blob: a4c7306ff43de4f1a928962251ffaea384dcb287 [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
Kent Overstreet81ab4192013-10-31 15:46:42 -0700180#include <linux/bcache.h>
Kent Overstreetcafe5632013-03-23 16:11:31 -0700181#include <linux/bio.h>
Kent Overstreetcafe5632013-03-23 16:11:31 -0700182#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
Kent Overstreet67539e82013-09-10 22:53:34 -0700190#include "bset.h"
Kent Overstreetcafe5632013-03-23 16:11:31 -0700191#include "util.h"
192#include "closure.h"
193
194struct bucket {
195 atomic_t pin;
196 uint16_t prio;
197 uint8_t gen;
198 uint8_t disk_gen;
199 uint8_t last_gc; /* Most out of date gen in the btree */
200 uint8_t gc_gen;
Nicholas Swenson981aa8c2013-11-07 17:53:19 -0800201 uint16_t gc_mark; /* Bitfield used by GC. See below for field */
Kent Overstreetcafe5632013-03-23 16:11:31 -0700202};
203
204/*
205 * I'd use bitfields for these, but I don't trust the compiler not to screw me
206 * as multiple threads touch struct bucket without locking
207 */
208
209BITMASK(GC_MARK, struct bucket, gc_mark, 0, 2);
210#define GC_MARK_RECLAIMABLE 0
211#define GC_MARK_DIRTY 1
212#define GC_MARK_METADATA 2
Darrick J. Wong94717442014-01-28 16:57:39 -0800213#define GC_SECTORS_USED_SIZE 13
214#define MAX_GC_SECTORS_USED (~(~0ULL << GC_SECTORS_USED_SIZE))
215BITMASK(GC_SECTORS_USED, struct bucket, gc_mark, 2, GC_SECTORS_USED_SIZE);
Nicholas Swenson981aa8c2013-11-07 17:53:19 -0800216BITMASK(GC_MOVE, struct bucket, gc_mark, 15, 1);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700217
Kent Overstreetcafe5632013-03-23 16:11:31 -0700218#include "journal.h"
219#include "stats.h"
220struct search;
221struct btree;
222struct keybuf;
223
224struct keybuf_key {
225 struct rb_node node;
226 BKEY_PADDED(key);
227 void *private;
228};
229
Kent Overstreetcafe5632013-03-23 16:11:31 -0700230struct keybuf {
Kent Overstreetcafe5632013-03-23 16:11:31 -0700231 struct bkey last_scanned;
232 spinlock_t lock;
233
234 /*
235 * Beginning and end of range in rb tree - so that we can skip taking
236 * lock and checking the rb tree when we need to check for overlapping
237 * keys.
238 */
239 struct bkey start;
240 struct bkey end;
241
242 struct rb_root keys;
243
Kent Overstreet48a915a2013-10-31 15:43:22 -0700244#define KEYBUF_NR 500
Kent Overstreetcafe5632013-03-23 16:11:31 -0700245 DECLARE_ARRAY_ALLOCATOR(struct keybuf_key, freelist, KEYBUF_NR);
246};
247
248struct bio_split_pool {
249 struct bio_set *bio_split;
250 mempool_t *bio_split_hook;
251};
252
253struct bio_split_hook {
254 struct closure cl;
255 struct bio_split_pool *p;
256 struct bio *bio;
257 bio_end_io_t *bi_end_io;
258 void *bi_private;
259};
260
261struct bcache_device {
262 struct closure cl;
263
264 struct kobject kobj;
265
266 struct cache_set *c;
267 unsigned id;
268#define BCACHEDEVNAME_SIZE 12
269 char name[BCACHEDEVNAME_SIZE];
270
271 struct gendisk *disk;
272
Kent Overstreetc4d951d2013-08-21 17:49:09 -0700273 unsigned long flags;
274#define BCACHE_DEV_CLOSING 0
275#define BCACHE_DEV_DETACHING 1
276#define BCACHE_DEV_UNLINK_DONE 2
Kent Overstreetcafe5632013-03-23 16:11:31 -0700277
Kent Overstreet48a915a2013-10-31 15:43:22 -0700278 unsigned nr_stripes;
Kent Overstreet2d679fc2013-08-17 02:13:15 -0700279 unsigned stripe_size;
Kent Overstreet279afba2013-06-05 06:21:07 -0700280 atomic_t *stripe_sectors_dirty;
Kent Overstreet48a915a2013-10-31 15:43:22 -0700281 unsigned long *full_dirty_stripes;
Kent Overstreet279afba2013-06-05 06:21:07 -0700282
