blob: d3169c0652f8cd67e1ae1a478eb687c157ed8ae3 [file] [log] [blame]
Kent Overstreetcafe5632013-03-23 16:11:31 -07001/*
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"
Kent Overstreet279afba2013-06-05 06:21:07 -070013#include "writeback.h"
Kent Overstreetcafe5632013-03-23 16:11:31 -070014
Kent Overstreetc37511b2013-04-26 15:39:55 -070015#include <linux/blkdev.h>
Kent Overstreetcafe5632013-03-23 16:11:31 -070016#include <linux/buffer_head.h>
17#include <linux/debugfs.h>
18#include <linux/genhd.h>
Kent Overstreet79826c32013-07-10 18:31:58 -070019#include <linux/kthread.h>
Kent Overstreetcafe5632013-03-23 16:11:31 -070020#include <linux/module.h>
21#include <linux/random.h>
22#include <linux/reboot.h>
23#include <linux/sysfs.h>
24
25MODULE_LICENSE("GPL");
26MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
27
28static const char bcache_magic[] = {
29 0xc6, 0x85, 0x73, 0xf6, 0x4e, 0x1a, 0x45, 0xca,
30 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81
31};
32
33static const char invalid_uuid[] = {
34 0xa0, 0x3e, 0xf8, 0xed, 0x3e, 0xe1, 0xb8, 0x78,
35 0xc8, 0x50, 0xfc, 0x5e, 0xcb, 0x16, 0xcd, 0x99
36};
37
38/* Default is -1; we skip past it for struct cached_dev's cache mode */
39const char * const bch_cache_modes[] = {
40 "default",
41 "writethrough",
42 "writeback",
43 "writearound",
44 "none",
45 NULL
46};
47
48struct uuid_entry_v0 {
49 uint8_t uuid[16];
50 uint8_t label[32];
51 uint32_t first_reg;
52 uint32_t last_reg;
53 uint32_t invalidated;
54 uint32_t pad;
55};
56
57static struct kobject *bcache_kobj;
58struct mutex bch_register_lock;
59LIST_HEAD(bch_cache_sets);
60static LIST_HEAD(uncached_devices);
61
62static int bcache_major, bcache_minor;
63static wait_queue_head_t unregister_wait;
64struct workqueue_struct *bcache_wq;
65
66#define BTREE_MAX_PAGES (256 * 1024 / PAGE_SIZE)
67
68static void bio_split_pool_free(struct bio_split_pool *p)
69{
Kent Overstreet8ef74792013-04-05 13:46:13 -070070 if (p->bio_split_hook)
71 mempool_destroy(p->bio_split_hook);
72
Kent Overstreetcafe5632013-03-23 16:11:31 -070073 if (p->bio_split)
74 bioset_free(p->bio_split);
Kent Overstreetcafe5632013-03-23 16:11:31 -070075}
76
77static int bio_split_pool_init(struct bio_split_pool *p)
78{
79 p->bio_split = bioset_create(4, 0);
80 if (!p->bio_split)
81 return -ENOMEM;
82
83 p->bio_split_hook = mempool_create_kmalloc_pool(4,
84 sizeof(struct bio_split_hook));
85 if (!p->bio_split_hook)
86 return -ENOMEM;
87
88 return 0;
89}
90
91/* Superblock */
92
93static const char *read_super(struct cache_sb *sb, struct block_device *bdev,
94 struct page **res)
95{
96 const char *err;
97 struct cache_sb *s;
98 struct buffer_head *bh = __bread(bdev, 1, SB_SIZE);
99 unsigned i;
100
101 if (!bh)
102 return "IO error";
103
104 s = (struct cache_sb *) bh->b_data;
105
106 sb->offset = le64_to_cpu(s->offset);
107 sb->version = le64_to_cpu(s->version);
108
109 memcpy(sb->magic, s->magic, 16);
110 memcpy(sb->uuid, s->uuid, 16);
111 memcpy(sb->set_uuid, s->set_uuid, 16);
112 memcpy(sb->label, s->label, SB_LABEL_SIZE);
113
114 sb->flags = le64_to_cpu(s->flags);
115 sb->seq = le64_to_cpu(s->seq);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700116 sb->last_mount = le32_to_cpu(s->last_mount);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700117 sb->first_bucket = le16_to_cpu(s->first_bucket);
118 sb->keys = le16_to_cpu(s->keys);
119
120 for (i = 0; i < SB_JOURNAL_BUCKETS; i++)
121 sb->d[i] = le64_to_cpu(s->d[i]);
122
123 pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u",
124 sb->version, sb->flags, sb->seq, sb->keys);
125
126 err = "Not a bcache superblock";
127 if (sb->offset != SB_SECTOR)
128 goto err;
129
130 if (memcmp(sb->magic, bcache_magic, 16))
131 goto err;
132
133 err = "Too many journal buckets";
134 if (sb->keys > SB_JOURNAL_BUCKETS)
135 goto err;
136
137 err = "Bad checksum";
138 if (s->csum != csum_set(s))
139 goto err;
140
141 err = "Bad UUID";
Kent Overstreet169ef1c2013-03-28 12:50:55 -0600142 if (bch_is_zero(sb->uuid, 16))
Kent Overstreetcafe5632013-03-23 16:11:31 -0700143 goto err;
144
Kent Overstreet8abb2a52013-04-23 21:51:48 -0700145 sb->block_size = le16_to_cpu(s->block_size);
146
147 err = "Superblock block size smaller than device block size";
148 if (sb->block_size << 9 < bdev_logical_block_size(bdev))
149 goto err;
150
Kent Overstreet29033812013-04-11 15:14:35 -0700151 switch (sb->version) {
152 case BCACHE_SB_VERSION_BDEV:
Kent Overstreet29033812013-04-11 15:14:35 -0700153 sb->data_offset = BDEV_DATA_START_DEFAULT;
154 break;
155 case BCACHE_SB_VERSION_BDEV_WITH_OFFSET:
Kent Overstreet29033812013-04-11 15:14:35 -0700156 sb->data_offset = le64_to_cpu(s->data_offset);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700157
Kent Overstreet29033812013-04-11 15:14:35 -0700158 err = "Bad data offset";
159 if (sb->data_offset < BDEV_DATA_START_DEFAULT)
Kent Overstreetcafe5632013-03-23 16:11:31 -0700160 goto err;
161
Kent Overstreet29033812013-04-11 15:14:35 -0700162 break;
163 case BCACHE_SB_VERSION_CDEV:
164 case BCACHE_SB_VERSION_CDEV_WITH_UUID:
165 sb->nbuckets = le64_to_cpu(s->nbuckets);
166 sb->block_size = le16_to_cpu(s->block_size);
167 sb->bucket_size = le16_to_cpu(s->bucket_size);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700168
Kent Overstreet29033812013-04-11 15:14:35 -0700169 sb->nr_in_set = le16_to_cpu(s->nr_in_set);
170 sb->nr_this_dev = le16_to_cpu(s->nr_this_dev);
171
172 err = "Too many buckets";
173 if (sb->nbuckets > LONG_MAX)
174 goto err;
175
176 err = "Not enough buckets";
177 if (sb->nbuckets < 1 << 7)
178 goto err;
179
180 err = "Bad block/bucket size";
181 if (!is_power_of_2(sb->block_size) ||
182 sb->block_size > PAGE_SECTORS ||
183 !is_power_of_2(sb->bucket_size) ||
184 sb->bucket_size < PAGE_SECTORS)
185 goto err;
186
187 err = "Invalid superblock: device too small";
188 if (get_capacity(bdev->bd_disk) < sb->bucket_size * sb->nbuckets)
189 goto err;
190
191 err = "Bad UUID";
192 if (bch_is_zero(sb->set_uuid, 16))
193 goto err;
194
195 err = "Bad cache device number in set";
196 if (!sb->nr_in_set ||
197 sb->nr_in_set <= sb->nr_this_dev ||
198 sb->nr_in_set > MAX_CACHES_PER_SET)
199 goto err;
200
201 err = "Journal buckets not sequential";
202 for (i = 0; i < sb->keys; i++)
203 if (sb->d[i] != sb->first_bucket + i)
204 goto err;
205
206 err = "Too many journal buckets";
207 if (sb->first_bucket + sb->keys > sb->nbuckets)
208 goto err;
209
210 err = "Invalid superblock: first bucket comes before end of super";
211 if (sb->first_bucket * sb->bucket_size < 16)
212 goto err;
213
214 break;
215 default:
216 err = "Unsupported superblock version";
Kent Overstreetcafe5632013-03-23 16:11:31 -0700217 goto err;
Kent Overstreet29033812013-04-11 15:14:35 -0700218 }
219
Kent Overstreetcafe5632013-03-23 16:11:31 -0700220 sb->last_mount = get_seconds();
221 err = NULL;
222
223 get_page(bh->b_page);
224 *res = bh->b_page;
225err:
226 put_bh(bh);
227 return err;
228}
229
230static void write_bdev_super_endio(struct bio *bio, int error)
231{
232 struct cached_dev *dc = bio->bi_private;
233 /* XXX: error checking */
234
235 closure_put(&dc->sb_write.cl);
236}
237
238static void __write_super(struct cache_sb *sb, struct bio *bio)
239{
240 struct cache_sb *out = page_address(bio->bi_io_vec[0].bv_page);
241 unsigned i;
242
