 28 * works much like a bcache device that has a backing device, except the 29 * "cached" data is always dirty. The end result is that we get thin 30 * provisioning with very little additional code. 31 * 32 * Flash only volumes work but they're not production ready because the moving 33 * garbage collector needs more work. More on that later. 34 * 35 * BUCKETS/ALLOCATION: 36 * 37 * Bcache is primarily designed for caching, which means that in normal 38 * operation all of our available space will be allocated. Thus, we need an 39 * efficient way of deleting things from the cache so we can write new things to 40 * it. 41 * 42 * To do this, we first divide the cache device up into buckets. A bucket is the 43 * unit of allocation; they're typically around 1 mb - anywhere from 128k to 2M+ 44 * works efficiently. 45 * 46 * Each bucket has a 16 bit priority, and an 8 bit generation associated with 47 * it. The gens and priorities for all the buckets are stored contiguously and 48 * packed on disk (in a linked list of buckets - aside from the superblock, all 49 * of bcache's metadata is stored in buckets). 50 * 51 * The priority is used to implement an LRU. We reset a bucket's priority when 52 * we allocate it or on cache it, and every so often we decrement the priority 53 * of each bucket. It could be used to implement something more sophisticated, 54 * if anyone ever gets around to it. 55 * 56 * The generation is used for invalidating buckets. Each pointer also has an 8 57 * bit generation embedded in it; for a pointer to be considered valid, its gen 58 * must match the gen of the bucket it points into. Thus, to reuse a bucket all 59 * we have to do is increment its gen (and write its new gen to disk; we batch 60 * this up). 61 * 62 * Bcache is entirely COW - we never write twice to a bucket, even buckets that 63 * contain metadata (including btree nodes). 64 * 65 * THE BTREE: 66 * 67 * Bcache is in large part design around the btree. 68 * 69 * At a high level, the btree is just an index of key -> ptr tuples. 70 * 71 * Keys represent extents, and thus have a size field. Keys also have a variable 72 * number of pointers attached to them (potentially zero, which is handy for 73 * invalidating the cache). 74 * 75 * The key itself is an inode:offset pair. The inode number corresponds to a 76 * backing device or a flash only volume. The offset is the ending offset of the 77 * extent within the inode - not the starting offset; this makes lookups 78 * slightly more convenient. 79 * 80 * Pointers contain the cache device id, the offset on that device, and an 8 bit 81 * generation number. More on the gen later. 82 * 83 * Index lookups are not fully abstracted - cache lookups in particular are 84 * still somewhat mixed in with the btree code, but things are headed in that 85 * direction. 86 * 87 * Updates are fairly well abstracted, though. There are two different ways of 88 * updating the btree; insert and replace. 89 * 90 * BTREE_INSERT will just take a list of keys and insert them into the btree - 91 * overwriting (possibly only partially) any extents they overlap with. This is 92 * used to update the index after a write. 93 * 94 * BTREE_REPLACE is really cmpxchg(); it inserts a key into the btree iff it is 95 * overwriting a key that matches another given key. This is used for inserting 96 * data into the cache after a cache miss, and for background writeback, and for 97 * the moving garbage collector. 98 * 99 * There is no "delete" operation; deleting things from the index is 100 * accomplished by either by invalidating pointers (by incrementing a bucket's 101 * gen) or by inserting a key with 0 pointers - which will overwrite anything 102 * previously present at that location in the index. 