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/refcount.h> 189 #include <linux/types.h> 190 #include <linux/workqueue.h> 191 192 #include "bset.h" 193 #include "util.h" 194 #include "closure.h" 195 196 struct bucket { 197 atomic_t pin; 198 uint16_t prio; 199 uint8_t gen; 200 uint8_t last_gc; /* Most out of date gen in the btree */ 201 uint16_t gc_mark; /* Bitfield used by GC. See below for field */ 202 }; 203 204 /* 205 * I'd use bitfields for these, but I don't trust the compiler not to screw me 206 * as multiple threads touch struct bucket without locking 207 */ 208 209 BITMASK(GC_MARK, struct bucket, gc_mark, 0, 2); 210 #define GC_MARK_RECLAIMABLE 1 211 #define GC_MARK_DIRTY 2 212 #define GC_MARK_METADATA 3 213 #define GC_SECTORS_USED_SIZE 13 214 #define MAX_GC_SECTORS_USED (~(~0ULL << GC_SECTORS_USED_SIZE)) 215 BITMASK(GC_SECTORS_USED, struct bucket, gc_mark, 2, GC_SECTORS_USED_SIZE); 216 BITMASK(GC_MOVE, struct bucket, gc_mark, 15, 1); 217 218 #include "journal.h" 219 #include "stats.h" 220 struct search; 221 struct btree; 222 struct keybuf; 223 224 struct keybuf_key { 225 struct rb_node node; 226 BKEY_PADDED(key); 227 void *private; 228 }; 229 230 struct keybuf { 231 struct bkey last_scanned; 232 spinlock_t lock; 233 234 /* 235 * Beginning and end of range in rb tree - so that we can skip taking 236 * lock and checking the rb tree when we need to check for overlapping 237 * keys. 238 */ 239 struct bkey start; 240 struct bkey end; 241 242 struct rb_root keys; 243 244 #define KEYBUF_NR 500 245 DECLARE_ARRAY_ALLOCATOR(struct keybuf_key, freelist, KEYBUF_NR); 246 }; 247 248 struct bcache_device { 249 struct closure cl; 250 251 struct kobject kobj; 252 253 struct cache_set *c; 254 unsigned id; 255 #define BCACHEDEVNAME_SIZE 12 256 char name[BCACHEDEVNAME_SIZE]; 257 258 struct gendisk *disk; 259 260 unsigned long flags; 261 #define BCACHE_DEV_CLOSING 0 262 #define BCACHE_DEV_DETACHING 1 263 #define BCACHE_DEV_UNLINK_DONE 2 264 265 unsigned nr_stripes; 266 unsigned stripe_size; 267 atomic_t *stripe_sectors_dirty; 268 unsigned long *full_dirty_stripes; 269 270 struct bio_set *bio_split; 271 272 unsigned data_csum:1; 273 274 int (*cache_miss)(struct btree *, struct search *, 275 struct bio *, unsigned); 276 int (*ioctl) (struct bcache_device *, fmode_t, unsigned, unsigned long); 277 }; 278 279 struct io { 280 /* Used to track sequential IO so it can be skipped */ 281 struct hlist_node hash; 282 struct list_head lru; 283 284 unsigned long jiffies; 285 unsigned sequential; 286 sector_t last; 287 }; 288 289 struct cached_dev { 290 struct list_head list; 291 struct bcache_device disk; 292 struct block_device *bdev; 293 294 struct cache_sb sb; 295 struct bio sb_bio; 296 struct bio_vec sb_bv[1]; 297 struct closure sb_write; 298 struct semaphore sb_write_mutex; 299 300 /* Refcount on the cache set. Always nonzero when we're caching. */ 301 refcount_t count; 302 struct work_struct detach; 303 304 /* 305 * Device might not be running if it's dirty and the cache set hasn't 306 * showed up yet. 307 */ 308 atomic_t running; 309 310 /* 311 * Writes take a shared lock from start to finish; scanning for dirty 312 * data to refill the rb tree requires an exclusive lock. 313 */ 314 struct rw_semaphore writeback_lock; 315 316 /* 317 * Nonzero, and writeback has a refcount (d->count), iff there is dirty 318 * data in the cache. Protected by writeback_lock; must have an 319 * shared lock to set and exclusive lock to clear. 320 */ 321 atomic_t has_dirty; 322 323 /* 324 * Set to zero by things that touch the backing volume-- except 325 * writeback. Incremented by writeback. Used to determine when to 326 * accelerate idle writeback. 