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->block_size = le16_to_cpu(s->block_size);
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_rw = REQ_SYNC|REQ_META;
216 bio->bi_iter.bi_size = SB_SIZE;
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);
241 submit_bio(REQ_WRITE, bio);
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->bi_bdev = 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_error, "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->bi_bdev = 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_error, 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, unsigned long rw,
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_rw = REQ_SYNC|REQ_META|rw;
352 bio->bi_iter.bi_size = KEY_SIZE(k) << 9;
354 bio->bi_end_io = uuid_endio;
355 bio->bi_private = cl;
356 bch_bio_map(bio, c->uuids);
358 bch_submit_bbio(bio, c, k, i);
364 bch_extent_to_text(buf, sizeof(buf), k);
365 pr_debug("%s UUIDs at %s", rw & REQ_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, 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_WRITE, &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, unsigned long rw)
503 struct closure *cl = &ca->prio;
504 struct bio *bio = bch_bbio_alloc(ca->set);
506 closure_init_stack(cl);
508 bio->bi_iter.bi_sector = bucket * ca->sb.bucket_size;
509 bio->bi_bdev = ca->bdev;
510 bio->bi_rw = REQ_SYNC|REQ_META|rw;
511 bio->bi_iter.bi_size = bucket_bytes(ca);
513 bio->bi_end_io = prio_endio;
514 bio->bi_private = ca;
515 bch_bio_map(bio, ca->disk_buckets);
517 closure_bio_submit(bio, &ca->prio);
521 void bch_prio_write(struct cache *ca)
527 closure_init_stack(&cl);
529 lockdep_assert_held(&ca->set->bucket_lock);
531 ca->disk_buckets->seq++;
533 atomic_long_add(ca->sb.bucket_size * prio_buckets(ca),
534 &ca->meta_sectors_written);
536 //pr_debug("free %zu, free_inc %zu, unused %zu", fifo_used(&ca->free),
537 // fifo_used(&ca->free_inc), fifo_used(&ca->unused));
539 for (i = prio_buckets(ca) - 1; i >= 0; --i) {
541 struct prio_set *p = ca->disk_buckets;
542 struct bucket_disk *d = p->data;
543 struct bucket_disk *end = d + prios_per_bucket(ca);
545 for (b = ca->buckets + i * prios_per_bucket(ca);
546 b < ca->buckets + ca->sb.nbuckets && d < end;
548 d->prio = cpu_to_le16(b->prio);
552 p->next_bucket = ca->prio_buckets[i + 1];
553 p->magic = pset_magic(&ca->sb);
554 p->csum = bch_crc64(&p->magic, bucket_bytes(ca) - 8);
556 bucket = bch_bucket_alloc(ca, RESERVE_PRIO, true);
557 BUG_ON(bucket == -1);
559 mutex_unlock(&ca->set->bucket_lock);
560 prio_io(ca, bucket, REQ_WRITE);
561 mutex_lock(&ca->set->bucket_lock);
563 ca->prio_buckets[i] = bucket;
564 atomic_dec_bug(&ca->buckets[bucket].pin);
567 mutex_unlock(&ca->set->bucket_lock);
569 bch_journal_meta(ca->set, &cl);
572 mutex_lock(&ca->set->bucket_lock);
575 * Don't want the old priorities to get garbage collected until after we
576 * finish writing the new ones, and they're journalled
578 for (i = 0; i < prio_buckets(ca); i++) {
579 if (ca->prio_last_buckets[i])
580 __bch_bucket_free(ca,
581 &ca->buckets[ca->prio_last_buckets[i]]);
583 ca->prio_last_buckets[i] = ca->prio_buckets[i];
587 static void prio_read(struct cache *ca, uint64_t bucket)
589 struct prio_set *p = ca->disk_buckets;
590 struct bucket_disk *d = p->data + prios_per_bucket(ca), *end = d;
592 unsigned bucket_nr = 0;
594 for (b = ca->buckets;
595 b < ca->buckets + ca->sb.nbuckets;
598 ca->prio_buckets[bucket_nr] = bucket;
599 ca->prio_last_buckets[bucket_nr] = bucket;
602 prio_io(ca, bucket, READ_SYNC);
604 if (p->csum != bch_crc64(&p->magic, bucket_bytes(ca) - 8))
605 pr_warn("bad csum reading priorities");
607 if (p->magic != pset_magic(&ca->sb))
608 pr_warn("bad magic reading priorities");
610 bucket = p->next_bucket;
614 b->prio = le16_to_cpu(d->prio);
615 b->gen = b->last_gc = d->gen;
621 static int open_dev(struct block_device *b, fmode_t mode)
623 struct bcache_device *d = b->bd_disk->private_data;
624 if (test_bit(BCACHE_DEV_CLOSING, &d->flags))
631 static void release_dev(struct gendisk *b, fmode_t mode)
633 struct bcache_device *d = b->private_data;
637 static int ioctl_dev(struct block_device *b, fmode_t mode,
638 unsigned int cmd, unsigned long arg)
640 struct bcache_device *d = b->bd_disk->private_data;
641 return d->ioctl(d, mode, cmd, arg);
644 static const struct block_device_operations bcache_ops = {
646 .release = release_dev,
648 .owner = THIS_MODULE,
651 void bcache_device_stop(struct bcache_device *d)
653 if (!test_and_set_bit(BCACHE_DEV_CLOSING, &d->flags))
654 closure_queue(&d->cl);
657 static void bcache_device_unlink(struct bcache_device *d)
