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 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()).
181 #define pr_fmt(fmt) "bcachefs: %s() " fmt "\n", __func__
183 #define pr_fmt(fmt) "%s() " fmt "\n", __func__
186 #include <linux/backing-dev-defs.h>
187 #include <linux/bug.h>
188 #include <linux/bio.h>
189 #include <linux/closure.h>
190 #include <linux/kobject.h>
191 #include <linux/list.h>
192 #include <linux/math64.h>
193 #include <linux/mutex.h>
194 #include <linux/percpu-refcount.h>
195 #include <linux/percpu-rwsem.h>
196 #include <linux/rhashtable.h>
197 #include <linux/rwsem.h>
198 #include <linux/semaphore.h>
199 #include <linux/seqlock.h>
200 #include <linux/shrinker.h>
201 #include <linux/srcu.h>
202 #include <linux/types.h>
203 #include <linux/workqueue.h>
204 #include <linux/zstd.h>
206 #include "bcachefs_format.h"
209 #include "nocow_locking_types.h"
211 #include "seqmutex.h"
214 #ifdef CONFIG_BCACHEFS_DEBUG
215 #define BCH_WRITE_REF_DEBUG
218 #ifndef dynamic_fault
219 #define dynamic_fault(...) 0
222 #define race_fault(...) dynamic_fault("bcachefs:race")
224 #define trace_and_count(_c, _name, ...) \
226 this_cpu_inc((_c)->counters[BCH_COUNTER_##_name]); \
227 trace_##_name(__VA_ARGS__); \
230 #define bch2_fs_init_fault(name) \
231 dynamic_fault("bcachefs:bch_fs_init:" name)
232 #define bch2_meta_read_fault(name) \
233 dynamic_fault("bcachefs:meta:read:" name)
234 #define bch2_meta_write_fault(name) \
235 dynamic_fault("bcachefs:meta:write:" name)
238 #define BCACHEFS_LOG_PREFIX
241 #ifdef BCACHEFS_LOG_PREFIX
243 #define bch2_log_msg(_c, fmt) "bcachefs (%s): " fmt, ((_c)->name)
244 #define bch2_fmt_dev(_ca, fmt) "bcachefs (%s): " fmt "\n", ((_ca)->name)
245 #define bch2_fmt_dev_offset(_ca, _offset, fmt) "bcachefs (%s sector %llu): " fmt "\n", ((_ca)->name), (_offset)
246 #define bch2_fmt_inum(_c, _inum, fmt) "bcachefs (%s inum %llu): " fmt "\n", ((_c)->name), (_inum)
247 #define bch2_fmt_inum_offset(_c, _inum, _offset, fmt) \
248 "bcachefs (%s inum %llu offset %llu): " fmt "\n", ((_c)->name), (_inum), (_offset)
252 #define bch2_log_msg(_c, fmt) fmt
253 #define bch2_fmt_dev(_ca, fmt) "%s: " fmt "\n", ((_ca)->name)
254 #define bch2_fmt_dev_offset(_ca, _offset, fmt) "%s sector %llu: " fmt "\n", ((_ca)->name), (_offset)
255 #define bch2_fmt_inum(_c, _inum, fmt) "inum %llu: " fmt "\n", (_inum)
256 #define bch2_fmt_inum_offset(_c, _inum, _offset, fmt) \
257 "inum %llu offset %llu: " fmt "\n", (_inum), (_offset)
261 #define bch2_fmt(_c, fmt) bch2_log_msg(_c, fmt "\n")
263 #define bch_info(c, fmt, ...) \
264 printk(KERN_INFO bch2_fmt(c, fmt), ##__VA_ARGS__)
265 #define bch_notice(c, fmt, ...) \
266 printk(KERN_NOTICE bch2_fmt(c, fmt), ##__VA_ARGS__)
267 #define bch_warn(c, fmt, ...) \
268 printk(KERN_WARNING bch2_fmt(c, fmt), ##__VA_ARGS__)
269 #define bch_warn_ratelimited(c, fmt, ...) \
270 printk_ratelimited(KERN_WARNING bch2_fmt(c, fmt), ##__VA_ARGS__)
