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)
61 #define BCACHE_MINORS 16 /* partition support */
65 static const char *read_super(struct cache_sb *sb, struct block_device *bdev,
70 struct buffer_head *bh = __bread(bdev, 1, SB_SIZE);
76 s = (struct cache_sb *) bh->b_data;
78 sb->offset = le64_to_cpu(s->offset);
79 sb->version = le64_to_cpu(s->version);
81 memcpy(sb->magic, s->magic, 16);
82 memcpy(sb->uuid, s->uuid, 16);
83 memcpy(sb->set_uuid, s->set_uuid, 16);
84 memcpy(sb->label, s->label, SB_LABEL_SIZE);
86 sb->flags = le64_to_cpu(s->flags);
87 sb->seq = le64_to_cpu(s->seq);
88 sb->last_mount = le32_to_cpu(s->last_mount);
89 sb->first_bucket = le16_to_cpu(s->first_bucket);
90 sb->keys = le16_to_cpu(s->keys);
92 for (i = 0; i < SB_JOURNAL_BUCKETS; i++)
93 sb->d[i] = le64_to_cpu(s->d[i]);
95 pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u",
96 sb->version, sb->flags, sb->seq, sb->keys);
98 err = "Not a bcache superblock";
99 if (sb->offset != SB_SECTOR)
102 if (memcmp(sb->magic, bcache_magic, 16))
105 err = "Too many journal buckets";
106 if (sb->keys > SB_JOURNAL_BUCKETS)
109 err = "Bad checksum";
110 if (s->csum != csum_set(s))
114 if (bch_is_zero(sb->uuid, 16))
117 sb->block_size = le16_to_cpu(s->block_size);
119 err = "Superblock block size smaller than device block size";
120 if (sb->block_size << 9 < bdev_logical_block_size(bdev))
123 switch (sb->version) {
124 case BCACHE_SB_VERSION_BDEV:
125 sb->data_offset = BDEV_DATA_START_DEFAULT;
127 case BCACHE_SB_VERSION_BDEV_WITH_OFFSET:
128 sb->data_offset = le64_to_cpu(s->data_offset);
130 err = "Bad data offset";
131 if (sb->data_offset < BDEV_DATA_START_DEFAULT)
135 case BCACHE_SB_VERSION_CDEV:
136 case BCACHE_SB_VERSION_CDEV_WITH_UUID:
137 sb->nbuckets = le64_to_cpu(s->nbuckets);
138 sb->bucket_size = le16_to_cpu(s->bucket_size);
140 sb->nr_in_set = le16_to_cpu(s->nr_in_set);
141 sb->nr_this_dev = le16_to_cpu(s->nr_this_dev);
143 err = "Too many buckets";
144 if (sb->nbuckets > LONG_MAX)
147 err = "Not enough buckets";
148 if (sb->nbuckets < 1 << 7)
151 err = "Bad block/bucket size";
152 if (!is_power_of_2(sb->block_size) ||
153 sb->block_size > PAGE_SECTORS ||
154 !is_power_of_2(sb->bucket_size) ||
155 sb->bucket_size < PAGE_SECTORS)
158 err = "Invalid superblock: device too small";
159 if (get_capacity(bdev->bd_disk) < sb->bucket_size * sb->nbuckets)
163 if (bch_is_zero(sb->set_uuid, 16))
166 err = "Bad cache device number in set";
167 if (!sb->nr_in_set ||
168 sb->nr_in_set <= sb->nr_this_dev ||
169 sb->nr_in_set > MAX_CACHES_PER_SET)
172 err = "Journal buckets not sequential";
173 for (i = 0; i < sb->keys; i++)
174 if (sb->d[i] != sb->first_bucket + i)
177 err = "Too many journal buckets";
178 if (sb->first_bucket + sb->keys > sb->nbuckets)
181 err = "Invalid superblock: first bucket comes before end of super";
182 if (sb->first_bucket * sb->bucket_size < 16)
187 err = "Unsupported superblock version";
191 sb->last_mount = get_seconds();
194 get_page(bh->b_page);
201 static void write_bdev_super_endio(struct bio *bio)
203 struct cached_dev *dc = bio->bi_private;
204 /* XXX: error checking */
206 closure_put(&dc->sb_write);
209 static void __write_super(struct cache_sb *sb, struct bio *bio)
211 struct cache_sb *out = page_address(bio->bi_io_vec[0].bv_page);
214 bio->bi_iter.bi_sector = SB_SECTOR;
215 bio->bi_iter.bi_size = SB_SIZE;
216 bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
217 bch_bio_map(bio, NULL);
219 out->offset = cpu_to_le64(sb->offset);
220 out->version = cpu_to_le64(sb->version);
222 memcpy(out->uuid, sb->uuid, 16);
223 memcpy(out->set_uuid, sb->set_uuid, 16);
224 memcpy(out->label, sb->label, SB_LABEL_SIZE);
226 out->flags = cpu_to_le64(sb->flags);
227 out->seq = cpu_to_le64(sb->seq);
229 out->last_mount = cpu_to_le32(sb->last_mount);
230 out->first_bucket = cpu_to_le16(sb->first_bucket);
231 out->keys = cpu_to_le16(sb->keys);
233 for (i = 0; i < sb->keys; i++)
234 out->d[i] = cpu_to_le64(sb->d[i]);
236 out->csum = csum_set(out);
238 pr_debug("ver %llu, flags %llu, seq %llu",
239 sb->version, sb->flags, sb->seq);
244 static void bch_write_bdev_super_unlock(struct closure *cl)
246 struct cached_dev *dc = container_of(cl, struct cached_dev, sb_write);
248 up(&dc->sb_write_mutex);
251 void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent)
253 struct closure *cl = &dc->sb_write;
254 struct bio *bio = &dc->sb_bio;
256 down(&dc->sb_write_mutex);
257 closure_init(cl, parent);
260 bio_set_dev(bio, dc->bdev);
261 bio->bi_end_io = write_bdev_super_endio;
262 bio->bi_private = dc;
265 __write_super(&dc->sb, bio);
