Commit | Line | Data |
---|---|---|
cafe5635 KO |
1 | #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 | ||
81ab4190 | 180 | #include <linux/bcache.h> |
cafe5635 | 181 | #include <linux/bio.h> |
cafe5635 KO |
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 | ||
67539e85 | 190 | #include "bset.h" |
cafe5635 KO |
191 | #include "util.h" |
192 | #include "closure.h" | |
193 | ||
194 | struct bucket { | |
195 | atomic_t pin; | |
196 | uint16_t prio; | |
197 | uint8_t gen; | |
cafe5635 | 198 | uint8_t last_gc; /* Most out of date gen in the btree */ |
981aa8c0 | 199 | uint16_t gc_mark; /* Bitfield used by GC. See below for field */ |
cafe5635 KO |
200 | }; |
201 | ||
202 | /* | |
203 | * I'd use bitfields for these, but I don't trust the compiler not to screw me | |
204 | * as multiple threads touch struct bucket without locking | |
205 | */ | |
206 | ||
207 | BITMASK(GC_MARK, struct bucket, gc_mark, 0, 2); | |
4fe6a816 KO |
208 | #define GC_MARK_RECLAIMABLE 1 |
209 | #define GC_MARK_DIRTY 2 | |
210 | #define GC_MARK_METADATA 3 | |
94717447 DW |
211 | #define GC_SECTORS_USED_SIZE 13 |
212 | #define MAX_GC_SECTORS_USED (~(~0ULL << GC_SECTORS_USED_SIZE)) | |
213 | BITMASK(GC_SECTORS_USED, struct bucket, gc_mark, 2, GC_SECTORS_USED_SIZE); | |
981aa8c0 | 214 | BITMASK(GC_MOVE, struct bucket, gc_mark, 15, 1); |
cafe5635 | 215 | |
cafe5635 KO |
216 | #include "journal.h" |
217 | #include "stats.h" | |
218 | struct search; | |
219 | struct btree; | |
220 | struct keybuf; | |
221 | ||
222 | struct keybuf_key { | |
223 | struct rb_node node; | |
224 | BKEY_PADDED(key); | |
225 | void *private; | |
226 | }; | |
227 | ||
cafe5635 | 228 | struct keybuf { |
cafe5635 KO |
229 | struct bkey last_scanned; |
230 | spinlock_t lock; | |
231 | ||
232 | /* | |
233 | * Beginning and end of range in rb tree - so that we can skip taking | |
234 | * lock and checking the rb tree when we need to check for overlapping | |
235 | * keys. | |
236 | */ | |
237 | struct bkey start; | |
238 | struct bkey end; | |
239 | ||
240 | struct rb_root keys; | |
241 | ||
48a915a8 | 242 | #define KEYBUF_NR 500 |
cafe5635 KO |
243 | DECLARE_ARRAY_ALLOCATOR(struct keybuf_key, freelist, KEYBUF_NR); |
244 | }; | |
245 | ||
246 | struct bio_split_pool { | |
247 | struct bio_set *bio_split; | |
248 | mempool_t *bio_split_hook; | |
249 | }; | |
250 | ||
251 | struct bio_split_hook { | |
252 | struct closure cl; | |
253 | struct bio_split_pool *p; | |
254 | struct bio *bio; | |
255 | bio_end_io_t *bi_end_io; | |
256 | void *bi_private; | |
257 | }; | |
258 | ||
259 | struct bcache_device { | |
260 | struct closure cl; | |
261 | ||
262 | struct kobject kobj; | |
263 | ||
264 | struct cache_set *c; | |
265 | unsigned id; | |
266 | #define BCACHEDEVNAME_SIZE 12 | |
267 | char name[BCACHEDEVNAME_SIZE]; | |
268 | ||
269 | struct gendisk *disk; | |
270 | ||
c4d951dd KO |
271 | unsigned long flags; |
272 | #define BCACHE_DEV_CLOSING 0 | |
273 | #define BCACHE_DEV_DETACHING 1 | |
274 | #define BCACHE_DEV_UNLINK_DONE 2 | |
cafe5635 | 275 | |
48a915a8 | 276 | unsigned nr_stripes; |
2d679fc7 | 277 | unsigned stripe_size; |
279afbad | 278 | atomic_t *stripe_sectors_dirty; |
48a915a8 | 279 | unsigned long *full_dirty_stripes; |
279afbad | 280 | |
cafe5635 KO |
281 | unsigned long sectors_dirty_last; |
282 | long sectors_dirty_derivative; | |
283 | ||
cafe5635 KO |
284 | struct bio_set *bio_split; |
285 | ||
286 | unsigned data_csum:1; | |
287 | ||
