xref: /openbmc/linux/drivers/md/bcache/bcache.h (revision 7e6f7d24)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _BCACHE_H
3 #define _BCACHE_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 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 #define pr_fmt(fmt) "bcache: %s() " fmt "\n", __func__
180 
181 #include <linux/bcache.h>
182 #include <linux/bio.h>
183 #include <linux/kobject.h>
184 #include <linux/list.h>
185 #include <linux/mutex.h>
186 #include <linux/rbtree.h>
187 #include <linux/rwsem.h>
188 #include <linux/refcount.h>
189 #include <linux/types.h>
190 #include <linux/workqueue.h>
191 #include <linux/kthread.h>
192 
193 #include "bset.h"
194 #include "util.h"
195 #include "closure.h"
196 
197 struct bucket {
198 	atomic_t	pin;
199 	uint16_t	prio;
200 	uint8_t		gen;
201 	uint8_t		last_gc; /* Most out of date gen in the btree */
202 	uint16_t	gc_mark; /* Bitfield used by GC. See below for field */
203 };
204 
205 /*
206  * I'd use bitfields for these, but I don't trust the compiler not to screw me
207  * as multiple threads touch struct bucket without locking
208  */
209 
210 BITMASK(GC_MARK,	 struct bucket, gc_mark, 0, 2);
211 #define GC_MARK_RECLAIMABLE	1
212 #define GC_MARK_DIRTY		2
213 #define GC_MARK_METADATA	3
214 #define GC_SECTORS_USED_SIZE	13
215 #define MAX_GC_SECTORS_USED	(~(~0ULL << GC_SECTORS_USED_SIZE))
216 BITMASK(GC_SECTORS_USED, struct bucket, gc_mark, 2, GC_SECTORS_USED_SIZE);
217 BITMASK(GC_MOVE, struct bucket, gc_mark, 15, 1);
218 
219 #include "journal.h"
220 #include "stats.h"
221 struct search;
222 struct btree;
223 struct keybuf;
224 
225 struct keybuf_key {
226 	struct rb_node		node;
227 	BKEY_PADDED(key);
228 	void			*private;
229 };
230 
231 struct keybuf {
232 	struct bkey		last_scanned;
233 	spinlock_t		lock;
234 
235 	/*
236 	 * Beginning and end of range in rb tree - so that we can skip taking
237 	 * lock and checking the rb tree when we need to check for overlapping
238 	 * keys.
239 	 */
240 	struct bkey		start;
241 	struct bkey		end;
242 
243 	struct rb_root		keys;
244 
245 #define KEYBUF_NR		500
246 	DECLARE_ARRAY_ALLOCATOR(struct keybuf_key, freelist, KEYBUF_NR);
247 };
248 
249 struct bcache_device {
250 	struct closure		cl;
251 
252 	struct kobject		kobj;
253 
254 	struct cache_set	*c;
255 	unsigned		id;
256 #define BCACHEDEVNAME_SIZE	12
257 	char			name[BCACHEDEVNAME_SIZE];
258 
259 	struct gendisk		*disk;
260 
261 	unsigned long		flags;
262 #define BCACHE_DEV_CLOSING		0
263 #define BCACHE_DEV_DETACHING		1
264 #define BCACHE_DEV_UNLINK_DONE		2
265 #define BCACHE_DEV_WB_RUNNING		3
266 #define BCACHE_DEV_RATE_DW_RUNNING	4
267 	unsigned		nr_stripes;
268 	unsigned		stripe_size;
269 	atomic_t		*stripe_sectors_dirty;
270 	unsigned long		*full_dirty_stripes;
271 
272 	struct bio_set		bio_split;
273 
274 	unsigned		data_csum:1;
275 
276 	int (*cache_miss)(struct btree *, struct search *,
277 			  struct bio *, unsigned);
278 	int (*ioctl) (struct bcache_device *, fmode_t, unsigned, unsigned long);
279 };
280 
281 struct io {
282 	/* Used to track sequential IO so it can be skipped */
283 	struct hlist_node	hash;
284 	struct list_head	lru;
285 
286 	unsigned long		jiffies;
287 	unsigned		sequential;
288 	sector_t		last;
289 };
290 
291 enum stop_on_failure {
292 	BCH_CACHED_DEV_STOP_AUTO = 0,
293 	BCH_CACHED_DEV_STOP_ALWAYS,
294 	BCH_CACHED_DEV_STOP_MODE_MAX,
295 };
296 
297 struct cached_dev {
298 	struct list_head	list;
299 	struct bcache_device	disk;
300 	struct block_device	*bdev;
301 
302 	struct cache_sb		sb;
303 	struct bio		sb_bio;
304 	struct bio_vec		sb_bv[1];
305 	struct closure		sb_write;
306 	struct semaphore	sb_write_mutex;
307 
308 	/* Refcount on the cache set. Always nonzero when we're caching. */
309 	refcount_t		count;
310 	struct work_struct	detach;
311 
312 	/*
313 	 * Device might not be running if it's dirty and the cache set hasn't
314 	 * showed up yet.
