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/thread_with_file_types.h>
204 #include <linux/time_stats.h>
205 #include <linux/types.h>
206 #include <linux/workqueue.h>
207 #include <linux/zstd.h>
209 #include "bcachefs_format.h"
212 #include "nocow_locking_types.h"
214 #include "recovery_types.h"
215 #include "sb-errors_types.h"
216 #include "seqmutex.h"
219 #ifdef CONFIG_BCACHEFS_DEBUG
220 #define BCH_WRITE_REF_DEBUG
223 #ifndef dynamic_fault
224 #define dynamic_fault(...) 0
227 #define race_fault(...) dynamic_fault("bcachefs:race")
229 #define count_event(_c, _name) this_cpu_inc((_c)->counters[BCH_COUNTER_##_name])
231 #define trace_and_count(_c, _name, ...) \
233 count_event(_c, _name); \
234 trace_##_name(__VA_ARGS__); \
237 #define bch2_fs_init_fault(name) \
238 dynamic_fault("bcachefs:bch_fs_init:" name)
239 #define bch2_meta_read_fault(name) \
240 dynamic_fault("bcachefs:meta:read:" name)
241 #define bch2_meta_write_fault(name) \
242 dynamic_fault("bcachefs:meta:write:" name)
245 #define BCACHEFS_LOG_PREFIX
248 #ifdef BCACHEFS_LOG_PREFIX
250 #define bch2_log_msg(_c, fmt) "bcachefs (%s): " fmt, ((_c)->name)
251 #define bch2_fmt_dev(_ca, fmt) "bcachefs (%s): " fmt "\n", ((_ca)->name)
252 #define bch2_fmt_dev_offset(_ca, _offset, fmt) "bcachefs (%s sector %llu): " fmt "\n", ((_ca)->name), (_offset)
253 #define bch2_fmt_inum(_c, _inum, fmt) "bcachefs (%s inum %llu): " fmt "\n", ((_c)->name), (_inum)
254 #define bch2_fmt_inum_offset(_c, _inum, _offset, fmt) \
255 "bcachefs (%s inum %llu offset %llu): " fmt "\n", ((_c)->name), (_inum), (_offset)
259 #define bch2_log_msg(_c, fmt) fmt
260 #define bch2_fmt_dev(_ca, fmt) "%s: " fmt "\n", ((_ca)->name)
261 #define bch2_fmt_dev_offset(_ca, _offset, fmt) "%s sector %llu: " fmt "\n", ((_ca)->name), (_offset)
262 #define bch2_fmt_inum(_c, _inum, fmt) "inum %llu: " fmt "\n", (_inum)
263 #define bch2_fmt_inum_offset(_c, _inum, _offset, fmt) \
264 "inum %llu offset %llu: " fmt "\n", (_inum), (_offset)
268 #define bch2_fmt(_c, fmt) bch2_log_msg(_c, fmt "\n")
271 void __bch2_print(struct bch_fs *c, const char *fmt, ...);
273 #define maybe_dev_to_fs(_c) _Generic((_c), \
274 struct bch_dev *: ((struct bch_dev *) (_c))->fs, \
275 struct bch_fs *: (_c))
277 #define bch2_print(_c, ...) __bch2_print(maybe_dev_to_fs(_c), __VA_ARGS__)
279 #define bch2_print_ratelimited(_c, ...) \
281 static DEFINE_RATELIMIT_STATE(_rs, \
282 DEFAULT_RATELIMIT_INTERVAL, \
283 DEFAULT_RATELIMIT_BURST); \
285 if (__ratelimit(&_rs)) \
286 bch2_print(_c, __VA_ARGS__); \
289 #define bch_info(c, fmt, ...) \
290 bch2_print(c, KERN_INFO bch2_fmt(c, fmt), ##__VA_ARGS__)
291 #define bch_notice(c, fmt, ...) \
292 bch2_print(c, KERN_NOTICE bch2_fmt(c, fmt), ##__VA_ARGS__)
293 #define bch_warn(c, fmt, ...) \
294 bch2_print(c, KERN_WARNING bch2_fmt(c, fmt), ##__VA_ARGS__)
295 #define bch_warn_ratelimited(c, fmt, ...) \
296 bch2_print_ratelimited(c, KERN_WARNING bch2_fmt(c, fmt), ##__VA_ARGS__)
298 #define bch_err(c, fmt, ...) \
299 bch2_print(c, KERN_ERR bch2_fmt(c, fmt), ##__VA_ARGS__)
300 #define bch_err_dev(ca, fmt, ...) \
