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 int 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 int nr_stripes;
268 unsigned int stripe_size;
269 atomic_t *stripe_sectors_dirty;
270 unsigned long *full_dirty_stripes;
271
272 struct bio_set bio_split;
273
274 unsigned int data_csum:1;
275
276 int (*cache_miss)(struct btree *b, struct search *s,
277 struct bio *bio, unsigned int sectors);
278 int (*ioctl)(struct bcache_device *d, fmode_t mode,
279 unsigned int cmd, unsigned long arg);
280 };
281
282 struct io {
283 /* Used to track sequential IO so it can be skipped */
284 struct hlist_node hash;
285 struct list_head lru;
286
287 unsigned long jiffies;
288 unsigned int sequential;
289 sector_t last;
290 };
291
292 enum stop_on_failure {
293 BCH_CACHED_DEV_STOP_AUTO = 0,
294 BCH_CACHED_DEV_STOP_ALWAYS,
295 BCH_CACHED_DEV_STOP_MODE_MAX,
296 };
297
298 struct cached_dev {
299 struct list_head list;
300 struct bcache_device disk;
301 struct block_device *bdev;
302
303 struct cache_sb sb;
304 struct bio sb_bio;
305 struct bio_vec sb_bv[1];
306 struct closure sb_write;
307 struct semaphore sb_write_mutex;
308
309 /* Refcount on the cache set. Always nonzero when we're caching. */
310 refcount_t count;
311 struct work_struct detach;
312
313 /*
314 * Device might not be running if it's dirty and the cache set hasn't
315 * showed up yet.
316 */
317 atomic_t running;
318
319 /*
320 * Writes take a shared lock from start to finish; scanning for dirty
321 * data to refill the rb tree requires an exclusive lock.
322 */
323 struct rw_semaphore writeback_lock;
324
325 /*
326 * Nonzero, and writeback has a refcount (d->count), iff there is dirty
327 * data in the cache. Protected by writeback_lock; must have an
328 * shared lock to set and exclusive lock to clear.
329 */
330 atomic_t has_dirty;
331
332 #define BCH_CACHE_READA_ALL 0
333 #define BCH_CACHE_READA_META_ONLY 1
334 unsigned int cache_readahead_policy;
335 struct bch_ratelimit writeback_rate;
336 struct delayed_work writeback_rate_update;
337
338 /* Limit number of writeback bios in flight */
339 struct semaphore in_flight;
340 struct task_struct *writeback_thread;
341 struct workqueue_struct *writeback_write_wq;
342
343 struct keybuf writeback_keys;
344
345 struct task_struct *status_update_thread;
346 /*
347 * Order the write-half of writeback operations strongly in dispatch
348 * order. (Maintain LBA order; don't allow reads completing out of
349 * order to re-order the writes...)
