?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
dm: Call proper helper to determine dax support
[linux/fpc-iii.git] / drivers / md / bcache / super.c
blob2cbfcd99b7ee792b5dcef89ffd2455d2e13c4eaf
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * bcache setup/teardown code, and some metadata io - read a superblock and
4 * figure out what to do with it.
5 *
6 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
7 * Copyright 2012 Google, Inc.
8 */
9
10 #include "bcache.h"
11 #include "btree.h"
12 #include "debug.h"
13 #include "extents.h"
14 #include "request.h"
15 #include "writeback.h"
16
17 #include <linux/blkdev.h>
18 #include <linux/buffer_head.h>
19 #include <linux/debugfs.h>
20 #include <linux/genhd.h>
21 #include <linux/idr.h>
22 #include <linux/kthread.h>
23 #include <linux/module.h>
24 #include <linux/random.h>
25 #include <linux/reboot.h>
26 #include <linux/sysfs.h>
27
28 unsigned int bch_cutoff_writeback;
29 unsigned int bch_cutoff_writeback_sync;
30
31 static const char bcache_magic[] = {
32 0xc6, 0x85, 0x73, 0xf6, 0x4e, 0x1a, 0x45, 0xca,
33 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81
34 };
35
36 static const char invalid_uuid[] = {
37 0xa0, 0x3e, 0xf8, 0xed, 0x3e, 0xe1, 0xb8, 0x78,
38 0xc8, 0x50, 0xfc, 0x5e, 0xcb, 0x16, 0xcd, 0x99
39 };
40
41 static struct kobject *bcache_kobj;
42 struct mutex bch_register_lock;
43 bool bcache_is_reboot;
44 LIST_HEAD(bch_cache_sets);
45 static LIST_HEAD(uncached_devices);
46
47 static int bcache_major;
48 static DEFINE_IDA(bcache_device_idx);
49 static wait_queue_head_t unregister_wait;
50 struct workqueue_struct *bcache_wq;
51 struct workqueue_struct *bch_journal_wq;
52
53
54 #define BTREE_MAX_PAGES (256 * 1024 / PAGE_SIZE)
55 /* limitation of partitions number on single bcache device */
56 #define BCACHE_MINORS 128
57 /* limitation of bcache devices number on single system */
58 #define BCACHE_DEVICE_IDX_MAX ((1U << MINORBITS)/BCACHE_MINORS)
59
60 /* Superblock */
61
62 static const char *read_super(struct cache_sb *sb, struct block_device *bdev,
63 struct page **res)
64 {
65 const char *err;
66 struct cache_sb *s;
67 struct buffer_head *bh = __bread(bdev, 1, SB_SIZE);
68 unsigned int i;
69
70 if (!bh)
71 return "IO error";
72
73 s = (struct cache_sb *) bh->b_data;
74
75 sb->offset = le64_to_cpu(s->offset);
76 sb->version = le64_to_cpu(s->version);
77
78 memcpy(sb->magic, s->magic, 16);
79 memcpy(sb->uuid, s->uuid, 16);
80 memcpy(sb->set_uuid, s->set_uuid, 16);
81 memcpy(sb->label, s->label, SB_LABEL_SIZE);
82
83 sb->flags = le64_to_cpu(s->flags);
84 sb->seq = le64_to_cpu(s->seq);
85 sb->last_mount = le32_to_cpu(s->last_mount);
86 sb->first_bucket = le16_to_cpu(s->first_bucket);
87 sb->keys = le16_to_cpu(s->keys);
88
89 for (i = 0; i < SB_JOURNAL_BUCKETS; i++)
90 sb->d[i] = le64_to_cpu(s->d[i]);
91
92 pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u",
93 sb->version, sb->flags, sb->seq, sb->keys);
94
95 err = "Not a bcache superblock";
96 if (sb->offset != SB_SECTOR)
97 goto err;
98
99 if (memcmp(sb->magic, bcache_magic, 16))
100 goto err;
101
102 err = "Too many journal buckets";
103 if (sb->keys > SB_JOURNAL_BUCKETS)
104 goto err;
105
106 err = "Bad checksum";
107 if (s->csum != csum_set(s))
108 goto err;
109
110 err = "Bad UUID";
111 if (bch_is_zero(sb->uuid, 16))
112 goto err;
113
114 sb->block_size = le16_to_cpu(s->block_size);
115
116 err = "Superblock block size smaller than device block size";
117 if (sb->block_size << 9 < bdev_logical_block_size(bdev))
118 goto err;
119
120 switch (sb->version) {
121 case BCACHE_SB_VERSION_BDEV:
122 sb->data_offset = BDEV_DATA_START_DEFAULT;
123 break;
124 case BCACHE_SB_VERSION_BDEV_WITH_OFFSET:
125 sb->data_offset = le64_to_cpu(s->data_offset);
126
127 err = "Bad data offset";
128 if (sb->data_offset < BDEV_DATA_START_DEFAULT)
129 goto err;
130
131 break;
132 case BCACHE_SB_VERSION_CDEV:
133 case BCACHE_SB_VERSION_CDEV_WITH_UUID:
134 sb->nbuckets = le64_to_cpu(s->nbuckets);
135 sb->bucket_size = le16_to_cpu(s->bucket_size);
136
137 sb->nr_in_set = le16_to_cpu(s->nr_in_set);
138 sb->nr_this_dev = le16_to_cpu(s->nr_this_dev);
139
140 err = "Too many buckets";
141 if (sb->nbuckets > LONG_MAX)
142 goto err;
143
144 err = "Not enough buckets";
145 if (sb->nbuckets < 1 << 7)
146 goto err;
147
148 err = "Bad block/bucket size";
149 if (!is_power_of_2(sb->block_size) ||
150 sb->block_size > PAGE_SECTORS ||
151 !is_power_of_2(sb->bucket_size) ||
152 sb->bucket_size < PAGE_SECTORS)
153 goto err;
154
155 err = "Invalid superblock: device too small";
156 if (get_capacity(bdev->bd_disk) <
157 sb->bucket_size * sb->nbuckets)
158 goto err;
159
160 err = "Bad UUID";
161 if (bch_is_zero(sb->set_uuid, 16))
162 goto err;
163
164 err = "Bad cache device number in set";
165 if (!sb->nr_in_set ||
166 sb->nr_in_set <= sb->nr_this_dev ||
167 sb->nr_in_set > MAX_CACHES_PER_SET)
168 goto err;
169
170 err = "Journal buckets not sequential";
171 for (i = 0; i < sb->keys; i++)
172 if (sb->d[i] != sb->first_bucket + i)
173 goto err;
174
175 err = "Too many journal buckets";
176 if (sb->first_bucket + sb->keys > sb->nbuckets)
177 goto err;
178
179 err = "Invalid superblock: first bucket comes before end of super";
180 if (sb->first_bucket * sb->bucket_size < 16)
181 goto err;
182
183 break;
184 default:
185 err = "Unsupported superblock version";
186 goto err;
187 }
188
189 sb->last_mount = (u32)ktime_get_real_seconds();
190 err = NULL;
191
192 get_page(bh->b_page);
193 *res = bh->b_page;
194 err:
195 put_bh(bh);
196 return err;
197 }
198
199 static void write_bdev_super_endio(struct bio *bio)
200 {
201 struct cached_dev *dc = bio->bi_private;
202
203 if (bio->bi_status)
204 bch_count_backing_io_errors(dc, bio);
205
206 closure_put(&dc->sb_write);
207 }
208
209 static void __write_super(struct cache_sb *sb, struct bio *bio)
210 {
211 struct cache_sb *out = page_address(bio_first_page_all(bio));
212 unsigned int i;
213
214 bio->bi_iter.bi_sector = SB_SECTOR;
215 bio->bi_iter.bi_size = SB_SIZE;
216 bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
217 bch_bio_map(bio, NULL);
218
219 out->offset = cpu_to_le64(sb->offset);
220 out->version = cpu_to_le64(sb->version);
221
222 memcpy(out->uuid, sb->uuid, 16);
223 memcpy(out->set_uuid, sb->set_uuid, 16);
224 memcpy(out->label, sb->label, SB_LABEL_SIZE);
225
226 out->flags = cpu_to_le64(sb->flags);
227 out->seq = cpu_to_le64(sb->seq);
228
229 out->last_mount = cpu_to_le32(sb->last_mount);
230 out->first_bucket = cpu_to_le16(sb->first_bucket);
231 out->keys = cpu_to_le16(sb->keys);
232
233 for (i = 0; i < sb->keys; i++)
234 out->d[i] = cpu_to_le64(sb->d[i]);
235
236 out->csum = csum_set(out);
237
238 pr_debug("ver %llu, flags %llu, seq %llu",
239 sb->version, sb->flags, sb->seq);
240
241 submit_bio(bio);
242 }
243
244 static void bch_write_bdev_super_unlock(struct closure *cl)
245 {
246 struct cached_dev *dc = container_of(cl, struct cached_dev, sb_write);
247
248 up(&dc->sb_write_mutex);
249 }
250
251 void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent)
252 {
253 struct closure *cl = &dc->sb_write;
254 struct bio *bio = &dc->sb_bio;
255
256 down(&dc->sb_write_mutex);
257 closure_init(cl, parent);
258
259 bio_reset(bio);
260 bio_set_dev(bio, dc->bdev);
261 bio->bi_end_io = write_bdev_super_endio;
262 bio->bi_private = dc;
263
264 closure_get(cl);
265 /* I/O request sent to backing device */
266 __write_super(&dc->sb, bio);
267
268 closure_return_with_destructor(cl, bch_write_bdev_super_unlock);
269 }
270
271 static void write_super_endio(struct bio *bio)
272 {