301 bch2_print(c, KERN_ERR bch2_fmt_dev(ca, fmt), ##__VA_ARGS__)
302 #define bch_err_dev_offset(ca, _offset, fmt, ...) \
303 bch2_print(c, KERN_ERR bch2_fmt_dev_offset(ca, _offset, fmt), ##__VA_ARGS__)
304 #define bch_err_inum(c, _inum, fmt, ...) \
305 bch2_print(c, KERN_ERR bch2_fmt_inum(c, _inum, fmt), ##__VA_ARGS__)
306 #define bch_err_inum_offset(c, _inum, _offset, fmt, ...) \
307 bch2_print(c, KERN_ERR bch2_fmt_inum_offset(c, _inum, _offset, fmt), ##__VA_ARGS__)
309 #define bch_err_ratelimited(c, fmt, ...) \
310 bch2_print_ratelimited(c, KERN_ERR bch2_fmt(c, fmt), ##__VA_ARGS__)
311 #define bch_err_dev_ratelimited(ca, fmt, ...) \
312 bch2_print_ratelimited(ca, KERN_ERR bch2_fmt_dev(ca, fmt), ##__VA_ARGS__)
313 #define bch_err_dev_offset_ratelimited(ca, _offset, fmt, ...) \
314 bch2_print_ratelimited(ca, KERN_ERR bch2_fmt_dev_offset(ca, _offset, fmt), ##__VA_ARGS__)
315 #define bch_err_inum_ratelimited(c, _inum, fmt, ...) \
316 bch2_print_ratelimited(c, KERN_ERR bch2_fmt_inum(c, _inum, fmt), ##__VA_ARGS__)
317 #define bch_err_inum_offset_ratelimited(c, _inum, _offset, fmt, ...) \
318 bch2_print_ratelimited(c, KERN_ERR bch2_fmt_inum_offset(c, _inum, _offset, fmt), ##__VA_ARGS__)
320 static inline bool should_print_err(int err)
322 return err && !bch2_err_matches(err, BCH_ERR_transaction_restart);
325 #define bch_err_fn(_c, _ret) \
327 if (should_print_err(_ret)) \
328 bch_err(_c, "%s(): error %s", __func__, bch2_err_str(_ret));\
331 #define bch_err_fn_ratelimited(_c, _ret) \
333 if (should_print_err(_ret)) \
334 bch_err_ratelimited(_c, "%s(): error %s", __func__, bch2_err_str(_ret));\
337 #define bch_err_msg(_c, _ret, _msg, ...) \
339 if (should_print_err(_ret)) \
340 bch_err(_c, "%s(): error " _msg " %s", __func__, \
341 ##__VA_ARGS__, bch2_err_str(_ret)); \
344 #define bch_verbose(c, fmt, ...) \
346 if ((c)->opts.verbose) \
347 bch_info(c, fmt, ##__VA_ARGS__); \
350 #define pr_verbose_init(opts, fmt, ...) \
352 if (opt_get(opts, verbose)) \
353 pr_info(fmt, ##__VA_ARGS__); \
356 /* Parameters that are useful for debugging, but should always be compiled in: */
357 #define BCH_DEBUG_PARAMS_ALWAYS() \
358 BCH_DEBUG_PARAM(key_merging_disabled, \
359 "Disables merging of extents") \
360 BCH_DEBUG_PARAM(btree_gc_always_rewrite, \
361 "Causes mark and sweep to compact and rewrite every " \
362 "btree node it traverses") \
363 BCH_DEBUG_PARAM(btree_gc_rewrite_disabled, \
364 "Disables rewriting of btree nodes during mark and sweep")\
365 BCH_DEBUG_PARAM(btree_shrinker_disabled, \
366 "Disables the shrinker callback for the btree node cache")\
367 BCH_DEBUG_PARAM(verify_btree_ondisk, \
368 "Reread btree nodes at various points to verify the " \
369 "mergesort in the read path against modifications " \
371 BCH_DEBUG_PARAM(verify_all_btree_replicas, \
372 "When reading btree nodes, read all replicas and " \
374 BCH_DEBUG_PARAM(backpointers_no_use_write_buffer, \
375 "Don't use the write buffer for backpointers, enabling "\
376 "extra runtime checks")
378 /* Parameters that should only be compiled in debug mode: */
379 #define BCH_DEBUG_PARAMS_DEBUG() \
380 BCH_DEBUG_PARAM(expensive_debug_checks, \
