1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _BCACHE_H
3 #define _BCACHE_H
4
5 /*
6 * SOME HIGH LEVEL CODE DOCUMENTATION:
7 *
8 * Bcache mostly works with cache sets, cache devices, and backing devices.
9 *
10 * Support for multiple cache devices hasn't quite been finished off yet, but
11 * it's about 95% plumbed through. A cache set and its cache devices is sort of
12 * like a md raid array and its component devices. Most of the code doesn't care
13 * about individual cache devices, the main abstraction is the cache set.
14 *
15 * Multiple cache devices is intended to give us the ability to mirror dirty
16 * cached data and metadata, without mirroring clean cached data.
17 *
18 * Backing devices are different, in that they have a lifetime independent of a
19 * cache set. When you register a newly formatted backing device it'll come up
20 * in passthrough mode, and then you can attach and detach a backing device from
21 * a cache set at runtime - while it's mounted and in use. Detaching implicitly
22 * invalidates any cached data for that backing device.
23 *
24 * A cache set can have multiple (many) backing devices attached to it.
25 *
26 * There's also flash only volumes - this is the reason for the distinction
27 * between struct cached_dev and struct bcache_device. A flash only volume
28 * works much like a bcache device that has a backing device, except the
29 * "cached" data is always dirty. The end result is that we get thin
30 * provisioning with very little additional code.
31 *
32 * Flash only volumes work but they're not production ready because the moving
33 * garbage collector needs more work. More on that later.
34 *
35 * BUCKETS/ALLOCATION:
36 *
37 * Bcache is primarily designed for caching, which means that in normal
38 * operation all of our available space will be allocated. Thus, we need an
39 * efficient way of deleting things from the cache so we can write new things to
40 * it.
41 *
42 * To do this, we first divide the cache device up into buckets. A bucket is the
43 * unit of allocation; they're typically around 1 mb - anywhere from 128k to 2M+
44 * works efficiently.
45 *
46 * Each bucket has a 16 bit priority, and an 8 bit generation associated with
47 * it. The gens and priorities for all the buckets are stored contiguously and
48 * packed on disk (in a linked list of buckets - aside from the superblock, all
49 * of bcache's metadata is stored in buckets).
50 *
51 * The priority is used to implement an LRU. We reset a bucket's priority when
52 * we allocate it or on cache it, and every so often we decrement the priority
53 * of each bucket. It could be used to implement something more sophisticated,
54 * if anyone ever gets around to it.
55 *
56 * The generation is used for invalidating buckets. Each pointer also has an 8
57 * bit generation embedded in it; for a pointer to be considered valid, its gen
58 * must match the gen of the bucket it points into. Thus, to reuse a bucket all
59 * we have to do is increment its gen (and write its new gen to disk; we batch
60 * this up).
61 *
62 * Bcache is entirely COW - we never write twice to a bucket, even buckets that
63 * contain metadata (including btree nodes).
64 *
65 * THE BTREE:
66 *
67 * Bcache is in large part design around the btree.
68 *
69 * At a high level, the btree is just an index of key -> ptr tuples.
70 *
71 * Keys represent extents, and thus have a size field. Keys also have a variable
72 * number of pointers attached to them (potentially zero, which is handy for
73 * invalidating the cache).
74 *
75 * The key itself is an inode:offset pair. The inode number corresponds to a
76 * backing device or a flash only volume. The offset is the ending offset of the
77 * extent within the inode - not the starting offset; this makes lookups
78 * slightly more convenient.
79 *
80 * Pointers contain the cache device id, the offset on that device, and an 8 bit
81 * generation number. More on the gen later.
82 *
83 * Index lookups are not fully abstracted - cache lookups in particular are
84 * still somewhat mixed in with the btree code, but things are headed in that
85 * direction.
86 *
87 * Updates are fairly well abstracted, though. There are two different ways of
88 * updating the btree; insert and replace.
89 *
90 * BTREE_INSERT will just take a list of keys and insert them into the btree -
91 * overwriting (possibly only partially) any extents they overlap with. This is
92 * used to update the index after a write.
93 *
94 * BTREE_REPLACE is really cmpxchg(); it inserts a key into the btree iff it is
95 * overwriting a key that matches another given key. This is used for inserting
96 * data into the cache after a cache miss, and for background writeback, and for
97 * the moving garbage collector.
98 *
99 * There is no "delete" operation; deleting things from the index is
100 * accomplished by either by invalidating pointers (by incrementing a bucket's
101 * gen) or by inserting a key with 0 pointers - which will overwrite anything
102 * previously present at that location in the index.
103 *
104 * This means that there are always stale/invalid keys in the btree. They're
105 * filtered out by the code that iterates through a btree node, and removed when
106 * a btree node is rewritten.
107 *
108 * BTREE NODES:
109 *
110 * Our unit of allocation is a bucket, and we can't arbitrarily allocate and
111 * free smaller than a bucket - so, that's how big our btree nodes are.