327 */ 328 atomic_t backing_idle; 329 330 struct bch_ratelimit writeback_rate; 331 struct delayed_work writeback_rate_update; 332 333 /* Limit number of writeback bios in flight */ 334 struct semaphore in_flight; 335 struct task_struct *writeback_thread; 336 struct workqueue_struct *writeback_write_wq; 337 338 struct keybuf writeback_keys; 339 340 /* 341 * Order the write-half of writeback operations strongly in dispatch 342 * order. (Maintain LBA order; don't allow reads completing out of 343 * order to re-order the writes...) 344 */ 345 struct closure_waitlist writeback_ordering_wait; 346 atomic_t writeback_sequence_next; 347 348 /* For tracking sequential IO */ 349 #define RECENT_IO_BITS 7 350 #define RECENT_IO (1 << RECENT_IO_BITS) 351 struct io io[RECENT_IO]; 352 struct hlist_head io_hash[RECENT_IO + 1]; 353 struct list_head io_lru; 354 spinlock_t io_lock; 355 356 struct cache_accounting accounting; 357 358 /* The rest of this all shows up in sysfs */ 359 unsigned sequential_cutoff; 360 unsigned readahead; 361 362 unsigned verify:1; 363 unsigned bypass_torture_test:1; 364 365 unsigned partial_stripes_expensive:1; 366 unsigned writeback_metadata:1; 367 unsigned writeback_running:1; 368 unsigned char writeback_percent; 369 unsigned writeback_delay; 370 371 uint64_t writeback_rate_target; 372 int64_t writeback_rate_proportional; 373 int64_t writeback_rate_integral; 374 int64_t writeback_rate_integral_scaled; 375 int32_t writeback_rate_change; 376 377 unsigned writeback_rate_update_seconds; 378 unsigned writeback_rate_i_term_inverse; 379 unsigned writeback_rate_p_term_inverse; 380 unsigned writeback_rate_minimum; 381 }; 382 383 enum alloc_reserve { 384 RESERVE_BTREE, 385 RESERVE_PRIO, 386 RESERVE_MOVINGGC, 387 RESERVE_NONE, 388 RESERVE_NR, 389 }; 390 391 struct cache { 392 struct cache_set *set; 393 struct cache_sb sb; 394 struct bio sb_bio; 395 struct bio_vec sb_bv[1]; 396 397 struct kobject kobj; 398 struct block_device *bdev; 399 400 struct task_struct *alloc_thread; 401 402 struct closure prio; 403 struct prio_set *disk_buckets; 404 405 /* 406 * When allocating new buckets, prio_write() gets first dibs - since we 407 * may not be allocate at all without writing priorities and gens. 408 * prio_buckets[] contains the last buckets we wrote priorities to (so 409 * gc can mark them as metadata), prio_next[] contains the buckets 410 * allocated for the next prio write. 411 */ 412 uint64_t *prio_buckets; 413 uint64_t *prio_last_buckets; 414 415 /* 416 * free: Buckets that are ready to be used 417 * 418 * free_inc: Incoming buckets - these are buckets that currently have 419 * cached data in them, and we can't reuse them until after we write 420 * their new gen to disk. After prio_write() finishes writing the new 421 * gens/prios, they'll be moved to the free list (and possibly discarded 422 * in the process) 423 */ 424 DECLARE_FIFO(long, free)[RESERVE_NR]; 425 DECLARE_FIFO(long, free_inc); 426 427 size_t fifo_last_bucket; 428 429 /* Allocation stuff: */ 430 struct bucket *buckets; 431 432 DECLARE_HEAP(struct bucket *, heap); 433 434 /* 435 * If nonzero, we know we aren't going to find any buckets to invalidate 436 * until a gc finishes - otherwise we could pointlessly burn a ton of 437 * cpu 438 */ 439 unsigned invalidate_needs_gc; 440 441 bool discard; /* Get rid of? */ 442 443 struct journal_device journal; 444 445 /* The rest of this all shows up in sysfs */ 446 #define IO_ERROR_SHIFT 20 447 atomic_t io_errors; 448 atomic_t io_count; 449 450 atomic_long_t meta_sectors_written; 451 atomic_long_t btree_sectors_written; 452 atomic_long_t sectors_written; 453 }; 454 455 struct gc_stat { 456 size_t nodes; 457 size_t key_bytes; 458 459 size_t nkeys; 460 uint64_t data; /* sectors */ 461 unsigned in_use; /* percent */ 462 }; 463 464 /* 465 * Flag bits, for how the cache set is shutting down, and what phase it's at: 466 * 467 * CACHE_SET_UNREGISTERING means we're not just shutting down, we're detaching 468 * all the backing devices first (their cached data gets invalidated, and they 469 * won't automatically reattach). 