659 lockdep_assert_held(&bch_register_lock);
661 if (d->c && !test_and_set_bit(BCACHE_DEV_UNLINK_DONE, &d->flags)) {
665 sysfs_remove_link(&d->c->kobj, d->name);
666 sysfs_remove_link(&d->kobj, "cache");
668 for_each_cache(ca, d->c, i)
669 bd_unlink_disk_holder(ca->bdev, d->disk);
673 static void bcache_device_link(struct bcache_device *d, struct cache_set *c,
679 for_each_cache(ca, d->c, i)
680 bd_link_disk_holder(ca->bdev, d->disk);
682 snprintf(d->name, BCACHEDEVNAME_SIZE,
683 "%s%u", name, d->id);
685 WARN(sysfs_create_link(&d->kobj, &c->kobj, "cache") ||
686 sysfs_create_link(&c->kobj, &d->kobj, d->name),
687 "Couldn't create device <-> cache set symlinks");
689 clear_bit(BCACHE_DEV_UNLINK_DONE, &d->flags);
692 static void bcache_device_detach(struct bcache_device *d)
694 lockdep_assert_held(&bch_register_lock);
696 if (test_bit(BCACHE_DEV_DETACHING, &d->flags)) {
697 struct uuid_entry *u = d->c->uuids + d->id;
699 SET_UUID_FLASH_ONLY(u, 0);
700 memcpy(u->uuid, invalid_uuid, 16);
701 u->invalidated = cpu_to_le32(get_seconds());
702 bch_uuid_write(d->c);
705 bcache_device_unlink(d);
707 d->c->devices[d->id] = NULL;
708 closure_put(&d->c->caching);
712 static void bcache_device_attach(struct bcache_device *d, struct cache_set *c,
719 closure_get(&c->caching);
722 static void bcache_device_free(struct bcache_device *d)
724 lockdep_assert_held(&bch_register_lock);
726 pr_info("%s stopped", d->disk->disk_name);
729 bcache_device_detach(d);
730 if (d->disk && d->disk->flags & GENHD_FL_UP)
731 del_gendisk(d->disk);
732 if (d->disk && d->disk->queue)
733 blk_cleanup_queue(d->disk->queue);
735 ida_simple_remove(&bcache_minor, d->disk->first_minor);
740 bioset_free(d->bio_split);
741 kvfree(d->full_dirty_stripes);
742 kvfree(d->stripe_sectors_dirty);
744 closure_debug_destroy(&d->cl);
747 static int bcache_device_init(struct bcache_device *d, unsigned block_size,
750 struct request_queue *q;
755 d->stripe_size = 1 << 31;
757 d->nr_stripes = DIV_ROUND_UP_ULL(sectors, d->stripe_size);
759 if (!d->nr_stripes ||
760 d->nr_stripes > INT_MAX ||
761 d->nr_stripes > SIZE_MAX / sizeof(atomic_t)) {
762 pr_err("nr_stripes too large");
766 n = d->nr_stripes * sizeof(atomic_t);
767 d->stripe_sectors_dirty = n < PAGE_SIZE << 6
768 ? kzalloc(n, GFP_KERNEL)
770 if (!d->stripe_sectors_dirty)
773 n = BITS_TO_LONGS(d->nr_stripes) * sizeof(unsigned long);
774 d->full_dirty_stripes = n < PAGE_SIZE << 6
775 ? kzalloc(n, GFP_KERNEL)
777 if (!d->full_dirty_stripes)
780 minor = ida_simple_get(&bcache_minor, 0, MINORMASK + 1, GFP_KERNEL);
784 if (!(d->bio_split = bioset_create(4, offsetof(struct bbio, bio))) ||
785 !(d->disk = alloc_disk(1))) {
786 ida_simple_remove(&bcache_minor, minor);
790 set_capacity(d->disk, sectors);
791 snprintf(d->disk->disk_name, DISK_NAME_LEN, "bcache%i", minor);
793 d->disk->major = bcache_major;
794 d->disk->first_minor = minor;
795 d->disk->fops = &bcache_ops;
796 d->disk->private_data = d;
798 q = blk_alloc_queue(GFP_KERNEL);
802 blk_queue_make_request(q, NULL);
805 q->backing_dev_info.congested_data = d;
806 q->limits.max_hw_sectors = UINT_MAX;
807 q->limits.max_sectors = UINT_MAX;
808 q->limits.max_segment_size = UINT_MAX;
809 q->limits.max_segments = BIO_MAX_PAGES;
810 blk_queue_max_discard_sectors(q, UINT_MAX);
811 q->limits.discard_granularity = 512;
812 q->limits.io_min = block_size;
813 q->limits.logical_block_size = block_size;
814 q->limits.physical_block_size = block_size;
815 set_bit(QUEUE_FLAG_NONROT, &d->disk->queue->queue_flags);
816 clear_bit(QUEUE_FLAG_ADD_RANDOM, &d->disk->queue->queue_flags);
817 set_bit(QUEUE_FLAG_DISCARD, &d->disk->queue->queue_flags);
819 blk_queue_flush(q, REQ_FLUSH|REQ_FUA);
826 static void calc_cached_dev_sectors(struct cache_set *c)
828 uint64_t sectors = 0;
829 struct cached_dev *dc;
831 list_for_each_entry(dc, &c->cached_devs, list)
832 sectors += bdev_sectors(dc->bdev);
834 c->cached_dev_sectors = sectors;
837 void bch_cached_dev_run(struct cached_dev *dc)
839 struct bcache_device *d = &dc->disk;
840 char buf[SB_LABEL_SIZE + 1];
843 kasprintf(GFP_KERNEL, "CACHED_UUID=%pU", dc->sb.uuid),
848 memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
849 buf[SB_LABEL_SIZE] = '\0';
850 env[2] = kasprintf(GFP_KERNEL, "CACHED_LABEL=%s", buf);
852 if (atomic_xchg(&dc->running, 1))
856 BDEV_STATE(&dc->sb) != BDEV_STATE_NONE) {
858 closure_init_stack(&cl);
860 SET_BDEV_STATE(&dc->sb, BDEV_STATE_STALE);
861 bch_write_bdev_super(dc, &cl);
866 bd_link_disk_holder(dc->bdev, dc->disk.disk);
867 /* won't show up in the uevent file, use udevadm monitor -e instead
868 * only class / kset properties are persistent */
