xref: /linux/fs/bcachefs/bcachefs.h (revision 4e94ddfe2aab72139acb8d5372fac9e6c3f3e383)
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/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>
205 
206 #include "bcachefs_format.h"
207 #include "errcode.h"
208 #include "fifo.h"
209 #include "nocow_locking_types.h"
210 #include "opts.h"
211 #include "recovery_types.h"
212 #include "sb-errors_types.h"
213 #include "seqmutex.h"
214 #include "util.h"
215 
216 #ifdef CONFIG_BCACHEFS_DEBUG
217 #define BCH_WRITE_REF_DEBUG
218 #endif
219 
220 #ifndef dynamic_fault
221 #define dynamic_fault(...)		0
222 #endif
223 
224 #define race_fault(...)			dynamic_fault("bcachefs:race")
225 
226 #define trace_and_count(_c, _name, ...)					\
227 do {									\
228 	this_cpu_inc((_c)->counters[BCH_COUNTER_##_name]);		\
229 	trace_##_name(__VA_ARGS__);					\
230 } while (0)
231 
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)
238 
239 #ifdef __KERNEL__
240 #define BCACHEFS_LOG_PREFIX
241 #endif
242 
243 #ifdef BCACHEFS_LOG_PREFIX
244 
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)
251 
252 #else
253 
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)
260 
261 #endif
262 
263 #define bch2_fmt(_c, fmt)		bch2_log_msg(_c, fmt "\n")
264 
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__)
273 
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__)
284 
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__)
295 
296 #define bch_err_fn(_c, _ret)						\
297 do {									\
298 	if (_ret && !bch2_err_matches(_ret, BCH_ERR_transaction_restart))\
299 		bch_err(_c, "%s(): error %s", __func__, bch2_err_str(_ret));\
300 } while (0)
301 
302 #define bch_err_msg(_c, _ret, _msg, ...)				\
303 do {									\
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));		\
307 } while (0)
308 
309 #define bch_verbose(c, fmt, ...)					\
310 do {									\
311 	if ((c)->opts.verbose)						\
312 		bch_info(c, fmt, ##__VA_ARGS__);			\
313 } while (0)
314 
315 #define pr_verbose_init(opts, fmt, ...)					\
316 do {									\
317 	if (opt_get(opts, verbose))					\
318 		pr_info(fmt, ##__VA_ARGS__);				\
319 } while (0)
320 
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 "	\
335 		"done in memory")					\
336 	BCH_DEBUG_PARAM(verify_all_btree_replicas,			\
337 		"When reading btree nodes, read all replicas and "	\
338 		"compare them")						\
339 	BCH_DEBUG_PARAM(backpointers_no_use_write_buffer,		\
340 		"Don't use the write buffer for backpointers, enabling "\
341 		"extra runtime checks")
342 
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"	\
363 		"cached data")						\
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")
369 
370 #define BCH_DEBUG_PARAMS_ALL() BCH_DEBUG_PARAMS_ALWAYS() BCH_DEBUG_PARAMS_DEBUG()
371 
372 #ifdef CONFIG_BCACHEFS_DEBUG
373 #define BCH_DEBUG_PARAMS() BCH_DEBUG_PARAMS_ALL()
374 #else
375 #define BCH_DEBUG_PARAMS() BCH_DEBUG_PARAMS_ALWAYS()
376 #endif
377 
378 #define BCH_DEBUG_PARAM(name, description) extern bool bch2_##name;
379 BCH_DEBUG_PARAMS()
380 #undef BCH_DEBUG_PARAM
381 
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
386 #endif
387 
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)			\
