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"
217 #ifdef CONFIG_BCACHEFS_DEBUG
218 #define BCH_WRITE_REF_DEBUG
221 #ifndef dynamic_fault
222 #define dynamic_fault(...) 0
225 #define race_fault(...) dynamic_fault("bcachefs:race")
227 #define count_event(_c, _name) this_cpu_inc((_c)->counters[BCH_COUNTER_##_name])
229 #define trace_and_count(_c, _name, ...) \
231 count_event(_c, _name); \
232 trace_##_name(__VA_ARGS__); \
235 #define bch2_fs_init_fault(name) \
236 dynamic_fault("bcachefs:bch_fs_init:" name)
237 #define bch2_meta_read_fault(name) \
238 dynamic_fault("bcachefs:meta:read:" name)
239 #define bch2_meta_write_fault(name) \
240 dynamic_fault("bcachefs:meta:write:" name)
243 #define BCACHEFS_LOG_PREFIX
246 #ifdef BCACHEFS_LOG_PREFIX
248 #define bch2_log_msg(_c, fmt) "bcachefs (%s): " fmt, ((_c)->name)
249 #define bch2_fmt_dev(_ca, fmt) "bcachefs (%s): " fmt "\n", ((_ca)->name)
250 #define bch2_fmt_dev_offset(_ca, _offset, fmt) "bcachefs (%s sector %llu): " fmt "\n", ((_ca)->name), (_offset)
251 #define bch2_fmt_inum(_c, _inum, fmt) "bcachefs (%s inum %llu): " fmt "\n", ((_c)->name), (_inum)
252 #define bch2_fmt_inum_offset(_c, _inum, _offset, fmt) \
253 "bcachefs (%s inum %llu offset %llu): " fmt "\n", ((_c)->name), (_inum), (_offset)
257 #define bch2_log_msg(_c, fmt) fmt
258 #define bch2_fmt_dev(_ca, fmt) "%s: " fmt "\n", ((_ca)->name)
259 #define bch2_fmt_dev_offset(_ca, _offset, fmt) "%s sector %llu: " fmt "\n", ((_ca)->name), (_offset)
260 #define bch2_fmt_inum(_c, _inum, fmt) "inum %llu: " fmt "\n", (_inum)
261 #define bch2_fmt_inum_offset(_c, _inum, _offset, fmt) \
262 "inum %llu offset %llu: " fmt "\n", (_inum), (_offset)
266 #define bch2_fmt(_c, fmt) bch2_log_msg(_c, fmt "\n")
269 void __bch2_print(struct bch_fs *c, const char *fmt, ...);
271 #define maybe_dev_to_fs(_c) _Generic((_c), \
272 struct bch_dev *: ((struct bch_dev *) (_c))->fs, \
273 struct bch_fs *: (_c))
275 #define bch2_print(_c, ...) __bch2_print(maybe_dev_to_fs(_c), __VA_ARGS__)
277 #define bch2_print_ratelimited(_c, ...) \
279 static DEFINE_RATELIMIT_STATE(_rs, \
280 DEFAULT_RATELIMIT_INTERVAL, \
281 DEFAULT_RATELIMIT_BURST); \
283 if (__ratelimit(&_rs)) \
284 bch2_print(_c, __VA_ARGS__); \
287 #define bch_info(c, fmt, ...) \
288 bch2_print(c, KERN_INFO bch2_fmt(c, fmt), ##__VA_ARGS__)
289 #define bch_notice(c, fmt, ...) \
290 bch2_print(c, KERN_NOTICE bch2_fmt(c, fmt), ##__VA_ARGS__)
291 #define bch_warn(c, fmt, ...) \
292 bch2_print(c, KERN_WARNING bch2_fmt(c, fmt), ##__VA_ARGS__)
293 #define bch_warn_ratelimited(c, fmt, ...) \
294 bch2_print_ratelimited(c, KERN_WARNING bch2_fmt(c, fmt), ##__VA_ARGS__)
296 #define bch_err(c, fmt, ...) \
297 bch2_print(c, KERN_ERR bch2_fmt(c, fmt), ##__VA_ARGS__)
298 #define bch_err_dev(ca, fmt, ...) \
299 bch2_print(c, KERN_ERR bch2_fmt_dev(ca, fmt), ##__VA_ARGS__)
300 #define bch_err_dev_offset(ca, _offset, fmt, ...) \
301 bch2_print(c, KERN_ERR bch2_fmt_dev_offset(ca, _offset, fmt), ##__VA_ARGS__)
