blob: f05723565f172b16a8fcd86c14940041a0b0a94a [file] [log] [blame]
Kent Overstreetcafe5632013-03-23 16:11:31 -07001#ifndef _BCACHE_H
2#define _BCACHE_H
3
4/*
5 * SOME HIGH LEVEL CODE DOCUMENTATION:
6 *
7 * Bcache mostly works with cache sets, cache devices, and backing devices.
8 *
9 * Support for multiple cache devices hasn't quite been finished off yet, but
10 * it's about 95% plumbed through. A cache set and its cache devices is sort of
11 * like a md raid array and its component devices. Most of the code doesn't care
12 * about individual cache devices, the main abstraction is the cache set.
13 *
14 * Multiple cache devices is intended to give us the ability to mirror dirty
15 * cached data and metadata, without mirroring clean cached data.
16 *
17 * Backing devices are different, in that they have a lifetime independent of a
18 * cache set. When you register a newly formatted backing device it'll come up
19 * in passthrough mode, and then you can attach and detach a backing device from
20 * a cache set at runtime - while it's mounted and in use. Detaching implicitly
21 * invalidates any cached data for that backing device.
22 *
23 * A cache set can have multiple (many) backing devices attached to it.
24 *
25 * There's also flash only volumes - this is the reason for the distinction
26 * between struct cached_dev and struct bcache_device. A flash only volume
27 * works much like a bcache device that has a backing device, except the
28 * "cached" data is always dirty. The end result is that we get thin
29 * provisioning with very little additional code.
30 *
31 * Flash only volumes work but they're not production ready because the moving
32 * garbage collector needs more work. More on that later.
33 *
34 * BUCKETS/ALLOCATION:
35 *
36 * Bcache is primarily designed for caching, which means that in normal
37 * operation all of our available space will be allocated. Thus, we need an
38 * efficient way of deleting things from the cache so we can write new things to
39 * it.
40 *
41 * To do this, we first divide the cache device up into buckets. A bucket is the
42 * unit of allocation; they're typically around 1 mb - anywhere from 128k to 2M+
43 * works efficiently.
44 *
45 * Each bucket has a 16 bit priority, and an 8 bit generation associated with
46 * it. The gens and priorities for all the buckets are stored contiguously and
47 * packed on disk (in a linked list of buckets - aside from the superblock, all
48 * of bcache's metadata is stored in buckets).
49 *
50 * The priority is used to implement an LRU. We reset a bucket's priority when
51 * we allocate it or on cache it, and every so often we decrement the priority
52 * of each bucket. It could be used to implement something more sophisticated,
53 * if anyone ever gets around to it.
54 *
55 * The generation is used for invalidating buckets. Each pointer also has an 8
56 * bit generation embedded in it; for a pointer to be considered valid, its gen
57 * must match the gen of the bucket it points into. Thus, to reuse a bucket all
58 * we have to do is increment its gen (and write its new gen to disk; we batch
59 * this up).
60 *
61 * Bcache is entirely COW - we never write twice to a bucket, even buckets that
62 * contain metadata (including btree nodes).
63 *
64 * THE BTREE:
65 *
66 * Bcache is in large part design around the btree.
67 *
68 * At a high level, the btree is just an index of key -> ptr tuples.
69 *
70 * Keys represent extents, and thus have a size field. Keys also have a variable
71 * number of pointers attached to them (potentially zero, which is handy for
72 * invalidating the cache).
73 *
74 * The key itself is an inode:offset pair. The inode number corresponds to a
75 * backing device or a flash only volume. The offset is the ending offset of the
76 * extent within the inode - not the starting offset; this makes lookups
77 * slightly more convenient.
78 *
79 * Pointers contain the cache device id, the offset on that device, and an 8 bit
80 * generation number. More on the gen later.
81 *
82 * Index lookups are not fully abstracted - cache lookups in particular are
83 * still somewhat mixed in with the btree code, but things are headed in that
84 * direction.
85 *
86 * Updates are fairly well abstracted, though. There are two different ways of
87 * updating the btree; insert and replace.
88 *
89 * BTREE_INSERT will just take a list of keys and insert them into the btree -
90 * overwriting (possibly only partially) any extents they overlap with. This is
91 * used to update the index after a write.
92 *
93 * BTREE_REPLACE is really cmpxchg(); it inserts a key into the btree iff it is
94 * overwriting a key that matches another given key. This is used for inserting
95 * data into the cache after a cache miss, and for background writeback, and for
96 * the moving garbage collector.
97 *
98 * There is no "delete" operation; deleting things from the index is
99 * accomplished by either by invalidating pointers (by incrementing a bucket's
100 * gen) or by inserting a key with 0 pointers - which will overwrite anything
101 * previously present at that location in the index.
102 *
103 * This means that there are always stale/invalid keys in the btree. They're
104 * filtered out by the code that iterates through a btree node, and removed when
105 * a btree node is rewritten.
106 *
107 * BTREE NODES:
108 *
109 * Our unit of allocation is a bucket, and we we can't arbitrarily allocate and
110 * free smaller than a bucket - so, that's how big our btree nodes are.
111 *
112 * (If buckets are really big we'll only use part of the bucket for a btree node
113 * - no less than 1/4th - but a bucket still contains no more than a single
114 * btree node. I'd actually like to change this, but for now we rely on the
115 * bucket's gen for deleting btree nodes when we rewrite/split a node.)
116 *
117 * Anyways, btree nodes are big - big enough to be inefficient with a textbook
118 * btree implementation.
119 *
120 * The way this is solved is that btree nodes are internally log structured; we
121 * can append new keys to an existing btree node without rewriting it. This
122 * means each set of keys we write is sorted, but the node is not.
