1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef _BCACHEFS_H
3#define _BCACHEFS_H
4
5/*
6 * SOME HIGH LEVEL CODE DOCUMENTATION:
7 *
8 * Bcache mostly works with cache sets, cache devices, and backing devices.
9 *
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.
14 *
15 * Multiple cache devices is intended to give us the ability to mirror dirty
16 * cached data and metadata, without mirroring clean cached data.
17 *
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.
23 *
24 * A cache set can have multiple (many) backing devices attached to it.
25 *
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.
31 *
32 * Flash only volumes work but they're not production ready because the moving
33 * garbage collector needs more work. More on that later.
34 *
35 * BUCKETS/ALLOCATION:
36 *
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
40 * it.
41 *
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+
44 * works efficiently.
45 *
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).
50 *
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.
55 *
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
60 * this up).
61 *
62 * Bcache is entirely COW - we never write twice to a bucket, even buckets that
63 * contain metadata (including btree nodes).
64 *
65 * THE BTREE:
66 *
67 * Bcache is in large part design around the btree.
68 *
69 * At a high level, the btree is just an index of key -> ptr tuples.
70 *
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).
74 *
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.
79 *
80 * Pointers contain the cache device id, the offset on that device, and an 8 bit
81 * generation number. More on the gen later.
82 *
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
85 * direction.
86 *
87 * Updates are fairly well abstracted, though. There are two different ways of
88 * updating the btree; insert and replace.
89 *
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.
93 *
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.
98 *
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.
103 *
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.
107 *
108 * BTREE NODES:
109 *
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.
112 *
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.)
117 *
118 * Anyways, btree nodes are big - big enough to be inefficient with a textbook
119 * btree implementation.
120 *
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.
124 *
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
131 * smaller).
132 *
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.
136 *
137 * GARBAGE COLLECTION:
138 *
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.
142 *
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.
146 *
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.
151 *
152 * THE JOURNAL:
153 *
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
158 * implemented.
159 *
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.
167 *
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
171 * writing them out.
172 *
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()).
177 */
178
179#undef pr_fmt
180#ifdef __KERNEL__
181#define pr_fmt(fmt) "bcachefs: %s() " fmt "\n", __func__
182#else
183#define pr_fmt(fmt) "%s() " fmt "\n", __func__
184#endif
185
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>
206
207#include "bcachefs_format.h"
208#include "errcode.h"
209#include "fifo.h"
210#include "nocow_locking_types.h"
211#include "opts.h"
212#include "recovery_passes_types.h"
213#include "sb-errors_types.h"
214#include "seqmutex.h"
215#include "time_stats.h"
216#include "util.h"
217
218#ifdef CONFIG_BCACHEFS_DEBUG
219#define BCH_WRITE_REF_DEBUG
220#endif
221
222#ifndef dynamic_fault
223#define dynamic_fault(...) 0
224#endif
225
226#define race_fault(...) dynamic_fault("bcachefs:race")
227
228#define count_event(_c, _name) this_cpu_inc((_c)->counters[BCH_COUNTER_##_name])
229
230#define trace_and_count(_c, _name, ...) \
231do { \
232 count_event(_c, _name); \
233 trace_##_name(__VA_ARGS__); \
234} while (0)
235
236#define bch2_fs_init_fault(name) \
237 dynamic_fault("bcachefs:bch_fs_init:" name)
238#define bch2_meta_read_fault(name) \
239 dynamic_fault("bcachefs:meta:read:" name)
240#define bch2_meta_write_fault(name) \
241 dynamic_fault("bcachefs:meta:write:" name)
242
243#ifdef __KERNEL__
244#define BCACHEFS_LOG_PREFIX
245#endif
246
247#ifdef BCACHEFS_LOG_PREFIX
248
249#define bch2_log_msg(_c, fmt) "bcachefs (%s): " fmt, ((_c)->name)
