1 /* SPDX-License-Identifier: GPL-2.0 */
6 * SOME HIGH LEVEL CODE DOCUMENTATION:
8 * Bcache mostly works with cache sets, cache devices, and backing devices.
10 * Support for multiple cache devices hasn't quite been finished off yet, but
11 * it's about 95% plumbed through. A cache set and its cache devices is sort of
12 * like a md raid array and its component devices. Most of the code doesn't care
13 * about individual cache devices, the main abstraction is the cache set.
15 * Multiple cache devices is intended to give us the ability to mirror dirty
16 * cached data and metadata, without mirroring clean cached data.
18 * Backing devices are different, in that they have a lifetime independent of a
19 * cache set. When you register a newly formatted backing device it'll come up
20 * in passthrough mode, and then you can attach and detach a backing device from
21 * a cache set at runtime - while it's mounted and in use. Detaching implicitly
22 * invalidates any cached data for that backing device.
24 * A cache set can have multiple (many) backing devices attached to it.
26 * There's also flash only volumes - this is the reason for the distinction
27 * between struct cached_dev and struct bcache_device. A flash only volume
28 * works much like a bcache device that has a backing device, except the
29 * "cached" data is always dirty. The end result is that we get thin
30 * provisioning with very little additional code.
32 * Flash only volumes work but they're not production ready because the moving
33 * garbage collector needs more work. More on that later.
37 * Bcache is primarily designed for caching, which means that in normal
38 * operation all of our available space will be allocated. Thus, we need an
39 * efficient way of deleting things from the cache so we can write new things to
42 * To do this, we first divide the cache device up into buckets. A bucket is the
43 * unit of allocation; they're typically around 1 mb - anywhere from 128k to 2M+
46 * Each bucket has a 16 bit priority, and an 8 bit generation associated with
47 * it. The gens and priorities for all the buckets are stored contiguously and
48 * packed on disk (in a linked list of buckets - aside from the superblock, all
49 * of bcache's metadata is stored in buckets).
51 * The priority is used to implement an LRU. We reset a bucket's priority when
52 * we allocate it or on cache it, and every so often we decrement the priority
53 * of each bucket. It could be used to implement something more sophisticated,
54 * if anyone ever gets around to it.
56 * The generation is used for invalidating buckets. Each pointer also has an 8
57 * bit generation embedded in it; for a pointer to be considered valid, its gen
58 * must match the gen of the bucket it points into. Thus, to reuse a bucket all
59 * we have to do is increment its gen (and write its new gen to disk; we batch
62 * Bcache is entirely COW - we never write twice to a bucket, even buckets that
63 * contain metadata (including btree nodes).
67 * Bcache is in large part design around the btree.
69 * At a high level, the btree is just an index of key -> ptr tuples.
71 * Keys represent extents, and thus have a size field. Keys also have a variable
72 * number of pointers attached to them (potentially zero, which is handy for
73 * invalidating the cache).
75 * The key itself is an inode:offset pair. The inode number corresponds to a
76 * backing device or a flash only volume. The offset is the ending offset of the
77 * extent within the inode - not the starting offset; this makes lookups
78 * slightly more convenient.
80 * Pointers contain the cache device id, the offset on that device, and an 8 bit
81 * generation number. More on the gen later.
83 * Index lookups are not fully abstracted - cache lookups in particular are
84 * still somewhat mixed in with the btree code, but things are headed in that
87 * Updates are fairly well abstracted, though. There are two different ways of
88 * updating the btree; insert and replace.
90 * BTREE_INSERT will just take a list of keys and insert them into the btree -
91 * overwriting (possibly only partially) any extents they overlap with. This is
92 * used to update the index after a write.
94 * BTREE_REPLACE is really cmpxchg(); it inserts a key into the btree iff it is
95 * overwriting a key that matches another given key. This is used for inserting
96 * data into the cache after a cache miss, and for background writeback, and for
97 * the moving garbage collector.
99 * There is no "delete" operation; deleting things from the index is
100 * accomplished by either by invalidating pointers (by incrementing a bucket's
101 * gen) or by inserting a key with 0 pointers - which will overwrite anything
102 * previously present at that location in the index.
104 * This means that there are always stale/invalid keys in the btree. They're
105 * filtered out by the code that iterates through a btree node, and removed when
106 * a btree node is rewritten.
110 * Our unit of allocation is a bucket, and we can't arbitrarily allocate and
111 * free smaller than a bucket - so, that's how big our btree nodes are.
113 * (If buckets are really big we'll only use part of the bucket for a btree node
114 * - no less than 1/4th - but a bucket still contains no more than a single
115 * btree node. I'd actually like to change this, but for now we rely on the
116 * bucket's gen for deleting btree nodes when we rewrite/split a node.)
118 * Anyways, btree nodes are big - big enough to be inefficient with a textbook
119 * btree implementation.
121 * The way this is solved is that btree nodes are internally log structured; we
122 * can append new keys to an existing btree node without rewriting it. This
123 * means each set of keys we write is sorted, but the node is not.