273 struct cache *ca = bio->bi_private;
274
275 /* is_read = 0 */
276 bch_count_io_errors(ca, bio->bi_status, 0,
277 "writing superblock");
278 closure_put(&ca->set->sb_write);
279 }
280
281 static void bcache_write_super_unlock(struct closure *cl)
282 {
283 struct cache_set *c = container_of(cl, struct cache_set, sb_write);
284
285 up(&c->sb_write_mutex);
286 }
287
288 void bcache_write_super(struct cache_set *c)
289 {
290 struct closure *cl = &c->sb_write;
291 struct cache *ca;
292 unsigned int i;
293
294 down(&c->sb_write_mutex);
295 closure_init(cl, &c->cl);
296
297 c->sb.seq++;
298
299 for_each_cache(ca, c, i) {
300 struct bio *bio = &ca->sb_bio;
301
302 ca->sb.version = BCACHE_SB_VERSION_CDEV_WITH_UUID;
303 ca->sb.seq = c->sb.seq;
304 ca->sb.last_mount = c->sb.last_mount;
305
306 SET_CACHE_SYNC(&ca->sb, CACHE_SYNC(&c->sb));
307
308 bio_reset(bio);
309 bio_set_dev(bio, ca->bdev);
310 bio->bi_end_io = write_super_endio;
311 bio->bi_private = ca;
312
313 closure_get(cl);
314 __write_super(&ca->sb, bio);
315 }
316
317 closure_return_with_destructor(cl, bcache_write_super_unlock);
318 }
319
320 /* UUID io */
321
322 static void uuid_endio(struct bio *bio)
323 {
324 struct closure *cl = bio->bi_private;
325 struct cache_set *c = container_of(cl, struct cache_set, uuid_write);
326
327 cache_set_err_on(bio->bi_status, c, "accessing uuids");
328 bch_bbio_free(bio, c);
329 closure_put(cl);
330 }
331
332 static void uuid_io_unlock(struct closure *cl)
333 {
334 struct cache_set *c = container_of(cl, struct cache_set, uuid_write);
335
336 up(&c->uuid_write_mutex);
337 }
338
339 static void uuid_io(struct cache_set *c, int op, unsigned long op_flags,
340 struct bkey *k, struct closure *parent)
341 {
342 struct closure *cl = &c->uuid_write;
343 struct uuid_entry *u;
344 unsigned int i;
345 char buf[80];
346
347 BUG_ON(!parent);
348 down(&c->uuid_write_mutex);
349 closure_init(cl, parent);
350
351 for (i = 0; i < KEY_PTRS(k); i++) {
352 struct bio *bio = bch_bbio_alloc(c);
353
354 bio->bi_opf = REQ_SYNC | REQ_META | op_flags;
355 bio->bi_iter.bi_size = KEY_SIZE(k) << 9;
356
357 bio->bi_end_io = uuid_endio;
358 bio->bi_private = cl;
359 bio_set_op_attrs(bio, op, REQ_SYNC|REQ_META|op_flags);
360 bch_bio_map(bio, c->uuids);
361
362 bch_submit_bbio(bio, c, k, i);
363
364 if (op != REQ_OP_WRITE)
365 break;
366 }
367
368 bch_extent_to_text(buf, sizeof(buf), k);
369 pr_debug("%s UUIDs at %s", op == REQ_OP_WRITE ? "wrote" : "read", buf);
370
371 for (u = c->uuids; u < c->uuids + c->nr_uuids; u++)
372 if (!bch_is_zero(u->uuid, 16))
373 pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u",
374 u - c->uuids, u->uuid, u->label,
375 u->first_reg, u->last_reg, u->invalidated);
376
377 closure_return_with_destructor(cl, uuid_io_unlock);
378 }
379
380 static char *uuid_read(struct cache_set *c, struct jset *j, struct closure *cl)
381 {
382 struct bkey *k = &j->uuid_bucket;
383
384 if (__bch_btree_ptr_invalid(c, k))
385 return "bad uuid pointer";
386
387 bkey_copy(&c->uuid_bucket, k);
388 uuid_io(c, REQ_OP_READ, 0, k, cl);
389
390 if (j->version < BCACHE_JSET_VERSION_UUIDv1) {
391 struct uuid_entry_v0 *u0 = (void *) c->uuids;
392 struct uuid_entry *u1 = (void *) c->uuids;
393 int i;
394
395 closure_sync(cl);
396
397 /*
398 * Since the new uuid entry is bigger than the old, we have to
399 * convert starting at the highest memory address and work down
400 * in order to do it in place
401 */
402
403 for (i = c->nr_uuids - 1;
404 i >= 0;
405 --i) {
406 memcpy(u1[i].uuid, u0[i].uuid, 16);
407 memcpy(u1[i].label, u0[i].label, 32);
408
409 u1[i].first_reg = u0[i].first_reg;
410 u1[i].last_reg = u0[i].last_reg;
411 u1[i].invalidated = u0[i].invalidated;
412
413 u1[i].flags = 0;
414 u1[i].sectors = 0;
415 }
416 }
417
418 return NULL;
419 }
420
421 static int __uuid_write(struct cache_set *c)
422 {
423 BKEY_PADDED(key) k;
424 struct closure cl;
425 struct cache *ca;
426
427 closure_init_stack(&cl);
428 lockdep_assert_held(&bch_register_lock);
429
430 if (bch_bucket_alloc_set(c, RESERVE_BTREE, &k.key, 1, true))
431 return 1;
432
433 SET_KEY_SIZE(&k.key, c->sb.bucket_size);
434 uuid_io(c, REQ_OP_WRITE, 0, &k.key, &cl);
435 closure_sync(&cl);
436
437 /* Only one bucket used for uuid write */
438 ca = PTR_CACHE(c, &k.key, 0);
439 atomic_long_add(ca->sb.bucket_size, &ca->meta_sectors_written);
440
441 bkey_copy(&c->uuid_bucket, &k.key);
442 bkey_put(c, &k.key);
443 return 0;
444 }
445
446 int bch_uuid_write(struct cache_set *c)
447 {
448 int ret = __uuid_write(c);
449
450 if (!ret)
451 bch_journal_meta(c, NULL);
452
453 return ret;
454 }
455
456 static struct uuid_entry *uuid_find(struct cache_set *c, const char *uuid)
457 {
458 struct uuid_entry *u;
459
460 for (u = c->uuids;
461 u < c->uuids + c->nr_uuids; u++)
462 if (!memcmp(u->uuid, uuid, 16))
463 return u;
464
465 return NULL;
466 }
467
468 static struct uuid_entry *uuid_find_empty(struct cache_set *c)
469 {
470 static const char zero_uuid[16] = "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0";
471
472 return uuid_find(c, zero_uuid);
473 }
474
475 /*
476 * Bucket priorities/gens:
477 *
478 * For each bucket, we store on disk its
479 * 8 bit gen
480 * 16 bit priority
481 *
482 * See alloc.c for an explanation of the gen. The priority is used to implement
483 * lru (and in the future other) cache replacement policies; for most purposes
484 * it's just an opaque integer.
485 *
486 * The gens and the priorities don't have a whole lot to do with each other, and
487 * it's actually the gens that must be written out at specific times - it's no
488 * big deal if the priorities don't get written, if we lose them we just reuse
489 * buckets in suboptimal order.
490 *
491 * On disk they're stored in a packed array, and in as many buckets are required
492 * to fit them all. The buckets we use to store them form a list; the journal
493 * header points to the first bucket, the first bucket points to the second
494 * bucket, et cetera.
495 *
496 * This code is used by the allocation code; periodically (whenever it runs out
497 * of buckets to allocate from) the allocation code will invalidate some
498 * buckets, but it can't use those buckets until their new gens are safely on
499 * disk.
500 */
501
502 static void prio_endio(struct bio *bio)
503 {
504 struct cache *ca = bio->bi_private;
505
506 cache_set_err_on(bio->bi_status, ca->set, "accessing priorities");
507 bch_bbio_free(bio, ca->set);
508 closure_put(&ca->prio);
509 }
510
511 static void prio_io(struct cache *ca, uint64_t bucket, int op,
512 unsigned long op_flags)
513 {
514 struct closure *cl = &ca->prio;
515 struct bio *bio = bch_bbio_alloc(ca->set);
516
517 closure_init_stack(cl);
518
519 bio->bi_iter.bi_sector = bucket * ca->sb.bucket_size;
520 bio_set_dev(bio, ca->bdev);
521 bio->bi_iter.bi_size = bucket_bytes(ca);
522
523 bio->bi_end_io = prio_endio;
524 bio->bi_private = ca;
525 bio_set_op_attrs(bio, op, REQ_SYNC|REQ_META|op_flags);
526 bch_bio_map(bio, ca->disk_buckets);
527
528 closure_bio_submit(ca->set, bio, &ca->prio);
529 closure_sync(cl);
530 }
531
532 int bch_prio_write(struct cache *ca, bool wait)
533 {
534 int i;
535 struct bucket *b;
536 struct closure cl;
537
538 pr_debug("free_prio=%zu, free_none=%zu, free_inc=%zu",
539 fifo_used(&ca->free[RESERVE_PRIO]),
540 fifo_used(&ca->free[RESERVE_NONE]),
541 fifo_used(&ca->free_inc));
542
543 /*
544 * Pre-check if there are enough free buckets. In the non-blocking
545 * scenario it's better to fail early rather than starting to allocate
546 * buckets and do a cleanup later in case of failure.
