2 * bcache setup/teardown code, and some metadata io - read a superblock and
3 * figure out what to do with it.
5 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
6 * Copyright 2012 Google, Inc.
14 #include "writeback.h"
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>
27 MODULE_LICENSE("GPL");
28 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
30 static const char bcache_magic[] = {
31 0xc6, 0x85, 0x73, 0xf6, 0x4e, 0x1a, 0x45, 0xca,
32 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81
35 static const char invalid_uuid[] = {
36 0xa0, 0x3e, 0xf8, 0xed, 0x3e, 0xe1, 0xb8, 0x78,
37 0xc8, 0x50, 0xfc, 0x5e, 0xcb, 0x16, 0xcd, 0x99
40 /* Default is -1; we skip past it for struct cached_dev's cache mode */
41 const char * const bch_cache_modes[] = {
50 static struct kobject *bcache_kobj;
51 struct mutex bch_register_lock;
52 LIST_HEAD(bch_cache_sets);
53 static LIST_HEAD(uncached_devices);
55 static int bcache_major;
56 static DEFINE_IDA(bcache_minor);
57 static wait_queue_head_t unregister_wait;
58 struct workqueue_struct *bcache_wq;
60 #define BTREE_MAX_PAGES (256 * 1024 / PAGE_SIZE)
64 static const char *read_super(struct cache_sb *sb, struct block_device *bdev,
69 struct buffer_head *bh = __bread(bdev, 1, SB_SIZE);
75 s = (struct cache_sb *) bh->b_data;
77 sb->offset = le64_to_cpu(s->offset);
78 sb->version = le64_to_cpu(s->version);
80 memcpy(sb->magic, s->magic, 16);
81 memcpy(sb->uuid, s->uuid, 16);
82 memcpy(sb->set_uuid, s->set_uuid, 16);
83 memcpy(sb->label, s->label, SB_LABEL_SIZE);
85 sb->flags = le64_to_cpu(s->flags);
86 sb->seq = le64_to_cpu(s->seq);
87 sb->last_mount = le32_to_cpu(s->last_mount);
88 sb->first_bucket = le16_to_cpu(s->first_bucket);
89 sb->keys = le16_to_cpu(s->keys);
91 for (i = 0; i < SB_JOURNAL_BUCKETS; i++)
92 sb->d[i] = le64_to_cpu(s->d[i]);
94 pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u",
95 sb->version, sb->flags, sb->seq, sb->keys);
97 err = "Not a bcache superblock";
98 if (sb->offset != SB_SECTOR)
101 if (memcmp(sb->magic, bcache_magic, 16))
104 err = "Too many journal buckets";
105 if (sb->keys > SB_JOURNAL_BUCKETS)
108 err = "Bad checksum";
109 if (s->csum != csum_set(s))
113 if (bch_is_zero(sb->uuid, 16))
116 sb->block_size = le16_to_cpu(s->block_size);
118 err = "Superblock block size smaller than device block size";
119 if (sb->block_size << 9 < bdev_logical_block_size(bdev))
122 switch (sb->version) {
123 case BCACHE_SB_VERSION_BDEV:
124 sb->data_offset = BDEV_DATA_START_DEFAULT;
126 case BCACHE_SB_VERSION_BDEV_WITH_OFFSET:
127 sb->data_offset = le64_to_cpu(s->data_offset);
129 err = "Bad data offset";
130 if (sb->data_offset < BDEV_DATA_START_DEFAULT)
134 case BCACHE_SB_VERSION_CDEV:
135 case BCACHE_SB_VERSION_CDEV_WITH_UUID:
136 sb->nbuckets = le64_to_cpu(s->nbuckets);
137 sb->bucket_size = le16_to_cpu(s->bucket_size);
139 sb->nr_in_set = le16_to_cpu(s->nr_in_set);
140 sb->nr_this_dev = le16_to_cpu(s->nr_this_dev);
142 err = "Too many buckets";
143 if (sb->nbuckets > LONG_MAX)
146 err = "Not enough buckets";
147 if (sb->nbuckets < 1 << 7)
150 err = "Bad block/bucket size";
151 if (!is_power_of_2(sb->block_size) ||
152 sb->block_size > PAGE_SECTORS ||
153 !is_power_of_2(sb->bucket_size) ||
154 sb->bucket_size < PAGE_SECTORS)
157 err = "Invalid superblock: device too small";
158 if (get_capacity(bdev->bd_disk) < sb->bucket_size * sb->nbuckets)
162 if (bch_is_zero(sb->set_uuid, 16))
165 err = "Bad cache device number in set";
166 if (!sb->nr_in_set ||
167 sb->nr_in_set <= sb->nr_this_dev ||
168 sb->nr_in_set > MAX_CACHES_PER_SET)
171 err = "Journal buckets not sequential";
172 for (i = 0; i < sb->keys; i++)
173 if (sb->d[i] != sb->first_bucket + i)
176 err = "Too many journal buckets";
177 if (sb->first_bucket + sb->keys > sb->nbuckets)
180 err = "Invalid superblock: first bucket comes before end of super";
181 if (sb->first_bucket * sb->bucket_size < 16)
186 err = "Unsupported superblock version";
190 sb->last_mount = get_seconds();
193 get_page(bh->b_page);
200 static void write_bdev_super_endio(struct bio *bio)
202 struct cached_dev *dc = bio->bi_private;
203 /* XXX: error checking */
205 closure_put(&dc->sb_write);
208 static void __write_super(struct cache_sb *sb, struct bio *bio)
210 struct cache_sb *out = page_address(bio->bi_io_vec[0].bv_page);
213 bio->bi_iter.bi_sector = SB_SECTOR;
214 bio->bi_iter.bi_size = SB_SIZE;
215 bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
216 bch_bio_map(bio, NULL);
218 out->offset = cpu_to_le64(sb->offset);
219 out->version = cpu_to_le64(sb->version);
221 memcpy(out->uuid, sb->uuid, 16);
222 memcpy(out->set_uuid, sb->set_uuid, 16);
223 memcpy(out->label, sb->label, SB_LABEL_SIZE);
225 out->flags = cpu_to_le64(sb->flags);
226 out->seq = cpu_to_le64(sb->seq);
228 out->last_mount = cpu_to_le32(sb->last_mount);
229 out->first_bucket = cpu_to_le16(sb->first_bucket);
230 out->keys = cpu_to_le16(sb->keys);
232 for (i = 0; i < sb->keys; i++)
233 out->d[i] = cpu_to_le64(sb->d[i]);
235 out->csum = csum_set(out);
237 pr_debug("ver %llu, flags %llu, seq %llu",
238 sb->version, sb->flags, sb->seq);
243 static void bch_write_bdev_super_unlock(struct closure *cl)
245 struct cached_dev *dc = container_of(cl, struct cached_dev, sb_write);
247 up(&dc->sb_write_mutex);
250 void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent)
252 struct closure *cl = &dc->sb_write;
253 struct bio *bio = &dc->sb_bio;
255 down(&dc->sb_write_mutex);
256 closure_init(cl, parent);
259 bio->bi_bdev = dc->bdev;
260 bio->bi_end_io = write_bdev_super_endio;
261 bio->bi_private = dc;
264 __write_super(&dc->sb, bio);
266 closure_return_with_destructor(cl, bch_write_bdev_super_unlock);
269 static void write_super_endio(struct bio *bio)
271 struct cache *ca = bio->bi_private;
273 bch_count_io_errors(ca, bio->bi_error, "writing superblock");
274 closure_put(&ca->set->sb_write);