393 	x(btree_node_sort)			\
394 	x(btree_node_read)			\
395 	x(btree_interior_update_foreground)	\
396 	x(btree_interior_update_total)		\
397 	x(btree_gc)				\
398 	x(data_write)				\
399 	x(data_read)				\
400 	x(data_promote)				\
401 	x(journal_flush_write)			\
402 	x(journal_noflush_write)		\
403 	x(journal_flush_seq)			\
404 	x(blocked_journal)			\
405 	x(blocked_allocate)			\
406 	x(blocked_allocate_open_bucket)		\
407 	x(nocow_lock_contended)
408 
409 enum bch_time_stats {
410 #define x(name) BCH_TIME_##name,
411 	BCH_TIME_STATS()
412 #undef x
413 	BCH_TIME_STAT_NR
414 };
415 
416 #include "alloc_types.h"
417 #include "btree_types.h"
418 #include "btree_write_buffer_types.h"
419 #include "buckets_types.h"
420 #include "buckets_waiting_for_journal_types.h"
421 #include "clock_types.h"
422 #include "disk_groups_types.h"
423 #include "ec_types.h"
424 #include "journal_types.h"
425 #include "keylist_types.h"
426 #include "quota_types.h"
427 #include "rebalance_types.h"
428 #include "replicas_types.h"
429 #include "subvolume_types.h"
430 #include "super_types.h"
431 
432 /* Number of nodes btree coalesce will try to coalesce at once */
433 #define GC_MERGE_NODES		4U
434 
435 /* Maximum number of nodes we might need to allocate atomically: */
436 #define BTREE_RESERVE_MAX	(BTREE_MAX_DEPTH + (BTREE_MAX_DEPTH - 1))
437 
438 /* Size of the freelist we allocate btree nodes from: */
439 #define BTREE_NODE_RESERVE	(BTREE_RESERVE_MAX * 4)
440 
441 #define BTREE_NODE_OPEN_BUCKET_RESERVE	(BTREE_RESERVE_MAX * BCH_REPLICAS_MAX)
442 
443 struct btree;
444 
445 enum gc_phase {
446 	GC_PHASE_NOT_RUNNING,
447 	GC_PHASE_START,
448 	GC_PHASE_SB,
449 
450 	GC_PHASE_BTREE_stripes,
451 	GC_PHASE_BTREE_extents,
452 	GC_PHASE_BTREE_inodes,
453 	GC_PHASE_BTREE_dirents,
454 	GC_PHASE_BTREE_xattrs,
455 	GC_PHASE_BTREE_alloc,
456 	GC_PHASE_BTREE_quotas,
457 	GC_PHASE_BTREE_reflink,
458 	GC_PHASE_BTREE_subvolumes,
459 	GC_PHASE_BTREE_snapshots,
460 	GC_PHASE_BTREE_lru,
461 	GC_PHASE_BTREE_freespace,
462 	GC_PHASE_BTREE_need_discard,
463 	GC_PHASE_BTREE_backpointers,
464 	GC_PHASE_BTREE_bucket_gens,
465 	GC_PHASE_BTREE_snapshot_trees,
466 	GC_PHASE_BTREE_deleted_inodes,
467 	GC_PHASE_BTREE_logged_ops,
468 	GC_PHASE_BTREE_rebalance_work,
469 
470 	GC_PHASE_PENDING_DELETE,
471 };
472 
473 struct gc_pos {
474 	enum gc_phase		phase;
475 	struct bpos		pos;
476 	unsigned		level;
477 };
478 
479 struct reflink_gc {
480 	u64		offset;
481 	u32		size;
482 	u32		refcount;
483 };
484 
485 typedef GENRADIX(struct reflink_gc) reflink_gc_table;
486 
487 struct io_count {
488 	u64			sectors[2][BCH_DATA_NR];
489 };
490 
491 struct bch_dev {
492 	struct kobject		kobj;
493 	struct percpu_ref	ref;
494 	struct completion	ref_completion;
495 	struct percpu_ref	io_ref;
496 	struct completion	io_ref_completion;
497 
498 	struct bch_fs		*fs;
499 
500 	u8			dev_idx;
501 	/*
502 	 * Cached version of this device's member info from superblock
503 	 * Committed by bch2_write_super() -> bch_fs_mi_update()
504 	 */
505 	struct bch_member_cpu	mi;
506 	atomic64_t		errors[BCH_MEMBER_ERROR_NR];
507 
508 	__uuid_t		uuid;
509 	char			name[BDEVNAME_SIZE];
510 