350 */
351 struct closure_waitlist writeback_ordering_wait;
352 atomic_t writeback_sequence_next;
353
354 /* For tracking sequential IO */
355 #define RECENT_IO_BITS 7
356 #define RECENT_IO (1 << RECENT_IO_BITS)
357 struct io io[RECENT_IO];
358 struct hlist_head io_hash[RECENT_IO + 1];
359 struct list_head io_lru;
360 spinlock_t io_lock;
361
362 struct cache_accounting accounting;
363
364 /* The rest of this all shows up in sysfs */
365 unsigned int sequential_cutoff;
366 unsigned int readahead;
367
368 unsigned int io_disable:1;
369 unsigned int verify:1;
370 unsigned int bypass_torture_test:1;
371
372 unsigned int partial_stripes_expensive:1;
373 unsigned int writeback_metadata:1;
374 unsigned int writeback_running:1;
375 unsigned char writeback_percent;
376 unsigned int writeback_delay;
377
378 uint64_t writeback_rate_target;
379 int64_t writeback_rate_proportional;
380 int64_t writeback_rate_integral;
381 int64_t writeback_rate_integral_scaled;
382 int32_t writeback_rate_change;
383
384 unsigned int writeback_rate_update_seconds;
385 unsigned int writeback_rate_i_term_inverse;
386 unsigned int writeback_rate_p_term_inverse;
387 unsigned int writeback_rate_minimum;
388
389 enum stop_on_failure stop_when_cache_set_failed;
390 #define DEFAULT_CACHED_DEV_ERROR_LIMIT 64
391 atomic_t io_errors;
392 unsigned int error_limit;
393 unsigned int offline_seconds;
394
395 char backing_dev_name[BDEVNAME_SIZE];
396 };
397
398 enum alloc_reserve {
399 RESERVE_BTREE,
400 RESERVE_PRIO,
401 RESERVE_MOVINGGC,
402 RESERVE_NONE,
403 RESERVE_NR,
404 };
405
406 struct cache {
407 struct cache_set *set;
408 struct cache_sb sb;
409 struct bio sb_bio;
410 struct bio_vec sb_bv[1];
411
412 struct kobject kobj;
413 struct block_device *bdev;
414
415 struct task_struct *alloc_thread;
416
417 struct closure prio;
418 struct prio_set *disk_buckets;
419
420 /*
421 * When allocating new buckets, prio_write() gets first dibs - since we
422 * may not be allocate at all without writing priorities and gens.
423 * prio_last_buckets[] contains the last buckets we wrote priorities to
424 * (so gc can mark them as metadata), prio_buckets[] contains the
425 * buckets allocated for the next prio write.
426 */
427 uint64_t *prio_buckets;
428 uint64_t *prio_last_buckets;
429
430 /*
431 * free: Buckets that are ready to be used
432 *
433 * free_inc: Incoming buckets - these are buckets that currently have
434 * cached data in them, and we can't reuse them until after we write
435 * their new gen to disk. After prio_write() finishes writing the new
436 * gens/prios, they'll be moved to the free list (and possibly discarded
437 * in the process)
438 */
439 DECLARE_FIFO(long, free)[RESERVE_NR];
440 DECLARE_FIFO(long, free_inc);
441
442 size_t fifo_last_bucket;
443
444 /* Allocation stuff: */
445 struct bucket *buckets;
446
447 DECLARE_HEAP(struct bucket *, heap);
448
449 /*
450 * If nonzero, we know we aren't going to find any buckets to invalidate
451 * until a gc finishes - otherwise we could pointlessly burn a ton of
452 * cpu
453 */
454 unsigned int invalidate_needs_gc;
455
456 bool discard; /* Get rid of? */
457
458 struct journal_device journal;
459
460 /* The rest of this all shows up in sysfs */
461 #define IO_ERROR_SHIFT 20
462 atomic_t io_errors;
463 atomic_t io_count;
464
465 atomic_long_t meta_sectors_written;
466 atomic_long_t btree_sectors_written;
467 atomic_long_t sectors_written;
468
469 char cache_dev_name[BDEVNAME_SIZE];
470 };
471
472 struct gc_stat {
473 size_t nodes;
474 size_t nodes_pre;
475 size_t key_bytes;
476
477 size_t nkeys;
478 uint64_t data; /* sectors */
479 unsigned int in_use; /* percent */
480 };
481
482 /*
483 * Flag bits, for how the cache set is shutting down, and what phase it's at:
484 *
485 * CACHE_SET_UNREGISTERING means we're not just shutting down, we're detaching
486 * all the backing devices first (their cached data gets invalidated, and they
487 * won't automatically reattach).
488 *
489 * CACHE_SET_STOPPING always gets set first when we're closing down a cache set;
490 * we'll continue to run normally for awhile with CACHE_SET_STOPPING set (i.e.
491 * flushing dirty data).
492 *
493 * CACHE_SET_RUNNING means all cache devices have been registered and journal
494 * replay is complete.
495 *
496 * CACHE_SET_IO_DISABLE is set when bcache is stopping the whold cache set, all
497 * external and internal I/O should be denied when this flag is set.