470 * 471 * CACHE_SET_STOPPING always gets set first when we're closing down a cache set; 472 * we'll continue to run normally for awhile with CACHE_SET_STOPPING set (i.e. 473 * flushing dirty data). 474 * 475 * CACHE_SET_RUNNING means all cache devices have been registered and journal 476 * replay is complete. 477 */ 478 #define CACHE_SET_UNREGISTERING 0 479 #define CACHE_SET_STOPPING 1 480 #define CACHE_SET_RUNNING 2 481 482 struct cache_set { 483 struct closure cl; 484 485 struct list_head list; 486 struct kobject kobj; 487 struct kobject internal; 488 struct dentry *debug; 489 struct cache_accounting accounting; 490 491 unsigned long flags; 492 493 struct cache_sb sb; 494 495 struct cache *cache[MAX_CACHES_PER_SET]; 496 struct cache *cache_by_alloc[MAX_CACHES_PER_SET]; 497 int caches_loaded; 498 499 struct bcache_device **devices; 500 struct list_head cached_devs; 501 uint64_t cached_dev_sectors; 502 struct closure caching; 503 504 struct closure sb_write; 505 struct semaphore sb_write_mutex; 506 507 mempool_t *search; 508 mempool_t *bio_meta; 509 struct bio_set *bio_split; 510 511 /* For the btree cache */ 512 struct shrinker shrink; 513 514 /* For the btree cache and anything allocation related */ 515 struct mutex bucket_lock; 516 517 /* log2(bucket_size), in sectors */ 518 unsigned short bucket_bits; 519 520 /* log2(block_size), in sectors */ 521 unsigned short block_bits; 522 523 /* 524 * Default number of pages for a new btree node - may be less than a 525 * full bucket 526 */ 527 unsigned btree_pages; 528 529 /* 530 * Lists of struct btrees; lru is the list for structs that have memory 531 * allocated for actual btree node, freed is for structs that do not. 532 * 533 * We never free a struct btree, except on shutdown - we just put it on 534 * the btree_cache_freed list and reuse it later. This simplifies the 535 * code, and it doesn't cost us much memory as the memory usage is 536 * dominated by buffers that hold the actual btree node data and those 537 * can be freed - and the number of struct btrees allocated is 538 * effectively bounded. 539 * 540 * btree_cache_freeable effectively is a small cache - we use it because 541 * high order page allocations can be rather expensive, and it's quite 542 * common to delete and allocate btree nodes in quick succession. It 543 * should never grow past ~2-3 nodes in practice. 544 */ 545 struct list_head btree_cache; 546 struct list_head btree_cache_freeable; 547 struct list_head btree_cache_freed; 548 549 /* Number of elements in btree_cache + btree_cache_freeable lists */ 550 unsigned btree_cache_used; 551 552 /* 553 * If we need to allocate memory for a new btree node and that 554 * allocation fails, we can cannibalize another node in the btree cache 555 * to satisfy the allocation - lock to guarantee only one thread does 556 * this at a time: 557 */ 558 wait_queue_head_t btree_cache_wait; 559 struct task_struct *btree_cache_alloc_lock; 560 561 /* 562 * When we free a btree node, we increment the gen of the bucket the 563 * node is in - but we can't rewrite the prios and gens until we 564 * finished whatever it is we were doing, otherwise after a crash the 565 * btree node would be freed but for say a split, we might not have the 566 * pointers to the new nodes inserted into the btree yet. 567 * 568 * This is a refcount that blocks prio_write() until the new keys are 569 * written. 