381 "Enables various runtime debugging checks that " \
382 "significantly affect performance") \
383 BCH_DEBUG_PARAM(debug_check_iterators, \
384 "Enables extra verification for btree iterators") \
385 BCH_DEBUG_PARAM(debug_check_btree_accounting, \
386 "Verify btree accounting for keys within a node") \
387 BCH_DEBUG_PARAM(journal_seq_verify, \
388 "Store the journal sequence number in the version " \
389 "number of every btree key, and verify that btree " \
390 "update ordering is preserved during recovery") \
391 BCH_DEBUG_PARAM(inject_invalid_keys, \
392 "Store the journal sequence number in the version " \
393 "number of every btree key, and verify that btree " \
394 "update ordering is preserved during recovery") \
395 BCH_DEBUG_PARAM(test_alloc_startup, \
396 "Force allocator startup to use the slowpath where it" \
397 "can't find enough free buckets without invalidating" \
399 BCH_DEBUG_PARAM(force_reconstruct_read, \
400 "Force reads to use the reconstruct path, when reading" \
401 "from erasure coded extents") \
402 BCH_DEBUG_PARAM(test_restart_gc, \
403 "Test restarting mark and sweep gc when bucket gens change")
405 #define BCH_DEBUG_PARAMS_ALL() BCH_DEBUG_PARAMS_ALWAYS() BCH_DEBUG_PARAMS_DEBUG()
407 #ifdef CONFIG_BCACHEFS_DEBUG
408 #define BCH_DEBUG_PARAMS() BCH_DEBUG_PARAMS_ALL()
410 #define BCH_DEBUG_PARAMS() BCH_DEBUG_PARAMS_ALWAYS()
413 #define BCH_DEBUG_PARAM(name, description) extern bool bch2_##name;
415 #undef BCH_DEBUG_PARAM
417 #ifndef CONFIG_BCACHEFS_DEBUG
418 #define BCH_DEBUG_PARAM(name, description) static const __maybe_unused bool bch2_##name;
419 BCH_DEBUG_PARAMS_DEBUG()
420 #undef BCH_DEBUG_PARAM
423 #define BCH_TIME_STATS() \
424 x(btree_node_mem_alloc) \
425 x(btree_node_split) \
426 x(btree_node_compact) \
427 x(btree_node_merge) \
430 x(btree_node_read_done) \
431 x(btree_interior_update_foreground) \
432 x(btree_interior_update_total) \
437 x(journal_flush_write) \
438 x(journal_noflush_write) \
439 x(journal_flush_seq) \
440 x(blocked_journal_low_on_space) \
441 x(blocked_journal_low_on_pin) \
442 x(blocked_journal_max_in_flight) \
443 x(blocked_allocate) \
444 x(blocked_allocate_open_bucket) \
445 x(blocked_write_buffer_full) \
446 x(nocow_lock_contended)
448 enum bch_time_stats {
449 #define x(name) BCH_TIME_##name,
455 #include "alloc_types.h"
456 #include "btree_types.h"
457 #include "btree_write_buffer_types.h"
458 #include "buckets_types.h"
459 #include "buckets_waiting_for_journal_types.h"
460 #include "clock_types.h"
461 #include "disk_groups_types.h"
462 #include "ec_types.h"
463 #include "journal_types.h"
464 #include "keylist_types.h"
465 #include "quota_types.h"
466 #include "rebalance_types.h"
467 #include "replicas_types.h"
468 #include "subvolume_types.h"
469 #include "super_types.h"
471 /* Number of nodes btree coalesce will try to coalesce at once */
472 #define GC_MERGE_NODES 4U
474 /* Maximum number of nodes we might need to allocate atomically: */
475 #define BTREE_RESERVE_MAX (BTREE_MAX_DEPTH + (BTREE_MAX_DEPTH - 1))
477 /* Size of the freelist we allocate btree nodes from: */