302 #define bch_err_inum(c, _inum, fmt, ...) \
303 bch2_print(c, KERN_ERR bch2_fmt_inum(c, _inum, fmt), ##__VA_ARGS__)
304 #define bch_err_inum_offset(c, _inum, _offset, fmt, ...) \
305 bch2_print(c, KERN_ERR bch2_fmt_inum_offset(c, _inum, _offset, fmt), ##__VA_ARGS__)
307 #define bch_err_ratelimited(c, fmt, ...) \
308 bch2_print_ratelimited(c, KERN_ERR bch2_fmt(c, fmt), ##__VA_ARGS__)
309 #define bch_err_dev_ratelimited(ca, fmt, ...) \
310 bch2_print_ratelimited(ca, KERN_ERR bch2_fmt_dev(ca, fmt), ##__VA_ARGS__)
311 #define bch_err_dev_offset_ratelimited(ca, _offset, fmt, ...) \
312 bch2_print_ratelimited(ca, KERN_ERR bch2_fmt_dev_offset(ca, _offset, fmt), ##__VA_ARGS__)
313 #define bch_err_inum_ratelimited(c, _inum, fmt, ...) \
314 bch2_print_ratelimited(c, KERN_ERR bch2_fmt_inum(c, _inum, fmt), ##__VA_ARGS__)
315 #define bch_err_inum_offset_ratelimited(c, _inum, _offset, fmt, ...) \
316 bch2_print_ratelimited(c, KERN_ERR bch2_fmt_inum_offset(c, _inum, _offset, fmt), ##__VA_ARGS__)
318 #define bch_err_fn(_c, _ret) \
320 if (_ret && !bch2_err_matches(_ret, BCH_ERR_transaction_restart))\
321 bch_err(_c, "%s(): error %s", __func__, bch2_err_str(_ret));\
324 #define bch_err_msg(_c, _ret, _msg, ...) \
326 if (_ret && !bch2_err_matches(_ret, BCH_ERR_transaction_restart))\
327 bch_err(_c, "%s(): error " _msg " %s", __func__, \
328 ##__VA_ARGS__, bch2_err_str(_ret)); \
331 #define bch_verbose(c, fmt, ...) \
333 if ((c)->opts.verbose) \
334 bch_info(c, fmt, ##__VA_ARGS__); \
337 #define pr_verbose_init(opts, fmt, ...) \
339 if (opt_get(opts, verbose)) \
340 pr_info(fmt, ##__VA_ARGS__); \
343 /* Parameters that are useful for debugging, but should always be compiled in: */
344 #define BCH_DEBUG_PARAMS_ALWAYS() \
345 BCH_DEBUG_PARAM(key_merging_disabled, \
346 "Disables merging of extents") \
347 BCH_DEBUG_PARAM(btree_gc_always_rewrite, \
348 "Causes mark and sweep to compact and rewrite every " \
349 "btree node it traverses") \
350 BCH_DEBUG_PARAM(btree_gc_rewrite_disabled, \
351 "Disables rewriting of btree nodes during mark and sweep")\
352 BCH_DEBUG_PARAM(btree_shrinker_disabled, \
353 "Disables the shrinker callback for the btree node cache")\
354 BCH_DEBUG_PARAM(verify_btree_ondisk, \
355 "Reread btree nodes at various points to verify the " \
356 "mergesort in the read path against modifications " \
358 BCH_DEBUG_PARAM(verify_all_btree_replicas, \
359 "When reading btree nodes, read all replicas and " \
361 BCH_DEBUG_PARAM(backpointers_no_use_write_buffer, \
362 "Don't use the write buffer for backpointers, enabling "\
363 "extra runtime checks")
365 /* Parameters that should only be compiled in debug mode: */
366 #define BCH_DEBUG_PARAMS_DEBUG() \
367 BCH_DEBUG_PARAM(expensive_debug_checks, \
368 "Enables various runtime debugging checks that " \
369 "significantly affect performance") \
370 BCH_DEBUG_PARAM(debug_check_iterators, \
371 "Enables extra verification for btree iterators") \
372 BCH_DEBUG_PARAM(debug_check_btree_accounting, \
