Loading...
Note: File does not exist in v3.1.
1/*
2 * bcache setup/teardown code, and some metadata io - read a superblock and
3 * figure out what to do with it.
4 *
5 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
6 * Copyright 2012 Google, Inc.
7 */
8
9#include "bcache.h"
10#include "btree.h"
11#include "debug.h"
12#include "extents.h"
13#include "request.h"
14#include "writeback.h"
15
16#include <linux/blkdev.h>
17#include <linux/buffer_head.h>
18#include <linux/debugfs.h>
19#include <linux/genhd.h>
20#include <linux/idr.h>
21#include <linux/kthread.h>
22#include <linux/module.h>
23#include <linux/random.h>
24#include <linux/reboot.h>
25#include <linux/sysfs.h>
26
27MODULE_LICENSE("GPL");
28MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
29
30static const char bcache_magic[] = {
31 0xc6, 0x85, 0x73, 0xf6, 0x4e, 0x1a, 0x45, 0xca,
32 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81
33};
34
35static const char invalid_uuid[] = {
36 0xa0, 0x3e, 0xf8, 0xed, 0x3e, 0xe1, 0xb8, 0x78,
37 0xc8, 0x50, 0xfc, 0x5e, 0xcb, 0x16, 0xcd, 0x99
38};
39
40/* Default is -1; we skip past it for struct cached_dev's cache mode */
41const char * const bch_cache_modes[] = {
42 "default",
43 "writethrough",
44 "writeback",
45 "writearound",
46 "none",
47 NULL
48};
49
50static struct kobject *bcache_kobj;
51struct mutex bch_register_lock;
52LIST_HEAD(bch_cache_sets);
53static LIST_HEAD(uncached_devices);
54
55static int bcache_major;
56static DEFINE_IDA(bcache_minor);
57static wait_queue_head_t unregister_wait;
58struct workqueue_struct *bcache_wq;
59
60#define BTREE_MAX_PAGES (256 * 1024 / PAGE_SIZE)
61
62static void bio_split_pool_free(struct bio_split_pool *p)
63{
64 if (p->bio_split_hook)
65 mempool_destroy(p->bio_split_hook);
66
67 if (p->bio_split)
68 bioset_free(p->bio_split);
69}
70
71static int bio_split_pool_init(struct bio_split_pool *p)
72{
73 p->bio_split = bioset_create(4, 0);
74 if (!p->bio_split)
75 return -ENOMEM;
76
77 p->bio_split_hook = mempool_create_kmalloc_pool(4,
78 sizeof(struct bio_split_hook));
79 if (!p->bio_split_hook)
80 return -ENOMEM;
81
82 return 0;
83}
84
85/* Superblock */
86
87static const char *read_super(struct cache_sb *sb, struct block_device *bdev,
88 struct page **res)
89{
90 const char *err;
91 struct cache_sb *s;
92 struct buffer_head *bh = __bread(bdev, 1, SB_SIZE);
93 unsigned i;
94
95 if (!bh)
96 return "IO error";
97
98 s = (struct cache_sb *) bh->b_data;
99
100 sb->offset = le64_to_cpu(s->offset);
101 sb->version = le64_to_cpu(s->version);
102
103 memcpy(sb->magic, s->magic, 16);
104 memcpy(sb->uuid, s->uuid, 16);
105 memcpy(sb->set_uuid, s->set_uuid, 16);
106 memcpy(sb->label, s->label, SB_LABEL_SIZE);
107
108 sb->flags = le64_to_cpu(s->flags);
109 sb->seq = le64_to_cpu(s->seq);
110 sb->last_mount = le32_to_cpu(s->last_mount);
111 sb->first_bucket = le16_to_cpu(s->first_bucket);
112 sb->keys = le16_to_cpu(s->keys);
113
114 for (i = 0; i < SB_JOURNAL_BUCKETS; i++)
115 sb->d[i] = le64_to_cpu(s->d[i]);
116
117 pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u",
118 sb->version, sb->flags, sb->seq, sb->keys);
119
120 err = "Not a bcache superblock";
121 if (sb->offset != SB_SECTOR)
122 goto err;
123
124 if (memcmp(sb->magic, bcache_magic, 16))
125 goto err;
126
127 err = "Too many journal buckets";
128 if (sb->keys > SB_JOURNAL_BUCKETS)
129 goto err;
130
131 err = "Bad checksum";
132 if (s->csum != csum_set(s))
133 goto err;
134
135 err = "Bad UUID";
136 if (bch_is_zero(sb->uuid, 16))
137 goto err;
138
139 sb->block_size = le16_to_cpu(s->block_size);
140
141 err = "Superblock block size smaller than device block size";
142 if (sb->block_size << 9 < bdev_logical_block_size(bdev))
143 goto err;
144
145 switch (sb->version) {
146 case BCACHE_SB_VERSION_BDEV:
147 sb->data_offset = BDEV_DATA_START_DEFAULT;
148 break;
149 case BCACHE_SB_VERSION_BDEV_WITH_OFFSET:
150 sb->data_offset = le64_to_cpu(s->data_offset);
151
152 err = "Bad data offset";
153 if (sb->data_offset < BDEV_DATA_START_DEFAULT)
154 goto err;
155
156 break;
157 case BCACHE_SB_VERSION_CDEV:
158 case BCACHE_SB_VERSION_CDEV_WITH_UUID:
159 sb->nbuckets = le64_to_cpu(s->nbuckets);
160 sb->block_size = le16_to_cpu(s->block_size);
161 sb->bucket_size = le16_to_cpu(s->bucket_size);
162
163 sb->nr_in_set = le16_to_cpu(s->nr_in_set);
164 sb->nr_this_dev = le16_to_cpu(s->nr_this_dev);
165
166 err = "Too many buckets";
167 if (sb->nbuckets > LONG_MAX)
168 goto err;
169
170 err = "Not enough buckets";
171 if (sb->nbuckets < 1 << 7)
172 goto err;
173
174 err = "Bad block/bucket size";
175 if (!is_power_of_2(sb->block_size) ||
176 sb->block_size > PAGE_SECTORS ||
177 !is_power_of_2(sb->bucket_size) ||
178 sb->bucket_size < PAGE_SECTORS)
179 goto err;
180
181 err = "Invalid superblock: device too small";
182 if (get_capacity(bdev->bd_disk) < sb->bucket_size * sb->nbuckets)
183 goto err;
184
185 err = "Bad UUID";
186 if (bch_is_zero(sb->set_uuid, 16))
187 goto err;
188
189 err = "Bad cache device number in set";
190 if (!sb->nr_in_set ||
191 sb->nr_in_set <= sb->nr_this_dev ||
192 sb->nr_in_set > MAX_CACHES_PER_SET)
193 goto err;
194
195 err = "Journal buckets not sequential";
196 for (i = 0; i < sb->keys; i++)
197 if (sb->d[i] != sb->first_bucket + i)
198 goto err;
199
200 err = "Too many journal buckets";
201 if (sb->first_bucket + sb->keys > sb->nbuckets)
202 goto err;
203
204 err = "Invalid superblock: first bucket comes before end of super";
205 if (sb->first_bucket * sb->bucket_size < 16)
206 goto err;
207
208 break;
209 default:
210 err = "Unsupported superblock version";
211 goto err;
212 }
213
214 sb->last_mount = get_seconds();
215 err = NULL;
216
217 get_page(bh->b_page);
218 *res = bh->b_page;
219err:
220 put_bh(bh);
221 return err;
222}
223
224static void write_bdev_super_endio(struct bio *bio, int error)
225{
226 struct cached_dev *dc = bio->bi_private;
227 /* XXX: error checking */
228
229 closure_put(&dc->sb_write);
230}
231
232static void __write_super(struct cache_sb *sb, struct bio *bio)
233{
234 struct cache_sb *out = page_address(bio->bi_io_vec[0].bv_page);
235 unsigned i;
236
237 bio->bi_iter.bi_sector = SB_SECTOR;
238 bio->bi_rw = REQ_SYNC|REQ_META;
239 bio->bi_iter.bi_size = SB_SIZE;
240 bch_bio_map(bio, NULL);
241
242 out->offset = cpu_to_le64(sb->offset);
243 out->version = cpu_to_le64(sb->version);
244
245 memcpy(out->uuid, sb->uuid, 16);
246 memcpy(out->set_uuid, sb->set_uuid, 16);
247 memcpy(out->label, sb->label, SB_LABEL_SIZE);
248
249 out->flags = cpu_to_le64(sb->flags);
250 out->seq = cpu_to_le64(sb->seq);
251
252 out->last_mount = cpu_to_le32(sb->last_mount);
253 out->first_bucket = cpu_to_le16(sb->first_bucket);
254 out->keys = cpu_to_le16(sb->keys);
255
256 for (i = 0; i < sb->keys; i++)
257 out->d[i] = cpu_to_le64(sb->d[i]);
