blob: fb343276beef7f2ab176bae1d7347822f3a87bca [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"
Kent Overstreet65d45232013-12-20 17:22:05 -080012#include "extents.h"
Kent Overstreetcafe5632013-03-23 16:11:31 -070013#include "request.h"
Kent Overstreet279afba2013-06-05 06:21:07 -070014#include "writeback.h"
Kent Overstreetcafe5632013-03-23 16:11:31 -070015
Kent Overstreetc37511b2013-04-26 15:39:55 -070016#include <linux/blkdev.h>
Kent Overstreetcafe5632013-03-23 16:11:31 -070017#include <linux/buffer_head.h>
18#include <linux/debugfs.h>
19#include <linux/genhd.h>
Kent Overstreet28935ab2013-07-31 01:12:02 -070020#include <linux/idr.h>
Kent Overstreet79826c32013-07-10 18:31:58 -070021#include <linux/kthread.h>
Kent Overstreetcafe5632013-03-23 16:11:31 -070022#include <linux/module.h>
23#include <linux/random.h>
24#include <linux/reboot.h>
25#include <linux/sysfs.h>
26
27MODULE_LICENSE("GPL");
28MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
29
30static const char bcache_magic[] = {
31 0xc6, 0x85, 0x73, 0xf6, 0x4e, 0x1a, 0x45, 0xca,
32 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81
33};
34
35static const char invalid_uuid[] = {
36 0xa0, 0x3e, 0xf8, 0xed, 0x3e, 0xe1, 0xb8, 0x78,
37 0xc8, 0x50, 0xfc, 0x5e, 0xcb, 0x16, 0xcd, 0x99
38};
39
40/* Default is -1; we skip past it for struct cached_dev's cache mode */
41const char * const bch_cache_modes[] = {
42 "default",
43 "writethrough",
44 "writeback",
45 "writearound",
46 "none",
47 NULL
48};
49
Kent Overstreetcafe5632013-03-23 16:11:31 -070050static struct kobject *bcache_kobj;
51struct mutex bch_register_lock;
52LIST_HEAD(bch_cache_sets);
53static LIST_HEAD(uncached_devices);
54
Kent Overstreet28935ab2013-07-31 01:12:02 -070055static int bcache_major;
56static DEFINE_IDA(bcache_minor);
Kent Overstreetcafe5632013-03-23 16:11:31 -070057static wait_queue_head_t unregister_wait;
58struct workqueue_struct *bcache_wq;
59
60#define BTREE_MAX_PAGES (256 * 1024 / PAGE_SIZE)
61
62static void bio_split_pool_free(struct bio_split_pool *p)
63{
Kent Overstreet8ef74792013-04-05 13:46:13 -070064 if (p->bio_split_hook)
65 mempool_destroy(p->bio_split_hook);
66
Kent Overstreetcafe5632013-03-23 16:11:31 -070067 if (p->bio_split)
68 bioset_free(p->bio_split);
Kent Overstreetcafe5632013-03-23 16:11:31 -070069}
70
71static int bio_split_pool_init(struct bio_split_pool *p)
72{
73 p->bio_split = bioset_create(4, 0);
74 if (!p->bio_split)
75 return -ENOMEM;
76
77 p->bio_split_hook = mempool_create_kmalloc_pool(4,
78 sizeof(struct bio_split_hook));
79 if (!p->bio_split_hook)
80 return -ENOMEM;
81
82 return 0;
83}
84
85/* Superblock */
86
87static const char *read_super(struct cache_sb *sb, struct block_device *bdev,
88 struct page **res)
89{
90 const char *err;
91 struct cache_sb *s;
92 struct buffer_head *bh = __bread(bdev, 1, SB_SIZE);
93 unsigned i;
94
95 if (!bh)
96 return "IO error";
97
98 s = (struct cache_sb *) bh->b_data;
99
100 sb->offset = le64_to_cpu(s->offset);
101 sb->version = le64_to_cpu(s->version);
102
103 memcpy(sb->magic, s->magic, 16);
104 memcpy(sb->uuid, s->uuid, 16);
105 memcpy(sb->set_uuid, s->set_uuid, 16);
106 memcpy(sb->label, s->label, SB_LABEL_SIZE);
107
108 sb->flags = le64_to_cpu(s->flags);
109 sb->seq = le64_to_cpu(s->seq);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700110 sb->last_mount = le32_to_cpu(s->last_mount);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700111 sb->first_bucket = le16_to_cpu(s->first_bucket);
112 sb->keys = le16_to_cpu(s->keys);
113
114 for (i = 0; i < SB_JOURNAL_BUCKETS; i++)
115 sb->d[i] = le64_to_cpu(s->d[i]);
116
117 pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u",
118 sb->version, sb->flags, sb->seq, sb->keys);
119
120 err = "Not a bcache superblock";
121 if (sb->offset != SB_SECTOR)
122 goto err;
123
124 if (memcmp(sb->magic, bcache_magic, 16))
125 goto err;
126
127 err = "Too many journal buckets";
128 if (sb->keys > SB_JOURNAL_BUCKETS)
129 goto err;
130
131 err = "Bad checksum";
132 if (s->csum != csum_set(s))
133 goto err;
134
135 err = "Bad UUID";
Kent Overstreet169ef1c2013-03-28 12:50:55 -0600136 if (bch_is_zero(sb->uuid, 16))
Kent Overstreetcafe5632013-03-23 16:11:31 -0700137 goto err;
138
Kent Overstreet8abb2a52013-04-23 21:51:48 -0700139 sb->block_size = le16_to_cpu(s->block_size);
140
141 err = "Superblock block size smaller than device block size";
142 if (sb->block_size << 9 < bdev_logical_block_size(bdev))
143 goto err;
144
Kent Overstreet29033812013-04-11 15:14:35 -0700145 switch (sb->version) {
146 case BCACHE_SB_VERSION_BDEV:
Kent Overstreet29033812013-04-11 15:14:35 -0700147 sb->data_offset = BDEV_DATA_START_DEFAULT;
148 break;
149 case BCACHE_SB_VERSION_BDEV_WITH_OFFSET:
Kent Overstreet29033812013-04-11 15:14:35 -0700150 sb->data_offset = le64_to_cpu(s->data_offset);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700151
Kent Overstreet29033812013-04-11 15:14:35 -0700152 err = "Bad data offset";
153 if (sb->data_offset < BDEV_DATA_START_DEFAULT)
Kent Overstreetcafe5632013-03-23 16:11:31 -0700154 goto err;
155
Kent Overstreet29033812013-04-11 15:14:35 -0700156 break;
157 case BCACHE_SB_VERSION_CDEV:
158 case BCACHE_SB_VERSION_CDEV_WITH_UUID:
159 sb->nbuckets = le64_to_cpu(s->nbuckets);
160 sb->block_size = le16_to_cpu(s->block_size);
161 sb->bucket_size = le16_to_cpu(s->bucket_size);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700162
Kent Overstreet29033812013-04-11 15:14:35 -0700163 sb->nr_in_set = le16_to_cpu(s->nr_in_set);
164 sb->nr_this_dev = le16_to_cpu(s->nr_this_dev);
165
166 err = "Too many buckets";
167 if (sb->nbuckets > LONG_MAX)
168 goto err;
169
170 err = "Not enough buckets";
171 if (sb->nbuckets < 1 << 7)
172 goto err;
173
174 err = "Bad block/bucket size";
175 if (!is_power_of_2(sb->block_size) ||
176 sb->block_size > PAGE_SECTORS ||
177 !is_power_of_2(sb->bucket_size) ||
178 sb->bucket_size < PAGE_SECTORS)
179 goto err;
180
181 err = "Invalid superblock: device too small";
182 if (get_capacity(bdev->bd_disk) < sb->bucket_size * sb->nbuckets)
183 goto err;
184
185 err = "Bad UUID";
186 if (bch_is_zero(sb->set_uuid, 16))
187 goto err;
188
189 err = "Bad cache device number in set";
190 if (!sb->nr_in_set ||
191 sb->nr_in_set <= sb->nr_this_dev ||
192 sb->nr_in_set > MAX_CACHES_PER_SET)
193 goto err;
194
195 err = "Journal buckets not sequential";
196 for (i = 0; i < sb->keys; i++)
197 if (sb->d[i] != sb->first_bucket + i)
198 goto err;
199
200 err = "Too many journal buckets";
201 if (sb->first_bucket + sb->keys > sb->nbuckets)
202 goto err;
203
204 err = "Invalid superblock: first bucket comes before end of super";
205 if (sb->first_bucket * sb->bucket_size < 16)
206 goto err;
207
208 break;
209 default:
210 err = "Unsupported superblock version";
Kent Overstreetcafe5632013-03-23 16:11:31 -0700211 goto err;
Kent Overstreet29033812013-04-11 15:14:35 -0700212 }
213
Kent Overstreetcafe5632013-03-23 16:11:31 -0700214 sb->last_mount = get_seconds();
215 err = NULL;
216
217 get_page(bh->b_page);
218 *res = bh->b_page;
219err:
220 put_bh(bh);
221 return err;
222}
223
224static void write_bdev_super_endio(struct bio *bio, int error)
225{
226 struct cached_dev *dc = bio->bi_private;
227 /* XXX: error checking */
228
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800229 closure_put(&dc->sb_write);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700230}
231
232static void __write_super(struct cache_sb *sb, struct bio *bio)
233{
234 struct cache_sb *out = page_address(bio->bi_io_vec[0].bv_page);
235 unsigned i;
236
Kent Overstreet4f024f32013-10-11 15:44:27 -0700237 bio->bi_iter.bi_sector = SB_SECTOR;
238 bio->bi_rw = REQ_SYNC|REQ_META;
239 bio->bi_iter.bi_size = SB_SIZE;
Kent Overstreet169ef1c2013-03-28 12:50:55 -0600240 bch_bio_map(bio, NULL);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700241
242 out->offset = cpu_to_le64(sb->offset);
243 out->version = cpu_to_le64(sb->version);
244
245 memcpy(out->uuid, sb->uuid, 16);
246 memcpy(out->set_uuid, sb->set_uuid, 16);
247 memcpy(out->label, sb->label, SB_LABEL_SIZE);
248
249 out->flags = cpu_to_le64(sb->flags);
250 out->seq = cpu_to_le64(sb->seq);
251
252 out->last_mount = cpu_to_le32(sb->last_mount);
253 out->first_bucket = cpu_to_le16(sb->first_bucket);
254 out->keys = cpu_to_le16(sb->keys);
255
256 for (i = 0; i < sb->keys; i++)
257 out->d[i] = cpu_to_le64(sb->d[i]);
258
259 out->csum = csum_set(out);
260
261 pr_debug("ver %llu, flags %llu, seq %llu",
262 sb->version, sb->flags, sb->seq);
263
264 submit_bio(REQ_WRITE, bio);
265}
266
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800267static void bch_write_bdev_super_unlock(struct closure *cl)
268{
269 struct cached_dev *dc = container_of(cl, struct cached_dev, sb_write);
270
271 up(&dc->sb_write_mutex);
272}
273
Kent Overstreetcafe5632013-03-23 16:11:31 -0700274void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent)
275{
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800276 struct closure *cl = &dc->sb_write;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700277 struct bio *bio = &dc->sb_bio;