123 *
124 * We maintain this log structure in memory - keeping 1Mb of keys sorted would
125 * be expensive, and we have to distinguish between the keys we have written and
126 * the keys we haven't. So to do a lookup in a btree node, we have to search
127 * each sorted set. But we do merge written sets together lazily, so the cost of
128 * these extra searches is quite low (normally most of the keys in a btree node
129 * will be in one big set, and then there'll be one or two sets that are much
130 * smaller).
131 *
132 * This log structure makes bcache's btree more of a hybrid between a
133 * conventional btree and a compacting data structure, with some of the
134 * advantages of both.
135 *
136 * GARBAGE COLLECTION:
137 *
138 * We can't just invalidate any bucket - it might contain dirty data or
139 * metadata. If it once contained dirty data, other writes might overwrite it
140 * later, leaving no valid pointers into that bucket in the index.
141 *
142 * Thus, the primary purpose of garbage collection is to find buckets to reuse.
143 * It also counts how much valid data it each bucket currently contains, so that
144 * allocation can reuse buckets sooner when they've been mostly overwritten.
145 *
146 * It also does some things that are really internal to the btree
147 * implementation. If a btree node contains pointers that are stale by more than
148 * some threshold, it rewrites the btree node to avoid the bucket's generation
149 * wrapping around. It also merges adjacent btree nodes if they're empty enough.
150 *
151 * THE JOURNAL:
152 *
153 * Bcache's journal is not necessary for consistency; we always strictly
154 * order metadata writes so that the btree and everything else is consistent on
155 * disk in the event of an unclean shutdown, and in fact bcache had writeback
156 * caching (with recovery from unclean shutdown) before journalling was
157 * implemented.
158 *
159 * Rather, the journal is purely a performance optimization; we can't complete a
160 * write until we've updated the index on disk, otherwise the cache would be
161 * inconsistent in the event of an unclean shutdown. This means that without the
162 * journal, on random write workloads we constantly have to update all the leaf
163 * nodes in the btree, and those writes will be mostly empty (appending at most
164 * a few keys each) - highly inefficient in terms of amount of metadata writes,
165 * and it puts more strain on the various btree resorting/compacting code.
166 *
167 * The journal is just a log of keys we've inserted; on startup we just reinsert
168 * all the keys in the open journal entries. That means that when we're updating
169 * a node in the btree, we can wait until a 4k block of keys fills up before
170 * writing them out.
171 *
172 * For simplicity, we only journal updates to leaf nodes; updates to parent
173 * nodes are rare enough (since our leaf nodes are huge) that it wasn't worth
174 * the complexity to deal with journalling them (in particular, journal replay)
175 * - updates to non leaf nodes just happen synchronously (see btree_split()).
176 */
177
178#define pr_fmt(fmt) "bcache: %s() " fmt "\n", __func__
179
180#include <linux/bio.h>
181#include <linux/blktrace_api.h>
182#include <linux/kobject.h>
183#include <linux/list.h>
184#include <linux/mutex.h>
185#include <linux/rbtree.h>
186#include <linux/rwsem.h>
187#include <linux/types.h>
188#include <linux/workqueue.h>
189
190#include "util.h"
191#include "closure.h"
192
193struct bucket {
194 atomic_t pin;
195 uint16_t prio;
196 uint8_t gen;
197 uint8_t disk_gen;
198 uint8_t last_gc; /* Most out of date gen in the btree */
199 uint8_t gc_gen;
200 uint16_t gc_mark;
201};
202
203/*
204 * I'd use bitfields for these, but I don't trust the compiler not to screw me
205 * as multiple threads touch struct bucket without locking
206 */
207
208BITMASK(GC_MARK, struct bucket, gc_mark, 0, 2);
209#define GC_MARK_RECLAIMABLE 0
210#define GC_MARK_DIRTY 1
211#define GC_MARK_METADATA 2
212BITMASK(GC_SECTORS_USED, struct bucket, gc_mark, 2, 14);
213
214struct bkey {
215 uint64_t high;
216 uint64_t low;
217 uint64_t ptr[];
218};
219
220/* Enough for a key with 6 pointers */
221#define BKEY_PAD 8
222
223#define BKEY_PADDED(key) \
224 union { struct bkey key; uint64_t key ## _pad[BKEY_PAD]; }
225
226/* Version 1: Backing device
227 * Version 2: Seed pointer into btree node checksum
228 * Version 3: New UUID format
229 */
230#define BCACHE_SB_VERSION 3
231
232#define SB_SECTOR 8
233#define SB_SIZE 4096
234#define SB_LABEL_SIZE 32
235#define SB_JOURNAL_BUCKETS 256U
236/* SB_JOURNAL_BUCKETS must be divisible by BITS_PER_LONG */
237#define MAX_CACHES_PER_SET 8
238
239#define BDEV_DATA_START 16 /* sectors */
240
241struct cache_sb {
242 uint64_t csum;
243 uint64_t offset; /* sector where this sb was written */
244 uint64_t version;
245#define CACHE_BACKING_DEV 1
246
247 uint8_t magic[16];
248
249 uint8_t uuid[16];
250 union {
251 uint8_t set_uuid[16];
252 uint64_t set_magic;
253 };
