]>
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 | ||
cafe5635 KO |
246 | struct bcache_device { |
247 | struct closure cl; | |
248 | ||
249 | struct kobject kobj; | |
250 | ||
251 | struct cache_set *c; | |
252 | unsigned id; | |
253 | #define BCACHEDEVNAME_SIZE 12 | |
254 | char name[BCACHEDEVNAME_SIZE]; | |
255 | ||
256 | struct gendisk *disk; | |
257 | ||
c4d951dd KO |
258 | unsigned long flags; |
259 | #define BCACHE_DEV_CLOSING 0 | |
260 | #define BCACHE_DEV_DETACHING 1 | |
261 | #define BCACHE_DEV_UNLINK_DONE 2 | |
cafe5635 | 262 | |
48a915a8 | 263 | unsigned nr_stripes; |
2d679fc7 | 264 | unsigned stripe_size; |
279afbad | 265 | atomic_t *stripe_sectors_dirty; |
48a915a8 | 266 | unsigned long *full_dirty_stripes; |
279afbad | 267 | |
cafe5635 KO |
268 | unsigned long sectors_dirty_last; |
269 | long sectors_dirty_derivative; | |
270 | ||
cafe5635 KO |
271 | struct bio_set *bio_split; |
272 | ||
273 | unsigned data_csum:1; | |
274 | ||
275 | int (*cache_miss)(struct btree *, struct search *, | |
276 | struct bio *, unsigned); | |
277 | int (*ioctl) (struct bcache_device *, fmode_t, unsigned, unsigned long); | |
cafe5635 KO |
278 | }; |
279 | ||
280 | struct io { | |
281 | /* Used to track sequential IO so it can be skipped */ | |
282 | struct hlist_node hash; | |
283 | struct list_head lru; | |
284 | ||
285 | unsigned long jiffies; | |
286 | unsigned sequential; | |
287 | sector_t last; | |
288 | }; | |
289 | ||
290 | struct cached_dev { | |
291 | struct list_head list; | |
292 | struct bcache_device disk; | |
293 | struct block_device *bdev; | |
294 | ||
295 | struct cache_sb sb; | |
296 | struct bio sb_bio; | |
297 | struct bio_vec sb_bv[1]; | |
cb7a583e KO |
298 | struct closure sb_write; |
299 | struct semaphore sb_write_mutex; | |
cafe5635 KO |
300 | |
301 | /* Refcount on the cache set. Always nonzero when we're caching. */ | |
302 | atomic_t count; | |
303 | struct work_struct detach; | |
304 | ||
305 | /* | |
306 | * Device might not be running if it's dirty and the cache set hasn't | |
307 | * showed up yet. | |
308 | */ | |
309 | atomic_t running; | |
310 | ||
311 | /* | |
312 | * Writes take a shared lock from start to finish; scanning for dirty | |
313 | * data to refill the rb tree requires an exclusive lock. | |
314 | */ | |
315 | struct rw_semaphore writeback_lock; | |
316 | ||
317 | /* | |
318 | * Nonzero, and writeback has a refcount (d->count), iff there is dirty | |
319 | * data in the cache. Protected by writeback_lock; must have an | |
320 | * shared lock to set and exclusive lock to clear. | |
321 | */ | |
322 | atomic_t has_dirty; | |
323 | ||
c2a4f318 | 324 | struct bch_ratelimit writeback_rate; |
cafe5635 KO |
325 | struct delayed_work writeback_rate_update; |
326 | ||
327 | /* | |
328 | * Internal to the writeback code, so read_dirty() can keep track of | |
329 | * where it's at. | |
330 | */ | |
331 | sector_t last_read; | |
332 | ||
c2a4f318 KO |
333 | /* Limit number of writeback bios in flight */ |
334 | struct semaphore in_flight; | |
5e6926da | 335 | struct task_struct *writeback_thread; |
cafe5635 KO |
336 | |
337 | struct keybuf writeback_keys; | |
338 | ||
339 | /* For tracking sequential IO */ | |
