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.
13 #include "writeback.h"
15 #include <linux/blkdev.h>
16 #include <linux/buffer_head.h>
17 #include <linux/debugfs.h>
18 #include <linux/genhd.h>
19 #include <linux/kthread.h>
20 #include <linux/module.h>
21 #include <linux/random.h>
22 #include <linux/reboot.h>
23 #include <linux/sysfs.h>
25 MODULE_LICENSE("GPL");
26 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
28 static const char bcache_magic
[] = {
29 0xc6, 0x85, 0x73, 0xf6, 0x4e, 0x1a, 0x45, 0xca,
30 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81
33 static const char invalid_uuid
[] = {
34 0xa0, 0x3e, 0xf8, 0xed, 0x3e, 0xe1, 0xb8, 0x78,
35 0xc8, 0x50, 0xfc, 0x5e, 0xcb, 0x16, 0xcd, 0x99
38 /* Default is -1; we skip past it for struct cached_dev's cache mode */
39 const char * const bch_cache_modes
[] = {
48 static struct kobject
*bcache_kobj
;
49 struct mutex bch_register_lock
;
50 LIST_HEAD(bch_cache_sets
);
51 static LIST_HEAD(uncached_devices
);
53 static int bcache_major
, bcache_minor
;
54 static wait_queue_head_t unregister_wait
;
55 struct workqueue_struct
*bcache_wq
;
57 #define BTREE_MAX_PAGES (256 * 1024 / PAGE_SIZE)
59 static void bio_split_pool_free(struct bio_split_pool
*p
)
61 if (p
->bio_split_hook
)
62 mempool_destroy(p
->bio_split_hook
);
65 bioset_free(p
->bio_split
);
68 static int bio_split_pool_init(struct bio_split_pool
*p
)
70 p
->bio_split
= bioset_create(4, 0);
74 p
->bio_split_hook
= mempool_create_kmalloc_pool(4,
75 sizeof(struct bio_split_hook
));
76 if (!p
->bio_split_hook
)
84 static const char *read_super(struct cache_sb
*sb
, struct block_device
*bdev
,
89 struct buffer_head
*bh
= __bread(bdev
, 1, SB_SIZE
);
95 s
= (struct cache_sb
*) bh
->b_data
;
97 sb
->offset
= le64_to_cpu(s
->offset
);
98 sb
->version
= le64_to_cpu(s
->version
);
100 memcpy(sb
->magic
, s
->magic
, 16);
101 memcpy(sb
->uuid
, s
->uuid
, 16);
102 memcpy(sb
->set_uuid
, s
->set_uuid
, 16);
103 memcpy(sb
->label
, s
->label
, SB_LABEL_SIZE
);
105 sb
->flags
= le64_to_cpu(s
->flags
);
106 sb
->seq
= le64_to_cpu(s
->seq
);
107 sb
->last_mount
= le32_to_cpu(s
->last_mount
);
108 sb
->first_bucket
= le16_to_cpu(s
->first_bucket
);
109 sb
->keys
= le16_to_cpu(s
->keys
);
111 for (i
= 0; i
< SB_JOURNAL_BUCKETS
; i
++)
112 sb
->d
[i
] = le64_to_cpu(s
->d
[i
]);
114 pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u",
115 sb
->version
, sb
->flags
, sb
->seq
, sb
->keys
);
117 err
= "Not a bcache superblock";
118 if (sb
->offset
!= SB_SECTOR
)
121 if (memcmp(sb
->magic
, bcache_magic
, 16))
124 err
= "Too many journal buckets";
125 if (sb
->keys
> SB_JOURNAL_BUCKETS
)
128 err
= "Bad checksum";
129 if (s
->csum
!= csum_set(s
))
133 if (bch_is_zero(sb
->uuid
, 16))
136 sb
->block_size
= le16_to_cpu(s
->block_size
);
138 err
= "Superblock block size smaller than device block size";
139 if (sb
->block_size
<< 9 < bdev_logical_block_size(bdev
))
142 switch (sb
->version
) {
143 case BCACHE_SB_VERSION_BDEV
:
144 sb
->data_offset
= BDEV_DATA_START_DEFAULT
;
146 case BCACHE_SB_VERSION_BDEV_WITH_OFFSET
:
147 sb
->data_offset
= le64_to_cpu(s
->data_offset
);
149 err
= "Bad data offset";
150 if (sb
->data_offset
< BDEV_DATA_START_DEFAULT
)
154 case BCACHE_SB_VERSION_CDEV
:
155 case BCACHE_SB_VERSION_CDEV_WITH_UUID
:
156 sb
->nbuckets
= le64_to_cpu(s
->nbuckets
);
157 sb
->block_size
= le16_to_cpu(s
->block_size
);
158 sb
->bucket_size
= le16_to_cpu(s
->bucket_size
);
160 sb
->nr_in_set
= le16_to_cpu(s
->nr_in_set
);
161 sb
->nr_this_dev
= le16_to_cpu(s
->nr_this_dev
);
163 err
= "Too many buckets";
164 if (sb
->nbuckets
> LONG_MAX
)
167 err
= "Not enough buckets";
168 if (sb
->nbuckets
< 1 << 7)
171 err
= "Bad block/bucket size";
172 if (!is_power_of_2(sb
->block_size
) ||
173 sb
->block_size
> PAGE_SECTORS
||
174 !is_power_of_2(sb
->bucket_size
) ||
175 sb
->bucket_size
< PAGE_SECTORS
)
178 err
= "Invalid superblock: device too small";
179 if (get_capacity(bdev
->bd_disk
) < sb
->bucket_size
* sb
->nbuckets
)
183 if (bch_is_zero(sb
->set_uuid
, 16))
186 err
= "Bad cache device number in set";
187 if (!sb
->nr_in_set
||
188 sb
->nr_in_set
<= sb
->nr_this_dev
||
189 sb
->nr_in_set
> MAX_CACHES_PER_SET
)
192 err
= "Journal buckets not sequential";
193 for (i
= 0; i
< sb
->keys
; i
++)
194 if (sb
->d
[i
] != sb
->first_bucket
+ i
)
197 err
= "Too many journal buckets";
198 if (sb
->first_bucket
+ sb
->keys
> sb
->nbuckets
)
201 err
= "Invalid superblock: first bucket comes before end of super";
202 if (sb
->first_bucket
* sb
->bucket_size
< 16)
207 err
= "Unsupported superblock version";
211 sb
->last_mount
= get_seconds();
214 get_page(bh
->b_page
);
221 static void write_bdev_super_endio(struct bio
*bio
, int error
)
223 struct cached_dev
*dc
= bio
->bi_private
;
224 /* XXX: error checking */
226 closure_put(&dc
->sb_write
.cl
);
229 static void __write_super(struct cache_sb
*sb
, struct bio
*bio
)
231 struct cache_sb
*out
= page_address(bio
->bi_io_vec
[0].bv_page
);
234 bio
->bi_sector
= SB_SECTOR
;
235 bio
->bi_rw
= REQ_SYNC
|REQ_META
;
236 bio
->bi_size
= SB_SIZE
;
237 bch_bio_map(bio
, NULL
);
239 out
->offset
= cpu_to_le64(sb
->offset
);
240 out
->version
