2 * bcache setup/teardown code, and some metadata io - read a superblock and
3 * figure out what to do with it.
5 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
6 * Copyright 2012 Google, Inc.
14 #include <linux/buffer_head.h>
15 #include <linux/debugfs.h>
16 #include <linux/genhd.h>
17 #include <linux/module.h>
18 #include <linux/random.h>
19 #include <linux/reboot.h>
20 #include <linux/sysfs.h>
22 MODULE_LICENSE("GPL");
23 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
25 static const char bcache_magic
[] = {
26 0xc6, 0x85, 0x73, 0xf6, 0x4e, 0x1a, 0x45, 0xca,
27 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81
30 static const char invalid_uuid
[] = {
31 0xa0, 0x3e, 0xf8, 0xed, 0x3e, 0xe1, 0xb8, 0x78,
32 0xc8, 0x50, 0xfc, 0x5e, 0xcb, 0x16, 0xcd, 0x99
35 /* Default is -1; we skip past it for struct cached_dev's cache mode */
36 const char * const bch_cache_modes
[] = {
45 struct uuid_entry_v0
{
54 static struct kobject
*bcache_kobj
;
55 struct mutex bch_register_lock
;
56 LIST_HEAD(bch_cache_sets
);
57 static LIST_HEAD(uncached_devices
);
59 static int bcache_major
, bcache_minor
;
60 static wait_queue_head_t unregister_wait
;
61 struct workqueue_struct
*bcache_wq
;
63 #define BTREE_MAX_PAGES (256 * 1024 / PAGE_SIZE)
65 static void bio_split_pool_free(struct bio_split_pool
*p
)
67 if (p
->bio_split_hook
)
68 mempool_destroy(p
->bio_split_hook
);
71 bioset_free(p
->bio_split
);
74 static int bio_split_pool_init(struct bio_split_pool
*p
)
76 p
->bio_split
= bioset_create(4, 0);
80 p
->bio_split_hook
= mempool_create_kmalloc_pool(4,
81 sizeof(struct bio_split_hook
));
82 if (!p
->bio_split_hook
)
90 static const char *read_super(struct cache_sb
*sb
, struct block_device
*bdev
,
95 struct buffer_head
*bh
= __bread(bdev
, 1, SB_SIZE
);
101 s
= (struct cache_sb
*) bh
->b_data
;
103 sb
->offset
= le64_to_cpu(s
->offset
);
104 sb
->version
= le64_to_cpu(s
->version
);
106 memcpy(sb
->magic
, s
->magic
, 16);
107 memcpy(sb
->uuid
, s
->uuid
, 16);
108 memcpy(sb
->set_uuid
, s
->set_uuid
, 16);
109 memcpy(sb
->label
, s
->label
, SB_LABEL_SIZE
);
111 sb
->flags
= le64_to_cpu(s
->flags
);
112 sb
->seq
= le64_to_cpu(s
->seq
);
113 sb
->last_mount
= le32_to_cpu(s
->last_mount
);
114 sb
->first_bucket
= le16_to_cpu(s
->first_bucket
);
115 sb
->keys
= le16_to_cpu(s
->keys
);
117 for (i
= 0; i
< SB_JOURNAL_BUCKETS
; i
++)
118 sb
->d
[i
] = le64_to_cpu(s
->d
[i
]);
120 pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u",
121 sb
->version
, sb
->flags
, sb
->seq
, sb
->keys
);
123 err
= "Not a bcache superblock";
124 if (sb
->offset
!= SB_SECTOR
)
127 if (memcmp(sb
->magic
, bcache_magic
, 16))
130 err
= "Too many journal buckets";
131 if (sb
->keys
> SB_JOURNAL_BUCKETS
)
134 err
= "Bad checksum";
135 if (s
->csum
!= csum_set(s
))
139 if (bch_is_zero(sb
->uuid
, 16))
142 sb
->block_size
= le16_to_cpu(s
->block_size
);
144 err
= "Superblock block size smaller than device block size";
145 if (sb
->block_size
<< 9 < bdev_logical_block_size(bdev
))
148 switch (sb
->version
) {
149 case BCACHE_SB_VERSION_BDEV
:
150 sb
->data_offset
= BDEV_DATA_START_DEFAULT
;
152 case BCACHE_SB_VERSION_BDEV_WITH_OFFSET
:
153 sb
->data_offset
= le64_to_cpu(s
->data_offset
);
155 err
= "Bad data offset";
156 if (sb
->data_offset
< BDEV_DATA_START_DEFAULT
)
160 case BCACHE_SB_VERSION_CDEV
:
161 case BCACHE_SB_VERSION_CDEV_WITH_UUID
:
162 sb
->nbuckets
= le64_to_cpu(s
->nbuckets
);
163 sb
->block_size
= le16_to_cpu(s
->block_size
);
164 sb
->bucket_size
= le16_to_cpu(s
->bucket_size
);
166 sb
->nr_in_set
= le16_to_cpu(s
->nr_in_set
);
167 sb
->nr_this_dev
= le16_to_cpu(s
->nr_this_dev
);
169 err
= "Too many buckets";
170 if (sb
->nbuckets
> LONG_MAX
)
173 err
= "Not enough buckets";
174 if (sb
->nbuckets
< 1 << 7)
177 err
= "Bad block/bucket size";
178 if (!is_power_of_2(sb
->block_size
) ||
179 sb
->block_size
> PAGE_SECTORS
||
180 !is_power_of_2(sb
->bucket_size
) ||
181 sb
->bucket_size
< PAGE_SECTORS
)
184 err
= "Invalid superblock: device too small";
185 if (get_capacity(bdev
->bd_disk
) < sb
->bucket_size
* sb
->nbuckets
)
189 if (bch_is_zero(sb
->set_uuid
, 16))
192 err
= "Bad cache device number in set";
193 if (!sb
->nr_in_set
||
194 sb
->nr_in_set
<= sb
->nr_this_dev
||
195 sb
->nr_in_set
> MAX_CACHES_PER_SET
)
198 err
= "Journal buckets not sequential";
199 for (i
= 0; i
< sb
->keys
; i
++)
200 if (sb
->d
[i
] != sb
->first_bucket
+ i
)
203 err
= "Too many journal buckets";
204 if (sb
->first_bucket
+ sb
->keys
> sb
->nbuckets
)
207 err
= "Invalid superblock: first bucket comes before end of super";
208 if (sb
->first_bucket
* sb
->bucket_size
< 16)
213 err
= "Unsupported superblock version";
217 sb
->last_mount
= get_seconds();
220 get_page(bh
->b_page
);
227 static void write_bdev_super_endio(struct bio
*bio
, int error
)
229 struct cached_dev
*dc
= bio
->bi_private
;
230 /* XXX: error checking */
232 closure_put(&dc
->sb_write
.cl
);
235 static void __write_super(struct cache_sb
*sb
, struct bio
*bio
)
237 struct cache_sb
*out
= page_address(bio
->bi_io_vec
[0].bv_page
);
240 bio
->bi_sector
= SB_SECTOR
;
241 bio
->bi_rw
= REQ_SYNC
|REQ_META
;
242 bio
->bi_size
= SB_SIZE
;
243 bch_bio_map(bio
, NULL
);
245 out
->offset
