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/idr.h>
20 #include <linux/kthread.h>
21 #include <linux/module.h>
22 #include <linux/random.h>
23 #include <linux/reboot.h>
24 #include <linux/sysfs.h>
26 MODULE_LICENSE("GPL");
27 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
29 static const char bcache_magic
[] = {
30 0xc6, 0x85, 0x73, 0xf6, 0x4e, 0x1a, 0x45, 0xca,
31 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81
34 static const char invalid_uuid
[] = {
35 0xa0, 0x3e, 0xf8, 0xed, 0x3e, 0xe1, 0xb8, 0x78,
36 0xc8, 0x50, 0xfc, 0x5e, 0xcb, 0x16, 0xcd, 0x99
39 /* Default is -1; we skip past it for struct cached_dev's cache mode */
40 const char * const bch_cache_modes
[] = {
49 static struct kobject
*bcache_kobj
;
50 struct mutex bch_register_lock
;
51 LIST_HEAD(bch_cache_sets
);
52 static LIST_HEAD(uncached_devices
);
54 static int bcache_major
;
55 static DEFINE_IDA(bcache_minor
);
56 static wait_queue_head_t unregister_wait
;
57 struct workqueue_struct
*bcache_wq
;
59 #define BTREE_MAX_PAGES (256 * 1024 / PAGE_SIZE)
61 static void bio_split_pool_free(struct bio_split_pool
*p
)
63 if (p
->bio_split_hook
)
64 mempool_destroy(p
->bio_split_hook
);
67 bioset_free(p
->bio_split
);
70 static int bio_split_pool_init(struct bio_split_pool
*p
)
72 p
->bio_split
= bioset_create(4, 0);
76 p
->bio_split_hook
= mempool_create_kmalloc_pool(4,
77 sizeof(struct bio_split_hook
));
78 if (!p
->bio_split_hook
)
86 static const char *read_super(struct cache_sb
*sb
, struct block_device
*bdev
,
91 struct buffer_head
*bh
= __bread(bdev
, 1, SB_SIZE
);
97 s
= (struct cache_sb
*) bh
->b_data
;
99 sb
->offset
= le64_to_cpu(s
->offset
);
100 sb
->version
= le64_to_cpu(s
->version
);
102 memcpy(sb
->magic
, s
->magic
, 16);
103 memcpy(sb
->uuid
, s
->uuid
, 16);
104 memcpy(sb
->set_uuid
, s
->set_uuid
, 16);
105 memcpy(sb
->label
, s
->label
, SB_LABEL_SIZE
);
107 sb
->flags
= le64_to_cpu(s
->flags
);
108 sb
->seq
= le64_to_cpu(s
->seq
);
109 sb
->last_mount
= le32_to_cpu(s
->last_mount
);
110 sb
->first_bucket
= le16_to_cpu(s
->first_bucket
);
111 sb
->keys
= le16_to_cpu(s
->keys
);
113 for (i
= 0; i
< SB_JOURNAL_BUCKETS
; i
++)
114 sb
->d
[i
] = le64_to_cpu(s
->d
[i
]);
116 pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u",
117 sb
->version
, sb
->flags
, sb
->seq
, sb
->keys
);
119 err
= "Not a bcache superblock";
120 if (sb
->offset
!= SB_SECTOR
)
123 if (memcmp(sb
->magic
, bcache_magic
, 16))
126 err
= "Too many journal buckets";
127 if (sb
->keys
> SB_JOURNAL_BUCKETS
)
130 err
= "Bad checksum";
131 if (s
->csum
!= csum_set(s
))
135 if (bch_is_zero(sb
->uuid
, 16))
138 sb
->block_size
= le16_to_cpu(s
->block_size
);
140 err
= "Superblock block size smaller than device block size";
141 if (sb
->block_size
<< 9 < bdev_logical_block_size(bdev
))
144 switch (sb
->version
) {
145 case BCACHE_SB_VERSION_BDEV
:
146 sb
->data_offset
= BDEV_DATA_START_DEFAULT
;
148 case BCACHE_SB_VERSION_BDEV_WITH_OFFSET
:
149 sb
->data_offset
= le64_to_cpu(s
->data_offset
);
151 err
= "Bad data offset";
152 if (sb
->data_offset
< BDEV_DATA_START_DEFAULT
)
156 case BCACHE_SB_VERSION_CDEV
:
157 case BCACHE_SB_VERSION_CDEV_WITH_UUID
:
158 sb
->nbuckets
= le64_to_cpu(s
->nbuckets
);
159 sb
->block_size
= le16_to_cpu(s
->block_size
);
160 sb
->bucket_size
= le16_to_cpu(s
->bucket_size
);
162 sb
->nr_in_set
= le16_to_cpu(s
->nr_in_set
);
163 sb
->nr_this_dev
= le16_to_cpu(s
->nr_this_dev
);
165 err
= "Too many buckets";
166 if (sb
->nbuckets
> LONG_MAX
)
169 err
= "Not enough buckets";
170 if (sb
->nbuckets
< 1 << 7)
173 err
= "Bad block/bucket size";
174 if (!is_power_of_2(sb
->block_size
) ||
175 sb
->block_size
> PAGE_SECTORS
||
176 !is_power_of_2(sb
->bucket_size
) ||
177 sb
->bucket_size
< PAGE_SECTORS
)
180 err
= "Invalid superblock: device too small";
181 if (get_capacity(bdev
->bd_disk
) < sb
->bucket_size
* sb
->nbuckets
)
185 if (bch_is_zero(sb
->set_uuid
, 16))
188 err
= "Bad cache device number in set";
189 if (!sb
->nr_in_set
||
190 sb
->nr_in_set
<= sb
->nr_this_dev
||
191 sb
->nr_in_set
> MAX_CACHES_PER_SET
)
194 err
= "Journal buckets not sequential";
195 for (i
= 0; i
< sb
->keys
; i
++)
196 if (sb
->d
[i
] != sb
->first_bucket
+ i
)
199 err
= "Too many journal buckets";
200 if (sb
->first_bucket
+ sb
->keys
> sb
->nbuckets
)
203 err
= "Invalid superblock: first bucket comes before end of super";
204 if (sb
->first_bucket
* sb
->bucket_size
< 16)
209 err
= "Unsupported superblock version";
213 sb
->last_mount
= get_seconds();
216 get_page(bh
->b_page
);
223 static void write_bdev_super_endio(struct bio
*bio
, int error
)
225 struct cached_dev
*dc
= bio
->bi_private
;
226 /* XXX: error checking */
228 closure_put(&dc
->sb_write
.cl
);
231 static void __write_super(struct cache_sb
*sb
, struct bio
*bio
)
233 struct cache_sb
*out
= page_address(bio
->bi_io_vec
[0].bv_page
);
236 bio
->bi_sector
= SB_SECTOR
;
237 bio
->bi_rw
= REQ_SYNC
|REQ_META
;
238 bio
->bi_size
= SB_SIZE
;
239 bch_bio_map(bio
, NULL
);
241 out
->offset
= cpu_to_le64(sb
->offset
);
242 out
->version
= cpu_to_le64(sb
->version
);
244 memcpy(out
->uuid
, sb
->uuid
, 16);
245 memcpy(out
->set_uuid
, sb
->set_uuid
, 16);
246 memcpy(out
->label
, sb
->label
, SB_LABEL_SIZE
);
248 out
->flags
= cpu_to_le64(sb
->flags
);
249 out
->seq
= cpu_to_le64(sb
->seq
);
251 out
->last_mount
= cpu_to_le32(sb
->last_mount
);
252 out
->first_bucket
= cpu_to_le16(sb
->first_bucket
);
253 out
->keys
= cpu_to_le16(sb
->keys
);
255 for (i
= 0; i
< sb
->keys
; i
++)
256 out
->d
[i
] = cpu_to_le64(sb
->d
[i
]);
258 out
->csum
= csum_set(out
);
260 pr_debug("ver %llu, flags %llu, seq %llu",
261 sb
->version
, sb
->flags
, sb
->seq
);
263 submit_bio(REQ_WRITE
, bio
);
266 void bch_write_bdev_super(struct cached_dev
*dc
, struct closure
*parent