277 static void bcache_write_super_unlock(struct closure *cl)
279 struct cache_set *c = container_of(cl, struct cache_set, sb_write);
281 up(&c->sb_write_mutex);
284 void bcache_write_super(struct cache_set *c)
286 struct closure *cl = &c->sb_write;
290 down(&c->sb_write_mutex);
291 closure_init(cl, &c->cl);
295 for_each_cache(ca, c, i) {
296 struct bio *bio = &ca->sb_bio;
298 ca->sb.version = BCACHE_SB_VERSION_CDEV_WITH_UUID;
299 ca->sb.seq = c->sb.seq;
300 ca->sb.last_mount = c->sb.last_mount;
302 SET_CACHE_SYNC(&ca->sb, CACHE_SYNC(&c->sb));
305 bio->bi_bdev = ca->bdev;
306 bio->bi_end_io = write_super_endio;
307 bio->bi_private = ca;
310 __write_super(&ca->sb, bio);
313 closure_return_with_destructor(cl, bcache_write_super_unlock);
318 static void uuid_endio(struct bio *bio)
320 struct closure *cl = bio->bi_private;
321 struct cache_set *c = container_of(cl, struct cache_set, uuid_write);
323 cache_set_err_on(bio->bi_error, c, "accessing uuids");
324 bch_bbio_free(bio, c);
328 static void uuid_io_unlock(struct closure *cl)
330 struct cache_set *c = container_of(cl, struct cache_set, uuid_write);
332 up(&c->uuid_write_mutex);
335 static void uuid_io(struct cache_set *c, int op, unsigned long op_flags,
336 struct bkey *k, struct closure *parent)
338 struct closure *cl = &c->uuid_write;
339 struct uuid_entry *u;
344 down(&c->uuid_write_mutex);
345 closure_init(cl, parent);
347 for (i = 0; i < KEY_PTRS(k); i++) {
348 struct bio *bio = bch_bbio_alloc(c);
350 bio->bi_opf = REQ_SYNC | REQ_META | op_flags;
351 bio->bi_iter.bi_size = KEY_SIZE(k) << 9;
353 bio->bi_end_io = uuid_endio;
354 bio->bi_private = cl;
355 bio_set_op_attrs(bio, op, REQ_SYNC|REQ_META|op_flags);
356 bch_bio_map(bio, c->uuids);
358 bch_submit_bbio(bio, c, k, i);
360 if (op != REQ_OP_WRITE)
364 bch_extent_to_text(buf, sizeof(buf), k);
365 pr_debug("%s UUIDs at %s", op == REQ_OP_WRITE ? "wrote" : "read", buf);
367 for (u = c->uuids; u < c->uuids + c->nr_uuids; u++)
368 if (!bch_is_zero(u->uuid, 16))
369 pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u",
370 u - c->uuids, u->uuid, u->label,
371 u->first_reg, u->last_reg, u->invalidated);
373 closure_return_with_destructor(cl, uuid_io_unlock);
376 static char *uuid_read(struct cache_set *c, struct jset *j, struct closure *cl)
378 struct bkey *k = &j->uuid_bucket;
380 if (__bch_btree_ptr_invalid(c, k))
381 return "bad uuid pointer";
383 bkey_copy(&c->uuid_bucket, k);
384 uuid_io(c, REQ_OP_READ, READ_SYNC, k, cl);
386 if (j->version < BCACHE_JSET_VERSION_UUIDv1) {
387 struct uuid_entry_v0 *u0 = (void *) c->uuids;
388 struct uuid_entry *u1 = (void *) c->uuids;
394 * Since the new uuid entry is bigger than the old, we have to
395 * convert starting at the highest memory address and work down
396 * in order to do it in place
399 for (i = c->nr_uuids - 1;
402 memcpy(u1[i].uuid, u0[i].uuid, 16);
403 memcpy(u1[i].label, u0[i].label, 32);
405 u1[i].first_reg = u0[i].first_reg;
406 u1[i].last_reg = u0[i].last_reg;
407 u1[i].invalidated = u0[i].invalidated;
417 static int __uuid_write(struct cache_set *c)
421 closure_init_stack(&cl);
423 lockdep_assert_held(&bch_register_lock);
425 if (bch_bucket_alloc_set(c, RESERVE_BTREE, &k.key, 1, true))
428 SET_KEY_SIZE(&k.key, c->sb.bucket_size);
429 uuid_io(c, REQ_OP_WRITE, 0, &k.key, &cl);
432 bkey_copy(&c->uuid_bucket, &k.key);
437 int bch_uuid_write(struct cache_set *c)
439 int ret = __uuid_write(c);
442 bch_journal_meta(c, NULL);
447 static struct uuid_entry *uuid_find(struct cache_set *c, const char *uuid)
449 struct uuid_entry *u;
452 u < c->uuids + c->nr_uuids; u++)
453 if (!memcmp(u->uuid, uuid, 16))
459 static struct uuid_entry *uuid_find_empty(struct cache_set *c)
461 static const char zero_uuid[16] = "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0";
462 return uuid_find(c, zero_uuid);
466 * Bucket priorities/gens:
468 * For each bucket, we store on disk its
472 * See alloc.c for an explanation of the gen. The priority is used to implement
473 * lru (and in the future other) cache replacement policies; for most purposes
474 * it's just an opaque integer.
476 * The gens and the priorities don't have a whole lot to do with each other, and
477 * it's actually the gens that must be written out at specific times - it's no
478 * big deal if the priorities don't get written, if we lose them we just reuse
479 * buckets in suboptimal order.
481 * On disk they're stored in a packed array, and in as many buckets are required
482 * to fit them all. The buckets we use to store them form a list; the journal
483 * header points to the first bucket, the first bucket points to the second
486 * This code is used by the allocation code; periodically (whenever it runs out
487 * of buckets to allocate from) the allocation code will invalidate some
488 * buckets, but it can't use those buckets until their new gens are safely on
492 static void prio_endio(struct bio *bio)
494 struct cache *ca = bio->bi_private;
496 cache_set_err_on(bio->bi_error, ca->set, "accessing priorities");
497 bch_bbio_free(bio, ca->set);
498 closure_put(&ca->prio);
501 static void prio_io(struct cache *ca, uint64_t bucket, int op,
502 unsigned long op_flags)
504 struct closure *cl = &ca->prio;
505 struct bio *bio = bch_bbio_alloc(ca->set);
507 closure_init_stack(cl);
509 bio->bi_iter.bi_sector = bucket * ca->sb.bucket_size;
510 bio->bi_bdev = ca->bdev;
511 bio->bi_iter.bi_size = bucket_bytes(ca);
513 bio->bi_end_io = prio_endio;
514 bio->bi_private = ca;
515 bio_set_op_attrs(bio, op, REQ_SYNC|REQ_META|op_flags);
516 bch_bio_map(bio, ca->disk_buckets);
518 closure_bio_submit(bio, &ca->prio);
522 void bch_prio_write(struct cache *ca)
528 closure_init_stack(&cl);
530 lockdep_assert_held(&ca->set->bucket_lock);
532 ca->disk_buckets->seq++;
534 atomic_long_add(ca->sb.bucket_size * prio_buckets(ca),
535 &ca->meta_sectors_written);
537 //pr_debug("free %zu, free_inc %zu, unused %zu", fifo_used(&ca->free),