498 *
499 */
500 #define CACHE_SET_UNREGISTERING 0
501 #define CACHE_SET_STOPPING 1
502 #define CACHE_SET_RUNNING 2
503 #define CACHE_SET_IO_DISABLE 3
504
505 struct cache_set {
506 struct closure cl;
507
508 struct list_head list;
509 struct kobject kobj;
510 struct kobject internal;
511 struct dentry *debug;
512 struct cache_accounting accounting;
513
514 unsigned long flags;
515 atomic_t idle_counter;
516 atomic_t at_max_writeback_rate;
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 int devices_max_used;
526 atomic_t attached_dev_nr;
527 struct list_head cached_devs;
528 uint64_t cached_dev_sectors;
529 atomic_long_t flash_dev_dirty_sectors;
530 struct closure caching;
531
532 struct closure sb_write;
533 struct semaphore sb_write_mutex;
534
535 mempool_t search;
536 mempool_t bio_meta;
537 struct bio_set bio_split;
538
539 /* For the btree cache */
540 struct shrinker shrink;
541
542 /* For the btree cache and anything allocation related */
543 struct mutex bucket_lock;
544
545 /* log2(bucket_size), in sectors */
546 unsigned short bucket_bits;
547
548 /* log2(block_size), in sectors */
549 unsigned short block_bits;
550
551 /*
552 * Default number of pages for a new btree node - may be less than a
553 * full bucket
554 */
555 unsigned int btree_pages;
556
557 /*
558 * Lists of struct btrees; lru is the list for structs that have memory
559 * allocated for actual btree node, freed is for structs that do not.
560 *
561 * We never free a struct btree, except on shutdown - we just put it on
562 * the btree_cache_freed list and reuse it later. This simplifies the
563 * code, and it doesn't cost us much memory as the memory usage is
564 * dominated by buffers that hold the actual btree node data and those
565 * can be freed - and the number of struct btrees allocated is
566 * effectively bounded.
567 *
568 * btree_cache_freeable effectively is a small cache - we use it because
569 * high order page allocations can be rather expensive, and it's quite
570 * common to delete and allocate btree nodes in quick succession. It
571 * should never grow past ~2-3 nodes in practice.
572 */
573 struct list_head btree_cache;
574 struct list_head btree_cache_freeable;
575 struct list_head btree_cache_freed;
576
577 /* Number of elements in btree_cache + btree_cache_freeable lists */
578 unsigned int btree_cache_used;
579
580 /*
581 * If we need to allocate memory for a new btree node and that
582 * allocation fails, we can cannibalize another node in the btree cache
583 * to satisfy the allocation - lock to guarantee only one thread does
584 * this at a time:
585 */
586 wait_queue_head_t btree_cache_wait;
587 struct task_struct *btree_cache_alloc_lock;
588
589 /*
590 * When we free a btree node, we increment the gen of the bucket the
591 * node is in - but we can't rewrite the prios and gens until we
592 * finished whatever it is we were doing, otherwise after a crash the
593 * btree node would be freed but for say a split, we might not have the
594 * pointers to the new nodes inserted into the btree yet.
595 *
596 * This is a refcount that blocks prio_write() until the new keys are
597 * written.
598 */
599 atomic_t prio_blocked;
600 wait_queue_head_t bucket_wait;
601
602 /*
603 * For any bio we don't skip we subtract the number of sectors from
604 * rescale; when it hits 0 we rescale all the bucket priorities.
605 */
606 atomic_t rescale;
607 /*
608 * used for GC, identify if any front side I/Os is inflight
609 */
610 atomic_t search_inflight;
611 /*
612 * When we invalidate buckets, we use both the priority and the amount
613 * of good data to determine which buckets to reuse first - to weight
614 * those together consistently we keep track of the smallest nonzero
615 * priority of any bucket.