)
268 struct closure
*cl
= &dc
->sb_write
.cl
;
269 struct bio
*bio
= &dc
->sb_bio
;
271 closure_lock(&dc
->sb_write
, parent
);
274 bio
->bi_bdev
= dc
->bdev
;
275 bio
->bi_end_io
= write_bdev_super_endio
;
276 bio
->bi_private
= dc
;
279 __write_super(&dc
->sb
, bio
);
284 static void write_super_endio(struct bio
*bio
, int error
)
286 struct cache
*ca
= bio
->bi_private
;
288 bch_count_io_errors(ca
, error
, "writing superblock");
289 closure_put(&ca
->set
->sb_write
.cl
);
292 void bcache_write_super(struct cache_set
*c
)
294 struct closure
*cl
= &c
->sb_write
.cl
;
298 closure_lock(&c
->sb_write
, &c
->cl
);
302 for_each_cache(ca
, c
, i
) {
303 struct bio
*bio
= &ca
->sb_bio
;
305 ca
->sb
.version
= BCACHE_SB_VERSION_CDEV_WITH_UUID
;
306 ca
->sb
.seq
= c
->sb
.seq
;
307 ca
->sb
.last_mount
= c
->sb
.last_mount
;
309 SET_CACHE_SYNC(&ca
->sb
, CACHE_SYNC(&c
->sb
));
312 bio
->bi_bdev
= ca
->bdev
;
313 bio
->bi_end_io
= write_super_endio
;
314 bio
->bi_private
= ca
;
317 __write_super(&ca
->sb
, bio
);
325 static void uuid_endio(struct bio
*bio
, int error
)
327 struct closure
*cl
= bio
->bi_private
;
328 struct cache_set
*c
= container_of(cl
, struct cache_set
, uuid_write
.cl
);
330 cache_set_err_on(error
, c
, "accessing uuids");
331 bch_bbio_free(bio
, c
);
335 static void uuid_io(struct cache_set
*c
, unsigned long rw
,
336 struct bkey
*k
, struct closure
*parent
)
338 struct closure
*cl
= &c
->uuid_write
.cl
;
339 struct uuid_entry
*u
;
344 closure_lock(&c
->uuid_write
, parent
);
346 for (i
= 0; i
< KEY_PTRS(k
); i
++) {
347 struct bio
*bio
= bch_bbio_alloc(c
);
349 bio
->bi_rw
= REQ_SYNC
|REQ_META
|rw
;
350 bio
->bi_size
= KEY_SIZE(k
) << 9;
352 bio
->bi_end_io
= uuid_endio
;
353 bio
->bi_private
= cl
;
354 bch_bio_map(bio
, c
->uuids
);
356 bch_submit_bbio(bio
, c
, k
, i
);
362 bch_bkey_to_text(buf
, sizeof(buf
), k
);
363 pr_debug("%s UUIDs at %s", rw
& REQ_WRITE
? "wrote" : "read", buf
);
365 for (u
= c
->uuids
; u
< c
->uuids
+ c
->nr_uuids
; u
++)
366 if (!bch_is_zero(u
->uuid
, 16))
367 pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u",
368 u
- c
->uuids
, u
->uuid
, u
->label
,
369 u
->first_reg
, u
->last_reg
, u
->invalidated
);
374 static char *uuid_read(struct cache_set
*c
, struct jset
*j
, struct closure
*cl
)
376 struct bkey
*k
= &j
->uuid_bucket
;
378 if (bch_btree_ptr_invalid(c
, k
))
379 return "bad uuid pointer";
381 bkey_copy(&c
->uuid_bucket
, k
);
382 uuid_io(c
, READ_SYNC
, k
, cl
);
384 if (j
->version
< BCACHE_JSET_VERSION_UUIDv1
) {
385 struct uuid_entry_v0
*u0
= (void *) c
->uuids
;
386 struct uuid_entry
*u1
= (void *) c
->uuids
;
392 * Since the new uuid entry is bigger than the old, we have to
393 * convert starting at the highest memory address and work down
394 * in order to do it in place
397 for (i
= c
->nr_uuids
- 1;
400 memcpy(u1
[i
].uuid
, u0
[i
].uuid
, 16);
401 memcpy(u1
[i
].label
, u0
[i
].label
, 32);
403 u1
[i
].first_reg
= u0
[i
].first_reg
;
404 u1
[i
].last_reg
= u0
[i
].last_reg
;
405 u1
[i
].invalidated
= u0
[i
].invalidated
;
415 static int __uuid_write(struct cache_set
*c
)
419 closure_init_stack(&cl
);
421 lockdep_assert_held(&bch_register_lock
);
423 if (bch_bucket_alloc_set(c
, WATERMARK_METADATA
, &k
.key
, 1, true))
426 SET_KEY_SIZE(&k
.key
, c
->sb
.bucket_size
);
427 uuid_io(c
, REQ_WRITE
, &k
.key
, &cl
);
430 bkey_copy(&c
->uuid_bucket
, &k
.key
);
435 int bch_uuid_write(struct cache_set
*c
)
437 int ret
= __uuid_write(c
);
440 bch_journal_meta(c
, NULL
);
445 static struct uuid_entry
*uuid_find(struct cache_set
*c
, const char *uuid
)
447 struct uuid_entry
*u
;
450 u
< c
->uuids
+ c
->nr_uuids
; u
++)
451 if (!memcmp(u
->uuid
, uuid
, 16))
457 static struct uuid_entry
*uuid_find_empty(struct cache_set
*c
)
459 static const char zero_uuid
[16] = "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0";
460 return uuid_find(c
, zero_uuid
);
464 * Bucket priorities/gens:
466 * For each bucket, we store on disk its
470 * See alloc.c for an explanation of the gen. The priority is used to implement
471 * lru (and in the future other) cache replacement policies; for most purposes
472 * it's just an opaque integer.
474 * The gens and the priorities don't have a whole lot to do with each other, and
475 * it's actually the gens that must be written out at specific times - it's no
476 * big deal if the priorities don't get written, if we lose them we just reuse
477 * buckets in suboptimal order.
479 * On disk they're stored in a packed array, and in as many buckets are required
480 * to fit them all. The buckets we use to store them form a list; the journal
481 * header points to the first bucket, the first bucket points to the second
484 * This code is used by the allocation code; periodically (whenever it runs out
485 * of buckets to allocate from) the allocation code will invalidate some
486 * buckets, but it can't use those buckets until their new gens are safely on
490 static void prio_endio(struct bio
*bio
, int error
)
492 struct cache
*ca
= bio
->bi_private
;
494 cache_set_err_on(error
, ca
->set
, "accessing priorities");
495 bch_bbio_free(bio
, ca
->set
);
496 closure_put(&ca
->prio
);
499 static void prio_io(struct cache
*ca
, uint64_t bucket
, unsigned long rw
)
501 struct closure
*cl
= &ca
->prio
;
502 struct bio
*bio
= bch_bbio_alloc(ca
->set
);
504 closure_init_stack(cl
);
506 bio
->bi_sector
= bucket
* ca
->sb
.bucket_size
;
507 bio
->bi_bdev
= ca
->bdev
;
508 bio
->bi_rw
= REQ_SYNC
|REQ_META
|rw
;
509 bio
->bi_size
= bucket_bytes(ca
);
511 bio
->bi_end_io
= prio_endio
;
512 bio
->bi_private
= ca
;
513 bch_bio_map(bio
, ca
->disk_buckets
);
515 closure_bio_submit(bio
, &ca
->prio
, ca
);
519 #define buckets_free(c) "free %zu, free_inc %zu, unused %zu", \