315 	 */
316 	atomic_t		running;
317 
318 	/*
319 	 * Writes take a shared lock from start to finish; scanning for dirty
320 	 * data to refill the rb tree requires an exclusive lock.
321 	 */
322 	struct rw_semaphore	writeback_lock;
323 
324 	/*
325 	 * Nonzero, and writeback has a refcount (d->count), iff there is dirty
326 	 * data in the cache. Protected by writeback_lock; must have an
327 	 * shared lock to set and exclusive lock to clear.
328 	 */
329 	atomic_t		has_dirty;
330 
331 	/*
332 	 * Set to zero by things that touch the backing volume-- except
333 	 * writeback.  Incremented by writeback.  Used to determine when to
334 	 * accelerate idle writeback.
335 	 */
336 	atomic_t		backing_idle;
337 
338 	struct bch_ratelimit	writeback_rate;
339 	struct delayed_work	writeback_rate_update;
340 
341 	/* Limit number of writeback bios in flight */
342 	struct semaphore	in_flight;
343 	struct task_struct	*writeback_thread;
344 	struct workqueue_struct	*writeback_write_wq;
345 
346 	struct keybuf		writeback_keys;
347 
348 	struct task_struct	*status_update_thread;
349 	/*
350 	 * Order the write-half of writeback operations strongly in dispatch
351 	 * order.  (Maintain LBA order; don't allow reads completing out of
352 	 * order to re-order the writes...)
353 	 */
354 	struct closure_waitlist writeback_ordering_wait;
355 	atomic_t		writeback_sequence_next;
356 
357 	/* For tracking sequential IO */
358 #define RECENT_IO_BITS	7
359 #define RECENT_IO	(1 << RECENT_IO_BITS)
360 	struct io		io[RECENT_IO];
361 	struct hlist_head	io_hash[RECENT_IO + 1];
362 	struct list_head	io_lru;
363 	spinlock_t		io_lock;
364 
365 	struct cache_accounting	accounting;
366 
367 	/* The rest of this all shows up in sysfs */
368 	unsigned		sequential_cutoff;
369 	unsigned		readahead;
370 
371 	unsigned		io_disable:1;
372 	unsigned		verify:1;
373 	unsigned		bypass_torture_test:1;
374 
375 	unsigned		partial_stripes_expensive:1;
376 	unsigned		writeback_metadata:1;
377 	unsigned		writeback_running:1;
378 	unsigned char		writeback_percent;
379 	unsigned		writeback_delay;
380 
381 	uint64_t		writeback_rate_target;
382 	int64_t			writeback_rate_proportional;
383 	int64_t			writeback_rate_integral;
384 	int64_t			writeback_rate_integral_scaled;
385 	int32_t			writeback_rate_change;
386 
387 	unsigned		writeback_rate_update_seconds;
388 	unsigned		writeback_rate_i_term_inverse;
389 	unsigned		writeback_rate_p_term_inverse;
390 	unsigned		writeback_rate_minimum;
391 
392 	enum stop_on_failure	stop_when_cache_set_failed;
393 #define DEFAULT_CACHED_DEV_ERROR_LIMIT	64
394 	atomic_t		io_errors;
395 	unsigned		error_limit;
396 	unsigned		offline_seconds;
397 
398 	char			backing_dev_name[BDEVNAME_SIZE];
399 };
400 
401 enum alloc_reserve {
402 	RESERVE_BTREE,
403 	RESERVE_PRIO,
404 	RESERVE_MOVINGGC,
405 	RESERVE_NONE,