278
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800279 down(&dc->sb_write_mutex);
280 closure_init(cl, parent);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700281
282 bio_reset(bio);
283 bio->bi_bdev = dc->bdev;
284 bio->bi_end_io = write_bdev_super_endio;
285 bio->bi_private = dc;
286
287 closure_get(cl);
288 __write_super(&dc->sb, bio);
289
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800290 closure_return_with_destructor(cl, bch_write_bdev_super_unlock);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700291}
292
293static void write_super_endio(struct bio *bio, int error)
294{
295 struct cache *ca = bio->bi_private;
296
297 bch_count_io_errors(ca, error, "writing superblock");
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800298 closure_put(&ca->set->sb_write);
299}
300
301static void bcache_write_super_unlock(struct closure *cl)
302{
303 struct cache_set *c = container_of(cl, struct cache_set, sb_write);
304
305 up(&c->sb_write_mutex);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700306}
307
308void bcache_write_super(struct cache_set *c)
309{
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800310 struct closure *cl = &c->sb_write;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700311 struct cache *ca;
312 unsigned i;
313
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800314 down(&c->sb_write_mutex);
315 closure_init(cl, &c->cl);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700316
317 c->sb.seq++;
318
319 for_each_cache(ca, c, i) {
320 struct bio *bio = &ca->sb_bio;
321
Kent Overstreet29033812013-04-11 15:14:35 -0700322 ca->sb.version = BCACHE_SB_VERSION_CDEV_WITH_UUID;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700323 ca->sb.seq = c->sb.seq;
324 ca->sb.last_mount = c->sb.last_mount;
325
326 SET_CACHE_SYNC(&ca->sb, CACHE_SYNC(&c->sb));
327
328 bio_reset(bio);
329 bio->bi_bdev = ca->bdev;
330 bio->bi_end_io = write_super_endio;
331 bio->bi_private = ca;
332
333 closure_get(cl);
334 __write_super(&ca->sb, bio);
335 }
336
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800337 closure_return_with_destructor(cl, bcache_write_super_unlock);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700338}
339
340/* UUID io */
341
342static void uuid_endio(struct bio *bio, int error)
343{
344 struct closure *cl = bio->bi_private;
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800345 struct cache_set *c = container_of(cl, struct cache_set, uuid_write);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700346
347 cache_set_err_on(error, c, "accessing uuids");
348 bch_bbio_free(bio, c);
349 closure_put(cl);
350}
351
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800352static void uuid_io_unlock(struct closure *cl)
353{
354 struct cache_set *c = container_of(cl, struct cache_set, uuid_write);
355
356 up(&c->uuid_write_mutex);
357}
358
Kent Overstreetcafe5632013-03-23 16:11:31 -0700359static void uuid_io(struct cache_set *c, unsigned long rw,
360 struct bkey *k, struct closure *parent)
361{
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800362 struct closure *cl = &c->uuid_write;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700363 struct uuid_entry *u;
364 unsigned i;
Kent Overstreet85b14922013-05-14 20:33:16 -0700365 char buf[80];
Kent Overstreetcafe5632013-03-23 16:11:31 -0700366
367 BUG_ON(!parent);
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800368 down(&c->uuid_write_mutex);
369 closure_init(cl, parent);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700370
371 for (i = 0; i < KEY_PTRS(k); i++) {
372 struct bio *bio = bch_bbio_alloc(c);
373
374 bio->bi_rw = REQ_SYNC|REQ_META|rw;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700375 bio->bi_iter.bi_size = KEY_SIZE(k) << 9;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700376
377 bio->bi_end_io = uuid_endio;
378 bio->bi_private = cl;
Kent Overstreet169ef1c2013-03-28 12:50:55 -0600379 bch_bio_map(bio, c->uuids);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700380
381 bch_submit_bbio(bio, c, k, i);
382
383 if (!(rw & WRITE))
384 break;
385 }
386
Kent Overstreetdc9d98d2013-12-17 23:47:33 -0800387 bch_extent_to_text(buf, sizeof(buf), k);
Kent Overstreet85b14922013-05-14 20:33:16 -0700388 pr_debug("%s UUIDs at %s", rw & REQ_WRITE ? "wrote" : "read", buf);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700389
390 for (u = c->uuids; u < c->uuids + c->nr_uuids; u++)
Kent Overstreet169ef1c2013-03-28 12:50:55 -0600391 if (!bch_is_zero(u->uuid, 16))
Kent Overstreetcafe5632013-03-23 16:11:31 -0700392 pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u",
393 u - c->uuids, u->uuid, u->label,
394 u->first_reg, u->last_reg, u->invalidated);
395
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800396 closure_return_with_destructor(cl, uuid_io_unlock);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700397}
398
399static char *uuid_read(struct cache_set *c, struct jset *j, struct closure *cl)
400{
401 struct bkey *k = &j->uuid_bucket;
402
Kent Overstreet65d45232013-12-20 17:22:05 -0800403 if (__bch_btree_ptr_invalid(c, k))
Kent Overstreetcafe5632013-03-23 16:11:31 -0700404 return "bad uuid pointer";
405
406 bkey_copy(&c->uuid_bucket, k);
407 uuid_io(c, READ_SYNC, k, cl);
408
409 if (j->version < BCACHE_JSET_VERSION_UUIDv1) {
410 struct uuid_entry_v0 *u0 = (void *) c->uuids;
411 struct uuid_entry *u1 = (void *) c->uuids;
412 int i;
413
414 closure_sync(cl);
415
416 /*
417 * Since the new uuid entry is bigger than the old, we have to
418 * convert starting at the highest memory address and work down
419 * in order to do it in place
420 */
421
422 for (i = c->nr_uuids - 1;
423 i >= 0;
424 --i) {
425 memcpy(u1[i].uuid, u0[i].uuid, 16);
426 memcpy(u1[i].label, u0[i].label, 32);
427
428 u1[i].first_reg = u0[i].first_reg;
429 u1[i].last_reg = u0[i].last_reg;
430 u1[i].invalidated = u0[i].invalidated;
431
432 u1[i].flags = 0;
433 u1[i].sectors = 0;
434 }
435 }
436
437 return NULL;
438}
439
440static int __uuid_write(struct cache_set *c)
441{
442 BKEY_PADDED(key) k;
443 struct closure cl;
444 closure_init_stack(&cl);
445
446 lockdep_assert_held(&bch_register_lock);
447
Kent Overstreet78365412013-12-17 01:29:34 -0800448 if (bch_bucket_alloc_set(c, RESERVE_BTREE, &k.key, 1, true))
Kent Overstreetcafe5632013-03-23 16:11:31 -0700449 return 1;
450
451 SET_KEY_SIZE(&k.key, c->sb.bucket_size);
452 uuid_io(c, REQ_WRITE, &k.key, &cl);
453 closure_sync(&cl);
454
455 bkey_copy(&c->uuid_bucket, &k.key);
Kent Overstreet3a3b6a42013-07-24 16:46:42 -0700456 bkey_put(c, &k.key);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700457 return 0;
458}
459
460int bch_uuid_write(struct cache_set *c)
461{
462 int ret = __uuid_write(c);
463
464 if (!ret)
465 bch_journal_meta(c, NULL);
466
467 return ret;
468}
469
470static struct uuid_entry *uuid_find(struct cache_set *c, const char *uuid)
471{
472 struct uuid_entry *u;
473
474 for (u = c->uuids;
475 u < c->uuids + c->nr_uuids; u++)
476 if (!memcmp(u->uuid, uuid, 16))
477 return u;
478
479 return NULL;
480}
481
482static struct uuid_entry *uuid_find_empty(struct cache_set *c)
483{
484 static const char zero_uuid[16] = "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0";
485 return uuid_find(c, zero_uuid);
486}
487
488/*
489 * Bucket priorities/gens:
490 *
491 * For each bucket, we store on disk its
492 * 8 bit gen
493 * 16 bit priority
494 *
495 * See alloc.c for an explanation of the gen. The priority is used to implement
496 * lru (and in the future other) cache replacement policies; for most purposes
497 * it's just an opaque integer.
498 *
499 * The gens and the priorities don't have a whole lot to do with each other, and
500 * it's actually the gens that must be written out at specific times - it's no
501 * big deal if the priorities don't get written, if we lose them we just reuse
502 * buckets in suboptimal order.
503 *
504 * On disk they're stored in a packed array, and in as many buckets are required
505 * to fit them all. The buckets we use to store them form a list; the journal
506 * header points to the first bucket, the first bucket points to the second
507 * bucket, et cetera.
508 *
509 * This code is used by the allocation code; periodically (whenever it runs out
510 * of buckets to allocate from) the allocation code will invalidate some
511 * buckets, but it can't use those buckets until their new gens are safely on
512 * disk.
513 */
514
515static void prio_endio(struct bio *bio, int error)
516{
517 struct cache *ca = bio->bi_private;
518
519 cache_set_err_on(error, ca->set, "accessing priorities");
520 bch_bbio_free(bio, ca->set);
521 closure_put(&ca->prio);
522}
523
524static void prio_io(struct cache *ca, uint64_t bucket, unsigned long rw)
525{
526 struct closure *cl = &ca->prio;
527 struct bio *bio = bch_bbio_alloc(ca->set);
528
529 closure_init_stack(cl);
530
Kent Overstreet4f024f32013-10-11 15:44:27 -0700531 bio->bi_iter.bi_sector = bucket * ca->sb.bucket_size;
532 bio->bi_bdev = ca->bdev;
533 bio->bi_rw = REQ_SYNC|REQ_META|rw;