340 | #define RECENT_IO_BITS 7 | |
341 | #define RECENT_IO (1 << RECENT_IO_BITS) | |
342 | struct io io[RECENT_IO]; | |
343 | struct hlist_head io_hash[RECENT_IO + 1]; | |
344 | struct list_head io_lru; | |
345 | spinlock_t io_lock; | |
346 | ||
347 | struct cache_accounting accounting; | |
348 | ||
349 | /* The rest of this all shows up in sysfs */ | |
350 | unsigned sequential_cutoff; | |
351 | unsigned readahead; | |
352 | ||
cafe5635 | 353 | unsigned verify:1; |
5ceaaad7 | 354 | unsigned bypass_torture_test:1; |
cafe5635 | 355 | |
72c27061 | 356 | unsigned partial_stripes_expensive:1; |
cafe5635 KO |
357 | unsigned writeback_metadata:1; |
358 | unsigned writeback_running:1; | |
359 | unsigned char writeback_percent; | |
360 | unsigned writeback_delay; | |
361 | ||
cafe5635 | 362 | uint64_t writeback_rate_target; |
16749c23 KO |
363 | int64_t writeback_rate_proportional; |
364 | int64_t writeback_rate_derivative; | |
365 | int64_t writeback_rate_change; | |
cafe5635 KO |
366 | |
367 | unsigned writeback_rate_update_seconds; | |
368 | unsigned writeback_rate_d_term; | |
369 | unsigned writeback_rate_p_term_inverse; | |
cafe5635 KO |
370 | }; |
371 | ||
78365411 KO |
372 | enum alloc_reserve { |
373 | RESERVE_BTREE, | |
374 | RESERVE_PRIO, | |
375 | RESERVE_MOVINGGC, | |
376 | RESERVE_NONE, | |
377 | RESERVE_NR, | |
cafe5635 KO |
378 | }; |
379 | ||
380 | struct cache { | |
381 | struct cache_set *set; | |
382 | struct cache_sb sb; | |
383 | struct bio sb_bio; | |
384 | struct bio_vec sb_bv[1]; | |
385 | ||
386 | struct kobject kobj; | |
387 | struct block_device *bdev; | |
388 | ||
119ba0f8 | 389 | struct task_struct *alloc_thread; |
cafe5635 KO |
390 | |
391 | struct closure prio; | |
392 | struct prio_set *disk_buckets; | |
393 | ||
394 | /* | |
395 | * When allocating new buckets, prio_write() gets first dibs - since we | |
396 | * may not be allocate at all without writing priorities and gens. | |
397 | * prio_buckets[] contains the last buckets we wrote priorities to (so | |
398 | * gc can mark them as metadata), prio_next[] contains the buckets | |
399 | * allocated for the next prio write. | |
400 | */ | |
401 | uint64_t *prio_buckets; | |
402 | uint64_t *prio_last_buckets; | |
403 | ||
404 | /* | |
405 | * free: Buckets that are ready to be used | |
406 | * | |
407 | * free_inc: Incoming buckets - these are buckets that currently have | |
408 | * cached data in them, and we can't reuse them until after we write | |
409 | * their new gen to disk. After prio_write() finishes writing the new | |
410 | * gens/prios, they'll be moved to the free list (and possibly discarded | |
411 | * in the process) | |
cafe5635 | 412 | */ |
78365411 | 413 | DECLARE_FIFO(long, free)[RESERVE_NR]; |
cafe5635 | 414 | DECLARE_FIFO(long, free_inc); |
cafe5635 KO |
415 | |
416 | size_t fifo_last_bucket; | |
417 | ||
418 | /* Allocation stuff: */ | |
419 | struct bucket *buckets; | |
420 | ||
421 | DECLARE_HEAP(struct bucket *, heap); | |
422 | ||
cafe5635 KO |
423 | /* |
424 | * If nonzero, we know we aren't going to find any buckets to invalidate | |
425 | * until a gc finishes - otherwise we could pointlessly burn a ton of | |
426 | * cpu | |
427 | */ | |