254 uint8_t label[SB_LABEL_SIZE];
255
256 uint64_t flags;
257 uint64_t seq;
258 uint64_t pad[8];
259
260 uint64_t nbuckets; /* device size */
261 uint16_t block_size; /* sectors */
262 uint16_t bucket_size; /* sectors */
263
264 uint16_t nr_in_set;
265 uint16_t nr_this_dev;
266
267 uint32_t last_mount; /* time_t */
268
269 uint16_t first_bucket;
270 union {
271 uint16_t njournal_buckets;
272 uint16_t keys;
273 };
274 uint64_t d[SB_JOURNAL_BUCKETS]; /* journal buckets */
275};
276
277BITMASK(CACHE_SYNC, struct cache_sb, flags, 0, 1);
278BITMASK(CACHE_DISCARD, struct cache_sb, flags, 1, 1);
279BITMASK(CACHE_REPLACEMENT, struct cache_sb, flags, 2, 3);
280#define CACHE_REPLACEMENT_LRU 0U
281#define CACHE_REPLACEMENT_FIFO 1U
282#define CACHE_REPLACEMENT_RANDOM 2U
283
284BITMASK(BDEV_CACHE_MODE, struct cache_sb, flags, 0, 4);
285#define CACHE_MODE_WRITETHROUGH 0U
286#define CACHE_MODE_WRITEBACK 1U
287#define CACHE_MODE_WRITEAROUND 2U
288#define CACHE_MODE_NONE 3U
289BITMASK(BDEV_STATE, struct cache_sb, flags, 61, 2);
290#define BDEV_STATE_NONE 0U
291#define BDEV_STATE_CLEAN 1U
292#define BDEV_STATE_DIRTY 2U
293#define BDEV_STATE_STALE 3U
294
295/* Version 1: Seed pointer into btree node checksum
296 */
297#define BCACHE_BSET_VERSION 1
298
299/*
300 * This is the on disk format for btree nodes - a btree node on disk is a list
301 * of these; within each set the keys are sorted
302 */
303struct bset {
304 uint64_t csum;
305 uint64_t magic;
306 uint64_t seq;
307 uint32_t version;
308 uint32_t keys;
309
310 union {
311 struct bkey start[0];
312 uint64_t d[0];
313 };
314};
315
316/*
317 * On disk format for priorities and gens - see super.c near prio_write() for
318 * more.
319 */
320struct prio_set {
321 uint64_t csum;
322 uint64_t magic;
323 uint64_t seq;
324 uint32_t version;
325 uint32_t pad;
326
327 uint64_t next_bucket;
328
329 struct bucket_disk {
330 uint16_t prio;
331 uint8_t gen;
332 } __attribute((packed)) data[];
333};
334
335struct uuid_entry {
336 union {
337 struct {
338 uint8_t uuid[16];
339 uint8_t label[32];
340 uint32_t first_reg;
341 uint32_t last_reg;
342 uint32_t invalidated;
343
344 uint32_t flags;
345 /* Size of flash only volumes */
346 uint64_t sectors;
347 };
348
349 uint8_t pad[128];
350 };
351};
352
353BITMASK(UUID_FLASH_ONLY, struct uuid_entry, flags, 0, 1);
354
355#include "journal.h"
356#include "stats.h"
357struct search;
358struct btree;
359struct keybuf;
360
361struct keybuf_key {
362 struct rb_node node;
363 BKEY_PADDED(key);
364 void *private;
365};
366
367typedef bool (keybuf_pred_fn)(struct keybuf *, struct bkey *);
368
369struct keybuf {
370 keybuf_pred_fn *key_predicate;
371
372 struct bkey last_scanned;
373 spinlock_t lock;
374
375 /*
376 * Beginning and end of range in rb tree - so that we can skip taking
377 * lock and checking the rb tree when we need to check for overlapping
378 * keys.
379 */
380 struct bkey start;
381 struct bkey end;
382
383 struct rb_root keys;
384
385#define KEYBUF_NR 100
386 DECLARE_ARRAY_ALLOCATOR(struct keybuf_key, freelist, KEYBUF_NR);
387};
388
389struct bio_split_pool {
390 struct bio_set *bio_split;
391 mempool_t *bio_split_hook;
392};
393
394struct bio_split_hook {
395 struct closure cl;
396 struct bio_split_pool *p;
397 struct bio *bio;
398 bio_end_io_t *bi_end_io;
399 void *bi_private;
400};
401
402struct bcache_device {
403 struct closure cl;
404
405 struct kobject kobj;
406
407 struct cache_set *c;
408 unsigned id;
409#define BCACHEDEVNAME_SIZE 12
410 char name[BCACHEDEVNAME_SIZE];
411
412 struct gendisk *disk;
413
414 /* If nonzero, we're closing */
415 atomic_t closing;
416
417 /* If nonzero, we're detaching/unregistering from cache set */
418 atomic_t detaching;
419
420 atomic_long_t sectors_dirty;
421 unsigned long sectors_dirty_gc;
422 unsigned long sectors_dirty_last;
423 long sectors_dirty_derivative;
424
425 mempool_t *unaligned_bvec;
426 struct bio_set *bio_split;
427
428 unsigned data_csum:1;
429
430 int (*cache_miss)(struct btree *, struct search *,
431 struct bio *, unsigned);
432 int (*ioctl) (struct bcache_device *, fmode_t, unsigned, unsigned long);
433
434 struct bio_split_pool bio_split_hook;
435};
436
437struct io {
438 /* Used to track sequential IO so it can be skipped */
439 struct hlist_node hash;
440 struct list_head lru;
441
442 unsigned long jiffies;
443 unsigned sequential;
444 sector_t last;
445};
446
447struct cached_dev {
448 struct list_head list;
449 struct bcache_device disk;
450 struct block_device *bdev;
451
452 struct cache_sb sb;
453 struct bio sb_bio;
454 struct bio_vec sb_bv[1];
455 struct closure_with_waitlist sb_write;
456
457 /* Refcount on the cache set. Always nonzero when we're caching. */
458 atomic_t count;
459 struct work_struct detach;
460
461 /*
462 * Device might not be running if it's dirty and the cache set hasn't
463 * showed up yet.
464 */
465 atomic_t running;
466
467 /*
468 * Writes take a shared lock from start to finish; scanning for dirty
469 * data to refill the rb tree requires an exclusive lock.