869 kobject_uevent_env(&disk_to_dev(d->disk)->kobj, KOBJ_CHANGE, env);
873 if (sysfs_create_link(&d->kobj, &disk_to_dev(d->disk)->kobj, "dev") ||
874 sysfs_create_link(&disk_to_dev(d->disk)->kobj, &d->kobj, "bcache"))
875 pr_debug("error creating sysfs link");
878 static void cached_dev_detach_finish(struct work_struct *w)
880 struct cached_dev *dc = container_of(w, struct cached_dev, detach);
881 char buf[BDEVNAME_SIZE];
883 closure_init_stack(&cl);
885 BUG_ON(!test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags));
886 BUG_ON(atomic_read(&dc->count));
888 mutex_lock(&bch_register_lock);
890 memset(&dc->sb.set_uuid, 0, 16);
891 SET_BDEV_STATE(&dc->sb, BDEV_STATE_NONE);
893 bch_write_bdev_super(dc, &cl);
896 bcache_device_detach(&dc->disk);
897 list_move(&dc->list, &uncached_devices);
899 clear_bit(BCACHE_DEV_DETACHING, &dc->disk.flags);
900 clear_bit(BCACHE_DEV_UNLINK_DONE, &dc->disk.flags);
902 mutex_unlock(&bch_register_lock);
904 pr_info("Caching disabled for %s", bdevname(dc->bdev, buf));
906 /* Drop ref we took in cached_dev_detach() */
907 closure_put(&dc->disk.cl);
910 void bch_cached_dev_detach(struct cached_dev *dc)
912 lockdep_assert_held(&bch_register_lock);
914 if (test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags))
917 if (test_and_set_bit(BCACHE_DEV_DETACHING, &dc->disk.flags))
921 * Block the device from being closed and freed until we're finished
924 closure_get(&dc->disk.cl);
926 bch_writeback_queue(dc);
930 int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c)
932 uint32_t rtime = cpu_to_le32(get_seconds());
933 struct uuid_entry *u;
934 char buf[BDEVNAME_SIZE];
936 bdevname(dc->bdev, buf);
938 if (memcmp(dc->sb.set_uuid, c->sb.set_uuid, 16))
942 pr_err("Can't attach %s: already attached", buf);
946 if (test_bit(CACHE_SET_STOPPING, &c->flags)) {
947 pr_err("Can't attach %s: shutting down", buf);
951 if (dc->sb.block_size < c->sb.block_size) {
953 pr_err("Couldn't attach %s: block size less than set's block size",
958 u = uuid_find(c, dc->sb.uuid);
961 (BDEV_STATE(&dc->sb) == BDEV_STATE_STALE ||
962 BDEV_STATE(&dc->sb) == BDEV_STATE_NONE)) {
963 memcpy(u->uuid, invalid_uuid, 16);
964 u->invalidated = cpu_to_le32(get_seconds());
969 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
970 pr_err("Couldn't find uuid for %s in set", buf);
974 u = uuid_find_empty(c);
976 pr_err("Not caching %s, no room for UUID", buf);
981 /* Deadlocks since we're called via sysfs...
982 sysfs_remove_file(&dc->kobj, &sysfs_attach);
985 if (bch_is_zero(u->uuid, 16)) {
987 closure_init_stack(&cl);
989 memcpy(u->uuid, dc->sb.uuid, 16);
990 memcpy(u->label, dc->sb.label, SB_LABEL_SIZE);
991 u->first_reg = u->last_reg = rtime;
994 memcpy(dc->sb.set_uuid, c->sb.set_uuid, 16);
995 SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN);
997 bch_write_bdev_super(dc, &cl);
1000 u->last_reg = rtime;
1004 bcache_device_attach(&dc->disk, c, u - c->uuids);
1005 list_move(&dc->list, &c->cached_devs);
1006 calc_cached_dev_sectors(c);
1010 * dc->c must be set before dc->count != 0 - paired with the mb in
1013 atomic_set(&dc->count, 1);
1015 if (bch_cached_dev_writeback_start(dc))
1018 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
1019 bch_sectors_dirty_init(dc);
1020 atomic_set(&dc->has_dirty, 1);
1021 atomic_inc(&dc->count);
1022 bch_writeback_queue(dc);
1025 bch_cached_dev_run(dc);
1026 bcache_device_link(&dc->disk, c, "bdev");
1028 pr_info("Caching %s as %s on set %pU",
1029 bdevname(dc->bdev, buf), dc->disk.disk->disk_name,
1030 dc->disk.c->sb.set_uuid);
1034 void bch_cached_dev_release(struct kobject *kobj)
1036 struct cached_dev *dc = container_of(kobj, struct cached_dev,
1039 module_put(THIS_MODULE);
1042 static void cached_dev_free(struct closure *cl)
1044 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1046 cancel_delayed_work_sync(&dc->writeback_rate_update);
1047 if (!IS_ERR_OR_NULL(dc->writeback_thread))
1048 kthread_stop(dc->writeback_thread);
1050 mutex_lock(&bch_register_lock);
1052 if (atomic_read(&dc->running))
1053 bd_unlink_disk_holder(dc->bdev, dc->disk.disk);
1054 bcache_device_free(&dc->disk);
1055 list_del(&dc->list);
1057 mutex_unlock(&bch_register_lock);
1059 if (!IS_ERR_OR_NULL(dc->bdev))
1060 blkdev_put(dc->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1062 wake_up(&unregister_wait);
1064 kobject_put(&dc->disk.kobj);
1067 static void cached_dev_flush(struct closure *cl)
1069 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1070 struct bcache_device *d = &dc->disk;
1072 mutex_lock(&bch_register_lock);
1073 bcache_device_unlink(d);
1074 mutex_unlock(&bch_register_lock);
1076 bch_cache_accounting_destroy(&dc->accounting);