be628be0 | 428 | unsigned invalidate_needs_gc; |
cafe5635 KO |
429 | |
430 | bool discard; /* Get rid of? */ | |
431 | ||
cafe5635 KO |
432 | struct journal_device journal; |
433 | ||
434 | /* The rest of this all shows up in sysfs */ | |
435 | #define IO_ERROR_SHIFT 20 | |
436 | atomic_t io_errors; | |
437 | atomic_t io_count; | |
438 | ||
439 | atomic_long_t meta_sectors_written; | |
440 | atomic_long_t btree_sectors_written; | |
441 | atomic_long_t sectors_written; | |
cafe5635 KO |
442 | }; |
443 | ||
444 | struct gc_stat { | |
445 | size_t nodes; | |
446 | size_t key_bytes; | |
447 | ||
448 | size_t nkeys; | |
449 | uint64_t data; /* sectors */ | |
cafe5635 KO |
450 | unsigned in_use; /* percent */ |
451 | }; | |
452 | ||
453 | /* | |
454 | * Flag bits, for how the cache set is shutting down, and what phase it's at: | |
455 | * | |
456 | * CACHE_SET_UNREGISTERING means we're not just shutting down, we're detaching | |
457 | * all the backing devices first (their cached data gets invalidated, and they | |
458 | * won't automatically reattach). | |
459 | * | |
460 | * CACHE_SET_STOPPING always gets set first when we're closing down a cache set; | |
461 | * we'll continue to run normally for awhile with CACHE_SET_STOPPING set (i.e. | |
462 | * flushing dirty data). | |
bf0c55c9 SP |
463 | * |
464 | * CACHE_SET_RUNNING means all cache devices have been registered and journal | |
465 | * replay is complete. | |
cafe5635 KO |
466 | */ |
467 | #define CACHE_SET_UNREGISTERING 0 | |
468 | #define CACHE_SET_STOPPING 1 | |
bf0c55c9 | 469 | #define CACHE_SET_RUNNING 2 |
cafe5635 KO |
470 | |
471 | struct cache_set { | |
472 | struct closure cl; | |
473 | ||
474 | struct list_head list; | |
475 | struct kobject kobj; | |
476 | struct kobject internal; | |
477 | struct dentry *debug; | |
478 | struct cache_accounting accounting; | |
479 | ||
480 | unsigned long flags; | |
481 | ||
482 | struct cache_sb sb; | |
483 | ||
484 | struct cache *cache[MAX_CACHES_PER_SET]; | |
485 | struct cache *cache_by_alloc[MAX_CACHES_PER_SET]; | |
486 | int caches_loaded; | |
487 | ||
488 | struct bcache_device **devices; | |
489 | struct list_head cached_devs; | |
490 | uint64_t cached_dev_sectors; | |
491 | struct closure caching; | |
492 | ||
cb7a583e KO |
493 | struct closure sb_write; |
494 | struct semaphore sb_write_mutex; | |
cafe5635 KO |
495 | |
496 | mempool_t *search; | |
497 | mempool_t *bio_meta; | |
498 | struct bio_set *bio_split; | |
499 | ||
500 | /* For the btree cache */ | |
501 | struct shrinker shrink; | |
502 | ||
cafe5635 KO |
503 | /* For the btree cache and anything allocation related */ |
504 | struct mutex bucket_lock; | |
505 | ||
506 | /* log2(bucket_size), in sectors */ | |
507 | unsigned short bucket_bits; | |
508 | ||
509 | /* log2(block_size), in sectors */ | |
510 | unsigned short block_bits; | |
511 | ||
512 | /* | |
513 | * Default number of pages for a new btree node - may be less than a | |
514 | * full bucket | |
515 | */ | |
516 | unsigned btree_pages; | |
517 | ||
518 | /* | |
519 | * Lists of struct btrees; lru is the list for structs that have memory | |
520 | * allocated for actual btree node, freed is for structs that do not. | |
521 | * | |