= cpu_to_le64(sb
->offset
);
246 out
->version
= cpu_to_le64(sb
->version
);
248 memcpy(out
->uuid
, sb
->uuid
, 16);
249 memcpy(out
->set_uuid
, sb
->set_uuid
, 16);
250 memcpy(out
->label
, sb
->label
, SB_LABEL_SIZE
);
252 out
->flags
= cpu_to_le64(sb
->flags
);
253 out
->seq
= cpu_to_le64(sb
->seq
);
255 out
->last_mount
= cpu_to_le32(sb
->last_mount
);
256 out
->first_bucket
= cpu_to_le16(sb
->first_bucket
);
257 out
->keys
= cpu_to_le16(sb
->keys
);
259 for (i
= 0; i
< sb
->keys
; i
++)
260 out
->d
[i
] = cpu_to_le64(sb
->d
[i
]);
262 out
->csum
= csum_set(out
);
264 pr_debug("ver %llu, flags %llu, seq %llu",
265 sb
->version
, sb
->flags
, sb
->seq
);
267 submit_bio(REQ_WRITE
, bio
);
270 void bch_write_bdev_super(struct cached_dev
*dc
, struct closure
*parent
)
272 struct closure
*cl
= &dc
->sb_write
.cl
;
273 struct bio
*bio
= &dc
->sb_bio
;
275 closure_lock(&dc
->sb_write
, parent
);
278 bio
->bi_bdev
= dc
->bdev
;
279 bio
->bi_end_io
= write_bdev_super_endio
;
280 bio
->bi_private
= dc
;
283 __write_super(&dc
->sb
, bio
);
288 static void write_super_endio(struct bio
*bio
, int error
)
290 struct cache
*ca
= bio
->bi_private
;
292 bch_count_io_errors(ca
, error
, "writing superblock");
293 closure_put(&ca
->set
->sb_write
.cl
);
296 void bcache_write_super(struct cache_set
*c
)
298 struct closure
*cl
= &c
->sb_write
.cl
;
302 closure_lock(&c
->sb_write
, &c
->cl
);
306 for_each_cache(ca
, c
, i
) {
307 struct bio
*bio
= &ca
->sb_bio
;
309 ca
->sb
.version
= BCACHE_SB_VERSION_CDEV_WITH_UUID
;
310 ca
->sb
.seq
= c
->sb
.seq
;
311 ca
->sb
.last_mount
= c
->sb
.last_mount
;
313 SET_CACHE_SYNC(&ca
->sb
, CACHE_SYNC(&c
->sb
));
316 bio
->bi_bdev
= ca
->bdev
;
317 bio
->bi_end_io
= write_super_endio
;
318 bio
->bi_private
= ca
;
321 __write_super(&ca
->sb
, bio
);
329 static void uuid_endio(struct bio
*bio
, int error
)
331 struct closure
*cl
= bio
->bi_private
;
332 struct cache_set
*c
= container_of(cl
, struct cache_set
, uuid_write
.cl
);
334 cache_set_err_on(error
, c
, "accessing uuids");
335 bch_bbio_free(bio
, c
);
339 static void uuid_io(struct cache_set
*c
, unsigned long rw
,
340 struct bkey
*k
, struct closure
*parent
)
342 struct closure
*cl
= &c
->uuid_write
.cl
;
343 struct uuid_entry
*u
;
347 closure_lock(&c
->uuid_write
, parent
);
349 for (i
= 0; i
< KEY_PTRS(k
); i
++) {
350 struct bio
*bio
= bch_bbio_alloc(c
);
352 bio
->bi_rw
= REQ_SYNC
|REQ_META
|rw
;
353 bio
->bi_size
= KEY_SIZE(k
) << 9;
355 bio
->bi_end_io
= uuid_endio
;
356 bio
->bi_private
= cl
;
357 bch_bio_map(bio
, c
->uuids
);
359 bch_submit_bbio(bio
, c
, k
, i
);
365 pr_debug("%s UUIDs at %s", rw
& REQ_WRITE
? "wrote" : "read",
366 pkey(&c
->uuid_bucket
));
368 for (u
= c
->uuids
; u
< c
->uuids
+ c
->nr_uuids
; u
++)
369 if (!bch_is_zero(u
->uuid
, 16))
370 pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u",
371 u
- c
->uuids
, u
->uuid
, u
->label
,
372 u
->first_reg
, u
->last_reg
, u
->invalidated
);
377 static char *uuid_read(struct cache_set
*c
, struct jset
*j
, struct closure
*cl
)
379 struct bkey
*k
= &j
->uuid_bucket
;
381 if (__bch_ptr_invalid(c
, 1, k
))
382 return "bad uuid pointer";
384 bkey_copy(&c
->uuid_bucket
, k
);
385 uuid_io(c
, READ_SYNC
, k
, cl
);
387 if (j
->version
< BCACHE_JSET_VERSION_UUIDv1
) {
388 struct uuid_entry_v0
*u0
= (void *) c
->uuids
;
389 struct uuid_entry
*u1
= (void *) c
->uuids
;
395 * Since the new uuid entry is bigger than the old, we have to
396 * convert starting at the highest memory address and work down
397 * in order to do it in place
400 for (i
= c
->nr_uuids
- 1;
403 memcpy(u1
[i
].uuid
, u0
[i
].uuid
, 16);
404 memcpy(u1
[i
].label
, u0
[i
].label
, 32);
406 u1
[i
].first_reg
= u0
[i
].first_reg
;
407 u1
[i
].last_reg
= u0
[i
].last_reg
;
408 u1
[i
].invalidated
= u0
[i
].invalidated
;
418 static int __uuid_write(struct cache_set
*c
)
422 closure_init_stack(&cl
);
424 lockdep_assert_held(&bch_register_lock
);
426 if (bch_bucket_alloc_set(c
, WATERMARK_METADATA
, &k
.key
, 1, &cl
))
429 SET_KEY_SIZE(&k
.key
, c
->sb
.bucket_size
);
430 uuid_io(c
, REQ_WRITE
, &k
.key
, &cl
);
433 bkey_copy(&c
->uuid_bucket
, &k
.key
);
434 __bkey_put(c
, &k
.key
);
438 int bch_uuid_write(struct cache_set
*c
)
440 int ret
= __uuid_write(c
);
443 bch_journal_meta(c
, NULL
);
448 static struct uuid_entry
*uuid_find(struct cache_set
*c
, const char *uuid
)
450 struct uuid_entry
*u
;
453 u
< c
->uuids
+ c
->nr_uuids
; u
++)
454 if (!memcmp(u
->uuid
, uuid
, 16))
460 static struct uuid_entry
*uuid_find_empty(struct cache_set
*c
)
462 static const char zero_uuid
[16] = "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0";
463 return uuid_find(c
, zero_uuid
);
467 * Bucket priorities/gens:
469 * For each bucket, we store on disk its
473 * See alloc.c for an explanation of the gen. The priority is used to implement
474 * lru (and in the future other) cache replacement policies; for most purposes
475 * it's just an opaque integer.
477 * The gens and the priorities don't have a whole lot to do with each other, and
478 * it's actually the gens that must be written out at specific times - it's no
479 * big deal if the priorities don't get written, if we lose them we just reuse
480 * buckets in suboptimal order.