112 *
113 * (If buckets are really big we'll only use part of the bucket for a btree node
114 * - no less than 1/4th - but a bucket still contains no more than a single
115 * btree node. I'd actually like to change this, but for now we rely on the
116 * bucket's gen for deleting btree nodes when we rewrite/split a node.)
117 *
118 * Anyways, btree nodes are big - big enough to be inefficient with a textbook
119 * btree implementation.
120 *
121 * The way this is solved is that btree nodes are internally log structured; we
122 * can append new keys to an existing btree node without rewriting it. This
123 * means each set of keys we write is sorted, but the node is not.
124 *
125 * We maintain this log structure in memory - keeping 1Mb of keys sorted would
126 * be expensive, and we have to distinguish between the keys we have written and
127 * the keys we haven't. So to do a lookup in a btree node, we have to search
128 * each sorted set. But we do merge written sets together lazily, so the cost of
129 * these extra searches is quite low (normally most of the keys in a btree node
130 * will be in one big set, and then there'll be one or two sets that are much
131 * smaller).
132 *
133 * This log structure makes bcache's btree more of a hybrid between a
134 * conventional btree and a compacting data structure, with some of the
135 * advantages of both.
136 *
137 * GARBAGE COLLECTION:
138 *
139 * We can't just invalidate any bucket - it might contain dirty data or
140 * metadata. If it once contained dirty data, other writes might overwrite it
141 * later, leaving no valid pointers into that bucket in the index.
142 *
143 * Thus, the primary purpose of garbage collection is to find buckets to reuse.
144 * It also counts how much valid data it each bucket currently contains, so that
145 * allocation can reuse buckets sooner when they've been mostly overwritten.
146 *
147 * It also does some things that are really internal to the btree
148 * implementation. If a btree node contains pointers that are stale by more than
149 * some threshold, it rewrites the btree node to avoid the bucket's generation
150 * wrapping around. It also merges adjacent btree nodes if they're empty enough.
151 *
152 * THE JOURNAL:
153 *
154 * Bcache's journal is not necessary for consistency; we always strictly
155 * order metadata writes so that the btree and everything else is consistent on
156 * disk in the event of an unclean shutdown, and in fact bcache had writeback
157 * caching (with recovery from unclean shutdown) before journalling was
158 * implemented.
159 *
160 * Rather, the journal is purely a performance optimization; we can't complete a
161 * write until we've updated the index on disk, otherwise the cache would be
162 * inconsistent in the event of an unclean shutdown. This means that without the
163 * journal, on random write workloads we constantly have to update all the leaf
164 * nodes in the btree, and those writes will be mostly empty (appending at most
165 * a few keys each) - highly inefficient in terms of amount of metadata writes,
166 * and it puts more strain on the various btree resorting/compacting code.
167 *
168 * The journal is just a log of keys we've inserted; on startup we just reinsert
169 * all the keys in the open journal entries. That means that when we're updating
170 * a node in the btree, we can wait until a 4k block of keys fills up before
171 * writing them out.
172 *
173 * For simplicity, we only journal updates to leaf nodes; updates to parent
174 * nodes are rare enough (since our leaf nodes are huge) that it wasn't worth
175 * the complexity to deal with journalling them (in particular, journal replay)
176 * - updates to non leaf nodes just happen synchronously (see btree_split()).
177 */
178
179 #define pr_fmt(fmt) "bcache: %s() " fmt, __func__
180
181 #include <linux/bio.h>
182 #include <linux/kobject.h>
183 #include <linux/list.h>
184 #include <linux/mutex.h>
185 #include <linux/rbtree.h>
186 #include <linux/rwsem.h>
187 #include <linux/refcount.h>
188 #include <linux/types.h>
189 #include <linux/workqueue.h>
190 #include <linux/kthread.h>
191
192 #include "bcache_ondisk.h"
193 #include "bset.h"
194 #include "util.h"
195 #include "closure.h"
196
197 struct bucket {
198 atomic_t pin;
199 uint16_t prio;
200 uint8_t gen;
201 uint8_t last_gc; /* Most out of date gen in the btree */
202 uint16_t gc_mark; /* Bitfield used by GC. See below for field */
203 };
204
205 /*
206 * I'd use bitfields for these, but I don't trust the compiler not to screw me
207 * as multiple threads touch struct bucket without locking
208 */
209
210 BITMASK(GC_MARK, struct bucket, gc_mark, 0, 2);
211 #define GC_MARK_RECLAIMABLE 1
212 #define GC_MARK_DIRTY 2
213 #define GC_MARK_METADATA 3
214 #define GC_SECTORS_USED_SIZE 13
215 #define MAX_GC_SECTORS_USED (~(~0ULL << GC_SECTORS_USED_SIZE))
216 BITMASK(GC_SECTORS_USED, struct bucket, gc_mark, 2, GC_SECTORS_USED_SIZE);
217 BITMASK(GC_MOVE, struct bucket, gc_mark, 15, 1);
218
219 #include "journal.h"
220 #include "stats.h"
221 struct search;
222 struct btree;
223 struct keybuf;
224
225 struct keybuf_key {
226 struct rb_node node;
227 BKEY_PADDED(key);
228 void *private;
229 };
230
231 struct keybuf {
232 struct bkey last_scanned;
233 spinlock_t lock;
234
235 /*
236 * Beginning and end of range in rb tree - so that we can skip taking
237 * lock and checking the rb tree when we need to check for overlapping
238 * keys.
