Linux 4.16.11
[linux/fpc-iii.git] / drivers / md / bcache / super.c
blobf2273143b3cb2384109ba47bf3518adf31337ad8
1 /*
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.
7 */
9 #include "bcache.h"
10 #include "btree.h"
11 #include "debug.h"
12 #include "extents.h"
13 #include "request.h"
14 #include "writeback.h"
16 #include <linux/blkdev.h>
17 #include <linux/buffer_head.h>
18 #include <linux/debugfs.h>
19 #include <linux/genhd.h>
20 #include <linux/idr.h>
21 #include <linux/kthread.h>
22 #include <linux/module.h>
23 #include <linux/random.h>
24 #include <linux/reboot.h>
25 #include <linux/sysfs.h>
27 MODULE_LICENSE("GPL");
28 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
30 static const char bcache_magic[] = {
31 0xc6, 0x85, 0x73, 0xf6, 0x4e, 0x1a, 0x45, 0xca,
32 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81
35 static const char invalid_uuid[] = {
36 0xa0, 0x3e, 0xf8, 0xed, 0x3e, 0xe1, 0xb8, 0x78,
37 0xc8, 0x50, 0xfc, 0x5e, 0xcb, 0x16, 0xcd, 0x99
40 /* Default is -1; we skip past it for struct cached_dev's cache mode */
41 const char * const bch_cache_modes[] = {
42 "default",
43 "writethrough",
44 "writeback",
45 "writearound",
46 "none",
47 NULL
50 static struct kobject *bcache_kobj;
51 struct mutex bch_register_lock;
52 LIST_HEAD(bch_cache_sets);
53 static LIST_HEAD(uncached_devices);
55 static int bcache_major;
56 static DEFINE_IDA(bcache_device_idx);
57 static wait_queue_head_t unregister_wait;
58 struct workqueue_struct *bcache_wq;
60 #define BTREE_MAX_PAGES (256 * 1024 / PAGE_SIZE)
61 /* limitation of partitions number on single bcache device */
62 #define BCACHE_MINORS 128
63 /* limitation of bcache devices number on single system */
64 #define BCACHE_DEVICE_IDX_MAX ((1U << MINORBITS)/BCACHE_MINORS)
66 /* Superblock */
68 static const char *read_super(struct cache_sb *sb, struct block_device *bdev,
69 struct page **res)
71 const char *err;
72 struct cache_sb *s;
73 struct buffer_head *bh = __bread(bdev, 1, SB_SIZE);
74 unsigned i;
76 if (!bh)
77 return "IO error";
79 s = (struct cache_sb *) bh->b_data;
81 sb->offset = le64_to_cpu(s->offset);
82 sb->version = le64_to_cpu(s->version);
84 memcpy(sb->magic, s->magic, 16);
85 memcpy(sb->uuid, s->uuid, 16);
86 memcpy(sb->set_uuid, s->set_uuid, 16);
87 memcpy(sb->label, s->label, SB_LABEL_SIZE);
89 sb->flags = le64_to_cpu(s->flags);
90 sb->seq = le64_to_cpu(s->seq);
91 sb->last_mount = le32_to_cpu(s->last_mount);
92 sb->first_bucket = le16_to_cpu(s->first_bucket);
93 sb->keys = le16_to_cpu(s->keys);
95 for (i = 0; i < SB_JOURNAL_BUCKETS; i++)
96 sb->d[i] = le64_to_cpu(s->d[i]);
98 pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u",
99 sb->version, sb->flags, sb->seq, sb->keys);
101 err = "Not a bcache superblock";
102 if (sb->offset != SB_SECTOR)
103 goto err;
105 if (memcmp(sb->magic, bcache_magic, 16))
106 goto err;
108 err = "Too many journal buckets";
109 if (sb->keys > SB_JOURNAL_BUCKETS)
110 goto err;
112 err = "Bad checksum";
113 if (s->csum != csum_set(s))
114 goto err;
116 err = "Bad UUID";
117 if (bch_is_zero(sb->uuid, 16))
118 goto err;
120 sb->block_size = le16_to_cpu(s->block_size);
122 err = "Superblock block size smaller than device block size";
123 if (sb->block_size << 9 < bdev_logical_block_size(bdev))
124 goto err;
126 switch (sb->version) {
127 case BCACHE_SB_VERSION_BDEV:
128 sb->data_offset = BDEV_DATA_START_DEFAULT;
129 break;
130 case BCACHE_SB_VERSION_BDEV_WITH_OFFSET:
131 sb->data_offset = le64_to_cpu(s->data_offset);
133 err = "Bad data offset";
134 if (sb->data_offset < BDEV_DATA_START_DEFAULT)
135 goto err;
137 break;
138 case BCACHE_SB_VERSION_CDEV:
139 case BCACHE_SB_VERSION_CDEV_WITH_UUID:
140 sb->nbuckets = le64_to_cpu(s->nbuckets);
141 sb->bucket_size = le16_to_cpu(s->bucket_size);
143 sb->nr_in_set = le16_to_cpu(s->nr_in_set);
144 sb->nr_this_dev = le16_to_cpu(s->nr_this_dev);
146 err = "Too many buckets";
147 if (sb->nbuckets > LONG_MAX)
148 goto err;
150 err = "Not enough buckets";
151 if (sb->nbuckets < 1 << 7)
152 goto err;
154 err = "Bad block/bucket size";
155 if (!is_power_of_2(sb->block_size) ||
156 sb->block_size > PAGE_SECTORS ||
157 !is_power_of_2(sb->bucket_size) ||
158 sb->bucket_size < PAGE_SECTORS)
159 goto err;
161 err = "Invalid superblock: device too small";
162 if (get_capacity(bdev->bd_disk) < sb->bucket_size * sb->nbuckets)
163 goto err;
165 err = "Bad UUID";
166 if (bch_is_zero(sb->set_uuid, 16))
167 goto err;
169 err = "Bad cache device number in set";
170 if (!sb->nr_in_set ||
171 sb->nr_in_set <= sb->nr_this_dev ||
172 sb->nr_in_set > MAX_CACHES_PER_SET)
173 goto err;
175 err = "Journal buckets not sequential";
176 for (i = 0; i < sb->keys; i++)
177 if (sb->d[i] != sb->first_bucket + i)
178 goto err;
180 err = "Too many journal buckets";
181 if (sb->first_bucket + sb->keys > sb->nbuckets)
182 goto err;
184 err = "Invalid superblock: first bucket comes before end of super";
185 if (sb->first_bucket * sb->bucket_size < 16)
186 goto err;
188 break;
189 default:
190 err = "Unsupported superblock version";
191 goto err;
194 sb->last_mount = get_seconds();
195 err = NULL;
197 get_page(bh->b_page);
198 *res = bh->b_page;
199 err:
200 put_bh(bh);
201 return err;
204 static void write_bdev_super_endio(struct bio *bio)
206 struct cached_dev *dc = bio->bi_private;
207 /* XXX: error checking */
209 closure_put(&dc->sb_write);
212 static void __write_super(struct cache_sb *sb, struct bio *bio)
214 struct cache_sb *out = page_address(bio_first_page_all(bio));
215 unsigned i;
217 bio->bi_iter.bi_sector = SB_SECTOR;
218 bio->bi_iter.bi_size = SB_SIZE;
219 bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
220 bch_bio_map(bio, NULL);
222 out->offset = cpu_to_le64(sb->offset);
223 out->version = cpu_to_le64(sb->version);
225 memcpy(out->uuid, sb->uuid, 16);
226 memcpy(out->set_uuid, sb->set_uuid, 16);
227 memcpy(out->label, sb->label, SB_LABEL_SIZE);
229 out->flags = cpu_to_le64(sb->flags);
230 out->seq = cpu_to_le64(sb->seq);
232 out->last_mount = cpu_to_le32(sb->last_mount);
233 out->first_bucket = cpu_to_le16(sb->first_bucket);
234 out->keys = cpu_to_le16(sb->keys);
236 for (i = 0; i < sb->keys; i++)
237 out->d[i] = cpu_to_le64(sb->d[i]);
239 out->csum = csum_set(out);
241 pr_debug("ver %llu, flags %llu, seq %llu",
242 sb->version, sb->flags, sb->seq);