103 * 104 * This means that there are always stale/invalid keys in the btree. They're 105 * filtered out by the code that iterates through a btree node, and removed when 106 * a btree node is rewritten. 107 * 108 * BTREE NODES: 109 * 110 * Our unit of allocation is a bucket, and we we can't arbitrarily allocate and 111 * free smaller than a bucket - so, that's how big our btree nodes are. 112 * 113 * (If buckets are really big we'll only use part of the bucket for a btree node 114 * - no less than 1/4th - but a bucket still contains no more than a single 115 * btree node. I'd actually like to change this, but for now we rely on the 116 * bucket's gen for deleting btree nodes when we rewrite/split a node.) 117 * 118 * Anyways, btree nodes are big - big enough to be inefficient with a textbook 119 * btree implementation. 120 * 121 * The way this is solved is that btree nodes are internally log structured; we 122 * can append new keys to an existing btree node without rewriting it. This 123 * means each set of keys we write is sorted, but the node is not. 124 * 125 * We maintain this log structure in memory - keeping 1Mb of keys sorted would 126 * be expensive, and we have to distinguish between the keys we have written and 127 * the keys we haven't. So to do a lookup in a btree node, we have to search 128 * each sorted set. But we do merge written sets together lazily, so the cost of 129 * these extra searches is quite low (normally most of the keys in a btree node 130 * will be in one big set, and then there'll be one or two sets that are much 131 * smaller). 132 * 133 * This log structure makes bcache's btree more of a hybrid between a 134 * conventional btree and a compacting data structure, with some of the 135 * advantages of both. 136 * 137 * GARBAGE COLLECTION: 138 * 139 * We can't just invalidate any bucket - it might contain dirty data or 140 * metadata. If it once contained dirty data, other writes might overwrite it 141 * later, leaving no valid pointers into that bucket in the index. 142 * 143 * Thus, the primary purpose of garbage collection is to find buckets to reuse. 144 * It also counts how much valid data it each bucket currently contains, so that 145 * allocation can reuse buckets sooner when they've been mostly overwritten. 146 * 147 * It also does some things that are really internal to the btree 148 * implementation. If a btree node contains pointers that are stale by more than 149 * some threshold, it rewrites the btree node to avoid the bucket's generation 150 * wrapping around. It also merges adjacent btree nodes if they're empty enough. 151 * 152 * THE JOURNAL: 153 * 154 * Bcache's journal is not necessary for consistency; we always strictly 155 * order metadata writes so that the btree and everything else is consistent on 156 * disk in the event of an unclean shutdown, and in fact bcache had writeback 157 * caching (with recovery from unclean shutdown) before journalling was 158 * implemented. 159 * 160 * Rather, the journal is purely a performance optimization; we can't complete a 161 * write until we've updated the index on disk, otherwise the cache would be 162 * inconsistent in the event of an unclean shutdown. This means that without the 163 * journal, on random write workloads we constantly have to update all the leaf 164 * nodes in the btree, and those writes will be mostly empty (appending at most 165 * a few keys each) - highly inefficient in terms of amount of metadata writes, 166 * and it puts more strain on the various btree resorting/compacting code. 167 * 168 * The journal is just a log of keys we've inserted; on startup we just reinsert 169 * all the keys in the open journal entries. That means that when we're updating 170 * a node in the btree, we can wait until a 4k block of keys fills up before 171 * writing them out. 172 * 173 * For simplicity, we only journal updates to leaf nodes; updates to parent 174 * nodes are rare enough (since our leaf nodes are huge) that it wasn't worth 175 * the complexity to deal with journalling them (in particular, journal replay) 176 * - updates to non leaf nodes just happen synchronously (see btree_split()). 