projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
bcache: Disable broken btree fuzz tester
[deliverable/linux.git] / drivers / md / bcache / super.c
CommitLineData
cafe5635
KO		 1/*
2 * bcache setup/teardown code, and some metadata io - read a superblock and
3 * figure out what to do with it.
4 *
5 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
6 * Copyright 2012 Google, Inc.
7 */
8
9#include "bcache.h"
10#include "btree.h"
11#include "debug.h"
12#include "request.h"
13
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>
21
22MODULE_LICENSE("GPL");
23MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
24
25static const char bcache_magic[] = {
26 0xc6, 0x85, 0x73, 0xf6, 0x4e, 0x1a, 0x45, 0xca,
27 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81
28};
29
30static const char invalid_uuid[] = {
31 0xa0, 0x3e, 0xf8, 0xed, 0x3e, 0xe1, 0xb8, 0x78,
32 0xc8, 0x50, 0xfc, 0x5e, 0xcb, 0x16, 0xcd, 0x99
33};
34
35/* Default is -1; we skip past it for struct cached_dev's cache mode */
36const char * const bch_cache_modes[] = {
37 "default",
38 "writethrough",
39 "writeback",
40 "writearound",
41 "none",
42 NULL
43};
44
45struct uuid_entry_v0 {
46 uint8_t uuid[16];
47 uint8_t label[32];
48 uint32_t first_reg;
49 uint32_t last_reg;
50 uint32_t invalidated;
51 uint32_t pad;
52};
53
54static struct kobject *bcache_kobj;
55struct mutex bch_register_lock;
56LIST_HEAD(bch_cache_sets);
57static LIST_HEAD(uncached_devices);
58
59static int bcache_major, bcache_minor;
60static wait_queue_head_t unregister_wait;
61struct workqueue_struct *bcache_wq;
62
63#define BTREE_MAX_PAGES (256 * 1024 / PAGE_SIZE)
64
65static void bio_split_pool_free(struct bio_split_pool *p)
66{
8ef74790
KO		 67 if (p->bio_split_hook)
68 mempool_destroy(p->bio_split_hook);
69
cafe5635
KO		 70 if (p->bio_split)
71 bioset_free(p->bio_split);
cafe5635
KO		 72}
73
74static int bio_split_pool_init(struct bio_split_pool *p)
75{
76 p->bio_split = bioset_create(4, 0);
77 if (!p->bio_split)
78 return -ENOMEM;
79
80 p->bio_split_hook = mempool_create_kmalloc_pool(4,
81 sizeof(struct bio_split_hook));
82 if (!p->bio_split_hook)
83 return -ENOMEM;
84
85 return 0;
86}
87
88/* Superblock */
89
90static const char *read_super(struct cache_sb *sb, struct block_device *bdev,
91 struct page **res)
92{
93 const char *err;
94 struct cache_sb *s;
95 struct buffer_head *bh = __bread(bdev, 1, SB_SIZE);
96 unsigned i;
97
98 if (!bh)
99 return "IO error";
100
101 s = (struct cache_sb *) bh->b_data;
102
103 sb->offset = le64_to_cpu(s->offset);
104 sb->version = le64_to_cpu(s->version);
105
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);
110
111 sb->flags = le64_to_cpu(s->flags);
112 sb->seq = le64_to_cpu(s->seq);
113
114 sb->nbuckets = le64_to_cpu(s->nbuckets);
115 sb->block_size = le16_to_cpu(s->block_size);
116 sb->bucket_size = le16_to_cpu(s->bucket_size);
117
118 sb->nr_in_set = le16_to_cpu(s->nr_in_set);
119 sb->nr_this_dev = le16_to_cpu(s->nr_this_dev);
120 sb->last_mount = le32_to_cpu(s->last_mount);
121
122 sb->first_bucket = le16_to_cpu(s->first_bucket);
123 sb->keys = le16_to_cpu(s->keys);
124
125 for (i = 0; i < SB_JOURNAL_BUCKETS; i++)
126 sb->d[i] = le64_to_cpu(s->d[i]);
127
128 pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u",
129 sb->version, sb->flags, sb->seq, sb->keys);
130
131 err = "Not a bcache superblock";
132 if (sb->offset != SB_SECTOR)
133 goto err;
134
135 if (memcmp(sb->magic, bcache_magic, 16))
136 goto err;
137
138 err = "Too many journal buckets";
139 if (sb->keys > SB_JOURNAL_BUCKETS)
140 goto err;
141
142 err = "Bad checksum";
143 if (s->csum != csum_set(s))
144 goto err;
145
146 err = "Bad UUID";
169ef1cf 147 if (bch_is_zero(sb->uuid, 16))
cafe5635
KO		 148 goto err;
149
150 err = "Unsupported superblock version";
151 if (sb->version > BCACHE_SB_VERSION)
152 goto err;
153
154 err = "Bad block/bucket size";
155 if (!is_power_of_2(sb->block_size) || sb->block_size > PAGE_SECTORS ||
156 !is_power_of_2(sb->bucket_size) || sb->bucket_size < PAGE_SECTORS)
157 goto err;
158
159 err = "Too many buckets";
160 if (sb->nbuckets > LONG_MAX)
161 goto err;
162
163 err = "Not enough buckets";
164 if (sb->nbuckets < 1 << 7)
165 goto err;
166
167 err = "Invalid superblock: device too small";
168 if (get_capacity(bdev->bd_disk) < sb->bucket_size * sb->nbuckets)
169 goto err;
170
171 if (sb->version == CACHE_BACKING_DEV)
172 goto out;
173
174 err = "Bad UUID";
169ef1cf 175 if (bch_is_zero(sb->set_uuid, 16))
cafe5635
KO		 176 goto err;
177
178 err = "Bad cache device number in set";
179 if (!sb->nr_in_set ||
180 sb->nr_in_set <= sb->nr_this_dev ||
181 sb->nr_in_set > MAX_CACHES_PER_SET)
182 goto err;
183
184 err = "Journal buckets not sequential";
185 for (i = 0; i < sb->keys; i++)
186 if (sb->d[i] != sb->first_bucket + i)
187 goto err;
188
189 err = "Too many journal buckets";
190 if (sb->first_bucket + sb->keys > sb->nbuckets)
191 goto err;
192
193 err = "Invalid superblock: first bucket comes before end of super";
194 if (sb->first_bucket * sb->bucket_size < 16)
195 goto err;
196out:
197 sb->last_mount = get_seconds();
198 err = NULL;
199
200 get_page(bh->b_page);
201 *res = bh->b_page;
202err:
203 put_bh(bh);
204 return err;
205}
206
207static void write_bdev_super_endio(struct bio *bio, int error)
208{
209 struct cached_dev *dc = bio->bi_private;
210 /* XXX: error checking */
211
212 closure_put(&dc->sb_write.cl);
213}
214
215static void __write_super(struct cache_sb *sb, struct bio *bio)
216{
217 struct cache_sb *out = page_address(bio->bi_io_vec[0].bv_page);
218 unsigned i;
219
220 bio->bi_sector = SB_SECTOR;
221 bio->bi_rw = REQ_SYNC|REQ_META;
222 bio->bi_size = SB_SIZE;
169ef1cf 223 bch_bio_map(bio, NULL);
cafe5635
KO		 224
225 out->offset = cpu_to_le64(sb->offset);
226 out->version = cpu_to_le64(sb->version);
227
228 memcpy(out->uuid, sb->uuid, 16);
229 memcpy(out->set_uuid, sb->set_uuid, 16);
230 memcpy(out->label, sb->label, SB_LABEL_SIZE);
231
232 out->flags = cpu_to_le64(sb->flags);
233 out->seq = cpu_to_le64(sb->seq);
234
235 out->last_mount = cpu_to_le32(sb->last_mount);
236 out->first_bucket = cpu_to_le16(sb->first_bucket);