522 | * We never free a struct btree, except on shutdown - we just put it on | |
523 | * the btree_cache_freed list and reuse it later. This simplifies the | |
524 | * code, and it doesn't cost us much memory as the memory usage is | |
525 | * dominated by buffers that hold the actual btree node data and those | |
526 | * can be freed - and the number of struct btrees allocated is | |
527 | * effectively bounded. | |
528 | * | |
529 | * btree_cache_freeable effectively is a small cache - we use it because | |
530 | * high order page allocations can be rather expensive, and it's quite | |
531 | * common to delete and allocate btree nodes in quick succession. It | |
532 | * should never grow past ~2-3 nodes in practice. | |
533 | */ | |
534 | struct list_head btree_cache; | |
535 | struct list_head btree_cache_freeable; | |
536 | struct list_head btree_cache_freed; | |
537 | ||
538 | /* Number of elements in btree_cache + btree_cache_freeable lists */ | |
0a63b66d | 539 | unsigned btree_cache_used; |
cafe5635 KO |
540 | |
541 | /* | |
542 | * If we need to allocate memory for a new btree node and that | |
543 | * allocation fails, we can cannibalize another node in the btree cache | |
0a63b66d KO |
544 | * to satisfy the allocation - lock to guarantee only one thread does |
545 | * this at a time: | |
cafe5635 | 546 | */ |
0a63b66d KO |
547 | wait_queue_head_t btree_cache_wait; |
548 | struct task_struct *btree_cache_alloc_lock; | |
cafe5635 KO |
549 | |
550 | /* | |
551 | * When we free a btree node, we increment the gen of the bucket the | |
552 | * node is in - but we can't rewrite the prios and gens until we | |
553 | * finished whatever it is we were doing, otherwise after a crash the | |
554 | * btree node would be freed but for say a split, we might not have the | |
555 | * pointers to the new nodes inserted into the btree yet. | |
556 | * | |
557 | * This is a refcount that blocks prio_write() until the new keys are | |
558 | * written. | |
559 | */ | |
560 | atomic_t prio_blocked; | |
35fcd848 | 561 | wait_queue_head_t bucket_wait; |
cafe5635 KO |
562 | |
563 | /* | |
564 | * For any bio we don't skip we subtract the number of sectors from | |
565 | * rescale; when it hits 0 we rescale all the bucket priorities. | |
566 | */ | |
567 | atomic_t rescale; | |
568 | /* | |
569 | * When we invalidate buckets, we use both the priority and the amount | |
570 | * of good data to determine which buckets to reuse first - to weight | |
571 | * those together consistently we keep track of the smallest nonzero | |
572 | * priority of any bucket. | |
573 | */ | |
574 | uint16_t min_prio; | |
575 | ||
576 | /* | |
3a2fd9d5 | 577 | * max(gen - last_gc) for all buckets. When it gets too big we have to gc |
cafe5635 KO |
578 | * to keep gens from wrapping around. |
579 | */ | |
580 | uint8_t need_gc; | |
581 | struct gc_stat gc_stats; | |
582 | size_t nbuckets; | |
583 | ||
72a44517 | 584 | struct task_struct *gc_thread; |
cafe5635 KO |
585 | /* Where in the btree gc currently is */ |
586 | struct bkey gc_done; | |
587 | ||
588 | /* | |
589 | * The allocation code needs gc_mark in struct bucket to be correct, but | |
590 | * it's not while a gc is in progress. Protected by bucket_lock. | |