177 */ 178 179 #define pr_fmt(fmt) "bcache: %s() " fmt "\n", __func__ 180 181 #include <linux/bcache.h> 182 #include <linux/bio.h> 183 #include <linux/kobject.h> 184 #include <linux/list.h> 185 #include <linux/mutex.h> 186 #include <linux/rbtree.h> 187 #include <linux/rwsem.h> 188 #include <linux/types.h> 189 #include <linux/workqueue.h> 190 191 #include "bset.h" 192 #include "util.h" 193 #include "closure.h" 194 195 struct bucket { 196 atomic_t pin; 197 uint16_t prio; 198 uint8_t gen; 199 uint8_t last_gc; /* Most out of date gen in the btree */ 200 uint16_t gc_mark; /* Bitfield used by GC. See below for field */ 201 }; 202 203 /* 204 * I'd use bitfields for these, but I don't trust the compiler not to screw me 205 * as multiple threads touch struct bucket without locking 206 */ 207 208 BITMASK(GC_MARK, struct bucket, gc_mark, 0, 2); 209 #define GC_MARK_RECLAIMABLE 1 210 #define GC_MARK_DIRTY 2 211 #define GC_MARK_METADATA 3 212 #define GC_SECTORS_USED_SIZE 13 213 #define MAX_GC_SECTORS_USED (~(~0ULL << GC_SECTORS_USED_SIZE)) 214 BITMASK(GC_SECTORS_USED, struct bucket, gc_mark, 2, GC_SECTORS_USED_SIZE); 215 BITMASK(GC_MOVE, struct bucket, gc_mark, 15, 1); 216 217 #include "journal.h" 218 #include "stats.h" 219 struct search; 220 struct btree; 221 struct keybuf; 222 223 struct keybuf_key { 224 struct rb_node node; 225 BKEY_PADDED(key); 226 void *private; 227 }; 228 229 struct keybuf { 230 struct bkey last_scanned; 231 spinlock_t lock; 232 233 /* 234 * Beginning and end of range in rb tree - so that we can skip taking 235 * lock and checking the rb tree when we need to check for overlapping 236 * keys. 237 */ 238 struct bkey start; 239 struct bkey end; 240 241 struct rb_root keys; 242 243 #define KEYBUF_NR 500 244 DECLARE_ARRAY_ALLOCATOR(struct keybuf_key, freelist, KEYBUF_NR); 245 }; 246 247 struct bcache_device { 248 struct closure cl; 249 250 struct kobject kobj; 251 252 struct cache_set *c; 253 unsigned id; 254 #define BCACHEDEVNAME_SIZE 12 255 char name[BCACHEDEVNAME_SIZE]; 256 257 struct gendisk *disk; 258 259 unsigned long flags; 260 #define BCACHE_DEV_CLOSING 0 261 #define BCACHE_DEV_DETACHING 1 262 #define BCACHE_DEV_UNLINK_DONE 2 263 264 unsigned nr_stripes; 265 unsigned stripe_size; 266 atomic_t *stripe_sectors_dirty; 267 unsigned long *full_dirty_stripes; 268 269 unsigned long sectors_dirty_last; 270 long sectors_dirty_derivative; 271 272 struct bio_set *bio_split; 273 274 unsigned data_csum:1; 275 276 int (*cache_miss)(struct btree *, struct search *, 277 struct bio *, unsigned); 278 int (*ioctl) (struct bcache_device *, fmode_t, unsigned, unsigned long); 279 }; 280 281 struct io { 282 /* Used to track sequential IO so it can be skipped */ 283 struct hlist_node hash; 284 struct list_head lru; 285 286 unsigned long jiffies; 287 unsigned sequential; 288 sector_t last; 289 }; 290 291 struct cached_dev { 292 struct list_head list; 293 struct bcache_device disk; 294 struct block_device *bdev; 295 296 struct cache_sb sb; 297 struct bio sb_bio; 298 struct bio_vec sb_bv[1]; 299 struct closure sb_write; 300 struct semaphore sb_write_mutex; 301 302 /* Refcount on the cache set. Always nonzero when we're caching. */ 303 atomic_t count; 304 struct work_struct detach; 305 306 /* 307 * Device might not be running if it's dirty and the cache set hasn't 308 * showed up yet. 309 */ 310 atomic_t running; 311 312 /* 313 * Writes take a shared lock from start to finish; scanning for dirty 314 * data to refill the rb tree requires an exclusive lock. 315 */ 316 struct rw_semaphore writeback_lock; 317 318 /* 319 * Nonzero, and writeback has a refcount (d->count), iff there is dirty 320 * data in the cache. Protected by writeback_lock; must have an 321 * shared lock to set and exclusive lock to clear. 