406 	RESERVE_NR,
407 };
408 
409 struct cache {
410 	struct cache_set	*set;
411 	struct cache_sb		sb;
412 	struct bio		sb_bio;
413 	struct bio_vec		sb_bv[1];
414 
415 	struct kobject		kobj;
416 	struct block_device	*bdev;
417 
418 	struct task_struct	*alloc_thread;
419 
420 	struct closure		prio;
421 	struct prio_set		*disk_buckets;
422 
423 	/*
424 	 * When allocating new buckets, prio_write() gets first dibs - since we
425 	 * may not be allocate at all without writing priorities and gens.
426 	 * prio_buckets[] contains the last buckets we wrote priorities to (so
427 	 * gc can mark them as metadata), prio_next[] contains the buckets
428 	 * allocated for the next prio write.
429 	 */
430 	uint64_t		*prio_buckets;
431 	uint64_t		*prio_last_buckets;
432 
433 	/*
434 	 * free: Buckets that are ready to be used
435 	 *
436 	 * free_inc: Incoming buckets - these are buckets that currently have
437 	 * cached data in them, and we can't reuse them until after we write
438 	 * their new gen to disk. After prio_write() finishes writing the new
439 	 * gens/prios, they'll be moved to the free list (and possibly discarded
440 	 * in the process)
441 	 */
442 	DECLARE_FIFO(long, free)[RESERVE_NR];
443 	DECLARE_FIFO(long, free_inc);
444 
445 	size_t			fifo_last_bucket;
446 
447 	/* Allocation stuff: */
448 	struct bucket		*buckets;
449 
450 	DECLARE_HEAP(struct bucket *, heap);
451 
452 	/*
453 	 * If nonzero, we know we aren't going to find any buckets to invalidate
454 	 * until a gc finishes - otherwise we could pointlessly burn a ton of
455 	 * cpu
456 	 */
457 	unsigned		invalidate_needs_gc;
458 
459 	bool			discard; /* Get rid of? */
460 
461 	struct journal_device	journal;
462 
463 	/* The rest of this all shows up in sysfs */
464 #define IO_ERROR_SHIFT		20
465 	atomic_t		io_errors;
466 	atomic_t		io_count;
467 
468 	atomic_long_t		meta_sectors_written;
469 	atomic_long_t		btree_sectors_written;
470 	atomic_long_t		sectors_written;
471 
472 	char			cache_dev_name[BDEVNAME_SIZE];
473 };
474 
475 struct gc_stat {
476 	size_t			nodes;
477 	size_t			key_bytes;
478 
479 	size_t			nkeys;
480 	uint64_t		data;	/* sectors */
481 	unsigned		in_use; /* percent */
482 };
483 
484 /*
485  * Flag bits, for how the cache set is shutting down, and what phase it's at:
486  *
487  * CACHE_SET_UNREGISTERING means we're not just shutting down, we're detaching
488  * all the backing devices first (their cached data gets invalidated, and they
489  * won't automatically reattach).
490  *
491  * CACHE_SET_STOPPING always gets set first when we're closing down a cache set;
492  * we'll continue to run normally for awhile with CACHE_SET_STOPPING set (i.e.
493  * flushing dirty data).
494  *
495  * CACHE_SET_RUNNING means all cache devices have been registered and journal
496  * replay is complete.