267 closure_return_with_destructor(cl, bch_write_bdev_super_unlock);
270 static void write_super_endio(struct bio *bio)
272 struct cache *ca = bio->bi_private;
274 bch_count_io_errors(ca, bio->bi_status, "writing superblock");
275 closure_put(&ca->set->sb_write);
278 static void bcache_write_super_unlock(struct closure *cl)
280 struct cache_set *c = container_of(cl, struct cache_set, sb_write);
282 up(&c->sb_write_mutex);
285 void bcache_write_super(struct cache_set *c)
287 struct closure *cl = &c->sb_write;
291 down(&c->sb_write_mutex);
292 closure_init(cl, &c->cl);
296 for_each_cache(ca, c, i) {
297 struct bio *bio = &ca->sb_bio;
299 ca->sb.version = BCACHE_SB_VERSION_CDEV_WITH_UUID;
300 ca->sb.seq = c->sb.seq;
301 ca->sb.last_mount = c->sb.last_mount;
303 SET_CACHE_SYNC(&ca->sb, CACHE_SYNC(&c->sb));
306 bio_set_dev(bio, ca->bdev);
307 bio->bi_end_io = write_super_endio;
308 bio->bi_private = ca;
311 __write_super(&ca->sb, bio);
314 closure_return_with_destructor(cl, bcache_write_super_unlock);
319 static void uuid_endio(struct bio *bio)
321 struct closure *cl = bio->bi_private;
322 struct cache_set *c = container_of(cl, struct cache_set, uuid_write);
324 cache_set_err_on(bio->bi_status, c, "accessing uuids");
325 bch_bbio_free(bio, c);
329 static void uuid_io_unlock(struct closure *cl)
331 struct cache_set *c = container_of(cl, struct cache_set, uuid_write);
333 up(&c->uuid_write_mutex);
336 static void uuid_io(struct cache_set *c, int op, unsigned long op_flags,
337 struct bkey *k, struct closure *parent)
339 struct closure *cl = &c->uuid_write;
340 struct uuid_entry *u;
345 down(&c->uuid_write_mutex);
346 closure_init(cl, parent);
348 for (i = 0; i < KEY_PTRS(k); i++) {
349 struct bio *bio = bch_bbio_alloc(c);
351 bio->bi_opf = REQ_SYNC | REQ_META | op_flags;
352 bio->bi_iter.bi_size = KEY_SIZE(k) << 9;
354 bio->bi_end_io = uuid_endio;
355 bio->bi_private = cl;
356 bio_set_op_attrs(bio, op, REQ_SYNC|REQ_META|op_flags);
357 bch_bio_map(bio, c->uuids);
359 bch_submit_bbio(bio, c, k, i);
361 if (op != REQ_OP_WRITE)
365 bch_extent_to_text(buf, sizeof(buf), k);
366 pr_debug("%s UUIDs at %s", op == REQ_OP_WRITE ? "wrote" : "read", buf);
368 for (u = c->uuids; u < c->uuids + c->nr_uuids; u++)
369 if (!bch_is_zero(u->uuid, 16))
370 pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u",
371 u - c->uuids, u->uuid, u->label,
372 u->first_reg, u->last_reg, u->invalidated);
374 closure_return_with_destructor(cl, uuid_io_unlock);
377 static char *uuid_read(struct cache_set *c, struct jset *j, struct closure *cl)
379 struct bkey *k = &j->uuid_bucket;
381 if (__bch_btree_ptr_invalid(c, k))
382 return "bad uuid pointer";
384 bkey_copy(&c->uuid_bucket, k);
385 uuid_io(c, REQ_OP_READ, 0, k, cl);
387 if (j->version < BCACHE_JSET_VERSION_UUIDv1) {
388 struct uuid_entry_v0 *u0 = (void *) c->uuids;
389 struct uuid_entry *u1 = (void *) c->uuids;
395 * Since the new uuid entry is bigger than the old, we have to
396 * convert starting at the highest memory address and work down
397 * in order to do it in place
400 for (i = c->nr_uuids - 1;
403 memcpy(u1[i].uuid, u0[i].uuid, 16);
404 memcpy(u1[i].label, u0[i].label, 32);
406 u1[i].first_reg = u0[i].first_reg;
407 u1[i].last_reg = u0[i].last_reg;
408 u1[i].invalidated = u0[i].invalidated;
418 static int __uuid_write(struct cache_set *c)
422 closure_init_stack(&cl);
424 lockdep_assert_held(&bch_register_lock);
426 if (bch_bucket_alloc_set(c, RESERVE_BTREE, &k.key, 1, true))
429 SET_KEY_SIZE(&k.key, c->sb.bucket_size);
430 uuid_io(c, REQ_OP_WRITE, 0, &k.key, &cl);
433 bkey_copy(&c->uuid_bucket, &k.key);
438 int bch_uuid_write(struct cache_set *c)
440 int ret = __uuid_write(c);
443 bch_journal_meta(c, NULL);
448 static struct uuid_entry *uuid_find(struct cache_set *c, const char *uuid)
450 struct uuid_entry *u;
453 u < c->uuids + c->nr_uuids; u++)
454 if (!memcmp(u->uuid, uuid, 16))
460 static struct uuid_entry *uuid_find_empty(struct cache_set *c)
462 static const char zero_uuid[16] = "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0";
463 return uuid_find(c, zero_uuid);
467 * Bucket priorities/gens:
469 * For each bucket, we store on disk its
473 * See alloc.c for an explanation of the gen. The priority is used to implement
474 * lru (and in the future other) cache replacement policies; for most purposes
475 * it's just an opaque integer.
477 * The gens and the priorities don't have a whole lot to do with each other, and
478 * it's actually the gens that must be written out at specific times - it's no
479 * big deal if the priorities don't get written, if we lose them we just reuse
480 * buckets in suboptimal order.