243 bio->bi_sector = SB_SECTOR;
244 bio->bi_rw = REQ_SYNC|REQ_META;
245 bio->bi_size = SB_SIZE;
Kent Overstreet169ef1c2013-03-28 12:50:55 -0600246 bch_bio_map(bio, NULL);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700247
248 out->offset = cpu_to_le64(sb->offset);
249 out->version = cpu_to_le64(sb->version);
250
251 memcpy(out->uuid, sb->uuid, 16);
252 memcpy(out->set_uuid, sb->set_uuid, 16);
253 memcpy(out->label, sb->label, SB_LABEL_SIZE);
254
255 out->flags = cpu_to_le64(sb->flags);
256 out->seq = cpu_to_le64(sb->seq);
257
258 out->last_mount = cpu_to_le32(sb->last_mount);
259 out->first_bucket = cpu_to_le16(sb->first_bucket);
260 out->keys = cpu_to_le16(sb->keys);
261
262 for (i = 0; i < sb->keys; i++)
263 out->d[i] = cpu_to_le64(sb->d[i]);
264
265 out->csum = csum_set(out);
266
267 pr_debug("ver %llu, flags %llu, seq %llu",
268 sb->version, sb->flags, sb->seq);
269
270 submit_bio(REQ_WRITE, bio);
271}
272
273void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent)
274{
275 struct closure *cl = &dc->sb_write.cl;
276 struct bio *bio = &dc->sb_bio;
277
278 closure_lock(&dc->sb_write, parent);
279
280 bio_reset(bio);
281 bio->bi_bdev = dc->bdev;
282 bio->bi_end_io = write_bdev_super_endio;
283 bio->bi_private = dc;
284
285 closure_get(cl);
286 __write_super(&dc->sb, bio);
287
288 closure_return(cl);
289}
290
291static void write_super_endio(struct bio *bio, int error)
292{
293 struct cache *ca = bio->bi_private;
294
295 bch_count_io_errors(ca, error, "writing superblock");
296 closure_put(&ca->set->sb_write.cl);
297}
298
299void bcache_write_super(struct cache_set *c)
300{
301 struct closure *cl = &c->sb_write.cl;
302 struct cache *ca;
303 unsigned i;
304
305 closure_lock(&c->sb_write, &c->cl);
306
307 c->sb.seq++;
308
309 for_each_cache(ca, c, i) {
310 struct bio *bio = &ca->sb_bio;
311
Kent Overstreet29033812013-04-11 15:14:35 -0700312 ca->sb.version = BCACHE_SB_VERSION_CDEV_WITH_UUID;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700313 ca->sb.seq = c->sb.seq;
314 ca->sb.last_mount = c->sb.last_mount;
315
316 SET_CACHE_SYNC(&ca->sb, CACHE_SYNC(&c->sb));
317
318 bio_reset(bio);
319 bio->bi_bdev = ca->bdev;
320 bio->bi_end_io = write_super_endio;
321 bio->bi_private = ca;
322
323 closure_get(cl);
324 __write_super(&ca->sb, bio);
325 }
326
327 closure_return(cl);
328}
329
330/* UUID io */
331
332static void uuid_endio(struct bio *bio, int error)
333{
334 struct closure *cl = bio->bi_private;
335 struct cache_set *c = container_of(cl, struct cache_set, uuid_write.cl);
336
337 cache_set_err_on(error, c, "accessing uuids");
338 bch_bbio_free(bio, c);
339 closure_put(cl);
340}
341
342static void uuid_io(struct cache_set *c, unsigned long rw,
343 struct bkey *k, struct closure *parent)
344{
345 struct closure *cl = &c->uuid_write.cl;
346 struct uuid_entry *u;
347 unsigned i;
Kent Overstreet85b14922013-05-14 20:33:16 -0700348 char buf[80];
Kent Overstreetcafe5632013-03-23 16:11:31 -0700349
350 BUG_ON(!parent);
351 closure_lock(&c->uuid_write, parent);
352
353 for (i = 0; i < KEY_PTRS(k); i++) {
354 struct bio *bio = bch_bbio_alloc(c);
355
356 bio->bi_rw = REQ_SYNC|REQ_META|rw;
357 bio->bi_size = KEY_SIZE(k) << 9;
358
359 bio->bi_end_io = uuid_endio;
360 bio->bi_private = cl;
Kent Overstreet169ef1c2013-03-28 12:50:55 -0600361 bch_bio_map(bio, c->uuids);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700362
363 bch_submit_bbio(bio, c, k, i);
364
365 if (!(rw & WRITE))
366 break;
367 }
368
Kent Overstreet85b14922013-05-14 20:33:16 -0700369 bch_bkey_to_text(buf, sizeof(buf), k);
370 pr_debug("%s UUIDs at %s", rw & REQ_WRITE ? "wrote" : "read", buf);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700371
372 for (u = c->uuids; u < c->uuids + c->nr_uuids; u++)
Kent Overstreet169ef1c2013-03-28 12:50:55 -0600373 if (!bch_is_zero(u->uuid, 16))
Kent Overstreetcafe5632013-03-23 16:11:31 -0700374 pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u",
375 u - c->uuids, u->uuid, u->label,
376 u->first_reg, u->last_reg, u->invalidated);
377
378 closure_return(cl);
379}
380
381static char *uuid_read(struct cache_set *c, struct jset *j, struct closure *cl)
382{
383 struct bkey *k = &j->uuid_bucket;
384
385 if (__bch_ptr_invalid(c, 1, k))
386 return "bad uuid pointer";
387
388 bkey_copy(&c->uuid_bucket, k);
389 uuid_io(c, READ_SYNC, k, cl);
390
391 if (j->version < BCACHE_JSET_VERSION_UUIDv1) {
392 struct uuid_entry_v0 *u0 = (void *) c->uuids;
393 struct uuid_entry *u1 = (void *) c->uuids;
394 int i;
395
396 closure_sync(cl);
397
398 /*
399 * Since the new uuid entry is bigger than the old, we have to
400 * convert starting at the highest memory address and work down
401 * in order to do it in place
402 */
403
404 for (i = c->nr_uuids - 1;
405 i >= 0;
406 --i) {
407 memcpy(u1[i].uuid, u0[i].uuid, 16);
408 memcpy(u1[i].label, u0[i].label, 32);
409
410 u1[i].first_reg = u0[i].first_reg;
411 u1[i].last_reg = u0[i].last_reg;
412 u1[i].invalidated = u0[i].invalidated;
413
414 u1[i].flags = 0;
415 u1[i].sectors = 0;
416 }
417 }
418
419 return NULL;
420}
421
422static int __uuid_write(struct cache_set *c)
423{
424 BKEY_PADDED(key) k;
425 struct closure cl;
426 closure_init_stack(&cl);
427
428 lockdep_assert_held(&bch_register_lock);
429
430 if (bch_bucket_alloc_set(c, WATERMARK_METADATA, &k.key, 1, &cl))
431 return 1;
432
433 SET_KEY_SIZE(&k.key, c->sb.bucket_size);
434 uuid_io(c, REQ_WRITE, &k.key, &cl);
435 closure_sync(&cl);
436
437 bkey_copy(&c->uuid_bucket, &k.key);
438 __bkey_put(c, &k.key);
439 return 0;
440}
441
442int bch_uuid_write(struct cache_set *c)
443{
444 int ret = __uuid_write(c);
445
446 if (!ret)
447 bch_journal_meta(c, NULL);
448
449 return ret;
450}
451
452static struct uuid_entry *uuid_find(struct cache_set *c, const char *uuid)
453{
454 struct uuid_entry *u;
455
456 for (u = c->uuids;
457 u < c->uuids + c->nr_uuids; u++)
458 if (!memcmp(u->uuid, uuid, 16))
459 return u;
460
461 return NULL;
462}
463
464static struct uuid_entry *uuid_find_empty(struct cache_set *c)
465{
466 static const char zero_uuid[16] = "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0";
467 return uuid_find(c, zero_uuid);
468}
469
470/*
471 * Bucket priorities/gens:
472 *
473 * For each bucket, we store on disk its
474 * 8 bit gen
475 * 16 bit priority
476 *
477 * See alloc.c for an explanation of the gen. The priority is used to implement
478 * lru (and in the future other) cache replacement policies; for most purposes
479 * it's just an opaque integer.
480 *
481 * The gens and the priorities don't have a whole lot to do with each other, and
482 * it's actually the gens that must be written out at specific times - it's no
483 * big deal if the priorities don't get written, if we lose them we just reuse
484 * buckets in suboptimal order.
485 *
486 * On disk they're stored in a packed array, and in as many buckets are required
487 * to fit them all. The buckets we use to store them form a list; the journal
488 * header points to the first bucket, the first bucket points to the second
489 * bucket, et cetera.
490 *
491 * This code is used by the allocation code; periodically (whenever it runs out
492 * of buckets to allocate from) the allocation code will invalidate some
493 * buckets, but it can't use those buckets until their new gens are safely on
494 * disk.
