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()).
180 #define pr_fmt(fmt) "bcachefs: %s() " fmt "\n", __func__
182 #include <linux/backing-dev-defs.h>
183 #include <linux/bug.h>
184 #include <linux/bio.h>
185 #include <linux/closure.h>
186 #include <linux/kobject.h>
187 #include <linux/list.h>
188 #include <linux/mutex.h>
189 #include <linux/percpu-refcount.h>
190 #include <linux/percpu-rwsem.h>
191 #include <linux/rhashtable.h>
192 #include <linux/rwsem.h>
193 #include <linux/seqlock.h>
194 #include <linux/shrinker.h>
195 #include <linux/types.h>
196 #include <linux/workqueue.h>
197 #include <linux/zstd.h>
199 #include "bcachefs_format.h"
204 #define dynamic_fault(...) 0
205 #define race_fault(...) 0
207 #define bch2_fs_init_fault(name) \
208 dynamic_fault("bcachefs:bch_fs_init:" name)
209 #define bch2_meta_read_fault(name) \
210 dynamic_fault("bcachefs:meta:read:" name)
211 #define bch2_meta_write_fault(name) \
212 dynamic_fault("bcachefs:meta:write:" name)
215 #define bch2_fmt(_c, fmt) "bcachefs (%s): " fmt "\n", ((_c)->name)
217 #define bch2_fmt(_c, fmt) fmt "\n"
220 #define bch_info(c, fmt, ...) \
221 printk(KERN_INFO bch2_fmt(c, fmt), ##__VA_ARGS__)
222 #define bch_notice(c, fmt, ...) \
223 printk(KERN_NOTICE bch2_fmt(c, fmt), ##__VA_ARGS__)
224 #define bch_warn(c, fmt, ...) \
225 printk(KERN_WARNING bch2_fmt(c, fmt), ##__VA_ARGS__)
226 #define bch_err(c, fmt, ...) \
227 printk(KERN_ERR bch2_fmt(c, fmt), ##__VA_ARGS__)
229 #define bch_verbose(c, fmt, ...) \
231 if ((c)->opts.verbose_recovery) \
232 bch_info(c, fmt, ##__VA_ARGS__); \
235 #define pr_verbose_init(opts, fmt, ...) \
237 if (opt_get(opts, verbose_init)) \
238 pr_info(fmt, ##__VA_ARGS__); \
241 /* Parameters that are useful for debugging, but should always be compiled in: */
242 #define BCH_DEBUG_PARAMS_ALWAYS() \
243 BCH_DEBUG_PARAM(key_merging_disabled, \
244 "Disables merging of extents") \
245 BCH_DEBUG_PARAM(btree_gc_always_rewrite, \
246 "Causes mark and sweep to compact and rewrite every " \
247 "btree node it traverses") \
248 BCH_DEBUG_PARAM(btree_gc_rewrite_disabled, \
249 "Disables rewriting of btree nodes during mark and sweep")\
250 BCH_DEBUG_PARAM(btree_shrinker_disabled, \
251 "Disables the shrinker callback for the btree node cache")
253 /* Parameters that should only be compiled in in debug mode: */
254 #define BCH_DEBUG_PARAMS_DEBUG() \
255 BCH_DEBUG_PARAM(expensive_debug_checks, \
256 "Enables various runtime debugging checks that " \
257 "significantly affect performance") \
258 BCH_DEBUG_PARAM(debug_check_bkeys, \
259 "Run bkey_debugcheck (primarily checking GC/allocation "\
260 "information) when iterating over keys") \
261 BCH_DEBUG_PARAM(verify_btree_ondisk, \
262 "Reread btree nodes at various points to verify the " \
263 "mergesort in the read path against modifications " \
265 BCH_DEBUG_PARAM(journal_seq_verify, \
266 "Store the journal sequence number in the version " \
267 "number of every btree key, and verify that btree " \
268 "update ordering is preserved during recovery") \
269 BCH_DEBUG_PARAM(inject_invalid_keys, \
270 "Store the journal sequence number in the version " \
271 "number of every btree key, and verify that btree " \
272 "update ordering is preserved during recovery") \
274 #define BCH_DEBUG_PARAMS_ALL() BCH_DEBUG_PARAMS_ALWAYS() BCH_DEBUG_PARAMS_DEBUG()
276 #ifdef CONFIG_BCACHEFS_DEBUG
277 #define BCH_DEBUG_PARAMS() BCH_DEBUG_PARAMS_ALL()
279 #define BCH_DEBUG_PARAMS() BCH_DEBUG_PARAMS_ALWAYS()
282 #define BCH_TIME_STATS() \
283 x(btree_node_mem_alloc) \