470 */
471 struct rw_semaphore writeback_lock;
472
473 /*
474 * Nonzero, and writeback has a refcount (d->count), iff there is dirty
475 * data in the cache. Protected by writeback_lock; must have an
476 * shared lock to set and exclusive lock to clear.
477 */
478 atomic_t has_dirty;
479
480 struct ratelimit writeback_rate;
481 struct delayed_work writeback_rate_update;
482
483 /*
484 * Internal to the writeback code, so read_dirty() can keep track of
485 * where it's at.
486 */
487 sector_t last_read;
488
489 /* Number of writeback bios in flight */
490 atomic_t in_flight;
491 struct closure_with_timer writeback;
492 struct closure_waitlist writeback_wait;
493
494 struct keybuf writeback_keys;
495
496 /* For tracking sequential IO */
497#define RECENT_IO_BITS 7
498#define RECENT_IO (1 << RECENT_IO_BITS)
499 struct io io[RECENT_IO];
500 struct hlist_head io_hash[RECENT_IO + 1];
501 struct list_head io_lru;
502 spinlock_t io_lock;
503
504 struct cache_accounting accounting;
505
506 /* The rest of this all shows up in sysfs */
507 unsigned sequential_cutoff;
508 unsigned readahead;
509
510 unsigned sequential_merge:1;
511 unsigned verify:1;
512
513 unsigned writeback_metadata:1;
514 unsigned writeback_running:1;
515 unsigned char writeback_percent;
516 unsigned writeback_delay;
517
518 int writeback_rate_change;
519 int64_t writeback_rate_derivative;
520 uint64_t writeback_rate_target;
521
522 unsigned writeback_rate_update_seconds;
523 unsigned writeback_rate_d_term;
524 unsigned writeback_rate_p_term_inverse;
525 unsigned writeback_rate_d_smooth;
526};
527
528enum alloc_watermarks {
529 WATERMARK_PRIO,
530 WATERMARK_METADATA,
531 WATERMARK_MOVINGGC,
532 WATERMARK_NONE,
533 WATERMARK_MAX
534};
535
536struct cache {
537 struct cache_set *set;
538 struct cache_sb sb;
539 struct bio sb_bio;
540 struct bio_vec sb_bv[1];
541
542 struct kobject kobj;
543 struct block_device *bdev;
544
545 unsigned watermark[WATERMARK_MAX];
546
547 struct closure alloc;
548 struct workqueue_struct *alloc_workqueue;
549
550 struct closure prio;
551 struct prio_set *disk_buckets;
552
553 /*
554 * When allocating new buckets, prio_write() gets first dibs - since we
555 * may not be allocate at all without writing priorities and gens.
556 * prio_buckets[] contains the last buckets we wrote priorities to (so
557 * gc can mark them as metadata), prio_next[] contains the buckets
558 * allocated for the next prio write.
559 */
560 uint64_t *prio_buckets;
561 uint64_t *prio_last_buckets;
562
563 /*
564 * free: Buckets that are ready to be used
565 *
566 * free_inc: Incoming buckets - these are buckets that currently have
567 * cached data in them, and we can't reuse them until after we write
568 * their new gen to disk. After prio_write() finishes writing the new
569 * gens/prios, they'll be moved to the free list (and possibly discarded
570 * in the process)
571 *
572 * unused: GC found nothing pointing into these buckets (possibly
573 * because all the data they contained was overwritten), so we only
574 * need to discard them before they can be moved to the free list.
575 */
576 DECLARE_FIFO(long, free);
577 DECLARE_FIFO(long, free_inc);
578 DECLARE_FIFO(long, unused);
579
580 size_t fifo_last_bucket;
581
582 /* Allocation stuff: */
583 struct bucket *buckets;
584
585 DECLARE_HEAP(struct bucket *, heap);
586
587 /*
588 * max(gen - disk_gen) for all buckets. When it gets too big we have to
589 * call prio_write() to keep gens from wrapping.
590 */
591 uint8_t need_save_prio;
592 unsigned gc_move_threshold;
593
594 /*
595 * If nonzero, we know we aren't going to find any buckets to invalidate
596 * until a gc finishes - otherwise we could pointlessly burn a ton of
597 * cpu
598 */
599 unsigned invalidate_needs_gc:1;
600
601 bool discard; /* Get rid of? */
602
603 /*
604 * We preallocate structs for issuing discards to buckets, and keep them
605 * on this list when they're not in use; do_discard() issues discards
606 * whenever there's work to do and is called by free_some_buckets() and
607 * when a discard finishes.
608 */
609 atomic_t discards_in_flight;
610 struct list_head discards;
611
612 struct journal_device journal;
613
614 /* The rest of this all shows up in sysfs */
615#define IO_ERROR_SHIFT 20
616 atomic_t io_errors;
617 atomic_t io_count;
618
619 atomic_long_t meta_sectors_written;
620 atomic_long_t btree_sectors_written;
621 atomic_long_t sectors_written;
622
623 struct bio_split_pool bio_split_hook;
624};
625
626struct gc_stat {
627 size_t nodes;
628 size_t key_bytes;
629
630 size_t nkeys;
631 uint64_t data; /* sectors */
632 uint64_t dirty; /* sectors */
633 unsigned in_use; /* percent */
634};
635
636/*
637 * Flag bits, for how the cache set is shutting down, and what phase it's at:
638 *
639 * CACHE_SET_UNREGISTERING means we're not just shutting down, we're detaching
640 * all the backing devices first (their cached data gets invalidated, and they
641 * won't automatically reattach).
642 *
643 * CACHE_SET_STOPPING always gets set first when we're closing down a cache set;
644 * we'll continue to run normally for awhile with CACHE_SET_STOPPING set (i.e.
