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/refcount.h>
197 #include <linux/rhashtable.h>
198 #include <linux/rwsem.h>
199 #include <linux/semaphore.h>
200 #include <linux/seqlock.h>
201 #include <linux/shrinker.h>
202 #include <linux/srcu.h>
203 #include <linux/types.h>
204 #include <linux/workqueue.h>
205 #include <linux/zstd.h>
207 #include "bcachefs_format.h"
210 #include "nocow_locking_types.h"
212 #include "recovery_types.h"
213 #include "sb-errors_types.h"
214 #include "seqmutex.h"
215 #include "time_stats.h"
218 #ifdef CONFIG_BCACHEFS_DEBUG
219 #define BCH_WRITE_REF_DEBUG
222 #ifndef dynamic_fault
223 #define dynamic_fault(...) 0
226 #define race_fault(...) dynamic_fault("bcachefs:race")
228 #define count_event(_c, _name) this_cpu_inc((_c)->counters[BCH_COUNTER_##_name])
230 #define trace_and_count(_c, _name, ...) \
232 count_event(_c, _name); \
233 trace_##_name(__VA_ARGS__); \
236 #define bch2_fs_init_fault(name) \
237 dynamic_fault("bcachefs:bch_fs_init:" name)
238 #define bch2_meta_read_fault(name) \
239 dynamic_fault("bcachefs:meta:read:" name)
240 #define bch2_meta_write_fault(name) \
241 dynamic_fault("bcachefs:meta:write:" name)
244 #define BCACHEFS_LOG_PREFIX
247 #ifdef BCACHEFS_LOG_PREFIX
249 #define bch2_log_msg(_c, fmt) "bcachefs (%s): " fmt, ((_c)->name)
250 #define bch2_fmt_dev(_ca, fmt) "bcachefs (%s): " fmt "\n", ((_ca)->name)
251 #define bch2_fmt_dev_offset(_ca, _offset, fmt) "bcachefs (%s sector %llu): " fmt "\n", ((_ca)->name), (_offset)
252 #define bch2_fmt_inum(_c, _inum, fmt) "bcachefs (%s inum %llu): " fmt "\n", ((_c)->name), (_inum)
253 #define bch2_fmt_inum_offset(_c, _inum, _offset, fmt) \
254 "bcachefs (%s inum %llu offset %llu): " fmt "\n", ((_c)->name), (_inum), (_offset)
258 #define bch2_log_msg(_c, fmt) fmt
259 #define bch2_fmt_dev(_ca, fmt) "%s: " fmt "\n", ((_ca)->name)
260 #define bch2_fmt_dev_offset(_ca, _offset, fmt) "%s sector %llu: " fmt "\n", ((_ca)->name), (_offset)
261 #define bch2_fmt_inum(_c, _inum, fmt) "inum %llu: " fmt "\n", (_inum)
262 #define bch2_fmt_inum_offset(_c, _inum, _offset, fmt) \
263 "inum %llu offset %llu: " fmt "\n", (_inum), (_offset)
267 #define bch2_fmt(_c, fmt) bch2_log_msg(_c, fmt "\n")
270 void bch2_print_opts(struct bch_opts *, const char *, ...);
273 void __bch2_print(struct bch_fs *c, const char *fmt, ...);
275 #define maybe_dev_to_fs(_c) _Generic((_c), \
276 struct bch_dev *: ((struct bch_dev *) (_c))->fs, \
277 struct bch_fs *: (_c))
279 #define bch2_print(_c, ...) __bch2_print(maybe_dev_to_fs(_c), __VA_ARGS__)
281 #define bch2_print_ratelimited(_c, ...) \
283 static DEFINE_RATELIMIT_STATE(_rs, \
284 DEFAULT_RATELIMIT_INTERVAL, \
285 DEFAULT_RATELIMIT_BURST); \
287 if (__ratelimit(&_rs)) \
288 bch2_print(_c, __VA_ARGS__); \
291 #define bch_info(c, fmt, ...) \
292 bch2_print(c, KERN_INFO bch2_fmt(c, fmt), ##__VA_ARGS__)
293 #define bch_notice(c, fmt, ...) \
294 bch2_print(c, KERN_NOTICE bch2_fmt(c, fmt), ##__VA_ARGS__)
295 #define bch_warn(c, fmt, ...) \
296 bch2_print(c, KERN_WARNING bch2_fmt(c, fmt), ##__VA_ARGS__)
297 #define bch_warn_ratelimited(c, fmt, ...) \
298 bch2_print_ratelimited(c, KERN_WARNING bch2_fmt(c, fmt), ##__VA_ARGS__)
300 #define bch_err(c, fmt, ...) \