1077 kobject_del(&d->kobj);
1079 continue_at(cl, cached_dev_free, system_wq);
1082 static int cached_dev_init(struct cached_dev *dc, unsigned block_size)
1086 struct request_queue *q = bdev_get_queue(dc->bdev);
1088 __module_get(THIS_MODULE);
1089 INIT_LIST_HEAD(&dc->list);
1090 closure_init(&dc->disk.cl, NULL);
1091 set_closure_fn(&dc->disk.cl, cached_dev_flush, system_wq);
1092 kobject_init(&dc->disk.kobj, &bch_cached_dev_ktype);
1093 INIT_WORK(&dc->detach, cached_dev_detach_finish);
1094 sema_init(&dc->sb_write_mutex, 1);
1095 INIT_LIST_HEAD(&dc->io_lru);
1096 spin_lock_init(&dc->io_lock);
1097 bch_cache_accounting_init(&dc->accounting, &dc->disk.cl);
1099 dc->sequential_cutoff = 4 << 20;
1101 for (io = dc->io; io < dc->io + RECENT_IO; io++) {
1102 list_add(&io->lru, &dc->io_lru);
1103 hlist_add_head(&io->hash, dc->io_hash + RECENT_IO);
1106 dc->disk.stripe_size = q->limits.io_opt >> 9;
1108 if (dc->disk.stripe_size)
1109 dc->partial_stripes_expensive =
1110 q->limits.raid_partial_stripes_expensive;
1112 ret = bcache_device_init(&dc->disk, block_size,
1113 dc->bdev->bd_part->nr_sects - dc->sb.data_offset);
1117 set_capacity(dc->disk.disk,
1118 dc->bdev->bd_part->nr_sects - dc->sb.data_offset);
1120 dc->disk.disk->queue->backing_dev_info.ra_pages =
1121 max(dc->disk.disk->queue->backing_dev_info.ra_pages,
1122 q->backing_dev_info.ra_pages);
1124 bch_cached_dev_request_init(dc);
1125 bch_cached_dev_writeback_init(dc);
1129 /* Cached device - bcache superblock */
1131 static void register_bdev(struct cache_sb *sb, struct page *sb_page,
1132 struct block_device *bdev,
1133 struct cached_dev *dc)
1135 char name[BDEVNAME_SIZE];
1136 const char *err = "cannot allocate memory";
1137 struct cache_set *c;
1139 memcpy(&dc->sb, sb, sizeof(struct cache_sb));
1141 dc->bdev->bd_holder = dc;
1143 bio_init(&dc->sb_bio);
1144 dc->sb_bio.bi_max_vecs = 1;
1145 dc->sb_bio.bi_io_vec = dc->sb_bio.bi_inline_vecs;
1146 dc->sb_bio.bi_io_vec[0].bv_page = sb_page;
1149 if (cached_dev_init(dc, sb->block_size << 9))
1152 err = "error creating kobject";
1153 if (kobject_add(&dc->disk.kobj, &part_to_dev(bdev->bd_part)->kobj,
1156 if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj))
1159 pr_info("registered backing device %s", bdevname(bdev, name));
1161 list_add(&dc->list, &uncached_devices);
1162 list_for_each_entry(c, &bch_cache_sets, list)
1163 bch_cached_dev_attach(dc, c);
1165 if (BDEV_STATE(&dc->sb) == BDEV_STATE_NONE ||
1166 BDEV_STATE(&dc->sb) == BDEV_STATE_STALE)
1167 bch_cached_dev_run(dc);
1171 pr_notice("error opening %s: %s", bdevname(bdev, name), err);
1172 bcache_device_stop(&dc->disk);
1175 /* Flash only volumes */
1177 void bch_flash_dev_release(struct kobject *kobj)
1179 struct bcache_device *d = container_of(kobj, struct bcache_device,
1184 static void flash_dev_free(struct closure *cl)
1186 struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1187 mutex_lock(&bch_register_lock);
1188 bcache_device_free(d);
1189 mutex_unlock(&bch_register_lock);
1190 kobject_put(&d->kobj);
1193 static void flash_dev_flush(struct closure *cl)
1195 struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1197 mutex_lock(&bch_register_lock);
1198 bcache_device_unlink(d);
1199 mutex_unlock(&bch_register_lock);
1200 kobject_del(&d->kobj);
1201 continue_at(cl, flash_dev_free, system_wq);
1204 static int flash_dev_run(struct cache_set *c, struct uuid_entry *u)
1206 struct bcache_device *d = kzalloc(sizeof(struct bcache_device),
1211 closure_init(&d->cl, NULL);
1212 set_closure_fn(&d->cl, flash_dev_flush, system_wq);
1214 kobject_init(&d->kobj, &bch_flash_dev_ktype);
1216 if (bcache_device_init(d, block_bytes(c), u->sectors))
1219 bcache_device_attach(d, c, u - c->uuids);
1220 bch_flash_dev_request_init(d);
1223 if (kobject_add(&d->kobj, &disk_to_dev(d->disk)->kobj, "bcache"))
1226 bcache_device_link(d, c, "volume");
1230 kobject_put(&d->kobj);
1234 static int flash_devs_run(struct cache_set *c)
1237 struct uuid_entry *u;
1240 u < c->uuids + c->nr_uuids && !ret;
1242 if (UUID_FLASH_ONLY(u))
1243 ret = flash_dev_run(c, u);
1248 int bch_flash_dev_create(struct cache_set *c, uint64_t size)
1250 struct uuid_entry *u;
1252 if (test_bit(CACHE_SET_STOPPING, &c->flags))
1255 if (!test_bit(CACHE_SET_RUNNING, &c->flags))
1258 u = uuid_find_empty(c);
1260 pr_err("Can't create volume, no room for UUID");
1264 get_random_bytes(u->uuid, 16);
1265 memset(u->label, 0, 32);
1266 u->first_reg = u->last_reg = cpu_to_le32(get_seconds());
1268 SET_UUID_FLASH_ONLY(u, 1);
1269 u->sectors = size >> 9;
1273 return flash_dev_run(c, u);
1279 bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...)
