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, __func__ 180 181 #include <linux/bio.h> 182 #include <linux/kobject.h> 183 #include <linux/list.h> 184 #include <linux/mutex.h> 185 #include <linux/rbtree.h> 186 #include <linux/rwsem.h> 187 #include <linux/refcount.h> 188 #include <linux/types.h> 189 #include <linux/workqueue.h> 190 #include <linux/kthread.h> 191 192 #include "bcache_ondisk.h" 193 #include "bset.h" 194 #include "util.h" 195 #include "closure.h" 196 197 struct bucket { 198 atomic_t pin; 199 uint16_t prio; 200 uint8_t gen; 201 uint8_t last_gc; /* Most out of date gen in the btree */ 202 uint16_t gc_mark; /* Bitfield used by GC. See below for field */ 203 }; 204 205 /* 206 * I'd use bitfields for these, but I don't trust the compiler not to screw me 207 * as multiple threads touch struct bucket without locking 208 */ 209 210 BITMASK(GC_MARK, struct bucket, gc_mark, 0, 2); 211 #define GC_MARK_RECLAIMABLE 1 212 #define GC_MARK_DIRTY 2 213 #define GC_MARK_METADATA 3 214 #define GC_SECTORS_USED_SIZE 13 215 #define MAX_GC_SECTORS_USED (~(~0ULL << GC_SECTORS_USED_SIZE)) 216 BITMASK(GC_SECTORS_USED, struct bucket, gc_mark, 2, GC_SECTORS_USED_SIZE); 217 BITMASK(GC_MOVE, struct bucket, gc_mark, 15, 1); 218 219 #include "journal.h" 220 #include "stats.h" 221 struct search; 222 struct btree; 223 struct keybuf; 224 225 struct keybuf_key { 226 struct rb_node node; 227 BKEY_PADDED(key); 228 void *private; 229 }; 230 231 struct keybuf { 232 struct bkey last_scanned; 233 spinlock_t lock; 234 235 /* 236 * Beginning and end of range in rb tree - so that we can skip taking 237 * lock and checking the rb tree when we need to check for overlapping 238 * keys. 239 */ 240 struct bkey start; 241 struct bkey end; 242 243 struct rb_root keys; 244 245 #define KEYBUF_NR 500 246 DECLARE_ARRAY_ALLOCATOR(struct keybuf_key, freelist, KEYBUF_NR); 247 }; 248 249 struct bcache_device { 250 struct closure cl; 251 252 struct kobject kobj; 253 254 struct cache_set *c; 255 unsigned int id; 256 #define BCACHEDEVNAME_SIZE 12 257 char name[BCACHEDEVNAME_SIZE]; 258 259 struct gendisk *disk; 260 261 unsigned long flags; 262 #define BCACHE_DEV_CLOSING 0 263 #define BCACHE_DEV_DETACHING 1 264 #define BCACHE_DEV_UNLINK_DONE 2 265 #define BCACHE_DEV_WB_RUNNING 3 266 #define BCACHE_DEV_RATE_DW_RUNNING 4 267 int nr_stripes; 268 unsigned int stripe_size; 269 atomic_t *stripe_sectors_dirty; 270 unsigned long *full_dirty_stripes; 271 272 struct bio_set bio_split; 273 274 unsigned int data_csum:1; 275 276 int (*cache_miss)(struct btree *b, struct search *s, 277 struct bio *bio, unsigned int sectors); 278 int (*ioctl)(struct bcache_device *d, fmode_t mode, 279 unsigned int cmd, unsigned long arg); 280 }; 281 282 struct io { 283 /* Used to track sequential IO so it can be skipped */ 284 struct hlist_node hash; 285 struct list_head lru; 286 287 unsigned long jiffies; 288 unsigned int sequential; 289 sector_t last; 290 }; 291 292 enum stop_on_failure { 293 BCH_CACHED_DEV_STOP_AUTO = 0, 294 BCH_CACHED_DEV_STOP_ALWAYS, 295 BCH_CACHED_DEV_STOP_MODE_MAX, 296 }; 297 298 struct cached_dev { 299 struct list_head list; 300 struct bcache_device disk; 301 struct block_device *bdev; 302 303 struct cache_sb sb; 304 struct cache_sb_disk *sb_disk; 305 struct bio