272 #define bch_err(c, fmt, ...) \
273 printk(KERN_ERR bch2_fmt(c, fmt), ##__VA_ARGS__)
274 #define bch_err_dev(ca, fmt, ...) \
275 printk(KERN_ERR bch2_fmt_dev(ca, fmt), ##__VA_ARGS__)
276 #define bch_err_dev_offset(ca, _offset, fmt, ...) \
277 printk(KERN_ERR bch2_fmt_dev_offset(ca, _offset, fmt), ##__VA_ARGS__)
278 #define bch_err_inum(c, _inum, fmt, ...) \
279 printk(KERN_ERR bch2_fmt_inum(c, _inum, fmt), ##__VA_ARGS__)
280 #define bch_err_inum_offset(c, _inum, _offset, fmt, ...) \
281 printk(KERN_ERR bch2_fmt_inum_offset(c, _inum, _offset, fmt), ##__VA_ARGS__)
283 #define bch_err_ratelimited(c, fmt, ...) \
284 printk_ratelimited(KERN_ERR bch2_fmt(c, fmt), ##__VA_ARGS__)
285 #define bch_err_dev_ratelimited(ca, fmt, ...) \
286 printk_ratelimited(KERN_ERR bch2_fmt_dev(ca, fmt), ##__VA_ARGS__)
287 #define bch_err_dev_offset_ratelimited(ca, _offset, fmt, ...) \
288 printk_ratelimited(KERN_ERR bch2_fmt_dev_offset(ca, _offset, fmt), ##__VA_ARGS__)
289 #define bch_err_inum_ratelimited(c, _inum, fmt, ...) \
290 printk_ratelimited(KERN_ERR bch2_fmt_inum(c, _inum, fmt), ##__VA_ARGS__)
291 #define bch_err_inum_offset_ratelimited(c, _inum, _offset, fmt, ...) \
292 printk_ratelimited(KERN_ERR bch2_fmt_inum_offset(c, _inum, _offset, fmt), ##__VA_ARGS__)
294 #define bch_err_fn(_c, _ret) \
295 bch_err(_c, "%s(): error %s", __func__, bch2_err_str(_ret))
296 #define bch_err_msg(_c, _ret, _msg) \
297 bch_err(_c, "%s(): error " _msg " %s", __func__, bch2_err_str(_ret))
299 #define bch_verbose(c, fmt, ...) \
301 if ((c)->opts.verbose) \
302 bch_info(c, fmt, ##__VA_ARGS__); \
305 #define pr_verbose_init(opts, fmt, ...) \
307 if (opt_get(opts, verbose)) \
308 pr_info(fmt, ##__VA_ARGS__); \
311 /* Parameters that are useful for debugging, but should always be compiled in: */
312 #define BCH_DEBUG_PARAMS_ALWAYS() \
313 BCH_DEBUG_PARAM(key_merging_disabled, \
314 "Disables merging of extents") \
315 BCH_DEBUG_PARAM(btree_gc_always_rewrite, \
316 "Causes mark and sweep to compact and rewrite every " \
317 "btree node it traverses") \
318 BCH_DEBUG_PARAM(btree_gc_rewrite_disabled, \
319 "Disables rewriting of btree nodes during mark and sweep")\
320 BCH_DEBUG_PARAM(btree_shrinker_disabled, \
321 "Disables the shrinker callback for the btree node cache")\
322 BCH_DEBUG_PARAM(verify_btree_ondisk, \
323 "Reread btree nodes at various points to verify the " \
324 "mergesort in the read path against modifications " \
326 BCH_DEBUG_PARAM(verify_all_btree_replicas, \
327 "When reading btree nodes, read all replicas and " \
329 BCH_DEBUG_PARAM(backpointers_no_use_write_buffer, \
330 "Don't use the write buffer for backpointers, enabling "\
331 "extra runtime checks")
333 /* Parameters that should only be compiled in debug mode: */
334 #define BCH_DEBUG_PARAMS_DEBUG() \
335 BCH_DEBUG_PARAM(expensive_debug_checks, \
336 "Enables various runtime debugging checks that " \
337 "significantly affect performance") \
