xref: /linux/fs/bcachefs/bcachefs.h (revision 031fba65fc202abf1f193e321be7a2c274fd88ba)
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 "seqmutex.h"
213 #include "util.h"
214 
215 #ifdef CONFIG_BCACHEFS_DEBUG
216 #define BCH_WRITE_REF_DEBUG
217 #endif
218 
219 #ifndef dynamic_fault
220 #define dynamic_fault(...)		0
221 #endif
222 
223 #define race_fault(...)			dynamic_fault("bcachefs:race")
224 
225 #define trace_and_count(_c, _name, ...)					\
226 do {									\
227 	this_cpu_inc((_c)->counters[BCH_COUNTER_##_name]);		\
228 	trace_##_name(__VA_ARGS__);					\
229 } while (0)
230 
231 #define bch2_fs_init_fault(name)					\
232 	dynamic_fault("bcachefs:bch_fs_init:" name)
233 #define bch2_meta_read_fault(name)					\
234 	 dynamic_fault("bcachefs:meta:read:" name)
235 #define bch2_meta_write_fault(name)					\
236 	 dynamic_fault("bcachefs:meta:write:" name)
237 
238 #ifdef __KERNEL__
239 #define BCACHEFS_LOG_PREFIX
240 #endif
241 
242 #ifdef BCACHEFS_LOG_PREFIX
243 
244 #define bch2_log_msg(_c, fmt)			"bcachefs (%s): " fmt, ((_c)->name)
245 #define bch2_fmt_dev(_ca, fmt)			"bcachefs (%s): " fmt "\n", ((_ca)->name)
246 #define bch2_fmt_dev_offset(_ca, _offset, fmt)	"bcachefs (%s sector %llu): " fmt "\n", ((_ca)->name), (_offset)
247 #define bch2_fmt_inum(_c, _inum, fmt)		"bcachefs (%s inum %llu): " fmt "\n", ((_c)->name), (_inum)
248 #define bch2_fmt_inum_offset(_c, _inum, _offset, fmt)			\
249 	 "bcachefs (%s inum %llu offset %llu): " fmt "\n", ((_c)->name), (_inum), (_offset)
250 
251 #else
252 
253 #define bch2_log_msg(_c, fmt)			fmt
254 #define bch2_fmt_dev(_ca, fmt)			"%s: " fmt "\n", ((_ca)->name)
255 #define bch2_fmt_dev_offset(_ca, _offset, fmt)	"%s sector %llu: " fmt "\n", ((_ca)->name), (_offset)
256 #define bch2_fmt_inum(_c, _inum, fmt)		"inum %llu: " fmt "\n", (_inum)
257 #define bch2_fmt_inum_offset(_c, _inum, _offset, fmt)				\
258 	 "inum %llu offset %llu: " fmt "\n", (_inum), (_offset)
259 
260 #endif
261 
262 #define bch2_fmt(_c, fmt)		bch2_log_msg(_c, fmt "\n")
263 
264 #define bch_info(c, fmt, ...) \
265 	printk(KERN_INFO bch2_fmt(c, fmt), ##__VA_ARGS__)
266 #define bch_notice(c, fmt, ...) \
267 	printk(KERN_NOTICE bch2_fmt(c, fmt), ##__VA_ARGS__)
268 #define bch_warn(c, fmt, ...) \
269 	printk(KERN_WARNING bch2_fmt(c, fmt), ##__VA_ARGS__)
270 #define bch_warn_ratelimited(c, fmt, ...) \
271 	printk_ratelimited(KERN_WARNING bch2_fmt(c, fmt), ##__VA_ARGS__)
272 
273 #define bch_err(c, fmt, ...) \
274 	printk(KERN_ERR bch2_fmt(c, fmt), ##__VA_ARGS__)
275 #define bch_err_dev(ca, fmt, ...) \
276 	printk(KERN_ERR bch2_fmt_dev(ca, fmt), ##__VA_ARGS__)
277 #define bch_err_dev_offset(ca, _offset, fmt, ...) \
278 	printk(KERN_ERR bch2_fmt_dev_offset(ca, _offset, fmt), ##__VA_ARGS__)
279 #define bch_err_inum(c, _inum, fmt, ...) \
280 	printk(KERN_ERR bch2_fmt_inum(c, _inum, fmt), ##__VA_ARGS__)
281 #define bch_err_inum_offset(c, _inum, _offset, fmt, ...) \
282 	printk(KERN_ERR bch2_fmt_inum_offset(c, _inum, _offset, fmt), ##__VA_ARGS__)
283 
284 #define bch_err_ratelimited(c, fmt, ...) \
285 	printk_ratelimited(KERN_ERR bch2_fmt(c, fmt), ##__VA_ARGS__)
286 #define bch_err_dev_ratelimited(ca, fmt, ...) \
287 	printk_ratelimited(KERN_ERR bch2_fmt_dev(ca, fmt), ##__VA_ARGS__)
288 #define bch_err_dev_offset_ratelimited(ca, _offset, fmt, ...) \
289 	printk_ratelimited(KERN_ERR bch2_fmt_dev_offset(ca, _offset, fmt), ##__VA_ARGS__)
290 #define bch_err_inum_ratelimited(c, _inum, fmt, ...) \
291 	printk_ratelimited(KERN_ERR bch2_fmt_inum(c, _inum, fmt), ##__VA_ARGS__)
