1 /* SPDX-License-Identifier: GPL-2.0 */
6 * SOME HIGH LEVEL CODE DOCUMENTATION:
8 * Bcache mostly works with cache sets, cache devices, and backing devices.
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.
15 * Multiple cache devices is intended to give us the ability to mirror dirty
16 * cached data and metadata, without mirroring clean cached data.
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.
24 * A cache set can have multiple (many) backing devices attached to it.
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.
32 * Flash only volumes work but they're not production ready because the moving
33 * garbage collector needs more work. More on that later.
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
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+
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).
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.
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
62 * Bcache is entirely COW - we never write twice to a bucket, even buckets that
63 * contain metadata (including btree nodes).
67 * Bcache is in large part design around the btree.
69 * At a high level, the btree is just an index of key -> ptr tuples.
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).
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.
80 * Pointers contain the cache device id, the offset on that device, and an 8 bit
81 * generation number. More on the gen later.
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
87 * Updates are fairly well abstracted, though. There are two different ways of
88 * updating the btree; insert and replace.
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.
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.
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.
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.
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.
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.)
118 * Anyways, btree nodes are big - big enough to be inefficient with a textbook
119 * btree implementation.
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.
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
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.
137 * GARBAGE COLLECTION:
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.
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.
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.
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
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.
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
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()).
179 #define pr_fmt(fmt) "bcache: %s() " fmt "\n", __func__
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 #include <linux/kthread.h>
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 */
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
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);
232 struct bkey last_scanned;
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
245 #define KEYBUF_NR 500
246 DECLARE_ARRAY_ALLOCATOR(struct keybuf_key, freelist, KEYBUF_NR);
249 struct bcache_device {
256 #define BCACHEDEVNAME_SIZE 12
257 char name[BCACHEDEVNAME_SIZE];
259 struct gendisk *disk;
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
268 #define BCH_MIN_STRIPE_SZ ((4 << 20) >> SECTOR_SHIFT)
269 unsigned int stripe_size;
270 atomic_t *stripe_sectors_dirty;
271 unsigned long *full_dirty_stripes;
273 struct bio_set bio_split;
275 unsigned int data_csum:1;
277 int (*cache_miss)(struct btree *b, struct search *s,
278 struct bio *bio, unsigned int sectors);
279 int (*ioctl)(struct bcache_device *d, fmode_t mode,
280 unsigned int cmd, unsigned long arg);
284 /* Used to track sequential IO so it can be skipped */
285 struct hlist_node hash;
286 struct list_head lru;
288 unsigned long jiffies;
289 unsigned int sequential;
293 enum stop_on_failure {
294 BCH_CACHED_DEV_STOP_AUTO = 0,
295 BCH_CACHED_DEV_STOP_ALWAYS,
296 BCH_CACHED_DEV_STOP_MODE_MAX,
300 struct list_head list;
301 struct bcache_device disk;
302 struct block_device *bdev;
306 struct bio_vec sb_bv[1];
307 struct closure sb_write;
308 struct semaphore sb_write_mutex;
310 /* Refcount on the cache set. Always nonzero when we're caching. */
312 struct work_struct detach;
315 * Device might not be running if it's dirty and the cache set hasn't
321 * Writes take a shared lock from start to finish; scanning for dirty
322 * data to refill the rb tree requires an exclusive lock.
324 struct rw_semaphore writeback_lock;
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.
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;
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;
344 struct keybuf writeback_keys;
346 struct task_struct *status_update_thread;
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...)
352 struct closure_waitlist writeback_ordering_wait;
353 atomic_t writeback_sequence_next;
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;
363 struct cache_accounting accounting;
365 /* The rest of this all shows up in sysfs */
366 unsigned int sequential_cutoff;
367 unsigned int readahead;
369 unsigned int io_disable:1;
370 unsigned int verify:1;
371 unsigned int bypass_torture_test:1;
373 unsigned int partial_stripes_expensive:1;
374 unsigned int writeback_metadata:1;
375 unsigned int writeback_running:1;
376 unsigned char writeback_percent;
377 unsigned int writeback_delay;
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;
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_minimum;
390 enum stop_on_failure stop_when_cache_set_failed;
391 #define DEFAULT_CACHED_DEV_ERROR_LIMIT 64
393 unsigned int error_limit;
394 unsigned int offline_seconds;
396 char backing_dev_name[BDEVNAME_SIZE];
408 struct cache_set *set;
411 struct bio_vec sb_bv[1];
414 struct block_device *bdev;
416 struct task_struct *alloc_thread;
419 struct prio_set *disk_buckets;
422 * When allocating new buckets, prio_write() gets first dibs - since we
423 * may not be allocate at all without writing priorities and gens.