482 * On disk they're stored in a packed array, and in as many buckets are required
483 * to fit them all. The buckets we use to store them form a list; the journal
484 * header points to the first bucket, the first bucket points to the second
487 * This code is used by the allocation code; periodically (whenever it runs out
488 * of buckets to allocate from) the allocation code will invalidate some
489 * buckets, but it can't use those buckets until their new gens are safely on
493 static void prio_endio(struct bio *bio)
495 struct cache *ca = bio->bi_private;
497 cache_set_err_on(bio->bi_status, ca->set, "accessing priorities");
498 bch_bbio_free(bio, ca->set);
499 closure_put(&ca->prio);
502 static void prio_io(struct cache *ca, uint64_t bucket, int op,
503 unsigned long op_flags)
505 struct closure *cl = &ca->prio;
506 struct bio *bio = bch_bbio_alloc(ca->set);
508 closure_init_stack(cl);
510 bio->bi_iter.bi_sector = bucket * ca->sb.bucket_size;
511 bio_set_dev(bio, ca->bdev);
512 bio->bi_iter.bi_size = bucket_bytes(ca);
514 bio->bi_end_io = prio_endio;
515 bio->bi_private = ca;
516 bio_set_op_attrs(bio, op, REQ_SYNC|REQ_META|op_flags);
517 bch_bio_map(bio, ca->disk_buckets);
519 closure_bio_submit(bio, &ca->prio);
523 void bch_prio_write(struct cache *ca)
529 closure_init_stack(&cl);
531 lockdep_assert_held(&ca->set->bucket_lock);
533 ca->disk_buckets->seq++;
535 atomic_long_add(ca->sb.bucket_size * prio_buckets(ca),
536 &ca->meta_sectors_written);
538 //pr_debug("free %zu, free_inc %zu, unused %zu", fifo_used(&ca->free),
539 // fifo_used(&ca->free_inc), fifo_used(&ca->unused));
541 for (i = prio_buckets(ca) - 1; i >= 0; --i) {
543 struct prio_set *p = ca->disk_buckets;
544 struct bucket_disk *d = p->data;
545 struct bucket_disk *end = d + prios_per_bucket(ca);
547 for (b = ca->buckets + i * prios_per_bucket(ca);
548 b < ca->buckets + ca->sb.nbuckets && d < end;
550 d->prio = cpu_to_le16(b->prio);
554 p->next_bucket = ca->prio_buckets[i + 1];
555 p->magic = pset_magic(&ca->sb);
556 p->csum = bch_crc64(&p->magic, bucket_bytes(ca) - 8);
558 bucket = bch_bucket_alloc(ca, RESERVE_PRIO, true);
559 BUG_ON(bucket == -1);
561 mutex_unlock(&ca->set->bucket_lock);
562 prio_io(ca, bucket, REQ_OP_WRITE, 0);
563 mutex_lock(&ca->set->bucket_lock);
565 ca->prio_buckets[i] = bucket;
566 atomic_dec_bug(&ca->buckets[bucket].pin);
569 mutex_unlock(&ca->set->bucket_lock);
571 bch_journal_meta(ca->set, &cl);
574 mutex_lock(&ca->set->bucket_lock);
577 * Don't want the old priorities to get garbage collected until after we
578 * finish writing the new ones, and they're journalled
580 for (i = 0; i < prio_buckets(ca); i++) {
581 if (ca->prio_last_buckets[i])
582 __bch_bucket_free(ca,
583 &ca->buckets[ca->prio_last_buckets[i]]);
585 ca->prio_last_buckets[i] = ca->prio_buckets[i];
589 static void prio_read(struct cache *ca, uint64_t bucket)
591 struct prio_set *p = ca->disk_buckets;
592 struct bucket_disk *d = p->data + prios_per_bucket(ca), *end = d;
594 unsigned bucket_nr = 0;
596 for (b = ca->buckets;
597 b < ca->buckets + ca->sb.nbuckets;
600 ca->prio_buckets[bucket_nr] = bucket;
601 ca->prio_last_buckets[bucket_nr] = bucket;
604 prio_io(ca, bucket, REQ_OP_READ, 0);
606 if (p->csum != bch_crc64(&p->magic, bucket_bytes(ca) - 8))
607 pr_warn("bad csum reading priorities");
609 if (p->magic != pset_magic(&ca->sb))
610 pr_warn("bad magic reading priorities");
612 bucket = p->next_bucket;
616 b->prio = le16_to_cpu(d->prio);
617 b->gen = b->last_gc = d->gen;
623 static int open_dev(struct block_device *b, fmode_t mode)
625 struct bcache_device *d = b->bd_disk->private_data;
626 if (test_bit(BCACHE_DEV_CLOSING, &d->flags))
633 static void release_dev(struct gendisk *b, fmode_t mode)
635 struct bcache_device *d = b->private_data;
639 static int ioctl_dev(struct block_device *b, fmode_t mode,
640 unsigned int cmd, unsigned long arg)
642 struct bcache_device *d = b->bd_disk->private_data;
643 return d->ioctl(d, mode, cmd, arg);
646 static const struct block_device_operations bcache_ops = {
648 .release = release_dev,
650 .owner = THIS_MODULE,
653 void bcache_device_stop(struct bcache_device *d)
655 if (!test_and_set_bit(BCACHE_DEV_CLOSING, &d->flags))
656 closure_queue(&d->cl);
659 static void bcache_device_unlink(struct bcache_device *d)
661 lockdep_assert_held(&bch_register_lock);
663 if (d->c && !test_and_set_bit(BCACHE_DEV_UNLINK_DONE, &d->flags)) {
667 sysfs_remove_link(&d->c->kobj, d->name);
668 sysfs_remove_link(&d->kobj, "cache");
670 for_each_cache(ca, d->c, i)
671 bd_unlink_disk_holder(ca->bdev, d->disk);
675 static void bcache_device_link(struct bcache_device *d, struct cache_set *c,