Kent Overstreetcafe5632013-03-23 16:11:31 -0700283 unsigned long sectors_dirty_last;
284 long sectors_dirty_derivative;
285
Kent Overstreetcafe5632013-03-23 16:11:31 -0700286 struct bio_set *bio_split;
287
288 unsigned data_csum:1;
289
290 int (*cache_miss)(struct btree *, struct search *,
291 struct bio *, unsigned);
292 int (*ioctl) (struct bcache_device *, fmode_t, unsigned, unsigned long);
293
294 struct bio_split_pool bio_split_hook;
295};
296
297struct io {
298 /* Used to track sequential IO so it can be skipped */
299 struct hlist_node hash;
300 struct list_head lru;
301
302 unsigned long jiffies;
303 unsigned sequential;
304 sector_t last;
305};
306
307struct cached_dev {
308 struct list_head list;
309 struct bcache_device disk;
310 struct block_device *bdev;
311
312 struct cache_sb sb;
313 struct bio sb_bio;
314 struct bio_vec sb_bv[1];
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800315 struct closure sb_write;
316 struct semaphore sb_write_mutex;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700317
318 /* Refcount on the cache set. Always nonzero when we're caching. */
319 atomic_t count;
320 struct work_struct detach;
321
322 /*
323 * Device might not be running if it's dirty and the cache set hasn't
324 * showed up yet.
325 */
326 atomic_t running;
327
328 /*
329 * Writes take a shared lock from start to finish; scanning for dirty
330 * data to refill the rb tree requires an exclusive lock.
331 */
332 struct rw_semaphore writeback_lock;
333
334 /*
335 * Nonzero, and writeback has a refcount (d->count), iff there is dirty
336 * data in the cache. Protected by writeback_lock; must have an
337 * shared lock to set and exclusive lock to clear.
338 */
339 atomic_t has_dirty;
340
Kent Overstreetc2a4f312013-09-23 23:17:31 -0700341 struct bch_ratelimit writeback_rate;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700342 struct delayed_work writeback_rate_update;
343
344 /*
345 * Internal to the writeback code, so read_dirty() can keep track of
346 * where it's at.
347 */
348 sector_t last_read;
349
Kent Overstreetc2a4f312013-09-23 23:17:31 -0700350 /* Limit number of writeback bios in flight */
351 struct semaphore in_flight;
Kent Overstreet5e6926d2013-07-24 17:50:06 -0700352 struct task_struct *writeback_thread;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700353
354 struct keybuf writeback_keys;
355
356 /* For tracking sequential IO */
357#define RECENT_IO_BITS 7
358#define RECENT_IO (1 << RECENT_IO_BITS)
359 struct io io[RECENT_IO];
360 struct hlist_head io_hash[RECENT_IO + 1];
361 struct list_head io_lru;
362 spinlock_t io_lock;
363
364 struct cache_accounting accounting;
365
366 /* The rest of this all shows up in sysfs */
367 unsigned sequential_cutoff;
368 unsigned readahead;
369
Kent Overstreetcafe5632013-03-23 16:11:31 -0700370 unsigned verify:1;
Kent Overstreet5ceaaad2013-09-10 14:27:42 -0700371 unsigned bypass_torture_test:1;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700372
Kent Overstreet72c27062013-06-05 06:24:39 -0700373 unsigned partial_stripes_expensive:1;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700374 unsigned writeback_metadata:1;
375 unsigned writeback_running:1;
376 unsigned char writeback_percent;
377 unsigned writeback_delay;
378
Kent Overstreetcafe5632013-03-23 16:11:31 -0700379 uint64_t writeback_rate_target;
Kent Overstreet16749c22013-11-11 13:58:34 -0800380 int64_t writeback_rate_proportional;
381 int64_t writeback_rate_derivative;
382 int64_t writeback_rate_change;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700383