301 bch2_print(c, KERN_ERR bch2_fmt(c, fmt), ##__VA_ARGS__)
302 #define bch_err_dev(ca, fmt, ...) \
303 bch2_print(c, KERN_ERR bch2_fmt_dev(ca, fmt), ##__VA_ARGS__)
304 #define bch_err_dev_offset(ca, _offset, fmt, ...) \
305 bch2_print(c, KERN_ERR bch2_fmt_dev_offset(ca, _offset, fmt), ##__VA_ARGS__)
306 #define bch_err_inum(c, _inum, fmt, ...) \
307 bch2_print(c, KERN_ERR bch2_fmt_inum(c, _inum, fmt), ##__VA_ARGS__)
308 #define bch_err_inum_offset(c, _inum, _offset, fmt, ...) \
309 bch2_print(c, KERN_ERR bch2_fmt_inum_offset(c, _inum, _offset, fmt), ##__VA_ARGS__)
311 #define bch_err_ratelimited(c, fmt, ...) \
312 bch2_print_ratelimited(c, KERN_ERR bch2_fmt(c, fmt), ##__VA_ARGS__)
313 #define bch_err_dev_ratelimited(ca, fmt, ...) \
314 bch2_print_ratelimited(ca, KERN_ERR bch2_fmt_dev(ca, fmt), ##__VA_ARGS__)
315 #define bch_err_dev_offset_ratelimited(ca, _offset, fmt, ...) \
316 bch2_print_ratelimited(ca, KERN_ERR bch2_fmt_dev_offset(ca, _offset, fmt), ##__VA_ARGS__)
317 #define bch_err_inum_ratelimited(c, _inum, fmt, ...) \
318 bch2_print_ratelimited(c, KERN_ERR bch2_fmt_inum(c, _inum, fmt), ##__VA_ARGS__)
319 #define bch_err_inum_offset_ratelimited(c, _inum, _offset, fmt, ...) \
320 bch2_print_ratelimited(c, KERN_ERR bch2_fmt_inum_offset(c, _inum, _offset, fmt), ##__VA_ARGS__)
322 static inline bool should_print_err(int err)
324 return err && !bch2_err_matches(err, BCH_ERR_transaction_restart);
327 #define bch_err_fn(_c, _ret) \
329 if (should_print_err(_ret)) \
330 bch_err(_c, "%s(): error %s", __func__, bch2_err_str(_ret));\
333 #define bch_err_fn_ratelimited(_c, _ret) \
335 if (should_print_err(_ret)) \
336 bch_err_ratelimited(_c, "%s(): error %s", __func__, bch2_err_str(_ret));\
339 #define bch_err_msg(_c, _ret, _msg, ...) \
341 if (should_print_err(_ret)) \
342 bch_err(_c, "%s(): error " _msg " %s", __func__, \
343 ##__VA_ARGS__, bch2_err_str(_ret)); \
346 #define bch_verbose(c, fmt, ...) \
348 if ((c)->opts.verbose) \
349 bch_info(c, fmt, ##__VA_ARGS__); \
352 #define pr_verbose_init(opts, fmt, ...) \
354 if (opt_get(opts, verbose)) \
355 pr_info(fmt, ##__VA_ARGS__); \
358 /* Parameters that are useful for debugging, but should always be compiled in: */
359 #define BCH_DEBUG_PARAMS_ALWAYS() \
360 BCH_DEBUG_PARAM(key_merging_disabled, \
361 "Disables merging of extents") \
362 BCH_DEBUG_PARAM(btree_gc_always_rewrite, \
363 "Causes mark and sweep to compact and rewrite every " \
364 "btree node it traverses") \
365 BCH_DEBUG_PARAM(btree_gc_rewrite_disabled, \
366 "Disables rewriting of btree nodes during mark and sweep")\
367 BCH_DEBUG_PARAM(btree_shrinker_disabled, \
368 "Disables the shrinker callback for the btree node cache")\
369 BCH_DEBUG_PARAM(verify_btree_ondisk, \
370 "Reread btree nodes at various points to verify the " \
371 "mergesort in the read path against modifications " \
373 BCH_DEBUG_PARAM(verify_all_btree_replicas, \
374 "When reading btree nodes, read all replicas and " \
376 BCH_DEBUG_PARAM(backpointers_no_use_write_buffer, \
377 "Don't use the write buffer for backpointers, enabling "\
378 "extra runtime checks")
380 /* Parameters that should only be compiled in debug mode: */
381 #define BCH_DEBUG_PARAMS_DEBUG() \