616 */
617 uint16_t min_prio;
618
619 /*
620 * max(gen - last_gc) for all buckets. When it gets too big we have to
621 * gc to keep gens from wrapping around.
622 */
623 uint8_t need_gc;
624 struct gc_stat gc_stats;
625 size_t nbuckets;
626 size_t avail_nbuckets;
627
628 struct task_struct *gc_thread;
629 /* Where in the btree gc currently is */
630 struct bkey gc_done;
631
632 /*
633 * For automatical garbage collection after writeback completed, this
634 * varialbe is used as bit fields,
635 * - 0000 0001b (BCH_ENABLE_AUTO_GC): enable gc after writeback
636 * - 0000 0010b (BCH_DO_AUTO_GC): do gc after writeback
637 * This is an optimization for following write request after writeback
638 * finished, but read hit rate dropped due to clean data on cache is
639 * discarded. Unless user explicitly sets it via sysfs, it won't be
640 * enabled.
641 */
642 #define BCH_ENABLE_AUTO_GC 1
643 #define BCH_DO_AUTO_GC 2
644 uint8_t gc_after_writeback;
645
646 /*
647 * The allocation code needs gc_mark in struct bucket to be correct, but
648 * it's not while a gc is in progress. Protected by bucket_lock.
649 */
650 int gc_mark_valid;
651
652 /* Counts how many sectors bio_insert has added to the cache */
653 atomic_t sectors_to_gc;
654 wait_queue_head_t gc_wait;
655
656 struct keybuf moving_gc_keys;
657 /* Number of moving GC bios in flight */
658 struct semaphore moving_in_flight;
659
660 struct workqueue_struct *moving_gc_wq;
661
662 struct btree *root;
663
664 #ifdef CONFIG_BCACHE_DEBUG
665 struct btree *verify_data;
666 struct bset *verify_ondisk;
667 struct mutex verify_lock;
668 #endif
669
670 unsigned int nr_uuids;
671 struct uuid_entry *uuids;
672 BKEY_PADDED(uuid_bucket);
673 struct closure uuid_write;
674 struct semaphore uuid_write_mutex;
675
676 /*
677 * A btree node on disk could have too many bsets for an iterator to fit
678 * on the stack - have to dynamically allocate them.
679 * bch_cache_set_alloc() will make sure the pool can allocate iterators
680 * equipped with enough room that can host
681 * (sb.bucket_size / sb.block_size)
682 * btree_iter_sets, which is more than static MAX_BSETS.
683 */
684 mempool_t fill_iter;
685
686 struct bset_sort_state sort;
687
688 /* List of buckets we're currently writing data to */
689 struct list_head data_buckets;
690 spinlock_t data_bucket_lock;
691
692 struct journal journal;
693
694 #define CONGESTED_MAX 1024
695 unsigned int congested_last_us;
696 atomic_t congested;
697
698 /* The rest of this all shows up in sysfs */
699 unsigned int congested_read_threshold_us;
700 unsigned int congested_write_threshold_us;
701
702 struct time_stats btree_gc_time;
703 struct time_stats btree_split_time;
704 struct time_stats btree_read_time;
705
706 atomic_long_t cache_read_races;
707 atomic_long_t writeback_keys_done;
708 atomic_long_t writeback_keys_failed;
709
710 atomic_long_t reclaim;
711 atomic_long_t reclaimed_journal_buckets;
712 atomic_long_t flush_write;
713
714 enum {
715 ON_ERROR_UNREGISTER,
716 ON_ERROR_PANIC,
717 } on_error;
718 #define DEFAULT_IO_ERROR_LIMIT 8
719 unsigned int error_limit;
720 unsigned int error_decay;
721
722 unsigned short journal_delay_ms;
723 bool expensive_debug_checks;
724 unsigned int verify:1;
725 unsigned int key_merging_disabled:1;
726 unsigned int gc_always_rewrite:1;
727 unsigned int shrinker_disabled:1;
728 unsigned int copy_gc_enabled:1;
729
730 #define BUCKET_HASH_BITS 12
731 struct hlist_head bucket_hash[1 << BUCKET_HASH_BITS];
732 };
733
734 struct bbio {
735 unsigned int submit_time_us;
736 union {
737 struct bkey key;
738 uint64_t _pad[3];
739 /*
740 * We only need pad = 3 here because we only ever carry around a
741 * single pointer - i.e. the pointer we're doing io to/from.