237 out->keys = cpu_to_le16(sb->keys);
238
239 for (i = 0; i < sb->keys; i++)
240 out->d[i] = cpu_to_le64(sb->d[i]);
241
242 out->csum = csum_set(out);
243
244 pr_debug("ver %llu, flags %llu, seq %llu",
245 sb->version, sb->flags, sb->seq);
246
247 submit_bio(REQ_WRITE, bio);
248}
249
250void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent)
251{
252 struct closure *cl = &dc->sb_write.cl;
253 struct bio *bio = &dc->sb_bio;
254
255 closure_lock(&dc->sb_write, parent);
256
257 bio_reset(bio);
258 bio->bi_bdev = dc->bdev;
259 bio->bi_end_io = write_bdev_super_endio;
260 bio->bi_private = dc;
261
262 closure_get(cl);
263 __write_super(&dc->sb, bio);
264
265 closure_return(cl);
266}
267
268static void write_super_endio(struct bio *bio, int error)
269{
270 struct cache *ca = bio->bi_private;
271
272 bch_count_io_errors(ca, error, "writing superblock");
273 closure_put(&ca->set->sb_write.cl);
274}
275
276void bcache_write_super(struct cache_set *c)
277{
278 struct closure *cl = &c->sb_write.cl;
279 struct cache *ca;
280 unsigned i;
281
282 closure_lock(&c->sb_write, &c->cl);
283
284 c->sb.seq++;
285
286 for_each_cache(ca, c, i) {
287 struct bio *bio = &ca->sb_bio;
288
289 ca->sb.version = BCACHE_SB_VERSION;
290 ca->sb.seq = c->sb.seq;
291 ca->sb.last_mount = c->sb.last_mount;
292
293 SET_CACHE_SYNC(&ca->sb, CACHE_SYNC(&c->sb));
294
295 bio_reset(bio);
296 bio->bi_bdev = ca->bdev;
297 bio->bi_end_io = write_super_endio;
298 bio->bi_private = ca;
299
300 closure_get(cl);
301 __write_super(&ca->sb, bio);
302 }
303
304 closure_return(cl);
305}
306
307/* UUID io */
308
309static void uuid_endio(struct bio *bio, int error)
310{
311 struct closure *cl = bio->bi_private;
312 struct cache_set *c = container_of(cl, struct cache_set, uuid_write.cl);
313
314 cache_set_err_on(error, c, "accessing uuids");
315 bch_bbio_free(bio, c);
316 closure_put(cl);
317}
318
319static void uuid_io(struct cache_set *c, unsigned long rw,
320 struct bkey *k, struct closure *parent)
321{
322 struct closure *cl = &c->uuid_write.cl;
323 struct uuid_entry *u;
324 unsigned i;
325
326 BUG_ON(!parent);
327 closure_lock(&c->uuid_write, parent);
328
329 for (i = 0; i < KEY_PTRS(k); i++) {
330 struct bio *bio = bch_bbio_alloc(c);
331
332 bio->bi_rw = REQ_SYNC|REQ_META|rw;
333 bio->bi_size = KEY_SIZE(k) << 9;
334
335 bio->bi_end_io = uuid_endio;
336 bio->bi_private = cl;
169ef1cf 337 bch_bio_map(bio, c->uuids);
cafe5635
KO		 338
339 bch_submit_bbio(bio, c, k, i);
340
341 if (!(rw & WRITE))
342 break;
343 }
344
345 pr_debug("%s UUIDs at %s", rw & REQ_WRITE ? "wrote" : "read",
346 pkey(&c->uuid_bucket));
347
348 for (u = c->uuids; u < c->uuids + c->nr_uuids; u++)
169ef1cf 349 if (!bch_is_zero(u->uuid, 16))
cafe5635
KO		 350 pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u",
351 u - c->uuids, u->uuid, u->label,
352 u->first_reg, u->last_reg, u->invalidated);
353
354 closure_return(cl);
355}
356
357static char *uuid_read(struct cache_set *c, struct jset *j, struct closure *cl)
358{
359 struct bkey *k = &j->uuid_bucket;
360
361 if (__bch_ptr_invalid(c, 1, k))
362 return "bad uuid pointer";
363
364 bkey_copy(&c->uuid_bucket, k);
365 uuid_io(c, READ_SYNC, k, cl);
366
367 if (j->version < BCACHE_JSET_VERSION_UUIDv1) {
368 struct uuid_entry_v0 *u0 = (void *) c->uuids;
369 struct uuid_entry *u1 = (void *) c->uuids;
370 int i;
371
372 closure_sync(cl);
373
374 /*
375 * Since the new uuid entry is bigger than the old, we have to
376 * convert starting at the highest memory address and work down
377 * in order to do it in place
378 */
379
380 for (i = c->nr_uuids - 1;
381 i >= 0;
382 --i) {
383 memcpy(u1[i].uuid, u0[i].uuid, 16);
384 memcpy(u1[i].label, u0[i].label, 32);
385
386 u1[i].first_reg = u0[i].first_reg;
387 u1[i].last_reg = u0[i].last_reg;
388 u1[i].invalidated = u0[i].invalidated;
389
390 u1[i].flags = 0;
391 u1[i].sectors = 0;
392 }
393 }
394
395 return NULL;
396}
397
398static int __uuid_write(struct cache_set *c)
399{
400 BKEY_PADDED(key) k;
401 struct closure cl;
402 closure_init_stack(&cl);
403
404 lockdep_assert_held(&bch_register_lock);
405
406 if (bch_bucket_alloc_set(c, WATERMARK_METADATA, &k.key, 1, &cl))
407 return 1;
408
409 SET_KEY_SIZE(&k.key, c->sb.bucket_size);
410 uuid_io(c, REQ_WRITE, &k.key, &cl);
411 closure_sync(&cl);
412
413 bkey_copy(&c->uuid_bucket, &k.key);
414 __bkey_put(c, &k.key);
415 return 0;
416}
417
418int bch_uuid_write(struct cache_set *c)
419{
420 int ret = __uuid_write(c);
421
422 if (!ret)
423 bch_journal_meta(c, NULL);
424
425 return ret;
426}
427
428static struct uuid_entry *uuid_find(struct cache_set *c, const char *uuid)
429{
430 struct uuid_entry *u;
431
432 for (u = c->uuids;
433 u < c->uuids + c->nr_uuids; u++)
434 if (!memcmp(u->uuid, uuid, 16))
435 return u;
436
437 return NULL;
438}
439
440static struct uuid_entry *uuid_find_empty(struct cache_set *c)
441{
442 static const char zero_uuid[16] = "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0";
443 return uuid_find(c, zero_uuid);
444}
445
446/*
447 * Bucket priorities/gens:
448 *
449 * For each bucket, we store on disk its
450 * 8 bit gen
451 * 16 bit priority
452 *
453 * See alloc.c for an explanation of the gen. The priority is used to implement
454 * lru (and in the future other) cache replacement policies; for most purposes
455 * it's just an opaque integer.
456 *
457 * The gens and the priorities don't have a whole lot to do with each other, and
458 * it's actually the gens that must be written out at specific times - it's no
459 * big deal if the priorities don't get written, if we lose them we just reuse
460 * buckets in suboptimal order.
461 *
462 * On disk they're stored in a packed array, and in as many buckets are required
463 * to fit them all. The buckets we use to store them form a list; the journal
464 * header points to the first bucket, the first bucket points to the second
465 * bucket, et cetera.
466 *
467 * This code is used by the allocation code; periodically (whenever it runs out
468 * of buckets to allocate from) the allocation code will invalidate some
469 * buckets, but it can't use those buckets until their new gens are safely on
470 * disk.