511 	struct bch_sb_handle	disk_sb;
512 	struct bch_sb		*sb_read_scratch;
513 	int			sb_write_error;
514 	dev_t			dev;
515 	atomic_t		flush_seq;
516 
517 	struct bch_devs_mask	self;
518 
519 	/* biosets used in cloned bios for writing multiple replicas */
520 	struct bio_set		replica_set;
521 
522 	/*
523 	 * Buckets:
524 	 * Per-bucket arrays are protected by c->mark_lock, bucket_lock and
525 	 * gc_lock, for device resize - holding any is sufficient for access:
526 	 * Or rcu_read_lock(), but only for ptr_stale():
527 	 */
528 	struct bucket_array __rcu *buckets_gc;
529 	struct bucket_gens __rcu *bucket_gens;
530 	u8			*oldest_gen;
531 	unsigned long		*buckets_nouse;
532 	struct rw_semaphore	bucket_lock;
533 
534 	struct bch_dev_usage		*usage_base;
535 	struct bch_dev_usage __percpu	*usage[JOURNAL_BUF_NR];
536 	struct bch_dev_usage __percpu	*usage_gc;
537 
538 	/* Allocator: */
539 	u64			new_fs_bucket_idx;
540 	u64			alloc_cursor;
541 
542 	unsigned		nr_open_buckets;
543 	unsigned		nr_btree_reserve;
544 
545 	size_t			inc_gen_needs_gc;
546 	size_t			inc_gen_really_needs_gc;
547 	size_t			buckets_waiting_on_journal;
548 
549 	atomic64_t		rebalance_work;
550 
551 	struct journal_device	journal;
552 	u64			prev_journal_sector;
553 
554 	struct work_struct	io_error_work;
555 
556 	/* The rest of this all shows up in sysfs */
557 	atomic64_t		cur_latency[2];
558 	struct bch2_time_stats	io_latency[2];
559 
560 #define CONGESTED_MAX		1024
561 	atomic_t		congested;
562 	u64			congested_last;
563 
564 	struct io_count __percpu *io_done;
565 };
566 
567 enum {
568 	/* startup: */
569 	BCH_FS_STARTED,
570 	BCH_FS_MAY_GO_RW,
571 	BCH_FS_RW,
572 	BCH_FS_WAS_RW,
573 
574 	/* shutdown: */
575 	BCH_FS_STOPPING,
576 	BCH_FS_EMERGENCY_RO,
577 	BCH_FS_GOING_RO,
578 	BCH_FS_WRITE_DISABLE_COMPLETE,
579 	BCH_FS_CLEAN_SHUTDOWN,
580 
581 	/* fsck passes: */
582 	BCH_FS_FSCK_DONE,
583 	BCH_FS_INITIAL_GC_UNFIXED,	/* kill when we enumerate fsck errors */
584 	BCH_FS_NEED_ANOTHER_GC,
585 
586 	BCH_FS_NEED_DELETE_DEAD_SNAPSHOTS,
587 
588 	/* errors: */
589 	BCH_FS_ERROR,
590 	BCH_FS_TOPOLOGY_ERROR,
591 	BCH_FS_ERRORS_FIXED,
592 	BCH_FS_ERRORS_NOT_FIXED,
593 };
594 
595 struct btree_debug {
596 	unsigned		id;
597 };
598 
599 #define BCH_TRANSACTIONS_NR 128
600 
601 struct btree_transaction_stats {
602 	struct bch2_time_stats	lock_hold_times;
603 	struct mutex		lock;
604 	unsigned		nr_max_paths;
605 	unsigned		wb_updates_size;
606 	unsigned		max_mem;
607 	char			*max_paths_text;
608 };
609 
610 struct bch_fs_pcpu {
611 	u64			sectors_available;
612 };
613 
614 struct journal_seq_blacklist_table {
615 	size_t			nr;
616 	struct journal_seq_blacklist_table_entry {
617 		u64		start;
618 		u64		end;
619 		bool		dirty;
620 	}			entries[0];
621 };
622 
623 struct journal_keys {
624 	struct journal_key {
625 		u64		journal_seq;
626 		u32		journal_offset;
627 		enum btree_id	btree_id:8;
628 		unsigned	level:8;
629 		bool		allocated;
630 		bool		overwritten;
631 		struct bkey_i	*k;
632 	}			*d;
633 	/*
634 	 * Gap buffer: instead of all the empty space in the array being at the
635 	 * end of the buffer - from @nr to @size - the empty space is at @gap.