288 | int (*cache_miss)(struct btree *, struct search *, | |
289 | struct bio *, unsigned); | |
290 | int (*ioctl) (struct bcache_device *, fmode_t, unsigned, unsigned long); | |
291 | ||
292 | struct bio_split_pool bio_split_hook; | |
293 | }; | |
294 | ||
295 | struct io { | |
296 | /* Used to track sequential IO so it can be skipped */ | |
297 | struct hlist_node hash; | |
298 | struct list_head lru; | |
299 | ||
300 | unsigned long jiffies; | |
301 | unsigned sequential; | |
302 | sector_t last; | |
303 | }; | |
304 | ||
305 | struct cached_dev { | |
306 | struct list_head list; | |
307 | struct bcache_device disk; | |
308 | struct block_device *bdev; | |
309 | ||
310 | struct cache_sb sb; | |
311 | struct bio sb_bio; | |
312 | struct bio_vec sb_bv[1]; | |
cb7a583e KO |
313 | struct closure sb_write; |
314 | struct semaphore sb_write_mutex; | |
cafe5635 KO |
315 | |
316 | /* Refcount on the cache set. Always nonzero when we're caching. */ | |
317 | atomic_t count; | |
318 | struct work_struct detach; | |
319 | ||
320 | /* | |
321 | * Device might not be running if it's dirty and the cache set hasn't | |
322 | * showed up yet. | |
323 | */ | |
324 | atomic_t running; | |
325 | ||
326 | /* | |
327 | * Writes take a shared lock from start to finish; scanning for dirty | |
328 | * data to refill the rb tree requires an exclusive lock. | |
329 | */ | |
330 | struct rw_semaphore writeback_lock; | |
331 | ||
332 | /* | |
333 | * Nonzero, and writeback has a refcount (d->count), iff there is dirty | |
334 | * data in the cache. Protected by writeback_lock; must have an | |
335 | * shared lock to set and exclusive lock to clear. | |
336 | */ | |
337 | atomic_t has_dirty; | |
338 | ||
c2a4f318 | 339 | struct bch_ratelimit writeback_rate; |
cafe5635 KO |
340 | struct delayed_work writeback_rate_update; |
341 | ||
342 | /* | |
343 | * Internal to the writeback code, so read_dirty() can keep track of | |
344 | * where it's at. | |
345 | */ | |
346 | sector_t last_read; | |
347 | ||
c2a4f318 KO |
348 | /* Limit number of writeback bios in flight */ |
349 | struct semaphore in_flight; | |
5e6926da | 350 | struct task_struct *writeback_thread; |
cafe5635 KO |
351 | |
352 | struct keybuf writeback_keys; | |
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 sequential_cutoff; | |
366 | unsigned readahead; | |
367 | ||
cafe5635 | 368 | unsigned verify:1; |
5ceaaad7 | 369 | unsigned bypass_torture_test:1; |
cafe5635 | 370 | |
72c27061 | 371 | unsigned partial_stripes_expensive:1; |
cafe5635 KO |
372 | unsigned writeback_metadata:1; |
373 | unsigned writeback_running:1; | |
374 | unsigned char writeback_percent; | |
375 | unsigned writeback_delay; | |
376 | ||
cafe5635 | 377 | uint64_t writeback_rate_target; |
16749c23 KO |
378 | int64_t writeback_rate_proportional; |
379 | int64_t writeback_rate_derivative; | |
380 | int64_t writeback_rate_change; | |
cafe5635 KO |
381 | |
382 | unsigned writeback_rate_update_seconds; | |
383 | unsigned writeback_rate_d_term; | |
384 | unsigned writeback_rate_p_term_inverse; | |
cafe5635 KO |
385 | }; |
386 | ||
78365411 KO |
387 | enum alloc_reserve { |
388 | RESERVE_BTREE, | |
389 | RESERVE_PRIO, | |
390 | RESERVE_MOVINGGC, | |
391 | RESERVE_NONE, | |
392 | RESERVE_NR, | |
cafe5635 KO |
393 | }; |
394 | ||
395 | struct cache { | |
396 | struct cache_set *set; | |
397 | struct cache_sb sb; | |
398 | struct bio sb_bio; | |
399 | struct bio_vec sb_bv[1]; | |
400 | ||
401 | struct kobject kobj; | |
402 | struct block_device *bdev; | |