538 // fifo_used(&ca->free_inc), fifo_used(&ca->unused));
540 for (i = prio_buckets(ca) - 1; i >= 0; --i) {
542 struct prio_set *p = ca->disk_buckets;
543 struct bucket_disk *d = p->data;
544 struct bucket_disk *end = d + prios_per_bucket(ca);
546 for (b = ca->buckets + i * prios_per_bucket(ca);
547 b < ca->buckets + ca->sb.nbuckets && d < end;
549 d->prio = cpu_to_le16(b->prio);
553 p->next_bucket = ca->prio_buckets[i + 1];
554 p->magic = pset_magic(&ca->sb);
555 p->csum = bch_crc64(&p->magic, bucket_bytes(ca) - 8);
557 bucket = bch_bucket_alloc(ca, RESERVE_PRIO, true);
558 BUG_ON(bucket == -1);
560 mutex_unlock(&ca->set->bucket_lock);
561 prio_io(ca, bucket, REQ_OP_WRITE, 0);
562 mutex_lock(&ca->set->bucket_lock);
564 ca->prio_buckets[i] = bucket;
565 atomic_dec_bug(&ca->buckets[bucket].pin);
568 mutex_unlock(&ca->set->bucket_lock);
570 bch_journal_meta(ca->set, &cl);
573 mutex_lock(&ca->set->bucket_lock);
576 * Don't want the old priorities to get garbage collected until after we
577 * finish writing the new ones, and they're journalled
579 for (i = 0; i < prio_buckets(ca); i++) {
580 if (ca->prio_last_buckets[i])
581 __bch_bucket_free(ca,
582 &ca->buckets[ca->prio_last_buckets[i]]);
584 ca->prio_last_buckets[i] = ca->prio_buckets[i];
588 static void prio_read(struct cache *ca, uint64_t bucket)
590 struct prio_set *p = ca->disk_buckets;
591 struct bucket_disk *d = p->data + prios_per_bucket(ca), *end = d;
593 unsigned bucket_nr = 0;
595 for (b = ca->buckets;
596 b < ca->buckets + ca->sb.nbuckets;
599 ca->prio_buckets[bucket_nr] = bucket;
600 ca->prio_last_buckets[bucket_nr] = bucket;
603 prio_io(ca, bucket, REQ_OP_READ, READ_SYNC);
605 if (p->csum != bch_crc64(&p->magic, bucket_bytes(ca) - 8))
606 pr_warn("bad csum reading priorities");
608 if (p->magic != pset_magic(&ca->sb))
609 pr_warn("bad magic reading priorities");
611 bucket = p->next_bucket;
615 b->prio = le16_to_cpu(d->prio);
616 b->gen = b->last_gc = d->gen;
622 static int open_dev(struct block_device *b, fmode_t mode)
624 struct bcache_device *d = b->bd_disk->private_data;
625 if (test_bit(BCACHE_DEV_CLOSING, &d->flags))
632 static void release_dev(struct gendisk *b, fmode_t mode)
634 struct bcache_device *d = b->private_data;
638 static int ioctl_dev(struct block_device *b, fmode_t mode,
639 unsigned int cmd, unsigned long arg)
641 struct bcache_device *d = b->bd_disk->private_data;
642 return d->ioctl(d, mode, cmd, arg);
645 static const struct block_device_operations bcache_ops = {
647 .release = release_dev,
649 .owner = THIS_MODULE,
652 void bcache_device_stop(struct bcache_device *d)
654 if (!test_and_set_bit(BCACHE_DEV_CLOSING, &d->flags))
655 closure_queue(&d->cl);
658 static void bcache_device_unlink(struct bcache_device *d)
660 lockdep_assert_held(&bch_register_lock);
662 if (d->c && !test_and_set_bit(BCACHE_DEV_UNLINK_DONE, &d->flags)) {
666 sysfs_remove_link(&d->c->kobj, d->name);
667 sysfs_remove_link(&d->kobj, "cache");
669 for_each_cache(ca, d->c, i)
670 bd_unlink_disk_holder(ca->bdev, d->disk);
674 static void bcache_device_link(struct bcache_device *d, struct cache_set *c,
680 for_each_cache(ca, d->c, i)
681 bd_link_disk_holder(ca->bdev, d->disk);
683 snprintf(d->name, BCACHEDEVNAME_SIZE,
684 "%s%u", name, d->id);
686 WARN(sysfs_create_link(&d->kobj, &c->kobj, "cache") ||
687 sysfs_create_link(&c->kobj, &d->kobj, d->name),
688 "Couldn't create device <-> cache set symlinks");
690 clear_bit(BCACHE_DEV_UNLINK_DONE, &d->flags);
693 static void bcache_device_detach(struct bcache_device *d)
695 lockdep_assert_held(&bch_register_lock);
697 if (test_bit(BCACHE_DEV_DETACHING, &d->flags)) {
698 struct uuid_entry *u = d->c->uuids + d->id;
700 SET_UUID_FLASH_ONLY(u, 0);
701 memcpy(u->uuid, invalid_uuid, 16);
702 u->invalidated = cpu_to_le32(get_seconds());
703 bch_uuid_write(d->c);
706 bcache_device_unlink(d);
708 d->c->devices[d->id] = NULL;
709 closure_put(&d->c->caching);
713 static void bcache_device_attach(struct bcache_device *d, struct cache_set *c,
720 closure_get(&c->caching);
723 static void bcache_device_free(struct bcache_device *d)
725 lockdep_assert_held(&bch_register_lock);
727 pr_info("%s stopped", d->disk->disk_name);
730 bcache_device_detach(d);
731 if (d->disk && d->disk->flags & GENHD_FL_UP)
732 del_gendisk(d->disk);
733 if (d->disk && d->disk->queue)
734 blk_cleanup_queue(d->disk->queue);
736 ida_simple_remove(&bcache_minor, d->disk->first_minor);
741 bioset_free(d->bio_split);
742 kvfree(d->full_dirty_stripes);
743 kvfree(d->stripe_sectors_dirty);
745 closure_debug_destroy(&d->cl);
748 static int bcache_device_init(struct bcache_device *d, unsigned block_size,
751 struct request_queue *q;
756 d->stripe_size = 1 << 31;
758 d->nr_stripes = DIV_ROUND_UP_ULL(sectors, d->stripe_size);
760 if (!d->nr_stripes ||
761 d->nr_stripes > INT_MAX ||
762 d->nr_stripes > SIZE_MAX / sizeof(atomic_t)) {
763 pr_err("nr_stripes too large or invalid: %u (start sector beyond end of disk?)",
764 (unsigned)d->nr_stripes);
768 n = d->nr_stripes * sizeof(atomic_t);
769 d->stripe_sectors_dirty = n < PAGE_SIZE << 6
770 ? kzalloc(n, GFP_KERNEL)
772 if (!d->stripe_sectors_dirty)
775 n = BITS_TO_LONGS(d->nr_stripes) * sizeof(unsigned long);
776 d->full_dirty_stripes = n < PAGE_SIZE << 6
777 ? kzalloc(n, GFP_KERNEL)
779 if (!d->full_dirty_stripes)
782 minor = ida_simple_get(&bcache_minor, 0, MINORMASK + 1, GFP_KERNEL);
786 if (!(d->bio_split = bioset_create(4, offsetof(struct bbio, bio))) ||
787 !(d->disk = alloc_disk(1))) {
788 ida_simple_remove(&bcache_minor, minor);
792 set_capacity(d->disk, sectors);
793 snprintf(d->disk->disk_name, DISK_NAME_LEN, "bcache%i", minor);
795 d->disk->major = bcache_major;
796 d->disk->first_minor = minor;
797 d->disk->fops = &bcache_ops;
798 d->disk->private_data = d;
800 q = blk_alloc_queue(GFP_KERNEL);
804 blk_queue_make_request(q, NULL);
807 q->backing_dev_info.congested_data = d;