681 for_each_cache(ca, d->c, i)
682 bd_link_disk_holder(ca->bdev, d->disk);
684 snprintf(d->name, BCACHEDEVNAME_SIZE,
685 "%s%u", name, d->id);
687 WARN(sysfs_create_link(&d->kobj, &c->kobj, "cache") ||
688 sysfs_create_link(&c->kobj, &d->kobj, d->name),
689 "Couldn't create device <-> cache set symlinks");
691 clear_bit(BCACHE_DEV_UNLINK_DONE, &d->flags);
694 static void bcache_device_detach(struct bcache_device *d)
696 lockdep_assert_held(&bch_register_lock);
698 if (test_bit(BCACHE_DEV_DETACHING, &d->flags)) {
699 struct uuid_entry *u = d->c->uuids + d->id;
701 SET_UUID_FLASH_ONLY(u, 0);
702 memcpy(u->uuid, invalid_uuid, 16);
703 u->invalidated = cpu_to_le32(get_seconds());
704 bch_uuid_write(d->c);
707 bcache_device_unlink(d);
709 d->c->devices[d->id] = NULL;
710 closure_put(&d->c->caching);
714 static void bcache_device_attach(struct bcache_device *d, struct cache_set *c,
721 closure_get(&c->caching);
724 static void bcache_device_free(struct bcache_device *d)
726 lockdep_assert_held(&bch_register_lock);
728 pr_info("%s stopped", d->disk->disk_name);
731 bcache_device_detach(d);
732 if (d->disk && d->disk->flags & GENHD_FL_UP)
733 del_gendisk(d->disk);
734 if (d->disk && d->disk->queue)
735 blk_cleanup_queue(d->disk->queue);
737 ida_simple_remove(&bcache_minor, d->disk->first_minor);
742 bioset_free(d->bio_split);
743 kvfree(d->full_dirty_stripes);
744 kvfree(d->stripe_sectors_dirty);
746 closure_debug_destroy(&d->cl);
749 static int bcache_device_init(struct bcache_device *d, unsigned block_size,
752 struct request_queue *q;
757 d->stripe_size = 1 << 31;
759 d->nr_stripes = DIV_ROUND_UP_ULL(sectors, d->stripe_size);
761 if (!d->nr_stripes ||
762 d->nr_stripes > INT_MAX ||
763 d->nr_stripes > SIZE_MAX / sizeof(atomic_t)) {
764 pr_err("nr_stripes too large or invalid: %u (start sector beyond end of disk?)",
765 (unsigned)d->nr_stripes);
769 n = d->nr_stripes * sizeof(atomic_t);
770 d->stripe_sectors_dirty = kvzalloc(n, GFP_KERNEL);
771 if (!d->stripe_sectors_dirty)
774 n = BITS_TO_LONGS(d->nr_stripes) * sizeof(unsigned long);
775 d->full_dirty_stripes = kvzalloc(n, GFP_KERNEL);
776 if (!d->full_dirty_stripes)
779 minor = ida_simple_get(&bcache_minor, 0, MINORMASK + 1, GFP_KERNEL);
783 minor *= BCACHE_MINORS;
785 if (!(d->bio_split = bioset_create(4, offsetof(struct bbio, bio),
787 BIOSET_NEED_RESCUER)) ||
788 !(d->disk = alloc_disk(BCACHE_MINORS))) {
789 ida_simple_remove(&bcache_minor, minor);
793 set_capacity(d->disk, sectors);
794 snprintf(d->disk->disk_name, DISK_NAME_LEN, "bcache%i", minor);
796 d->disk->major = bcache_major;
797 d->disk->first_minor = minor;
798 d->disk->fops = &bcache_ops;
799 d->disk->private_data = d;
801 q = blk_alloc_queue(GFP_KERNEL);
805 blk_queue_make_request(q, NULL);
808 q->backing_dev_info->congested_data = d;
809 q->limits.max_hw_sectors = UINT_MAX;
810 q->limits.max_sectors = UINT_MAX;
811 q->limits.max_segment_size = UINT_MAX;
812 q->limits.max_segments = BIO_MAX_PAGES;
813 blk_queue_max_discard_sectors(q, UINT_MAX);
814 q->limits.discard_granularity = 512;
815 q->limits.io_min = block_size;
816 q->limits.logical_block_size = block_size;
817 q->limits.physical_block_size = block_size;
818 set_bit(QUEUE_FLAG_NONROT, &d->disk->queue->queue_flags);
819 clear_bit(QUEUE_FLAG_ADD_RANDOM, &d->disk->queue->queue_flags);
820 set_bit(QUEUE_FLAG_DISCARD, &d->disk->queue->queue_flags);
822 blk_queue_write_cache(q, true, true);
829 static void calc_cached_dev_sectors(struct cache_set *c)
831 uint64_t sectors = 0;
832 struct cached_dev *dc;
834 list_for_each_entry(dc, &c->cached_devs, list)
835 sectors += bdev_sectors(dc->bdev);
837 c->cached_dev_sectors = sectors;
840 void bch_cached_dev_run(struct cached_dev *dc)
842 struct bcache_device *d = &dc->disk;
843 char buf[SB_LABEL_SIZE + 1];
846 kasprintf(GFP_KERNEL, "CACHED_UUID=%pU", dc->sb.uuid),
851 memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
852 buf[SB_LABEL_SIZE] = '\0';
853 env[2] = kasprintf(GFP_KERNEL, "CACHED_LABEL=%s", buf);
855 if (atomic_xchg(&dc->running, 1)) {
862 BDEV_STATE(&dc->sb) != BDEV_STATE_NONE) {
864 closure_init_stack(&cl);
866 SET_BDEV_STATE(&dc->sb, BDEV_STATE_STALE);
867 bch_write_bdev_super(dc, &cl);
872 bd_link_disk_holder(dc->bdev, dc->disk.disk);
873 /* won't show up in the uevent file, use udevadm monitor -e instead
874 * only class / kset properties are persistent */
875 kobject_uevent_env(&disk_to_dev(d->disk)->kobj, KOBJ_CHANGE, env);
879 if (sysfs_create_link(&d->kobj, &disk_to_dev(d->disk)->kobj, "dev") ||
880 sysfs_create_link(&disk_to_dev(d->disk)->kobj, &d->kobj, "bcache"))