239 */
240 struct bkey start;
241 struct bkey end;
242
243 struct rb_root keys;
244
245 #define KEYBUF_NR 500
246 DECLARE_ARRAY_ALLOCATOR(struct keybuf_key, freelist, KEYBUF_NR);
247 };
248
249 struct bcache_device {
250 struct closure cl;
251
252 struct kobject kobj;
253
254 struct cache_set *c;
255 unsigned int id;
256 #define BCACHEDEVNAME_SIZE 12
257 char name[BCACHEDEVNAME_SIZE];
258
259 struct gendisk *disk;
260
261 unsigned long flags;
262 #define BCACHE_DEV_CLOSING 0
263 #define BCACHE_DEV_DETACHING 1
264 #define BCACHE_DEV_UNLINK_DONE 2
265 #define BCACHE_DEV_WB_RUNNING 3
266 #define BCACHE_DEV_RATE_DW_RUNNING 4
267 int nr_stripes;
268 #define BCH_MIN_STRIPE_SZ ((4 << 20) >> SECTOR_SHIFT)
269 unsigned int stripe_size;
270 atomic_t *stripe_sectors_dirty;
271 unsigned long *full_dirty_stripes;
272
273 struct bio_set bio_split;
274
275 unsigned int data_csum:1;
276
277 int (*cache_miss)(struct btree *b, struct search *s,
278 struct bio *bio, unsigned int sectors);
279 int (*ioctl)(struct bcache_device *d, fmode_t mode,
280 unsigned int cmd, unsigned long arg);
281 };
282
283 struct io {
284 /* Used to track sequential IO so it can be skipped */
285 struct hlist_node hash;
286 struct list_head lru;
287
288 unsigned long jiffies;
289 unsigned int sequential;
290 sector_t last;
291 };
292
293 enum stop_on_failure {
294 BCH_CACHED_DEV_STOP_AUTO = 0,
295 BCH_CACHED_DEV_STOP_ALWAYS,
296 BCH_CACHED_DEV_STOP_MODE_MAX,
297 };
298
299 struct cached_dev {
300 struct list_head list;
301 struct bcache_device disk;
302 struct block_device *bdev;
303
304 struct cache_sb sb;
305 struct cache_sb_disk *sb_disk;
306 struct bio sb_bio;
307 struct bio_vec sb_bv[1];
308 struct closure sb_write;
309 struct semaphore sb_write_mutex;
310
311 /* Refcount on the cache set. Always nonzero when we're caching. */
312 refcount_t count;
313 struct work_struct detach;
314
315 /*
316 * Device might not be running if it's dirty and the cache set hasn't
317 * showed up yet.
318 */
319 atomic_t running;
320
321 /*
322 * Writes take a shared lock from start to finish; scanning for dirty
323 * data to refill the rb tree requires an exclusive lock.
324 */
325 struct rw_semaphore writeback_lock;
326
327 /*
328 * Nonzero, and writeback has a refcount (d->count), iff there is dirty
329 * data in the cache. Protected by writeback_lock; must have an
330 * shared lock to set and exclusive lock to clear.
331 */
332 atomic_t has_dirty;
333
334 #define BCH_CACHE_READA_ALL 0
335 #define BCH_CACHE_READA_META_ONLY 1
336 unsigned int cache_readahead_policy;
337 struct bch_ratelimit writeback_rate;
338 struct delayed_work writeback_rate_update;
339
340 /* Limit number of writeback bios in flight */
341 struct semaphore in_flight;
342 struct task_struct *writeback_thread;
343 struct workqueue_struct *writeback_write_wq;
344
345 struct keybuf writeback_keys;
346
347 struct task_struct *status_update_thread;
348 /*
349 * Order the write-half of writeback operations strongly in dispatch
350 * order. (Maintain LBA order; don't allow reads completing out of
351 * order to re-order the writes...)