482 * On disk they're stored in a packed array, and in as many buckets are required
483 * to fit them all. The buckets we use to store them form a list; the journal
484 * header points to the first bucket, the first bucket points to the second
487 * This code is used by the allocation code; periodically (whenever it runs out
488 * of buckets to allocate from) the allocation code will invalidate some
489 * buckets, but it can't use those buckets until their new gens are safely on
493 static void prio_endio(struct bio
*bio
, int error
)
495 struct cache
*ca
= bio
->bi_private
;
497 cache_set_err_on(error
, ca
->set
, "accessing priorities");
498 bch_bbio_free(bio
, ca
->set
);
499 closure_put(&ca
->prio
);
502 static void prio_io(struct cache
*ca
, uint64_t bucket
, unsigned long rw
)
504 struct closure
*cl
= &ca
->prio
;
505 struct bio
*bio
= bch_bbio_alloc(ca
->set
);
507 closure_init_stack(cl
);
509 bio
->bi_sector
= bucket
* ca
->sb
.bucket_size
;
510 bio
->bi_bdev
= ca
->bdev
;
511 bio
->bi_rw
= REQ_SYNC
|REQ_META
|rw
;
512 bio
->bi_size
= bucket_bytes(ca
);
514 bio
->bi_end_io
= prio_endio
;
515 bio
->bi_private
= ca
;
516 bch_bio_map(bio
, ca
->disk_buckets
);
518 closure_bio_submit(bio
, &ca
->prio
, ca
);
522 #define buckets_free(c) "free %zu, free_inc %zu, unused %zu", \
523 fifo_used(&c->free), fifo_used(&c->free_inc), fifo_used(&c->unused)
525 void bch_prio_write(struct cache
*ca
)
531 closure_init_stack(&cl
);
533 lockdep_assert_held(&ca
->set
->bucket_lock
);
535 for (b
= ca
->buckets
;
536 b
< ca
->buckets
+ ca
->sb
.nbuckets
; b
++)
537 b
->disk_gen
= b
->gen
;
539 ca
->disk_buckets
->seq
++;
541 atomic_long_add(ca
->sb
.bucket_size
* prio_buckets(ca
),
542 &ca
->meta_sectors_written
);
544 pr_debug("free %zu, free_inc %zu, unused %zu", fifo_used(&ca
->free
),
545 fifo_used(&ca
->free_inc
), fifo_used(&ca
->unused
));
546 blktrace_msg(ca
, "Starting priorities: " buckets_free(ca
));
548 for (i
= prio_buckets(ca
) - 1; i
>= 0; --i
) {
550 struct prio_set
*p
= ca
->disk_buckets
;
551 struct bucket_disk
*d
= p
->data
;
552 struct bucket_disk
*end
= d
+ prios_per_bucket(ca
);
554 for (b
= ca
->buckets
+ i
* prios_per_bucket(ca
);
555 b
< ca
->buckets
+ ca
->sb
.nbuckets
&& d
< end
;
557 d
->prio
= cpu_to_le16(b
->prio
);
561 p
->next_bucket
= ca
->prio_buckets
[i
+ 1];
562 p
->magic
= pset_magic(ca
);
563 p
->csum
= bch_crc64(&p
->magic
, bucket_bytes(ca
) - 8);
565 bucket
= bch_bucket_alloc(ca
, WATERMARK_PRIO
, &cl
);
566 BUG_ON(bucket
== -1);
568 mutex_unlock(&ca
->set
->bucket_lock
);
569 prio_io(ca
, bucket
, REQ_WRITE
);
570 mutex_lock(&ca
->set
->bucket_lock
);
572 ca
->prio_buckets
[i
] = bucket
;
573 atomic_dec_bug(&ca
->buckets
[bucket
].pin
);
576 mutex_unlock(&ca
->set
->bucket_lock
);
578 bch_journal_meta(ca
->set
, &cl
);
581 mutex_lock(&ca
->set
->bucket_lock
);
583 ca
->need_save_prio
= 0;
586 * Don't want the old priorities to get garbage collected until after we
587 * finish writing the new ones, and they're journalled
589 for (i
= 0; i
< prio_buckets(ca
); i
++)
590 ca
->prio_last_buckets
[i
] = ca
->prio_buckets
[i
];
593 static void prio_read(struct cache
*ca
, uint64_t bucket
)
595 struct prio_set
*p
= ca
->disk_buckets
;
596 struct bucket_disk
*d
= p
->data
+ prios_per_bucket(ca
), *end
= d
;
598 unsigned bucket_nr
= 0;
600 for (b
= ca
->buckets
;
601 b
< ca
->buckets
+ ca
->sb
.nbuckets
;
604 ca
->prio_buckets
[bucket_nr
] = bucket
;
605 ca
->prio_last_buckets
[bucket_nr
] = bucket
;
608 prio_io(ca
, bucket
, READ_SYNC
);
610 if (p
->csum
!= bch_crc64(&p
->magic
, bucket_bytes(ca
) - 8))
611 pr_warn("bad csum reading priorities");
613 if (p
->magic
!= pset_magic(ca
))
614 pr_warn("bad magic reading priorities");
616 bucket
= p
->next_bucket
;
620 b
->prio
= le16_to_cpu(d
->prio
);
621 b
->gen
= b
->disk_gen
= b
->last_gc
= b
->gc_gen
= d
->gen
;
627 static int open_dev(struct block_device
*b
, fmode_t mode
)
629 struct bcache_device
*d
= b
->bd_disk
->private_data
;
630 if (atomic_read(&d
->closing
))
637 static void release_dev(struct gendisk
*b
, fmode_t mode
)
639 struct bcache_device
*d
= b
->private_data
;
643 static int ioctl_dev(struct block_device
*b
, fmode_t mode
,
644 unsigned int cmd
, unsigned long arg
)
646 struct bcache_device
*d
= b
->bd_disk
->private_data
;
647 return d
->ioctl(d
, mode
, cmd
, arg
);
650 static const struct block_device_operations bcache_ops
= {
652 .release
= release_dev
,
654 .owner
= THIS_MODULE
,
657 void bcache_device_stop(struct bcache_device
*d
)
659 if (!atomic_xchg(&d
->closing
, 1))
660 closure_queue(&d
->cl
);
663 static void bcache_device_unlink(struct bcache_device
*d
)
668 sysfs_remove_link(&d
->c
->kobj
, d
->name
);
669 sysfs_remove_link(&d
->kobj
, "cache");
671 for_each_cache(ca
, d
->c
, i
)
672 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 (atomic_read(&d
->detaching
)) {
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
);
704 atomic_set(&d
->detaching
, 0);
707 bcache_device_unlink(d
);
709 d
->c
->devices
[d
->id
] = NULL
;
710 closure_put(&d
->c
->caching
);
714 static void bcache_device_attach(struct bcache_device
*d
, struct cache_set
*c
,
717 BUG_ON(test_bit(CACHE_SET_STOPPING
, &c
->flags
));
723 closure_get(&c
->caching
);
726 static void bcache_device_free(struct bcache_device
*d
)
728 lockdep_assert_held(&bch_register_lock
);
730 pr_info("%s stopped", d
->disk
->disk_name
);
733 bcache_device_detach(d
);
734 if (d
->disk
&& d
->disk
->flags
& GENHD_FL_UP
)
735 del_gendisk(d
->disk
);
736 if (d
->disk