244 submit_bio(bio);
247 static void bch_write_bdev_super_unlock(struct closure *cl)
249 struct cached_dev *dc = container_of(cl, struct cached_dev, sb_write);
251 up(&dc->sb_write_mutex);
254 void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent)
256 struct closure *cl = &dc->sb_write;
257 struct bio *bio = &dc->sb_bio;
259 down(&dc->sb_write_mutex);
260 closure_init(cl, parent);
262 bio_reset(bio);
263 bio_set_dev(bio, dc->bdev);
264 bio->bi_end_io = write_bdev_super_endio;
265 bio->bi_private = dc;
267 closure_get(cl);
268 __write_super(&dc->sb, bio);
270 closure_return_with_destructor(cl, bch_write_bdev_super_unlock);
273 static void write_super_endio(struct bio *bio)
275 struct cache *ca = bio->bi_private;
277 /* is_read = 0 */
278 bch_count_io_errors(ca, bio->bi_status, 0,
279 "writing superblock");
280 closure_put(&ca->set->sb_write);
283 static void bcache_write_super_unlock(struct closure *cl)
285 struct cache_set *c = container_of(cl, struct cache_set, sb_write);
287 up(&c->sb_write_mutex);
290 void bcache_write_super(struct cache_set *c)
292 struct closure *cl = &c->sb_write;
293 struct cache *ca;
294 unsigned i;
296 down(&c->sb_write_mutex);
297 closure_init(cl, &c->cl);
299 c->sb.seq++;
301 for_each_cache(ca, c, i) {
302 struct bio *bio = &ca->sb_bio;
304 ca->sb.version = BCACHE_SB_VERSION_CDEV_WITH_UUID;
305 ca->sb.seq = c->sb.seq;
306 ca->sb.last_mount = c->sb.last_mount;
308 SET_CACHE_SYNC(&ca->sb, CACHE_SYNC(&c->sb));
310 bio_reset(bio);
311 bio_set_dev(bio, ca->bdev);
312 bio->bi_end_io = write_super_endio;
313 bio->bi_private = ca;
315 closure_get(cl);
316 __write_super(&ca->sb, bio);
319 closure_return_with_destructor(cl, bcache_write_super_unlock);
322 /* UUID io */
324 static void uuid_endio(struct bio *bio)
326 struct closure *cl = bio->bi_private;
327 struct cache_set *c = container_of(cl, struct cache_set, uuid_write);
329 cache_set_err_on(bio->bi_status, c, "accessing uuids");
330 bch_bbio_free(bio, c);
331 closure_put(cl);
334 static void uuid_io_unlock(struct closure *cl)
336 struct cache_set *c = container_of(cl, struct cache_set, uuid_write);
338 up(&c->uuid_write_mutex);
341 static void uuid_io(struct cache_set *c, int op, unsigned long op_flags,
342 struct bkey *k, struct closure *parent)
344 struct closure *cl = &c->uuid_write;
345 struct uuid_entry *u;
346 unsigned i;
347 char buf[80];
349 BUG_ON(!parent);
350 down(&c->uuid_write_mutex);
351 closure_init(cl, parent);
353 for (i = 0; i < KEY_PTRS(k); i++) {
354 struct bio *bio = bch_bbio_alloc(c);
356 bio->bi_opf = REQ_SYNC | REQ_META | op_flags;
357 bio->bi_iter.bi_size = KEY_SIZE(k) << 9;
359 bio->bi_end_io = uuid_endio;
360 bio->bi_private = cl;
361 bio_set_op_attrs(bio, op, REQ_SYNC|REQ_META|op_flags);
362 bch_bio_map(bio, c->uuids);
364 bch_submit_bbio(bio, c, k, i);
366 if (op != REQ_OP_WRITE)
367 break;
370 bch_extent_to_text(buf, sizeof(buf), k);
371 pr_debug("%s UUIDs at %s", op == REQ_OP_WRITE ? "wrote" : "read", buf);
373 for (u = c->uuids; u < c->uuids + c->nr_uuids; u++)
374 if (!bch_is_zero(u->uuid, 16))
375 pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u",
376 u - c->uuids, u->uuid, u->label,
377 u->first_reg, u->last_reg, u->invalidated);
379 closure_return_with_destructor(cl, uuid_io_unlock);
382 static char *uuid_read(struct cache_set *c, struct jset *j, struct closure *cl)
384 struct bkey *k = &j->uuid_bucket;
386 if (__bch_btree_ptr_invalid(c, k))
387 return "bad uuid pointer";
389 bkey_copy(&c->uuid_bucket, k);
390 uuid_io(c, REQ_OP_READ, 0, k, cl);
392 if (j->version < BCACHE_JSET_VERSION_UUIDv1) {
393 struct uuid_entry_v0 *u0 = (void *) c->uuids;
394 struct uuid_entry *u1 = (void *) c->uuids;
395 int i;
397 closure_sync(cl);
400 * Since the new uuid entry is bigger than the old, we have to
401 * convert starting at the highest memory address and work down
402 * in order to do it in place
405 for (i = c->nr_uuids - 1;
406 i >= 0;
407 --i) {
408 memcpy(u1[i].uuid, u0[i].uuid, 16);
409 memcpy(u1[i].label, u0[i].label, 32);
411 u1[i].first_reg = u0[i].first_reg;
412 u1[i].last_reg = u0[i].last_reg;
413 u1[i].invalidated = u0[i].invalidated;
415 u1[i].flags = 0;
416 u1[i].sectors = 0;
420 return NULL;
423 static int __uuid_write(struct cache_set *c)
425 BKEY_PADDED(key) k;
426 struct closure cl;
427 closure_init_stack(&cl);
429 lockdep_assert_held(&bch_register_lock);
431 if (bch_bucket_alloc_set(c, RESERVE_BTREE, &k.key, 1, true))
432 return 1;
434 SET_KEY_SIZE(&k.key, c->sb.bucket_size);
435 uuid_io(c, REQ_OP_WRITE, 0, &k.key, &cl);
436 closure_sync(&cl);
438 bkey_copy(&c->uuid_bucket, &k.key);
439 bkey_put(c, &k.key);
440 return 0;
443 int bch_uuid_write(struct cache_set *c)
445 int ret = __uuid_write(c);
447 if (!ret)
448 bch_journal_meta(c, NULL);
450 return ret;
453 static struct uuid_entry *uuid_find(struct cache_set *c, const char *uuid)
455 struct uuid_entry *u;
457 for (u = c->uuids;
458 u < c->uuids + c->nr_uuids; u++)
459 if (!memcmp(u->uuid, uuid, 16))
460 return u;
462 return NULL;
465 static struct uuid_entry *uuid_find_empty(struct cache_set *c)
467 static const char zero_uuid[16] = "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0";
468 return uuid_find(c, zero_uuid);
472 * Bucket priorities/gens:
474 * For each bucket, we store on disk its
475 * 8 bit gen
476 * 16 bit priority
478 * See alloc.c for an explanation of the gen. The priority is used to implement
479 * lru (and in the future other) cache replacement policies; for most purposes
480 * it's just an opaque integer.
482 * The gens and the priorities don't have a whole lot to do with each other, and
483 * it's actually the gens that must be written out at specific times - it's no
484 * big deal if the priorities don't get written, if we lose them we just reuse
485 * buckets in suboptimal order.
487 * On disk they're stored in a packed array, and in as many buckets are required
488 * to fit them all. The buckets we use to store them form a list; the journal
489 * header points to the first bucket, the first bucket points to the second
490 * bucket, et cetera.
492 * This code is used by the allocation code; periodically (whenever it runs out
493 * of buckets to allocate from) the allocation code will invalidate some
494 * buckets, but it can't use those buckets until their new gens are safely on
495 * disk.