258
259 out->csum = csum_set(out);
260
261 pr_debug("ver %llu, flags %llu, seq %llu",
262 sb->version, sb->flags, sb->seq);
263
264 submit_bio(REQ_WRITE, bio);
265}
266
267static void bch_write_bdev_super_unlock(struct closure *cl)
268{
269 struct cached_dev *dc = container_of(cl, struct cached_dev, sb_write);
270
271 up(&dc->sb_write_mutex);
272}
273
274void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent)
275{
276 struct closure *cl = &dc->sb_write;
277 struct bio *bio = &dc->sb_bio;
278
279 down(&dc->sb_write_mutex);
280 closure_init(cl, parent);
281
282 bio_reset(bio);
283 bio->bi_bdev = dc->bdev;
284 bio->bi_end_io = write_bdev_super_endio;
285 bio->bi_private = dc;
286
287 closure_get(cl);
288 __write_super(&dc->sb, bio);
289
290 closure_return_with_destructor(cl, bch_write_bdev_super_unlock);
291}
292
293static void write_super_endio(struct bio *bio, int error)
294{
295 struct cache *ca = bio->bi_private;
296
297 bch_count_io_errors(ca, error, "writing superblock");
298 closure_put(&ca->set->sb_write);
299}
300
301static void bcache_write_super_unlock(struct closure *cl)
302{
303 struct cache_set *c = container_of(cl, struct cache_set, sb_write);
304
305 up(&c->sb_write_mutex);
306}
307
308void bcache_write_super(struct cache_set *c)
309{
310 struct closure *cl = &c->sb_write;
311 struct cache *ca;
312 unsigned i;
313
314 down(&c->sb_write_mutex);
315 closure_init(cl, &c->cl);
316
317 c->sb.seq++;
318
319 for_each_cache(ca, c, i) {
320 struct bio *bio = &ca->sb_bio;
321
322 ca->sb.version = BCACHE_SB_VERSION_CDEV_WITH_UUID;
323 ca->sb.seq = c->sb.seq;
324 ca->sb.last_mount = c->sb.last_mount;
325
326 SET_CACHE_SYNC(&ca->sb, CACHE_SYNC(&c->sb));
327
328 bio_reset(bio);
329 bio->bi_bdev = ca->bdev;
330 bio->bi_end_io = write_super_endio;
331 bio->bi_private = ca;
332
333 closure_get(cl);
334 __write_super(&ca->sb, bio);
335 }
336
337 closure_return_with_destructor(cl, bcache_write_super_unlock);
338}
339
340/* UUID io */
341
342static void uuid_endio(struct bio *bio, int error)
343{
344 struct closure *cl = bio->bi_private;
345 struct cache_set *c = container_of(cl, struct cache_set, uuid_write);
346
347 cache_set_err_on(error, c, "accessing uuids");
348 bch_bbio_free(bio, c);
349 closure_put(cl);
350}
351
352static void uuid_io_unlock(struct closure *cl)
353{
354 struct cache_set *c = container_of(cl, struct cache_set, uuid_write);
355
356 up(&c->uuid_write_mutex);
357}
358
359static void uuid_io(struct cache_set *c, unsigned long rw,
360 struct bkey *k, struct closure *parent)
361{
362 struct closure *cl = &c->uuid_write;
363 struct uuid_entry *u;
364 unsigned i;
365 char buf[80];
366
367 BUG_ON(!parent);
368 down(&c->uuid_write_mutex);
369 closure_init(cl, parent);
370
371 for (i = 0; i < KEY_PTRS(k); i++) {
372 struct bio *bio = bch_bbio_alloc(c);
373
374 bio->bi_rw = REQ_SYNC|REQ_META|rw;
375 bio->bi_iter.bi_size = KEY_SIZE(k) << 9;
376
377 bio->bi_end_io = uuid_endio;
378 bio->bi_private = cl;
379 bch_bio_map(bio, c->uuids);
380
381 bch_submit_bbio(bio, c, k, i);
382
383 if (!(rw & WRITE))
384 break;
385 }
386
387 bch_extent_to_text(buf, sizeof(buf), k);
388 pr_debug("%s UUIDs at %s", rw & REQ_WRITE ? "wrote" : "read", buf);
389
390 for (u = c->uuids; u < c->uuids + c->nr_uuids; u++)
391 if (!bch_is_zero(u->uuid, 16))
392 pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u",
393 u - c->uuids, u->uuid, u->label,
394 u->first_reg, u->last_reg, u->invalidated);
395
396 closure_return_with_destructor(cl, uuid_io_unlock);
397}
398
399static char *uuid_read(struct cache_set *c, struct jset *j, struct closure *cl)
400{
401 struct bkey *k = &j->uuid_bucket;
402
403 if (__bch_btree_ptr_invalid(c, k))
404 return "bad uuid pointer";
405
406 bkey_copy(&c->uuid_bucket, k);
407 uuid_io(c, READ_SYNC, k, cl);
408
409 if (j->version < BCACHE_JSET_VERSION_UUIDv1) {
410 struct uuid_entry_v0 *u0 = (void *) c->uuids;
411 struct uuid_entry *u1 = (void *) c->uuids;
412 int i;
413
414 closure_sync(cl);
415
416 /*
417 * Since the new uuid entry is bigger than the old, we have to
418 * convert starting at the highest memory address and work down
419 * in order to do it in place
420 */
421
422 for (i = c->nr_uuids - 1;
423 i >= 0;
424 --i) {
425 memcpy(u1[i].uuid, u0[i].uuid, 16);
426 memcpy(u1[i].label, u0[i].label, 32);
427
428 u1[i].first_reg = u0[i].first_reg;
429 u1[i].last_reg = u0[i].last_reg;
430 u1[i].invalidated = u0[i].invalidated;
431
432 u1[i].flags = 0;
433 u1[i].sectors = 0;
434 }
435 }
436
437 return NULL;
438}
439
440static int __uuid_write(struct cache_set *c)
441{
442 BKEY_PADDED(key) k;
443 struct closure cl;
444 closure_init_stack(&cl);
445
446 lockdep_assert_held(&bch_register_lock);
447
448 if (bch_bucket_alloc_set(c, RESERVE_BTREE, &k.key, 1, true))
449 return 1;
450
451 SET_KEY_SIZE(&k.key, c->sb.bucket_size);
452 uuid_io(c, REQ_WRITE, &k.key, &cl);
453 closure_sync(&cl);
454
455 bkey_copy(&c->uuid_bucket, &k.key);
456 bkey_put(c, &k.key);
457 return 0;
458}
459
460int bch_uuid_write(struct cache_set *c)
461{
462 int ret = __uuid_write(c);
463
464 if (!ret)
465 bch_journal_meta(c, NULL);
466
467 return ret;
468}
469
470static struct uuid_entry *uuid_find(struct cache_set *c, const char *uuid)
471{
472 struct uuid_entry *u;
473
474 for (u = c->uuids;
475 u < c->uuids + c->nr_uuids; u++)
476 if (!memcmp(u->uuid, uuid, 16))
477 return u;
478
479 return NULL;
480}
481
482static struct uuid_entry *uuid_find_empty(struct cache_set *c)
483{
484 static const char zero_uuid[16] = "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0";
485 return uuid_find(c, zero_uuid);
486}
487
488/*
489 * Bucket priorities/gens:
490 *
491 * For each bucket, we store on disk its
492 * 8 bit gen
493 * 16 bit priority
494 *
495 * See alloc.c for an explanation of the gen. The priority is used to implement
496 * lru (and in the future other) cache replacement policies; for most purposes
497 * it's just an opaque integer.
498 *
499 * The gens and the priorities don't have a whole lot to do with each other, and
500 * it's actually the gens that must be written out at specific times - it's no
501 * big deal if the priorities don't get written, if we lose them we just reuse
502 * buckets in suboptimal order.
503 *
504 * On disk they're stored in a packed array, and in as many buckets are required
505 * to fit them all. The buckets we use to store them form a list; the journal
506 * header points to the first bucket, the first bucket points to the second
507 * bucket, et cetera.
508 *
509 * This code is used by the allocation code; periodically (whenever it runs out
510 * of buckets to allocate from) the allocation code will invalidate some
511 * buckets, but it can't use those buckets until their new gens are safely on
512 * disk.