373 "Verify btree accounting for keys within a node") \
374 BCH_DEBUG_PARAM(journal_seq_verify, \
375 "Store the journal sequence number in the version " \
376 "number of every btree key, and verify that btree " \
377 "update ordering is preserved during recovery") \
378 BCH_DEBUG_PARAM(inject_invalid_keys, \
379 "Store the journal sequence number in the version " \
380 "number of every btree key, and verify that btree " \
381 "update ordering is preserved during recovery") \
382 BCH_DEBUG_PARAM(test_alloc_startup, \
383 "Force allocator startup to use the slowpath where it" \
384 "can't find enough free buckets without invalidating" \
386 BCH_DEBUG_PARAM(force_reconstruct_read, \
387 "Force reads to use the reconstruct path, when reading" \
388 "from erasure coded extents") \
389 BCH_DEBUG_PARAM(test_restart_gc, \
390 "Test restarting mark and sweep gc when bucket gens change")
392 #define BCH_DEBUG_PARAMS_ALL() BCH_DEBUG_PARAMS_ALWAYS() BCH_DEBUG_PARAMS_DEBUG()
394 #ifdef CONFIG_BCACHEFS_DEBUG
395 #define BCH_DEBUG_PARAMS() BCH_DEBUG_PARAMS_ALL()
397 #define BCH_DEBUG_PARAMS() BCH_DEBUG_PARAMS_ALWAYS()
400 #define BCH_DEBUG_PARAM(name, description) extern bool bch2_##name;
402 #undef BCH_DEBUG_PARAM
404 #ifndef CONFIG_BCACHEFS_DEBUG
405 #define BCH_DEBUG_PARAM(name, description) static const __maybe_unused bool bch2_##name;
406 BCH_DEBUG_PARAMS_DEBUG()
407 #undef BCH_DEBUG_PARAM
410 #define BCH_TIME_STATS() \
411 x(btree_node_mem_alloc) \
412 x(btree_node_split) \
413 x(btree_node_compact) \
414 x(btree_node_merge) \
417 x(btree_interior_update_foreground) \
418 x(btree_interior_update_total) \
423 x(journal_flush_write) \
424 x(journal_noflush_write) \
425 x(journal_flush_seq) \
426 x(blocked_journal_low_on_space) \
427 x(blocked_journal_low_on_pin) \
428 x(blocked_journal_max_in_flight) \
429 x(blocked_allocate) \
430 x(blocked_allocate_open_bucket) \
431 x(blocked_write_buffer_full) \
432 x(nocow_lock_contended)
434 enum bch_time_stats {
435 #define x(name) BCH_TIME_##name,
441 #include "alloc_types.h"
442 #include "btree_types.h"
443 #include "btree_write_buffer_types.h"
444 #include "buckets_types.h"
445 #include "buckets_waiting_for_journal_types.h"
446 #include "clock_types.h"
447 #include "disk_groups_types.h"
448 #include "ec_types.h"
449 #include "journal_types.h"
450 #include "keylist_types.h"
451 #include "quota_types.h"
452 #include "rebalance_types.h"
453 #include "replicas_types.h"
454 #include "subvolume_types.h"
455 #include "super_types.h"
457 /* Number of nodes btree coalesce will try to coalesce at once */
458 #define GC_MERGE_NODES 4U
460 /* Maximum number of nodes we might need to allocate atomically: */
461 #define BTREE_RESERVE_MAX (BTREE_MAX_DEPTH + (BTREE_MAX_DEPTH - 1))
463 /* Size of the freelist we allocate btree nodes from: */
464 #define BTREE_NODE_RESERVE (BTREE_RESERVE_MAX * 4)
466 #define BTREE_NODE_OPEN_BUCKET_RESERVE (BTREE_RESERVE_MAX * BCH_REPLICAS_MAX)
472 wait_queue_head_t wait;
477 GC_PHASE_NOT_RUNNING,