424 * prio_last_buckets[] contains the last buckets we wrote priorities to
425 * (so gc can mark them as metadata), prio_buckets[] contains the
426 * buckets allocated for the next prio write.
428 uint64_t *prio_buckets;
429 uint64_t *prio_last_buckets;
432 * free: Buckets that are ready to be used
434 * free_inc: Incoming buckets - these are buckets that currently have
435 * cached data in them, and we can't reuse them until after we write
436 * their new gen to disk. After prio_write() finishes writing the new
437 * gens/prios, they'll be moved to the free list (and possibly discarded
440 DECLARE_FIFO(long, free)[RESERVE_NR];
441 DECLARE_FIFO(long, free_inc);
443 size_t fifo_last_bucket;
445 /* Allocation stuff: */
446 struct bucket *buckets;
448 DECLARE_HEAP(struct bucket *, heap);
451 * If nonzero, we know we aren't going to find any buckets to invalidate
452 * until a gc finishes - otherwise we could pointlessly burn a ton of
455 unsigned int invalidate_needs_gc;
457 bool discard; /* Get rid of? */
459 struct journal_device journal;
461 /* The rest of this all shows up in sysfs */
462 #define IO_ERROR_SHIFT 20
466 atomic_long_t meta_sectors_written;
467 atomic_long_t btree_sectors_written;
468 atomic_long_t sectors_written;
470 char cache_dev_name[BDEVNAME_SIZE];
479 uint64_t data; /* sectors */
480 unsigned int in_use; /* percent */
484 * Flag bits, for how the cache set is shutting down, and what phase it's at:
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).
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).
494 * CACHE_SET_RUNNING means all cache devices have been registered and journal
495 * replay is complete.
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.
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
509 struct list_head list;
511 struct kobject internal;
512 struct dentry *debug;
513 struct cache_accounting accounting;
516 atomic_t idle_counter;
517 atomic_t at_max_writeback_rate;
521 struct cache *cache[MAX_CACHES_PER_SET];
522 struct cache *cache_by_alloc[MAX_CACHES_PER_SET];
525 struct bcache_device **devices;
526 unsigned int devices_max_used;
527 atomic_t attached_dev_nr;
528 struct list_head cached_devs;
529 uint64_t cached_dev_sectors;
530 atomic_long_t flash_dev_dirty_sectors;
531 struct closure caching;
533 struct closure sb_write;
534 struct semaphore sb_write_mutex;
538 struct bio_set bio_split;
540 /* For the btree cache */
541 struct shrinker shrink;
543 /* For the btree cache and anything allocation related */
544 struct mutex bucket_lock;
546 /* log2(bucket_size), in sectors */
547 unsigned short bucket_bits;
549 /* log2(block_size), in sectors */
550 unsigned short block_bits;
553 * Default number of pages for a new btree node - may be less than a
556 unsigned int btree_pages;
559 * Lists of struct btrees; lru is the list for structs that have memory
560 * allocated for actual btree node, freed is for structs that do not.
562 * We never free a struct btree, except on shutdown - we just put it on
563 * the btree_cache_freed list and reuse it later. This simplifies the
564 * code, and it doesn't cost us much memory as the memory usage is
565 * dominated by buffers that hold the actual btree node data and those
566 * can be freed - and the number of struct btrees allocated is
567 * effectively bounded.
569 * btree_cache_freeable effectively is a small cache - we use it because
570 * high order page allocations can be rather expensive, and it's quite
571 * common to delete and allocate btree nodes in quick succession. It
572 * should never grow past ~2-3 nodes in practice.
574 struct list_head btree_cache;
575 struct list_head btree_cache_freeable;
576 struct list_head btree_cache_freed;
578 /* Number of elements in btree_cache + btree_cache_freeable lists */
579 unsigned int btree_cache_used;
582 * If we need to allocate memory for a new btree node and that
583 * allocation fails, we can cannibalize another node in the btree cache
584 * to satisfy the allocation - lock to guarantee only one thread does
587 wait_queue_head_t btree_cache_wait;
588 struct task_struct *btree_cache_alloc_lock;
589 spinlock_t btree_cannibalize_lock;
592 * When we free a btree node, we increment the gen of the bucket the
593 * node is in - but we can't rewrite the prios and gens until we
594 * finished whatever it is we were doing, otherwise after a crash the
595 * btree node would be freed but for say a split, we might not have the
596 * pointers to the new nodes inserted into the btree yet.