881 pr_debug("error creating sysfs link");
884 static void cached_dev_detach_finish(struct work_struct *w)
886 struct cached_dev *dc = container_of(w, struct cached_dev, detach);
887 char buf[BDEVNAME_SIZE];
889 closure_init_stack(&cl);
891 BUG_ON(!test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags));
892 BUG_ON(atomic_read(&dc->count));
894 mutex_lock(&bch_register_lock);
896 cancel_delayed_work_sync(&dc->writeback_rate_update);
897 if (!IS_ERR_OR_NULL(dc->writeback_thread)) {
898 kthread_stop(dc->writeback_thread);
899 dc->writeback_thread = NULL;
902 memset(&dc->sb.set_uuid, 0, 16);
903 SET_BDEV_STATE(&dc->sb, BDEV_STATE_NONE);
905 bch_write_bdev_super(dc, &cl);
908 calc_cached_dev_sectors(dc->disk.c);
909 bcache_device_detach(&dc->disk);
910 list_move(&dc->list, &uncached_devices);
912 clear_bit(BCACHE_DEV_DETACHING, &dc->disk.flags);
913 clear_bit(BCACHE_DEV_UNLINK_DONE, &dc->disk.flags);
915 mutex_unlock(&bch_register_lock);
917 pr_info("Caching disabled for %s", bdevname(dc->bdev, buf));
919 /* Drop ref we took in cached_dev_detach() */
920 closure_put(&dc->disk.cl);
923 void bch_cached_dev_detach(struct cached_dev *dc)
925 lockdep_assert_held(&bch_register_lock);
927 if (test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags))
930 if (test_and_set_bit(BCACHE_DEV_DETACHING, &dc->disk.flags))
934 * Block the device from being closed and freed until we're finished
937 closure_get(&dc->disk.cl);
939 bch_writeback_queue(dc);
943 int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c,
946 uint32_t rtime = cpu_to_le32(get_seconds());
947 struct uuid_entry *u;
948 char buf[BDEVNAME_SIZE];
949 struct cached_dev *exist_dc, *t;
951 bdevname(dc->bdev, buf);
953 if ((set_uuid && memcmp(set_uuid, c->sb.set_uuid, 16)) ||
954 (!set_uuid && memcmp(dc->sb.set_uuid, c->sb.set_uuid, 16)))
958 pr_err("Can't attach %s: already attached", buf);
962 if (test_bit(CACHE_SET_STOPPING, &c->flags)) {
963 pr_err("Can't attach %s: shutting down", buf);
967 if (dc->sb.block_size < c->sb.block_size) {
969 pr_err("Couldn't attach %s: block size less than set's block size",
974 /* Check whether already attached */
975 list_for_each_entry_safe(exist_dc, t, &c->cached_devs, list) {
976 if (!memcmp(dc->sb.uuid, exist_dc->sb.uuid, 16)) {
977 pr_err("Tried to attach %s but duplicate UUID already attached",
984 u = uuid_find(c, dc->sb.uuid);
987 (BDEV_STATE(&dc->sb) == BDEV_STATE_STALE ||
988 BDEV_STATE(&dc->sb) == BDEV_STATE_NONE)) {
989 memcpy(u->uuid, invalid_uuid, 16);
990 u->invalidated = cpu_to_le32(get_seconds());
995 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
996 pr_err("Couldn't find uuid for %s in set", buf);
1000 u = uuid_find_empty(c);
1002 pr_err("Not caching %s, no room for UUID", buf);
1007 /* Deadlocks since we're called via sysfs...
1008 sysfs_remove_file(&dc->kobj, &sysfs_attach);
1011 if (bch_is_zero(u->uuid, 16)) {
1013 closure_init_stack(&cl);
1015 memcpy(u->uuid, dc->sb.uuid, 16);
1016 memcpy(u->label, dc->sb.label, SB_LABEL_SIZE);
1017 u->first_reg = u->last_reg = rtime;
1020 memcpy(dc->sb.set_uuid, c->sb.set_uuid, 16);
1021 SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN);
1023 bch_write_bdev_super(dc, &cl);
1026 u->last_reg = rtime;
1030 bcache_device_attach(&dc->disk, c, u - c->uuids);
1031 list_move(&dc->list, &c->cached_devs);
1032 calc_cached_dev_sectors(c);
1036 * dc->c must be set before dc->count != 0 - paired with the mb in
1039 atomic_set(&dc->count, 1);
1041 /* Block writeback thread, but spawn it */
1042 down_write(&dc->writeback_lock);
1043 if (bch_cached_dev_writeback_start(dc)) {
1044 up_write(&dc->writeback_lock);
1048 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
1049 atomic_set(&dc->has_dirty, 1);
1050 atomic_inc(&dc->count);
1051 bch_writeback_queue(dc);
1054 bch_sectors_dirty_init(&dc->disk);
1056 bch_cached_dev_run(dc);
1057 bcache_device_link(&dc->disk, c, "bdev");
1059 /* Allow the writeback thread to proceed */
1060 up_write(&dc->writeback_lock);
1062 pr_info("Caching %s as %s on set %pU",
1063 bdevname(dc->bdev, buf), dc->disk.disk->disk_name,
1064 dc->disk.c->sb.set_uuid);
1068 void bch_cached_dev_release(struct kobject *kobj)
1070 struct cached_dev *dc = container_of(kobj, struct cached_dev,
1073 module_put(THIS_MODULE);
1076 static void cached_dev_free(struct closure *cl)
1078 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1080 cancel_delayed_work_sync(&dc->writeback_rate_update);
1081 if (!IS_ERR_OR_NULL(dc->writeback_thread))
1082 kthread_stop(dc->writeback_thread);
1083 if (dc->writeback_write_wq)
1084 destroy_workqueue(dc->writeback_write_wq);