547 */
548 if (!wait) {
549 size_t avail = fifo_used(&ca->free[RESERVE_PRIO]) +
550 fifo_used(&ca->free[RESERVE_NONE]);
551 if (prio_buckets(ca) > avail)
552 return -ENOMEM;
553 }
554
555 closure_init_stack(&cl);
556
557 lockdep_assert_held(&ca->set->bucket_lock);
558
559 ca->disk_buckets->seq++;
560
561 atomic_long_add(ca->sb.bucket_size * prio_buckets(ca),
562 &ca->meta_sectors_written);
563
564 for (i = prio_buckets(ca) - 1; i >= 0; --i) {
565 long bucket;
566 struct prio_set *p = ca->disk_buckets;
567 struct bucket_disk *d = p->data;
568 struct bucket_disk *end = d + prios_per_bucket(ca);
569
570 for (b = ca->buckets + i * prios_per_bucket(ca);
571 b < ca->buckets + ca->sb.nbuckets && d < end;
572 b++, d++) {
573 d->prio = cpu_to_le16(b->prio);
574 d->gen = b->gen;
575 }
576
577 p->next_bucket = ca->prio_buckets[i + 1];
578 p->magic = pset_magic(&ca->sb);
579 p->csum = bch_crc64(&p->magic, bucket_bytes(ca) - 8);
580
581 bucket = bch_bucket_alloc(ca, RESERVE_PRIO, wait);
582 BUG_ON(bucket == -1);
583
584 mutex_unlock(&ca->set->bucket_lock);
585 prio_io(ca, bucket, REQ_OP_WRITE, 0);
586 mutex_lock(&ca->set->bucket_lock);
587
588 ca->prio_buckets[i] = bucket;
589 atomic_dec_bug(&ca->buckets[bucket].pin);
590 }
591
592 mutex_unlock(&ca->set->bucket_lock);
593
594 bch_journal_meta(ca->set, &cl);
595 closure_sync(&cl);
596
597 mutex_lock(&ca->set->bucket_lock);
598
599 /*
600 * Don't want the old priorities to get garbage collected until after we
601 * finish writing the new ones, and they're journalled
602 */
603 for (i = 0; i < prio_buckets(ca); i++) {
604 if (ca->prio_last_buckets[i])
605 __bch_bucket_free(ca,
606 &ca->buckets[ca->prio_last_buckets[i]]);
607
608 ca->prio_last_buckets[i] = ca->prio_buckets[i];
609 }
610 return 0;
611 }
612
613 static void prio_read(struct cache *ca, uint64_t bucket)
614 {
615 struct prio_set *p = ca->disk_buckets;
616 struct bucket_disk *d = p->data + prios_per_bucket(ca), *end = d;
617 struct bucket *b;
618 unsigned int bucket_nr = 0;
619
620 for (b = ca->buckets;
621 b < ca->buckets + ca->sb.nbuckets;
622 b++, d++) {
623 if (d == end) {
624 ca->prio_buckets[bucket_nr] = bucket;
625 ca->prio_last_buckets[bucket_nr] = bucket;
626 bucket_nr++;
627
628 prio_io(ca, bucket, REQ_OP_READ, 0);
629
630 if (p->csum !=
631 bch_crc64(&p->magic, bucket_bytes(ca) - 8))
632 pr_warn("bad csum reading priorities");
633
634 if (p->magic != pset_magic(&ca->sb))
635 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->last_gc = d->gen;
643 }
644 }
645
646 /* Bcache device */
647
648 static int open_dev(struct block_device *b, fmode_t mode)
649 {
650 struct bcache_device *d = b->bd_disk->private_data;
651
652 if (test_bit(BCACHE_DEV_CLOSING, &d->flags))
653 return -ENXIO;
654
655 closure_get(&d->cl);
656 return 0;
657 }
658
659 static void release_dev(struct gendisk *b, fmode_t mode)
660 {
661 struct bcache_device *d = b->private_data;
662
663 closure_put(&d->cl);
664 }
665
666 static int ioctl_dev(struct block_device *b, fmode_t mode,
667 unsigned int cmd, unsigned long arg)
668 {
669 struct bcache_device *d = b->bd_disk->private_data;
670
671 return d->ioctl(d, mode, cmd, arg);
672 }
673
674 static const struct block_device_operations bcache_ops = {
675 .open = open_dev,
676 .release = release_dev,
677 .ioctl = ioctl_dev,
678 .owner = THIS_MODULE,
679 };
680
681 void bcache_device_stop(struct bcache_device *d)
682 {
683 if (!test_and_set_bit(BCACHE_DEV_CLOSING, &d->flags))
684 /*
685 * closure_fn set to
686 * - cached device: cached_dev_flush()
687 * - flash dev: flash_dev_flush()
688 */
689 closure_queue(&d->cl);
690 }
691
692 static void bcache_device_unlink(struct bcache_device *d)
693 {
694 lockdep_assert_held(&bch_register_lock);
695
696 if (d->c && !test_and_set_bit(BCACHE_DEV_UNLINK_DONE, &d->flags)) {
697 unsigned int i;
698 struct cache *ca;
699
700 sysfs_remove_link(&d->c->kobj, d->name);
701 sysfs_remove_link(&d->kobj, "cache");
702
703 for_each_cache(ca, d->c, i)
704 bd_unlink_disk_holder(ca->bdev, d->disk);
705 }
706 }
707
708 static void bcache_device_link(struct bcache_device *d, struct cache_set *c,
709 const char *name)
710 {
711 unsigned int i;
712 struct cache *ca;
713 int ret;
714
715 for_each_cache(ca, d->c, i)
716 bd_link_disk_holder(ca->bdev, d->disk);
717
718 snprintf(d->name, BCACHEDEVNAME_SIZE,
719 "%s%u", name, d->id);
720
721 ret = sysfs_create_link(&d->kobj, &c->kobj, "cache");
722 if (ret < 0)
723 pr_err("Couldn't create device -> cache set symlink");
724
725 ret = sysfs_create_link(&c->kobj, &d->kobj, d->name);
726 if (ret < 0)
727 pr_err("Couldn't create cache set -> device symlink");
728
729 clear_bit(BCACHE_DEV_UNLINK_DONE, &d->flags);
730 }
731
732 static void bcache_device_detach(struct bcache_device *d)
733 {
734 lockdep_assert_held(&bch_register_lock);
735
736 atomic_dec(&d->c->attached_dev_nr);
737
738 if (test_bit(BCACHE_DEV_DETACHING, &d->flags)) {
739 struct uuid_entry *u = d->c->uuids + d->id;
740
741 SET_UUID_FLASH_ONLY(u, 0);
742 memcpy(u->uuid, invalid_uuid, 16);
743 u->invalidated = cpu_to_le32((u32)ktime_get_real_seconds());
744 bch_uuid_write(d->c);
745 }
746
747 bcache_device_unlink(d);
748
749 d->c->devices[d->id] = NULL;
750 closure_put(&d->c->caching);
751 d->c = NULL;
752 }
753
754 static void bcache_device_attach(struct bcache_device *d, struct cache_set *c,
755 unsigned int id)
756 {
757 d->id = id;
758 d->c = c;
759 c->devices[id] = d;
760
761 if (id >= c->devices_max_used)
762 c->devices_max_used = id + 1;
763
764 closure_get(&c->caching);
765 }
766
767 static inline int first_minor_to_idx(int first_minor)
768 {
769 return (first_minor/BCACHE_MINORS);
770 }
771
772 static inline int idx_to_first_minor(int idx)
773 {
774 return (idx * BCACHE_MINORS);
775 }
776
777 static void bcache_device_free(struct bcache_device *d)
778 {
779 struct gendisk *disk = d->disk;
780
781 lockdep_assert_held(&bch_register_lock);
782
783 if (disk)
784 pr_info("%s stopped", disk->disk_name);
785 else
786 pr_err("bcache device (NULL gendisk) stopped");
787
788 if (d->c)
789 bcache_device_detach(d);
790
791 if (disk) {
792 bool disk_added = (disk->flags & GENHD_FL_UP) != 0;
793
794 if (disk_added)
795 del_gendisk(disk);
796
797 if (disk->queue)
798 blk_cleanup_queue(disk->queue);
799
800 ida_simple_remove(&bcache_device_idx,
801 first_minor_to_idx(disk->first_minor));
802 if (disk_added)
803 put_disk(disk);
804 }
805
806 bioset_exit(&d->bio_split);
807 kvfree(d->full_dirty_stripes);
808 kvfree(d->stripe_sectors_dirty);
809
810 closure_debug_destroy(&d->cl);
811 }
812
813 static int bcache_device_init(struct bcache_device *d, unsigned int block_size,
814 sector_t sectors)
815 {
816 struct request_queue *q;
817 const size_t max_stripes = min_t(size_t, INT_MAX,
818 SIZE_MAX / sizeof(atomic_t));
819 uint64_t n;
820 int idx;
821
822 if (!d->stripe_size)
823 d->stripe_size = 1 << 31;
824
825 n = DIV_ROUND_UP_ULL(sectors, d->stripe_size);
826 if (!n || n > max_stripes) {
827 pr_err("nr_stripes too large or invalid: %llu (start sector beyond end of disk?)\n",
828 n);
829 return -ENOMEM;
830 }
831 d->nr_stripes = n;
832
833 n = d->nr_stripes * sizeof(atomic_t);
834 d->stripe_sectors_dirty = kvzalloc(n, GFP_KERNEL);
835 if (!d->stripe_sectors_dirty)
836 return -ENOMEM;
837
838 n = BITS_TO_LONGS(d->nr_stripes) * sizeof(unsigned long);
839 d->full_dirty_stripes = kvzalloc(n, GFP_KERNEL);
840 if (!d->full_dirty_stripes)
841 return -ENOMEM;
842
843 idx = ida_simple_get(&bcache_device_idx, 0,
844 BCACHE_DEVICE_IDX_MAX, GFP_KERNEL);
845 if (idx < 0)
846 return idx;
847
848 if (bioset_init(&d->bio_split, 4, offsetof(struct bbio, bio),
849 BIOSET_NEED_BVECS|BIOSET_NEED_RESCUER))
850 goto err;
851
852 d->disk = alloc_disk(BCACHE_MINORS);
853 if (!d->disk)
854 goto err;
855
856 set_capacity(d->disk, sectors);
857 snprintf(d->disk->disk_name, DISK_NAME_LEN, "bcache%i", idx);
858
859 d->disk->major = bcache_major;
860 d->disk->first_minor = idx_to_first_minor(idx);
861 d->disk->fops = &bcache_ops;
862 d->disk->private_data = d;
863
864 q = blk_alloc_queue(GFP_KERNEL);
865 if (!q)
866 return -ENOMEM;
867
868 blk_queue_make_request(q, NULL);
869 d->disk->queue = q;
870 q->queuedata = d;
871 q->backing_dev_info->congested_data = d;
872 q->limits.max_hw_sectors = UINT_MAX;
873 q->limits.max_sectors = UINT_MAX;
874 q->limits.max_segment_size = UINT_MAX;
875 q->limits.max_segments = BIO_MAX_PAGES;
876 blk_queue_max_discard_sectors(q, UINT_MAX);
877 q->limits.discard_granularity = 512;
878 q->limits.io_min = block_size;
879 q->limits.logical_block_size = block_size;
880 q->limits.physical_block_size = block_size;
881 blk_queue_flag_set(QUEUE_FLAG_NONROT, d->disk->queue);
882 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, d->disk->queue);
883 blk_queue_flag_set(QUEUE_FLAG_DISCARD, d->disk->queue);
884
885 blk_queue_write_cache(q, true, true);
886
887 return 0;
888
889 err:
890 ida_simple_remove(&bcache_device_idx, idx);
891 return -ENOMEM;
892
893 }
894
895 /* Cached device */
896
897 static void calc_cached_dev_sectors(struct cache_set *c)
898 {
899 uint64_t sectors = 0;
900 struct cached_dev *dc;
901
902 list_for_each_entry(dc, &c->cached_devs, list)
903 sectors += bdev_sectors(dc->bdev);
904
905 c->cached_dev_sectors = sectors;
906 }
907
908 #define BACKING_DEV_OFFLINE_TIMEOUT 5
909 static int cached_dev_status_update(void *arg)
910 {
911 struct cached_dev *dc = arg;
912 struct request_queue *q;
913
914 /*
915 * If this delayed worker is stopping outside, directly quit here.
916 * dc->io_disable might be set via sysfs interface, so check it
917 * here too.