598 * This is a refcount that blocks prio_write() until the new keys are
601 atomic_t prio_blocked;
602 wait_queue_head_t bucket_wait;
605 * For any bio we don't skip we subtract the number of sectors from
606 * rescale; when it hits 0 we rescale all the bucket priorities.
610 * used for GC, identify if any front side I/Os is inflight
612 atomic_t search_inflight;
614 * When we invalidate buckets, we use both the priority and the amount
615 * of good data to determine which buckets to reuse first - to weight
616 * those together consistently we keep track of the smallest nonzero
617 * priority of any bucket.
622 * max(gen - last_gc) for all buckets. When it gets too big we have to
623 * gc to keep gens from wrapping around.
626 struct gc_stat gc_stats;
628 size_t avail_nbuckets;
630 struct task_struct *gc_thread;
631 /* Where in the btree gc currently is */
635 * The allocation code needs gc_mark in struct bucket to be correct, but
636 * it's not while a gc is in progress. Protected by bucket_lock.
640 /* Counts how many sectors bio_insert has added to the cache */
641 atomic_t sectors_to_gc;
642 wait_queue_head_t gc_wait;
644 struct keybuf moving_gc_keys;
645 /* Number of moving GC bios in flight */
646 struct semaphore moving_in_flight;
648 struct workqueue_struct *moving_gc_wq;
652 #ifdef CONFIG_BCACHE_DEBUG
653 struct btree *verify_data;
654 struct bset *verify_ondisk;
655 struct mutex verify_lock;
658 unsigned int nr_uuids;
659 struct uuid_entry *uuids;
660 BKEY_PADDED(uuid_bucket);
661 struct closure uuid_write;
662 struct semaphore uuid_write_mutex;
665 * A btree node on disk could have too many bsets for an iterator to fit
666 * on the stack - have to dynamically allocate them
670 struct bset_sort_state sort;
672 /* List of buckets we're currently writing data to */
673 struct list_head data_buckets;
674 spinlock_t data_bucket_lock;
676 struct journal journal;
678 #define CONGESTED_MAX 1024
679 unsigned int congested_last_us;
682 /* The rest of this all shows up in sysfs */
683 unsigned int congested_read_threshold_us;
684 unsigned int congested_write_threshold_us;
686 struct time_stats btree_gc_time;
687 struct time_stats btree_split_time;
688 struct time_stats btree_read_time;
690 atomic_long_t cache_read_races;
691 atomic_long_t writeback_keys_done;
692 atomic_long_t writeback_keys_failed;
694 atomic_long_t reclaim;
695 atomic_long_t flush_write;
696 atomic_long_t retry_flush_write;
702 #define DEFAULT_IO_ERROR_LIMIT 8
703 unsigned int error_limit;
704 unsigned int error_decay;
706 unsigned short journal_delay_ms;
707 bool expensive_debug_checks;
708 unsigned int verify:1;
709 unsigned int key_merging_disabled:1;
710 unsigned int gc_always_rewrite:1;
711 unsigned int shrinker_disabled:1;
712 unsigned int copy_gc_enabled:1;
714 #define BUCKET_HASH_BITS 12
715 struct hlist_head bucket_hash[1 << BUCKET_HASH_BITS];
719 unsigned int submit_time_us;
724 * We only need pad = 3 here because we only ever carry around a
725 * single pointer - i.e. the pointer we're doing io to/from.