1086 mutex_lock(&bch_register_lock);
1088 if (atomic_read(&dc->running))
1089 bd_unlink_disk_holder(dc->bdev, dc->disk.disk);
1090 bcache_device_free(&dc->disk);
1091 list_del(&dc->list);
1093 mutex_unlock(&bch_register_lock);
1095 if (!IS_ERR_OR_NULL(dc->bdev))
1096 blkdev_put(dc->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1098 wake_up(&unregister_wait);
1100 kobject_put(&dc->disk.kobj);
1103 static void cached_dev_flush(struct closure *cl)
1105 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1106 struct bcache_device *d = &dc->disk;
1108 mutex_lock(&bch_register_lock);
1109 bcache_device_unlink(d);
1110 mutex_unlock(&bch_register_lock);
1112 bch_cache_accounting_destroy(&dc->accounting);
1113 kobject_del(&d->kobj);
1115 continue_at(cl, cached_dev_free, system_wq);
1118 static int cached_dev_init(struct cached_dev *dc, unsigned block_size)
1122 struct request_queue *q = bdev_get_queue(dc->bdev);
1124 __module_get(THIS_MODULE);
1125 INIT_LIST_HEAD(&dc->list);
1126 closure_init(&dc->disk.cl, NULL);
1127 set_closure_fn(&dc->disk.cl, cached_dev_flush, system_wq);
1128 kobject_init(&dc->disk.kobj, &bch_cached_dev_ktype);
1129 INIT_WORK(&dc->detach, cached_dev_detach_finish);
1130 sema_init(&dc->sb_write_mutex, 1);
1131 INIT_LIST_HEAD(&dc->io_lru);
1132 spin_lock_init(&dc->io_lock);
1133 bch_cache_accounting_init(&dc->accounting, &dc->disk.cl);
1135 dc->sequential_cutoff = 4 << 20;
1137 for (io = dc->io; io < dc->io + RECENT_IO; io++) {
1138 list_add(&io->lru, &dc->io_lru);
1139 hlist_add_head(&io->hash, dc->io_hash + RECENT_IO);
1142 dc->disk.stripe_size = q->limits.io_opt >> 9;
1144 if (dc->disk.stripe_size)
1145 dc->partial_stripes_expensive =
1146 q->limits.raid_partial_stripes_expensive;
1148 ret = bcache_device_init(&dc->disk, block_size,
1149 dc->bdev->bd_part->nr_sects - dc->sb.data_offset);
1153 set_capacity(dc->disk.disk,
1154 dc->bdev->bd_part->nr_sects - dc->sb.data_offset);
1156 dc->disk.disk->queue->backing_dev_info->ra_pages =
1157 max(dc->disk.disk->queue->backing_dev_info->ra_pages,
1158 q->backing_dev_info->ra_pages);
1160 bch_cached_dev_request_init(dc);
1161 bch_cached_dev_writeback_init(dc);
1165 /* Cached device - bcache superblock */
1167 static void register_bdev(struct cache_sb *sb, struct page *sb_page,
1168 struct block_device *bdev,
1169 struct cached_dev *dc)
1171 char name[BDEVNAME_SIZE];
1172 const char *err = "cannot allocate memory";
1173 struct cache_set *c;
1175 memcpy(&dc->sb, sb, sizeof(struct cache_sb));
1177 dc->bdev->bd_holder = dc;
1179 bio_init(&dc->sb_bio, dc->sb_bio.bi_inline_vecs, 1);
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),
1545 BIOSET_NEED_RESCUER)) ||
1546 !(c->uuids = alloc_bucket_pages(GFP_KERNEL, c)) ||
1547 !(c->moving_gc_wq = alloc_workqueue("bcache_gc",
1548 WQ_MEM_RECLAIM, 0)) ||
1549 bch_journal_alloc(c) ||
1550 bch_btree_cache_alloc(c) ||
1551 bch_open_buckets_alloc(c) ||
1552 bch_bset_sort_state_init(&c->sort, ilog2(c->btree_pages)))
1555 c->congested_read_threshold_us = 2000;
1556 c->congested_write_threshold_us = 20000;
1557 c->error_limit = 8 << IO_ERROR_SHIFT;
1561 bch_cache_set_unregister(c);
1565 static int run_cache_set(struct cache_set *c)
1567 const char *err = "cannot allocate memory";
1568 struct cached_dev *dc, *t;
1573 closure_init_stack(&cl);
1575 for_each_cache(ca, c, i)
1576 c->nbuckets += ca->sb.nbuckets;
1579 if (CACHE_SYNC(&c->sb)) {
1584 err = "cannot allocate memory for journal";
1585 if (bch_journal_read(c, &journal))
1588 pr_debug("btree_journal_read() done");
1590 err = "no journal entries found";
1591 if (list_empty(&journal))
1594 j = &list_entry(journal.prev, struct journal_replay, list)->j;
1596 err = "IO error reading priorities";
1597 for_each_cache(ca, c, i)
1598 prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev]);
1601 * If prio_read() fails it'll call cache_set_error and we'll
1602 * tear everything down right away, but if we perhaps checked
1603 * sooner we could avoid journal replay.
1608 err = "bad btree root";
1609 if (__bch_btree_ptr_invalid(c, k))
1612 err = "error reading btree root";
1613 c->root = bch_btree_node_get(c, NULL, k, j->btree_level, true, NULL);
1614 if (IS_ERR_OR_NULL(c->root))
1617 list_del_init(&c->root->list);
1618 rw_unlock(true, c->root);
1620 err = uuid_read(c, j, &cl);
1624 err = "error in recovery";
1625 if (bch_btree_check(c))
1628 bch_journal_mark(c, &journal);
1629 bch_initial_gc_finish(c);
1630 pr_debug("btree_check() done");
1633 * bcache_journal_next() can't happen sooner, or
1634 * btree_gc_finish() will give spurious errors about last_gc >
1635 * gc_gen - this is a hack but oh well.