591 | */ | |
592 | int gc_mark_valid; | |
593 | ||
594 | /* Counts how many sectors bio_insert has added to the cache */ | |
595 | atomic_t sectors_to_gc; | |
be628be0 | 596 | wait_queue_head_t gc_wait; |
cafe5635 | 597 | |
cafe5635 KO |
598 | struct keybuf moving_gc_keys; |
599 | /* Number of moving GC bios in flight */ | |
72a44517 | 600 | struct semaphore moving_in_flight; |
cafe5635 | 601 | |
da415a09 NS |
602 | struct workqueue_struct *moving_gc_wq; |
603 | ||
cafe5635 KO |
604 | struct btree *root; |
605 | ||
606 | #ifdef CONFIG_BCACHE_DEBUG | |
607 | struct btree *verify_data; | |
78b77bf8 | 608 | struct bset *verify_ondisk; |
cafe5635 KO |
609 | struct mutex verify_lock; |
610 | #endif | |
611 | ||
612 | unsigned nr_uuids; | |
613 | struct uuid_entry *uuids; | |
614 | BKEY_PADDED(uuid_bucket); | |
cb7a583e KO |
615 | struct closure uuid_write; |
616 | struct semaphore uuid_write_mutex; | |
cafe5635 KO |
617 | |
618 | /* | |
619 | * A btree node on disk could have too many bsets for an iterator to fit | |
57943511 | 620 | * on the stack - have to dynamically allocate them |
cafe5635 | 621 | */ |
57943511 | 622 | mempool_t *fill_iter; |
cafe5635 | 623 | |
67539e85 | 624 | struct bset_sort_state sort; |
cafe5635 KO |
625 | |
626 | /* List of buckets we're currently writing data to */ | |
627 | struct list_head data_buckets; | |
628 | spinlock_t data_bucket_lock; | |
629 | ||
630 | struct journal journal; | |
631 | ||
632 | #define CONGESTED_MAX 1024 | |
633 | unsigned congested_last_us; | |
634 | atomic_t congested; | |
635 | ||
636 | /* The rest of this all shows up in sysfs */ | |
637 | unsigned congested_read_threshold_us; | |
638 | unsigned congested_write_threshold_us; | |
639 | ||
cafe5635 KO |
640 | struct time_stats btree_gc_time; |
641 | struct time_stats btree_split_time; | |
cafe5635 | 642 | struct time_stats btree_read_time; |
cafe5635 KO |
643 | |
644 | atomic_long_t cache_read_races; | |
645 | atomic_long_t writeback_keys_done; | |
646 | atomic_long_t writeback_keys_failed; | |
77c320eb KO |
647 | |
648 | enum { | |
649 | ON_ERROR_UNREGISTER, | |
650 | ON_ERROR_PANIC, | |
651 | } on_error; | |
cafe5635 KO |
652 | unsigned error_limit; |
653 | unsigned error_decay; | |
77c320eb | 654 | |
cafe5635 | 655 | unsigned short journal_delay_ms; |
a85e968e | 656 | bool expensive_debug_checks; |
cafe5635 KO |
657 | unsigned verify:1; |
658 | unsigned key_merging_disabled:1; | |
659 | unsigned gc_always_rewrite:1; | |
660 | unsigned shrinker_disabled:1; | |
661 | unsigned copy_gc_enabled:1; | |
662 | ||
663 | #define BUCKET_HASH_BITS 12 | |
664 | struct hlist_head bucket_hash[1 << BUCKET_HASH_BITS]; | |
665 | }; | |
666 | ||
cafe5635 KO |
667 | struct bbio { |
668 | unsigned submit_time_us; | |
669 | union { | |
670 | struct bkey key; | |
671 | uint64_t _pad[3]; | |
672 | /* | |
673 | * We only need pad = 3 here because we only ever carry around a | |
674 | * single pointer - i.e. the pointer we're doing io to/from. | |
675 | */ | |
676 | }; | |
677 | struct bio bio; | |
678 | }; | |
679 | ||
cafe5635 | 680 | #define BTREE_PRIO USHRT_MAX |
e0a985a4 | 681 | #define INITIAL_PRIO 32768U |
cafe5635 KO |
682 | |