1283 if (c->on_error != ON_ERROR_PANIC &&
1284 test_bit(CACHE_SET_STOPPING, &c->flags))
1287 /* XXX: we can be called from atomic context
1288 acquire_console_sem();
1291 printk(KERN_ERR "bcache: error on %pU: ", c->sb.set_uuid);
1293 va_start(args, fmt);
1297 printk(", disabling caching\n");
1299 if (c->on_error == ON_ERROR_PANIC)
1300 panic("panic forced after error\n");
1302 bch_cache_set_unregister(c);
1306 void bch_cache_set_release(struct kobject *kobj)
1308 struct cache_set *c = container_of(kobj, struct cache_set, kobj);
1310 module_put(THIS_MODULE);
1313 static void cache_set_free(struct closure *cl)
1315 struct cache_set *c = container_of(cl, struct cache_set, cl);
1319 if (!IS_ERR_OR_NULL(c->debug))
1320 debugfs_remove(c->debug);
1322 bch_open_buckets_free(c);
1323 bch_btree_cache_free(c);
1324 bch_journal_free(c);
1326 for_each_cache(ca, c, i)
1329 c->cache[ca->sb.nr_this_dev] = NULL;
1330 kobject_put(&ca->kobj);
1333 bch_bset_sort_state_free(&c->sort);
1334 free_pages((unsigned long) c->uuids, ilog2(bucket_pages(c)));
1336 if (c->moving_gc_wq)
1337 destroy_workqueue(c->moving_gc_wq);
1339 bioset_free(c->bio_split);
1341 mempool_destroy(c->fill_iter);
1343 mempool_destroy(c->bio_meta);
1345 mempool_destroy(c->search);
1348 mutex_lock(&bch_register_lock);
1350 mutex_unlock(&bch_register_lock);
1352 pr_info("Cache set %pU unregistered", c->sb.set_uuid);
1353 wake_up(&unregister_wait);
1355 closure_debug_destroy(&c->cl);
1356 kobject_put(&c->kobj);
1359 static void cache_set_flush(struct closure *cl)
1361 struct cache_set *c = container_of(cl, struct cache_set, caching);
1366 bch_cache_accounting_destroy(&c->accounting);
1368 kobject_put(&c->internal);
1369 kobject_del(&c->kobj);
1372 kthread_stop(c->gc_thread);
1374 if (!IS_ERR_OR_NULL(c->root))
1375 list_add(&c->root->list, &c->btree_cache);
1377 /* Should skip this if we're unregistering because of an error */
1378 list_for_each_entry(b, &c->btree_cache, list) {
1379 mutex_lock(&b->write_lock);
1380 if (btree_node_dirty(b))
1381 __bch_btree_node_write(b, NULL);
1382 mutex_unlock(&b->write_lock);
1385 for_each_cache(ca, c, i)
1386 if (ca->alloc_thread)
1387 kthread_stop(ca->alloc_thread);
1389 if (c->journal.cur) {
1390 cancel_delayed_work_sync(&c->journal.work);
1391 /* flush last journal entry if needed */
1392 c->journal.work.work.func(&c->journal.work.work);
1398 static void __cache_set_unregister(struct closure *cl)
1400 struct cache_set *c = container_of(cl, struct cache_set, caching);
1401 struct cached_dev *dc;
1404 mutex_lock(&bch_register_lock);
1406 for (i = 0; i < c->nr_uuids; i++)
1407 if (c->devices[i]) {
1408 if (!UUID_FLASH_ONLY(&c->uuids[i]) &&
1409 test_bit(CACHE_SET_UNREGISTERING, &c->flags)) {
1410 dc = container_of(c->devices[i],
1411 struct cached_dev, disk);
1412 bch_cached_dev_detach(dc);
1414 bcache_device_stop(c->devices[i]);
1418 mutex_unlock(&bch_register_lock);
1420 continue_at(cl, cache_set_flush, system_wq);
1423 void bch_cache_set_stop(struct cache_set *c)
1425 if (!test_and_set_bit(CACHE_SET_STOPPING, &c->flags))
1426 closure_queue(&c->caching);
1429 void bch_cache_set_unregister(struct cache_set *c)
1431 set_bit(CACHE_SET_UNREGISTERING, &c->flags);
1432 bch_cache_set_stop(c);
1435 #define alloc_bucket_pages(gfp, c) \
1436 ((void *) __get_free_pages(__GFP_ZERO|gfp, ilog2(bucket_pages(c))))
1438 struct cache_set *bch_cache_set_alloc(struct cache_sb *sb)
1441 struct cache_set *c = kzalloc(sizeof(struct cache_set), GFP_KERNEL);
1445 __module_get(THIS_MODULE);
1446 closure_init(&c->cl, NULL);
1447 set_closure_fn(&c->cl, cache_set_free, system_wq);
1449 closure_init(&c->caching, &c->cl);
1450 set_closure_fn(&c->caching, __cache_set_unregister, system_wq);
1452 /* Maybe create continue_at_noreturn() and use it here? */
1453 closure_set_stopped(&c->cl);
1454 closure_put(&c->cl);
1456 kobject_init(&c->kobj, &bch_cache_set_ktype);
1457 kobject_init(&c->internal, &bch_cache_set_internal_ktype);
1459 bch_cache_accounting_init(&c->accounting, &c->cl);
1461 memcpy(c->sb.set_uuid, sb->set_uuid, 16);
1462 c->sb.block_size = sb->block_size;
1463 c->sb.bucket_size = sb->bucket_size;
1464 c->sb.nr_in_set = sb->nr_in_set;
1465 c->sb.last_mount = sb->last_mount;
1466 c->bucket_bits = ilog2(sb->bucket_size);
1467 c->block_bits = ilog2(sb->block_size);
1468 c->nr_uuids = bucket_bytes(c) / sizeof(struct uuid_entry);
1470 c->btree_pages = bucket_pages(c);
1471 if (c->btree_pages > BTREE_MAX_PAGES)
1472 c->btree_pages = max_t(int, c->btree_pages / 4,
1475 sema_init(&c->sb_write_mutex, 1);
1476 mutex_init(&c->bucket_lock);
1477 init_waitqueue_head(&c->btree_cache_wait);
1478 init_waitqueue_head(&c->bucket_wait);