645 * flushing dirty data).
646 *
Kent Overstreetb1a67b02013-03-25 11:46:44 -0700647 * CACHE_SET_STOPPING_2 gets set at the last phase, when it's time to shut down
648 * the allocation thread.
Kent Overstreetcafe5632013-03-23 16:11:31 -0700649 */
650#define CACHE_SET_UNREGISTERING 0
651#define CACHE_SET_STOPPING 1
652#define CACHE_SET_STOPPING_2 2
653
654struct cache_set {
655 struct closure cl;
656
657 struct list_head list;
658 struct kobject kobj;
659 struct kobject internal;
660 struct dentry *debug;
661 struct cache_accounting accounting;
662
663 unsigned long flags;
664
665 struct cache_sb sb;
666
667 struct cache *cache[MAX_CACHES_PER_SET];
668 struct cache *cache_by_alloc[MAX_CACHES_PER_SET];
669 int caches_loaded;
670
671 struct bcache_device **devices;
672 struct list_head cached_devs;
673 uint64_t cached_dev_sectors;
674 struct closure caching;
675
676 struct closure_with_waitlist sb_write;
677
678 mempool_t *search;
679 mempool_t *bio_meta;
680 struct bio_set *bio_split;
681
682 /* For the btree cache */
683 struct shrinker shrink;
684
685 /* For the allocator itself */
686 wait_queue_head_t alloc_wait;
687
688 /* For the btree cache and anything allocation related */
689 struct mutex bucket_lock;
690
691 /* log2(bucket_size), in sectors */
692 unsigned short bucket_bits;
693
694 /* log2(block_size), in sectors */
695 unsigned short block_bits;
696
697 /*
698 * Default number of pages for a new btree node - may be less than a
699 * full bucket
700 */
701 unsigned btree_pages;
702
703 /*
704 * Lists of struct btrees; lru is the list for structs that have memory
705 * allocated for actual btree node, freed is for structs that do not.
706 *
707 * We never free a struct btree, except on shutdown - we just put it on
708 * the btree_cache_freed list and reuse it later. This simplifies the
709 * code, and it doesn't cost us much memory as the memory usage is
710 * dominated by buffers that hold the actual btree node data and those
711 * can be freed - and the number of struct btrees allocated is
712 * effectively bounded.
713 *
714 * btree_cache_freeable effectively is a small cache - we use it because
715 * high order page allocations can be rather expensive, and it's quite
716 * common to delete and allocate btree nodes in quick succession. It
717 * should never grow past ~2-3 nodes in practice.
718 */
719 struct list_head btree_cache;
720 struct list_head btree_cache_freeable;
721 struct list_head btree_cache_freed;
722
723 /* Number of elements in btree_cache + btree_cache_freeable lists */
724 unsigned bucket_cache_used;
725
726 /*
727 * If we need to allocate memory for a new btree node and that
728 * allocation fails, we can cannibalize another node in the btree cache
729 * to satisfy the allocation. However, only one thread can be doing this
730 * at a time, for obvious reasons - try_harder and try_wait are
731 * basically a lock for this that we can wait on asynchronously. The
732 * btree_root() macro releases the lock when it returns.
733 */
734 struct closure *try_harder;
735 struct closure_waitlist try_wait;
736 uint64_t try_harder_start;
737
738 /*
739 * When we free a btree node, we increment the gen of the bucket the
740 * node is in - but we can't rewrite the prios and gens until we
741 * finished whatever it is we were doing, otherwise after a crash the
742 * btree node would be freed but for say a split, we might not have the
743 * pointers to the new nodes inserted into the btree yet.
744 *
745 * This is a refcount that blocks prio_write() until the new keys are
746 * written.
747 */
748 atomic_t prio_blocked;
749 struct closure_waitlist bucket_wait;
750
751 /*
752 * For any bio we don't skip we subtract the number of sectors from
753 * rescale; when it hits 0 we rescale all the bucket priorities.
754 */
755 atomic_t rescale;
756 /*
757 * When we invalidate buckets, we use both the priority and the amount
758 * of good data to determine which buckets to reuse first - to weight
759 * those together consistently we keep track of the smallest nonzero
760 * priority of any bucket.
761 */
762 uint16_t min_prio;
763
764 /*
765 * max(gen - gc_gen) for all buckets. When it gets too big we have to gc
766 * to keep gens from wrapping around.
767 */
768 uint8_t need_gc;
769 struct gc_stat gc_stats;
770 size_t nbuckets;
771
772 struct closure_with_waitlist gc;
773 /* Where in the btree gc currently is */
774 struct bkey gc_done;
775
776 /*
777 * The allocation code needs gc_mark in struct bucket to be correct, but
778 * it's not while a gc is in progress. Protected by bucket_lock.