125 * We maintain this log structure in memory - keeping 1Mb of keys sorted would
126 * be expensive, and we have to distinguish between the keys we have written and
127 * the keys we haven't. So to do a lookup in a btree node, we have to search
128 * each sorted set. But we do merge written sets together lazily, so the cost of
129 * these extra searches is quite low (normally most of the keys in a btree node
130 * will be in one big set, and then there'll be one or two sets that are much
133 * This log structure makes bcache's btree more of a hybrid between a
134 * conventional btree and a compacting data structure, with some of the
135 * advantages of both.
137 * GARBAGE COLLECTION:
139 * We can't just invalidate any bucket - it might contain dirty data or
140 * metadata. If it once contained dirty data, other writes might overwrite it
141 * later, leaving no valid pointers into that bucket in the index.
143 * Thus, the primary purpose of garbage collection is to find buckets to reuse.
144 * It also counts how much valid data it each bucket currently contains, so that
145 * allocation can reuse buckets sooner when they've been mostly overwritten.
147 * It also does some things that are really internal to the btree
148 * implementation. If a btree node contains pointers that are stale by more than
149 * some threshold, it rewrites the btree node to avoid the bucket's generation
150 * wrapping around. It also merges adjacent btree nodes if they're empty enough.
154 * Bcache's journal is not necessary for consistency; we always strictly
155 * order metadata writes so that the btree and everything else is consistent on
156 * disk in the event of an unclean shutdown, and in fact bcache had writeback
157 * caching (with recovery from unclean shutdown) before journalling was
160 * Rather, the journal is purely a performance optimization; we can't complete a
161 * write until we've updated the index on disk, otherwise the cache would be
162 * inconsistent in the event of an unclean shutdown. This means that without the
163 * journal, on random write workloads we constantly have to update all the leaf
164 * nodes in the btree, and those writes will be mostly empty (appending at most
165 * a few keys each) - highly inefficient in terms of amount of metadata writes,
166 * and it puts more strain on the various btree resorting/compacting code.
168 * The journal is just a log of keys we've inserted; on startup we just reinsert
169 * all the keys in the open journal entries. That means that when we're updating
170 * a node in the btree, we can wait until a 4k block of keys fills up before
173 * For simplicity, we only journal updates to leaf nodes; updates to parent
174 * nodes are rare enough (since our leaf nodes are huge) that it wasn't worth
175 * the complexity to deal with journalling them (in particular, journal replay)
176 * - updates to non leaf nodes just happen synchronously (see btree_split()).
181 #define pr_fmt(fmt) "bcachefs: %s() " fmt "\n", __func__
183 #define pr_fmt(fmt) "%s() " fmt "\n", __func__
186 #include <linux/backing-dev-defs.h>
187 #include <linux/bug.h>
188 #include <linux/bio.h>
189 #include <linux/closure.h>
190 #include <linux/kobject.h>
191 #include <linux/list.h>
192 #include <linux/math64.h>
193 #include <linux/mutex.h>
194 #include <linux/percpu-refcount.h>
195 #include <linux/percpu-rwsem.h>
196 #include <linux/rhashtable.h>
197 #include <linux/rwsem.h>
198 #include <linux/semaphore.h>
199 #include <linux/seqlock.h>
200 #include <linux/shrinker.h>
201 #include <linux/srcu.h>
202 #include <linux/types.h>
203 #include <linux/workqueue.h>
204 #include <linux/zstd.h>
206 #include "bcachefs_format.h"
209 #include "nocow_locking_types.h"
211 #include "recovery_types.h"
212 #include "sb-errors_types.h"