513 */
514
515static void prio_endio(struct bio *bio, int error)
516{
517 struct cache *ca = bio->bi_private;
518
519 cache_set_err_on(error, ca->set, "accessing priorities");
520 bch_bbio_free(bio, ca->set);
521 closure_put(&ca->prio);
522}
523
524static void prio_io(struct cache *ca, uint64_t bucket, unsigned long rw)
525{
526 struct closure *cl = &ca->prio;
527 struct bio *bio = bch_bbio_alloc(ca->set);
528
529 closure_init_stack(cl);
530
531 bio->bi_iter.bi_sector = bucket * ca->sb.bucket_size;
532 bio->bi_bdev = ca->bdev;
533 bio->bi_rw = REQ_SYNC|REQ_META|rw;
534 bio->bi_iter.bi_size = bucket_bytes(ca);
535
536 bio->bi_end_io = prio_endio;
537 bio->bi_private = ca;
538 bch_bio_map(bio, ca->disk_buckets);
539
540 closure_bio_submit(bio, &ca->prio, ca);
541 closure_sync(cl);
542}
543
544void bch_prio_write(struct cache *ca)
545{
546 int i;
547 struct bucket *b;
548 struct closure cl;
549
550 closure_init_stack(&cl);
551
552 lockdep_assert_held(&ca->set->bucket_lock);
553
554 ca->disk_buckets->seq++;
555
556 atomic_long_add(ca->sb.bucket_size * prio_buckets(ca),
557 &ca->meta_sectors_written);
558
559 //pr_debug("free %zu, free_inc %zu, unused %zu", fifo_used(&ca->free),
560 // fifo_used(&ca->free_inc), fifo_used(&ca->unused));
561
562 for (i = prio_buckets(ca) - 1; i >= 0; --i) {
563 long bucket;
564 struct prio_set *p = ca->disk_buckets;
565 struct bucket_disk *d = p->data;
566 struct bucket_disk *end = d + prios_per_bucket(ca);
567
568 for (b = ca->buckets + i * prios_per_bucket(ca);
569 b < ca->buckets + ca->sb.nbuckets && d < end;
570 b++, d++) {
571 d->prio = cpu_to_le16(b->prio);
572 d->gen = b->gen;
573 }
574
575 p->next_bucket = ca->prio_buckets[i + 1];
576 p->magic = pset_magic(&ca->sb);
577 p->csum = bch_crc64(&p->magic, bucket_bytes(ca) - 8);
578
579 bucket = bch_bucket_alloc(ca, RESERVE_PRIO, true);
580 BUG_ON(bucket == -1);
581
582 mutex_unlock(&ca->set->bucket_lock);
583 prio_io(ca, bucket, REQ_WRITE);
584 mutex_lock(&ca->set->bucket_lock);
585
586 ca->prio_buckets[i] = bucket;
587 atomic_dec_bug(&ca->buckets[bucket].pin);
588 }
589
590 mutex_unlock(&ca->set->bucket_lock);
591
592 bch_journal_meta(ca->set, &cl);
593 closure_sync(&cl);
594
595 mutex_lock(&ca->set->bucket_lock);
596
597 /*
598 * Don't want the old priorities to get garbage collected until after we
599 * finish writing the new ones, and they're journalled
600 */
601 for (i = 0; i < prio_buckets(ca); i++) {
602 if (ca->prio_last_buckets[i])
603 __bch_bucket_free(ca,
604 &ca->buckets[ca->prio_last_buckets[i]]);
605
606 ca->prio_last_buckets[i] = ca->prio_buckets[i];
607 }
608}
609
610static void prio_read(struct cache *ca, uint64_t bucket)
611{
612 struct prio_set *p = ca->disk_buckets;
613 struct bucket_disk *d = p->data + prios_per_bucket(ca), *end = d;
614 struct bucket *b;
615 unsigned bucket_nr = 0;
616
617 for (b = ca->buckets;
618 b < ca->buckets + ca->sb.nbuckets;
619 b++, d++) {
620 if (d == end) {
621 ca->prio_buckets[bucket_nr] = bucket;
622 ca->prio_last_buckets[bucket_nr] = bucket;
623 bucket_nr++;
624
625 prio_io(ca, bucket, READ_SYNC);
626
627 if (p->csum != bch_crc64(&p->magic, bucket_bytes(ca) - 8))
628 pr_warn("bad csum reading priorities");
629
630 if (p->magic != pset_magic(&ca->sb))
631 pr_warn("bad magic reading priorities");
632
633 bucket = p->next_bucket;
634 d = p->data;
635 }
636
637 b->prio = le16_to_cpu(d->prio);
638 b->gen = b->last_gc = d->gen;
639 }
640}
641
642/* Bcache device */
643
644static int open_dev(struct block_device *b, fmode_t mode)
645{
646 struct bcache_device *d = b->bd_disk->private_data;
647 if (test_bit(BCACHE_DEV_CLOSING, &d->flags))
648 return -ENXIO;
649
650 closure_get(&d->cl);
651 return 0;
652}
653
654static void release_dev(struct gendisk *b, fmode_t mode)
655{
656 struct bcache_device *d = b->private_data;
657 closure_put(&d->cl);
658}
659
660static int ioctl_dev(struct block_device *b, fmode_t mode,
661 unsigned int cmd, unsigned long arg)
662{
663 struct bcache_device *d = b->bd_disk->private_data;
664 return d->ioctl(d, mode, cmd, arg);
665}
666
667static const struct block_device_operations bcache_ops = {
668 .open = open_dev,
669 .release = release_dev,
670 .ioctl = ioctl_dev,
671 .owner = THIS_MODULE,
672};
673
674void bcache_device_stop(struct bcache_device *d)
675{
676 if (!test_and_set_bit(BCACHE_DEV_CLOSING, &d->flags))
677 closure_queue(&d->cl);
678}
679
680static void bcache_device_unlink(struct bcache_device *d)
681{
682 lockdep_assert_held(&bch_register_lock);
683
684 if (d->c && !test_and_set_bit(BCACHE_DEV_UNLINK_DONE, &d->flags)) {
685 unsigned i;
686 struct cache *ca;
687
688 sysfs_remove_link(&d->c->kobj, d->name);
689 sysfs_remove_link(&d->kobj, "cache");
690
691 for_each_cache(ca, d->c, i)
692 bd_unlink_disk_holder(ca->bdev, d->disk);
693 }
694}
695
696static void bcache_device_link(struct bcache_device *d, struct cache_set *c,
697 const char *name)
698{
699 unsigned i;
700 struct cache *ca;
701
702 for_each_cache(ca, d->c, i)
703 bd_link_disk_holder(ca->bdev, d->disk);
704
705 snprintf(d->name, BCACHEDEVNAME_SIZE,
706 "%s%u", name, d->id);
707
708 WARN(sysfs_create_link(&d->kobj, &c->kobj, "cache") ||
709 sysfs_create_link(&c->kobj, &d->kobj, d->name),
710 "Couldn't create device <-> cache set symlinks");
711}
712
713static void bcache_device_detach(struct bcache_device *d)
714{
715 lockdep_assert_held(&bch_register_lock);
716
717 if (test_bit(BCACHE_DEV_DETACHING, &d->flags)) {
718 struct uuid_entry *u = d->c->uuids + d->id;
719
720 SET_UUID_FLASH_ONLY(u, 0);
721 memcpy(u->uuid, invalid_uuid, 16);
722 u->invalidated = cpu_to_le32(get_seconds());
723 bch_uuid_write(d->c);
724 }
725
726 bcache_device_unlink(d);
727
728 d->c->devices[d->id] = NULL;
729 closure_put(&d->c->caching);
730 d->c = NULL;
731}
732
733static void bcache_device_attach(struct bcache_device *d, struct cache_set *c,
734 unsigned id)
735{
736 BUG_ON(test_bit(CACHE_SET_STOPPING, &c->flags));
737
738 d->id = id;
739 d->c = c;
740 c->devices[id] = d;
741
742 closure_get(&c->caching);
743}
744
745static void bcache_device_free(struct bcache_device *d)
746{
747 lockdep_assert_held(&bch_register_lock);
748
749 pr_info("%s stopped", d->disk->disk_name);
750
751 if (d->c)
752 bcache_device_detach(d);
753 if (d->disk && d->disk->flags & GENHD_FL_UP)
754 del_gendisk(d->disk);
755 if (d->disk && d->disk->queue)
756 blk_cleanup_queue(d->disk->queue);
757 if (d->disk) {
758 ida_simple_remove(&bcache_minor, d->disk->first_minor);
759 put_disk(d->disk);
760 }
761
762 bio_split_pool_free(&d->bio_split_hook);
763 if (d->bio_split)
764 bioset_free(d->bio_split);
765 if (is_vmalloc_addr(d->full_dirty_stripes))