&& d
->disk
->queue
)
737 blk_cleanup_queue(d
->disk
->queue
);
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
);
747 closure_debug_destroy(&d
->cl
);
750 static int bcache_device_init(struct bcache_device
*d
, unsigned block_size
)
752 struct request_queue
*q
;
754 if (!(d
->bio_split
= bioset_create(4, offsetof(struct bbio
, bio
))) ||
755 !(d
->unaligned_bvec
= mempool_create_kmalloc_pool(1,
756 sizeof(struct bio_vec
) * BIO_MAX_PAGES
)) ||
757 bio_split_pool_init(&d
->bio_split_hook
) ||
758 !(d
->disk
= alloc_disk(1)) ||
759 !(q
= blk_alloc_queue(GFP_KERNEL
)))
762 snprintf(d
->disk
->disk_name
, DISK_NAME_LEN
, "bcache%i", bcache_minor
);
764 d
->disk
->major
= bcache_major
;
765 d
->disk
->first_minor
= bcache_minor
++;
766 d
->disk
->fops
= &bcache_ops
;
767 d
->disk
->private_data
= d
;
769 blk_queue_make_request(q
, NULL
);
772 q
->backing_dev_info
.congested_data
= d
;
773 q
->limits
.max_hw_sectors
= UINT_MAX
;
774 q
->limits
.max_sectors
= UINT_MAX
;
775 q
->limits
.max_segment_size
= UINT_MAX
;
776 q
->limits
.max_segments
= BIO_MAX_PAGES
;
777 q
->limits
.max_discard_sectors
= UINT_MAX
;
778 q
->limits
.io_min
= block_size
;
779 q
->limits
.logical_block_size
= block_size
;
780 q
->limits
.physical_block_size
= block_size
;
781 set_bit(QUEUE_FLAG_NONROT
, &d
->disk
->queue
->queue_flags
);
782 set_bit(QUEUE_FLAG_DISCARD
, &d
->disk
->queue
->queue_flags
);
789 static void calc_cached_dev_sectors(struct cache_set
*c
)
791 uint64_t sectors
= 0;
792 struct cached_dev
*dc
;
794 list_for_each_entry(dc
, &c
->cached_devs
, list
)
795 sectors
+= bdev_sectors(dc
->bdev
);
797 c
->cached_dev_sectors
= sectors
;
800 void bch_cached_dev_run(struct cached_dev
*dc
)
802 struct bcache_device
*d
= &dc
->disk
;
804 if (atomic_xchg(&dc
->running
, 1))
808 BDEV_STATE(&dc
->sb
) != BDEV_STATE_NONE
) {
810 closure_init_stack(&cl
);
812 SET_BDEV_STATE(&dc
->sb
, BDEV_STATE_STALE
);
813 bch_write_bdev_super(dc
, &cl
);
818 bd_link_disk_holder(dc
->bdev
, dc
->disk
.disk
);
820 char *env
[] = { "SYMLINK=label" , NULL
};
821 kobject_uevent_env(&disk_to_dev(d
->disk
)->kobj
, KOBJ_CHANGE
, env
);
823 if (sysfs_create_link(&d
->kobj
, &disk_to_dev(d
->disk
)->kobj
, "dev") ||
824 sysfs_create_link(&disk_to_dev(d
->disk
)->kobj
, &d
->kobj
, "bcache"))
825 pr_debug("error creating sysfs link");
828 static void cached_dev_detach_finish(struct work_struct
*w
)
830 struct cached_dev
*dc
= container_of(w
, struct cached_dev
, detach
);
831 char buf
[BDEVNAME_SIZE
];
833 closure_init_stack(&cl
);
835 BUG_ON(!atomic_read(&dc
->disk
.detaching
));
836 BUG_ON(atomic_read(&dc
->count
));
838 mutex_lock(&bch_register_lock
);
840 memset(&dc
->sb
.set_uuid
, 0, 16);
841 SET_BDEV_STATE(&dc
->sb
, BDEV_STATE_NONE
);
843 bch_write_bdev_super(dc
, &cl
);
846 bcache_device_detach(&dc
->disk
);
847 list_move(&dc
->list
, &uncached_devices
);
849 mutex_unlock(&bch_register_lock
);
851 pr_info("Caching disabled for %s", bdevname(dc
->bdev
, buf
));
853 /* Drop ref we took in cached_dev_detach() */
854 closure_put(&dc
->disk
.cl
);
857 void bch_cached_dev_detach(struct cached_dev
*dc
)
859 lockdep_assert_held(&bch_register_lock
);
861 if (atomic_read(&dc
->disk
.closing
))
864 if (atomic_xchg(&dc
->disk
.detaching
, 1))
868 * Block the device from being closed and freed until we're finished
871 closure_get(&dc
->disk
.cl
);
873 bch_writeback_queue(dc
);
877 int bch_cached_dev_attach(struct cached_dev
*dc
, struct cache_set
*c
)
879 uint32_t rtime
= cpu_to_le32(get_seconds());
880 struct uuid_entry
*u
;
881 char buf
[BDEVNAME_SIZE
];
883 bdevname(dc
->bdev
, buf
);
885 if (memcmp(dc
->sb
.set_uuid
, c
->sb
.set_uuid
, 16))
889 pr_err("Can't attach %s: already attached", buf
);
893 if (test_bit(CACHE_SET_STOPPING
, &c
->flags
)) {
894 pr_err("Can't attach %s: shutting down", buf
);
898 if (dc
->sb
.block_size
< c
->sb
.block_size
) {
900 pr_err("Couldn't attach %s: block size less than set's block size",
905 u
= uuid_find(c
, dc
->sb
.uuid
);
908 (BDEV_STATE(&dc
->sb
) == BDEV_STATE_STALE
||
909 BDEV_STATE(&dc
->sb
) == BDEV_STATE_NONE
)) {
910 memcpy(u
->uuid
, invalid_uuid
, 16);
911 u
->invalidated
= cpu_to_le32(get_seconds());
916 if (BDEV_STATE(&dc
->sb
) == BDEV_STATE_DIRTY
) {
917 pr_err("Couldn't find uuid for %s in set", buf
);
921 u
= uuid_find_empty(c
);
923 pr_err("Not caching %s, no room for UUID", buf
);
928 /* Deadlocks since we're called via sysfs...
929 sysfs_remove_file(&dc->kobj, &sysfs_attach);
932 if (bch_is_zero(u
->uuid
, 16)) {
934 closure_init_stack(&cl
);
936 memcpy(u
->uuid
, dc
->sb
.uuid
, 16);
937 memcpy(u
->label
, dc
->sb
.label
, SB_LABEL_SIZE
);
938 u
->first_reg
= u
->last_reg
= rtime
;
941 memcpy(dc
->sb
.set_uuid
, c
->sb
.set_uuid
, 16);
942 SET_BDEV_STATE(&dc
->sb
, BDEV_STATE_CLEAN
);
944 bch_write_bdev_super(dc
, &cl
);
951 bcache_device_attach(&dc
->disk
, c
, u
- c
->uuids
);
952 list_move(&dc
->list
, &c
->cached_devs
);
953 calc_cached_dev_sectors(c
);
957 * dc->c must be set before dc->count != 0 - paired with the mb in
960 atomic_set(&dc
->count
, 1);
962 if (BDEV_STATE(&dc
->sb
) == BDEV_STATE_DIRTY
) {
963 atomic_set(&dc
->has_dirty
, 1);
964 atomic_inc(&dc
->count
);
965 bch_writeback_queue(dc
);
968 bch_cached_dev_run(dc
);
969 bcache_device_link(&dc
->disk
, c
, "bdev");
971 pr_info("Caching %s as %s on set %pU",
972 bdevname(dc
->bdev
, buf
), dc
->disk
.disk
->disk_name
,
973 dc
->disk
.c
->sb
.set_uuid
);
977 void bch_cached_dev_release(struct kobject
*kobj
)
979 struct cached_dev
*dc