471 */
472
473static void prio_endio(struct bio *bio, int error)
474{
475 struct cache *ca = bio->bi_private;
476
477 cache_set_err_on(error, ca->set, "accessing priorities");
478 bch_bbio_free(bio, ca->set);
479 closure_put(&ca->prio);
480}
481
482static void prio_io(struct cache *ca, uint64_t bucket, unsigned long rw)
483{
484 struct closure *cl = &ca->prio;
485 struct bio *bio = bch_bbio_alloc(ca->set);
486
487 closure_init_stack(cl);
488
489 bio->bi_sector = bucket * ca->sb.bucket_size;
490 bio->bi_bdev = ca->bdev;
491 bio->bi_rw = REQ_SYNC|REQ_META|rw;
492 bio->bi_size = bucket_bytes(ca);
493
494 bio->bi_end_io = prio_endio;
495 bio->bi_private = ca;
169ef1cf 496 bch_bio_map(bio, ca->disk_buckets);
cafe5635
KO		 497
498 closure_bio_submit(bio, &ca->prio, ca);
499 closure_sync(cl);
500}
501
502#define buckets_free(c) "free %zu, free_inc %zu, unused %zu", \
503 fifo_used(&c->free), fifo_used(&c->free_inc), fifo_used(&c->unused)
504
505void bch_prio_write(struct cache *ca)
506{
507 int i;
508 struct bucket *b;
509 struct closure cl;
510
511 closure_init_stack(&cl);
512
513 lockdep_assert_held(&ca->set->bucket_lock);
514
515 for (b = ca->buckets;
516 b < ca->buckets + ca->sb.nbuckets; b++)
517 b->disk_gen = b->gen;
518
519 ca->disk_buckets->seq++;
520
521 atomic_long_add(ca->sb.bucket_size * prio_buckets(ca),
522 &ca->meta_sectors_written);
523
524 pr_debug("free %zu, free_inc %zu, unused %zu", fifo_used(&ca->free),
525 fifo_used(&ca->free_inc), fifo_used(&ca->unused));
526 blktrace_msg(ca, "Starting priorities: " buckets_free(ca));
527
528 for (i = prio_buckets(ca) - 1; i >= 0; --i) {
529 long bucket;
530 struct prio_set *p = ca->disk_buckets;
b1a67b0f
KO		 531 struct bucket_disk *d = p->data;
532 struct bucket_disk *end = d + prios_per_bucket(ca);
cafe5635
KO		 533
534 for (b = ca->buckets + i * prios_per_bucket(ca);
535 b < ca->buckets + ca->sb.nbuckets && d < end;
536 b++, d++) {
537 d->prio = cpu_to_le16(b->prio);
538 d->gen = b->gen;
539 }
540
541 p->next_bucket = ca->prio_buckets[i + 1];
542 p->magic = pset_magic(ca);
169ef1cf 543 p->csum = bch_crc64(&p->magic, bucket_bytes(ca) - 8);
cafe5635
KO		 544
545 bucket = bch_bucket_alloc(ca, WATERMARK_PRIO, &cl);
546 BUG_ON(bucket == -1);
547
548 mutex_unlock(&ca->set->bucket_lock);
549 prio_io(ca, bucket, REQ_WRITE);
550 mutex_lock(&ca->set->bucket_lock);
551
552 ca->prio_buckets[i] = bucket;
553 atomic_dec_bug(&ca->buckets[bucket].pin);
554 }
555
556 mutex_unlock(&ca->set->bucket_lock);
557
558 bch_journal_meta(ca->set, &cl);
559 closure_sync(&cl);
560
561 mutex_lock(&ca->set->bucket_lock);
562
563 ca->need_save_prio = 0;
564
565 /*
566 * Don't want the old priorities to get garbage collected until after we
567 * finish writing the new ones, and they're journalled
568 */
569 for (i = 0; i < prio_buckets(ca); i++)
570 ca->prio_last_buckets[i] = ca->prio_buckets[i];
571}
572
573static void prio_read(struct cache *ca, uint64_t bucket)
574{
575 struct prio_set *p = ca->disk_buckets;
576 struct bucket_disk *d = p->data + prios_per_bucket(ca), *end = d;
577 struct bucket *b;
578 unsigned bucket_nr = 0;
579
580 for (b = ca->buckets;
581 b < ca->buckets + ca->sb.nbuckets;
582 b++, d++) {
583 if (d == end) {
584 ca->prio_buckets[bucket_nr] = bucket;
585 ca->prio_last_buckets[bucket_nr] = bucket;
586 bucket_nr++;
587
588 prio_io(ca, bucket, READ_SYNC);
589
169ef1cf 590 if (p->csum != bch_crc64(&p->magic, bucket_bytes(ca) - 8))
cafe5635
KO		 591 pr_warn("bad csum reading priorities");
592
593 if (p->magic != pset_magic(ca))
594 pr_warn("bad magic reading priorities");
595
596 bucket = p->next_bucket;
597 d = p->data;
598 }
599
600 b->prio = le16_to_cpu(d->prio);
601 b->gen = b->disk_gen = b->last_gc = b->gc_gen = d->gen;
602 }
603}
604
605/* Bcache device */
606
607static int open_dev(struct block_device *b, fmode_t mode)
608{
609 struct bcache_device *d = b->bd_disk->private_data;
610 if (atomic_read(&d->closing))
611 return -ENXIO;
612
613 closure_get(&d->cl);
614 return 0;
615}
616
617static int release_dev(struct gendisk *b, fmode_t mode)
618{
619 struct bcache_device *d = b->private_data;
620 closure_put(&d->cl);
621 return 0;
622}
623
624static int ioctl_dev(struct block_device *b, fmode_t mode,
625 unsigned int cmd, unsigned long arg)
626{
627 struct bcache_device *d = b->bd_disk->private_data;
628 return d->ioctl(d, mode, cmd, arg);
629}
630
631static const struct block_device_operations bcache_ops = {
632 .open = open_dev,
633 .release = release_dev,
634 .ioctl = ioctl_dev,
635 .owner = THIS_MODULE,
636};
637
638void bcache_device_stop(struct bcache_device *d)
639{
640 if (!atomic_xchg(&d->closing, 1))
641 closure_queue(&d->cl);
642}
643
644static void bcache_device_detach(struct bcache_device *d)
645{
646 lockdep_assert_held(&bch_register_lock);
647
648 if (atomic_read(&d->detaching)) {
649 struct uuid_entry *u = d->c->uuids + d->id;
650
651 SET_UUID_FLASH_ONLY(u, 0);
652 memcpy(u->uuid, invalid_uuid, 16);
653 u->invalidated = cpu_to_le32(get_seconds());
654 bch_uuid_write(d->c);
655
656 atomic_set(&d->detaching, 0);
657 }
658
659 d->c->devices[d->id] = NULL;
660 closure_put(&d->c->caching);
661 d->c = NULL;
662}
663
664static void bcache_device_attach(struct bcache_device *d, struct cache_set *c,
665 unsigned id)
666{
667 BUG_ON(test_bit(CACHE_SET_STOPPING, &c->flags));
668
669 d->id = id;
670 d->c = c;
671 c->devices[id] = d;
672
673 closure_get(&c->caching);
674}
675
676static void bcache_device_link(struct bcache_device *d, struct cache_set *c,
677 const char *name)
678{
679 snprintf(d->name, BCACHEDEVNAME_SIZE,
680 "%s%u", name, d->id);
681
682 WARN(sysfs_create_link(&d->kobj, &c->kobj, "cache") ||
683 sysfs_create_link(&c->kobj, &d->kobj, d->name),
684 "Couldn't create device <-> cache set symlinks");
685}
686
687static void bcache_device_free(struct bcache_device *d)
688{
689 lockdep_assert_held(&bch_register_lock);
690
691 pr_info("%s stopped", d->disk->disk_name);
692
693 if (d->c)
694 bcache_device_detach(d);
695
696 if (d->disk)
697 del_gendisk(d->disk);
698 if (d->disk && d->disk->queue)
699 blk_cleanup_queue(d->disk->queue);
700 if (d->disk)
701 put_disk(d->disk);
702
703 bio_split_pool_free(&d->bio_split_hook);
704 if (d->unaligned_bvec)
705 mempool_destroy(d->unaligned_bvec);
706 if (d->bio_split)
707 bioset_free(d->bio_split);
708
709 closure_debug_destroy(&d->cl);
710}
711
712static int bcache_device_init(struct bcache_device *d, unsigned block_size)
713{
714 struct request_queue *q;
715
716 if (!(d->bio_split = bioset_create(4, offsetof(struct bbio, bio))) ||
717 !(d->unaligned_bvec = mempool_create_kmalloc_pool(1,
718 sizeof(struct bio_vec) * BIO_MAX_PAGES)) ||
719 bio_split_pool_init(&d->bio_split_hook))
720
721 return -ENOMEM;
722
723 d->disk = alloc_disk(1);