382 BCH_DEBUG_PARAM(expensive_debug_checks, \
383 "Enables various runtime debugging checks that " \
384 "significantly affect performance") \
385 BCH_DEBUG_PARAM(debug_check_iterators, \
386 "Enables extra verification for btree iterators") \
387 BCH_DEBUG_PARAM(debug_check_btree_accounting, \
388 "Verify btree accounting for keys within a node") \
389 BCH_DEBUG_PARAM(journal_seq_verify, \
390 "Store the journal sequence number in the version " \
391 "number of every btree key, and verify that btree " \
392 "update ordering is preserved during recovery") \
393 BCH_DEBUG_PARAM(inject_invalid_keys, \
394 "Store the journal sequence number in the version " \
395 "number of every btree key, and verify that btree " \
396 "update ordering is preserved during recovery") \
397 BCH_DEBUG_PARAM(test_alloc_startup, \
398 "Force allocator startup to use the slowpath where it" \
399 "can't find enough free buckets without invalidating" \
401 BCH_DEBUG_PARAM(force_reconstruct_read, \
402 "Force reads to use the reconstruct path, when reading" \
403 "from erasure coded extents") \
404 BCH_DEBUG_PARAM(test_restart_gc, \
405 "Test restarting mark and sweep gc when bucket gens change")
407 #define BCH_DEBUG_PARAMS_ALL() BCH_DEBUG_PARAMS_ALWAYS() BCH_DEBUG_PARAMS_DEBUG()
409 #ifdef CONFIG_BCACHEFS_DEBUG
410 #define BCH_DEBUG_PARAMS() BCH_DEBUG_PARAMS_ALL()
412 #define BCH_DEBUG_PARAMS() BCH_DEBUG_PARAMS_ALWAYS()
415 #define BCH_DEBUG_PARAM(name, description) extern bool bch2_##name;
417 #undef BCH_DEBUG_PARAM
419 #ifndef CONFIG_BCACHEFS_DEBUG
420 #define BCH_DEBUG_PARAM(name, description) static const __maybe_unused bool bch2_##name;
421 BCH_DEBUG_PARAMS_DEBUG()
422 #undef BCH_DEBUG_PARAM
425 #define BCH_TIME_STATS() \
426 x(btree_node_mem_alloc) \
427 x(btree_node_split) \
428 x(btree_node_compact) \
429 x(btree_node_merge) \
432 x(btree_node_read_done) \
433 x(btree_interior_update_foreground) \
434 x(btree_interior_update_total) \
439 x(journal_flush_write) \
440 x(journal_noflush_write) \
441 x(journal_flush_seq) \
442 x(blocked_journal_low_on_space) \
443 x(blocked_journal_low_on_pin) \
444 x(blocked_journal_max_in_flight) \
445 x(blocked_allocate) \
446 x(blocked_allocate_open_bucket) \
447 x(blocked_write_buffer_full) \
448 x(nocow_lock_contended)
450 enum bch_time_stats {
451 #define x(name) BCH_TIME_##name,
457 #include "alloc_types.h"
458 #include "btree_types.h"
459 #include "btree_write_buffer_types.h"
460 #include "buckets_types.h"
461 #include "buckets_waiting_for_journal_types.h"
462 #include "clock_types.h"
463 #include "disk_groups_types.h"
464 #include "ec_types.h"
465 #include "journal_types.h"
466 #include "keylist_types.h"
467 #include "quota_types.h"
468 #include "rebalance_types.h"
469 #include "replicas_types.h"
470 #include "subvolume_types.h"
471 #include "super_types.h"
472 #include "thread_with_file_types.h"
474 /* Number of nodes btree coalesce will try to coalesce at once */
475 #define GC_MERGE_NODES 4U
477 /* Maximum number of nodes we might need to allocate atomically: */
478 #define BTREE_RESERVE_MAX (BTREE_MAX_DEPTH + (BTREE_MAX_DEPTH - 1))
480 /* Size of the freelist we allocate btree nodes from: */
481 #define BTREE_NODE_RESERVE (BTREE_RESERVE_MAX * 4)