766 vfree(d->full_dirty_stripes);
767 else
768 kfree(d->full_dirty_stripes);
769 if (is_vmalloc_addr(d->stripe_sectors_dirty))
770 vfree(d->stripe_sectors_dirty);
771 else
772 kfree(d->stripe_sectors_dirty);
773
774 closure_debug_destroy(&d->cl);
775}
776
777static int bcache_device_init(struct bcache_device *d, unsigned block_size,
778 sector_t sectors)
779{
780 struct request_queue *q;
781 size_t n;
782 int minor;
783
784 if (!d->stripe_size)
785 d->stripe_size = 1 << 31;
786
787 d->nr_stripes = DIV_ROUND_UP_ULL(sectors, d->stripe_size);
788
789 if (!d->nr_stripes ||
790 d->nr_stripes > INT_MAX ||
791 d->nr_stripes > SIZE_MAX / sizeof(atomic_t)) {
792 pr_err("nr_stripes too large");
793 return -ENOMEM;
794 }
795
796 n = d->nr_stripes * sizeof(atomic_t);
797 d->stripe_sectors_dirty = n < PAGE_SIZE << 6
798 ? kzalloc(n, GFP_KERNEL)
799 : vzalloc(n);
800 if (!d->stripe_sectors_dirty)
801 return -ENOMEM;
802
803 n = BITS_TO_LONGS(d->nr_stripes) * sizeof(unsigned long);
804 d->full_dirty_stripes = n < PAGE_SIZE << 6
805 ? kzalloc(n, GFP_KERNEL)
806 : vzalloc(n);
807 if (!d->full_dirty_stripes)
808 return -ENOMEM;
809
810 minor = ida_simple_get(&bcache_minor, 0, MINORMASK + 1, GFP_KERNEL);
811 if (minor < 0)
812 return minor;
813
814 if (!(d->bio_split = bioset_create(4, offsetof(struct bbio, bio))) ||
815 bio_split_pool_init(&d->bio_split_hook) ||
816 !(d->disk = alloc_disk(1))) {
817 ida_simple_remove(&bcache_minor, minor);
818 return -ENOMEM;
819 }
820
821 set_capacity(d->disk, sectors);
822 snprintf(d->disk->disk_name, DISK_NAME_LEN, "bcache%i", minor);
823
824 d->disk->major = bcache_major;
825 d->disk->first_minor = minor;
826 d->disk->fops = &bcache_ops;
827 d->disk->private_data = d;
828
829 q = blk_alloc_queue(GFP_KERNEL);
830 if (!q)
831 return -ENOMEM;
832
833 blk_queue_make_request(q, NULL);
834 d->disk->queue = q;
835 q->queuedata = d;
836 q->backing_dev_info.congested_data = d;
837 q->limits.max_hw_sectors = UINT_MAX;
838 q->limits.max_sectors = UINT_MAX;
839 q->limits.max_segment_size = UINT_MAX;
840 q->limits.max_segments = BIO_MAX_PAGES;
841 q->limits.max_discard_sectors = UINT_MAX;
842 q->limits.discard_granularity = 512;
843 q->limits.io_min = block_size;
844 q->limits.logical_block_size = block_size;
845 q->limits.physical_block_size = block_size;
846 set_bit(QUEUE_FLAG_NONROT, &d->disk->queue->queue_flags);
847 set_bit(QUEUE_FLAG_DISCARD, &d->disk->queue->queue_flags);
848
849 blk_queue_flush(q, REQ_FLUSH|REQ_FUA);
850
851 return 0;
852}
853
854/* Cached device */
855
856static void calc_cached_dev_sectors(struct cache_set *c)
857{
858 uint64_t sectors = 0;
859 struct cached_dev *dc;
860
861 list_for_each_entry(dc, &c->cached_devs, list)
862 sectors += bdev_sectors(dc->bdev);
863
864 c->cached_dev_sectors = sectors;
865}
866
867void bch_cached_dev_run(struct cached_dev *dc)
868{
869 struct bcache_device *d = &dc->disk;
870 char buf[SB_LABEL_SIZE + 1];
871 char *env[] = {
872 "DRIVER=bcache",
873 kasprintf(GFP_KERNEL, "CACHED_UUID=%pU", dc->sb.uuid),
874 NULL,
875 NULL,
876 };
877
878 memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
879 buf[SB_LABEL_SIZE] = '\0';
880 env[2] = kasprintf(GFP_KERNEL, "CACHED_LABEL=%s", buf);
881
882 if (atomic_xchg(&dc->running, 1))
883 return;
884
885 if (!d->c &&
886 BDEV_STATE(&dc->sb) != BDEV_STATE_NONE) {
887 struct closure cl;
888 closure_init_stack(&cl);
889
890 SET_BDEV_STATE(&dc->sb, BDEV_STATE_STALE);
891 bch_write_bdev_super(dc, &cl);
892 closure_sync(&cl);
893 }
894
895 add_disk(d->disk);
896 bd_link_disk_holder(dc->bdev, dc->disk.disk);
897 /* won't show up in the uevent file, use udevadm monitor -e instead
898 * only class / kset properties are persistent */
899 kobject_uevent_env(&disk_to_dev(d->disk)->kobj, KOBJ_CHANGE, env);
900 kfree(env[1]);
901 kfree(env[2]);
902
903 if (sysfs_create_link(&d->kobj, &disk_to_dev(d->disk)->kobj, "dev") ||
904 sysfs_create_link(&disk_to_dev(d->disk)->kobj, &d->kobj, "bcache"))
905 pr_debug("error creating sysfs link");
906}
907
908static void cached_dev_detach_finish(struct work_struct *w)
909{
910 struct cached_dev *dc = container_of(w, struct cached_dev, detach);
911 char buf[BDEVNAME_SIZE];
912 struct closure cl;
913 closure_init_stack(&cl);
914
915 BUG_ON(!test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags));
916 BUG_ON(atomic_read(&dc->count));
917
918 mutex_lock(&bch_register_lock);
919
920 memset(&dc->sb.set_uuid, 0, 16);
921 SET_BDEV_STATE(&dc->sb, BDEV_STATE_NONE);
922
923 bch_write_bdev_super(dc, &cl);
924 closure_sync(&cl);
925
926 bcache_device_detach(&dc->disk);
927 list_move(&dc->list, &uncached_devices);
928
929 clear_bit(BCACHE_DEV_DETACHING, &dc->disk.flags);
930
931 mutex_unlock(&bch_register_lock);
932
933 pr_info("Caching disabled for %s", bdevname(dc->bdev, buf));
934
935 /* Drop ref we took in cached_dev_detach() */
936 closure_put(&dc->disk.cl);
937}
938
939void bch_cached_dev_detach(struct cached_dev *dc)
940{
941 lockdep_assert_held(&bch_register_lock);
942
943 if (test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags))
944 return;
945
946 if (test_and_set_bit(BCACHE_DEV_DETACHING, &dc->disk.flags))
947 return;
948
949 /*
950 * Block the device from being closed and freed until we're finished
951 * detaching
952 */
953 closure_get(&dc->disk.cl);
954
955 bch_writeback_queue(dc);
956 cached_dev_put(dc);
957}
958
959int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c)
960{
961 uint32_t rtime = cpu_to_le32(get_seconds());
962 struct uuid_entry *u;
963 char buf[BDEVNAME_SIZE];
964
965 bdevname(dc->bdev, buf);
966
967 if (memcmp(dc->sb.set_uuid, c->sb.set_uuid, 16))
968 return -ENOENT;
969
970 if (dc->disk.c) {
971 pr_err("Can't attach %s: already attached", buf);
972 return -EINVAL;
973 }
974
975 if (test_bit(CACHE_SET_STOPPING, &c->flags)) {
976 pr_err("Can't attach %s: shutting down", buf);
977 return -EINVAL;
978 }
979
980 if (dc->sb.block_size < c->sb.block_size) {
981 /* Will die */
982 pr_err("Couldn't attach %s: block size less than set's block size",
983 buf);
984 return -EINVAL;
985 }
986
987 u = uuid_find(c, dc->sb.uuid);
988
989 if (u &&
990 (BDEV_STATE(&dc->sb) == BDEV_STATE_STALE ||
991 BDEV_STATE(&dc->sb) == BDEV_STATE_NONE)) {
992 memcpy(u->uuid, invalid_uuid, 16);
993 u->invalidated = cpu_to_le32(get_seconds());
994 u = NULL;
995 }
996
997 if (!u) {
998 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
999 pr_err("Couldn't find uuid for %s in set", buf);
1000 return -ENOENT;
1001 }
1002
1003 u = uuid_find_empty(c);
1004 if (!u) {
1005 pr_err("Not caching %s, no room for UUID", buf);
1006 return -EINVAL;
1007 }
1008 }
1009
1010 /* Deadlocks since we're called via sysfs...