534 bio->bi_iter.bi_size = bucket_bytes(ca);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700535
536 bio->bi_end_io = prio_endio;
537 bio->bi_private = ca;
Kent Overstreet169ef1c2013-03-28 12:50:55 -0600538 bch_bio_map(bio, ca->disk_buckets);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700539
540 closure_bio_submit(bio, &ca->prio, ca);
541 closure_sync(cl);
542}
543
544#define buckets_free(c) "free %zu, free_inc %zu, unused %zu", \
545 fifo_used(&c->free), fifo_used(&c->free_inc), fifo_used(&c->unused)
546
547void bch_prio_write(struct cache *ca)
548{
549 int i;
550 struct bucket *b;
551 struct closure cl;
552
553 closure_init_stack(&cl);
554
555 lockdep_assert_held(&ca->set->bucket_lock);
556
557 for (b = ca->buckets;
558 b < ca->buckets + ca->sb.nbuckets; b++)
559 b->disk_gen = b->gen;
560
561 ca->disk_buckets->seq++;
562
563 atomic_long_add(ca->sb.bucket_size * prio_buckets(ca),
564 &ca->meta_sectors_written);
565
Kent Overstreet78365412013-12-17 01:29:34 -0800566 //pr_debug("free %zu, free_inc %zu, unused %zu", fifo_used(&ca->free),
567 // fifo_used(&ca->free_inc), fifo_used(&ca->unused));
Kent Overstreetcafe5632013-03-23 16:11:31 -0700568
569 for (i = prio_buckets(ca) - 1; i >= 0; --i) {
570 long bucket;
571 struct prio_set *p = ca->disk_buckets;
Kent Overstreetb1a67b02013-03-25 11:46:44 -0700572 struct bucket_disk *d = p->data;
573 struct bucket_disk *end = d + prios_per_bucket(ca);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700574
575 for (b = ca->buckets + i * prios_per_bucket(ca);
576 b < ca->buckets + ca->sb.nbuckets && d < end;
577 b++, d++) {
578 d->prio = cpu_to_le16(b->prio);
579 d->gen = b->gen;
580 }
581
582 p->next_bucket = ca->prio_buckets[i + 1];
Kent Overstreet81ab4192013-10-31 15:46:42 -0700583 p->magic = pset_magic(&ca->sb);
Kent Overstreet169ef1c2013-03-28 12:50:55 -0600584 p->csum = bch_crc64(&p->magic, bucket_bytes(ca) - 8);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700585
Kent Overstreet78365412013-12-17 01:29:34 -0800586 bucket = bch_bucket_alloc(ca, RESERVE_PRIO, true);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700587 BUG_ON(bucket == -1);
588
589 mutex_unlock(&ca->set->bucket_lock);
590 prio_io(ca, bucket, REQ_WRITE);
591 mutex_lock(&ca->set->bucket_lock);
592
593 ca->prio_buckets[i] = bucket;
594 atomic_dec_bug(&ca->buckets[bucket].pin);
595 }
596
597 mutex_unlock(&ca->set->bucket_lock);
598
599 bch_journal_meta(ca->set, &cl);
600 closure_sync(&cl);
601
602 mutex_lock(&ca->set->bucket_lock);
603
604 ca->need_save_prio = 0;
605
606 /*
607 * Don't want the old priorities to get garbage collected until after we
608 * finish writing the new ones, and they're journalled
609 */
610 for (i = 0; i < prio_buckets(ca); i++)
611 ca->prio_last_buckets[i] = ca->prio_buckets[i];
612}
613
614static void prio_read(struct cache *ca, uint64_t bucket)
615{
616 struct prio_set *p = ca->disk_buckets;
617 struct bucket_disk *d = p->data + prios_per_bucket(ca), *end = d;
618 struct bucket *b;
619 unsigned bucket_nr = 0;
620
621 for (b = ca->buckets;
622 b < ca->buckets + ca->sb.nbuckets;
623 b++, d++) {
624 if (d == end) {
625 ca->prio_buckets[bucket_nr] = bucket;
626 ca->prio_last_buckets[bucket_nr] = bucket;
627 bucket_nr++;
628
629 prio_io(ca, bucket, READ_SYNC);
630
Kent Overstreet169ef1c2013-03-28 12:50:55 -0600631 if (p->csum != bch_crc64(&p->magic, bucket_bytes(ca) - 8))
Kent Overstreetcafe5632013-03-23 16:11:31 -0700632 pr_warn("bad csum reading priorities");
633
Kent Overstreet81ab4192013-10-31 15:46:42 -0700634 if (p->magic != pset_magic(&ca->sb))
Kent Overstreetcafe5632013-03-23 16:11:31 -0700635 pr_warn("bad magic reading priorities");
636
637 bucket = p->next_bucket;
638 d = p->data;
639 }
640
641 b->prio = le16_to_cpu(d->prio);
642 b->gen = b->disk_gen = b->last_gc = b->gc_gen = d->gen;
643 }
644}
645
646/* Bcache device */
647
648static int open_dev(struct block_device *b, fmode_t mode)
649{
650 struct bcache_device *d = b->bd_disk->private_data;
Kent Overstreetc4d951d2013-08-21 17:49:09 -0700651 if (test_bit(BCACHE_DEV_CLOSING, &d->flags))
Kent Overstreetcafe5632013-03-23 16:11:31 -0700652 return -ENXIO;
653
654 closure_get(&d->cl);
655 return 0;
656}
657
Emil Goode867e1162013-05-09 22:39:26 +0200658static void release_dev(struct gendisk *b, fmode_t mode)
Kent Overstreetcafe5632013-03-23 16:11:31 -0700659{
660 struct bcache_device *d = b->private_data;
661 closure_put(&d->cl);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700662}
663
664static int ioctl_dev(struct block_device *b, fmode_t mode,
665 unsigned int cmd, unsigned long arg)
666{
667 struct bcache_device *d = b->bd_disk->private_data;
668 return d->ioctl(d, mode, cmd, arg);
669}
670
671static const struct block_device_operations bcache_ops = {
672 .open = open_dev,
673 .release = release_dev,
674 .ioctl = ioctl_dev,
675 .owner = THIS_MODULE,
676};
677
678void bcache_device_stop(struct bcache_device *d)
679{
Kent Overstreetc4d951d2013-08-21 17:49:09 -0700680 if (!test_and_set_bit(BCACHE_DEV_CLOSING, &d->flags))
Kent Overstreetcafe5632013-03-23 16:11:31 -0700681 closure_queue(&d->cl);
682}
683
Kent Overstreetee668502013-02-01 07:29:41 -0800684static void bcache_device_unlink(struct bcache_device *d)
685{
Kent Overstreetc4d951d2013-08-21 17:49:09 -0700686 lockdep_assert_held(&bch_register_lock);
Kent Overstreetee668502013-02-01 07:29:41 -0800687
Kent Overstreetc4d951d2013-08-21 17:49:09 -0700688 if (d->c && !test_and_set_bit(BCACHE_DEV_UNLINK_DONE, &d->flags)) {
689 unsigned i;
690 struct cache *ca;
Kent Overstreetee668502013-02-01 07:29:41 -0800691
Kent Overstreetc4d951d2013-08-21 17:49:09 -0700692 sysfs_remove_link(&d->c->kobj, d->name);
693 sysfs_remove_link(&d->kobj, "cache");
694
695 for_each_cache(ca, d->c, i)
696 bd_unlink_disk_holder(ca->bdev, d->disk);
697 }
Kent Overstreetee668502013-02-01 07:29:41 -0800698}
699
700static void bcache_device_link(struct bcache_device *d, struct cache_set *c,
701 const char *name)
702{
703 unsigned i;
704 struct cache *ca;
705
706 for_each_cache(ca, d->c, i)
707 bd_link_disk_holder(ca->bdev, d->disk);
708
709 snprintf(d->name, BCACHEDEVNAME_SIZE,
710 "%s%u", name, d->id);
711
712 WARN(sysfs_create_link(&d->kobj, &c->kobj, "cache") ||
713 sysfs_create_link(&c->kobj, &d->kobj, d->name),
714 "Couldn't create device <-> cache set symlinks");
715}
716
Kent Overstreetcafe5632013-03-23 16:11:31 -0700717static void bcache_device_detach(struct bcache_device *d)
718{
719 lockdep_assert_held(&bch_register_lock);
720
Kent Overstreetc4d951d2013-08-21 17:49:09 -0700721 if (test_bit(BCACHE_DEV_DETACHING, &d->flags)) {
Kent Overstreetcafe5632013-03-23 16:11:31 -0700722 struct uuid_entry *u = d->c->uuids + d->id;
723
724 SET_UUID_FLASH_ONLY(u, 0);
725 memcpy(u->uuid, invalid_uuid, 16);
726 u->invalidated = cpu_to_le32(get_seconds());
727 bch_uuid_write(d->c);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700728 }
729
Kent Overstreetc4d951d2013-08-21 17:49:09 -0700730 bcache_device_unlink(d);
Kent Overstreetee668502013-02-01 07:29:41 -0800731
Kent Overstreetcafe5632013-03-23 16:11:31 -0700732 d->c->devices[d->id] = NULL;
733 closure_put(&d->c->caching);
734 d->c = NULL;
735}
736
737static void bcache_device_attach(struct bcache_device *d, struct cache_set *c,
738 unsigned id)
739{
740 BUG_ON(test_bit(CACHE_SET_STOPPING, &c->flags));
741
742 d->id = id;
743 d->c = c;
744 c->devices[id] = d;
745
746 closure_get(&c->caching);
747}
748
Kent Overstreetcafe5632013-03-23 16:11:31 -0700749static void bcache_device_free(struct bcache_device *d)
750{
751 lockdep_assert_held(&bch_register_lock);
752
753 pr_info("%s stopped", d->disk->disk_name);
754
755 if (d->c)
756 bcache_device_detach(d);
Kent Overstreetf59fce82013-05-15 00:11:26 -0700757 if (d->disk && d->disk->flags & GENHD_FL_UP)
Kent Overstreetcafe5632013-03-23 16:11:31 -0700758 del_gendisk(d->disk);
759 if (d->disk && d->disk->queue)
760 blk_cleanup_queue(d->disk->queue);
Kent Overstreet28935ab2013-07-31 01:12:02 -0700761 if (d->disk) {
762 ida_simple_remove(&bcache_minor, d->disk->first_minor);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700763 put_disk(d->disk);
Kent Overstreet28935ab2013-07-31 01:12:02 -0700764 }
Kent Overstreetcafe5632013-03-23 16:11:31 -0700765
766 bio_split_pool_free(&d->bio_split_hook);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700767 if (d->bio_split)
768 bioset_free(d->bio_split);
Kent Overstreet48a915a2013-10-31 15:43:22 -0700769 if (is_vmalloc_addr(d->full_dirty_stripes))
770 vfree(d->full_dirty_stripes);
771 else
772 kfree(d->full_dirty_stripes);
Kent Overstreet279afba2013-06-05 06:21:07 -0700773 if (is_vmalloc_addr(d->stripe_sectors_dirty))
774 vfree(d->stripe_sectors_dirty);
775 else
776 kfree(d->stripe_sectors_dirty);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700777
778 closure_debug_destroy(&d->cl);
779}
780
Kent Overstreet279afba2013-06-05 06:21:07 -0700781static int bcache_device_init(struct bcache_device *d, unsigned block_size,
782 sector_t sectors)
Kent Overstreetcafe5632013-03-23 16:11:31 -0700783{