779 */
780 int gc_mark_valid;
781
782 /* Counts how many sectors bio_insert has added to the cache */
783 atomic_t sectors_to_gc;
784
785 struct closure moving_gc;
786 struct closure_waitlist moving_gc_wait;
787 struct keybuf moving_gc_keys;
788 /* Number of moving GC bios in flight */
789 atomic_t in_flight;
790
791 struct btree *root;
792
793#ifdef CONFIG_BCACHE_DEBUG
794 struct btree *verify_data;
795 struct mutex verify_lock;
796#endif
797
798 unsigned nr_uuids;
799 struct uuid_entry *uuids;
800 BKEY_PADDED(uuid_bucket);
801 struct closure_with_waitlist uuid_write;
802
803 /*
804 * A btree node on disk could have too many bsets for an iterator to fit
805 * on the stack - this is a single element mempool for btree_read_work()
806 */
807 struct mutex fill_lock;
808 struct btree_iter *fill_iter;
809
810 /*
811 * btree_sort() is a merge sort and requires temporary space - single
812 * element mempool
813 */
814 struct mutex sort_lock;
815 struct bset *sort;
816
817 /* List of buckets we're currently writing data to */
818 struct list_head data_buckets;
819 spinlock_t data_bucket_lock;
820
821 struct journal journal;
822
823#define CONGESTED_MAX 1024
824 unsigned congested_last_us;
825 atomic_t congested;
826
827 /* The rest of this all shows up in sysfs */
828 unsigned congested_read_threshold_us;
829 unsigned congested_write_threshold_us;
830
831 spinlock_t sort_time_lock;
832 struct time_stats sort_time;
833 struct time_stats btree_gc_time;
834 struct time_stats btree_split_time;
835 spinlock_t btree_read_time_lock;
836 struct time_stats btree_read_time;
837 struct time_stats try_harder_time;
838
839 atomic_long_t cache_read_races;
840 atomic_long_t writeback_keys_done;
841 atomic_long_t writeback_keys_failed;
842 unsigned error_limit;
843 unsigned error_decay;
844 unsigned short journal_delay_ms;
845 unsigned verify:1;
846 unsigned key_merging_disabled:1;
847 unsigned gc_always_rewrite:1;
848 unsigned shrinker_disabled:1;
849 unsigned copy_gc_enabled:1;
850
851#define BUCKET_HASH_BITS 12
852 struct hlist_head bucket_hash[1 << BUCKET_HASH_BITS];
853};
854
855static inline bool key_merging_disabled(struct cache_set *c)
856{
857#ifdef CONFIG_BCACHE_DEBUG
858 return c->key_merging_disabled;
859#else
860 return 0;
861#endif
862}
863
864struct bbio {
865 unsigned submit_time_us;
866 union {
867 struct bkey key;
868 uint64_t _pad[3];
869 /*
870 * We only need pad = 3 here because we only ever carry around a
871 * single pointer - i.e. the pointer we're doing io to/from.
872 */
873 };
874 struct bio bio;
875};
876
877static inline unsigned local_clock_us(void)
878{
879 return local_clock() >> 10;
880}
881
882#define MAX_BSETS 4U
883
884#define BTREE_PRIO USHRT_MAX
885#define INITIAL_PRIO 32768
886
887#define btree_bytes(c) ((c)->btree_pages * PAGE_SIZE)
888#define btree_blocks(b) \
889 ((unsigned) (KEY_SIZE(&b->key) >> (b)->c->block_bits))
890
891#define btree_default_blocks(c) \
892 ((unsigned) ((PAGE_SECTORS * (c)->btree_pages) >> (c)->block_bits))
893
894#define bucket_pages(c) ((c)->sb.bucket_size / PAGE_SECTORS)
895#define bucket_bytes(c) ((c)->sb.bucket_size << 9)
896#define block_bytes(c) ((c)->sb.block_size << 9)
897
898#define __set_bytes(i, k) (sizeof(*(i)) + (k) * sizeof(uint64_t))
899#define set_bytes(i) __set_bytes(i, i->keys)
900
901#define __set_blocks(i, k, c) DIV_ROUND_UP(__set_bytes(i, k), block_bytes(c))
902#define set_blocks(i, c) __set_blocks(i, (i)->keys, c)
903
904#define node(i, j) ((struct bkey *) ((i)->d + (j)))
905#define end(i) node(i, (i)->keys)
906
907#define index(i, b) \
908 ((size_t) (((void *) i - (void *) (b)->sets[0].data) / \
909 block_bytes(b->c)))
910
911#define btree_data_space(b) (PAGE_SIZE << (b)->page_order)
912
913#define prios_per_bucket(c) \
914 ((bucket_bytes(c) - sizeof(struct prio_set)) / \
915 sizeof(struct bucket_disk))
916#define prio_buckets(c) \
917 DIV_ROUND_UP((size_t) (c)->sb.nbuckets, prios_per_bucket(c))
918
919#define JSET_MAGIC 0x245235c1a3625032ULL
920#define PSET_MAGIC 0x6750e15f87337f91ULL
921#define BSET_MAGIC 0x90135c78b99e07f5ULL
922
923#define jset_magic(c) ((c)->sb.set_magic ^ JSET_MAGIC)
924#define pset_magic(c) ((c)->sb.set_magic ^ PSET_MAGIC)
925#define bset_magic(c) ((c)->sb.set_magic ^ BSET_MAGIC)
926
927/* Bkey fields: all units are in sectors */
928
929#define KEY_FIELD(name, field, offset, size) \
930 BITMASK(name, struct bkey, field, offset, size)
931
932#define PTR_FIELD(name, offset, size) \
933 static inline uint64_t name(const struct bkey *k, unsigned i) \
934 { return (k->ptr[i] >> offset) & ~(((uint64_t) ~0) << size); } \
935 \
936 static inline void SET_##name(struct bkey *k, unsigned i, uint64_t v)\
937 { \
938 k->ptr[i] &= ~(~((uint64_t) ~0 << size) << offset); \
939 k->ptr[i] |= v << offset; \
940 }
941
942KEY_FIELD(KEY_PTRS, high, 60, 3)
943KEY_FIELD(HEADER_SIZE, high, 58, 2)
944KEY_FIELD(KEY_CSUM, high, 56, 2)
945KEY_FIELD(KEY_PINNED, high, 55, 1)
946KEY_FIELD(KEY_DIRTY, high, 36, 1)
947