483 #define BTREE_NODE_OPEN_BUCKET_RESERVE (BTREE_RESERVE_MAX * BCH_REPLICAS_MAX)
488 GC_PHASE_NOT_RUNNING,
492 GC_PHASE_BTREE_stripes,
493 GC_PHASE_BTREE_extents,
494 GC_PHASE_BTREE_inodes,
495 GC_PHASE_BTREE_dirents,
496 GC_PHASE_BTREE_xattrs,
497 GC_PHASE_BTREE_alloc,
498 GC_PHASE_BTREE_quotas,
499 GC_PHASE_BTREE_reflink,
500 GC_PHASE_BTREE_subvolumes,
501 GC_PHASE_BTREE_snapshots,
503 GC_PHASE_BTREE_freespace,
504 GC_PHASE_BTREE_need_discard,
505 GC_PHASE_BTREE_backpointers,
506 GC_PHASE_BTREE_bucket_gens,
507 GC_PHASE_BTREE_snapshot_trees,
508 GC_PHASE_BTREE_deleted_inodes,
509 GC_PHASE_BTREE_logged_ops,
510 GC_PHASE_BTREE_rebalance_work,
511 GC_PHASE_BTREE_subvolume_children,
513 GC_PHASE_PENDING_DELETE,
528 typedef GENRADIX(struct reflink_gc) reflink_gc_table;
531 u64 sectors[2][BCH_DATA_NR];
536 struct percpu_ref ref;
537 struct completion ref_completion;
538 struct percpu_ref io_ref;
539 struct completion io_ref_completion;
545 * Cached version of this device's member info from superblock
546 * Committed by bch2_write_super() -> bch_fs_mi_update()
548 struct bch_member_cpu mi;
549 atomic64_t errors[BCH_MEMBER_ERROR_NR];
552 char name[BDEVNAME_SIZE];
554 struct bch_sb_handle disk_sb;
555 struct bch_sb *sb_read_scratch;
560 struct bch_devs_mask self;
562 /* biosets used in cloned bios for writing multiple replicas */
563 struct bio_set replica_set;
567 * Per-bucket arrays are protected by c->mark_lock, bucket_lock and
568 * gc_lock, for device resize - holding any is sufficient for access:
569 * Or rcu_read_lock(), but only for ptr_stale():
571 struct bucket_array __rcu *buckets_gc;
572 struct bucket_gens __rcu *bucket_gens;
574 unsigned long *buckets_nouse;
575 struct rw_semaphore bucket_lock;
577 struct bch_dev_usage *usage_base;
578 struct bch_dev_usage __percpu *usage[JOURNAL_BUF_NR];
579 struct bch_dev_usage __percpu *usage_gc;
582 u64 new_fs_bucket_idx;
585 unsigned nr_open_buckets;
586 unsigned nr_btree_reserve;
588 size_t inc_gen_needs_gc;
589 size_t inc_gen_really_needs_gc;
590 size_t buckets_waiting_on_journal;
592 atomic64_t rebalance_work;
594 struct journal_device journal;
595 u64 prev_journal_sector;
597 struct work_struct io_error_work;
599 /* The rest of this all shows up in sysfs */
600 atomic64_t cur_latency[2];
601 struct bch2_time_stats_quantiles io_latency[2];
603 #define CONGESTED_MAX 1024
607 struct io_count __percpu *io_done;
616 #define BCH_FS_FLAGS() \
624 x(write_disable_complete) \
627 x(initial_gc_unfixed) \
629 x(need_delete_dead_snapshots) \
636 #define x(n) BCH_FS_##n,
645 #define BCH_TRANSACTIONS_NR 128
647 struct btree_transaction_stats {
648 struct bch2_time_stats duration;
649 struct bch2_time_stats lock_hold_times;
651 unsigned nr_max_paths;
652 unsigned journal_entries_size;
654 char *max_paths_text;
658 u64 sectors_available;
661 struct journal_seq_blacklist_table {
663 struct journal_seq_blacklist_table_entry {
670 struct journal_keys {
671 /* must match layout in darray_types.h */
676 enum btree_id btree_id:8;
683 * Gap buffer: instead of all the empty space in the array being at the
684 * end of the buffer - from @nr to @size - the empty space is at @gap.
685 * This means that sequential insertions are O(n) instead of O(n^2).