570 */ 571 atomic_t prio_blocked; 572 wait_queue_head_t bucket_wait; 573 574 /* 575 * For any bio we don't skip we subtract the number of sectors from 576 * rescale; when it hits 0 we rescale all the bucket priorities. 577 */ 578 atomic_t rescale; 579 /* 580 * When we invalidate buckets, we use both the priority and the amount 581 * of good data to determine which buckets to reuse first - to weight 582 * those together consistently we keep track of the smallest nonzero 583 * priority of any bucket. 584 */ 585 uint16_t min_prio; 586 587 /* 588 * max(gen - last_gc) for all buckets. When it gets too big we have to gc 589 * to keep gens from wrapping around. 590 */ 591 uint8_t need_gc; 592 struct gc_stat gc_stats; 593 size_t nbuckets; 594 size_t avail_nbuckets; 595 596 struct task_struct *gc_thread; 597 /* Where in the btree gc currently is */ 598 struct bkey gc_done; 599 600 /* 601 * The allocation code needs gc_mark in struct bucket to be correct, but 602 * it's not while a gc is in progress. Protected by bucket_lock. 603 */ 604 int gc_mark_valid; 605 606 /* Counts how many sectors bio_insert has added to the cache */ 607 atomic_t sectors_to_gc; 608 wait_queue_head_t gc_wait; 609 610 struct keybuf moving_gc_keys; 611 /* Number of moving GC bios in flight */ 612 struct semaphore moving_in_flight; 613 614 struct workqueue_struct *moving_gc_wq; 615 616 struct btree *root; 617 618 #ifdef CONFIG_BCACHE_DEBUG 619 struct btree *verify_data; 620 struct bset *verify_ondisk; 621 struct mutex verify_lock; 622 #endif 623 624 unsigned nr_uuids; 625 struct uuid_entry *uuids; 626 BKEY_PADDED(uuid_bucket); 627 struct closure uuid_write; 628 struct semaphore uuid_write_mutex; 629 630 /* 631 * A btree node on disk could have too many bsets for an iterator to fit 632 * on the stack - have to dynamically allocate them 633 */ 634 mempool_t *fill_iter; 635 636 struct bset_sort_state sort; 637 638 /* List of buckets we're currently writing data to */ 639 struct list_head data_buckets; 640 spinlock_t data_bucket_lock; 641 642 struct journal journal; 643 644 #define CONGESTED_MAX 1024 645 unsigned congested_last_us; 646 atomic_t congested; 647 648 /* The rest of this all shows up in sysfs */ 649 unsigned congested_read_threshold_us; 650 unsigned congested_write_threshold_us; 651 652 struct time_stats btree_gc_time; 653 struct time_stats btree_split_time; 654 struct time_stats btree_read_time; 655 656 atomic_long_t cache_read_races; 657 atomic_long_t writeback_keys_done; 658 atomic_long_t writeback_keys_failed; 659 660 enum { 661 ON_ERROR_UNREGISTER, 662 ON_ERROR_PANIC, 663 } on_error; 664 unsigned error_limit; 665 unsigned error_decay; 666 667 unsigned short journal_delay_ms; 668 bool expensive_debug_checks; 669 unsigned verify:1; 670 unsigned key_merging_disabled:1; 671 unsigned gc_always_rewrite:1; 672 unsigned shrinker_disabled:1; 673 unsigned copy_gc_enabled:1; 674 675 #define BUCKET_HASH_BITS 12 676 struct hlist_head bucket_hash[1 << BUCKET_HASH_BITS]; 677 }; 678 679 struct bbio { 680 unsigned submit_time_us; 681 union { 682 struct bkey key; 683 uint64_t _pad[3]; 684 /* 685 * We only need pad = 3 here because we only ever carry around a 686 * single pointer - i.e. the pointer we're doing io to/from. 687 */ 688 }; 689 struct bio bio; 690 }; 691 692 #define BTREE_PRIO USHRT_MAX 693 #define INITIAL_PRIO 32768U 694 695 #define btree_bytes(c) ((c)->btree_pages * PAGE_SIZE) 696 #define btree_blocks(b) \ 697 ((unsigned) (KEY_SIZE(&b->key) >> (b)->c->block_bits)) 698 699 #define btree_default_blocks(c) \ 700 ((unsigned) ((PAGE_SECTORS * (c)->btree_pages) >> (c)->block_bits)) 701 702 #define bucket_pages(c) ((c)->sb.bucket_size / PAGE_SECTORS) 703 #define