1637 bch_journal_next(&c->journal);
1639 err = "error starting allocator thread";
1640 for_each_cache(ca, c, i)
1641 if (bch_cache_allocator_start(ca))
1645 * First place it's safe to allocate: btree_check() and
1646 * btree_gc_finish() have to run before we have buckets to
1647 * allocate, and bch_bucket_alloc_set() might cause a journal
1648 * entry to be written so bcache_journal_next() has to be called
1651 * If the uuids were in the old format we have to rewrite them
1652 * before the next journal entry is written:
1654 if (j->version < BCACHE_JSET_VERSION_UUID)
1657 err = "bcache: replay journal failed";
1658 if (bch_journal_replay(c, &journal))
1661 pr_notice("invalidating existing data");
1663 for_each_cache(ca, c, i) {
1666 ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7,
1667 2, SB_JOURNAL_BUCKETS);
1669 for (j = 0; j < ca->sb.keys; j++)
1670 ca->sb.d[j] = ca->sb.first_bucket + j;
1673 bch_initial_gc_finish(c);
1675 err = "error starting allocator thread";
1676 for_each_cache(ca, c, i)
1677 if (bch_cache_allocator_start(ca))
1680 mutex_lock(&c->bucket_lock);
1681 for_each_cache(ca, c, i)
1683 mutex_unlock(&c->bucket_lock);
1685 err = "cannot allocate new UUID bucket";
1686 if (__uuid_write(c))
1689 err = "cannot allocate new btree root";
1690 c->root = __bch_btree_node_alloc(c, NULL, 0, true, NULL);
1691 if (IS_ERR_OR_NULL(c->root))
1694 mutex_lock(&c->root->write_lock);
1695 bkey_copy_key(&c->root->key, &MAX_KEY);
1696 bch_btree_node_write(c->root, &cl);
1697 mutex_unlock(&c->root->write_lock);
1699 bch_btree_set_root(c->root);
1700 rw_unlock(true, c->root);
1703 * We don't want to write the first journal entry until
1704 * everything is set up - fortunately journal entries won't be
1705 * written until the SET_CACHE_SYNC() here:
1707 SET_CACHE_SYNC(&c->sb, true);
1709 bch_journal_next(&c->journal);
1710 bch_journal_meta(c, &cl);
1713 err = "error starting gc thread";
1714 if (bch_gc_thread_start(c))
1718 c->sb.last_mount = get_seconds();
1719 bcache_write_super(c);
1721 list_for_each_entry_safe(dc, t, &uncached_devices, list)
1722 bch_cached_dev_attach(dc, c, NULL);
1726 set_bit(CACHE_SET_RUNNING, &c->flags);
1730 /* XXX: test this, it's broken */
1731 bch_cache_set_error(c, "%s", err);
1736 static bool can_attach_cache(struct cache *ca, struct cache_set *c)
1738 return ca->sb.block_size == c->sb.block_size &&
1739 ca->sb.bucket_size == c->sb.bucket_size &&
1740 ca->sb.nr_in_set == c->sb.nr_in_set;
1743 static const char *register_cache_set(struct cache *ca)
1746 const char *err = "cannot allocate memory";
1747 struct cache_set *c;
1749 list_for_each_entry(c, &bch_cache_sets, list)
1750 if (!memcmp(c->sb.set_uuid, ca->sb.set_uuid, 16)) {
1751 if (c->cache[ca->sb.nr_this_dev])
1752 return "duplicate cache set member";
1754 if (!can_attach_cache(ca, c))
1755 return "cache sb does not match set";
1757 if (!CACHE_SYNC(&ca->sb))
1758 SET_CACHE_SYNC(&c->sb, false);
1763 c = bch_cache_set_alloc(&ca->sb);
1767 err = "error creating kobject";
1768 if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->sb.set_uuid) ||
1769 kobject_add(&c->internal, &c->kobj, "internal"))
1772 if (bch_cache_accounting_add_kobjs(&c->accounting, &c->kobj))
1775 bch_debug_init_cache_set(c);
1777 list_add(&c->list, &bch_cache_sets);
1779 sprintf(buf, "cache%i", ca->sb.nr_this_dev);
1780 if (sysfs_create_link(&ca->kobj, &c->kobj, "set") ||
1781 sysfs_create_link(&c->kobj, &ca->kobj, buf))
1785 * A special case is both ca->sb.seq and c->sb.seq are 0,
1786 * such condition happens on a new created cache device whose
1787 * super block is never flushed yet. In this case c->sb.version
1788 * and other members should be updated too, otherwise we will
1789 * have a mistaken super block version in cache set.
1791 if (ca->sb.seq > c->sb.seq || c->sb.seq == 0) {
1792 c->sb.version = ca->sb.version;
1793 memcpy(c->sb.set_uuid, ca->sb.set_uuid, 16);
1794 c->sb.flags = ca->sb.flags;
1795 c->sb.seq = ca->sb.seq;
1796 pr_debug("set version = %llu", c->sb.version);
1799 kobject_get(&ca->kobj);
1801 ca->set->cache[ca->sb.nr_this_dev] = ca;
1802 c->cache_by_alloc[c->caches_loaded++] = ca;
1804 if (c->caches_loaded == c->sb.nr_in_set) {
1805 err = "failed to run cache set";
1806 if (run_cache_set(c) < 0)
1812 bch_cache_set_unregister(c);
1818 void bch_cache_release(struct kobject *kobj)
1820 struct cache *ca = container_of(kobj, struct cache, kobj);
1824 BUG_ON(ca->set->cache[ca->sb.nr_this_dev] != ca);
1825 ca->set->cache[ca->sb.nr_this_dev] = NULL;
1828 free_pages((unsigned long) ca->disk_buckets, ilog2(bucket_pages(ca)));
1829 kfree(ca->prio_buckets);
1832 free_heap(&ca->heap);
1833 free_fifo(&ca->free_inc);
1835 for (i = 0; i < RESERVE_NR; i++)
1836 free_fifo(&ca->free[i]);
1838 if (ca->sb_bio.bi_inline_vecs[0].bv_page)
1839 put_page(ca->sb_bio.bi_io_vec[0].bv_page);
1841 if (!IS_ERR_OR_NULL(ca->bdev))
1842 blkdev_put(ca->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1845 module_put(THIS_MODULE);
1848 static int cache_alloc(struct cache *ca)
1851 size_t btree_buckets;
1854 __module_get(THIS_MODULE);
1855 kobject_init(&ca->kobj, &bch_cache_ktype);
1857 bio_init(&ca->journal.bio, ca->journal.bio.bi_inline_vecs, 8);
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, ca->sb_bio.bi_inline_vecs, 1);
1906 ca->sb_bio.bi_io_vec[0].bv_page = sb_page;