= cpu_to_le64(sb
->version
);
242 memcpy(out
->uuid
, sb
->uuid
, 16);
243 memcpy(out
->set_uuid
, sb
->set_uuid
, 16);
244 memcpy(out
->label
, sb
->label
, SB_LABEL_SIZE
);
246 out
->flags
= cpu_to_le64(sb
->flags
);
247 out
->seq
= cpu_to_le64(sb
->seq
);
249 out
->last_mount
= cpu_to_le32(sb
->last_mount
);
250 out
->first_bucket
= cpu_to_le16(sb
->first_bucket
);
251 out
->keys
= cpu_to_le16(sb
->keys
);
253 for (i
= 0; i
< sb
->keys
; i
++)
254 out
->d
[i
] = cpu_to_le64(sb
->d
[i
]);
256 out
->csum
= csum_set(out
);
258 pr_debug("ver %llu, flags %llu, seq %llu",
259 sb
->version
, sb
->flags
, sb
->seq
);
261 submit_bio(REQ_WRITE
, bio
);
264 void bch_write_bdev_super(struct cached_dev
*dc
, struct closure
*parent
)
266 struct closure
*cl
= &dc
->sb_write
.cl
;
267 struct bio
*bio
= &dc
->sb_bio
;
269 closure_lock(&dc
->sb_write
, parent
);
272 bio
->bi_bdev
= dc
->bdev
;
273 bio
->bi_end_io
= write_bdev_super_endio
;
274 bio
->bi_private
= dc
;
277 __write_super(&dc
->sb
, bio
);
282 static void write_super_endio(struct bio
*bio
, int error
)
284 struct cache
*ca
= bio
->bi_private
;
286 bch_count_io_errors(ca
, error
, "writing superblock");
287 closure_put(&ca
->set
->sb_write
.cl
);
290 void bcache_write_super(struct cache_set
*c
)
292 struct closure
*cl
= &c
->sb_write
.cl
;
296 closure_lock(&c
->sb_write
, &c
->cl
);
300 for_each_cache(ca
, c
, i
) {
301 struct bio
*bio
= &ca
->sb_bio
;
303 ca
->sb
.version
= BCACHE_SB_VERSION_CDEV_WITH_UUID
;
304 ca
->sb
.seq
= c
->sb
.seq
;
305 ca
->sb
.last_mount
= c
->sb
.last_mount
;
307 SET_CACHE_SYNC(&ca
->sb
, CACHE_SYNC(&c
->sb
));
310 bio
->bi_bdev
= ca
->bdev
;
311 bio
->bi_end_io
= write_super_endio
;
312 bio
->bi_private
= ca
;
315 __write_super(&ca
->sb
, bio
);
323 static void uuid_endio(struct bio
*bio
, int error
)
325 struct closure
*cl
= bio
->bi_private
;
326 struct cache_set
*c
= container_of(cl
, struct cache_set
, uuid_write
.cl
);
328 cache_set_err_on(error
, c
, "accessing uuids");
329 bch_bbio_free(bio
, c
);
333 static void uuid_io(struct cache_set
*c
, unsigned long rw
,
334 struct bkey
*k
, struct closure
*parent
)
336 struct closure
*cl
= &c
->uuid_write
.cl
;
337 struct uuid_entry
*u
;
342 closure_lock(&c
->uuid_write
, parent
);
344 for (i
= 0; i
< KEY_PTRS(k
); i
++) {
345 struct bio
*bio
= bch_bbio_alloc(c
);
347 bio
->bi_rw
= REQ_SYNC
|REQ_META
|rw
;
348 bio
->bi_size
= KEY_SIZE(k
) << 9;
350 bio
->bi_end_io
= uuid_endio
;
351 bio
->bi_private
= cl
;
352 bch_bio_map(bio
, c
->uuids
);
354 bch_submit_bbio(bio
, c
, k
, i
);
360 bch_bkey_to_text(buf
, sizeof(buf
), k
);
361 pr_debug("%s UUIDs at %s", rw
& REQ_WRITE
? "wrote" : "read", buf
);
363 for (u
= c
->uuids
; u
< c
->uuids
+ c
->nr_uuids
; u
++)
364 if (!bch_is_zero(u
->uuid
, 16))
365 pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u",
366 u
- c
->uuids
, u
->uuid
, u
->label
,
367 u
->first_reg
, u
->last_reg
, u
->invalidated
);
372 static char *uuid_read(struct cache_set
*c
, struct jset
*j
, struct closure
*cl
)
374 struct bkey
*k
= &j
->uuid_bucket
;
376 if (bch_btree_ptr_invalid(c
, k
))
377 return "bad uuid pointer";
379 bkey_copy(&c
->uuid_bucket
, k
);
380 uuid_io(c
, READ_SYNC
, k
, cl
);
382 if (j
->version
< BCACHE_JSET_VERSION_UUIDv1
) {
383 struct uuid_entry_v0
*u0
= (void *) c
->uuids
;
384 struct uuid_entry
*u1
= (void *) c
->uuids
;
390 * Since the new uuid entry is bigger than the old, we have to
391 * convert starting at the highest memory address and work down
392 * in order to do it in place
395 for (i
= c
->nr_uuids
- 1;
398 memcpy(u1
[i
].uuid
, u0
[i
].uuid
, 16);
399 memcpy(u1
[i
].label
, u0
[i
].label
, 32);
401 u1
[i
].first_reg
= u0
[i
].first_reg
;
402 u1
[i
].last_reg
= u0
[i
].last_reg
;
403 u1
[i
].invalidated
= u0
[i
].invalidated
;
413 static int __uuid_write(struct cache_set
*c
)
417 closure_init_stack(&cl
);
419 lockdep_assert_held(&bch_register_lock
);
421 if (bch_bucket_alloc_set(c
, WATERMARK_METADATA
, &k
.key
, 1, true))
424 SET_KEY_SIZE(&k
.key
, c
->sb
.bucket_size
);
425 uuid_io(c
, REQ_WRITE
, &k
.key
, &cl
);
428 bkey_copy(&c
->uuid_bucket
, &k
.key
);
433 int bch_uuid_write(struct cache_set
*c
)
435 int ret
= __uuid_write(c
);
438 bch_journal_meta(c
, NULL
);
443 static struct uuid_entry
*uuid_find(struct cache_set
*c
, const char *uuid
)
445 struct uuid_entry
*u
;
448 u
< c
->uuids
+ c
->nr_uuids
; u
++)
449 if (!memcmp(u
->uuid
, uuid
, 16))
455 static struct uuid_entry
*uuid_find_empty(struct cache_set
*c
)
457 static const char zero_uuid
[16] = "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0";
458 return uuid_find(c
, zero_uuid
);
462 * Bucket priorities/gens:
464 * For each bucket, we store on disk its
468 * See alloc.c for an explanation of the gen. The priority is used to implement
469 * lru (and in the future other) cache replacement policies; for most purposes
470 * it's just an opaque integer.
472 * The gens and the priorities don't have a whole lot to do with each other, and
473 * it's actually the gens that must be written out at specific times - it's no
474 * big deal if the priorities don't get written, if we lose them we just reuse
475 * buckets in suboptimal order.