1479 sema_init(&c->uuid_write_mutex, 1);
1481 spin_lock_init(&c->btree_gc_time.lock);
1482 spin_lock_init(&c->btree_split_time.lock);
1483 spin_lock_init(&c->btree_read_time.lock);
1485 bch_moving_init_cache_set(c);
1487 INIT_LIST_HEAD(&c->list);
1488 INIT_LIST_HEAD(&c->cached_devs);
1489 INIT_LIST_HEAD(&c->btree_cache);
1490 INIT_LIST_HEAD(&c->btree_cache_freeable);
1491 INIT_LIST_HEAD(&c->btree_cache_freed);
1492 INIT_LIST_HEAD(&c->data_buckets);
1494 c->search = mempool_create_slab_pool(32, bch_search_cache);
1498 iter_size = (sb->bucket_size / sb->block_size + 1) *
1499 sizeof(struct btree_iter_set);
1501 if (!(c->devices = kzalloc(c->nr_uuids * sizeof(void *), GFP_KERNEL)) ||
1502 !(c->bio_meta = mempool_create_kmalloc_pool(2,
1503 sizeof(struct bbio) + sizeof(struct bio_vec) *
1504 bucket_pages(c))) ||
1505 !(c->fill_iter = mempool_create_kmalloc_pool(1, iter_size)) ||
1506 !(c->bio_split = bioset_create(4, offsetof(struct bbio, bio))) ||
1507 !(c->uuids = alloc_bucket_pages(GFP_KERNEL, c)) ||
1508 !(c->moving_gc_wq = create_workqueue("bcache_gc")) ||
1509 bch_journal_alloc(c) ||
1510 bch_btree_cache_alloc(c) ||
1511 bch_open_buckets_alloc(c) ||
1512 bch_bset_sort_state_init(&c->sort, ilog2(c->btree_pages)))
1515 c->congested_read_threshold_us = 2000;
1516 c->congested_write_threshold_us = 20000;
1517 c->error_limit = 8 << IO_ERROR_SHIFT;
1521 bch_cache_set_unregister(c);
1525 static void run_cache_set(struct cache_set *c)
1527 const char *err = "cannot allocate memory";
1528 struct cached_dev *dc, *t;
1533 closure_init_stack(&cl);
1535 for_each_cache(ca, c, i)
1536 c->nbuckets += ca->sb.nbuckets;
1538 if (CACHE_SYNC(&c->sb)) {
1543 err = "cannot allocate memory for journal";
1544 if (bch_journal_read(c, &journal))
1547 pr_debug("btree_journal_read() done");
1549 err = "no journal entries found";
1550 if (list_empty(&journal))
1553 j = &list_entry(journal.prev, struct journal_replay, list)->j;
1555 err = "IO error reading priorities";
1556 for_each_cache(ca, c, i)
1557 prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev]);
1560 * If prio_read() fails it'll call cache_set_error and we'll
1561 * tear everything down right away, but if we perhaps checked
1562 * sooner we could avoid journal replay.
1567 err = "bad btree root";
1568 if (__bch_btree_ptr_invalid(c, k))
1571 err = "error reading btree root";
1572 c->root = bch_btree_node_get(c, NULL, k, j->btree_level, true, NULL);
1573 if (IS_ERR_OR_NULL(c->root))
1576 list_del_init(&c->root->list);
1577 rw_unlock(true, c->root);
1579 err = uuid_read(c, j, &cl);
1583 err = "error in recovery";
1584 if (bch_btree_check(c))
1587 bch_journal_mark(c, &journal);
1588 bch_initial_gc_finish(c);
1589 pr_debug("btree_check() done");
1592 * bcache_journal_next() can't happen sooner, or
1593 * btree_gc_finish() will give spurious errors about last_gc >
1594 * gc_gen - this is a hack but oh well.
1596 bch_journal_next(&c->journal);
1598 err = "error starting allocator thread";
1599 for_each_cache(ca, c, i)
1600 if (bch_cache_allocator_start(ca))
1604 * First place it's safe to allocate: btree_check() and
1605 * btree_gc_finish() have to run before we have buckets to
1606 * allocate, and bch_bucket_alloc_set() might cause a journal
1607 * entry to be written so bcache_journal_next() has to be called
1610 * If the uuids were in the old format we have to rewrite them
1611 * before the next journal entry is written:
1613 if (j->version < BCACHE_JSET_VERSION_UUID)
1616 bch_journal_replay(c, &journal);
1618 pr_notice("invalidating existing data");
1620 for_each_cache(ca, c, i) {
1623 ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7,
1624 2, SB_JOURNAL_BUCKETS);
1626 for (j = 0; j < ca->sb.keys; j++)
1627 ca->sb.d[j] = ca->sb.first_bucket + j;
1630 bch_initial_gc_finish(c);
1632 err = "error starting allocator thread";
1633 for_each_cache(ca, c, i)
1634 if (bch_cache_allocator_start(ca))
1637 mutex_lock(&c->bucket_lock);
1638 for_each_cache(ca, c, i)
1640 mutex_unlock(&c->bucket_lock);
1642 err = "cannot allocate new UUID bucket";
1643 if (__uuid_write(c))
1646 err = "cannot allocate new btree root";
1647 c->root = __bch_btree_node_alloc(c, NULL, 0, true, NULL);
1648 if (IS_ERR_OR_NULL(c->root))
1651 mutex_lock(&c->root->write_lock);
1652 bkey_copy_key(&c->root->key, &MAX_KEY);
1653 bch_btree_node_write(c->root, &cl);
1654 mutex_unlock(&c->root->write_lock);
1656 bch_btree_set_root(c->root);
1657 rw_unlock(true, c->root);
1660 * We don't want to write the first journal entry until
1661 * everything is set up - fortunately journal entries won't be
1662 * written until the SET_CACHE_SYNC() here:
1664 SET_CACHE_SYNC(&c->sb, true);