742 */
743 };
744 struct bio bio;
745 };
746
747 #define BTREE_PRIO USHRT_MAX
748 #define INITIAL_PRIO 32768U
749
750 #define btree_bytes(c) ((c)->btree_pages * PAGE_SIZE)
751 #define btree_blocks(b) \
752 ((unsigned int) (KEY_SIZE(&b->key) >> (b)->c->block_bits))
753
754 #define btree_default_blocks(c) \
755 ((unsigned int) ((PAGE_SECTORS * (c)->btree_pages) >> (c)->block_bits))
756
757 #define bucket_pages(c) ((c)->sb.bucket_size / PAGE_SECTORS)
758 #define bucket_bytes(c) ((c)->sb.bucket_size << 9)
759 #define block_bytes(c) ((c)->sb.block_size << 9)
760
761 #define prios_per_bucket(c) \
762 ((bucket_bytes(c) - sizeof(struct prio_set)) / \
763 sizeof(struct bucket_disk))
764 #define prio_buckets(c) \
765 DIV_ROUND_UP((size_t) (c)->sb.nbuckets, prios_per_bucket(c))
766
767 static inline size_t sector_to_bucket(struct cache_set *c, sector_t s)
768 {
769 return s >> c->bucket_bits;
770 }
771
772 static inline sector_t bucket_to_sector(struct cache_set *c, size_t b)
773 {
774 return ((sector_t) b) << c->bucket_bits;
775 }
776
777 static inline sector_t bucket_remainder(struct cache_set *c, sector_t s)
778 {
779 return s & (c->sb.bucket_size - 1);
780 }
781
782 static inline struct cache *PTR_CACHE(struct cache_set *c,
783 const struct bkey *k,
784 unsigned int ptr)
785 {
786 return c->cache[PTR_DEV(k, ptr)];
787 }
788
789 static inline size_t PTR_BUCKET_NR(struct cache_set *c,
790 const struct bkey *k,
791 unsigned int ptr)
792 {
793 return sector_to_bucket(c, PTR_OFFSET(k, ptr));
794 }
795
796 static inline struct bucket *PTR_BUCKET(struct cache_set *c,
797 const struct bkey *k,
798 unsigned int ptr)
799 {
800 return PTR_CACHE(c, k, ptr)->buckets + PTR_BUCKET_NR(c, k, ptr);
801 }
802
803 static inline uint8_t gen_after(uint8_t a, uint8_t b)
804 {
805 uint8_t r = a - b;
806
807 return r > 128U ? 0 : r;
808 }
809
810 static inline uint8_t ptr_stale(struct cache_set *c, const struct bkey *k,
811 unsigned int i)
812 {
813 return gen_after(PTR_BUCKET(c, k, i)->gen, PTR_GEN(k, i));
814 }
815
816 static inline bool ptr_available(struct cache_set *c, const struct bkey *k,
817 unsigned int i)
818 {
819 return (PTR_DEV(k, i) < MAX_CACHES_PER_SET) && PTR_CACHE(c, k, i);
820 }
821
822 /* Btree key macros */
823
824 /*
825 * This is used for various on disk data structures - cache_sb, prio_set, bset,
826 * jset: The checksum is _always_ the first 8 bytes of these structs
827 */
828 #define csum_set(i) \
829 bch_crc64(((void *) (i)) + sizeof(uint64_t), \
830 ((void *) bset_bkey_last(i)) - \
831 (((void *) (i)) + sizeof(uint64_t)))
832
833 /* Error handling macros */
834
835 #define btree_bug(b, ...) \
836 do { \