784 struct request_queue *q;
Kent Overstreet279afba2013-06-05 06:21:07 -0700785 size_t n;
Kent Overstreet28935ab2013-07-31 01:12:02 -0700786 int minor;
Kent Overstreet279afba2013-06-05 06:21:07 -0700787
Kent Overstreet2d679fc2013-08-17 02:13:15 -0700788 if (!d->stripe_size)
789 d->stripe_size = 1 << 31;
Kent Overstreet279afba2013-06-05 06:21:07 -0700790
Kent Overstreet2d679fc2013-08-17 02:13:15 -0700791 d->nr_stripes = DIV_ROUND_UP_ULL(sectors, d->stripe_size);
Kent Overstreet279afba2013-06-05 06:21:07 -0700792
Kent Overstreet48a915a2013-10-31 15:43:22 -0700793 if (!d->nr_stripes ||
794 d->nr_stripes > INT_MAX ||
795 d->nr_stripes > SIZE_MAX / sizeof(atomic_t)) {
796 pr_err("nr_stripes too large");
Kent Overstreet279afba2013-06-05 06:21:07 -0700797 return -ENOMEM;
Kent Overstreet48a915a2013-10-31 15:43:22 -0700798 }
Kent Overstreet279afba2013-06-05 06:21:07 -0700799
800 n = d->nr_stripes * sizeof(atomic_t);
801 d->stripe_sectors_dirty = n < PAGE_SIZE << 6
802 ? kzalloc(n, GFP_KERNEL)
803 : vzalloc(n);
804 if (!d->stripe_sectors_dirty)
805 return -ENOMEM;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700806
Kent Overstreet48a915a2013-10-31 15:43:22 -0700807 n = BITS_TO_LONGS(d->nr_stripes) * sizeof(unsigned long);
808 d->full_dirty_stripes = n < PAGE_SIZE << 6
809 ? kzalloc(n, GFP_KERNEL)
810 : vzalloc(n);
811 if (!d->full_dirty_stripes)
812 return -ENOMEM;
813
Kent Overstreet28935ab2013-07-31 01:12:02 -0700814 minor = ida_simple_get(&bcache_minor, 0, MINORMASK + 1, GFP_KERNEL);
815 if (minor < 0)
816 return minor;
817
Kent Overstreetcafe5632013-03-23 16:11:31 -0700818 if (!(d->bio_split = bioset_create(4, offsetof(struct bbio, bio))) ||
Kent Overstreetf59fce82013-05-15 00:11:26 -0700819 bio_split_pool_init(&d->bio_split_hook) ||
Kent Overstreet28935ab2013-07-31 01:12:02 -0700820 !(d->disk = alloc_disk(1))) {
821 ida_simple_remove(&bcache_minor, minor);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700822 return -ENOMEM;
Kent Overstreet28935ab2013-07-31 01:12:02 -0700823 }
Kent Overstreetcafe5632013-03-23 16:11:31 -0700824
Kent Overstreet279afba2013-06-05 06:21:07 -0700825 set_capacity(d->disk, sectors);
Kent Overstreet28935ab2013-07-31 01:12:02 -0700826 snprintf(d->disk->disk_name, DISK_NAME_LEN, "bcache%i", minor);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700827
828 d->disk->major = bcache_major;
Kent Overstreet28935ab2013-07-31 01:12:02 -0700829 d->disk->first_minor = minor;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700830 d->disk->fops = &bcache_ops;
831 d->disk->private_data = d;
832
Kent Overstreet28935ab2013-07-31 01:12:02 -0700833 q = blk_alloc_queue(GFP_KERNEL);
834 if (!q)
835 return -ENOMEM;
836
Kent Overstreetcafe5632013-03-23 16:11:31 -0700837 blk_queue_make_request(q, NULL);
838 d->disk->queue = q;
839 q->queuedata = d;
840 q->backing_dev_info.congested_data = d;
841 q->limits.max_hw_sectors = UINT_MAX;
842 q->limits.max_sectors = UINT_MAX;
843 q->limits.max_segment_size = UINT_MAX;
844 q->limits.max_segments = BIO_MAX_PAGES;
845 q->limits.max_discard_sectors = UINT_MAX;
Kent Overstreet90db6912014-02-10 17:26:40 -0800846 q->limits.discard_granularity = 512;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700847 q->limits.io_min = block_size;
848 q->limits.logical_block_size = block_size;
849 q->limits.physical_block_size = block_size;
850 set_bit(QUEUE_FLAG_NONROT, &d->disk->queue->queue_flags);
851 set_bit(QUEUE_FLAG_DISCARD, &d->disk->queue->queue_flags);
852
Kent Overstreet54d12f22013-07-10 18:44:40 -0700853 blk_queue_flush(q, REQ_FLUSH|REQ_FUA);
854
Kent Overstreetcafe5632013-03-23 16:11:31 -0700855 return 0;
856}
857
858/* Cached device */
859
860static void calc_cached_dev_sectors(struct cache_set *c)
861{
862 uint64_t sectors = 0;
863 struct cached_dev *dc;
864
865 list_for_each_entry(dc, &c->cached_devs, list)
866 sectors += bdev_sectors(dc->bdev);
867
868 c->cached_dev_sectors = sectors;
869}
870
871void bch_cached_dev_run(struct cached_dev *dc)
872{
873 struct bcache_device *d = &dc->disk;
Gabriel de Perthuisab9e1402013-06-09 00:54:48 +0200874 char buf[SB_LABEL_SIZE + 1];
Gabriel de Perthuisa25c32b2013-06-07 23:27:01 +0200875 char *env[] = {
876 "DRIVER=bcache",
877 kasprintf(GFP_KERNEL, "CACHED_UUID=%pU", dc->sb.uuid),
Gabriel de Perthuisab9e1402013-06-09 00:54:48 +0200878 NULL,
879 NULL,
Gabriel de Perthuisa25c32b2013-06-07 23:27:01 +0200880 };
Kent Overstreetcafe5632013-03-23 16:11:31 -0700881
Gabriel de Perthuisab9e1402013-06-09 00:54:48 +0200882 memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
883 buf[SB_LABEL_SIZE] = '\0';
884 env[2] = kasprintf(GFP_KERNEL, "CACHED_LABEL=%s", buf);
885
Kent Overstreetcafe5632013-03-23 16:11:31 -0700886 if (atomic_xchg(&dc->running, 1))
887 return;
888
889 if (!d->c &&
890 BDEV_STATE(&dc->sb) != BDEV_STATE_NONE) {
891 struct closure cl;
892 closure_init_stack(&cl);
893
894 SET_BDEV_STATE(&dc->sb, BDEV_STATE_STALE);
895 bch_write_bdev_super(dc, &cl);
896 closure_sync(&cl);
897 }
898
899 add_disk(d->disk);
Kent Overstreetee668502013-02-01 07:29:41 -0800900 bd_link_disk_holder(dc->bdev, dc->disk.disk);
Gabriel de Perthuisa25c32b2013-06-07 23:27:01 +0200901 /* won't show up in the uevent file, use udevadm monitor -e instead
902 * only class / kset properties are persistent */
Kent Overstreetcafe5632013-03-23 16:11:31 -0700903 kobject_uevent_env(&disk_to_dev(d->disk)->kobj, KOBJ_CHANGE, env);
Gabriel de Perthuisa25c32b2013-06-07 23:27:01 +0200904 kfree(env[1]);
Gabriel de Perthuisab9e1402013-06-09 00:54:48 +0200905 kfree(env[2]);
Gabriel de Perthuisa25c32b2013-06-07 23:27:01 +0200906
Kent Overstreetcafe5632013-03-23 16:11:31 -0700907 if (sysfs_create_link(&d->kobj, &disk_to_dev(d->disk)->kobj, "dev") ||
908 sysfs_create_link(&disk_to_dev(d->disk)->kobj, &d->kobj, "bcache"))
909 pr_debug("error creating sysfs link");
910}
911
912static void cached_dev_detach_finish(struct work_struct *w)
913{
914 struct cached_dev *dc = container_of(w, struct cached_dev, detach);
915 char buf[BDEVNAME_SIZE];
916 struct closure cl;
917 closure_init_stack(&cl);
918
Kent Overstreetc4d951d2013-08-21 17:49:09 -0700919 BUG_ON(!test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags));
Kent Overstreetcafe5632013-03-23 16:11:31 -0700920 BUG_ON(atomic_read(&dc->count));
921
Kent Overstreetcafe5632013-03-23 16:11:31 -0700922 mutex_lock(&bch_register_lock);
923
924 memset(&dc->sb.set_uuid, 0, 16);
925 SET_BDEV_STATE(&dc->sb, BDEV_STATE_NONE);
926
927 bch_write_bdev_super(dc, &cl);
928 closure_sync(&cl);
929
930 bcache_device_detach(&dc->disk);
931 list_move(&dc->list, &uncached_devices);
932
Kent Overstreetc4d951d2013-08-21 17:49:09 -0700933 clear_bit(BCACHE_DEV_DETACHING, &dc->disk.flags);
934
Kent Overstreetcafe5632013-03-23 16:11:31 -0700935 mutex_unlock(&bch_register_lock);
936
937 pr_info("Caching disabled for %s", bdevname(dc->bdev, buf));
938
939 /* Drop ref we took in cached_dev_detach() */
940 closure_put(&dc->disk.cl);
941}
942
943void bch_cached_dev_detach(struct cached_dev *dc)
944{
945 lockdep_assert_held(&bch_register_lock);
946
Kent Overstreetc4d951d2013-08-21 17:49:09 -0700947 if (test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags))
Kent Overstreetcafe5632013-03-23 16:11:31 -0700948 return;
949
Kent Overstreetc4d951d2013-08-21 17:49:09 -0700950 if (test_and_set_bit(BCACHE_DEV_DETACHING, &dc->disk.flags))
Kent Overstreetcafe5632013-03-23 16:11:31 -0700951 return;
952
953 /*
954 * Block the device from being closed and freed until we're finished
955 * detaching
956 */
957 closure_get(&dc->disk.cl);
958
959 bch_writeback_queue(dc);
960 cached_dev_put(dc);
961}
962
963int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c)
964{
965 uint32_t rtime = cpu_to_le32(get_seconds());
966 struct uuid_entry *u;
967 char buf[BDEVNAME_SIZE];
968
969 bdevname(dc->bdev, buf);
970
971 if (memcmp(dc->sb.set_uuid, c->sb.set_uuid, 16))
972 return -ENOENT;
973
974 if (dc->disk.c) {
975 pr_err("Can't attach %s: already attached", buf);
976 return -EINVAL;
977 }
978
979 if (test_bit(CACHE_SET_STOPPING, &c->flags)) {
980 pr_err("Can't attach %s: shutting down", buf);
981 return -EINVAL;
982 }
983
984 if (dc->sb.block_size < c->sb.block_size) {
985 /* Will die */
Kent Overstreetb1a67b02013-03-25 11:46:44 -0700986 pr_err("Couldn't attach %s: block size less than set's block size",
987 buf);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700988 return -EINVAL;
989 }
990
991 u = uuid_find(c, dc->sb.uuid);
992
993 if (u &&
994 (BDEV_STATE(&dc->sb) == BDEV_STATE_STALE ||
995 BDEV_STATE(&dc->sb) == BDEV_STATE_NONE)) {
996 memcpy(u->uuid, invalid_uuid, 16);
997 u->invalidated = cpu_to_le32(get_seconds());
998 u = NULL;
999 }
1000
1001 if (!u) {
1002 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
1003 pr_err("Couldn't find uuid for %s in set", buf);
1004 return -ENOENT;
1005 }
1006
1007 u = uuid_find_empty(c);
1008 if (!u) {
1009 pr_err("Not caching %s, no room for UUID", buf);
1010 return -EINVAL;
1011 }
1012 }
1013
1014 /* Deadlocks since we're called via sysfs...