497  *
498  * CACHE_SET_IO_DISABLE is set when bcache is stopping the whold cache set, all
499  * external and internal I/O should be denied when this flag is set.
500  *
501  */
502 #define CACHE_SET_UNREGISTERING		0
503 #define	CACHE_SET_STOPPING		1
504 #define	CACHE_SET_RUNNING		2
505 #define CACHE_SET_IO_DISABLE		3
506 
507 struct cache_set {
508 	struct closure		cl;
509 
510 	struct list_head	list;
511 	struct kobject		kobj;
512 	struct kobject		internal;
513 	struct dentry		*debug;
514 	struct cache_accounting accounting;
515 
516 	unsigned long		flags;
517 
518 	struct cache_sb		sb;
519 
520 	struct cache		*cache[MAX_CACHES_PER_SET];
521 	struct cache		*cache_by_alloc[MAX_CACHES_PER_SET];
522 	int			caches_loaded;
523 
524 	struct bcache_device	**devices;
525 	unsigned		devices_max_used;
526 	struct list_head	cached_devs;
527 	uint64_t		cached_dev_sectors;
528 	struct closure		caching;
529 
530 	struct closure		sb_write;
531 	struct semaphore	sb_write_mutex;
532 
533 	mempool_t		search;
534 	mempool_t		bio_meta;
535 	struct bio_set		bio_split;
536 
537 	/* For the btree cache */
538 	struct shrinker		shrink;
539 
540 	/* For the btree cache and anything allocation related */
541 	struct mutex		bucket_lock;
542 
543 	/* log2(bucket_size), in sectors */
544 	unsigned short		bucket_bits;
545 
546 	/* log2(block_size), in sectors */
547 	unsigned short		block_bits;
548 
549 	/*
550 	 * Default number of pages for a new btree node - may be less than a
551 	 * full bucket
552 	 */
553 	unsigned		btree_pages;
554 
555 	/*
556 	 * Lists of struct btrees; lru is the list for structs that have memory
557 	 * allocated for actual btree node, freed is for structs that do not.
558 	 *
559 	 * We never free a struct btree, except on shutdown - we just put it on
560 	 * the btree_cache_freed list and reuse it later. This simplifies the
561 	 * code, and it doesn't cost us much memory as the memory usage is
562 	 * dominated by buffers that hold the actual btree node data and those
563 	 * can be freed - and the number of struct btrees allocated is
564 	 * effectively bounded.
565 	 *
566 	 * btree_cache_freeable effectively is a small cache - we use it because
567 	 * high order page allocations can be rather expensive, and it's quite
568 	 * common to delete and allocate btree nodes in quick succession. It
569 	 * should never grow past ~2-3 nodes in practice.
570 	 */
571 	struct list_head	btree_cache;
572 	struct list_head	btree_cache_freeable;
573 	struct list_head	btree_cache_freed;
574 
575 	/* Number of elements in btree_cache + btree_cache_freeable lists */
576 	unsigned		btree_cache_used;
577 
578 	/*
579 	 * If we need to allocate memory for a new btree node and that
580 	 * allocation fails, we can cannibalize another node in the btree cache
581 	 * to satisfy the allocation - lock to guarantee only one thread does
582 	 * this at a time:
583 	 */
584 	wait_queue_head_t	btree_cache_wait;
585 	struct task_struct	*btree_cache_alloc_lock;
586 
587 	/*
588 	 * When we free a btree node, we increment the gen of the bucket the
589 	 * node is in - but we can't rewrite the prios and gens until we
590 	 * finished whatever it is we were doing, otherwise after a crash the
591 	 * btree node would be freed but for say a split, we might not have the
592 	 * pointers to the new nodes inserted into the btree yet.
593 	 *
594 	 * This is a refcount that blocks prio_write() until the new keys are
595 	 * written.
596 	 */
597 	atomic_t		prio_blocked;
598 	wait_queue_head_t	bucket_wait;
599 
600 	/*
601 	 * For any bio we don't skip we subtract the number of sectors from
602 	 * rescale; when it hits 0 we rescale all the bucket priorities.