498 static void prio_endio(struct bio *bio)
500 struct cache *ca = bio->bi_private;
502 cache_set_err_on(bio->bi_status, ca->set, "accessing priorities");
503 bch_bbio_free(bio, ca->set);
504 closure_put(&ca->prio);
507 static void prio_io(struct cache *ca, uint64_t bucket, int op,
508 unsigned long op_flags)
510 struct closure *cl = &ca->prio;
511 struct bio *bio = bch_bbio_alloc(ca->set);
513 closure_init_stack(cl);
515 bio->bi_iter.bi_sector = bucket * ca->sb.bucket_size;
516 bio_set_dev(bio, ca->bdev);
517 bio->bi_iter.bi_size = bucket_bytes(ca);
519 bio->bi_end_io = prio_endio;
520 bio->bi_private = ca;
521 bio_set_op_attrs(bio, op, REQ_SYNC|REQ_META|op_flags);
522 bch_bio_map(bio, ca->disk_buckets);
524 closure_bio_submit(bio, &ca->prio);
525 closure_sync(cl);
528 void bch_prio_write(struct cache *ca)
530 int i;
531 struct bucket *b;
532 struct closure cl;
534 closure_init_stack(&cl);
536 lockdep_assert_held(&ca->set->bucket_lock);
538 ca->disk_buckets->seq++;
540 atomic_long_add(ca->sb.bucket_size * prio_buckets(ca),
541 &ca->meta_sectors_written);
543 //pr_debug("free %zu, free_inc %zu, unused %zu", fifo_used(&ca->free),
544 // fifo_used(&ca->free_inc), fifo_used(&ca->unused));
546 for (i = prio_buckets(ca) - 1; i >= 0; --i) {
547 long bucket;
548 struct prio_set *p = ca->disk_buckets;
549 struct bucket_disk *d = p->data;
550 struct bucket_disk *end = d + prios_per_bucket(ca);
552 for (b = ca->buckets + i * prios_per_bucket(ca);
553 b < ca->buckets + ca->sb.nbuckets && d < end;
554 b++, d++) {
555 d->prio = cpu_to_le16(b->prio);
556 d->gen = b->gen;
559 p->next_bucket = ca->prio_buckets[i + 1];
560 p->magic = pset_magic(&ca->sb);
561 p->csum = bch_crc64(&p->magic, bucket_bytes(ca) - 8);
563 bucket = bch_bucket_alloc(ca, RESERVE_PRIO, true);
564 BUG_ON(bucket == -1);
566 mutex_unlock(&ca->set->bucket_lock);
567 prio_io(ca, bucket, REQ_OP_WRITE, 0);
568 mutex_lock(&ca->set->bucket_lock);
570 ca->prio_buckets[i] = bucket;
571 atomic_dec_bug(&ca->buckets[bucket].pin);
574 mutex_unlock(&ca->set->bucket_lock);
576 bch_journal_meta(ca->set, &cl);
577 closure_sync(&cl);
579 mutex_lock(&ca->set->bucket_lock);
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 if (ca->prio_last_buckets[i])
587 __bch_bucket_free(ca,
588 &ca->buckets[ca->prio_last_buckets[i]]);
590 ca->prio_last_buckets[i] = ca->prio_buckets[i];
594 static void prio_read(struct cache *ca, uint64_t bucket)
596 struct prio_set *p = ca->disk_buckets;
597 struct bucket_disk *d = p->data + prios_per_bucket(ca), *end = d;
598 struct bucket *b;
599 unsigned bucket_nr = 0;
601 for (b = ca->buckets;
602 b < ca->buckets + ca->sb.nbuckets;
603 b++, d++) {
604 if (d == end) {
605 ca->prio_buckets[bucket_nr] = bucket;
606 ca->prio_last_buckets[bucket_nr] = bucket;
607 bucket_nr++;
609 prio_io(ca, bucket, REQ_OP_READ, 0);
611 if (p->csum != bch_crc64(&p->magic, bucket_bytes(ca) - 8))
612 pr_warn("bad csum reading priorities");
614 if (p->magic != pset_magic(&ca->sb))
615 pr_warn("bad magic reading priorities");
617 bucket = p->next_bucket;
618 d = p->data;
621 b->prio = le16_to_cpu(d->prio);
622 b->gen = b->last_gc = d->gen;
626 /* Bcache device */
628 static int open_dev(struct block_device *b, fmode_t mode)
630 struct bcache_device *d = b->bd_disk->private_data;
631 if (test_bit(BCACHE_DEV_CLOSING, &d->flags))
632 return -ENXIO;
634 closure_get(&d->cl);
635 return 0;
638 static void release_dev(struct gendisk *b, fmode_t mode)
640 struct bcache_device *d = b->private_data;
641 closure_put(&d->cl);
644 static int ioctl_dev(struct block_device *b, fmode_t mode,
645 unsigned int cmd, unsigned long arg)
647 struct bcache_device *d = b->bd_disk->private_data;
648 return d->ioctl(d, mode, cmd, arg);
651 static const struct block_device_operations bcache_ops = {
652 .open = open_dev,
653 .release = release_dev,
654 .ioctl = ioctl_dev,
655 .owner = THIS_MODULE,
658 void bcache_device_stop(struct bcache_device *d)
660 if (!test_and_set_bit(BCACHE_DEV_CLOSING, &d->flags))
661 closure_queue(&d->cl);
664 static void bcache_device_unlink(struct bcache_device *d)
666 lockdep_assert_held(&bch_register_lock);
668 if (d->c && !test_and_set_bit(BCACHE_DEV_UNLINK_DONE, &d->flags)) {
669 unsigned i;
670 struct cache *ca;
672 sysfs_remove_link(&d->c->kobj, d->name);
673 sysfs_remove_link(&d->kobj, "cache");
675 for_each_cache(ca, d->c, i)
676 bd_unlink_disk_holder(ca->bdev, d->disk);
680 static void bcache_device_link(struct bcache_device *d, struct cache_set *c,
681 const char *name)
683 unsigned i;
684 struct cache *ca;
686 for_each_cache(ca, d->c, i)
687 bd_link_disk_holder(ca->bdev, d->disk);
689 snprintf(d->name, BCACHEDEVNAME_SIZE,
690 "%s%u", name, d->id);
692 WARN(sysfs_create_link(&d->kobj, &c->kobj, "cache") ||
693 sysfs_create_link(&c->kobj, &d->kobj, d->name),
694 "Couldn't create device <-> cache set symlinks");
696 clear_bit(BCACHE_DEV_UNLINK_DONE, &d->flags);
699 static void bcache_device_detach(struct bcache_device *d)
701 lockdep_assert_held(&bch_register_lock);
703 if (test_bit(BCACHE_DEV_DETACHING, &d->flags)) {
704 struct uuid_entry *u = d->c->uuids + d->id;
706 SET_UUID_FLASH_ONLY(u, 0);
707 memcpy(u->uuid, invalid_uuid, 16);
708 u->invalidated = cpu_to_le32(get_seconds());
709 bch_uuid_write(d->c);
712 bcache_device_unlink(d);
714 d->c->devices[d->id] = NULL;
715 closure_put(&d->c->caching);
716 d->c = NULL;
719 static void bcache_device_attach(struct bcache_device *d, struct cache_set *c,
720 unsigned id)
722 d->id = id;
723 d->c = c;
724 c->devices[id] = d;
726 if (id >= c->devices_max_used)
727 c->devices_max_used = id + 1;
729 closure_get(&c->caching);
732 static inline int first_minor_to_idx(int first_minor)
734 return (first_minor/BCACHE_MINORS);
737 static inline int idx_to_first_minor(int idx)
739 return (idx * BCACHE_MINORS);
742 static void bcache_device_free(struct bcache_device *d)
744 lockdep_assert_held(&bch_register_lock);
746 pr_info("%s stopped", d->disk->disk_name);
748 if (d->c)
749 bcache_device_detach(d);
750 if (d->disk && d->disk->flags & GENHD_FL_UP)
751 del_gendisk(d->disk);
752 if (d->disk && d->disk->queue)
753 blk_cleanup_queue(d->disk->queue);
754 if (d->disk) {
755 ida_simple_remove(&bcache_device_idx,
756 first_minor_to_idx(d->disk->first_minor));
757 put_disk(d->disk);
760 if (d->bio_split)
761 bioset_free(d->bio_split);
762 kvfree(d->full_dirty_stripes);
763 kvfree(d->stripe_sectors_dirty);
765 closure_debug_destroy(&d->cl);
768 static int bcache_device_init(struct bcache_device *d, unsigned block_size,
769 sector_t sectors)
771 struct request_queue *q;
772 size_t n;
773 int idx;
775 if (!d->stripe_size)
776 d->stripe_size = 1 << 31;
778 d->nr_stripes = DIV_ROUND_UP_ULL(sectors, d->stripe_size);
780 if (!d->nr_stripes ||
781 d->nr_stripes > INT_MAX ||
782 d->nr_stripes > SIZE_MAX / sizeof(atomic_t)) {
783 pr_err("nr_stripes too large or invalid: %u (start sector beyond end of disk?)",
784 (unsigned)d->nr_stripes);
785 return -ENOMEM;
788 n = d->nr_stripes * sizeof(atomic_t);
789 d->stripe_sectors_dirty = kvzalloc(n, GFP_KERNEL);