213 #include "seqmutex.h"
216 #ifdef CONFIG_BCACHEFS_DEBUG
217 #define BCH_WRITE_REF_DEBUG
220 #ifndef dynamic_fault
221 #define dynamic_fault(...) 0
224 #define race_fault(...) dynamic_fault("bcachefs:race")
226 #define trace_and_count(_c, _name, ...) \
228 this_cpu_inc((_c)->counters[BCH_COUNTER_##_name]); \
229 trace_##_name(__VA_ARGS__); \
232 #define bch2_fs_init_fault(name) \
233 dynamic_fault("bcachefs:bch_fs_init:" name)
234 #define bch2_meta_read_fault(name) \
235 dynamic_fault("bcachefs:meta:read:" name)
236 #define bch2_meta_write_fault(name) \
237 dynamic_fault("bcachefs:meta:write:" name)
240 #define BCACHEFS_LOG_PREFIX
243 #ifdef BCACHEFS_LOG_PREFIX
245 #define bch2_log_msg(_c, fmt) "bcachefs (%s): " fmt, ((_c)->name)
246 #define bch2_fmt_dev(_ca, fmt) "bcachefs (%s): " fmt "\n", ((_ca)->name)
247 #define bch2_fmt_dev_offset(_ca, _offset, fmt) "bcachefs (%s sector %llu): " fmt "\n", ((_ca)->name), (_offset)
248 #define bch2_fmt_inum(_c, _inum, fmt) "bcachefs (%s inum %llu): " fmt "\n", ((_c)->name), (_inum)
249 #define bch2_fmt_inum_offset(_c, _inum, _offset, fmt) \
250 "bcachefs (%s inum %llu offset %llu): " fmt "\n", ((_c)->name), (_inum), (_offset)
254 #define bch2_log_msg(_c, fmt) fmt
255 #define bch2_fmt_dev(_ca, fmt) "%s: " fmt "\n", ((_ca)->name)
256 #define bch2_fmt_dev_offset(_ca, _offset, fmt) "%s sector %llu: " fmt "\n", ((_ca)->name), (_offset)
257 #define bch2_fmt_inum(_c, _inum, fmt) "inum %llu: " fmt "\n", (_inum)
258 #define bch2_fmt_inum_offset(_c, _inum, _offset, fmt) \
259 "inum %llu offset %llu: " fmt "\n", (_inum), (_offset)
263 #define bch2_fmt(_c, fmt) bch2_log_msg(_c, fmt "\n")
265 #define bch_info(c, fmt, ...) \
266 printk(KERN_INFO bch2_fmt(c, fmt), ##__VA_ARGS__)
267 #define bch_notice(c, fmt, ...) \
268 printk(KERN_NOTICE bch2_fmt(c, fmt), ##__VA_ARGS__)
269 #define bch_warn(c, fmt, ...) \
270 printk(KERN_WARNING bch2_fmt(c, fmt), ##__VA_ARGS__)
271 #define bch_warn_ratelimited(c, fmt, ...) \
272 printk_ratelimited(KERN_WARNING bch2_fmt(c, fmt), ##__VA_ARGS__)
274 #define bch_err(c, fmt, ...) \
275 printk(KERN_ERR bch2_fmt(c, fmt), ##__VA_ARGS__)
276 #define bch_err_dev(ca, fmt, ...) \
277 printk(KERN_ERR bch2_fmt_dev(ca, fmt), ##__VA_ARGS__)
278 #define bch_err_dev_offset(ca, _offset, fmt, ...) \
279 printk(KERN_ERR bch2_fmt_dev_offset(ca, _offset, fmt), ##__VA_ARGS__)
280 #define bch_err_inum(c, _inum, fmt, ...) \
281 printk(KERN_ERR bch2_fmt_inum(c, _inum, fmt), ##__VA_ARGS__)
282 #define bch_err_inum_offset(c, _inum, _offset, fmt, ...) \
283 printk(KERN_ERR bch2_fmt_inum_offset(c, _inum, _offset, fmt), ##__VA_ARGS__)
285 #define bch_err_ratelimited(c, fmt, ...) \
286 printk_ratelimited(KERN_ERR bch2_fmt(c, fmt), ##__VA_ARGS__)
287 #define bch_err_dev_ratelimited(ca, fmt, ...) \
288 printk_ratelimited(KERN_ERR bch2_fmt_dev(ca, fmt), ##__VA_ARGS__)
289 #define bch_err_dev_offset_ratelimited(ca, _offset, fmt, ...) \
290 printk_ratelimited(KERN_ERR bch2_fmt_dev_offset(ca, _offset, fmt), ##__VA_ARGS__)
291 #define bch_err_inum_ratelimited(c, _inum, fmt, ...) \
292 printk_ratelimited(KERN_ERR bch2_fmt_inum(c, _inum, fmt), ##__VA_ARGS__)
293 #define bch_err_inum_offset_ratelimited(c, _inum, _offset, fmt, ...) \
294 printk_ratelimited(KERN_ERR bch2_fmt_inum_offset(c, _inum, _offset, fmt), ##__VA_ARGS__)
296 #define bch_err_fn(_c, _ret) \
298 if (_ret && !bch2_err_matches(_ret, BCH_ERR_transaction_restart))\
299 bch_err(_c, "%s(): error %s", __func__, bch2_err_str(_ret));\
302 #define bch_err_msg(_c, _ret, _msg, ...) \
304 if (_ret && !bch2_err_matches(_ret, BCH_ERR_transaction_restart))\
305 bch_err(_c, "%s(): error " _msg " %s", __func__, \
306 ##__VA_ARGS__, bch2_err_str(_ret)); \
309 #define bch_verbose(c, fmt, ...) \
311 if ((c)->opts.verbose) \
312 bch_info(c, fmt, ##__VA_ARGS__); \