918 */
919 while (!kthread_should_stop() && !dc->io_disable) {
920 q = bdev_get_queue(dc->bdev);
921 if (blk_queue_dying(q))
922 dc->offline_seconds++;
923 else
924 dc->offline_seconds = 0;
925
926 if (dc->offline_seconds >= BACKING_DEV_OFFLINE_TIMEOUT) {
927 pr_err("%s: device offline for %d seconds",
928 dc->backing_dev_name,
929 BACKING_DEV_OFFLINE_TIMEOUT);
930 pr_err("%s: disable I/O request due to backing "
931 "device offline", dc->disk.name);
932 dc->io_disable = true;
933 /* let others know earlier that io_disable is true */
934 smp_mb();
935 bcache_device_stop(&dc->disk);
936 break;
937 }
938 schedule_timeout_interruptible(HZ);
939 }
940
941 wait_for_kthread_stop();
942 return 0;
943 }
944
945
946 int bch_cached_dev_run(struct cached_dev *dc)
947 {
948 struct bcache_device *d = &dc->disk;
949 char *buf = kmemdup_nul(dc->sb.label, SB_LABEL_SIZE, GFP_KERNEL);
950 char *env[] = {
951 "DRIVER=bcache",
952 kasprintf(GFP_KERNEL, "CACHED_UUID=%pU", dc->sb.uuid),
953 kasprintf(GFP_KERNEL, "CACHED_LABEL=%s", buf ? : ""),
954 NULL,
955 };
956
957 if (dc->io_disable) {
958 pr_err("I/O disabled on cached dev %s",
959 dc->backing_dev_name);
960 kfree(env[1]);
961 kfree(env[2]);
962 kfree(buf);
963 return -EIO;
964 }
965
966 if (atomic_xchg(&dc->running, 1)) {
967 kfree(env[1]);
968 kfree(env[2]);
969 kfree(buf);
970 pr_info("cached dev %s is running already",
971 dc->backing_dev_name);
972 return -EBUSY;
973 }
974
975 if (!d->c &&
976 BDEV_STATE(&dc->sb) != BDEV_STATE_NONE) {
977 struct closure cl;
978
979 closure_init_stack(&cl);
980
981 SET_BDEV_STATE(&dc->sb, BDEV_STATE_STALE);
982 bch_write_bdev_super(dc, &cl);
983 closure_sync(&cl);
984 }
985
986 add_disk(d->disk);
987 bd_link_disk_holder(dc->bdev, dc->disk.disk);
988 /*
989 * won't show up in the uevent file, use udevadm monitor -e instead
990 * only class / kset properties are persistent
991 */
992 kobject_uevent_env(&disk_to_dev(d->disk)->kobj, KOBJ_CHANGE, env);
993 kfree(env[1]);
994 kfree(env[2]);
995 kfree(buf);
996
997 if (sysfs_create_link(&d->kobj, &disk_to_dev(d->disk)->kobj, "dev") ||
998 sysfs_create_link(&disk_to_dev(d->disk)->kobj,
999 &d->kobj, "bcache")) {
1000 pr_err("Couldn't create bcache dev <-> disk sysfs symlinks");
1001 return -ENOMEM;
1002 }
1003
1004 dc->status_update_thread = kthread_run(cached_dev_status_update,
1005 dc, "bcache_status_update");
1006 if (IS_ERR(dc->status_update_thread)) {
1007 pr_warn("failed to create bcache_status_update kthread, "
1008 "continue to run without monitoring backing "
1009 "device status");
1010 }
1011
1012 return 0;
1013 }
1014
1015 /*
1016 * If BCACHE_DEV_RATE_DW_RUNNING is set, it means routine of the delayed
1017 * work dc->writeback_rate_update is running. Wait until the routine
1018 * quits (BCACHE_DEV_RATE_DW_RUNNING is clear), then continue to
1019 * cancel it. If BCACHE_DEV_RATE_DW_RUNNING is not clear after time_out
1020 * seconds, give up waiting here and continue to cancel it too.
1021 */
1022 static void cancel_writeback_rate_update_dwork(struct cached_dev *dc)
1023 {
1024 int time_out = WRITEBACK_RATE_UPDATE_SECS_MAX * HZ;
1025
1026 do {
1027 if (!test_bit(BCACHE_DEV_RATE_DW_RUNNING,
1028 &dc->disk.flags))
1029 break;
1030 time_out--;
1031 schedule_timeout_interruptible(1);
1032 } while (time_out > 0);
1033
1034 if (time_out == 0)
1035 pr_warn("give up waiting for dc->writeback_write_update to quit");
1036
1037 cancel_delayed_work_sync(&dc->writeback_rate_update);
1038 }
1039
1040 static void cached_dev_detach_finish(struct work_struct *w)
1041 {
1042 struct cached_dev *dc = container_of(w, struct cached_dev, detach);
1043 struct closure cl;
1044
1045 closure_init_stack(&cl);
1046
1047 BUG_ON(!test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags));
1048 BUG_ON(refcount_read(&dc->count));
1049
1050
1051 if (test_and_clear_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags))
1052 cancel_writeback_rate_update_dwork(dc);
1053
1054 if (!IS_ERR_OR_NULL(dc->writeback_thread)) {
1055 kthread_stop(dc->writeback_thread);
1056 dc->writeback_thread = NULL;
1057 }
1058
1059 memset(&dc->sb.set_uuid, 0, 16);
1060 SET_BDEV_STATE(&dc->sb, BDEV_STATE_NONE);
1061
1062 bch_write_bdev_super(dc, &cl);
1063 closure_sync(&cl);
1064
1065 mutex_lock(&bch_register_lock);
1066
1067 calc_cached_dev_sectors(dc->disk.c);
1068 bcache_device_detach(&dc->disk);
1069 list_move(&dc->list, &uncached_devices);
1070
1071 clear_bit(BCACHE_DEV_DETACHING, &dc->disk.flags);
1072 clear_bit(BCACHE_DEV_UNLINK_DONE, &dc->disk.flags);
1073
1074 mutex_unlock(&bch_register_lock);
1075
1076 pr_info("Caching disabled for %s", dc->backing_dev_name);
1077
1078 /* Drop ref we took in cached_dev_detach() */
1079 closure_put(&dc->disk.cl);
1080 }
1081
1082 void bch_cached_dev_detach(struct cached_dev *dc)
1083 {
1084 lockdep_assert_held(&bch_register_lock);
1085
1086 if (test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags))
1087 return;
1088
1089 if (test_and_set_bit(BCACHE_DEV_DETACHING, &dc->disk.flags))
1090 return;
1091
1092 /*
1093 * Block the device from being closed and freed until we're finished
1094 * detaching
1095 */
1096 closure_get(&dc->disk.cl);
1097
1098 bch_writeback_queue(dc);
1099
1100 cached_dev_put(dc);
1101 }
1102
1103 int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c,
1104 uint8_t *set_uuid)
1105 {
1106 uint32_t rtime = cpu_to_le32((u32)ktime_get_real_seconds());
1107 struct uuid_entry *u;
1108 struct cached_dev *exist_dc, *t;
1109 int ret = 0;
1110
1111 if ((set_uuid && memcmp(set_uuid, c->sb.set_uuid, 16)) ||
1112 (!set_uuid && memcmp(dc->sb.set_uuid, c->sb.set_uuid, 16)))
1113 return -ENOENT;
1114
1115 if (dc->disk.c) {
1116 pr_err("Can't attach %s: already attached",
1117 dc->backing_dev_name);
1118 return -EINVAL;
1119 }
1120
1121 if (test_bit(CACHE_SET_STOPPING, &c->flags)) {
1122 pr_err("Can't attach %s: shutting down",
1123 dc->backing_dev_name);
1124 return -EINVAL;
1125 }
1126
1127 if (dc->sb.block_size < c->sb.block_size) {
1128 /* Will die */
1129 pr_err("Couldn't attach %s: block size less than set's block size",
1130 dc->backing_dev_name);
1131 return -EINVAL;
1132 }
1133
1134 /* Check whether already attached */
1135 list_for_each_entry_safe(exist_dc, t, &c->cached_devs, list) {
1136 if (!memcmp(dc->sb.uuid, exist_dc->sb.uuid, 16)) {
1137 pr_err("Tried to attach %s but duplicate UUID already attached",
1138 dc->backing_dev_name);
1139
1140 return -EINVAL;
1141 }
1142 }
1143
1144 u = uuid_find(c, dc->sb.uuid);
1145
1146 if (u &&
1147 (BDEV_STATE(&dc->sb) == BDEV_STATE_STALE ||
1148 BDEV_STATE(&dc->sb) == BDEV_STATE_NONE)) {
1149 memcpy(u->uuid, invalid_uuid, 16);
1150 u->invalidated = cpu_to_le32((u32)ktime_get_real_seconds());
1151 u = NULL;
1152 }
1153
1154 if (!u) {
1155 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
1156 pr_err("Couldn't find uuid for %s in set",
1157 dc->backing_dev_name);
1158 return -ENOENT;
1159 }
1160
1161 u = uuid_find_empty(c);
1162 if (!u) {
1163 pr_err("Not caching %s, no room for UUID",
1164 dc->backing_dev_name);
1165 return -EINVAL;
1166 }
1167 }
1168
1169 /*
1170 * Deadlocks since we're called via sysfs...
1171 * sysfs_remove_file(&dc->kobj, &sysfs_attach);
1172 */
1173
1174 if (bch_is_zero(u->uuid, 16)) {
1175 struct closure cl;
1176
1177 closure_init_stack(&cl);
1178
1179 memcpy(u->uuid, dc->sb.uuid, 16);
1180 memcpy(u->label, dc->sb.label, SB_LABEL_SIZE);
1181 u->first_reg = u->last_reg = rtime;
1182 bch_uuid_write(c);
1183
1184 memcpy(dc->sb.set_uuid, c->sb.set_uuid, 16);
1185 SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN);
1186
1187 bch_write_bdev_super(dc, &cl);
1188 closure_sync(&cl);
1189 } else {
1190 u->last_reg = rtime;
1191 bch_uuid_write(c);
1192 }
1193
1194 bcache_device_attach(&dc->disk, c, u - c->uuids);
1195 list_move(&dc->list, &c->cached_devs);
1196 calc_cached_dev_sectors(c);
1197
1198 /*
1199 * dc->c must be set before dc->count != 0 - paired with the mb in
1200 * cached_dev_get()
1201 */
1202 smp_wmb();
1203 refcount_set(&dc->count, 1);
1204
1205 /* Block writeback thread, but spawn it */
1206 down_write(&dc->writeback_lock);
1207 if (bch_cached_dev_writeback_start(dc)) {
1208 up_write(&dc->writeback_lock);
1209 pr_err("Couldn't start writeback facilities for %s",
1210 dc->disk.disk->disk_name);
1211 return -ENOMEM;
1212 }
1213
1214 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
1215 atomic_set(&dc->has_dirty, 1);
1216 bch_writeback_queue(dc);
1217 }
1218
1219 bch_sectors_dirty_init(&dc->disk);
1220
1221 ret = bch_cached_dev_run(dc);
1222 if (ret && (ret != -EBUSY)) {
1223 up_write(&dc->writeback_lock);
1224 /*
1225 * bch_register_lock is held, bcache_device_stop() is not
1226 * able to be directly called. The kthread and kworker
1227 * created previously in bch_cached_dev_writeback_start()
1228 * have to be stopped manually here.