790 if (!d->stripe_sectors_dirty)
791 return -ENOMEM;
793 n = BITS_TO_LONGS(d->nr_stripes) * sizeof(unsigned long);
794 d->full_dirty_stripes = kvzalloc(n, GFP_KERNEL);
795 if (!d->full_dirty_stripes)
796 return -ENOMEM;
798 idx = ida_simple_get(&bcache_device_idx, 0,
799 BCACHE_DEVICE_IDX_MAX, GFP_KERNEL);
800 if (idx < 0)
801 return idx;
803 if (!(d->bio_split = bioset_create(4, offsetof(struct bbio, bio),
804 BIOSET_NEED_BVECS |
805 BIOSET_NEED_RESCUER)) ||
806 !(d->disk = alloc_disk(BCACHE_MINORS))) {
807 ida_simple_remove(&bcache_device_idx, idx);
808 return -ENOMEM;
811 set_capacity(d->disk, sectors);
812 snprintf(d->disk->disk_name, DISK_NAME_LEN, "bcache%i", idx);
814 d->disk->major = bcache_major;
815 d->disk->first_minor = idx_to_first_minor(idx);
816 d->disk->fops = &bcache_ops;
817 d->disk->private_data = d;
819 q = blk_alloc_queue(GFP_KERNEL);
820 if (!q)
821 return -ENOMEM;
823 blk_queue_make_request(q, NULL);
824 d->disk->queue = q;
825 q->queuedata = d;
826 q->backing_dev_info->congested_data = d;
827 q->limits.max_hw_sectors = UINT_MAX;
828 q->limits.max_sectors = UINT_MAX;
829 q->limits.max_segment_size = UINT_MAX;
830 q->limits.max_segments = BIO_MAX_PAGES;
831 blk_queue_max_discard_sectors(q, UINT_MAX);
832 q->limits.discard_granularity = 512;
833 q->limits.io_min = block_size;
834 q->limits.logical_block_size = block_size;
835 q->limits.physical_block_size = block_size;
836 set_bit(QUEUE_FLAG_NONROT, &d->disk->queue->queue_flags);
837 clear_bit(QUEUE_FLAG_ADD_RANDOM, &d->disk->queue->queue_flags);
838 set_bit(QUEUE_FLAG_DISCARD, &d->disk->queue->queue_flags);
840 blk_queue_write_cache(q, true, true);
842 return 0;
845 /* Cached device */
847 static void calc_cached_dev_sectors(struct cache_set *c)
849 uint64_t sectors = 0;
850 struct cached_dev *dc;
852 list_for_each_entry(dc, &c->cached_devs, list)
853 sectors += bdev_sectors(dc->bdev);
855 c->cached_dev_sectors = sectors;
858 void bch_cached_dev_run(struct cached_dev *dc)
860 struct bcache_device *d = &dc->disk;
861 char buf[SB_LABEL_SIZE + 1];
862 char *env[] = {
863 "DRIVER=bcache",
864 kasprintf(GFP_KERNEL, "CACHED_UUID=%pU", dc->sb.uuid),
865 NULL,
866 NULL,
869 memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
870 buf[SB_LABEL_SIZE] = '\0';
871 env[2] = kasprintf(GFP_KERNEL, "CACHED_LABEL=%s", buf);
873 if (atomic_xchg(&dc->running, 1)) {
874 kfree(env[1]);
875 kfree(env[2]);
876 return;
879 if (!d->c &&
880 BDEV_STATE(&dc->sb) != BDEV_STATE_NONE) {
881 struct closure cl;
882 closure_init_stack(&cl);
884 SET_BDEV_STATE(&dc->sb, BDEV_STATE_STALE);
885 bch_write_bdev_super(dc, &cl);
886 closure_sync(&cl);
889 add_disk(d->disk);
890 bd_link_disk_holder(dc->bdev, dc->disk.disk);
891 /* won't show up in the uevent file, use udevadm monitor -e instead
892 * only class / kset properties are persistent */
893 kobject_uevent_env(&disk_to_dev(d->disk)->kobj, KOBJ_CHANGE, env);
894 kfree(env[1]);
895 kfree(env[2]);
897 if (sysfs_create_link(&d->kobj, &disk_to_dev(d->disk)->kobj, "dev") ||
898 sysfs_create_link(&disk_to_dev(d->disk)->kobj, &d->kobj, "bcache"))
899 pr_debug("error creating sysfs link");
902 static void cached_dev_detach_finish(struct work_struct *w)
904 struct cached_dev *dc = container_of(w, struct cached_dev, detach);
905 char buf[BDEVNAME_SIZE];
906 struct closure cl;
907 closure_init_stack(&cl);
909 BUG_ON(!test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags));
910 BUG_ON(refcount_read(&dc->count));
912 mutex_lock(&bch_register_lock);
914 cancel_delayed_work_sync(&dc->writeback_rate_update);
915 if (!IS_ERR_OR_NULL(dc->writeback_thread)) {
916 kthread_stop(dc->writeback_thread);
917 dc->writeback_thread = NULL;
920 memset(&dc->sb.set_uuid, 0, 16);
921 SET_BDEV_STATE(&dc->sb, BDEV_STATE_NONE);
923 bch_write_bdev_super(dc, &cl);
924 closure_sync(&cl);
926 bcache_device_detach(&dc->disk);
927 list_move(&dc->list, &uncached_devices);
929 clear_bit(BCACHE_DEV_DETACHING, &dc->disk.flags);
930 clear_bit(BCACHE_DEV_UNLINK_DONE, &dc->disk.flags);
932 mutex_unlock(&bch_register_lock);
934 pr_info("Caching disabled for %s", bdevname(dc->bdev, buf));
936 /* Drop ref we took in cached_dev_detach() */
937 closure_put(&dc->disk.cl);
940 void bch_cached_dev_detach(struct cached_dev *dc)
942 lockdep_assert_held(&bch_register_lock);
944 if (test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags))
945 return;
947 if (test_and_set_bit(BCACHE_DEV_DETACHING, &dc->disk.flags))
948 return;
951 * Block the device from being closed and freed until we're finished
952 * detaching
954 closure_get(&dc->disk.cl);
956 bch_writeback_queue(dc);
957 cached_dev_put(dc);
960 int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c,
961 uint8_t *set_uuid)
963 uint32_t rtime = cpu_to_le32(get_seconds());
964 struct uuid_entry *u;
965 char buf[BDEVNAME_SIZE];
966 struct cached_dev *exist_dc, *t;
968 bdevname(dc->bdev, buf);
970 if ((set_uuid && memcmp(set_uuid, c->sb.set_uuid, 16)) ||
971 (!set_uuid && memcmp(dc->sb.set_uuid, c->sb.set_uuid, 16)))
972 return -ENOENT;
974 if (dc->disk.c) {
975 pr_err("Can't attach %s: already attached", buf);
976 return -EINVAL;
979 if (test_bit(CACHE_SET_STOPPING, &c->flags)) {
980 pr_err("Can't attach %s: shutting down", buf);
981 return -EINVAL;
984 if (dc->sb.block_size < c->sb.block_size) {
985 /* Will die */
986 pr_err("Couldn't attach %s: block size less than set's block size",
987 buf);
988 return -EINVAL;
991 /* Check whether already attached */
992 list_for_each_entry_safe(exist_dc, t, &c->cached_devs, list) {
993 if (!memcmp(dc->sb.uuid, exist_dc->sb.uuid, 16)) {
994 pr_err("Tried to attach %s but duplicate UUID already attached",
995 buf);
997 return -EINVAL;
1001 u = uuid_find(c, dc->sb.uuid);
1003 if (u &&
1004 (BDEV_STATE(&dc->sb) == BDEV_STATE_STALE ||
1005 BDEV_STATE(&dc->sb) == BDEV_STATE_NONE)) {
1006 memcpy(u->uuid, invalid_uuid, 16);
1007 u->invalidated = cpu_to_le32(get_seconds());
1008 u = NULL;
1011 if (!u) {
1012 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
1013 pr_err("Couldn't find uuid for %s in set", buf);
1014 return -ENOENT;
1017 u = uuid_find_empty(c);
1018 if (!u) {
1019 pr_err("Not caching %s, no room for UUID", buf);
1020 return -EINVAL;
1024 /* Deadlocks since we're called via sysfs...
1025 sysfs_remove_file(&dc->kobj, &sysfs_attach);
1028 if (bch_is_zero(u->uuid, 16)) {
1029 struct closure cl;
1030 closure_init_stack(&cl);
1032 memcpy(u->uuid, dc->sb.uuid, 16);
1033 memcpy(u->label, dc->sb.label, SB_LABEL_SIZE);
1034 u->first_reg = u->last_reg = rtime;
1035 bch_uuid_write(c);
1037 memcpy(dc->sb.set_uuid, c->sb.set_uuid, 16);
1038 SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN);
1040 bch_write_bdev_super(dc, &cl);
1041 closure_sync(&cl);
1042 } else {
1043 u->last_reg = rtime;
1044 bch_uuid_write(c);
1047 bcache_device_attach(&dc->disk, c, u - c->uuids);
1048 list_move(&dc->list, &c->cached_devs);
1049 calc_cached_dev_sectors(c);
1051 smp_wmb();
1053 * dc->c must be set before dc->count != 0 - paired with the mb in
1054 * cached_dev_get()
1056 refcount_set(&dc->count, 1);