1015 sysfs_remove_file(&dc->kobj, &sysfs_attach);
1016 */
1017
Kent Overstreet169ef1c2013-03-28 12:50:55 -06001018 if (bch_is_zero(u->uuid, 16)) {
Kent Overstreetcafe5632013-03-23 16:11:31 -07001019 struct closure cl;
1020 closure_init_stack(&cl);
1021
1022 memcpy(u->uuid, dc->sb.uuid, 16);
1023 memcpy(u->label, dc->sb.label, SB_LABEL_SIZE);
1024 u->first_reg = u->last_reg = rtime;
1025 bch_uuid_write(c);
1026
1027 memcpy(dc->sb.set_uuid, c->sb.set_uuid, 16);
1028 SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN);
1029
1030 bch_write_bdev_super(dc, &cl);
1031 closure_sync(&cl);
1032 } else {
1033 u->last_reg = rtime;
1034 bch_uuid_write(c);
1035 }
1036
1037 bcache_device_attach(&dc->disk, c, u - c->uuids);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001038 list_move(&dc->list, &c->cached_devs);
1039 calc_cached_dev_sectors(c);
1040
1041 smp_wmb();
1042 /*
1043 * dc->c must be set before dc->count != 0 - paired with the mb in
1044 * cached_dev_get()
1045 */
1046 atomic_set(&dc->count, 1);
1047
1048 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
Kent Overstreet444fc0b2013-05-11 17:07:26 -07001049 bch_sectors_dirty_init(dc);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001050 atomic_set(&dc->has_dirty, 1);
1051 atomic_inc(&dc->count);
1052 bch_writeback_queue(dc);
1053 }
1054
1055 bch_cached_dev_run(dc);
Kent Overstreetee668502013-02-01 07:29:41 -08001056 bcache_device_link(&dc->disk, c, "bdev");
Kent Overstreetcafe5632013-03-23 16:11:31 -07001057
1058 pr_info("Caching %s as %s on set %pU",
1059 bdevname(dc->bdev, buf), dc->disk.disk->disk_name,
1060 dc->disk.c->sb.set_uuid);
1061 return 0;
1062}
1063
1064void bch_cached_dev_release(struct kobject *kobj)
1065{
1066 struct cached_dev *dc = container_of(kobj, struct cached_dev,
1067 disk.kobj);
1068 kfree(dc);
1069 module_put(THIS_MODULE);
1070}
1071
1072static void cached_dev_free(struct closure *cl)
1073{
1074 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1075
1076 cancel_delayed_work_sync(&dc->writeback_rate_update);
Kent Overstreet5e6926d2013-07-24 17:50:06 -07001077 kthread_stop(dc->writeback_thread);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001078
1079 mutex_lock(&bch_register_lock);
1080
Kent Overstreetf59fce82013-05-15 00:11:26 -07001081 if (atomic_read(&dc->running))
1082 bd_unlink_disk_holder(dc->bdev, dc->disk.disk);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001083 bcache_device_free(&dc->disk);
1084 list_del(&dc->list);
1085
1086 mutex_unlock(&bch_register_lock);
1087
1088 if (!IS_ERR_OR_NULL(dc->bdev)) {
Kent Overstreetf59fce82013-05-15 00:11:26 -07001089 if (dc->bdev->bd_disk)
1090 blk_sync_queue(bdev_get_queue(dc->bdev));
1091
Kent Overstreetcafe5632013-03-23 16:11:31 -07001092 blkdev_put(dc->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1093 }
1094
1095 wake_up(&unregister_wait);
1096
1097 kobject_put(&dc->disk.kobj);
1098}
1099
1100static void cached_dev_flush(struct closure *cl)
1101{
1102 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1103 struct bcache_device *d = &dc->disk;
1104
Kent Overstreetc9502ea2013-07-10 21:25:02 -07001105 mutex_lock(&bch_register_lock);
Kent Overstreetc4d951d2013-08-21 17:49:09 -07001106 bcache_device_unlink(d);
Kent Overstreetc9502ea2013-07-10 21:25:02 -07001107 mutex_unlock(&bch_register_lock);
1108
Kent Overstreetcafe5632013-03-23 16:11:31 -07001109 bch_cache_accounting_destroy(&dc->accounting);
1110 kobject_del(&d->kobj);
1111
1112 continue_at(cl, cached_dev_free, system_wq);
1113}
1114
1115static int cached_dev_init(struct cached_dev *dc, unsigned block_size)
1116{
Kent Overstreetf59fce82013-05-15 00:11:26 -07001117 int ret;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001118 struct io *io;
Kent Overstreetf59fce82013-05-15 00:11:26 -07001119 struct request_queue *q = bdev_get_queue(dc->bdev);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001120
1121 __module_get(THIS_MODULE);
1122 INIT_LIST_HEAD(&dc->list);
Kent Overstreetf59fce82013-05-15 00:11:26 -07001123 closure_init(&dc->disk.cl, NULL);
1124 set_closure_fn(&dc->disk.cl, cached_dev_flush, system_wq);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001125 kobject_init(&dc->disk.kobj, &bch_cached_dev_ktype);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001126 INIT_WORK(&dc->detach, cached_dev_detach_finish);
Kent Overstreetcb7a5832013-12-16 15:27:25 -08001127 sema_init(&dc->sb_write_mutex, 1);
Kent Overstreetf59fce82013-05-15 00:11:26 -07001128 INIT_LIST_HEAD(&dc->io_lru);
1129 spin_lock_init(&dc->io_lock);
1130 bch_cache_accounting_init(&dc->accounting, &dc->disk.cl);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001131
Kent Overstreetcafe5632013-03-23 16:11:31 -07001132 dc->sequential_cutoff = 4 << 20;
1133
Kent Overstreetcafe5632013-03-23 16:11:31 -07001134 for (io = dc->io; io < dc->io + RECENT_IO; io++) {
1135 list_add(&io->lru, &dc->io_lru);
1136 hlist_add_head(&io->hash, dc->io_hash + RECENT_IO);
1137 }
1138
Kent Overstreetc78afc62013-07-11 22:39:53 -07001139 dc->disk.stripe_size = q->limits.io_opt >> 9;
1140
1141 if (dc->disk.stripe_size)
1142 dc->partial_stripes_expensive =
1143 q->limits.raid_partial_stripes_expensive;
1144
Kent Overstreet279afba2013-06-05 06:21:07 -07001145 ret = bcache_device_init(&dc->disk, block_size,
1146 dc->bdev->bd_part->nr_sects - dc->sb.data_offset);
Kent Overstreetf59fce82013-05-15 00:11:26 -07001147 if (ret)
1148 return ret;
1149
1150 set_capacity(dc->disk.disk,
1151 dc->bdev->bd_part->nr_sects - dc->sb.data_offset);
1152
1153 dc->disk.disk->queue->backing_dev_info.ra_pages =
1154 max(dc->disk.disk->queue->backing_dev_info.ra_pages,
1155 q->backing_dev_info.ra_pages);
1156
1157 bch_cached_dev_request_init(dc);
1158 bch_cached_dev_writeback_init(dc);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001159 return 0;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001160}
1161
1162/* Cached device - bcache superblock */
1163
Kent Overstreetf59fce82013-05-15 00:11:26 -07001164static void register_bdev(struct cache_sb *sb, struct page *sb_page,
Kent Overstreetcafe5632013-03-23 16:11:31 -07001165 struct block_device *bdev,
1166 struct cached_dev *dc)
1167{
1168 char name[BDEVNAME_SIZE];
1169 const char *err = "cannot allocate memory";
Kent Overstreetcafe5632013-03-23 16:11:31 -07001170 struct cache_set *c;
1171
Kent Overstreetcafe5632013-03-23 16:11:31 -07001172 memcpy(&dc->sb, sb, sizeof(struct cache_sb));
Kent Overstreetcafe5632013-03-23 16:11:31 -07001173 dc->bdev = bdev;
1174 dc->bdev->bd_holder = dc;
1175
Kent Overstreetf59fce82013-05-15 00:11:26 -07001176 bio_init(&dc->sb_bio);
1177 dc->sb_bio.bi_max_vecs = 1;
1178 dc->sb_bio.bi_io_vec = dc->sb_bio.bi_inline_vecs;
1179 dc->sb_bio.bi_io_vec[0].bv_page = sb_page;
1180 get_page(sb_page);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001181
Kent Overstreetf59fce82013-05-15 00:11:26 -07001182 if (cached_dev_init(dc, sb->block_size << 9))
1183 goto err;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001184
1185 err = "error creating kobject";
1186 if (kobject_add(&dc->disk.kobj, &part_to_dev(bdev->bd_part)->kobj,
1187 "bcache"))
1188 goto err;
1189 if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj))
1190 goto err;
1191
Kent Overstreetf59fce82013-05-15 00:11:26 -07001192 pr_info("registered backing device %s", bdevname(bdev, name));
1193
Kent Overstreetcafe5632013-03-23 16:11:31 -07001194 list_add(&dc->list, &uncached_devices);
1195 list_for_each_entry(c, &bch_cache_sets, list)
1196 bch_cached_dev_attach(dc, c);
1197
1198 if (BDEV_STATE(&dc->sb) == BDEV_STATE_NONE ||
1199 BDEV_STATE(&dc->sb) == BDEV_STATE_STALE)
1200 bch_cached_dev_run(dc);
1201
Kent Overstreetf59fce82013-05-15 00:11:26 -07001202 return;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001203err:
Kent Overstreetcafe5632013-03-23 16:11:31 -07001204 pr_notice("error opening %s: %s", bdevname(bdev, name), err);
Kent Overstreetf59fce82013-05-15 00:11:26 -07001205 bcache_device_stop(&dc->disk);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001206}
1207
1208/* Flash only volumes */
1209
1210void bch_flash_dev_release(struct kobject *kobj)
1211{
1212 struct bcache_device *d = container_of(kobj, struct bcache_device,
1213 kobj);
1214 kfree(d);
1215}
1216
1217static void flash_dev_free(struct closure *cl)
1218{
1219 struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1220 bcache_device_free(d);
1221 kobject_put(&d->kobj);
1222}
1223
1224static void flash_dev_flush(struct closure *cl)
1225{
1226 struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1227
Kent Overstreetee668502013-02-01 07:29:41 -08001228 bcache_device_unlink(d);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001229 kobject_del(&d->kobj);
1230 continue_at(cl, flash_dev_free, system_wq);
1231}
1232
1233static int flash_dev_run(struct cache_set *c, struct uuid_entry *u)
1234{
1235 struct bcache_device *d = kzalloc(sizeof(struct bcache_device),
1236 GFP_KERNEL);
1237 if (!d)
1238 return -ENOMEM;
1239
1240 closure_init(&d->cl, NULL);
1241 set_closure_fn(&d->cl, flash_dev_flush, system_wq);
1242
1243 kobject_init(&d->kobj, &bch_flash_dev_ktype);
1244
Kent Overstreet279afba2013-06-05 06:21:07 -07001245 if (bcache_device_init(d, block_bytes(c), u->sectors))
Kent Overstreetcafe5632013-03-23 16:11:31 -07001246 goto err;
1247
1248 bcache_device_attach(d, c, u - c->uuids);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001249 bch_flash_dev_request_init(d);
1250 add_disk(d->disk);
1251
1252 if (kobject_add(&d->kobj, &disk_to_dev(d->disk)->kobj, "bcache"))
1253 goto err;
1254
1255 bcache_device_link(d, c, "volume");
1256
1257 return 0;
1258err:
1259 kobject_put(&d->kobj);
1260 return -ENOMEM;
1261}
1262
1263static int flash_devs_run(struct cache_set *c)
1264{
1265 int ret = 0;
1266 struct uuid_entry *u;
1267
1268 for (u = c->uuids;
1269 u < c->uuids + c->nr_uuids && !ret;
1270 u++)
1271 if (UUID_FLASH_ONLY(u))
1272 ret = flash_dev_run(c, u);
1273
1274 return ret;
1275}
1276
1277int bch_flash_dev_create(struct cache_set *c, uint64_t size)
1278{
1279 struct uuid_entry *u;
1280
1281 if (test_bit(CACHE_SET_STOPPING, &c->flags))
1282 return -EINTR;
1283
1284 u = uuid_find_empty(c);
1285 if (!u) {
1286 pr_err("Can't create volume, no room for UUID");
1287 return -EINVAL;
1288 }
1289
1290 get_random_bytes(u->uuid, 16);
1291 memset(u->label, 0, 32);
1292 u->first_reg = u->last_reg = cpu_to_le32(get_seconds());
1293
1294 SET_UUID_FLASH_ONLY(u, 1);
1295 u->sectors = size >> 9;
1296
1297 bch_uuid_write(c);
1298
1299 return flash_dev_run(c, u);
1300}
1301
1302/* Cache set */
1303
1304__printf(2, 3)
1305bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...)