1229 */
1230 kthread_stop(dc->writeback_thread);
1231 cancel_writeback_rate_update_dwork(dc);
1232 pr_err("Couldn't run cached device %s",
1233 dc->backing_dev_name);
1234 return ret;
1235 }
1236
1237 bcache_device_link(&dc->disk, c, "bdev");
1238 atomic_inc(&c->attached_dev_nr);
1239
1240 /* Allow the writeback thread to proceed */
1241 up_write(&dc->writeback_lock);
1242
1243 pr_info("Caching %s as %s on set %pU",
1244 dc->backing_dev_name,
1245 dc->disk.disk->disk_name,
1246 dc->disk.c->sb.set_uuid);
1247 return 0;
1248 }
1249
1250 /* when dc->disk.kobj released */
1251 void bch_cached_dev_release(struct kobject *kobj)
1252 {
1253 struct cached_dev *dc = container_of(kobj, struct cached_dev,
1254 disk.kobj);
1255 kfree(dc);
1256 module_put(THIS_MODULE);
1257 }
1258
1259 static void cached_dev_free(struct closure *cl)
1260 {
1261 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1262
1263 if (test_and_clear_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags))
1264 cancel_writeback_rate_update_dwork(dc);
1265
1266 if (!IS_ERR_OR_NULL(dc->writeback_thread))
1267 kthread_stop(dc->writeback_thread);
1268 if (!IS_ERR_OR_NULL(dc->status_update_thread))
1269 kthread_stop(dc->status_update_thread);
1270
1271 mutex_lock(&bch_register_lock);
1272
1273 if (atomic_read(&dc->running))
1274 bd_unlink_disk_holder(dc->bdev, dc->disk.disk);
1275 bcache_device_free(&dc->disk);
1276 list_del(&dc->list);
1277
1278 mutex_unlock(&bch_register_lock);
1279
1280 if (dc->sb_bio.bi_inline_vecs[0].bv_page)
1281 put_page(bio_first_page_all(&dc->sb_bio));
1282
1283 if (!IS_ERR_OR_NULL(dc->bdev))
1284 blkdev_put(dc->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1285
1286 wake_up(&unregister_wait);
1287
1288 kobject_put(&dc->disk.kobj);
1289 }
1290
1291 static void cached_dev_flush(struct closure *cl)
1292 {
1293 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1294 struct bcache_device *d = &dc->disk;
1295
1296 mutex_lock(&bch_register_lock);
1297 bcache_device_unlink(d);
1298 mutex_unlock(&bch_register_lock);
1299
1300 bch_cache_accounting_destroy(&dc->accounting);
1301 kobject_del(&d->kobj);
1302
1303 continue_at(cl, cached_dev_free, system_wq);
1304 }
1305
1306 static int cached_dev_init(struct cached_dev *dc, unsigned int block_size)
1307 {
1308 int ret;
1309 struct io *io;
1310 struct request_queue *q = bdev_get_queue(dc->bdev);
1311
1312 __module_get(THIS_MODULE);
1313 INIT_LIST_HEAD(&dc->list);
1314 closure_init(&dc->disk.cl, NULL);
1315 set_closure_fn(&dc->disk.cl, cached_dev_flush, system_wq);
1316 kobject_init(&dc->disk.kobj, &bch_cached_dev_ktype);
1317 INIT_WORK(&dc->detach, cached_dev_detach_finish);
1318 sema_init(&dc->sb_write_mutex, 1);
1319 INIT_LIST_HEAD(&dc->io_lru);
1320 spin_lock_init(&dc->io_lock);
1321 bch_cache_accounting_init(&dc->accounting, &dc->disk.cl);
1322
1323 dc->sequential_cutoff = 4 << 20;
1324
1325 for (io = dc->io; io < dc->io + RECENT_IO; io++) {
1326 list_add(&io->lru, &dc->io_lru);
1327 hlist_add_head(&io->hash, dc->io_hash + RECENT_IO);
1328 }
1329
1330 dc->disk.stripe_size = q->limits.io_opt >> 9;
1331
1332 if (dc->disk.stripe_size)
1333 dc->partial_stripes_expensive =
1334 q->limits.raid_partial_stripes_expensive;
1335
1336 ret = bcache_device_init(&dc->disk, block_size,
1337 dc->bdev->bd_part->nr_sects - dc->sb.data_offset);
1338 if (ret)
1339 return ret;
1340
1341 dc->disk.disk->queue->backing_dev_info->ra_pages =
1342 max(dc->disk.disk->queue->backing_dev_info->ra_pages,
1343 q->backing_dev_info->ra_pages);
1344
1345 atomic_set(&dc->io_errors, 0);
1346 dc->io_disable = false;
1347 dc->error_limit = DEFAULT_CACHED_DEV_ERROR_LIMIT;
1348 /* default to auto */
1349 dc->stop_when_cache_set_failed = BCH_CACHED_DEV_STOP_AUTO;
1350
1351 bch_cached_dev_request_init(dc);
1352 bch_cached_dev_writeback_init(dc);
1353 return 0;
1354 }
1355
1356 /* Cached device - bcache superblock */
1357
1358 static int register_bdev(struct cache_sb *sb, struct page *sb_page,
1359 struct block_device *bdev,
1360 struct cached_dev *dc)
1361 {
1362 const char *err = "cannot allocate memory";
1363 struct cache_set *c;
1364 int ret = -ENOMEM;
1365
1366 bdevname(bdev, dc->backing_dev_name);
1367 memcpy(&dc->sb, sb, sizeof(struct cache_sb));
1368 dc->bdev = bdev;
1369 dc->bdev->bd_holder = dc;
1370
1371 bio_init(&dc->sb_bio, dc->sb_bio.bi_inline_vecs, 1);
1372 bio_first_bvec_all(&dc->sb_bio)->bv_page = sb_page;
1373 get_page(sb_page);
1374
1375
1376 if (cached_dev_init(dc, sb->block_size << 9))
1377 goto err;
1378
1379 err = "error creating kobject";
1380 if (kobject_add(&dc->disk.kobj, &part_to_dev(bdev->bd_part)->kobj,
1381 "bcache"))
1382 goto err;
1383 if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj))
1384 goto err;
1385
1386 pr_info("registered backing device %s", dc->backing_dev_name);
1387
1388 list_add(&dc->list, &uncached_devices);
1389 /* attach to a matched cache set if it exists */
1390 list_for_each_entry(c, &bch_cache_sets, list)
1391 bch_cached_dev_attach(dc, c, NULL);
1392
1393 if (BDEV_STATE(&dc->sb) == BDEV_STATE_NONE ||
1394 BDEV_STATE(&dc->sb) == BDEV_STATE_STALE) {
1395 err = "failed to run cached device";
1396 ret = bch_cached_dev_run(dc);
1397 if (ret)
1398 goto err;
1399 }
1400
1401 return 0;
1402 err:
1403 pr_notice("error %s: %s", dc->backing_dev_name, err);
1404 bcache_device_stop(&dc->disk);
1405 return ret;
1406 }
1407
1408 /* Flash only volumes */
1409
1410 /* When d->kobj released */
1411 void bch_flash_dev_release(struct kobject *kobj)
1412 {
1413 struct bcache_device *d = container_of(kobj, struct bcache_device,
1414 kobj);
1415 kfree(d);
1416 }
1417
1418 static void flash_dev_free(struct closure *cl)
1419 {
1420 struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1421
1422 mutex_lock(&bch_register_lock);
1423 atomic_long_sub(bcache_dev_sectors_dirty(d),
1424 &d->c->flash_dev_dirty_sectors);
1425 bcache_device_free(d);
1426 mutex_unlock(&bch_register_lock);
1427 kobject_put(&d->kobj);
1428 }
1429
1430 static void flash_dev_flush(struct closure *cl)
1431 {
1432 struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1433
1434 mutex_lock(&bch_register_lock);
1435 bcache_device_unlink(d);
1436 mutex_unlock(&bch_register_lock);
1437 kobject_del(&d->kobj);
1438 continue_at(cl, flash_dev_free, system_wq);
1439 }
1440
1441 static int flash_dev_run(struct cache_set *c, struct uuid_entry *u)
1442 {
1443 struct bcache_device *d = kzalloc(sizeof(struct bcache_device),
1444 GFP_KERNEL);
1445 if (!d)
1446 return -ENOMEM;
1447
1448 closure_init(&d->cl, NULL);
1449 set_closure_fn(&d->cl, flash_dev_flush, system_wq);
1450
1451 kobject_init(&d->kobj, &bch_flash_dev_ktype);
1452
1453 if (bcache_device_init(d, block_bytes(c), u->sectors))
1454 goto err;
1455
1456 bcache_device_attach(d, c, u - c->uuids);
1457 bch_sectors_dirty_init(d);
1458 bch_flash_dev_request_init(d);
1459 add_disk(d->disk);
1460
1461 if (kobject_add(&d->kobj, &disk_to_dev(d->disk)->kobj, "bcache"))
1462 goto err;
1463
1464 bcache_device_link(d, c, "volume");
1465
1466 return 0;
1467 err:
1468 kobject_put(&d->kobj);
1469 return -ENOMEM;
1470 }
1471
1472 static int flash_devs_run(struct cache_set *c)
1473 {
1474 int ret = 0;
1475 struct uuid_entry *u;
1476
1477 for (u = c->uuids;
1478 u < c->uuids + c->nr_uuids && !ret;
1479 u++)
1480 if (UUID_FLASH_ONLY(u))
1481 ret = flash_dev_run(c, u);
1482
1483 return ret;
1484 }
1485
1486 int bch_flash_dev_create(struct cache_set *c, uint64_t size)
1487 {
1488 struct uuid_entry *u;
1489
1490 if (test_bit(CACHE_SET_STOPPING, &c->flags))
1491 return -EINTR;
1492
1493 if (!test_bit(CACHE_SET_RUNNING, &c->flags))
1494 return -EPERM;
1495
1496 u = uuid_find_empty(c);
1497 if (!u) {
1498 pr_err("Can't create volume, no room for UUID");
1499 return -EINVAL;
1500 }
1501
1502 get_random_bytes(u->uuid, 16);
1503 memset(u->label, 0, 32);
1504 u->first_reg = u->last_reg = cpu_to_le32((u32)ktime_get_real_seconds());
1505
1506 SET_UUID_FLASH_ONLY(u, 1);
1507 u->sectors = size >> 9;
1508
1509 bch_uuid_write(c);
1510
1511 return flash_dev_run(c, u);
1512 }
1513
1514 bool bch_cached_dev_error(struct cached_dev *dc)
1515 {
1516 if (!dc || test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags))
1517 return false;
1518
1519 dc->io_disable = true;
1520 /* make others know io_disable is true earlier */
1521 smp_mb();
1522
1523 pr_err("stop %s: too many IO errors on backing device %s\n",
1524 dc->disk.disk->disk_name, dc->backing_dev_name);
1525
1526 bcache_device_stop(&dc->disk);
1527 return true;
1528 }
1529
1530 /* Cache set */
1531
1532 __printf(2, 3)
1533 bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...)