724 if (!d->disk)
725 return -ENOMEM;
726
727 snprintf(d->disk->disk_name, DISK_NAME_LEN, "bcache%i", bcache_minor);
728
729 d->disk->major = bcache_major;
730 d->disk->first_minor = bcache_minor++;
731 d->disk->fops = &bcache_ops;
732 d->disk->private_data = d;
733
734 q = blk_alloc_queue(GFP_KERNEL);
735 if (!q)
736 return -ENOMEM;
737
738 blk_queue_make_request(q, NULL);
739 d->disk->queue = q;
740 q->queuedata = d;
741 q->backing_dev_info.congested_data = d;
742 q->limits.max_hw_sectors = UINT_MAX;
743 q->limits.max_sectors = UINT_MAX;
744 q->limits.max_segment_size = UINT_MAX;
745 q->limits.max_segments = BIO_MAX_PAGES;
746 q->limits.max_discard_sectors = UINT_MAX;
747 q->limits.io_min = block_size;
748 q->limits.logical_block_size = block_size;
749 q->limits.physical_block_size = block_size;
750 set_bit(QUEUE_FLAG_NONROT, &d->disk->queue->queue_flags);
751 set_bit(QUEUE_FLAG_DISCARD, &d->disk->queue->queue_flags);
752
753 return 0;
754}
755
756/* Cached device */
757
758static void calc_cached_dev_sectors(struct cache_set *c)
759{
760 uint64_t sectors = 0;
761 struct cached_dev *dc;
762
763 list_for_each_entry(dc, &c->cached_devs, list)
764 sectors += bdev_sectors(dc->bdev);
765
766 c->cached_dev_sectors = sectors;
767}
768
769void bch_cached_dev_run(struct cached_dev *dc)
770{
771 struct bcache_device *d = &dc->disk;
772
773 if (atomic_xchg(&dc->running, 1))
774 return;
775
776 if (!d->c &&
777 BDEV_STATE(&dc->sb) != BDEV_STATE_NONE) {
778 struct closure cl;
779 closure_init_stack(&cl);
780
781 SET_BDEV_STATE(&dc->sb, BDEV_STATE_STALE);
782 bch_write_bdev_super(dc, &cl);
783 closure_sync(&cl);
784 }
785
786 add_disk(d->disk);
787#if 0
788 char *env[] = { "SYMLINK=label" , NULL };
789 kobject_uevent_env(&disk_to_dev(d->disk)->kobj, KOBJ_CHANGE, env);
790#endif
791 if (sysfs_create_link(&d->kobj, &disk_to_dev(d->disk)->kobj, "dev") ||
792 sysfs_create_link(&disk_to_dev(d->disk)->kobj, &d->kobj, "bcache"))
793 pr_debug("error creating sysfs link");
794}
795
796static void cached_dev_detach_finish(struct work_struct *w)
797{
798 struct cached_dev *dc = container_of(w, struct cached_dev, detach);
799 char buf[BDEVNAME_SIZE];
800 struct closure cl;
801 closure_init_stack(&cl);
802
803 BUG_ON(!atomic_read(&dc->disk.detaching));
804 BUG_ON(atomic_read(&dc->count));
805
806 sysfs_remove_link(&dc->disk.c->kobj, dc->disk.name);
807 sysfs_remove_link(&dc->disk.kobj, "cache");
808
809 mutex_lock(&bch_register_lock);
810
811 memset(&dc->sb.set_uuid, 0, 16);
812 SET_BDEV_STATE(&dc->sb, BDEV_STATE_NONE);
813
814 bch_write_bdev_super(dc, &cl);
815 closure_sync(&cl);
816
817 bcache_device_detach(&dc->disk);
818 list_move(&dc->list, &uncached_devices);
819
820 mutex_unlock(&bch_register_lock);
821
822 pr_info("Caching disabled for %s", bdevname(dc->bdev, buf));
823
824 /* Drop ref we took in cached_dev_detach() */
825 closure_put(&dc->disk.cl);
826}
827
828void bch_cached_dev_detach(struct cached_dev *dc)
829{
830 lockdep_assert_held(&bch_register_lock);
831
832 if (atomic_read(&dc->disk.closing))
833 return;
834
835 if (atomic_xchg(&dc->disk.detaching, 1))
836 return;
837
838 /*
839 * Block the device from being closed and freed until we're finished
840 * detaching
841 */
842 closure_get(&dc->disk.cl);
843
844 bch_writeback_queue(dc);
845 cached_dev_put(dc);
846}
847
848int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c)
849{
850 uint32_t rtime = cpu_to_le32(get_seconds());
851 struct uuid_entry *u;
852 char buf[BDEVNAME_SIZE];
853
854 bdevname(dc->bdev, buf);
855
856 if (memcmp(dc->sb.set_uuid, c->sb.set_uuid, 16))
857 return -ENOENT;
858
859 if (dc->disk.c) {
860 pr_err("Can't attach %s: already attached", buf);
861 return -EINVAL;
862 }
863
864 if (test_bit(CACHE_SET_STOPPING, &c->flags)) {
865 pr_err("Can't attach %s: shutting down", buf);
866 return -EINVAL;
867 }
868
869 if (dc->sb.block_size < c->sb.block_size) {
870 /* Will die */
b1a67b0f
KO		 871 pr_err("Couldn't attach %s: block size less than set's block size",
872 buf);
cafe5635
KO		 873 return -EINVAL;
874 }
875
876 u = uuid_find(c, dc->sb.uuid);
877
878 if (u &&
879 (BDEV_STATE(&dc->sb) == BDEV_STATE_STALE ||
880 BDEV_STATE(&dc->sb) == BDEV_STATE_NONE)) {
881 memcpy(u->uuid, invalid_uuid, 16);
882 u->invalidated = cpu_to_le32(get_seconds());
883 u = NULL;
884 }
885
886 if (!u) {
887 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
888 pr_err("Couldn't find uuid for %s in set", buf);
889 return -ENOENT;
890 }
891
892 u = uuid_find_empty(c);
893 if (!u) {
894 pr_err("Not caching %s, no room for UUID", buf);
895 return -EINVAL;
896 }
897 }
898
899 /* Deadlocks since we're called via sysfs...
900 sysfs_remove_file(&dc->kobj, &sysfs_attach);
901 */
902
169ef1cf 903 if (bch_is_zero(u->uuid, 16)) {
cafe5635
KO		 904 struct closure cl;
905 closure_init_stack(&cl);
906
907 memcpy(u->uuid, dc->sb.uuid, 16);
908 memcpy(u->label, dc->sb.label, SB_LABEL_SIZE);
909 u->first_reg = u->last_reg = rtime;
910 bch_uuid_write(c);
911
912 memcpy(dc->sb.set_uuid, c->sb.set_uuid, 16);
913 SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN);
914
915 bch_write_bdev_super(dc, &cl);
916 closure_sync(&cl);
917 } else {
918 u->last_reg = rtime;
919 bch_uuid_write(c);
920 }
921
922 bcache_device_attach(&dc->disk, c, u - c->uuids);
923 bcache_device_link(&dc->disk, c, "bdev");
924 list_move(&dc->list, &c->cached_devs);
925 calc_cached_dev_sectors(c);
926
927 smp_wmb();
928 /*
929 * dc->c must be set before dc->count != 0 - paired with the mb in
930 * cached_dev_get()
931 */
932 atomic_set(&dc->count, 1);
933
934 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
935 atomic_set(&dc->has_dirty, 1);
936 atomic_inc(&dc->count);
937 bch_writeback_queue(dc);
938 }
939
940 bch_cached_dev_run(dc);
941
942 pr_info("Caching %s as %s on set %pU",
943 bdevname(dc->bdev, buf), dc->disk.disk->disk_name,
944 dc->disk.c->sb.set_uuid);
945 return 0;
946}
947
948void bch_cached_dev_release(struct kobject *kobj)
949{
950 struct cached_dev *dc = container_of(kobj, struct cached_dev,
951 disk.kobj);
952 kfree(dc);
953 module_put(THIS_MODULE);
954}
955
956static void cached_dev_free(struct closure *cl)
957{
958 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
959
960 cancel_delayed_work_sync(&dc->writeback_rate_update);
961
962 mutex_lock(&bch_register_lock);
963
964 bcache_device_free(&dc->disk);
965 list_del(&dc->list);
966
967 mutex_unlock(&bch_register_lock);
968
969 if (!IS_ERR_OR_NULL(dc->bdev)) {
970 blk_sync_queue(bdev_get_queue(dc->bdev));
971 blkdev_put(dc->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
972 }
973
974 wake_up(&unregister_wait);
975
976 kobject_put(&dc->disk.kobj);
977}
978
979static void cached_dev_flush(struct closure *cl)
980{
981 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