481 GC_PHASE_BTREE_stripes,
482 GC_PHASE_BTREE_extents,
483 GC_PHASE_BTREE_inodes,
484 GC_PHASE_BTREE_dirents,
485 GC_PHASE_BTREE_xattrs,
486 GC_PHASE_BTREE_alloc,
487 GC_PHASE_BTREE_quotas,
488 GC_PHASE_BTREE_reflink,
489 GC_PHASE_BTREE_subvolumes,
490 GC_PHASE_BTREE_snapshots,
492 GC_PHASE_BTREE_freespace,
493 GC_PHASE_BTREE_need_discard,
494 GC_PHASE_BTREE_backpointers,
495 GC_PHASE_BTREE_bucket_gens,
496 GC_PHASE_BTREE_snapshot_trees,
497 GC_PHASE_BTREE_deleted_inodes,
498 GC_PHASE_BTREE_logged_ops,
499 GC_PHASE_BTREE_rebalance_work,
501 GC_PHASE_PENDING_DELETE,
516 typedef GENRADIX(struct reflink_gc) reflink_gc_table;
519 u64 sectors[2][BCH_DATA_NR];
524 struct percpu_ref ref;
525 struct completion ref_completion;
526 struct percpu_ref io_ref;
527 struct completion io_ref_completion;
533 * Cached version of this device's member info from superblock
534 * Committed by bch2_write_super() -> bch_fs_mi_update()
536 struct bch_member_cpu mi;
537 atomic64_t errors[BCH_MEMBER_ERROR_NR];
540 char name[BDEVNAME_SIZE];
542 struct bch_sb_handle disk_sb;
543 struct bch_sb *sb_read_scratch;
548 struct bch_devs_mask self;
550 /* biosets used in cloned bios for writing multiple replicas */
551 struct bio_set replica_set;
555 * Per-bucket arrays are protected by c->mark_lock, bucket_lock and
556 * gc_lock, for device resize - holding any is sufficient for access:
557 * Or rcu_read_lock(), but only for ptr_stale():
559 struct bucket_array __rcu *buckets_gc;
560 struct bucket_gens __rcu *bucket_gens;
562 unsigned long *buckets_nouse;
563 struct rw_semaphore bucket_lock;
565 struct bch_dev_usage *usage_base;
566 struct bch_dev_usage __percpu *usage[JOURNAL_BUF_NR];
567 struct bch_dev_usage __percpu *usage_gc;
570 u64 new_fs_bucket_idx;
573 unsigned nr_open_buckets;
574 unsigned nr_btree_reserve;
576 size_t inc_gen_needs_gc;
577 size_t inc_gen_really_needs_gc;
578 size_t buckets_waiting_on_journal;
580 atomic64_t rebalance_work;
582 struct journal_device journal;
583 u64 prev_journal_sector;
585 struct work_struct io_error_work;
587 /* The rest of this all shows up in sysfs */
588 atomic64_t cur_latency[2];
589 struct bch2_time_stats io_latency[2];
591 #define CONGESTED_MAX 1024
595 struct io_count __percpu *io_done;
601 * replace with something more general from enumated fsck passes/errors:
607 #define BCH_FS_FLAGS() \
615 x(write_disable_complete) \
618 x(initial_gc_unfixed) \
620 x(need_delete_dead_snapshots) \
627 #define x(n) BCH_FS_##n,
636 #define BCH_TRANSACTIONS_NR 128
638 struct btree_transaction_stats {
639 struct bch2_time_stats lock_hold_times;
641 unsigned nr_max_paths;
642 unsigned journal_entries_size;
644 char *max_paths_text;
648 u64 sectors_available;
651 struct journal_seq_blacklist_table {
653 struct journal_seq_blacklist_table_entry {
660 struct journal_keys {
664 enum btree_id btree_id:8;
671 * Gap buffer: instead of all the empty space in the array being at the
672 * end of the buffer - from @nr to @size - the empty space is at @gap.