250#define bch2_fmt_dev(_ca, fmt) "bcachefs (%s): " fmt "\n", ((_ca)->name)
251#define bch2_fmt_dev_offset(_ca, _offset, fmt) "bcachefs (%s sector %llu): " fmt "\n", ((_ca)->name), (_offset)
252#define bch2_fmt_inum(_c, _inum, fmt) "bcachefs (%s inum %llu): " fmt "\n", ((_c)->name), (_inum)
253#define bch2_fmt_inum_offset(_c, _inum, _offset, fmt) \
254 "bcachefs (%s inum %llu offset %llu): " fmt "\n", ((_c)->name), (_inum), (_offset)
255
256#else
257
258#define bch2_log_msg(_c, fmt) fmt
259#define bch2_fmt_dev(_ca, fmt) "%s: " fmt "\n", ((_ca)->name)
260#define bch2_fmt_dev_offset(_ca, _offset, fmt) "%s sector %llu: " fmt "\n", ((_ca)->name), (_offset)
261#define bch2_fmt_inum(_c, _inum, fmt) "inum %llu: " fmt "\n", (_inum)
262#define bch2_fmt_inum_offset(_c, _inum, _offset, fmt) \
263 "inum %llu offset %llu: " fmt "\n", (_inum), (_offset)
264
265#endif
266
267#define bch2_fmt(_c, fmt) bch2_log_msg(_c, fmt "\n")
268
269__printf(2, 3)
270void bch2_print_opts(struct bch_opts *, const char *, ...);
271
272__printf(2, 3)
273void __bch2_print(struct bch_fs *c, const char *fmt, ...);
274
275#define maybe_dev_to_fs(_c) _Generic((_c), \
276 struct bch_dev *: ((struct bch_dev *) (_c))->fs, \
277 struct bch_fs *: (_c))
278
279#define bch2_print(_c, ...) __bch2_print(maybe_dev_to_fs(_c), __VA_ARGS__)
280
281#define bch2_print_ratelimited(_c, ...) \
282do { \
283 static DEFINE_RATELIMIT_STATE(_rs, \
284 DEFAULT_RATELIMIT_INTERVAL, \
285 DEFAULT_RATELIMIT_BURST); \
286 \
287 if (__ratelimit(&_rs)) \
288 bch2_print(_c, __VA_ARGS__); \
289} while (0)
290
291#define bch_info(c, fmt, ...) \
292 bch2_print(c, KERN_INFO bch2_fmt(c, fmt), ##__VA_ARGS__)
293#define bch_notice(c, fmt, ...) \
294 bch2_print(c, KERN_NOTICE bch2_fmt(c, fmt), ##__VA_ARGS__)
295#define bch_warn(c, fmt, ...) \
296 bch2_print(c, KERN_WARNING bch2_fmt(c, fmt), ##__VA_ARGS__)
297#define bch_warn_ratelimited(c, fmt, ...) \
298 bch2_print_ratelimited(c, KERN_WARNING bch2_fmt(c, fmt), ##__VA_ARGS__)
299
300#define bch_err(c, fmt, ...) \
301 bch2_print(c, KERN_ERR bch2_fmt(c, fmt), ##__VA_ARGS__)
302#define bch_err_dev(ca, fmt, ...) \
303 bch2_print(c, KERN_ERR bch2_fmt_dev(ca, fmt), ##__VA_ARGS__)
304#define bch_err_dev_offset(ca, _offset, fmt, ...) \
305 bch2_print(c, KERN_ERR bch2_fmt_dev_offset(ca, _offset, fmt), ##__VA_ARGS__)
306#define bch_err_inum(c, _inum, fmt, ...) \
307 bch2_print(c, KERN_ERR bch2_fmt_inum(c, _inum, fmt), ##__VA_ARGS__)
308#define bch_err_inum_offset(c, _inum, _offset, fmt, ...) \
309 bch2_print(c, KERN_ERR bch2_fmt_inum_offset(c, _inum, _offset, fmt), ##__VA_ARGS__)
310
311#define bch_err_ratelimited(c, fmt, ...) \
312 bch2_print_ratelimited(c, KERN_ERR bch2_fmt(c, fmt), ##__VA_ARGS__)
313#define bch_err_dev_ratelimited(ca, fmt, ...) \
314 bch2_print_ratelimited(ca, KERN_ERR bch2_fmt_dev(ca, fmt), ##__VA_ARGS__)
315#define bch_err_dev_offset_ratelimited(ca, _offset, fmt, ...) \
316 bch2_print_ratelimited(ca, KERN_ERR bch2_fmt_dev_offset(ca, _offset, fmt), ##__VA_ARGS__)
317#define bch_err_inum_ratelimited(c, _inum, fmt, ...) \
318 bch2_print_ratelimited(c, KERN_ERR bch2_fmt_inum(c, _inum, fmt), ##__VA_ARGS__)
319#define bch_err_inum_offset_ratelimited(c, _inum, _offset, fmt, ...) \
320 bch2_print_ratelimited(c, KERN_ERR bch2_fmt_inum_offset(c, _inum, _offset, fmt), ##__VA_ARGS__)
321
322static inline bool should_print_err(int err)
323{
324 return err && !bch2_err_matches(err, BCH_ERR_transaction_restart);
325}
326
327#define bch_err_fn(_c, _ret) \
328do { \
329 if (should_print_err(_ret)) \
330 bch_err(_c, "%s(): error %s", __func__, bch2_err_str(_ret));\
331} while (0)
332
333#define bch_err_fn_ratelimited(_c, _ret) \
334do { \
335 if (should_print_err(_ret)) \
336 bch_err_ratelimited(_c, "%s(): error %s", __func__, bch2_err_str(_ret));\
337} while (0)
338
339#define bch_err_msg(_c, _ret, _msg, ...) \
340do { \
341 if (should_print_err(_ret)) \
342 bch_err(_c, "%s(): error " _msg " %s", __func__, \
343 ##__VA_ARGS__, bch2_err_str(_ret)); \
344} while (0)
345
346#define bch_verbose(c, fmt, ...) \
347do { \
348 if ((c)->opts.verbose) \
349 bch_info(c, fmt, ##__VA_ARGS__); \
350} while (0)
351
352#define pr_verbose_init(opts, fmt, ...) \
353do { \
354 if (opt_get(opts, verbose)) \
355 pr_info(fmt, ##__VA_ARGS__); \
356} while (0)
357
358/* Parameters that are useful for debugging, but should always be compiled in: */
359#define BCH_DEBUG_PARAMS_ALWAYS() \
360 BCH_DEBUG_PARAM(key_merging_disabled, \
361 "Disables merging of extents") \
362 BCH_DEBUG_PARAM(btree_gc_always_rewrite, \
363 "Causes mark and sweep to compact and rewrite every " \