689 bool initial_ref_held;
692 struct btree_trans_buf {
693 struct btree_trans *trans;
696 #define REPLICAS_DELTA_LIST_MAX (1U << 16)
698 #define BCACHEFS_ROOT_SUBVOL_INUM \
699 ((subvol_inum) { BCACHEFS_ROOT_SUBVOL, BCACHEFS_ROOT_INO })
701 #define BCH_WRITE_REFS() \
713 x(delete_dead_snapshots) \
714 x(snapshot_delete_pagecache) \
716 x(btree_write_buffer)
719 #define x(n) BCH_WRITE_REF_##n,
728 struct list_head list;
730 struct kobject counters_kobj;
731 struct kobject internal;
732 struct kobject opts_dir;
733 struct kobject time_stats;
737 struct device *chardev;
738 struct super_block *vfs_sb;
741 struct stdio_redirect *stdio;
742 struct task_struct *stdio_filter;
744 /* ro/rw, add/remove/resize devices: */
745 struct rw_semaphore state_lock;
747 /* Counts outstanding writes, for clean transition to read-only */
748 #ifdef BCH_WRITE_REF_DEBUG
749 atomic_long_t writes[BCH_WRITE_REF_NR];
751 struct percpu_ref writes;
754 * Analagous to c->writes, for asynchronous ops that don't necessarily
755 * need fs to be read-write
758 wait_queue_head_t ro_ref_wait;
760 struct work_struct read_only_work;
762 struct bch_dev __rcu *devs[BCH_SB_MEMBERS_MAX];
764 struct bch_replicas_cpu replicas;
765 struct bch_replicas_cpu replicas_gc;
766 struct mutex replicas_gc_lock;
767 mempool_t replicas_delta_pool;
769 struct journal_entry_res btree_root_journal_res;
770 struct journal_entry_res replicas_journal_res;
771 struct journal_entry_res clock_journal_res;
772 struct journal_entry_res dev_usage_journal_res;
774 struct bch_disk_groups_cpu __rcu *disk_groups;
776 struct bch_opts opts;
778 /* Updated by bch2_sb_update():*/
785 u16 version_upgrade_complete;
794 unsigned time_units_per_sec;
795 unsigned nsec_per_time_unit;
798 unsigned long errors_silent[BITS_TO_LONGS(BCH_SB_ERR_MAX)];
802 struct bch_sb_handle disk_sb;
804 unsigned short block_bits; /* ilog2(block_size) */
806 u16 btree_foreground_merge_threshold;
808 struct closure sb_write;
809 struct mutex sb_lock;
812 struct snapshot_table __rcu *snapshots;
813 size_t snapshot_table_size;
814 struct mutex snapshot_table_lock;
815 struct rw_semaphore snapshot_create_lock;
817 struct work_struct snapshot_delete_work;
818 struct work_struct snapshot_wait_for_pagecache_and_delete_work;
819 snapshot_id_list snapshots_unlinked;
820 struct mutex snapshots_unlinked_lock;
823 struct bio_set btree_bio;
824 struct workqueue_struct *io_complete_wq;
826 struct btree_root btree_roots_known[BTREE_ID_NR];
827 DARRAY(struct btree_root) btree_roots_extra;
828 struct mutex btree_root_lock;
830 struct btree_cache btree_cache;
833 * Cache of allocated btree nodes - if we allocate a btree node and
834 * don't use it, if we free it that space can't be reused until going
835 * _all_ the way through the allocator (which exposes us to a livelock
836 * when allocating btree reserves fail halfway through) - instead, we
837 * can stick them here:
839 struct btree_alloc btree_reserve_cache[BTREE_NODE_RESERVE * 2];
840 unsigned btree_reserve_cache_nr;
841 struct mutex btree_reserve_cache_lock;
843 mempool_t btree_interior_update_pool;
844 struct list_head btree_interior_update_list;
845 struct list_head btree_interior_updates_unwritten;
846 struct mutex btree_interior_update_lock;
847 struct closure_waitlist btree_interior_update_wait;
849 struct workqueue_struct *btree_interior_update_worker;
850 struct work_struct btree_interior_update_work;
852 struct workqueue_struct *btree_node_rewrite_worker;
854 struct list_head pending_node_rewrites;
855 struct mutex pending_node_rewrites_lock;
858 spinlock_t btree_write_error_lock;
859 struct btree_write_stats {
862 } btree_write_stats[BTREE_WRITE_TYPE_NR];
865 struct seqmutex btree_trans_lock;
866 struct list_head btree_trans_list;
867 mempool_t btree_trans_pool;
868 mempool_t btree_trans_mem_pool;
869 struct btree_trans_buf __percpu *btree_trans_bufs;
871 struct srcu_struct btree_trans_barrier;
872 bool btree_trans_barrier_initialized;
874 struct btree_key_cache btree_key_cache;
875 unsigned btree_key_cache_btrees;
877 struct btree_write_buffer btree_write_buffer;
879 struct workqueue_struct *btree_update_wq;
880 struct workqueue_struct *btree_io_complete_wq;
881 /* copygc needs its own workqueue for index updates.. */
882 struct workqueue_struct *copygc_wq;
884 * Use a dedicated wq for write ref holder tasks. Required to avoid
885 * dependency problems with other wq tasks that can block on ref
886 * draining, such as read-only transition.