603 	 */
604 	atomic_t		rescale;
605 	/*
606 	 * When we invalidate buckets, we use both the priority and the amount
607 	 * of good data to determine which buckets to reuse first - to weight
608 	 * those together consistently we keep track of the smallest nonzero
609 	 * priority of any bucket.
610 	 */
611 	uint16_t		min_prio;
612 
613 	/*
614 	 * max(gen - last_gc) for all buckets. When it gets too big we have to gc
615 	 * to keep gens from wrapping around.
616 	 */
617 	uint8_t			need_gc;
618 	struct gc_stat		gc_stats;
619 	size_t			nbuckets;
620 	size_t			avail_nbuckets;
621 
622 	struct task_struct	*gc_thread;
623 	/* Where in the btree gc currently is */
624 	struct bkey		gc_done;
625 
626 	/*
627 	 * The allocation code needs gc_mark in struct bucket to be correct, but
628 	 * it's not while a gc is in progress. Protected by bucket_lock.
629 	 */
630 	int			gc_mark_valid;
631 
632 	/* Counts how many sectors bio_insert has added to the cache */
633 	atomic_t		sectors_to_gc;
634 	wait_queue_head_t	gc_wait;
635 
636 	struct keybuf		moving_gc_keys;
637 	/* Number of moving GC bios in flight */
638 	struct semaphore	moving_in_flight;
639 
640 	struct workqueue_struct	*moving_gc_wq;
641 
642 	struct btree		*root;
643 
644 #ifdef CONFIG_BCACHE_DEBUG
645 	struct btree		*verify_data;
646 	struct bset		*verify_ondisk;
647 	struct mutex		verify_lock;
648 #endif
649 
650 	unsigned		nr_uuids;
651 	struct uuid_entry	*uuids;
652 	BKEY_PADDED(uuid_bucket);
653 	struct closure		uuid_write;
654 	struct semaphore	uuid_write_mutex;
655 
656 	/*
657 	 * A btree node on disk could have too many bsets for an iterator to fit
658 	 * on the stack - have to dynamically allocate them
659 	 */
660 	mempool_t		fill_iter;
661 
662 	struct bset_sort_state	sort;
663 
664 	/* List of buckets we're currently writing data to */
665 	struct list_head	data_buckets;
666 	spinlock_t		data_bucket_lock;
667 
668 	struct journal		journal;
669 
670 #define CONGESTED_MAX		1024
671 	unsigned		congested_last_us;
672 	atomic_t		congested;
673 
674 	/* The rest of this all shows up in sysfs */
675 	unsigned		congested_read_threshold_us;
676 	unsigned		congested_write_threshold_us;
677 
678 	struct time_stats	btree_gc_time;
679 	struct time_stats	btree_split_time;
680 	struct time_stats	btree_read_time;
681 
682 	atomic_long_t		cache_read_races;
683 	atomic_long_t		writeback_keys_done;
684 	atomic_long_t		writeback_keys_failed;
685 
686 	atomic_long_t		reclaim;
687 	atomic_long_t		flush_write;
688 	atomic_long_t		retry_flush_write;
689 
690 	enum			{
691 		ON_ERROR_UNREGISTER,
692 		ON_ERROR_PANIC,
693 	}			on_error;
694 #define DEFAULT_IO_ERROR_LIMIT 8
695 	unsigned		error_limit;
696 	unsigned		error_decay;
697 
698 	unsigned short		journal_delay_ms;
699 	bool			expensive_debug_checks;
700 	unsigned		verify:1;
701 	unsigned		key_merging_disabled:1;
702 	unsigned		gc_always_rewrite:1;
703 	unsigned		shrinker_disabled:1;
704 	unsigned		copy_gc_enabled:1;
705 
706 #define BUCKET_HASH_BITS	12
707 	struct hlist_head	bucket_hash[1 << BUCKET_HASH_BITS];
708 
709 	DECLARE_HEAP(struct btree *, flush_btree);
710 };
711 
712 struct bbio {
713 	unsigned		submit_time_us;
714 	union {
715 		struct bkey	key;
716 		uint64_t	_pad[3];
717 		/*
718 		 * We only need pad = 3 here because we only ever carry around a
719 		 * single pointer - i.e. the pointer we're doing io to/from.