364 "btree node it traverses") \
365 BCH_DEBUG_PARAM(btree_gc_rewrite_disabled, \
366 "Disables rewriting of btree nodes during mark and sweep")\
367 BCH_DEBUG_PARAM(btree_shrinker_disabled, \
368 "Disables the shrinker callback for the btree node cache")\
369 BCH_DEBUG_PARAM(verify_btree_ondisk, \
370 "Reread btree nodes at various points to verify the " \
371 "mergesort in the read path against modifications " \
372 "done in memory") \
373 BCH_DEBUG_PARAM(verify_all_btree_replicas, \
374 "When reading btree nodes, read all replicas and " \
375 "compare them") \
376 BCH_DEBUG_PARAM(backpointers_no_use_write_buffer, \
377 "Don't use the write buffer for backpointers, enabling "\
378 "extra runtime checks")
379
380/* Parameters that should only be compiled in debug mode: */
381#define BCH_DEBUG_PARAMS_DEBUG() \
382 BCH_DEBUG_PARAM(expensive_debug_checks, \
383 "Enables various runtime debugging checks that " \
384 "significantly affect performance") \
385 BCH_DEBUG_PARAM(debug_check_iterators, \
386 "Enables extra verification for btree iterators") \
387 BCH_DEBUG_PARAM(debug_check_btree_accounting, \
388 "Verify btree accounting for keys within a node") \
389 BCH_DEBUG_PARAM(journal_seq_verify, \
390 "Store the journal sequence number in the version " \
391 "number of every btree key, and verify that btree " \
392 "update ordering is preserved during recovery") \
393 BCH_DEBUG_PARAM(inject_invalid_keys, \
394 "Store the journal sequence number in the version " \
395 "number of every btree key, and verify that btree " \
396 "update ordering is preserved during recovery") \
397 BCH_DEBUG_PARAM(test_alloc_startup, \
398 "Force allocator startup to use the slowpath where it" \
399 "can't find enough free buckets without invalidating" \
400 "cached data") \
401 BCH_DEBUG_PARAM(force_reconstruct_read, \
402 "Force reads to use the reconstruct path, when reading" \
403 "from erasure coded extents") \
404 BCH_DEBUG_PARAM(test_restart_gc, \
405 "Test restarting mark and sweep gc when bucket gens change")
406
407#define BCH_DEBUG_PARAMS_ALL() BCH_DEBUG_PARAMS_ALWAYS() BCH_DEBUG_PARAMS_DEBUG()
408
409#ifdef CONFIG_BCACHEFS_DEBUG
410#define BCH_DEBUG_PARAMS() BCH_DEBUG_PARAMS_ALL()
411#else
412#define BCH_DEBUG_PARAMS() BCH_DEBUG_PARAMS_ALWAYS()
413#endif
414
415#define BCH_DEBUG_PARAM(name, description) extern bool bch2_##name;
416BCH_DEBUG_PARAMS()
417#undef BCH_DEBUG_PARAM
418
419#ifndef CONFIG_BCACHEFS_DEBUG
420#define BCH_DEBUG_PARAM(name, description) static const __maybe_unused bool bch2_##name;
421BCH_DEBUG_PARAMS_DEBUG()
422#undef BCH_DEBUG_PARAM
423#endif
424
425#define BCH_TIME_STATS() \
426 x(btree_node_mem_alloc) \
427 x(btree_node_split) \
428 x(btree_node_compact) \
429 x(btree_node_merge) \
430 x(btree_node_sort) \
431 x(btree_node_read) \
432 x(btree_node_read_done) \
433 x(btree_interior_update_foreground) \
434 x(btree_interior_update_total) \
435 x(btree_gc) \
436 x(data_write) \
437 x(data_read) \
438 x(data_promote) \
439 x(journal_flush_write) \
440 x(journal_noflush_write) \
441 x(journal_flush_seq) \
442 x(blocked_journal_low_on_space) \
443 x(blocked_journal_low_on_pin) \
444 x(blocked_journal_max_in_flight) \
445 x(blocked_allocate) \
446 x(blocked_allocate_open_bucket) \
447 x(blocked_write_buffer_full) \
448 x(nocow_lock_contended)
449
450enum bch_time_stats {
451#define x(name) BCH_TIME_##name,
452 BCH_TIME_STATS()
453#undef x
454 BCH_TIME_STAT_NR
455};
456
457#include "alloc_types.h"
458#include "btree_types.h"
459#include "btree_node_scan_types.h"
460#include "btree_write_buffer_types.h"
461#include "buckets_types.h"
462#include "buckets_waiting_for_journal_types.h"
463#include "clock_types.h"
464#include "disk_groups_types.h"
465#include "ec_types.h"
466#include "journal_types.h"
467#include "keylist_types.h"
468#include "quota_types.h"
469#include "rebalance_types.h"
470#include "replicas_types.h"
471#include "subvolume_types.h"
472#include "super_types.h"
473#include "thread_with_file_types.h"
474
475/* Number of nodes btree coalesce will try to coalesce at once */
476#define GC_MERGE_NODES 4U
477
478/* Maximum number of nodes we might need to allocate atomically: */
479#define BTREE_RESERVE_MAX (BTREE_MAX_DEPTH + (BTREE_MAX_DEPTH - 1))
480
481/* Size of the freelist we allocate btree nodes from: */
482#define BTREE_NODE_RESERVE (BTREE_RESERVE_MAX * 4)
483
484#define BTREE_NODE_OPEN_BUCKET_RESERVE (BTREE_RESERVE_MAX * BCH_REPLICAS_MAX)
485
486struct btree;
487
488enum gc_phase {
489 GC_PHASE_NOT_RUNNING,
490 GC_PHASE_START,
491 GC_PHASE_SB,