322 */ 323 atomic_t has_dirty; 324 325 struct bch_ratelimit writeback_rate; 326 struct delayed_work writeback_rate_update; 327 328 /* 329 * Internal to the writeback code, so read_dirty() can keep track of 330 * where it's at. 331 */ 332 sector_t last_read; 333 334 /* Limit number of writeback bios in flight */ 335 struct semaphore in_flight; 336 struct task_struct *writeback_thread; 337 struct workqueue_struct *writeback_write_wq; 338 339 struct keybuf writeback_keys; 340 341 /* For tracking sequential IO */ 342 #define RECENT_IO_BITS 7 343 #define RECENT_IO (1 << RECENT_IO_BITS) 344 struct io io[RECENT_IO]; 345 struct hlist_head io_hash[RECENT_IO + 1]; 346 struct list_head io_lru; 347 spinlock_t io_lock; 348 349 struct cache_accounting accounting; 350 351 /* The rest of this all shows up in sysfs */ 352 unsigned sequential_cutoff; 353 unsigned readahead; 354 355 unsigned verify:1; 356 unsigned bypass_torture_test:1; 357 358 unsigned partial_stripes_expensive:1; 359 unsigned writeback_metadata:1; 360 unsigned writeback_running:1; 361 unsigned char writeback_percent; 362 unsigned writeback_delay; 363 364 uint64_t writeback_rate_target; 365 int64_t writeback_rate_proportional; 366 int64_t writeback_rate_derivative; 367 int64_t writeback_rate_change; 368 369 unsigned writeback_rate_update_seconds; 370 unsigned writeback_rate_d_term; 371 unsigned writeback_rate_p_term_inverse; 372 }; 373 374 enum alloc_reserve { 375 RESERVE_BTREE, 376 RESERVE_PRIO, 377 RESERVE_MOVINGGC, 378 RESERVE_NONE, 379 RESERVE_NR, 380 }; 381 382 struct cache { 383 struct cache_set *set; 384 struct cache_sb sb; 385 struct bio sb_bio; 386 struct bio_vec sb_bv[1]; 387 388 struct kobject kobj; 389 struct block_device *bdev; 390 391 struct task_struct *alloc_thread; 392 393 struct closure prio; 394 struct prio_set *disk_buckets; 395 396 /* 397 * When allocating new buckets, prio_write() gets first dibs - since we 398 * may not be allocate at all without writing priorities and gens. 399 * prio_buckets[] contains the last buckets we wrote priorities to (so 400 * gc can mark them as metadata), prio_next[] contains the buckets 401 * allocated for the next prio write. 402 */ 403 uint64_t *prio_buckets; 404 uint64_t *prio_last_buckets; 405 406 /* 407 * free: Buckets that are ready to be used 408 * 409 * free_inc: Incoming buckets - these are buckets that currently have 410 * cached data in them, and we can't reuse them until after we write 411 * their new gen to disk. After prio_write() finishes writing the new 412 * gens/prios, they'll be moved to the free list (and possibly discarded 413 * in the process) 414 */ 415 DECLARE_FIFO(long, free)[RESERVE_NR]; 416 DECLARE_FIFO(long, free_inc); 417 418 size_t fifo_last_bucket; 419 420 /* Allocation stuff: */ 421 struct bucket *buckets; 422 423 DECLARE_HEAP(struct bucket *, heap); 424 425 /* 426 * If nonzero, we know we aren't going to find any buckets to invalidate 427 * until a gc finishes - otherwise we could pointlessly burn a ton of 428 * cpu 429 */ 430 unsigned invalidate_needs_gc; 431 432 bool discard; /* Get rid of? */ 433 434 struct journal_device journal; 435 436 /* The rest of this all shows up in sysfs */ 437 #define IO_ERROR_SHIFT 20 438 atomic_t io_errors; 439 atomic_t io_count; 440 441 atomic_long_t meta_sectors_written; 442 atomic_long_t btree_sectors_written; 443 atomic_long_t sectors_written; 444 }; 445 446 struct gc_stat { 447 size_t nodes; 448 size_t key_bytes; 449 450 size_t nkeys; 451 uint64_t data; /* sectors */ 452 unsigned in_use; /* percent */ 453 }; 454 455 /* 456 * Flag bits, for how the cache set is shutting down, and what phase it's at: 457 * 458 * CACHE_SET_UNREGISTERING means we're not just shutting down, we're detaching 459 * all the backing devices first (their cached data gets invalidated, and they 460 * won't automatically reattach). 