888 struct workqueue_struct *write_ref_wq;
891 struct bch_devs_mask rw_devs[BCH_DATA_NR];
893 u64 capacity; /* sectors */
896 * When capacity _decreases_ (due to a disk being removed), we
897 * increment capacity_gen - this invalidates outstanding reservations
898 * and forces them to be revalidated
901 unsigned bucket_size_max;
903 atomic64_t sectors_available;
904 struct mutex sectors_available_lock;
906 struct bch_fs_pcpu __percpu *pcpu;
908 struct percpu_rw_semaphore mark_lock;
910 seqcount_t usage_lock;
911 struct bch_fs_usage *usage_base;
912 struct bch_fs_usage __percpu *usage[JOURNAL_BUF_NR];
913 struct bch_fs_usage __percpu *usage_gc;
914 u64 __percpu *online_reserved;
916 /* single element mempool: */
917 struct mutex usage_scratch_lock;
918 struct bch_fs_usage_online *usage_scratch;
920 struct io_clock io_clock[2];
922 /* JOURNAL SEQ BLACKLIST */
923 struct journal_seq_blacklist_table *
924 journal_seq_blacklist_table;
925 struct work_struct journal_seq_blacklist_gc_work;
928 spinlock_t freelist_lock;
929 struct closure_waitlist freelist_wait;
931 open_bucket_idx_t open_buckets_freelist;
932 open_bucket_idx_t open_buckets_nr_free;
933 struct closure_waitlist open_buckets_wait;
934 struct open_bucket open_buckets[OPEN_BUCKETS_COUNT];
935 open_bucket_idx_t open_buckets_hash[OPEN_BUCKETS_COUNT];
937 open_bucket_idx_t open_buckets_partial[OPEN_BUCKETS_COUNT];
938 open_bucket_idx_t open_buckets_partial_nr;
940 struct write_point btree_write_point;
941 struct write_point rebalance_write_point;
943 struct write_point write_points[WRITE_POINT_MAX];
944 struct hlist_head write_points_hash[WRITE_POINT_HASH_NR];
945 struct mutex write_points_hash_lock;
946 unsigned write_points_nr;
948 struct buckets_waiting_for_journal buckets_waiting_for_journal;
949 struct work_struct invalidate_work;
950 struct work_struct discard_work;
951 struct mutex discard_buckets_in_flight_lock;
952 DARRAY(struct bpos) discard_buckets_in_flight;
953 struct work_struct discard_fast_work;
955 /* GARBAGE COLLECTION */
956 struct task_struct *gc_thread;
958 unsigned long gc_count;
960 enum btree_id gc_gens_btree;
961 struct bpos gc_gens_pos;
964 * Tracks GC's progress - everything in the range [ZERO_KEY..gc_cur_pos]
965 * has been marked by GC.
967 * gc_cur_phase is a superset of btree_ids (BTREE_ID_extents etc.)
969 * Protected by gc_pos_lock. Only written to by GC thread, so GC thread
970 * can read without a lock.
972 seqcount_t gc_pos_lock;
973 struct gc_pos gc_pos;
976 * The allocation code needs gc_mark in struct bucket to be correct, but
977 * it's not while a gc is in progress.