338 BCH_DEBUG_PARAM(debug_check_iterators, \
339 "Enables extra verification for btree iterators") \
340 BCH_DEBUG_PARAM(debug_check_btree_accounting, \
341 "Verify btree accounting for keys within a node") \
342 BCH_DEBUG_PARAM(journal_seq_verify, \
343 "Store the journal sequence number in the version " \
344 "number of every btree key, and verify that btree " \
345 "update ordering is preserved during recovery") \
346 BCH_DEBUG_PARAM(inject_invalid_keys, \
347 "Store the journal sequence number in the version " \
348 "number of every btree key, and verify that btree " \
349 "update ordering is preserved during recovery") \
350 BCH_DEBUG_PARAM(test_alloc_startup, \
351 "Force allocator startup to use the slowpath where it" \
352 "can't find enough free buckets without invalidating" \
354 BCH_DEBUG_PARAM(force_reconstruct_read, \
355 "Force reads to use the reconstruct path, when reading" \
356 "from erasure coded extents") \
357 BCH_DEBUG_PARAM(test_restart_gc, \
358 "Test restarting mark and sweep gc when bucket gens change")
360 #define BCH_DEBUG_PARAMS_ALL() BCH_DEBUG_PARAMS_ALWAYS() BCH_DEBUG_PARAMS_DEBUG()
362 #ifdef CONFIG_BCACHEFS_DEBUG
363 #define BCH_DEBUG_PARAMS() BCH_DEBUG_PARAMS_ALL()
365 #define BCH_DEBUG_PARAMS() BCH_DEBUG_PARAMS_ALWAYS()
368 #define BCH_DEBUG_PARAM(name, description) extern bool bch2_##name;
370 #undef BCH_DEBUG_PARAM
372 #ifndef CONFIG_BCACHEFS_DEBUG
373 #define BCH_DEBUG_PARAM(name, description) static const bool bch2_##name;
374 BCH_DEBUG_PARAMS_DEBUG()
375 #undef BCH_DEBUG_PARAM
378 #define BCH_TIME_STATS() \
379 x(btree_node_mem_alloc) \
380 x(btree_node_split) \
381 x(btree_node_compact) \
382 x(btree_node_merge) \
385 x(btree_interior_update_foreground) \
386 x(btree_interior_update_total) \
391 x(journal_flush_write) \
392 x(journal_noflush_write) \
393 x(journal_flush_seq) \
395 x(blocked_allocate) \
396 x(blocked_allocate_open_bucket) \
397 x(nocow_lock_contended)
399 enum bch_time_stats {
400 #define x(name) BCH_TIME_##name,
406 #include "alloc_types.h"
407 #include "btree_types.h"
408 #include "btree_write_buffer_types.h"
409 #include "buckets_types.h"
410 #include "buckets_waiting_for_journal_types.h"
411 #include "clock_types.h"
412 #include "ec_types.h"
413 #include "journal_types.h"
414 #include "keylist_types.h"
415 #include "quota_types.h"
416 #include "rebalance_types.h"
417 #include "replicas_types.h"
418 #include "subvolume_types.h"
419 #include "super_types.h"
421 /* Number of nodes btree coalesce will try to coalesce at once */
422 #define GC_MERGE_NODES 4U
424 /* Maximum number of nodes we might need to allocate atomically: */
425 #define BTREE_RESERVE_MAX (BTREE_MAX_DEPTH + (BTREE_MAX_DEPTH - 1))
427 /* Size of the freelist we allocate btree nodes from: */
428 #define BTREE_NODE_RESERVE (BTREE_RESERVE_MAX * 4)
430 #define BTREE_NODE_OPEN_BUCKET_RESERVE (BTREE_RESERVE_MAX * BCH_REPLICAS_MAX)
435 GC_PHASE_NOT_RUNNING,