461 * 462 * CACHE_SET_STOPPING always gets set first when we're closing down a cache set; 463 * we'll continue to run normally for awhile with CACHE_SET_STOPPING set (i.e. 464 * flushing dirty data). 465 * 466 * CACHE_SET_RUNNING means all cache devices have been registered and journal 467 * replay is complete. 468 */ 469 #define CACHE_SET_UNREGISTERING 0 470 #define CACHE_SET_STOPPING 1 471 #define CACHE_SET_RUNNING 2 472 473 struct cache_set { 474 struct closure cl; 475 476 struct list_head list; 477 struct kobject kobj; 478 struct kobject internal; 479 struct dentry *debug; 480 struct cache_accounting accounting; 481 482 unsigned long flags; 483 484 struct cache_sb sb; 485 486 struct cache *cache[MAX_CACHES_PER_SET]; 487 struct cache *cache_by_alloc[MAX_CACHES_PER_SET]; 488 int caches_loaded; 489 490 struct bcache_device **devices; 491 struct list_head cached_devs; 492 uint64_t cached_dev_sectors; 493 struct closure caching; 494 495 struct closure sb_write; 496 struct semaphore sb_write_mutex; 497 498 mempool_t *search; 499 mempool_t *bio_meta; 500 struct bio_set *bio_split; 501 502 /* For the btree cache */ 503 struct shrinker shrink; 504 505 /* For the btree cache and anything allocation related */ 506 struct mutex bucket_lock; 507 508 /* log2(bucket_size), in sectors */ 509 unsigned short bucket_bits; 510 511 /* log2(block_size), in sectors */ 512 unsigned short block_bits; 513 514 /* 515 * Default number of pages for a new btree node - may be less than a 516 * full bucket 517 */ 518 unsigned btree_pages; 519 520 /* 521 * Lists of struct btrees; lru is the list for structs that have memory 522 * allocated for actual btree node, freed is for structs that do not. 523 * 524 * We never free a struct btree, except on shutdown - we just put it on 525 * the btree_cache_freed list and reuse it later. This simplifies the 526 * code, and it doesn't cost us much memory as the memory usage is 527 * dominated by buffers that hold the actual btree node data and those 528 * can be freed - and the number of struct btrees allocated is 529 * effectively bounded. 530 * 531 * btree_cache_freeable effectively is a small cache - we use it because 532 * high order page allocations can be rather expensive, and it's quite 533 * common to delete and allocate btree nodes in quick succession. It 534 * should never grow past ~2-3 nodes in practice. 535 */ 536 struct list_head btree_cache; 537 struct list_head btree_cache_freeable; 538 struct list_head btree_cache_freed; 539 540 /* Number of elements in btree_cache + btree_cache_freeable lists */ 541 unsigned btree_cache_used; 542 543 /* 544 * If we need to allocate memory for a new btree node and that 545 * allocation fails, we can cannibalize another node in the btree cache 546 * to satisfy the allocation - lock to guarantee only one thread does 547 * this at a time: 548 */ 549 wait_queue_head_t btree_cache_wait; 550 struct task_struct *btree_cache_alloc_lock; 551 552 /* 553 * When we free a btree node, we increment the gen of the bucket the 554 * node is in - but we can't rewrite the prios and gens until we 555 * finished whatever it is we were doing, otherwise after a crash the 556 * btree node would be freed but for say a split, we might not have the 557 * pointers to the new nodes inserted into the btree yet. 558 * 559 * This is a refcount that blocks prio_write() until the new keys are 560 * written. 