292 #define bch_err_inum_offset_ratelimited(c, _inum, _offset, fmt, ...) \
293 	printk_ratelimited(KERN_ERR bch2_fmt_inum_offset(c, _inum, _offset, fmt), ##__VA_ARGS__)
294 
295 #define bch_err_fn(_c, _ret)						\
296 do {									\
297 	if (_ret && !bch2_err_matches(_ret, BCH_ERR_transaction_restart))\
298 		bch_err(_c, "%s(): error %s", __func__, bch2_err_str(_ret));\
299 } while (0)
300 
301 #define bch_err_msg(_c, _ret, _msg, ...)				\
302 do {									\
303 	if (_ret && !bch2_err_matches(_ret, BCH_ERR_transaction_restart))\
304 		bch_err(_c, "%s(): error " _msg " %s", __func__,	\
305 			##__VA_ARGS__, bch2_err_str(_ret));		\
306 } while (0)
307 
308 #define bch_verbose(c, fmt, ...)					\
309 do {									\
310 	if ((c)->opts.verbose)						\
311 		bch_info(c, fmt, ##__VA_ARGS__);			\
312 } while (0)
313 
314 #define pr_verbose_init(opts, fmt, ...)					\
315 do {									\
316 	if (opt_get(opts, verbose))					\
317 		pr_info(fmt, ##__VA_ARGS__);				\
318 } while (0)
319 
320 /* Parameters that are useful for debugging, but should always be compiled in: */
321 #define BCH_DEBUG_PARAMS_ALWAYS()					\
322 	BCH_DEBUG_PARAM(key_merging_disabled,				\
323 		"Disables merging of extents")				\
324 	BCH_DEBUG_PARAM(btree_gc_always_rewrite,			\
325 		"Causes mark and sweep to compact and rewrite every "	\
326 		"btree node it traverses")				\
327 	BCH_DEBUG_PARAM(btree_gc_rewrite_disabled,			\
328 		"Disables rewriting of btree nodes during mark and sweep")\
329 	BCH_DEBUG_PARAM(btree_shrinker_disabled,			\
330 		"Disables the shrinker callback for the btree node cache")\
331 	BCH_DEBUG_PARAM(verify_btree_ondisk,				\
332 		"Reread btree nodes at various points to verify the "	\
333 		"mergesort in the read path against modifications "	\
334 		"done in memory")					\
335 	BCH_DEBUG_PARAM(verify_all_btree_replicas,			\
336 		"When reading btree nodes, read all replicas and "	\
337 		"compare them")						\
338 	BCH_DEBUG_PARAM(backpointers_no_use_write_buffer,		\
339 		"Don't use the write buffer for backpointers, enabling "\
340 		"extra runtime checks")
341 
342 /* Parameters that should only be compiled in debug mode: */
343 #define BCH_DEBUG_PARAMS_DEBUG()					\
344 	BCH_DEBUG_PARAM(expensive_debug_checks,				\
345 		"Enables various runtime debugging checks that "	\
346 		"significantly affect performance")			\
347 	BCH_DEBUG_PARAM(debug_check_iterators,				\
348 		"Enables extra verification for btree iterators")	\
349 	BCH_DEBUG_PARAM(debug_check_btree_accounting,			\
350 		"Verify btree accounting for keys within a node")	\
351 	BCH_DEBUG_PARAM(journal_seq_verify,				\
352 		"Store the journal sequence number in the version "	\
353 		"number of every btree key, and verify that btree "	\
354 		"update ordering is preserved during recovery")		\
355 	BCH_DEBUG_PARAM(inject_invalid_keys,				\
356 		"Store the journal sequence number in the version "	\
357 		"number of every btree key, and verify that btree "	\
358 		"update ordering is preserved during recovery")		\
359 	BCH_DEBUG_PARAM(test_alloc_startup,				\
360 		"Force allocator startup to use the slowpath where it"	\
361 		"can't find enough free buckets without invalidating"	\
362 		"cached data")						\
363 	BCH_DEBUG_PARAM(force_reconstruct_read,				\
364 		"Force reads to use the reconstruct path, when reading"	\
365 		"from erasure coded extents")				\
366 	BCH_DEBUG_PARAM(test_restart_gc,				\
367 		"Test restarting mark and sweep gc when bucket gens change")
368 