= container_of(kobj
, struct cached_dev
,
982 module_put(THIS_MODULE
);
985 static void cached_dev_free(struct closure
*cl
)
987 struct cached_dev
*dc
= container_of(cl
, struct cached_dev
, disk
.cl
);
989 cancel_delayed_work_sync(&dc
->writeback_rate_update
);
991 mutex_lock(&bch_register_lock
);
993 if (atomic_read(&dc
->running
))
994 bd_unlink_disk_holder(dc
->bdev
, dc
->disk
.disk
);
995 bcache_device_free(&dc
->disk
);
998 mutex_unlock(&bch_register_lock
);
1000 if (!IS_ERR_OR_NULL(dc
->bdev
)) {
1001 if (dc
->bdev
->bd_disk
)
1002 blk_sync_queue(bdev_get_queue(dc
->bdev
));
1004 blkdev_put(dc
->bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
1007 wake_up(&unregister_wait
);
1009 kobject_put(&dc
->disk
.kobj
);
1012 static void cached_dev_flush(struct closure
*cl
)
1014 struct cached_dev
*dc
= container_of(cl
, struct cached_dev
, disk
.cl
);
1015 struct bcache_device
*d
= &dc
->disk
;
1017 bch_cache_accounting_destroy(&dc
->accounting
);
1018 kobject_del(&d
->kobj
);
1020 continue_at(cl
, cached_dev_free
, system_wq
);
1023 static int cached_dev_init(struct cached_dev
*dc
, unsigned block_size
)
1027 struct request_queue
*q
= bdev_get_queue(dc
->bdev
);
1029 __module_get(THIS_MODULE
);
1030 INIT_LIST_HEAD(&dc
->list
);
1031 closure_init(&dc
->disk
.cl
, NULL
);
1032 set_closure_fn(&dc
->disk
.cl
, cached_dev_flush
, system_wq
);
1033 kobject_init(&dc
->disk
.kobj
, &bch_cached_dev_ktype
);
1034 INIT_WORK(&dc
->detach
, cached_dev_detach_finish
);
1035 closure_init_unlocked(&dc
->sb_write
);
1036 INIT_LIST_HEAD(&dc
->io_lru
);
1037 spin_lock_init(&dc
->io_lock
);
1038 bch_cache_accounting_init(&dc
->accounting
, &dc
->disk
.cl
);
1040 dc
->sequential_merge
= true;
1041 dc
->sequential_cutoff
= 4 << 20;
1043 for (io
= dc
->io
; io
< dc
->io
+ RECENT_IO
; io
++) {
1044 list_add(&io
->lru
, &dc
->io_lru
);
1045 hlist_add_head(&io
->hash
, dc
->io_hash
+ RECENT_IO
);
1048 ret
= bcache_device_init(&dc
->disk
, block_size
);
1052 set_capacity(dc
->disk
.disk
,
1053 dc
->bdev
->bd_part
->nr_sects
- dc
->sb
.data_offset
);
1055 dc
->disk
.disk
->queue
->backing_dev_info
.ra_pages
=
1056 max(dc
->disk
.disk
->queue
->backing_dev_info
.ra_pages
,
1057 q
->backing_dev_info
.ra_pages
);
1059 bch_cached_dev_request_init(dc
);
1060 bch_cached_dev_writeback_init(dc
);
1064 /* Cached device - bcache superblock */
1066 static void register_bdev(struct cache_sb
*sb
, struct page
*sb_page
,
1067 struct block_device
*bdev
,
1068 struct cached_dev
*dc
)
1070 char name
[BDEVNAME_SIZE
];
1071 const char *err
= "cannot allocate memory";
1072 struct cache_set
*c
;
1074 memcpy(&dc
->sb
, sb
, sizeof(struct cache_sb
));
1076 dc
->bdev
->bd_holder
= dc
;
1078 bio_init(&dc
->sb_bio
);
1079 dc
->sb_bio
.bi_max_vecs
= 1;
1080 dc
->sb_bio
.bi_io_vec
= dc
->sb_bio
.bi_inline_vecs
;
1081 dc
->sb_bio
.bi_io_vec
[0].bv_page
= sb_page
;
1084 if (cached_dev_init(dc
, sb
->block_size
<< 9))
1087 err
= "error creating kobject";
1088 if (kobject_add(&dc
->disk
.kobj
, &part_to_dev(bdev
->bd_part
)->kobj
,
1091 if (bch_cache_accounting_add_kobjs(&dc
->accounting
, &dc
->disk
.kobj
))
1094 pr_info("registered backing device %s", bdevname(bdev
, name
));
1096 list_add(&dc
->list
, &uncached_devices
);
1097 list_for_each_entry(c
, &bch_cache_sets
, list
)
1098 bch_cached_dev_attach(dc
, c
);
1100 if (BDEV_STATE(&dc
->sb
) == BDEV_STATE_NONE
||
1101 BDEV_STATE(&dc
->sb
) == BDEV_STATE_STALE
)
1102 bch_cached_dev_run(dc
);
1106 pr_notice("error opening %s: %s", bdevname(bdev
, name
), err
);
1107 bcache_device_stop(&dc
->disk
);
1110 /* Flash only volumes */
1112 void bch_flash_dev_release(struct kobject
*kobj
)
1114 struct bcache_device
*d
= container_of(kobj
, struct bcache_device
,
1119 static void flash_dev_free(struct closure
*cl
)
1121 struct bcache_device
*d
= container_of(cl
, struct bcache_device
, cl
);
1122 bcache_device_free(d
);
1123 kobject_put(&d
->kobj
);
1126 static void flash_dev_flush(struct closure
*cl
)
1128 struct bcache_device
*d
= container_of(cl
, struct bcache_device
, cl
);
1130 bcache_device_unlink(d
);
1131 kobject_del(&d
->kobj
);
1132 continue_at(cl
, flash_dev_free
, system_wq
);
1135 static int flash_dev_run(struct cache_set
*c
, struct uuid_entry
*u
)
1137 struct bcache_device
*d
= kzalloc(sizeof(struct bcache_device
),
1142 closure_init(&d
->cl
, NULL
);
1143 set_closure_fn(&d
->cl
, flash_dev_flush
, system_wq
);
1145 kobject_init(&d
->kobj
, &bch_flash_dev_ktype
);
1147 if (bcache_device_init(d
, block_bytes(c
)))
1150 bcache_device_attach(d
, c
, u
- c
->uuids
);
1151 set_capacity(d
->disk
, u
->sectors
);
1152 bch_flash_dev_request_init(d
);
1155 if (kobject_add(&d
->kobj
, &disk_to_dev(d
->disk
)->kobj
, "bcache"))
1158 bcache_device_link(d
, c
, "volume");
1162 kobject_put(&d
->kobj
);
1166 static int flash_devs_run(struct cache_set
*c
)
1169 struct uuid_entry
*u
;
1172 u
< c
->uuids
+ c
->nr_uuids
&& !ret
;
1174 if (UUID_FLASH_ONLY(u
))
1175 ret
= flash_dev_run(c
, u
);
1180 int bch_flash_dev_create(struct cache_set
*c
, uint64_t size
)
1182 struct uuid_entry
*u
;
1184 if (test_bit(CACHE_SET_STOPPING
, &c
->flags
))
1187 u
= uuid_find_empty(c
);
1189 pr_err("Can't create volume, no room for UUID");
1193 get_random_bytes(u
->uuid
, 16);
1194 memset(u
->label
, 0, 32);
1195 u
->first_reg
= u
->last_reg
= cpu_to_le32(get_seconds());
1197 SET_UUID_FLASH_ONLY(u
, 1);
1198 u
->sectors
= size
>> 9;
1202 return flash_dev_run(c
, u
);
1208 bool bch_cache_set_error(struct cache_set
*c
, const char *fmt
, ...)