439 GC_PHASE_BTREE_stripes,
440 GC_PHASE_BTREE_extents,
441 GC_PHASE_BTREE_inodes,
442 GC_PHASE_BTREE_dirents,
443 GC_PHASE_BTREE_xattrs,
444 GC_PHASE_BTREE_alloc,
445 GC_PHASE_BTREE_quotas,
446 GC_PHASE_BTREE_reflink,
447 GC_PHASE_BTREE_subvolumes,
448 GC_PHASE_BTREE_snapshots,
450 GC_PHASE_BTREE_freespace,
451 GC_PHASE_BTREE_need_discard,
452 GC_PHASE_BTREE_backpointers,
453 GC_PHASE_BTREE_bucket_gens,
454 GC_PHASE_BTREE_snapshot_trees,
456 GC_PHASE_PENDING_DELETE,
471 typedef GENRADIX(struct reflink_gc) reflink_gc_table;
474 u64 sectors[2][BCH_DATA_NR];
479 struct percpu_ref ref;
480 struct completion ref_completion;
481 struct percpu_ref io_ref;
482 struct completion io_ref_completion;
488 * Cached version of this device's member info from superblock
489 * Committed by bch2_write_super() -> bch_fs_mi_update()
491 struct bch_member_cpu mi;
493 char name[BDEVNAME_SIZE];
495 struct bch_sb_handle disk_sb;
496 struct bch_sb *sb_read_scratch;
501 struct bch_devs_mask self;
503 /* biosets used in cloned bios for writing multiple replicas */
504 struct bio_set replica_set;
508 * Per-bucket arrays are protected by c->mark_lock, bucket_lock and
509 * gc_lock, for device resize - holding any is sufficient for access:
510 * Or rcu_read_lock(), but only for ptr_stale():
512 struct bucket_array __rcu *buckets_gc;
513 struct bucket_gens __rcu *bucket_gens;
515 unsigned long *buckets_nouse;
516 struct rw_semaphore bucket_lock;
518 struct bch_dev_usage *usage_base;
519 struct bch_dev_usage __percpu *usage[JOURNAL_BUF_NR];
520 struct bch_dev_usage __percpu *usage_gc;
523 u64 new_fs_bucket_idx;
526 unsigned nr_open_buckets;
527 unsigned nr_btree_reserve;
529 size_t inc_gen_needs_gc;
530 size_t inc_gen_really_needs_gc;
531 size_t buckets_waiting_on_journal;
533 atomic64_t rebalance_work;
535 struct journal_device journal;
536 u64 prev_journal_sector;
538 struct work_struct io_error_work;
540 /* The rest of this all shows up in sysfs */
541 atomic64_t cur_latency[2];
542 struct bch2_time_stats io_latency[2];
544 #define CONGESTED_MAX 1024
548 struct io_count __percpu *io_done;
562 BCH_FS_WRITE_DISABLE_COMPLETE,
563 BCH_FS_CLEAN_SHUTDOWN,
567 BCH_FS_INITIAL_GC_UNFIXED, /* kill when we enumerate fsck errors */
568 BCH_FS_NEED_ANOTHER_GC,
570 BCH_FS_HAVE_DELETED_SNAPSHOTS,
574 BCH_FS_TOPOLOGY_ERROR,
576 BCH_FS_ERRORS_NOT_FIXED,
583 #define BCH_TRANSACTIONS_NR 128
585 struct btree_transaction_stats {
586 struct bch2_time_stats lock_hold_times;
588 unsigned nr_max_paths;
589 unsigned wb_updates_size;
591 char *max_paths_text;
595 u64 sectors_available;
598 struct journal_seq_blacklist_table {
600 struct journal_seq_blacklist_table_entry {
607 struct journal_keys {
611 enum btree_id btree_id:8;
618 * Gap buffer: instead of all the empty space in the array being at the
619 * end of the buffer - from @nr to @size - the empty space is at @gap.
620 * This means that sequential insertions are O(n) instead of O(n^2).