673 * This means that sequential insertions are O(n) instead of O(n^2).
679 bool initial_ref_held;
682 struct btree_trans_buf {
683 struct btree_trans *trans;
686 #define REPLICAS_DELTA_LIST_MAX (1U << 16)
688 #define BCACHEFS_ROOT_SUBVOL_INUM \
689 ((subvol_inum) { BCACHEFS_ROOT_SUBVOL, BCACHEFS_ROOT_INO })
691 #define BCH_WRITE_REFS() \
702 x(delete_dead_snapshots) \
703 x(snapshot_delete_pagecache) \
705 x(btree_write_buffer)
708 #define x(n) BCH_WRITE_REF_##n,
717 struct list_head list;
719 struct kobject counters_kobj;
720 struct kobject internal;
721 struct kobject opts_dir;
722 struct kobject time_stats;
726 struct device *chardev;
727 struct super_block *vfs_sb;
730 struct log_output *output;
731 struct task_struct *output_filter;
733 /* ro/rw, add/remove/resize devices: */
734 struct rw_semaphore state_lock;
736 /* Counts outstanding writes, for clean transition to read-only */
737 #ifdef BCH_WRITE_REF_DEBUG
738 atomic_long_t writes[BCH_WRITE_REF_NR];
740 struct percpu_ref writes;
743 * Analagous to c->writes, for asynchronous ops that don't necessarily
744 * need fs to be read-write
747 wait_queue_head_t ro_ref_wait;
749 struct work_struct read_only_work;
751 struct bch_dev __rcu *devs[BCH_SB_MEMBERS_MAX];
753 struct bch_replicas_cpu replicas;
754 struct bch_replicas_cpu replicas_gc;
755 struct mutex replicas_gc_lock;
756 mempool_t replicas_delta_pool;
758 struct journal_entry_res btree_root_journal_res;
759 struct journal_entry_res replicas_journal_res;
760 struct journal_entry_res clock_journal_res;
761 struct journal_entry_res dev_usage_journal_res;
763 struct bch_disk_groups_cpu __rcu *disk_groups;
765 struct bch_opts opts;
767 /* Updated by bch2_sb_update():*/
774 u16 version_upgrade_complete;
783 unsigned time_units_per_sec;
784 unsigned nsec_per_time_unit;
790 struct bch_sb_handle disk_sb;
792 unsigned short block_bits; /* ilog2(block_size) */
794 u16 btree_foreground_merge_threshold;
796 struct closure sb_write;
797 struct mutex sb_lock;
800 struct snapshot_table __rcu *snapshots;
801 size_t snapshot_table_size;
802 struct mutex snapshot_table_lock;
803 struct rw_semaphore snapshot_create_lock;
805 struct work_struct snapshot_delete_work;
806 struct work_struct snapshot_wait_for_pagecache_and_delete_work;
807 snapshot_id_list snapshots_unlinked;
808 struct mutex snapshots_unlinked_lock;
811 struct bio_set btree_bio;
812 struct workqueue_struct *io_complete_wq;
814 struct btree_root btree_roots_known[BTREE_ID_NR];
815 DARRAY(struct btree_root) btree_roots_extra;
816 struct mutex btree_root_lock;
818 struct btree_cache btree_cache;
821 * Cache of allocated btree nodes - if we allocate a btree node and
822 * don't use it, if we free it that space can't be reused until going
823 * _all_ the way through the allocator (which exposes us to a livelock
824 * when allocating btree reserves fail halfway through) - instead, we
825 * can stick them here:
827 struct btree_alloc btree_reserve_cache[BTREE_NODE_RESERVE * 2];
828 unsigned btree_reserve_cache_nr;
829 struct mutex btree_reserve_cache_lock;
831 mempool_t btree_interior_update_pool;
832 struct list_head btree_interior_update_list;
833 struct list_head btree_interior_updates_unwritten;
834 struct mutex btree_interior_update_lock;
835 struct closure_waitlist btree_interior_update_wait;