492
493 GC_PHASE_BTREE_stripes,
494 GC_PHASE_BTREE_extents,
495 GC_PHASE_BTREE_inodes,
496 GC_PHASE_BTREE_dirents,
497 GC_PHASE_BTREE_xattrs,
498 GC_PHASE_BTREE_alloc,
499 GC_PHASE_BTREE_quotas,
500 GC_PHASE_BTREE_reflink,
501 GC_PHASE_BTREE_subvolumes,
502 GC_PHASE_BTREE_snapshots,
503 GC_PHASE_BTREE_lru,
504 GC_PHASE_BTREE_freespace,
505 GC_PHASE_BTREE_need_discard,
506 GC_PHASE_BTREE_backpointers,
507 GC_PHASE_BTREE_bucket_gens,
508 GC_PHASE_BTREE_snapshot_trees,
509 GC_PHASE_BTREE_deleted_inodes,
510 GC_PHASE_BTREE_logged_ops,
511 GC_PHASE_BTREE_rebalance_work,
512 GC_PHASE_BTREE_subvolume_children,
513
514 GC_PHASE_PENDING_DELETE,
515};
516
517struct gc_pos {
518 enum gc_phase phase;
519 struct bpos pos;
520 unsigned level;
521};
522
523struct reflink_gc {
524 u64 offset;
525 u32 size;
526 u32 refcount;
527};
528
529typedef GENRADIX(struct reflink_gc) reflink_gc_table;
530
531struct io_count {
532 u64 sectors[2][BCH_DATA_NR];
533};
534
535struct bch_dev {
536 struct kobject kobj;
537 struct percpu_ref ref;
538 struct completion ref_completion;
539 struct percpu_ref io_ref;
540 struct completion io_ref_completion;
541
542 struct bch_fs *fs;
543
544 u8 dev_idx;
545 /*
546 * Cached version of this device's member info from superblock
547 * Committed by bch2_write_super() -> bch_fs_mi_update()
548 */
549 struct bch_member_cpu mi;
550 atomic64_t errors[BCH_MEMBER_ERROR_NR];
551
552 __uuid_t uuid;
553 char name[BDEVNAME_SIZE];
554
555 struct bch_sb_handle disk_sb;
556 struct bch_sb *sb_read_scratch;
557 int sb_write_error;
558 dev_t dev;
559 atomic_t flush_seq;
560
561 struct bch_devs_mask self;
562
563 /* biosets used in cloned bios for writing multiple replicas */
564 struct bio_set replica_set;
565
566 /*
567 * Buckets:
568 * Per-bucket arrays are protected by c->mark_lock, bucket_lock and
569 * gc_lock, for device resize - holding any is sufficient for access:
570 * Or rcu_read_lock(), but only for ptr_stale():
571 */
572 struct bucket_array __rcu *buckets_gc;
573 struct bucket_gens __rcu *bucket_gens;
574 u8 *oldest_gen;
575 unsigned long *buckets_nouse;
576 struct rw_semaphore bucket_lock;
577
578 struct bch_dev_usage *usage_base;
579 struct bch_dev_usage __percpu *usage[JOURNAL_BUF_NR];
580 struct bch_dev_usage __percpu *usage_gc;
581
582 /* Allocator: */
583 u64 new_fs_bucket_idx;
584 u64 alloc_cursor;
585
586 unsigned nr_open_buckets;
587 unsigned nr_btree_reserve;
588
589 size_t inc_gen_needs_gc;
590 size_t inc_gen_really_needs_gc;
591 size_t buckets_waiting_on_journal;
592
593 atomic64_t rebalance_work;
594
595 struct journal_device journal;
596 u64 prev_journal_sector;
597
598 struct work_struct io_error_work;
599
600 /* The rest of this all shows up in sysfs */
601 atomic64_t cur_latency[2];
602 struct bch2_time_stats_quantiles io_latency[2];
603
604#define CONGESTED_MAX 1024
605 atomic_t congested;
606 u64 congested_last;
607
608 struct io_count __percpu *io_done;
609};
610
611/*
612 * initial_gc_unfixed
613 * error
614 * topology error
615 */
616
617#define BCH_FS_FLAGS() \
618 x(new_fs) \
619 x(started) \
620 x(may_go_rw) \
621 x(rw) \
622 x(was_rw) \
623 x(stopping) \
624 x(emergency_ro) \
625 x(going_ro) \
626 x(write_disable_complete) \
627 x(clean_shutdown) \
628 x(fsck_running) \
629 x(initial_gc_unfixed) \
630 x(need_another_gc) \
631 x(need_delete_dead_snapshots) \
632 x(error) \
633 x(topology_error) \
634 x(errors_fixed) \
635 x(errors_not_fixed)
636
637enum bch_fs_flags {
638#define x(n) BCH_FS_##n,
639 BCH_FS_FLAGS()
640#undef x
641};
642
643struct btree_debug {
644 unsigned id;
645};
646
647#define BCH_TRANSACTIONS_NR 128
648
649struct btree_transaction_stats {
650 struct bch2_time_stats duration;
651 struct bch2_time_stats lock_hold_times;
652 struct mutex lock;
653 unsigned nr_max_paths;
654 unsigned journal_entries_size;
655 unsigned max_mem;
656 char *max_paths_text;
657};
658
659struct bch_fs_pcpu {
660 u64 sectors_available;
661};
662
663struct journal_seq_blacklist_table {
664 size_t nr;
665 struct journal_seq_blacklist_table_entry {
666 u64 start;
667 u64 end;
668 bool dirty;
669 } entries[];
670};
671
672struct journal_keys {
673 /* must match layout in darray_types.h */
674 size_t nr, size;
675 struct journal_key {
676 u64 journal_seq;
677 u32 journal_offset;
678 enum btree_id btree_id:8;
679 unsigned level:8;
680 bool allocated;
681 bool overwritten;
682 struct bkey_i *k;
683 } *data;
684 /*
685 * Gap buffer: instead of all the empty space in the array being at the
686 * end of the buffer - from @nr to @size - the empty space is at @gap.