1011 sysfs_remove_file(&dc->kobj, &sysfs_attach);
1012 */
1013
1014 if (bch_is_zero(u->uuid, 16)) {
1015 struct closure cl;
1016 closure_init_stack(&cl);
1017
1018 memcpy(u->uuid, dc->sb.uuid, 16);
1019 memcpy(u->label, dc->sb.label, SB_LABEL_SIZE);
1020 u->first_reg = u->last_reg = rtime;
1021 bch_uuid_write(c);
1022
1023 memcpy(dc->sb.set_uuid, c->sb.set_uuid, 16);
1024 SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN);
1025
1026 bch_write_bdev_super(dc, &cl);
1027 closure_sync(&cl);
1028 } else {
1029 u->last_reg = rtime;
1030 bch_uuid_write(c);
1031 }
1032
1033 bcache_device_attach(&dc->disk, c, u - c->uuids);
1034 list_move(&dc->list, &c->cached_devs);
1035 calc_cached_dev_sectors(c);
1036
1037 smp_wmb();
1038 /*
1039 * dc->c must be set before dc->count != 0 - paired with the mb in
1040 * cached_dev_get()
1041 */
1042 atomic_set(&dc->count, 1);
1043
1044 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
1045 bch_sectors_dirty_init(dc);
1046 atomic_set(&dc->has_dirty, 1);
1047 atomic_inc(&dc->count);
1048 bch_writeback_queue(dc);
1049 }
1050
1051 bch_cached_dev_run(dc);
1052 bcache_device_link(&dc->disk, c, "bdev");
1053
1054 pr_info("Caching %s as %s on set %pU",
1055 bdevname(dc->bdev, buf), dc->disk.disk->disk_name,
1056 dc->disk.c->sb.set_uuid);
1057 return 0;
1058}
1059
1060void bch_cached_dev_release(struct kobject *kobj)
1061{
1062 struct cached_dev *dc = container_of(kobj, struct cached_dev,
1063 disk.kobj);
1064 kfree(dc);
1065 module_put(THIS_MODULE);
1066}
1067
1068static void cached_dev_free(struct closure *cl)
1069{
1070 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1071
1072 cancel_delayed_work_sync(&dc->writeback_rate_update);
1073 kthread_stop(dc->writeback_thread);
1074
1075 mutex_lock(&bch_register_lock);
1076
1077 if (atomic_read(&dc->running))
1078 bd_unlink_disk_holder(dc->bdev, dc->disk.disk);
1079 bcache_device_free(&dc->disk);
1080 list_del(&dc->list);
1081
1082 mutex_unlock(&bch_register_lock);
1083
1084 if (!IS_ERR_OR_NULL(dc->bdev)) {
1085 if (dc->bdev->bd_disk)
1086 blk_sync_queue(bdev_get_queue(dc->bdev));
1087
1088 blkdev_put(dc->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1089 }
1090
1091 wake_up(&unregister_wait);
1092
1093 kobject_put(&dc->disk.kobj);
1094}
1095
1096static void cached_dev_flush(struct closure *cl)
1097{
1098 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1099 struct bcache_device *d = &dc->disk;
1100
1101 mutex_lock(&bch_register_lock);
1102 bcache_device_unlink(d);
1103 mutex_unlock(&bch_register_lock);
1104
1105 bch_cache_accounting_destroy(&dc->accounting);
1106 kobject_del(&d->kobj);
1107
1108 continue_at(cl, cached_dev_free, system_wq);
1109}
1110
1111static int cached_dev_init(struct cached_dev *dc, unsigned block_size)
1112{
1113 int ret;
1114 struct io *io;
1115 struct request_queue *q = bdev_get_queue(dc->bdev);
1116
1117 __module_get(THIS_MODULE);
1118 INIT_LIST_HEAD(&dc->list);
1119 closure_init(&dc->disk.cl, NULL);
1120 set_closure_fn(&dc->disk.cl, cached_dev_flush, system_wq);
1121 kobject_init(&dc->disk.kobj, &bch_cached_dev_ktype);
1122 INIT_WORK(&dc->detach, cached_dev_detach_finish);
1123 sema_init(&dc->sb_write_mutex, 1);
1124 INIT_LIST_HEAD(&dc->io_lru);
1125 spin_lock_init(&dc->io_lock);
1126 bch_cache_accounting_init(&dc->accounting, &dc->disk.cl);
1127
1128 dc->sequential_cutoff = 4 << 20;
1129
1130 for (io = dc->io; io < dc->io + RECENT_IO; io++) {
1131 list_add(&io->lru, &dc->io_lru);
1132 hlist_add_head(&io->hash, dc->io_hash + RECENT_IO);
1133 }
1134
1135 dc->disk.stripe_size = q->limits.io_opt >> 9;
1136
1137 if (dc->disk.stripe_size)
1138 dc->partial_stripes_expensive =
1139 q->limits.raid_partial_stripes_expensive;
1140
1141 ret = bcache_device_init(&dc->disk, block_size,
1142 dc->bdev->bd_part->nr_sects - dc->sb.data_offset);
1143 if (ret)
1144 return ret;
1145
1146 set_capacity(dc->disk.disk,
1147 dc->bdev->bd_part->nr_sects - dc->sb.data_offset);
1148
1149 dc->disk.disk->queue->backing_dev_info.ra_pages =
1150 max(dc->disk.disk->queue->backing_dev_info.ra_pages,
1151 q->backing_dev_info.ra_pages);
1152
1153 bch_cached_dev_request_init(dc);
1154 bch_cached_dev_writeback_init(dc);
1155 return 0;
1156}
1157
1158/* Cached device - bcache superblock */
1159
1160static void register_bdev(struct cache_sb *sb, struct page *sb_page,
1161 struct block_device *bdev,
1162 struct cached_dev *dc)
1163{
1164 char name[BDEVNAME_SIZE];
1165 const char *err = "cannot allocate memory";
1166 struct cache_set *c;
1167
1168 memcpy(&dc->sb, sb, sizeof(struct cache_sb));
1169 dc->bdev = bdev;
1170 dc->bdev->bd_holder = dc;
1171
1172 bio_init(&dc->sb_bio);
1173 dc->sb_bio.bi_max_vecs = 1;
1174 dc->sb_bio.bi_io_vec = dc->sb_bio.bi_inline_vecs;
1175 dc->sb_bio.bi_io_vec[0].bv_page = sb_page;
1176 get_page(sb_page);
1177
1178 if (cached_dev_init(dc, sb->block_size << 9))
1179 goto err;
1180
1181 err = "error creating kobject";
1182 if (kobject_add(&dc->disk.kobj, &part_to_dev(bdev->bd_part)->kobj,
1183 "bcache"))
1184 goto err;
1185 if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj))
1186 goto err;
1187
1188 pr_info("registered backing device %s", bdevname(bdev, name));
1189
1190 list_add(&dc->list, &uncached_devices);
1191 list_for_each_entry(c, &bch_cache_sets, list)
1192 bch_cached_dev_attach(dc, c);
1193
1194 if (BDEV_STATE(&dc->sb) == BDEV_STATE_NONE ||
1195 BDEV_STATE(&dc->sb) == BDEV_STATE_STALE)
1196 bch_cached_dev_run(dc);
1197
1198 return;
1199err:
1200 pr_notice("error opening %s: %s", bdevname(bdev, name), err);
1201 bcache_device_stop(&dc->disk);
1202}
1203
1204/* Flash only volumes */
1205
1206void bch_flash_dev_release(struct kobject *kobj)
1207{
1208 struct bcache_device *d = container_of(kobj, struct bcache_device,
1209 kobj);
1210 kfree(d);
1211}
1212
1213static void flash_dev_free(struct closure *cl)
1214{
1215 struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1216 bcache_device_free(d);
1217 kobject_put(&d->kobj);
1218}
1219
1220static void flash_dev_flush(struct closure *cl)
1221{
1222 struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1223
1224 bcache_device_unlink(d);
1225 kobject_del(&d->kobj);
1226 continue_at(cl, flash_dev_free, system_wq);
1227}
1228
1229static int flash_dev_run(struct cache_set *c, struct uuid_entry *u)
1230{
1231 struct bcache_device *d = kzalloc(sizeof(struct bcache_device),
1232 GFP_KERNEL);
1233 if (!d)
1234 return -ENOMEM;
1235
1236 closure_init(&d->cl, NULL);
1237 set_closure_fn(&d->cl, flash_dev_flush, system_wq);
1238
1239 kobject_init(&d->kobj, &bch_flash_dev_ktype);
1240
1241 if (bcache_device_init(d, block_bytes(c), u->sectors))
1242 goto err;
1243
1244 bcache_device_attach(d, c, u - c->uuids);
1245 bch_flash_dev_request_init(d);
1246 add_disk(d->disk);
1247
1248 if (kobject_add(&d->kobj, &disk_to_dev(d->disk)->kobj, "bcache"))
1249 goto err;
1250
1251 bcache_device_link(d, c, "volume");
1252
1253 return 0;
1254err:
1255 kobject_put(&d->kobj);
1256 return -ENOMEM;
1257}
1258
1259static int flash_devs_run(struct cache_set *c)
1260{
1261 int ret = 0;
1262 struct uuid_entry *u;
1263
1264 for (u = c->uuids;
1265 u < c->uuids + c->nr_uuids && !ret;
1266 u++)
1267 if (UUID_FLASH_ONLY(u))
1268 ret = flash_dev_run(c, u);
1269
1270 return ret;
1271}
1272
1273int bch_flash_dev_create(struct cache_set *c, uint64_t size)
1274{
1275 struct uuid_entry *u;
1276
1277 if (test_bit(CACHE_SET_STOPPING, &c->flags))
1278 return -EINTR;
1279
1280 u = uuid_find_empty(c);
1281 if (!u) {
1282 pr_err("Can't create volume, no room for UUID");
1283 return -EINVAL;
1284 }
1285
1286 get_random_bytes(u->uuid, 16);
1287 memset(u->label, 0, 32);
1288 u->first_reg = u->last_reg = cpu_to_le32(get_seconds());
1289
1290 SET_UUID_FLASH_ONLY(u, 1);
1291 u->sectors = size >> 9;
1292
1293 bch_uuid_write(c);
1294
1295 return flash_dev_run(c, u);
1296}
1297
1298/* Cache set */
1299
1300__printf(2, 3)
1301bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...)