1534 {
1535 va_list args;
1536
1537 if (c->on_error != ON_ERROR_PANIC &&
1538 test_bit(CACHE_SET_STOPPING, &c->flags))
1539 return false;
1540
1541 if (test_and_set_bit(CACHE_SET_IO_DISABLE, &c->flags))
1542 pr_info("CACHE_SET_IO_DISABLE already set");
1543
1544 /*
1545 * XXX: we can be called from atomic context
1546 * acquire_console_sem();
1547 */
1548
1549 pr_err("bcache: error on %pU: ", c->sb.set_uuid);
1550
1551 va_start(args, fmt);
1552 vprintk(fmt, args);
1553 va_end(args);
1554
1555 pr_err(", disabling caching\n");
1556
1557 if (c->on_error == ON_ERROR_PANIC)
1558 panic("panic forced after error\n");
1559
1560 bch_cache_set_unregister(c);
1561 return true;
1562 }
1563
1564 /* When c->kobj released */
1565 void bch_cache_set_release(struct kobject *kobj)
1566 {
1567 struct cache_set *c = container_of(kobj, struct cache_set, kobj);
1568
1569 kfree(c);
1570 module_put(THIS_MODULE);
1571 }
1572
1573 static void cache_set_free(struct closure *cl)
1574 {
1575 struct cache_set *c = container_of(cl, struct cache_set, cl);
1576 struct cache *ca;
1577 unsigned int i;
1578
1579 debugfs_remove(c->debug);
1580
1581 bch_open_buckets_free(c);
1582 bch_btree_cache_free(c);
1583 bch_journal_free(c);
1584
1585 mutex_lock(&bch_register_lock);
1586 for_each_cache(ca, c, i)
1587 if (ca) {
1588 ca->set = NULL;
1589 c->cache[ca->sb.nr_this_dev] = NULL;
1590 kobject_put(&ca->kobj);
1591 }
1592
1593 bch_bset_sort_state_free(&c->sort);
1594 free_pages((unsigned long) c->uuids, ilog2(bucket_pages(c)));
1595
1596 if (c->moving_gc_wq)
1597 destroy_workqueue(c->moving_gc_wq);
1598 bioset_exit(&c->bio_split);
1599 mempool_exit(&c->fill_iter);
1600 mempool_exit(&c->bio_meta);
1601 mempool_exit(&c->search);
1602 kfree(c->devices);
1603
1604 list_del(&c->list);
1605 mutex_unlock(&bch_register_lock);
1606
1607 pr_info("Cache set %pU unregistered", c->sb.set_uuid);
1608 wake_up(&unregister_wait);
1609
1610 closure_debug_destroy(&c->cl);
1611 kobject_put(&c->kobj);
1612 }
1613
1614 static void cache_set_flush(struct closure *cl)
1615 {
1616 struct cache_set *c = container_of(cl, struct cache_set, caching);
1617 struct cache *ca;
1618 struct btree *b;
1619 unsigned int i;
1620
1621 bch_cache_accounting_destroy(&c->accounting);
1622
1623 kobject_put(&c->internal);
1624 kobject_del(&c->kobj);
1625
1626 if (!IS_ERR_OR_NULL(c->gc_thread))
1627 kthread_stop(c->gc_thread);
1628
1629 if (!IS_ERR_OR_NULL(c->root))
1630 list_add(&c->root->list, &c->btree_cache);
1631
1632 /*
1633 * Avoid flushing cached nodes if cache set is retiring
1634 * due to too many I/O errors detected.
1635 */
1636 if (!test_bit(CACHE_SET_IO_DISABLE, &c->flags))
1637 list_for_each_entry(b, &c->btree_cache, list) {
1638 mutex_lock(&b->write_lock);
1639 if (btree_node_dirty(b))
1640 __bch_btree_node_write(b, NULL);
1641 mutex_unlock(&b->write_lock);
1642 }
1643
1644 for_each_cache(ca, c, i)
1645 if (ca->alloc_thread)
1646 kthread_stop(ca->alloc_thread);
1647
1648 if (c->journal.cur) {
1649 cancel_delayed_work_sync(&c->journal.work);
1650 /* flush last journal entry if needed */
1651 c->journal.work.work.func(&c->journal.work.work);
1652 }
1653
1654 closure_return(cl);
1655 }
1656
1657 /*
1658 * This function is only called when CACHE_SET_IO_DISABLE is set, which means
1659 * cache set is unregistering due to too many I/O errors. In this condition,
1660 * the bcache device might be stopped, it depends on stop_when_cache_set_failed
1661 * value and whether the broken cache has dirty data:
1662 *
1663 * dc->stop_when_cache_set_failed dc->has_dirty stop bcache device
1664 * BCH_CACHED_STOP_AUTO 0 NO
1665 * BCH_CACHED_STOP_AUTO 1 YES
1666 * BCH_CACHED_DEV_STOP_ALWAYS 0 YES
1667 * BCH_CACHED_DEV_STOP_ALWAYS 1 YES
1668 *
1669 * The expected behavior is, if stop_when_cache_set_failed is configured to
1670 * "auto" via sysfs interface, the bcache device will not be stopped if the
1671 * backing device is clean on the broken cache device.
1672 */
1673 static void conditional_stop_bcache_device(struct cache_set *c,
1674 struct bcache_device *d,
1675 struct cached_dev *dc)
1676 {
1677 if (dc->stop_when_cache_set_failed == BCH_CACHED_DEV_STOP_ALWAYS) {
1678 pr_warn("stop_when_cache_set_failed of %s is \"always\", stop it for failed cache set %pU.",
1679 d->disk->disk_name, c->sb.set_uuid);
1680 bcache_device_stop(d);
1681 } else if (atomic_read(&dc->has_dirty)) {
1682 /*
1683 * dc->stop_when_cache_set_failed == BCH_CACHED_STOP_AUTO
1684 * and dc->has_dirty == 1
1685 */
1686 pr_warn("stop_when_cache_set_failed of %s is \"auto\" and cache is dirty, stop it to avoid potential data corruption.",
1687 d->disk->disk_name);
1688 /*
1689 * There might be a small time gap that cache set is
1690 * released but bcache device is not. Inside this time
1691 * gap, regular I/O requests will directly go into
1692 * backing device as no cache set attached to. This
1693 * behavior may also introduce potential inconsistence
1694 * data in writeback mode while cache is dirty.
1695 * Therefore before calling bcache_device_stop() due
1696 * to a broken cache device, dc->io_disable should be
1697 * explicitly set to true.
1698 */
1699 dc->io_disable = true;
1700 /* make others know io_disable is true earlier */
1701 smp_mb();
1702 bcache_device_stop(d);
1703 } else {
1704 /*
1705 * dc->stop_when_cache_set_failed == BCH_CACHED_STOP_AUTO
1706 * and dc->has_dirty == 0
1707 */
1708 pr_warn("stop_when_cache_set_failed of %s is \"auto\" and cache is clean, keep it alive.",
1709 d->disk->disk_name);
1710 }
1711 }
1712
1713 static void __cache_set_unregister(struct closure *cl)
1714 {
1715 struct cache_set *c = container_of(cl, struct cache_set, caching);
1716 struct cached_dev *dc;
1717 struct bcache_device *d;
1718 size_t i;
1719
1720 mutex_lock(&bch_register_lock);
1721
1722 for (i = 0; i < c->devices_max_used; i++) {
1723 d = c->devices[i];
1724 if (!d)
1725 continue;
1726
1727 if (!UUID_FLASH_ONLY(&c->uuids[i]) &&
1728 test_bit(CACHE_SET_UNREGISTERING, &c->flags)) {
1729 dc = container_of(d, struct cached_dev, disk);
1730 bch_cached_dev_detach(dc);
1731 if (test_bit(CACHE_SET_IO_DISABLE, &c->flags))
1732 conditional_stop_bcache_device(c, d, dc);
1733 } else {
1734 bcache_device_stop(d);
1735 }
1736 }
1737
1738 mutex_unlock(&bch_register_lock);
1739
1740 continue_at(cl, cache_set_flush, system_wq);
1741 }
1742
1743 void bch_cache_set_stop(struct cache_set *c)
1744 {
1745 if (!test_and_set_bit(CACHE_SET_STOPPING, &c->flags))
1746 /* closure_fn set to __cache_set_unregister() */
1747 closure_queue(&c->caching);
1748 }
1749
1750 void bch_cache_set_unregister(struct cache_set *c)
1751 {
1752 set_bit(CACHE_SET_UNREGISTERING, &c->flags);
1753 bch_cache_set_stop(c);
1754 }
1755
1756 #define alloc_bucket_pages(gfp, c) \
1757 ((void *) __get_free_pages(__GFP_ZERO|__GFP_COMP|gfp, ilog2(bucket_pages(c))))
1758
1759 struct cache_set *bch_cache_set_alloc(struct cache_sb *sb)
1760 {
1761 int iter_size;
1762 struct cache_set *c = kzalloc(sizeof(struct cache_set), GFP_KERNEL);
1763
1764 if (!c)
1765 return NULL;
1766
1767 __module_get(THIS_MODULE);
1768 closure_init(&c->cl, NULL);
1769 set_closure_fn(&c->cl, cache_set_free, system_wq);
1770
1771 closure_init(&c->caching, &c->cl);
1772 set_closure_fn(&c->caching, __cache_set_unregister, system_wq);
1773
1774 /* Maybe create continue_at_noreturn() and use it here? */
1775 closure_set_stopped(&c->cl);
1776 closure_put(&c->cl);
1777
1778 kobject_init(&c->kobj, &bch_cache_set_ktype);
1779 kobject_init(&c->internal, &bch_cache_set_internal_ktype);
1780
1781 bch_cache_accounting_init(&c->accounting, &c->cl);
1782
1783 memcpy(c->sb.set_uuid, sb->set_uuid, 16);
1784 c->sb.block_size = sb->block_size;
1785 c->sb.bucket_size = sb->bucket_size;
1786 c->sb.nr_in_set = sb->nr_in_set;
1787 c->sb.last_mount = sb->last_mount;
1788 c->bucket_bits = ilog2(sb->bucket_size);
1789 c->block_bits = ilog2(sb->block_size);
1790 c->nr_uuids = bucket_bytes(c) / sizeof(struct uuid_entry);
1791 c->devices_max_used = 0;
1792 atomic_set(&c->attached_dev_nr, 0);
1793 c->btree_pages = bucket_pages(c);
1794 if (c->btree_pages > BTREE_MAX_PAGES)
1795 c->btree_pages = max_t(int, c->btree_pages / 4,
1796 BTREE_MAX_PAGES);
1797
1798 sema_init(&c->sb_write_mutex, 1);
1799 mutex_init(&c->bucket_lock);
1800 init_waitqueue_head(&c->btree_cache_wait);
1801 init_waitqueue_head(&c->bucket_wait);
1802 init_waitqueue_head(&c->gc_wait);
1803 sema_init(&c->uuid_write_mutex, 1);
1804
1805 spin_lock_init(&c->btree_gc_time.lock);
1806 spin_lock_init(&c->btree_split_time.lock);
1807 spin_lock_init(&c->btree_read_time.lock);
1808