687 * This means that sequential insertions are O(n) instead of O(n^2).
688 */
689 size_t gap;
690 atomic_t ref;
691 bool initial_ref_held;
692};
693
694struct btree_trans_buf {
695 struct btree_trans *trans;
696};
697
698#define REPLICAS_DELTA_LIST_MAX (1U << 16)
699
700#define BCACHEFS_ROOT_SUBVOL_INUM \
701 ((subvol_inum) { BCACHEFS_ROOT_SUBVOL, BCACHEFS_ROOT_INO })
702
703#define BCH_WRITE_REFS() \
704 x(trans) \
705 x(write) \
706 x(promote) \
707 x(node_rewrite) \
708 x(stripe_create) \
709 x(stripe_delete) \
710 x(reflink) \
711 x(fallocate) \
712 x(fsync) \
713 x(dio_write) \
714 x(discard) \
715 x(discard_fast) \
716 x(invalidate) \
717 x(delete_dead_snapshots) \
718 x(snapshot_delete_pagecache) \
719 x(sysfs) \
720 x(btree_write_buffer)
721
722enum bch_write_ref {
723#define x(n) BCH_WRITE_REF_##n,
724 BCH_WRITE_REFS()
725#undef x
726 BCH_WRITE_REF_NR,
727};
728
729struct bch_fs {
730 struct closure cl;
731
732 struct list_head list;
733 struct kobject kobj;
734 struct kobject counters_kobj;
735 struct kobject internal;
736 struct kobject opts_dir;
737 struct kobject time_stats;
738 unsigned long flags;
739
740 int minor;
741 struct device *chardev;
742 struct super_block *vfs_sb;
743 dev_t dev;
744 char name[40];
745 struct stdio_redirect *stdio;
746 struct task_struct *stdio_filter;
747
748 /* ro/rw, add/remove/resize devices: */
749 struct rw_semaphore state_lock;
750
751 /* Counts outstanding writes, for clean transition to read-only */
752#ifdef BCH_WRITE_REF_DEBUG
753 atomic_long_t writes[BCH_WRITE_REF_NR];
754#else
755 struct percpu_ref writes;
756#endif
757 /*
758 * Analagous to c->writes, for asynchronous ops that don't necessarily
759 * need fs to be read-write
760 */
761 refcount_t ro_ref;
762 wait_queue_head_t ro_ref_wait;
763
764 struct work_struct read_only_work;
765
766 struct bch_dev __rcu *devs[BCH_SB_MEMBERS_MAX];
767
768 struct bch_replicas_cpu replicas;
769 struct bch_replicas_cpu replicas_gc;
770 struct mutex replicas_gc_lock;
771 mempool_t replicas_delta_pool;
772
773 struct journal_entry_res btree_root_journal_res;
774 struct journal_entry_res replicas_journal_res;
775 struct journal_entry_res clock_journal_res;
776 struct journal_entry_res dev_usage_journal_res;
777
778 struct bch_disk_groups_cpu __rcu *disk_groups;
779
780 struct bch_opts opts;
781
782 /* Updated by bch2_sb_update():*/
783 struct {
784 __uuid_t uuid;
785 __uuid_t user_uuid;
786
787 u16 version;
788 u16 version_min;
789 u16 version_upgrade_complete;
790
791 u8 nr_devices;
792 u8 clean;
793
794 u8 encryption_type;
795
796 u64 time_base_lo;
797 u32 time_base_hi;
798 unsigned time_units_per_sec;
799 unsigned nsec_per_time_unit;
800 u64 features;
801 u64 compat;
802 unsigned long errors_silent[BITS_TO_LONGS(BCH_SB_ERR_MAX)];
803 u64 btrees_lost_data;
804 } sb;
805
806
807 struct bch_sb_handle disk_sb;
808
809 unsigned short block_bits; /* ilog2(block_size) */
810
811 u16 btree_foreground_merge_threshold;
812
813 struct closure sb_write;
814 struct mutex sb_lock;
815
816 /* snapshot.c: */
817 struct snapshot_table __rcu *snapshots;
818 struct mutex snapshot_table_lock;
819 struct rw_semaphore snapshot_create_lock;
820
821 struct work_struct snapshot_delete_work;
822 struct work_struct snapshot_wait_for_pagecache_and_delete_work;
823 snapshot_id_list snapshots_unlinked;
824 struct mutex snapshots_unlinked_lock;
825
826 /* BTREE CACHE */
827 struct bio_set btree_bio;
828 struct workqueue_struct *io_complete_wq;
829
830 struct btree_root btree_roots_known[BTREE_ID_NR];
831 DARRAY(struct btree_root) btree_roots_extra;
832 struct mutex btree_root_lock;
833
834 struct btree_cache btree_cache;
835
836 /*
837 * Cache of allocated btree nodes - if we allocate a btree node and
838 * don't use it, if we free it that space can't be reused until going
839 * _all_ the way through the allocator (which exposes us to a livelock
840 * when allocating btree reserves fail halfway through) - instead, we
841 * can stick them here:
842 */
843 struct btree_alloc btree_reserve_cache[BTREE_NODE_RESERVE * 2];
844 unsigned btree_reserve_cache_nr;
845 struct mutex btree_reserve_cache_lock;
846
847 mempool_t btree_interior_update_pool;
848 struct list_head btree_interior_update_list;