sb_bio; 306 struct bio_vec sb_bv[1]; 307 struct closure sb_write; 308 struct semaphore sb_write_mutex; 309 310 /* Refcount on the cache set. Always nonzero when we're caching. */ 311 refcount_t count; 312 struct work_struct detach; 313 314 /* 315 * Device might not be running if it's dirty and the cache set hasn't 316 * showed up yet. 317 */ 318 atomic_t running; 319 320 /* 321 * Writes take a shared lock from start to finish; scanning for dirty 322 * data to refill the rb tree requires an exclusive lock. 323 */ 324 struct rw_semaphore writeback_lock; 325 326 /* 327 * Nonzero, and writeback has a refcount (d->count), iff there is dirty 328 * data in the cache. Protected by writeback_lock; must have an 329 * shared lock to set and exclusive lock to clear. 330 */ 331 atomic_t has_dirty; 332 333 #define BCH_CACHE_READA_ALL 0 334 #define BCH_CACHE_READA_META_ONLY 1 335 unsigned int cache_readahead_policy; 336 struct bch_ratelimit writeback_rate; 337 struct delayed_work writeback_rate_update; 338 339 /* Limit number of writeback bios in flight */ 340 struct semaphore in_flight; 341 struct task_struct *writeback_thread; 342 struct workqueue_struct *writeback_write_wq; 343 344 struct keybuf writeback_keys; 345 346 struct task_struct *status_update_thread; 347 /* 348 * Order the write-half of writeback operations strongly in dispatch 349 * order. (Maintain LBA order; don't allow reads completing out of 350 * order to re-order the writes...) 351 */ 352 struct closure_waitlist writeback_ordering_wait; 353 atomic_t writeback_sequence_next; 354 355 /* For tracking sequential IO */ 356 #define RECENT_IO_BITS 7 357 #define RECENT_IO (1 << RECENT_IO_BITS) 358 struct io io[RECENT_IO]; 359 struct hlist_head io_hash[RECENT_IO + 1]; 360 struct list_head io_lru; 361 spinlock_t io_lock; 362 363 struct cache_accounting accounting; 364 365 /* The rest of this all shows up in sysfs */ 366 unsigned int sequential_cutoff; 367 368 unsigned int io_disable:1; 369 unsigned int verify:1; 370 unsigned int bypass_torture_test:1; 371 372 unsigned int partial_stripes_expensive:1; 373 unsigned int writeback_metadata:1; 374 unsigned int writeback_running:1; 375 unsigned int writeback_consider_fragment:1; 376 unsigned char writeback_percent; 377 unsigned int writeback_delay; 378 379 uint64_t writeback_rate_target; 380 int64_t writeback_rate_proportional; 381 int64_t writeback_rate_integral; 382 int64_t writeback_rate_integral_scaled; 383 int32_t writeback_rate_change; 384 385 unsigned int writeback_rate_update_seconds; 386 unsigned int writeback_rate_i_term_inverse; 387 unsigned int writeback_rate_p_term_inverse; 388 unsigned int writeback_rate_fp_term_low; 389 unsigned int writeback_rate_fp_term_mid; 390 unsigned int writeback_rate_fp_term_high; 391 unsigned int writeback_rate_minimum; 392 393 enum stop_on_failure stop_when_cache_set_failed; 394 #define DEFAULT_CACHED_DEV_ERROR_LIMIT 64 395 atomic_t io_errors; 396 unsigned int error_limit; 397 unsigned int offline_seconds; 398 }; 399 400 enum alloc_reserve { 401 RESERVE_BTREE, 402 RESERVE_PRIO, 403 RESERVE_MOVINGGC, 404 RESERVE_NONE, 405 RESERVE_NR, 406 }; 407 408 struct cache { 409 struct cache_set *set; 410 struct cache_sb sb; 411 struct cache_sb_disk *sb_disk; 412 struct bio sb_bio; 413 struct bio_vec