369 #define BCH_DEBUG_PARAMS_ALL() BCH_DEBUG_PARAMS_ALWAYS() BCH_DEBUG_PARAMS_DEBUG()
370 
371 #ifdef CONFIG_BCACHEFS_DEBUG
372 #define BCH_DEBUG_PARAMS() BCH_DEBUG_PARAMS_ALL()
373 #else
374 #define BCH_DEBUG_PARAMS() BCH_DEBUG_PARAMS_ALWAYS()
375 #endif
376 
377 #define BCH_DEBUG_PARAM(name, description) extern bool bch2_##name;
378 BCH_DEBUG_PARAMS()
379 #undef BCH_DEBUG_PARAM
380 
381 #ifndef CONFIG_BCACHEFS_DEBUG
382 #define BCH_DEBUG_PARAM(name, description) static const __maybe_unused bool bch2_##name;
383 BCH_DEBUG_PARAMS_DEBUG()
384 #undef BCH_DEBUG_PARAM
385 #endif
386 
387 #define BCH_TIME_STATS()			\
388 	x(btree_node_mem_alloc)			\
389 	x(btree_node_split)			\
390 	x(btree_node_compact)			\
391 	x(btree_node_merge)			\
392 	x(btree_node_sort)			\
393 	x(btree_node_read)			\
394 	x(btree_interior_update_foreground)	\
395 	x(btree_interior_update_total)		\
396 	x(btree_gc)				\
397 	x(data_write)				\
398 	x(data_read)				\
399 	x(data_promote)				\
400 	x(journal_flush_write)			\
401 	x(journal_noflush_write)		\
402 	x(journal_flush_seq)			\
403 	x(blocked_journal)			\
404 	x(blocked_allocate)			\
405 	x(blocked_allocate_open_bucket)		\
406 	x(nocow_lock_contended)
407 
408 enum bch_time_stats {
409 #define x(name) BCH_TIME_##name,
410 	BCH_TIME_STATS()
411 #undef x
412 	BCH_TIME_STAT_NR
413 };
414 
415 #include "alloc_types.h"
416 #include "btree_types.h"
417 #include "btree_write_buffer_types.h"
418 #include "buckets_types.h"
419 #include "buckets_waiting_for_journal_types.h"
420 #include "clock_types.h"
421 #include "ec_types.h"
422 #include "journal_types.h"
423 #include "keylist_types.h"
424 #include "quota_types.h"
425 #include "rebalance_types.h"
426 #include "replicas_types.h"
427 #include "subvolume_types.h"
428 #include "super_types.h"
429 
430 /* Number of nodes btree coalesce will try to coalesce at once */
431 #define GC_MERGE_NODES		4U
432 
433 /* Maximum number of nodes we might need to allocate atomically: */
434 #define BTREE_RESERVE_MAX	(BTREE_MAX_DEPTH + (BTREE_MAX_DEPTH - 1))
435 
436 /* Size of the freelist we allocate btree nodes from: */
437 #define BTREE_NODE_RESERVE	(BTREE_RESERVE_MAX * 4)
438 
439 #define BTREE_NODE_OPEN_BUCKET_RESERVE	(BTREE_RESERVE_MAX * BCH_REPLICAS_MAX)
440 
441 struct btree;
442 
443 enum gc_phase {
444 	GC_PHASE_NOT_RUNNING,
445 	GC_PHASE_START,
446 	GC_PHASE_SB,
447 
448 	GC_PHASE_BTREE_stripes,
449 	GC_PHASE_BTREE_extents,
450 	GC_PHASE_BTREE_inodes,
451 	GC_PHASE_BTREE_dirents,
452 	GC_PHASE_BTREE_xattrs,
453 	GC_PHASE_BTREE_alloc,
454 	GC_PHASE_BTREE_quotas,
455 	GC_PHASE_BTREE_reflink,
456 	GC_PHASE_BTREE_subvolumes,
457 	GC_PHASE_BTREE_snapshots,
458 	GC_PHASE_BTREE_lru,
459 	GC_PHASE_BTREE_freespace,
460 	GC_PHASE_BTREE_need_discard,
461 	GC_PHASE_BTREE_backpointers,
462 	GC_PHASE_BTREE_bucket_gens,
463 	GC_PHASE_BTREE_snapshot_trees,
464 	GC_PHASE_BTREE_deleted_inodes,
465 	GC_PHASE_BTREE_logged_ops,
466 
467 	GC_PHASE_PENDING_DELETE,
468 };
469 
470 struct gc_pos {
471 	enum gc_phase		phase;
472 	struct bpos		pos;
473 	unsigned		level;
474 };
475 
476 struct reflink_gc {
477 	u64		offset;
478 	u32		size;
479 	u32		refcount;
480 };
481 
482 typedef GENRADIX(struct reflink_gc) reflink_gc_table;
483 
484 struct io_count {
485 	u64			sectors[2][BCH_DATA_NR];
486 };
487 
488 struct bch_dev {
489 	struct kobject		kobj;
490 	struct percpu_ref	ref;
491 	struct completion	ref_completion;
492 	struct percpu_ref	io_ref;
493 	struct completion	io_ref_completion;
494 
495 	struct bch_fs		*fs;