1058 /* Block writeback thread, but spawn it */
1059 down_write(&dc->writeback_lock);
1060 if (bch_cached_dev_writeback_start(dc)) {
1061 up_write(&dc->writeback_lock);
1062 return -ENOMEM;
1065 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
1066 bch_sectors_dirty_init(&dc->disk);
1067 atomic_set(&dc->has_dirty, 1);
1068 refcount_inc(&dc->count);
1069 bch_writeback_queue(dc);
1072 bch_cached_dev_run(dc);
1073 bcache_device_link(&dc->disk, c, "bdev");
1075 /* Allow the writeback thread to proceed */
1076 up_write(&dc->writeback_lock);
1078 pr_info("Caching %s as %s on set %pU",
1079 bdevname(dc->bdev, buf), dc->disk.disk->disk_name,
1080 dc->disk.c->sb.set_uuid);
1081 return 0;
1084 void bch_cached_dev_release(struct kobject *kobj)
1086 struct cached_dev *dc = container_of(kobj, struct cached_dev,
1087 disk.kobj);
1088 kfree(dc);
1089 module_put(THIS_MODULE);
1092 static void cached_dev_free(struct closure *cl)
1094 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1096 cancel_delayed_work_sync(&dc->writeback_rate_update);
1097 if (!IS_ERR_OR_NULL(dc->writeback_thread))
1098 kthread_stop(dc->writeback_thread);
1099 if (dc->writeback_write_wq)
1100 destroy_workqueue(dc->writeback_write_wq);
1102 mutex_lock(&bch_register_lock);
1104 if (atomic_read(&dc->running))
1105 bd_unlink_disk_holder(dc->bdev, dc->disk.disk);
1106 bcache_device_free(&dc->disk);
1107 list_del(&dc->list);
1109 mutex_unlock(&bch_register_lock);
1111 if (!IS_ERR_OR_NULL(dc->bdev))
1112 blkdev_put(dc->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1114 wake_up(&unregister_wait);
1116 kobject_put(&dc->disk.kobj);
1119 static void cached_dev_flush(struct closure *cl)
1121 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1122 struct bcache_device *d = &dc->disk;
1124 mutex_lock(&bch_register_lock);
1125 bcache_device_unlink(d);
1126 mutex_unlock(&bch_register_lock);
1128 bch_cache_accounting_destroy(&dc->accounting);
1129 kobject_del(&d->kobj);
1131 continue_at(cl, cached_dev_free, system_wq);
1134 static int cached_dev_init(struct cached_dev *dc, unsigned block_size)
1136 int ret;
1137 struct io *io;
1138 struct request_queue *q = bdev_get_queue(dc->bdev);
1140 __module_get(THIS_MODULE);
1141 INIT_LIST_HEAD(&dc->list);
1142 closure_init(&dc->disk.cl, NULL);
1143 set_closure_fn(&dc->disk.cl, cached_dev_flush, system_wq);
1144 kobject_init(&dc->disk.kobj, &bch_cached_dev_ktype);
1145 INIT_WORK(&dc->detach, cached_dev_detach_finish);
1146 sema_init(&dc->sb_write_mutex, 1);
1147 INIT_LIST_HEAD(&dc->io_lru);
1148 spin_lock_init(&dc->io_lock);
1149 bch_cache_accounting_init(&dc->accounting, &dc->disk.cl);
1151 dc->sequential_cutoff = 4 << 20;
1153 for (io = dc->io; io < dc->io + RECENT_IO; io++) {
1154 list_add(&io->lru, &dc->io_lru);
1155 hlist_add_head(&io->hash, dc->io_hash + RECENT_IO);
1158 dc->disk.stripe_size = q->limits.io_opt >> 9;
1160 if (dc->disk.stripe_size)
1161 dc->partial_stripes_expensive =
1162 q->limits.raid_partial_stripes_expensive;
1164 ret = bcache_device_init(&dc->disk, block_size,
1165 dc->bdev->bd_part->nr_sects - dc->sb.data_offset);
1166 if (ret)
1167 return ret;
1169 dc->disk.disk->queue->backing_dev_info->ra_pages =
1170 max(dc->disk.disk->queue->backing_dev_info->ra_pages,
1171 q->backing_dev_info->ra_pages);
1173 bch_cached_dev_request_init(dc);
1174 bch_cached_dev_writeback_init(dc);
1175 return 0;
1178 /* Cached device - bcache superblock */
1180 static void register_bdev(struct cache_sb *sb, struct page *sb_page,
1181 struct block_device *bdev,
1182 struct cached_dev *dc)
1184 char name[BDEVNAME_SIZE];
1185 const char *err = "cannot allocate memory";
1186 struct cache_set *c;
1188 memcpy(&dc->sb, sb, sizeof(struct cache_sb));
1189 dc->bdev = bdev;
1190 dc->bdev->bd_holder = dc;
1192 bio_init(&dc->sb_bio, dc->sb_bio.bi_inline_vecs, 1);
1193 bio_first_bvec_all(&dc->sb_bio)->bv_page = sb_page;
1194 get_page(sb_page);
1196 if (cached_dev_init(dc, sb->block_size << 9))
1197 goto err;
1199 err = "error creating kobject";
1200 if (kobject_add(&dc->disk.kobj, &part_to_dev(bdev->bd_part)->kobj,
1201 "bcache"))
1202 goto err;
1203 if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj))
1204 goto err;
1206 pr_info("registered backing device %s", bdevname(bdev, name));
1208 list_add(&dc->list, &uncached_devices);
1209 list_for_each_entry(c, &bch_cache_sets, list)
1210 bch_cached_dev_attach(dc, c, NULL);
1212 if (BDEV_STATE(&dc->sb) == BDEV_STATE_NONE ||
1213 BDEV_STATE(&dc->sb) == BDEV_STATE_STALE)
1214 bch_cached_dev_run(dc);
1216 return;
1217 err:
1218 pr_notice("error %s: %s", bdevname(bdev, name), err);
1219 bcache_device_stop(&dc->disk);
1222 /* Flash only volumes */
1224 void bch_flash_dev_release(struct kobject *kobj)
1226 struct bcache_device *d = container_of(kobj, struct bcache_device,
1227 kobj);
1228 kfree(d);
1231 static void flash_dev_free(struct closure *cl)
1233 struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1234 mutex_lock(&bch_register_lock);
1235 bcache_device_free(d);
1236 mutex_unlock(&bch_register_lock);
1237 kobject_put(&d->kobj);
1240 static void flash_dev_flush(struct closure *cl)
1242 struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1244 mutex_lock(&bch_register_lock);
1245 bcache_device_unlink(d);
1246 mutex_unlock(&bch_register_lock);
1247 kobject_del(&d->kobj);
1248 continue_at(cl, flash_dev_free, system_wq);
1251 static int flash_dev_run(struct cache_set *c, struct uuid_entry *u)
1253 struct bcache_device *d = kzalloc(sizeof(struct bcache_device),
1254 GFP_KERNEL);
1255 if (!d)
1256 return -ENOMEM;
1258 closure_init(&d->cl, NULL);
1259 set_closure_fn(&d->cl, flash_dev_flush, system_wq);
1261 kobject_init(&d->kobj, &bch_flash_dev_ktype);
1263 if (bcache_device_init(d, block_bytes(c), u->sectors))
1264 goto err;
1266 bcache_device_attach(d, c, u - c->uuids);
1267 bch_sectors_dirty_init(d);
1268 bch_flash_dev_request_init(d);
1269 add_disk(d->disk);
1271 if (kobject_add(&d->kobj, &disk_to_dev(d->disk)->kobj, "bcache"))
1272 goto err;
1274 bcache_device_link(d, c, "volume");
1276 return 0;
1277 err:
1278 kobject_put(&d->kobj);
1279 return -ENOMEM;
1282 static int flash_devs_run(struct cache_set *c)
1284 int ret = 0;
1285 struct uuid_entry *u;
1287 for (u = c->uuids;
1288 u < c->uuids + c->nr_uuids && !ret;
1289 u++)
1290 if (UUID_FLASH_ONLY(u))
1291 ret = flash_dev_run(c, u);
1293 return ret;
1296 int bch_flash_dev_create(struct cache_set *c, uint64_t size)
1298 struct uuid_entry *u;
1300 if (test_bit(CACHE_SET_STOPPING, &c->flags))
1301 return -EINTR;
1303 if (!test_bit(CACHE_SET_RUNNING, &c->flags))
1304 return -EPERM;
1306 u = uuid_find_empty(c);
1307 if (!u) {
1308 pr_err("Can't create volume, no room for UUID");
1309 return -EINVAL;
1312 get_random_bytes(u->uuid, 16);
1313 memset(u->label, 0, 32);
1314 u->first_reg = u->last_reg = cpu_to_le32(get_seconds());
1316 SET_UUID_FLASH_ONLY(u, 1);
1317 u->sectors = size >> 9;
1319 bch_uuid_write(c);
1321 return flash_dev_run(c, u);
1324 /* Cache set */
1326 __printf(2, 3)
1327 bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...)