478 #define BTREE_NODE_RESERVE (BTREE_RESERVE_MAX * 4)
480 #define BTREE_NODE_OPEN_BUCKET_RESERVE (BTREE_RESERVE_MAX * BCH_REPLICAS_MAX)
485 GC_PHASE_NOT_RUNNING,
489 GC_PHASE_BTREE_stripes,
490 GC_PHASE_BTREE_extents,
491 GC_PHASE_BTREE_inodes,
492 GC_PHASE_BTREE_dirents,
493 GC_PHASE_BTREE_xattrs,
494 GC_PHASE_BTREE_alloc,
495 GC_PHASE_BTREE_quotas,
496 GC_PHASE_BTREE_reflink,
497 GC_PHASE_BTREE_subvolumes,
498 GC_PHASE_BTREE_snapshots,
500 GC_PHASE_BTREE_freespace,
501 GC_PHASE_BTREE_need_discard,
502 GC_PHASE_BTREE_backpointers,
503 GC_PHASE_BTREE_bucket_gens,
504 GC_PHASE_BTREE_snapshot_trees,
505 GC_PHASE_BTREE_deleted_inodes,
506 GC_PHASE_BTREE_logged_ops,
507 GC_PHASE_BTREE_rebalance_work,
509 GC_PHASE_PENDING_DELETE,
524 typedef GENRADIX(struct reflink_gc) reflink_gc_table;
527 u64 sectors[2][BCH_DATA_NR];
532 struct percpu_ref ref;
533 struct completion ref_completion;
534 struct percpu_ref io_ref;
535 struct completion io_ref_completion;
541 * Cached version of this device's member info from superblock
542 * Committed by bch2_write_super() -> bch_fs_mi_update()
544 struct bch_member_cpu mi;
545 atomic64_t errors[BCH_MEMBER_ERROR_NR];
548 char name[BDEVNAME_SIZE];
550 struct bch_sb_handle disk_sb;
551 struct bch_sb *sb_read_scratch;
556 struct bch_devs_mask self;
558 /* biosets used in cloned bios for writing multiple replicas */
559 struct bio_set replica_set;
563 * Per-bucket arrays are protected by c->mark_lock, bucket_lock and
564 * gc_lock, for device resize - holding any is sufficient for access:
565 * Or rcu_read_lock(), but only for ptr_stale():
567 struct bucket_array __rcu *buckets_gc;
568 struct bucket_gens __rcu *bucket_gens;
570 unsigned long *buckets_nouse;
571 struct rw_semaphore bucket_lock;
573 struct bch_dev_usage *usage_base;
574 struct bch_dev_usage __percpu *usage[JOURNAL_BUF_NR];
575 struct bch_dev_usage __percpu *usage_gc;
578 u64 new_fs_bucket_idx;
581 unsigned nr_open_buckets;
582 unsigned nr_btree_reserve;
584 size_t inc_gen_needs_gc;
585 size_t inc_gen_really_needs_gc;
586 size_t buckets_waiting_on_journal;
588 atomic64_t rebalance_work;
590 struct journal_device journal;
591 u64 prev_journal_sector;
593 struct work_struct io_error_work;
595 /* The rest of this all shows up in sysfs */
596 atomic64_t cur_latency[2];
597 struct time_stats_quantiles io_latency[2];
599 #define CONGESTED_MAX 1024
603 struct io_count __percpu *io_done;
612 #define BCH_FS_FLAGS() \
620 x(write_disable_complete) \
623 x(initial_gc_unfixed) \
625 x(need_delete_dead_snapshots) \
632 #define x(n) BCH_FS_##n,
641 #define BCH_TRANSACTIONS_NR 128
643 struct btree_transaction_stats {
644 struct time_stats duration;
645 struct time_stats lock_hold_times;
647 unsigned nr_max_paths;
648 unsigned journal_entries_size;
650 char *max_paths_text;
654 u64 sectors_available;
657 struct journal_seq_blacklist_table {
659 struct journal_seq_blacklist_table_entry {
666 struct journal_keys {
670 enum btree_id btree_id:8;
677 * Gap buffer: instead of all the empty space in the array being at the
678 * end of the buffer - from @nr to @size - the empty space is at @gap.