520 fifo_used(&c->free), fifo_used(&c->free_inc), fifo_used(&c->unused)
522 void bch_prio_write(struct cache
*ca
)
528 closure_init_stack(&cl
);
530 lockdep_assert_held(&ca
->set
->bucket_lock
);
532 for (b
= ca
->buckets
;
533 b
< ca
->buckets
+ ca
->sb
.nbuckets
; b
++)
534 b
->disk_gen
= b
->gen
;
536 ca
->disk_buckets
->seq
++;
538 atomic_long_add(ca
->sb
.bucket_size
* prio_buckets(ca
),
539 &ca
->meta_sectors_written
);
541 pr_debug("free %zu, free_inc %zu, unused %zu", fifo_used(&ca
->free
),
542 fifo_used(&ca
->free_inc
), fifo_used(&ca
->unused
));
544 for (i
= prio_buckets(ca
) - 1; i
>= 0; --i
) {
546 struct prio_set
*p
= ca
->disk_buckets
;
547 struct bucket_disk
*d
= p
->data
;
548 struct bucket_disk
*end
= d
+ prios_per_bucket(ca
);
550 for (b
= ca
->buckets
+ i
* prios_per_bucket(ca
);
551 b
< ca
->buckets
+ ca
->sb
.nbuckets
&& d
< end
;
553 d
->prio
= cpu_to_le16(b
->prio
);
557 p
->next_bucket
= ca
->prio_buckets
[i
+ 1];
558 p
->magic
= pset_magic(&ca
->sb
);
559 p
->csum
= bch_crc64(&p
->magic
, bucket_bytes(ca
) - 8);
561 bucket
= bch_bucket_alloc(ca
, WATERMARK_PRIO
, true);
562 BUG_ON(bucket
== -1);
564 mutex_unlock(&ca
->set
->bucket_lock
);
565 prio_io(ca
, bucket
, REQ_WRITE
);
566 mutex_lock(&ca
->set
->bucket_lock
);
568 ca
->prio_buckets
[i
] = bucket
;
569 atomic_dec_bug(&ca
->buckets
[bucket
].pin
);
572 mutex_unlock(&ca
->set
->bucket_lock
);
574 bch_journal_meta(ca
->set
, &cl
);
577 mutex_lock(&ca
->set
->bucket_lock
);
579 ca
->need_save_prio
= 0;
582 * Don't want the old priorities to get garbage collected until after we
583 * finish writing the new ones, and they're journalled
585 for (i
= 0; i
< prio_buckets(ca
); i
++)
586 ca
->prio_last_buckets
[i
] = ca
->prio_buckets
[i
];
589 static void prio_read(struct cache
*ca
, uint64_t bucket
)
591 struct prio_set
*p
= ca
->disk_buckets
;
592 struct bucket_disk
*d
= p
->data
+ prios_per_bucket(ca
), *end
= d
;
594 unsigned bucket_nr
= 0;
596 for (b
= ca
->buckets
;
597 b
< ca
->buckets
+ ca
->sb
.nbuckets
;
600 ca
->prio_buckets
[bucket_nr
] = bucket
;
601 ca
->prio_last_buckets
[bucket_nr
] = bucket
;
604 prio_io(ca
, bucket
, READ_SYNC
);
606 if (p
->csum
!= bch_crc64(&p
->magic
, bucket_bytes(ca
) - 8))
607 pr_warn("bad csum reading priorities");
609 if (p
->magic
!= pset_magic(&ca
->sb
))
610 pr_warn("bad magic reading priorities");
612 bucket
= p
->next_bucket
;
616 b
->prio
= le16_to_cpu(d
->prio
);
617 b
->gen
= b
->disk_gen
= b
->last_gc
= b
->gc_gen
= d
->gen
;
623 static int open_dev(struct block_device
*b
, fmode_t mode
)
625 struct bcache_device
*d
= b
->bd_disk
->private_data
;
626 if (test_bit(BCACHE_DEV_CLOSING
, &d
->flags
))
633 static void release_dev(struct gendisk
*b
, fmode_t mode
)
635 struct bcache_device
*d
= b
->private_data
;
639 static int ioctl_dev(struct block_device
*b
, fmode_t mode
,
640 unsigned int cmd
, unsigned long arg
)
642 struct bcache_device
*d
= b
->bd_disk
->private_data
;
643 return d
->ioctl(d
, mode
, cmd
, arg
);
646 static const struct block_device_operations bcache_ops
= {
648 .release
= release_dev
,
650 .owner
= THIS_MODULE
,
653 void bcache_device_stop(struct bcache_device
*d
)
655 if (!test_and_set_bit(BCACHE_DEV_CLOSING
, &d
->flags
))
656 closure_queue(&d
->cl
);
659 static void bcache_device_unlink(struct bcache_device
*d
)
661 lockdep_assert_held(&bch_register_lock
);
663 if (d
->c
&& !test_and_set_bit(BCACHE_DEV_UNLINK_DONE
, &d
->flags
)) {
667 sysfs_remove_link(&d
->c
->kobj
, d
->name
);
668 sysfs_remove_link(&d
->kobj
, "cache");
670 for_each_cache(ca
, d
->c
, i
)
671 bd_unlink_disk_holder(ca
->bdev
, d
->disk
);
675 static void bcache_device_link(struct bcache_device
*d
, struct cache_set
*c
,
681 for_each_cache(ca
, d
->c
, i
)
682 bd_link_disk_holder(ca
->bdev
, d
->disk
);
684 snprintf(d
->name
, BCACHEDEVNAME_SIZE
,
685 "%s%u", name
, d
->id
);
687 WARN(sysfs_create_link(&d
->kobj
, &c
->kobj
, "cache") ||
688 sysfs_create_link(&c
->kobj
, &d
->kobj
, d
->name
),
689 "Couldn't create device <-> cache set symlinks");
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
,
715 BUG_ON(test_bit(CACHE_SET_STOPPING
, &c
->flags
));
721 closure_get(&c
->caching
);
724 static void bcache_device_free(struct bcache_device
*d
)
726 lockdep_assert_held(&bch_register_lock
);
728 pr_info("%s stopped", d
->disk
->disk_name
);
731 bcache_device_detach(d
);
732 if (d
->disk
&& d
->disk
->flags
& GENHD_FL_UP
)
733 del_gendisk(d
->disk
);
734 if (d
->disk
&& d
->disk
->queue
)
735 blk_cleanup_queue(d
->disk
->queue
);
737 ida_simple_remove(&bcache_minor
, d
->disk
->first_minor
);
741 bio_split_pool_free(&d
->bio_split_hook
);
742 if (d
->unaligned_bvec
)
743 mempool_destroy(d
->unaligned_bvec
);
745 bioset_free(d
->bio_split
);
746 if (is_vmalloc_addr(d
->full_dirty_stripes
))
747 vfree(d
->full_dirty_stripes
);
749 kfree(d
->full_dirty_stripes
);
750 if (is_vmalloc_addr(d
->stripe_sectors_dirty
))
751 vfree(d
->stripe_sectors_dirty
);
753 kfree(d
->stripe_sectors_dirty
);
755 closure_debug_destroy(&d
->cl
);
758 static int bcache_device_init(struct bcache_device
*d
, unsigned block_size
,
761 struct request_queue
*q
;
766 d
->stripe_size
= 1 << 31;
768 d
->nr_stripes
= DIV_ROUND_UP_ULL(sectors
, d
->stripe_size
);
770 if (!d
->nr_stripes
||
771 d
->nr_stripes
> INT_MAX
||
772 d
->nr_stripes
> SIZE_MAX
/ sizeof(atomic_t
)) {
773 pr_err("nr_stripes too large");
777 n
= d
->nr_stripes
* sizeof(atomic_t
);
778 d
->stripe_sectors_dirty
= n
< PAGE_SIZE
<< 6
779 ? kzalloc(n
, GFP_KERNEL
)
781 if (!d
->stripe_sectors_dirty
)
784 n
= BITS_TO_LONGS(d
->nr_stripes
) * sizeof(unsigned long);
785 d
->full_dirty_stripes
= n
< PAGE_SIZE
<< 6
786 ? kzalloc(n
, GFP_KERNEL
)
788 if (!d
->full_dirty_stripes
)
791 minor
= ida_simple_get(&bcache_minor
, 0, MINORMASK
+ 1, GFP_KERNEL
);
795 if (!(d