982 struct bcache_device *d = &dc->disk;
983
984 bch_cache_accounting_destroy(&dc->accounting);
985 kobject_del(&d->kobj);
986
987 continue_at(cl, cached_dev_free, system_wq);
988}
989
990static int cached_dev_init(struct cached_dev *dc, unsigned block_size)
991{
992 int err;
993 struct io *io;
994
995 closure_init(&dc->disk.cl, NULL);
996 set_closure_fn(&dc->disk.cl, cached_dev_flush, system_wq);
997
998 __module_get(THIS_MODULE);
999 INIT_LIST_HEAD(&dc->list);
1000 kobject_init(&dc->disk.kobj, &bch_cached_dev_ktype);
1001
1002 bch_cache_accounting_init(&dc->accounting, &dc->disk.cl);
1003
1004 err = bcache_device_init(&dc->disk, block_size);
1005 if (err)
1006 goto err;
1007
1008 spin_lock_init(&dc->io_lock);
1009 closure_init_unlocked(&dc->sb_write);
1010 INIT_WORK(&dc->detach, cached_dev_detach_finish);
1011
1012 dc->sequential_merge = true;
1013 dc->sequential_cutoff = 4 << 20;
1014
1015 INIT_LIST_HEAD(&dc->io_lru);
1016 dc->sb_bio.bi_max_vecs = 1;
1017 dc->sb_bio.bi_io_vec = dc->sb_bio.bi_inline_vecs;
1018
1019 for (io = dc->io; io < dc->io + RECENT_IO; io++) {
1020 list_add(&io->lru, &dc->io_lru);
1021 hlist_add_head(&io->hash, dc->io_hash + RECENT_IO);
1022 }
1023
1024 bch_writeback_init_cached_dev(dc);
1025 return 0;
1026err:
1027 bcache_device_stop(&dc->disk);
1028 return err;
1029}
1030
1031/* Cached device - bcache superblock */
1032
1033static const char *register_bdev(struct cache_sb *sb, struct page *sb_page,
1034 struct block_device *bdev,
1035 struct cached_dev *dc)
1036{
1037 char name[BDEVNAME_SIZE];
1038 const char *err = "cannot allocate memory";
1039 struct gendisk *g;
1040 struct cache_set *c;
1041
1042 if (!dc || cached_dev_init(dc, sb->block_size << 9) != 0)
1043 return err;
1044
1045 memcpy(&dc->sb, sb, sizeof(struct cache_sb));
1046 dc->sb_bio.bi_io_vec[0].bv_page = sb_page;
1047 dc->bdev = bdev;
1048 dc->bdev->bd_holder = dc;
1049
1050 g = dc->disk.disk;
1051
1052 set_capacity(g, dc->bdev->bd_part->nr_sects - 16);
1053
1054 bch_cached_dev_request_init(dc);
1055
1056 err = "error creating kobject";
1057 if (kobject_add(&dc->disk.kobj, &part_to_dev(bdev->bd_part)->kobj,
1058 "bcache"))
1059 goto err;
1060 if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj))
1061 goto err;
1062
1063 list_add(&dc->list, &uncached_devices);
1064 list_for_each_entry(c, &bch_cache_sets, list)
1065 bch_cached_dev_attach(dc, c);
1066
1067 if (BDEV_STATE(&dc->sb) == BDEV_STATE_NONE ||
1068 BDEV_STATE(&dc->sb) == BDEV_STATE_STALE)
1069 bch_cached_dev_run(dc);
1070
1071 return NULL;
1072err:
1073 kobject_put(&dc->disk.kobj);
1074 pr_notice("error opening %s: %s", bdevname(bdev, name), err);
1075 /*
1076 * Return NULL instead of an error because kobject_put() cleans
1077 * everything up
1078 */
1079 return NULL;
1080}
1081
1082/* Flash only volumes */
1083
1084void bch_flash_dev_release(struct kobject *kobj)
1085{
1086 struct bcache_device *d = container_of(kobj, struct bcache_device,
1087 kobj);
1088 kfree(d);
1089}
1090
1091static void flash_dev_free(struct closure *cl)
1092{
1093 struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1094 bcache_device_free(d);
1095 kobject_put(&d->kobj);
1096}
1097
1098static void flash_dev_flush(struct closure *cl)
1099{
1100 struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1101
1102 sysfs_remove_link(&d->c->kobj, d->name);
1103 sysfs_remove_link(&d->kobj, "cache");
1104 kobject_del(&d->kobj);
1105 continue_at(cl, flash_dev_free, system_wq);
1106}
1107
1108static int flash_dev_run(struct cache_set *c, struct uuid_entry *u)
1109{
1110 struct bcache_device *d = kzalloc(sizeof(struct bcache_device),
1111 GFP_KERNEL);
1112 if (!d)
1113 return -ENOMEM;
1114
1115 closure_init(&d->cl, NULL);
1116 set_closure_fn(&d->cl, flash_dev_flush, system_wq);
1117
1118 kobject_init(&d->kobj, &bch_flash_dev_ktype);
1119
1120 if (bcache_device_init(d, block_bytes(c)))
1121 goto err;
1122
1123 bcache_device_attach(d, c, u - c->uuids);
1124 set_capacity(d->disk, u->sectors);
1125 bch_flash_dev_request_init(d);
1126 add_disk(d->disk);
1127
1128 if (kobject_add(&d->kobj, &disk_to_dev(d->disk)->kobj, "bcache"))
1129 goto err;
1130
1131 bcache_device_link(d, c, "volume");
1132
1133 return 0;
1134err:
1135 kobject_put(&d->kobj);
1136 return -ENOMEM;
1137}
1138
1139static int flash_devs_run(struct cache_set *c)
1140{
1141 int ret = 0;
1142 struct uuid_entry *u;
1143
1144 for (u = c->uuids;
1145 u < c->uuids + c->nr_uuids && !ret;
1146 u++)
1147 if (UUID_FLASH_ONLY(u))
1148 ret = flash_dev_run(c, u);
1149
1150 return ret;
1151}
1152
1153int bch_flash_dev_create(struct cache_set *c, uint64_t size)
1154{
1155 struct uuid_entry *u;
1156
1157 if (test_bit(CACHE_SET_STOPPING, &c->flags))
1158 return -EINTR;
1159
1160 u = uuid_find_empty(c);
1161 if (!u) {
1162 pr_err("Can't create volume, no room for UUID");
1163 return -EINVAL;
1164 }
1165
1166 get_random_bytes(u->uuid, 16);
1167 memset(u->label, 0, 32);
1168 u->first_reg = u->last_reg = cpu_to_le32(get_seconds());
1169
1170 SET_UUID_FLASH_ONLY(u, 1);
1171 u->sectors = size >> 9;
1172
1173 bch_uuid_write(c);
1174
1175 return flash_dev_run(c, u);
1176}
1177
1178/* Cache set */
1179
1180__printf(2, 3)
1181bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...)
1182{
1183 va_list args;
1184
1185 if (test_bit(CACHE_SET_STOPPING, &c->flags))
1186 return false;
1187
1188 /* XXX: we can be called from atomic context
1189 acquire_console_sem();
1190 */
1191
1192 printk(KERN_ERR "bcache: error on %pU: ", c->sb.set_uuid);
1193
1194 va_start(args, fmt);
1195 vprintk(fmt, args);
1196 va_end(args);
1197
1198 printk(", disabling caching\n");
1199
1200 bch_cache_set_unregister(c);
1201 return true;
1202}
1203
1204void bch_cache_set_release(struct kobject *kobj)
1205{
1206 struct cache_set *c = container_of(kobj, struct cache_set, kobj);
1207 kfree(c);
1208 module_put(THIS_MODULE);
1209}
1210
1211static void cache_set_free(struct closure *cl)
1212{
1213 struct cache_set *c = container_of(cl, struct cache_set, cl);
1214 struct cache *ca;
1215 unsigned i;
1216
1217 if (!IS_ERR_OR_NULL(c->debug))
1218 debugfs_remove(c->debug);
1219
1220 bch_open_buckets_free(c);
1221 bch_btree_cache_free(c);
1222 bch_journal_free(c);
1223
1224 for_each_cache(ca, c, i)
1225 if (ca)
1226 kobject_put(&ca->kobj);
1227
1228 free_pages((unsigned long) c->uuids, ilog2(bucket_pages(c)));
1229 free_pages((unsigned long) c->sort, ilog2(bucket_pages(c)));
1230
1231 kfree(c->fill_iter);
1232 if (c->bio_split)
1233 bioset_free(c->bio_split);
1234 if (c->bio_meta)
1235 mempool_destroy(c->bio_meta);
1236 if (c->search)
1237 mempool_destroy(c->search);
1238 kfree(c->devices);
1239
1240 mutex_lock(&bch_register_lock);
1241 list_del(&c->list);
1242 mutex_unlock(&bch_register_lock);
1243
1244 pr_info("Cache set %pU unregistered", c->sb.set_uuid);