636 	 * This means that sequential insertions are O(n) instead of O(n^2).
637 	 */
638 	size_t			gap;
639 	size_t			nr;
640 	size_t			size;
641 };
642 
643 struct btree_trans_buf {
644 	struct btree_trans	*trans;
645 };
646 
647 #define REPLICAS_DELTA_LIST_MAX	(1U << 16)
648 
649 #define BCACHEFS_ROOT_SUBVOL_INUM					\
650 	((subvol_inum) { BCACHEFS_ROOT_SUBVOL,	BCACHEFS_ROOT_INO })
651 
652 #define BCH_WRITE_REFS()						\
653 	x(trans)							\
654 	x(write)							\
655 	x(promote)							\
656 	x(node_rewrite)							\
657 	x(stripe_create)						\
658 	x(stripe_delete)						\
659 	x(reflink)							\
660 	x(fallocate)							\
661 	x(discard)							\
662 	x(invalidate)							\
663 	x(delete_dead_snapshots)					\
664 	x(snapshot_delete_pagecache)					\
665 	x(sysfs)
666 
667 enum bch_write_ref {
668 #define x(n) BCH_WRITE_REF_##n,
669 	BCH_WRITE_REFS()
670 #undef x
671 	BCH_WRITE_REF_NR,
672 };
673 
674 struct bch_fs {
675 	struct closure		cl;
676 
677 	struct list_head	list;
678 	struct kobject		kobj;
679 	struct kobject		counters_kobj;
680 	struct kobject		internal;
681 	struct kobject		opts_dir;
682 	struct kobject		time_stats;
683 	unsigned long		flags;
684 
685 	int			minor;
686 	struct device		*chardev;
687 	struct super_block	*vfs_sb;
688 	dev_t			dev;
689 	char			name[40];
690 
691 	/* ro/rw, add/remove/resize devices: */
692 	struct rw_semaphore	state_lock;
693 
694 	/* Counts outstanding writes, for clean transition to read-only */
695 #ifdef BCH_WRITE_REF_DEBUG
696 	atomic_long_t		writes[BCH_WRITE_REF_NR];
697 #else
698 	struct percpu_ref	writes;
699 #endif
700 	struct work_struct	read_only_work;
701 
702 	struct bch_dev __rcu	*devs[BCH_SB_MEMBERS_MAX];
703 
704 	struct bch_replicas_cpu replicas;
705 	struct bch_replicas_cpu replicas_gc;
706 	struct mutex		replicas_gc_lock;
707 	mempool_t		replicas_delta_pool;
708 
709 	struct journal_entry_res btree_root_journal_res;
710 	struct journal_entry_res replicas_journal_res;
711 	struct journal_entry_res clock_journal_res;
712 	struct journal_entry_res dev_usage_journal_res;
713 
714 	struct bch_disk_groups_cpu __rcu *disk_groups;
715 
716 	struct bch_opts		opts;
717 
718 	/* Updated by bch2_sb_update():*/
719 	struct {
720 		__uuid_t	uuid;
721 		__uuid_t	user_uuid;
722 
723 		u16		version;
724 		u16		version_min;
725 		u16		version_upgrade_complete;
726 
727 		u8		nr_devices;
728 		u8		clean;
729 
730 		u8		encryption_type;
731 
732 		u64		time_base_lo;
733 		u32		time_base_hi;
734 		unsigned	time_units_per_sec;
735 		unsigned	nsec_per_time_unit;
736 		u64		features;
737 		u64		compat;
738 	}			sb;
739 
740 
741 	struct bch_sb_handle	disk_sb;
742 
743 	unsigned short		block_bits;	/* ilog2(block_size) */
744 
745 	u16			btree_foreground_merge_threshold;
746 
747 	struct closure		sb_write;