495 */
496
497static void prio_endio(struct bio *bio, int error)
498{
499 struct cache *ca = bio->bi_private;
500
501 cache_set_err_on(error, ca->set, "accessing priorities");
502 bch_bbio_free(bio, ca->set);
503 closure_put(&ca->prio);
504}
505
506static void prio_io(struct cache *ca, uint64_t bucket, unsigned long rw)
507{
508 struct closure *cl = &ca->prio;
509 struct bio *bio = bch_bbio_alloc(ca->set);
510
511 closure_init_stack(cl);
512
513 bio->bi_sector = bucket * ca->sb.bucket_size;
514 bio->bi_bdev = ca->bdev;
515 bio->bi_rw = REQ_SYNC|REQ_META|rw;
516 bio->bi_size = bucket_bytes(ca);
517
518 bio->bi_end_io = prio_endio;
519 bio->bi_private = ca;
Kent Overstreet169ef1c2013-03-28 12:50:55 -0600520 bch_bio_map(bio, ca->disk_buckets);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700521
522 closure_bio_submit(bio, &ca->prio, ca);
523 closure_sync(cl);
524}
525
526#define buckets_free(c) "free %zu, free_inc %zu, unused %zu", \
527 fifo_used(&c->free), fifo_used(&c->free_inc), fifo_used(&c->unused)
528
529void bch_prio_write(struct cache *ca)
530{
531 int i;
532 struct bucket *b;
533 struct closure cl;
534
535 closure_init_stack(&cl);
536
537 lockdep_assert_held(&ca->set->bucket_lock);
538
539 for (b = ca->buckets;
540 b < ca->buckets + ca->sb.nbuckets; b++)
541 b->disk_gen = b->gen;
542
543 ca->disk_buckets->seq++;
544
545 atomic_long_add(ca->sb.bucket_size * prio_buckets(ca),
546 &ca->meta_sectors_written);
547
548 pr_debug("free %zu, free_inc %zu, unused %zu", fifo_used(&ca->free),
549 fifo_used(&ca->free_inc), fifo_used(&ca->unused));
Kent Overstreetcafe5632013-03-23 16:11:31 -0700550
551 for (i = prio_buckets(ca) - 1; i >= 0; --i) {
552 long bucket;
553 struct prio_set *p = ca->disk_buckets;
Kent Overstreetb1a67b02013-03-25 11:46:44 -0700554 struct bucket_disk *d = p->data;
555 struct bucket_disk *end = d + prios_per_bucket(ca);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700556
557 for (b = ca->buckets + i * prios_per_bucket(ca);
558 b < ca->buckets + ca->sb.nbuckets && d < end;
559 b++, d++) {
560 d->prio = cpu_to_le16(b->prio);
561 d->gen = b->gen;
562 }
563
564 p->next_bucket = ca->prio_buckets[i + 1];
565 p->magic = pset_magic(ca);
Kent Overstreet169ef1c2013-03-28 12:50:55 -0600566 p->csum = bch_crc64(&p->magic, bucket_bytes(ca) - 8);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700567
568 bucket = bch_bucket_alloc(ca, WATERMARK_PRIO, &cl);
569 BUG_ON(bucket == -1);
570
571 mutex_unlock(&ca->set->bucket_lock);
572 prio_io(ca, bucket, REQ_WRITE);
573 mutex_lock(&ca->set->bucket_lock);
574
575 ca->prio_buckets[i] = bucket;
576 atomic_dec_bug(&ca->buckets[bucket].pin);
577 }
578
579 mutex_unlock(&ca->set->bucket_lock);
580
581 bch_journal_meta(ca->set, &cl);
582 closure_sync(&cl);
583
584 mutex_lock(&ca->set->bucket_lock);
585
586 ca->need_save_prio = 0;
587
588 /*
589 * Don't want the old priorities to get garbage collected until after we
590 * finish writing the new ones, and they're journalled
591 */
592 for (i = 0; i < prio_buckets(ca); i++)
593 ca->prio_last_buckets[i] = ca->prio_buckets[i];
594}
595
596static void prio_read(struct cache *ca, uint64_t bucket)
597{
598 struct prio_set *p = ca->disk_buckets;
599 struct bucket_disk *d = p->data + prios_per_bucket(ca), *end = d;
600 struct bucket *b;
601 unsigned bucket_nr = 0;
602
603 for (b = ca->buckets;
604 b < ca->buckets + ca->sb.nbuckets;
605 b++, d++) {
606 if (d == end) {
607 ca->prio_buckets[bucket_nr] = bucket;
608 ca->prio_last_buckets[bucket_nr] = bucket;
609 bucket_nr++;
610
611 prio_io(ca, bucket, READ_SYNC);
612
Kent Overstreet169ef1c2013-03-28 12:50:55 -0600613 if (p->csum != bch_crc64(&p->magic, bucket_bytes(ca) - 8))
Kent Overstreetcafe5632013-03-23 16:11:31 -0700614 pr_warn("bad csum reading priorities");
615
616 if (p->magic != pset_magic(ca))
617 pr_warn("bad magic reading priorities");
618
619 bucket = p->next_bucket;
620 d = p->data;
621 }
622
623 b->prio = le16_to_cpu(d->prio);
624 b->gen = b->disk_gen = b->last_gc = b->gc_gen = d->gen;
625 }
626}
627
628/* Bcache device */
629
630static int open_dev(struct block_device *b, fmode_t mode)
631{
632 struct bcache_device *d = b->bd_disk->private_data;
633 if (atomic_read(&d->closing))
634 return -ENXIO;
635
636 closure_get(&d->cl);
637 return 0;
638}
639
Emil Goode867e1162013-05-09 22:39:26 +0200640static void release_dev(struct gendisk *b, fmode_t mode)
Kent Overstreetcafe5632013-03-23 16:11:31 -0700641{
642 struct bcache_device *d = b->private_data;
643 closure_put(&d->cl);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700644}
645
646static int ioctl_dev(struct block_device *b, fmode_t mode,
647 unsigned int cmd, unsigned long arg)
648{
649 struct bcache_device *d = b->bd_disk->private_data;
650 return d->ioctl(d, mode, cmd, arg);
651}
652
653static const struct block_device_operations bcache_ops = {
654 .open = open_dev,
655 .release = release_dev,
656 .ioctl = ioctl_dev,
657 .owner = THIS_MODULE,
658};
659
660void bcache_device_stop(struct bcache_device *d)
661{
662 if (!atomic_xchg(&d->closing, 1))
663 closure_queue(&d->cl);
664}
665
Kent Overstreetee668502013-02-01 07:29:41 -0800666static void bcache_device_unlink(struct bcache_device *d)
667{
668 unsigned i;
669 struct cache *ca;
670
671 sysfs_remove_link(&d->c->kobj, d->name);
672 sysfs_remove_link(&d->kobj, "cache");
673
674 for_each_cache(ca, d->c, i)
675 bd_unlink_disk_holder(ca->bdev, d->disk);
676}
677
678static void bcache_device_link(struct bcache_device *d, struct cache_set *c,
679 const char *name)
680{
681 unsigned i;
682 struct cache *ca;
683
684 for_each_cache(ca, d->c, i)
685 bd_link_disk_holder(ca->bdev, d->disk);
686
687 snprintf(d->name, BCACHEDEVNAME_SIZE,
688 "%s%u", name, d->id);
689
690 WARN(sysfs_create_link(&d->kobj, &c->kobj, "cache") ||
691 sysfs_create_link(&c->kobj, &d->kobj, d->name),
692 "Couldn't create device <-> cache set symlinks");
693}
694
Kent Overstreetcafe5632013-03-23 16:11:31 -0700695static void bcache_device_detach(struct bcache_device *d)
696{
697 lockdep_assert_held(&bch_register_lock);
698
699 if (atomic_read(&d->detaching)) {
700 struct uuid_entry *u = d->c->uuids + d->id;
701
702 SET_UUID_FLASH_ONLY(u, 0);
703 memcpy(u->uuid, invalid_uuid, 16);
704 u->invalidated = cpu_to_le32(get_seconds());
705 bch_uuid_write(d->c);
706
707 atomic_set(&d->detaching, 0);
708 }
709
Kent Overstreetc9502ea2013-07-10 21:25:02 -0700710 if (!d->flush_done)
711 bcache_device_unlink(d);
Kent Overstreetee668502013-02-01 07:29:41 -0800712
Kent Overstreetcafe5632013-03-23 16:11:31 -0700713 d->c->devices[d->id] = NULL;
714 closure_put(&d->c->caching);
715 d->c = NULL;
716}
717
718static void bcache_device_attach(struct bcache_device *d, struct cache_set *c,
719 unsigned id)
720{
721 BUG_ON(test_bit(CACHE_SET_STOPPING, &c->flags));
722
723 d->id = id;
724 d->c = c;
725 c->devices[id] = d;
726
727 closure_get(&c->caching);
728}
729
Kent Overstreetcafe5632013-03-23 16:11:31 -0700730static void bcache_device_free(struct bcache_device *d)
731{
732 lockdep_assert_held(&bch_register_lock);
733
734 pr_info("%s stopped", d->disk->disk_name);
735
736 if (d->c)
737 bcache_device_detach(d);
Kent Overstreetf59fce82013-05-15 00:11:26 -0700738 if (d->disk && d->disk->flags & GENHD_FL_UP)
Kent Overstreetcafe5632013-03-23 16:11:31 -0700739 del_gendisk(d->disk);
740 if (d->disk && d->disk->queue)
741 blk_cleanup_queue(d->disk->queue);
742 if (d->disk)
743 put_disk(d->disk);
744
745 bio_split_pool_free(&d->bio_split_hook);
746 if (d->unaligned_bvec)
747 mempool_destroy(d->unaligned_bvec);
748 if (d->bio_split)
749 bioset_free(d->bio_split);
Kent Overstreet279afba2013-06-05 06:21:07 -0700750 if (is_vmalloc_addr(d->stripe_sectors_dirty))
751 vfree(d->stripe_sectors_dirty);
752 else
753 kfree(d->stripe_sectors_dirty);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700754
755 closure_debug_destroy(&d->cl);
756}
757
Kent Overstreet279afba2013-06-05 06:21:07 -0700758static int bcache_device_init(struct bcache_device *d, unsigned block_size,
759 sector_t sectors)
Kent Overstreetcafe5632013-03-23 16:11:31 -0700760{
761 struct request_queue *q;
Kent Overstreet279afba2013-06-05 06:21:07 -0700762 size_t n;
763
Kent Overstreet2d679fc2013-08-17 02:13:15 -0700764 if (!d->stripe_size)
765 d->stripe_size = 1 << 31;
Kent Overstreet279afba2013-06-05 06:21:07 -0700766
Kent Overstreet2d679fc2013-08-17 02:13:15 -0700767 d->nr_stripes = DIV_ROUND_UP_ULL(sectors, d->stripe_size);
Kent Overstreet279afba2013-06-05 06:21:07 -0700768