352 */
353 struct closure_waitlist writeback_ordering_wait;
354 atomic_t writeback_sequence_next;
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 int sequential_cutoff;
368
369 unsigned int io_disable:1;
370 unsigned int verify:1;
371 unsigned int bypass_torture_test:1;
372
373 unsigned int partial_stripes_expensive:1;
374 unsigned int writeback_metadata:1;
375 unsigned int writeback_running:1;
376 unsigned int writeback_consider_fragment:1;
377 unsigned char writeback_percent;
378 unsigned int writeback_delay;
379
380 uint64_t writeback_rate_target;
381 int64_t writeback_rate_proportional;
382 int64_t writeback_rate_integral;
383 int64_t writeback_rate_integral_scaled;
384 int32_t writeback_rate_change;
385
386 unsigned int writeback_rate_update_seconds;
387 unsigned int writeback_rate_i_term_inverse;
388 unsigned int writeback_rate_p_term_inverse;
389 unsigned int writeback_rate_fp_term_low;
390 unsigned int writeback_rate_fp_term_mid;
391 unsigned int writeback_rate_fp_term_high;
392 unsigned int writeback_rate_minimum;
393
394 enum stop_on_failure stop_when_cache_set_failed;
395 #define DEFAULT_CACHED_DEV_ERROR_LIMIT 64
396 atomic_t io_errors;
397 unsigned int error_limit;
398 unsigned int offline_seconds;
399
400 /*
401 * Retry to update writeback_rate if contention happens for
402 * down_read(dc->writeback_lock) in update_writeback_rate()
403 */
404 #define BCH_WBRATE_UPDATE_MAX_SKIPS 15
405 unsigned int rate_update_retry;
406 };
407
408 enum alloc_reserve {
409 RESERVE_BTREE,
410 RESERVE_PRIO,
411 RESERVE_MOVINGGC,
412 RESERVE_NONE,
413 RESERVE_NR,
414 };
415
416 struct cache {
417 struct cache_set *set;
418 struct cache_sb sb;
419 struct cache_sb_disk *sb_disk;
420 struct bio sb_bio;
421 struct bio_vec sb_bv[1];
422
423 struct kobject kobj;
424 struct block_device *bdev;
425
426 struct task_struct *alloc_thread;
427
428 struct closure prio;
429 struct prio_set *disk_buckets;
430
431 /*
432 * When allocating new buckets, prio_write() gets first dibs - since we
433 * may not be allocate at all without writing priorities and gens.
434 * prio_last_buckets[] contains the last buckets we wrote priorities to
435 * (so gc can mark them as metadata), prio_buckets[] contains the
436 * buckets allocated for the next prio write.
437 */
438 uint64_t *prio_buckets;
439 uint64_t *prio_last_buckets;
440
441 /*
442 * free: Buckets that are ready to be used
443 *
444 * free_inc: Incoming buckets - these are buckets that currently have
445 * cached data in them, and we can't reuse them until after we write
446 * their new gen to disk. After prio_write() finishes writing the new
447 * gens/prios, they'll be moved to the free list (and possibly discarded
448 * in the process)
449 */
450 DECLARE_FIFO(long, free)[RESERVE_NR];
451 DECLARE_FIFO(long, free_inc);
452
453 size_t fifo_last_bucket;
454
455 /* Allocation stuff: */
456 struct bucket *buckets;
457
458 DECLARE_HEAP(struct bucket *, heap);
459
460 /*
461 * If nonzero, we know we aren't going to find any buckets to invalidate
462 * until a gc finishes - otherwise we could pointlessly burn a ton of
463 * cpu
464 */
465 unsigned int invalidate_needs_gc;
466
467 bool discard; /* Get rid of? */
468
469 struct journal_device journal;
470
471 /* The rest of this all shows up in sysfs */
472 #define IO_ERROR_SHIFT 20
473 atomic_t io_errors;
474 atomic_t io_count;
475
476 atomic_long_t meta_sectors_written;
477 atomic_long_t btree_sectors_written;
478 atomic_long_t sectors_written;
479 };
480
481 struct gc_stat {
482 size_t nodes;
483 size_t nodes_pre;
484 size_t key_bytes;
485
486 size_t nkeys;
487 uint64_t data; /* sectors */
488 unsigned int in_use; /* percent */
489 };
490
491 /*
492 * Flag bits, for how the cache set is shutting down, and what phase it's at:
493 *
494 * CACHE_SET_UNREGISTERING means we're not just shutting down, we're detaching
495 * all the backing devices first (their cached data gets invalidated, and they
496 * won't automatically reattach).
497 *
498 * CACHE_SET_STOPPING always gets set first when we're closing down a cache set;
499 * we'll continue to run normally for awhile with CACHE_SET_STOPPING set (i.e.
500 * flushing dirty data).
501 *
502 * CACHE_SET_RUNNING means all cache devices have been registered and journal
503 * replay is complete.
504 *
505 * CACHE_SET_IO_DISABLE is set when bcache is stopping the whold cache set, all
506 * external and internal I/O should be denied when this flag is set.
507 *
508 */
509 #define CACHE_SET_UNREGISTERING 0
510 #define CACHE_SET_STOPPING 1
511 #define CACHE_SET_RUNNING 2
512 #define CACHE_SET_IO_DISABLE 3
513
514 struct cache_set {
515 struct closure cl;
516
517 struct list_head list;
518 struct kobject kobj;
519 struct kobject internal;
520 struct dentry *debug;
521 struct cache_accounting accounting;
522
523 unsigned long flags;
524 atomic_t idle_counter;
525 atomic_t at_max_writeback_rate;
526
527 struct cache *cache;
528
529 struct bcache_device **devices;
530 unsigned int devices_max_used;
531 atomic_t attached_dev_nr;
532 struct list_head cached_devs;
533 uint64_t cached_dev_sectors;
534 atomic_long_t flash_dev_dirty_sectors;
535 struct closure caching;
536
537 struct closure sb_write;
538 struct semaphore sb_write_mutex;
539
540 mempool_t search;
541 mempool_t bio_meta;
542 struct bio_set bio_split;
543
544 /* For the btree cache */
545 struct shrinker shrink;
546
547 /* For the btree cache and anything allocation related */
548 struct mutex bucket_lock;
549
550 /* log2(bucket_size), in sectors */
551 unsigned short bucket_bits;
552
553 /* log2(block_size), in sectors */
554 unsigned short block_bits;
555
556 /*
557 * Default number of pages for a new btree node - may be less than a
558 * full bucket
559 */
560 unsigned int btree_pages;
561
562 /*
563 * Lists of struct btrees; lru is the list for structs that have memory
564 * allocated for actual btree node, freed is for structs that do not.