496 
497 	u8			dev_idx;
498 	/*
499 	 * Cached version of this device's member info from superblock
500 	 * Committed by bch2_write_super() -> bch_fs_mi_update()
501 	 */
502 	struct bch_member_cpu	mi;
503 	__uuid_t		uuid;
504 	char			name[BDEVNAME_SIZE];
505 
506 	struct bch_sb_handle	disk_sb;
507 	struct bch_sb		*sb_read_scratch;
508 	int			sb_write_error;
509 	dev_t			dev;
510 	atomic_t		flush_seq;
511 
512 	struct bch_devs_mask	self;
513 
514 	/* biosets used in cloned bios for writing multiple replicas */
515 	struct bio_set		replica_set;
516 
517 	/*
518 	 * Buckets:
519 	 * Per-bucket arrays are protected by c->mark_lock, bucket_lock and
520 	 * gc_lock, for device resize - holding any is sufficient for access:
521 	 * Or rcu_read_lock(), but only for ptr_stale():
522 	 */
523 	struct bucket_array __rcu *buckets_gc;
524 	struct bucket_gens __rcu *bucket_gens;
525 	u8			*oldest_gen;
526 	unsigned long		*buckets_nouse;
527 	struct rw_semaphore	bucket_lock;
528 
529 	struct bch_dev_usage		*usage_base;
530 	struct bch_dev_usage __percpu	*usage[JOURNAL_BUF_NR];
531 	struct bch_dev_usage __percpu	*usage_gc;
532 
533 	/* Allocator: */
534 	u64			new_fs_bucket_idx;
535 	u64			alloc_cursor;
536 
537 	unsigned		nr_open_buckets;
538 	unsigned		nr_btree_reserve;
539 
540 	size_t			inc_gen_needs_gc;
541 	size_t			inc_gen_really_needs_gc;
542 	size_t			buckets_waiting_on_journal;
543 
544 	atomic64_t		rebalance_work;
545 
546 	struct journal_device	journal;
547 	u64			prev_journal_sector;
548 
549 	struct work_struct	io_error_work;
550 
551 	/* The rest of this all shows up in sysfs */
552 	atomic64_t		cur_latency[2];
553 	struct bch2_time_stats	io_latency[2];
554 
555 #define CONGESTED_MAX		1024
556 	atomic_t		congested;
557 	u64			congested_last;
558 
559 	struct io_count __percpu *io_done;
560 };
561 
562 enum {
563 	/* startup: */
564 	BCH_FS_STARTED,
565 	BCH_FS_MAY_GO_RW,
566 	BCH_FS_RW,
567 	BCH_FS_WAS_RW,
568 
569 	/* shutdown: */
570 	BCH_FS_STOPPING,
571 	BCH_FS_EMERGENCY_RO,
572 	BCH_FS_GOING_RO,
573 	BCH_FS_WRITE_DISABLE_COMPLETE,
574 	BCH_FS_CLEAN_SHUTDOWN,
575 
576 	/* fsck passes: */
577 	BCH_FS_FSCK_DONE,
578 	BCH_FS_INITIAL_GC_UNFIXED,	/* kill when we enumerate fsck errors */
579 	BCH_FS_NEED_ANOTHER_GC,
580 
581 	BCH_FS_HAVE_DELETED_SNAPSHOTS,
582 
583 	/* errors: */
584 	BCH_FS_ERROR,
585 	BCH_FS_TOPOLOGY_ERROR,
586 	BCH_FS_ERRORS_FIXED,
587 	BCH_FS_ERRORS_NOT_FIXED,
588 };
589 
590 struct btree_debug {
591 	unsigned		id;
592 };
593 
594 #define BCH_TRANSACTIONS_NR 128
595 
596 struct btree_transaction_stats {
597 	struct bch2_time_stats	lock_hold_times;
598 	struct mutex		lock;
599 	unsigned		nr_max_paths;
600 	unsigned		wb_updates_size;
601 	unsigned		max_mem;
602 	char			*max_paths_text;
603 };
604 
605 struct bch_fs_pcpu {
606 	u64			sectors_available;
607 };
608 
609 struct journal_seq_blacklist_table {
610 	size_t			nr;
611 	struct journal_seq_blacklist_table_entry {
612 		u64		start;
613 		u64		end;
614 		bool		dirty;
615 	}			entries[0];
616 };
617 
618 struct journal_keys {
619 	struct journal_key {
620 		u64		journal_seq;
621 		u32		journal_offset;
622 		enum btree_id	btree_id:8;
623 		unsigned	level:8;
624 		bool		allocated;
625 		bool		overwritten;
626 		struct bkey_i	*k;
627 	}			*d;
628 	/*
629 	 * Gap buffer: instead of all the empty space in the array being at the
630 	 * end of the buffer - from @nr to @size - the empty space is at @gap.
631 	 * This means that sequential insertions are O(n) instead of O(n^2).