731 #define BTREE_PRIO USHRT_MAX
732 #define INITIAL_PRIO 32768U
734 #define btree_bytes(c) ((c)->btree_pages * PAGE_SIZE)
735 #define btree_blocks(b) \
736 ((unsigned int) (KEY_SIZE(&b->key) >> (b)->c->block_bits))
738 #define btree_default_blocks(c) \
739 ((unsigned int) ((PAGE_SECTORS * (c)->btree_pages) >> (c)->block_bits))
741 #define bucket_pages(c) ((c)->sb.bucket_size / PAGE_SECTORS)
742 #define bucket_bytes(c) ((c)->sb.bucket_size << 9)
743 #define block_bytes(c) ((c)->sb.block_size << 9)
745 #define prios_per_bucket(c) \
746 ((bucket_bytes(c) - sizeof(struct prio_set)) / \
747 sizeof(struct bucket_disk))
748 #define prio_buckets(c) \
749 DIV_ROUND_UP((size_t) (c)->sb.nbuckets, prios_per_bucket(c))
751 static inline size_t sector_to_bucket(struct cache_set *c, sector_t s)
753 return s >> c->bucket_bits;
756 static inline sector_t bucket_to_sector(struct cache_set *c, size_t b)
758 return ((sector_t) b) << c->bucket_bits;
761 static inline sector_t bucket_remainder(struct cache_set *c, sector_t s)
763 return s & (c->sb.bucket_size - 1);
766 static inline struct cache *PTR_CACHE(struct cache_set *c,
767 const struct bkey *k,
770 return c->cache[PTR_DEV(k, ptr)];
773 static inline size_t PTR_BUCKET_NR(struct cache_set *c,
774 const struct bkey *k,
777 return sector_to_bucket(c, PTR_OFFSET(k, ptr));
780 static inline struct bucket *PTR_BUCKET(struct cache_set *c,
781 const struct bkey *k,
784 return PTR_CACHE(c, k, ptr)->buckets + PTR_BUCKET_NR(c, k, ptr);
787 static inline uint8_t gen_after(uint8_t a, uint8_t b)
791 return r > 128U ? 0 : r;
794 static inline uint8_t ptr_stale(struct cache_set *c, const struct bkey *k,
797 return gen_after(PTR_BUCKET(c, k, i)->gen, PTR_GEN(k, i));
800 static inline bool ptr_available(struct cache_set *c, const struct bkey *k,
803 return (PTR_DEV(k, i) < MAX_CACHES_PER_SET) && PTR_CACHE(c, k, i);
806 /* Btree key macros */
809 * This is used for various on disk data structures - cache_sb, prio_set, bset,
810 * jset: The checksum is _always_ the first 8 bytes of these structs
812 #define csum_set(i) \
813 bch_crc64(((void *) (i)) + sizeof(uint64_t), \
814 ((void *) bset_bkey_last(i)) - \
815 (((void *) (i)) + sizeof(uint64_t)))
817 /* Error handling macros */
819 #define btree_bug(b, ...) \
821 if (bch_cache_set_error((b)->c, __VA_ARGS__)) \
825 #define cache_bug(c, ...) \
827 if (bch_cache_set_error(c, __VA_ARGS__)) \
831 #define btree_bug_on(cond, b, ...) \
834 btree_bug(b, __VA_ARGS__); \
837 #define cache_bug_on(cond, c, ...) \
840 cache_bug(c, __VA_ARGS__); \
843 #define cache_set_err_on(cond, c, ...) \
846 bch_cache_set_error(c, __VA_ARGS__); \
851 #define for_each_cache(ca, cs, iter) \
852 for (iter = 0; ca = cs->cache[iter], iter < (cs)->sb.nr_in_set; iter++)
854 #define for_each_bucket(b, ca) \
855 for (b = (ca)->buckets + (ca)->sb.first_bucket; \
856 b < (ca)->buckets + (ca)->sb.nbuckets; b++)
858 static inline void cached_dev_put(struct cached_dev *dc)
860 if (refcount_dec_and_test(&dc->count))
861 schedule_work(&dc->detach);
864 static inline bool cached_dev_get(struct cached_dev *dc)
866 if (!refcount_inc_not_zero(&dc->count))
869 /* Paired with the mb in cached_dev_attach */
870 smp_mb__after_atomic();
875 * bucket_gc_gen() returns the difference between the bucket's current gen and
876 * the oldest gen of any pointer into that bucket in the btree (last_gc).
879 static inline uint8_t bucket_gc_gen(struct bucket *b)
881 return b->gen - b->last_gc;
884 #define BUCKET_GC_GEN_MAX 96U
886 #define kobj_attribute_write(n, fn) \
887 static struct kobj_attribute ksysfs_##n = __ATTR(n, 0200, NULL, fn)
889 #define kobj_attribute_rw(n, show, store) \
890 static struct kobj_attribute ksysfs_##n = \
891 __ATTR(n, 0600, show, store)
893 static inline void wake_up_allocators(struct cache_set *c)
898 for_each_cache(ca, c, i)
899 wake_up_process(ca->alloc_thread);
902 static inline void closure_bio_submit(struct cache_set *c,
907 if (unlikely(test_bit(CACHE_SET_IO_DISABLE, &c->flags))) {
908 bio->bi_status = BLK_STS_IOERR;
912 generic_make_request(bio);
916 * Prevent the kthread exits directly, and make sure when kthread_stop()
917 * is called to stop a kthread, it is still alive. If a kthread might be
918 * stopped by CACHE_SET_IO_DISABLE bit set, wait_for_kthread_stop() is
919 * necessary before the kthread returns.