627 struct btree_path_buf {
628 struct btree_path *path;
631 #define REPLICAS_DELTA_LIST_MAX (1U << 16)
633 #define BCACHEFS_ROOT_SUBVOL_INUM \
634 ((subvol_inum) { BCACHEFS_ROOT_SUBVOL, BCACHEFS_ROOT_INO })
636 #define BCH_WRITE_REFS() \
647 x(delete_dead_snapshots) \
648 x(snapshot_delete_pagecache) \
652 #define x(n) BCH_WRITE_REF_##n,
658 #define PASS_SILENT BIT(0)
659 #define PASS_FSCK BIT(1)
660 #define PASS_UNCLEAN BIT(2)
661 #define PASS_ALWAYS BIT(3)
663 #define BCH_RECOVERY_PASSES() \
664 x(alloc_read, PASS_ALWAYS) \
665 x(stripes_read, PASS_ALWAYS) \
666 x(initialize_subvolumes, 0) \
667 x(snapshots_read, PASS_ALWAYS) \
668 x(check_topology, 0) \
669 x(check_allocations, PASS_FSCK) \
670 x(set_may_go_rw, PASS_ALWAYS|PASS_SILENT) \
671 x(journal_replay, PASS_ALWAYS) \
672 x(check_alloc_info, PASS_FSCK) \
673 x(check_lrus, PASS_FSCK) \
674 x(check_btree_backpointers, PASS_FSCK) \
675 x(check_backpointers_to_extents,PASS_FSCK) \
676 x(check_extents_to_backpointers,PASS_FSCK) \
677 x(check_alloc_to_lru_refs, PASS_FSCK) \
678 x(fs_freespace_init, PASS_ALWAYS|PASS_SILENT) \
679 x(bucket_gens_init, 0) \
680 x(check_snapshot_trees, PASS_FSCK) \
681 x(check_snapshots, PASS_FSCK) \
682 x(check_subvols, PASS_FSCK) \
683 x(delete_dead_snapshots, PASS_FSCK|PASS_UNCLEAN) \
684 x(fs_upgrade_for_subvolumes, 0) \
685 x(check_inodes, PASS_FSCK|PASS_UNCLEAN) \
686 x(check_extents, PASS_FSCK) \
687 x(check_dirents, PASS_FSCK) \
688 x(check_xattrs, PASS_FSCK) \
689 x(check_root, PASS_FSCK) \
690 x(check_directory_structure, PASS_FSCK) \
691 x(check_nlinks, PASS_FSCK) \
692 x(fix_reflink_p, 0) \
694 enum bch_recovery_pass {
695 #define x(n, when) BCH_RECOVERY_PASS_##n,
696 BCH_RECOVERY_PASSES()
703 struct list_head list;
705 struct kobject counters_kobj;
706 struct kobject internal;
707 struct kobject opts_dir;
708 struct kobject time_stats;
712 struct device *chardev;
713 struct super_block *vfs_sb;
717 /* ro/rw, add/remove/resize devices: */
718 struct rw_semaphore state_lock;
720 /* Counts outstanding writes, for clean transition to read-only */
721 #ifdef BCH_WRITE_REF_DEBUG
722 atomic_long_t writes[BCH_WRITE_REF_NR];
724 struct percpu_ref writes;
726 struct work_struct read_only_work;
728 struct bch_dev __rcu *devs[BCH_SB_MEMBERS_MAX];
730 struct bch_replicas_cpu replicas;
731 struct bch_replicas_cpu replicas_gc;
732 struct mutex replicas_gc_lock;
733 mempool_t replicas_delta_pool;
735 struct journal_entry_res btree_root_journal_res;
736 struct journal_entry_res replicas_journal_res;
737 struct journal_entry_res clock_journal_res;
738 struct journal_entry_res dev_usage_journal_res;
740 struct bch_disk_groups_cpu __rcu *disk_groups;
742 struct bch_opts opts;
744 /* Updated by bch2_sb_update():*/
751 u16 version_upgrade_complete;
760 unsigned time_units_per_sec;
761 unsigned nsec_per_time_unit;
767 struct bch_sb_handle disk_sb;
769 unsigned short block_bits; /* ilog2(block_size) */
771 u16 btree_foreground_merge_threshold;
773 struct closure sb_write;
774 struct mutex sb_lock;
777 struct snapshot_table __rcu *snapshots;
778 size_t snapshot_table_size;
779 struct mutex snapshot_table_lock;
781 struct work_struct snapshot_delete_work;
782 struct work_struct snapshot_wait_for_pagecache_and_delete_work;
783 snapshot_id_list snapshots_unlinked;
784 struct mutex snapshots_unlinked_lock;
787 struct bio_set btree_bio;
788 struct workqueue_struct *io_complete_wq;
790 struct btree_root btree_roots_known[BTREE_ID_NR];
791 DARRAY(struct btree_root) btree_roots_extra;
792 struct mutex btree_root_lock;
794 struct btree_cache btree_cache;
797 * Cache of allocated btree nodes - if we allocate a btree node and
798 * don't use it, if we free it that space can't be reused until going
799 * _all_ the way through the allocator (which exposes us to a livelock