384 unsigned writeback_rate_update_seconds;
385 unsigned writeback_rate_d_term;
386 unsigned writeback_rate_p_term_inverse;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700387};
388
Kent Overstreet78365412013-12-17 01:29:34 -0800389enum alloc_reserve {
390 RESERVE_BTREE,
391 RESERVE_PRIO,
392 RESERVE_MOVINGGC,
393 RESERVE_NONE,
394 RESERVE_NR,
Kent Overstreetcafe5632013-03-23 16:11:31 -0700395};
396
397struct cache {
398 struct cache_set *set;
399 struct cache_sb sb;
400 struct bio sb_bio;
401 struct bio_vec sb_bv[1];
402
403 struct kobject kobj;
404 struct block_device *bdev;
405
Kent Overstreet119ba0f2013-04-24 19:01:12 -0700406 struct task_struct *alloc_thread;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700407
408 struct closure prio;
409 struct prio_set *disk_buckets;
410
411 /*
412 * When allocating new buckets, prio_write() gets first dibs - since we
413 * may not be allocate at all without writing priorities and gens.
414 * prio_buckets[] contains the last buckets we wrote priorities to (so
415 * gc can mark them as metadata), prio_next[] contains the buckets
416 * allocated for the next prio write.
417 */
418 uint64_t *prio_buckets;
419 uint64_t *prio_last_buckets;
420
421 /*
422 * free: Buckets that are ready to be used
423 *
424 * free_inc: Incoming buckets - these are buckets that currently have
425 * cached data in them, and we can't reuse them until after we write
426 * their new gen to disk. After prio_write() finishes writing the new
427 * gens/prios, they'll be moved to the free list (and possibly discarded
428 * in the process)
429 *
430 * unused: GC found nothing pointing into these buckets (possibly
431 * because all the data they contained was overwritten), so we only
432 * need to discard them before they can be moved to the free list.
433 */
Kent Overstreet78365412013-12-17 01:29:34 -0800434 DECLARE_FIFO(long, free)[RESERVE_NR];
Kent Overstreetcafe5632013-03-23 16:11:31 -0700435 DECLARE_FIFO(long, free_inc);
436 DECLARE_FIFO(long, unused);
437
438 size_t fifo_last_bucket;
439
440 /* Allocation stuff: */
441 struct bucket *buckets;
442
443 DECLARE_HEAP(struct bucket *, heap);
444
445 /*
446 * max(gen - disk_gen) for all buckets. When it gets too big we have to
447 * call prio_write() to keep gens from wrapping.
448 */
449 uint8_t need_save_prio;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700450
451 /*
452 * If nonzero, we know we aren't going to find any buckets to invalidate
453 * until a gc finishes - otherwise we could pointlessly burn a ton of
454 * cpu
455 */
456 unsigned invalidate_needs_gc:1;
457
458 bool discard; /* Get rid of? */
459
Kent Overstreetcafe5632013-03-23 16:11:31 -0700460 struct journal_device journal;
461
462 /* The rest of this all shows up in sysfs */
463#define IO_ERROR_SHIFT 20
464 atomic_t io_errors;
465 atomic_t io_count;
466
467 atomic_long_t meta_sectors_written;
468 atomic_long_t btree_sectors_written;
469 atomic_long_t sectors_written;
470
471 struct bio_split_pool bio_split_hook;
472};
473
474struct gc_stat {
475 size_t nodes;
476 size_t key_bytes;
477
478 size_t nkeys;
479 uint64_t data; /* sectors */
Kent Overstreetcafe5632013-03-23 16:11:31 -0700480 unsigned in_use; /* percent */
481};
482
483/*
484 * Flag bits, for how the cache set is shutting down, and what phase it's at:
485 *
486 * CACHE_SET_UNREGISTERING means we're not just shutting down, we're detaching
487 * all the backing devices first (their cached data gets invalidated, and they
488 * won't automatically reattach).