769 if (!d->nr_stripes || d->nr_stripes > SIZE_MAX / sizeof(atomic_t))
770 return -ENOMEM;
771
772 n = d->nr_stripes * sizeof(atomic_t);
773 d->stripe_sectors_dirty = n < PAGE_SIZE << 6
774 ? kzalloc(n, GFP_KERNEL)
775 : vzalloc(n);
776 if (!d->stripe_sectors_dirty)
777 return -ENOMEM;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700778
779 if (!(d->bio_split = bioset_create(4, offsetof(struct bbio, bio))) ||
780 !(d->unaligned_bvec = mempool_create_kmalloc_pool(1,
781 sizeof(struct bio_vec) * BIO_MAX_PAGES)) ||
Kent Overstreetf59fce82013-05-15 00:11:26 -0700782 bio_split_pool_init(&d->bio_split_hook) ||
783 !(d->disk = alloc_disk(1)) ||
784 !(q = blk_alloc_queue(GFP_KERNEL)))
Kent Overstreetcafe5632013-03-23 16:11:31 -0700785 return -ENOMEM;
786
Kent Overstreet279afba2013-06-05 06:21:07 -0700787 set_capacity(d->disk, sectors);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700788 snprintf(d->disk->disk_name, DISK_NAME_LEN, "bcache%i", bcache_minor);
789
790 d->disk->major = bcache_major;
791 d->disk->first_minor = bcache_minor++;
792 d->disk->fops = &bcache_ops;
793 d->disk->private_data = d;
794
Kent Overstreetcafe5632013-03-23 16:11:31 -0700795 blk_queue_make_request(q, NULL);
796 d->disk->queue = q;
797 q->queuedata = d;
798 q->backing_dev_info.congested_data = d;
799 q->limits.max_hw_sectors = UINT_MAX;
800 q->limits.max_sectors = UINT_MAX;
801 q->limits.max_segment_size = UINT_MAX;
802 q->limits.max_segments = BIO_MAX_PAGES;
803 q->limits.max_discard_sectors = UINT_MAX;
804 q->limits.io_min = block_size;
805 q->limits.logical_block_size = block_size;
806 q->limits.physical_block_size = block_size;
807 set_bit(QUEUE_FLAG_NONROT, &d->disk->queue->queue_flags);
808 set_bit(QUEUE_FLAG_DISCARD, &d->disk->queue->queue_flags);
809
Kent Overstreet54d12f22013-07-10 18:44:40 -0700810 blk_queue_flush(q, REQ_FLUSH|REQ_FUA);
811
Kent Overstreetcafe5632013-03-23 16:11:31 -0700812 return 0;
813}
814
815/* Cached device */
816
817static void calc_cached_dev_sectors(struct cache_set *c)
818{
819 uint64_t sectors = 0;
820 struct cached_dev *dc;
821
822 list_for_each_entry(dc, &c->cached_devs, list)
823 sectors += bdev_sectors(dc->bdev);
824
825 c->cached_dev_sectors = sectors;
826}
827
828void bch_cached_dev_run(struct cached_dev *dc)
829{
830 struct bcache_device *d = &dc->disk;
Gabriel de Perthuisab9e1402013-06-09 00:54:48 +0200831 char buf[SB_LABEL_SIZE + 1];
Gabriel de Perthuisa25c32b2013-06-07 23:27:01 +0200832 char *env[] = {
833 "DRIVER=bcache",
834 kasprintf(GFP_KERNEL, "CACHED_UUID=%pU", dc->sb.uuid),
Gabriel de Perthuisab9e1402013-06-09 00:54:48 +0200835 NULL,
836 NULL,
Gabriel de Perthuisa25c32b2013-06-07 23:27:01 +0200837 };
Kent Overstreetcafe5632013-03-23 16:11:31 -0700838
Gabriel de Perthuisab9e1402013-06-09 00:54:48 +0200839 memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
840 buf[SB_LABEL_SIZE] = '\0';
841 env[2] = kasprintf(GFP_KERNEL, "CACHED_LABEL=%s", buf);
842
Kent Overstreetcafe5632013-03-23 16:11:31 -0700843 if (atomic_xchg(&dc->running, 1))
844 return;
845
846 if (!d->c &&
847 BDEV_STATE(&dc->sb) != BDEV_STATE_NONE) {
848 struct closure cl;
849 closure_init_stack(&cl);
850
851 SET_BDEV_STATE(&dc->sb, BDEV_STATE_STALE);
852 bch_write_bdev_super(dc, &cl);
853 closure_sync(&cl);
854 }
855
856 add_disk(d->disk);
Kent Overstreetee668502013-02-01 07:29:41 -0800857 bd_link_disk_holder(dc->bdev, dc->disk.disk);
Gabriel de Perthuisa25c32b2013-06-07 23:27:01 +0200858 /* won't show up in the uevent file, use udevadm monitor -e instead
859 * only class / kset properties are persistent */
Kent Overstreetcafe5632013-03-23 16:11:31 -0700860 kobject_uevent_env(&disk_to_dev(d->disk)->kobj, KOBJ_CHANGE, env);
Gabriel de Perthuisa25c32b2013-06-07 23:27:01 +0200861 kfree(env[1]);
Gabriel de Perthuisab9e1402013-06-09 00:54:48 +0200862 kfree(env[2]);
Gabriel de Perthuisa25c32b2013-06-07 23:27:01 +0200863
Kent Overstreetcafe5632013-03-23 16:11:31 -0700864 if (sysfs_create_link(&d->kobj, &disk_to_dev(d->disk)->kobj, "dev") ||
865 sysfs_create_link(&disk_to_dev(d->disk)->kobj, &d->kobj, "bcache"))
866 pr_debug("error creating sysfs link");
867}
868
869static void cached_dev_detach_finish(struct work_struct *w)
870{
871 struct cached_dev *dc = container_of(w, struct cached_dev, detach);
872 char buf[BDEVNAME_SIZE];
873 struct closure cl;
874 closure_init_stack(&cl);
875
876 BUG_ON(!atomic_read(&dc->disk.detaching));
877 BUG_ON(atomic_read(&dc->count));
878
Kent Overstreetcafe5632013-03-23 16:11:31 -0700879 mutex_lock(&bch_register_lock);
880
881 memset(&dc->sb.set_uuid, 0, 16);
882 SET_BDEV_STATE(&dc->sb, BDEV_STATE_NONE);
883
884 bch_write_bdev_super(dc, &cl);
885 closure_sync(&cl);
886
887 bcache_device_detach(&dc->disk);
888 list_move(&dc->list, &uncached_devices);
889
890 mutex_unlock(&bch_register_lock);
891
892 pr_info("Caching disabled for %s", bdevname(dc->bdev, buf));
893
894 /* Drop ref we took in cached_dev_detach() */
895 closure_put(&dc->disk.cl);
896}
897
898void bch_cached_dev_detach(struct cached_dev *dc)
899{
900 lockdep_assert_held(&bch_register_lock);
901
902 if (atomic_read(&dc->disk.closing))
903 return;
904
905 if (atomic_xchg(&dc->disk.detaching, 1))
906 return;
907
908 /*
909 * Block the device from being closed and freed until we're finished
910 * detaching
911 */
912 closure_get(&dc->disk.cl);
913
914 bch_writeback_queue(dc);
915 cached_dev_put(dc);
916}
917
918int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c)
919{
920 uint32_t rtime = cpu_to_le32(get_seconds());
921 struct uuid_entry *u;
922 char buf[BDEVNAME_SIZE];
923
924 bdevname(dc->bdev, buf);
925
926 if (memcmp(dc->sb.set_uuid, c->sb.set_uuid, 16))
927 return -ENOENT;
928
929 if (dc->disk.c) {
930 pr_err("Can't attach %s: already attached", buf);
931 return -EINVAL;
932 }
933
934 if (test_bit(CACHE_SET_STOPPING, &c->flags)) {
935 pr_err("Can't attach %s: shutting down", buf);
936 return -EINVAL;
937 }
938
939 if (dc->sb.block_size < c->sb.block_size) {
940 /* Will die */
Kent Overstreetb1a67b02013-03-25 11:46:44 -0700941 pr_err("Couldn't attach %s: block size less than set's block size",
942 buf);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700943 return -EINVAL;
944 }
945
946 u = uuid_find(c, dc->sb.uuid);
947
948 if (u &&
949 (BDEV_STATE(&dc->sb) == BDEV_STATE_STALE ||
950 BDEV_STATE(&dc->sb) == BDEV_STATE_NONE)) {
951 memcpy(u->uuid, invalid_uuid, 16);
952 u->invalidated = cpu_to_le32(get_seconds());
953 u = NULL;
954 }
955
956 if (!u) {
957 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
958 pr_err("Couldn't find uuid for %s in set", buf);
959 return -ENOENT;
960 }
961
962 u = uuid_find_empty(c);
963 if (!u) {
964 pr_err("Not caching %s, no room for UUID", buf);
965 return -EINVAL;
966 }
967 }
968
969 /* Deadlocks since we're called via sysfs...
970 sysfs_remove_file(&dc->kobj, &sysfs_attach);
971 */
972
Kent Overstreet169ef1c2013-03-28 12:50:55 -0600973 if (bch_is_zero(u->uuid, 16)) {
Kent Overstreetcafe5632013-03-23 16:11:31 -0700974 struct closure cl;
975 closure_init_stack(&cl);
976
977 memcpy(u->uuid, dc->sb.uuid, 16);
978 memcpy(u->label, dc->sb.label, SB_LABEL_SIZE);
979 u->first_reg = u->last_reg = rtime;
980 bch_uuid_write(c);
981
982 memcpy(dc->sb.set_uuid, c->sb.set_uuid, 16);
983 SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN);
984
985 bch_write_bdev_super(dc, &cl);
986 closure_sync(&cl);
987 } else {
988 u->last_reg = rtime;
989 bch_uuid_write(c);
990 }
991
992 bcache_device_attach(&dc->disk, c, u - c->uuids);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700993 list_move(&dc->list, &c->cached_devs);
994 calc_cached_dev_sectors(c);
995
996 smp_wmb();
997 /*
998 * dc->c must be set before dc->count != 0 - paired with the mb in
999 * cached_dev_get()
1000 */
1001 atomic_set(&dc->count, 1);
1002
1003 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
Kent Overstreet444fc0b2013-05-11 17:07:26 -07001004 bch_sectors_dirty_init(dc);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001005 atomic_set(&dc->has_dirty, 1);
1006 atomic_inc(&dc->count);
1007 bch_writeback_queue(dc);
1008 }
1009
1010 bch_cached_dev_run(dc);
Kent Overstreetee668502013-02-01 07:29:41 -08001011 bcache_device_link(&dc->disk, c, "bdev");
Kent Overstreetcafe5632013-03-23 16:11:31 -07001012
1013 pr_info("Caching %s as %s on set %pU",
1014 bdevname(dc->bdev, buf), dc->disk.disk->disk_name,
1015 dc->disk.c->sb.set_uuid);
1016 return 0;
1017}
1018
1019void bch_cached_dev_release(struct kobject *kobj)