sb_bv[1]; 414 415 struct kobject kobj; 416 struct block_device *bdev; 417 418 struct task_struct *alloc_thread; 419 420 struct closure prio; 421 struct prio_set *disk_buckets; 422 423 /* 424 * When allocating new buckets, prio_write() gets first dibs - since we 425 * may not be allocate at all without writing priorities and gens. 426 * prio_last_buckets[] contains the last buckets we wrote priorities to 427 * (so gc can mark them as metadata), prio_buckets[] contains the 428 * buckets allocated for the next prio write. 429 */ 430 uint64_t *prio_buckets; 431 uint64_t *prio_last_buckets; 432 433 /* 434 * free: Buckets that are ready to be used 435 * 436 * free_inc: Incoming buckets - these are buckets that currently have 437 * cached data in them, and we can't reuse them until after we write 438 * their new gen to disk. After prio_write() finishes writing the new 439 * gens/prios, they'll be moved to the free list (and possibly discarded 440 * in the process) 441 */ 442 DECLARE_FIFO(long, free)[RESERVE_NR]; 443 DECLARE_FIFO(long, free_inc); 444 445 size_t fifo_last_bucket; 446 447 /* Allocation stuff: */ 448 struct bucket *buckets; 449 450 DECLARE_HEAP(struct bucket *, heap); 451 452 /* 453 * If nonzero, we know we aren't going to find any buckets to invalidate 454 * until a gc finishes - otherwise we could pointlessly burn a ton of 455 * cpu 456 */ 457 unsigned int invalidate_needs_gc; 458 459 bool discard; /* Get rid of? */ 460 461 struct journal_device journal; 462 463 /* The rest of this all shows up in sysfs */ 464 #define IO_ERROR_SHIFT 20 465 atomic_t io_errors; 466 atomic_t io_count; 467 468 atomic_long_t meta_sectors_written; 469 atomic_long_t btree_sectors_written; 470 atomic_long_t sectors_written; 471 }; 472 473 struct gc_stat { 474 size_t nodes; 475 size_t nodes_pre; 476 size_t key_bytes; 477 478 size_t nkeys; 479 uint64_t data; /* sectors */ 480 unsigned int in_use; /* percent */ 481 }; 482 483 /* 484 * Flag bits, for how the cache set is shutting down, and what phase it's at: 485 * 486 * CACHE_SET_UNREGISTERING means we're not just shutting down, we're detaching 487 * all the backing devices first (their cached data gets invalidated, and they 488 * won't automatically reattach). 489 * 490 * CACHE_SET_STOPPING always gets set first when we're closing down a cache set; 491 * we'll continue to run normally for awhile with CACHE_SET_STOPPING set (i.e. 492 * flushing dirty data). 493 * 494 * CACHE_SET_RUNNING means all cache devices have been registered and journal 495 * replay is complete. 496 * 497 * CACHE_SET_IO_DISABLE is set when bcache is stopping the whold cache set, all 498 * external and internal I/O should be denied when this flag is set. 499 * 500 */ 501 #define CACHE_SET_UNREGISTERING 0 502 #define CACHE_SET_STOPPING 1 503 #define CACHE_SET_RUNNING 2 504 #define CACHE_SET_IO_DISABLE 3 505 506 struct cache_set { 507 struct closure cl; 508 509 struct list_head list; 510 struct kobject kobj; 511 struct kobject internal; 512 struct dentry *debug; 513 struct cache_accounting accounting; 514 515 unsigned long flags; 516 atomic_t idle_counter; 517 atomic_t at_max_writeback_rate; 518 519 struct cache *cache; 520 521 struct bcache_device **devices; 522 unsigned int devices_max_used; 523 atomic_t