1245 wake_up(&unregister_wait);
1246
1247 closure_debug_destroy(&c->cl);
1248 kobject_put(&c->kobj);
1249}
1250
1251static void cache_set_flush(struct closure *cl)
1252{
1253 struct cache_set *c = container_of(cl, struct cache_set, caching);
1254 struct btree *b;
1255
1256 /* Shut down allocator threads */
1257 set_bit(CACHE_SET_STOPPING_2, &c->flags);
1258 wake_up(&c->alloc_wait);
1259
1260 bch_cache_accounting_destroy(&c->accounting);
1261
1262 kobject_put(&c->internal);
1263 kobject_del(&c->kobj);
1264
1265 if (!IS_ERR_OR_NULL(c->root))
1266 list_add(&c->root->list, &c->btree_cache);
1267
1268 /* Should skip this if we're unregistering because of an error */
1269 list_for_each_entry(b, &c->btree_cache, list)
1270 if (btree_node_dirty(b))
1271 bch_btree_write(b, true, NULL);
1272
1273 closure_return(cl);
1274}
1275
1276static void __cache_set_unregister(struct closure *cl)
1277{
1278 struct cache_set *c = container_of(cl, struct cache_set, caching);
1279 struct cached_dev *dc, *t;
1280 size_t i;
1281
1282 mutex_lock(&bch_register_lock);
1283
1284 if (test_bit(CACHE_SET_UNREGISTERING, &c->flags))
1285 list_for_each_entry_safe(dc, t, &c->cached_devs, list)
1286 bch_cached_dev_detach(dc);
1287
1288 for (i = 0; i < c->nr_uuids; i++)
1289 if (c->devices[i] && UUID_FLASH_ONLY(&c->uuids[i]))
1290 bcache_device_stop(c->devices[i]);
1291
1292 mutex_unlock(&bch_register_lock);
1293
1294 continue_at(cl, cache_set_flush, system_wq);
1295}
1296
1297void bch_cache_set_stop(struct cache_set *c)
1298{
1299 if (!test_and_set_bit(CACHE_SET_STOPPING, &c->flags))
1300 closure_queue(&c->caching);
1301}
1302
1303void bch_cache_set_unregister(struct cache_set *c)
1304{
1305 set_bit(CACHE_SET_UNREGISTERING, &c->flags);
1306 bch_cache_set_stop(c);
1307}
1308
1309#define alloc_bucket_pages(gfp, c) \
1310 ((void *) __get_free_pages(__GFP_ZERO|gfp, ilog2(bucket_pages(c))))
1311
1312struct cache_set *bch_cache_set_alloc(struct cache_sb *sb)
1313{
1314 int iter_size;
1315 struct cache_set *c = kzalloc(sizeof(struct cache_set), GFP_KERNEL);
1316 if (!c)
1317 return NULL;
1318
1319 __module_get(THIS_MODULE);
1320 closure_init(&c->cl, NULL);
1321 set_closure_fn(&c->cl, cache_set_free, system_wq);
1322
1323 closure_init(&c->caching, &c->cl);
1324 set_closure_fn(&c->caching, __cache_set_unregister, system_wq);
1325
1326 /* Maybe create continue_at_noreturn() and use it here? */
1327 closure_set_stopped(&c->cl);
1328 closure_put(&c->cl);
1329
1330 kobject_init(&c->kobj, &bch_cache_set_ktype);
1331 kobject_init(&c->internal, &bch_cache_set_internal_ktype);
1332
1333 bch_cache_accounting_init(&c->accounting, &c->cl);
1334
1335 memcpy(c->sb.set_uuid, sb->set_uuid, 16);
1336 c->sb.block_size = sb->block_size;
1337 c->sb.bucket_size = sb->bucket_size;
1338 c->sb.nr_in_set = sb->nr_in_set;
1339 c->sb.last_mount = sb->last_mount;
1340 c->bucket_bits = ilog2(sb->bucket_size);
1341 c->block_bits = ilog2(sb->block_size);
1342 c->nr_uuids = bucket_bytes(c) / sizeof(struct uuid_entry);
1343
1344 c->btree_pages = c->sb.bucket_size / PAGE_SECTORS;
1345 if (c->btree_pages > BTREE_MAX_PAGES)
1346 c->btree_pages = max_t(int, c->btree_pages / 4,
1347 BTREE_MAX_PAGES);
1348
1349 init_waitqueue_head(&c->alloc_wait);
1350 mutex_init(&c->bucket_lock);
1351 mutex_init(&c->fill_lock);
1352 mutex_init(&c->sort_lock);
1353 spin_lock_init(&c->sort_time_lock);
1354 closure_init_unlocked(&c->sb_write);
1355 closure_init_unlocked(&c->uuid_write);
1356 spin_lock_init(&c->btree_read_time_lock);
1357 bch_moving_init_cache_set(c);
1358
1359 INIT_LIST_HEAD(&c->list);
1360 INIT_LIST_HEAD(&c->cached_devs);
1361 INIT_LIST_HEAD(&c->btree_cache);
1362 INIT_LIST_HEAD(&c->btree_cache_freeable);
1363 INIT_LIST_HEAD(&c->btree_cache_freed);
1364 INIT_LIST_HEAD(&c->data_buckets);
1365
1366 c->search = mempool_create_slab_pool(32, bch_search_cache);
1367 if (!c->search)
1368 goto err;
1369
1370 iter_size = (sb->bucket_size / sb->block_size + 1) *
1371 sizeof(struct btree_iter_set);
1372
1373 if (!(c->devices = kzalloc(c->nr_uuids * sizeof(void *), GFP_KERNEL)) ||
1374 !(c->bio_meta = mempool_create_kmalloc_pool(2,
1375 sizeof(struct bbio) + sizeof(struct bio_vec) *
1376 bucket_pages(c))) ||
1377 !(c->bio_split = bioset_create(4, offsetof(struct bbio, bio))) ||
1378 !(c->fill_iter = kmalloc(iter_size, GFP_KERNEL)) ||
1379 !(c->sort = alloc_bucket_pages(GFP_KERNEL, c)) ||
1380 !(c->uuids = alloc_bucket_pages(GFP_KERNEL, c)) ||
1381 bch_journal_alloc(c) ||
1382 bch_btree_cache_alloc(c) ||
1383 bch_open_buckets_alloc(c))
1384 goto err;
1385
1386 c->fill_iter->size = sb->bucket_size / sb->block_size;
1387
1388 c->congested_read_threshold_us = 2000;
1389 c->congested_write_threshold_us = 20000;
1390 c->error_limit = 8 << IO_ERROR_SHIFT;
1391
1392 return c;
1393err:
1394 bch_cache_set_unregister(c);
1395 return NULL;
1396}
1397
1398static void run_cache_set(struct cache_set *c)
1399{
1400 const char *err = "cannot allocate memory";
1401 struct cached_dev *dc, *t;
1402 struct cache *ca;
1403 unsigned i;
1404
1405 struct btree_op op;
1406 bch_btree_op_init_stack(&op);
1407 op.lock = SHRT_MAX;
1408
1409 for_each_cache(ca, c, i)
1410 c->nbuckets += ca->sb.nbuckets;
1411
1412 if (CACHE_SYNC(&c->sb)) {
1413 LIST_HEAD(journal);
1414 struct bkey *k;
1415 struct jset *j;
1416
1417 err = "cannot allocate memory for journal";
1418 if (bch_journal_read(c, &journal, &op))
1419 goto err;
1420
1421 pr_debug("btree_journal_read() done");
1422
1423 err = "no journal entries found";
1424 if (list_empty(&journal))
1425 goto err;
1426
1427 j = &list_entry(journal.prev, struct journal_replay, list)->j;
1428
1429 err = "IO error reading priorities";
1430 for_each_cache(ca, c, i)
1431 prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev]);
1432
1433 /*
1434 * If prio_read() fails it'll call cache_set_error and we'll
1435 * tear everything down right away, but if we perhaps checked
1436 * sooner we could avoid journal replay.
1437 */
1438
1439 k = &j->btree_root;
1440
1441 err = "bad btree root";
1442 if (__bch_ptr_invalid(c, j->btree_level + 1, k))
1443 goto err;
1444
1445 err = "error reading btree root";
1446 c->root = bch_btree_node_get(c, k, j->btree_level, &op);
1447 if (IS_ERR_OR_NULL(c->root))
1448 goto err;
1449
1450 list_del_init(&c->root->list);
1451 rw_unlock(true, c->root);
1452
1453 err = uuid_read(c, j, &op.cl);
1454 if (err)
1455 goto err;
1456
1457 err = "error in recovery";
1458 if (bch_btree_check(c, &op))
1459 goto err;
1460
1461 bch_journal_mark(c, &journal);
1462 bch_btree_gc_finish(c);
1463 pr_debug("btree_check() done");
1464
1465 /*
1466 * bcache_journal_next() can't happen sooner, or
1467 * btree_gc_finish() will give spurious errors about last_gc >
1468 * gc_gen - this is a hack but oh well.