403 | ||
119ba0f8 | 404 | struct task_struct *alloc_thread; |
cafe5635 KO |
405 | |
406 | struct closure prio; | |
407 | struct prio_set *disk_buckets; | |
408 | ||
409 | /* | |
410 | * When allocating new buckets, prio_write() gets first dibs - since we | |
411 | * may not be allocate at all without writing priorities and gens. | |
412 | * prio_buckets[] contains the last buckets we wrote priorities to (so | |
413 | * gc can mark them as metadata), prio_next[] contains the buckets | |
414 | * allocated for the next prio write. | |
415 | */ | |
416 | uint64_t *prio_buckets; | |
417 | uint64_t *prio_last_buckets; | |
418 | ||
419 | /* | |
420 | * free: Buckets that are ready to be used | |
421 | * | |
422 | * free_inc: Incoming buckets - these are buckets that currently have | |
423 | * cached data in them, and we can't reuse them until after we write | |
424 | * their new gen to disk. After prio_write() finishes writing the new | |
425 | * gens/prios, they'll be moved to the free list (and possibly discarded | |
426 | * in the process) | |
cafe5635 | 427 | */ |
78365411 | 428 | DECLARE_FIFO(long, free)[RESERVE_NR]; |
cafe5635 | 429 | DECLARE_FIFO(long, free_inc); |
cafe5635 KO |
430 | |
431 | size_t fifo_last_bucket; | |
432 | ||
433 | /* Allocation stuff: */ | |
434 | struct bucket *buckets; | |
435 | ||
436 | DECLARE_HEAP(struct bucket *, heap); | |
437 | ||
cafe5635 KO |
438 | /* |
439 | * If nonzero, we know we aren't going to find any buckets to invalidate | |
440 | * until a gc finishes - otherwise we could pointlessly burn a ton of | |
441 | * cpu | |
442 | */ | |
443 | unsigned invalidate_needs_gc:1; | |
444 | ||
445 | bool discard; /* Get rid of? */ | |
446 | ||
cafe5635 KO |
447 | struct journal_device journal; |
448 | ||
449 | /* The rest of this all shows up in sysfs */ | |
450 | #define IO_ERROR_SHIFT 20 | |
451 | atomic_t io_errors; | |
452 | atomic_t io_count; | |
453 | ||
454 | atomic_long_t meta_sectors_written; | |
455 | atomic_long_t btree_sectors_written; | |
456 | atomic_long_t sectors_written; | |
457 | ||
458 | struct bio_split_pool bio_split_hook; | |
459 | }; | |
460 | ||
461 | struct gc_stat { | |
462 | size_t nodes; | |
463 | size_t key_bytes; | |
464 | ||
465 | size_t nkeys; | |
466 | uint64_t data; /* sectors */ | |
cafe5635 KO |
467 | unsigned in_use; /* percent */ |
468 | }; | |
469 | ||
470 | /* | |
471 | * Flag bits, for how the cache set is shutting down, and what phase it's at: | |
472 | * | |
473 | * CACHE_SET_UNREGISTERING means we're not just shutting down, we're detaching | |
474 | * all the backing devices first (their cached data gets invalidated, and they | |
475 | * won't automatically reattach). | |
476 | * | |
477 | * CACHE_SET_STOPPING always gets set first when we're closing down a cache set; | |
478 | * we'll continue to run normally for awhile with CACHE_SET_STOPPING set (i.e. | |
479 | * flushing dirty data). | |
bf0c55c9 SP |
480 | * |
481 | * CACHE_SET_RUNNING means all cache devices have been registered and journal | |
482 | * replay is complete. | |
cafe5635 KO |
483 | */ |
484 | #define CACHE_SET_UNREGISTERING 0 | |
485 | #define CACHE_SET_STOPPING 1 | |
bf0c55c9 | 486 | #define CACHE_SET_RUNNING 2 |
cafe5635 KO |
487 | |
488 | struct cache_set { | |
489 | struct closure cl; | |
490 | ||
491 | struct list_head list; | |
492 | struct kobject kobj; | |
493 | struct kobject internal; | |
494 | struct dentry *debug; | |
495 | struct cache_accounting accounting; | |