979 struct rw_semaphore gc_lock;
980 struct mutex gc_gens_lock;
983 struct semaphore io_in_flight;
984 struct bio_set bio_read;
985 struct bio_set bio_read_split;
986 struct bio_set bio_write;
987 struct mutex bio_bounce_pages_lock;
988 mempool_t bio_bounce_pages;
989 struct bucket_nocow_lock_table
991 struct rhashtable promote_table;
993 mempool_t compression_bounce[2];
994 mempool_t compress_workspace[BCH_COMPRESSION_TYPE_NR];
995 mempool_t decompress_workspace;
996 size_t zstd_workspace_size;
998 struct crypto_shash *sha256;
999 struct crypto_sync_skcipher *chacha20;
1000 struct crypto_shash *poly1305;
1002 atomic64_t key_version;
1004 mempool_t large_bkey_pool;
1007 struct list_head moving_context_list;
1008 struct mutex moving_context_lock;
1011 struct bch_fs_rebalance rebalance;
1014 struct task_struct *copygc_thread;
1015 struct write_point copygc_write_point;
1018 bool copygc_running;
1019 wait_queue_head_t copygc_running_wq;
1022 GENRADIX(struct stripe) stripes;
1023 GENRADIX(struct gc_stripe) gc_stripes;
1025 struct hlist_head ec_stripes_new[32];
1026 spinlock_t ec_stripes_new_lock;
1028 ec_stripes_heap ec_stripes_heap;
1029 struct mutex ec_stripes_heap_lock;
1031 /* ERASURE CODING */
1032 struct list_head ec_stripe_head_list;
1033 struct mutex ec_stripe_head_lock;
1035 struct list_head ec_stripe_new_list;
1036 struct mutex ec_stripe_new_lock;
1037 wait_queue_head_t ec_stripe_new_wait;
1039 struct work_struct ec_stripe_create_work;
1042 struct work_struct ec_stripe_delete_work;
1044 struct bio_set ec_bioset;
1047 reflink_gc_table reflink_gc_table;
1048 size_t reflink_gc_nr;
1051 struct list_head vfs_inodes_list;
1052 struct mutex vfs_inodes_lock;
1054 /* VFS IO PATH - fs-io.c */
1055 struct bio_set writepage_bioset;
1056 struct bio_set dio_write_bioset;
1057 struct bio_set dio_read_bioset;
1058 struct bio_set nocow_flush_bioset;
1061 struct bch_memquota_type quotas[QTYP_NR];
1064 u64 journal_replay_seq_start;
1065 u64 journal_replay_seq_end;
1067 * Two different uses:
1068 * "Has this fsck pass?" - i.e. should this type of error be an
1069 * emergency read-only
1070 * And, in certain situations fsck will rewind to an earlier pass: used
1071 * for signaling to the toplevel code which pass we want to run now.
1073 enum bch_recovery_pass curr_recovery_pass;
1074 /* bitmap of explicitly enabled recovery passes: */
1075 u64 recovery_passes_explicit;
1076 /* bitmask of recovery passes that we actually ran */
1077 u64 recovery_passes_complete;
1078 /* never rewinds version of curr_recovery_pass */
1079 enum bch_recovery_pass recovery_pass_done;
1080 struct semaphore online_fsck_mutex;
1083 struct dentry *fs_debug_dir;
1084 struct dentry *btree_debug_dir;
1085 struct btree_debug btree_debug[BTREE_ID_NR];
1086 struct btree *verify_data;
1087 struct btree_node *verify_ondisk;
1088 struct mutex verify_lock;
1090 u64 *unused_inode_hints;
1091 unsigned inode_shard_bits;
1094 * A btree node on disk could have too many bsets for an iterator to fit
1095 * on the stack - have to dynamically allocate them
1097 mempool_t fill_iter;
1099 mempool_t btree_bounce_pool;
1101 struct journal journal;
1102 GENRADIX(struct journal_replay *) journal_entries;
1103 u64 journal_entries_base_seq;
1104 struct journal_keys journal_keys;
1105 struct list_head journal_iters;
1107 u64 last_bucket_seq_cleanup;
1109 u64 counters_on_mount[BCH_COUNTER_NR];
1110 u64 __percpu *counters;
1112 unsigned btree_gc_periodic:1;
1113 unsigned copy_gc_enabled:1;
1114 bool promote_whole_extents;
1116 struct bch2_time_stats times[BCH_TIME_STAT_NR];
1118 struct btree_transaction_stats btree_transaction_stats[BCH_TRANSACTIONS_NR];