808 q->limits.max_hw_sectors = UINT_MAX;
809 q->limits.max_sectors = UINT_MAX;
810 q->limits.max_segment_size = UINT_MAX;
811 q->limits.max_segments = BIO_MAX_PAGES;
812 blk_queue_max_discard_sectors(q, UINT_MAX);
813 q->limits.discard_granularity = 512;
814 q->limits.io_min = block_size;
815 q->limits.logical_block_size = block_size;
816 q->limits.physical_block_size = block_size;
817 set_bit(QUEUE_FLAG_NONROT, &d->disk->queue->queue_flags);
818 clear_bit(QUEUE_FLAG_ADD_RANDOM, &d->disk->queue->queue_flags);
819 set_bit(QUEUE_FLAG_DISCARD, &d->disk->queue->queue_flags);
821 blk_queue_write_cache(q, true, true);
828 static void calc_cached_dev_sectors(struct cache_set *c)
830 uint64_t sectors = 0;
831 struct cached_dev *dc;
833 list_for_each_entry(dc, &c->cached_devs, list)
834 sectors += bdev_sectors(dc->bdev);
836 c->cached_dev_sectors = sectors;
839 void bch_cached_dev_run(struct cached_dev *dc)
841 struct bcache_device *d = &dc->disk;
842 char buf[SB_LABEL_SIZE + 1];
845 kasprintf(GFP_KERNEL, "CACHED_UUID=%pU", dc->sb.uuid),
850 memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
851 buf[SB_LABEL_SIZE] = '\0';
852 env[2] = kasprintf(GFP_KERNEL, "CACHED_LABEL=%s", buf);
854 if (atomic_xchg(&dc->running, 1)) {
861 BDEV_STATE(&dc->sb) != BDEV_STATE_NONE) {
863 closure_init_stack(&cl);
865 SET_BDEV_STATE(&dc->sb, BDEV_STATE_STALE);
866 bch_write_bdev_super(dc, &cl);
871 bd_link_disk_holder(dc->bdev, dc->disk.disk);
872 /* won't show up in the uevent file, use udevadm monitor -e instead
873 * only class / kset properties are persistent */
874 kobject_uevent_env(&disk_to_dev(d->disk)->kobj, KOBJ_CHANGE, env);
878 if (sysfs_create_link(&d->kobj, &disk_to_dev(d->disk)->kobj, "dev") ||
879 sysfs_create_link(&disk_to_dev(d->disk)->kobj, &d->kobj, "bcache"))
880 pr_debug("error creating sysfs link");
883 static void cached_dev_detach_finish(struct work_struct *w)
885 struct cached_dev *dc = container_of(w, struct cached_dev, detach);
886 char buf[BDEVNAME_SIZE];
888 closure_init_stack(&cl);
890 BUG_ON(!test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags));
891 BUG_ON(atomic_read(&dc->count));
893 mutex_lock(&bch_register_lock);
895 cancel_delayed_work_sync(&dc->writeback_rate_update);
896 if (!IS_ERR_OR_NULL(dc->writeback_thread)) {
897 kthread_stop(dc->writeback_thread);
898 dc->writeback_thread = NULL;
901 memset(&dc->sb.set_uuid, 0, 16);
902 SET_BDEV_STATE(&dc->sb, BDEV_STATE_NONE);
904 bch_write_bdev_super(dc, &cl);
907 calc_cached_dev_sectors(dc->disk.c);
908 bcache_device_detach(&dc->disk);
909 list_move(&dc->list, &uncached_devices);
911 clear_bit(BCACHE_DEV_DETACHING, &dc->disk.flags);
912 clear_bit(BCACHE_DEV_UNLINK_DONE, &dc->disk.flags);
914 mutex_unlock(&bch_register_lock);
916 pr_info("Caching disabled for %s", bdevname(dc->bdev, buf));
918 /* Drop ref we took in cached_dev_detach() */
919 closure_put(&dc->disk.cl);
922 void bch_cached_dev_detach(struct cached_dev *dc)
924 lockdep_assert_held(&bch_register_lock);
926 if (test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags))
929 if (test_and_set_bit(BCACHE_DEV_DETACHING, &dc->disk.flags))
933 * Block the device from being closed and freed until we're finished
936 closure_get(&dc->disk.cl);
938 bch_writeback_queue(dc);
942 int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c,
945 uint32_t rtime = cpu_to_le32(get_seconds());
946 struct uuid_entry *u;
947 char buf[BDEVNAME_SIZE];
948 struct cached_dev *exist_dc, *t;
950 bdevname(dc->bdev, buf);
952 if ((set_uuid && memcmp(set_uuid, c->sb.set_uuid, 16)) ||
953 (!set_uuid && memcmp(dc->sb.set_uuid, c->sb.set_uuid, 16)))
957 pr_err("Can't attach %s: already attached", buf);
961 if (test_bit(CACHE_SET_STOPPING, &c->flags)) {
962 pr_err("Can't attach %s: shutting down", buf);
966 if (dc->sb.block_size < c->sb.block_size) {
968 pr_err("Couldn't attach %s: block size less than set's block size",
973 /* Check whether already attached */
974 list_for_each_entry_safe(exist_dc, t, &c->cached_devs, list) {
975 if (!memcmp(dc->sb.uuid, exist_dc->sb.uuid, 16)) {
976 pr_err("Tried to attach %s but duplicate UUID already attached",
983 u = uuid_find(c, dc->sb.uuid);
986 (BDEV_STATE(&dc->sb) == BDEV_STATE_STALE ||
987 BDEV_STATE(&dc->sb) == BDEV_STATE_NONE)) {
988 memcpy(u->uuid, invalid_uuid, 16);
989 u->invalidated = cpu_to_le32(get_seconds());
994 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
995 pr_err("Couldn't find uuid for %s in set", buf);
999 u = uuid_find_empty(c);
1001 pr_err("Not caching %s, no room for UUID", buf);
1006 /* Deadlocks since we're called via sysfs...
1007 sysfs_remove_file(&dc->kobj, &sysfs_attach);
1010 if (bch_is_zero(u->uuid, 16)) {
1012 closure_init_stack(&cl);
1014 memcpy(u->uuid, dc->sb.uuid, 16);
1015 memcpy(u->label, dc->sb.label, SB_LABEL_SIZE);
1016 u->first_reg = u->last_reg = rtime;
1019 memcpy(dc->sb.set_uuid, c->sb.set_uuid, 16);
1020 SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN);
1022 bch_write_bdev_super(dc, &cl);
1025 u->last_reg = rtime;
1029 bcache_device_attach(&dc->disk, c, u - c->uuids);
1030 list_move(&dc->list, &c->cached_devs);
1031 calc_cached_dev_sectors(c);
1035 * dc->c must be set before dc->count != 0 - paired with the mb in
1038 atomic_set(&dc->count, 1);
1040 /* Block writeback thread, but spawn it */
1041 down_write(&dc->writeback_lock);
1042 if (bch_cached_dev_writeback_start(dc)) {
1043 up_write(&dc->writeback_lock);
1047 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
1048 bch_sectors_dirty_init(&dc->disk);
1049 atomic_set(&dc->has_dirty, 1);
1050 atomic_inc(&dc->count);
1051 bch_writeback_queue(dc);
1054 bch_cached_dev_run(dc);
1055 bcache_device_link(&dc->disk, c, "bdev");
1057 /* Allow the writeback thread to proceed */
1058 up_write(&dc->writeback_lock);
1060 pr_info("Caching %s as %s on set %pU",
1061 bdevname(dc->bdev, buf), dc->disk.disk->disk_name,