720 		 */
721 	};
722 	struct bio		bio;
723 };
724 
725 #define BTREE_PRIO		USHRT_MAX
726 #define INITIAL_PRIO		32768U
727 
728 #define btree_bytes(c)		((c)->btree_pages * PAGE_SIZE)
729 #define btree_blocks(b)							\
730 	((unsigned) (KEY_SIZE(&b->key) >> (b)->c->block_bits))
731 
732 #define btree_default_blocks(c)						\
733 	((unsigned) ((PAGE_SECTORS * (c)->btree_pages) >> (c)->block_bits))
734 
735 #define bucket_pages(c)		((c)->sb.bucket_size / PAGE_SECTORS)
736 #define bucket_bytes(c)		((c)->sb.bucket_size << 9)
737 #define block_bytes(c)		((c)->sb.block_size << 9)
738 
739 #define prios_per_bucket(c)				\
740 	((bucket_bytes(c) - sizeof(struct prio_set)) /	\
741 	 sizeof(struct bucket_disk))
742 #define prio_buckets(c)					\
743 	DIV_ROUND_UP((size_t) (c)->sb.nbuckets, prios_per_bucket(c))
744 
745 static inline size_t sector_to_bucket(struct cache_set *c, sector_t s)
746 {
747 	return s >> c->bucket_bits;
748 }
749 
750 static inline sector_t bucket_to_sector(struct cache_set *c, size_t b)
751 {
752 	return ((sector_t) b) << c->bucket_bits;
753 }
754 
755 static inline sector_t bucket_remainder(struct cache_set *c, sector_t s)
756 {
757 	return s & (c->sb.bucket_size - 1);
758 }
759 
760 static inline struct cache *PTR_CACHE(struct cache_set *c,
761 				      const struct bkey *k,
762 				      unsigned ptr)
763 {
764 	return c->cache[PTR_DEV(k, ptr)];
765 }
766 
767 static inline size_t PTR_BUCKET_NR(struct cache_set *c,
768 				   const struct bkey *k,
769 				   unsigned ptr)
770 {
771 	return sector_to_bucket(c, PTR_OFFSET(k, ptr));
772 }
773 
774 static inline struct bucket *PTR_BUCKET(struct cache_set *c,
775 					const struct bkey *k,
776 					unsigned ptr)
777 {
778 	return PTR_CACHE(c, k, ptr)->buckets + PTR_BUCKET_NR(c, k, ptr);
779 }
780 
781 static inline uint8_t gen_after(uint8_t a, uint8_t b)
782 {
783 	uint8_t r = a - b;
784 	return r > 128U ? 0 : r;
785 }
786 
787 static inline uint8_t ptr_stale(struct cache_set *c, const struct bkey *k,
788 				unsigned i)
789 {
790 	return gen_after(PTR_BUCKET(c, k, i)->gen, PTR_GEN(k, i));
791 }
792 
793 static inline bool ptr_available(struct cache_set *c, const struct bkey *k,
794 				 unsigned i)
795 {
796 	return (PTR_DEV(k, i) < MAX_CACHES_PER_SET) && PTR_CACHE(c, k, i);
797 }
798 
799 /* Btree key macros */
800 
801 /*
802  * This is used for various on disk data structures - cache_sb, prio_set, bset,
803  * jset: The checksum is _always_ the first 8 bytes of these structs
804  */
805 #define csum_set(i)							\
806 	bch_crc64(((void *) (i)) + sizeof(uint64_t),			\
807 		  ((void *) bset_bkey_last(i)) -			\
808 		  (((void *) (i)) + sizeof(uint64_t)))
809 
810 /* Error handling macros */
811 
812 #define btree_bug(b, ...)						\
813 do {									\
814 	if (bch_cache_set_error((b)->c, __VA_ARGS__))			\