->bio_split
= bioset_create(4, offsetof(struct bbio
, bio
))) ||
796 !(d
->unaligned_bvec
= mempool_create_kmalloc_pool(1,
797 sizeof(struct bio_vec
) * BIO_MAX_PAGES
)) ||
798 bio_split_pool_init(&d
->bio_split_hook
) ||
799 !(d
->disk
= alloc_disk(1))) {
800 ida_simple_remove(&bcache_minor
, minor
);
804 set_capacity(d
->disk
, sectors
);
805 snprintf(d
->disk
->disk_name
, DISK_NAME_LEN
, "bcache%i", minor
);
807 d
->disk
->major
= bcache_major
;
808 d
->disk
->first_minor
= minor
;
809 d
->disk
->fops
= &bcache_ops
;
810 d
->disk
->private_data
= d
;
812 q
= blk_alloc_queue(GFP_KERNEL
);
816 blk_queue_make_request(q
, NULL
);
819 q
->backing_dev_info
.congested_data
= d
;
820 q
->limits
.max_hw_sectors
= UINT_MAX
;
821 q
->limits
.max_sectors
= UINT_MAX
;
822 q
->limits
.max_segment_size
= UINT_MAX
;
823 q
->limits
.max_segments
= BIO_MAX_PAGES
;
824 q
->limits
.max_discard_sectors
= UINT_MAX
;
825 q
->limits
.io_min
= block_size
;
826 q
->limits
.logical_block_size
= block_size
;
827 q
->limits
.physical_block_size
= block_size
;
828 set_bit(QUEUE_FLAG_NONROT
, &d
->disk
->queue
->queue_flags
);
829 set_bit(QUEUE_FLAG_DISCARD
, &d
->disk
->queue
->queue_flags
);
831 blk_queue_flush(q
, REQ_FLUSH
|REQ_FUA
);
838 static void calc_cached_dev_sectors(struct cache_set
*c
)
840 uint64_t sectors
= 0;
841 struct cached_dev
*dc
;
843 list_for_each_entry(dc
, &c
->cached_devs
, list
)
844 sectors
+= bdev_sectors(dc
->bdev
);
846 c
->cached_dev_sectors
= sectors
;
849 void bch_cached_dev_run(struct cached_dev
*dc
)
851 struct bcache_device
*d
= &dc
->disk
;
852 char buf
[SB_LABEL_SIZE
+ 1];
855 kasprintf(GFP_KERNEL
, "CACHED_UUID=%pU", dc
->sb
.uuid
),
860 memcpy(buf
, dc
->sb
.label
, SB_LABEL_SIZE
);
861 buf
[SB_LABEL_SIZE
] = '\0';
862 env
[2] = kasprintf(GFP_KERNEL
, "CACHED_LABEL=%s", buf
);
864 if (atomic_xchg(&dc
->running
, 1))
868 BDEV_STATE(&dc
->sb
) != BDEV_STATE_NONE
) {
870 closure_init_stack(&cl
);
872 SET_BDEV_STATE(&dc
->sb
, BDEV_STATE_STALE
);
873 bch_write_bdev_super(dc
, &cl
);
878 bd_link_disk_holder(dc
->bdev
, dc
->disk
.disk
);
879 /* won't show up in the uevent file, use udevadm monitor -e instead
880 * only class / kset properties are persistent */
881 kobject_uevent_env(&disk_to_dev(d
->disk
)->kobj
, KOBJ_CHANGE
, env
);
885 if (sysfs_create_link(&d
->kobj
, &disk_to_dev(d
->disk
)->kobj
, "dev") ||
886 sysfs_create_link(&disk_to_dev(d
->disk
)->kobj
, &d
->kobj
, "bcache"))
887 pr_debug("error creating sysfs link");
890 static void cached_dev_detach_finish(struct work_struct
*w
)
892 struct cached_dev
*dc
= container_of(w
, struct cached_dev
, detach
);
893 char buf
[BDEVNAME_SIZE
];
895 closure_init_stack(&cl
);
897 BUG_ON(!test_bit(BCACHE_DEV_DETACHING
, &dc
->disk
.flags
));
898 BUG_ON(atomic_read(&dc
->count
));
900 mutex_lock(&bch_register_lock
);
902 memset(&dc
->sb
.set_uuid
, 0, 16);
903 SET_BDEV_STATE(&dc
->sb
, BDEV_STATE_NONE
);
905 bch_write_bdev_super(dc
, &cl
);
908 bcache_device_detach(&dc
->disk
);
909 list_move(&dc
->list
, &uncached_devices
);
911 clear_bit(BCACHE_DEV_DETACHING
, &dc
->disk
.flags
);
913 mutex_unlock(&bch_register_lock
);
915 pr_info("Caching disabled for %s", bdevname(dc
->bdev
, buf
));
917 /* Drop ref we took in cached_dev_detach() */
918 closure_put(&dc
->disk
.cl
);
921 void bch_cached_dev_detach(struct cached_dev
*dc
)
923 lockdep_assert_held(&bch_register_lock
);
925 if (test_bit(BCACHE_DEV_CLOSING
, &dc
->disk
.flags
))
928 if (test_and_set_bit(BCACHE_DEV_DETACHING
, &dc
->disk
.flags
))
932 * Block the device from being closed and freed until we're finished
935 closure_get(&dc
->disk
.cl
);
937 bch_writeback_queue(dc
);
941 int bch_cached_dev_attach(struct cached_dev
*dc
, struct cache_set
*c
)
943 uint32_t rtime
= cpu_to_le32(get_seconds());
944 struct uuid_entry
*u
;
945 char buf
[BDEVNAME_SIZE
];
947 bdevname(dc
->bdev
, buf
);
949 if (memcmp(dc
->sb
.set_uuid
, c
->sb
.set_uuid
, 16))
953 pr_err("Can't attach %s: already attached", buf
);
957 if (test_bit(CACHE_SET_STOPPING
, &c
->flags
)) {
958 pr_err("Can't attach %s: shutting down", buf
);
962 if (dc
->sb
.block_size
< c
->sb
.block_size
) {
964 pr_err("Couldn't attach %s: block size less than set's block size",
969 u
= uuid_find(c
, dc
->sb
.uuid
);
972 (BDEV_STATE(&dc
->sb
) == BDEV_STATE_STALE
||
973 BDEV_STATE(&dc
->sb
) == BDEV_STATE_NONE
)) {
974 memcpy(u
->uuid
, invalid_uuid
, 16);
975 u
->invalidated
= cpu_to_le32(get_seconds());
980 if (BDEV_STATE(&dc
->sb
) == BDEV_STATE_DIRTY
) {
981 pr_err("Couldn't find uuid for %s in set", buf
);
985 u
= uuid_find_empty(c
);
987 pr_err("Not caching %s, no room for UUID", buf
);
992 /* Deadlocks since we're called via sysfs...
993 sysfs_remove_file(&dc->kobj, &sysfs_attach);
996 if (bch_is_zero(u
->uuid
, 16)) {
998 closure_init_stack(&cl
);
1000 memcpy(u
->uuid
, dc
->sb
.uuid
, 16);
1001 memcpy(u
->label
, dc
->sb
.label
, SB_LABEL_SIZE
);
1002 u
->first_reg
= u
->last_reg
= rtime
;
1005 memcpy(dc
->sb
.set_uuid
, c
->sb
.set_uuid
, 16);
1006 SET_BDEV_STATE(&dc
->sb
, BDEV_STATE_CLEAN
);
1008 bch_write_bdev_super(dc
, &cl
);
1011 u
->last_reg
= rtime
;
1015 bcache_device_attach(&dc
->disk
, c
, u
- c
->uuids
);
1016 list_move(&dc
->list
, &c
->cached_devs
);
1017 calc_cached_dev_sectors(c
);
1021 * dc->c must be set before dc->count != 0 - paired with the mb in
1024 atomic_set(&dc
->count
, 1);
1026 if (BDEV_STATE(&dc
->sb
) == BDEV_STATE_DIRTY
) {
1027 bch_sectors_dirty_init(dc
);
1028 atomic_set(&dc
->has_dirty
, 1);
1029 atomic_inc(&dc
->count
);
1030 bch_writeback_queue(dc
);
1033 bch_cached_dev_run(dc
);
1034 bcache_device_link(&dc
->disk
, c
, "bdev");
1036 pr_info("Caching %s as %s on set %pU",
1037 bdevname(dc
->bdev
, buf
), dc
->disk
.disk
->disk_name
,
1038 dc
->disk
.c
->sb
.set_uuid
);