1062 dc->disk.c->sb.set_uuid);
1066 void bch_cached_dev_release(struct kobject *kobj)
1068 struct cached_dev *dc = container_of(kobj, struct cached_dev,
1071 module_put(THIS_MODULE);
1074 static void cached_dev_free(struct closure *cl)
1076 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1078 cancel_delayed_work_sync(&dc->writeback_rate_update);
1079 if (!IS_ERR_OR_NULL(dc->writeback_thread))
1080 kthread_stop(dc->writeback_thread);
1081 if (dc->writeback_write_wq)
1082 destroy_workqueue(dc->writeback_write_wq);
1084 mutex_lock(&bch_register_lock);
1086 if (atomic_read(&dc->running))
1087 bd_unlink_disk_holder(dc->bdev, dc->disk.disk);
1088 bcache_device_free(&dc->disk);
1089 list_del(&dc->list);
1091 mutex_unlock(&bch_register_lock);
1093 if (!IS_ERR_OR_NULL(dc->bdev))
1094 blkdev_put(dc->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1096 wake_up(&unregister_wait);
1098 kobject_put(&dc->disk.kobj);
1101 static void cached_dev_flush(struct closure *cl)
1103 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1104 struct bcache_device *d = &dc->disk;
1106 mutex_lock(&bch_register_lock);
1107 bcache_device_unlink(d);
1108 mutex_unlock(&bch_register_lock);
1110 bch_cache_accounting_destroy(&dc->accounting);
1111 kobject_del(&d->kobj);
1113 continue_at(cl, cached_dev_free, system_wq);
1116 static int cached_dev_init(struct cached_dev *dc, unsigned block_size)
1120 struct request_queue *q = bdev_get_queue(dc->bdev);
1122 __module_get(THIS_MODULE);
1123 INIT_LIST_HEAD(&dc->list);
1124 closure_init(&dc->disk.cl, NULL);
1125 set_closure_fn(&dc->disk.cl, cached_dev_flush, system_wq);
1126 kobject_init(&dc->disk.kobj, &bch_cached_dev_ktype);
1127 INIT_WORK(&dc->detach, cached_dev_detach_finish);
1128 sema_init(&dc->sb_write_mutex, 1);
1129 INIT_LIST_HEAD(&dc->io_lru);
1130 spin_lock_init(&dc->io_lock);
1131 bch_cache_accounting_init(&dc->accounting, &dc->disk.cl);
1133 dc->sequential_cutoff = 4 << 20;
1135 for (io = dc->io; io < dc->io + RECENT_IO; io++) {
1136 list_add(&io->lru, &dc->io_lru);
1137 hlist_add_head(&io->hash, dc->io_hash + RECENT_IO);
1140 dc->disk.stripe_size = q->limits.io_opt >> 9;
1142 if (dc->disk.stripe_size)
1143 dc->partial_stripes_expensive =
1144 q->limits.raid_partial_stripes_expensive;
1146 ret = bcache_device_init(&dc->disk, block_size,
1147 dc->bdev->bd_part->nr_sects - dc->sb.data_offset);
1151 set_capacity(dc->disk.disk,
1152 dc->bdev->bd_part->nr_sects - dc->sb.data_offset);
1154 dc->disk.disk->queue->backing_dev_info.ra_pages =
1155 max(dc->disk.disk->queue->backing_dev_info.ra_pages,
1156 q->backing_dev_info.ra_pages);
1158 bch_cached_dev_request_init(dc);
1159 bch_cached_dev_writeback_init(dc);
1163 /* Cached device - bcache superblock */
1165 static void register_bdev(struct cache_sb *sb, struct page *sb_page,
1166 struct block_device *bdev,
1167 struct cached_dev *dc)
1169 char name[BDEVNAME_SIZE];
1170 const char *err = "cannot allocate memory";
1171 struct cache_set *c;
1173 memcpy(&dc->sb, sb, sizeof(struct cache_sb));
1175 dc->bdev->bd_holder = dc;
1177 bio_init(&dc->sb_bio);
1178 dc->sb_bio.bi_max_vecs = 1;
1179 dc->sb_bio.bi_io_vec = dc->sb_bio.bi_inline_vecs;
1180 dc->sb_bio.bi_io_vec[0].bv_page = sb_page;
1183 if (cached_dev_init(dc, sb->block_size << 9))
1186 err = "error creating kobject";
1187 if (kobject_add(&dc->disk.kobj, &part_to_dev(bdev->bd_part)->kobj,
1190 if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj))
1193 pr_info("registered backing device %s", bdevname(bdev, name));
1195 list_add(&dc->list, &uncached_devices);
1196 list_for_each_entry(c, &bch_cache_sets, list)
1197 bch_cached_dev_attach(dc, c, NULL);
1199 if (BDEV_STATE(&dc->sb) == BDEV_STATE_NONE ||
1200 BDEV_STATE(&dc->sb) == BDEV_STATE_STALE)
1201 bch_cached_dev_run(dc);
1205 pr_notice("error %s: %s", bdevname(bdev, name), err);
1206 bcache_device_stop(&dc->disk);
1209 /* Flash only volumes */
1211 void bch_flash_dev_release(struct kobject *kobj)
1213 struct bcache_device *d = container_of(kobj, struct bcache_device,
1218 static void flash_dev_free(struct closure *cl)
1220 struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1221 mutex_lock(&bch_register_lock);
1222 bcache_device_free(d);
1223 mutex_unlock(&bch_register_lock);
1224 kobject_put(&d->kobj);
1227 static void flash_dev_flush(struct closure *cl)
1229 struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1231 mutex_lock(&bch_register_lock);
1232 bcache_device_unlink(d);
1233 mutex_unlock(&bch_register_lock);
1234 kobject_del(&d->kobj);
1235 continue_at(cl, flash_dev_free, system_wq);
1238 static int flash_dev_run(struct cache_set *c, struct uuid_entry *u)
1240 struct bcache_device *d = kzalloc(sizeof(struct bcache_device),
1245 closure_init(&d->cl, NULL);
1246 set_closure_fn(&d->cl, flash_dev_flush, system_wq);
1248 kobject_init(&d->kobj, &bch_flash_dev_ktype);
1250 if (bcache_device_init(d, block_bytes(c), u->sectors))
1253 bcache_device_attach(d, c, u - c->uuids);
1254 bch_sectors_dirty_init(d);
1255 bch_flash_dev_request_init(d);
1258 if (kobject_add(&d->kobj, &disk_to_dev(d->disk)->kobj, "bcache"))
1261 bcache_device_link(d, c, "volume");
1265 kobject_put(&d->kobj);
1269 static int flash_devs_run(struct cache_set *c)
1272 struct uuid_entry *u;
1275 u < c->uuids + c->nr_uuids && !ret;
1277 if (UUID_FLASH_ONLY(u))
1278 ret = flash_dev_run(c, u);
1283 int bch_flash_dev_create(struct cache_set *c, uint64_t size)
1285 struct uuid_entry *u;
1287 if (test_bit(CACHE_SET_STOPPING, &c->flags))
1290 if (!test_bit(CACHE_SET_RUNNING, &c->flags))
1293 u = uuid_find_empty(c);
1295 pr_err("Can't create volume, no room for UUID");
1299 get_random_bytes(u->uuid, 16);
1300 memset(u->label, 0, 32);
1301 u->first_reg = u->last_reg = cpu_to_le32(get_seconds());
1303 SET_UUID_FLASH_ONLY(u, 1);
1304 u->sectors = size >> 9;
1308 return flash_dev_run(c, u);
1314 bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...)