748 	struct mutex		sb_lock;
749 
750 	/* snapshot.c: */
751 	struct snapshot_table __rcu *snapshots;
752 	size_t			snapshot_table_size;
753 	struct mutex		snapshot_table_lock;
754 	struct rw_semaphore	snapshot_create_lock;
755 
756 	struct work_struct	snapshot_delete_work;
757 	struct work_struct	snapshot_wait_for_pagecache_and_delete_work;
758 	snapshot_id_list	snapshots_unlinked;
759 	struct mutex		snapshots_unlinked_lock;
760 
761 	/* BTREE CACHE */
762 	struct bio_set		btree_bio;
763 	struct workqueue_struct	*io_complete_wq;
764 
765 	struct btree_root	btree_roots_known[BTREE_ID_NR];
766 	DARRAY(struct btree_root) btree_roots_extra;
767 	struct mutex		btree_root_lock;
768 
769 	struct btree_cache	btree_cache;
770 
771 	/*
772 	 * Cache of allocated btree nodes - if we allocate a btree node and
773 	 * don't use it, if we free it that space can't be reused until going
774 	 * _all_ the way through the allocator (which exposes us to a livelock
775 	 * when allocating btree reserves fail halfway through) - instead, we
776 	 * can stick them here:
777 	 */
778 	struct btree_alloc	btree_reserve_cache[BTREE_NODE_RESERVE * 2];
779 	unsigned		btree_reserve_cache_nr;
780 	struct mutex		btree_reserve_cache_lock;
781 
782 	mempool_t		btree_interior_update_pool;
783 	struct list_head	btree_interior_update_list;
784 	struct list_head	btree_interior_updates_unwritten;
785 	struct mutex		btree_interior_update_lock;
786 	struct closure_waitlist	btree_interior_update_wait;
787 
788 	struct workqueue_struct	*btree_interior_update_worker;
789 	struct work_struct	btree_interior_update_work;
790 
791 	struct list_head	pending_node_rewrites;
792 	struct mutex		pending_node_rewrites_lock;
793 
794 	/* btree_io.c: */
795 	spinlock_t		btree_write_error_lock;
796 	struct btree_write_stats {
797 		atomic64_t	nr;
798 		atomic64_t	bytes;
799 	}			btree_write_stats[BTREE_WRITE_TYPE_NR];
800 
801 	/* btree_iter.c: */
802 	struct seqmutex		btree_trans_lock;
803 	struct list_head	btree_trans_list;
804 	mempool_t		btree_trans_pool;
805 	mempool_t		btree_trans_mem_pool;
806 	struct btree_trans_buf  __percpu	*btree_trans_bufs;
807 
808 	struct srcu_struct	btree_trans_barrier;
809 	bool			btree_trans_barrier_initialized;
810 
811 	struct btree_key_cache	btree_key_cache;
812 	unsigned		btree_key_cache_btrees;
813 
814 	struct btree_write_buffer btree_write_buffer;
815 
816 	struct workqueue_struct	*btree_update_wq;
817 	struct workqueue_struct	*btree_io_complete_wq;
818 	/* copygc needs its own workqueue for index updates.. */
819 	struct workqueue_struct	*copygc_wq;
820 	/*
821 	 * Use a dedicated wq for write ref holder tasks. Required to avoid
822 	 * dependency problems with other wq tasks that can block on ref
823 	 * draining, such as read-only transition.