1020{
1021 struct cached_dev *dc = container_of(kobj, struct cached_dev,
1022 disk.kobj);
1023 kfree(dc);
1024 module_put(THIS_MODULE);
1025}
1026
1027static void cached_dev_free(struct closure *cl)
1028{
1029 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1030
1031 cancel_delayed_work_sync(&dc->writeback_rate_update);
1032
1033 mutex_lock(&bch_register_lock);
1034
Kent Overstreetf59fce82013-05-15 00:11:26 -07001035 if (atomic_read(&dc->running))
1036 bd_unlink_disk_holder(dc->bdev, dc->disk.disk);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001037 bcache_device_free(&dc->disk);
1038 list_del(&dc->list);
1039
1040 mutex_unlock(&bch_register_lock);
1041
1042 if (!IS_ERR_OR_NULL(dc->bdev)) {
Kent Overstreetf59fce82013-05-15 00:11:26 -07001043 if (dc->bdev->bd_disk)
1044 blk_sync_queue(bdev_get_queue(dc->bdev));
1045
Kent Overstreetcafe5632013-03-23 16:11:31 -07001046 blkdev_put(dc->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1047 }
1048
1049 wake_up(&unregister_wait);
1050
1051 kobject_put(&dc->disk.kobj);
1052}
1053
1054static void cached_dev_flush(struct closure *cl)
1055{
1056 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1057 struct bcache_device *d = &dc->disk;
1058
Kent Overstreetc9502ea2013-07-10 21:25:02 -07001059 mutex_lock(&bch_register_lock);
1060 d->flush_done = 1;
1061
1062 if (d->c)
1063 bcache_device_unlink(d);
1064
1065 mutex_unlock(&bch_register_lock);
1066
Kent Overstreetcafe5632013-03-23 16:11:31 -07001067 bch_cache_accounting_destroy(&dc->accounting);
1068 kobject_del(&d->kobj);
1069
1070 continue_at(cl, cached_dev_free, system_wq);
1071}
1072
1073static int cached_dev_init(struct cached_dev *dc, unsigned block_size)
1074{
Kent Overstreetf59fce82013-05-15 00:11:26 -07001075 int ret;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001076 struct io *io;
Kent Overstreetf59fce82013-05-15 00:11:26 -07001077 struct request_queue *q = bdev_get_queue(dc->bdev);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001078
1079 __module_get(THIS_MODULE);
1080 INIT_LIST_HEAD(&dc->list);
Kent Overstreetf59fce82013-05-15 00:11:26 -07001081 closure_init(&dc->disk.cl, NULL);
1082 set_closure_fn(&dc->disk.cl, cached_dev_flush, system_wq);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001083 kobject_init(&dc->disk.kobj, &bch_cached_dev_ktype);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001084 INIT_WORK(&dc->detach, cached_dev_detach_finish);
Kent Overstreetf59fce82013-05-15 00:11:26 -07001085 closure_init_unlocked(&dc->sb_write);
1086 INIT_LIST_HEAD(&dc->io_lru);
1087 spin_lock_init(&dc->io_lock);
1088 bch_cache_accounting_init(&dc->accounting, &dc->disk.cl);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001089
1090 dc->sequential_merge = true;
1091 dc->sequential_cutoff = 4 << 20;
1092
Kent Overstreetcafe5632013-03-23 16:11:31 -07001093 for (io = dc->io; io < dc->io + RECENT_IO; io++) {
1094 list_add(&io->lru, &dc->io_lru);
1095 hlist_add_head(&io->hash, dc->io_hash + RECENT_IO);
1096 }
1097
Kent Overstreet279afba2013-06-05 06:21:07 -07001098 ret = bcache_device_init(&dc->disk, block_size,
1099 dc->bdev->bd_part->nr_sects - dc->sb.data_offset);
Kent Overstreetf59fce82013-05-15 00:11:26 -07001100 if (ret)
1101 return ret;
1102
1103 set_capacity(dc->disk.disk,
1104 dc->bdev->bd_part->nr_sects - dc->sb.data_offset);
1105
1106 dc->disk.disk->queue->backing_dev_info.ra_pages =
1107 max(dc->disk.disk->queue->backing_dev_info.ra_pages,
1108 q->backing_dev_info.ra_pages);
1109
1110 bch_cached_dev_request_init(dc);
1111 bch_cached_dev_writeback_init(dc);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001112 return 0;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001113}
1114
1115/* Cached device - bcache superblock */
1116
Kent Overstreetf59fce82013-05-15 00:11:26 -07001117static void register_bdev(struct cache_sb *sb, struct page *sb_page,
Kent Overstreetcafe5632013-03-23 16:11:31 -07001118 struct block_device *bdev,
1119 struct cached_dev *dc)
1120{
1121 char name[BDEVNAME_SIZE];
1122 const char *err = "cannot allocate memory";
Kent Overstreetcafe5632013-03-23 16:11:31 -07001123 struct cache_set *c;
1124
Kent Overstreetcafe5632013-03-23 16:11:31 -07001125 memcpy(&dc->sb, sb, sizeof(struct cache_sb));
Kent Overstreetcafe5632013-03-23 16:11:31 -07001126 dc->bdev = bdev;
1127 dc->bdev->bd_holder = dc;
1128
Kent Overstreetf59fce82013-05-15 00:11:26 -07001129 bio_init(&dc->sb_bio);
1130 dc->sb_bio.bi_max_vecs = 1;
1131 dc->sb_bio.bi_io_vec = dc->sb_bio.bi_inline_vecs;
1132 dc->sb_bio.bi_io_vec[0].bv_page = sb_page;
1133 get_page(sb_page);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001134
Kent Overstreetf59fce82013-05-15 00:11:26 -07001135 if (cached_dev_init(dc, sb->block_size << 9))
1136 goto err;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001137
1138 err = "error creating kobject";
1139 if (kobject_add(&dc->disk.kobj, &part_to_dev(bdev->bd_part)->kobj,
1140 "bcache"))
1141 goto err;
1142 if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj))
1143 goto err;
1144
Kent Overstreetf59fce82013-05-15 00:11:26 -07001145 pr_info("registered backing device %s", bdevname(bdev, name));
1146
Kent Overstreetcafe5632013-03-23 16:11:31 -07001147 list_add(&dc->list, &uncached_devices);
1148 list_for_each_entry(c, &bch_cache_sets, list)
1149 bch_cached_dev_attach(dc, c);
1150
1151 if (BDEV_STATE(&dc->sb) == BDEV_STATE_NONE ||
1152 BDEV_STATE(&dc->sb) == BDEV_STATE_STALE)
1153 bch_cached_dev_run(dc);
1154
Kent Overstreetf59fce82013-05-15 00:11:26 -07001155 return;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001156err:
Kent Overstreetcafe5632013-03-23 16:11:31 -07001157 pr_notice("error opening %s: %s", bdevname(bdev, name), err);
Kent Overstreetf59fce82013-05-15 00:11:26 -07001158 bcache_device_stop(&dc->disk);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001159}
1160
1161/* Flash only volumes */
1162
1163void bch_flash_dev_release(struct kobject *kobj)
1164{
1165 struct bcache_device *d = container_of(kobj, struct bcache_device,
1166 kobj);
1167 kfree(d);
1168}
1169
1170static void flash_dev_free(struct closure *cl)
1171{
1172 struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1173 bcache_device_free(d);
1174 kobject_put(&d->kobj);
1175}
1176
1177static void flash_dev_flush(struct closure *cl)
1178{
1179 struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1180
Kent Overstreetee668502013-02-01 07:29:41 -08001181 bcache_device_unlink(d);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001182 kobject_del(&d->kobj);
1183 continue_at(cl, flash_dev_free, system_wq);
1184}
1185
1186static int flash_dev_run(struct cache_set *c, struct uuid_entry *u)
1187{
1188 struct bcache_device *d = kzalloc(sizeof(struct bcache_device),
1189 GFP_KERNEL);
1190 if (!d)
1191 return -ENOMEM;
1192
1193 closure_init(&d->cl, NULL);
1194 set_closure_fn(&d->cl, flash_dev_flush, system_wq);
1195
1196 kobject_init(&d->kobj, &bch_flash_dev_ktype);
1197
Kent Overstreet279afba2013-06-05 06:21:07 -07001198 if (bcache_device_init(d, block_bytes(c), u->sectors))
Kent Overstreetcafe5632013-03-23 16:11:31 -07001199 goto err;
1200
1201 bcache_device_attach(d, c, u - c->uuids);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001202 bch_flash_dev_request_init(d);
1203 add_disk(d->disk);
1204
1205 if (kobject_add(&d->kobj, &disk_to_dev(d->disk)->kobj, "bcache"))
1206 goto err;
1207
1208 bcache_device_link(d, c, "volume");
1209
1210 return 0;
1211err:
1212 kobject_put(&d->kobj);
1213 return -ENOMEM;
1214}
1215
1216static int flash_devs_run(struct cache_set *c)
1217{
1218 int ret = 0;
1219 struct uuid_entry *u;
1220
1221 for (u = c->uuids;
1222 u < c->uuids + c->nr_uuids && !ret;
1223 u++)
1224 if (UUID_FLASH_ONLY(u))
1225 ret = flash_dev_run(c, u);
1226
1227 return ret;
1228}
1229
1230int bch_flash_dev_create(struct cache_set *c, uint64_t size)
1231{
1232 struct uuid_entry *u;
1233
1234 if (test_bit(CACHE_SET_STOPPING, &c->flags))
1235 return -EINTR;
1236
1237 u = uuid_find_empty(c);
1238 if (!u) {
1239 pr_err("Can't create volume, no room for UUID");
1240 return -EINVAL;
1241 }
1242
1243 get_random_bytes(u->uuid, 16);
1244 memset(u->label, 0, 32);
1245 u->first_reg = u->last_reg = cpu_to_le32(get_seconds());
1246
1247 SET_UUID_FLASH_ONLY(u, 1);
1248 u->sectors = size >> 9;
1249
1250 bch_uuid_write(c);
1251
1252 return flash_dev_run(c, u);
1253}
1254
1255/* Cache set */
1256
1257__printf(2, 3)
1258bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...)