attached_dev_nr; 524 struct list_head cached_devs; 525 uint64_t cached_dev_sectors; 526 atomic_long_t flash_dev_dirty_sectors; 527 struct closure caching; 528 529 struct closure sb_write; 530 struct semaphore sb_write_mutex; 531 532 mempool_t search; 533 mempool_t bio_meta; 534 struct bio_set bio_split; 535 536 /* For the btree cache */ 537 struct shrinker shrink; 538 539 /* For the btree cache and anything allocation related */ 540 struct mutex bucket_lock; 541 542 /* log2(bucket_size), in sectors */ 543 unsigned short bucket_bits; 544 545 /* log2(block_size), in sectors */ 546 unsigned short block_bits; 547 548 /* 549 * Default number of pages for a new btree node - may be less than a 550 * full bucket 551 */ 552 unsigned int btree_pages; 553 554 /* 555 * Lists of struct btrees; lru is the list for structs that have memory 556 * allocated for actual btree node, freed is for structs that do not. 557 * 558 * We never free a struct btree, except on shutdown - we just put it on 559 * the btree_cache_freed list and reuse it later. This simplifies the 560 * code, and it doesn't cost us much memory as the memory usage is 561 * dominated by buffers that hold the actual btree node data and those 562 * can be freed - and the number of struct btrees allocated is 563 * effectively bounded. 564 * 565 * btree_cache_freeable effectively is a small cache - we use it because 566 * high order page allocations can be rather expensive, and it's quite 567 * common to delete and allocate btree nodes in quick succession. It 568 * should never grow past ~2-3 nodes in practice. 569 */ 570 struct list_head btree_cache; 571 struct list_head btree_cache_freeable; 572 struct list_head btree_cache_freed; 573 574 /* Number of elements in btree_cache + btree_cache_freeable lists */ 575 unsigned int btree_cache_used; 576 577 /* 578 * If we need to allocate memory for a new btree node and that 579 * allocation fails, we can cannibalize another node in the btree cache 580 * to satisfy the allocation - lock to guarantee only one thread does 581 * this at a time: 582 */ 583 wait_queue_head_t btree_cache_wait; 584 struct task_struct *btree_cache_alloc_lock; 585 spinlock_t btree_cannibalize_lock; 586 587 /* 588 * When we free a btree node, we increment the gen of the bucket the 589 * node is in - but we can't rewrite the prios and gens until we 590 * finished whatever it is we were doing, otherwise after a crash the 591 * btree node would be freed but for say a split, we might not have the 592 * pointers to the new nodes inserted into the btree yet. 593 * 594 * This is a refcount that blocks prio_write() until the new keys are 595 * written. 596 */ 597 atomic_t prio_blocked; 598 wait_queue_head_t bucket_wait; 599 600 /* 601 * For any bio we don't skip we subtract the number of sectors from 602 * rescale; when it hits 0 we rescale all the bucket priorities. 603 */ 604 atomic_t rescale; 605 /* 606 * used for GC, identify if any front side I/Os is inflight 607 */ 608 atomic_t search_inflight; 609 /* 610 * When we invalidate buckets, we use both the priority and the amount 611 * of good data to determine which buckets to reuse first - to weight 612 * those together consistently we keep track of the smallest nonzero 613 * priority of any bucket. 