1121 struct list_head fsck_error_msgs;
1122 struct mutex fsck_error_msgs_lock;
1123 bool fsck_alloc_msgs_err;
1125 bch_sb_errors_cpu fsck_error_counts;
1126 struct mutex fsck_error_counts_lock;
1129 extern struct wait_queue_head bch2_read_only_wait;
1131 static inline void bch2_write_ref_get(struct bch_fs *c, enum bch_write_ref ref)
1133 #ifdef BCH_WRITE_REF_DEBUG
1134 atomic_long_inc(&c->writes[ref]);
1136 percpu_ref_get(&c->writes);
1140 static inline bool __bch2_write_ref_tryget(struct bch_fs *c, enum bch_write_ref ref)
1142 #ifdef BCH_WRITE_REF_DEBUG
1143 return !test_bit(BCH_FS_going_ro, &c->flags) &&
1144 atomic_long_inc_not_zero(&c->writes[ref]);
1146 return percpu_ref_tryget(&c->writes);
1150 static inline bool bch2_write_ref_tryget(struct bch_fs *c, enum bch_write_ref ref)
1152 #ifdef BCH_WRITE_REF_DEBUG
1153 return !test_bit(BCH_FS_going_ro, &c->flags) &&
1154 atomic_long_inc_not_zero(&c->writes[ref]);
1156 return percpu_ref_tryget_live(&c->writes);
1160 static inline void bch2_write_ref_put(struct bch_fs *c, enum bch_write_ref ref)
1162 #ifdef BCH_WRITE_REF_DEBUG
1163 long v = atomic_long_dec_return(&c->writes[ref]);
1168 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
1169 if (atomic_long_read(&c->writes[i]))
1172 set_bit(BCH_FS_write_disable_complete, &c->flags);
1173 wake_up(&bch2_read_only_wait);
1175 percpu_ref_put(&c->writes);
1179 static inline bool bch2_ro_ref_tryget(struct bch_fs *c)
1181 if (test_bit(BCH_FS_stopping, &c->flags))
1184 return refcount_inc_not_zero(&c->ro_ref);
1187 static inline void bch2_ro_ref_put(struct bch_fs *c)
1189 if (refcount_dec_and_test(&c->ro_ref))
1190 wake_up(&c->ro_ref_wait);
1193 static inline void bch2_set_ra_pages(struct bch_fs *c, unsigned ra_pages)
1195 #ifndef NO_BCACHEFS_FS
1197 c->vfs_sb->s_bdi->ra_pages = ra_pages;
1201 static inline unsigned bucket_bytes(const struct bch_dev *ca)
1203 return ca->mi.bucket_size << 9;
1206 static inline unsigned block_bytes(const struct bch_fs *c)
1208 return c->opts.block_size;
1211 static inline unsigned block_sectors(const struct bch_fs *c)
1213 return c->opts.block_size >> 9;
1216 static inline bool btree_id_cached(const struct bch_fs *c, enum btree_id btree)
1218 return c->btree_key_cache_btrees & (1U << btree);
1221 static inline struct timespec64 bch2_time_to_timespec(const struct bch_fs *c, s64 time)
1223 struct timespec64 t;
1226 time += c->sb.time_base_lo;
1228 t.tv_sec = div_s64_rem(time, c->sb.time_units_per_sec, &rem);
1229 t.tv_nsec = rem * c->sb.nsec_per_time_unit;
1233 static inline s64 timespec_to_bch2_time(const struct bch_fs *c, struct timespec64 ts)
1235 return (ts.tv_sec * c->sb.time_units_per_sec +
1236 (int) ts.tv_nsec / c->sb.nsec_per_time_unit) - c->sb.time_base_lo;
1239 static inline s64 bch2_current_time(const struct bch_fs *c)
1241 struct timespec64 now;
1243 ktime_get_coarse_real_ts64(&now);
1244 return timespec_to_bch2_time(c, now);
1247 static inline bool bch2_dev_exists2(const struct bch_fs *c, unsigned dev)
1249 return dev < c->sb.nr_devices && c->devs[dev];
1252 static inline struct stdio_redirect *bch2_fs_stdio_redirect(struct bch_fs *c)
1254 struct stdio_redirect *stdio = c->stdio;
1256 if (c->stdio_filter && c->stdio_filter != current)
1261 static inline unsigned metadata_replicas_required(struct bch_fs *c)
1263 return min(c->opts.metadata_replicas,
1264 c->opts.metadata_replicas_required);
1267 static inline unsigned data_replicas_required(struct bch_fs *c)
1269 return min(c->opts.data_replicas,
1270 c->opts.data_replicas_required);
1273 #define BKEY_PADDED_ONSTACK(key, pad) \
1274 struct { struct bkey_i key; __u64 key ## _pad[pad]; }
1276 #endif /* _BCACHEFS_H */