1469 */
1470 bch_journal_next(&c->journal);
1471
1472 for_each_cache(ca, c, i)
1473 closure_call(&ca->alloc, bch_allocator_thread,
1474 system_wq, &c->cl);
1475
1476 /*
1477 * First place it's safe to allocate: btree_check() and
1478 * btree_gc_finish() have to run before we have buckets to
1479 * allocate, and bch_bucket_alloc_set() might cause a journal
1480 * entry to be written so bcache_journal_next() has to be called
1481 * first.
1482 *
1483 * If the uuids were in the old format we have to rewrite them
1484 * before the next journal entry is written:
1485 */
1486 if (j->version < BCACHE_JSET_VERSION_UUID)
1487 __uuid_write(c);
1488
1489 bch_journal_replay(c, &journal, &op);
1490 } else {
1491 pr_notice("invalidating existing data");
1492 /* Don't want invalidate_buckets() to queue a gc yet */
1493 closure_lock(&c->gc, NULL);
1494
1495 for_each_cache(ca, c, i) {
1496 unsigned j;
1497
1498 ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7,
1499 2, SB_JOURNAL_BUCKETS);
1500
1501 for (j = 0; j < ca->sb.keys; j++)
1502 ca->sb.d[j] = ca->sb.first_bucket + j;
1503 }
1504
1505 bch_btree_gc_finish(c);
1506
1507 for_each_cache(ca, c, i)
1508 closure_call(&ca->alloc, bch_allocator_thread,
1509 ca->alloc_workqueue, &c->cl);
1510
1511 mutex_lock(&c->bucket_lock);
1512 for_each_cache(ca, c, i)
1513 bch_prio_write(ca);
1514 mutex_unlock(&c->bucket_lock);
1515
1516 wake_up(&c->alloc_wait);
1517
1518 err = "cannot allocate new UUID bucket";
1519 if (__uuid_write(c))
1520 goto err_unlock_gc;
1521
1522 err = "cannot allocate new btree root";
1523 c->root = bch_btree_node_alloc(c, 0, &op.cl);
1524 if (IS_ERR_OR_NULL(c->root))
1525 goto err_unlock_gc;
1526
1527 bkey_copy_key(&c->root->key, &MAX_KEY);
1528 bch_btree_write(c->root, true, &op);
1529
1530 bch_btree_set_root(c->root);
1531 rw_unlock(true, c->root);
1532
1533 /*
1534 * We don't want to write the first journal entry until
1535 * everything is set up - fortunately journal entries won't be
1536 * written until the SET_CACHE_SYNC() here:
1537 */
1538 SET_CACHE_SYNC(&c->sb, true);
1539
1540 bch_journal_next(&c->journal);
1541 bch_journal_meta(c, &op.cl);
1542
1543 /* Unlock */
1544 closure_set_stopped(&c->gc.cl);
1545 closure_put(&c->gc.cl);
1546 }
1547
1548 closure_sync(&op.cl);
1549 c->sb.last_mount = get_seconds();
1550 bcache_write_super(c);
1551
1552 list_for_each_entry_safe(dc, t, &uncached_devices, list)
1553 bch_cached_dev_attach(dc, c);
1554
1555 flash_devs_run(c);
1556
1557 return;
1558err_unlock_gc:
1559 closure_set_stopped(&c->gc.cl);
1560 closure_put(&c->gc.cl);
1561err:
1562 closure_sync(&op.cl);
1563 /* XXX: test this, it's broken */
1564 bch_cache_set_error(c, err);
1565}
1566
1567static bool can_attach_cache(struct cache *ca, struct cache_set *c)
1568{
1569 return ca->sb.block_size == c->sb.block_size &&
1570 ca->sb.bucket_size == c->sb.block_size &&
1571 ca->sb.nr_in_set == c->sb.nr_in_set;
1572}
1573
1574static const char *register_cache_set(struct cache *ca)
1575{
1576 char buf[12];
1577 const char *err = "cannot allocate memory";
1578 struct cache_set *c;
1579
1580 list_for_each_entry(c, &bch_cache_sets, list)
1581 if (!memcmp(c->sb.set_uuid, ca->sb.set_uuid, 16)) {
1582 if (c->cache[ca->sb.nr_this_dev])
1583 return "duplicate cache set member";
1584
1585 if (!can_attach_cache(ca, c))
1586 return "cache sb does not match set";
1587
1588 if (!CACHE_SYNC(&ca->sb))
1589 SET_CACHE_SYNC(&c->sb, false);
1590
1591 goto found;
1592 }
1593
1594 c = bch_cache_set_alloc(&ca->sb);
1595 if (!c)
1596 return err;
1597
1598 err = "error creating kobject";
1599 if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->sb.set_uuid) ||
1600 kobject_add(&c->internal, &c->kobj, "internal"))
1601 goto err;
1602
1603 if (bch_cache_accounting_add_kobjs(&c->accounting, &c->kobj))
1604 goto err;
1605
1606 bch_debug_init_cache_set(c);
1607
1608 list_add(&c->list, &bch_cache_sets);
1609found:
1610 sprintf(buf, "cache%i", ca->sb.nr_this_dev);
1611 if (sysfs_create_link(&ca->kobj, &c->kobj, "set") ||
1612 sysfs_create_link(&c->kobj, &ca->kobj, buf))
1613 goto err;
1614
1615 if (ca->sb.seq > c->sb.seq) {
1616 c->sb.version = ca->sb.version;
1617 memcpy(c->sb.set_uuid, ca->sb.set_uuid, 16);
1618 c->sb.flags = ca->sb.flags;
1619 c->sb.seq = ca->sb.seq;
1620 pr_debug("set version = %llu", c->sb.version);
1621 }
1622
1623 ca->set = c;
1624 ca->set->cache[ca->sb.nr_this_dev] = ca;
1625 c->cache_by_alloc[c->caches_loaded++] = ca;
1626
1627 if (c->caches_loaded == c->sb.nr_in_set)
1628 run_cache_set(c);
1629
1630 return NULL;
1631err:
1632 bch_cache_set_unregister(c);
1633 return err;
1634}
1635
1636/* Cache device */
1637
1638void bch_cache_release(struct kobject *kobj)
1639{
1640 struct cache *ca = container_of(kobj, struct cache, kobj);
1641
1642 if (ca->set)
1643 ca->set->cache[ca->sb.nr_this_dev] = NULL;
1644
1645 bch_cache_allocator_exit(ca);
1646
1647 bio_split_pool_free(&ca->bio_split_hook);
1648
1649 if (ca->alloc_workqueue)
1650 destroy_workqueue(ca->alloc_workqueue);
1651
1652 free_pages((unsigned long) ca->disk_buckets, ilog2(bucket_pages(ca)));
1653 kfree(ca->prio_buckets);
1654 vfree(ca->buckets);
1655
1656 free_heap(&ca->heap);
1657 free_fifo(&ca->unused);
1658 free_fifo(&ca->free_inc);
1659 free_fifo(&ca->free);
1660
1661 if (ca->sb_bio.bi_inline_vecs[0].bv_page)
1662 put_page(ca->sb_bio.bi_io_vec[0].bv_page);
1663
1664 if (!IS_ERR_OR_NULL(ca->bdev)) {
1665 blk_sync_queue(bdev_get_queue(ca->bdev));
1666 blkdev_put(ca->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1667 }
1668
1669 kfree(ca);
1670 module_put(THIS_MODULE);
1671}
1672
1673static int cache_alloc(struct cache_sb *sb, struct cache *ca)
1674{
1675 size_t free;
1676 struct bucket *b;
1677
1678 if (!ca)
1679 return -ENOMEM;
1680
1681 __module_get(THIS_MODULE);
1682 kobject_init(&ca->kobj, &bch_cache_ktype);
1683
1684 memcpy(&ca->sb, sb, sizeof(struct cache_sb));
1685
1686 INIT_LIST_HEAD(&ca->discards);
1687
1688 bio_init(&ca->sb_bio);
1689 ca->sb_bio.bi_max_vecs = 1;
1690 ca->sb_bio.bi_io_vec = ca->sb_bio.bi_inline_vecs;
1691
1692 bio_init(&ca->journal.bio);
1693 ca->journal.bio.bi_max_vecs = 8;
1694 ca->journal.bio.bi_io_vec = ca->journal.bio.bi_inline_vecs;
1695
1696 free = roundup_pow_of_two(ca->sb.nbuckets) >> 9;
1697 free = max_t(size_t, free, (prio_buckets(ca) + 8) * 2);
1698
1699 if (!init_fifo(&ca->free, free, GFP_KERNEL) ||
1700 !init_fifo(&ca->free_inc, free << 2, GFP_KERNEL) ||
1701 !init_fifo(&ca->unused, free << 2, GFP_KERNEL) ||
1702 !init_heap(&ca->heap, free << 3, GFP_KERNEL) ||
1703 !(ca->buckets = vmalloc(sizeof(struct bucket) *
1704 ca->sb.nbuckets)) ||