837 if (bch_cache_set_error((b)->c, __VA_ARGS__)) \
838 dump_stack(); \
839 } while (0)
840
841 #define cache_bug(c, ...) \
842 do { \
843 if (bch_cache_set_error(c, __VA_ARGS__)) \
844 dump_stack(); \
845 } while (0)
846
847 #define btree_bug_on(cond, b, ...) \
848 do { \
849 if (cond) \
850 btree_bug(b, __VA_ARGS__); \
851 } while (0)
852
853 #define cache_bug_on(cond, c, ...) \
854 do { \
855 if (cond) \
856 cache_bug(c, __VA_ARGS__); \
857 } while (0)
858
859 #define cache_set_err_on(cond, c, ...) \
860 do { \
861 if (cond) \
862 bch_cache_set_error(c, __VA_ARGS__); \
863 } while (0)
864
865 /* Looping macros */
866
867 #define for_each_cache(ca, cs, iter) \
868 for (iter = 0; ca = cs->cache[iter], iter < (cs)->sb.nr_in_set; iter++)
869
870 #define for_each_bucket(b, ca) \
871 for (b = (ca)->buckets + (ca)->sb.first_bucket; \
872 b < (ca)->buckets + (ca)->sb.nbuckets; b++)
873
874 static inline void cached_dev_put(struct cached_dev *dc)
875 {
876 if (refcount_dec_and_test(&dc->count))
877 schedule_work(&dc->detach);
878 }
879
880 static inline bool cached_dev_get(struct cached_dev *dc)
881 {
882 if (!refcount_inc_not_zero(&dc->count))
883 return false;
884
885 /* Paired with the mb in cached_dev_attach */
886 smp_mb__after_atomic();
887 return true;
888 }
889
890 /*
891 * bucket_gc_gen() returns the difference between the bucket's current gen and
892 * the oldest gen of any pointer into that bucket in the btree (last_gc).
893 */
894
895 static inline uint8_t bucket_gc_gen(struct bucket *b)
896 {
897 return b->gen - b->last_gc;
898 }
899
900 #define BUCKET_GC_GEN_MAX 96U
901
902 #define kobj_attribute_write(n, fn) \
903 static struct kobj_attribute ksysfs_##n = __ATTR(n, 0200, NULL, fn)
904
905 #define kobj_attribute_rw(n, show, store) \
906 static struct kobj_attribute ksysfs_##n = \
907 __ATTR(n, 0600, show, store)
908
909 static inline void wake_up_allocators(struct cache_set *c)
910 {
911 struct cache *ca;
912 unsigned int i;
913
914 for_each_cache(ca, c, i)
915 wake_up_process(ca->alloc_thread);
916 }
917
918 static inline void closure_bio_submit(struct cache_set *c,
919 struct bio *bio,
920 struct closure *cl)
921 {
922 closure_get(cl);
923 if (unlikely(test_bit(CACHE_SET_IO_DISABLE, &c->flags))) {
924 bio->bi_status = BLK_STS_IOERR;
925 bio_endio(bio);
926 return;
927 }
928 generic_make_request(bio);
929 }
930
931 /*
932 * Prevent the kthread exits directly, and make sure when kthread_stop()
933 * is called to stop a kthread, it is still alive. If a kthread might be
934 * stopped by CACHE_SET_IO_DISABLE bit set, wait_for_kthread_stop() is
935 * necessary before the kthread returns.