496 | ||
497 | unsigned long flags; | |
498 | ||
499 | struct cache_sb sb; | |
500 | ||
501 | struct cache *cache[MAX_CACHES_PER_SET]; | |
502 | struct cache *cache_by_alloc[MAX_CACHES_PER_SET]; | |
503 | int caches_loaded; | |
504 | ||
505 | struct bcache_device **devices; | |
506 | struct list_head cached_devs; | |
507 | uint64_t cached_dev_sectors; | |
508 | struct closure caching; | |
509 | ||
cb7a583e KO |
510 | struct closure sb_write; |
511 | struct semaphore sb_write_mutex; | |
cafe5635 KO |
512 | |
513 | mempool_t *search; | |
514 | mempool_t *bio_meta; | |
515 | struct bio_set *bio_split; | |
516 | ||
517 | /* For the btree cache */ | |
518 | struct shrinker shrink; | |
519 | ||
cafe5635 KO |
520 | /* For the btree cache and anything allocation related */ |
521 | struct mutex bucket_lock; | |
522 | ||
523 | /* log2(bucket_size), in sectors */ | |
524 | unsigned short bucket_bits; | |
525 | ||
526 | /* log2(block_size), in sectors */ | |
527 | unsigned short block_bits; | |
528 | ||
529 | /* | |
530 | * Default number of pages for a new btree node - may be less than a | |
531 | * full bucket | |
532 | */ | |
533 | unsigned btree_pages; | |
534 | ||
535 | /* | |
536 | * Lists of struct btrees; lru is the list for structs that have memory | |
537 | * allocated for actual btree node, freed is for structs that do not. | |
538 | * | |
539 | * We never free a struct btree, except on shutdown - we just put it on | |
540 | * the btree_cache_freed list and reuse it later. This simplifies the | |
541 | * code, and it doesn't cost us much memory as the memory usage is | |
542 | * dominated by buffers that hold the actual btree node data and those | |
543 | * can be freed - and the number of struct btrees allocated is | |
544 | * effectively bounded. | |
545 | * | |
546 | * btree_cache_freeable effectively is a small cache - we use it because | |
547 | * high order page allocations can be rather expensive, and it's quite | |
548 | * common to delete and allocate btree nodes in quick succession. It | |
549 | * should never grow past ~2-3 nodes in practice. | |
550 | */ | |
551 | struct list_head btree_cache; | |
552 | struct list_head btree_cache_freeable; | |
553 | struct list_head btree_cache_freed; | |
554 | ||
555 | /* Number of elements in btree_cache + btree_cache_freeable lists */ | |
0a63b66d | 556 | unsigned btree_cache_used; |
cafe5635 KO |
557 | |
558 | /* | |
559 | * If we need to allocate memory for a new btree node and that | |
560 | * allocation fails, we can cannibalize another node in the btree cache | |
0a63b66d KO |
561 | * to satisfy the allocation - lock to guarantee only one thread does |
562 | * this at a time: | |
cafe5635 | 563 | */ |
0a63b66d KO |
564 | wait_queue_head_t btree_cache_wait; |
565 | struct task_struct *btree_cache_alloc_lock; | |
cafe5635 KO |
566 | |
567 | /* | |
568 | * When we free a btree node, we increment the gen of the bucket the | |
569 | * node is in - but we can't rewrite the prios and gens until we | |
570 | * finished whatever it is we were doing, otherwise after a crash the | |
571 | * btree node would be freed but for say a split, we might not have the | |
572 | * pointers to the new nodes inserted into the btree yet. | |
573 | * | |
574 | * This is a refcount that blocks prio_write() until the new keys are | |
575 | * written. | |
576 | */ | |
577 | atomic_t prio_blocked; | |
35fcd848 | 578 | wait_queue_head_t bucket_wait; |
cafe5635 KO |
579 | |
580 | /* | |