614 */ 615 uint16_t min_prio; 616 617 /* 618 * max(gen - last_gc) for all buckets. When it gets too big we have to 619 * gc to keep gens from wrapping around. 620 */ 621 uint8_t need_gc; 622 struct gc_stat gc_stats; 623 size_t nbuckets; 624 size_t avail_nbuckets; 625 626 struct task_struct *gc_thread; 627 /* Where in the btree gc currently is */ 628 struct bkey gc_done; 629 630 /* 631 * For automatical garbage collection after writeback completed, this 632 * varialbe is used as bit fields, 633 * - 0000 0001b (BCH_ENABLE_AUTO_GC): enable gc after writeback 634 * - 0000 0010b (BCH_DO_AUTO_GC): do gc after writeback 635 * This is an optimization for following write request after writeback 636 * finished, but read hit rate dropped due to clean data on cache is 637 * discarded. Unless user explicitly sets it via sysfs, it won't be 638 * enabled. 639 */ 640 #define BCH_ENABLE_AUTO_GC 1 641 #define BCH_DO_AUTO_GC 2 642 uint8_t gc_after_writeback; 643 644 /* 645 * The allocation code needs gc_mark in struct bucket to be correct, but 646 * it's not while a gc is in progress. Protected by bucket_lock. 647 */ 648 int gc_mark_valid; 649 650 /* Counts how many sectors bio_insert has added to the cache */ 651 atomic_t sectors_to_gc; 652 wait_queue_head_t gc_wait; 653 654 struct keybuf moving_gc_keys; 655 /* Number of moving GC bios in flight */ 656 struct semaphore moving_in_flight; 657 658 struct workqueue_struct *moving_gc_wq; 659 660 struct btree *root; 661 662 #ifdef CONFIG_BCACHE_DEBUG 663 struct btree *verify_data; 664 struct bset *verify_ondisk; 665 struct mutex verify_lock; 666 #endif 667 668 uint8_t set_uuid[16]; 669 unsigned int nr_uuids; 670 struct uuid_entry *uuids; 671 BKEY_PADDED(uuid_bucket); 672 struct closure uuid_write; 673 struct semaphore uuid_write_mutex; 674 675 /* 676 * A btree node on disk could have too many bsets for an iterator to fit 677 * on the stack - have to dynamically allocate them. 678 * bch_cache_set_alloc() will make sure the pool can allocate iterators 679 * equipped with enough room that can host 680 * (sb.bucket_size / sb.block_size) 681 * btree_iter_sets, which is more than static MAX_BSETS. 682 */ 683 mempool_t fill_iter; 684 685 struct bset_sort_state sort; 686 687 /* List of buckets we're currently writing data to */ 688 struct list_head data_buckets; 689 spinlock_t data_bucket_lock; 690 691 struct journal journal; 692 693 #define CONGESTED_MAX 1024 694 unsigned int congested_last_us; 695 atomic_t congested; 696 697 /* The rest of this all shows up in sysfs */ 698 unsigned int congested_read_threshold_us; 699 unsigned int congested_write_threshold_us; 700 701 struct time_stats btree_gc_time; 702 struct time_stats btree_split_time; 703 struct time_stats btree_read_time; 704 705 atomic_long_t cache_read_races; 706 atomic_long_t writeback_keys_done; 707 atomic_long_t writeback_keys_failed; 708 709 atomic_long_t reclaim; 710 atomic_long_t reclaimed_journal_buckets; 711 atomic_long_t flush_write; 712 713 enum { 714 ON_ERROR_UNREGISTER, 715 ON_ERROR_PANIC, 716 } on_error; 717 #define DEFAULT_IO_ERROR_LIMIT 8 718 unsigned int error_limit; 719 unsigned int error_decay; 720 721 unsigned short journal_delay_ms; 722 bool expensive_debug_checks; 723 unsigned int verify:1; 724 unsigned int key_merging_disabled:1; 