837 struct workqueue_struct *btree_interior_update_worker;
838 struct work_struct btree_interior_update_work;
840 struct list_head pending_node_rewrites;
841 struct mutex pending_node_rewrites_lock;
844 spinlock_t btree_write_error_lock;
845 struct btree_write_stats {
848 } btree_write_stats[BTREE_WRITE_TYPE_NR];
851 struct seqmutex btree_trans_lock;
852 struct list_head btree_trans_list;
853 mempool_t btree_trans_pool;
854 mempool_t btree_trans_mem_pool;
855 struct btree_trans_buf __percpu *btree_trans_bufs;
857 struct srcu_struct btree_trans_barrier;
858 bool btree_trans_barrier_initialized;
860 struct btree_key_cache btree_key_cache;
861 unsigned btree_key_cache_btrees;
863 struct btree_write_buffer btree_write_buffer;
865 struct workqueue_struct *btree_update_wq;
866 struct workqueue_struct *btree_io_complete_wq;
867 /* copygc needs its own workqueue for index updates.. */
868 struct workqueue_struct *copygc_wq;
870 * Use a dedicated wq for write ref holder tasks. Required to avoid
871 * dependency problems with other wq tasks that can block on ref
872 * draining, such as read-only transition.
874 struct workqueue_struct *write_ref_wq;
877 struct bch_devs_mask rw_devs[BCH_DATA_NR];
879 u64 capacity; /* sectors */
882 * When capacity _decreases_ (due to a disk being removed), we
883 * increment capacity_gen - this invalidates outstanding reservations
884 * and forces them to be revalidated
887 unsigned bucket_size_max;
889 atomic64_t sectors_available;
890 struct mutex sectors_available_lock;
892 struct bch_fs_pcpu __percpu *pcpu;
894 struct percpu_rw_semaphore mark_lock;
896 seqcount_t usage_lock;
897 struct bch_fs_usage *usage_base;
898 struct bch_fs_usage __percpu *usage[JOURNAL_BUF_NR];
899 struct bch_fs_usage __percpu *usage_gc;
900 u64 __percpu *online_reserved;
902 /* single element mempool: */
903 struct mutex usage_scratch_lock;
904 struct bch_fs_usage_online *usage_scratch;
906 struct io_clock io_clock[2];
908 /* JOURNAL SEQ BLACKLIST */
909 struct journal_seq_blacklist_table *
910 journal_seq_blacklist_table;
911 struct work_struct journal_seq_blacklist_gc_work;
914 spinlock_t freelist_lock;
915 struct closure_waitlist freelist_wait;
916 u64 blocked_allocate;
917 u64 blocked_allocate_open_bucket;
919 open_bucket_idx_t open_buckets_freelist;
920 open_bucket_idx_t open_buckets_nr_free;
921 struct closure_waitlist open_buckets_wait;
922 struct open_bucket open_buckets[OPEN_BUCKETS_COUNT];
923 open_bucket_idx_t open_buckets_hash[OPEN_BUCKETS_COUNT];
925 open_bucket_idx_t open_buckets_partial[OPEN_BUCKETS_COUNT];
926 open_bucket_idx_t open_buckets_partial_nr;
928 struct write_point btree_write_point;
929 struct write_point rebalance_write_point;
931 struct write_point write_points[WRITE_POINT_MAX];
932 struct hlist_head write_points_hash[WRITE_POINT_HASH_NR];
933 struct mutex write_points_hash_lock;
934 unsigned write_points_nr;
936 struct buckets_waiting_for_journal buckets_waiting_for_journal;
937 struct work_struct discard_work;
938 struct work_struct invalidate_work;
940 /* GARBAGE COLLECTION */
941 struct task_struct *gc_thread;
943 unsigned long gc_count;
945 enum btree_id gc_gens_btree;
946 struct bpos gc_gens_pos;
949 * Tracks GC's progress - everything in the range [ZERO_KEY..gc_cur_pos]
950 * has been marked by GC.