800 * when allocating btree reserves fail halfway through) - instead, we
801 * can stick them here:
803 struct btree_alloc btree_reserve_cache[BTREE_NODE_RESERVE * 2];
804 unsigned btree_reserve_cache_nr;
805 struct mutex btree_reserve_cache_lock;
807 mempool_t btree_interior_update_pool;
808 struct list_head btree_interior_update_list;
809 struct list_head btree_interior_updates_unwritten;
810 struct mutex btree_interior_update_lock;
811 struct closure_waitlist btree_interior_update_wait;
813 struct workqueue_struct *btree_interior_update_worker;
814 struct work_struct btree_interior_update_work;
816 struct list_head pending_node_rewrites;
817 struct mutex pending_node_rewrites_lock;
820 spinlock_t btree_write_error_lock;
821 struct btree_write_stats {
824 } btree_write_stats[BTREE_WRITE_TYPE_NR];
827 struct seqmutex btree_trans_lock;
828 struct list_head btree_trans_list;
829 mempool_t btree_paths_pool;
830 mempool_t btree_trans_mem_pool;
831 struct btree_path_buf __percpu *btree_paths_bufs;
833 struct srcu_struct btree_trans_barrier;
834 bool btree_trans_barrier_initialized;
836 struct btree_key_cache btree_key_cache;
837 unsigned btree_key_cache_btrees;
839 struct btree_write_buffer btree_write_buffer;
841 struct workqueue_struct *btree_update_wq;
842 struct workqueue_struct *btree_io_complete_wq;
843 /* copygc needs its own workqueue for index updates.. */
844 struct workqueue_struct *copygc_wq;
846 * Use a dedicated wq for write ref holder tasks. Required to avoid
847 * dependency problems with other wq tasks that can block on ref
848 * draining, such as read-only transition.
850 struct workqueue_struct *write_ref_wq;
853 struct bch_devs_mask rw_devs[BCH_DATA_NR];
855 u64 capacity; /* sectors */
858 * When capacity _decreases_ (due to a disk being removed), we
859 * increment capacity_gen - this invalidates outstanding reservations
860 * and forces them to be revalidated
863 unsigned bucket_size_max;
865 atomic64_t sectors_available;
866 struct mutex sectors_available_lock;
868 struct bch_fs_pcpu __percpu *pcpu;
870 struct percpu_rw_semaphore mark_lock;
872 seqcount_t usage_lock;
873 struct bch_fs_usage *usage_base;
874 struct bch_fs_usage __percpu *usage[JOURNAL_BUF_NR];
875 struct bch_fs_usage __percpu *usage_gc;
876 u64 __percpu *online_reserved;
878 /* single element mempool: */
879 struct mutex usage_scratch_lock;
880 struct bch_fs_usage_online *usage_scratch;
882 struct io_clock io_clock[2];
884 /* JOURNAL SEQ BLACKLIST */
885 struct journal_seq_blacklist_table *
886 journal_seq_blacklist_table;
887 struct work_struct journal_seq_blacklist_gc_work;
890 spinlock_t freelist_lock;
891 struct closure_waitlist freelist_wait;
892 u64 blocked_allocate;
893 u64 blocked_allocate_open_bucket;
895 open_bucket_idx_t open_buckets_freelist;
896 open_bucket_idx_t open_buckets_nr_free;
897 struct closure_waitlist open_buckets_wait;
898 struct open_bucket open_buckets[OPEN_BUCKETS_COUNT];
899 open_bucket_idx_t open_buckets_hash[OPEN_BUCKETS_COUNT];
901 open_bucket_idx_t open_buckets_partial[OPEN_BUCKETS_COUNT];
902 open_bucket_idx_t open_buckets_partial_nr;
904 struct write_point btree_write_point;
905 struct write_point rebalance_write_point;
907 struct write_point write_points[WRITE_POINT_MAX];
908 struct hlist_head write_points_hash[WRITE_POINT_HASH_NR];
909 struct mutex write_points_hash_lock;
910 unsigned write_points_nr;
912 struct buckets_waiting_for_journal buckets_waiting_for_journal;
913 struct work_struct discard_work;
914 struct work_struct invalidate_work;
916 /* GARBAGE COLLECTION */
917 struct task_struct *gc_thread;
919 unsigned long gc_count;
921 enum btree_id gc_gens_btree;
922 struct bpos gc_gens_pos;
925 * Tracks GC's progress - everything in the range [ZERO_KEY..gc_cur_pos]
926 * has been marked by GC.