1259{
1260 va_list args;
1261
Kent Overstreet77c320e2013-07-11 19:42:51 -07001262 if (c->on_error != ON_ERROR_PANIC &&
1263 test_bit(CACHE_SET_STOPPING, &c->flags))
Kent Overstreetcafe5632013-03-23 16:11:31 -07001264 return false;
1265
1266 /* XXX: we can be called from atomic context
1267 acquire_console_sem();
1268 */
1269
1270 printk(KERN_ERR "bcache: error on %pU: ", c->sb.set_uuid);
1271
1272 va_start(args, fmt);
1273 vprintk(fmt, args);
1274 va_end(args);
1275
1276 printk(", disabling caching\n");
1277
Kent Overstreet77c320e2013-07-11 19:42:51 -07001278 if (c->on_error == ON_ERROR_PANIC)
1279 panic("panic forced after error\n");
1280
Kent Overstreetcafe5632013-03-23 16:11:31 -07001281 bch_cache_set_unregister(c);
1282 return true;
1283}
1284
1285void bch_cache_set_release(struct kobject *kobj)
1286{
1287 struct cache_set *c = container_of(kobj, struct cache_set, kobj);
1288 kfree(c);
1289 module_put(THIS_MODULE);
1290}
1291
1292static void cache_set_free(struct closure *cl)
1293{
1294 struct cache_set *c = container_of(cl, struct cache_set, cl);
1295 struct cache *ca;
1296 unsigned i;
1297
1298 if (!IS_ERR_OR_NULL(c->debug))
1299 debugfs_remove(c->debug);
1300
1301 bch_open_buckets_free(c);
1302 bch_btree_cache_free(c);
1303 bch_journal_free(c);
1304
1305 for_each_cache(ca, c, i)
1306 if (ca)
1307 kobject_put(&ca->kobj);
1308
1309 free_pages((unsigned long) c->uuids, ilog2(bucket_pages(c)));
1310 free_pages((unsigned long) c->sort, ilog2(bucket_pages(c)));
1311
Kent Overstreetcafe5632013-03-23 16:11:31 -07001312 if (c->bio_split)
1313 bioset_free(c->bio_split);
Kent Overstreet57943512013-04-25 13:58:35 -07001314 if (c->fill_iter)
1315 mempool_destroy(c->fill_iter);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001316 if (c->bio_meta)
1317 mempool_destroy(c->bio_meta);
1318 if (c->search)
1319 mempool_destroy(c->search);
1320 kfree(c->devices);
1321
1322 mutex_lock(&bch_register_lock);
1323 list_del(&c->list);
1324 mutex_unlock(&bch_register_lock);
1325
1326 pr_info("Cache set %pU unregistered", c->sb.set_uuid);
1327 wake_up(&unregister_wait);
1328
1329 closure_debug_destroy(&c->cl);
1330 kobject_put(&c->kobj);
1331}
1332
1333static void cache_set_flush(struct closure *cl)
1334{
1335 struct cache_set *c = container_of(cl, struct cache_set, caching);
Kent Overstreet79826c32013-07-10 18:31:58 -07001336 struct cache *ca;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001337 struct btree *b;
Kent Overstreet79826c32013-07-10 18:31:58 -07001338 unsigned i;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001339
1340 bch_cache_accounting_destroy(&c->accounting);
1341
1342 kobject_put(&c->internal);
1343 kobject_del(&c->kobj);
1344
1345 if (!IS_ERR_OR_NULL(c->root))
1346 list_add(&c->root->list, &c->btree_cache);
1347
1348 /* Should skip this if we're unregistering because of an error */
1349 list_for_each_entry(b, &c->btree_cache, list)
1350 if (btree_node_dirty(b))
Kent Overstreet57943512013-04-25 13:58:35 -07001351 bch_btree_node_write(b, NULL);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001352
Kent Overstreet79826c32013-07-10 18:31:58 -07001353 for_each_cache(ca, c, i)
1354 if (ca->alloc_thread)
1355 kthread_stop(ca->alloc_thread);
1356
Kent Overstreetcafe5632013-03-23 16:11:31 -07001357 closure_return(cl);
1358}
1359
1360static void __cache_set_unregister(struct closure *cl)
1361{
1362 struct cache_set *c = container_of(cl, struct cache_set, caching);
Kent Overstreet5caa52a2013-07-10 21:03:25 -07001363 struct cached_dev *dc;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001364 size_t i;
1365
1366 mutex_lock(&bch_register_lock);
1367
Kent Overstreetcafe5632013-03-23 16:11:31 -07001368 for (i = 0; i < c->nr_uuids; i++)
Kent Overstreet5caa52a2013-07-10 21:03:25 -07001369 if (c->devices[i]) {
1370 if (!UUID_FLASH_ONLY(&c->uuids[i]) &&
1371 test_bit(CACHE_SET_UNREGISTERING, &c->flags)) {
1372 dc = container_of(c->devices[i],
1373 struct cached_dev, disk);
1374 bch_cached_dev_detach(dc);
1375 } else {
1376 bcache_device_stop(c->devices[i]);
1377 }
1378 }
Kent Overstreetcafe5632013-03-23 16:11:31 -07001379
1380 mutex_unlock(&bch_register_lock);
1381
1382 continue_at(cl, cache_set_flush, system_wq);
1383}
1384
1385void bch_cache_set_stop(struct cache_set *c)
1386{
1387 if (!test_and_set_bit(CACHE_SET_STOPPING, &c->flags))
1388 closure_queue(&c->caching);
1389}
1390
1391void bch_cache_set_unregister(struct cache_set *c)
1392{
1393 set_bit(CACHE_SET_UNREGISTERING, &c->flags);
1394 bch_cache_set_stop(c);
1395}
1396
1397#define alloc_bucket_pages(gfp, c) \
1398 ((void *) __get_free_pages(__GFP_ZERO|gfp, ilog2(bucket_pages(c))))
1399
1400struct cache_set *bch_cache_set_alloc(struct cache_sb *sb)
1401{
1402 int iter_size;
1403 struct cache_set *c = kzalloc(sizeof(struct cache_set), GFP_KERNEL);
1404 if (!c)
1405 return NULL;
1406
1407 __module_get(THIS_MODULE);
1408 closure_init(&c->cl, NULL);
1409 set_closure_fn(&c->cl, cache_set_free, system_wq);
1410
1411 closure_init(&c->caching, &c->cl);
1412 set_closure_fn(&c->caching, __cache_set_unregister, system_wq);
1413
1414 /* Maybe create continue_at_noreturn() and use it here? */
1415 closure_set_stopped(&c->cl);
1416 closure_put(&c->cl);
1417
1418 kobject_init(&c->kobj, &bch_cache_set_ktype);
1419 kobject_init(&c->internal, &bch_cache_set_internal_ktype);
1420
1421 bch_cache_accounting_init(&c->accounting, &c->cl);
1422
1423 memcpy(c->sb.set_uuid, sb->set_uuid, 16);
1424 c->sb.block_size = sb->block_size;
1425 c->sb.bucket_size = sb->bucket_size;
1426 c->sb.nr_in_set = sb->nr_in_set;
1427 c->sb.last_mount = sb->last_mount;
1428 c->bucket_bits = ilog2(sb->bucket_size);
1429 c->block_bits = ilog2(sb->block_size);
1430 c->nr_uuids = bucket_bytes(c) / sizeof(struct uuid_entry);
1431
1432 c->btree_pages = c->sb.bucket_size / PAGE_SECTORS;
1433 if (c->btree_pages > BTREE_MAX_PAGES)
1434 c->btree_pages = max_t(int, c->btree_pages / 4,
1435 BTREE_MAX_PAGES);
1436
Kent Overstreet6ded34d2013-05-11 15:59:37 -07001437 c->sort_crit_factor = int_sqrt(c->btree_pages);
1438
Kent Overstreetcafe5632013-03-23 16:11:31 -07001439 mutex_init(&c->bucket_lock);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001440 mutex_init(&c->sort_lock);
1441 spin_lock_init(&c->sort_time_lock);
1442 closure_init_unlocked(&c->sb_write);
1443 closure_init_unlocked(&c->uuid_write);
1444 spin_lock_init(&c->btree_read_time_lock);
1445 bch_moving_init_cache_set(c);
1446
1447 INIT_LIST_HEAD(&c->list);
1448 INIT_LIST_HEAD(&c->cached_devs);
1449 INIT_LIST_HEAD(&c->btree_cache);
1450 INIT_LIST_HEAD(&c->btree_cache_freeable);
1451 INIT_LIST_HEAD(&c->btree_cache_freed);
1452 INIT_LIST_HEAD(&c->data_buckets);
1453
1454 c->search = mempool_create_slab_pool(32, bch_search_cache);
1455 if (!c->search)
1456 goto err;
1457
1458 iter_size = (sb->bucket_size / sb->block_size + 1) *
1459 sizeof(struct btree_iter_set);
1460
1461 if (!(c->devices = kzalloc(c->nr_uuids * sizeof(void *), GFP_KERNEL)) ||
1462 !(c->bio_meta = mempool_create_kmalloc_pool(2,
1463 sizeof(struct bbio) + sizeof(struct bio_vec) *
1464 bucket_pages(c))) ||
Kent Overstreet57943512013-04-25 13:58:35 -07001465 !(c->fill_iter = mempool_create_kmalloc_pool(1, iter_size)) ||
Kent Overstreetcafe5632013-03-23 16:11:31 -07001466 !(c->bio_split = bioset_create(4, offsetof(struct bbio, bio))) ||
Kent Overstreetcafe5632013-03-23 16:11:31 -07001467 !(c->sort = alloc_bucket_pages(GFP_KERNEL, c)) ||
1468 !(c->uuids = alloc_bucket_pages(GFP_KERNEL, c)) ||
1469 bch_journal_alloc(c) ||
1470 bch_btree_cache_alloc(c) ||
1471 bch_open_buckets_alloc(c))
1472 goto err;
1473
Kent Overstreetcafe5632013-03-23 16:11:31 -07001474 c->congested_read_threshold_us = 2000;
1475 c->congested_write_threshold_us = 20000;
1476 c->error_limit = 8 << IO_ERROR_SHIFT;
1477
1478 return c;
1479err:
1480 bch_cache_set_unregister(c);
1481 return NULL;
1482}
1483
1484static void run_cache_set(struct cache_set *c)
1485{
1486 const char *err = "cannot allocate memory";
1487 struct cached_dev *dc, *t;
1488 struct cache *ca;
1489 unsigned i;
1490
1491 struct btree_op op;
1492 bch_btree_op_init_stack(&op);
1493 op.lock = SHRT_MAX;
1494
1495 for_each_cache(ca, c, i)
1496 c->nbuckets += ca->sb.nbuckets;
1497
1498 if (CACHE_SYNC(&c->sb)) {
1499 LIST_HEAD(journal);
1500 struct bkey *k;
1501 struct jset *j;
1502
1503 err = "cannot allocate memory for journal";
1504 if (bch_journal_read(c, &journal, &op))
1505 goto err;
1506
1507 pr_debug("btree_journal_read() done");
1508
1509 err = "no journal entries found";
1510 if (list_empty(&journal))
1511 goto err;
1512
1513 j = &list_entry(journal.prev, struct journal_replay, list)->j;
1514
1515 err = "IO error reading priorities";
1516 for_each_cache(ca, c, i)
1517 prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev]);
1518
1519 /*
1520 * If prio_read() fails it'll call cache_set_error and we'll
1521 * tear everything down right away, but if we perhaps checked
1522 * sooner we could avoid journal replay.
1523 */
1524
1525 k = &j->btree_root;
1526
1527 err = "bad btree root";
1528 if (__bch_ptr_invalid(c, j->btree_level + 1, k))
1529 goto err;
1530
1531 err = "error reading btree root";
1532 c->root = bch_btree_node_get(c, k, j->btree_level, &op);
1533 if (IS_ERR_OR_NULL(c->root))
1534 goto err;
1535
1536 list_del_init(&c->root->list);
1537 rw_unlock(true, c->root);
1538
1539 err = uuid_read(c, j, &op.cl);
1540 if (err)
1541 goto err;
1542
1543 err = "error in recovery";
1544 if (bch_btree_check(c, &op))
1545 goto err;
1546
1547 bch_journal_mark(c, &journal);
1548 bch_btree_gc_finish(c);
1549 pr_debug("btree_check() done");
1550
1551 /*
1552 * bcache_journal_next() can't happen sooner, or
1553 * btree_gc_finish() will give spurious errors about last_gc >
1554 * gc_gen - this is a hack but oh well.