477 * On disk they're stored in a packed array, and in as many buckets are required
478 * to fit them all. The buckets we use to store them form a list; the journal
479 * header points to the first bucket, the first bucket points to the second
482 * This code is used by the allocation code; periodically (whenever it runs out
483 * of buckets to allocate from) the allocation code will invalidate some
484 * buckets, but it can't use those buckets until their new gens are safely on
488 static void prio_endio(struct bio
*bio
, int error
)
490 struct cache
*ca
= bio
->bi_private
;
492 cache_set_err_on(error
, ca
->set
, "accessing priorities");
493 bch_bbio_free(bio
, ca
->set
);
494 closure_put(&ca
->prio
);
497 static void prio_io(struct cache
*ca
, uint64_t bucket
, unsigned long rw
)
499 struct closure
*cl
= &ca
->prio
;
500 struct bio
*bio
= bch_bbio_alloc(ca
->set
);
502 closure_init_stack(cl
);
504 bio
->bi_sector
= bucket
* ca
->sb
.bucket_size
;
505 bio
->bi_bdev
= ca
->bdev
;
506 bio
->bi_rw
= REQ_SYNC
|REQ_META
|rw
;
507 bio
->bi_size
= bucket_bytes(ca
);
509 bio
->bi_end_io
= prio_endio
;
510 bio
->bi_private
= ca
;
511 bch_bio_map(bio
, ca
->disk_buckets
);
513 closure_bio_submit(bio
, &ca
->prio
, ca
);
517 #define buckets_free(c) "free %zu, free_inc %zu, unused %zu", \
518 fifo_used(&c->free), fifo_used(&c->free_inc), fifo_used(&c->unused)
520 void bch_prio_write(struct cache
*ca
)
526 closure_init_stack(&cl
);
528 lockdep_assert_held(&ca
->set
->bucket_lock
);
530 for (b
= ca
->buckets
;
531 b
< ca
->buckets
+ ca
->sb
.nbuckets
; b
++)
532 b
->disk_gen
= b
->gen
;
534 ca
->disk_buckets
->seq
++;
536 atomic_long_add(ca
->sb
.bucket_size
* prio_buckets(ca
),
537 &ca
->meta_sectors_written
);
539 pr_debug("free %zu, free_inc %zu, unused %zu", fifo_used(&ca
->free
),
540 fifo_used(&ca
->free_inc
), fifo_used(&ca
->unused
));
542 for (i
= prio_buckets(ca
) - 1; i
>= 0; --i
) {
544 struct prio_set
*p
= ca
->disk_buckets
;
545 struct bucket_disk
*d
= p
->data
;
546 struct bucket_disk
*end
= d
+ prios_per_bucket(ca
);
548 for (b
= ca
->buckets
+ i
* prios_per_bucket(ca
);
549 b
< ca
->buckets
+ ca
->sb
.nbuckets
&& d
< end
;
551 d
->prio
= cpu_to_le16(b
->prio
);
555 p
->next_bucket
= ca
->prio_buckets
[i
+ 1];
556 p
->magic
= pset_magic(&ca
->sb
);
557 p
->csum
= bch_crc64(&p
->magic
, bucket_bytes(ca
) - 8);
559 bucket
= bch_bucket_alloc(ca
, WATERMARK_PRIO
, true);
560 BUG_ON(bucket
== -1);
562 mutex_unlock(&ca
->set
->bucket_lock
);
563 prio_io(ca
, bucket
, REQ_WRITE
);
564 mutex_lock(&ca
->set
->bucket_lock
);
566 ca
->prio_buckets
[i
] = bucket
;
567 atomic_dec_bug(&ca
->buckets
[bucket
].pin
);
570 mutex_unlock(&ca
->set
->bucket_lock
);
572 bch_journal_meta(ca
->set
, &cl
);
575 mutex_lock(&ca
->set
->bucket_lock
);
577 ca
->need_save_prio
= 0;
580 * Don't want the old priorities to get garbage collected until after we
581 * finish writing the new ones, and they're journalled
583 for (i
= 0; i
< prio_buckets(ca
); i
++)
584 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
->disk_gen
= b
->last_gc
= b
->gc_gen
= 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");
690 static void bcache_device_detach(struct bcache_device
*d
)
692 lockdep_assert_held(&bch_register_lock
);
694 if (test_bit(BCACHE_DEV_DETACHING
, &d
->flags
)) {
695 struct uuid_entry
*u
= d
->c
->uuids
+ d
->id
;
697 SET_UUID_FLASH_ONLY(u
, 0);
698 memcpy(u
->uuid
, invalid_uuid
, 16);
699 u
->invalidated
= cpu_to_le32(get_seconds());
700 bch_uuid_write(d
->c
);
703 bcache_device_unlink(d
);
705 d
->c
->devices
[d
->id
] = NULL
;
706 closure_put(&d
->c
->caching
);
710 static void bcache_device_attach(struct bcache_device
*d
, struct cache_set
*c
,
713 BUG_ON(test_bit(CACHE_SET_STOPPING
, &c
->flags
));
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
);
737 bio_split_pool_free(&d
->bio_split_hook
);
738 if (d
->unaligned_bvec
)
739 mempool_destroy(d
->unaligned_bvec
);
741 bioset_free(d
->bio_split
);
742 if (is_vmalloc_addr(d
->full_dirty_stripes
))
743 vfree(d
->full_dirty_stripes
);
745 kfree(d
->full_dirty_stripes
);
746 if (is_vmalloc_addr(d
->stripe_sectors_dirty
))
747 vfree(d
->stripe_sectors_dirty
);
749 kfree(d
->stripe_sectors_dirty
);
751 closure_debug_destroy(&d
->cl
);
754 static int bcache_device_init(struct bcache_device
*d
, unsigned block_size
,
757 struct request_queue
*q
;
761 d
->stripe_size
= 1 << 31;
763 d
->nr_stripes
= DIV_ROUND_UP_ULL(sectors
, d
->stripe_size
);
765 if (!d
->nr_stripes
||
766 d
->nr_stripes
> INT_MAX
||
767 d
->nr_stripes
> SIZE_MAX
/ sizeof(atomic_t
)) {
768 pr_err("nr_stripes too large");
772 n
= d
->nr_stripes
* sizeof(atomic_t
);
773 d
->stripe_sectors_dirty
= n
< PAGE_SIZE
<< 6
774 ? kzalloc(n
, GFP_KERNEL
)
776 if (!d
->stripe_sectors_dirty
)
779 n
= BITS_TO_LONGS(d
->nr_stripes
) * sizeof(unsigned long);
780 d
->full_dirty_stripes
= n
< PAGE_SIZE
<< 6
781 ? kzalloc(n
, GFP_KERNEL
)
783 if (!d
->full_dirty_stripes
)
786 if (!(d
->bio_split
= bioset_create(4, offsetof(struct bbio
, bio
))) ||
787 !(d
->unaligned_bvec
= mempool_create_kmalloc_pool(1,
788 sizeof(struct bio_vec
) * BIO_MAX_PAGES
)) ||
789 bio_split_pool_init(&d
->bio_split_hook
) ||
790 !(d
->disk
= alloc_disk(1)) ||
791 !(q
= blk_alloc_queue(GFP_KERNEL
)))
794 set_capacity(d
->disk
, sectors
);
795 snprintf(d
->disk
->disk_name
, DISK_NAME_LEN
, "bcache%i", bcache_minor
);
797 d
->disk
->major
= bcache_major
;
798 d
->disk
->first_minor
= bcache_minor
++;
799 d
->disk
->fops
= &bcache_ops
;
800 d
->disk
->private_data
= d
;
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 q
->limits
.max_discard_sectors
= UINT_MAX
;
811 q
->limits
.io_min
= block_size
;
812 q
->limits
.logical_block_size
= block_size
;
813 q
->limits
.physical_block_size
= block_size
;
814 set_bit(QUEUE_FLAG_NONROT
, &d
->disk
->queue
->queue_flags
);
815 set_bit(QUEUE_FLAG_DISCARD
, &d
->disk
->queue
->queue_flags
);
817 blk_queue_flush(q
, REQ_FLUSH
|REQ_FUA
);