581 | * For any bio we don't skip we subtract the number of sectors from | |
582 | * rescale; when it hits 0 we rescale all the bucket priorities. | |
583 | */ | |
584 | atomic_t rescale; | |
585 | /* | |
586 | * When we invalidate buckets, we use both the priority and the amount | |
587 | * of good data to determine which buckets to reuse first - to weight | |
588 | * those together consistently we keep track of the smallest nonzero | |
589 | * priority of any bucket. | |
590 | */ | |
591 | uint16_t min_prio; | |
592 | ||
593 | /* | |
3a2fd9d5 | 594 | * max(gen - last_gc) for all buckets. When it gets too big we have to gc |
cafe5635 KO |
595 | * to keep gens from wrapping around. |
596 | */ | |
597 | uint8_t need_gc; | |
598 | struct gc_stat gc_stats; | |
599 | size_t nbuckets; | |
600 | ||
72a44517 | 601 | struct task_struct *gc_thread; |
cafe5635 KO |
602 | /* Where in the btree gc currently is */ |
603 | struct bkey gc_done; | |
604 | ||
605 | /* | |
606 | * The allocation code needs gc_mark in struct bucket to be correct, but | |
607 | * it's not while a gc is in progress. Protected by bucket_lock. | |
608 | */ | |
609 | int gc_mark_valid; | |
610 | ||
611 | /* Counts how many sectors bio_insert has added to the cache */ | |
612 | atomic_t sectors_to_gc; | |
613 | ||
72a44517 | 614 | wait_queue_head_t moving_gc_wait; |
cafe5635 KO |
615 | struct keybuf moving_gc_keys; |
616 | /* Number of moving GC bios in flight */ | |
72a44517 | 617 | struct semaphore moving_in_flight; |
cafe5635 | 618 | |
da415a09 NS |
619 | struct workqueue_struct *moving_gc_wq; |
620 | ||
cafe5635 KO |
621 | struct btree *root; |
622 | ||
623 | #ifdef CONFIG_BCACHE_DEBUG | |
624 | struct btree *verify_data; | |
78b77bf8 | 625 | struct bset *verify_ondisk; |
cafe5635 KO |
626 | struct mutex verify_lock; |
627 | #endif | |
628 | ||
629 | unsigned nr_uuids; | |
630 | struct uuid_entry *uuids; | |
631 | BKEY_PADDED(uuid_bucket); | |
cb7a583e KO |
632 | struct closure uuid_write; |
633 | struct semaphore uuid_write_mutex; | |
cafe5635 KO |
634 | |
635 | /* | |
636 | * A btree node on disk could have too many bsets for an iterator to fit | |
57943511 | 637 | * on the stack - have to dynamically allocate them |
cafe5635 | 638 | */ |
57943511 | 639 | mempool_t *fill_iter; |
cafe5635 | 640 | |
67539e85 | 641 | struct bset_sort_state sort; |
cafe5635 KO |
642 | |
643 | /* List of buckets we're currently writing data to */ | |
644 | struct list_head data_buckets; | |
645 | spinlock_t data_bucket_lock; | |
646 | ||
647 | struct journal journal; | |
648 | ||
649 | #define CONGESTED_MAX 1024 | |
650 | unsigned congested_last_us; | |
651 | atomic_t congested; | |
652 | ||
653 | /* The rest of this all shows up in sysfs */ | |
654 | unsigned congested_read_threshold_us; | |
655 | unsigned congested_write_threshold_us; | |
656 | ||
cafe5635 KO |
657 | struct time_stats btree_gc_time; |
658 | struct time_stats btree_split_time; | |
cafe5635 | 659 | struct time_stats btree_read_time; |
cafe5635 KO |
660 | |
661 | atomic_long_t cache_read_races; | |
662 | atomic_long_t writeback_keys_done; | |
663 | atomic_long_t writeback_keys_failed; | |
77c320eb KO |
664 | |
665 | enum { | |
666 | ON_ERROR_UNREGISTER, | |
667 | ON_ERROR_PANIC, | |
668 | } on_error; | |
cafe5635 KO |
669 | unsigned error_limit; |
670 | unsigned error_decay; | |
77c320eb | 671 | |
cafe5635 | 672 | unsigned short journal_delay_ms; |
a85e968e | 673 | bool expensive_debug_checks; |
cafe5635 KO |