561 */ 562 atomic_t prio_blocked; 563 wait_queue_head_t bucket_wait; 564 565 /* 566 * For any bio we don't skip we subtract the number of sectors from 567 * rescale; when it hits 0 we rescale all the bucket priorities. 568 */ 569 atomic_t rescale; 570 /* 571 * When we invalidate buckets, we use both the priority and the amount 572 * of good data to determine which buckets to reuse first - to weight 573 * those together consistently we keep track of the smallest nonzero 574 * priority of any bucket. 575 */ 576 uint16_t min_prio; 577 578 /* 579 * max(gen - last_gc) for all buckets. When it gets too big we have to gc 580 * to keep gens from wrapping around. 581 */ 582 uint8_t need_gc; 583 struct gc_stat gc_stats; 584 size_t nbuckets; 585 586 struct task_struct *gc_thread; 587 /* Where in the btree gc currently is */ 588 struct bkey gc_done; 589 590 /* 591 * The allocation code needs gc_mark in struct bucket to be correct, but 592 * it's not while a gc is in progress. Protected by bucket_lock. 593 */ 594 int gc_mark_valid; 595 596 /* Counts how many sectors bio_insert has added to the cache */ 597 atomic_t sectors_to_gc; 598 wait_queue_head_t gc_wait; 599 600 struct keybuf moving_gc_keys; 601 /* Number of moving GC bios in flight */ 602 struct semaphore moving_in_flight; 603 604 struct workqueue_struct *moving_gc_wq; 605 606 struct btree *root; 607 608 #ifdef CONFIG_BCACHE_DEBUG 609 struct btree *verify_data; 610 struct bset *verify_ondisk; 611 struct mutex verify_lock; 612 #endif 613 614 unsigned nr_uuids; 615 struct uuid_entry *uuids; 616 BKEY_PADDED(uuid_bucket); 617 struct closure uuid_write; 618 struct semaphore uuid_write_mutex; 619 620 /* 621 * A btree node on disk could have too many bsets for an iterator to fit 622 * on the stack - have to dynamically allocate them 623 */ 624 mempool_t *fill_iter; 625 626 struct bset_sort_state sort; 627 628 /* List of buckets we're currently writing data to */ 629 struct list_head data_buckets; 630 spinlock_t data_bucket_lock; 631 632 struct journal journal; 633 634 #define CONGESTED_MAX 1024 635 unsigned congested_last_us; 636 atomic_t congested; 637 638 /* The rest of this all shows up in sysfs */ 639 unsigned congested_read_threshold_us; 640 unsigned congested_write_threshold_us; 641 642 struct time_stats btree_gc_time; 643 struct time_stats btree_split_time; 644 struct time_stats btree_read_time; 645 646 atomic_long_t cache_read_races; 647 atomic_long_t writeback_keys_done; 648 atomic_long_t writeback_keys_failed; 649 650 enum { 651 ON_ERROR_UNREGISTER, 652 ON_ERROR_PANIC, 653 } on_error; 654 unsigned error_limit; 655 unsigned error_decay; 656 657 unsigned short journal_delay_ms; 658 bool expensive_debug_checks; 659 unsigned verify:1; 660 unsigned key_merging_disabled:1; 661 unsigned gc_always_rewrite:1; 662 unsigned shrinker_disabled:1; 663 unsigned copy_gc_enabled:1; 664 665 #define BUCKET_HASH_BITS 12 666 struct hlist_head bucket_hash[1 << BUCKET_HASH_BITS]; 667 }; 668 669 struct bbio { 670 unsigned submit_time_us; 671 union { 672 struct bkey key; 673 uint64_t _pad[3]; 674 /* 675 * We only need pad = 3 here because we only ever carry around a 676 * single pointer - i.e. the pointer we're doing io to/from. 677 */ 678 }; 679 struct bio bio; 680 }; 681 682 #define BTREE_PRIO USHRT_MAX 683 #define INITIAL_PRIO 32768U 684 685 #define btree_bytes(c) ((c)->btree_pages * PAGE_SIZE) 686 #define btree_blocks(b) \ 687 ((unsigned) (KEY_SIZE(&b->key) >> (b)->c->block_bits)) 688 689 #define btree_default_blocks(c) \ 690 ((unsigned) ((PAGE_SECTORS * (c)->btree_pages) >> (c)->block_bits)) 691 692 #define bucket_pages(c) ((c)->sb.bucket_size / PAGE_SECTORS) 693 #define bucket_bytes(c) ((c)->sb.bucket_size << 9) 694 #define block_bytes(c) ((c)->sb.block_size << 9) 695 696 #define prios_per_bucket(c) \ 697 ((bucket_bytes(c) - sizeof(struct prio_set)) / \ 698 sizeof(struct bucket_disk)) 699 #define prio_buckets(c) \ 700 DIV_ROUND_UP((size_t) (c)->sb.nbuckets, prios_per_bucket(c)) 701 702 static inline size_t sector_to_bucket(struct cache_set *c, sector_t s) 703 { 704 return s >> c->bucket_bits; 705 } 706 707 static inline sector_t bucket_to_sector(struct cache_set *c, size_t b) 708 { 709 return ((sector_t) b) << c->bucket_bits; 710 } 711 712 static inline sector_t bucket_remainder(struct cache_set *c, sector_t s) 713 { 714 return s & (c->sb.bucket_size - 1); 715 } 716 717 static inline struct cache *PTR_CACHE(struct cache_set *c, 718 const struct bkey *k, 719 unsigned ptr) 720 { 721 return c->cache[PTR_DEV(k, ptr)]; 722 } 723 724 static inline size_t PTR_BUCKET_NR(struct cache_set *c, 725 const struct bkey *k, 726 unsigned ptr) 727 { 728 return sector_to_bucket(c, PTR_OFFSET(k, ptr)); 729 } 730 731 static inline struct bucket *PTR_BUCKET(struct cache_set *c, 732 const struct bkey *k, 733 unsigned ptr) 734 { 735 return PTR_CACHE(c, k, ptr)->buckets + PTR_BUCKET_NR(c, k, ptr); 736 } 737 738 static inline uint8_t gen_after(uint8_t a, uint8_t b) 739 { 740 uint8_t r = a - b; 741 return r > 128U ? 0 : r; 742 } 743 744 static inline uint8_t ptr_stale(struct cache_set *c, const struct bkey *k, 745 unsigned i) 746 { 747 return gen_after(PTR_BUCKET(c, k, i)->gen, PTR_GEN(k, i)); 748 } 749 750 static inline bool ptr_available(struct cache_set *c, const struct bkey *k, 751 unsigned i) 752 { 753 return (PTR_DEV(k, i) < MAX_CACHES_PER_SET) && PTR_CACHE(c, k, i); 754 } 755 756 /* Btree key macros */ 757 758 /* 759 * This is used for various on disk data structures - cache_sb, prio_set, bset, 760 * jset: The checksum is _always_ the first 8 bytes of these structs 761 */ 762 #define csum_set(i) \ 763 bch_crc64(((void *) (i)) + sizeof(uint64_t), \ 764 ((void *) bset_bkey_last(i)) - \ 765 (((void *) (i)) + sizeof(uint64_t))) 766 767 /* Error handling macros */ 768 769 #define btree_bug(b, ...) \ 770 do { \ 771 if (bch_cache_set_error((b)->c, __VA_ARGS__)) \ 772 dump_stack(); \ 773 } while (0) 774 775 #define cache_bug(c, ...) \ 776 do { \ 777 if (bch_cache_set_error(c, __VA_ARGS__)) \ 778 dump_stack(); \ 779 } while (0) 780 781 #define btree_bug_on(cond, b, ...) \ 782 do { \ 783 if (cond) \ 784 btree_bug(b, __VA_ARGS__); \ 785 } while (0) 786 787 #define cache_bug_on(cond, c, ...) \ 788 do { \ 789 if (cond) \ 790 cache_bug(c, __VA_ARGS__); \ 791 } while (0) 792 793 #define cache_set_err_on(cond, c, ...) \ 794 do { \ 795 if (cond) \ 796 bch_cache_set_error(c, __VA_ARGS__); \ 797 } while (0) 798 799 /* Looping macros */ 800 801 #define for_each_cache(ca, cs, iter) \ 802 for (iter = 0; ca = cs->cache[iter], iter < (cs)->sb.nr_in_set; iter++) 803 804 #define for_each_bucket(b, ca) \ 805 for (b = (ca)->buckets + (ca)->sb.first_bucket; \ 806 b < (ca)->buckets + (ca)->sb.nbuckets; b++) 807 808 static inline void cached_dev_put(struct cached_dev *dc) 809 { 810 if (atomic_dec_and_test(&dc->count)) 811 schedule_work(&dc->detach); 812 } 813 814 static inline bool cached_dev_get(struct cached_dev *dc) 815 { 816 if (!atomic_inc_not_zero(&dc->count)) 817 return false; 818 819 /* Paired with the mb in cached_dev_attach */ 820 smp_mb__after_atomic(); 821 return true; 822 } 823 824 /* 825 * bucket_gc_gen() returns the difference between the bucket's current gen and 826 * the oldest gen of any pointer into that bucket in the btree (last_gc). 