683 | #define btree_bytes(c) ((c)->btree_pages * PAGE_SIZE) | |
684 | #define btree_blocks(b) \ | |
685 | ((unsigned) (KEY_SIZE(&b->key) >> (b)->c->block_bits)) | |
686 | ||
687 | #define btree_default_blocks(c) \ | |
688 | ((unsigned) ((PAGE_SECTORS * (c)->btree_pages) >> (c)->block_bits)) | |
689 | ||
690 | #define bucket_pages(c) ((c)->sb.bucket_size / PAGE_SECTORS) | |
691 | #define bucket_bytes(c) ((c)->sb.bucket_size << 9) | |
692 | #define block_bytes(c) ((c)->sb.block_size << 9) | |
693 | ||
cafe5635 KO |
694 | #define prios_per_bucket(c) \ |
695 | ((bucket_bytes(c) - sizeof(struct prio_set)) / \ | |
696 | sizeof(struct bucket_disk)) | |
697 | #define prio_buckets(c) \ | |
698 | DIV_ROUND_UP((size_t) (c)->sb.nbuckets, prios_per_bucket(c)) | |
699 | ||
cafe5635 KO |
700 | static inline size_t sector_to_bucket(struct cache_set *c, sector_t s) |
701 | { | |
702 | return s >> c->bucket_bits; | |
703 | } | |
704 | ||
705 | static inline sector_t bucket_to_sector(struct cache_set *c, size_t b) | |
706 | { | |
707 | return ((sector_t) b) << c->bucket_bits; | |
708 | } | |
709 | ||
710 | static inline sector_t bucket_remainder(struct cache_set *c, sector_t s) | |
711 | { | |
712 | return s & (c->sb.bucket_size - 1); | |
713 | } | |
714 | ||
715 | static inline struct cache *PTR_CACHE(struct cache_set *c, | |
716 | const struct bkey *k, | |
717 | unsigned ptr) | |
718 | { | |
719 | return c->cache[PTR_DEV(k, ptr)]; | |
720 | } | |
721 | ||
722 | static inline size_t PTR_BUCKET_NR(struct cache_set *c, | |
723 | const struct bkey *k, | |
724 | unsigned ptr) | |
725 | { | |
726 | return sector_to_bucket(c, PTR_OFFSET(k, ptr)); | |
727 | } | |
728 | ||
729 | static inline struct bucket *PTR_BUCKET(struct cache_set *c, | |
730 | const struct bkey *k, | |
731 | unsigned ptr) | |
732 | { | |
733 | return PTR_CACHE(c, k, ptr)->buckets + PTR_BUCKET_NR(c, k, ptr); | |
734 | } | |
735 | ||
9a02b7ee KO |
736 | static inline uint8_t gen_after(uint8_t a, uint8_t b) |
737 | { | |
738 | uint8_t r = a - b; | |
739 | return r > 128U ? 0 : r; | |
740 | } | |
741 | ||
742 | static inline uint8_t ptr_stale(struct cache_set *c, const struct bkey *k, | |
743 | unsigned i) | |
744 | { | |
745 | return gen_after(PTR_BUCKET(c, k, i)->gen, PTR_GEN(k, i)); | |
746 | } | |
747 | ||
748 | static inline bool ptr_available(struct cache_set *c, const struct bkey *k, | |
749 | unsigned i) | |
750 | { | |
751 | return (PTR_DEV(k, i) < MAX_CACHES_PER_SET) && PTR_CACHE(c, k, i); | |
752 | } | |
753 | ||
cafe5635 KO |
754 | /* Btree key macros */ |
755 | ||
cafe5635 KO |
756 | /* |
757 | * This is used for various on disk data structures - cache_sb, prio_set, bset, | |
758 | * jset: The checksum is _always_ the first 8 bytes of these structs | |
759 | */ | |
760 | #define csum_set(i) \ | |
169ef1cf | 761 | bch_crc64(((void *) (i)) + sizeof(uint64_t), \ |
fafff81c KO |
762 | ((void *) bset_bkey_last(i)) - \ |
763 | (((void *) (i)) + sizeof(uint64_t))) | |
cafe5635 KO |
764 | |
765 | /* Error handling macros */ | |
766 | ||
767 | #define btree_bug(b, ...) \ | |
768 | do { \ | |
769 | if (bch_cache_set_error((b)->c, __VA_ARGS__)) \ | |
770 | dump_stack(); \ | |
771 | } while (0) | |
772 | ||