489 *
490 * CACHE_SET_STOPPING always gets set first when we're closing down a cache set;
491 * we'll continue to run normally for awhile with CACHE_SET_STOPPING set (i.e.
492 * flushing dirty data).
Kent Overstreetcafe5632013-03-23 16:11:31 -0700493 */
494#define CACHE_SET_UNREGISTERING 0
495#define CACHE_SET_STOPPING 1
Kent Overstreetcafe5632013-03-23 16:11:31 -0700496
497struct cache_set {
498 struct closure cl;
499
500 struct list_head list;
501 struct kobject kobj;
502 struct kobject internal;
503 struct dentry *debug;
504 struct cache_accounting accounting;
505
506 unsigned long flags;
507
508 struct cache_sb sb;
509
510 struct cache *cache[MAX_CACHES_PER_SET];
511 struct cache *cache_by_alloc[MAX_CACHES_PER_SET];
512 int caches_loaded;
513
514 struct bcache_device **devices;
515 struct list_head cached_devs;
516 uint64_t cached_dev_sectors;
517 struct closure caching;
518
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800519 struct closure sb_write;
520 struct semaphore sb_write_mutex;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700521
522 mempool_t *search;
523 mempool_t *bio_meta;
524 struct bio_set *bio_split;
525
526 /* For the btree cache */
527 struct shrinker shrink;
528
Kent Overstreetcafe5632013-03-23 16:11:31 -0700529 /* For the btree cache and anything allocation related */
530 struct mutex bucket_lock;
531
532 /* log2(bucket_size), in sectors */
533 unsigned short bucket_bits;
534
535 /* log2(block_size), in sectors */
536 unsigned short block_bits;
537
538 /*
539 * Default number of pages for a new btree node - may be less than a
540 * full bucket
541 */
542 unsigned btree_pages;
543
544 /*
545 * Lists of struct btrees; lru is the list for structs that have memory
546 * allocated for actual btree node, freed is for structs that do not.
547 *
548 * We never free a struct btree, except on shutdown - we just put it on
549 * the btree_cache_freed list and reuse it later. This simplifies the
550 * code, and it doesn't cost us much memory as the memory usage is
551 * dominated by buffers that hold the actual btree node data and those
552 * can be freed - and the number of struct btrees allocated is
553 * effectively bounded.
554 *
555 * btree_cache_freeable effectively is a small cache - we use it because
556 * high order page allocations can be rather expensive, and it's quite
557 * common to delete and allocate btree nodes in quick succession. It
558 * should never grow past ~2-3 nodes in practice.
559 */
560 struct list_head btree_cache;
561 struct list_head btree_cache_freeable;
562 struct list_head btree_cache_freed;
563
564 /* Number of elements in btree_cache + btree_cache_freeable lists */
565 unsigned bucket_cache_used;
566
567 /*
568 * If we need to allocate memory for a new btree node and that
569 * allocation fails, we can cannibalize another node in the btree cache
570 * to satisfy the allocation. However, only one thread can be doing this
571 * at a time, for obvious reasons - try_harder and try_wait are
572 * basically a lock for this that we can wait on asynchronously. The
573 * btree_root() macro releases the lock when it returns.
574 */
Kent Overstreete8e1d462013-07-24 17:27:07 -0700575 struct task_struct *try_harder;
576 wait_queue_head_t try_wait;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700577 uint64_t try_harder_start;
578
579 /*
580 * When we free a btree node, we increment the gen of the bucket the
581 * node is in - but we can't rewrite the prios and gens until we
582 * finished whatever it is we were doing, otherwise after a crash the
583 * btree node would be freed but for say a split, we might not have the
584 * pointers to the new nodes inserted into the btree yet.
585 *
586 * This is a refcount that blocks prio_write() until the new keys are
587 * written.
588 */
589 atomic_t prio_blocked;
Kent Overstreet35fcd842013-07-24 17:29:09 -0700590 wait_queue_head_t bucket_wait;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700591
592 /*
593 * For any bio we don't skip we subtract the number of sectors from
594 * rescale; when it hits 0 we rescale all the bucket priorities.