827 */ 828 829 static inline uint8_t bucket_gc_gen(struct bucket *b) 830 { 831 return b->gen - b->last_gc; 832 } 833 834 #define BUCKET_GC_GEN_MAX 96U 835 836 #define kobj_attribute_write(n, fn) \ 837 static struct kobj_attribute ksysfs_##n = __ATTR(n, S_IWUSR, NULL, fn) 838 839 #define kobj_attribute_rw(n, show, store) \ 840 static struct kobj_attribute ksysfs_##n = \ 841 __ATTR(n, S_IWUSR|S_IRUSR, show, store) 842 843 static inline void wake_up_allocators(struct cache_set *c) 844 { 845 struct cache *ca; 846 unsigned i; 847 848 for_each_cache(ca, c, i) 849 wake_up_process(ca->alloc_thread); 850 } 851 852 /* Forward declarations */ 853 854 void bch_count_io_errors(struct cache *, blk_status_t, const char *); 855 void bch_bbio_count_io_errors(struct cache_set *, struct bio *, 856 blk_status_t, const char *); 857 void bch_bbio_endio(struct cache_set *, struct bio *, blk_status_t, 858 const char *); 859 void bch_bbio_free(struct bio *, struct cache_set *); 860 struct bio *bch_bbio_alloc(struct cache_set *); 861 862 void __bch_submit_bbio(struct bio *, struct cache_set *); 863 void bch_submit_bbio(struct bio *, struct cache_set *, struct bkey *, unsigned); 864 865 uint8_t bch_inc_gen(struct cache *, struct bucket *); 866 void bch_rescale_priorities(struct cache_set *, int); 867 868 bool bch_can_invalidate_bucket(struct cache *, struct bucket *); 869 void __bch_invalidate_one_bucket(struct cache *, struct bucket *); 870 871 void __bch_bucket_free(struct cache *, struct bucket *); 872 void bch_bucket_free(struct cache_set *, struct bkey *); 873 874 long bch_bucket_alloc(struct cache *, unsigned, bool); 875 int __bch_bucket_alloc_set(struct cache_set *, unsigned, 876 struct bkey *, int, bool); 877 int bch_bucket_alloc_set(struct cache_set *, unsigned, 878 struct bkey *, int, bool); 879 bool bch_alloc_sectors(struct cache_set *, struct bkey *, unsigned, 880 unsigned, unsigned, bool); 881 882 __printf(2, 3) 883 bool bch_cache_set_error(struct cache_set *, const char *, ...); 884 885 void bch_prio_write(struct cache *); 886 void bch_write_bdev_super(struct cached_dev *, struct closure *); 887 888 extern struct workqueue_struct *bcache_wq; 889 extern const char * const bch_cache_modes[]; 890 extern struct mutex bch_register_lock; 891 extern struct list_head bch_cache_sets; 892 893 extern struct kobj_type bch_cached_dev_ktype; 894 extern struct kobj_type bch_flash_dev_ktype; 895 extern struct kobj_type bch_cache_set_ktype; 896 extern struct kobj_type bch_cache_set_internal_ktype; 897 extern struct kobj_type bch_cache_ktype; 898 899 void bch_cached_dev_release(struct kobject *); 900 void bch_flash_dev_release(struct kobject *); 901 void bch_cache_set_release(struct kobject *); 902 void bch_cache_release(struct kobject *); 903 904 int bch_uuid_write(struct cache_set *); 905 void bcache_write_super(struct cache_set *); 906 907 int bch_flash_dev_create(struct cache_set *c, uint64_t size); 908 909 int bch_cached_dev_attach(struct cached_dev *, struct cache_set *); 910 void bch_cached_dev_detach(struct cached_dev *); 911 void bch_cached_dev_run(struct cached_dev *); 912 void bcache_device_stop(struct bcache_device *); 913 914 void bch_cache_set_unregister(struct cache_set *); 915 void bch_cache_set_stop(struct cache_set *); 916 917 struct cache_set *bch_cache_set_alloc(struct cache_sb *); 918 void bch_btree_cache_free(struct cache_set *); 919 int bch_btree_cache_alloc(struct cache_set *); 920 void bch_moving_init_cache_set(struct cache_set *); 921 int bch_open_buckets_alloc(struct cache_set *); 922 void bch_open_buckets_free(struct cache_set *); 923 924 int bch_cache_allocator_start(struct cache *ca); 925 926 void bch_debug_exit(void); 927 int bch_debug_init(struct kobject *); 928 void bch_request_exit(void); 929 int bch_request_init(void); 930 931 #endif /* _BCACHE_H */ 932