1318 if (c->on_error != ON_ERROR_PANIC &&
1319 test_bit(CACHE_SET_STOPPING, &c->flags))
1322 /* XXX: we can be called from atomic context
1323 acquire_console_sem();
1326 printk(KERN_ERR "bcache: error on %pU: ", c->sb.set_uuid);
1328 va_start(args, fmt);
1332 printk(", disabling caching\n");
1334 if (c->on_error == ON_ERROR_PANIC)
1335 panic("panic forced after error\n");
1337 bch_cache_set_unregister(c);
1341 void bch_cache_set_release(struct kobject *kobj)
1343 struct cache_set *c = container_of(kobj, struct cache_set, kobj);
1345 module_put(THIS_MODULE);
1348 static void cache_set_free(struct closure *cl)
1350 struct cache_set *c = container_of(cl, struct cache_set, cl);
1354 if (!IS_ERR_OR_NULL(c->debug))
1355 debugfs_remove(c->debug);
1357 bch_open_buckets_free(c);
1358 bch_btree_cache_free(c);
1359 bch_journal_free(c);
1361 mutex_lock(&bch_register_lock);
1362 for_each_cache(ca, c, i)
1365 c->cache[ca->sb.nr_this_dev] = NULL;
1366 kobject_put(&ca->kobj);
1369 bch_bset_sort_state_free(&c->sort);
1370 free_pages((unsigned long) c->uuids, ilog2(bucket_pages(c)));
1372 if (c->moving_gc_wq)
1373 destroy_workqueue(c->moving_gc_wq);
1375 bioset_free(c->bio_split);
1377 mempool_destroy(c->fill_iter);
1379 mempool_destroy(c->bio_meta);
1381 mempool_destroy(c->search);
1385 mutex_unlock(&bch_register_lock);
1387 pr_info("Cache set %pU unregistered", c->sb.set_uuid);
1388 wake_up(&unregister_wait);
1390 closure_debug_destroy(&c->cl);
1391 kobject_put(&c->kobj);
1394 static void cache_set_flush(struct closure *cl)
1396 struct cache_set *c = container_of(cl, struct cache_set, caching);
1401 bch_cache_accounting_destroy(&c->accounting);
1403 kobject_put(&c->internal);
1404 kobject_del(&c->kobj);
1406 if (!IS_ERR_OR_NULL(c->gc_thread))
1407 kthread_stop(c->gc_thread);
1409 if (!IS_ERR_OR_NULL(c->root))
1410 list_add(&c->root->list, &c->btree_cache);
1412 /* Should skip this if we're unregistering because of an error */
1413 list_for_each_entry(b, &c->btree_cache, list) {
1414 mutex_lock(&b->write_lock);
1415 if (btree_node_dirty(b))
1416 __bch_btree_node_write(b, NULL);
1417 mutex_unlock(&b->write_lock);
1420 for_each_cache(ca, c, i)
1421 if (ca->alloc_thread)
1422 kthread_stop(ca->alloc_thread);
1424 if (c->journal.cur) {
1425 cancel_delayed_work_sync(&c->journal.work);
1426 /* flush last journal entry if needed */
1427 c->journal.work.work.func(&c->journal.work.work);
1433 static void __cache_set_unregister(struct closure *cl)
1435 struct cache_set *c = container_of(cl, struct cache_set, caching);
1436 struct cached_dev *dc;
1439 mutex_lock(&bch_register_lock);
1441 for (i = 0; i < c->nr_uuids; i++)
1442 if (c->devices[i]) {
1443 if (!UUID_FLASH_ONLY(&c->uuids[i]) &&
1444 test_bit(CACHE_SET_UNREGISTERING, &c->flags)) {
1445 dc = container_of(c->devices[i],
1446 struct cached_dev, disk);
1447 bch_cached_dev_detach(dc);
1449 bcache_device_stop(c->devices[i]);
1453 mutex_unlock(&bch_register_lock);
1455 continue_at(cl, cache_set_flush, system_wq);
1458 void bch_cache_set_stop(struct cache_set *c)
1460 if (!test_and_set_bit(CACHE_SET_STOPPING, &c->flags))
1461 closure_queue(&c->caching);
1464 void bch_cache_set_unregister(struct cache_set *c)
1466 set_bit(CACHE_SET_UNREGISTERING, &c->flags);
1467 bch_cache_set_stop(c);
1470 #define alloc_bucket_pages(gfp, c) \
1471 ((void *) __get_free_pages(__GFP_ZERO|__GFP_COMP|gfp, ilog2(bucket_pages(c))))
1473 struct cache_set *bch_cache_set_alloc(struct cache_sb *sb)
1476 struct cache_set *c = kzalloc(sizeof(struct cache_set), GFP_KERNEL);
1480 __module_get(THIS_MODULE);
1481 closure_init(&c->cl, NULL);
1482 set_closure_fn(&c->cl, cache_set_free, system_wq);
1484 closure_init(&c->caching, &c->cl);
1485 set_closure_fn(&c->caching, __cache_set_unregister, system_wq);
1487 /* Maybe create continue_at_noreturn() and use it here? */
1488 closure_set_stopped(&c->cl);
1489 closure_put(&c->cl);
1491 kobject_init(&c->kobj, &bch_cache_set_ktype);
1492 kobject_init(&c->internal, &bch_cache_set_internal_ktype);
1494 bch_cache_accounting_init(&c->accounting, &c->cl);
1496 memcpy(c->sb.set_uuid, sb->set_uuid, 16);
1497 c->sb.block_size = sb->block_size;
1498 c->sb.bucket_size = sb->bucket_size;
1499 c->sb.nr_in_set = sb->nr_in_set;
1500 c->sb.last_mount = sb->last_mount;
1501 c->bucket_bits = ilog2(sb->bucket_size);
1502 c->block_bits = ilog2(sb->block_size);
1503 c->nr_uuids = bucket_bytes(c) / sizeof(struct uuid_entry);
1505 c->btree_pages = bucket_pages(c);
1506 if (c->btree_pages > BTREE_MAX_PAGES)
1507 c->btree_pages = max_t(int, c->btree_pages / 4,
1510 sema_init(&c->sb_write_mutex, 1);
1511 mutex_init(&c->bucket_lock);
1512 init_waitqueue_head(&c->btree_cache_wait);
1513 spin_lock_init(&c->btree_cannibalize_lock);
1514 init_waitqueue_head(&c->bucket_wait);
1515 init_waitqueue_head(&c->gc_wait);
1516 sema_init(&c->uuid_write_mutex, 1);
1518 spin_lock_init(&c->btree_gc_time.lock);
1519 spin_lock_init(&c->btree_split_time.lock);
1520 spin_lock_init(&c->btree_read_time.lock);
1522 bch_moving_init_cache_set(c);
1524 INIT_LIST_HEAD(&c->list);
1525 INIT_LIST_HEAD(&c->cached_devs);
1526 INIT_LIST_HEAD(&c->btree_cache);
1527 INIT_LIST_HEAD(&c->btree_cache_freeable);
1528 INIT_LIST_HEAD(&c->btree_cache_freed);
1529 INIT_LIST_HEAD(&c->data_buckets);
1531 c->search = mempool_create_slab_pool(32, bch_search_cache);
1535 iter_size = (sb->bucket_size / sb->block_size + 1) *
1536 sizeof(struct btree_iter_set);
1538 if (!(c->devices = kzalloc(c->nr_uuids * sizeof(void *), GFP_KERNEL)) ||
1539 !(c->bio_meta = mempool_create_kmalloc_pool(2,
1540 sizeof(struct bbio) + sizeof(struct bio_vec) *
1541 bucket_pages(c))) ||
1542 !(c->fill_iter = mempool_create_kmalloc_pool(1, iter_size)) ||
1543 !(c->bio_split = bioset_create(4, offsetof(struct bbio, bio))) ||
1544 !(c->uuids = alloc_bucket_pages(GFP_KERNEL, c)) ||
1545 !(c->moving_gc_wq = alloc_workqueue("bcache_gc",
1546 WQ_MEM_RECLAIM, 0)) ||
1547 bch_journal_alloc(c) ||
1548 bch_btree_cache_alloc(c) ||
1549 bch_open_buckets_alloc(c) ||
1550 bch_bset_sort_state_init(&c->sort, ilog2(c->btree_pages)))
1553 c->congested_read_threshold_us = 2000;
1554 c->congested_write_threshold_us = 20000;
1555 c->error_limit = 8 << IO_ERROR_SHIFT;
1559 bch_cache_set_unregister(c);
1563 static int run_cache_set(struct cache_set *c)
1565 const char *err = "cannot allocate memory";
1566 struct cached_dev *dc, *t;
1571 closure_init_stack(&cl);
1573 for_each_cache(ca, c, i)
1574 c->nbuckets += ca->sb.nbuckets;
1577 if (CACHE_SYNC(&c->sb)) {
1582 err = "cannot allocate memory for journal";
1583 if (bch_journal_read(c, &journal))
1586 pr_debug("btree_journal_read() done");
1588 err = "no journal entries found";
1589 if (list_empty(&journal))
1592 j = &list_entry(journal.prev, struct journal_replay, list)->j;
1594 err = "IO error reading priorities";
1595 for_each_cache(ca, c, i)
1596 prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev]);
1599 * If prio_read() fails it'll call cache_set_error and we'll
1600 * tear everything down right away, but if we perhaps checked
1601 * sooner we could avoid journal replay.