849 struct list_head btree_interior_updates_unwritten;
850 struct mutex btree_interior_update_lock;
851 struct closure_waitlist btree_interior_update_wait;
852
853 struct workqueue_struct *btree_interior_update_worker;
854 struct work_struct btree_interior_update_work;
855
856 struct workqueue_struct *btree_node_rewrite_worker;
857
858 struct list_head pending_node_rewrites;
859 struct mutex pending_node_rewrites_lock;
860
861 /* btree_io.c: */
862 spinlock_t btree_write_error_lock;
863 struct btree_write_stats {
864 atomic64_t nr;
865 atomic64_t bytes;
866 } btree_write_stats[BTREE_WRITE_TYPE_NR];
867
868 /* btree_iter.c: */
869 struct seqmutex btree_trans_lock;
870 struct list_head btree_trans_list;
871 mempool_t btree_trans_pool;
872 mempool_t btree_trans_mem_pool;
873 struct btree_trans_buf __percpu *btree_trans_bufs;
874
875 struct srcu_struct btree_trans_barrier;
876 bool btree_trans_barrier_initialized;
877
878 struct btree_key_cache btree_key_cache;
879 unsigned btree_key_cache_btrees;
880
881 struct btree_write_buffer btree_write_buffer;
882
883 struct workqueue_struct *btree_update_wq;
884 struct workqueue_struct *btree_io_complete_wq;
885 /* copygc needs its own workqueue for index updates.. */
886 struct workqueue_struct *copygc_wq;
887 /*
888 * Use a dedicated wq for write ref holder tasks. Required to avoid
889 * dependency problems with other wq tasks that can block on ref
890 * draining, such as read-only transition.
891 */
892 struct workqueue_struct *write_ref_wq;
893
894 /* ALLOCATION */
895 struct bch_devs_mask rw_devs[BCH_DATA_NR];
896
897 u64 capacity; /* sectors */
898
899 /*
900 * When capacity _decreases_ (due to a disk being removed), we
901 * increment capacity_gen - this invalidates outstanding reservations
902 * and forces them to be revalidated
903 */
904 u32 capacity_gen;
905 unsigned bucket_size_max;
906
907 atomic64_t sectors_available;
908 struct mutex sectors_available_lock;
909
910 struct bch_fs_pcpu __percpu *pcpu;
911
912 struct percpu_rw_semaphore mark_lock;
913
914 seqcount_t usage_lock;
915 struct bch_fs_usage *usage_base;
916 struct bch_fs_usage __percpu *usage[JOURNAL_BUF_NR];
917 struct bch_fs_usage __percpu *usage_gc;
918 u64 __percpu *online_reserved;
919
920 /* single element mempool: */
921 struct mutex usage_scratch_lock;
922 struct bch_fs_usage_online *usage_scratch;
923
924 struct io_clock io_clock[2];
925
926 /* JOURNAL SEQ BLACKLIST */
927 struct journal_seq_blacklist_table *
928 journal_seq_blacklist_table;
929 struct work_struct journal_seq_blacklist_gc_work;
930
931 /* ALLOCATOR */
932 spinlock_t freelist_lock;
933 struct closure_waitlist freelist_wait;
934
935 open_bucket_idx_t open_buckets_freelist;
936 open_bucket_idx_t open_buckets_nr_free;
937 struct closure_waitlist open_buckets_wait;
938 struct open_bucket open_buckets[OPEN_BUCKETS_COUNT];
939 open_bucket_idx_t open_buckets_hash[OPEN_BUCKETS_COUNT];
940
941 open_bucket_idx_t open_buckets_partial[OPEN_BUCKETS_COUNT];
942 open_bucket_idx_t open_buckets_partial_nr;
943
944 struct write_point btree_write_point;
945 struct write_point rebalance_write_point;
946
947 struct write_point write_points[WRITE_POINT_MAX];
948 struct hlist_head write_points_hash[WRITE_POINT_HASH_NR];
949 struct mutex write_points_hash_lock;
950 unsigned write_points_nr;
951
952 struct buckets_waiting_for_journal buckets_waiting_for_journal;
953 struct work_struct invalidate_work;
954 struct work_struct discard_work;
955 struct mutex discard_buckets_in_flight_lock;
956 DARRAY(struct bpos) discard_buckets_in_flight;
957 struct work_struct discard_fast_work;
958
959 /* GARBAGE COLLECTION */
960 struct task_struct *gc_thread;
961 atomic_t kick_gc;
962 unsigned long gc_count;
963
964 enum btree_id gc_gens_btree;
965 struct bpos gc_gens_pos;
966
967 /*
968 * Tracks GC's progress - everything in the range [ZERO_KEY..gc_cur_pos]
969 * has been marked by GC.
970 *
971 * gc_cur_phase is a superset of btree_ids (BTREE_ID_extents etc.)
972 *
973 * Protected by gc_pos_lock. Only written to by GC thread, so GC thread
974 * can read without a lock.
975 */
976 seqcount_t gc_pos_lock;
977 struct gc_pos gc_pos;
978
979 /*
980 * The allocation code needs gc_mark in struct bucket to be correct, but
981 * it's not while a gc is in progress.