674 | unsigned verify:1; |
675 | unsigned key_merging_disabled:1; | |
676 | unsigned gc_always_rewrite:1; | |
677 | unsigned shrinker_disabled:1; | |
678 | unsigned copy_gc_enabled:1; | |
679 | ||
680 | #define BUCKET_HASH_BITS 12 | |
681 | struct hlist_head bucket_hash[1 << BUCKET_HASH_BITS]; | |
682 | }; | |
683 | ||
cafe5635 KO |
684 | struct bbio { |
685 | unsigned submit_time_us; | |
686 | union { | |
687 | struct bkey key; | |
688 | uint64_t _pad[3]; | |
689 | /* | |
690 | * We only need pad = 3 here because we only ever carry around a | |
691 | * single pointer - i.e. the pointer we're doing io to/from. | |
692 | */ | |
693 | }; | |
694 | struct bio bio; | |
695 | }; | |
696 | ||
cafe5635 | 697 | #define BTREE_PRIO USHRT_MAX |
e0a985a4 | 698 | #define INITIAL_PRIO 32768U |
cafe5635 KO |
699 | |
700 | #define btree_bytes(c) ((c)->btree_pages * PAGE_SIZE) | |
701 | #define btree_blocks(b) \ | |
702 | ((unsigned) (KEY_SIZE(&b->key) >> (b)->c->block_bits)) | |
703 | ||
704 | #define btree_default_blocks(c) \ | |
705 | ((unsigned) ((PAGE_SECTORS * (c)->btree_pages) >> (c)->block_bits)) | |
706 | ||
707 | #define bucket_pages(c) ((c)->sb.bucket_size / PAGE_SECTORS) | |
708 | #define bucket_bytes(c) ((c)->sb.bucket_size << 9) | |
709 | #define block_bytes(c) ((c)->sb.block_size << 9) | |
710 | ||
cafe5635 KO |
711 | #define prios_per_bucket(c) \ |
712 | ((bucket_bytes(c) - sizeof(struct prio_set)) / \ | |
713 | sizeof(struct bucket_disk)) | |
714 | #define prio_buckets(c) \ | |
715 | DIV_ROUND_UP((size_t) (c)->sb.nbuckets, prios_per_bucket(c)) | |
716 | ||
cafe5635 KO |
717 | static inline size_t sector_to_bucket(struct cache_set *c, sector_t s) |
718 | { | |
719 | return s >> c->bucket_bits; | |
720 | } | |
721 | ||
722 | static inline sector_t bucket_to_sector(struct cache_set *c, size_t b) | |
723 | { | |
724 | return ((sector_t) b) << c->bucket_bits; | |
725 | } | |
726 | ||
727 | static inline sector_t bucket_remainder(struct cache_set *c, sector_t s) | |
728 | { | |
729 | return s & (c->sb.bucket_size - 1); | |
730 | } | |
731 | ||
732 | static inline struct cache *PTR_CACHE(struct cache_set *c, | |
733 | const struct bkey *k, | |
734 | unsigned ptr) | |
735 | { | |
736 | return c->cache[PTR_DEV(k, ptr)]; | |
737 | } | |
738 | ||
739 | static inline size_t PTR_BUCKET_NR(struct cache_set *c, | |
740 | const struct bkey *k, | |
741 | unsigned ptr) | |
742 | { | |
743 | return sector_to_bucket(c, PTR_OFFSET(k, ptr)); | |
744 | } | |
745 | ||
746 | static inline struct bucket *PTR_BUCKET(struct cache_set *c, | |
747 | const struct bkey *k, | |
748 | unsigned ptr) | |
749 | { | |
750 | return PTR_CACHE(c, k, ptr)->buckets + PTR_BUCKET_NR(c, k, ptr); | |
751 | } | |
752 | ||
9a02b7ee KO |
753 | static inline uint8_t gen_after(uint8_t a, uint8_t b) |
754 | { | |
755 | uint8_t r = a - b; | |
756 | return r > 128U ? 0 : r; | |
757 | } | |
758 | ||
759 | static inline uint8_t ptr_stale(struct cache_set *c, const struct bkey *k, | |
760 | unsigned i) | |
761 | { | |
762 | return gen_after(PTR_BUCKET(c, k, i)->gen, PTR_GEN(k, i)); | |
763 | } | |
764 | ||
765 | static inline bool ptr_available(struct cache_set *c, const struct bkey *k, | |
766 | unsigned i) | |
767 | { | |
768 | return (PTR_DEV(k, i) < MAX_CACHES_PER_SET) && PTR_CACHE(c, k, i); | |
769 | } | |
770 | ||
cafe5635 KO |
771 | /* Btree key macros */ |
772 | ||
cafe5635 KO |
773 | /* |
774 | * This is used for various on disk data structures - cache_sb, prio_set, bset, | |