948KEY_FIELD(KEY_SIZE, high, 20, 16)
949KEY_FIELD(KEY_INODE, high, 0, 20)
950
951/* Next time I change the on disk format, KEY_OFFSET() won't be 64 bits */
952
953static inline uint64_t KEY_OFFSET(const struct bkey *k)
954{
955 return k->low;
956}
957
958static inline void SET_KEY_OFFSET(struct bkey *k, uint64_t v)
959{
960 k->low = v;
961}
962
963PTR_FIELD(PTR_DEV, 51, 12)
964PTR_FIELD(PTR_OFFSET, 8, 43)
965PTR_FIELD(PTR_GEN, 0, 8)
966
967#define PTR_CHECK_DEV ((1 << 12) - 1)
968
969#define PTR(gen, offset, dev) \
970 ((((uint64_t) dev) << 51) | ((uint64_t) offset) << 8 | gen)
971
972static inline size_t sector_to_bucket(struct cache_set *c, sector_t s)
973{
974 return s >> c->bucket_bits;
975}
976
977static inline sector_t bucket_to_sector(struct cache_set *c, size_t b)
978{
979 return ((sector_t) b) << c->bucket_bits;
980}
981
982static inline sector_t bucket_remainder(struct cache_set *c, sector_t s)
983{
984 return s & (c->sb.bucket_size - 1);
985}
986
987static inline struct cache *PTR_CACHE(struct cache_set *c,
988 const struct bkey *k,
989 unsigned ptr)
990{
991 return c->cache[PTR_DEV(k, ptr)];
992}
993
994static inline size_t PTR_BUCKET_NR(struct cache_set *c,
995 const struct bkey *k,
996 unsigned ptr)
997{
998 return sector_to_bucket(c, PTR_OFFSET(k, ptr));
999}
1000
1001static inline struct bucket *PTR_BUCKET(struct cache_set *c,
1002 const struct bkey *k,
1003 unsigned ptr)
1004{
1005 return PTR_CACHE(c, k, ptr)->buckets + PTR_BUCKET_NR(c, k, ptr);
1006}
1007
1008/* Btree key macros */
1009
1010/*
1011 * The high bit being set is a relic from when we used it to do binary
1012 * searches - it told you where a key started. It's not used anymore,
1013 * and can probably be safely dropped.
1014 */
Kent Overstreetb1a67b02013-03-25 11:46:44 -07001015#define KEY(dev, sector, len) \
1016((struct bkey) { \
Kent Overstreetcafe5632013-03-23 16:11:31 -07001017 .high = (1ULL << 63) | ((uint64_t) (len) << 20) | (dev), \
1018 .low = (sector) \
Kent Overstreetb1a67b02013-03-25 11:46:44 -07001019})
Kent Overstreetcafe5632013-03-23 16:11:31 -07001020
1021static inline void bkey_init(struct bkey *k)
1022{
1023 *k = KEY(0, 0, 0);
1024}
1025
1026#define KEY_START(k) (KEY_OFFSET(k) - KEY_SIZE(k))
1027#define START_KEY(k) KEY(KEY_INODE(k), KEY_START(k), 0)
1028#define MAX_KEY KEY(~(~0 << 20), ((uint64_t) ~0) >> 1, 0)
1029#define ZERO_KEY KEY(0, 0, 0)
1030
1031/*
1032 * This is used for various on disk data structures - cache_sb, prio_set, bset,
1033 * jset: The checksum is _always_ the first 8 bytes of these structs
1034 */
1035#define csum_set(i) \
Kent Overstreet169ef1c2013-03-28 12:50:55 -06001036 bch_crc64(((void *) (i)) + sizeof(uint64_t), \
Kent Overstreetcafe5632013-03-23 16:11:31 -07001037 ((void *) end(i)) - (((void *) (i)) + sizeof(uint64_t)))
1038
1039/* Error handling macros */
1040
1041#define btree_bug(b, ...) \
1042do { \
1043 if (bch_cache_set_error((b)->c, __VA_ARGS__)) \
1044 dump_stack(); \
1045} while (0)
1046
1047#define cache_bug(c, ...) \
1048do { \
1049 if (bch_cache_set_error(c, __VA_ARGS__)) \
1050 dump_stack(); \
1051} while (0)
1052
1053#define btree_bug_on(cond, b, ...) \
1054do { \
1055 if (cond) \
1056 btree_bug(b, __VA_ARGS__); \
1057} while (0)
1058
1059#define cache_bug_on(cond, c, ...) \
1060do { \
1061 if (cond) \
1062 cache_bug(c, __VA_ARGS__); \
1063} while (0)
1064
1065#define cache_set_err_on(cond, c, ...) \
1066do { \
1067 if (cond) \
1068 bch_cache_set_error(c, __VA_ARGS__); \
1069} while (0)
1070
1071/* Looping macros */
1072
1073#define for_each_cache(ca, cs, iter) \
1074 for (iter = 0; ca = cs->cache[iter], iter < (cs)->sb.nr_in_set; iter++)
1075
1076#define for_each_bucket(b, ca) \
1077 for (b = (ca)->buckets + (ca)->sb.first_bucket; \
1078 b < (ca)->buckets + (ca)->sb.nbuckets; b++)
1079
1080static inline void __bkey_put(struct cache_set *c, struct bkey *k)
1081{
1082 unsigned i;
1083
1084 for (i = 0; i < KEY_PTRS(k); i++)
1085 atomic_dec_bug(&PTR_BUCKET(c, k, i)->pin);
1086}
1087
1088/* Blktrace macros */
1089
1090#define blktrace_msg(c, fmt, ...) \
1091do { \
1092 struct request_queue *q = bdev_get_queue(c->bdev); \
1093 if (q) \
1094 blk_add_trace_msg(q, fmt, ##__VA_ARGS__); \
1095} while (0)
1096
1097#define blktrace_msg_all(s, fmt, ...) \
1098do { \
1099 struct cache *_c; \
1100 unsigned i; \
1101 for_each_cache(_c, (s), i) \
1102 blktrace_msg(_c, fmt, ##__VA_ARGS__); \
1103} while (0)
1104
1105static inline void cached_dev_put(struct cached_dev *dc)
1106{
1107 if (atomic_dec_and_test(&dc->count))
1108 schedule_work(&dc->detach);
1109}
1110
1111static inline bool cached_dev_get(struct cached_dev *dc)
1112{
1113 if (!atomic_inc_not_zero(&dc->count))
1114 return false;
1115
1116 /* Paired with the mb in cached_dev_attach */
1117 smp_mb__after_atomic_inc();
1118 return true;
1119}
1120
1121/*
1122 * bucket_gc_gen() returns the difference between the bucket's current gen and
1123 * the oldest gen of any pointer into that bucket in the btree (last_gc).