824 static void calc_cached_dev_sectors(struct cache_set
*c
)
826 uint64_t sectors
= 0;
827 struct cached_dev
*dc
;
829 list_for_each_entry(dc
, &c
->cached_devs
, list
)
830 sectors
+= bdev_sectors(dc
->bdev
);
832 c
->cached_dev_sectors
= sectors
;
835 void bch_cached_dev_run(struct cached_dev
*dc
)
837 struct bcache_device
*d
= &dc
->disk
;
838 char buf
[SB_LABEL_SIZE
+ 1];
841 kasprintf(GFP_KERNEL
, "CACHED_UUID=%pU", dc
->sb
.uuid
),
846 memcpy(buf
, dc
->sb
.label
, SB_LABEL_SIZE
);
847 buf
[SB_LABEL_SIZE
] = '\0';
848 env
[2] = kasprintf(GFP_KERNEL
, "CACHED_LABEL=%s", buf
);
850 if (atomic_xchg(&dc
->running
, 1))
854 BDEV_STATE(&dc
->sb
) != BDEV_STATE_NONE
) {
856 closure_init_stack(&cl
);
858 SET_BDEV_STATE(&dc
->sb
, BDEV_STATE_STALE
);
859 bch_write_bdev_super(dc
, &cl
);
864 bd_link_disk_holder(dc
->bdev
, dc
->disk
.disk
);
865 /* won't show up in the uevent file, use udevadm monitor -e instead
866 * only class / kset properties are persistent */
867 kobject_uevent_env(&disk_to_dev(d
->disk
)->kobj
, KOBJ_CHANGE
, env
);
871 if (sysfs_create_link(&d
->kobj
, &disk_to_dev(d
->disk
)->kobj
, "dev") ||
872 sysfs_create_link(&disk_to_dev(d
->disk
)->kobj
, &d
->kobj
, "bcache"))
873 pr_debug("error creating sysfs link");
876 static void cached_dev_detach_finish(struct work_struct
*w
)
878 struct cached_dev
*dc
= container_of(w
, struct cached_dev
, detach
);
879 char buf
[BDEVNAME_SIZE
];
881 closure_init_stack(&cl
);
883 BUG_ON(!test_bit(BCACHE_DEV_DETACHING
, &dc
->disk
.flags
));
884 BUG_ON(atomic_read(&dc
->count
));
886 mutex_lock(&bch_register_lock
);
888 memset(&dc
->sb
.set_uuid
, 0, 16);
889 SET_BDEV_STATE(&dc
->sb
, BDEV_STATE_NONE
);
891 bch_write_bdev_super(dc
, &cl
);
894 bcache_device_detach(&dc
->disk
);
895 list_move(&dc
->list
, &uncached_devices
);
897 clear_bit(BCACHE_DEV_DETACHING
, &dc
->disk
.flags
);
899 mutex_unlock(&bch_register_lock
);
901 pr_info("Caching disabled for %s", bdevname(dc
->bdev
, buf
));
903 /* Drop ref we took in cached_dev_detach() */
904 closure_put(&dc
->disk
.cl
);
907 void bch_cached_dev_detach(struct cached_dev
*dc
)
909 lockdep_assert_held(&bch_register_lock
);
911 if (test_bit(BCACHE_DEV_CLOSING
, &dc
->disk
.flags
))
914 if (test_and_set_bit(BCACHE_DEV_DETACHING
, &dc
->disk
.flags
))
918 * Block the device from being closed and freed until we're finished
921 closure_get(&dc
->disk
.cl
);
923 bch_writeback_queue(dc
);
927 int bch_cached_dev_attach(struct cached_dev
*dc
, struct cache_set
*c
)
929 uint32_t rtime
= cpu_to_le32(get_seconds());
930 struct uuid_entry
*u
;
931 char buf
[BDEVNAME_SIZE
];
933 bdevname(dc
->bdev
, buf
);
935 if (memcmp(dc
->sb
.set_uuid
, c
->sb
.set_uuid
, 16))
939 pr_err("Can't attach %s: already attached", buf
);
943 if (test_bit(CACHE_SET_STOPPING
, &c
->flags
)) {
944 pr_err("Can't attach %s: shutting down", buf
);
948 if (dc
->sb
.block_size
< c
->sb
.block_size
) {
950 pr_err("Couldn't attach %s: block size less than set's block size",
955 u
= uuid_find(c
, dc
->sb
.uuid
);
958 (BDEV_STATE(&dc
->sb
) == BDEV_STATE_STALE
||
959 BDEV_STATE(&dc
->sb
) == BDEV_STATE_NONE
)) {
960 memcpy(u
->uuid
, invalid_uuid
, 16);
961 u
->invalidated
= cpu_to_le32(get_seconds());
966 if (BDEV_STATE(&dc
->sb
) == BDEV_STATE_DIRTY
) {
967 pr_err("Couldn't find uuid for %s in set", buf
);
971 u
= uuid_find_empty(c
);
973 pr_err("Not caching %s, no room for UUID", buf
);
978 /* Deadlocks since we're called via sysfs...
979 sysfs_remove_file(&dc->kobj, &sysfs_attach);
982 if (bch_is_zero(u
->uuid
, 16)) {
984 closure_init_stack(&cl
);
986 memcpy(u
->uuid
, dc
->sb
.uuid
, 16);
987 memcpy(u
->label
, dc
->sb
.label
, SB_LABEL_SIZE
);
988 u
->first_reg
= u
->last_reg
= rtime
;
991 memcpy(dc
->sb
.set_uuid
, c
->sb
.set_uuid
, 16);
992 SET_BDEV_STATE(&dc
->sb
, BDEV_STATE_CLEAN
);
994 bch_write_bdev_super(dc
, &cl
);
1001 bcache_device_attach(&dc
->disk
, c
, u
- c
->uuids
);
1002 list_move(&dc
->list
, &c
->cached_devs
);
1003 calc_cached_dev_sectors(c
);
1007 * dc->c must be set before dc->count != 0 - paired with the mb in
1010 atomic_set(&dc
->count
, 1);
1012 if (BDEV_STATE(&dc
->sb
) == BDEV_STATE_DIRTY
) {
1013 bch_sectors_dirty_init(dc
);
1014 atomic_set(&dc
->has_dirty
, 1);
1015 atomic_inc(&dc
->count
);
1016 bch_writeback_queue(dc
);
1019 bch_cached_dev_run(dc
);
1020 bcache_device_link(&dc
->disk
, c
, "bdev");
1022 pr_info("Caching %s as %s on set %pU",
1023 bdevname(dc
->bdev
, buf
), dc
->disk
.disk
->disk_name
,
1024 dc
->disk
.c
->sb
.set_uuid
);
1028 void bch_cached_dev_release(struct kobject
*kobj
)
1030 struct cached_dev
*dc
= container_of(kobj
, struct cached_dev
,
1033 module_put(THIS_MODULE
);
1036 static void cached_dev_free(struct closure
*cl
)
1038 struct cached_dev
*dc
= container_of(cl
, struct cached_dev
, disk
.cl
);
1040 cancel_delayed_work_sync(&dc
->writeback_rate_update
);
1041 kthread_stop(dc
->writeback_thread
);
1043 mutex_lock(&bch_register_lock
);
1045 if (atomic_read(&dc
->running
))
1046 bd_unlink_disk_holder(dc
->bdev
, dc
->disk
.disk
);
1047 bcache_device_free(&dc
->disk
);
1048 list_del(&dc
->list
);
1050 mutex_unlock(&bch_register_lock
);
1052 if (!IS_ERR_OR_NULL(dc
->bdev
)) {
1053 if (dc
->bdev
->bd_disk
)
1054 blk_sync_queue(bdev_get_queue(dc
->bdev
));
1056 blkdev_put(dc
->bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
1059 wake_up(&unregister_wait
);
1061 kobject_put(&dc
->disk
.kobj
);
1064 static void cached_dev_flush(struct closure
*cl
)
1066 struct cached_dev
*dc
= container_of(cl
, struct cached_dev
, disk
.cl
);
1067 struct bcache_device
*d
= &dc
->disk
;
1069 mutex_lock(&bch_register_lock
);
1070 bcache_device_unlink(d
);
1071 mutex_unlock(&bch_register_lock
);
1073 bch_cache_accounting_destroy(&dc
->accounting
);
1074 kobject_del(&d
->kobj
);
1076 continue_at(cl
, cached_dev_free
, system_wq
);
1079 static int cached_dev_init(struct cached_dev
*dc
, unsigned block_size
)
1083 struct request_queue
*q
= bdev_get_queue(dc
->bdev
);
1085 __module_get(THIS_MODULE