773 | #define cache_bug(c, ...) \ | |
774 | do { \ | |
775 | if (bch_cache_set_error(c, __VA_ARGS__)) \ | |
776 | dump_stack(); \ | |
777 | } while (0) | |
778 | ||
779 | #define btree_bug_on(cond, b, ...) \ | |
780 | do { \ | |
781 | if (cond) \ | |
782 | btree_bug(b, __VA_ARGS__); \ | |
783 | } while (0) | |
784 | ||
785 | #define cache_bug_on(cond, c, ...) \ | |
786 | do { \ | |
787 | if (cond) \ | |
788 | cache_bug(c, __VA_ARGS__); \ | |
789 | } while (0) | |
790 | ||
791 | #define cache_set_err_on(cond, c, ...) \ | |
792 | do { \ | |
793 | if (cond) \ | |
794 | bch_cache_set_error(c, __VA_ARGS__); \ | |
795 | } while (0) | |
796 | ||
797 | /* Looping macros */ | |
798 | ||
799 | #define for_each_cache(ca, cs, iter) \ | |
800 | for (iter = 0; ca = cs->cache[iter], iter < (cs)->sb.nr_in_set; iter++) | |
801 | ||
802 | #define for_each_bucket(b, ca) \ | |
803 | for (b = (ca)->buckets + (ca)->sb.first_bucket; \ | |
804 | b < (ca)->buckets + (ca)->sb.nbuckets; b++) | |
805 | ||
cafe5635 KO |
806 | static inline void cached_dev_put(struct cached_dev *dc) |
807 | { | |
808 | if (atomic_dec_and_test(&dc->count)) | |
809 | schedule_work(&dc->detach); | |
810 | } | |
811 | ||
812 | static inline bool cached_dev_get(struct cached_dev *dc) | |
813 | { | |
814 | if (!atomic_inc_not_zero(&dc->count)) | |
815 | return false; | |
816 | ||
817 | /* Paired with the mb in cached_dev_attach */ | |
4e857c58 | 818 | smp_mb__after_atomic(); |
cafe5635 KO |
819 | return true; |
820 | } | |
821 | ||
822 | /* | |
823 | * bucket_gc_gen() returns the difference between the bucket's current gen and | |
824 | * the oldest gen of any pointer into that bucket in the btree (last_gc). | |
cafe5635 KO |
825 | */ |
826 | ||
827 | static inline uint8_t bucket_gc_gen(struct bucket *b) | |
828 | { | |
829 | return b->gen - b->last_gc; | |
830 | } | |
831 | ||
cafe5635 | 832 | #define BUCKET_GC_GEN_MAX 96U |
cafe5635 KO |
833 | |
834 | #define kobj_attribute_write(n, fn) \ | |
835 | static struct kobj_attribute ksysfs_##n = __ATTR(n, S_IWUSR, NULL, fn) | |
836 | ||
837 | #define kobj_attribute_rw(n, show, store) \ | |
838 | static struct kobj_attribute ksysfs_##n = \ | |
839 | __ATTR(n, S_IWUSR|S_IRUSR, show, store) | |
840 | ||
119ba0f8 KO |
841 | static inline void wake_up_allocators(struct cache_set *c) |
842 | { | |
843 | struct cache *ca; | |
844 | unsigned i; | |
845 | ||
846 | for_each_cache(ca, c, i) | |
847 | wake_up_process(ca->alloc_thread); | |
848 | } | |
849 | ||
cafe5635 KO |
850 | /* Forward declarations */ |
851 | ||
4e4cbee9 | 852 | void bch_count_io_errors(struct cache *, blk_status_t, const char *); |
cafe5635 | 853 | void bch_bbio_count_io_errors(struct cache_set *, struct bio *, |
4e4cbee9 CH |
854 | blk_status_t, const char *); |
855 | void bch_bbio_endio(struct cache_set *, struct bio *, blk_status_t, | |
856 | const char *); | |
cafe5635 KO |
857 | void bch_bbio_free(struct bio *, struct cache_set *); |
858 | struct bio *bch_bbio_alloc(struct cache_set *); | |
859 | ||
cafe5635 KO |
860 | void __bch_submit_bbio(struct bio *, struct cache_set *); |
861 | void bch_submit_bbio(struct bio *, struct cache_set *, struct bkey *, unsigned); | |
862 | ||