679 * This means that sequential insertions are O(n) instead of O(n^2).
685 bool initial_ref_held;
688 struct btree_trans_buf {
689 struct btree_trans *trans;
692 #define REPLICAS_DELTA_LIST_MAX (1U << 16)
694 #define BCACHEFS_ROOT_SUBVOL_INUM \
695 ((subvol_inum) { BCACHEFS_ROOT_SUBVOL, BCACHEFS_ROOT_INO })
697 #define BCH_WRITE_REFS() \
708 x(delete_dead_snapshots) \
709 x(snapshot_delete_pagecache) \
711 x(btree_write_buffer)
714 #define x(n) BCH_WRITE_REF_##n,
723 struct list_head list;
725 struct kobject counters_kobj;
726 struct kobject internal;
727 struct kobject opts_dir;
728 struct kobject time_stats;
732 struct device *chardev;
733 struct super_block *vfs_sb;
736 struct stdio_redirect *stdio;
737 struct task_struct *stdio_filter;
739 /* ro/rw, add/remove/resize devices: */
740 struct rw_semaphore state_lock;
742 /* Counts outstanding writes, for clean transition to read-only */
743 #ifdef BCH_WRITE_REF_DEBUG
744 atomic_long_t writes[BCH_WRITE_REF_NR];
746 struct percpu_ref writes;
749 * Analagous to c->writes, for asynchronous ops that don't necessarily
750 * need fs to be read-write
753 wait_queue_head_t ro_ref_wait;
755 struct work_struct read_only_work;
757 struct bch_dev __rcu *devs[BCH_SB_MEMBERS_MAX];
759 struct bch_replicas_cpu replicas;
760 struct bch_replicas_cpu replicas_gc;
761 struct mutex replicas_gc_lock;
762 mempool_t replicas_delta_pool;
764 struct journal_entry_res btree_root_journal_res;
765 struct journal_entry_res replicas_journal_res;
766 struct journal_entry_res clock_journal_res;
767 struct journal_entry_res dev_usage_journal_res;
769 struct bch_disk_groups_cpu __rcu *disk_groups;
771 struct bch_opts opts;
773 /* Updated by bch2_sb_update():*/
780 u16 version_upgrade_complete;
789 unsigned time_units_per_sec;
790 unsigned nsec_per_time_unit;
793 unsigned long errors_silent[BITS_TO_LONGS(BCH_SB_ERR_MAX)];
797 struct bch_sb_handle disk_sb;
799 unsigned short block_bits; /* ilog2(block_size) */
801 u16 btree_foreground_merge_threshold;
803 struct closure sb_write;
804 struct mutex sb_lock;
807 struct snapshot_table __rcu *snapshots;
808 size_t snapshot_table_size;
809 struct mutex snapshot_table_lock;
810 struct rw_semaphore snapshot_create_lock;
812 struct work_struct snapshot_delete_work;
813 struct work_struct snapshot_wait_for_pagecache_and_delete_work;
814 snapshot_id_list snapshots_unlinked;
815 struct mutex snapshots_unlinked_lock;
818 struct bio_set btree_bio;
819 struct workqueue_struct *io_complete_wq;
821 struct btree_root btree_roots_known[BTREE_ID_NR];
822 DARRAY(struct btree_root) btree_roots_extra;
823 struct mutex btree_root_lock;
825 struct btree_cache btree_cache;
828 * Cache of allocated btree nodes - if we allocate a btree node and
829 * don't use it, if we free it that space can't be reused until going
830 * _all_ the way through the allocator (which exposes us to a livelock
831 * when allocating btree reserves fail halfway through) - instead, we
832 * can stick them here:
834 struct btree_alloc btree_reserve_cache[BTREE_NODE_RESERVE * 2];
835 unsigned btree_reserve_cache_nr;
836 struct mutex btree_reserve_cache_lock;
838 mempool_t btree_interior_update_pool;
839 struct list_head btree_interior_update_list;
840 struct list_head btree_interior_updates_unwritten;
841 struct mutex btree_interior_update_lock;
842 struct closure_waitlist btree_interior_update_wait;
844 struct workqueue_struct *btree_interior_update_worker;
845 struct work_struct btree_interior_update_work;
847 struct list_head pending_node_rewrites;
848 struct mutex pending_node_rewrites_lock;
851 spinlock_t btree_write_error_lock;
852 struct btree_write_stats {
855 } btree_write_stats[BTREE_WRITE_TYPE_NR];
858 struct seqmutex btree_trans_lock;
859 struct list_head btree_trans_list;
860 mempool_t btree_trans_pool;
861 mempool_t btree_trans_mem_pool;
862 struct btree_trans_buf __percpu *btree_trans_bufs;
864 struct srcu_struct btree_trans_barrier;
865 bool btree_trans_barrier_initialized;
867 struct btree_key_cache btree_key_cache;
868 unsigned btree_key_cache_btrees;
870 struct btree_write_buffer btree_write_buffer;
872 struct workqueue_struct *btree_update_wq;
873 struct workqueue_struct *btree_io_complete_wq;
874 /* copygc needs its own workqueue for index updates.. */
875 struct workqueue_struct *copygc_wq;
877 * Use a dedicated wq for write ref holder tasks. Required to avoid
878 * dependency problems with other wq tasks that can block on ref
879 * draining, such as read-only transition.