565 *
566 * We never free a struct btree, except on shutdown - we just put it on
567 * the btree_cache_freed list and reuse it later. This simplifies the
568 * code, and it doesn't cost us much memory as the memory usage is
569 * dominated by buffers that hold the actual btree node data and those
570 * can be freed - and the number of struct btrees allocated is
571 * effectively bounded.
572 *
573 * btree_cache_freeable effectively is a small cache - we use it because
574 * high order page allocations can be rather expensive, and it's quite
575 * common to delete and allocate btree nodes in quick succession. It
576 * should never grow past ~2-3 nodes in practice.
577 */
578 struct list_head btree_cache;
579 struct list_head btree_cache_freeable;
580 struct list_head btree_cache_freed;
581
582 /* Number of elements in btree_cache + btree_cache_freeable lists */
583 unsigned int btree_cache_used;
584
585 /*
586 * If we need to allocate memory for a new btree node and that
587 * allocation fails, we can cannibalize another node in the btree cache
588 * to satisfy the allocation - lock to guarantee only one thread does
589 * this at a time:
590 */
591 wait_queue_head_t btree_cache_wait;
592 struct task_struct *btree_cache_alloc_lock;
593 spinlock_t btree_cannibalize_lock;
594
595 /*
596 * When we free a btree node, we increment the gen of the bucket the
597 * node is in - but we can't rewrite the prios and gens until we
598 * finished whatever it is we were doing, otherwise after a crash the
599 * btree node would be freed but for say a split, we might not have the
600 * pointers to the new nodes inserted into the btree yet.
601 *
602 * This is a refcount that blocks prio_write() until the new keys are
603 * written.
604 */
605 atomic_t prio_blocked;
606 wait_queue_head_t bucket_wait;
607
608 /*
609 * For any bio we don't skip we subtract the number of sectors from
610 * rescale; when it hits 0 we rescale all the bucket priorities.
611 */
612 atomic_t rescale;
613 /*
614 * used for GC, identify if any front side I/Os is inflight
615 */
616 atomic_t search_inflight;
617 /*
618 * When we invalidate buckets, we use both the priority and the amount
619 * of good data to determine which buckets to reuse first - to weight
620 * those together consistently we keep track of the smallest nonzero
621 * priority of any bucket.
622 */
623 uint16_t min_prio;
624
625 /*
626 * max(gen - last_gc) for all buckets. When it gets too big we have to
627 * gc to keep gens from wrapping around.
628 */
629 uint8_t need_gc;
630 struct gc_stat gc_stats;
631 size_t nbuckets;
632 size_t avail_nbuckets;
633
634 struct task_struct *gc_thread;
635 /* Where in the btree gc currently is */
636 struct bkey gc_done;
637
638 /*
639 * For automatical garbage collection after writeback completed, this
640 * varialbe is used as bit fields,
641 * - 0000 0001b (BCH_ENABLE_AUTO_GC): enable gc after writeback
642 * - 0000 0010b (BCH_DO_AUTO_GC): do gc after writeback
643 * This is an optimization for following write request after writeback
644 * finished, but read hit rate dropped due to clean data on cache is
645 * discarded. Unless user explicitly sets it via sysfs, it won't be
646 * enabled.
647 */
648 #define BCH_ENABLE_AUTO_GC 1
649 #define BCH_DO_AUTO_GC 2
650 uint8_t gc_after_writeback;
651
652 /*
653 * The allocation code needs gc_mark in struct bucket to be correct, but
654 * it's not while a gc is in progress. Protected by bucket_lock.
655 */
656 int gc_mark_valid;
657
658 /* Counts how many sectors bio_insert has added to the cache */
659 atomic_t sectors_to_gc;
660 wait_queue_head_t gc_wait;
661
662 struct keybuf moving_gc_keys;
663 /* Number of moving GC bios in flight */
664 struct semaphore moving_in_flight;
665
666 struct workqueue_struct *moving_gc_wq;
667
668 struct btree *root;
669
670 #ifdef CONFIG_BCACHE_DEBUG
671 struct btree *verify_data;
672 struct bset *verify_ondisk;
673 struct mutex verify_lock;
674 #endif
675
676 uint8_t set_uuid[16];
677 unsigned int nr_uuids;
678 struct uuid_entry *uuids;
679 BKEY_PADDED(uuid_bucket);
680 struct closure uuid_write;
681 struct semaphore uuid_write_mutex;
682
683 /*
684 * A btree node on disk could have too many bsets for an iterator to fit
685 * on the stack - have to dynamically allocate them.