1809 bch_moving_init_cache_set(c);
1810
1811 INIT_LIST_HEAD(&c->list);
1812 INIT_LIST_HEAD(&c->cached_devs);
1813 INIT_LIST_HEAD(&c->btree_cache);
1814 INIT_LIST_HEAD(&c->btree_cache_freeable);
1815 INIT_LIST_HEAD(&c->btree_cache_freed);
1816 INIT_LIST_HEAD(&c->data_buckets);
1817
1818 iter_size = (sb->bucket_size / sb->block_size + 1) *
1819 sizeof(struct btree_iter_set);
1820
1821 if (!(c->devices = kcalloc(c->nr_uuids, sizeof(void *), GFP_KERNEL)) ||
1822 mempool_init_slab_pool(&c->search, 32, bch_search_cache) ||
1823 mempool_init_kmalloc_pool(&c->bio_meta, 2,
1824 sizeof(struct bbio) + sizeof(struct bio_vec) *
1825 bucket_pages(c)) ||
1826 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
1827 bioset_init(&c->bio_split, 4, offsetof(struct bbio, bio),
1828 BIOSET_NEED_BVECS|BIOSET_NEED_RESCUER) ||
1829 !(c->uuids = alloc_bucket_pages(GFP_KERNEL, c)) ||
1830 !(c->moving_gc_wq = alloc_workqueue("bcache_gc",
1831 WQ_MEM_RECLAIM, 0)) ||
1832 bch_journal_alloc(c) ||
1833 bch_btree_cache_alloc(c) ||
1834 bch_open_buckets_alloc(c) ||
1835 bch_bset_sort_state_init(&c->sort, ilog2(c->btree_pages)))
1836 goto err;
1837
1838 c->congested_read_threshold_us = 2000;
1839 c->congested_write_threshold_us = 20000;
1840 c->error_limit = DEFAULT_IO_ERROR_LIMIT;
1841 WARN_ON(test_and_clear_bit(CACHE_SET_IO_DISABLE, &c->flags));
1842
1843 return c;
1844 err:
1845 bch_cache_set_unregister(c);
1846 return NULL;
1847 }
1848
1849 static int run_cache_set(struct cache_set *c)
1850 {
1851 const char *err = "cannot allocate memory";
1852 struct cached_dev *dc, *t;
1853 struct cache *ca;
1854 struct closure cl;
1855 unsigned int i;
1856 LIST_HEAD(journal);
1857 struct journal_replay *l;
1858
1859 closure_init_stack(&cl);
1860
1861 for_each_cache(ca, c, i)
1862 c->nbuckets += ca->sb.nbuckets;
1863 set_gc_sectors(c);
1864
1865 if (CACHE_SYNC(&c->sb)) {
1866 struct bkey *k;
1867 struct jset *j;
1868
1869 err = "cannot allocate memory for journal";
1870 if (bch_journal_read(c, &journal))
1871 goto err;
1872
1873 pr_debug("btree_journal_read() done");
1874
1875 err = "no journal entries found";
1876 if (list_empty(&journal))
1877 goto err;
1878
1879 j = &list_entry(journal.prev, struct journal_replay, list)->j;
1880
1881 err = "IO error reading priorities";
1882 for_each_cache(ca, c, i)
1883 prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev]);
1884
1885 /*
1886 * If prio_read() fails it'll call cache_set_error and we'll
1887 * tear everything down right away, but if we perhaps checked
1888 * sooner we could avoid journal replay.
1889 */
1890
1891 k = &j->btree_root;
1892
1893 err = "bad btree root";
1894 if (__bch_btree_ptr_invalid(c, k))
1895 goto err;
1896
1897 err = "error reading btree root";
1898 c->root = bch_btree_node_get(c, NULL, k,
1899 j->btree_level,
1900 true, NULL);
1901 if (IS_ERR_OR_NULL(c->root))
1902 goto err;
1903
1904 list_del_init(&c->root->list);
1905 rw_unlock(true, c->root);
1906
1907 err = uuid_read(c, j, &cl);
1908 if (err)
1909 goto err;
1910
1911 err = "error in recovery";
1912 if (bch_btree_check(c))
1913 goto err;
1914
1915 /*
1916 * bch_btree_check() may occupy too much system memory which
1917 * has negative effects to user space application (e.g. data
1918 * base) performance. Shrink the mca cache memory proactively
1919 * here to avoid competing memory with user space workloads..
1920 */
1921 if (!c->shrinker_disabled) {
1922 struct shrink_control sc;
1923
1924 sc.gfp_mask = GFP_KERNEL;
1925 sc.nr_to_scan = c->btree_cache_used * c->btree_pages;
1926 /* first run to clear b->accessed tag */
1927 c->shrink.scan_objects(&c->shrink, &sc);
1928 /* second run to reap non-accessed nodes */
1929 c->shrink.scan_objects(&c->shrink, &sc);
1930 }
1931
1932 bch_journal_mark(c, &journal);
1933 bch_initial_gc_finish(c);
1934 pr_debug("btree_check() done");
1935
1936 /*
1937 * bcache_journal_next() can't happen sooner, or
1938 * btree_gc_finish() will give spurious errors about last_gc >
1939 * gc_gen - this is a hack but oh well.
1940 */
1941 bch_journal_next(&c->journal);
1942
1943 err = "error starting allocator thread";
1944 for_each_cache(ca, c, i)
1945 if (bch_cache_allocator_start(ca))
1946 goto err;
1947
1948 /*
1949 * First place it's safe to allocate: btree_check() and
1950 * btree_gc_finish() have to run before we have buckets to
1951 * allocate, and bch_bucket_alloc_set() might cause a journal
1952 * entry to be written so bcache_journal_next() has to be called
1953 * first.
1954 *
1955 * If the uuids were in the old format we have to rewrite them
1956 * before the next journal entry is written:
1957 */
1958 if (j->version < BCACHE_JSET_VERSION_UUID)
1959 __uuid_write(c);
1960
1961 err = "bcache: replay journal failed";
1962 if (bch_journal_replay(c, &journal))
1963 goto err;
1964 } else {
1965 pr_notice("invalidating existing data");
1966
1967 for_each_cache(ca, c, i) {
1968 unsigned int j;
1969
1970 ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7,
1971 2, SB_JOURNAL_BUCKETS);
1972
1973 for (j = 0; j < ca->sb.keys; j++)
1974 ca->sb.d[j] = ca->sb.first_bucket + j;
1975 }
1976
1977 bch_initial_gc_finish(c);
1978
1979 err = "error starting allocator thread";
1980 for_each_cache(ca, c, i)
1981 if (bch_cache_allocator_start(ca))
1982 goto err;
1983
1984 mutex_lock(&c->bucket_lock);
1985 for_each_cache(ca, c, i)
1986 bch_prio_write(ca, true);
1987 mutex_unlock(&c->bucket_lock);
1988
1989 err = "cannot allocate new UUID bucket";
1990 if (__uuid_write(c))
1991 goto err;
1992
1993 err = "cannot allocate new btree root";
1994 c->root = __bch_btree_node_alloc(c, NULL, 0, true, NULL);
1995 if (IS_ERR_OR_NULL(c->root))
1996 goto err;
1997
1998 mutex_lock(&c->root->write_lock);
1999 bkey_copy_key(&c->root->key, &MAX_KEY);
2000 bch_btree_node_write(c->root, &cl);
2001 mutex_unlock(&c->root->write_lock);
2002
2003 bch_btree_set_root(c->root);
2004 rw_unlock(true, c->root);
2005
2006 /*
2007 * We don't want to write the first journal entry until
2008 * everything is set up - fortunately journal entries won't be
2009 * written until the SET_CACHE_SYNC() here:
2010 */
2011 SET_CACHE_SYNC(&c->sb, true);
2012
2013 bch_journal_next(&c->journal);
2014 bch_journal_meta(c, &cl);
2015 }
2016
2017 err = "error starting gc thread";
2018 if (bch_gc_thread_start(c))
2019 goto err;
2020
2021 closure_sync(&cl);
2022 c->sb.last_mount = (u32)ktime_get_real_seconds();
2023 bcache_write_super(c);
2024
2025 list_for_each_entry_safe(dc, t, &uncached_devices, list)
2026 bch_cached_dev_attach(dc, c, NULL);
2027
2028 flash_devs_run(c);
2029
2030 set_bit(CACHE_SET_RUNNING, &c->flags);
2031 return 0;
2032 err:
2033 while (!list_empty(&journal)) {
2034 l = list_first_entry(&journal, struct journal_replay, list);
2035 list_del(&l->list);
2036 kfree(l);
2037 }
2038
2039 closure_sync(&cl);
2040
2041 bch_cache_set_error(c, "%s", err);
2042
2043 return -EIO;
2044 }
2045
2046 static bool can_attach_cache(struct cache *ca, struct cache_set *c)
2047 {
2048 return ca->sb.block_size == c->sb.block_size &&
2049 ca->sb.bucket_size == c->sb.bucket_size &&
2050 ca->sb.nr_in_set == c->sb.nr_in_set;
2051 }
2052
2053 static const char *register_cache_set(struct cache *ca)
2054 {
2055 char buf[12];
2056 const char *err = "cannot allocate memory";
2057 struct cache_set *c;
2058
2059 list_for_each_entry(c, &bch_cache_sets, list)
2060 if (!memcmp(c->sb.set_uuid, ca->sb.set_uuid, 16)) {
2061 if (c->cache[ca->sb.nr_this_dev])
2062 return "duplicate cache set member";
2063
2064 if (!can_attach_cache(ca, c))
2065 return "cache sb does not match set";
2066
2067 if (!CACHE_SYNC(&ca->sb))
2068 SET_CACHE_SYNC(&c->sb, false);
2069
2070 goto found;
2071 }
2072
2073 c = bch_cache_set_alloc(&ca->sb);
2074 if (!c)
2075 return err;
2076
2077 err = "error creating kobject";
2078 if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->sb.set_uuid) ||
2079 kobject_add(&c->internal, &c->kobj, "internal"))
2080 goto err;
2081
2082 if (bch_cache_accounting_add_kobjs(&c->accounting, &c->kobj))
2083 goto err;
2084
2085 bch_debug_init_cache_set(c);
2086
2087 list_add(&c->list, &bch_cache_sets);
2088 found:
2089 sprintf(buf, "cache%i", ca->sb.nr_this_dev);
2090 if (sysfs_create_link(&ca->kobj, &c->kobj, "set") ||
2091 sysfs_create_link(&c->kobj, &ca->kobj, buf))
2092 goto err;
2093
2094 /*
2095 * A special case is both ca->sb.seq and c->sb.seq are 0,
2096 * such condition happens on a new created cache device whose
2097 * super block is never flushed yet. In this case c->sb.version
2098 * and other members should be updated too, otherwise we will
2099 * have a mistaken super block version in cache set.