315 #define pr_verbose_init(opts, fmt, ...) \
317 if (opt_get(opts, verbose)) \
318 pr_info(fmt, ##__VA_ARGS__); \
321 /* Parameters that are useful for debugging, but should always be compiled in: */
322 #define BCH_DEBUG_PARAMS_ALWAYS() \
323 BCH_DEBUG_PARAM(key_merging_disabled, \
324 "Disables merging of extents") \
325 BCH_DEBUG_PARAM(btree_gc_always_rewrite, \
326 "Causes mark and sweep to compact and rewrite every " \
327 "btree node it traverses") \
328 BCH_DEBUG_PARAM(btree_gc_rewrite_disabled, \
329 "Disables rewriting of btree nodes during mark and sweep")\
330 BCH_DEBUG_PARAM(btree_shrinker_disabled, \
331 "Disables the shrinker callback for the btree node cache")\
332 BCH_DEBUG_PARAM(verify_btree_ondisk, \
333 "Reread btree nodes at various points to verify the " \
334 "mergesort in the read path against modifications " \
336 BCH_DEBUG_PARAM(verify_all_btree_replicas, \
337 "When reading btree nodes, read all replicas and " \
339 BCH_DEBUG_PARAM(backpointers_no_use_write_buffer, \
340 "Don't use the write buffer for backpointers, enabling "\
341 "extra runtime checks")
343 /* Parameters that should only be compiled in debug mode: */
344 #define BCH_DEBUG_PARAMS_DEBUG() \
345 BCH_DEBUG_PARAM(expensive_debug_checks, \
346 "Enables various runtime debugging checks that " \
347 "significantly affect performance") \
348 BCH_DEBUG_PARAM(debug_check_iterators, \
349 "Enables extra verification for btree iterators") \
350 BCH_DEBUG_PARAM(debug_check_btree_accounting, \
351 "Verify btree accounting for keys within a node") \
352 BCH_DEBUG_PARAM(journal_seq_verify, \
353 "Store the journal sequence number in the version " \
354 "number of every btree key, and verify that btree " \
355 "update ordering is preserved during recovery") \
356 BCH_DEBUG_PARAM(inject_invalid_keys, \
357 "Store the journal sequence number in the version " \
358 "number of every btree key, and verify that btree " \
359 "update ordering is preserved during recovery") \
360 BCH_DEBUG_PARAM(test_alloc_startup, \
361 "Force allocator startup to use the slowpath where it" \
362 "can't find enough free buckets without invalidating" \
364 BCH_DEBUG_PARAM(force_reconstruct_read, \
365 "Force reads to use the reconstruct path, when reading" \
366 "from erasure coded extents") \
367 BCH_DEBUG_PARAM(test_restart_gc, \
368 "Test restarting mark and sweep gc when bucket gens change")
370 #define BCH_DEBUG_PARAMS_ALL() BCH_DEBUG_PARAMS_ALWAYS() BCH_DEBUG_PARAMS_DEBUG()
372 #ifdef CONFIG_BCACHEFS_DEBUG
373 #define BCH_DEBUG_PARAMS() BCH_DEBUG_PARAMS_ALL()
375 #define BCH_DEBUG_PARAMS() BCH_DEBUG_PARAMS_ALWAYS()
378 #define BCH_DEBUG_PARAM(name, description) extern bool bch2_##name;
380 #undef BCH_DEBUG_PARAM
382 #ifndef CONFIG_BCACHEFS_DEBUG
383 #define BCH_DEBUG_PARAM(name, description) static const __maybe_unused bool bch2_##name;
384 BCH_DEBUG_PARAMS_DEBUG()
385 #undef BCH_DEBUG_PARAM
388 #define BCH_TIME_STATS() \
389 x(btree_node_mem_alloc) \
390 x(btree_node_split) \
391 x(btree_node_compact) \
392 x(btree_node_merge) \
395 x(btree_interior_update_foreground) \
396 x(btree_interior_update_total) \
401 x(journal_flush_write) \
402 x(journal_noflush_write) \
403 x(journal_flush_seq) \
404 x(blocked_journal_low_on_space) \
405 x(blocked_journal_low_on_pin) \
406 x(blocked_journal_max_in_flight) \
407 x(blocked_allocate) \
408 x(blocked_allocate_open_bucket) \
409 x(blocked_write_buffer_full) \
410 x(nocow_lock_contended)
412 enum bch_time_stats {
413 #define x(name) BCH_TIME_##name,
419 #include "alloc_types.h"
420 #include "btree_types.h"
421 #include "btree_write_buffer_types.h"
422 #include "buckets_types.h"
423 #include "buckets_waiting_for_journal_types.h"
424 #include "clock_types.h"
425 #include "disk_groups_types.h"
426 #include "ec_types.h"