815 		dump_stack();						\
816 } while (0)
817 
818 #define cache_bug(c, ...)						\
819 do {									\
820 	if (bch_cache_set_error(c, __VA_ARGS__))			\
821 		dump_stack();						\
822 } while (0)
823 
824 #define btree_bug_on(cond, b, ...)					\
825 do {									\
826 	if (cond)							\
827 		btree_bug(b, __VA_ARGS__);				\
828 } while (0)
829 
830 #define cache_bug_on(cond, c, ...)					\
831 do {									\
832 	if (cond)							\
833 		cache_bug(c, __VA_ARGS__);				\
834 } while (0)
835 
836 #define cache_set_err_on(cond, c, ...)					\
837 do {									\
838 	if (cond)							\
839 		bch_cache_set_error(c, __VA_ARGS__);			\
840 } while (0)
841 
842 /* Looping macros */
843 
844 #define for_each_cache(ca, cs, iter)					\
845 	for (iter = 0; ca = cs->cache[iter], iter < (cs)->sb.nr_in_set; iter++)
846 
847 #define for_each_bucket(b, ca)						\
848 	for (b = (ca)->buckets + (ca)->sb.first_bucket;			\
849 	     b < (ca)->buckets + (ca)->sb.nbuckets; b++)
850 
851 static inline void cached_dev_put(struct cached_dev *dc)
852 {
853 	if (refcount_dec_and_test(&dc->count))
854 		schedule_work(&dc->detach);
855 }
856 
857 static inline bool cached_dev_get(struct cached_dev *dc)
858 {
859 	if (!refcount_inc_not_zero(&dc->count))
860 		return false;
861 
862 	/* Paired with the mb in cached_dev_attach */
863 	smp_mb__after_atomic();
864 	return true;
865 }
866 
867 /*
868  * bucket_gc_gen() returns the difference between the bucket's current gen and
869  * the oldest gen of any pointer into that bucket in the btree (last_gc).
870  */
871 
872 static inline uint8_t bucket_gc_gen(struct bucket *b)
873 {
874 	return b->gen - b->last_gc;
875 }
876 
877 #define BUCKET_GC_GEN_MAX	96U
878 
879 #define kobj_attribute_write(n, fn)					\
880 	static struct kobj_attribute ksysfs_##n = __ATTR(n, S_IWUSR, NULL, fn)
881 
882 #define kobj_attribute_rw(n, show, store)				\
883 	static struct kobj_attribute ksysfs_##n =			\
884 		__ATTR(n, S_IWUSR|S_IRUSR, show, store)
885 
886 static inline void wake_up_allocators(struct cache_set *c)
887 {
888 	struct cache *ca;
889 	unsigned i;
890 
891 	for_each_cache(ca, c, i)
892 		wake_up_process(ca->alloc_thread);
893 }
894 
895 static inline void closure_bio_submit(struct cache_set *c,
896 				      struct bio *bio,
897 				      struct closure *cl)
898 {
899 	closure_get(cl);
900 	if (unlikely(test_bit(CACHE_SET_IO_DISABLE, &c->flags))) {
901 		bio->bi_status = BLK_STS_IOERR;
902 		bio_endio(bio);
903 		return;
904 	}
905 	generic_make_request(bio);
906 }
907 
908 /*
909  * Prevent the kthread exits directly, and make sure when kthread_stop()
910  * is called to stop a kthread, it is still alive. If a kthread might be
911  * stopped by CACHE_SET_IO_DISABLE bit set, wait_for_kthread_stop() is
912  * necessary before the kthread returns.