595 */
596 atomic_t rescale;
597 /*
598 * When we invalidate buckets, we use both the priority and the amount
599 * of good data to determine which buckets to reuse first - to weight
600 * those together consistently we keep track of the smallest nonzero
601 * priority of any bucket.
602 */
603 uint16_t min_prio;
604
605 /*
606 * max(gen - gc_gen) for all buckets. When it gets too big we have to gc
607 * to keep gens from wrapping around.
608 */
609 uint8_t need_gc;
610 struct gc_stat gc_stats;
611 size_t nbuckets;
612
Kent Overstreet72a44512013-10-24 17:19:26 -0700613 struct task_struct *gc_thread;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700614 /* Where in the btree gc currently is */
615 struct bkey gc_done;
616
617 /*
618 * The allocation code needs gc_mark in struct bucket to be correct, but
619 * it's not while a gc is in progress. Protected by bucket_lock.
620 */
621 int gc_mark_valid;
622
623 /* Counts how many sectors bio_insert has added to the cache */
624 atomic_t sectors_to_gc;
625
Kent Overstreet72a44512013-10-24 17:19:26 -0700626 wait_queue_head_t moving_gc_wait;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700627 struct keybuf moving_gc_keys;
628 /* Number of moving GC bios in flight */
Kent Overstreet72a44512013-10-24 17:19:26 -0700629 struct semaphore moving_in_flight;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700630
631 struct btree *root;
632
633#ifdef CONFIG_BCACHE_DEBUG
634 struct btree *verify_data;
Kent Overstreet78b77bf2013-12-17 22:49:08 -0800635 struct bset *verify_ondisk;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700636 struct mutex verify_lock;
637#endif
638
639 unsigned nr_uuids;
640 struct uuid_entry *uuids;
641 BKEY_PADDED(uuid_bucket);
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800642 struct closure uuid_write;
643 struct semaphore uuid_write_mutex;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700644
645 /*
646 * A btree node on disk could have too many bsets for an iterator to fit
Kent Overstreet57943512013-04-25 13:58:35 -0700647 * on the stack - have to dynamically allocate them
Kent Overstreetcafe5632013-03-23 16:11:31 -0700648 */
Kent Overstreet57943512013-04-25 13:58:35 -0700649 mempool_t *fill_iter;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700650
Kent Overstreet67539e82013-09-10 22:53:34 -0700651 struct bset_sort_state sort;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700652
653 /* List of buckets we're currently writing data to */
654 struct list_head data_buckets;
655 spinlock_t data_bucket_lock;
656
657 struct journal journal;
658
659#define CONGESTED_MAX 1024
660 unsigned congested_last_us;
661 atomic_t congested;
662
663 /* The rest of this all shows up in sysfs */
664 unsigned congested_read_threshold_us;
665 unsigned congested_write_threshold_us;
666
Kent Overstreetcafe5632013-03-23 16:11:31 -0700667 struct time_stats btree_gc_time;
668 struct time_stats btree_split_time;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700669 struct time_stats btree_read_time;
670 struct time_stats try_harder_time;
671
672 atomic_long_t cache_read_races;
673 atomic_long_t writeback_keys_done;
674 atomic_long_t writeback_keys_failed;
Kent Overstreet77c320e2013-07-11 19:42:51 -0700675
676 enum {
677 ON_ERROR_UNREGISTER,
678 ON_ERROR_PANIC,
679 } on_error;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700680 unsigned error_limit;
681 unsigned error_decay;
Kent Overstreet77c320e2013-07-11 19:42:51 -0700682
Kent Overstreetcafe5632013-03-23 16:11:31 -0700683 unsigned short journal_delay_ms;
Kent Overstreeta85e9682013-12-20 17:28:16 -0800684 bool expensive_debug_checks;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700685 unsigned verify:1;
686 unsigned key_merging_disabled:1;
687 unsigned gc_always_rewrite:1;
688 unsigned shrinker_disabled:1;
689 unsigned copy_gc_enabled:1;
690
691#define BUCKET_HASH_BITS 12
692 struct hlist_head bucket_hash[1 << BUCKET_HASH_BITS];
693};
694
Kent Overstreetcafe5632013-03-23 16:11:31 -0700695struct bbio {
696 unsigned submit_time_us;
697 union {
698 struct bkey key;
699 uint64_t _pad[3];
700 /*
701 * We only need pad = 3 here because we only ever carry around a
702 * single pointer - i.e. the pointer we're doing io to/from.