1555 */
1556 bch_journal_next(&c->journal);
1557
Kent Overstreet119ba0f2013-04-24 19:01:12 -07001558 err = "error starting allocator thread";
Kent Overstreetcafe5632013-03-23 16:11:31 -07001559 for_each_cache(ca, c, i)
Kent Overstreet119ba0f2013-04-24 19:01:12 -07001560 if (bch_cache_allocator_start(ca))
1561 goto err;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001562
1563 /*
1564 * First place it's safe to allocate: btree_check() and
1565 * btree_gc_finish() have to run before we have buckets to
1566 * allocate, and bch_bucket_alloc_set() might cause a journal
1567 * entry to be written so bcache_journal_next() has to be called
1568 * first.
1569 *
1570 * If the uuids were in the old format we have to rewrite them
1571 * before the next journal entry is written:
1572 */
1573 if (j->version < BCACHE_JSET_VERSION_UUID)
1574 __uuid_write(c);
1575
1576 bch_journal_replay(c, &journal, &op);
1577 } else {
1578 pr_notice("invalidating existing data");
1579 /* Don't want invalidate_buckets() to queue a gc yet */
1580 closure_lock(&c->gc, NULL);
1581
1582 for_each_cache(ca, c, i) {
1583 unsigned j;
1584
1585 ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7,
1586 2, SB_JOURNAL_BUCKETS);
1587
1588 for (j = 0; j < ca->sb.keys; j++)
1589 ca->sb.d[j] = ca->sb.first_bucket + j;
1590 }
1591
1592 bch_btree_gc_finish(c);
1593
Kent Overstreet119ba0f2013-04-24 19:01:12 -07001594 err = "error starting allocator thread";
Kent Overstreetcafe5632013-03-23 16:11:31 -07001595 for_each_cache(ca, c, i)
Kent Overstreet119ba0f2013-04-24 19:01:12 -07001596 if (bch_cache_allocator_start(ca))
1597 goto err;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001598
1599 mutex_lock(&c->bucket_lock);
1600 for_each_cache(ca, c, i)
1601 bch_prio_write(ca);
1602 mutex_unlock(&c->bucket_lock);
1603
Kent Overstreetcafe5632013-03-23 16:11:31 -07001604 err = "cannot allocate new UUID bucket";
1605 if (__uuid_write(c))
1606 goto err_unlock_gc;
1607
1608 err = "cannot allocate new btree root";
1609 c->root = bch_btree_node_alloc(c, 0, &op.cl);
1610 if (IS_ERR_OR_NULL(c->root))
1611 goto err_unlock_gc;
1612
1613 bkey_copy_key(&c->root->key, &MAX_KEY);
Kent Overstreet57943512013-04-25 13:58:35 -07001614 bch_btree_node_write(c->root, &op.cl);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001615
1616 bch_btree_set_root(c->root);
1617 rw_unlock(true, c->root);
1618
1619 /*
1620 * We don't want to write the first journal entry until
1621 * everything is set up - fortunately journal entries won't be
1622 * written until the SET_CACHE_SYNC() here:
1623 */
1624 SET_CACHE_SYNC(&c->sb, true);
1625
1626 bch_journal_next(&c->journal);
1627 bch_journal_meta(c, &op.cl);
1628
1629 /* Unlock */
1630 closure_set_stopped(&c->gc.cl);
1631 closure_put(&c->gc.cl);
1632 }
1633
1634 closure_sync(&op.cl);
1635 c->sb.last_mount = get_seconds();
1636 bcache_write_super(c);
1637
1638 list_for_each_entry_safe(dc, t, &uncached_devices, list)
1639 bch_cached_dev_attach(dc, c);
1640
1641 flash_devs_run(c);
1642
1643 return;
1644err_unlock_gc:
1645 closure_set_stopped(&c->gc.cl);
1646 closure_put(&c->gc.cl);
1647err:
1648 closure_sync(&op.cl);
1649 /* XXX: test this, it's broken */
1650 bch_cache_set_error(c, err);
1651}
1652
1653static bool can_attach_cache(struct cache *ca, struct cache_set *c)
1654{
1655 return ca->sb.block_size == c->sb.block_size &&
1656 ca->sb.bucket_size == c->sb.block_size &&
1657 ca->sb.nr_in_set == c->sb.nr_in_set;
1658}
1659
1660static const char *register_cache_set(struct cache *ca)
1661{
1662 char buf[12];
1663 const char *err = "cannot allocate memory";
1664 struct cache_set *c;
1665
1666 list_for_each_entry(c, &bch_cache_sets, list)
1667 if (!memcmp(c->sb.set_uuid, ca->sb.set_uuid, 16)) {
1668 if (c->cache[ca->sb.nr_this_dev])
1669 return "duplicate cache set member";
1670
1671 if (!can_attach_cache(ca, c))
1672 return "cache sb does not match set";
1673
1674 if (!CACHE_SYNC(&ca->sb))
1675 SET_CACHE_SYNC(&c->sb, false);
1676
1677 goto found;
1678 }
1679
1680 c = bch_cache_set_alloc(&ca->sb);
1681 if (!c)
1682 return err;
1683
1684 err = "error creating kobject";
1685 if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->sb.set_uuid) ||
1686 kobject_add(&c->internal, &c->kobj, "internal"))
1687 goto err;
1688
1689 if (bch_cache_accounting_add_kobjs(&c->accounting, &c->kobj))
1690 goto err;
1691
1692 bch_debug_init_cache_set(c);
1693
1694 list_add(&c->list, &bch_cache_sets);
1695found:
1696 sprintf(buf, "cache%i", ca->sb.nr_this_dev);
1697 if (sysfs_create_link(&ca->kobj, &c->kobj, "set") ||
1698 sysfs_create_link(&c->kobj, &ca->kobj, buf))
1699 goto err;
1700
1701 if (ca->sb.seq > c->sb.seq) {
1702 c->sb.version = ca->sb.version;
1703 memcpy(c->sb.set_uuid, ca->sb.set_uuid, 16);
1704 c->sb.flags = ca->sb.flags;
1705 c->sb.seq = ca->sb.seq;
1706 pr_debug("set version = %llu", c->sb.version);
1707 }
1708
1709 ca->set = c;
1710 ca->set->cache[ca->sb.nr_this_dev] = ca;
1711 c->cache_by_alloc[c->caches_loaded++] = ca;
1712
1713 if (c->caches_loaded == c->sb.nr_in_set)
1714 run_cache_set(c);
1715
1716 return NULL;
1717err:
1718 bch_cache_set_unregister(c);
1719 return err;
1720}
1721
1722/* Cache device */
1723
1724void bch_cache_release(struct kobject *kobj)
1725{
1726 struct cache *ca = container_of(kobj, struct cache, kobj);
1727
1728 if (ca->set)
1729 ca->set->cache[ca->sb.nr_this_dev] = NULL;
1730
Kent Overstreetcafe5632013-03-23 16:11:31 -07001731 bio_split_pool_free(&ca->bio_split_hook);
1732
Kent Overstreetcafe5632013-03-23 16:11:31 -07001733 free_pages((unsigned long) ca->disk_buckets, ilog2(bucket_pages(ca)));
1734 kfree(ca->prio_buckets);
1735 vfree(ca->buckets);
1736
1737 free_heap(&ca->heap);
1738 free_fifo(&ca->unused);
1739 free_fifo(&ca->free_inc);
1740 free_fifo(&ca->free);
1741
1742 if (ca->sb_bio.bi_inline_vecs[0].bv_page)
1743 put_page(ca->sb_bio.bi_io_vec[0].bv_page);
1744
1745 if (!IS_ERR_OR_NULL(ca->bdev)) {
1746 blk_sync_queue(bdev_get_queue(ca->bdev));
1747 blkdev_put(ca->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1748 }
1749
1750 kfree(ca);
1751 module_put(THIS_MODULE);
1752}
1753
1754static int cache_alloc(struct cache_sb *sb, struct cache *ca)
1755{
1756 size_t free;
1757 struct bucket *b;
1758
Kent Overstreetcafe5632013-03-23 16:11:31 -07001759 __module_get(THIS_MODULE);
1760 kobject_init(&ca->kobj, &bch_cache_ktype);
1761
Kent Overstreetcafe5632013-03-23 16:11:31 -07001762 bio_init(&ca->journal.bio);
1763 ca->journal.bio.bi_max_vecs = 8;
1764 ca->journal.bio.bi_io_vec = ca->journal.bio.bi_inline_vecs;
1765
1766 free = roundup_pow_of_two(ca->sb.nbuckets) >> 9;
1767 free = max_t(size_t, free, (prio_buckets(ca) + 8) * 2);
1768
1769 if (!init_fifo(&ca->free, free, GFP_KERNEL) ||
1770 !init_fifo(&ca->free_inc, free << 2, GFP_KERNEL) ||
1771 !init_fifo(&ca->unused, free << 2, GFP_KERNEL) ||
1772 !init_heap(&ca->heap, free << 3, GFP_KERNEL) ||
Kent Overstreetf59fce82013-05-15 00:11:26 -07001773 !(ca->buckets = vzalloc(sizeof(struct bucket) *
Kent Overstreetcafe5632013-03-23 16:11:31 -07001774 ca->sb.nbuckets)) ||
1775 !(ca->prio_buckets = kzalloc(sizeof(uint64_t) * prio_buckets(ca) *
1776 2, GFP_KERNEL)) ||
1777 !(ca->disk_buckets = alloc_bucket_pages(GFP_KERNEL, ca)) ||
Kent Overstreetcafe5632013-03-23 16:11:31 -07001778 bio_split_pool_init(&ca->bio_split_hook))
Kent Overstreetf59fce82013-05-15 00:11:26 -07001779 return -ENOMEM;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001780
1781 ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca);
1782
Kent Overstreetcafe5632013-03-23 16:11:31 -07001783 for_each_bucket(b, ca)
1784 atomic_set(&b->pin, 0);
1785
1786 if (bch_cache_allocator_init(ca))
1787 goto err;
1788
1789 return 0;
1790err:
1791 kobject_put(&ca->kobj);
1792 return -ENOMEM;
1793}
1794
Kent Overstreetf59fce82013-05-15 00:11:26 -07001795static void register_cache(struct cache_sb *sb, struct page *sb_page,
Kent Overstreetcafe5632013-03-23 16:11:31 -07001796 struct block_device *bdev, struct cache *ca)
1797{
1798 char name[BDEVNAME_SIZE];
1799 const char *err = "cannot allocate memory";
1800
Kent Overstreetf59fce82013-05-15 00:11:26 -07001801 memcpy(&ca->sb, sb, sizeof(struct cache_sb));
Kent Overstreetcafe5632013-03-23 16:11:31 -07001802 ca->bdev = bdev;