1909 if (blk_queue_discard(bdev_get_queue(bdev)))
1910 ca->discard = CACHE_DISCARD(&ca->sb);
1912 ret = cache_alloc(ca);
1914 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1916 err = "cache_alloc(): -ENOMEM";
1918 err = "cache_alloc(): unknown error";
1922 if (kobject_add(&ca->kobj, &part_to_dev(bdev->bd_part)->kobj, "bcache")) {
1923 err = "error calling kobject_add";
1928 mutex_lock(&bch_register_lock);
1929 err = register_cache_set(ca);
1930 mutex_unlock(&bch_register_lock);
1937 pr_info("registered cache device %s", name);
1940 kobject_put(&ca->kobj);
1944 pr_notice("error %s: %s", name, err);
1949 /* Global interfaces/init */
1951 static ssize_t register_bcache(struct kobject *, struct kobj_attribute *,
1952 const char *, size_t);
1954 kobj_attribute_write(register, register_bcache);
1955 kobj_attribute_write(register_quiet, register_bcache);
1957 static bool bch_is_open_backing(struct block_device *bdev) {
1958 struct cache_set *c, *tc;
1959 struct cached_dev *dc, *t;
1961 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
1962 list_for_each_entry_safe(dc, t, &c->cached_devs, list)
1963 if (dc->bdev == bdev)
1965 list_for_each_entry_safe(dc, t, &uncached_devices, list)
1966 if (dc->bdev == bdev)
1971 static bool bch_is_open_cache(struct block_device *bdev) {
1972 struct cache_set *c, *tc;
1976 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
1977 for_each_cache(ca, c, i)
1978 if (ca->bdev == bdev)
1983 static bool bch_is_open(struct block_device *bdev) {
1984 return bch_is_open_cache(bdev) || bch_is_open_backing(bdev);
1987 static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr,
1988 const char *buffer, size_t size)
1991 const char *err = "cannot allocate memory";
1993 struct cache_sb *sb = NULL;
1994 struct block_device *bdev = NULL;
1995 struct page *sb_page = NULL;
1997 if (!try_module_get(THIS_MODULE))
2000 if (!(path = kstrndup(buffer, size, GFP_KERNEL)) ||
2001 !(sb = kmalloc(sizeof(struct cache_sb), GFP_KERNEL)))
2004 err = "failed to open device";
2005 bdev = blkdev_get_by_path(strim(path),
2006 FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2009 if (bdev == ERR_PTR(-EBUSY)) {
2010 bdev = lookup_bdev(strim(path));
2011 mutex_lock(&bch_register_lock);
2012 if (!IS_ERR(bdev) && bch_is_open(bdev))
2013 err = "device already registered";
2015 err = "device busy";
2016 mutex_unlock(&bch_register_lock);
2019 if (attr == &ksysfs_register_quiet)
2025 err = "failed to set blocksize";
2026 if (set_blocksize(bdev, 4096))
2029 err = read_super(sb, bdev, &sb_page);
2033 err = "failed to register device";
2034 if (SB_IS_BDEV(sb)) {
2035 struct cached_dev *dc = kzalloc(sizeof(*dc), GFP_KERNEL);
2039 mutex_lock(&bch_register_lock);
2040 register_bdev(sb, sb_page, bdev, dc);
2041 mutex_unlock(&bch_register_lock);
2043 struct cache *ca = kzalloc(sizeof(*ca), GFP_KERNEL);
2047 if (register_cache(sb, sb_page, bdev, ca) != 0)
2055 module_put(THIS_MODULE);
2059 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2061 pr_info("error %s: %s", path, err);
2066 static int bcache_reboot(struct notifier_block *n, unsigned long code, void *x)
2068 if (code == SYS_DOWN ||
2070 code == SYS_POWER_OFF) {
2072 unsigned long start = jiffies;
2073 bool stopped = false;
2075 struct cache_set *c, *tc;
2076 struct cached_dev *dc, *tdc;
2078 mutex_lock(&bch_register_lock);
2080 if (list_empty(&bch_cache_sets) &&
2081 list_empty(&uncached_devices))
2084 pr_info("Stopping all devices:");
2086 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
2087 bch_cache_set_stop(c);
2089 list_for_each_entry_safe(dc, tdc, &uncached_devices, list)
2090 bcache_device_stop(&dc->disk);
2092 /* What's a condition variable? */
2094 long timeout = start + 2 * HZ - jiffies;
2096 stopped = list_empty(&bch_cache_sets) &&
2097 list_empty(&uncached_devices);
2099 if (timeout < 0 || stopped)
2102 prepare_to_wait(&unregister_wait, &wait,
2103 TASK_UNINTERRUPTIBLE);
2105 mutex_unlock(&bch_register_lock);
2106 schedule_timeout(timeout);
2107 mutex_lock(&bch_register_lock);
2110 finish_wait(&unregister_wait, &wait);
2113 pr_info("All devices stopped");
2115 pr_notice("Timeout waiting for devices to be closed");
2117 mutex_unlock(&bch_register_lock);
2123 static struct notifier_block reboot = {
2124 .notifier_call = bcache_reboot,
2125 .priority = INT_MAX, /* before any real devices */
2128 static void bcache_exit(void)
2133 kobject_put(bcache_kobj);
2135 destroy_workqueue(bcache_wq);
2137 unregister_blkdev(bcache_major, "bcache");
2138 unregister_reboot_notifier(&reboot);
2139 mutex_destroy(&bch_register_lock);
2142 static int __init bcache_init(void)
2144 static const struct attribute *files[] = {
2145 &ksysfs_register.attr,
2146 &ksysfs_register_quiet.attr,
2150 mutex_init(&bch_register_lock);
2151 init_waitqueue_head(&unregister_wait);
2152 register_reboot_notifier(&reboot);
2153 closure_debug_init();
2155 bcache_major = register_blkdev(0, "bcache");
2156 if (bcache_major < 0) {
2157 unregister_reboot_notifier(&reboot);
2158 mutex_destroy(&bch_register_lock);
2159 return bcache_major;
2162 if (!(bcache_wq = alloc_workqueue("bcache", WQ_MEM_RECLAIM, 0)) ||
2163 !(bcache_kobj = kobject_create_and_add("bcache", fs_kobj)) ||
2164 bch_request_init() ||
2165 bch_debug_init(bcache_kobj) ||
2166 sysfs_create_files(bcache_kobj, files))
2175 module_exit(bcache_exit);
2176 module_init(bcache_init);