);
1086 INIT_LIST_HEAD(&dc
->list
);
1087 closure_init(&dc
->disk
.cl
, NULL
);
1088 set_closure_fn(&dc
->disk
.cl
, cached_dev_flush
, system_wq
);
1089 kobject_init(&dc
->disk
.kobj
, &bch_cached_dev_ktype
);
1090 INIT_WORK(&dc
->detach
, cached_dev_detach_finish
);
1091 closure_init_unlocked(&dc
->sb_write
);
1092 INIT_LIST_HEAD(&dc
->io_lru
);
1093 spin_lock_init(&dc
->io_lock
);
1094 bch_cache_accounting_init(&dc
->accounting
, &dc
->disk
.cl
);
1096 dc
->sequential_cutoff
= 4 << 20;
1098 for (io
= dc
->io
; io
< dc
->io
+ RECENT_IO
; io
++) {
1099 list_add(&io
->lru
, &dc
->io_lru
);
1100 hlist_add_head(&io
->hash
, dc
->io_hash
+ RECENT_IO
);
1103 ret
= bcache_device_init(&dc
->disk
, block_size
,
1104 dc
->bdev
->bd_part
->nr_sects
- dc
->sb
.data_offset
);
1108 set_capacity(dc
->disk
.disk
,
1109 dc
->bdev
->bd_part
->nr_sects
- dc
->sb
.data_offset
);
1111 dc
->disk
.disk
->queue
->backing_dev_info
.ra_pages
=
1112 max(dc
->disk
.disk
->queue
->backing_dev_info
.ra_pages
,
1113 q
->backing_dev_info
.ra_pages
);
1115 bch_cached_dev_request_init(dc
);
1116 bch_cached_dev_writeback_init(dc
);
1120 /* Cached device - bcache superblock */
1122 static void register_bdev(struct cache_sb
*sb
, struct page
*sb_page
,
1123 struct block_device
*bdev
,
1124 struct cached_dev
*dc
)
1126 char name
[BDEVNAME_SIZE
];
1127 const char *err
= "cannot allocate memory";
1128 struct cache_set
*c
;
1130 memcpy(&dc
->sb
, sb
, sizeof(struct cache_sb
));
1132 dc
->bdev
->bd_holder
= dc
;
1134 bio_init(&dc
->sb_bio
);
1135 dc
->sb_bio
.bi_max_vecs
= 1;
1136 dc
->sb_bio
.bi_io_vec
= dc
->sb_bio
.bi_inline_vecs
;
1137 dc
->sb_bio
.bi_io_vec
[0].bv_page
= sb_page
;
1140 if (cached_dev_init(dc
, sb
->block_size
<< 9))
1143 err
= "error creating kobject";
1144 if (kobject_add(&dc
->disk
.kobj
, &part_to_dev(bdev
->bd_part
)->kobj
,
1147 if (bch_cache_accounting_add_kobjs(&dc
->accounting
, &dc
->disk
.kobj
))
1150 pr_info("registered backing device %s", bdevname(bdev
, name
));
1152 list_add(&dc
->list
, &uncached_devices
);
1153 list_for_each_entry(c
, &bch_cache_sets
, list
)
1154 bch_cached_dev_attach(dc
, c
);
1156 if (BDEV_STATE(&dc
->sb
) == BDEV_STATE_NONE
||
1157 BDEV_STATE(&dc
->sb
) == BDEV_STATE_STALE
)
1158 bch_cached_dev_run(dc
);
1162 pr_notice("error opening %s: %s", bdevname(bdev
, name
), err
);
1163 bcache_device_stop(&dc
->disk
);
1166 /* Flash only volumes */
1168 void bch_flash_dev_release(struct kobject
*kobj
)
1170 struct bcache_device
*d
= container_of(kobj
, struct bcache_device
,
1175 static void flash_dev_free(struct closure
*cl
)
1177 struct bcache_device
*d
= container_of(cl
, struct bcache_device
, cl
);
1178 bcache_device_free(d
);
1179 kobject_put(&d
->kobj
);
1182 static void flash_dev_flush(struct closure
*cl
)
1184 struct bcache_device
*d
= container_of(cl
, struct bcache_device
, cl
);
1186 bcache_device_unlink(d
);
1187 kobject_del(&d
->kobj
);
1188 continue_at(cl
, flash_dev_free
, system_wq
);
1191 static int flash_dev_run(struct cache_set
*c
, struct uuid_entry
*u
)
1193 struct bcache_device
*d
= kzalloc(sizeof(struct bcache_device
),
1198 closure_init(&d
->cl
, NULL
);
1199 set_closure_fn(&d
->cl
, flash_dev_flush
, system_wq
);
1201 kobject_init(&d
->kobj
, &bch_flash_dev_ktype
);
1203 if (bcache_device_init(d
, block_bytes(c
), u
->sectors
))
1206 bcache_device_attach(d
, c
, u
- c
->uuids
);
1207 bch_flash_dev_request_init(d
);
1210 if (kobject_add(&d
->kobj
, &disk_to_dev(d
->disk
)->kobj
, "bcache"))
1213 bcache_device_link(d
, c
, "volume");
1217 kobject_put(&d
->kobj
);
1221 static int flash_devs_run(struct cache_set
*c
)
1224 struct uuid_entry
*u
;
1227 u
< c
->uuids
+ c
->nr_uuids
&& !ret
;
1229 if (UUID_FLASH_ONLY(u
))
1230 ret
= flash_dev_run(c
, u
);
1235 int bch_flash_dev_create(struct cache_set
*c
, uint64_t size
)
1237 struct uuid_entry
*u
;
1239 if (test_bit(CACHE_SET_STOPPING
, &c
->flags
))
1242 u
= uuid_find_empty(c
);
1244 pr_err("Can't create volume, no room for UUID");
1248 get_random_bytes(u
->uuid
, 16);
1249 memset(u
->label
, 0, 32);
1250 u
->first_reg
= u
->last_reg
= cpu_to_le32(get_seconds());
1252 SET_UUID_FLASH_ONLY(u
, 1);
1253 u
->sectors
= size
>> 9;
1257 return flash_dev_run(c
, u
);
1263 bool bch_cache_set_error(struct cache_set
*c
, const char *fmt
, ...)
1267 if (c
->on_error
!= ON_ERROR_PANIC
&&
1268 test_bit(CACHE_SET_STOPPING
, &c
->flags
))
1271 /* XXX: we can be called from atomic context
1272 acquire_console_sem();
1275 printk(KERN_ERR
"bcache: error on %pU: ", c
->sb
.set_uuid
);
1277 va_start(args
, fmt
);
1281 printk(", disabling caching\n");
1283 if (c
->on_error
== ON_ERROR_PANIC
)
1284 panic("panic forced after error\n");
1286 bch_cache_set_unregister(c
);
1290 void bch_cache_set_release(struct kobject
*kobj
)
1292 struct cache_set
*c
= container_of(kobj
, struct cache_set
, kobj
);
1294 module_put(THIS_MODULE
);
1297 static void cache_set_free(struct closure
*cl
)
1299 struct cache_set
*c
= container_of(cl
, struct cache_set
, cl
);
1303 if (!IS_ERR_OR_NULL(c
->debug
))
1304 debugfs_remove(c
->debug
);
1306 bch_open_buckets_free(c
);
1307 bch_btree_cache_free(c
);
1308 bch_journal_free(c
);
1310 for_each_cache(ca
, c
, i
)
1312 kobject_put(&ca
->kobj
);
1314 free_pages((unsigned long) c
->uuids
, ilog2(bucket_pages(c
)));
1315 free_pages((unsigned long) c
->sort
, ilog2(bucket_pages(c
)));
1318 bioset_free(c
->bio_split
);
1320 mempool_destroy(c
->fill_iter
);
1322 mempool_destroy(c
->bio_meta
);
1324 mempool_destroy(c
->search
);
1327 mutex_lock(&bch_register_lock
);
1329 mutex_unlock(&bch_register_lock
);
1331 pr_info("Cache set %pU unregistered", c
->sb
.set_uuid
);
1332 wake_up(&unregister_wait
);
1334 closure_debug_destroy(&c
->cl
);
1335 kobject_put(&c
->kobj
);
1338 static void cache_set_flush(struct closure
*cl
)
1340 struct cache_set
*c
= container_of(cl
, struct cache_set
, caching
);
1345 bch_cache_accounting_destroy(&c
->accounting
);
1347 kobject_put(&c
->internal
);