1329 va_list args;
1331 if (c->on_error != ON_ERROR_PANIC &&
1332 test_bit(CACHE_SET_STOPPING, &c->flags))
1333 return false;
1335 /* XXX: we can be called from atomic context
1336 acquire_console_sem();
1339 printk(KERN_ERR "bcache: error on %pU: ", c->sb.set_uuid);
1341 va_start(args, fmt);
1342 vprintk(fmt, args);
1343 va_end(args);
1345 printk(", disabling caching\n");
1347 if (c->on_error == ON_ERROR_PANIC)
1348 panic("panic forced after error\n");
1350 bch_cache_set_unregister(c);
1351 return true;
1354 void bch_cache_set_release(struct kobject *kobj)
1356 struct cache_set *c = container_of(kobj, struct cache_set, kobj);
1357 kfree(c);
1358 module_put(THIS_MODULE);
1361 static void cache_set_free(struct closure *cl)
1363 struct cache_set *c = container_of(cl, struct cache_set, cl);
1364 struct cache *ca;
1365 unsigned i;
1367 if (!IS_ERR_OR_NULL(c->debug))
1368 debugfs_remove(c->debug);
1370 bch_open_buckets_free(c);
1371 bch_btree_cache_free(c);
1372 bch_journal_free(c);
1374 for_each_cache(ca, c, i)
1375 if (ca) {
1376 ca->set = NULL;
1377 c->cache[ca->sb.nr_this_dev] = NULL;
1378 kobject_put(&ca->kobj);
1381 bch_bset_sort_state_free(&c->sort);
1382 free_pages((unsigned long) c->uuids, ilog2(bucket_pages(c)));
1384 if (c->moving_gc_wq)
1385 destroy_workqueue(c->moving_gc_wq);
1386 if (c->bio_split)
1387 bioset_free(c->bio_split);
1388 if (c->fill_iter)
1389 mempool_destroy(c->fill_iter);
1390 if (c->bio_meta)
1391 mempool_destroy(c->bio_meta);
1392 if (c->search)
1393 mempool_destroy(c->search);
1394 kfree(c->devices);
1396 mutex_lock(&bch_register_lock);
1397 list_del(&c->list);
1398 mutex_unlock(&bch_register_lock);
1400 pr_info("Cache set %pU unregistered", c->sb.set_uuid);
1401 wake_up(&unregister_wait);
1403 closure_debug_destroy(&c->cl);
1404 kobject_put(&c->kobj);
1407 static void cache_set_flush(struct closure *cl)
1409 struct cache_set *c = container_of(cl, struct cache_set, caching);
1410 struct cache *ca;
1411 struct btree *b;
1412 unsigned i;
1414 bch_cache_accounting_destroy(&c->accounting);
1416 kobject_put(&c->internal);
1417 kobject_del(&c->kobj);
1419 if (c->gc_thread)
1420 kthread_stop(c->gc_thread);
1422 if (!IS_ERR_OR_NULL(c->root))
1423 list_add(&c->root->list, &c->btree_cache);
1425 /* Should skip this if we're unregistering because of an error */
1426 list_for_each_entry(b, &c->btree_cache, list) {
1427 mutex_lock(&b->write_lock);
1428 if (btree_node_dirty(b))
1429 __bch_btree_node_write(b, NULL);
1430 mutex_unlock(&b->write_lock);
1433 for_each_cache(ca, c, i)
1434 if (ca->alloc_thread)
1435 kthread_stop(ca->alloc_thread);
1437 if (c->journal.cur) {
1438 cancel_delayed_work_sync(&c->journal.work);
1439 /* flush last journal entry if needed */
1440 c->journal.work.work.func(&c->journal.work.work);
1443 closure_return(cl);
1446 static void __cache_set_unregister(struct closure *cl)
1448 struct cache_set *c = container_of(cl, struct cache_set, caching);
1449 struct cached_dev *dc;
1450 size_t i;
1452 mutex_lock(&bch_register_lock);
1454 for (i = 0; i < c->devices_max_used; i++)
1455 if (c->devices[i]) {
1456 if (!UUID_FLASH_ONLY(&c->uuids[i]) &&
1457 test_bit(CACHE_SET_UNREGISTERING, &c->flags)) {
1458 dc = container_of(c->devices[i],
1459 struct cached_dev, disk);
1460 bch_cached_dev_detach(dc);
1461 } else {
1462 bcache_device_stop(c->devices[i]);
1466 mutex_unlock(&bch_register_lock);
1468 continue_at(cl, cache_set_flush, system_wq);
1471 void bch_cache_set_stop(struct cache_set *c)
1473 if (!test_and_set_bit(CACHE_SET_STOPPING, &c->flags))
1474 closure_queue(&c->caching);
1477 void bch_cache_set_unregister(struct cache_set *c)
1479 set_bit(CACHE_SET_UNREGISTERING, &c->flags);
1480 bch_cache_set_stop(c);
1483 #define alloc_bucket_pages(gfp, c) \
1484 ((void *) __get_free_pages(__GFP_ZERO|gfp, ilog2(bucket_pages(c))))
1486 struct cache_set *bch_cache_set_alloc(struct cache_sb *sb)
1488 int iter_size;
1489 struct cache_set *c = kzalloc(sizeof(struct cache_set), GFP_KERNEL);
1490 if (!c)
1491 return NULL;
1493 __module_get(THIS_MODULE);
1494 closure_init(&c->cl, NULL);
1495 set_closure_fn(&c->cl, cache_set_free, system_wq);
1497 closure_init(&c->caching, &c->cl);
1498 set_closure_fn(&c->caching, __cache_set_unregister, system_wq);
1500 /* Maybe create continue_at_noreturn() and use it here? */
1501 closure_set_stopped(&c->cl);
1502 closure_put(&c->cl);
1504 kobject_init(&c->kobj, &bch_cache_set_ktype);
1505 kobject_init(&c->internal, &bch_cache_set_internal_ktype);
1507 bch_cache_accounting_init(&c->accounting, &c->cl);
1509 memcpy(c->sb.set_uuid, sb->set_uuid, 16);
1510 c->sb.block_size = sb->block_size;
1511 c->sb.bucket_size = sb->bucket_size;
1512 c->sb.nr_in_set = sb->nr_in_set;
1513 c->sb.last_mount = sb->last_mount;
1514 c->bucket_bits = ilog2(sb->bucket_size);
1515 c->block_bits = ilog2(sb->block_size);
1516 c->nr_uuids = bucket_bytes(c) / sizeof(struct uuid_entry);
1517 c->devices_max_used = 0;
1518 c->btree_pages = bucket_pages(c);
1519 if (c->btree_pages > BTREE_MAX_PAGES)
1520 c->btree_pages = max_t(int, c->btree_pages / 4,
1521 BTREE_MAX_PAGES);
1523 sema_init(&c->sb_write_mutex, 1);
1524 mutex_init(&c->bucket_lock);
1525 init_waitqueue_head(&c->btree_cache_wait);
1526 init_waitqueue_head(&c->bucket_wait);
1527 init_waitqueue_head(&c->gc_wait);
1528 sema_init(&c->uuid_write_mutex, 1);
1530 spin_lock_init(&c->btree_gc_time.lock);
1531 spin_lock_init(&c->btree_split_time.lock);
1532 spin_lock_init(&c->btree_read_time.lock);
1534 bch_moving_init_cache_set(c);
1536 INIT_LIST_HEAD(&c->list);
1537 INIT_LIST_HEAD(&c->cached_devs);
1538 INIT_LIST_HEAD(&c->btree_cache);
1539 INIT_LIST_HEAD(&c->btree_cache_freeable);
1540 INIT_LIST_HEAD(&c->btree_cache_freed);
1541 INIT_LIST_HEAD(&c->data_buckets);
1543 c->search = mempool_create_slab_pool(32, bch_search_cache);
1544 if (!c->search)
1545 goto err;
1547 iter_size = (sb->bucket_size / sb->block_size + 1) *
1548 sizeof(struct btree_iter_set);
1550 if (!(c->devices = kzalloc(c->nr_uuids * sizeof(void *), GFP_KERNEL)) ||
1551 !(c->bio_meta = mempool_create_kmalloc_pool(2,
1552 sizeof(struct bbio) + sizeof(struct bio_vec) *
1553 bucket_pages(c))) ||
1554 !(c->fill_iter = mempool_create_kmalloc_pool(1, iter_size)) ||
1555 !(c->bio_split = bioset_create(4, offsetof(struct bbio, bio),
1556 BIOSET_NEED_BVECS |
1557 BIOSET_NEED_RESCUER)) ||
1558 !(c->uuids = alloc_bucket_pages(GFP_KERNEL, c)) ||
1559 !(c->moving_gc_wq = alloc_workqueue("bcache_gc",
1560 WQ_MEM_RECLAIM, 0)) ||
1561 bch_journal_alloc(c) ||
1562 bch_btree_cache_alloc(c) ||
1563 bch_open_buckets_alloc(c) ||
1564 bch_bset_sort_state_init(&c->sort, ilog2(c->btree_pages)))
1565 goto err;
1567 c->congested_read_threshold_us = 2000;
1568 c->congested_write_threshold_us = 20000;
1569 c->error_limit = DEFAULT_IO_ERROR_LIMIT;
1571 return c;
1572 err:
1573 bch_cache_set_unregister(c);
1574 return NULL;
1577 static void run_cache_set(struct cache_set *c)
1579 const char *err = "cannot allocate memory";
1580 struct cached_dev *dc, *t;
1581 struct cache *ca;
1582 struct closure cl;
1583 unsigned i;
1585 closure_init_stack(&cl);
1587 for_each_cache(ca, c, i)
1588 c->nbuckets += ca->sb.nbuckets;
1589 set_gc_sectors(c);
1591 if (CACHE_SYNC(&c->sb)) {
1592 LIST_HEAD(journal);
1593 struct bkey *k;
1594 struct jset *j;
1596 err = "cannot allocate memory for journal";
1597 if (bch_journal_read(c, &journal))
1598 goto err;
1600 pr_debug("btree_journal_read() done");
1602 err = "no journal entries found";
1603 if (list_empty(&journal))
1604 goto err;
1606 j = &list_entry(journal.prev, struct journal_replay, list)->j;
1608 err = "IO error reading priorities";
1609 for_each_cache(ca, c, i)
1610 prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev]);
1613 * If prio_read() fails it'll call cache_set_error and we'll
1614 * tear everything down right away, but if we perhaps checked
1615 * sooner we could avoid journal replay.