1042 void bch_cached_dev_release(struct kobject
*kobj
)
1044 struct cached_dev
*dc
= container_of(kobj
, struct cached_dev
,
1047 module_put(THIS_MODULE
);
1050 static void cached_dev_free(struct closure
*cl
)
1052 struct cached_dev
*dc
= container_of(cl
, struct cached_dev
, disk
.cl
);
1054 cancel_delayed_work_sync(&dc
->writeback_rate_update
);
1055 kthread_stop(dc
->writeback_thread
);
1057 mutex_lock(&bch_register_lock
);
1059 if (atomic_read(&dc
->running
))
1060 bd_unlink_disk_holder(dc
->bdev
, dc
->disk
.disk
);
1061 bcache_device_free(&dc
->disk
);
1062 list_del(&dc
->list
);
1064 mutex_unlock(&bch_register_lock
);
1066 if (!IS_ERR_OR_NULL(dc
->bdev
)) {
1067 if (dc
->bdev
->bd_disk
)
1068 blk_sync_queue(bdev_get_queue(dc
->bdev
));
1070 blkdev_put(dc
->bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
1073 wake_up(&unregister_wait
);
1075 kobject_put(&dc
->disk
.kobj
);
1078 static void cached_dev_flush(struct closure
*cl
)
1080 struct cached_dev
*dc
= container_of(cl
, struct cached_dev
, disk
.cl
);
1081 struct bcache_device
*d
= &dc
->disk
;
1083 mutex_lock(&bch_register_lock
);
1084 bcache_device_unlink(d
);
1085 mutex_unlock(&bch_register_lock
);
1087 bch_cache_accounting_destroy(&dc
->accounting
);
1088 kobject_del(&d
->kobj
);
1090 continue_at(cl
, cached_dev_free
, system_wq
);
1093 static int cached_dev_init(struct cached_dev
*dc
, unsigned block_size
)
1097 struct request_queue
*q
= bdev_get_queue(dc
->bdev
);
1099 __module_get(THIS_MODULE
);
1100 INIT_LIST_HEAD(&dc
->list
);
1101 closure_init(&dc
->disk
.cl
, NULL
);
1102 set_closure_fn(&dc
->disk
.cl
, cached_dev_flush
, system_wq
);
1103 kobject_init(&dc
->disk
.kobj
, &bch_cached_dev_ktype
);
1104 INIT_WORK(&dc
->detach
, cached_dev_detach_finish
);
1105 closure_init_unlocked(&dc
->sb_write
);
1106 INIT_LIST_HEAD(&dc
->io_lru
);
1107 spin_lock_init(&dc
->io_lock
);
1108 bch_cache_accounting_init(&dc
->accounting
, &dc
->disk
.cl
);
1110 dc
->sequential_cutoff
= 4 << 20;
1112 for (io
= dc
->io
; io
< dc
->io
+ RECENT_IO
; io
++) {
1113 list_add(&io
->lru
, &dc
->io_lru
);
1114 hlist_add_head(&io
->hash
, dc
->io_hash
+ RECENT_IO
);
1117 ret
= bcache_device_init(&dc
->disk
, block_size
,
1118 dc
->bdev
->bd_part
->nr_sects
- dc
->sb
.data_offset
);
1122 set_capacity(dc
->disk
.disk
,
1123 dc
->bdev
->bd_part
->nr_sects
- dc
->sb
.data_offset
);
1125 dc
->disk
.disk
->queue
->backing_dev_info
.ra_pages
=
1126 max(dc
->disk
.disk
->queue
->backing_dev_info
.ra_pages
,
1127 q
->backing_dev_info
.ra_pages
);
1129 bch_cached_dev_request_init(dc
);
1130 bch_cached_dev_writeback_init(dc
);
1134 /* Cached device - bcache superblock */
1136 static void register_bdev(struct cache_sb
*sb
, struct page
*sb_page
,
1137 struct block_device
*bdev
,
1138 struct cached_dev
*dc
)
1140 char name
[BDEVNAME_SIZE
];
1141 const char *err
= "cannot allocate memory";
1142 struct cache_set
*c
;
1144 memcpy(&dc
->sb
, sb
, sizeof(struct cache_sb
));
1146 dc
->bdev
->bd_holder
= dc
;
1148 bio_init(&dc
->sb_bio
);
1149 dc
->sb_bio
.bi_max_vecs
= 1;
1150 dc
->sb_bio
.bi_io_vec
= dc
->sb_bio
.bi_inline_vecs
;
1151 dc
->sb_bio
.bi_io_vec
[0].bv_page
= sb_page
;
1154 if (cached_dev_init(dc
, sb
->block_size
<< 9))
1157 err
= "error creating kobject";
1158 if (kobject_add(&dc
->disk
.kobj
, &part_to_dev(bdev
->bd_part
)->kobj
,
1161 if (bch_cache_accounting_add_kobjs(&dc
->accounting
, &dc
->disk
.kobj
))
1164 pr_info("registered backing device %s", bdevname(bdev
, name
));
1166 list_add(&dc
->list
, &uncached_devices
);
1167 list_for_each_entry(c
, &bch_cache_sets
, list
)
1168 bch_cached_dev_attach(dc
, c
);
1170 if (BDEV_STATE(&dc
->sb
) == BDEV_STATE_NONE
||
1171 BDEV_STATE(&dc
->sb
) == BDEV_STATE_STALE
)
1172 bch_cached_dev_run(dc
);
1176 pr_notice("error opening %s: %s", bdevname(bdev
, name
), err
);
1177 bcache_device_stop(&dc
->disk
);
1180 /* Flash only volumes */
1182 void bch_flash_dev_release(struct kobject
*kobj
)
1184 struct bcache_device
*d
= container_of(kobj
, struct bcache_device
,
1189 static void flash_dev_free(struct closure
*cl
)
1191 struct bcache_device
*d
= container_of(cl
, struct bcache_device
, cl
);
1192 bcache_device_free(d
);
1193 kobject_put(&d
->kobj
);
1196 static void flash_dev_flush(struct closure
*cl
)
1198 struct bcache_device
*d
= container_of(cl
, struct bcache_device
, cl
);
1200 bcache_device_unlink(d
);
1201 kobject_del(&d
->kobj
);
1202 continue_at(cl
, flash_dev_free
, system_wq
);
1205 static int flash_dev_run(struct cache_set
*c
, struct uuid_entry
*u
)
1207 struct bcache_device
*d
= kzalloc(sizeof(struct bcache_device
),
1212 closure_init(&d
->cl
, NULL
);
1213 set_closure_fn(&d
->cl
, flash_dev_flush
, system_wq
);
1215 kobject_init(&d
->kobj
, &bch_flash_dev_ktype
);
1217 if (bcache_device_init(d
, block_bytes(c
), u
->sectors
))
1220 bcache_device_attach(d
, c
, u
- c
->uuids
);
1221 bch_flash_dev_request_init(d
);
1224 if (kobject_add(&d
->kobj
, &disk_to_dev(d
->disk
)->kobj
, "bcache"))
1227 bcache_device_link(d
, c
, "volume");
1231 kobject_put(&d
->kobj
);
1235 static int flash_devs_run(struct cache_set
*c
)
1238 struct uuid_entry
*u
;
1241 u
< c
->uuids
+ c
->nr_uuids
&& !ret
;
1243 if (UUID_FLASH_ONLY(u
))
1244 ret
= flash_dev_run(c
, u
);
1249 int bch_flash_dev_create(struct cache_set
*c
, uint64_t size
)
1251 struct uuid_entry
*u
;
1253 if (test_bit(CACHE_SET_STOPPING
, &c
->flags
))
1256 u
= uuid_find_empty(c
);
1258 pr_err("Can't create volume, no room for UUID");
1262 get_random_bytes(u
->uuid
, 16);
1263 memset(u
->label
, 0, 32);
1264 u
->first_reg
= u
->last_reg
= cpu_to_le32(get_seconds());
1266 SET_UUID_FLASH_ONLY(u
, 1);
1267 u
->sectors
= size
>> 9;
1271 return flash_dev_run(c
, u
);
1277 bool bch_cache_set_error(struct cache_set
*c
, const char *fmt
, ...)