427 #include "journal_types.h"
428 #include "keylist_types.h"
429 #include "quota_types.h"
430 #include "rebalance_types.h"
431 #include "replicas_types.h"
432 #include "subvolume_types.h"
433 #include "super_types.h"
435 /* Number of nodes btree coalesce will try to coalesce at once */
436 #define GC_MERGE_NODES 4U
438 /* Maximum number of nodes we might need to allocate atomically: */
439 #define BTREE_RESERVE_MAX (BTREE_MAX_DEPTH + (BTREE_MAX_DEPTH - 1))
441 /* Size of the freelist we allocate btree nodes from: */
442 #define BTREE_NODE_RESERVE (BTREE_RESERVE_MAX * 4)
444 #define BTREE_NODE_OPEN_BUCKET_RESERVE (BTREE_RESERVE_MAX * BCH_REPLICAS_MAX)
449 GC_PHASE_NOT_RUNNING,
453 GC_PHASE_BTREE_stripes,
454 GC_PHASE_BTREE_extents,
455 GC_PHASE_BTREE_inodes,
456 GC_PHASE_BTREE_dirents,
457 GC_PHASE_BTREE_xattrs,
458 GC_PHASE_BTREE_alloc,
459 GC_PHASE_BTREE_quotas,
460 GC_PHASE_BTREE_reflink,
461 GC_PHASE_BTREE_subvolumes,
462 GC_PHASE_BTREE_snapshots,
464 GC_PHASE_BTREE_freespace,
465 GC_PHASE_BTREE_need_discard,
466 GC_PHASE_BTREE_backpointers,
467 GC_PHASE_BTREE_bucket_gens,
468 GC_PHASE_BTREE_snapshot_trees,
469 GC_PHASE_BTREE_deleted_inodes,
470 GC_PHASE_BTREE_logged_ops,
471 GC_PHASE_BTREE_rebalance_work,
473 GC_PHASE_PENDING_DELETE,
488 typedef GENRADIX(struct reflink_gc) reflink_gc_table;
491 u64 sectors[2][BCH_DATA_NR];
496 struct percpu_ref ref;
497 struct completion ref_completion;
498 struct percpu_ref io_ref;
499 struct completion io_ref_completion;
505 * Cached version of this device's member info from superblock
506 * Committed by bch2_write_super() -> bch_fs_mi_update()
508 struct bch_member_cpu mi;
509 atomic64_t errors[BCH_MEMBER_ERROR_NR];
512 char name[BDEVNAME_SIZE];
514 struct bch_sb_handle disk_sb;
515 struct bch_sb *sb_read_scratch;
520 struct bch_devs_mask self;
522 /* biosets used in cloned bios for writing multiple replicas */
523 struct bio_set replica_set;
527 * Per-bucket arrays are protected by c->mark_lock, bucket_lock and
528 * gc_lock, for device resize - holding any is sufficient for access:
529 * Or rcu_read_lock(), but only for ptr_stale():
531 struct bucket_array __rcu *buckets_gc;
532 struct bucket_gens __rcu *bucket_gens;
534 unsigned long *buckets_nouse;
535 struct rw_semaphore bucket_lock;
537 struct bch_dev_usage *usage_base;
538 struct bch_dev_usage __percpu *usage[JOURNAL_BUF_NR];
539 struct bch_dev_usage __percpu *usage_gc;
542 u64 new_fs_bucket_idx;
545 unsigned nr_open_buckets;
546 unsigned nr_btree_reserve;
548 size_t inc_gen_needs_gc;
549 size_t inc_gen_really_needs_gc;
550 size_t buckets_waiting_on_journal;
552 atomic64_t rebalance_work;
554 struct journal_device journal;
555 u64 prev_journal_sector;
557 struct work_struct io_error_work;
559 /* The rest of this all shows up in sysfs */
560 atomic64_t cur_latency[2];
561 struct bch2_time_stats io_latency[2];
563 #define CONGESTED_MAX 1024
567 struct io_count __percpu *io_done;
581 BCH_FS_WRITE_DISABLE_COMPLETE,
582 BCH_FS_CLEAN_SHUTDOWN,
586 BCH_FS_INITIAL_GC_UNFIXED, /* kill when we enumerate fsck errors */
587 BCH_FS_NEED_ANOTHER_GC,
589 BCH_FS_NEED_DELETE_DEAD_SNAPSHOTS,
593 BCH_FS_TOPOLOGY_ERROR,
595 BCH_FS_ERRORS_NOT_FIXED,
602 #define BCH_TRANSACTIONS_NR 128
604 struct btree_transaction_stats {
605 struct bch2_time_stats lock_hold_times;
607 unsigned nr_max_paths;
608 unsigned wb_updates_size;
610 char *max_paths_text;
614 u64 sectors_available;
617 struct journal_seq_blacklist_table {
619 struct journal_seq_blacklist_table_entry {
626 struct journal_keys {
630 enum btree_id btree_id:8;
637 * Gap buffer: instead of all the empty space in the array being at the
638 * end of the buffer - from @nr to @size - the empty space is at @gap.
639 * This means that sequential insertions are O(n) instead of O(n^2).