bucket_bytes(c) ((c)->sb.bucket_size << 9) 704 #define block_bytes(c) ((c)->sb.block_size << 9) 705 706 #define prios_per_bucket(c) \ 707 ((bucket_bytes(c) - sizeof(struct prio_set)) / \ 708 sizeof(struct bucket_disk)) 709 #define prio_buckets(c) \ 710 DIV_ROUND_UP((size_t) (c)->sb.nbuckets, prios_per_bucket(c)) 711 712 static inline size_t sector_to_bucket(struct cache_set *c, sector_t s) 713 { 714 return s >> c->bucket_bits; 715 } 716 717 static inline sector_t bucket_to_sector(struct cache_set *c, size_t b) 718 { 719 return ((sector_t) b) << c->bucket_bits; 720 } 721 722 static inline sector_t bucket_remainder(struct cache_set *c, sector_t s) 723 { 724 return s & (c->sb.bucket_size - 1); 725 } 726 727 static inline struct cache *PTR_CACHE(struct cache_set *c, 728 const struct bkey *k, 729 unsigned ptr) 730 { 731 return c->cache[PTR_DEV(k, ptr)]; 732 } 733 734 static inline size_t PTR_BUCKET_NR(struct cache_set *c, 735 const struct bkey *k, 736 unsigned ptr) 737 { 738 return sector_to_bucket(c, PTR_OFFSET(k, ptr)); 739 } 740 741 static inline struct bucket *PTR_BUCKET(struct cache_set *c, 742 const struct bkey *k, 743 unsigned ptr) 744 { 745 return PTR_CACHE(c, k, ptr)->buckets + PTR_BUCKET_NR(c, k, ptr); 746 } 747 748 static inline uint8_t gen_after(uint8_t a, uint8_t b) 749 { 750 uint8_t r = a - b; 751 return r > 128U ? 0 : r; 752 } 753 754 static inline uint8_t ptr_stale(struct cache_set *c, const struct bkey *k, 755 unsigned i) 756 { 757 return gen_after(PTR_BUCKET(c, k, i)->gen, PTR_GEN(k, i)); 758 } 759 760 static inline bool ptr_available(struct cache_set *c, const struct bkey *k, 761 unsigned i) 762 { 763 return (PTR_DEV(k, i) < MAX_CACHES_PER_SET) && PTR_CACHE(c, k, i); 764 } 765 766 /* Btree key macros */ 767 768 /* 769 * This is used for various on disk data structures - cache_sb, prio_set, bset, 770 * jset: The checksum is _always_ the first 8 bytes of these structs 771 */ 772 #define csum_set(i) \ 773 bch_crc64(((void *) (i)) + sizeof(uint64_t), \ 774 ((void *) bset_bkey_last(i)) - \ 775 (((void *) (i)) + sizeof(uint64_t))) 776 777 /* Error handling macros */ 778 779 #define btree_bug(b, ...) \ 780 do { \ 781 if (bch_cache_set_error((b)->c, __VA_ARGS__)) \ 782 dump_stack(); \ 783 } while (0) 784 785 #define cache_bug(c, ...) \ 786 do { \ 787 if (bch_cache_set_error(c, __VA_ARGS__)) \ 788 dump_stack(); \ 789 } while (0) 790 791 #define btree_bug_on(cond, b, ...) \ 792 do { \ 793 if (cond) \ 794 btree_bug(b, __VA_ARGS__); \ 795 } while (0) 796 797 #define cache_bug_on(cond, c, ...) \ 798 do { \ 799 if (cond) \ 800 cache_bug(c, __VA_ARGS__); \ 801 } while (0) 802 803 #define cache_set_err_on(cond, c, ...) \ 804 do { \ 805 if (cond) \ 806 bch_cache_set_error(c, __VA_ARGS__); \ 807 } while (0) 808 809 /* Looping macros */ 810 811 #define for_each_cache(ca, cs, iter) \ 812 for (iter = 0; ca = cs->cache[iter], iter < (cs)->sb.nr_in_set; iter++) 813 814 #define for_each_bucket(b, ca) \ 815 for (b = (ca)->buckets + (ca)->sb.first_bucket; \ 816 b < (ca)->buckets + (ca)->sb.nbuckets; b++) 817 818 static inline void cached_dev_put(struct cached_dev *dc) 819 { 820 if (refcount_dec_and_test(&dc->count)) 821 schedule_work(&dc->detach); 822 } 823 824 static inline bool cached_dev_get(struct cached_dev *dc) 825 { 826 if (!refcount_inc_not_zero(&dc->count)) 827 return false; 828 829 /* Paired with the mb in cached_dev_attach */ 830 smp_mb__after_atomic(); 831 return true; 832 } 833 834 /* 835 * bucket_gc_gen() returns the difference between the bucket's current gen and 836 * the oldest gen of any pointer into that bucket in the btree (last_gc). 