725 unsigned int gc_always_rewrite:1; 726 unsigned int shrinker_disabled:1; 727 unsigned int copy_gc_enabled:1; 728 unsigned int idle_max_writeback_rate_enabled:1; 729 730 #define BUCKET_HASH_BITS 12 731 struct hlist_head bucket_hash[1 << BUCKET_HASH_BITS]; 732 }; 733 734 struct bbio { 735 unsigned int submit_time_us; 736 union { 737 struct bkey key; 738 uint64_t _pad[3]; 739 /* 740 * We only need pad = 3 here because we only ever carry around a 741 * single pointer - i.e. the pointer we're doing io to/from. 742 */ 743 }; 744 struct bio bio; 745 }; 746 747 #define BTREE_PRIO USHRT_MAX 748 #define INITIAL_PRIO 32768U 749 750 #define btree_bytes(c) ((c)->btree_pages * PAGE_SIZE) 751 #define btree_blocks(b) \ 752 ((unsigned int) (KEY_SIZE(&b->key) >> (b)->c->block_bits)) 753 754 #define btree_default_blocks(c) \ 755 ((unsigned int) ((PAGE_SECTORS * (c)->btree_pages) >> (c)->block_bits)) 756 757 #define bucket_bytes(ca) ((ca)->sb.bucket_size << 9) 758 #define block_bytes(ca) ((ca)->sb.block_size << 9) 759 760 static inline unsigned int meta_bucket_pages(struct cache_sb *sb) 761 { 762 unsigned int n, max_pages; 763 764 max_pages = min_t(unsigned int, 765 __rounddown_pow_of_two(USHRT_MAX) / PAGE_SECTORS, 766 MAX_ORDER_NR_PAGES); 767 768 n = sb->bucket_size / PAGE_SECTORS; 769 if (n > max_pages) 770 n = max_pages; 771 772 return n; 773 } 774 775 static inline unsigned int meta_bucket_bytes(struct cache_sb *sb) 776 { 777 return meta_bucket_pages(sb) << PAGE_SHIFT; 778 } 779 780 #define prios_per_bucket(ca) \ 781 ((meta_bucket_bytes(&(ca)->sb) - sizeof(struct prio_set)) / \ 782 sizeof(struct bucket_disk)) 783 784 #define prio_buckets(ca) \ 785 DIV_ROUND_UP((size_t) (ca)->sb.nbuckets, prios_per_bucket(ca)) 786 787 static inline size_t sector_to_bucket(struct cache_set *c, sector_t s) 788 { 789 return s >> c->bucket_bits; 790 } 791 792 static inline sector_t bucket_to_sector(struct cache_set *c, size_t b) 793 { 794 return ((sector_t) b) << c->bucket_bits; 795 } 796 797 static inline sector_t bucket_remainder(struct cache_set *c, sector_t s) 798 { 799 return s & (c->cache->sb.bucket_size - 1); 800 } 801 802 static inline size_t PTR_BUCKET_NR(struct cache_set *c, 803 const struct bkey *k, 804 unsigned int ptr) 805 { 806 return sector_to_bucket(c, PTR_OFFSET(k, ptr)); 807 } 808 809 static inline struct bucket *PTR_BUCKET(struct cache_set *c, 810 const struct bkey *k, 811 unsigned int ptr) 812 { 813 return c->cache->buckets + PTR_BUCKET_NR(c, k, ptr); 814 } 815 816 static inline uint8_t gen_after(uint8_t a, uint8_t b) 817 { 818 uint8_t r = a - b; 819 820 return r > 128U ? 0 : r; 821 } 822 823 static inline uint8_t ptr_stale(struct cache_set *c, const struct bkey *k, 824 unsigned int i) 825 { 826 return gen_after(PTR_BUCKET(c, k, i)->gen, PTR_GEN(k, i)); 827 } 828 829 static inline bool ptr_available(struct cache_set *c, const struct bkey *k, 830 unsigned int i) 831 { 832 return (PTR_DEV(k, i) < MAX_CACHES_PER_SET) && c->cache; 833 } 834 835 /* Btree key macros */ 836 837 /* 838 * This is used for various on disk data structures - cache_sb, prio_set, bset, 839 * jset: The checksum is _always_ the first 8 bytes of these structs 840 */ 841 #define csum_set(i) \ 842 bch_crc64(((void *) (i)) + sizeof(uint64_t), \ 843 ((void *) bset_bkey_last(i)) - \ 844 (((void *) (i)) + sizeof(uint64_t))) 845 846 /* Error handling macros */ 847 848 #define btree_bug(b, ...) \ 849 do { \ 850 if (bch_cache_set_error((b)->c, __VA_ARGS__)) \ 851 dump_stack(); \ 852 } while (0) 853 854 #define cache_bug(c, ...) \ 855 do { \ 856 if (bch_cache_set_error(c, __VA_ARGS__)) \ 857 dump_stack(); \ 858 } while (0) 859 860 #define btree_bug_on(cond, b, ...) \ 861 do { \ 862 if (cond) \ 863 btree_bug(b, __VA_ARGS__); \ 864 } while (0) 865 866 #define cache_bug_on(cond, c, ...) \ 867 do { \ 868 if (cond) \ 869 cache_bug(c, __VA_ARGS__); \ 870 } while (0) 871 872 #define cache_set_err_on(cond, c, ...) \ 873 do { \ 874 if (cond) \ 875 bch_cache_set_error(c, __VA_ARGS__); \ 876 } while (0) 877 878 /* Looping macros */ 879 880 #define for_each_bucket(b, ca) \ 881 for (b = (ca)->buckets + (ca)->sb.first_bucket; \ 882 b < (ca)->buckets + (ca)->sb.nbuckets; b++) 883 884 static inline void cached_dev_put(struct cached_dev *dc) 885 { 886 if (refcount_dec_and_test(&dc->count)) 887 schedule_work(&dc->detach); 888 } 889 890 static inline bool cached_dev_get(struct cached_dev *dc) 891 { 892 if (!refcount_inc_not_zero(&dc->count)) 893 return false; 894 895 /* Paired with the mb in cached_dev_attach */ 896 smp_mb__after_atomic(); 897 return true; 898 } 899 900 /* 901 * bucket_gc_gen() returns the difference between the bucket's current gen and 902 * the oldest gen of any pointer into that bucket in the btree (last_gc). 903 */ 904 905 static inline uint8_t bucket_gc_gen(struct bucket *b) 906 { 907 return b->gen - b->last_gc; 908 } 909 910 #define BUCKET_GC_GEN_MAX 96U 911 912 #define kobj_attribute_write(n, fn) \ 913 static struct kobj_attribute ksysfs_##n = __ATTR(n, 0200, NULL, fn) 914 915 #define kobj_attribute_rw(n, show, store) \ 916 static struct kobj_attribute ksysfs_##n = \ 917 __ATTR(n, 0600, show, store) 918 919 static inline void wake_up_allocators(struct cache_set *c) 920 { 921 struct cache *ca = c->cache; 922 923 wake_up_process(ca->alloc_thread); 924 } 925 926 static inline void closure_bio_submit(struct cache_set *c, 927 struct bio *bio, 928 struct closure *cl) 929 { 930 closure_get(cl); 931 if (unlikely(test_bit(CACHE_SET_IO_DISABLE, &c->flags))) { 932 bio->bi_status = BLK_STS_IOERR; 933 bio_endio(bio); 934 return; 935 } 936 submit_bio_noacct(bio); 937 } 938 939 /* 940 * Prevent the kthread exits directly, and make sure when kthread_stop() 941 * is called to stop a kthread, it is still alive. If a kthread might be 942 * stopped by CACHE_SET_IO_DISABLE bit set, wait_for_kthread_stop() is 943 * necessary before the kthread returns. 944 */ 945 static inline void wait_for_kthread_stop(void) 946 { 947 while (!kthread_should_stop()) { 948 set_current_state(TASK_INTERRUPTIBLE); 949 schedule(); 950 } 951 } 952 953 /* Forward declarations */ 954 955 void bch_count_backing_io_errors(struct cached_dev *dc, struct bio *bio); 956 void bch_count_io_errors(struct cache *ca, blk_status_t error, 957 int is_read, const char *m); 958 void bch_bbio_count_io_errors(struct cache_set *c, struct bio *bio, 959 blk_status_t error, const char *m); 960 void bch_bbio_endio(struct cache_set *c, struct bio *bio, 961 blk_status_t error, const char *m); 962 void bch_bbio_free(struct bio *bio, struct