645 bool initial_ref_held;
648 struct btree_trans_buf {
649 struct btree_trans *trans;
652 #define REPLICAS_DELTA_LIST_MAX (1U << 16)
654 #define BCACHEFS_ROOT_SUBVOL_INUM \
655 ((subvol_inum) { BCACHEFS_ROOT_SUBVOL, BCACHEFS_ROOT_INO })
657 #define BCH_WRITE_REFS() \
668 x(delete_dead_snapshots) \
669 x(snapshot_delete_pagecache) \
671 x(btree_write_buffer)
674 #define x(n) BCH_WRITE_REF_##n,
683 struct list_head list;
685 struct kobject counters_kobj;
686 struct kobject internal;
687 struct kobject opts_dir;
688 struct kobject time_stats;
692 struct device *chardev;
693 struct super_block *vfs_sb;
697 /* ro/rw, add/remove/resize devices: */
698 struct rw_semaphore state_lock;
700 /* Counts outstanding writes, for clean transition to read-only */
701 #ifdef BCH_WRITE_REF_DEBUG
702 atomic_long_t writes[BCH_WRITE_REF_NR];
704 struct percpu_ref writes;
706 struct work_struct read_only_work;
708 struct bch_dev __rcu *devs[BCH_SB_MEMBERS_MAX];
710 struct bch_replicas_cpu replicas;
711 struct bch_replicas_cpu replicas_gc;
712 struct mutex replicas_gc_lock;
713 mempool_t replicas_delta_pool;
715 struct journal_entry_res btree_root_journal_res;
716 struct journal_entry_res replicas_journal_res;
717 struct journal_entry_res clock_journal_res;
718 struct journal_entry_res dev_usage_journal_res;
720 struct bch_disk_groups_cpu __rcu *disk_groups;
722 struct bch_opts opts;
724 /* Updated by bch2_sb_update():*/
731 u16 version_upgrade_complete;
740 unsigned time_units_per_sec;
741 unsigned nsec_per_time_unit;
747 struct bch_sb_handle disk_sb;
749 unsigned short block_bits; /* ilog2(block_size) */
751 u16 btree_foreground_merge_threshold;
753 struct closure sb_write;
754 struct mutex sb_lock;
757 struct snapshot_table __rcu *snapshots;
758 size_t snapshot_table_size;
759 struct mutex snapshot_table_lock;
760 struct rw_semaphore snapshot_create_lock;
762 struct work_struct snapshot_delete_work;
763 struct work_struct snapshot_wait_for_pagecache_and_delete_work;
764 snapshot_id_list snapshots_unlinked;
765 struct mutex snapshots_unlinked_lock;
768 struct bio_set btree_bio;
769 struct workqueue_struct *io_complete_wq;
771 struct btree_root btree_roots_known[BTREE_ID_NR];
772 DARRAY(struct btree_root) btree_roots_extra;
773 struct mutex btree_root_lock;
775 struct btree_cache btree_cache;
778 * Cache of allocated btree nodes - if we allocate a btree node and
779 * don't use it, if we free it that space can't be reused until going
780 * _all_ the way through the allocator (which exposes us to a livelock
781 * when allocating btree reserves fail halfway through) - instead, we
782 * can stick them here:
784 struct btree_alloc btree_reserve_cache[BTREE_NODE_RESERVE * 2];
785 unsigned btree_reserve_cache_nr;
786 struct mutex btree_reserve_cache_lock;
788 mempool_t btree_interior_update_pool;
789 struct list_head btree_interior_update_list;
790 struct list_head btree_interior_updates_unwritten;
791 struct mutex btree_interior_update_lock;
792 struct closure_waitlist btree_interior_update_wait;
794 struct workqueue_struct *btree_interior_update_worker;
795 struct work_struct btree_interior_update_work;
797 struct list_head pending_node_rewrites;
798 struct mutex pending_node_rewrites_lock;
801 spinlock_t btree_write_error_lock;
802 struct btree_write_stats {
805 } btree_write_stats[BTREE_WRITE_TYPE_NR];
808 struct seqmutex btree_trans_lock;
809 struct list_head btree_trans_list;
810 mempool_t btree_trans_pool;
811 mempool_t btree_trans_mem_pool;
812 struct btree_trans_buf __percpu *btree_trans_bufs;
814 struct srcu_struct btree_trans_barrier;
815 bool btree_trans_barrier_initialized;
817 struct btree_key_cache btree_key_cache;
818 unsigned btree_key_cache_btrees;
820 struct btree_write_buffer btree_write_buffer;
822 struct workqueue_struct *btree_update_wq;
823 struct workqueue_struct *btree_io_complete_wq;
824 /* copygc needs its own workqueue for index updates.. */
825 struct workqueue_struct *copygc_wq;
827 * Use a dedicated wq for write ref holder tasks. Required to avoid
828 * dependency problems with other wq tasks that can block on ref
829 * draining, such as read-only transition.