1302{
1303 va_list args;
1304
1305 if (c->on_error != ON_ERROR_PANIC &&
1306 test_bit(CACHE_SET_STOPPING, &c->flags))
1307 return false;
1308
1309 /* XXX: we can be called from atomic context
1310 acquire_console_sem();
1311 */
1312
1313 printk(KERN_ERR "bcache: error on %pU: ", c->sb.set_uuid);
1314
1315 va_start(args, fmt);
1316 vprintk(fmt, args);
1317 va_end(args);
1318
1319 printk(", disabling caching\n");
1320
1321 if (c->on_error == ON_ERROR_PANIC)
1322 panic("panic forced after error\n");
1323
1324 bch_cache_set_unregister(c);
1325 return true;
1326}
1327
1328void bch_cache_set_release(struct kobject *kobj)
1329{
1330 struct cache_set *c = container_of(kobj, struct cache_set, kobj);
1331 kfree(c);
1332 module_put(THIS_MODULE);
1333}
1334
1335static void cache_set_free(struct closure *cl)
1336{
1337 struct cache_set *c = container_of(cl, struct cache_set, cl);
1338 struct cache *ca;
1339 unsigned i;
1340
1341 if (!IS_ERR_OR_NULL(c->debug))
1342 debugfs_remove(c->debug);
1343
1344 bch_open_buckets_free(c);
1345 bch_btree_cache_free(c);
1346 bch_journal_free(c);
1347
1348 for_each_cache(ca, c, i)
1349 if (ca)
1350 kobject_put(&ca->kobj);
1351
1352 bch_bset_sort_state_free(&c->sort);
1353 free_pages((unsigned long) c->uuids, ilog2(bucket_pages(c)));
1354
1355 if (c->moving_gc_wq)
1356 destroy_workqueue(c->moving_gc_wq);
1357 if (c->bio_split)
1358 bioset_free(c->bio_split);
1359 if (c->fill_iter)
1360 mempool_destroy(c->fill_iter);
1361 if (c->bio_meta)
1362 mempool_destroy(c->bio_meta);
1363 if (c->search)
1364 mempool_destroy(c->search);
1365 kfree(c->devices);
1366
1367 mutex_lock(&bch_register_lock);
1368 list_del(&c->list);
1369 mutex_unlock(&bch_register_lock);
1370
1371 pr_info("Cache set %pU unregistered", c->sb.set_uuid);
1372 wake_up(&unregister_wait);
1373
1374 closure_debug_destroy(&c->cl);
1375 kobject_put(&c->kobj);
1376}
1377
1378static void cache_set_flush(struct closure *cl)
1379{
1380 struct cache_set *c = container_of(cl, struct cache_set, caching);
1381 struct cache *ca;
1382 struct btree *b;
1383 unsigned i;
1384
1385 bch_cache_accounting_destroy(&c->accounting);
1386
1387 kobject_put(&c->internal);
1388 kobject_del(&c->kobj);
1389
1390 if (c->gc_thread)
1391 kthread_stop(c->gc_thread);
1392
1393 if (!IS_ERR_OR_NULL(c->root))
1394 list_add(&c->root->list, &c->btree_cache);
1395
1396 /* Should skip this if we're unregistering because of an error */
1397 list_for_each_entry(b, &c->btree_cache, list) {
1398 mutex_lock(&b->write_lock);
1399 if (btree_node_dirty(b))
1400 __bch_btree_node_write(b, NULL);
1401 mutex_unlock(&b->write_lock);
1402 }
1403
1404 for_each_cache(ca, c, i)
1405 if (ca->alloc_thread)
1406 kthread_stop(ca->alloc_thread);
1407
1408 cancel_delayed_work_sync(&c->journal.work);
1409 /* flush last journal entry if needed */
1410 c->journal.work.work.func(&c->journal.work.work);
1411
1412 closure_return(cl);
1413}
1414
1415static void __cache_set_unregister(struct closure *cl)
1416{
1417 struct cache_set *c = container_of(cl, struct cache_set, caching);
1418 struct cached_dev *dc;
1419 size_t i;
1420
1421 mutex_lock(&bch_register_lock);
1422
1423 for (i = 0; i < c->nr_uuids; i++)
1424 if (c->devices[i]) {
1425 if (!UUID_FLASH_ONLY(&c->uuids[i]) &&
1426 test_bit(CACHE_SET_UNREGISTERING, &c->flags)) {
1427 dc = container_of(c->devices[i],
1428 struct cached_dev, disk);
1429 bch_cached_dev_detach(dc);
1430 } else {
1431 bcache_device_stop(c->devices[i]);
1432 }
1433 }
1434
1435 mutex_unlock(&bch_register_lock);
1436
1437 continue_at(cl, cache_set_flush, system_wq);
1438}
1439
1440void bch_cache_set_stop(struct cache_set *c)
1441{
1442 if (!test_and_set_bit(CACHE_SET_STOPPING, &c->flags))
1443 closure_queue(&c->caching);
1444}
1445
1446void bch_cache_set_unregister(struct cache_set *c)
1447{
1448 set_bit(CACHE_SET_UNREGISTERING, &c->flags);
1449 bch_cache_set_stop(c);
1450}
1451
1452#define alloc_bucket_pages(gfp, c) \
1453 ((void *) __get_free_pages(__GFP_ZERO|gfp, ilog2(bucket_pages(c))))
1454
1455struct cache_set *bch_cache_set_alloc(struct cache_sb *sb)
1456{
1457 int iter_size;
1458 struct cache_set *c = kzalloc(sizeof(struct cache_set), GFP_KERNEL);
1459 if (!c)
1460 return NULL;
1461
1462 __module_get(THIS_MODULE);
1463 closure_init(&c->cl, NULL);
1464 set_closure_fn(&c->cl, cache_set_free, system_wq);
1465
1466 closure_init(&c->caching, &c->cl);
1467 set_closure_fn(&c->caching, __cache_set_unregister, system_wq);
1468
1469 /* Maybe create continue_at_noreturn() and use it here? */
1470 closure_set_stopped(&c->cl);
1471 closure_put(&c->cl);
1472
1473 kobject_init(&c->kobj, &bch_cache_set_ktype);
1474 kobject_init(&c->internal, &bch_cache_set_internal_ktype);
1475
1476 bch_cache_accounting_init(&c->accounting, &c->cl);
1477
1478 memcpy(c->sb.set_uuid, sb->set_uuid, 16);
1479 c->sb.block_size = sb->block_size;
1480 c->sb.bucket_size = sb->bucket_size;
1481 c->sb.nr_in_set = sb->nr_in_set;
1482 c->sb.last_mount = sb->last_mount;
1483 c->bucket_bits = ilog2(sb->bucket_size);
1484 c->block_bits = ilog2(sb->block_size);
1485 c->nr_uuids = bucket_bytes(c) / sizeof(struct uuid_entry);
1486
1487 c->btree_pages = bucket_pages(c);
1488 if (c->btree_pages > BTREE_MAX_PAGES)
1489 c->btree_pages = max_t(int, c->btree_pages / 4,
1490 BTREE_MAX_PAGES);
1491
1492 sema_init(&c->sb_write_mutex, 1);
1493 mutex_init(&c->bucket_lock);
1494 init_waitqueue_head(&c->btree_cache_wait);
1495 init_waitqueue_head(&c->bucket_wait);
1496 sema_init(&c->uuid_write_mutex, 1);
1497
1498 spin_lock_init(&c->btree_gc_time.lock);
1499 spin_lock_init(&c->btree_split_time.lock);
1500 spin_lock_init(&c->btree_read_time.lock);
1501
1502 bch_moving_init_cache_set(c);
1503
1504 INIT_LIST_HEAD(&c->list);
1505 INIT_LIST_HEAD(&c->cached_devs);
1506 INIT_LIST_HEAD(&c->btree_cache);
1507 INIT_LIST_HEAD(&c->btree_cache_freeable);
1508 INIT_LIST_HEAD(&c->btree_cache_freed);
1509 INIT_LIST_HEAD(&c->data_buckets);
1510
1511 c->search = mempool_create_slab_pool(32, bch_search_cache);
1512 if (!c->search)
1513 goto err;
1514
1515 iter_size = (sb->bucket_size / sb->block_size + 1) *
1516 sizeof(struct btree_iter_set);
1517
1518 if (!(c->devices = kzalloc(c->nr_uuids * sizeof(void *), GFP_KERNEL)) ||
1519 !(c->bio_meta = mempool_create_kmalloc_pool(2,
1520 sizeof(struct bbio) + sizeof(struct bio_vec) *
1521 bucket_pages(c))) ||
1522 !(c->fill_iter = mempool_create_kmalloc_pool(1, iter_size)) ||
1523 !(c->bio_split = bioset_create(4, offsetof(struct bbio, bio))) ||
1524 !(c->uuids = alloc_bucket_pages(GFP_KERNEL, c)) ||
1525 !(c->moving_gc_wq = create_workqueue("bcache_gc")) ||
1526 bch_journal_alloc(c) ||
1527 bch_btree_cache_alloc(c) ||
1528 bch_open_buckets_alloc(c) ||
1529 bch_bset_sort_state_init(&c->sort, ilog2(c->btree_pages)))
1530 goto err;
1531
1532 c->congested_read_threshold_us = 2000;
1533 c->congested_write_threshold_us = 20000;
1534 c->error_limit = 8 << IO_ERROR_SHIFT;
1535
1536 return c;
1537err:
1538 bch_cache_set_unregister(c);
1539 return NULL;
1540}
1541
1542static void run_cache_set(struct cache_set *c)
1543{
1544 const char *err = "cannot allocate memory";
1545 struct cached_dev *dc, *t;
1546 struct cache *ca;
1547 struct closure cl;
1548 unsigned i;
1549
1550 closure_init_stack(&cl);
1551
1552 for_each_cache(ca, c, i)
1553 c->nbuckets += ca->sb.nbuckets;
1554
1555 if (CACHE_SYNC(&c->sb)) {
1556 LIST_HEAD(journal);
1557 struct bkey *k;
1558 struct jset *j;
1559
1560 err = "cannot allocate memory for journal";
1561 if (bch_journal_read(c, &journal))
1562 goto err;
1563
1564 pr_debug("btree_journal_read() done");
1565
1566 err = "no journal entries found";
1567 if (list_empty(&journal))
1568 goto err;
1569
1570 j = &list_entry(journal.prev, struct journal_replay, list)->j;
1571
1572 err = "IO error reading priorities";
1573 for_each_cache(ca, c, i)
1574 prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev]);
1575
1576 /*
1577 * If prio_read() fails it'll call cache_set_error and we'll
1578 * tear everything down right away, but if we perhaps checked
1579 * sooner we could avoid journal replay.