1212 if (test_bit(CACHE_SET_STOPPING
, &c
->flags
))
1215 /* XXX: we can be called from atomic context
1216 acquire_console_sem();
1219 printk(KERN_ERR
"bcache: error on %pU: ", c
->sb
.set_uuid
);
1221 va_start(args
, fmt
);
1225 printk(", disabling caching\n");
1227 bch_cache_set_unregister(c
);
1231 void bch_cache_set_release(struct kobject
*kobj
)
1233 struct cache_set
*c
= container_of(kobj
, struct cache_set
, kobj
);
1235 module_put(THIS_MODULE
);
1238 static void cache_set_free(struct closure
*cl
)
1240 struct cache_set
*c
= container_of(cl
, struct cache_set
, cl
);
1244 if (!IS_ERR_OR_NULL(c
->debug
))
1245 debugfs_remove(c
->debug
);
1247 bch_open_buckets_free(c
);
1248 bch_btree_cache_free(c
);
1249 bch_journal_free(c
);
1251 for_each_cache(ca
, c
, i
)
1253 kobject_put(&ca
->kobj
);
1255 free_pages((unsigned long) c
->uuids
, ilog2(bucket_pages(c
)));
1256 free_pages((unsigned long) c
->sort
, ilog2(bucket_pages(c
)));
1258 kfree(c
->fill_iter
);
1260 bioset_free(c
->bio_split
);
1262 mempool_destroy(c
->bio_meta
);
1264 mempool_destroy(c
->search
);
1267 mutex_lock(&bch_register_lock
);
1269 mutex_unlock(&bch_register_lock
);
1271 pr_info("Cache set %pU unregistered", c
->sb
.set_uuid
);
1272 wake_up(&unregister_wait
);
1274 closure_debug_destroy(&c
->cl
);
1275 kobject_put(&c
->kobj
);
1278 static void cache_set_flush(struct closure
*cl
)
1280 struct cache_set
*c
= container_of(cl
, struct cache_set
, caching
);
1283 /* Shut down allocator threads */
1284 set_bit(CACHE_SET_STOPPING_2
, &c
->flags
);
1285 wake_up(&c
->alloc_wait
);
1287 bch_cache_accounting_destroy(&c
->accounting
);
1289 kobject_put(&c
->internal
);
1290 kobject_del(&c
->kobj
);
1292 if (!IS_ERR_OR_NULL(c
->root
))
1293 list_add(&c
->root
->list
, &c
->btree_cache
);
1295 /* Should skip this if we're unregistering because of an error */
1296 list_for_each_entry(b
, &c
->btree_cache
, list
)
1297 if (btree_node_dirty(b
))
1298 bch_btree_write(b
, true, NULL
);
1303 static void __cache_set_unregister(struct closure
*cl
)
1305 struct cache_set
*c
= container_of(cl
, struct cache_set
, caching
);
1306 struct cached_dev
*dc
, *t
;
1309 mutex_lock(&bch_register_lock
);
1311 if (test_bit(CACHE_SET_UNREGISTERING
, &c
->flags
))
1312 list_for_each_entry_safe(dc
, t
, &c
->cached_devs
, list
)
1313 bch_cached_dev_detach(dc
);
1315 for (i
= 0; i
< c
->nr_uuids
; i
++)
1316 if (c
->devices
[i
] && UUID_FLASH_ONLY(&c
->uuids
[i
]))
1317 bcache_device_stop(c
->devices
[i
]);
1319 mutex_unlock(&bch_register_lock
);
1321 continue_at(cl
, cache_set_flush
, system_wq
);
1324 void bch_cache_set_stop(struct cache_set
*c
)
1326 if (!test_and_set_bit(CACHE_SET_STOPPING
, &c
->flags
))
1327 closure_queue(&c
->caching
);
1330 void bch_cache_set_unregister(struct cache_set
*c
)
1332 set_bit(CACHE_SET_UNREGISTERING
, &c
->flags
);
1333 bch_cache_set_stop(c
);
1336 #define alloc_bucket_pages(gfp, c) \
1337 ((void *) __get_free_pages(__GFP_ZERO|gfp, ilog2(bucket_pages(c))))
1339 struct cache_set
*bch_cache_set_alloc(struct cache_sb
*sb
)
1342 struct cache_set
*c
= kzalloc(sizeof(struct cache_set
), GFP_KERNEL
);
1346 __module_get(THIS_MODULE
);
1347 closure_init(&c
->cl
, NULL
);
1348 set_closure_fn(&c
->cl
, cache_set_free
, system_wq
);
1350 closure_init(&c
->caching
, &c
->cl
);
1351 set_closure_fn(&c
->caching
, __cache_set_unregister
, system_wq
);
1353 /* Maybe create continue_at_noreturn() and use it here? */
1354 closure_set_stopped(&c
->cl
);
1355 closure_put(&c
->cl
);
1357 kobject_init(&c
->kobj
, &bch_cache_set_ktype
);
1358 kobject_init(&c
->internal
, &bch_cache_set_internal_ktype
);
1360 bch_cache_accounting_init(&c
->accounting
, &c
->cl
);
1362 memcpy(c
->sb
.set_uuid
, sb
->set_uuid
, 16);
1363 c
->sb
.block_size
= sb
->block_size
;
1364 c
->sb
.bucket_size
= sb
->bucket_size
;
1365 c
->sb
.nr_in_set
= sb
->nr_in_set
;
1366 c
->sb
.last_mount
= sb
->last_mount
;
1367 c
->bucket_bits
= ilog2(sb
->bucket_size
);
1368 c
->block_bits
= ilog2(sb
->block_size
);
1369 c
->nr_uuids
= bucket_bytes(c
) / sizeof(struct uuid_entry
);
1371 c
->btree_pages
= c
->sb
.bucket_size
/ PAGE_SECTORS
;
1372 if (c
->btree_pages
> BTREE_MAX_PAGES
)
1373 c
->btree_pages
= max_t(int, c
->btree_pages
/ 4,
1376 init_waitqueue_head(&c
->alloc_wait
);
1377 mutex_init(&c
->bucket_lock
);
1378 mutex_init(&c
->fill_lock
);
1379 mutex_init(&c
->sort_lock
);
1380 spin_lock_init(&c
->sort_time_lock
);
1381 closure_init_unlocked(&c
->sb_write
);
1382 closure_init_unlocked(&c
->uuid_write
);
1383 spin_lock_init(&c
->btree_read_time_lock
);
1384 bch_moving_init_cache_set(c
);
1386 INIT_LIST_HEAD(&c
->list
);
1387 INIT_LIST_HEAD(&c
->cached_devs
);
1388 INIT_LIST_HEAD(&c
->btree_cache
);
1389 INIT_LIST_HEAD(&c
->btree_cache_freeable
);
1390 INIT_LIST_HEAD(&c
->btree_cache_freed
);
1391 INIT_LIST_HEAD(&c
->data_buckets
);
1393 c
->search
= mempool_create_slab_pool(32, bch_search_cache
);
1397 iter_size
= (sb
->bucket_size
/ sb
->block_size
+ 1) *
1398 sizeof(struct btree_iter_set
);
1400 if (!(c
->devices
= kzalloc(c
->nr_uuids
* sizeof(void *), GFP_KERNEL
)) ||
1401 !(c
->bio_meta
= mempool_create_kmalloc_pool(2,
1402 sizeof(struct bbio
) + sizeof(struct bio_vec
) *
1403 bucket_pages(c
))) ||
1404 !(c
->bio_split
= bioset_create(4, offsetof(struct bbio
, bio
))) ||
1405 !(c
->fill_iter
= kmalloc(iter_size
, GFP_KERNEL
)) ||
1406 !(c
->sort
= alloc_bucket_pages(GFP_KERNEL
, c
)) ||
1407 !(c
->uuids
= alloc_bucket_pages(GFP_KERNEL
, c
)) ||
1408 bch_journal_alloc(c
) ||
1409 bch_btree_cache_alloc(c
) ||
1410 bch_open_buckets_alloc(c
))
1413 c
->fill_iter
->size
= sb
->bucket_size
/ sb
->block_size
;
1415 c
->congested_read_threshold_us
= 2000;
1416 c
->congested_write_threshold_us
= 20000;
1417 c
->error_limit
= 8 << IO_ERROR_SHIFT
;
1421 bch_cache_set_unregister(c
);
1425 static void run_cache_set(struct cache_set
*c
)
1427 const char *err
= "cannot allocate memory";
1428 struct cached_dev
*dc
, *t
;
1433 bch_btree_op_init_stack(&op
);
1436 for_each_cache(ca
, c
, i
)
1437 c
->nbuckets
+= ca
->sb
.nbuckets
;
1439 if (CACHE_SYNC(&c
->sb
)) {
1444 err
= "cannot allocate memory for journal";
1445 if (bch_journal_read(c
, &journal
, &op
))
1448 pr_debug("btree_journal_read() done");
1450 err
= "no journal entries found";
1451 if (list_empty(&journal
))
1454 j
= &list_entry(journal
.prev
, struct journal_replay
, list
)->j
;
1456 err
= "IO error reading priorities";
1457 for_each_cache(ca
, c
, i
)
1458 prio_read(ca
, j
->prio_bucket
[ca
->sb
.nr_this_dev
]);
1461 * If prio_read() fails it'll call cache_set_error and we'll
1462 * tear everything down right away, but if we perhaps checked
1463 * sooner we could avoid journal replay.