824 	 */
825 	struct workqueue_struct *write_ref_wq;
826 
827 	/* ALLOCATION */
828 	struct bch_devs_mask	rw_devs[BCH_DATA_NR];
829 
830 	u64			capacity; /* sectors */
831 
832 	/*
833 	 * When capacity _decreases_ (due to a disk being removed), we
834 	 * increment capacity_gen - this invalidates outstanding reservations
835 	 * and forces them to be revalidated
836 	 */
837 	u32			capacity_gen;
838 	unsigned		bucket_size_max;
839 
840 	atomic64_t		sectors_available;
841 	struct mutex		sectors_available_lock;
842 
843 	struct bch_fs_pcpu __percpu	*pcpu;
844 
845 	struct percpu_rw_semaphore	mark_lock;
846 
847 	seqcount_t			usage_lock;
848 	struct bch_fs_usage		*usage_base;
849 	struct bch_fs_usage __percpu	*usage[JOURNAL_BUF_NR];
850 	struct bch_fs_usage __percpu	*usage_gc;
851 	u64 __percpu		*online_reserved;
852 
853 	/* single element mempool: */
854 	struct mutex		usage_scratch_lock;
855 	struct bch_fs_usage_online *usage_scratch;
856 
857 	struct io_clock		io_clock[2];
858 
859 	/* JOURNAL SEQ BLACKLIST */
860 	struct journal_seq_blacklist_table *
861 				journal_seq_blacklist_table;
862 	struct work_struct	journal_seq_blacklist_gc_work;
863 
864 	/* ALLOCATOR */
865 	spinlock_t		freelist_lock;
866 	struct closure_waitlist	freelist_wait;
867 	u64			blocked_allocate;
868 	u64			blocked_allocate_open_bucket;
869 
870 	open_bucket_idx_t	open_buckets_freelist;
871 	open_bucket_idx_t	open_buckets_nr_free;
872 	struct closure_waitlist	open_buckets_wait;
873 	struct open_bucket	open_buckets[OPEN_BUCKETS_COUNT];
874 	open_bucket_idx_t	open_buckets_hash[OPEN_BUCKETS_COUNT];
875 
876 	open_bucket_idx_t	open_buckets_partial[OPEN_BUCKETS_COUNT];
877 	open_bucket_idx_t	open_buckets_partial_nr;
878 
879 	struct write_point	btree_write_point;
880 	struct write_point	rebalance_write_point;
881 
882 	struct write_point	write_points[WRITE_POINT_MAX];
883 	struct hlist_head	write_points_hash[WRITE_POINT_HASH_NR];
884 	struct mutex		write_points_hash_lock;
885 	unsigned		write_points_nr;
886 
887 	struct buckets_waiting_for_journal buckets_waiting_for_journal;
888 	struct work_struct	discard_work;
889 	struct work_struct	invalidate_work;
890 
891 	/* GARBAGE COLLECTION */
892 	struct task_struct	*gc_thread;
893 	atomic_t		kick_gc;
894 	unsigned long		gc_count;
895 
896 	enum btree_id		gc_gens_btree;
897 	struct bpos		gc_gens_pos;
898 
899 	/*
900 	 * Tracks GC's progress - everything in the range [ZERO_KEY..gc_cur_pos]
901 	 * has been marked by GC.
902 	 *
903 	 * gc_cur_phase is a superset of btree_ids (BTREE_ID_extents etc.)
904 	 *
905 	 * Protected by gc_pos_lock. Only written to by GC thread, so GC thread
906 	 * can read without a lock.
907 	 */
908 	seqcount_t		gc_pos_lock;
909 	struct gc_pos		gc_pos;
910 
911 	/*
912 	 * The allocation code needs gc_mark in struct bucket to be correct, but
913 	 * it's not while a gc is in progress.
914 	 */
915 	struct rw_semaphore	gc_lock;
916 	struct mutex		gc_gens_lock;
917 
918 	/* IO PATH */
919 	struct semaphore	io_in_flight;
920 	struct bio_set		bio_read;
921 	struct bio_set		bio_read_split;
922 	struct bio_set		bio_write;
923 	struct mutex		bio_bounce_pages_lock;
924 	mempool_t		bio_bounce_pages;
925 	struct bucket_nocow_lock_table
926 				nocow_locks;
927 	struct rhashtable	promote_table;
928 
929 	mempool_t		compression_bounce[2];