837 */ 838 839 static inline uint8_t bucket_gc_gen(struct bucket *b) 840 { 841 return b->gen - b->last_gc; 842 } 843 844 #define BUCKET_GC_GEN_MAX 96U 845 846 #define kobj_attribute_write(n, fn) \ 847 static struct kobj_attribute ksysfs_##n = __ATTR(n, S_IWUSR, NULL, fn) 848 849 #define kobj_attribute_rw(n, show, store) \ 850 static struct kobj_attribute ksysfs_##n = \ 851 __ATTR(n, S_IWUSR|S_IRUSR, show, store) 852 853 static inline void wake_up_allocators(struct cache_set *c) 854 { 855 struct cache *ca; 856 unsigned i; 857 858 for_each_cache(ca, c, i) 859 wake_up_process(ca->alloc_thread); 860 } 861 862 /* Forward declarations */ 863 864 void bch_count_io_errors(struct cache *, blk_status_t, const char *); 865 void bch_bbio_count_io_errors(struct cache_set *, struct bio *, 866 blk_status_t, const char *); 867 void bch_bbio_endio(struct cache_set *, struct bio *, blk_status_t, 868 const char *); 869 void bch_bbio_free(struct bio *, struct cache_set *); 870 struct bio *bch_bbio_alloc(struct cache_set *); 871 872 void __bch_submit_bbio(struct bio *, struct cache_set *); 873 void bch_submit_bbio(struct bio *, struct cache_set *, struct bkey *, unsigned); 874 875 uint8_t bch_inc_gen(struct cache *, struct bucket *); 876 void bch_rescale_priorities(struct cache_set *, int); 877 878 bool bch_can_invalidate_bucket(struct cache *, struct bucket *); 879 void __bch_invalidate_one_bucket(struct cache *, struct bucket *); 880 881 void __bch_bucket_free(struct cache *, struct bucket *); 882 void bch_bucket_free(struct cache_set *, struct bkey *); 883 884 long bch_bucket_alloc(struct cache *, unsigned, bool); 885 int __bch_bucket_alloc_set(struct cache_set *, unsigned, 886 struct bkey *, int, bool); 887 int bch_bucket_alloc_set(struct cache_set *, unsigned, 888 struct bkey *, int, bool); 889 bool bch_alloc_sectors(struct cache_set *, struct bkey *, unsigned, 890 unsigned, unsigned, bool); 891 892 __printf(2, 3) 893 bool bch_cache_set_error(struct cache_set *, const char *, ...); 894 895 void bch_prio_write(struct cache *); 896 void bch_write_bdev_super(struct cached_dev *, struct closure *); 897 898 extern struct workqueue_struct *bcache_wq; 899 extern const char * const bch_cache_modes[]; 900 extern struct mutex bch_register_lock; 901 extern struct list_head bch_cache_sets; 902 903 extern struct kobj_type bch_cached_dev_ktype; 904 extern struct kobj_type bch_flash_dev_ktype; 905 extern struct kobj_type bch_cache_set_ktype; 906 extern struct kobj_type bch_cache_set_internal_ktype; 907 extern struct kobj_type bch_cache_ktype; 908 909 void bch_cached_dev_release(struct kobject *); 910 void bch_flash_dev_release(struct kobject *); 911 void bch_cache_set_release(struct kobject *); 912 void bch_cache_release(struct kobject *); 913 914 int bch_uuid_write(struct cache_set *); 915 void bcache_write_super(struct cache_set *); 916 917 int bch_flash_dev_create(struct cache_set *c, uint64_t size); 918 919 int bch_cached_dev_attach(struct cached_dev *, struct cache_set *); 920 void bch_cached_dev_detach(struct cached_dev *); 921 void bch_cached_dev_run(struct cached_dev *); 922 void bcache_device_stop(struct bcache_device *); 923 924 void bch_cache_set_unregister(struct cache_set *); 925 void bch_cache_set_stop(struct cache_set *); 926 927 struct cache_set *bch_cache_set_alloc(struct cache_sb *); 928 void bch_btree_cache_free(struct cache_set *); 929 int bch_btree_cache_alloc(struct cache_set *); 930 void bch_moving_init_cache_set(struct cache_set *); 931 int bch_open_buckets_alloc(struct cache_set *); 932 void bch_open_buckets_free(struct cache_set *); 933 934 int bch_cache_allocator_start(struct cache *ca); 935 936 void bch_debug_exit(void); 937 int bch_debug_init(struct kobject *); 938 void bch_request_exit(void); 939 int bch_request_init(void); 940 941 #endif /* _BCACHE_H */ 942