928 * gc_cur_phase is a superset of btree_ids (BTREE_ID_extents etc.)
930 * Protected by gc_pos_lock. Only written to by GC thread, so GC thread
931 * can read without a lock.
933 seqcount_t gc_pos_lock;
934 struct gc_pos gc_pos;
937 * The allocation code needs gc_mark in struct bucket to be correct, but
938 * it's not while a gc is in progress.
940 struct rw_semaphore gc_lock;
941 struct mutex gc_gens_lock;
944 struct semaphore io_in_flight;
945 struct bio_set bio_read;
946 struct bio_set bio_read_split;
947 struct bio_set bio_write;
948 struct mutex bio_bounce_pages_lock;
949 mempool_t bio_bounce_pages;
950 struct bucket_nocow_lock_table
952 struct rhashtable promote_table;
954 mempool_t compression_bounce[2];
955 mempool_t compress_workspace[BCH_COMPRESSION_TYPE_NR];
956 mempool_t decompress_workspace;
957 ZSTD_parameters zstd_params;
959 struct crypto_shash *sha256;
960 struct crypto_sync_skcipher *chacha20;
961 struct crypto_shash *poly1305;
963 atomic64_t key_version;
965 mempool_t large_bkey_pool;
968 struct list_head moving_context_list;
969 struct mutex moving_context_lock;
971 struct list_head data_progress_list;
972 struct mutex data_progress_lock;
975 struct bch_fs_rebalance rebalance;
978 struct task_struct *copygc_thread;
979 struct write_point copygc_write_point;
983 wait_queue_head_t copygc_running_wq;
986 GENRADIX(struct stripe) stripes;
987 GENRADIX(struct gc_stripe) gc_stripes;
989 struct hlist_head ec_stripes_new[32];
990 spinlock_t ec_stripes_new_lock;
992 ec_stripes_heap ec_stripes_heap;
993 struct mutex ec_stripes_heap_lock;
996 struct list_head ec_stripe_head_list;
997 struct mutex ec_stripe_head_lock;
999 struct list_head ec_stripe_new_list;
1000 struct mutex ec_stripe_new_lock;
1001 wait_queue_head_t ec_stripe_new_wait;
1003 struct work_struct ec_stripe_create_work;
1006 struct work_struct ec_stripe_delete_work;
1008 struct bio_set ec_bioset;
1011 reflink_gc_table reflink_gc_table;
1012 size_t reflink_gc_nr;
1015 struct list_head vfs_inodes_list;
1016 struct mutex vfs_inodes_lock;
1018 /* VFS IO PATH - fs-io.c */
1019 struct bio_set writepage_bioset;
1020 struct bio_set dio_write_bioset;
1021 struct bio_set dio_read_bioset;
1022 struct bio_set nocow_flush_bioset;
1025 struct list_head fsck_errors;
1026 struct mutex fsck_error_lock;
1027 bool fsck_alloc_err;
1030 struct bch_memquota_type quotas[QTYP_NR];
1033 u64 journal_replay_seq_start;
1034 u64 journal_replay_seq_end;
1035 enum bch_recovery_pass curr_recovery_pass;
1036 /* bitmap of explicitly enabled recovery passes: */
1037 u64 recovery_passes_explicit;
1040 struct dentry *fs_debug_dir;
1041 struct dentry *btree_debug_dir;
1042 struct btree_debug btree_debug[BTREE_ID_NR];
1043 struct btree *verify_data;
1044 struct btree_node *verify_ondisk;
1045 struct mutex verify_lock;
1047 u64 *unused_inode_hints;
1048 unsigned inode_shard_bits;
1051 * A btree node on disk could have too many bsets for an iterator to fit
1052 * on the stack - have to dynamically allocate them
1054 mempool_t fill_iter;
1056 mempool_t btree_bounce_pool;
1058 struct journal journal;
1059 GENRADIX(struct journal_replay *) journal_entries;
1060 u64 journal_entries_base_seq;
1061 struct journal_keys journal_keys;
1062 struct list_head journal_iters;