632 	 */
633 	size_t			gap;
634 	size_t			nr;
635 	size_t			size;
636 };
637 
638 struct btree_trans_buf {
639 	struct btree_trans	*trans;
640 };
641 
642 #define REPLICAS_DELTA_LIST_MAX	(1U << 16)
643 
644 #define BCACHEFS_ROOT_SUBVOL_INUM					\
645 	((subvol_inum) { BCACHEFS_ROOT_SUBVOL,	BCACHEFS_ROOT_INO })
646 
647 #define BCH_WRITE_REFS()						\
648 	x(trans)							\
649 	x(write)							\
650 	x(promote)							\
651 	x(node_rewrite)							\
652 	x(stripe_create)						\
653 	x(stripe_delete)						\
654 	x(reflink)							\
655 	x(fallocate)							\
656 	x(discard)							\
657 	x(invalidate)							\
658 	x(delete_dead_snapshots)					\
659 	x(snapshot_delete_pagecache)					\
660 	x(sysfs)
661 
662 enum bch_write_ref {
663 #define x(n) BCH_WRITE_REF_##n,
664 	BCH_WRITE_REFS()
665 #undef x
666 	BCH_WRITE_REF_NR,
667 };
668 
669 struct bch_fs {
670 	struct closure		cl;
671 
672 	struct list_head	list;
673 	struct kobject		kobj;
674 	struct kobject		counters_kobj;
675 	struct kobject		internal;
676 	struct kobject		opts_dir;
677 	struct kobject		time_stats;
678 	unsigned long		flags;
679 
680 	int			minor;
681 	struct device		*chardev;
682 	struct super_block	*vfs_sb;
683 	dev_t			dev;
684 	char			name[40];
685 
686 	/* ro/rw, add/remove/resize devices: */
687 	struct rw_semaphore	state_lock;
688 
689 	/* Counts outstanding writes, for clean transition to read-only */
690 #ifdef BCH_WRITE_REF_DEBUG
691 	atomic_long_t		writes[BCH_WRITE_REF_NR];
692 #else
693 	struct percpu_ref	writes;
694 #endif
695 	struct work_struct	read_only_work;
696 
697 	struct bch_dev __rcu	*devs[BCH_SB_MEMBERS_MAX];
698 
699 	struct bch_replicas_cpu replicas;
700 	struct bch_replicas_cpu replicas_gc;
701 	struct mutex		replicas_gc_lock;
702 	mempool_t		replicas_delta_pool;
703 
704 	struct journal_entry_res btree_root_journal_res;
705 	struct journal_entry_res replicas_journal_res;
706 	struct journal_entry_res clock_journal_res;
707 	struct journal_entry_res dev_usage_journal_res;
708 
709 	struct bch_disk_groups_cpu __rcu *disk_groups;
710 
711 	struct bch_opts		opts;
712 
713 	/* Updated by bch2_sb_update():*/
714 	struct {
715 		__uuid_t	uuid;
716 		__uuid_t	user_uuid;
717 
718 		u16		version;
719 		u16		version_min;
720 		u16		version_upgrade_complete;
721 
722 		u8		nr_devices;
723 		u8		clean;
724 
725 		u8		encryption_type;
726 
727 		u64		time_base_lo;
728 		u32		time_base_hi;
729 		unsigned	time_units_per_sec;
730 		unsigned	nsec_per_time_unit;
731 		u64		features;
732 		u64		compat;
733 	}			sb;
734 
735 
736 	struct bch_sb_handle	disk_sb;
737 
738 	unsigned short		block_bits;	/* ilog2(block_size) */
739 
740 	u16			btree_foreground_merge_threshold;
741 
742 	struct closure		sb_write;
743 	struct mutex		sb_lock;
744 
745 	/* snapshot.c: */
746 	struct snapshot_table __rcu *snapshots;
747 	size_t			snapshot_table_size;
748 	struct mutex		snapshot_table_lock;
749 	struct rw_semaphore	snapshot_create_lock;
750 
751 	struct work_struct	snapshot_delete_work;
752 	struct work_struct	snapshot_wait_for_pagecache_and_delete_work;
753 	snapshot_id_list	snapshots_unlinked;
754 	struct mutex		snapshots_unlinked_lock;
755 
756 	/* BTREE CACHE */
757 	struct bio_set		btree_bio;
758 	struct workqueue_struct	*io_complete_wq;
759 
760 	struct btree_root	btree_roots_known[BTREE_ID_NR];
761 	DARRAY(struct btree_root) btree_roots_extra;
762 	struct mutex		btree_root_lock;
763 
764 	struct btree_cache	btree_cache;
765 
766 	/*
767 	 * Cache of allocated btree nodes - if we allocate a btree node and
768 	 * don't use it, if we free it that space can't be reused until going
769 	 * _all_ the way through the allocator (which exposes us to a livelock
770 	 * when allocating btree reserves fail halfway through) - instead, we
771 	 * can stick them here:
772 	 */
773 	struct btree_alloc	btree_reserve_cache[BTREE_NODE_RESERVE * 2];
774 	unsigned		btree_reserve_cache_nr;
775 	struct mutex		btree_reserve_cache_lock;
776 
777 	mempool_t		btree_interior_update_pool;