1666 bch_journal_next(&c->journal);
1667 bch_journal_meta(c, &cl);
1670 err = "error starting gc thread";
1671 if (bch_gc_thread_start(c))
1675 c->sb.last_mount = get_seconds();
1676 bcache_write_super(c);
1678 list_for_each_entry_safe(dc, t, &uncached_devices, list)
1679 bch_cached_dev_attach(dc, c);
1683 set_bit(CACHE_SET_RUNNING, &c->flags);
1687 /* XXX: test this, it's broken */
1688 bch_cache_set_error(c, "%s", err);
1691 static bool can_attach_cache(struct cache *ca, struct cache_set *c)
1693 return ca->sb.block_size == c->sb.block_size &&
1694 ca->sb.bucket_size == c->sb.bucket_size &&
1695 ca->sb.nr_in_set == c->sb.nr_in_set;
1698 static const char *register_cache_set(struct cache *ca)
1701 const char *err = "cannot allocate memory";
1702 struct cache_set *c;
1704 list_for_each_entry(c, &bch_cache_sets, list)
1705 if (!memcmp(c->sb.set_uuid, ca->sb.set_uuid, 16)) {
1706 if (c->cache[ca->sb.nr_this_dev])
1707 return "duplicate cache set member";
1709 if (!can_attach_cache(ca, c))
1710 return "cache sb does not match set";
1712 if (!CACHE_SYNC(&ca->sb))
1713 SET_CACHE_SYNC(&c->sb, false);
1718 c = bch_cache_set_alloc(&ca->sb);
1722 err = "error creating kobject";
1723 if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->sb.set_uuid) ||
1724 kobject_add(&c->internal, &c->kobj, "internal"))
1727 if (bch_cache_accounting_add_kobjs(&c->accounting, &c->kobj))
1730 bch_debug_init_cache_set(c);
1732 list_add(&c->list, &bch_cache_sets);
1734 sprintf(buf, "cache%i", ca->sb.nr_this_dev);
1735 if (sysfs_create_link(&ca->kobj, &c->kobj, "set") ||
1736 sysfs_create_link(&c->kobj, &ca->kobj, buf))
1739 if (ca->sb.seq > c->sb.seq) {
1740 c->sb.version = ca->sb.version;
1741 memcpy(c->sb.set_uuid, ca->sb.set_uuid, 16);
1742 c->sb.flags = ca->sb.flags;
1743 c->sb.seq = ca->sb.seq;
1744 pr_debug("set version = %llu", c->sb.version);
1747 kobject_get(&ca->kobj);
1749 ca->set->cache[ca->sb.nr_this_dev] = ca;
1750 c->cache_by_alloc[c->caches_loaded++] = ca;
1752 if (c->caches_loaded == c->sb.nr_in_set)
1757 bch_cache_set_unregister(c);
1763 void bch_cache_release(struct kobject *kobj)
1765 struct cache *ca = container_of(kobj, struct cache, kobj);
1769 BUG_ON(ca->set->cache[ca->sb.nr_this_dev] != ca);
1770 ca->set->cache[ca->sb.nr_this_dev] = NULL;
1773 free_pages((unsigned long) ca->disk_buckets, ilog2(bucket_pages(ca)));
1774 kfree(ca->prio_buckets);
1777 free_heap(&ca->heap);
1778 free_fifo(&ca->free_inc);
1780 for (i = 0; i < RESERVE_NR; i++)
1781 free_fifo(&ca->free[i]);
1783 if (ca->sb_bio.bi_inline_vecs[0].bv_page)
1784 put_page(ca->sb_bio.bi_io_vec[0].bv_page);
1786 if (!IS_ERR_OR_NULL(ca->bdev))
1787 blkdev_put(ca->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1790 module_put(THIS_MODULE);
1793 static int cache_alloc(struct cache_sb *sb, struct cache *ca)
1798 __module_get(THIS_MODULE);
1799 kobject_init(&ca->kobj, &bch_cache_ktype);
1801 bio_init(&ca->journal.bio);
1802 ca->journal.bio.bi_max_vecs = 8;
1803 ca->journal.bio.bi_io_vec = ca->journal.bio.bi_inline_vecs;
1805 free = roundup_pow_of_two(ca->sb.nbuckets) >> 10;
1807 if (!init_fifo(&ca->free[RESERVE_BTREE], 8, GFP_KERNEL) ||
1808 !init_fifo(&ca->free[RESERVE_PRIO], prio_buckets(ca), GFP_KERNEL) ||
1809 !init_fifo(&ca->free[RESERVE_MOVINGGC], free, GFP_KERNEL) ||
1810 !init_fifo(&ca->free[RESERVE_NONE], free, GFP_KERNEL) ||
1811 !init_fifo(&ca->free_inc, free << 2, GFP_KERNEL) ||
1812 !init_heap(&ca->heap, free << 3, GFP_KERNEL) ||
1813 !(ca->buckets = vzalloc(sizeof(struct bucket) *
1814 ca->sb.nbuckets)) ||
1815 !(ca->prio_buckets = kzalloc(sizeof(uint64_t) * prio_buckets(ca) *
1817 !(ca->disk_buckets = alloc_bucket_pages(GFP_KERNEL, ca)))
1820 ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca);
1822 for_each_bucket(b, ca)
1823 atomic_set(&b->pin, 0);
1828 static void register_cache(struct cache_sb *sb, struct page *sb_page,
1829 struct block_device *bdev, struct cache *ca)
1831 char name[BDEVNAME_SIZE];
1832 const char *err = "cannot allocate memory";
1834 memcpy(&ca->sb, sb, sizeof(struct cache_sb));
1836 ca->bdev->bd_holder = ca;
1838 bio_init(&ca->sb_bio);
1839 ca->sb_bio.bi_max_vecs = 1;
1840 ca->sb_bio.bi_io_vec = ca->sb_bio.bi_inline_vecs;
1841 ca->sb_bio.bi_io_vec[0].bv_page = sb_page;
1844 if (blk_queue_discard(bdev_get_queue(ca->bdev)))
1845 ca->discard = CACHE_DISCARD(&ca->sb);
1847 if (cache_alloc(sb, ca) != 0)
1850 err = "error creating kobject";
1851 if (kobject_add(&ca->kobj, &part_to_dev(bdev->bd_part)->kobj, "bcache"))
1854 mutex_lock(&bch_register_lock);
1855 err = register_cache_set(ca);
1856 mutex_unlock(&bch_register_lock);
1861 pr_info("registered cache device %s", bdevname(bdev, name));