1281 if (c
->on_error
!= ON_ERROR_PANIC
&&
1282 test_bit(CACHE_SET_STOPPING
, &c
->flags
))
1285 /* XXX: we can be called from atomic context
1286 acquire_console_sem();
1289 printk(KERN_ERR
"bcache: error on %pU: ", c
->sb
.set_uuid
);
1291 va_start(args
, fmt
);
1295 printk(", disabling caching\n");
1297 if (c
->on_error
== ON_ERROR_PANIC
)
1298 panic("panic forced after error\n");
1300 bch_cache_set_unregister(c
);
1304 void bch_cache_set_release(struct kobject
*kobj
)
1306 struct cache_set
*c
= container_of(kobj
, struct cache_set
, kobj
);
1308 module_put(THIS_MODULE
);
1311 static void cache_set_free(struct closure
*cl
)
1313 struct cache_set
*c
= container_of(cl
, struct cache_set
, cl
);
1317 if (!IS_ERR_OR_NULL(c
->debug
))
1318 debugfs_remove(c
->debug
);
1320 bch_open_buckets_free(c
);
1321 bch_btree_cache_free(c
);
1322 bch_journal_free(c
);
1324 for_each_cache(ca
, c
, i
)
1326 kobject_put(&ca
->kobj
);
1328 free_pages((unsigned long) c
->uuids
, ilog2(bucket_pages(c
)));
1329 free_pages((unsigned long) c
->sort
, ilog2(bucket_pages(c
)));
1332 bioset_free(c
->bio_split
);
1334 mempool_destroy(c
->fill_iter
);
1336 mempool_destroy(c
->bio_meta
);
1338 mempool_destroy(c
->search
);
1341 mutex_lock(&bch_register_lock
);
1343 mutex_unlock(&bch_register_lock
);
1345 pr_info("Cache set %pU unregistered", c
->sb
.set_uuid
);
1346 wake_up(&unregister_wait
);
1348 closure_debug_destroy(&c
->cl
);
1349 kobject_put(&c
->kobj
);
1352 static void cache_set_flush(struct closure
*cl
)
1354 struct cache_set
*c
= container_of(cl
, struct cache_set
, caching
);
1359 bch_cache_accounting_destroy(&c
->accounting
);
1361 kobject_put(&c
->internal
);
1362 kobject_del(&c
->kobj
);
1365 kthread_stop(c
->gc_thread
);
1367 if (!IS_ERR_OR_NULL(c
->root
))
1368 list_add(&c
->root
->list
, &c
->btree_cache
);
1370 /* Should skip this if we're unregistering because of an error */
1371 list_for_each_entry(b
, &c
->btree_cache
, list
)
1372 if (btree_node_dirty(b
))
1373 bch_btree_node_write(b
, NULL
);
1375 for_each_cache(ca
, c
, i
)
1376 if (ca
->alloc_thread
)
1377 kthread_stop(ca
->alloc_thread
);
1382 static void __cache_set_unregister(struct closure
*cl
)
1384 struct cache_set
*c
= container_of(cl
, struct cache_set
, caching
);
1385 struct cached_dev
*dc
;
1388 mutex_lock(&bch_register_lock
);
1390 for (i
= 0; i
< c
->nr_uuids
; i
++)
1391 if (c
->devices
[i
]) {
1392 if (!UUID_FLASH_ONLY(&c
->uuids
[i
]) &&
1393 test_bit(CACHE_SET_UNREGISTERING
, &c
->flags
)) {
1394 dc
= container_of(c
->devices
[i
],
1395 struct cached_dev
, disk
);
1396 bch_cached_dev_detach(dc
);
1398 bcache_device_stop(c
->devices
[i
]);
1402 mutex_unlock(&bch_register_lock
);
1404 continue_at(cl
, cache_set_flush
, system_wq
);
1407 void bch_cache_set_stop(struct cache_set
*c
)
1409 if (!test_and_set_bit(CACHE_SET_STOPPING
, &c
->flags
))
1410 closure_queue(&c
->caching
);
1413 void bch_cache_set_unregister(struct cache_set
*c
)
1415 set_bit(CACHE_SET_UNREGISTERING
, &c
->flags
);
1416 bch_cache_set_stop(c
);
1419 #define alloc_bucket_pages(gfp, c) \
1420 ((void *) __get_free_pages(__GFP_ZERO|gfp, ilog2(bucket_pages(c))))
1422 struct cache_set
*bch_cache_set_alloc(struct cache_sb
*sb
)
1425 struct cache_set
*c
= kzalloc(sizeof(struct cache_set
), GFP_KERNEL
);
1429 __module_get(THIS_MODULE
);
1430 closure_init(&c
->cl
, NULL
);
1431 set_closure_fn(&c
->cl
, cache_set_free
, system_wq
);
1433 closure_init(&c
->caching
, &c
->cl
);
1434 set_closure_fn(&c
->caching
, __cache_set_unregister
, system_wq
);
1436 /* Maybe create continue_at_noreturn() and use it here? */
1437 closure_set_stopped(&c
->cl
);
1438 closure_put(&c
->cl
);
1440 kobject_init(&c
->kobj
, &bch_cache_set_ktype
);
1441 kobject_init(&c
->internal
, &bch_cache_set_internal_ktype
);
1443 bch_cache_accounting_init(&c
->accounting
, &c
->cl
);
1445 memcpy(c
->sb
.set_uuid
, sb
->set_uuid
, 16);
1446 c
->sb
.block_size
= sb
->block_size
;
1447 c
->sb
.bucket_size
= sb
->bucket_size
;
1448 c
->sb
.nr_in_set
= sb
->nr_in_set
;
1449 c
->sb
.last_mount
= sb
->last_mount
;
1450 c
->bucket_bits
= ilog2(sb
->bucket_size
);
1451 c
->block_bits
= ilog2(sb
->block_size
);
1452 c
->nr_uuids
= bucket_bytes(c
) / sizeof(struct uuid_entry
);
1454 c
->btree_pages
= c
->sb
.bucket_size
/ PAGE_SECTORS
;
1455 if (c
->btree_pages
> BTREE_MAX_PAGES
)
1456 c
->btree_pages
= max_t(int, c
->btree_pages
/ 4,
1459 c
->sort_crit_factor
= int_sqrt(c
->btree_pages
);
1461 closure_init_unlocked(&c
->sb_write
);
1462 mutex_init(&c
->bucket_lock
);
1463 init_waitqueue_head(&c
->try_wait
);
1464 init_waitqueue_head(&c
->bucket_wait
);
1465 closure_init_unlocked(&c
->uuid_write
);
1466 mutex_init(&c
->sort_lock
);
1468 spin_lock_init(&c
->sort_time
.lock
);
1469 spin_lock_init(&c
->btree_gc_time
.lock
);
1470 spin_lock_init(&c
->btree_split_time
.lock
);
1471 spin_lock_init(&c
->btree_read_time
.lock
);
1472 spin_lock_init(&c
->try_harder_time
.lock
);
1474 bch_moving_init_cache_set(c
);
1476 INIT_LIST_HEAD(&c
->list
);
1477 INIT_LIST_HEAD(&c
->cached_devs
);
1478 INIT_LIST_HEAD(&c
->btree_cache
);
1479 INIT_LIST_HEAD(&c
->btree_cache_freeable
);
1480 INIT_LIST_HEAD(&c
->btree_cache_freed
);
1481 INIT_LIST_HEAD(&c
->data_buckets
);
1483 c
->search
= mempool_create_slab_pool(32, bch_search_cache
);
1487 iter_size
= (sb
->bucket_size
/ sb
->block_size
+ 1) *
1488 sizeof(struct btree_iter_set
);
1490 if (!(c
->devices
= kzalloc(c
->nr_uuids
* sizeof(void *), GFP_KERNEL
)) ||
1491 !(c
->bio_meta
= mempool_create_kmalloc_pool(2,
1492 sizeof(struct bbio
) + sizeof(struct bio_vec
) *
1493 bucket_pages(c
))) ||
1494 !(c
->fill_iter
= mempool_create_kmalloc_pool(1, iter_size
)) ||
1495 !(c
->bio_split
= bioset_create(4, offsetof(struct bbio
, bio
))) ||
1496 !(c
->sort
= alloc_bucket_pages(GFP_KERNEL
, c
)) ||
1497 !(c
->uuids
= alloc_bucket_pages(GFP_KERNEL
, c
)) ||
1498 bch_journal_alloc(c
) ||
1499 bch_btree_cache_alloc(c
) ||
1500 bch_open_buckets_alloc(c
))
1503 c
->congested_read_threshold_us
= 2000;
1504 c
->congested_write_threshold_us
= 20000;
1505 c
->error_limit
= 8 << IO_ERROR_SHIFT
;
1509 bch_cache_set_unregister(c
);
1513 static void run_cache_set(struct cache_set
*c
)
1515 const char *err
= "cannot allocate memory";
1516 struct cached_dev
*dc
, *t
;
1521 closure_init_stack(&cl
);
1523 for_each_cache(ca
, c
, i
)
1524 c
->nbuckets
+= ca
->sb
.nbuckets
;
1526 if (CACHE_SYNC(&c
->sb
)) {
1531 err
= "cannot allocate memory for journal";
1532 if (bch_journal_read(c
, &journal
))
1535 pr_debug("btree_journal_read() done");
1537 err
= "no journal entries found";
1538 if (list_empty(&journal
))
1541 j
= &list_entry(journal
.prev
, struct journal_replay
, list
)->j
;
1543 err
= "IO error reading priorities";
1544 for_each_cache(ca
, c
, i
)
1545 prio_read(ca
, j
->prio_bucket
[ca
->sb
.nr_this_dev
]);
1548 * If prio_read() fails it'll call cache_set_error and we'll
1549 * tear everything down right away, but if we perhaps checked
1550 * sooner we could avoid journal replay.