930 	mempool_t		compress_workspace[BCH_COMPRESSION_TYPE_NR];
931 	mempool_t		decompress_workspace;
932 	ZSTD_parameters		zstd_params;
933 
934 	struct crypto_shash	*sha256;
935 	struct crypto_sync_skcipher *chacha20;
936 	struct crypto_shash	*poly1305;
937 
938 	atomic64_t		key_version;
939 
940 	mempool_t		large_bkey_pool;
941 
942 	/* MOVE.C */
943 	struct list_head	moving_context_list;
944 	struct mutex		moving_context_lock;
945 
946 	/* REBALANCE */
947 	struct bch_fs_rebalance	rebalance;
948 
949 	/* COPYGC */
950 	struct task_struct	*copygc_thread;
951 	struct write_point	copygc_write_point;
952 	s64			copygc_wait_at;
953 	s64			copygc_wait;
954 	bool			copygc_running;
955 	wait_queue_head_t	copygc_running_wq;
956 
957 	/* STRIPES: */
958 	GENRADIX(struct stripe) stripes;
959 	GENRADIX(struct gc_stripe) gc_stripes;
960 
961 	struct hlist_head	ec_stripes_new[32];
962 	spinlock_t		ec_stripes_new_lock;
963 
964 	ec_stripes_heap		ec_stripes_heap;
965 	struct mutex		ec_stripes_heap_lock;
966 
967 	/* ERASURE CODING */
968 	struct list_head	ec_stripe_head_list;
969 	struct mutex		ec_stripe_head_lock;
970 
971 	struct list_head	ec_stripe_new_list;
972 	struct mutex		ec_stripe_new_lock;
973 	wait_queue_head_t	ec_stripe_new_wait;
974 
975 	struct work_struct	ec_stripe_create_work;
976 	u64			ec_stripe_hint;
977 
978 	struct work_struct	ec_stripe_delete_work;
979 
980 	struct bio_set		ec_bioset;
981 
982 	/* REFLINK */
983 	reflink_gc_table	reflink_gc_table;
984 	size_t			reflink_gc_nr;
985 
986 	/* fs.c */
987 	struct list_head	vfs_inodes_list;
988 	struct mutex		vfs_inodes_lock;
989 
990 	/* VFS IO PATH - fs-io.c */
991 	struct bio_set		writepage_bioset;
992 	struct bio_set		dio_write_bioset;
993 	struct bio_set		dio_read_bioset;
994 	struct bio_set		nocow_flush_bioset;
995 
996 	/* QUOTAS */
997 	struct bch_memquota_type quotas[QTYP_NR];
998 
999 	/* RECOVERY */
1000 	u64			journal_replay_seq_start;
1001 	u64			journal_replay_seq_end;
1002 	enum bch_recovery_pass	curr_recovery_pass;
1003 	/* bitmap of explicitly enabled recovery passes: */
1004 	u64			recovery_passes_explicit;
1005 	u64			recovery_passes_complete;
1006 
1007 	/* DEBUG JUNK */
1008 	struct dentry		*fs_debug_dir;
1009 	struct dentry		*btree_debug_dir;
1010 	struct btree_debug	btree_debug[BTREE_ID_NR];
1011 	struct btree		*verify_data;
1012 	struct btree_node	*verify_ondisk;
1013 	struct mutex		verify_lock;
1014 
1015 	u64			*unused_inode_hints;
1016 	unsigned		inode_shard_bits;
1017 
1018 	/*
1019 	 * A btree node on disk could have too many bsets for an iterator to fit
1020 	 * on the stack - have to dynamically allocate them
1021 	 */
1022 	mempool_t		fill_iter;
1023 
1024 	mempool_t		btree_bounce_pool;
1025 
1026 	struct journal		journal;
1027 	GENRADIX(struct journal_replay *) journal_entries;
1028 	u64			journal_entries_base_seq;
1029 	struct journal_keys	journal_keys;
1030 	struct list_head	journal_iters;
1031 
1032 	u64			last_bucket_seq_cleanup;
1033 
1034 	u64			counters_on_mount[BCH_COUNTER_NR];
1035 	u64 __percpu		*counters;
1036 
1037 	unsigned		btree_gc_periodic:1;
1038 	unsigned		copy_gc_enabled:1;
1039 	bool			promote_whole_extents;
1040 
1041 	struct bch2_time_stats	times[BCH_TIME_STAT_NR];
1042 
1043 	struct btree_transaction_stats btree_transaction_stats[BCH_TRANSACTIONS_NR];
1044 