881 struct workqueue_struct *write_ref_wq;
884 struct bch_devs_mask rw_devs[BCH_DATA_NR];
886 u64 capacity; /* sectors */
889 * When capacity _decreases_ (due to a disk being removed), we
890 * increment capacity_gen - this invalidates outstanding reservations
891 * and forces them to be revalidated
894 unsigned bucket_size_max;
896 atomic64_t sectors_available;
897 struct mutex sectors_available_lock;
899 struct bch_fs_pcpu __percpu *pcpu;
901 struct percpu_rw_semaphore mark_lock;
903 seqcount_t usage_lock;
904 struct bch_fs_usage *usage_base;
905 struct bch_fs_usage __percpu *usage[JOURNAL_BUF_NR];
906 struct bch_fs_usage __percpu *usage_gc;
907 u64 __percpu *online_reserved;
909 /* single element mempool: */
910 struct mutex usage_scratch_lock;
911 struct bch_fs_usage_online *usage_scratch;
913 struct io_clock io_clock[2];
915 /* JOURNAL SEQ BLACKLIST */
916 struct journal_seq_blacklist_table *
917 journal_seq_blacklist_table;
918 struct work_struct journal_seq_blacklist_gc_work;
921 spinlock_t freelist_lock;
922 struct closure_waitlist freelist_wait;
924 open_bucket_idx_t open_buckets_freelist;
925 open_bucket_idx_t open_buckets_nr_free;
926 struct closure_waitlist open_buckets_wait;
927 struct open_bucket open_buckets[OPEN_BUCKETS_COUNT];
928 open_bucket_idx_t open_buckets_hash[OPEN_BUCKETS_COUNT];
930 open_bucket_idx_t open_buckets_partial[OPEN_BUCKETS_COUNT];
931 open_bucket_idx_t open_buckets_partial_nr;
933 struct write_point btree_write_point;
934 struct write_point rebalance_write_point;
936 struct write_point write_points[WRITE_POINT_MAX];
937 struct hlist_head write_points_hash[WRITE_POINT_HASH_NR];
938 struct mutex write_points_hash_lock;
939 unsigned write_points_nr;
941 struct buckets_waiting_for_journal buckets_waiting_for_journal;
942 struct work_struct discard_work;
943 struct work_struct invalidate_work;
945 /* GARBAGE COLLECTION */
946 struct task_struct *gc_thread;
948 unsigned long gc_count;
950 enum btree_id gc_gens_btree;
951 struct bpos gc_gens_pos;
954 * Tracks GC's progress - everything in the range [ZERO_KEY..gc_cur_pos]
955 * has been marked by GC.
957 * gc_cur_phase is a superset of btree_ids (BTREE_ID_extents etc.)
959 * Protected by gc_pos_lock. Only written to by GC thread, so GC thread
960 * can read without a lock.
962 seqcount_t gc_pos_lock;
963 struct gc_pos gc_pos;
966 * The allocation code needs gc_mark in struct bucket to be correct, but
967 * it's not while a gc is in progress.