288 x(btree_lock_contended_read) \
289 x(btree_lock_contended_intent) \
290 x(btree_lock_contended_write) \
299 enum bch_time_stats {
300 #define x(name) BCH_TIME_##name,
306 #include "alloc_types.h"
307 #include "btree_types.h"
308 #include "buckets_types.h"
309 #include "clock_types.h"
310 #include "journal_types.h"
311 #include "keylist_types.h"
312 #include "quota_types.h"
313 #include "rebalance_types.h"
314 #include "super_types.h"
316 /* Number of nodes btree coalesce will try to coalesce at once */
317 #define GC_MERGE_NODES 4U
319 /* Maximum number of nodes we might need to allocate atomically: */
320 #define BTREE_RESERVE_MAX (BTREE_MAX_DEPTH + (BTREE_MAX_DEPTH - 1))
322 /* Size of the freelist we allocate btree nodes from: */
323 #define BTREE_NODE_RESERVE (BTREE_RESERVE_MAX * 4)
331 #define DEF_BTREE_ID(kwd, val, name) GC_PHASE_BTREE_##kwd,
332 DEFINE_BCH_BTREE_IDS()
335 GC_PHASE_PENDING_DELETE,
347 u64 sectors[2][BCH_DATA_NR];
352 struct percpu_ref ref;
353 struct completion ref_completion;
354 struct percpu_ref io_ref;
355 struct completion io_ref_completion;
361 * Cached version of this device's member info from superblock
362 * Committed by bch2_write_super() -> bch_fs_mi_update()
364 struct bch_member_cpu mi;
366 char name[BDEVNAME_SIZE];
368 struct bch_sb_handle disk_sb;
371 struct bch_devs_mask self;
373 /* biosets used in cloned bios for writing multiple replicas */
374 struct bio_set replica_set;
378 * Per-bucket arrays are protected by c->usage_lock, bucket_lock and
379 * gc_lock, for device resize - holding any is sufficient for access:
380 * Or rcu_read_lock(), but only for ptr_stale():
382 struct bucket_array __rcu *buckets;
383 unsigned long *buckets_dirty;
384 /* most out of date gen in the btree */
386 struct rw_semaphore bucket_lock;
388 struct bch_dev_usage __percpu *usage_percpu;
389 struct bch_dev_usage usage_cached;
392 struct task_struct __rcu *alloc_thread;
395 * free: Buckets that are ready to be used
397 * free_inc: Incoming buckets - these are buckets that currently have
398 * cached data in them, and we can't reuse them until after we write
399 * their new gen to disk. After prio_write() finishes writing the new
400 * gens/prios, they'll be moved to the free list (and possibly discarded
403 alloc_fifo free[RESERVE_NR];
405 spinlock_t freelist_lock;
406 size_t nr_invalidated;
408 u8 open_buckets_partial[OPEN_BUCKETS_COUNT];
409 unsigned open_buckets_partial_nr;
411 size_t fifo_last_bucket;
413 /* last calculated minimum prio */
414 u16 max_last_bucket_io[2];
416 atomic_long_t saturated_count;
417 size_t inc_gen_needs_gc;
418 size_t inc_gen_really_needs_gc;
419 u64 allocator_journal_seq_flush;
420 bool allocator_invalidating_data;
421 bool allocator_blocked;
423 alloc_heap alloc_heap;
426 struct task_struct *copygc_thread;
427 copygc_heap copygc_heap;
428 struct bch_pd_controller copygc_pd;
429 struct write_point copygc_write_point;
431 atomic64_t rebalance_work;
433 struct journal_device journal;
435 struct work_struct io_error_work;
437 /* The rest of this all shows up in sysfs */
438 atomic64_t cur_latency[2];
439 struct bch2_time_stats io_latency[2];
441 #define CONGESTED_MAX 1024
445 struct io_count __percpu *io_done;
449 * Flag bits for what phase of startup/shutdown the cache set is at, how we're
450 * shutting down, etc.:
452 * BCH_FS_UNREGISTERING means we're not just shutting down, we're detaching
453 * all the backing devices first (their cached data gets invalidated, and they
454 * won't automatically reattach).