1045 	/* ERRORS */
1046 	struct list_head	fsck_error_msgs;
1047 	struct mutex		fsck_error_msgs_lock;
1048 	bool			fsck_alloc_msgs_err;
1049 
1050 	bch_sb_errors_cpu	fsck_error_counts;
1051 	struct mutex		fsck_error_counts_lock;
1052 };
1053 
1054 extern struct wait_queue_head bch2_read_only_wait;
1055 
1056 static inline void bch2_write_ref_get(struct bch_fs *c, enum bch_write_ref ref)
1057 {
1058 #ifdef BCH_WRITE_REF_DEBUG
1059 	atomic_long_inc(&c->writes[ref]);
1060 #else
1061 	percpu_ref_get(&c->writes);
1062 #endif
1063 }
1064 
1065 static inline bool bch2_write_ref_tryget(struct bch_fs *c, enum bch_write_ref ref)
1066 {
1067 #ifdef BCH_WRITE_REF_DEBUG
1068 	return !test_bit(BCH_FS_GOING_RO, &c->flags) &&
1069 		atomic_long_inc_not_zero(&c->writes[ref]);
1070 #else
1071 	return percpu_ref_tryget_live(&c->writes);
1072 #endif
1073 }
1074 
1075 static inline void bch2_write_ref_put(struct bch_fs *c, enum bch_write_ref ref)
1076 {
1077 #ifdef BCH_WRITE_REF_DEBUG
1078 	long v = atomic_long_dec_return(&c->writes[ref]);
1079 
1080 	BUG_ON(v < 0);
1081 	if (v)
1082 		return;
1083 	for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
1084 		if (atomic_long_read(&c->writes[i]))
1085 			return;
1086 
1087 	set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
1088 	wake_up(&bch2_read_only_wait);
1089 #else
1090 	percpu_ref_put(&c->writes);
1091 #endif
1092 }
1093 
1094 static inline void bch2_set_ra_pages(struct bch_fs *c, unsigned ra_pages)
1095 {
1096 #ifndef NO_BCACHEFS_FS
1097 	if (c->vfs_sb)
1098 		c->vfs_sb->s_bdi->ra_pages = ra_pages;
1099 #endif
1100 }
1101 
1102 static inline unsigned bucket_bytes(const struct bch_dev *ca)
1103 {
1104 	return ca->mi.bucket_size << 9;
1105 }
1106 
1107 static inline unsigned block_bytes(const struct bch_fs *c)
1108 {
1109 	return c->opts.block_size;
1110 }
1111 
1112 static inline unsigned block_sectors(const struct bch_fs *c)
1113 {
1114 	return c->opts.block_size >> 9;
1115 }
1116 
1117 static inline size_t btree_sectors(const struct bch_fs *c)
1118 {
1119 	return c->opts.btree_node_size >> 9;
1120 }
1121 
1122 static inline bool btree_id_cached(const struct bch_fs *c, enum btree_id btree)
1123 {
1124 	return c->btree_key_cache_btrees & (1U << btree);
1125 }
1126 
1127 static inline struct timespec64 bch2_time_to_timespec(const struct bch_fs *c, s64 time)
1128 {
1129 	struct timespec64 t;
1130 	s32 rem;
1131 
1132 	time += c->sb.time_base_lo;
1133 
1134 	t.tv_sec = div_s64_rem(time, c->sb.time_units_per_sec, &rem);
1135 	t.tv_nsec = rem * c->sb.nsec_per_time_unit;
1136 	return t;
1137 }
1138 
1139 static inline s64 timespec_to_bch2_time(const struct bch_fs *c, struct timespec64 ts)
1140 {
1141 	return (ts.tv_sec * c->sb.time_units_per_sec +
1142 		(int) ts.tv_nsec / c->sb.nsec_per_time_unit) - c->sb.time_base_lo;
1143 }
1144 
1145 static inline s64 bch2_current_time(const struct bch_fs *c)
1146 {
1147 	struct timespec64 now;
1148 
1149 	ktime_get_coarse_real_ts64(&now);
1150 	return timespec_to_bch2_time(c, now);
1151 }
1152 
1153 static inline bool bch2_dev_exists2(const struct bch_fs *c, unsigned dev)
1154 {
1155 	return dev < c->sb.nr_devices && c->devs[dev];
1156 }
1157 
1158 #define BKEY_PADDED_ONSTACK(key, pad)				\
1159 	struct { struct bkey_i key; __u64 key ## _pad[pad]; }
1160 
1161 #endif /* _BCACHEFS_H */
1162