952 * gc_cur_phase is a superset of btree_ids (BTREE_ID_extents etc.)
954 * Protected by gc_pos_lock. Only written to by GC thread, so GC thread
955 * can read without a lock.
957 seqcount_t gc_pos_lock;
958 struct gc_pos gc_pos;
961 * The allocation code needs gc_mark in struct bucket to be correct, but
962 * it's not while a gc is in progress.
964 struct rw_semaphore gc_lock;
965 struct mutex gc_gens_lock;
968 struct semaphore io_in_flight;
969 struct bio_set bio_read;
970 struct bio_set bio_read_split;
971 struct bio_set bio_write;
972 struct mutex bio_bounce_pages_lock;
973 mempool_t bio_bounce_pages;
974 struct bucket_nocow_lock_table
976 struct rhashtable promote_table;
978 mempool_t compression_bounce[2];
979 mempool_t compress_workspace[BCH_COMPRESSION_TYPE_NR];
980 mempool_t decompress_workspace;
981 size_t zstd_workspace_size;
983 struct crypto_shash *sha256;
984 struct crypto_sync_skcipher *chacha20;
985 struct crypto_shash *poly1305;
987 atomic64_t key_version;
989 mempool_t large_bkey_pool;
992 struct list_head moving_context_list;
993 struct mutex moving_context_lock;
996 struct bch_fs_rebalance rebalance;
999 struct task_struct *copygc_thread;
1000 struct write_point copygc_write_point;
1003 bool copygc_running;
1004 wait_queue_head_t copygc_running_wq;
1007 GENRADIX(struct stripe) stripes;
1008 GENRADIX(struct gc_stripe) gc_stripes;
1010 struct hlist_head ec_stripes_new[32];
1011 spinlock_t ec_stripes_new_lock;
1013 ec_stripes_heap ec_stripes_heap;
1014 struct mutex ec_stripes_heap_lock;
1016 /* ERASURE CODING */
1017 struct list_head ec_stripe_head_list;
1018 struct mutex ec_stripe_head_lock;
1020 struct list_head ec_stripe_new_list;
1021 struct mutex ec_stripe_new_lock;
1022 wait_queue_head_t ec_stripe_new_wait;
1024 struct work_struct ec_stripe_create_work;
1027 struct work_struct ec_stripe_delete_work;
1029 struct bio_set ec_bioset;
1032 reflink_gc_table reflink_gc_table;
1033 size_t reflink_gc_nr;
1036 struct list_head vfs_inodes_list;
1037 struct mutex vfs_inodes_lock;
1039 /* VFS IO PATH - fs-io.c */
1040 struct bio_set writepage_bioset;
1041 struct bio_set dio_write_bioset;
1042 struct bio_set dio_read_bioset;
1043 struct bio_set nocow_flush_bioset;
1046 struct bch_memquota_type quotas[QTYP_NR];
1049 u64 journal_replay_seq_start;
1050 u64 journal_replay_seq_end;
1052 * Two different uses:
1053 * "Has this fsck pass?" - i.e. should this type of error be an
1054 * emergency read-only
1055 * And, in certain situations fsck will rewind to an earlier pass: used
1056 * for signaling to the toplevel code which pass we want to run now.