458 BCH_FS_ALLOC_READ_DONE,
459 BCH_FS_ALLOCATOR_STARTED,
460 BCH_FS_INITIAL_GC_DONE,
466 BCH_FS_WRITE_DISABLE_COMPLETE,
474 BCH_FS_FSCK_FIXED_ERRORS,
476 BCH_FS_REBUILD_REPLICAS,
477 BCH_FS_HOLD_BTREE_WRITES,
482 struct dentry *btree;
483 struct dentry *btree_format;
484 struct dentry *failed;
497 struct list_head list;
499 struct kobject internal;
500 struct kobject opts_dir;
501 struct kobject time_stats;
505 struct device *chardev;
506 struct super_block *vfs_sb;
509 /* ro/rw, add/remove devices: */
510 struct mutex state_lock;
511 enum bch_fs_state state;
513 /* Counts outstanding writes, for clean transition to read-only */
514 struct percpu_ref writes;
515 struct work_struct read_only_work;
517 struct bch_dev __rcu *devs[BCH_SB_MEMBERS_MAX];
519 struct bch_replicas_cpu __rcu *replicas;
520 struct bch_replicas_cpu __rcu *replicas_gc;
521 struct mutex replicas_gc_lock;
523 struct bch_disk_groups_cpu __rcu *disk_groups;
525 struct bch_opts opts;
527 /* Updated by bch2_sb_update():*/
532 u16 encoded_extent_max;
545 struct bch_sb_handle disk_sb;
547 unsigned short block_bits; /* ilog2(block_size) */
549 u16 btree_foreground_merge_threshold;
551 struct closure sb_write;
552 struct mutex sb_lock;
555 struct bio_set btree_bio;
557 struct btree_root btree_roots[BTREE_ID_NR];
558 bool btree_roots_dirty;
559 struct mutex btree_root_lock;
561 struct btree_cache btree_cache;
563 mempool_t btree_reserve_pool;
566 * Cache of allocated btree nodes - if we allocate a btree node and
567 * don't use it, if we free it that space can't be reused until going
568 * _all_ the way through the allocator (which exposes us to a livelock
569 * when allocating btree reserves fail halfway through) - instead, we
570 * can stick them here:
572 struct btree_alloc btree_reserve_cache[BTREE_NODE_RESERVE * 2];
573 unsigned btree_reserve_cache_nr;
574 struct mutex btree_reserve_cache_lock;
576 mempool_t btree_interior_update_pool;
577 struct list_head btree_interior_update_list;
578 struct mutex btree_interior_update_lock;
579 struct closure_waitlist btree_interior_update_wait;
581 struct workqueue_struct *wq;
582 /* copygc needs its own workqueue for index updates.. */
583 struct workqueue_struct *copygc_wq;
586 struct delayed_work pd_controllers_update;
587 unsigned pd_controllers_update_seconds;
589 struct bch_devs_mask rw_devs[BCH_DATA_NR];
591 u64 capacity; /* sectors */
594 * When capacity _decreases_ (due to a disk being removed), we
595 * increment capacity_gen - this invalidates outstanding reservations
596 * and forces them to be revalidated
600 atomic64_t sectors_available;
602 struct bch_fs_usage __percpu *usage_percpu;
603 struct bch_fs_usage usage_cached;
604 struct percpu_rw_semaphore usage_lock;
606 struct closure_waitlist freelist_wait;
609 * When we invalidate buckets, we use both the priority and the amount
610 * of good data to determine which buckets to reuse first - to weight
611 * those together consistently we keep track of the smallest nonzero
612 * priority of any bucket.
614 struct bucket_clock bucket_clock[2];
616 struct io_clock io_clock[2];
619 spinlock_t freelist_lock;
620 u8 open_buckets_freelist;
621 u8 open_buckets_nr_free;
622 struct closure_waitlist open_buckets_wait;
623 struct open_bucket open_buckets[OPEN_BUCKETS_COUNT];
625 struct write_point btree_write_point;
626 struct write_point rebalance_write_point;
628 struct write_point write_points[WRITE_POINT_COUNT];
629 struct hlist_head write_points_hash[WRITE_POINT_COUNT];
630 struct mutex write_points_hash_lock;
632 /* GARBAGE COLLECTION */
633 struct task_struct *gc_thread;
635 unsigned long gc_count;
638 * Tracks GC's progress - everything in the range [ZERO_KEY..gc_cur_pos]
639 * has been marked by GC.