863 | uint8_t bch_inc_gen(struct cache *, struct bucket *); | |
864 | void bch_rescale_priorities(struct cache_set *, int); | |
cafe5635 | 865 | |
2531d9ee KO |
866 | bool bch_can_invalidate_bucket(struct cache *, struct bucket *); |
867 | void __bch_invalidate_one_bucket(struct cache *, struct bucket *); | |
868 | ||
869 | void __bch_bucket_free(struct cache *, struct bucket *); | |
cafe5635 KO |
870 | void bch_bucket_free(struct cache_set *, struct bkey *); |
871 | ||
2531d9ee | 872 | long bch_bucket_alloc(struct cache *, unsigned, bool); |
cafe5635 | 873 | int __bch_bucket_alloc_set(struct cache_set *, unsigned, |
35fcd848 | 874 | struct bkey *, int, bool); |
cafe5635 | 875 | int bch_bucket_alloc_set(struct cache_set *, unsigned, |
35fcd848 | 876 | struct bkey *, int, bool); |
2599b53b KO |
877 | bool bch_alloc_sectors(struct cache_set *, struct bkey *, unsigned, |
878 | unsigned, unsigned, bool); | |
cafe5635 KO |
879 | |
880 | __printf(2, 3) | |
881 | bool bch_cache_set_error(struct cache_set *, const char *, ...); | |
882 | ||
883 | void bch_prio_write(struct cache *); | |
884 | void bch_write_bdev_super(struct cached_dev *, struct closure *); | |
885 | ||
72a44517 | 886 | extern struct workqueue_struct *bcache_wq; |
cafe5635 KO |
887 | extern const char * const bch_cache_modes[]; |
888 | extern struct mutex bch_register_lock; | |
889 | extern struct list_head bch_cache_sets; | |
890 | ||
891 | extern struct kobj_type bch_cached_dev_ktype; | |
892 | extern struct kobj_type bch_flash_dev_ktype; | |
893 | extern struct kobj_type bch_cache_set_ktype; | |
894 | extern struct kobj_type bch_cache_set_internal_ktype; | |
895 | extern struct kobj_type bch_cache_ktype; | |
896 | ||
897 | void bch_cached_dev_release(struct kobject *); | |
898 | void bch_flash_dev_release(struct kobject *); | |
899 | void bch_cache_set_release(struct kobject *); | |
900 | void bch_cache_release(struct kobject *); | |
901 | ||
902 | int bch_uuid_write(struct cache_set *); | |
903 | void bcache_write_super(struct cache_set *); | |
904 | ||
905 | int bch_flash_dev_create(struct cache_set *c, uint64_t size); | |
906 | ||
907 | int bch_cached_dev_attach(struct cached_dev *, struct cache_set *); | |
908 | void bch_cached_dev_detach(struct cached_dev *); | |
909 | void bch_cached_dev_run(struct cached_dev *); | |
910 | void bcache_device_stop(struct bcache_device *); | |
911 | ||
912 | void bch_cache_set_unregister(struct cache_set *); | |
913 | void bch_cache_set_stop(struct cache_set *); | |
914 | ||
915 | struct cache_set *bch_cache_set_alloc(struct cache_sb *); | |
916 | void bch_btree_cache_free(struct cache_set *); | |
917 | int bch_btree_cache_alloc(struct cache_set *); | |
cafe5635 | 918 | void bch_moving_init_cache_set(struct cache_set *); |
2599b53b KO |
919 | int bch_open_buckets_alloc(struct cache_set *); |
920 | void bch_open_buckets_free(struct cache_set *); | |
cafe5635 | 921 | |
119ba0f8 | 922 | int bch_cache_allocator_start(struct cache *ca); |
cafe5635 KO |
923 | |
924 | void bch_debug_exit(void); | |
925 | int bch_debug_init(struct kobject *); | |
cafe5635 KO |
926 | void bch_request_exit(void); |
927 | int bch_request_init(void); | |
cafe5635 KO |
928 | |
929 | #endif /* _BCACHE_H */ |