1306{
1307 va_list args;
1308
Kent Overstreet77c320e2013-07-11 19:42:51 -07001309 if (c->on_error != ON_ERROR_PANIC &&
1310 test_bit(CACHE_SET_STOPPING, &c->flags))
Kent Overstreetcafe5632013-03-23 16:11:31 -07001311 return false;
1312
1313 /* XXX: we can be called from atomic context
1314 acquire_console_sem();
1315 */
1316
1317 printk(KERN_ERR "bcache: error on %pU: ", c->sb.set_uuid);
1318
1319 va_start(args, fmt);
1320 vprintk(fmt, args);
1321 va_end(args);
1322
1323 printk(", disabling caching\n");
1324
Kent Overstreet77c320e2013-07-11 19:42:51 -07001325 if (c->on_error == ON_ERROR_PANIC)
1326 panic("panic forced after error\n");
1327
Kent Overstreetcafe5632013-03-23 16:11:31 -07001328 bch_cache_set_unregister(c);
1329 return true;
1330}
1331
1332void bch_cache_set_release(struct kobject *kobj)
1333{
1334 struct cache_set *c = container_of(kobj, struct cache_set, kobj);
1335 kfree(c);
1336 module_put(THIS_MODULE);
1337}
1338
1339static void cache_set_free(struct closure *cl)
1340{
1341 struct cache_set *c = container_of(cl, struct cache_set, cl);
1342 struct cache *ca;
1343 unsigned i;
1344
1345 if (!IS_ERR_OR_NULL(c->debug))
1346 debugfs_remove(c->debug);
1347
1348 bch_open_buckets_free(c);
1349 bch_btree_cache_free(c);
1350 bch_journal_free(c);
1351
1352 for_each_cache(ca, c, i)
1353 if (ca)
1354 kobject_put(&ca->kobj);
1355
Kent Overstreet67539e82013-09-10 22:53:34 -07001356 bch_bset_sort_state_free(&c->sort);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001357 free_pages((unsigned long) c->uuids, ilog2(bucket_pages(c)));
Kent Overstreetcafe5632013-03-23 16:11:31 -07001358
Kent Overstreetcafe5632013-03-23 16:11:31 -07001359 if (c->bio_split)
1360 bioset_free(c->bio_split);
Kent Overstreet57943512013-04-25 13:58:35 -07001361 if (c->fill_iter)
1362 mempool_destroy(c->fill_iter);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001363 if (c->bio_meta)
1364 mempool_destroy(c->bio_meta);
1365 if (c->search)
1366 mempool_destroy(c->search);
1367 kfree(c->devices);
1368
1369 mutex_lock(&bch_register_lock);
1370 list_del(&c->list);
1371 mutex_unlock(&bch_register_lock);
1372
1373 pr_info("Cache set %pU unregistered", c->sb.set_uuid);
1374 wake_up(&unregister_wait);
1375
1376 closure_debug_destroy(&c->cl);
1377 kobject_put(&c->kobj);
1378}
1379
1380static void cache_set_flush(struct closure *cl)
1381{
1382 struct cache_set *c = container_of(cl, struct cache_set, caching);
Kent Overstreet79826c32013-07-10 18:31:58 -07001383 struct cache *ca;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001384 struct btree *b;
Kent Overstreet79826c32013-07-10 18:31:58 -07001385 unsigned i;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001386
1387 bch_cache_accounting_destroy(&c->accounting);
1388
1389 kobject_put(&c->internal);
1390 kobject_del(&c->kobj);
1391
Kent Overstreet72a44512013-10-24 17:19:26 -07001392 if (c->gc_thread)
1393 kthread_stop(c->gc_thread);
1394
Kent Overstreetcafe5632013-03-23 16:11:31 -07001395 if (!IS_ERR_OR_NULL(c->root))
1396 list_add(&c->root->list, &c->btree_cache);
1397
1398 /* Should skip this if we're unregistering because of an error */
1399 list_for_each_entry(b, &c->btree_cache, list)
1400 if (btree_node_dirty(b))
Kent Overstreet57943512013-04-25 13:58:35 -07001401 bch_btree_node_write(b, NULL);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001402
Kent Overstreet79826c32013-07-10 18:31:58 -07001403 for_each_cache(ca, c, i)
1404 if (ca->alloc_thread)
1405 kthread_stop(ca->alloc_thread);
1406
Kent Overstreetdabb4432014-02-19 19:48:26 -08001407 cancel_delayed_work_sync(&c->journal.work);
1408 /* flush last journal entry if needed */
1409 c->journal.work.work.func(&c->journal.work.work);
1410
Kent Overstreetcafe5632013-03-23 16:11:31 -07001411 closure_return(cl);
1412}
1413
1414static void __cache_set_unregister(struct closure *cl)
1415{
1416 struct cache_set *c = container_of(cl, struct cache_set, caching);
Kent Overstreet5caa52a2013-07-10 21:03:25 -07001417 struct cached_dev *dc;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001418 size_t i;
1419
1420 mutex_lock(&bch_register_lock);
1421
Kent Overstreetcafe5632013-03-23 16:11:31 -07001422 for (i = 0; i < c->nr_uuids; i++)
Kent Overstreet5caa52a2013-07-10 21:03:25 -07001423 if (c->devices[i]) {
1424 if (!UUID_FLASH_ONLY(&c->uuids[i]) &&
1425 test_bit(CACHE_SET_UNREGISTERING, &c->flags)) {
1426 dc = container_of(c->devices[i],
1427 struct cached_dev, disk);
1428 bch_cached_dev_detach(dc);
1429 } else {
1430 bcache_device_stop(c->devices[i]);
1431 }
1432 }
Kent Overstreetcafe5632013-03-23 16:11:31 -07001433
1434 mutex_unlock(&bch_register_lock);
1435
1436 continue_at(cl, cache_set_flush, system_wq);
1437}
1438
1439void bch_cache_set_stop(struct cache_set *c)
1440{
1441 if (!test_and_set_bit(CACHE_SET_STOPPING, &c->flags))
1442 closure_queue(&c->caching);
1443}
1444
1445void bch_cache_set_unregister(struct cache_set *c)
1446{
1447 set_bit(CACHE_SET_UNREGISTERING, &c->flags);
1448 bch_cache_set_stop(c);
1449}
1450
1451#define alloc_bucket_pages(gfp, c) \
1452 ((void *) __get_free_pages(__GFP_ZERO|gfp, ilog2(bucket_pages(c))))
1453
1454struct cache_set *bch_cache_set_alloc(struct cache_sb *sb)
1455{
1456 int iter_size;
1457 struct cache_set *c = kzalloc(sizeof(struct cache_set), GFP_KERNEL);
1458 if (!c)
1459 return NULL;
1460
1461 __module_get(THIS_MODULE);
1462 closure_init(&c->cl, NULL);
1463 set_closure_fn(&c->cl, cache_set_free, system_wq);
1464
1465 closure_init(&c->caching, &c->cl);
1466 set_closure_fn(&c->caching, __cache_set_unregister, system_wq);
1467
1468 /* Maybe create continue_at_noreturn() and use it here? */
1469 closure_set_stopped(&c->cl);
1470 closure_put(&c->cl);
1471
1472 kobject_init(&c->kobj, &bch_cache_set_ktype);
1473 kobject_init(&c->internal, &bch_cache_set_internal_ktype);
1474
1475 bch_cache_accounting_init(&c->accounting, &c->cl);
1476
1477 memcpy(c->sb.set_uuid, sb->set_uuid, 16);
1478 c->sb.block_size = sb->block_size;
1479 c->sb.bucket_size = sb->bucket_size;
1480 c->sb.nr_in_set = sb->nr_in_set;
1481 c->sb.last_mount = sb->last_mount;
1482 c->bucket_bits = ilog2(sb->bucket_size);
1483 c->block_bits = ilog2(sb->block_size);
1484 c->nr_uuids = bucket_bytes(c) / sizeof(struct uuid_entry);
1485
Kent Overstreetee811282013-12-17 23:49:49 -08001486 c->btree_pages = bucket_pages(c);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001487 if (c->btree_pages > BTREE_MAX_PAGES)
1488 c->btree_pages = max_t(int, c->btree_pages / 4,
1489 BTREE_MAX_PAGES);
1490
Kent Overstreetcb7a5832013-12-16 15:27:25 -08001491 sema_init(&c->sb_write_mutex, 1);
Kent Overstreete8e1d462013-07-24 17:27:07 -07001492 mutex_init(&c->bucket_lock);
1493 init_waitqueue_head(&c->try_wait);
Kent Overstreet35fcd842013-07-24 17:29:09 -07001494 init_waitqueue_head(&c->bucket_wait);
Kent Overstreetcb7a5832013-12-16 15:27:25 -08001495 sema_init(&c->uuid_write_mutex, 1);
Kent Overstreet65d22e92013-07-31 00:03:54 -07001496
Kent Overstreet65d22e92013-07-31 00:03:54 -07001497 spin_lock_init(&c->btree_gc_time.lock);
1498 spin_lock_init(&c->btree_split_time.lock);
1499 spin_lock_init(&c->btree_read_time.lock);
1500 spin_lock_init(&c->try_harder_time.lock);
Kent Overstreete8e1d462013-07-24 17:27:07 -07001501
Kent Overstreetcafe5632013-03-23 16:11:31 -07001502 bch_moving_init_cache_set(c);
1503
1504 INIT_LIST_HEAD(&c->list);
1505 INIT_LIST_HEAD(&c->cached_devs);
1506 INIT_LIST_HEAD(&c->btree_cache);
1507 INIT_LIST_HEAD(&c->btree_cache_freeable);
1508 INIT_LIST_HEAD(&c->btree_cache_freed);
1509 INIT_LIST_HEAD(&c->data_buckets);
1510
1511 c->search = mempool_create_slab_pool(32, bch_search_cache);
1512 if (!c->search)
1513 goto err;
1514
1515 iter_size = (sb->bucket_size / sb->block_size + 1) *
1516 sizeof(struct btree_iter_set);
1517
1518 if (!(c->devices = kzalloc(c->nr_uuids * sizeof(void *), GFP_KERNEL)) ||
1519 !(c->bio_meta = mempool_create_kmalloc_pool(2,
1520 sizeof(struct bbio) + sizeof(struct bio_vec) *
1521 bucket_pages(c))) ||
Kent Overstreet57943512013-04-25 13:58:35 -07001522 !(c->fill_iter = mempool_create_kmalloc_pool(1, iter_size)) ||
Kent Overstreetcafe5632013-03-23 16:11:31 -07001523 !(c->bio_split = bioset_create(4, offsetof(struct bbio, bio))) ||
Kent Overstreetcafe5632013-03-23 16:11:31 -07001524 !(c->uuids = alloc_bucket_pages(GFP_KERNEL, c)) ||
1525 bch_journal_alloc(c) ||
1526 bch_btree_cache_alloc(c) ||
Kent Overstreet67539e82013-09-10 22:53:34 -07001527 bch_open_buckets_alloc(c) ||
1528 bch_bset_sort_state_init(&c->sort, ilog2(c->btree_pages)))
Kent Overstreetcafe5632013-03-23 16:11:31 -07001529 goto err;
1530
Kent Overstreetcafe5632013-03-23 16:11:31 -07001531 c->congested_read_threshold_us = 2000;
1532 c->congested_write_threshold_us = 20000;
1533 c->error_limit = 8 << IO_ERROR_SHIFT;
1534
1535 return c;
1536err:
1537 bch_cache_set_unregister(c);
1538 return NULL;
1539}
1540
1541static void run_cache_set(struct cache_set *c)
1542{
1543 const char *err = "cannot allocate memory";
1544 struct cached_dev *dc, *t;
1545 struct cache *ca;
Kent Overstreetc18536a2013-07-24 17:44:17 -07001546 struct closure cl;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001547 unsigned i;
1548
Kent Overstreetc18536a2013-07-24 17:44:17 -07001549 closure_init_stack(&cl);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001550
1551 for_each_cache(ca, c, i)
1552 c->nbuckets += ca->sb.nbuckets;
1553
1554 if (CACHE_SYNC(&c->sb)) {
1555 LIST_HEAD(journal);
1556 struct bkey *k;
1557 struct jset *j;
1558
1559 err = "cannot allocate memory for journal";
Kent Overstreetc18536a2013-07-24 17:44:17 -07001560 if (bch_journal_read(c, &journal))
Kent Overstreetcafe5632013-03-23 16:11:31 -07001561 goto err;
1562
1563 pr_debug("btree_journal_read() done");
1564
1565 err = "no journal entries found";
1566 if (list_empty(&journal))
1567 goto err;
1568
1569 j = &list_entry(journal.prev, struct journal_replay, list)->j;
1570
1571 err = "IO error reading priorities";
1572 for_each_cache(ca, c, i)
1573 prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev]);
1574
1575 /*
1576 * If prio_read() fails it'll call cache_set_error and we'll
1577 * tear everything down right away, but if we perhaps checked
1578 * sooner we could avoid journal replay.