1555 err
= "bad btree root";
1556 if (bch_btree_ptr_invalid(c
, k
))
1559 err
= "error reading btree root";
1560 c
->root
= bch_btree_node_get(c
, k
, j
->btree_level
, true);
1561 if (IS_ERR_OR_NULL(c
->root
))
1564 list_del_init(&c
->root
->list
);
1565 rw_unlock(true, c
->root
);
1567 err
= uuid_read(c
, j
, &cl
);
1571 err
= "error in recovery";
1572 if (bch_btree_check(c
))
1575 bch_journal_mark(c
, &journal
);
1576 bch_btree_gc_finish(c
);
1577 pr_debug("btree_check() done");
1580 * bcache_journal_next() can't happen sooner, or
1581 * btree_gc_finish() will give spurious errors about last_gc >
1582 * gc_gen - this is a hack but oh well.
1584 bch_journal_next(&c
->journal
);
1586 err
= "error starting allocator thread";
1587 for_each_cache(ca
, c
, i
)
1588 if (bch_cache_allocator_start(ca
))
1592 * First place it's safe to allocate: btree_check() and
1593 * btree_gc_finish() have to run before we have buckets to
1594 * allocate, and bch_bucket_alloc_set() might cause a journal
1595 * entry to be written so bcache_journal_next() has to be called
1598 * If the uuids were in the old format we have to rewrite them
1599 * before the next journal entry is written:
1601 if (j
->version
< BCACHE_JSET_VERSION_UUID
)
1604 bch_journal_replay(c
, &journal
);
1606 pr_notice("invalidating existing data");
1608 for_each_cache(ca
, c
, i
) {
1611 ca
->sb
.keys
= clamp_t(int, ca
->sb
.nbuckets
>> 7,
1612 2, SB_JOURNAL_BUCKETS
);
1614 for (j
= 0; j
< ca
->sb
.keys
; j
++)
1615 ca
->sb
.d
[j
] = ca
->sb
.first_bucket
+ j
;
1618 bch_btree_gc_finish(c
);
1620 err
= "error starting allocator thread";
1621 for_each_cache(ca
, c
, i
)
1622 if (bch_cache_allocator_start(ca
))
1625 mutex_lock(&c
->bucket_lock
);
1626 for_each_cache(ca
, c
, i
)
1628 mutex_unlock(&c
->bucket_lock
);
1630 err
= "cannot allocate new UUID bucket";
1631 if (__uuid_write(c
))
1634 err
= "cannot allocate new btree root";
1635 c
->root
= bch_btree_node_alloc(c
, 0, true);
1636 if (IS_ERR_OR_NULL(c
->root
))
1639 bkey_copy_key(&c
->root
->key
, &MAX_KEY
);
1640 bch_btree_node_write(c
->root
, &cl
);
1642 bch_btree_set_root(c
->root
);
1643 rw_unlock(true, c
->root
);
1646 * We don't want to write the first journal entry until
1647 * everything is set up - fortunately journal entries won't be
1648 * written until the SET_CACHE_SYNC() here:
1650 SET_CACHE_SYNC(&c
->sb
, true);
1652 bch_journal_next(&c
->journal
);
1653 bch_journal_meta(c
, &cl
);
1656 err
= "error starting gc thread";
1657 if (bch_gc_thread_start(c
))
1661 c
->sb
.last_mount
= get_seconds();
1662 bcache_write_super(c
);
1664 list_for_each_entry_safe(dc
, t
, &uncached_devices
, list
)
1665 bch_cached_dev_attach(dc
, c
);
1672 /* XXX: test this, it's broken */
1673 bch_cache_set_error(c
, "%s", err
);
1676 static bool can_attach_cache(struct cache
*ca
, struct cache_set
*c
)
1678 return ca
->sb
.block_size
== c
->sb
.block_size
&&
1679 ca
->sb
.bucket_size
== c
->sb
.block_size
&&
1680 ca
->sb
.nr_in_set
== c
->sb
.nr_in_set
;
1683 static const char *register_cache_set(struct cache
*ca
)
1686 const char *err
= "cannot allocate memory";
1687 struct cache_set
*c
;
1689 list_for_each_entry(c
, &bch_cache_sets
, list
)
1690 if (!memcmp(c
->sb
.set_uuid
, ca
->sb
.set_uuid
, 16)) {
1691 if (c
->cache
[ca
->sb
.nr_this_dev
])
1692 return "duplicate cache set member";
1694 if (!can_attach_cache(ca
, c
))
1695 return "cache sb does not match set";
1697 if (!CACHE_SYNC(&ca
->sb
))
1698 SET_CACHE_SYNC(&c
->sb
, false);
1703 c
= bch_cache_set_alloc(&ca
->sb
);
1707 err
= "error creating kobject";
1708 if (kobject_add(&c
->kobj
, bcache_kobj
, "%pU", c
->sb
.set_uuid
) ||
1709 kobject_add(&c
->internal
, &c
->kobj
, "internal"))
1712 if (bch_cache_accounting_add_kobjs(&c
->accounting
, &c
->kobj
))
1715 bch_debug_init_cache_set(c
);
1717 list_add(&c
->list
, &bch_cache_sets
);
1719 sprintf(buf
, "cache%i", ca
->sb
.nr_this_dev
);
1720 if (sysfs_create_link(&ca
->kobj
, &c
->kobj
, "set") ||
1721 sysfs_create_link(&c
->kobj
, &ca
->kobj
, buf
))
1724 if (ca
->sb
.seq
> c
->sb
.seq
) {
1725 c
->sb
.version
= ca
->sb
.version
;
1726 memcpy(c
->sb
.set_uuid
, ca
->sb
.set_uuid
, 16);
1727 c
->sb
.flags
= ca
->sb
.flags
;
1728 c
->sb
.seq
= ca
->sb
.seq
;
1729 pr_debug("set version = %llu", c
->sb
.version
);
1733 ca
->set
->cache
[ca
->sb
.nr_this_dev
] = ca
;
1734 c
->cache_by_alloc
[c
->caches_loaded
++] = ca
;
1736 if (c
->caches_loaded
== c
->sb
.nr_in_set
)
1741 bch_cache_set_unregister(c
);
1747 void bch_cache_release(struct kobject
*kobj
)
1749 struct cache
*ca
= container_of(kobj
, struct cache
, kobj
);
1752 ca
->set
->cache
[ca
->sb
.nr_this_dev
] = NULL
;
1754 bio_split_pool_free(&ca
->bio_split_hook
);
1756 free_pages((unsigned long) ca
->disk_buckets
, ilog2(bucket_pages(ca
)));
1757 kfree(ca
->prio_buckets
);
1760 free_heap(&ca
->heap
);
1761 free_fifo(&ca
->unused
);
1762 free_fifo(&ca
->free_inc
);
1763 free_fifo(&ca
->free
);
1765 if (ca
->sb_bio
.bi_inline_vecs
[0].bv_page
)
1766 put_page(ca
->sb_bio
.bi_io_vec
[0].bv_page
);
1768 if (!IS_ERR_OR_NULL(ca
->bdev
)) {
1769 blk_sync_queue(bdev_get_queue(ca
->bdev
));
1770 blkdev_put(ca
->bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
1774 module_put(THIS_MODULE
);
1777 static int cache_alloc(struct cache_sb
*sb
, struct cache
*ca
)
1782 __module_get(THIS_MODULE
);
1783 kobject_init(&ca
->kobj
, &bch_cache_ktype
);
1785 bio_init(&ca
->journal
.bio
);
1786 ca
->journal
.bio
.bi_max_vecs
= 8;
1787 ca
->journal
.bio
.bi_io_vec
= ca
->journal
.bio
.bi_inline_vecs
;
1789 free
= roundup_pow_of_two(ca
->sb
.nbuckets
) >> 9;
1790 free
= max_t(size_t, free
, (prio_buckets(ca
) + 8) * 2);
1792 if (!init_fifo(&ca
->free
, free
, GFP_KERNEL
) ||
1793 !init_fifo(&ca
->free_inc
, free
<< 2, GFP_KERNEL
) ||
1794 !init_fifo(&ca
->unused
, free
<< 2, GFP_KERNEL
) ||
1795 !init_heap(&ca
->heap
, free
<< 3, GFP_KERNEL
) ||
1796 !(ca
->buckets
= vzalloc(sizeof(struct bucket
) *
1797 ca
->sb
.nbuckets
)) ||
1798 !(ca