1058 enum bch_recovery_pass curr_recovery_pass;
1059 /* bitmap of explicitly enabled recovery passes: */
1060 u64 recovery_passes_explicit;
1061 /* bitmask of recovery passes that we actually ran */
1062 u64 recovery_passes_complete;
1063 /* never rewinds version of curr_recovery_pass */
1064 enum bch_recovery_pass recovery_pass_done;
1065 struct semaphore online_fsck_mutex;
1068 struct dentry *fs_debug_dir;
1069 struct dentry *btree_debug_dir;
1070 struct btree_debug btree_debug[BTREE_ID_NR];
1071 struct btree *verify_data;
1072 struct btree_node *verify_ondisk;
1073 struct mutex verify_lock;
1075 u64 *unused_inode_hints;
1076 unsigned inode_shard_bits;
1079 * A btree node on disk could have too many bsets for an iterator to fit
1080 * on the stack - have to dynamically allocate them
1082 mempool_t fill_iter;
1084 mempool_t btree_bounce_pool;
1086 struct journal journal;
1087 GENRADIX(struct journal_replay *) journal_entries;
1088 u64 journal_entries_base_seq;
1089 struct journal_keys journal_keys;
1090 struct list_head journal_iters;
1092 u64 last_bucket_seq_cleanup;
1094 u64 counters_on_mount[BCH_COUNTER_NR];
1095 u64 __percpu *counters;
1097 unsigned btree_gc_periodic:1;
1098 unsigned copy_gc_enabled:1;
1099 bool promote_whole_extents;
1101 struct bch2_time_stats times[BCH_TIME_STAT_NR];
1103 struct btree_transaction_stats btree_transaction_stats[BCH_TRANSACTIONS_NR];
1106 struct list_head fsck_error_msgs;
1107 struct mutex fsck_error_msgs_lock;
1108 bool fsck_alloc_msgs_err;
1110 bch_sb_errors_cpu fsck_error_counts;
1111 struct mutex fsck_error_counts_lock;
1114 extern struct wait_queue_head bch2_read_only_wait;
1116 static inline void bch2_write_ref_get(struct bch_fs *c, enum bch_write_ref ref)
1118 #ifdef BCH_WRITE_REF_DEBUG
1119 atomic_long_inc(&c->writes[ref]);
1121 percpu_ref_get(&c->writes);
1125 static inline bool __bch2_write_ref_tryget(struct bch_fs *c, enum bch_write_ref ref)
1127 #ifdef BCH_WRITE_REF_DEBUG
1128 return !test_bit(BCH_FS_going_ro, &c->flags) &&
1129 atomic_long_inc_not_zero(&c->writes[ref]);
1131 return percpu_ref_tryget(&c->writes);
1135 static inline bool bch2_write_ref_tryget(struct bch_fs *c, enum bch_write_ref ref)
1137 #ifdef BCH_WRITE_REF_DEBUG
1138 return !test_bit(BCH_FS_going_ro, &c->flags) &&
1139 atomic_long_inc_not_zero(&c->writes[ref]);
1141 return percpu_ref_tryget_live(&c->writes);
1145 static inline void bch2_write_ref_put(struct bch_fs *c, enum bch_write_ref ref)
1147 #ifdef BCH_WRITE_REF_DEBUG
1148 long v = atomic_long_dec_return(&c->writes[ref]);
1153 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
1154 if (atomic_long_read(&c->writes[i]))
1157 set_bit(BCH_FS_write_disable_complete, &c->flags);
1158 wake_up(&bch2_read_only_wait);
1160 percpu_ref_put(&c->writes);
1164 static inline bool bch2_ro_ref_tryget(struct bch_fs *c)
1166 if (test_bit(BCH_FS_stopping, &c->flags))
1169 return refcount_inc_not_zero(&c->ro_ref);
1172 static inline void bch2_ro_ref_put(struct bch_fs *c)
1174 if (refcount_dec_and_test(&c->ro_ref))
1175 wake_up(&c->ro_ref_wait);
1178 static inline void bch2_set_ra_pages(struct bch_fs *c, unsigned ra_pages)
1180 #ifndef NO_BCACHEFS_FS
1182 c->vfs_sb->s_bdi->ra_pages = ra_pages;
1186 static inline unsigned bucket_bytes(const struct bch_dev *ca)
1188 return ca->mi.bucket_size << 9;
1191 static inline unsigned block_bytes(const struct bch_fs *c)
1193 return c->opts.block_size;
1196 static inline unsigned block_sectors(const struct bch_fs *c)
1198 return c->opts.block_size >> 9;
1201 static inline size_t btree_sectors(const struct bch_fs *c)
1203 return c->opts.btree_node_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 #define BKEY_PADDED_ONSTACK(key, pad) \
1243 struct { struct bkey_i key; __u64 key ## _pad[pad]; }
1245 #endif /* _BCACHEFS_H */