703 */
704 };
705 struct bio bio;
706};
707
Kent Overstreetcafe5632013-03-23 16:11:31 -0700708#define BTREE_PRIO USHRT_MAX
Kent Overstreete0a985a2013-11-12 13:49:10 -0800709#define INITIAL_PRIO 32768U
Kent Overstreetcafe5632013-03-23 16:11:31 -0700710
711#define btree_bytes(c) ((c)->btree_pages * PAGE_SIZE)
712#define btree_blocks(b) \
713 ((unsigned) (KEY_SIZE(&b->key) >> (b)->c->block_bits))
714
715#define btree_default_blocks(c) \
716 ((unsigned) ((PAGE_SECTORS * (c)->btree_pages) >> (c)->block_bits))
717
718#define bucket_pages(c) ((c)->sb.bucket_size / PAGE_SECTORS)
719#define bucket_bytes(c) ((c)->sb.bucket_size << 9)
720#define block_bytes(c) ((c)->sb.block_size << 9)
721
Kent Overstreetcafe5632013-03-23 16:11:31 -0700722#define prios_per_bucket(c) \
723 ((bucket_bytes(c) - sizeof(struct prio_set)) / \
724 sizeof(struct bucket_disk))
725#define prio_buckets(c) \
726 DIV_ROUND_UP((size_t) (c)->sb.nbuckets, prios_per_bucket(c))
727
Kent Overstreetcafe5632013-03-23 16:11:31 -0700728static inline size_t sector_to_bucket(struct cache_set *c, sector_t s)
729{
730 return s >> c->bucket_bits;
731}
732
733static inline sector_t bucket_to_sector(struct cache_set *c, size_t b)
734{
735 return ((sector_t) b) << c->bucket_bits;
736}
737
738static inline sector_t bucket_remainder(struct cache_set *c, sector_t s)
739{
740 return s & (c->sb.bucket_size - 1);
741}
742
743static inline struct cache *PTR_CACHE(struct cache_set *c,
744 const struct bkey *k,
745 unsigned ptr)
746{
747 return c->cache[PTR_DEV(k, ptr)];
748}
749
750static inline size_t PTR_BUCKET_NR(struct cache_set *c,
751 const struct bkey *k,
752 unsigned ptr)
753{
754 return sector_to_bucket(c, PTR_OFFSET(k, ptr));
755}
756
757static inline struct bucket *PTR_BUCKET(struct cache_set *c,
758 const struct bkey *k,
759 unsigned ptr)
760{
761 return PTR_CACHE(c, k, ptr)->buckets + PTR_BUCKET_NR(c, k, ptr);
762}
763
Kent Overstreet9a02b7e2013-12-20 17:24:46 -0800764static inline uint8_t gen_after(uint8_t a, uint8_t b)
765{
766 uint8_t r = a - b;
767 return r > 128U ? 0 : r;
768}
769
770static inline uint8_t ptr_stale(struct cache_set *c, const struct bkey *k,
771 unsigned i)
772{
773 return gen_after(PTR_BUCKET(c, k, i)->gen, PTR_GEN(k, i));
774}
775
776static inline bool ptr_available(struct cache_set *c, const struct bkey *k,
777 unsigned i)
778{
779 return (PTR_DEV(k, i) < MAX_CACHES_PER_SET) && PTR_CACHE(c, k, i);
780}
781
Kent Overstreetcafe5632013-03-23 16:11:31 -0700782/* Btree key macros */
783
Kent Overstreetcafe5632013-03-23 16:11:31 -0700784/*
785 * This is used for various on disk data structures - cache_sb, prio_set, bset,
786 * jset: The checksum is _always_ the first 8 bytes of these structs
787 */
788#define csum_set(i) \
Kent Overstreet169ef1c2013-03-28 12:50:55 -0600789 bch_crc64(((void *) (i)) + sizeof(uint64_t), \
Kent Overstreetfafff812013-12-17 21:56:21 -0800790 ((void *) bset_bkey_last(i)) - \
791 (((void *) (i)) + sizeof(uint64_t)))
Kent Overstreetcafe5632013-03-23 16:11:31 -0700792
793/* Error handling macros */
794
795#define btree_bug(b, ...) \
796do { \
797 if (bch_cache_set_error((b)->c, __VA_ARGS__)) \
798 dump_stack(); \
799} while (0)
800
801#define cache_bug(c, ...) \
802do { \