1579 */
1580
1581 k = &j->btree_root;
1582
1583 err = "bad btree root";
Kent Overstreet65d45232013-12-20 17:22:05 -08001584 if (__bch_btree_ptr_invalid(c, k))
Kent Overstreetcafe5632013-03-23 16:11:31 -07001585 goto err;
1586
1587 err = "error reading btree root";
Kent Overstreete8e1d462013-07-24 17:27:07 -07001588 c->root = bch_btree_node_get(c, k, j->btree_level, true);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001589 if (IS_ERR_OR_NULL(c->root))
1590 goto err;
1591
1592 list_del_init(&c->root->list);
1593 rw_unlock(true, c->root);
1594
Kent Overstreetc18536a2013-07-24 17:44:17 -07001595 err = uuid_read(c, j, &cl);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001596 if (err)
1597 goto err;
1598
1599 err = "error in recovery";
Kent Overstreetc18536a2013-07-24 17:44:17 -07001600 if (bch_btree_check(c))
Kent Overstreetcafe5632013-03-23 16:11:31 -07001601 goto err;
1602
1603 bch_journal_mark(c, &journal);
1604 bch_btree_gc_finish(c);
1605 pr_debug("btree_check() done");
1606
1607 /*
1608 * bcache_journal_next() can't happen sooner, or
1609 * btree_gc_finish() will give spurious errors about last_gc >
1610 * gc_gen - this is a hack but oh well.
1611 */
1612 bch_journal_next(&c->journal);
1613
Kent Overstreet119ba0f2013-04-24 19:01:12 -07001614 err = "error starting allocator thread";
Kent Overstreetcafe5632013-03-23 16:11:31 -07001615 for_each_cache(ca, c, i)
Kent Overstreet119ba0f2013-04-24 19:01:12 -07001616 if (bch_cache_allocator_start(ca))
1617 goto err;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001618
1619 /*
1620 * First place it's safe to allocate: btree_check() and
1621 * btree_gc_finish() have to run before we have buckets to
1622 * allocate, and bch_bucket_alloc_set() might cause a journal
1623 * entry to be written so bcache_journal_next() has to be called
1624 * first.
1625 *
1626 * If the uuids were in the old format we have to rewrite them
1627 * before the next journal entry is written:
1628 */
1629 if (j->version < BCACHE_JSET_VERSION_UUID)
1630 __uuid_write(c);
1631
Kent Overstreetc18536a2013-07-24 17:44:17 -07001632 bch_journal_replay(c, &journal);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001633 } else {
1634 pr_notice("invalidating existing data");
Kent Overstreetcafe5632013-03-23 16:11:31 -07001635
1636 for_each_cache(ca, c, i) {
1637 unsigned j;
1638
1639 ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7,
1640 2, SB_JOURNAL_BUCKETS);
1641
1642 for (j = 0; j < ca->sb.keys; j++)
1643 ca->sb.d[j] = ca->sb.first_bucket + j;
1644 }
1645
1646 bch_btree_gc_finish(c);
1647
Kent Overstreet119ba0f2013-04-24 19:01:12 -07001648 err = "error starting allocator thread";
Kent Overstreetcafe5632013-03-23 16:11:31 -07001649 for_each_cache(ca, c, i)
Kent Overstreet119ba0f2013-04-24 19:01:12 -07001650 if (bch_cache_allocator_start(ca))
1651 goto err;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001652
1653 mutex_lock(&c->bucket_lock);
1654 for_each_cache(ca, c, i)
1655 bch_prio_write(ca);
1656 mutex_unlock(&c->bucket_lock);
1657
Kent Overstreetcafe5632013-03-23 16:11:31 -07001658 err = "cannot allocate new UUID bucket";
1659 if (__uuid_write(c))
Kent Overstreet72a44512013-10-24 17:19:26 -07001660 goto err;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001661
1662 err = "cannot allocate new btree root";
Kent Overstreetbc9389e2013-09-10 19:07:35 -07001663 c->root = bch_btree_node_alloc(c, 0, true);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001664 if (IS_ERR_OR_NULL(c->root))
Kent Overstreet72a44512013-10-24 17:19:26 -07001665 goto err;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001666
1667 bkey_copy_key(&c->root->key, &MAX_KEY);
Kent Overstreetc18536a2013-07-24 17:44:17 -07001668 bch_btree_node_write(c->root, &cl);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001669
1670 bch_btree_set_root(c->root);
1671 rw_unlock(true, c->root);
1672
1673 /*
1674 * We don't want to write the first journal entry until
1675 * everything is set up - fortunately journal entries won't be
1676 * written until the SET_CACHE_SYNC() here:
1677 */
1678 SET_CACHE_SYNC(&c->sb, true);
1679
1680 bch_journal_next(&c->journal);
Kent Overstreetc18536a2013-07-24 17:44:17 -07001681 bch_journal_meta(c, &cl);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001682 }
1683
Kent Overstreet72a44512013-10-24 17:19:26 -07001684 err = "error starting gc thread";
1685 if (bch_gc_thread_start(c))
1686 goto err;
1687
Kent Overstreetc18536a2013-07-24 17:44:17 -07001688 closure_sync(&cl);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001689 c->sb.last_mount = get_seconds();
1690 bcache_write_super(c);
1691
1692 list_for_each_entry_safe(dc, t, &uncached_devices, list)
1693 bch_cached_dev_attach(dc, c);
1694
1695 flash_devs_run(c);
1696
1697 return;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001698err:
Kent Overstreetc18536a2013-07-24 17:44:17 -07001699 closure_sync(&cl);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001700 /* XXX: test this, it's broken */
Kees Cookc8694942013-09-10 21:41:34 -07001701 bch_cache_set_error(c, "%s", err);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001702}
1703
1704static bool can_attach_cache(struct cache *ca, struct cache_set *c)
1705{
1706 return ca->sb.block_size == c->sb.block_size &&
Nicholas Swenson9eb8ebe2013-10-22 13:19:23 -07001707 ca->sb.bucket_size == c->sb.bucket_size &&
Kent Overstreetcafe5632013-03-23 16:11:31 -07001708 ca->sb.nr_in_set == c->sb.nr_in_set;
1709}
1710
1711static const char *register_cache_set(struct cache *ca)
1712{
1713 char buf[12];
1714 const char *err = "cannot allocate memory";
1715 struct cache_set *c;
1716
1717 list_for_each_entry(c, &bch_cache_sets, list)
1718 if (!memcmp(c->sb.set_uuid, ca->sb.set_uuid, 16)) {
1719 if (c->cache[ca->sb.nr_this_dev])
1720 return "duplicate cache set member";
1721
1722 if (!can_attach_cache(ca, c))
1723 return "cache sb does not match set";
1724
1725 if (!CACHE_SYNC(&ca->sb))
1726 SET_CACHE_SYNC(&c->sb, false);
1727
1728 goto found;
1729 }
1730
1731 c = bch_cache_set_alloc(&ca->sb);
1732 if (!c)
1733 return err;
1734
1735 err = "error creating kobject";
1736 if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->sb.set_uuid) ||
1737 kobject_add(&c->internal, &c->kobj, "internal"))
1738 goto err;
1739
1740 if (bch_cache_accounting_add_kobjs(&c->accounting, &c->kobj))
1741 goto err;
1742
1743 bch_debug_init_cache_set(c);
1744
1745 list_add(&c->list, &bch_cache_sets);
1746found:
1747 sprintf(buf, "cache%i", ca->sb.nr_this_dev);
1748 if (sysfs_create_link(&ca->kobj, &c->kobj, "set") ||
1749 sysfs_create_link(&c->kobj, &ca->kobj, buf))
1750 goto err;
1751
1752 if (ca->sb.seq > c->sb.seq) {
1753 c->sb.version = ca->sb.version;
1754 memcpy(c->sb.set_uuid, ca->sb.set_uuid, 16);
1755 c->sb.flags = ca->sb.flags;
1756 c->sb.seq = ca->sb.seq;
1757 pr_debug("set version = %llu", c->sb.version);
1758 }
1759
1760 ca->set = c;
1761 ca->set->cache[ca->sb.nr_this_dev] = ca;
1762 c->cache_by_alloc[c->caches_loaded++] = ca;
1763
1764 if (c->caches_loaded == c->sb.nr_in_set)
1765 run_cache_set(c);
1766
1767 return NULL;
1768err:
1769 bch_cache_set_unregister(c);
1770 return err;
1771}
1772
1773/* Cache device */
1774
1775void bch_cache_release(struct kobject *kobj)
1776{
1777 struct cache *ca = container_of(kobj, struct cache, kobj);
Kent Overstreet78365412013-12-17 01:29:34 -08001778 unsigned i;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001779
1780 if (ca->set)
1781 ca->set->cache[ca->sb.nr_this_dev] = NULL;
1782
Kent Overstreetcafe5632013-03-23 16:11:31 -07001783 bio_split_pool_free(&ca->bio_split_hook);
1784
Kent Overstreetcafe5632013-03-23 16:11:31 -07001785 free_pages((unsigned long) ca->disk_buckets, ilog2(bucket_pages(ca)));
1786 kfree(ca->prio_buckets);
1787 vfree(ca->buckets);
1788
1789 free_heap(&ca->heap);
1790 free_fifo(&ca->unused);
1791 free_fifo(&ca->free_inc);
Kent Overstreet78365412013-12-17 01:29:34 -08001792
1793 for (i = 0; i < RESERVE_NR; i++)
1794 free_fifo(&ca->free[i]);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001795
1796 if (ca->sb_bio.bi_inline_vecs[0].bv_page)
1797 put_page(ca->sb_bio.bi_io_vec[0].bv_page);
1798
1799 if (!IS_ERR_OR_NULL(ca->bdev)) {
1800 blk_sync_queue(bdev_get_queue(ca->bdev));
1801 blkdev_put(ca->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1802 }
1803
1804 kfree(ca);
1805 module_put(THIS_MODULE);
1806}
1807
1808static int cache_alloc(struct cache_sb *sb, struct cache *ca)
1809{
1810 size_t free;
1811 struct bucket *b;
1812
Kent Overstreetcafe5632013-03-23 16:11:31 -07001813 __module_get(THIS_MODULE);
1814 kobject_init(&ca->kobj, &bch_cache_ktype);
1815
Kent Overstreetcafe5632013-03-23 16:11:31 -07001816 bio_init(&ca->journal.bio);
1817 ca->journal.bio.bi_max_vecs = 8;
1818 ca->journal.bio.bi_io_vec = ca->journal.bio.bi_inline_vecs;
1819
Kent Overstreet78365412013-12-17 01:29:34 -08001820 free = roundup_pow_of_two(ca->sb.nbuckets) >> 10;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001821
Kent Overstreet78365412013-12-17 01:29:34 -08001822 if (!init_fifo(&ca->free[RESERVE_BTREE], 8, GFP_KERNEL) ||
1823 !init_fifo(&ca->free[RESERVE_PRIO], prio_buckets(ca), GFP_KERNEL) ||
1824 !init_fifo(&ca->free[RESERVE_MOVINGGC], free, GFP_KERNEL) ||
1825 !init_fifo(&ca->free[RESERVE_NONE], free, GFP_KERNEL) ||
Kent Overstreetcafe5632013-03-23 16:11:31 -07001826 !init_fifo(&ca->free_inc, free << 2, GFP_KERNEL) ||
1827 !init_fifo(&ca->unused, free << 2, GFP_KERNEL) ||
1828 !init_heap(&ca->heap, free << 3, GFP_KERNEL) ||
Kent Overstreetf59fce82013-05-15 00:11:26 -07001829 !(ca->buckets = vzalloc(sizeof(struct bucket) *
Kent Overstreetcafe5632013-03-23 16:11:31 -07001830 ca->sb.nbuckets)) ||
1831 !(ca->prio_buckets = kzalloc(sizeof(uint64_t) * prio_buckets(ca) *
1832 2, GFP_KERNEL)) ||
1833 !(ca->disk_buckets = alloc_bucket_pages(GFP_KERNEL, ca)) ||