1580 */
1581
1582 k = &j->btree_root;
1583
1584 err = "bad btree root";
1585 if (__bch_btree_ptr_invalid(c, k))
1586 goto err;
1587
1588 err = "error reading btree root";
1589 c->root = bch_btree_node_get(c, NULL, k, j->btree_level, true);
1590 if (IS_ERR_OR_NULL(c->root))
1591 goto err;
1592
1593 list_del_init(&c->root->list);
1594 rw_unlock(true, c->root);
1595
1596 err = uuid_read(c, j, &cl);
1597 if (err)
1598 goto err;
1599
1600 err = "error in recovery";
1601 if (bch_btree_check(c))
1602 goto err;
1603
1604 bch_journal_mark(c, &journal);
1605 bch_initial_gc_finish(c);
1606 pr_debug("btree_check() done");
1607
1608 /*
1609 * bcache_journal_next() can't happen sooner, or
1610 * btree_gc_finish() will give spurious errors about last_gc >
1611 * gc_gen - this is a hack but oh well.
1612 */
1613 bch_journal_next(&c->journal);
1614
1615 err = "error starting allocator thread";
1616 for_each_cache(ca, c, i)
1617 if (bch_cache_allocator_start(ca))
1618 goto err;
1619
1620 /*
1621 * First place it's safe to allocate: btree_check() and
1622 * btree_gc_finish() have to run before we have buckets to
1623 * allocate, and bch_bucket_alloc_set() might cause a journal
1624 * entry to be written so bcache_journal_next() has to be called
1625 * first.
1626 *
1627 * If the uuids were in the old format we have to rewrite them
1628 * before the next journal entry is written:
1629 */
1630 if (j->version < BCACHE_JSET_VERSION_UUID)
1631 __uuid_write(c);
1632
1633 bch_journal_replay(c, &journal);
1634 } else {
1635 pr_notice("invalidating existing data");
1636
1637 for_each_cache(ca, c, i) {
1638 unsigned j;
1639
1640 ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7,
1641 2, SB_JOURNAL_BUCKETS);
1642
1643 for (j = 0; j < ca->sb.keys; j++)
1644 ca->sb.d[j] = ca->sb.first_bucket + j;
1645 }
1646
1647 bch_initial_gc_finish(c);
1648
1649 err = "error starting allocator thread";
1650 for_each_cache(ca, c, i)
1651 if (bch_cache_allocator_start(ca))
1652 goto err;
1653
1654 mutex_lock(&c->bucket_lock);
1655 for_each_cache(ca, c, i)
1656 bch_prio_write(ca);
1657 mutex_unlock(&c->bucket_lock);
1658
1659 err = "cannot allocate new UUID bucket";
1660 if (__uuid_write(c))
1661 goto err;
1662
1663 err = "cannot allocate new btree root";
1664 c->root = bch_btree_node_alloc(c, NULL, 0);
1665 if (IS_ERR_OR_NULL(c->root))
1666 goto err;
1667
1668 mutex_lock(&c->root->write_lock);
1669 bkey_copy_key(&c->root->key, &MAX_KEY);
1670 bch_btree_node_write(c->root, &cl);
1671 mutex_unlock(&c->root->write_lock);
1672
1673 bch_btree_set_root(c->root);
1674 rw_unlock(true, c->root);
1675
1676 /*
1677 * We don't want to write the first journal entry until
1678 * everything is set up - fortunately journal entries won't be
1679 * written until the SET_CACHE_SYNC() here:
1680 */
1681 SET_CACHE_SYNC(&c->sb, true);
1682
1683 bch_journal_next(&c->journal);
1684 bch_journal_meta(c, &cl);
1685 }
1686
1687 err = "error starting gc thread";
1688 if (bch_gc_thread_start(c))
1689 goto err;
1690
1691 closure_sync(&cl);
1692 c->sb.last_mount = get_seconds();
1693 bcache_write_super(c);
1694
1695 list_for_each_entry_safe(dc, t, &uncached_devices, list)
1696 bch_cached_dev_attach(dc, c);
1697
1698 flash_devs_run(c);
1699
1700 return;
1701err:
1702 closure_sync(&cl);
1703 /* XXX: test this, it's broken */
1704 bch_cache_set_error(c, "%s", err);
1705}
1706
1707static bool can_attach_cache(struct cache *ca, struct cache_set *c)
1708{
1709 return ca->sb.block_size == c->sb.block_size &&
1710 ca->sb.bucket_size == c->sb.bucket_size &&
1711 ca->sb.nr_in_set == c->sb.nr_in_set;
1712}
1713
1714static const char *register_cache_set(struct cache *ca)
1715{
1716 char buf[12];
1717 const char *err = "cannot allocate memory";
1718 struct cache_set *c;
1719
1720 list_for_each_entry(c, &bch_cache_sets, list)
1721 if (!memcmp(c->sb.set_uuid, ca->sb.set_uuid, 16)) {
1722 if (c->cache[ca->sb.nr_this_dev])
1723 return "duplicate cache set member";
1724
1725 if (!can_attach_cache(ca, c))
1726 return "cache sb does not match set";
1727
1728 if (!CACHE_SYNC(&ca->sb))
1729 SET_CACHE_SYNC(&c->sb, false);
1730
1731 goto found;
1732 }
1733
1734 c = bch_cache_set_alloc(&ca->sb);
1735 if (!c)
1736 return err;
1737
1738 err = "error creating kobject";
1739 if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->sb.set_uuid) ||
1740 kobject_add(&c->internal, &c->kobj, "internal"))
1741 goto err;
1742
1743 if (bch_cache_accounting_add_kobjs(&c->accounting, &c->kobj))
1744 goto err;
1745
1746 bch_debug_init_cache_set(c);
1747
1748 list_add(&c->list, &bch_cache_sets);
1749found:
1750 sprintf(buf, "cache%i", ca->sb.nr_this_dev);
1751 if (sysfs_create_link(&ca->kobj, &c->kobj, "set") ||
1752 sysfs_create_link(&c->kobj, &ca->kobj, buf))
1753 goto err;
1754
1755 if (ca->sb.seq > c->sb.seq) {
1756 c->sb.version = ca->sb.version;
1757 memcpy(c->sb.set_uuid, ca->sb.set_uuid, 16);
1758 c->sb.flags = ca->sb.flags;
1759 c->sb.seq = ca->sb.seq;
1760 pr_debug("set version = %llu", c->sb.version);
1761 }
1762
1763 ca->set = c;
1764 ca->set->cache[ca->sb.nr_this_dev] = ca;
1765 c->cache_by_alloc[c->caches_loaded++] = ca;
1766
1767 if (c->caches_loaded == c->sb.nr_in_set)
1768 run_cache_set(c);
1769
1770 return NULL;
1771err:
1772 bch_cache_set_unregister(c);
1773 return err;
1774}
1775
1776/* Cache device */
1777
1778void bch_cache_release(struct kobject *kobj)
1779{
1780 struct cache *ca = container_of(kobj, struct cache, kobj);
1781 unsigned i;
1782
1783 if (ca->set)
1784 ca->set->cache[ca->sb.nr_this_dev] = NULL;
1785
1786 bio_split_pool_free(&ca->bio_split_hook);
1787
1788 free_pages((unsigned long) ca->disk_buckets, ilog2(bucket_pages(ca)));
1789 kfree(ca->prio_buckets);
1790 vfree(ca->buckets);
1791
1792 free_heap(&ca->heap);
1793 free_fifo(&ca->free_inc);
1794
1795 for (i = 0; i < RESERVE_NR; i++)
1796 free_fifo(&ca->free[i]);
1797
1798 if (ca->sb_bio.bi_inline_vecs[0].bv_page)
1799 put_page(ca->sb_bio.bi_io_vec[0].bv_page);
1800
1801 if (!IS_ERR_OR_NULL(ca->bdev)) {
1802 blk_sync_queue(bdev_get_queue(ca->bdev));
1803 blkdev_put(ca->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1804 }
1805
1806 kfree(ca);
1807 module_put(THIS_MODULE);
1808}
1809
1810static int cache_alloc(struct cache_sb *sb, struct cache *ca)
1811{
1812 size_t free;
1813 struct bucket *b;
1814
1815 __module_get(THIS_MODULE);
1816 kobject_init(&ca->kobj, &bch_cache_ktype);
1817
1818 bio_init(&ca->journal.bio);
1819 ca->journal.bio.bi_max_vecs = 8;
1820 ca->journal.bio.bi_io_vec = ca->journal.bio.bi_inline_vecs;
1821
1822 free = roundup_pow_of_two(ca->sb.nbuckets) >> 10;
1823
1824 if (!init_fifo(&ca->free[RESERVE_BTREE], 8, GFP_KERNEL) ||
1825 !init_fifo(&ca->free[RESERVE_PRIO], prio_buckets(ca), GFP_KERNEL) ||
1826 !init_fifo(&ca->free[RESERVE_MOVINGGC], free, GFP_KERNEL) ||
1827 !init_fifo(&ca->free[RESERVE_NONE], free, GFP_KERNEL) ||
1828 !init_fifo(&ca->free_inc, free << 2, GFP_KERNEL) ||
1829 !init_heap(&ca->heap, free << 3, GFP_KERNEL) ||
1830 !(ca->buckets = vzalloc(sizeof(struct bucket) *
1831 ca->sb.nbuckets)) ||
1832 !(ca->prio_buckets = kzalloc(sizeof(uint64_t) * prio_buckets(ca) *
1833 2, GFP_KERNEL)) ||