->prio_buckets
= kzalloc(sizeof(uint64_t) * prio_buckets(ca
) *
1800 !(ca
->disk_buckets
= alloc_bucket_pages(GFP_KERNEL
, ca
)) ||
1801 bio_split_pool_init(&ca
->bio_split_hook
))
1804 ca
->prio_last_buckets
= ca
->prio_buckets
+ prio_buckets(ca
);
1806 for_each_bucket(b
, ca
)
1807 atomic_set(&b
->pin
, 0);
1809 if (bch_cache_allocator_init(ca
))
1814 kobject_put(&ca
->kobj
);
1818 static void register_cache(struct cache_sb
*sb
, struct page
*sb_page
,
1819 struct block_device
*bdev
, struct cache
*ca
)
1821 char name
[BDEVNAME_SIZE
];
1822 const char *err
= "cannot allocate memory";
1824 memcpy(&ca
->sb
, sb
, sizeof(struct cache_sb
));
1826 ca
->bdev
->bd_holder
= ca
;
1828 bio_init(&ca
->sb_bio
);
1829 ca
->sb_bio
.bi_max_vecs
= 1;
1830 ca
->sb_bio
.bi_io_vec
= ca
->sb_bio
.bi_inline_vecs
;
1831 ca
->sb_bio
.bi_io_vec
[0].bv_page
= sb_page
;
1834 if (blk_queue_discard(bdev_get_queue(ca
->bdev
)))
1835 ca
->discard
= CACHE_DISCARD(&ca
->sb
);
1837 if (cache_alloc(sb
, ca
) != 0)
1840 err
= "error creating kobject";
1841 if (kobject_add(&ca
->kobj
, &part_to_dev(bdev
->bd_part
)->kobj
, "bcache"))
1844 err
= register_cache_set(ca
);
1848 pr_info("registered cache device %s", bdevname(bdev
, name
));
1851 pr_notice("error opening %s: %s", bdevname(bdev
, name
), err
);
1852 kobject_put(&ca
->kobj
);
1855 /* Global interfaces/init */
1857 static ssize_t
register_bcache(struct kobject
*, struct kobj_attribute
*,
1858 const char *, size_t);
1860 kobj_attribute_write(register, register_bcache
);
1861 kobj_attribute_write(register_quiet
, register_bcache
);
1863 static bool bch_is_open_backing(struct block_device
*bdev
) {
1864 struct cache_set
*c
, *tc
;
1865 struct cached_dev
*dc
, *t
;
1867 list_for_each_entry_safe(c
, tc
, &bch_cache_sets
, list
)
1868 list_for_each_entry_safe(dc
, t
, &c
->cached_devs
, list
)
1869 if (dc
->bdev
== bdev
)
1871 list_for_each_entry_safe(dc
, t
, &uncached_devices
, list
)
1872 if (dc
->bdev
== bdev
)
1877 static bool bch_is_open_cache(struct block_device
*bdev
) {
1878 struct cache_set
*c
, *tc
;
1882 list_for_each_entry_safe(c
, tc
, &bch_cache_sets
, list
)
1883 for_each_cache(ca
, c
, i
)
1884 if (ca
->bdev
== bdev
)
1889 static bool bch_is_open(struct block_device
*bdev
) {
1890 return bch_is_open_cache(bdev
) || bch_is_open_backing(bdev
);
1893 static ssize_t
register_bcache(struct kobject
*k
, struct kobj_attribute
*attr
,
1894 const char *buffer
, size_t size
)
1897 const char *err
= "cannot allocate memory";
1899 struct cache_sb
*sb
= NULL
;
1900 struct block_device
*bdev
= NULL
;
1901 struct page
*sb_page
= NULL
;
1903 if (!try_module_get(THIS_MODULE
))
1906 mutex_lock(&bch_register_lock
);
1908 if (!(path
= kstrndup(buffer
, size
, GFP_KERNEL
)) ||
1909 !(sb
= kmalloc(sizeof(struct cache_sb
), GFP_KERNEL
)))
1912 err
= "failed to open device";
1913 bdev
= blkdev_get_by_path(strim(path
),
1914 FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
,
1917 if (bdev
== ERR_PTR(-EBUSY
)) {
1918 bdev
= lookup_bdev(strim(path
));
1919 if (!IS_ERR(bdev
) && bch_is_open(bdev
))
1920 err
= "device already registered";
1922 err
= "device busy";
1927 err
= "failed to set blocksize";
1928 if (set_blocksize(bdev
, 4096))
1931 err
= read_super(sb
, bdev
, &sb_page
);
1935 if (SB_IS_BDEV(sb
)) {
1936 struct cached_dev
*dc
= kzalloc(sizeof(*dc
), GFP_KERNEL
);
1940 register_bdev(sb
, sb_page
, bdev
, dc
);
1942 struct cache
*ca
= kzalloc(sizeof(*ca
), GFP_KERNEL
);
1946 register_cache(sb
, sb_page
, bdev
, ca
);
1953 mutex_unlock(&bch_register_lock
);
1954 module_put(THIS_MODULE
);
1958 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
1960 if (attr
!= &ksysfs_register_quiet
)
1961 pr_info("error opening %s: %s", path
, err
);
1966 static int bcache_reboot(struct notifier_block
*n
, unsigned long code
, void *x
)
1968 if (code
== SYS_DOWN
||
1970 code
== SYS_POWER_OFF
) {
1972 unsigned long start
= jiffies
;
1973 bool stopped
= false;
1975 struct cache_set
*c
, *tc
;
1976 struct cached_dev
*dc
, *tdc
;
1978 mutex_lock(&bch_register_lock
);
1980 if (list_empty(&bch_cache_sets
) &&
1981 list_empty(&uncached_devices
))
1984 pr_info("Stopping all devices:");
1986 list_for_each_entry_safe(c
, tc
, &bch_cache_sets
, list
)
1987 bch_cache_set_stop(c
);
1989 list_for_each_entry_safe(dc
, tdc
, &uncached_devices
, list
)
1990 bcache_device_stop(&dc
->disk
);
1992 /* What's a condition variable? */
1994 long timeout
= start
+ 2 * HZ
- jiffies
;
1996 stopped
= list_empty(&bch_cache_sets
) &&
1997 list_empty(&uncached_devices
);
1999 if (timeout
< 0 || stopped
)
2002 prepare_to_wait(&unregister_wait
, &wait
,
2003 TASK_UNINTERRUPTIBLE
);
2005 mutex_unlock(&bch_register_lock
);
2006 schedule_timeout(timeout
);
2007 mutex_lock(&bch_register_lock
);
2010 finish_wait(&unregister_wait
, &wait
);
2013 pr_info("All devices stopped");
2015 pr_notice("Timeout waiting for devices to be closed");
2017 mutex_unlock(&bch_register_lock
);
2023 static struct notifier_block reboot
= {
2024 .notifier_call
= bcache_reboot
,
2025 .priority
= INT_MAX
, /* before any real devices */
2028 static void bcache_exit(void)
2034 kobject_put(bcache_kobj
);
2036 destroy_workqueue(bcache_wq
);
2037 unregister_blkdev(bcache_major
, "bcache");
2038 unregister_reboot_notifier(&reboot
);
2041 static int __init
bcache_init(void)
2043 static const struct attribute
*files
[] = {
2044 &ksysfs_register
.attr
,
2045 &ksysfs_register_quiet
.attr
,
2049 mutex_init(&bch_register_lock
);
2050 init_waitqueue_head(&unregister_wait
);
2051 register_reboot_notifier(&reboot
);
2052 closure_debug_init();
2054 bcache_major
= register_blkdev(0, "bcache");
2055 if (bcache_major
< 0)
2056 return bcache_major
;
2058 if (!(bcache_wq
= create_workqueue("bcache")) ||
2059 !(bcache_kobj
= kobject_create_and_add("bcache", fs_kobj
)) ||
2060 sysfs_create_files(bcache_kobj
, files
) ||
2062 bch_request_init() ||
2063 bch_debug_init(bcache_kobj
))
2072 module_exit(bcache_exit
);
2073 module_init(bcache_init
);