641 * gc_cur_phase is a superset of btree_ids (BTREE_ID_EXTENTS etc.)
643 * gc_cur_phase == GC_PHASE_DONE indicates that gc is finished/not
644 * currently running, and gc marks are currently valid
646 * Protected by gc_pos_lock. Only written to by GC thread, so GC thread
647 * can read without a lock.
649 seqcount_t gc_pos_lock;
650 struct gc_pos gc_pos;
653 * The allocation code needs gc_mark in struct bucket to be correct, but
654 * it's not while a gc is in progress.
656 struct rw_semaphore gc_lock;
659 struct bio_set bio_read;
660 struct bio_set bio_read_split;
661 struct bio_set bio_write;
662 struct mutex bio_bounce_pages_lock;
663 mempool_t bio_bounce_pages;
664 struct rhashtable promote_table;
666 mempool_t compression_bounce[2];
667 mempool_t compress_workspace[BCH_COMPRESSION_NR];
668 mempool_t decompress_workspace;
669 ZSTD_parameters zstd_params;
671 struct crypto_shash *sha256;
672 struct crypto_sync_skcipher *chacha20;
673 struct crypto_shash *poly1305;
675 atomic64_t key_version;
678 struct bch_fs_rebalance rebalance;
680 /* VFS IO PATH - fs-io.c */
681 struct bio_set writepage_bioset;
682 struct bio_set dio_write_bioset;
683 struct bio_set dio_read_bioset;
685 struct bio_list btree_write_error_list;
686 struct work_struct btree_write_error_work;
687 spinlock_t btree_write_error_lock;
690 struct list_head fsck_errors;
691 struct mutex fsck_error_lock;
695 atomic_long_t nr_inodes;
698 struct bch_memquota_type quotas[QTYP_NR];
701 struct dentry *debug;
702 struct btree_debug btree_debug[BTREE_ID_NR];
703 #ifdef CONFIG_BCACHEFS_DEBUG
704 struct btree *verify_data;
705 struct btree_node *verify_ondisk;
706 struct mutex verify_lock;
709 u64 unused_inode_hint;
712 * A btree node on disk could have too many bsets for an iterator to fit
713 * on the stack - have to dynamically allocate them
717 mempool_t btree_bounce_pool;
719 struct journal journal;
721 unsigned bucket_journal_seq;
723 /* The rest of this all shows up in sysfs */
724 atomic_long_t read_realloc_races;
725 atomic_long_t extent_migrate_done;
726 atomic_long_t extent_migrate_raced;
728 unsigned btree_gc_periodic:1;
729 unsigned copy_gc_enabled:1;
730 bool promote_whole_extents;
732 #define BCH_DEBUG_PARAM(name, description) bool name;
733 BCH_DEBUG_PARAMS_ALL()
734 #undef BCH_DEBUG_PARAM
736 struct bch2_time_stats times[BCH_TIME_STAT_NR];
739 static inline void bch2_set_ra_pages(struct bch_fs *c, unsigned ra_pages)
741 #ifndef NO_BCACHEFS_FS
743 c->vfs_sb->s_bdi->ra_pages = ra_pages;
747 static inline bool bch2_fs_running(struct bch_fs *c)
749 return c->state == BCH_FS_RO || c->state == BCH_FS_RW;
752 static inline unsigned bucket_bytes(const struct bch_dev *ca)
754 return ca->mi.bucket_size << 9;
757 static inline unsigned block_bytes(const struct bch_fs *c)
759 return c->opts.block_size << 9;
762 static inline struct timespec64 bch2_time_to_timespec(struct bch_fs *c, u64 time)
764 return ns_to_timespec64(time * c->sb.time_precision + c->sb.time_base_lo);
767 static inline s64 timespec_to_bch2_time(struct bch_fs *c, struct timespec64 ts)
769 s64 ns = timespec64_to_ns(&ts) - c->sb.time_base_lo;
771 if (c->sb.time_precision == 1)
774 return div_s64(ns, c->sb.time_precision);
777 static inline s64 bch2_current_time(struct bch_fs *c)
779 struct timespec64 now;
781 ktime_get_real_ts64(&now);
782 return timespec_to_bch2_time(c, now);
785 #endif /* _BCACHEFS_H */