686 * bch_cache_set_alloc() will make sure the pool can allocate iterators
687 * equipped with enough room that can host
688 * (sb.bucket_size / sb.block_size)
689 * btree_iter_sets, which is more than static MAX_BSETS.
690 */
691 mempool_t fill_iter;
692
693 struct bset_sort_state sort;
694
695 /* List of buckets we're currently writing data to */
696 struct list_head data_buckets;
697 spinlock_t data_bucket_lock;
698
699 struct journal journal;
700
701 #define CONGESTED_MAX 1024
702 unsigned int congested_last_us;
703 atomic_t congested;
704
705 /* The rest of this all shows up in sysfs */
706 unsigned int congested_read_threshold_us;
707 unsigned int congested_write_threshold_us;
708
709 struct time_stats btree_gc_time;
710 struct time_stats btree_split_time;
711 struct time_stats btree_read_time;
712
713 atomic_long_t cache_read_races;
714 atomic_long_t writeback_keys_done;
715 atomic_long_t writeback_keys_failed;
716
717 atomic_long_t reclaim;
718 atomic_long_t reclaimed_journal_buckets;
719 atomic_long_t flush_write;
720
721 enum {
722 ON_ERROR_UNREGISTER,
723 ON_ERROR_PANIC,
724 } on_error;
725 #define DEFAULT_IO_ERROR_LIMIT 8
726 unsigned int error_limit;
727 unsigned int error_decay;
728
729 unsigned short journal_delay_ms;
730 bool expensive_debug_checks;
731 unsigned int verify:1;
732 unsigned int key_merging_disabled:1;
733 unsigned int gc_always_rewrite:1;
734 unsigned int shrinker_disabled:1;
735 unsigned int copy_gc_enabled:1;
736 unsigned int idle_max_writeback_rate_enabled:1;
737
738 #define BUCKET_HASH_BITS 12
739 struct hlist_head bucket_hash[1 << BUCKET_HASH_BITS];
740 };
741
742 struct bbio {
743 unsigned int submit_time_us;
744 union {
745 struct bkey key;
746 uint64_t _pad[3];
747 /*
748 * We only need pad = 3 here because we only ever carry around a
749 * single pointer - i.e. the pointer we're doing io to/from.
750 */
751 };
752 struct bio bio;
753 };
754
755 #define BTREE_PRIO USHRT_MAX
756 #define INITIAL_PRIO 32768U
757
758 #define btree_bytes(c) ((c)->btree_pages * PAGE_SIZE)
759 #define btree_blocks(b) \
760 ((unsigned int) (KEY_SIZE(&b->key) >> (b)->c->block_bits))
761
762 #define btree_default_blocks(c) \
763 ((unsigned int) ((PAGE_SECTORS * (c)->btree_pages) >> (c)->block_bits))
764
765 #define bucket_bytes(ca) ((ca)->sb.bucket_size << 9)
766 #define block_bytes(ca) ((ca)->sb.block_size << 9)
767
meta_bucket_pages(struct cache_sb * sb)768 static inline unsigned int meta_bucket_pages(struct cache_sb *sb)
769 {
770 unsigned int n, max_pages;
771
772 max_pages = min_t(unsigned int,
773 __rounddown_pow_of_two(USHRT_MAX) / PAGE_SECTORS,
774 MAX_ORDER_NR_PAGES);
775
776 n = sb->bucket_size / PAGE_SECTORS;
777 if (n > max_pages)
778 n = max_pages;
779
780 return n;
781 }
782
meta_bucket_bytes(struct cache_sb * sb)783 static inline unsigned int meta_bucket_bytes(struct cache_sb *sb)
784 {
785 return meta_bucket_pages(sb) << PAGE_SHIFT;
786 }
787
788 #define prios_per_bucket(ca) \
789 ((meta_bucket_bytes(&(ca)->sb) - sizeof(struct prio_set)) / \
790 sizeof(struct bucket_disk))
791
792 #define prio_buckets(ca) \
793 DIV_ROUND_UP((size_t) (ca)->sb.nbuckets, prios_per_bucket(ca))
794
sector_to_bucket(struct cache_set * c,sector_t s)795 static inline size_t sector_to_bucket(struct cache_set *c, sector_t s)
796 {
797 return s >> c->bucket_bits;
798 }
799
bucket_to_sector(struct cache_set * c,size_t b)800 static inline sector_t bucket_to_sector(struct cache_set *c, size_t b)
801 {
802 return ((sector_t) b) << c->bucket_bits;
803 }
804
bucket_remainder(struct cache_set * c,sector_t s)805 static inline sector_t bucket_remainder(struct cache_set *c, sector_t s)
806 {
807 return s & (c->cache->sb.bucket_size - 1);
808 }
809
PTR_BUCKET_NR(struct cache_set * c,const struct bkey * k,unsigned int ptr)810 static inline size_t PTR_BUCKET_NR(struct cache_set *c,
811 const struct bkey *k,
812 unsigned int ptr)
813 {
814 return sector_to_bucket(c, PTR_OFFSET(k, ptr));
815 }
816