1618 k = &j->btree_root;
1620 err = "bad btree root";
1621 if (__bch_btree_ptr_invalid(c, k))
1622 goto err;
1624 err = "error reading btree root";
1625 c->root = bch_btree_node_get(c, NULL, k, j->btree_level, true, NULL);
1626 if (IS_ERR_OR_NULL(c->root))
1627 goto err;
1629 list_del_init(&c->root->list);
1630 rw_unlock(true, c->root);
1632 err = uuid_read(c, j, &cl);
1633 if (err)
1634 goto err;
1636 err = "error in recovery";
1637 if (bch_btree_check(c))
1638 goto err;
1640 bch_journal_mark(c, &journal);
1641 bch_initial_gc_finish(c);
1642 pr_debug("btree_check() done");
1645 * bcache_journal_next() can't happen sooner, or
1646 * btree_gc_finish() will give spurious errors about last_gc >
1647 * gc_gen - this is a hack but oh well.
1649 bch_journal_next(&c->journal);
1651 err = "error starting allocator thread";
1652 for_each_cache(ca, c, i)
1653 if (bch_cache_allocator_start(ca))
1654 goto err;
1657 * First place it's safe to allocate: btree_check() and
1658 * btree_gc_finish() have to run before we have buckets to
1659 * allocate, and bch_bucket_alloc_set() might cause a journal
1660 * entry to be written so bcache_journal_next() has to be called
1661 * first.
1663 * If the uuids were in the old format we have to rewrite them
1664 * before the next journal entry is written:
1666 if (j->version < BCACHE_JSET_VERSION_UUID)
1667 __uuid_write(c);
1669 bch_journal_replay(c, &journal);
1670 } else {
1671 pr_notice("invalidating existing data");
1673 for_each_cache(ca, c, i) {
1674 unsigned j;
1676 ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7,
1677 2, SB_JOURNAL_BUCKETS);
1679 for (j = 0; j < ca->sb.keys; j++)
1680 ca->sb.d[j] = ca->sb.first_bucket + j;
1683 bch_initial_gc_finish(c);
1685 err = "error starting allocator thread";
1686 for_each_cache(ca, c, i)
1687 if (bch_cache_allocator_start(ca))
1688 goto err;
1690 mutex_lock(&c->bucket_lock);
1691 for_each_cache(ca, c, i)
1692 bch_prio_write(ca);
1693 mutex_unlock(&c->bucket_lock);
1695 err = "cannot allocate new UUID bucket";
1696 if (__uuid_write(c))
1697 goto err;
1699 err = "cannot allocate new btree root";
1700 c->root = __bch_btree_node_alloc(c, NULL, 0, true, NULL);
1701 if (IS_ERR_OR_NULL(c->root))
1702 goto err;
1704 mutex_lock(&c->root->write_lock);
1705 bkey_copy_key(&c->root->key, &MAX_KEY);
1706 bch_btree_node_write(c->root, &cl);
1707 mutex_unlock(&c->root->write_lock);
1709 bch_btree_set_root(c->root);
1710 rw_unlock(true, c->root);
1713 * We don't want to write the first journal entry until
1714 * everything is set up - fortunately journal entries won't be
1715 * written until the SET_CACHE_SYNC() here:
1717 SET_CACHE_SYNC(&c->sb, true);
1719 bch_journal_next(&c->journal);
1720 bch_journal_meta(c, &cl);
1723 err = "error starting gc thread";
1724 if (bch_gc_thread_start(c))
1725 goto err;
1727 closure_sync(&cl);
1728 c->sb.last_mount = get_seconds();
1729 bcache_write_super(c);
1731 list_for_each_entry_safe(dc, t, &uncached_devices, list)
1732 bch_cached_dev_attach(dc, c, NULL);
1734 flash_devs_run(c);
1736 set_bit(CACHE_SET_RUNNING, &c->flags);
1737 return;
1738 err:
1739 closure_sync(&cl);
1740 /* XXX: test this, it's broken */
1741 bch_cache_set_error(c, "%s", err);
1744 static bool can_attach_cache(struct cache *ca, struct cache_set *c)
1746 return ca->sb.block_size == c->sb.block_size &&
1747 ca->sb.bucket_size == c->sb.bucket_size &&
1748 ca->sb.nr_in_set == c->sb.nr_in_set;
1751 static const char *register_cache_set(struct cache *ca)
1753 char buf[12];
1754 const char *err = "cannot allocate memory";
1755 struct cache_set *c;
1757 list_for_each_entry(c, &bch_cache_sets, list)
1758 if (!memcmp(c->sb.set_uuid, ca->sb.set_uuid, 16)) {
1759 if (c->cache[ca->sb.nr_this_dev])
1760 return "duplicate cache set member";
1762 if (!can_attach_cache(ca, c))
1763 return "cache sb does not match set";
1765 if (!CACHE_SYNC(&ca->sb))
1766 SET_CACHE_SYNC(&c->sb, false);
1768 goto found;
1771 c = bch_cache_set_alloc(&ca->sb);
1772 if (!c)
1773 return err;
1775 err = "error creating kobject";
1776 if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->sb.set_uuid) ||
1777 kobject_add(&c->internal, &c->kobj, "internal"))
1778 goto err;
1780 if (bch_cache_accounting_add_kobjs(&c->accounting, &c->kobj))
1781 goto err;
1783 bch_debug_init_cache_set(c);
1785 list_add(&c->list, &bch_cache_sets);
1786 found:
1787 sprintf(buf, "cache%i", ca->sb.nr_this_dev);
1788 if (sysfs_create_link(&ca->kobj, &c->kobj, "set") ||
1789 sysfs_create_link(&c->kobj, &ca->kobj, buf))
1790 goto err;
1792 if (ca->sb.seq > c->sb.seq) {
1793 c->sb.version = ca->sb.version;
1794 memcpy(c->sb.set_uuid, ca->sb.set_uuid, 16);
1795 c->sb.flags = ca->sb.flags;
1796 c->sb.seq = ca->sb.seq;
1797 pr_debug("set version = %llu", c->sb.version);
1800 kobject_get(&ca->kobj);
1801 ca->set = c;
1802 ca->set->cache[ca->sb.nr_this_dev] = ca;
1803 c->cache_by_alloc[c->caches_loaded++] = ca;
1805 if (c->caches_loaded == c->sb.nr_in_set)
1806 run_cache_set(c);
1808 return NULL;
1809 err:
1810 bch_cache_set_unregister(c);
1811 return err;
1814 /* Cache device */
1816 void bch_cache_release(struct kobject *kobj)
1818 struct cache *ca = container_of(kobj, struct cache, kobj);
1819 unsigned i;
1821 if (ca->set) {
1822 BUG_ON(ca->set->cache[ca->sb.nr_this_dev] != ca);
1823 ca->set->cache[ca->sb.nr_this_dev] = NULL;
1826 free_pages((unsigned long) ca->disk_buckets, ilog2(bucket_pages(ca)));
1827 kfree(ca->prio_buckets);
1828 vfree(ca->buckets);
1830 free_heap(&ca->heap);
1831 free_fifo(&ca->free_inc);
1833 for (i = 0; i < RESERVE_NR; i++)
1834 free_fifo(&ca->free[i]);
1836 if (ca->sb_bio.bi_inline_vecs[0].bv_page)
1837 put_page(bio_first_page_all(&ca->sb_bio));
1839 if (!IS_ERR_OR_NULL(ca->bdev))
1840 blkdev_put(ca->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1842 kfree(ca);
1843 module_put(THIS_MODULE);
1846 static int cache_alloc(struct cache *ca)
1848 size_t free;
1849 size_t btree_buckets;
1850 struct bucket *b;
1852 __module_get(THIS_MODULE);
1853 kobject_init(&ca->kobj, &bch_cache_ktype);
1855 bio_init(&ca->journal.bio, ca->journal.bio.bi_inline_vecs, 8);
1858 * when ca->sb.njournal_buckets is not zero, journal exists,
1859 * and in bch_journal_replay(), tree node may split,
1860 * so bucket of RESERVE_BTREE type is needed,
1861 * the worst situation is all journal buckets are valid journal,
1862 * and all the keys need to replay,
1863 * so the number of RESERVE_BTREE type buckets should be as much
1864 * as journal buckets
1866 btree_buckets = ca->sb.njournal_buckets ?: 8;
1867 free = roundup_pow_of_two(ca->sb.nbuckets) >> 10;
1869 if (!init_fifo(&ca->free[RESERVE_BTREE], btree_buckets, GFP_KERNEL) ||
1870 !init_fifo_exact(&ca->free[RESERVE_PRIO], prio_buckets(ca), GFP_KERNEL) ||
1871 !init_fifo(&ca->free[RESERVE_MOVINGGC], free, GFP_KERNEL) ||
1872 !init_fifo(&ca->free[RESERVE_NONE], free, GFP_KERNEL) ||
1873 !init_fifo(&ca->free_inc, free << 2, GFP_KERNEL) ||
1874 !init_heap(&ca->heap, free << 3, GFP_KERNEL) ||
1875 !(ca->buckets = vzalloc(sizeof(struct bucket) *
1876 ca->sb.nbuckets)) ||
1877 !(ca->prio_buckets = kzalloc(sizeof(uint64_t) * prio_buckets(ca) *
1878 2, GFP_KERNEL)) ||
1879 !(ca->disk_buckets = alloc_bucket_pages(GFP_KERNEL, ca)))
1880 return -ENOMEM;