803 if (bch_cache_set_error(c, __VA_ARGS__)) \
804 dump_stack(); \
805} while (0)
806
807#define btree_bug_on(cond, b, ...) \
808do { \
809 if (cond) \
810 btree_bug(b, __VA_ARGS__); \
811} while (0)
812
813#define cache_bug_on(cond, c, ...) \
814do { \
815 if (cond) \
816 cache_bug(c, __VA_ARGS__); \
817} while (0)
818
819#define cache_set_err_on(cond, c, ...) \
820do { \
821 if (cond) \
822 bch_cache_set_error(c, __VA_ARGS__); \
823} while (0)
824
825/* Looping macros */
826
827#define for_each_cache(ca, cs, iter) \
828 for (iter = 0; ca = cs->cache[iter], iter < (cs)->sb.nr_in_set; iter++)
829
830#define for_each_bucket(b, ca) \
831 for (b = (ca)->buckets + (ca)->sb.first_bucket; \
832 b < (ca)->buckets + (ca)->sb.nbuckets; b++)
833
Kent Overstreetcafe5632013-03-23 16:11:31 -0700834static inline void cached_dev_put(struct cached_dev *dc)
835{
836 if (atomic_dec_and_test(&dc->count))
837 schedule_work(&dc->detach);
838}
839
840static inline bool cached_dev_get(struct cached_dev *dc)
841{
842 if (!atomic_inc_not_zero(&dc->count))
843 return false;
844
845 /* Paired with the mb in cached_dev_attach */
846 smp_mb__after_atomic_inc();
847 return true;
848}
849
850/*
851 * bucket_gc_gen() returns the difference between the bucket's current gen and
852 * the oldest gen of any pointer into that bucket in the btree (last_gc).
853 *
854 * bucket_disk_gen() returns the difference between the current gen and the gen
855 * on disk; they're both used to make sure gens don't wrap around.
856 */
857
858static inline uint8_t bucket_gc_gen(struct bucket *b)
859{
860 return b->gen - b->last_gc;
861}
862
863static inline uint8_t bucket_disk_gen(struct bucket *b)
864{
865 return b->gen - b->disk_gen;
866}
867
868#define BUCKET_GC_GEN_MAX 96U
869#define BUCKET_DISK_GEN_MAX 64U
870
871#define kobj_attribute_write(n, fn) \
872 static struct kobj_attribute ksysfs_##n = __ATTR(n, S_IWUSR, NULL, fn)
873
874#define kobj_attribute_rw(n, show, store) \
875 static struct kobj_attribute ksysfs_##n = \
876 __ATTR(n, S_IWUSR|S_IRUSR, show, store)
877
Kent Overstreet119ba0f2013-04-24 19:01:12 -0700878static inline void wake_up_allocators(struct cache_set *c)
879{
880 struct cache *ca;
881 unsigned i;
882
883 for_each_cache(ca, c, i)
884 wake_up_process(ca->alloc_thread);
885}
886
Kent Overstreetcafe5632013-03-23 16:11:31 -0700887/* Forward declarations */
888
Kent Overstreetcafe5632013-03-23 16:11:31 -0700889void bch_count_io_errors(struct cache *, int, const char *);
890void bch_bbio_count_io_errors(struct cache_set *, struct bio *,
891 int, const char *);
892void bch_bbio_endio(struct cache_set *, struct bio *, int, const char *);
893void bch_bbio_free(struct bio *, struct cache_set *);
894struct bio *bch_bbio_alloc(struct cache_set *);
895
Kent Overstreetcafe5632013-03-23 16:11:31 -0700896void bch_generic_make_request(struct bio *, struct bio_split_pool *);
897void __bch_submit_bbio(struct bio *, struct cache_set *);
898void bch_submit_bbio(struct bio *, struct cache_set *, struct bkey *, unsigned);
899
900uint8_t bch_inc_gen(struct cache *, struct bucket *);
901void bch_rescale_priorities(struct cache_set *, int);
902bool bch_bucket_add_unused(struct cache *, struct bucket *);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700903