1124 *
1125 * bucket_disk_gen() returns the difference between the current gen and the gen
1126 * on disk; they're both used to make sure gens don't wrap around.
1127 */
1128
1129static inline uint8_t bucket_gc_gen(struct bucket *b)
1130{
1131 return b->gen - b->last_gc;
1132}
1133
1134static inline uint8_t bucket_disk_gen(struct bucket *b)
1135{
1136 return b->gen - b->disk_gen;
1137}
1138
1139#define BUCKET_GC_GEN_MAX 96U
1140#define BUCKET_DISK_GEN_MAX 64U
1141
1142#define kobj_attribute_write(n, fn) \
1143 static struct kobj_attribute ksysfs_##n = __ATTR(n, S_IWUSR, NULL, fn)
1144
1145#define kobj_attribute_rw(n, show, store) \
1146 static struct kobj_attribute ksysfs_##n = \
1147 __ATTR(n, S_IWUSR|S_IRUSR, show, store)
1148
1149/* Forward declarations */
1150
1151void bch_writeback_queue(struct cached_dev *);
1152void bch_writeback_add(struct cached_dev *, unsigned);
1153
1154void bch_count_io_errors(struct cache *, int, const char *);
1155void bch_bbio_count_io_errors(struct cache_set *, struct bio *,
1156 int, const char *);
1157void bch_bbio_endio(struct cache_set *, struct bio *, int, const char *);
1158void bch_bbio_free(struct bio *, struct cache_set *);
1159struct bio *bch_bbio_alloc(struct cache_set *);
1160
1161struct bio *bch_bio_split(struct bio *, int, gfp_t, struct bio_set *);
1162void bch_generic_make_request(struct bio *, struct bio_split_pool *);
1163void __bch_submit_bbio(struct bio *, struct cache_set *);
1164void bch_submit_bbio(struct bio *, struct cache_set *, struct bkey *, unsigned);
1165
1166uint8_t bch_inc_gen(struct cache *, struct bucket *);
1167void bch_rescale_priorities(struct cache_set *, int);
1168bool bch_bucket_add_unused(struct cache *, struct bucket *);
1169void bch_allocator_thread(struct closure *);
1170
1171long bch_bucket_alloc(struct cache *, unsigned, struct closure *);
1172void bch_bucket_free(struct cache_set *, struct bkey *);
1173
1174int __bch_bucket_alloc_set(struct cache_set *, unsigned,
1175 struct bkey *, int, struct closure *);
1176int bch_bucket_alloc_set(struct cache_set *, unsigned,
1177 struct bkey *, int, struct closure *);
1178
1179__printf(2, 3)
1180bool bch_cache_set_error(struct cache_set *, const char *, ...);
1181
1182void bch_prio_write(struct cache *);
1183void bch_write_bdev_super(struct cached_dev *, struct closure *);
1184
1185extern struct workqueue_struct *bcache_wq, *bch_gc_wq;
1186extern const char * const bch_cache_modes[];
1187extern struct mutex bch_register_lock;
1188extern struct list_head bch_cache_sets;
1189
1190extern struct kobj_type bch_cached_dev_ktype;
1191extern struct kobj_type bch_flash_dev_ktype;
1192extern struct kobj_type bch_cache_set_ktype;
1193extern struct kobj_type bch_cache_set_internal_ktype;
1194extern struct kobj_type bch_cache_ktype;
1195
1196void bch_cached_dev_release(struct kobject *);
1197void bch_flash_dev_release(struct kobject *);
1198void bch_cache_set_release(struct kobject *);
1199void bch_cache_release(struct kobject *);
1200
1201int bch_uuid_write(struct cache_set *);
1202void bcache_write_super(struct cache_set *);
1203
1204int bch_flash_dev_create(struct cache_set *c, uint64_t size);
1205
1206int bch_cached_dev_attach(struct cached_dev *, struct cache_set *);
1207void bch_cached_dev_detach(struct cached_dev *);
1208void bch_cached_dev_run(struct cached_dev *);
1209void bcache_device_stop(struct bcache_device *);
1210
1211void bch_cache_set_unregister(struct cache_set *);
1212void bch_cache_set_stop(struct cache_set *);
1213
1214struct cache_set *bch_cache_set_alloc(struct cache_sb *);
1215void bch_btree_cache_free(struct cache_set *);
1216int bch_btree_cache_alloc(struct cache_set *);
1217void bch_writeback_init_cached_dev(struct cached_dev *);
1218void bch_moving_init_cache_set(struct cache_set *);
1219
1220void bch_cache_allocator_exit(struct cache *ca);
1221int bch_cache_allocator_init(struct cache *ca);
1222
1223void bch_debug_exit(void);
1224int bch_debug_init(struct kobject *);
1225void bch_writeback_exit(void);
1226int bch_writeback_init(void);
1227void bch_request_exit(void);
1228int bch_request_init(void);
1229void bch_btree_exit(void);
1230int bch_btree_init(void);
1231
1232#endif /* _BCACHE_H */