1064 u64 last_bucket_seq_cleanup;
1066 u64 counters_on_mount[BCH_COUNTER_NR];
1067 u64 __percpu *counters;
1069 unsigned btree_gc_periodic:1;
1070 unsigned copy_gc_enabled:1;
1071 bool promote_whole_extents;
1073 struct bch2_time_stats times[BCH_TIME_STAT_NR];
1075 struct btree_transaction_stats btree_transaction_stats[BCH_TRANSACTIONS_NR];
1078 extern struct wait_queue_head bch2_read_only_wait;
1080 static inline void bch2_write_ref_get(struct bch_fs *c, enum bch_write_ref ref)
1082 #ifdef BCH_WRITE_REF_DEBUG
1083 atomic_long_inc(&c->writes[ref]);
1085 percpu_ref_get(&c->writes);
1089 static inline bool bch2_write_ref_tryget(struct bch_fs *c, enum bch_write_ref ref)
1091 #ifdef BCH_WRITE_REF_DEBUG
1092 return !test_bit(BCH_FS_GOING_RO, &c->flags) &&
1093 atomic_long_inc_not_zero(&c->writes[ref]);
1095 return percpu_ref_tryget_live(&c->writes);
1099 static inline void bch2_write_ref_put(struct bch_fs *c, enum bch_write_ref ref)
1101 #ifdef BCH_WRITE_REF_DEBUG
1102 long v = atomic_long_dec_return(&c->writes[ref]);
1107 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
1108 if (atomic_long_read(&c->writes[i]))
1111 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
1112 wake_up(&bch2_read_only_wait);
1114 percpu_ref_put(&c->writes);
1118 static inline void bch2_set_ra_pages(struct bch_fs *c, unsigned ra_pages)
1120 #ifndef NO_BCACHEFS_FS
1122 c->vfs_sb->s_bdi->ra_pages = ra_pages;
1126 static inline unsigned bucket_bytes(const struct bch_dev *ca)
1128 return ca->mi.bucket_size << 9;
1131 static inline unsigned block_bytes(const struct bch_fs *c)
1133 return c->opts.block_size;
1136 static inline unsigned block_sectors(const struct bch_fs *c)
1138 return c->opts.block_size >> 9;
1141 static inline size_t btree_sectors(const struct bch_fs *c)
1143 return c->opts.btree_node_size >> 9;
1146 static inline bool btree_id_cached(const struct bch_fs *c, enum btree_id btree)
1148 return c->btree_key_cache_btrees & (1U << btree);
1151 static inline struct timespec64 bch2_time_to_timespec(const struct bch_fs *c, s64 time)
1153 struct timespec64 t;
1156 time += c->sb.time_base_lo;
1158 t.tv_sec = div_s64_rem(time, c->sb.time_units_per_sec, &rem);
1159 t.tv_nsec = rem * c->sb.nsec_per_time_unit;
1163 static inline s64 timespec_to_bch2_time(const struct bch_fs *c, struct timespec64 ts)
1165 return (ts.tv_sec * c->sb.time_units_per_sec +
1166 (int) ts.tv_nsec / c->sb.nsec_per_time_unit) - c->sb.time_base_lo;
1169 static inline s64 bch2_current_time(const struct bch_fs *c)
1171 struct timespec64 now;
1173 ktime_get_coarse_real_ts64(&now);
1174 return timespec_to_bch2_time(c, now);
1177 static inline bool bch2_dev_exists2(const struct bch_fs *c, unsigned dev)
1179 return dev < c->sb.nr_devices && c->devs[dev];
1183 * For when we need to rewind recovery passes and run a pass we skipped:
1185 static inline int bch2_run_explicit_recovery_pass(struct bch_fs *c,
1186 enum bch_recovery_pass pass)
1188 c->recovery_passes_explicit |= BIT_ULL(pass);
1190 if (c->curr_recovery_pass >= pass) {
1191 c->curr_recovery_pass = pass;
1192 return -BCH_ERR_restart_recovery;
1198 #define BKEY_PADDED_ONSTACK(key, pad) \
1199 struct { struct bkey_i key; __u64 key ## _pad[pad]; }
1201 #endif /* _BCACHEFS_H */