PTR_BUCKET(struct cache_set * c,const struct bkey * k,unsigned int ptr)817 static inline struct bucket *PTR_BUCKET(struct cache_set *c,
818 const struct bkey *k,
819 unsigned int ptr)
820 {
821 return c->cache->buckets + PTR_BUCKET_NR(c, k, ptr);
822 }
823
gen_after(uint8_t a,uint8_t b)824 static inline uint8_t gen_after(uint8_t a, uint8_t b)
825 {
826 uint8_t r = a - b;
827
828 return r > 128U ? 0 : r;
829 }
830
ptr_stale(struct cache_set * c,const struct bkey * k,unsigned int i)831 static inline uint8_t ptr_stale(struct cache_set *c, const struct bkey *k,
832 unsigned int i)
833 {
834 return gen_after(PTR_BUCKET(c, k, i)->gen, PTR_GEN(k, i));
835 }
836
ptr_available(struct cache_set * c,const struct bkey * k,unsigned int i)837 static inline bool ptr_available(struct cache_set *c, const struct bkey *k,
838 unsigned int i)
839 {
840 return (PTR_DEV(k, i) < MAX_CACHES_PER_SET) && c->cache;
841 }
842
843 /* Btree key macros */
844
845 /*
846 * This is used for various on disk data structures - cache_sb, prio_set, bset,
847 * jset: The checksum is _always_ the first 8 bytes of these structs
848 */
849 #define csum_set(i) \
850 bch_crc64(((void *) (i)) + sizeof(uint64_t), \
851 ((void *) bset_bkey_last(i)) - \
852 (((void *) (i)) + sizeof(uint64_t)))
853
854 /* Error handling macros */
855
856 #define btree_bug(b, ...) \
857 do { \
858 if (bch_cache_set_error((b)->c, __VA_ARGS__)) \
859 dump_stack(); \
860 } while (0)
861
862 #define cache_bug(c, ...) \
863 do { \
864 if (bch_cache_set_error(c, __VA_ARGS__)) \
865 dump_stack(); \
866 } while (0)
867
868 #define btree_bug_on(cond, b, ...) \
869 do { \
870 if (cond) \
871 btree_bug(b, __VA_ARGS__); \
872 } while (0)
873
874 #define cache_bug_on(cond, c, ...) \
875 do { \
876 if (cond) \
877 cache_bug(c, __VA_ARGS__); \
878 } while (0)
879
880 #define cache_set_err_on(cond, c, ...) \
881 do { \
882 if (cond) \
883 bch_cache_set_error(c, __VA_ARGS__); \
884 } while (0)
885
886 /* Looping macros */
887
888 #define for_each_bucket(b, ca) \
889 for (b = (ca)->buckets + (ca)->sb.first_bucket; \
890 b < (ca)->buckets + (ca)->sb.nbuckets; b++)
891
cached_dev_put(struct cached_dev * dc)892 static inline void cached_dev_put(struct cached_dev *dc)
893 {
894 if (refcount_dec_and_test(&dc->count))
895 schedule_work(&dc->detach);
896 }
897
cached_dev_get(struct cached_dev * dc)898 static inline bool cached_dev_get(struct cached_dev *dc)
899 {
900 if (!refcount_inc_not_zero(&dc->count))
901 return false;
902
903 /* Paired with the mb in cached_dev_attach */
904 smp_mb__after_atomic();
905 return true;
906 }
907
908 /*
909 * bucket_gc_gen() returns the difference between the bucket's current gen and
910 * the oldest gen of any pointer into that bucket in the btree (last_gc).
911 */
912
bucket_gc_gen(struct bucket * b)913 static inline uint8_t bucket_gc_gen(struct bucket *b)
914 {
915 return b->gen - b->last_gc;
916 }
917
918 #define BUCKET_GC_GEN_MAX 96U
919
920 #define kobj_attribute_write(n, fn) \
921 static struct kobj_attribute ksysfs_##n = __ATTR(n, 0200, NULL, fn)
922
923 #define kobj_attribute_rw(n, show, store) \
924 static struct kobj_attribute ksysfs_##n = \
925 __ATTR(n, 0600, show, store)
926
wake_up_allocators(struct cache_set * c)927 static inline void wake_up_allocators(struct cache_set *c)
928 {
929 struct cache *ca = c->cache;
930
931 wake_up_process(ca->alloc_thread);
932 }
933
closure_bio_submit(struct cache_set * c,struct bio * bio,struct closure * cl)934 static inline void closure_bio_submit(struct cache_set *c,
935 struct bio *bio,
936 struct closure *cl)
937 {
938 closure_get(cl);
939 if (unlikely(test_bit(CACHE_SET_IO_DISABLE, &c->flags))) {
940 bio->bi_status = BLK_STS_IOERR;
941 bio_endio(bio);
942 return;
943 }
944 submit_bio_noacct(bio);
945 }
946
947 /*
948 * Prevent the kthread exits directly, and make sure when kthread_stop()
949 * is called to stop a kthread, it is still alive. If a kthread might be
950 * stopped by CACHE_SET_IO_DISABLE bit set, wait_for_kthread_stop() is
951 * necessary before the kthread returns.