1606 err = "bad btree root";
1607 if (__bch_btree_ptr_invalid(c, k))
1610 err = "error reading btree root";
1611 c->root = bch_btree_node_get(c, NULL, k, j->btree_level, true, NULL);
1612 if (IS_ERR_OR_NULL(c->root))
1615 list_del_init(&c->root->list);
1616 rw_unlock(true, c->root);
1618 err = uuid_read(c, j, &cl);
1622 err = "error in recovery";
1623 if (bch_btree_check(c))
1626 bch_journal_mark(c, &journal);
1627 bch_initial_gc_finish(c);
1628 pr_debug("btree_check() done");
1631 * bcache_journal_next() can't happen sooner, or
1632 * btree_gc_finish() will give spurious errors about last_gc >
1633 * gc_gen - this is a hack but oh well.
1635 bch_journal_next(&c->journal);
1637 err = "error starting allocator thread";
1638 for_each_cache(ca, c, i)
1639 if (bch_cache_allocator_start(ca))
1643 * First place it's safe to allocate: btree_check() and
1644 * btree_gc_finish() have to run before we have buckets to
1645 * allocate, and bch_bucket_alloc_set() might cause a journal
1646 * entry to be written so bcache_journal_next() has to be called
1649 * If the uuids were in the old format we have to rewrite them
1650 * before the next journal entry is written:
1652 if (j->version < BCACHE_JSET_VERSION_UUID)
1655 err = "bcache: replay journal failed";
1656 if (bch_journal_replay(c, &journal))
1659 pr_notice("invalidating existing data");
1661 for_each_cache(ca, c, i) {
1664 ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7,
1665 2, SB_JOURNAL_BUCKETS);
1667 for (j = 0; j < ca->sb.keys; j++)
1668 ca->sb.d[j] = ca->sb.first_bucket + j;
1671 bch_initial_gc_finish(c);
1673 err = "error starting allocator thread";
1674 for_each_cache(ca, c, i)
1675 if (bch_cache_allocator_start(ca))
1678 mutex_lock(&c->bucket_lock);
1679 for_each_cache(ca, c, i)
1681 mutex_unlock(&c->bucket_lock);
1683 err = "cannot allocate new UUID bucket";
1684 if (__uuid_write(c))
1687 err = "cannot allocate new btree root";
1688 c->root = __bch_btree_node_alloc(c, NULL, 0, true, NULL);
1689 if (IS_ERR_OR_NULL(c->root))
1692 mutex_lock(&c->root->write_lock);
1693 bkey_copy_key(&c->root->key, &MAX_KEY);
1694 bch_btree_node_write(c->root, &cl);
1695 mutex_unlock(&c->root->write_lock);
1697 bch_btree_set_root(c->root);
1698 rw_unlock(true, c->root);
1701 * We don't want to write the first journal entry until
1702 * everything is set up - fortunately journal entries won't be
1703 * written until the SET_CACHE_SYNC() here:
1705 SET_CACHE_SYNC(&c->sb, true);
1707 bch_journal_next(&c->journal);
1708 bch_journal_meta(c, &cl);
1711 err = "error starting gc thread";
1712 if (bch_gc_thread_start(c))
1716 c->sb.last_mount = get_seconds();
1717 bcache_write_super(c);
1719 list_for_each_entry_safe(dc, t, &uncached_devices, list)
1720 bch_cached_dev_attach(dc, c, NULL);
1724 set_bit(CACHE_SET_RUNNING, &c->flags);
1728 /* XXX: test this, it's broken */
1729 bch_cache_set_error(c, "%s", err);
1734 static bool can_attach_cache(struct cache *ca, struct cache_set *c)
1736 return ca->sb.block_size == c->sb.block_size &&
1737 ca->sb.bucket_size == c->sb.bucket_size &&
1738 ca->sb.nr_in_set == c->sb.nr_in_set;
1741 static const char *register_cache_set(struct cache *ca)
1744 const char *err = "cannot allocate memory";
1745 struct cache_set *c;
1747 list_for_each_entry(c, &bch_cache_sets, list)
1748 if (!memcmp(c->sb.set_uuid, ca->sb.set_uuid, 16)) {
1749 if (c->cache[ca->sb.nr_this_dev])
1750 return "duplicate cache set member";
1752 if (!can_attach_cache(ca, c))
1753 return "cache sb does not match set";
1755 if (!CACHE_SYNC(&ca->sb))
1756 SET_CACHE_SYNC(&c->sb, false);
1761 c = bch_cache_set_alloc(&ca->sb);
1765 err = "error creating kobject";
1766 if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->sb.set_uuid) ||
1767 kobject_add(&c->internal, &c->kobj, "internal"))
1770 if (bch_cache_accounting_add_kobjs(&c->accounting, &c->kobj))
1773 bch_debug_init_cache_set(c);
1775 list_add(&c->list, &bch_cache_sets);
1777 sprintf(buf, "cache%i", ca->sb.nr_this_dev);
1778 if (sysfs_create_link(&ca->kobj, &c->kobj, "set") ||
1779 sysfs_create_link(&c->kobj, &ca->kobj, buf))
1783 * A special case is both ca->sb.seq and c->sb.seq are 0,
1784 * such condition happens on a new created cache device whose
1785 * super block is never flushed yet. In this case c->sb.version
1786 * and other members should be updated too, otherwise we will
1787 * have a mistaken super block version in cache set.
1789 if (ca->sb.seq > c->sb.seq || c->sb.seq == 0) {
1790 c->sb.version = ca->sb.version;
1791 memcpy(c->sb.set_uuid, ca->sb.set_uuid, 16);
1792 c->sb.flags = ca->sb.flags;
1793 c->sb.seq = ca->sb.seq;
1794 pr_debug("set version = %llu", c->sb.version);
1797 kobject_get(&ca->kobj);
1799 ca->set->cache[ca->sb.nr_this_dev] = ca;
1800 c->cache_by_alloc[c->caches_loaded++] = ca;
1802 if (c->caches_loaded == c->sb.nr_in_set) {
1803 err = "failed to run cache set";
1804 if (run_cache_set(c) < 0)
1810 bch_cache_set_unregister(c);
1816 void bch_cache_release(struct kobject *kobj)
1818 struct cache *ca = container_of(kobj, struct cache, kobj);
1822 BUG_ON(ca->set->cache[ca->sb.nr_this_dev] != ca);
1823 ca->set->cache[ca->sb.nr_this_dev] = NULL;
1826 free_pages((unsigned long) ca->disk_buckets, ilog2(bucket_pages(ca)));
1827 kfree(ca->prio_buckets);
1830 free_heap(&ca->heap);
1831 free_fifo(&ca->free_inc);
1833 for (i = 0; i < RESERVE_NR; i++)
1834 free_fifo(&ca->free[i]);
1836 if (ca->sb_bio.bi_inline_vecs[0].bv_page)
1837 put_page(ca->sb_bio.bi_io_vec[0].bv_page);
1839 if (!IS_ERR_OR_NULL(ca->bdev))
1840 blkdev_put(ca->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1843 module_put(THIS_MODULE);
1846 static int cache_alloc(struct cache *ca)
1849 size_t btree_buckets;
1852 __module_get(THIS_MODULE);
1853 kobject_init(&ca->kobj, &bch_cache_ktype);
1855 bio_init(&ca->journal.bio);
1856 ca->journal.bio.bi_max_vecs = 8;
1857 ca->journal.bio.bi_io_vec = ca->journal.bio.bi_inline_vecs;
1860 * when ca->sb.njournal_buckets is not zero, journal exists,
1861 * and in bch_journal_replay(), tree node may split,
1862 * so bucket of RESERVE_BTREE type is needed,
1863 * the worst situation is all journal buckets are valid journal,
1864 * and all the keys need to replay,
1865 * so the number of RESERVE_BTREE type buckets should be as much
1866 * as journal buckets
1868 btree_buckets = ca->sb.njournal_buckets ?: 8;
1869 free = roundup_pow_of_two(ca->sb.nbuckets) >> 10;
1871 if (!init_fifo(&ca->free[RESERVE_BTREE], btree_buckets, GFP_KERNEL) ||