1468 err
= "bad btree root";
1469 if (__bch_ptr_invalid(c
, j
->btree_level
+ 1, k
))
1472 err
= "error reading btree root";
1473 c
->root
= bch_btree_node_get(c
, k
, j
->btree_level
, &op
);
1474 if (IS_ERR_OR_NULL(c
->root
))
1477 list_del_init(&c
->root
->list
);
1478 rw_unlock(true, c
->root
);
1480 err
= uuid_read(c
, j
, &op
.cl
);
1484 err
= "error in recovery";
1485 if (bch_btree_check(c
, &op
))
1488 bch_journal_mark(c
, &journal
);
1489 bch_btree_gc_finish(c
);
1490 pr_debug("btree_check() done");
1493 * bcache_journal_next() can't happen sooner, or
1494 * btree_gc_finish() will give spurious errors about last_gc >
1495 * gc_gen - this is a hack but oh well.
1497 bch_journal_next(&c
->journal
);
1499 for_each_cache(ca
, c
, i
)
1500 closure_call(&ca
->alloc
, bch_allocator_thread
,
1504 * First place it's safe to allocate: btree_check() and
1505 * btree_gc_finish() have to run before we have buckets to
1506 * allocate, and bch_bucket_alloc_set() might cause a journal
1507 * entry to be written so bcache_journal_next() has to be called
1510 * If the uuids were in the old format we have to rewrite them
1511 * before the next journal entry is written:
1513 if (j
->version
< BCACHE_JSET_VERSION_UUID
)
1516 bch_journal_replay(c
, &journal
, &op
);
1518 pr_notice("invalidating existing data");
1519 /* Don't want invalidate_buckets() to queue a gc yet */
1520 closure_lock(&c
->gc
, NULL
);
1522 for_each_cache(ca
, c
, i
) {
1525 ca
->sb
.keys
= clamp_t(int, ca
->sb
.nbuckets
>> 7,
1526 2, SB_JOURNAL_BUCKETS
);
1528 for (j
= 0; j
< ca
->sb
.keys
; j
++)
1529 ca
->sb
.d
[j
] = ca
->sb
.first_bucket
+ j
;
1532 bch_btree_gc_finish(c
);
1534 for_each_cache(ca
, c
, i
)
1535 closure_call(&ca
->alloc
, bch_allocator_thread
,
1536 ca
->alloc_workqueue
, &c
->cl
);
1538 mutex_lock(&c
->bucket_lock
);
1539 for_each_cache(ca
, c
, i
)
1541 mutex_unlock(&c
->bucket_lock
);
1543 wake_up(&c
->alloc_wait
);
1545 err
= "cannot allocate new UUID bucket";
1546 if (__uuid_write(c
))
1549 err
= "cannot allocate new btree root";
1550 c
->root
= bch_btree_node_alloc(c
, 0, &op
.cl
);
1551 if (IS_ERR_OR_NULL(c
->root
))
1554 bkey_copy_key(&c
->root
->key
, &MAX_KEY
);
1555 bch_btree_write(c
->root
, true, &op
);
1557 bch_btree_set_root(c
->root
);
1558 rw_unlock(true, c
->root
);
1561 * We don't want to write the first journal entry until
1562 * everything is set up - fortunately journal entries won't be
1563 * written until the SET_CACHE_SYNC() here:
1565 SET_CACHE_SYNC(&c
->sb
, true);
1567 bch_journal_next(&c
->journal
);
1568 bch_journal_meta(c
, &op
.cl
);
1571 closure_set_stopped(&c
->gc
.cl
);
1572 closure_put(&c
->gc
.cl
);
1575 closure_sync(&op
.cl
);
1576 c
->sb
.last_mount
= get_seconds();
1577 bcache_write_super(c
);
1579 list_for_each_entry_safe(dc
, t
, &uncached_devices
, list
)
1580 bch_cached_dev_attach(dc
, c
);
1586 closure_set_stopped(&c
->gc
.cl
);
1587 closure_put(&c
->gc
.cl
);
1589 closure_sync(&op
.cl
);
1590 /* XXX: test this, it's broken */
1591 bch_cache_set_error(c
, err
);
1594 static bool can_attach_cache(struct cache
*ca
, struct cache_set
*c
)
1596 return ca
->sb
.block_size
== c
->sb
.block_size
&&
1597 ca
->sb
.bucket_size
== c
->sb
.block_size
&&
1598 ca
->sb
.nr_in_set
== c
->sb
.nr_in_set
;
1601 static const char *register_cache_set(struct cache
*ca
)
1604 const char *err
= "cannot allocate memory";
1605 struct cache_set
*c
;
1607 list_for_each_entry(c
, &bch_cache_sets
, list
)
1608 if (!memcmp(c
->sb
.set_uuid
, ca
->sb
.set_uuid
, 16)) {
1609 if (c
->cache
[ca
->sb
.nr_this_dev
])
1610 return "duplicate cache set member";
1612 if (!can_attach_cache(ca
, c
))
1613 return "cache sb does not match set";
1615 if (!CACHE_SYNC(&ca
->sb
))
1616 SET_CACHE_SYNC(&c
->sb
, false);
1621 c
= bch_cache_set_alloc(&ca
->sb
);
1625 err
= "error creating kobject";
1626 if (kobject_add(&c
->kobj
, bcache_kobj
, "%pU", c
->sb
.set_uuid
) ||
1627 kobject_add(&c
->internal
, &c
->kobj
, "internal"))
1630 if (bch_cache_accounting_add_kobjs(&c
->accounting
, &c
->kobj
))
1633 bch_debug_init_cache_set(c
);
1635 list_add(&c
->list
, &bch_cache_sets
);
1637 sprintf(buf
, "cache%i", ca
->sb
.nr_this_dev
);
1638 if (sysfs_create_link(&ca
->kobj
, &c
->kobj
, "set") ||
1639 sysfs_create_link(&c
->kobj
, &ca
->kobj
, buf
))
1642 if (ca
->sb
.seq
> c
->sb
.seq
) {
1643 c
->sb
.version
= ca
->sb
.version
;
1644 memcpy(c
->sb
.set_uuid
, ca
->sb
.set_uuid
, 16);
1645 c
->sb
.flags
= ca
->sb
.flags
;
1646 c
->sb
.seq
= ca
->sb
.seq
;
1647 pr_debug("set version = %llu", c
->sb
.version
);
1651 ca
->set
->cache
[ca
->sb
.nr_this_dev
] = ca
;
1652 c
->cache_by_alloc
[c
->caches_loaded
++] = ca
;
1654 if (c
->caches_loaded
== c
->sb
.nr_in_set
)
1659 bch_cache_set_unregister(c
);
1665 void bch_cache_release(struct kobject
*kobj
)
1667 struct cache
*ca
= container_of(kobj
, struct cache
, kobj
);
1670 ca
->set
->cache
[ca
->sb
.nr_this_dev
] = NULL
;
1672 bch_cache_allocator_exit(ca
);
1674 bio_split_pool_free(&ca
->bio_split_hook
);
1676 if (ca
->alloc_workqueue
)
1677 destroy_workqueue(ca
->alloc_workqueue
);
1679 free_pages((unsigned long) ca
->disk_buckets
, ilog2(bucket_pages(ca
)));
1680 kfree(ca
->prio_buckets
);
1683 free_heap(&ca
->heap
);
1684 free_fifo(&ca
->unused
);
1685 free_fifo(&ca
->free_inc
);
1686 free_fifo(&ca
->free
);
1688 if (ca
->sb_bio
.bi_inline_vecs
[0].bv_page
)
1689 put_page(ca
->sb_bio
.bi_io_vec
[0].bv_page
);
1691 if (!IS_ERR_OR_NULL(ca
->bdev
)) {
1692 blk_sync_queue(bdev_get_queue(ca
->bdev
));
1693 blkdev_put(ca
->bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
1697 module_put(THIS_MODULE
);
1700 static int cache_alloc(struct cache_sb
*sb
, struct cache
*ca
)
1705 __module_get(THIS_MODULE
);
1706 kobject_init(&ca
->kobj
, &bch_cache_ktype
);
1708 INIT_LIST_HEAD(&ca
->discards
);
1710 bio_init(&ca