778 	struct list_head	btree_interior_update_list;
779 	struct list_head	btree_interior_updates_unwritten;
780 	struct mutex		btree_interior_update_lock;
781 	struct closure_waitlist	btree_interior_update_wait;
782 
783 	struct workqueue_struct	*btree_interior_update_worker;
784 	struct work_struct	btree_interior_update_work;
785 
786 	struct list_head	pending_node_rewrites;
787 	struct mutex		pending_node_rewrites_lock;
788 
789 	/* btree_io.c: */
790 	spinlock_t		btree_write_error_lock;
791 	struct btree_write_stats {
792 		atomic64_t	nr;
793 		atomic64_t	bytes;
794 	}			btree_write_stats[BTREE_WRITE_TYPE_NR];
795 
796 	/* btree_iter.c: */
797 	struct seqmutex		btree_trans_lock;
798 	struct list_head	btree_trans_list;
799 	mempool_t		btree_trans_pool;
800 	mempool_t		btree_trans_mem_pool;
801 	struct btree_trans_buf  __percpu	*btree_trans_bufs;
802 
803 	struct srcu_struct	btree_trans_barrier;
804 	bool			btree_trans_barrier_initialized;
805 
806 	struct btree_key_cache	btree_key_cache;
807 	unsigned		btree_key_cache_btrees;
808 
809 	struct btree_write_buffer btree_write_buffer;
810 
811 	struct workqueue_struct	*btree_update_wq;
812 	struct workqueue_struct	*btree_io_complete_wq;
813 	/* copygc needs its own workqueue for index updates.. */
814 	struct workqueue_struct	*copygc_wq;
815 	/*
816 	 * Use a dedicated wq for write ref holder tasks. Required to avoid
817 	 * dependency problems with other wq tasks that can block on ref
818 	 * draining, such as read-only transition.
819 	 */
820 	struct workqueue_struct *write_ref_wq;
821 
822 	/* ALLOCATION */
823 	struct bch_devs_mask	rw_devs[BCH_DATA_NR];
824 
825 	u64			capacity; /* sectors */
826 
827 	/*
828 	 * When capacity _decreases_ (due to a disk being removed), we
829 	 * increment capacity_gen - this invalidates outstanding reservations
830 	 * and forces them to be revalidated
831 	 */
832 	u32			capacity_gen;
833 	unsigned		bucket_size_max;
834 
835 	atomic64_t		sectors_available;
836 	struct mutex		sectors_available_lock;
837 
838 	struct bch_fs_pcpu __percpu	*pcpu;
839 
840 	struct percpu_rw_semaphore	mark_lock;
841 
842 	seqcount_t			usage_lock;
843 	struct bch_fs_usage		*usage_base;
844 	struct bch_fs_usage __percpu	*usage[JOURNAL_BUF_NR];
845 	struct bch_fs_usage __percpu	*usage_gc;
846 	u64 __percpu		*online_reserved;
847 
848 	/* single element mempool: */
849 	struct mutex		usage_scratch_lock;
850 	struct bch_fs_usage_online *usage_scratch;
851 
852 	struct io_clock		io_clock[2];
853 
854 	/* JOURNAL SEQ BLACKLIST */
855 	struct journal_seq_blacklist_table *
856 				journal_seq_blacklist_table;
857 	struct work_struct	journal_seq_blacklist_gc_work;
858 
859 	/* ALLOCATOR */
860 	spinlock_t		freelist_lock;
861 	struct closure_waitlist	freelist_wait;
862 	u64			blocked_allocate;
863 	u64			blocked_allocate_open_bucket;
864 
865 	open_bucket_idx_t	open_buckets_freelist;
866 	open_bucket_idx_t	open_buckets_nr_free;
867 	struct closure_waitlist	open_buckets_wait;
868 	struct open_bucket	open_buckets[OPEN_BUCKETS_COUNT];
869 	open_bucket_idx_t	open_buckets_hash[OPEN_BUCKETS_COUNT];
870 
871 	open_bucket_idx_t	open_buckets_partial[OPEN_BUCKETS_COUNT];
872 	open_bucket_idx_t	open_buckets_partial_nr;
873 
874 	struct write_point	btree_write_point;
875 	struct write_point	rebalance_write_point;
876 
877 	struct write_point	write_points[WRITE_POINT_MAX];
878 	struct hlist_head	write_points_hash[WRITE_POINT_HASH_NR];
879 	struct mutex		write_points_hash_lock;
880 	unsigned		write_points_nr;
881 
882 	struct buckets_waiting_for_journal buckets_waiting_for_journal;
883 	struct work_struct	discard_work;
884 	struct work_struct	invalidate_work;
885 
886 	/* GARBAGE COLLECTION */
887 	struct task_struct	*gc_thread;
888 	atomic_t		kick_gc;
889 	unsigned long		gc_count;
890 
891 	enum btree_id		gc_gens_btree;
892 	struct bpos		gc_gens_pos;
893 
894 	/*
895 	 * Tracks GC's progress - everything in the range [ZERO_KEY..gc_cur_pos]
896 	 * has been marked by GC.