831 struct workqueue_struct *write_ref_wq;
834 struct bch_devs_mask rw_devs[BCH_DATA_NR];
836 u64 capacity; /* sectors */
839 * When capacity _decreases_ (due to a disk being removed), we
840 * increment capacity_gen - this invalidates outstanding reservations
841 * and forces them to be revalidated
844 unsigned bucket_size_max;
846 atomic64_t sectors_available;
847 struct mutex sectors_available_lock;
849 struct bch_fs_pcpu __percpu *pcpu;
851 struct percpu_rw_semaphore mark_lock;
853 seqcount_t usage_lock;
854 struct bch_fs_usage *usage_base;
855 struct bch_fs_usage __percpu *usage[JOURNAL_BUF_NR];
856 struct bch_fs_usage __percpu *usage_gc;
857 u64 __percpu *online_reserved;
859 /* single element mempool: */
860 struct mutex usage_scratch_lock;
861 struct bch_fs_usage_online *usage_scratch;
863 struct io_clock io_clock[2];
865 /* JOURNAL SEQ BLACKLIST */
866 struct journal_seq_blacklist_table *
867 journal_seq_blacklist_table;
868 struct work_struct journal_seq_blacklist_gc_work;
871 spinlock_t freelist_lock;
872 struct closure_waitlist freelist_wait;
873 u64 blocked_allocate;
874 u64 blocked_allocate_open_bucket;
876 open_bucket_idx_t open_buckets_freelist;
877 open_bucket_idx_t open_buckets_nr_free;
878 struct closure_waitlist open_buckets_wait;
879 struct open_bucket open_buckets[OPEN_BUCKETS_COUNT];
880 open_bucket_idx_t open_buckets_hash[OPEN_BUCKETS_COUNT];
882 open_bucket_idx_t open_buckets_partial[OPEN_BUCKETS_COUNT];
883 open_bucket_idx_t open_buckets_partial_nr;
885 struct write_point btree_write_point;
886 struct write_point rebalance_write_point;
888 struct write_point write_points[WRITE_POINT_MAX];
889 struct hlist_head write_points_hash[WRITE_POINT_HASH_NR];
890 struct mutex write_points_hash_lock;
891 unsigned write_points_nr;
893 struct buckets_waiting_for_journal buckets_waiting_for_journal;
894 struct work_struct discard_work;
895 struct work_struct invalidate_work;
897 /* GARBAGE COLLECTION */
898 struct task_struct *gc_thread;
900 unsigned long gc_count;
902 enum btree_id gc_gens_btree;
903 struct bpos gc_gens_pos;
906 * Tracks GC's progress - everything in the range [ZERO_KEY..gc_cur_pos]
907 * has been marked by GC.
909 * gc_cur_phase is a superset of btree_ids (BTREE_ID_extents etc.)
911 * Protected by gc_pos_lock. Only written to by GC thread, so GC thread
912 * can read without a lock.
914 seqcount_t gc_pos_lock;
915 struct gc_pos gc_pos;
918 * The allocation code needs gc_mark in struct bucket to be correct, but
919 * it's not while a gc is in progress.
921 struct rw_semaphore gc_lock;
922 struct mutex gc_gens_lock;
925 struct semaphore io_in_flight;
926 struct bio_set bio_read;
927 struct bio_set bio_read_split;
928 struct bio_set bio_write;
929 struct mutex bio_bounce_pages_lock;
930 mempool_t bio_bounce_pages;
931 struct bucket_nocow_lock_table
933 struct rhashtable promote_table;
935 mempool_t compression_bounce[2];
936 mempool_t compress_workspace[BCH_COMPRESSION_TYPE_NR];
937 mempool_t decompress_workspace;
938 size_t zstd_workspace_size;
940 struct crypto_shash *sha256;
941 struct crypto_sync_skcipher *chacha20;
942 struct crypto_shash *poly1305;
944 atomic64_t key_version;
946 mempool_t large_bkey_pool;
949 struct list_head moving_context_list;
950 struct mutex moving_context_lock;
953 struct bch_fs_rebalance rebalance;
956 struct task_struct *copygc_thread;
957 struct write_point copygc_write_point;
961 wait_queue_head_t copygc_running_wq;
964 GENRADIX(struct stripe) stripes;
965 GENRADIX(struct gc_stripe) gc_stripes;
967 struct hlist_head ec_stripes_new[32];
968 spinlock_t ec_stripes_new_lock;
970 ec_stripes_heap ec_stripes_heap;
971 struct mutex ec_stripes_heap_lock;
974 struct list_head ec_stripe_head_list;
975 struct mutex ec_stripe_head_lock;
977 struct list_head ec_stripe_new_list;
978 struct mutex ec_stripe_new_lock;
979 wait_queue_head_t ec_stripe_new_wait;
981 struct work_struct ec_stripe_create_work;
984 struct work_struct ec_stripe_delete_work;
986 struct bio_set ec_bioset;
989 reflink_gc_table reflink_gc_table;
990 size_t reflink_gc_nr;
993 struct list_head vfs_inodes_list;
994 struct mutex vfs_inodes_lock;
996 /* VFS IO PATH - fs-io.c */
997 struct bio_set writepage_bioset;
998 struct bio_set dio_write_bioset;
999 struct bio_set dio_read_bioset;
1000 struct bio_set nocow_flush_bioset;