982 */
983 struct rw_semaphore gc_lock;
984 struct mutex gc_gens_lock;
985
986 /* IO PATH */
987 struct semaphore io_in_flight;
988 struct bio_set bio_read;
989 struct bio_set bio_read_split;
990 struct bio_set bio_write;
991 struct mutex bio_bounce_pages_lock;
992 mempool_t bio_bounce_pages;
993 struct bucket_nocow_lock_table
994 nocow_locks;
995 struct rhashtable promote_table;
996
997 mempool_t compression_bounce[2];
998 mempool_t compress_workspace[BCH_COMPRESSION_TYPE_NR];
999 mempool_t decompress_workspace;
1000 size_t zstd_workspace_size;
1001
1002 struct crypto_shash *sha256;
1003 struct crypto_sync_skcipher *chacha20;
1004 struct crypto_shash *poly1305;
1005
1006 atomic64_t key_version;
1007
1008 mempool_t large_bkey_pool;
1009
1010 /* MOVE.C */
1011 struct list_head moving_context_list;
1012 struct mutex moving_context_lock;
1013
1014 /* REBALANCE */
1015 struct bch_fs_rebalance rebalance;
1016
1017 /* COPYGC */
1018 struct task_struct *copygc_thread;
1019 struct write_point copygc_write_point;
1020 s64 copygc_wait_at;
1021 s64 copygc_wait;
1022 bool copygc_running;
1023 wait_queue_head_t copygc_running_wq;
1024
1025 /* STRIPES: */
1026 GENRADIX(struct stripe) stripes;
1027 GENRADIX(struct gc_stripe) gc_stripes;
1028
1029 struct hlist_head ec_stripes_new[32];
1030 spinlock_t ec_stripes_new_lock;
1031
1032 ec_stripes_heap ec_stripes_heap;
1033 struct mutex ec_stripes_heap_lock;
1034
1035 /* ERASURE CODING */
1036 struct list_head ec_stripe_head_list;
1037 struct mutex ec_stripe_head_lock;
1038
1039 struct list_head ec_stripe_new_list;
1040 struct mutex ec_stripe_new_lock;
1041 wait_queue_head_t ec_stripe_new_wait;
1042
1043 struct work_struct ec_stripe_create_work;
1044 u64 ec_stripe_hint;
1045
1046 struct work_struct ec_stripe_delete_work;
1047
1048 struct bio_set ec_bioset;
1049
1050 /* REFLINK */
1051 reflink_gc_table reflink_gc_table;
1052 size_t reflink_gc_nr;
1053
1054 /* fs.c */
1055 struct list_head vfs_inodes_list;
1056 struct mutex vfs_inodes_lock;
1057
1058 /* VFS IO PATH - fs-io.c */
1059 struct bio_set writepage_bioset;
1060 struct bio_set dio_write_bioset;
1061 struct bio_set dio_read_bioset;
1062 struct bio_set nocow_flush_bioset;
1063
1064 /* QUOTAS */
1065 struct bch_memquota_type quotas[QTYP_NR];
1066
1067 /* RECOVERY */
1068 u64 journal_replay_seq_start;
1069 u64 journal_replay_seq_end;
1070 /*
1071 * Two different uses:
1072 * "Has this fsck pass?" - i.e. should this type of error be an
1073 * emergency read-only
1074 * And, in certain situations fsck will rewind to an earlier pass: used
1075 * for signaling to the toplevel code which pass we want to run now.
1076 */
1077 enum bch_recovery_pass curr_recovery_pass;
1078 /* bitmap of explicitly enabled recovery passes: */
1079 u64 recovery_passes_explicit;
1080 /* bitmask of recovery passes that we actually ran */
1081 u64 recovery_passes_complete;
1082 /* never rewinds version of curr_recovery_pass */
1083 enum bch_recovery_pass recovery_pass_done;
1084 struct semaphore online_fsck_mutex;
1085
1086 /* DEBUG JUNK */
1087 struct dentry *fs_debug_dir;
1088 struct dentry *btree_debug_dir;
1089 struct btree_debug btree_debug[BTREE_ID_NR];
1090 struct btree *verify_data;
1091 struct btree_node *verify_ondisk;
1092 struct mutex verify_lock;
1093
1094 u64 *unused_inode_hints;
1095 unsigned inode_shard_bits;
1096
1097 /*
1098 * A btree node on disk could have too many bsets for an iterator to fit
1099 * on the stack - have to dynamically allocate them
1100 */
1101 mempool_t fill_iter;
1102
1103 mempool_t btree_bounce_pool;
1104
1105 struct journal journal;
1106 GENRADIX(struct journal_replay *) journal_entries;
1107 u64 journal_entries_base_seq;
1108 struct journal_keys journal_keys;
1109 struct list_head journal_iters;
1110
1111 struct find_btree_nodes found_btree_nodes;
1112
1113 u64 last_bucket_seq_cleanup;
1114
1115 u64 counters_on_mount[BCH_COUNTER_NR];
1116 u64 __percpu *counters;
1117
1118 unsigned btree_gc_periodic:1;
1119 unsigned copy_gc_enabled:1;
1120 bool promote_whole_extents;
1121
1122 struct bch2_time_stats times[BCH_TIME_STAT_NR];
1123
1124 struct btree_transaction_stats btree_transaction_stats[BCH_TRANSACTIONS_NR];
1125
1126 /* ERRORS */
1127 struct list_head fsck_error_msgs;
1128 struct mutex fsck_error_msgs_lock;