1872 !init_fifo_exact(&ca->free[RESERVE_PRIO], prio_buckets(ca), GFP_KERNEL) ||
1873 !init_fifo(&ca->free[RESERVE_MOVINGGC], free, GFP_KERNEL) ||
1874 !init_fifo(&ca->free[RESERVE_NONE], free, GFP_KERNEL) ||
1875 !init_fifo(&ca->free_inc, free << 2, GFP_KERNEL) ||
1876 !init_heap(&ca->heap, free << 3, GFP_KERNEL) ||
1877 !(ca->buckets = vzalloc(sizeof(struct bucket) *
1878 ca->sb.nbuckets)) ||
1879 !(ca->prio_buckets = kzalloc(sizeof(uint64_t) * prio_buckets(ca) *
1881 !(ca->disk_buckets = alloc_bucket_pages(GFP_KERNEL, ca)))
1884 ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca);
1886 for_each_bucket(b, ca)
1887 atomic_set(&b->pin, 0);
1892 static int register_cache(struct cache_sb *sb, struct page *sb_page,
1893 struct block_device *bdev, struct cache *ca)
1895 char name[BDEVNAME_SIZE];
1896 const char *err = NULL; /* must be set for any error case */
1899 bdevname(bdev, name);
1901 memcpy(&ca->sb, sb, sizeof(struct cache_sb));
1903 ca->bdev->bd_holder = ca;
1905 bio_init(&ca->sb_bio);
1906 ca->sb_bio.bi_max_vecs = 1;
1907 ca->sb_bio.bi_io_vec = ca->sb_bio.bi_inline_vecs;
1908 ca->sb_bio.bi_io_vec[0].bv_page = sb_page;
1911 if (blk_queue_discard(bdev_get_queue(bdev)))
1912 ca->discard = CACHE_DISCARD(&ca->sb);
1914 ret = cache_alloc(ca);
1916 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1918 err = "cache_alloc(): -ENOMEM";
1920 err = "cache_alloc(): unknown error";
1924 if (kobject_add(&ca->kobj, &part_to_dev(bdev->bd_part)->kobj, "bcache")) {
1925 err = "error calling kobject_add";
1930 mutex_lock(&bch_register_lock);
1931 err = register_cache_set(ca);
1932 mutex_unlock(&bch_register_lock);
1939 pr_info("registered cache device %s", name);
1942 kobject_put(&ca->kobj);
1946 pr_notice("error %s: %s", name, err);
1951 /* Global interfaces/init */
1953 static ssize_t register_bcache(struct kobject *, struct kobj_attribute *,
1954 const char *, size_t);
1956 kobj_attribute_write(register, register_bcache);
1957 kobj_attribute_write(register_quiet, register_bcache);
1959 static bool bch_is_open_backing(struct block_device *bdev) {
1960 struct cache_set *c, *tc;
1961 struct cached_dev *dc, *t;
1963 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
1964 list_for_each_entry_safe(dc, t, &c->cached_devs, list)
1965 if (dc->bdev == bdev)
1967 list_for_each_entry_safe(dc, t, &uncached_devices, list)
1968 if (dc->bdev == bdev)
1973 static bool bch_is_open_cache(struct block_device *bdev) {
1974 struct cache_set *c, *tc;
1978 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
1979 for_each_cache(ca, c, i)
1980 if (ca->bdev == bdev)
1985 static bool bch_is_open(struct block_device *bdev) {
1986 return bch_is_open_cache(bdev) || bch_is_open_backing(bdev);
1989 static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr,
1990 const char *buffer, size_t size)
1993 const char *err = "cannot allocate memory";
1995 struct cache_sb *sb = NULL;
1996 struct block_device *bdev = NULL;
1997 struct page *sb_page = NULL;
1999 if (!try_module_get(THIS_MODULE))
2002 if (!(path = kstrndup(buffer, size, GFP_KERNEL)) ||
2003 !(sb = kmalloc(sizeof(struct cache_sb), GFP_KERNEL)))
2006 err = "failed to open device";
2007 bdev = blkdev_get_by_path(strim(path),
2008 FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2011 if (bdev == ERR_PTR(-EBUSY)) {
2012 bdev = lookup_bdev(strim(path));
2013 mutex_lock(&bch_register_lock);
2014 if (!IS_ERR(bdev) && bch_is_open(bdev))
2015 err = "device already registered";
2017 err = "device busy";
2018 mutex_unlock(&bch_register_lock);
2021 if (attr == &ksysfs_register_quiet)
2027 err = "failed to set blocksize";
2028 if (set_blocksize(bdev, 4096))
2031 err = read_super(sb, bdev, &sb_page);
2035 err = "failed to register device";
2036 if (SB_IS_BDEV(sb)) {
2037 struct cached_dev *dc = kzalloc(sizeof(*dc), GFP_KERNEL);
2041 mutex_lock(&bch_register_lock);
2042 register_bdev(sb, sb_page, bdev, dc);
2043 mutex_unlock(&bch_register_lock);
2045 struct cache *ca = kzalloc(sizeof(*ca), GFP_KERNEL);
2049 if (register_cache(sb, sb_page, bdev, ca) != 0)
2057 module_put(THIS_MODULE);
2061 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2063 pr_info("error %s: %s", path, err);
2068 static int bcache_reboot(struct notifier_block *n, unsigned long code, void *x)
2070 if (code == SYS_DOWN ||
2072 code == SYS_POWER_OFF) {
2074 unsigned long start = jiffies;
2075 bool stopped = false;
2077 struct cache_set *c, *tc;
2078 struct cached_dev *dc, *tdc;
2080 mutex_lock(&bch_register_lock);
2082 if (list_empty(&bch_cache_sets) &&
2083 list_empty(&uncached_devices))
2086 pr_info("Stopping all devices:");
2088 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
2089 bch_cache_set_stop(c);
2091 list_for_each_entry_safe(dc, tdc, &uncached_devices, list)
2092 bcache_device_stop(&dc->disk);
2094 /* What's a condition variable? */
2096 long timeout = start + 2 * HZ - jiffies;
2098 stopped = list_empty(&bch_cache_sets) &&
2099 list_empty(&uncached_devices);
2101 if (timeout < 0 || stopped)
2104 prepare_to_wait(&unregister_wait, &wait,
2105 TASK_UNINTERRUPTIBLE);
2107 mutex_unlock(&bch_register_lock);
2108 schedule_timeout(timeout);
2109 mutex_lock(&bch_register_lock);
2112 finish_wait(&unregister_wait, &wait);
2115 pr_info("All devices stopped");
2117 pr_notice("Timeout waiting for devices to be closed");
2119 mutex_unlock(&bch_register_lock);
2125 static struct notifier_block reboot = {
2126 .notifier_call = bcache_reboot,
2127 .priority = INT_MAX, /* before any real devices */
2130 static void bcache_exit(void)
2135 kobject_put(bcache_kobj);
2137 destroy_workqueue(bcache_wq);
2139 unregister_blkdev(bcache_major, "bcache");
2140 unregister_reboot_notifier(&reboot);
2141 mutex_destroy(&bch_register_lock);
2144 static int __init bcache_init(void)
2146 static const struct attribute *files[] = {
2147 &ksysfs_register.attr,
2148 &ksysfs_register_quiet.attr,
2152 mutex_init(&bch_register_lock);
2153 init_waitqueue_head(&unregister_wait);
2154 register_reboot_notifier(&reboot);
2155 closure_debug_init();
2157 bcache_major = register_blkdev(0, "bcache");
2158 if (bcache_major < 0) {
2159 unregister_reboot_notifier(&reboot);
2160 mutex_destroy(&bch_register_lock);
2161 return bcache_major;
2164 if (!(bcache_wq = alloc_workqueue("bcache", WQ_MEM_RECLAIM, 0)) ||
2165 !(bcache_kobj = kobject_create_and_add("bcache", fs_kobj)) ||
2166 bch_request_init() ||
2167 bch_debug_init(bcache_kobj) ||
2168 sysfs_create_files(bcache_kobj, files))
2177 module_exit(bcache_exit);
2178 module_init(bcache_init);