969 struct rw_semaphore gc_lock;
970 struct mutex gc_gens_lock;
973 struct semaphore io_in_flight;
974 struct bio_set bio_read;
975 struct bio_set bio_read_split;
976 struct bio_set bio_write;
977 struct mutex bio_bounce_pages_lock;
978 mempool_t bio_bounce_pages;
979 struct bucket_nocow_lock_table
981 struct rhashtable promote_table;
983 mempool_t compression_bounce[2];
984 mempool_t compress_workspace[BCH_COMPRESSION_TYPE_NR];
985 mempool_t decompress_workspace;
986 size_t zstd_workspace_size;
988 struct crypto_shash *sha256;
989 struct crypto_sync_skcipher *chacha20;
990 struct crypto_shash *poly1305;
992 atomic64_t key_version;
994 mempool_t large_bkey_pool;
997 struct list_head moving_context_list;
998 struct mutex moving_context_lock;
1001 struct bch_fs_rebalance rebalance;
1004 struct task_struct *copygc_thread;
1005 struct write_point copygc_write_point;
1008 bool copygc_running;
1009 wait_queue_head_t copygc_running_wq;
1012 GENRADIX(struct stripe) stripes;
1013 GENRADIX(struct gc_stripe) gc_stripes;
1015 struct hlist_head ec_stripes_new[32];
1016 spinlock_t ec_stripes_new_lock;
1018 ec_stripes_heap ec_stripes_heap;
1019 struct mutex ec_stripes_heap_lock;
1021 /* ERASURE CODING */
1022 struct list_head ec_stripe_head_list;
1023 struct mutex ec_stripe_head_lock;
1025 struct list_head ec_stripe_new_list;
1026 struct mutex ec_stripe_new_lock;
1027 wait_queue_head_t ec_stripe_new_wait;
1029 struct work_struct ec_stripe_create_work;
1032 struct work_struct ec_stripe_delete_work;
1034 struct bio_set ec_bioset;
1037 reflink_gc_table reflink_gc_table;
1038 size_t reflink_gc_nr;
1041 struct list_head vfs_inodes_list;
1042 struct mutex vfs_inodes_lock;
1044 /* VFS IO PATH - fs-io.c */
1045 struct bio_set writepage_bioset;
1046 struct bio_set dio_write_bioset;
1047 struct bio_set dio_read_bioset;
1048 struct bio_set nocow_flush_bioset;
1051 struct bch_memquota_type quotas[QTYP_NR];
1054 u64 journal_replay_seq_start;
1055 u64 journal_replay_seq_end;
1057 * Two different uses:
1058 * "Has this fsck pass?" - i.e. should this type of error be an
1059 * emergency read-only
1060 * And, in certain situations fsck will rewind to an earlier pass: used
1061 * for signaling to the toplevel code which pass we want to run now.
1063 enum bch_recovery_pass curr_recovery_pass;
1064 /* bitmap of explicitly enabled recovery passes: */
1065 u64 recovery_passes_explicit;
1066 /* bitmask of recovery passes that we actually ran */
1067 u64 recovery_passes_complete;
1068 /* never rewinds version of curr_recovery_pass */
1069 enum bch_recovery_pass recovery_pass_done;
1070 struct semaphore online_fsck_mutex;
1073 struct dentry *fs_debug_dir;
1074 struct dentry *btree_debug_dir;
1075 struct btree_debug btree_debug[BTREE_ID_NR];
1076 struct btree *verify_data;
1077 struct btree_node *verify_ondisk;
1078 struct mutex verify_lock;
1080 u64 *unused_inode_hints;
1081 unsigned inode_shard_bits;
1084 * A btree node on disk could have too many bsets for an iterator to fit
1085 * on the stack - have to dynamically allocate them
1087 mempool_t fill_iter;
1089 mempool_t btree_bounce_pool;
1091 struct journal journal;
1092 GENRADIX(struct journal_replay *) journal_entries;
1093 u64 journal_entries_base_seq;
1094 struct journal_keys journal_keys;
1095 struct list_head journal_iters;
1097 u64 last_bucket_seq_cleanup;
1099 u64 counters_on_mount[BCH_COUNTER_NR];
1100 u64 __percpu *counters;