1348 kobject_del(&c
->kobj
);
1351 kthread_stop(c
->gc_thread
);
1353 if (!IS_ERR_OR_NULL(c
->root
))
1354 list_add(&c
->root
->list
, &c
->btree_cache
);
1356 /* Should skip this if we're unregistering because of an error */
1357 list_for_each_entry(b
, &c
->btree_cache
, list
)
1358 if (btree_node_dirty(b
))
1359 bch_btree_node_write(b
, NULL
);
1361 for_each_cache(ca
, c
, i
)
1362 if (ca
->alloc_thread
)
1363 kthread_stop(ca
->alloc_thread
);
1368 static void __cache_set_unregister(struct closure
*cl
)
1370 struct cache_set
*c
= container_of(cl
, struct cache_set
, caching
);
1371 struct cached_dev
*dc
;
1374 mutex_lock(&bch_register_lock
);
1376 for (i
= 0; i
< c
->nr_uuids
; i
++)
1377 if (c
->devices
[i
]) {
1378 if (!UUID_FLASH_ONLY(&c
->uuids
[i
]) &&
1379 test_bit(CACHE_SET_UNREGISTERING
, &c
->flags
)) {
1380 dc
= container_of(c
->devices
[i
],
1381 struct cached_dev
, disk
);
1382 bch_cached_dev_detach(dc
);
1384 bcache_device_stop(c
->devices
[i
]);
1388 mutex_unlock(&bch_register_lock
);
1390 continue_at(cl
, cache_set_flush
, system_wq
);
1393 void bch_cache_set_stop(struct cache_set
*c
)
1395 if (!test_and_set_bit(CACHE_SET_STOPPING
, &c
->flags
))
1396 closure_queue(&c
->caching
);
1399 void bch_cache_set_unregister(struct cache_set
*c
)
1401 set_bit(CACHE_SET_UNREGISTERING
, &c
->flags
);
1402 bch_cache_set_stop(c
);
1405 #define alloc_bucket_pages(gfp, c) \
1406 ((void *) __get_free_pages(__GFP_ZERO|gfp, ilog2(bucket_pages(c))))
1408 struct cache_set
*bch_cache_set_alloc(struct cache_sb
*sb
)
1411 struct cache_set
*c
= kzalloc(sizeof(struct cache_set
), GFP_KERNEL
);
1415 __module_get(THIS_MODULE
);
1416 closure_init(&c
->cl
, NULL
);
1417 set_closure_fn(&c
->cl
, cache_set_free
, system_wq
);
1419 closure_init(&c
->caching
, &c
->cl
);
1420 set_closure_fn(&c
->caching
, __cache_set_unregister
, system_wq
);
1422 /* Maybe create continue_at_noreturn() and use it here? */
1423 closure_set_stopped(&c
->cl
);
1424 closure_put(&c
->cl
);
1426 kobject_init(&c
->kobj
, &bch_cache_set_ktype
);
1427 kobject_init(&c
->internal
, &bch_cache_set_internal_ktype
);
1429 bch_cache_accounting_init(&c
->accounting
, &c
->cl
);
1431 memcpy(c
->sb
.set_uuid
, sb
->set_uuid
, 16);
1432 c
->sb
.block_size
= sb
->block_size
;
1433 c
->sb
.bucket_size
= sb
->bucket_size
;
1434 c
->sb
.nr_in_set
= sb
->nr_in_set
;
1435 c
->sb
.last_mount
= sb
->last_mount
;
1436 c
->bucket_bits
= ilog2(sb
->bucket_size
);
1437 c
->block_bits
= ilog2(sb
->block_size
);
1438 c
->nr_uuids
= bucket_bytes(c
) / sizeof(struct uuid_entry
);
1440 c
->btree_pages
= c
->sb
.bucket_size
/ PAGE_SECTORS
;
1441 if (c
->btree_pages
> BTREE_MAX_PAGES
)
1442 c
->btree_pages
= max_t(int, c
->btree_pages
/ 4,
1445 c
->sort_crit_factor
= int_sqrt(c
->btree_pages
);
1447 closure_init_unlocked(&c
->sb_write
);
1448 mutex_init(&c
->bucket_lock
);
1449 init_waitqueue_head(&c
->try_wait
);
1450 init_waitqueue_head(&c
->bucket_wait
);
1451 closure_init_unlocked(&c
->uuid_write
);
1452 mutex_init(&c
->sort_lock
);
1454 spin_lock_init(&c
->sort_time
.lock
);
1455 spin_lock_init(&c
->btree_gc_time
.lock
);
1456 spin_lock_init(&c
->btree_split_time
.lock
);
1457 spin_lock_init(&c
->btree_read_time
.lock
);
1458 spin_lock_init(&c
->try_harder_time
.lock
);
1460 bch_moving_init_cache_set(c
);
1462 INIT_LIST_HEAD(&c
->list
);
1463 INIT_LIST_HEAD(&c
->cached_devs
);
1464 INIT_LIST_HEAD(&c
->btree_cache
);
1465 INIT_LIST_HEAD(&c
->btree_cache_freeable
);
1466 INIT_LIST_HEAD(&c
->btree_cache_freed
);
1467 INIT_LIST_HEAD(&c
->data_buckets
);
1469 c
->search
= mempool_create_slab_pool(32, bch_search_cache
);
1473 iter_size
= (sb
->bucket_size
/ sb
->block_size
+ 1) *
1474 sizeof(struct btree_iter_set
);
1476 if (!(c
->devices
= kzalloc(c
->nr_uuids
* sizeof(void *), GFP_KERNEL
)) ||
1477 !(c
->bio_meta
= mempool_create_kmalloc_pool(2,
1478 sizeof(struct bbio
) + sizeof(struct bio_vec
) *
1479 bucket_pages(c
))) ||
1480 !(c
->fill_iter
= mempool_create_kmalloc_pool(1, iter_size
)) ||
1481 !(c
->bio_split
= bioset_create(4, offsetof(struct bbio
, bio
))) ||
1482 !(c
->sort
= alloc_bucket_pages(GFP_KERNEL
, c
)) ||
1483 !(c
->uuids
= alloc_bucket_pages(GFP_KERNEL
, c
)) ||
1484 bch_journal_alloc(c
) ||
1485 bch_btree_cache_alloc(c
) ||
1486 bch_open_buckets_alloc(c
))
1489 c
->congested_read_threshold_us
= 2000;
1490 c
->congested_write_threshold_us
= 20000;
1491 c
->error_limit
= 8 << IO_ERROR_SHIFT
;
1495 bch_cache_set_unregister(c
);
1499 static void run_cache_set(struct cache_set
*c
)
1501 const char *err
= "cannot allocate memory";
1502 struct cached_dev
*dc
, *t
;
1507 closure_init_stack(&cl
);
1509 for_each_cache(ca
, c
, i
)
1510 c
->nbuckets
+= ca
->sb
.nbuckets
;
1512 if (CACHE_SYNC(&c
->sb
)) {
1517 err
= "cannot allocate memory for journal";
1518 if (bch_journal_read(c
, &journal
))
1521 pr_debug("btree_journal_read() done");
1523 err
= "no journal entries found";
1524 if (list_empty(&journal
))
1527 j
= &list_entry(journal
.prev
, struct journal_replay
, list
)->j
;
1529 err
= "IO error reading priorities";
1530 for_each_cache(ca
, c
, i
)
1531 prio_read(ca
, j
->prio_bucket
[ca
->sb
.nr_this_dev
]);
1534 * If prio_read() fails it'll call cache_set_error and we'll
1535 * tear everything down right away, but if we perhaps checked
1536 * sooner we could avoid journal replay.
1541 err
= "bad btree root";
1542 if (bch_btree_ptr_invalid(c
, k
))
1545 err
= "error reading btree root";
1546 c
->root
= bch_btree_node_get(c
, k
, j
->btree_level
, true);
1547 if (IS_ERR_OR_NULL(c
->root
))
1550 list_del_init(&c
->root
->list
);
1551 rw_unlock(true, c
->root
);
1553 err
= uuid_read(c
, j
, &cl
);
1557 err
= "error in recovery";
1558 if (bch_btree_check(c
))
1561 bch_journal_mark(c
, &journal
);
1562 bch_btree_gc_finish(c
);
1563 pr_debug("btree_check() done");
1566 * bcache_journal_next() can't happen sooner, or
1567 * btree_gc_finish() will give spurious errors about last_gc >
1568 * gc_gen - this is a hack but oh well.