897 	 *
898 	 * gc_cur_phase is a superset of btree_ids (BTREE_ID_extents etc.)
899 	 *
900 	 * Protected by gc_pos_lock. Only written to by GC thread, so GC thread
901 	 * can read without a lock.
902 	 */
903 	seqcount_t		gc_pos_lock;
904 	struct gc_pos		gc_pos;
905 
906 	/*
907 	 * The allocation code needs gc_mark in struct bucket to be correct, but
908 	 * it's not while a gc is in progress.
909 	 */
910 	struct rw_semaphore	gc_lock;
911 	struct mutex		gc_gens_lock;
912 
913 	/* IO PATH */
914 	struct semaphore	io_in_flight;
915 	struct bio_set		bio_read;
916 	struct bio_set		bio_read_split;
917 	struct bio_set		bio_write;
918 	struct mutex		bio_bounce_pages_lock;
919 	mempool_t		bio_bounce_pages;
920 	struct bucket_nocow_lock_table
921 				nocow_locks;
922 	struct rhashtable	promote_table;
923 
924 	mempool_t		compression_bounce[2];
925 	mempool_t		compress_workspace[BCH_COMPRESSION_TYPE_NR];
926 	mempool_t		decompress_workspace;
927 	ZSTD_parameters		zstd_params;
928 
929 	struct crypto_shash	*sha256;
930 	struct crypto_sync_skcipher *chacha20;
931 	struct crypto_shash	*poly1305;
932 
933 	atomic64_t		key_version;
934 
935 	mempool_t		large_bkey_pool;
936 
937 	/* MOVE.C */
938 	struct list_head	moving_context_list;
939 	struct mutex		moving_context_lock;
940 
941 	struct list_head	data_progress_list;
942 	struct mutex		data_progress_lock;
943 
944 	/* REBALANCE */
945 	struct bch_fs_rebalance	rebalance;
946 
947 	/* COPYGC */
948 	struct task_struct	*copygc_thread;
949 	struct write_point	copygc_write_point;
950 	s64			copygc_wait_at;
951 	s64			copygc_wait;
952 	bool			copygc_running;
953 	wait_queue_head_t	copygc_running_wq;
954 
955 	/* STRIPES: */
956 	GENRADIX(struct stripe) stripes;
957 	GENRADIX(struct gc_stripe) gc_stripes;
958 
959 	struct hlist_head	ec_stripes_new[32];
960 	spinlock_t		ec_stripes_new_lock;
961 
962 	ec_stripes_heap		ec_stripes_heap;
963 	struct mutex		ec_stripes_heap_lock;
964 
965 	/* ERASURE CODING */
966 	struct list_head	ec_stripe_head_list;
967 	struct mutex		ec_stripe_head_lock;
968 
969 	struct list_head	ec_stripe_new_list;
970 	struct mutex		ec_stripe_new_lock;
971 	wait_queue_head_t	ec_stripe_new_wait;
972 
973 	struct work_struct	ec_stripe_create_work;
974 	u64			ec_stripe_hint;
975 
976 	struct work_struct	ec_stripe_delete_work;
977 
978 	struct bio_set		ec_bioset;
979 
980 	/* REFLINK */
981 	reflink_gc_table	reflink_gc_table;
982 	size_t			reflink_gc_nr;
983 
984 	/* fs.c */
985 	struct list_head	vfs_inodes_list;
986 	struct mutex		vfs_inodes_lock;
987 
988 	/* VFS IO PATH - fs-io.c */
989 	struct bio_set		writepage_bioset;
990 	struct bio_set		dio_write_bioset;
991 	struct bio_set		dio_read_bioset;
992 	struct bio_set		nocow_flush_bioset;
993 
994 	/* ERRORS */
995 	struct list_head	fsck_errors;
996 	struct mutex		fsck_error_lock;
997 	bool			fsck_alloc_err;
998 
999 	/* QUOTAS */
1000 	struct bch_memquota_type quotas[QTYP_NR];
1001 
1002 	/* RECOVERY */
1003 	u64			journal_replay_seq_start;
1004 	u64			journal_replay_seq_end;
1005 	enum bch_recovery_pass	curr_recovery_pass;
1006 	/* bitmap of explicitly enabled recovery passes: */
1007 	u64			recovery_passes_explicit;
1008 	u64			recovery_passes_complete;
1009 
1010 	/* DEBUG JUNK */
1011 	struct dentry		*fs_debug_dir;
1012 	struct dentry		*btree_debug_dir;
1013 	struct btree_debug	btree_debug[BTREE_ID_NR];
1014 	struct btree		*verify_data;
1015 	struct btree_node	*verify_ondisk;
1016 	struct mutex		verify_lock;
1017 
1018 	u64			*unused_inode_hints;
1019 	unsigned		inode_shard_bits;
1020 
1021 	/*
1022 	 * A btree node on disk could have too many bsets for an iterator to fit
1023 	 * on the stack - have to dynamically allocate them
1024 	 */
1025 	mempool_t		fill_iter;