1129 bool fsck_alloc_msgs_err;
1130
1131 bch_sb_errors_cpu fsck_error_counts;
1132 struct mutex fsck_error_counts_lock;
1133};
1134
1135extern struct wait_queue_head bch2_read_only_wait;
1136
1137static inline void bch2_write_ref_get(struct bch_fs *c, enum bch_write_ref ref)
1138{
1139#ifdef BCH_WRITE_REF_DEBUG
1140 atomic_long_inc(v: &c->writes[ref]);
1141#else
1142 percpu_ref_get(&c->writes);
1143#endif
1144}
1145
1146static inline bool __bch2_write_ref_tryget(struct bch_fs *c, enum bch_write_ref ref)
1147{
1148#ifdef BCH_WRITE_REF_DEBUG
1149 return !test_bit(BCH_FS_going_ro, &c->flags) &&
1150 atomic_long_inc_not_zero(v: &c->writes[ref]);
1151#else
1152 return percpu_ref_tryget(&c->writes);
1153#endif
1154}
1155
1156static inline bool bch2_write_ref_tryget(struct bch_fs *c, enum bch_write_ref ref)
1157{
1158#ifdef BCH_WRITE_REF_DEBUG
1159 return !test_bit(BCH_FS_going_ro, &c->flags) &&
1160 atomic_long_inc_not_zero(v: &c->writes[ref]);
1161#else
1162 return percpu_ref_tryget_live(&c->writes);
1163#endif
1164}
1165
1166static inline void bch2_write_ref_put(struct bch_fs *c, enum bch_write_ref ref)
1167{
1168#ifdef BCH_WRITE_REF_DEBUG
1169 long v = atomic_long_dec_return(v: &c->writes[ref]);
1170
1171 BUG_ON(v < 0);
1172 if (v)
1173 return;
1174 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
1175 if (atomic_long_read(v: &c->writes[i]))
1176 return;
1177
1178 set_bit(nr: BCH_FS_write_disable_complete, addr: &c->flags);
1179 wake_up(&bch2_read_only_wait);
1180#else
1181 percpu_ref_put(&c->writes);
1182#endif
1183}
1184
1185static inline bool bch2_ro_ref_tryget(struct bch_fs *c)
1186{
1187 if (test_bit(BCH_FS_stopping, &c->flags))
1188 return false;
1189
1190 return refcount_inc_not_zero(r: &c->ro_ref);
1191}
1192
1193static inline void bch2_ro_ref_put(struct bch_fs *c)
1194{
1195 if (refcount_dec_and_test(r: &c->ro_ref))
1196 wake_up(&c->ro_ref_wait);
1197}
1198
1199static inline void bch2_set_ra_pages(struct bch_fs *c, unsigned ra_pages)
1200{
1201#ifndef NO_BCACHEFS_FS
1202 if (c->vfs_sb)
1203 c->vfs_sb->s_bdi->ra_pages = ra_pages;
1204#endif
1205}
1206
1207static inline unsigned bucket_bytes(const struct bch_dev *ca)
1208{
1209 return ca->mi.bucket_size << 9;
1210}
1211
1212static inline unsigned block_bytes(const struct bch_fs *c)
1213{
1214 return c->opts.block_size;
1215}
1216
1217static inline unsigned block_sectors(const struct bch_fs *c)
1218{
1219 return c->opts.block_size >> 9;
1220}
1221
1222static inline bool btree_id_cached(const struct bch_fs *c, enum btree_id btree)
1223{
1224 return c->btree_key_cache_btrees & (1U << btree);
1225}
1226
1227static inline struct timespec64 bch2_time_to_timespec(const struct bch_fs *c, s64 time)
1228{
1229 struct timespec64 t;
1230 s32 rem;
1231
1232 time += c->sb.time_base_lo;
1233
1234 t.tv_sec = div_s64_rem(dividend: time, divisor: c->sb.time_units_per_sec, remainder: &rem);
1235 t.tv_nsec = rem * c->sb.nsec_per_time_unit;
1236 return t;
1237}
1238
1239static inline s64 timespec_to_bch2_time(const struct bch_fs *c, struct timespec64 ts)
1240{
1241 return (ts.tv_sec * c->sb.time_units_per_sec +
1242 (int) ts.tv_nsec / c->sb.nsec_per_time_unit) - c->sb.time_base_lo;
1243}
1244
1245static inline s64 bch2_current_time(const struct bch_fs *c)
1246{
1247 struct timespec64 now;
1248
1249 ktime_get_coarse_real_ts64(ts: &now);
1250 return timespec_to_bch2_time(c, ts: now);
1251}
1252
1253static inline bool bch2_dev_exists2(const struct bch_fs *c, unsigned dev)
1254{
1255 return dev < c->sb.nr_devices && c->devs[dev];
1256}
1257
1258static inline struct stdio_redirect *bch2_fs_stdio_redirect(struct bch_fs *c)
1259{
1260 struct stdio_redirect *stdio = c->stdio;
1261
1262 if (c->stdio_filter && c->stdio_filter != current)
1263 stdio = NULL;
1264 return stdio;
1265}
1266
1267static inline unsigned metadata_replicas_required(struct bch_fs *c)
1268{
1269 return min(c->opts.metadata_replicas,
1270 c->opts.metadata_replicas_required);
1271}
1272
1273static inline unsigned data_replicas_required(struct bch_fs *c)
1274{
1275 return min(c->opts.data_replicas,
1276 c->opts.data_replicas_required);
1277}
1278
1279#define BKEY_PADDED_ONSTACK(key, pad) \
1280 struct { struct bkey_i key; __u64 key ## _pad[pad]; }
1281
1282#endif /* _BCACHEFS_H */
1283

source code of linux/fs/bcachefs/bcachefs.h