921 static inline void wait_for_kthread_stop(void)
923 while (!kthread_should_stop()) {
924 set_current_state(TASK_INTERRUPTIBLE);
929 /* Forward declarations */
931 void bch_count_backing_io_errors(struct cached_dev *dc, struct bio *bio);
932 void bch_count_io_errors(struct cache *ca, blk_status_t error,
933 int is_read, const char *m);
934 void bch_bbio_count_io_errors(struct cache_set *c, struct bio *bio,
935 blk_status_t error, const char *m);
936 void bch_bbio_endio(struct cache_set *c, struct bio *bio,
937 blk_status_t error, const char *m);
938 void bch_bbio_free(struct bio *bio, struct cache_set *c);
939 struct bio *bch_bbio_alloc(struct cache_set *c);
941 void __bch_submit_bbio(struct bio *bio, struct cache_set *c);
942 void bch_submit_bbio(struct bio *bio, struct cache_set *c,
943 struct bkey *k, unsigned int ptr);
945 uint8_t bch_inc_gen(struct cache *ca, struct bucket *b);
946 void bch_rescale_priorities(struct cache_set *c, int sectors);
948 bool bch_can_invalidate_bucket(struct cache *ca, struct bucket *b);
949 void __bch_invalidate_one_bucket(struct cache *ca, struct bucket *b);
951 void __bch_bucket_free(struct cache *ca, struct bucket *b);
952 void bch_bucket_free(struct cache_set *c, struct bkey *k);
954 long bch_bucket_alloc(struct cache *ca, unsigned int reserve, bool wait);
955 int __bch_bucket_alloc_set(struct cache_set *c, unsigned int reserve,
956 struct bkey *k, bool wait);
957 int bch_bucket_alloc_set(struct cache_set *c, unsigned int reserve,
958 struct bkey *k, bool wait);
959 bool bch_alloc_sectors(struct cache_set *c, struct bkey *k,
960 unsigned int sectors, unsigned int write_point,
961 unsigned int write_prio, bool wait);
962 bool bch_cached_dev_error(struct cached_dev *dc);
965 bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...);
967 int bch_prio_write(struct cache *ca, bool wait);
968 void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent);
970 extern struct workqueue_struct *bcache_wq;
971 extern struct workqueue_struct *bch_journal_wq;
972 extern struct mutex bch_register_lock;
973 extern struct list_head bch_cache_sets;
975 extern struct kobj_type bch_cached_dev_ktype;
976 extern struct kobj_type bch_flash_dev_ktype;
977 extern struct kobj_type bch_cache_set_ktype;
978 extern struct kobj_type bch_cache_set_internal_ktype;
979 extern struct kobj_type bch_cache_ktype;
981 void bch_cached_dev_release(struct kobject *kobj);
982 void bch_flash_dev_release(struct kobject *kobj);
983 void bch_cache_set_release(struct kobject *kobj);
984 void bch_cache_release(struct kobject *kobj);
986 int bch_uuid_write(struct cache_set *c);
987 void bcache_write_super(struct cache_set *c);
989 int bch_flash_dev_create(struct cache_set *c, uint64_t size);
991 int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c,
993 void bch_cached_dev_detach(struct cached_dev *dc);
994 void bch_cached_dev_run(struct cached_dev *dc);
995 void bcache_device_stop(struct bcache_device *d);
997 void bch_cache_set_unregister(struct cache_set *c);
998 void bch_cache_set_stop(struct cache_set *c);
1000 struct cache_set *bch_cache_set_alloc(struct cache_sb *sb);
1001 void bch_btree_cache_free(struct cache_set *c);
1002 int bch_btree_cache_alloc(struct cache_set *c);
1003 void bch_moving_init_cache_set(struct cache_set *c);
1004 int bch_open_buckets_alloc(struct cache_set *c);
1005 void bch_open_buckets_free(struct cache_set *c);
1007 int bch_cache_allocator_start(struct cache *ca);
1009 void bch_debug_exit(void);
1010 void bch_debug_init(struct kobject *kobj);
1011 void bch_request_exit(void);
1012 int bch_request_init(void);
1014 #endif /* _BCACHE_H */