775 | * jset: The checksum is _always_ the first 8 bytes of these structs | |
776 | */ | |
777 | #define csum_set(i) \ | |
169ef1cf | 778 | bch_crc64(((void *) (i)) + sizeof(uint64_t), \ |
fafff81c KO |
779 | ((void *) bset_bkey_last(i)) - \ |
780 | (((void *) (i)) + sizeof(uint64_t))) | |
cafe5635 KO |
781 | |
782 | /* Error handling macros */ | |
783 | ||
784 | #define btree_bug(b, ...) \ | |
785 | do { \ | |
786 | if (bch_cache_set_error((b)->c, __VA_ARGS__)) \ | |
787 | dump_stack(); \ | |
788 | } while (0) | |
789 | ||
790 | #define cache_bug(c, ...) \ | |
791 | do { \ | |
792 | if (bch_cache_set_error(c, __VA_ARGS__)) \ | |
793 | dump_stack(); \ | |
794 | } while (0) | |
795 | ||
796 | #define btree_bug_on(cond, b, ...) \ | |
797 | do { \ | |
798 | if (cond) \ | |
799 | btree_bug(b, __VA_ARGS__); \ | |
800 | } while (0) | |
801 | ||
802 | #define cache_bug_on(cond, c, ...) \ | |
803 | do { \ | |
804 | if (cond) \ | |
805 | cache_bug(c, __VA_ARGS__); \ | |
806 | } while (0) | |
807 | ||
808 | #define cache_set_err_on(cond, c, ...) \ | |
809 | do { \ | |
810 | if (cond) \ | |
811 | bch_cache_set_error(c, __VA_ARGS__); \ | |
812 | } while (0) | |
813 | ||
814 | /* Looping macros */ | |
815 | ||
816 | #define for_each_cache(ca, cs, iter) \ | |
817 | for (iter = 0; ca = cs->cache[iter], iter < (cs)->sb.nr_in_set; iter++) | |
818 | ||
819 | #define for_each_bucket(b, ca) \ | |
820 | for (b = (ca)->buckets + (ca)->sb.first_bucket; \ | |
821 | b < (ca)->buckets + (ca)->sb.nbuckets; b++) | |
822 | ||
cafe5635 KO |
823 | static inline void cached_dev_put(struct cached_dev *dc) |
824 | { | |
825 | if (atomic_dec_and_test(&dc->count)) | |
826 | schedule_work(&dc->detach); | |
827 | } | |
828 | ||
829 | static inline bool cached_dev_get(struct cached_dev *dc) | |
830 | { | |
831 | if (!atomic_inc_not_zero(&dc->count)) | |
832 | return false; | |
833 | ||
834 | /* Paired with the mb in cached_dev_attach */ | |
4e857c58 | 835 | smp_mb__after_atomic(); |
cafe5635 KO |
836 | return true; |
837 | } | |
838 | ||
839 | /* | |
840 | * bucket_gc_gen() returns the difference between the bucket's current gen and | |
841 | * the oldest gen of any pointer into that bucket in the btree (last_gc). | |
cafe5635 KO |
842 | */ |
843 | ||
844 | static inline uint8_t bucket_gc_gen(struct bucket *b) | |
845 | { | |
846 | return b->gen - b->last_gc; | |
847 | } | |
848 | ||
cafe5635 | 849 | #define BUCKET_GC_GEN_MAX 96U |
cafe5635 KO |
850 | |
851 | #define kobj_attribute_write(n, fn) \ | |
852 | static struct kobj_attribute ksysfs_##n = __ATTR(n, S_IWUSR, NULL, fn) | |
853 | ||
854 | #define kobj_attribute_rw(n, show, store) \ | |
855 | static struct kobj_attribute ksysfs_##n = \ | |
856 | __ATTR(n, S_IWUSR|S_IRUSR, show, store) | |
857 | ||
119ba0f8 KO |
858 | static inline void wake_up_allocators(struct cache_set *c) |
859 | { | |
860 | struct cache *ca; | |
861 | unsigned i; | |
862 | ||
863 | for_each_cache(ca, c, i) | |
864 | wake_up_process(ca->alloc_thread); | |
865 | } | |
866 | ||
cafe5635 KO |
867 | /* Forward declarations */ |
868 | ||
cafe5635 KO |
869 | void bch_count_io_errors(struct cache *, int, const char *); |
870 | void bch_bbio_count_io_errors(struct cache_set *, struct bio *, | |
871 | int, const char *); | |
872 | void bch_bbio_endio(struct cache_set *, struct bio *, int, const char *); | |
873 | void bch_bbio_free(struct bio *, struct cache_set *); | |
874 | struct bio *bch_bbio_alloc(struct cache_set *); | |
875 | ||
cafe5635 KO |