1003 struct bch_memquota_type quotas[QTYP_NR];
1006 u64 journal_replay_seq_start;
1007 u64 journal_replay_seq_end;
1008 enum bch_recovery_pass curr_recovery_pass;
1009 /* bitmap of explicitly enabled recovery passes: */
1010 u64 recovery_passes_explicit;
1011 u64 recovery_passes_complete;
1014 struct dentry *fs_debug_dir;
1015 struct dentry *btree_debug_dir;
1016 struct btree_debug btree_debug[BTREE_ID_NR];
1017 struct btree *verify_data;
1018 struct btree_node *verify_ondisk;
1019 struct mutex verify_lock;
1021 u64 *unused_inode_hints;
1022 unsigned inode_shard_bits;
1025 * A btree node on disk could have too many bsets for an iterator to fit
1026 * on the stack - have to dynamically allocate them
1028 mempool_t fill_iter;
1030 mempool_t btree_bounce_pool;
1032 struct journal journal;
1033 GENRADIX(struct journal_replay *) journal_entries;
1034 u64 journal_entries_base_seq;
1035 struct journal_keys journal_keys;
1036 struct list_head journal_iters;
1038 u64 last_bucket_seq_cleanup;
1040 u64 counters_on_mount[BCH_COUNTER_NR];
1041 u64 __percpu *counters;
1043 unsigned btree_gc_periodic:1;
1044 unsigned copy_gc_enabled:1;
1045 bool promote_whole_extents;
1047 struct bch2_time_stats times[BCH_TIME_STAT_NR];
1049 struct btree_transaction_stats btree_transaction_stats[BCH_TRANSACTIONS_NR];
1052 struct list_head fsck_error_msgs;
1053 struct mutex fsck_error_msgs_lock;
1054 bool fsck_alloc_msgs_err;
1056 bch_sb_errors_cpu fsck_error_counts;
1057 struct mutex fsck_error_counts_lock;
1060 extern struct wait_queue_head bch2_read_only_wait;
1062 static inline void bch2_write_ref_get(struct bch_fs *c, enum bch_write_ref ref)
1064 #ifdef BCH_WRITE_REF_DEBUG
1065 atomic_long_inc(&c->writes[ref]);
1067 percpu_ref_get(&c->writes);
1071 static inline bool __bch2_write_ref_tryget(struct bch_fs *c, enum bch_write_ref ref)
1073 #ifdef BCH_WRITE_REF_DEBUG
1074 return !test_bit(BCH_FS_GOING_RO, &c->flags) &&
1075 atomic_long_inc_not_zero(&c->writes[ref]);
1077 return percpu_ref_tryget(&c->writes);
1081 static inline bool bch2_write_ref_tryget(struct bch_fs *c, enum bch_write_ref ref)
1083 #ifdef BCH_WRITE_REF_DEBUG
1084 return !test_bit(BCH_FS_GOING_RO, &c->flags) &&
1085 atomic_long_inc_not_zero(&c->writes[ref]);
1087 return percpu_ref_tryget_live(&c->writes);
1091 static inline void bch2_write_ref_put(struct bch_fs *c, enum bch_write_ref ref)
1093 #ifdef BCH_WRITE_REF_DEBUG
1094 long v = atomic_long_dec_return(&c->writes[ref]);
1099 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
1100 if (atomic_long_read(&c->writes[i]))
1103 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
1104 wake_up(&bch2_read_only_wait);
1106 percpu_ref_put(&c->writes);
1110 static inline void bch2_set_ra_pages(struct bch_fs *c, unsigned ra_pages)
1112 #ifndef NO_BCACHEFS_FS
1114 c->vfs_sb->s_bdi->ra_pages = ra_pages;
1118 static inline unsigned bucket_bytes(const struct bch_dev *ca)
1120 return ca->mi.bucket_size << 9;
1123 static inline unsigned block_bytes(const struct bch_fs *c)
1125 return c->opts.block_size;
1128 static inline unsigned block_sectors(const struct bch_fs *c)
1130 return c->opts.block_size >> 9;
1133 static inline size_t btree_sectors(const struct bch_fs *c)
1135 return c->opts.btree_node_size >> 9;
1138 static inline bool btree_id_cached(const struct bch_fs *c, enum btree_id btree)
1140 return c->btree_key_cache_btrees & (1U << btree);
1143 static inline struct timespec64 bch2_time_to_timespec(const struct bch_fs *c, s64 time)
1145 struct timespec64 t;
1148 time += c->sb.time_base_lo;
1150 t.tv_sec = div_s64_rem(time, c->sb.time_units_per_sec, &rem);
1151 t.tv_nsec = rem * c->sb.nsec_per_time_unit;
1155 static inline s64 timespec_to_bch2_time(const struct bch_fs *c, struct timespec64 ts)
1157 return (ts.tv_sec * c->sb.time_units_per_sec +
1158 (int) ts.tv_nsec / c->sb.nsec_per_time_unit) - c->sb.time_base_lo;
1161 static inline s64 bch2_current_time(const struct bch_fs *c)
1163 struct timespec64 now;
1165 ktime_get_coarse_real_ts64(&now);
1166 return timespec_to_bch2_time(c, now);
1169 static inline bool bch2_dev_exists2(const struct bch_fs *c, unsigned dev)
1171 return dev < c->sb.nr_devices && c->devs[dev];
1174 #define BKEY_PADDED_ONSTACK(key, pad) \
1175 struct { struct bkey_i key; __u64 key ## _pad[pad]; }
1177 #endif /* _BCACHEFS_H */