952 */
wait_for_kthread_stop(void)953 static inline void wait_for_kthread_stop(void)
954 {
955 while (!kthread_should_stop()) {
956 set_current_state(TASK_INTERRUPTIBLE);
957 schedule();
958 }
959 }
960
961 /* Forward declarations */
962
963 void bch_count_backing_io_errors(struct cached_dev *dc, struct bio *bio);
964 void bch_count_io_errors(struct cache *ca, blk_status_t error,
965 int is_read, const char *m);
966 void bch_bbio_count_io_errors(struct cache_set *c, struct bio *bio,
967 blk_status_t error, const char *m);
968 void bch_bbio_endio(struct cache_set *c, struct bio *bio,
969 blk_status_t error, const char *m);
970 void bch_bbio_free(struct bio *bio, struct cache_set *c);
971 struct bio *bch_bbio_alloc(struct cache_set *c);
972
973 void __bch_submit_bbio(struct bio *bio, struct cache_set *c);
974 void bch_submit_bbio(struct bio *bio, struct cache_set *c,
975 struct bkey *k, unsigned int ptr);
976
977 uint8_t bch_inc_gen(struct cache *ca, struct bucket *b);
978 void bch_rescale_priorities(struct cache_set *c, int sectors);
979
980 bool bch_can_invalidate_bucket(struct cache *ca, struct bucket *b);
981 void __bch_invalidate_one_bucket(struct cache *ca, struct bucket *b);
982
983 void __bch_bucket_free(struct cache *ca, struct bucket *b);
984 void bch_bucket_free(struct cache_set *c, struct bkey *k);
985
986 long bch_bucket_alloc(struct cache *ca, unsigned int reserve, bool wait);
987 int __bch_bucket_alloc_set(struct cache_set *c, unsigned int reserve,
988 struct bkey *k, bool wait);
989 int bch_bucket_alloc_set(struct cache_set *c, unsigned int reserve,
990 struct bkey *k, bool wait);
991 bool bch_alloc_sectors(struct cache_set *c, struct bkey *k,
992 unsigned int sectors, unsigned int write_point,
993 unsigned int write_prio, bool wait);
994 bool bch_cached_dev_error(struct cached_dev *dc);
995
996 __printf(2, 3)
997 bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...);
998
999 int bch_prio_write(struct cache *ca, bool wait);
1000 void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent);
1001
1002 extern struct workqueue_struct *bcache_wq;
1003 extern struct workqueue_struct *bch_journal_wq;
1004 extern struct workqueue_struct *bch_flush_wq;
1005 extern struct mutex bch_register_lock;
1006 extern struct list_head bch_cache_sets;
1007
1008 extern struct kobj_type bch_cached_dev_ktype;
1009 extern struct kobj_type bch_flash_dev_ktype;
1010 extern struct kobj_type bch_cache_set_ktype;
1011 extern struct kobj_type bch_cache_set_internal_ktype;
1012 extern struct kobj_type bch_cache_ktype;
1013
1014 void bch_cached_dev_release(struct kobject *kobj);
1015 void bch_flash_dev_release(struct kobject *kobj);
1016 void bch_cache_set_release(struct kobject *kobj);
1017 void bch_cache_release(struct kobject *kobj);
1018
1019 int bch_uuid_write(struct cache_set *c);
1020 void bcache_write_super(struct cache_set *c);
1021
1022 int bch_flash_dev_create(struct cache_set *c, uint64_t size);
1023
1024 int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c,
1025 uint8_t *set_uuid);
1026 void bch_cached_dev_detach(struct cached_dev *dc);
1027 int bch_cached_dev_run(struct cached_dev *dc);
1028 void bcache_device_stop(struct bcache_device *d);
1029
1030 void bch_cache_set_unregister(struct cache_set *c);
1031 void bch_cache_set_stop(struct cache_set *c);
1032
1033 struct cache_set *bch_cache_set_alloc(struct cache_sb *sb);
1034 void bch_btree_cache_free(struct cache_set *c);
1035 int bch_btree_cache_alloc(struct cache_set *c);
1036 void bch_moving_init_cache_set(struct cache_set *c);
1037 int bch_open_buckets_alloc(struct cache_set *c);
1038 void bch_open_buckets_free(struct cache_set *c);
1039
1040 int bch_cache_allocator_start(struct cache *ca);
1041
1042 void bch_debug_exit(void);
1043 void bch_debug_init(void);
1044 void bch_request_exit(void);
1045 int bch_request_init(void);
1046 void bch_btree_exit(void);
1047 int bch_btree_init(void);
1048
1049 #endif /* _BCACHE_H */
1050