876 | void bch_generic_make_request(struct bio *, struct bio_split_pool *); |
877 | void __bch_submit_bbio(struct bio *, struct cache_set *); | |
878 | void bch_submit_bbio(struct bio *, struct cache_set *, struct bkey *, unsigned); | |
879 | ||
880 | uint8_t bch_inc_gen(struct cache *, struct bucket *); | |
881 | void bch_rescale_priorities(struct cache_set *, int); | |
cafe5635 | 882 | |
2531d9ee KO |
883 | bool bch_can_invalidate_bucket(struct cache *, struct bucket *); |
884 | void __bch_invalidate_one_bucket(struct cache *, struct bucket *); | |
885 | ||
886 | void __bch_bucket_free(struct cache *, struct bucket *); | |
cafe5635 KO |
887 | void bch_bucket_free(struct cache_set *, struct bkey *); |
888 | ||
2531d9ee | 889 | long bch_bucket_alloc(struct cache *, unsigned, bool); |
cafe5635 | 890 | int __bch_bucket_alloc_set(struct cache_set *, unsigned, |
35fcd848 | 891 | struct bkey *, int, bool); |
cafe5635 | 892 | int bch_bucket_alloc_set(struct cache_set *, unsigned, |
35fcd848 | 893 | struct bkey *, int, bool); |
2599b53b KO |
894 | bool bch_alloc_sectors(struct cache_set *, struct bkey *, unsigned, |
895 | unsigned, unsigned, bool); | |
cafe5635 KO |
896 | |
897 | __printf(2, 3) | |
898 | bool bch_cache_set_error(struct cache_set *, const char *, ...); | |
899 | ||
900 | void bch_prio_write(struct cache *); | |
901 | void bch_write_bdev_super(struct cached_dev *, struct closure *); | |
902 | ||
72a44517 | 903 | extern struct workqueue_struct *bcache_wq; |
cafe5635 KO |
904 | extern const char * const bch_cache_modes[]; |
905 | extern struct mutex bch_register_lock; | |
906 | extern struct list_head bch_cache_sets; | |
907 | ||
908 | extern struct kobj_type bch_cached_dev_ktype; | |
909 | extern struct kobj_type bch_flash_dev_ktype; | |
910 | extern struct kobj_type bch_cache_set_ktype; | |
911 | extern struct kobj_type bch_cache_set_internal_ktype; | |
912 | extern struct kobj_type bch_cache_ktype; | |
913 | ||
914 | void bch_cached_dev_release(struct kobject *); | |
915 | void bch_flash_dev_release(struct kobject *); | |
916 | void bch_cache_set_release(struct kobject *); | |
917 | void bch_cache_release(struct kobject *); | |
918 | ||
919 | int bch_uuid_write(struct cache_set *); | |
920 | void bcache_write_super(struct cache_set *); | |
921 | ||
922 | int bch_flash_dev_create(struct cache_set *c, uint64_t size); | |
923 | ||
924 | int bch_cached_dev_attach(struct cached_dev *, struct cache_set *); | |
925 | void bch_cached_dev_detach(struct cached_dev *); | |
926 | void bch_cached_dev_run(struct cached_dev *); | |
927 | void bcache_device_stop(struct bcache_device *); | |
928 | ||
929 | void bch_cache_set_unregister(struct cache_set *); | |
930 | void bch_cache_set_stop(struct cache_set *); | |
931 | ||
932 | struct cache_set *bch_cache_set_alloc(struct cache_sb *); | |
933 | void bch_btree_cache_free(struct cache_set *); | |
934 | int bch_btree_cache_alloc(struct cache_set *); | |
cafe5635 | 935 | void bch_moving_init_cache_set(struct cache_set *); |
2599b53b KO |
936 | int bch_open_buckets_alloc(struct cache_set *); |
937 | void bch_open_buckets_free(struct cache_set *); | |
cafe5635 | 938 | |
119ba0f8 | 939 | int bch_cache_allocator_start(struct cache *ca); |
cafe5635 KO |
940 | |
941 | void bch_debug_exit(void); | |
942 | int bch_debug_init(struct kobject *); | |
cafe5635 KO |
943 | void bch_request_exit(void); |
944 | int bch_request_init(void); | |
cafe5635 KO |
945 | |
946 | #endif /* _BCACHE_H */ |