1803 ca->bdev->bd_holder = ca;
1804
Kent Overstreetf59fce82013-05-15 00:11:26 -07001805 bio_init(&ca->sb_bio);
1806 ca->sb_bio.bi_max_vecs = 1;
1807 ca->sb_bio.bi_io_vec = ca->sb_bio.bi_inline_vecs;
1808 ca->sb_bio.bi_io_vec[0].bv_page = sb_page;
1809 get_page(sb_page);
1810
Kent Overstreetcafe5632013-03-23 16:11:31 -07001811 if (blk_queue_discard(bdev_get_queue(ca->bdev)))
1812 ca->discard = CACHE_DISCARD(&ca->sb);
1813
Kent Overstreetf59fce82013-05-15 00:11:26 -07001814 if (cache_alloc(sb, ca) != 0)
1815 goto err;
1816
Kent Overstreetcafe5632013-03-23 16:11:31 -07001817 err = "error creating kobject";
1818 if (kobject_add(&ca->kobj, &part_to_dev(bdev->bd_part)->kobj, "bcache"))
1819 goto err;
1820
1821 err = register_cache_set(ca);
1822 if (err)
1823 goto err;
1824
1825 pr_info("registered cache device %s", bdevname(bdev, name));
Kent Overstreetf59fce82013-05-15 00:11:26 -07001826 return;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001827err:
Kent Overstreetf59fce82013-05-15 00:11:26 -07001828 pr_notice("error opening %s: %s", bdevname(bdev, name), err);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001829 kobject_put(&ca->kobj);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001830}
1831
1832/* Global interfaces/init */
1833
1834static ssize_t register_bcache(struct kobject *, struct kobj_attribute *,
1835 const char *, size_t);
1836
1837kobj_attribute_write(register, register_bcache);
1838kobj_attribute_write(register_quiet, register_bcache);
1839
Gabriel de Perthuisa9dd53a2013-05-04 12:19:41 +02001840static bool bch_is_open_backing(struct block_device *bdev) {
1841 struct cache_set *c, *tc;
1842 struct cached_dev *dc, *t;
1843
1844 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
1845 list_for_each_entry_safe(dc, t, &c->cached_devs, list)
1846 if (dc->bdev == bdev)
1847 return true;
1848 list_for_each_entry_safe(dc, t, &uncached_devices, list)
1849 if (dc->bdev == bdev)
1850 return true;
1851 return false;
1852}
1853
1854static bool bch_is_open_cache(struct block_device *bdev) {
1855 struct cache_set *c, *tc;
1856 struct cache *ca;
1857 unsigned i;
1858
1859 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
1860 for_each_cache(ca, c, i)
1861 if (ca->bdev == bdev)
1862 return true;
1863 return false;
1864}
1865
1866static bool bch_is_open(struct block_device *bdev) {
1867 return bch_is_open_cache(bdev) || bch_is_open_backing(bdev);
1868}
1869
Kent Overstreetcafe5632013-03-23 16:11:31 -07001870static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr,
1871 const char *buffer, size_t size)
1872{
1873 ssize_t ret = size;
1874 const char *err = "cannot allocate memory";
1875 char *path = NULL;
1876 struct cache_sb *sb = NULL;
1877 struct block_device *bdev = NULL;
1878 struct page *sb_page = NULL;
1879
1880 if (!try_module_get(THIS_MODULE))
1881 return -EBUSY;
1882
1883 mutex_lock(&bch_register_lock);
1884
1885 if (!(path = kstrndup(buffer, size, GFP_KERNEL)) ||
1886 !(sb = kmalloc(sizeof(struct cache_sb), GFP_KERNEL)))
1887 goto err;
1888
1889 err = "failed to open device";
1890 bdev = blkdev_get_by_path(strim(path),
1891 FMODE_READ|FMODE_WRITE|FMODE_EXCL,
1892 sb);
Kent Overstreetf59fce82013-05-15 00:11:26 -07001893 if (IS_ERR(bdev)) {
Gabriel de Perthuisa9dd53a2013-05-04 12:19:41 +02001894 if (bdev == ERR_PTR(-EBUSY)) {
1895 bdev = lookup_bdev(strim(path));
1896 if (!IS_ERR(bdev) && bch_is_open(bdev))
1897 err = "device already registered";
1898 else
1899 err = "device busy";
1900 }
Kent Overstreetcafe5632013-03-23 16:11:31 -07001901 goto err;
Kent Overstreetf59fce82013-05-15 00:11:26 -07001902 }
1903
1904 err = "failed to set blocksize";
1905 if (set_blocksize(bdev, 4096))
1906 goto err_close;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001907
1908 err = read_super(sb, bdev, &sb_page);
1909 if (err)
1910 goto err_close;
1911
Kent Overstreet29033812013-04-11 15:14:35 -07001912 if (SB_IS_BDEV(sb)) {
Kent Overstreetcafe5632013-03-23 16:11:31 -07001913 struct cached_dev *dc = kzalloc(sizeof(*dc), GFP_KERNEL);
Kent Overstreetf59fce82013-05-15 00:11:26 -07001914 if (!dc)
1915 goto err_close;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001916
Kent Overstreetf59fce82013-05-15 00:11:26 -07001917 register_bdev(sb, sb_page, bdev, dc);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001918 } else {
1919 struct cache *ca = kzalloc(sizeof(*ca), GFP_KERNEL);
Kent Overstreetf59fce82013-05-15 00:11:26 -07001920 if (!ca)
1921 goto err_close;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001922
Kent Overstreetf59fce82013-05-15 00:11:26 -07001923 register_cache(sb, sb_page, bdev, ca);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001924 }
Kent Overstreetf59fce82013-05-15 00:11:26 -07001925out:
1926 if (sb_page)
Kent Overstreetcafe5632013-03-23 16:11:31 -07001927 put_page(sb_page);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001928 kfree(sb);
1929 kfree(path);
1930 mutex_unlock(&bch_register_lock);
1931 module_put(THIS_MODULE);
1932 return ret;
Kent Overstreetf59fce82013-05-15 00:11:26 -07001933
1934err_close:
1935 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1936err:
1937 if (attr != &ksysfs_register_quiet)
1938 pr_info("error opening %s: %s", path, err);
1939 ret = -EINVAL;
1940 goto out;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001941}
1942
1943static int bcache_reboot(struct notifier_block *n, unsigned long code, void *x)
1944{
1945 if (code == SYS_DOWN ||
1946 code == SYS_HALT ||
1947 code == SYS_POWER_OFF) {
1948 DEFINE_WAIT(wait);
1949 unsigned long start = jiffies;
1950 bool stopped = false;
1951
1952 struct cache_set *c, *tc;
1953 struct cached_dev *dc, *tdc;
1954
1955 mutex_lock(&bch_register_lock);
1956
1957 if (list_empty(&bch_cache_sets) &&
1958 list_empty(&uncached_devices))
1959 goto out;
1960
1961 pr_info("Stopping all devices:");
1962
1963 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
1964 bch_cache_set_stop(c);
1965
1966 list_for_each_entry_safe(dc, tdc, &uncached_devices, list)
1967 bcache_device_stop(&dc->disk);
1968
1969 /* What's a condition variable? */
1970 while (1) {
1971 long timeout = start + 2 * HZ - jiffies;
1972
1973 stopped = list_empty(&bch_cache_sets) &&
1974 list_empty(&uncached_devices);
1975
1976 if (timeout < 0 || stopped)
1977 break;
1978
1979 prepare_to_wait(&unregister_wait, &wait,
1980 TASK_UNINTERRUPTIBLE);
1981
1982 mutex_unlock(&bch_register_lock);
1983 schedule_timeout(timeout);
1984 mutex_lock(&bch_register_lock);
1985 }
1986
1987 finish_wait(&unregister_wait, &wait);
1988
1989 if (stopped)
1990 pr_info("All devices stopped");
1991 else
1992 pr_notice("Timeout waiting for devices to be closed");
1993out:
1994 mutex_unlock(&bch_register_lock);
1995 }
1996
1997 return NOTIFY_DONE;
1998}
1999
2000static struct notifier_block reboot = {
2001 .notifier_call = bcache_reboot,
2002 .priority = INT_MAX, /* before any real devices */
2003};
2004
2005static void bcache_exit(void)
2006{
2007 bch_debug_exit();
2008 bch_writeback_exit();
2009 bch_request_exit();
2010 bch_btree_exit();
2011 if (bcache_kobj)
2012 kobject_put(bcache_kobj);
2013 if (bcache_wq)
2014 destroy_workqueue(bcache_wq);
2015 unregister_blkdev(bcache_major, "bcache");
2016 unregister_reboot_notifier(&reboot);
2017}
2018
2019static int __init bcache_init(void)
2020{
2021 static const struct attribute *files[] = {
2022 &ksysfs_register.attr,
2023 &ksysfs_register_quiet.attr,
2024 NULL
2025 };
2026
2027 mutex_init(&bch_register_lock);
2028 init_waitqueue_head(&unregister_wait);
2029 register_reboot_notifier(&reboot);
Kent Overstreet07e86cc2013-03-25 11:46:43 -07002030 closure_debug_init();
Kent Overstreetcafe5632013-03-23 16:11:31 -07002031
2032 bcache_major = register_blkdev(0, "bcache");
2033 if (bcache_major < 0)
2034 return bcache_major;
2035
2036 if (!(bcache_wq = create_workqueue("bcache")) ||
2037 !(bcache_kobj = kobject_create_and_add("bcache", fs_kobj)) ||
2038 sysfs_create_files(bcache_kobj, files) ||
2039 bch_btree_init() ||
2040 bch_request_init() ||
2041 bch_writeback_init() ||
2042 bch_debug_init(bcache_kobj))
2043 goto err;
2044
2045 return 0;
2046err:
2047 bcache_exit();
2048 return -ENOMEM;
2049}
2050
2051module_exit(bcache_exit);
2052module_init(bcache_init);