1882 ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca);
1884 for_each_bucket(b, ca)
1885 atomic_set(&b->pin, 0);
1887 return 0;
1890 static int register_cache(struct cache_sb *sb, struct page *sb_page,
1891 struct block_device *bdev, struct cache *ca)
1893 char name[BDEVNAME_SIZE];
1894 const char *err = NULL; /* must be set for any error case */
1895 int ret = 0;
1897 bdevname(bdev, name);
1899 memcpy(&ca->sb, sb, sizeof(struct cache_sb));
1900 ca->bdev = bdev;
1901 ca->bdev->bd_holder = ca;
1903 bio_init(&ca->sb_bio, ca->sb_bio.bi_inline_vecs, 1);
1904 bio_first_bvec_all(&ca->sb_bio)->bv_page = sb_page;
1905 get_page(sb_page);
1907 if (blk_queue_discard(bdev_get_queue(bdev)))
1908 ca->discard = CACHE_DISCARD(&ca->sb);
1910 ret = cache_alloc(ca);
1911 if (ret != 0) {
1912 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1913 if (ret == -ENOMEM)
1914 err = "cache_alloc(): -ENOMEM";
1915 else
1916 err = "cache_alloc(): unknown error";
1917 goto err;
1920 if (kobject_add(&ca->kobj, &part_to_dev(bdev->bd_part)->kobj, "bcache")) {
1921 err = "error calling kobject_add";
1922 ret = -ENOMEM;
1923 goto out;
1926 mutex_lock(&bch_register_lock);
1927 err = register_cache_set(ca);
1928 mutex_unlock(&bch_register_lock);
1930 if (err) {
1931 ret = -ENODEV;
1932 goto out;
1935 pr_info("registered cache device %s", name);
1937 out:
1938 kobject_put(&ca->kobj);
1940 err:
1941 if (err)
1942 pr_notice("error %s: %s", name, err);
1944 return ret;
1947 /* Global interfaces/init */
1949 static ssize_t register_bcache(struct kobject *, struct kobj_attribute *,
1950 const char *, size_t);
1952 kobj_attribute_write(register, register_bcache);
1953 kobj_attribute_write(register_quiet, register_bcache);
1955 static bool bch_is_open_backing(struct block_device *bdev) {
1956 struct cache_set *c, *tc;
1957 struct cached_dev *dc, *t;
1959 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
1960 list_for_each_entry_safe(dc, t, &c->cached_devs, list)
1961 if (dc->bdev == bdev)
1962 return true;
1963 list_for_each_entry_safe(dc, t, &uncached_devices, list)
1964 if (dc->bdev == bdev)
1965 return true;
1966 return false;
1969 static bool bch_is_open_cache(struct block_device *bdev) {
1970 struct cache_set *c, *tc;
1971 struct cache *ca;
1972 unsigned i;
1974 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
1975 for_each_cache(ca, c, i)
1976 if (ca->bdev == bdev)
1977 return true;
1978 return false;
1981 static bool bch_is_open(struct block_device *bdev) {
1982 return bch_is_open_cache(bdev) || bch_is_open_backing(bdev);
1985 static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr,
1986 const char *buffer, size_t size)
1988 ssize_t ret = size;
1989 const char *err = "cannot allocate memory";
1990 char *path = NULL;
1991 struct cache_sb *sb = NULL;
1992 struct block_device *bdev = NULL;
1993 struct page *sb_page = NULL;
1995 if (!try_module_get(THIS_MODULE))
1996 return -EBUSY;
1998 if (!(path = kstrndup(buffer, size, GFP_KERNEL)) ||
1999 !(sb = kmalloc(sizeof(struct cache_sb), GFP_KERNEL)))
2000 goto err;
2002 err = "failed to open device";
2003 bdev = blkdev_get_by_path(strim(path),
2004 FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2005 sb);
2006 if (IS_ERR(bdev)) {
2007 if (bdev == ERR_PTR(-EBUSY)) {
2008 bdev = lookup_bdev(strim(path));
2009 mutex_lock(&bch_register_lock);
2010 if (!IS_ERR(bdev) && bch_is_open(bdev))
2011 err = "device already registered";
2012 else
2013 err = "device busy";
2014 mutex_unlock(&bch_register_lock);
2015 if (!IS_ERR(bdev))
2016 bdput(bdev);
2017 if (attr == &ksysfs_register_quiet)
2018 goto out;
2020 goto err;
2023 err = "failed to set blocksize";
2024 if (set_blocksize(bdev, 4096))
2025 goto err_close;
2027 err = read_super(sb, bdev, &sb_page);
2028 if (err)
2029 goto err_close;
2031 err = "failed to register device";
2032 if (SB_IS_BDEV(sb)) {
2033 struct cached_dev *dc = kzalloc(sizeof(*dc), GFP_KERNEL);
2034 if (!dc)
2035 goto err_close;
2037 mutex_lock(&bch_register_lock);
2038 register_bdev(sb, sb_page, bdev, dc);
2039 mutex_unlock(&bch_register_lock);
2040 } else {
2041 struct cache *ca = kzalloc(sizeof(*ca), GFP_KERNEL);
2042 if (!ca)
2043 goto err_close;
2045 if (register_cache(sb, sb_page, bdev, ca) != 0)
2046 goto err;
2048 out:
2049 if (sb_page)
2050 put_page(sb_page);
2051 kfree(sb);
2052 kfree(path);
2053 module_put(THIS_MODULE);
2054 return ret;
2056 err_close:
2057 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2058 err:
2059 pr_info("error %s: %s", path, err);
2060 ret = -EINVAL;
2061 goto out;
2064 static int bcache_reboot(struct notifier_block *n, unsigned long code, void *x)
2066 if (code == SYS_DOWN ||
2067 code == SYS_HALT ||
2068 code == SYS_POWER_OFF) {
2069 DEFINE_WAIT(wait);
2070 unsigned long start = jiffies;
2071 bool stopped = false;
2073 struct cache_set *c, *tc;
2074 struct cached_dev *dc, *tdc;
2076 mutex_lock(&bch_register_lock);
2078 if (list_empty(&bch_cache_sets) &&
2079 list_empty(&uncached_devices))
2080 goto out;
2082 pr_info("Stopping all devices:");
2084 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
2085 bch_cache_set_stop(c);
2087 list_for_each_entry_safe(dc, tdc, &uncached_devices, list)
2088 bcache_device_stop(&dc->disk);
2090 /* What's a condition variable? */
2091 while (1) {
2092 long timeout = start + 2 * HZ - jiffies;
2094 stopped = list_empty(&bch_cache_sets) &&
2095 list_empty(&uncached_devices);
2097 if (timeout < 0 || stopped)
2098 break;
2100 prepare_to_wait(&unregister_wait, &wait,
2101 TASK_UNINTERRUPTIBLE);
2103 mutex_unlock(&bch_register_lock);
2104 schedule_timeout(timeout);
2105 mutex_lock(&bch_register_lock);
2108 finish_wait(&unregister_wait, &wait);
2110 if (stopped)
2111 pr_info("All devices stopped");
2112 else
2113 pr_notice("Timeout waiting for devices to be closed");
2114 out:
2115 mutex_unlock(&bch_register_lock);
2118 return NOTIFY_DONE;
2121 static struct notifier_block reboot = {
2122 .notifier_call = bcache_reboot,
2123 .priority = INT_MAX, /* before any real devices */
2126 static void bcache_exit(void)
2128 bch_debug_exit();
2129 bch_request_exit();
2130 if (bcache_kobj)
2131 kobject_put(bcache_kobj);
2132 if (bcache_wq)
2133 destroy_workqueue(bcache_wq);
2134 if (bcache_major)
2135 unregister_blkdev(bcache_major, "bcache");
2136 unregister_reboot_notifier(&reboot);
2137 mutex_destroy(&bch_register_lock);
2140 static int __init bcache_init(void)
2142 static const struct attribute *files[] = {
2143 &ksysfs_register.attr,
2144 &ksysfs_register_quiet.attr,
2145 NULL
2148 mutex_init(&bch_register_lock);
2149 init_waitqueue_head(&unregister_wait);
2150 register_reboot_notifier(&reboot);
2151 closure_debug_init();
2153 bcache_major = register_blkdev(0, "bcache");
2154 if (bcache_major < 0) {
2155 unregister_reboot_notifier(&reboot);
2156 mutex_destroy(&bch_register_lock);
2157 return bcache_major;
2160 if (!(bcache_wq = alloc_workqueue("bcache", WQ_MEM_RECLAIM, 0)) ||
2161 !(bcache_kobj = kobject_create_and_add("bcache", fs_kobj)) ||
2162 bch_request_init() ||
2163 bch_debug_init(bcache_kobj) ||
2164 sysfs_create_files(bcache_kobj, files))
2165 goto err;
2167 return 0;
2168 err:
2169 bcache_exit();
2170 return -ENOMEM;
2173 module_exit(bcache_exit);
2174 module_init(bcache_init);