1705 !(ca->prio_buckets = kzalloc(sizeof(uint64_t) * prio_buckets(ca) *
1706 2, GFP_KERNEL)) ||
1707 !(ca->disk_buckets = alloc_bucket_pages(GFP_KERNEL, ca)) ||
1708 !(ca->alloc_workqueue = alloc_workqueue("bch_allocator", 0, 1)) ||
1709 bio_split_pool_init(&ca->bio_split_hook))
1710 goto err;
1711
1712 ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca);
1713
1714 memset(ca->buckets, 0, ca->sb.nbuckets * sizeof(struct bucket));
1715 for_each_bucket(b, ca)
1716 atomic_set(&b->pin, 0);
1717
1718 if (bch_cache_allocator_init(ca))
1719 goto err;
1720
1721 return 0;
1722err:
1723 kobject_put(&ca->kobj);
1724 return -ENOMEM;
1725}
1726
1727static const char *register_cache(struct cache_sb *sb, struct page *sb_page,
1728 struct block_device *bdev, struct cache *ca)
1729{
1730 char name[BDEVNAME_SIZE];
1731 const char *err = "cannot allocate memory";
1732
1733 if (cache_alloc(sb, ca) != 0)
1734 return err;
1735
1736 ca->sb_bio.bi_io_vec[0].bv_page = sb_page;
1737 ca->bdev = bdev;
1738 ca->bdev->bd_holder = ca;
1739
1740 if (blk_queue_discard(bdev_get_queue(ca->bdev)))
1741 ca->discard = CACHE_DISCARD(&ca->sb);
1742
1743 err = "error creating kobject";
1744 if (kobject_add(&ca->kobj, &part_to_dev(bdev->bd_part)->kobj, "bcache"))
1745 goto err;
1746
1747 err = register_cache_set(ca);
1748 if (err)
1749 goto err;
1750
1751 pr_info("registered cache device %s", bdevname(bdev, name));
1752
1753 return NULL;
1754err:
1755 kobject_put(&ca->kobj);
1756 pr_info("error opening %s: %s", bdevname(bdev, name), err);
1757 /* Return NULL instead of an error because kobject_put() cleans
1758 * everything up
1759 */
1760 return NULL;
1761}
1762
1763/* Global interfaces/init */
1764
1765static ssize_t register_bcache(struct kobject *, struct kobj_attribute *,
1766 const char *, size_t);
1767
1768kobj_attribute_write(register, register_bcache);
1769kobj_attribute_write(register_quiet, register_bcache);
1770
1771static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr,
1772 const char *buffer, size_t size)
1773{
1774 ssize_t ret = size;
1775 const char *err = "cannot allocate memory";
1776 char *path = NULL;
1777 struct cache_sb *sb = NULL;
1778 struct block_device *bdev = NULL;
1779 struct page *sb_page = NULL;
1780
1781 if (!try_module_get(THIS_MODULE))
1782 return -EBUSY;
1783
1784 mutex_lock(&bch_register_lock);
1785
1786 if (!(path = kstrndup(buffer, size, GFP_KERNEL)) ||
1787 !(sb = kmalloc(sizeof(struct cache_sb), GFP_KERNEL)))
1788 goto err;
1789
1790 err = "failed to open device";
1791 bdev = blkdev_get_by_path(strim(path),
1792 FMODE_READ|FMODE_WRITE|FMODE_EXCL,
1793 sb);
1794 if (bdev == ERR_PTR(-EBUSY))
1795 err = "device busy";
1796
1797 if (IS_ERR(bdev) ||
1798 set_blocksize(bdev, 4096))
1799 goto err;
1800
1801 err = read_super(sb, bdev, &sb_page);
1802 if (err)
1803 goto err_close;
1804
1805 if (sb->version == CACHE_BACKING_DEV) {
1806 struct cached_dev *dc = kzalloc(sizeof(*dc), GFP_KERNEL);
1807
1808 err = register_bdev(sb, sb_page, bdev, dc);
1809 } else {
1810 struct cache *ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1811
1812 err = register_cache(sb, sb_page, bdev, ca);
1813 }
1814
1815 if (err) {
1816 /* register_(bdev|cache) will only return an error if they
1817 * didn't get far enough to create the kobject - if they did,
1818 * the kobject destructor will do this cleanup.
1819 */
1820 put_page(sb_page);
1821err_close:
1822 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1823err:
1824 if (attr != &ksysfs_register_quiet)
1825 pr_info("error opening %s: %s", path, err);
1826 ret = -EINVAL;
1827 }
1828
1829 kfree(sb);
1830 kfree(path);
1831 mutex_unlock(&bch_register_lock);
1832 module_put(THIS_MODULE);
1833 return ret;
1834}
1835
1836static int bcache_reboot(struct notifier_block *n, unsigned long code, void *x)
1837{
1838 if (code == SYS_DOWN ||
1839 code == SYS_HALT ||
1840 code == SYS_POWER_OFF) {
1841 DEFINE_WAIT(wait);
1842 unsigned long start = jiffies;
1843 bool stopped = false;
1844
1845 struct cache_set *c, *tc;
1846 struct cached_dev *dc, *tdc;
1847
1848 mutex_lock(&bch_register_lock);
1849
1850 if (list_empty(&bch_cache_sets) &&
1851 list_empty(&uncached_devices))
1852 goto out;
1853
1854 pr_info("Stopping all devices:");
1855
1856 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
1857 bch_cache_set_stop(c);
1858
1859 list_for_each_entry_safe(dc, tdc, &uncached_devices, list)
1860 bcache_device_stop(&dc->disk);
1861
1862 /* What's a condition variable? */
1863 while (1) {
1864 long timeout = start + 2 * HZ - jiffies;
1865
1866 stopped = list_empty(&bch_cache_sets) &&
1867 list_empty(&uncached_devices);
1868
1869 if (timeout < 0 || stopped)
1870 break;
1871
1872 prepare_to_wait(&unregister_wait, &wait,
1873 TASK_UNINTERRUPTIBLE);
1874
1875 mutex_unlock(&bch_register_lock);
1876 schedule_timeout(timeout);
1877 mutex_lock(&bch_register_lock);
1878 }
1879
1880 finish_wait(&unregister_wait, &wait);
1881
1882 if (stopped)
1883 pr_info("All devices stopped");
1884 else
1885 pr_notice("Timeout waiting for devices to be closed");
1886out:
1887 mutex_unlock(&bch_register_lock);
1888 }
1889
1890 return NOTIFY_DONE;
1891}
1892
1893static struct notifier_block reboot = {
1894 .notifier_call = bcache_reboot,
1895 .priority = INT_MAX, /* before any real devices */
1896};
1897
1898static void bcache_exit(void)
1899{
1900 bch_debug_exit();
1901 bch_writeback_exit();
1902 bch_request_exit();
1903 bch_btree_exit();
1904 if (bcache_kobj)
1905 kobject_put(bcache_kobj);
1906 if (bcache_wq)
1907 destroy_workqueue(bcache_wq);
1908 unregister_blkdev(bcache_major, "bcache");
1909 unregister_reboot_notifier(&reboot);
1910}
1911
1912static int __init bcache_init(void)
1913{
1914 static const struct attribute *files[] = {
1915 &ksysfs_register.attr,
1916 &ksysfs_register_quiet.attr,
1917 NULL
1918 };
1919
1920 mutex_init(&bch_register_lock);
1921 init_waitqueue_head(&unregister_wait);
1922 register_reboot_notifier(&reboot);
07e86ccb 1923 closure_debug_init();
cafe5635
KO		 1924
1925 bcache_major = register_blkdev(0, "bcache");
1926 if (bcache_major < 0)
1927 return bcache_major;
1928
1929 if (!(bcache_wq = create_workqueue("bcache")) ||
1930 !(bcache_kobj = kobject_create_and_add("bcache", fs_kobj)) ||
1931 sysfs_create_files(bcache_kobj, files) ||
1932 bch_btree_init() ||
1933 bch_request_init() ||
1934 bch_writeback_init() ||
1935 bch_debug_init(bcache_kobj))
1936 goto err;
1937
1938 return 0;
1939err:
1940 bcache_exit();
1941 return -ENOMEM;
1942}
1943
1944module_exit(bcache_exit);
1945module_init(bcache_init);
Linux kernel - Deliverables
This page took 0.09813 seconds and 5 git commands to generate.