913  */
914 static inline void wait_for_kthread_stop(void)
915 {
916 	while (!kthread_should_stop()) {
917 		set_current_state(TASK_INTERRUPTIBLE);
918 		schedule();
919 	}
920 }
921 
922 /* Forward declarations */
923 
924 void bch_count_backing_io_errors(struct cached_dev *dc, struct bio *bio);
925 void bch_count_io_errors(struct cache *, blk_status_t, int, const char *);
926 void bch_bbio_count_io_errors(struct cache_set *, struct bio *,
927 			      blk_status_t, const char *);
928 void bch_bbio_endio(struct cache_set *, struct bio *, blk_status_t,
929 		const char *);
930 void bch_bbio_free(struct bio *, struct cache_set *);
931 struct bio *bch_bbio_alloc(struct cache_set *);
932 
933 void __bch_submit_bbio(struct bio *, struct cache_set *);
934 void bch_submit_bbio(struct bio *, struct cache_set *, struct bkey *, unsigned);
935 
936 uint8_t bch_inc_gen(struct cache *, struct bucket *);
937 void bch_rescale_priorities(struct cache_set *, int);
938 
939 bool bch_can_invalidate_bucket(struct cache *, struct bucket *);
940 void __bch_invalidate_one_bucket(struct cache *, struct bucket *);
941 
942 void __bch_bucket_free(struct cache *, struct bucket *);
943 void bch_bucket_free(struct cache_set *, struct bkey *);
944 
945 long bch_bucket_alloc(struct cache *, unsigned, bool);
946 int __bch_bucket_alloc_set(struct cache_set *, unsigned,
947 			   struct bkey *, int, bool);
948 int bch_bucket_alloc_set(struct cache_set *, unsigned,
949 			 struct bkey *, int, bool);
950 bool bch_alloc_sectors(struct cache_set *, struct bkey *, unsigned,
951 		       unsigned, unsigned, bool);
952 bool bch_cached_dev_error(struct cached_dev *dc);
953 
954 __printf(2, 3)
955 bool bch_cache_set_error(struct cache_set *, const char *, ...);
956 
957 void bch_prio_write(struct cache *);
958 void bch_write_bdev_super(struct cached_dev *, struct closure *);
959 
960 extern struct workqueue_struct *bcache_wq;
961 extern struct mutex bch_register_lock;
962 extern struct list_head bch_cache_sets;
963 
964 extern struct kobj_type bch_cached_dev_ktype;
965 extern struct kobj_type bch_flash_dev_ktype;
966 extern struct kobj_type bch_cache_set_ktype;
967 extern struct kobj_type bch_cache_set_internal_ktype;
968 extern struct kobj_type bch_cache_ktype;
969 
970 void bch_cached_dev_release(struct kobject *);
971 void bch_flash_dev_release(struct kobject *);
972 void bch_cache_set_release(struct kobject *);
973 void bch_cache_release(struct kobject *);
974 
975 int bch_uuid_write(struct cache_set *);
976 void bcache_write_super(struct cache_set *);
977 
978 int bch_flash_dev_create(struct cache_set *c, uint64_t size);
979 
980 int bch_cached_dev_attach(struct cached_dev *, struct cache_set *, uint8_t *);
981 void bch_cached_dev_detach(struct cached_dev *);
982 void bch_cached_dev_run(struct cached_dev *);
983 void bcache_device_stop(struct bcache_device *);
984 
985 void bch_cache_set_unregister(struct cache_set *);
986 void bch_cache_set_stop(struct cache_set *);
987 
988 struct cache_set *bch_cache_set_alloc(struct cache_sb *);
989 void bch_btree_cache_free(struct cache_set *);
990 int bch_btree_cache_alloc(struct cache_set *);
991 void bch_moving_init_cache_set(struct cache_set *);
992 int bch_open_buckets_alloc(struct cache_set *);
993 void bch_open_buckets_free(struct cache_set *);
994 
995 int bch_cache_allocator_start(struct cache *ca);
996 
997 void bch_debug_exit(void);
998 int bch_debug_init(struct kobject *);
999 void bch_request_exit(void);
1000 int bch_request_init(void);
1001 
1002 #endif /* _BCACHE_H */
1003