1570 bch_journal_next(&c
->journal
);
1572 err
= "error starting allocator thread";
1573 for_each_cache(ca
, c
, i
)
1574 if (bch_cache_allocator_start(ca
))
1578 * First place it's safe to allocate: btree_check() and
1579 * btree_gc_finish() have to run before we have buckets to
1580 * allocate, and bch_bucket_alloc_set() might cause a journal
1581 * entry to be written so bcache_journal_next() has to be called
1584 * If the uuids were in the old format we have to rewrite them
1585 * before the next journal entry is written:
1587 if (j
->version
< BCACHE_JSET_VERSION_UUID
)
1590 bch_journal_replay(c
, &journal
);
1592 pr_notice("invalidating existing data");
1594 for_each_cache(ca
, c
, i
) {
1597 ca
->sb
.keys
= clamp_t(int, ca
->sb
.nbuckets
>> 7,
1598 2, SB_JOURNAL_BUCKETS
);
1600 for (j
= 0; j
< ca
->sb
.keys
; j
++)
1601 ca
->sb
.d
[j
] = ca
->sb
.first_bucket
+ j
;
1604 bch_btree_gc_finish(c
);
1606 err
= "error starting allocator thread";
1607 for_each_cache(ca
, c
, i
)
1608 if (bch_cache_allocator_start(ca
))
1611 mutex_lock(&c
->bucket_lock
);
1612 for_each_cache(ca
, c
, i
)
1614 mutex_unlock(&c
->bucket_lock
);
1616 err
= "cannot allocate new UUID bucket";
1617 if (__uuid_write(c
))
1620 err
= "cannot allocate new btree root";
1621 c
->root
= bch_btree_node_alloc(c
, 0, true);
1622 if (IS_ERR_OR_NULL(c
->root
))
1625 bkey_copy_key(&c
->root
->key
, &MAX_KEY
);
1626 bch_btree_node_write(c
->root
, &cl
);
1628 bch_btree_set_root(c
->root
);
1629 rw_unlock(true, c
->root
);
1632 * We don't want to write the first journal entry until
1633 * everything is set up - fortunately journal entries won't be
1634 * written until the SET_CACHE_SYNC() here:
1636 SET_CACHE_SYNC(&c
->sb
, true);
1638 bch_journal_next(&c
->journal
);
1639 bch_journal_meta(c
, &cl
);
1642 err
= "error starting gc thread";
1643 if (bch_gc_thread_start(c
))
1647 c
->sb
.last_mount
= get_seconds();
1648 bcache_write_super(c
);
1650 list_for_each_entry_safe(dc
, t
, &uncached_devices
, list
)
1651 bch_cached_dev_attach(dc
, c
);
1658 /* XXX: test this, it's broken */
1659 bch_cache_set_error(c
, err
);
1662 static bool can_attach_cache(struct cache
*ca
, struct cache_set
*c
)
1664 return ca
->sb
.block_size
== c
->sb
.block_size
&&
1665 ca
->sb
.bucket_size
== c
->sb
.block_size
&&
1666 ca
->sb
.nr_in_set
== c
->sb
.nr_in_set
;
1669 static const char *register_cache_set(struct cache
*ca
)
1672 const char *err
= "cannot allocate memory";
1673 struct cache_set
*c
;
1675 list_for_each_entry(c
, &bch_cache_sets
, list
)
1676 if (!memcmp(c
->sb
.set_uuid
, ca
->sb
.set_uuid
, 16)) {
1677 if (c
->cache
[ca
->sb
.nr_this_dev
])
1678 return "duplicate cache set member";
1680 if (!can_attach_cache(ca
, c
))
1681 return "cache sb does not match set";
1683 if (!CACHE_SYNC(&ca
->sb
))
1684 SET_CACHE_SYNC(&c
->sb
, false);
1689 c
= bch_cache_set_alloc(&ca
->sb
);
1693 err
= "error creating kobject";
1694 if (kobject_add(&c
->kobj
, bcache_kobj
, "%pU", c
->sb
.set_uuid
) ||
1695 kobject_add(&c
->internal
, &c
->kobj
, "internal"))
1698 if (bch_cache_accounting_add_kobjs(&c
->accounting
, &c
->kobj
))
1701 bch_debug_init_cache_set(c
);
1703 list_add(&c
->list
, &bch_cache_sets
);
1705 sprintf(buf
, "cache%i", ca
->sb
.nr_this_dev
);
1706 if (sysfs_create_link(&ca
->kobj
, &c
->kobj
, "set") ||
1707 sysfs_create_link(&c
->kobj
, &ca
->kobj
, buf
))
1710 if (ca
->sb
.seq
> c
->sb
.seq
) {
1711 c
->sb
.version
= ca
->sb
.version
;
1712 memcpy(c
->sb
.set_uuid
, ca
->sb
.set_uuid
, 16);
1713 c
->sb
.flags
= ca
->sb
.flags
;
1714 c
->sb
.seq
= ca
->sb
.seq
;
1715 pr_debug("set version = %llu", c
->sb
.version
);
1719 ca
->set
->cache
[ca
->sb
.nr_this_dev
] = ca
;
1720 c
->cache_by_alloc
[c
->caches_loaded
++] = ca
;
1722 if (c
->caches_loaded
== c
->sb
.nr_in_set
)
1727 bch_cache_set_unregister(c
);
1733 void bch_cache_release(struct kobject
*kobj
)
1735 struct cache
*ca
= container_of(kobj
, struct cache
, kobj
);
1738 ca
->set
->cache
[ca
->sb
.nr_this_dev
] = NULL
;
1740 bio_split_pool_free(&ca
->bio_split_hook
);
1742 free_pages((unsigned long) ca
->disk_buckets
, ilog2(bucket_pages(ca
)));
1743 kfree(ca
->prio_buckets
);
1746 free_heap(&ca
->heap
);
1747 free_fifo(&ca
->unused
);
1748 free_fifo(&ca
->free_inc
);
1749 free_fifo(&ca
->free
);
1751 if (ca
->sb_bio
.bi_inline_vecs
[0].bv_page
)
1752 put_page(ca
->sb_bio
.bi_io_vec
[0].bv_page
);
1754 if (!IS_ERR_OR_NULL(ca
->bdev
)) {
1755 blk_sync_queue(bdev_get_queue(ca
->bdev
));
1756 blkdev_put(ca
->bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
1760 module_put(THIS_MODULE
);
1763 static int cache_alloc(struct cache_sb
*sb
, struct cache
*ca
)
1768 __module_get(THIS_MODULE
);
1769 kobject_init(&ca
->kobj
, &bch_cache_ktype
);
1771 bio_init(&ca
->journal
.bio
);
1772 ca
->journal
.bio
.bi_max_vecs
= 8;
1773 ca
->journal
.bio
.bi_io_vec
= ca
->journal
.bio
.bi_inline_vecs
;
1775 free
= roundup_pow_of_two(ca
->sb
.nbuckets
) >> 9;
1776 free
= max_t(size_t, free
, (prio_buckets(ca
) + 8) * 2);
1778 if (!init_fifo(&ca
->free
, free
, GFP_KERNEL
) ||
1779 !init_fifo(&ca
->free_inc
, free
<< 2, GFP_KERNEL
) ||
1780 !init_fifo(&ca
->unused
, free
<< 2, GFP_KERNEL
) ||
1781 !init_heap(&ca
->heap
, free
<< 3, GFP_KERNEL
) ||
1782 !(ca
->buckets
= vzalloc(sizeof(struct bucket
) *
1783 ca
->sb
.nbuckets
)) ||
1784 !(ca
->prio_buckets
= kzalloc(sizeof(uint64_t) * prio_buckets(ca
) *
1786 !(ca
->disk_buckets
= alloc_bucket_pages(GFP_KERNEL
, ca
)) ||
1787 bio_split_pool_init(&ca
->bio_split_hook
))
1790 ca
->prio_last_buckets
= ca
->prio_buckets
+ prio_buckets(ca
);
1792 for_each_bucket(b
, ca
)
1793 atomic_set(&b
->pin
, 0);