cache_set *c); 963 struct bio *bch_bbio_alloc(struct cache_set *c); 964 965 void __bch_submit_bbio(struct bio *bio, struct cache_set *c); 966 void bch_submit_bbio(struct bio *bio, struct cache_set *c, 967 struct bkey *k, unsigned int ptr); 968 969 uint8_t bch_inc_gen(struct cache *ca, struct bucket *b); 970 void bch_rescale_priorities(struct cache_set *c, int sectors); 971 972 bool bch_can_invalidate_bucket(struct cache *ca, struct bucket *b); 973 void __bch_invalidate_one_bucket(struct cache *ca, struct bucket *b); 974 975 void __bch_bucket_free(struct cache *ca, struct bucket *b); 976 void bch_bucket_free(struct cache_set *c, struct bkey *k); 977 978 long bch_bucket_alloc(struct cache *ca, unsigned int reserve, bool wait); 979 int __bch_bucket_alloc_set(struct cache_set *c, unsigned int reserve, 980 struct bkey *k, bool wait); 981 int bch_bucket_alloc_set(struct cache_set *c, unsigned int reserve, 982 struct bkey *k, bool wait); 983 bool bch_alloc_sectors(struct cache_set *c, struct bkey *k, 984 unsigned int sectors, unsigned int write_point, 985 unsigned int write_prio, bool wait); 986 bool bch_cached_dev_error(struct cached_dev *dc); 987 988 __printf(2, 3) 989 bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...); 990 991 int bch_prio_write(struct cache *ca, bool wait); 992 void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent); 993 994 extern struct workqueue_struct *bcache_wq; 995 extern struct workqueue_struct *bch_journal_wq; 996 extern struct workqueue_struct *bch_flush_wq; 997 extern struct mutex bch_register_lock; 998 extern struct list_head bch_cache_sets; 999 1000 extern struct kobj_type bch_cached_dev_ktype; 1001 extern struct kobj_type bch_flash_dev_ktype; 1002 extern struct kobj_type bch_cache_set_ktype; 1003 extern struct kobj_type bch_cache_set_internal_ktype; 1004 extern struct kobj_type bch_cache_ktype; 1005 1006 void bch_cached_dev_release(struct kobject *kobj); 1007 void bch_flash_dev_release(struct kobject *kobj); 1008 void bch_cache_set_release(struct kobject *kobj); 1009 void bch_cache_release(struct kobject *kobj); 1010 1011 int bch_uuid_write(struct cache_set *c); 1012 void bcache_write_super(struct cache_set *c); 1013 1014 int bch_flash_dev_create(struct cache_set *c, uint64_t size); 1015 1016 int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c, 1017 uint8_t *set_uuid); 1018 void bch_cached_dev_detach(struct cached_dev *dc); 1019 int bch_cached_dev_run(struct cached_dev *dc); 1020 void bcache_device_stop(struct bcache_device *d); 1021 1022 void bch_cache_set_unregister(struct cache_set *c); 1023 void bch_cache_set_stop(struct cache_set *c); 1024 1025 struct cache_set *bch_cache_set_alloc(struct cache_sb *sb); 1026 void bch_btree_cache_free(struct cache_set *c); 1027 int bch_btree_cache_alloc(struct cache_set *c); 1028 void bch_moving_init_cache_set(struct cache_set *c); 1029 int bch_open_buckets_alloc(struct cache_set *c); 1030 void bch_open_buckets_free(struct cache_set *c); 1031 1032 int bch_cache_allocator_start(struct cache *ca); 1033 1034 void bch_debug_exit(void); 1035 void bch_debug_init(void); 1036 void bch_request_exit(void); 1037 int bch_request_init(void); 1038 void bch_btree_exit(void); 1039 int bch_btree_init(void); 1040 1041 #endif /* _BCACHE_H */ 1042