Kent Overstreet35fcd842013-07-24 17:29:09 -0700904long bch_bucket_alloc(struct cache *, unsigned, bool);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700905void bch_bucket_free(struct cache_set *, struct bkey *);
906
907int __bch_bucket_alloc_set(struct cache_set *, unsigned,
Kent Overstreet35fcd842013-07-24 17:29:09 -0700908 struct bkey *, int, bool);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700909int bch_bucket_alloc_set(struct cache_set *, unsigned,
Kent Overstreet35fcd842013-07-24 17:29:09 -0700910 struct bkey *, int, bool);
Kent Overstreet2599b532013-07-24 18:11:11 -0700911bool bch_alloc_sectors(struct cache_set *, struct bkey *, unsigned,
912 unsigned, unsigned, bool);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700913
914__printf(2, 3)
915bool bch_cache_set_error(struct cache_set *, const char *, ...);
916
917void bch_prio_write(struct cache *);
918void bch_write_bdev_super(struct cached_dev *, struct closure *);
919
Kent Overstreet72a44512013-10-24 17:19:26 -0700920extern struct workqueue_struct *bcache_wq;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700921extern const char * const bch_cache_modes[];
922extern struct mutex bch_register_lock;
923extern struct list_head bch_cache_sets;
924
925extern struct kobj_type bch_cached_dev_ktype;
926extern struct kobj_type bch_flash_dev_ktype;
927extern struct kobj_type bch_cache_set_ktype;
928extern struct kobj_type bch_cache_set_internal_ktype;
929extern struct kobj_type bch_cache_ktype;
930
931void bch_cached_dev_release(struct kobject *);
932void bch_flash_dev_release(struct kobject *);
933void bch_cache_set_release(struct kobject *);
934void bch_cache_release(struct kobject *);
935
936int bch_uuid_write(struct cache_set *);
937void bcache_write_super(struct cache_set *);
938
939int bch_flash_dev_create(struct cache_set *c, uint64_t size);
940
941int bch_cached_dev_attach(struct cached_dev *, struct cache_set *);
942void bch_cached_dev_detach(struct cached_dev *);
943void bch_cached_dev_run(struct cached_dev *);
944void bcache_device_stop(struct bcache_device *);
945
946void bch_cache_set_unregister(struct cache_set *);
947void bch_cache_set_stop(struct cache_set *);
948
949struct cache_set *bch_cache_set_alloc(struct cache_sb *);
950void bch_btree_cache_free(struct cache_set *);
951int bch_btree_cache_alloc(struct cache_set *);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700952void bch_moving_init_cache_set(struct cache_set *);
Kent Overstreet2599b532013-07-24 18:11:11 -0700953int bch_open_buckets_alloc(struct cache_set *);
954void bch_open_buckets_free(struct cache_set *);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700955
Kent Overstreet119ba0f2013-04-24 19:01:12 -0700956int bch_cache_allocator_start(struct cache *ca);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700957int bch_cache_allocator_init(struct cache *ca);
958
959void bch_debug_exit(void);
960int bch_debug_init(struct kobject *);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700961void bch_request_exit(void);
962int bch_request_init(void);
963void bch_btree_exit(void);
964int bch_btree_init(void);
965
966#endif /* _BCACHE_H */