1795 if (bch_cache_allocator_init(ca
))
1800 kobject_put(&ca
->kobj
);
1804 static void register_cache(struct cache_sb
*sb
, struct page
*sb_page
,
1805 struct block_device
*bdev
, struct cache
*ca
)
1807 char name
[BDEVNAME_SIZE
];
1808 const char *err
= "cannot allocate memory";
1810 memcpy(&ca
->sb
, sb
, sizeof(struct cache_sb
));
1812 ca
->bdev
->bd_holder
= ca
;
1814 bio_init(&ca
->sb_bio
);
1815 ca
->sb_bio
.bi_max_vecs
= 1;
1816 ca
->sb_bio
.bi_io_vec
= ca
->sb_bio
.bi_inline_vecs
;
1817 ca
->sb_bio
.bi_io_vec
[0].bv_page
= sb_page
;
1820 if (blk_queue_discard(bdev_get_queue(ca
->bdev
)))
1821 ca
->discard
= CACHE_DISCARD(&ca
->sb
);
1823 if (cache_alloc(sb
, ca
) != 0)
1826 err
= "error creating kobject";
1827 if (kobject_add(&ca
->kobj
, &part_to_dev(bdev
->bd_part
)->kobj
, "bcache"))
1830 err
= register_cache_set(ca
);
1834 pr_info("registered cache device %s", bdevname(bdev
, name
));
1837 pr_notice("error opening %s: %s", bdevname(bdev
, name
), err
);
1838 kobject_put(&ca
->kobj
);
1841 /* Global interfaces/init */
1843 static ssize_t
register_bcache(struct kobject
*, struct kobj_attribute
*,
1844 const char *, size_t);
1846 kobj_attribute_write(register, register_bcache
);
1847 kobj_attribute_write(register_quiet
, register_bcache
);
1849 static bool bch_is_open_backing(struct block_device
*bdev
) {
1850 struct cache_set
*c
, *tc
;
1851 struct cached_dev
*dc
, *t
;
1853 list_for_each_entry_safe(c
, tc
, &bch_cache_sets
, list
)
1854 list_for_each_entry_safe(dc
, t
, &c
->cached_devs
, list
)
1855 if (dc
->bdev
== bdev
)
1857 list_for_each_entry_safe(dc
, t
, &uncached_devices
, list
)
1858 if (dc
->bdev
== bdev
)
1863 static bool bch_is_open_cache(struct block_device
*bdev
) {
1864 struct cache_set
*c
, *tc
;
1868 list_for_each_entry_safe(c
, tc
, &bch_cache_sets
, list
)
1869 for_each_cache(ca
, c
, i
)
1870 if (ca
->bdev
== bdev
)
1875 static bool bch_is_open(struct block_device
*bdev
) {
1876 return bch_is_open_cache(bdev
) || bch_is_open_backing(bdev
);
1879 static ssize_t
register_bcache(struct kobject
*k
, struct kobj_attribute
*attr
,
1880 const char *buffer
, size_t size
)
1883 const char *err
= "cannot allocate memory";
1885 struct cache_sb
*sb
= NULL
;
1886 struct block_device
*bdev
= NULL
;
1887 struct page
*sb_page
= NULL
;
1889 if (!try_module_get(THIS_MODULE
))
1892 mutex_lock(&bch_register_lock
);
1894 if (!(path
= kstrndup(buffer
, size
, GFP_KERNEL
)) ||
1895 !(sb
= kmalloc(sizeof(struct cache_sb
), GFP_KERNEL
)))
1898 err
= "failed to open device";
1899 bdev
= blkdev_get_by_path(strim(path
),
1900 FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
,
1903 if (bdev
== ERR_PTR(-EBUSY
)) {
1904 bdev
= lookup_bdev(strim(path
));
1905 if (!IS_ERR(bdev
) && bch_is_open(bdev
))
1906 err
= "device already registered";
1908 err
= "device busy";
1913 err
= "failed to set blocksize";
1914 if (set_blocksize(bdev
, 4096))
1917 err
= read_super(sb
, bdev
, &sb_page
);
1921 if (SB_IS_BDEV(sb
)) {
1922 struct cached_dev
*dc
= kzalloc(sizeof(*dc
), GFP_KERNEL
);
1926 register_bdev(sb
, sb_page
, bdev
, dc
);
1928 struct cache
*ca
= kzalloc(sizeof(*ca
), GFP_KERNEL
);
1932 register_cache(sb
, sb_page
, bdev
, ca
);
1939 mutex_unlock(&bch_register_lock
);
1940 module_put(THIS_MODULE
);
1944 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
1946 if (attr
!= &ksysfs_register_quiet
)
1947 pr_info("error opening %s: %s", path
, err
);
1952 static int bcache_reboot(struct notifier_block
*n
, unsigned long code
, void *x
)
1954 if (code
== SYS_DOWN
||
1956 code
== SYS_POWER_OFF
) {
1958 unsigned long start
= jiffies
;
1959 bool stopped
= false;
1961 struct cache_set
*c
, *tc
;
1962 struct cached_dev
*dc
, *tdc
;
1964 mutex_lock(&bch_register_lock
);
1966 if (list_empty(&bch_cache_sets
) &&
1967 list_empty(&uncached_devices
))
1970 pr_info("Stopping all devices:");
1972 list_for_each_entry_safe(c
, tc
, &bch_cache_sets
, list
)
1973 bch_cache_set_stop(c
);
1975 list_for_each_entry_safe(dc
, tdc
, &uncached_devices
, list
)
1976 bcache_device_stop(&dc
->disk
);
1978 /* What's a condition variable? */
1980 long timeout
= start
+ 2 * HZ
- jiffies
;
1982 stopped
= list_empty(&bch_cache_sets
) &&
1983 list_empty(&uncached_devices
);
1985 if (timeout
< 0 || stopped
)
1988 prepare_to_wait(&unregister_wait
, &wait
,
1989 TASK_UNINTERRUPTIBLE
);
1991 mutex_unlock(&bch_register_lock
);
1992 schedule_timeout(timeout
);
1993 mutex_lock(&bch_register_lock
);
1996 finish_wait(&unregister_wait
, &wait
);
1999 pr_info("All devices stopped");
2001 pr_notice("Timeout waiting for devices to be closed");
2003 mutex_unlock(&bch_register_lock
);
2009 static struct notifier_block reboot
= {
2010 .notifier_call
= bcache_reboot
,
2011 .priority
= INT_MAX
, /* before any real devices */
2014 static void bcache_exit(void)
2020 kobject_put(bcache_kobj
);
2022 destroy_workqueue(bcache_wq
);
2023 unregister_blkdev(bcache_major
, "bcache");
2024 unregister_reboot_notifier(&reboot
);
2027 static int __init
bcache_init(void)
2029 static const struct attribute
*files
[] = {
2030 &ksysfs_register
.attr
,
2031 &ksysfs_register_quiet
.attr
,
2035 mutex_init(&bch_register_lock
);
2036 init_waitqueue_head(&unregister_wait
);
2037 register_reboot_notifier(&reboot
);
2038 closure_debug_init();
2040 bcache_major
= register_blkdev(0, "bcache");
2041 if (bcache_major
< 0)
2042 return bcache_major
;
2044 if (!(bcache_wq
= create_workqueue("bcache")) ||
2045 !(bcache_kobj
= kobject_create_and_add("bcache", fs_kobj
)) ||
2046 sysfs_create_files(bcache_kobj
, files
) ||
2048 bch_request_init() ||
2049 bch_debug_init(bcache_kobj
))
2058 module_exit(bcache_exit
);
2059 module_init(bcache_init
);