Kent Overstreetcafe5632013-03-23 16:11:31 -07001834 bio_split_pool_init(&ca->bio_split_hook))
Kent Overstreetf59fce82013-05-15 00:11:26 -07001835 return -ENOMEM;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001836
1837 ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca);
1838
Kent Overstreetcafe5632013-03-23 16:11:31 -07001839 for_each_bucket(b, ca)
1840 atomic_set(&b->pin, 0);
1841
1842 if (bch_cache_allocator_init(ca))
1843 goto err;
1844
1845 return 0;
1846err:
1847 kobject_put(&ca->kobj);
1848 return -ENOMEM;
1849}
1850
Kent Overstreetf59fce82013-05-15 00:11:26 -07001851static void register_cache(struct cache_sb *sb, struct page *sb_page,
Kent Overstreetcafe5632013-03-23 16:11:31 -07001852 struct block_device *bdev, struct cache *ca)
1853{
1854 char name[BDEVNAME_SIZE];
1855 const char *err = "cannot allocate memory";
1856
Kent Overstreetf59fce82013-05-15 00:11:26 -07001857 memcpy(&ca->sb, sb, sizeof(struct cache_sb));
Kent Overstreetcafe5632013-03-23 16:11:31 -07001858 ca->bdev = bdev;
1859 ca->bdev->bd_holder = ca;
1860
Kent Overstreetf59fce82013-05-15 00:11:26 -07001861 bio_init(&ca->sb_bio);
1862 ca->sb_bio.bi_max_vecs = 1;
1863 ca->sb_bio.bi_io_vec = ca->sb_bio.bi_inline_vecs;
1864 ca->sb_bio.bi_io_vec[0].bv_page = sb_page;
1865 get_page(sb_page);
1866
Kent Overstreetcafe5632013-03-23 16:11:31 -07001867 if (blk_queue_discard(bdev_get_queue(ca->bdev)))
1868 ca->discard = CACHE_DISCARD(&ca->sb);
1869
Kent Overstreetf59fce82013-05-15 00:11:26 -07001870 if (cache_alloc(sb, ca) != 0)
1871 goto err;
1872
Kent Overstreetcafe5632013-03-23 16:11:31 -07001873 err = "error creating kobject";
1874 if (kobject_add(&ca->kobj, &part_to_dev(bdev->bd_part)->kobj, "bcache"))
1875 goto err;
1876
Kent Overstreet4fa03402014-03-17 18:58:55 -07001877 mutex_lock(&bch_register_lock);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001878 err = register_cache_set(ca);
Kent Overstreet4fa03402014-03-17 18:58:55 -07001879 mutex_unlock(&bch_register_lock);
1880
Kent Overstreetcafe5632013-03-23 16:11:31 -07001881 if (err)
1882 goto err;
1883
1884 pr_info("registered cache device %s", bdevname(bdev, name));
Kent Overstreetf59fce82013-05-15 00:11:26 -07001885 return;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001886err:
Kent Overstreetf59fce82013-05-15 00:11:26 -07001887 pr_notice("error opening %s: %s", bdevname(bdev, name), err);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001888 kobject_put(&ca->kobj);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001889}
1890
1891/* Global interfaces/init */
1892
1893static ssize_t register_bcache(struct kobject *, struct kobj_attribute *,
1894 const char *, size_t);
1895
1896kobj_attribute_write(register, register_bcache);
1897kobj_attribute_write(register_quiet, register_bcache);
1898
Gabriel de Perthuisa9dd53a2013-05-04 12:19:41 +02001899static bool bch_is_open_backing(struct block_device *bdev) {
1900 struct cache_set *c, *tc;
1901 struct cached_dev *dc, *t;
1902
1903 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
1904 list_for_each_entry_safe(dc, t, &c->cached_devs, list)
1905 if (dc->bdev == bdev)
1906 return true;
1907 list_for_each_entry_safe(dc, t, &uncached_devices, list)
1908 if (dc->bdev == bdev)
1909 return true;
1910 return false;
1911}
1912
1913static bool bch_is_open_cache(struct block_device *bdev) {
1914 struct cache_set *c, *tc;
1915 struct cache *ca;
1916 unsigned i;
1917
1918 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
1919 for_each_cache(ca, c, i)
1920 if (ca->bdev == bdev)
1921 return true;
1922 return false;
1923}
1924
1925static bool bch_is_open(struct block_device *bdev) {
1926 return bch_is_open_cache(bdev) || bch_is_open_backing(bdev);
1927}
1928
Kent Overstreetcafe5632013-03-23 16:11:31 -07001929static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr,
1930 const char *buffer, size_t size)
1931{
1932 ssize_t ret = size;
1933 const char *err = "cannot allocate memory";
1934 char *path = NULL;
1935 struct cache_sb *sb = NULL;
1936 struct block_device *bdev = NULL;
1937 struct page *sb_page = NULL;
1938
1939 if (!try_module_get(THIS_MODULE))
1940 return -EBUSY;
1941
Kent Overstreetcafe5632013-03-23 16:11:31 -07001942 if (!(path = kstrndup(buffer, size, GFP_KERNEL)) ||
1943 !(sb = kmalloc(sizeof(struct cache_sb), GFP_KERNEL)))
1944 goto err;
1945
1946 err = "failed to open device";
1947 bdev = blkdev_get_by_path(strim(path),
1948 FMODE_READ|FMODE_WRITE|FMODE_EXCL,
1949 sb);
Kent Overstreetf59fce82013-05-15 00:11:26 -07001950 if (IS_ERR(bdev)) {
Gabriel de Perthuisa9dd53a2013-05-04 12:19:41 +02001951 if (bdev == ERR_PTR(-EBUSY)) {
1952 bdev = lookup_bdev(strim(path));
1953 if (!IS_ERR(bdev) && bch_is_open(bdev))
1954 err = "device already registered";
1955 else
1956 err = "device busy";
1957 }
Kent Overstreetcafe5632013-03-23 16:11:31 -07001958 goto err;
Kent Overstreetf59fce82013-05-15 00:11:26 -07001959 }
1960
1961 err = "failed to set blocksize";
1962 if (set_blocksize(bdev, 4096))
1963 goto err_close;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001964
1965 err = read_super(sb, bdev, &sb_page);
1966 if (err)
1967 goto err_close;
1968
Kent Overstreet29033812013-04-11 15:14:35 -07001969 if (SB_IS_BDEV(sb)) {
Kent Overstreetcafe5632013-03-23 16:11:31 -07001970 struct cached_dev *dc = kzalloc(sizeof(*dc), GFP_KERNEL);
Kent Overstreetf59fce82013-05-15 00:11:26 -07001971 if (!dc)
1972 goto err_close;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001973
Kent Overstreet4fa03402014-03-17 18:58:55 -07001974 mutex_lock(&bch_register_lock);
Kent Overstreetf59fce82013-05-15 00:11:26 -07001975 register_bdev(sb, sb_page, bdev, dc);
Kent Overstreet4fa03402014-03-17 18:58:55 -07001976 mutex_unlock(&bch_register_lock);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001977 } else {
1978 struct cache *ca = kzalloc(sizeof(*ca), GFP_KERNEL);
Kent Overstreetf59fce82013-05-15 00:11:26 -07001979 if (!ca)
1980 goto err_close;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001981
Kent Overstreetf59fce82013-05-15 00:11:26 -07001982 register_cache(sb, sb_page, bdev, ca);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001983 }
Kent Overstreetf59fce82013-05-15 00:11:26 -07001984out:
1985 if (sb_page)
Kent Overstreetcafe5632013-03-23 16:11:31 -07001986 put_page(sb_page);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001987 kfree(sb);
1988 kfree(path);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001989 module_put(THIS_MODULE);
1990 return ret;
Kent Overstreetf59fce82013-05-15 00:11:26 -07001991
1992err_close:
1993 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1994err:
1995 if (attr != &ksysfs_register_quiet)
1996 pr_info("error opening %s: %s", path, err);
1997 ret = -EINVAL;
1998 goto out;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001999}
2000
2001static int bcache_reboot(struct notifier_block *n, unsigned long code, void *x)
2002{
2003 if (code == SYS_DOWN ||
2004 code == SYS_HALT ||
2005 code == SYS_POWER_OFF) {
2006 DEFINE_WAIT(wait);
2007 unsigned long start = jiffies;
2008 bool stopped = false;
2009
2010 struct cache_set *c, *tc;
2011 struct cached_dev *dc, *tdc;
2012
2013 mutex_lock(&bch_register_lock);
2014
2015 if (list_empty(&bch_cache_sets) &&
2016 list_empty(&uncached_devices))
2017 goto out;
2018
2019 pr_info("Stopping all devices:");
2020
2021 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
2022 bch_cache_set_stop(c);
2023
2024 list_for_each_entry_safe(dc, tdc, &uncached_devices, list)
2025 bcache_device_stop(&dc->disk);
2026
2027 /* What's a condition variable? */
2028 while (1) {
2029 long timeout = start + 2 * HZ - jiffies;
2030
2031 stopped = list_empty(&bch_cache_sets) &&
2032 list_empty(&uncached_devices);
2033
2034 if (timeout < 0 || stopped)
2035 break;
2036
2037 prepare_to_wait(&unregister_wait, &wait,
2038 TASK_UNINTERRUPTIBLE);
2039
2040 mutex_unlock(&bch_register_lock);
2041 schedule_timeout(timeout);
2042 mutex_lock(&bch_register_lock);
2043 }
2044
2045 finish_wait(&unregister_wait, &wait);
2046
2047 if (stopped)
2048 pr_info("All devices stopped");
2049 else
2050 pr_notice("Timeout waiting for devices to be closed");
2051out:
2052 mutex_unlock(&bch_register_lock);
2053 }
2054
2055 return NOTIFY_DONE;
2056}
2057
2058static struct notifier_block reboot = {
2059 .notifier_call = bcache_reboot,
2060 .priority = INT_MAX, /* before any real devices */
2061};
2062
2063static void bcache_exit(void)
2064{
2065 bch_debug_exit();
Kent Overstreetcafe5632013-03-23 16:11:31 -07002066 bch_request_exit();
2067 bch_btree_exit();
2068 if (bcache_kobj)
2069 kobject_put(bcache_kobj);
2070 if (bcache_wq)
2071 destroy_workqueue(bcache_wq);
Kent Overstreet5c41c8a2013-07-08 17:53:26 -07002072 if (bcache_major)
2073 unregister_blkdev(bcache_major, "bcache");
Kent Overstreetcafe5632013-03-23 16:11:31 -07002074 unregister_reboot_notifier(&reboot);
2075}
2076
2077static int __init bcache_init(void)
2078{
2079 static const struct attribute *files[] = {
2080 &ksysfs_register.attr,
2081 &ksysfs_register_quiet.attr,
2082 NULL
2083 };
2084
2085 mutex_init(&bch_register_lock);
2086 init_waitqueue_head(&unregister_wait);
2087 register_reboot_notifier(&reboot);
Kent Overstreet07e86cc2013-03-25 11:46:43 -07002088 closure_debug_init();
Kent Overstreetcafe5632013-03-23 16:11:31 -07002089
2090 bcache_major = register_blkdev(0, "bcache");
2091 if (bcache_major < 0)
2092 return bcache_major;
2093
2094 if (!(bcache_wq = create_workqueue("bcache")) ||
2095 !(bcache_kobj = kobject_create_and_add("bcache", fs_kobj)) ||
2096 sysfs_create_files(bcache_kobj, files) ||
2097 bch_btree_init() ||
2098 bch_request_init() ||
Kent Overstreetcafe5632013-03-23 16:11:31 -07002099 bch_debug_init(bcache_kobj))
2100 goto err;
2101
2102 return 0;
2103err:
2104 bcache_exit();
2105 return -ENOMEM;
2106}
2107
2108module_exit(bcache_exit);
2109module_init(bcache_init);