936 */
937 static inline void wait_for_kthread_stop(void)
938 {
939 while (!kthread_should_stop()) {
940 set_current_state(TASK_INTERRUPTIBLE);
941 schedule();
942 }
943 }
944
945 /* Forward declarations */
946
947 void bch_count_backing_io_errors(struct cached_dev *dc, struct bio *bio);
948 void bch_count_io_errors(struct cache *ca, blk_status_t error,
949 int is_read, const char *m);
950 void bch_bbio_count_io_errors(struct cache_set *c, struct bio *bio,
951 blk_status_t error, const char *m);
952 void bch_bbio_endio(struct cache_set *c, struct bio *bio,
953 blk_status_t error, const char *m);
954 void bch_bbio_free(struct bio *bio, struct cache_set *c);
955 struct bio *bch_bbio_alloc(struct cache_set *c);
956
957 void __bch_submit_bbio(struct bio *bio, struct cache_set *c);
958 void bch_submit_bbio(struct bio *bio, struct cache_set *c,
959 struct bkey *k, unsigned int ptr);
960
961 uint8_t bch_inc_gen(struct cache *ca, struct bucket *b);
962 void bch_rescale_priorities(struct cache_set *c, int sectors);
963
964 bool bch_can_invalidate_bucket(struct cache *ca, struct bucket *b);
965 void __bch_invalidate_one_bucket(struct cache *ca, struct bucket *b);
966
967 void __bch_bucket_free(struct cache *ca, struct bucket *b);
968 void bch_bucket_free(struct cache_set *c, struct bkey *k);
969
970 long bch_bucket_alloc(struct cache *ca, unsigned int reserve, bool wait);
971 int __bch_bucket_alloc_set(struct cache_set *c, unsigned int reserve,
972 struct bkey *k, int n, bool wait);
973 int bch_bucket_alloc_set(struct cache_set *c, unsigned int reserve,
974 struct bkey *k, int n, bool wait);
975 bool bch_alloc_sectors(struct cache_set *c, struct bkey *k,
976 unsigned int sectors, unsigned int write_point,
977 unsigned int write_prio, bool wait);
978 bool bch_cached_dev_error(struct cached_dev *dc);
979
980 __printf(2, 3)
981 bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...);
982
983 int bch_prio_write(struct cache *ca, bool wait);
984 void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent);
985
986 extern struct workqueue_struct *bcache_wq;
987 extern struct workqueue_struct *bch_journal_wq;
988 extern struct mutex bch_register_lock;
989 extern struct list_head bch_cache_sets;
990
991 extern struct kobj_type bch_cached_dev_ktype;
992 extern struct kobj_type bch_flash_dev_ktype;
993 extern struct kobj_type bch_cache_set_ktype;
994 extern struct kobj_type bch_cache_set_internal_ktype;
995 extern struct kobj_type bch_cache_ktype;
996
997 void bch_cached_dev_release(struct kobject *kobj);
998 void bch_flash_dev_release(struct kobject *kobj);
999 void bch_cache_set_release(struct kobject *kobj);
1000 void bch_cache_release(struct kobject *kobj);
1001
1002 int bch_uuid_write(struct cache_set *c);
1003 void bcache_write_super(struct cache_set *c);
1004
1005 int bch_flash_dev_create(struct cache_set *c, uint64_t size);
1006
1007 int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c,
1008 uint8_t *set_uuid);
1009 void bch_cached_dev_detach(struct cached_dev *dc);
1010 int bch_cached_dev_run(struct cached_dev *dc);
1011 void bcache_device_stop(struct bcache_device *d);
1012
1013 void bch_cache_set_unregister(struct cache_set *c);
1014 void bch_cache_set_stop(struct cache_set *c);
1015
1016 struct cache_set *bch_cache_set_alloc(struct cache_sb *sb);
1017 void bch_btree_cache_free(struct cache_set *c);
1018 int bch_btree_cache_alloc(struct cache_set *c);
1019 void bch_moving_init_cache_set(struct cache_set *c);
1020 int bch_open_buckets_alloc(struct cache_set *c);
1021 void bch_open_buckets_free(struct cache_set *c);
1022
1023 int bch_cache_allocator_start(struct cache *ca);
1024
1025 void bch_debug_exit(void);
1026 void bch_debug_init(void);
1027 void bch_request_exit(void);
1028 int bch_request_init(void);
1029
1030 #endif /* _BCACHE_H */