1834 !(ca->disk_buckets = alloc_bucket_pages(GFP_KERNEL, ca)) ||
1835 bio_split_pool_init(&ca->bio_split_hook))
1836 return -ENOMEM;
1837
1838 ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca);
1839
1840 for_each_bucket(b, ca)
1841 atomic_set(&b->pin, 0);
1842
1843 return 0;
1844}
1845
1846static void register_cache(struct cache_sb *sb, struct page *sb_page,
1847 struct block_device *bdev, struct cache *ca)
1848{
1849 char name[BDEVNAME_SIZE];
1850 const char *err = "cannot allocate memory";
1851
1852 memcpy(&ca->sb, sb, sizeof(struct cache_sb));
1853 ca->bdev = bdev;
1854 ca->bdev->bd_holder = ca;
1855
1856 bio_init(&ca->sb_bio);
1857 ca->sb_bio.bi_max_vecs = 1;
1858 ca->sb_bio.bi_io_vec = ca->sb_bio.bi_inline_vecs;
1859 ca->sb_bio.bi_io_vec[0].bv_page = sb_page;
1860 get_page(sb_page);
1861
1862 if (blk_queue_discard(bdev_get_queue(ca->bdev)))
1863 ca->discard = CACHE_DISCARD(&ca->sb);
1864
1865 if (cache_alloc(sb, ca) != 0)
1866 goto err;
1867
1868 err = "error creating kobject";
1869 if (kobject_add(&ca->kobj, &part_to_dev(bdev->bd_part)->kobj, "bcache"))
1870 goto err;
1871
1872 mutex_lock(&bch_register_lock);
1873 err = register_cache_set(ca);
1874 mutex_unlock(&bch_register_lock);
1875
1876 if (err)
1877 goto err;
1878
1879 pr_info("registered cache device %s", bdevname(bdev, name));
1880 return;
1881err:
1882 pr_notice("error opening %s: %s", bdevname(bdev, name), err);
1883 kobject_put(&ca->kobj);
1884}
1885
1886/* Global interfaces/init */
1887
1888static ssize_t register_bcache(struct kobject *, struct kobj_attribute *,
1889 const char *, size_t);
1890
1891kobj_attribute_write(register, register_bcache);
1892kobj_attribute_write(register_quiet, register_bcache);
1893
1894static bool bch_is_open_backing(struct block_device *bdev) {
1895 struct cache_set *c, *tc;
1896 struct cached_dev *dc, *t;
1897
1898 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
1899 list_for_each_entry_safe(dc, t, &c->cached_devs, list)
1900 if (dc->bdev == bdev)
1901 return true;
1902 list_for_each_entry_safe(dc, t, &uncached_devices, list)
1903 if (dc->bdev == bdev)
1904 return true;
1905 return false;
1906}
1907
1908static bool bch_is_open_cache(struct block_device *bdev) {
1909 struct cache_set *c, *tc;
1910 struct cache *ca;
1911 unsigned i;
1912
1913 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
1914 for_each_cache(ca, c, i)
1915 if (ca->bdev == bdev)
1916 return true;
1917 return false;
1918}
1919
1920static bool bch_is_open(struct block_device *bdev) {
1921 return bch_is_open_cache(bdev) || bch_is_open_backing(bdev);
1922}
1923
1924static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr,
1925 const char *buffer, size_t size)
1926{
1927 ssize_t ret = size;
1928 const char *err = "cannot allocate memory";
1929 char *path = NULL;
1930 struct cache_sb *sb = NULL;
1931 struct block_device *bdev = NULL;
1932 struct page *sb_page = NULL;
1933
1934 if (!try_module_get(THIS_MODULE))
1935 return -EBUSY;
1936
1937 if (!(path = kstrndup(buffer, size, GFP_KERNEL)) ||
1938 !(sb = kmalloc(sizeof(struct cache_sb), GFP_KERNEL)))
1939 goto err;
1940
1941 err = "failed to open device";
1942 bdev = blkdev_get_by_path(strim(path),
1943 FMODE_READ|FMODE_WRITE|FMODE_EXCL,
1944 sb);
1945 if (IS_ERR(bdev)) {
1946 if (bdev == ERR_PTR(-EBUSY)) {
1947 bdev = lookup_bdev(strim(path));
1948 if (!IS_ERR(bdev) && bch_is_open(bdev))
1949 err = "device already registered";
1950 else
1951 err = "device busy";
1952 }
1953 goto err;
1954 }
1955
1956 err = "failed to set blocksize";
1957 if (set_blocksize(bdev, 4096))
1958 goto err_close;
1959
1960 err = read_super(sb, bdev, &sb_page);
1961 if (err)
1962 goto err_close;
1963
1964 if (SB_IS_BDEV(sb)) {
1965 struct cached_dev *dc = kzalloc(sizeof(*dc), GFP_KERNEL);
1966 if (!dc)
1967 goto err_close;
1968
1969 mutex_lock(&bch_register_lock);
1970 register_bdev(sb, sb_page, bdev, dc);
1971 mutex_unlock(&bch_register_lock);
1972 } else {
1973 struct cache *ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1974 if (!ca)
1975 goto err_close;
1976
1977 register_cache(sb, sb_page, bdev, ca);
1978 }
1979out:
1980 if (sb_page)
1981 put_page(sb_page);
1982 kfree(sb);
1983 kfree(path);
1984 module_put(THIS_MODULE);
1985 return ret;
1986
1987err_close:
1988 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1989err:
1990 if (attr != &ksysfs_register_quiet)
1991 pr_info("error opening %s: %s", path, err);
1992 ret = -EINVAL;
1993 goto out;
1994}
1995
1996static int bcache_reboot(struct notifier_block *n, unsigned long code, void *x)
1997{
1998 if (code == SYS_DOWN ||
1999 code == SYS_HALT ||
2000 code == SYS_POWER_OFF) {
2001 DEFINE_WAIT(wait);
2002 unsigned long start = jiffies;
2003 bool stopped = false;
2004
2005 struct cache_set *c, *tc;
2006 struct cached_dev *dc, *tdc;
2007
2008 mutex_lock(&bch_register_lock);
2009
2010 if (list_empty(&bch_cache_sets) &&
2011 list_empty(&uncached_devices))
2012 goto out;
2013
2014 pr_info("Stopping all devices:");
2015
2016 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
2017 bch_cache_set_stop(c);
2018
2019 list_for_each_entry_safe(dc, tdc, &uncached_devices, list)
2020 bcache_device_stop(&dc->disk);
2021
2022 /* What's a condition variable? */
2023 while (1) {
2024 long timeout = start + 2 * HZ - jiffies;
2025
2026 stopped = list_empty(&bch_cache_sets) &&
2027 list_empty(&uncached_devices);
2028
2029 if (timeout < 0 || stopped)
2030 break;
2031
2032 prepare_to_wait(&unregister_wait, &wait,
2033 TASK_UNINTERRUPTIBLE);
2034
2035 mutex_unlock(&bch_register_lock);
2036 schedule_timeout(timeout);
2037 mutex_lock(&bch_register_lock);
2038 }
2039
2040 finish_wait(&unregister_wait, &wait);
2041
2042 if (stopped)
2043 pr_info("All devices stopped");
2044 else
2045 pr_notice("Timeout waiting for devices to be closed");
2046out:
2047 mutex_unlock(&bch_register_lock);
2048 }
2049
2050 return NOTIFY_DONE;
2051}
2052
2053static struct notifier_block reboot = {
2054 .notifier_call = bcache_reboot,
2055 .priority = INT_MAX, /* before any real devices */
2056};
2057
2058static void bcache_exit(void)
2059{
2060 bch_debug_exit();
2061 bch_request_exit();
2062 if (bcache_kobj)
2063 kobject_put(bcache_kobj);
2064 if (bcache_wq)
2065 destroy_workqueue(bcache_wq);
2066 if (bcache_major)
2067 unregister_blkdev(bcache_major, "bcache");
2068 unregister_reboot_notifier(&reboot);
2069}
2070
2071static int __init bcache_init(void)
2072{
2073 static const struct attribute *files[] = {
2074 &ksysfs_register.attr,
2075 &ksysfs_register_quiet.attr,
2076 NULL
2077 };
2078
2079 mutex_init(&bch_register_lock);
2080 init_waitqueue_head(&unregister_wait);
2081 register_reboot_notifier(&reboot);
2082 closure_debug_init();
2083
2084 bcache_major = register_blkdev(0, "bcache");
2085 if (bcache_major < 0)
2086 return bcache_major;
2087
2088 if (!(bcache_wq = create_workqueue("bcache")) ||
2089 !(bcache_kobj = kobject_create_and_add("bcache", fs_kobj)) ||
2090 sysfs_create_files(bcache_kobj, files) ||
2091 bch_request_init() ||
2092 bch_debug_init(bcache_kobj))
2093 goto err;
2094
2095 return 0;
2096err:
2097 bcache_exit();
2098 return -ENOMEM;
2099}
2100
2101module_exit(bcache_exit);
2102module_init(bcache_init);