1102 unsigned btree_gc_periodic:1;
1103 unsigned copy_gc_enabled:1;
1104 bool promote_whole_extents;
1106 struct time_stats times[BCH_TIME_STAT_NR];
1108 struct btree_transaction_stats btree_transaction_stats[BCH_TRANSACTIONS_NR];
1111 struct list_head fsck_error_msgs;
1112 struct mutex fsck_error_msgs_lock;
1113 bool fsck_alloc_msgs_err;
1115 bch_sb_errors_cpu fsck_error_counts;
1116 struct mutex fsck_error_counts_lock;
1119 extern struct wait_queue_head bch2_read_only_wait;
1121 static inline void bch2_write_ref_get(struct bch_fs *c, enum bch_write_ref ref)
1123 #ifdef BCH_WRITE_REF_DEBUG
1124 atomic_long_inc(&c->writes[ref]);
1126 percpu_ref_get(&c->writes);
1130 static inline bool __bch2_write_ref_tryget(struct bch_fs *c, enum bch_write_ref ref)
1132 #ifdef BCH_WRITE_REF_DEBUG
1133 return !test_bit(BCH_FS_going_ro, &c->flags) &&
1134 atomic_long_inc_not_zero(&c->writes[ref]);
1136 return percpu_ref_tryget(&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_live(&c->writes);
1150 static inline void bch2_write_ref_put(struct bch_fs *c, enum bch_write_ref ref)
1152 #ifdef BCH_WRITE_REF_DEBUG
1153 long v = atomic_long_dec_return(&c->writes[ref]);
1158 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
1159 if (atomic_long_read(&c->writes[i]))
1162 set_bit(BCH_FS_write_disable_complete, &c->flags);
1163 wake_up(&bch2_read_only_wait);
1165 percpu_ref_put(&c->writes);
1169 static inline bool bch2_ro_ref_tryget(struct bch_fs *c)
1171 if (test_bit(BCH_FS_stopping, &c->flags))
1174 return refcount_inc_not_zero(&c->ro_ref);
1177 static inline void bch2_ro_ref_put(struct bch_fs *c)
1179 if (refcount_dec_and_test(&c->ro_ref))
1180 wake_up(&c->ro_ref_wait);
1183 static inline void bch2_set_ra_pages(struct bch_fs *c, unsigned ra_pages)
1185 #ifndef NO_BCACHEFS_FS
1187 c->vfs_sb->s_bdi->ra_pages = ra_pages;
1191 static inline unsigned bucket_bytes(const struct bch_dev *ca)
1193 return ca->mi.bucket_size << 9;
1196 static inline unsigned block_bytes(const struct bch_fs *c)
1198 return c->opts.block_size;
1201 static inline unsigned block_sectors(const struct bch_fs *c)
1203 return c->opts.block_size >> 9;
1206 static inline bool btree_id_cached(const struct bch_fs *c, enum btree_id btree)
1208 return c->btree_key_cache_btrees & (1U << btree);
1211 static inline struct timespec64 bch2_time_to_timespec(const struct bch_fs *c, s64 time)
1213 struct timespec64 t;
1216 time += c->sb.time_base_lo;
1218 t.tv_sec = div_s64_rem(time, c->sb.time_units_per_sec, &rem);
1219 t.tv_nsec = rem * c->sb.nsec_per_time_unit;
1223 static inline s64 timespec_to_bch2_time(const struct bch_fs *c, struct timespec64 ts)
1225 return (ts.tv_sec * c->sb.time_units_per_sec +
1226 (int) ts.tv_nsec / c->sb.nsec_per_time_unit) - c->sb.time_base_lo;
1229 static inline s64 bch2_current_time(const struct bch_fs *c)
1231 struct timespec64 now;
1233 ktime_get_coarse_real_ts64(&now);
1234 return timespec_to_bch2_time(c, now);
1237 static inline bool bch2_dev_exists2(const struct bch_fs *c, unsigned dev)
1239 return dev < c->sb.nr_devices && c->devs[dev];
1242 static inline struct stdio_redirect *bch2_fs_stdio_redirect(struct bch_fs *c)
1244 struct stdio_redirect *stdio = c->stdio;
1246 if (c->stdio_filter && c->stdio_filter != current)
1251 #define BKEY_PADDED_ONSTACK(key, pad) \
1252 struct { struct bkey_i key; __u64 key ## _pad[pad]; }
1254 #endif /* _BCACHEFS_H */