1863 kobject_put(&ca->kobj);
1866 pr_notice("error opening %s: %s", bdevname(bdev, name), err);
1870 /* Global interfaces/init */
1872 static ssize_t register_bcache(struct kobject *, struct kobj_attribute *,
1873 const char *, size_t);
1875 kobj_attribute_write(register, register_bcache);
1876 kobj_attribute_write(register_quiet, register_bcache);
1878 static bool bch_is_open_backing(struct block_device *bdev) {
1879 struct cache_set *c, *tc;
1880 struct cached_dev *dc, *t;
1882 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
1883 list_for_each_entry_safe(dc, t, &c->cached_devs, list)
1884 if (dc->bdev == bdev)
1886 list_for_each_entry_safe(dc, t, &uncached_devices, list)
1887 if (dc->bdev == bdev)
1892 static bool bch_is_open_cache(struct block_device *bdev) {
1893 struct cache_set *c, *tc;
1897 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
1898 for_each_cache(ca, c, i)
1899 if (ca->bdev == bdev)
1904 static bool bch_is_open(struct block_device *bdev) {
1905 return bch_is_open_cache(bdev) || bch_is_open_backing(bdev);
1908 static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr,
1909 const char *buffer, size_t size)
1912 const char *err = "cannot allocate memory";
1914 struct cache_sb *sb = NULL;
1915 struct block_device *bdev = NULL;
1916 struct page *sb_page = NULL;
1918 if (!try_module_get(THIS_MODULE))
1921 if (!(path = kstrndup(buffer, size, GFP_KERNEL)) ||
1922 !(sb = kmalloc(sizeof(struct cache_sb), GFP_KERNEL)))
1925 err = "failed to open device";
1926 bdev = blkdev_get_by_path(strim(path),
1927 FMODE_READ|FMODE_WRITE|FMODE_EXCL,
1930 if (bdev == ERR_PTR(-EBUSY)) {
1931 bdev = lookup_bdev(strim(path));
1932 mutex_lock(&bch_register_lock);
1933 if (!IS_ERR(bdev) && bch_is_open(bdev))
1934 err = "device already registered";
1936 err = "device busy";
1937 mutex_unlock(&bch_register_lock);
1942 err = "failed to set blocksize";
1943 if (set_blocksize(bdev, 4096))
1946 err = read_super(sb, bdev, &sb_page);
1950 if (SB_IS_BDEV(sb)) {
1951 struct cached_dev *dc = kzalloc(sizeof(*dc), GFP_KERNEL);
1955 mutex_lock(&bch_register_lock);
1956 register_bdev(sb, sb_page, bdev, dc);
1957 mutex_unlock(&bch_register_lock);
1959 struct cache *ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1963 register_cache(sb, sb_page, bdev, ca);
1970 module_put(THIS_MODULE);
1974 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1976 if (attr != &ksysfs_register_quiet)
1977 pr_info("error opening %s: %s", path, err);
1982 static int bcache_reboot(struct notifier_block *n, unsigned long code, void *x)
1984 if (code == SYS_DOWN ||
1986 code == SYS_POWER_OFF) {
1988 unsigned long start = jiffies;
1989 bool stopped = false;
1991 struct cache_set *c, *tc;
1992 struct cached_dev *dc, *tdc;
1994 mutex_lock(&bch_register_lock);
1996 if (list_empty(&bch_cache_sets) &&
1997 list_empty(&uncached_devices))
2000 pr_info("Stopping all devices:");
2002 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
2003 bch_cache_set_stop(c);
2005 list_for_each_entry_safe(dc, tdc, &uncached_devices, list)
2006 bcache_device_stop(&dc->disk);
2008 /* What's a condition variable? */
2010 long timeout = start + 2 * HZ - jiffies;
2012 stopped = list_empty(&bch_cache_sets) &&
2013 list_empty(&uncached_devices);
2015 if (timeout < 0 || stopped)
2018 prepare_to_wait(&unregister_wait, &wait,
2019 TASK_UNINTERRUPTIBLE);
2021 mutex_unlock(&bch_register_lock);
2022 schedule_timeout(timeout);
2023 mutex_lock(&bch_register_lock);
2026 finish_wait(&unregister_wait, &wait);
2029 pr_info("All devices stopped");
2031 pr_notice("Timeout waiting for devices to be closed");
2033 mutex_unlock(&bch_register_lock);
2039 static struct notifier_block reboot = {
2040 .notifier_call = bcache_reboot,
2041 .priority = INT_MAX, /* before any real devices */
2044 static void bcache_exit(void)
2049 kobject_put(bcache_kobj);
2051 destroy_workqueue(bcache_wq);
2053 unregister_blkdev(bcache_major, "bcache");
2054 unregister_reboot_notifier(&reboot);
2057 static int __init bcache_init(void)
2059 static const struct attribute *files[] = {
2060 &ksysfs_register.attr,
2061 &ksysfs_register_quiet.attr,
2065 mutex_init(&bch_register_lock);
2066 init_waitqueue_head(&unregister_wait);
2067 register_reboot_notifier(&reboot);
2068 closure_debug_init();
2070 bcache_major = register_blkdev(0, "bcache");
2071 if (bcache_major < 0)
2072 return bcache_major;
2074 if (!(bcache_wq = create_workqueue("bcache")) ||
2075 !(bcache_kobj = kobject_create_and_add("bcache", fs_kobj)) ||
2076 sysfs_create_files(bcache_kobj, files) ||
2077 bch_request_init() ||
2078 bch_debug_init(bcache_kobj))
2087 module_exit(bcache_exit);
2088 module_init(bcache_init);