->journal
.bio
);
1711 ca
->journal
.bio
.bi_max_vecs
= 8;
1712 ca
->journal
.bio
.bi_io_vec
= ca
->journal
.bio
.bi_inline_vecs
;
1714 free
= roundup_pow_of_two(ca
->sb
.nbuckets
) >> 9;
1715 free
= max_t(size_t, free
, (prio_buckets(ca
) + 8) * 2);
1717 if (!init_fifo(&ca
->free
, free
, GFP_KERNEL
) ||
1718 !init_fifo(&ca
->free_inc
, free
<< 2, GFP_KERNEL
) ||
1719 !init_fifo(&ca
->unused
, free
<< 2, GFP_KERNEL
) ||
1720 !init_heap(&ca
->heap
, free
<< 3, GFP_KERNEL
) ||
1721 !(ca
->buckets
= vzalloc(sizeof(struct bucket
) *
1722 ca
->sb
.nbuckets
)) ||
1723 !(ca
->prio_buckets
= kzalloc(sizeof(uint64_t) * prio_buckets(ca
) *
1725 !(ca
->disk_buckets
= alloc_bucket_pages(GFP_KERNEL
, ca
)) ||
1726 !(ca
->alloc_workqueue
= alloc_workqueue("bch_allocator", 0, 1)) ||
1727 bio_split_pool_init(&ca
->bio_split_hook
))
1730 ca
->prio_last_buckets
= ca
->prio_buckets
+ prio_buckets(ca
);
1732 for_each_bucket(b
, ca
)
1733 atomic_set(&b
->pin
, 0);
1735 if (bch_cache_allocator_init(ca
))
1740 kobject_put(&ca
->kobj
);
1744 static void register_cache(struct cache_sb
*sb
, struct page
*sb_page
,
1745 struct block_device
*bdev
, struct cache
*ca
)
1747 char name
[BDEVNAME_SIZE
];
1748 const char *err
= "cannot allocate memory";
1750 memcpy(&ca
->sb
, sb
, sizeof(struct cache_sb
));
1752 ca
->bdev
->bd_holder
= ca
;
1754 bio_init(&ca
->sb_bio
);
1755 ca
->sb_bio
.bi_max_vecs
= 1;
1756 ca
->sb_bio
.bi_io_vec
= ca
->sb_bio
.bi_inline_vecs
;
1757 ca
->sb_bio
.bi_io_vec
[0].bv_page
= sb_page
;
1760 if (blk_queue_discard(bdev_get_queue(ca
->bdev
)))
1761 ca
->discard
= CACHE_DISCARD(&ca
->sb
);
1763 if (cache_alloc(sb
, ca
) != 0)
1766 err
= "error creating kobject";
1767 if (kobject_add(&ca
->kobj
, &part_to_dev(bdev
->bd_part
)->kobj
, "bcache"))
1770 err
= register_cache_set(ca
);
1774 pr_info("registered cache device %s", bdevname(bdev
, name
));
1777 pr_notice("error opening %s: %s", bdevname(bdev
, name
), err
);
1778 kobject_put(&ca
->kobj
);
1781 /* Global interfaces/init */
1783 static ssize_t
register_bcache(struct kobject
*, struct kobj_attribute
*,
1784 const char *, size_t);
1786 kobj_attribute_write(register, register_bcache
);
1787 kobj_attribute_write(register_quiet
, register_bcache
);
1789 static ssize_t
register_bcache(struct kobject
*k
, struct kobj_attribute
*attr
,
1790 const char *buffer
, size_t size
)
1793 const char *err
= "cannot allocate memory";
1795 struct cache_sb
*sb
= NULL
;
1796 struct block_device
*bdev
= NULL
;
1797 struct page
*sb_page
= NULL
;
1799 if (!try_module_get(THIS_MODULE
))
1802 mutex_lock(&bch_register_lock
);
1804 if (!(path
= kstrndup(buffer
, size
, GFP_KERNEL
)) ||
1805 !(sb
= kmalloc(sizeof(struct cache_sb
), GFP_KERNEL
)))
1808 err
= "failed to open device";
1809 bdev
= blkdev_get_by_path(strim(path
),
1810 FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
,
1813 if (bdev
== ERR_PTR(-EBUSY
))
1814 err
= "device busy";
1818 err
= "failed to set blocksize";
1819 if (set_blocksize(bdev
, 4096))
1822 err
= read_super(sb
, bdev
, &sb_page
);
1826 if (SB_IS_BDEV(sb
)) {
1827 struct cached_dev
*dc
= kzalloc(sizeof(*dc
), GFP_KERNEL
);
1831 register_bdev(sb
, sb_page
, bdev
, dc
);
1833 struct cache
*ca
= kzalloc(sizeof(*ca
), GFP_KERNEL
);
1837 register_cache(sb
, sb_page
, bdev
, ca
);
1844 mutex_unlock(&bch_register_lock
);
1845 module_put(THIS_MODULE
);
1849 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
1851 if (attr
!= &ksysfs_register_quiet
)
1852 pr_info("error opening %s: %s", path
, err
);
1857 static int bcache_reboot(struct notifier_block
*n
, unsigned long code
, void *x
)
1859 if (code
== SYS_DOWN
||
1861 code
== SYS_POWER_OFF
) {
1863 unsigned long start
= jiffies
;
1864 bool stopped
= false;
1866 struct cache_set
*c
, *tc
;
1867 struct cached_dev
*dc
, *tdc
;
1869 mutex_lock(&bch_register_lock
);
1871 if (list_empty(&bch_cache_sets
) &&
1872 list_empty(&uncached_devices
))
1875 pr_info("Stopping all devices:");
1877 list_for_each_entry_safe(c
, tc
, &bch_cache_sets
, list
)
1878 bch_cache_set_stop(c
);
1880 list_for_each_entry_safe(dc
, tdc
, &uncached_devices
, list
)
1881 bcache_device_stop(&dc
->disk
);
1883 /* What's a condition variable? */
1885 long timeout
= start
+ 2 * HZ
- jiffies
;
1887 stopped
= list_empty(&bch_cache_sets
) &&
1888 list_empty(&uncached_devices
);
1890 if (timeout
< 0 || stopped
)
1893 prepare_to_wait(&unregister_wait
, &wait
,
1894 TASK_UNINTERRUPTIBLE
);
1896 mutex_unlock(&bch_register_lock
);
1897 schedule_timeout(timeout
);
1898 mutex_lock(&bch_register_lock
);
1901 finish_wait(&unregister_wait
, &wait
);
1904 pr_info("All devices stopped");
1906 pr_notice("Timeout waiting for devices to be closed");
1908 mutex_unlock(&bch_register_lock
);
1914 static struct notifier_block reboot
= {
1915 .notifier_call
= bcache_reboot
,
1916 .priority
= INT_MAX
, /* before any real devices */
1919 static void bcache_exit(void)
1922 bch_writeback_exit();
1926 kobject_put(bcache_kobj
);
1928 destroy_workqueue(bcache_wq
);
1929 unregister_blkdev(bcache_major
, "bcache");
1930 unregister_reboot_notifier(&reboot
);
1933 static int __init
bcache_init(void)
1935 static const struct attribute
*files
[] = {
1936 &ksysfs_register
.attr
,
1937 &ksysfs_register_quiet
.attr
,
1941 mutex_init(&bch_register_lock
);
1942 init_waitqueue_head(&unregister_wait
);
1943 register_reboot_notifier(&reboot
);
1944 closure_debug_init();
1946 bcache_major
= register_blkdev(0, "bcache");
1947 if (bcache_major
< 0)
1948 return bcache_major
;
1950 if (!(bcache_wq
= create_workqueue("bcache")) ||
1951 !(bcache_kobj
= kobject_create_and_add("bcache", fs_kobj
)) ||
1952 sysfs_create_files(bcache_kobj
, files
) ||
1954 bch_request_init() ||
1955 bch_writeback_init() ||
1956 bch_debug_init(bcache_kobj
))
1965 module_exit(bcache_exit
);
1966 module_init(bcache_init
);