1026 
1027 	mempool_t		btree_bounce_pool;
1028 
1029 	struct journal		journal;
1030 	GENRADIX(struct journal_replay *) journal_entries;
1031 	u64			journal_entries_base_seq;
1032 	struct journal_keys	journal_keys;
1033 	struct list_head	journal_iters;
1034 
1035 	u64			last_bucket_seq_cleanup;
1036 
1037 	u64			counters_on_mount[BCH_COUNTER_NR];
1038 	u64 __percpu		*counters;
1039 
1040 	unsigned		btree_gc_periodic:1;
1041 	unsigned		copy_gc_enabled:1;
1042 	bool			promote_whole_extents;
1043 
1044 	struct bch2_time_stats	times[BCH_TIME_STAT_NR];
1045 
1046 	struct btree_transaction_stats btree_transaction_stats[BCH_TRANSACTIONS_NR];
1047 };
1048 
1049 extern struct wait_queue_head bch2_read_only_wait;
1050 
1051 static inline void bch2_write_ref_get(struct bch_fs *c, enum bch_write_ref ref)
1052 {
1053 #ifdef BCH_WRITE_REF_DEBUG
1054 	atomic_long_inc(&c->writes[ref]);
1055 #else
1056 	percpu_ref_get(&c->writes);
1057 #endif
1058 }
1059 
1060 static inline bool bch2_write_ref_tryget(struct bch_fs *c, enum bch_write_ref ref)
1061 {
1062 #ifdef BCH_WRITE_REF_DEBUG
1063 	return !test_bit(BCH_FS_GOING_RO, &c->flags) &&
1064 		atomic_long_inc_not_zero(&c->writes[ref]);
1065 #else
1066 	return percpu_ref_tryget_live(&c->writes);
1067 #endif
1068 }
1069 
1070 static inline void bch2_write_ref_put(struct bch_fs *c, enum bch_write_ref ref)
1071 {
1072 #ifdef BCH_WRITE_REF_DEBUG
1073 	long v = atomic_long_dec_return(&c->writes[ref]);
1074 
1075 	BUG_ON(v < 0);
1076 	if (v)
1077 		return;
1078 	for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
1079 		if (atomic_long_read(&c->writes[i]))
1080 			return;
1081 
1082 	set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
1083 	wake_up(&bch2_read_only_wait);
1084 #else
1085 	percpu_ref_put(&c->writes);
1086 #endif
1087 }
1088 
1089 static inline void bch2_set_ra_pages(struct bch_fs *c, unsigned ra_pages)
1090 {
1091 #ifndef NO_BCACHEFS_FS
1092 	if (c->vfs_sb)
1093 		c->vfs_sb->s_bdi->ra_pages = ra_pages;
1094 #endif
1095 }
1096 
1097 static inline unsigned bucket_bytes(const struct bch_dev *ca)
1098 {
1099 	return ca->mi.bucket_size << 9;
1100 }
1101 
1102 static inline unsigned block_bytes(const struct bch_fs *c)
1103 {
1104 	return c->opts.block_size;
1105 }
1106 
1107 static inline unsigned block_sectors(const struct bch_fs *c)
1108 {
1109 	return c->opts.block_size >> 9;
1110 }
1111 
1112 static inline size_t btree_sectors(const struct bch_fs *c)
1113 {
1114 	return c->opts.btree_node_size >> 9;
1115 }
1116 
1117 static inline bool btree_id_cached(const struct bch_fs *c, enum btree_id btree)
1118 {
1119 	return c->btree_key_cache_btrees & (1U << btree);
1120 }
1121 
1122 static inline struct timespec64 bch2_time_to_timespec(const struct bch_fs *c, s64 time)
1123 {
1124 	struct timespec64 t;
1125 	s32 rem;
1126 
1127 	time += c->sb.time_base_lo;
1128 
1129 	t.tv_sec = div_s64_rem(time, c->sb.time_units_per_sec, &rem);
1130 	t.tv_nsec = rem * c->sb.nsec_per_time_unit;
1131 	return t;
1132 }
1133 
1134 static inline s64 timespec_to_bch2_time(const struct bch_fs *c, struct timespec64 ts)
1135 {
1136 	return (ts.tv_sec * c->sb.time_units_per_sec +
1137 		(int) ts.tv_nsec / c->sb.nsec_per_time_unit) - c->sb.time_base_lo;
1138 }
1139 
1140 static inline s64 bch2_current_time(const struct bch_fs *c)
1141 {
1142 	struct timespec64 now;
1143 
1144 	ktime_get_coarse_real_ts64(&now);
1145 	return timespec_to_bch2_time(c, now);
1146 }
1147 
1148 static inline bool bch2_dev_exists2(const struct bch_fs *c, unsigned dev)
1149 {
1150 	return dev < c->sb.nr_devices && c->devs[dev];
1151 }
1152 
1153 #define BKEY_PADDED_ONSTACK(key, pad)				\
1154 	struct { struct bkey_i key; __u64 key ## _pad[pad]; }
1155 
1156 #endif /* _BCACHEFS_H */
1157