2100 */
2101 if (ca->sb.seq > c->sb.seq || c->sb.seq == 0) {
2102 c->sb.version = ca->sb.version;
2103 memcpy(c->sb.set_uuid, ca->sb.set_uuid, 16);
2104 c->sb.flags = ca->sb.flags;
2105 c->sb.seq = ca->sb.seq;
2106 pr_debug("set version = %llu", c->sb.version);
2107 }
2108
2109 kobject_get(&ca->kobj);
2110 ca->set = c;
2111 ca->set->cache[ca->sb.nr_this_dev] = ca;
2112 c->cache_by_alloc[c->caches_loaded++] = ca;
2113
2114 if (c->caches_loaded == c->sb.nr_in_set) {
2115 err = "failed to run cache set";
2116 if (run_cache_set(c) < 0)
2117 goto err;
2118 }
2119
2120 return NULL;
2121 err:
2122 bch_cache_set_unregister(c);
2123 return err;
2124 }
2125
2126 /* Cache device */
2127
2128 /* When ca->kobj released */
2129 void bch_cache_release(struct kobject *kobj)
2130 {
2131 struct cache *ca = container_of(kobj, struct cache, kobj);
2132 unsigned int i;
2133
2134 if (ca->set) {
2135 BUG_ON(ca->set->cache[ca->sb.nr_this_dev] != ca);
2136 ca->set->cache[ca->sb.nr_this_dev] = NULL;
2137 }
2138
2139 free_pages((unsigned long) ca->disk_buckets, ilog2(bucket_pages(ca)));
2140 kfree(ca->prio_buckets);
2141 vfree(ca->buckets);
2142
2143 free_heap(&ca->heap);
2144 free_fifo(&ca->free_inc);
2145
2146 for (i = 0; i < RESERVE_NR; i++)
2147 free_fifo(&ca->free[i]);
2148
2149 if (ca->sb_bio.bi_inline_vecs[0].bv_page)
2150 put_page(bio_first_page_all(&ca->sb_bio));
2151
2152 if (!IS_ERR_OR_NULL(ca->bdev))
2153 blkdev_put(ca->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2154
2155 kfree(ca);
2156 module_put(THIS_MODULE);
2157 }
2158
2159 static int cache_alloc(struct cache *ca)
2160 {
2161 size_t free;
2162 size_t btree_buckets;
2163 struct bucket *b;
2164 int ret = -ENOMEM;
2165 const char *err = NULL;
2166
2167 __module_get(THIS_MODULE);
2168 kobject_init(&ca->kobj, &bch_cache_ktype);
2169
2170 bio_init(&ca->journal.bio, ca->journal.bio.bi_inline_vecs, 8);
2171
2172 /*
2173 * when ca->sb.njournal_buckets is not zero, journal exists,
2174 * and in bch_journal_replay(), tree node may split,
2175 * so bucket of RESERVE_BTREE type is needed,
2176 * the worst situation is all journal buckets are valid journal,
2177 * and all the keys need to replay,
2178 * so the number of RESERVE_BTREE type buckets should be as much
2179 * as journal buckets
2180 */
2181 btree_buckets = ca->sb.njournal_buckets ?: 8;
2182 free = roundup_pow_of_two(ca->sb.nbuckets) >> 10;
2183 if (!free) {
2184 ret = -EPERM;
2185 err = "ca->sb.nbuckets is too small";
2186 goto err_free;
2187 }
2188
2189 if (!init_fifo(&ca->free[RESERVE_BTREE], btree_buckets,
2190 GFP_KERNEL)) {
2191 err = "ca->free[RESERVE_BTREE] alloc failed";
2192 goto err_btree_alloc;
2193 }
2194
2195 if (!init_fifo_exact(&ca->free[RESERVE_PRIO], prio_buckets(ca),
2196 GFP_KERNEL)) {
2197 err = "ca->free[RESERVE_PRIO] alloc failed";
2198 goto err_prio_alloc;
2199 }
2200
2201 if (!init_fifo(&ca->free[RESERVE_MOVINGGC], free, GFP_KERNEL)) {
2202 err = "ca->free[RESERVE_MOVINGGC] alloc failed";
2203 goto err_movinggc_alloc;
2204 }
2205
2206 if (!init_fifo(&ca->free[RESERVE_NONE], free, GFP_KERNEL)) {
2207 err = "ca->free[RESERVE_NONE] alloc failed";
2208 goto err_none_alloc;
2209 }
2210
2211 if (!init_fifo(&ca->free_inc, free << 2, GFP_KERNEL)) {
2212 err = "ca->free_inc alloc failed";
2213 goto err_free_inc_alloc;
2214 }
2215
2216 if (!init_heap(&ca->heap, free << 3, GFP_KERNEL)) {
2217 err = "ca->heap alloc failed";
2218 goto err_heap_alloc;
2219 }
2220
2221 ca->buckets = vzalloc(array_size(sizeof(struct bucket),
2222 ca->sb.nbuckets));
2223 if (!ca->buckets) {
2224 err = "ca->buckets alloc failed";
2225 goto err_buckets_alloc;
2226 }
2227
2228 ca->prio_buckets = kzalloc(array3_size(sizeof(uint64_t),
2229 prio_buckets(ca), 2),
2230 GFP_KERNEL);
2231 if (!ca->prio_buckets) {
2232 err = "ca->prio_buckets alloc failed";
2233 goto err_prio_buckets_alloc;
2234 }
2235
2236 ca->disk_buckets = alloc_bucket_pages(GFP_KERNEL, ca);
2237 if (!ca->disk_buckets) {
2238 err = "ca->disk_buckets alloc failed";
2239 goto err_disk_buckets_alloc;
2240 }
2241
2242 ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca);
2243
2244 for_each_bucket(b, ca)
2245 atomic_set(&b->pin, 0);
2246 return 0;
2247
2248 err_disk_buckets_alloc:
2249 kfree(ca->prio_buckets);
2250 err_prio_buckets_alloc:
2251 vfree(ca->buckets);
2252 err_buckets_alloc:
2253 free_heap(&ca->heap);
2254 err_heap_alloc:
2255 free_fifo(&ca->free_inc);
2256 err_free_inc_alloc:
2257 free_fifo(&ca->free[RESERVE_NONE]);
2258 err_none_alloc:
2259 free_fifo(&ca->free[RESERVE_MOVINGGC]);
2260 err_movinggc_alloc:
2261 free_fifo(&ca->free[RESERVE_PRIO]);
2262 err_prio_alloc:
2263 free_fifo(&ca->free[RESERVE_BTREE]);
2264 err_btree_alloc:
2265 err_free:
2266 module_put(THIS_MODULE);
2267 if (err)
2268 pr_notice("error %s: %s", ca->cache_dev_name, err);
2269 return ret;
2270 }
2271
2272 static int register_cache(struct cache_sb *sb, struct page *sb_page,
2273 struct block_device *bdev, struct cache *ca)
2274 {
2275 const char *err = NULL; /* must be set for any error case */
2276 int ret = 0;
2277
2278 bdevname(bdev, ca->cache_dev_name);
2279 memcpy(&ca->sb, sb, sizeof(struct cache_sb));
2280 ca->bdev = bdev;
2281 ca->bdev->bd_holder = ca;
2282
2283 bio_init(&ca->sb_bio, ca->sb_bio.bi_inline_vecs, 1);
2284 bio_first_bvec_all(&ca->sb_bio)->bv_page = sb_page;
2285 get_page(sb_page);
2286
2287 if (blk_queue_discard(bdev_get_queue(bdev)))
2288 ca->discard = CACHE_DISCARD(&ca->sb);
2289
2290 ret = cache_alloc(ca);
2291 if (ret != 0) {
2292 /*
2293 * If we failed here, it means ca->kobj is not initialized yet,
2294 * kobject_put() won't be called and there is no chance to
2295 * call blkdev_put() to bdev in bch_cache_release(). So we
2296 * explicitly call blkdev_put() here.
2297 */
2298 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2299 if (ret == -ENOMEM)
2300 err = "cache_alloc(): -ENOMEM";
2301 else if (ret == -EPERM)
2302 err = "cache_alloc(): cache device is too small";
2303 else
2304 err = "cache_alloc(): unknown error";
2305 goto err;
2306 }
2307
2308 if (kobject_add(&ca->kobj,
2309 &part_to_dev(bdev->bd_part)->kobj,
2310 "bcache")) {
2311 err = "error calling kobject_add";
2312 ret = -ENOMEM;
2313 goto out;
2314 }
2315
2316 mutex_lock(&bch_register_lock);
2317 err = register_cache_set(ca);
2318 mutex_unlock(&bch_register_lock);
2319
2320 if (err) {
2321 ret = -ENODEV;
2322 goto out;
2323 }
2324
2325 pr_info("registered cache device %s", ca->cache_dev_name);
2326
2327 out:
2328 kobject_put(&ca->kobj);
2329
2330 err:
2331 if (err)
2332 pr_notice("error %s: %s", ca->cache_dev_name, err);
2333
2334 return ret;
2335 }
2336
2337 /* Global interfaces/init */
2338
2339 static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr,
2340 const char *buffer, size_t size);
2341 static ssize_t bch_pending_bdevs_cleanup(struct kobject *k,
2342 struct kobj_attribute *attr,
2343 const char *buffer, size_t size);
2344
2345 kobj_attribute_write(register, register_bcache);
2346 kobj_attribute_write(register_quiet, register_bcache);
2347 kobj_attribute_write(pendings_cleanup, bch_pending_bdevs_cleanup);
2348
2349 static bool bch_is_open_backing(struct block_device *bdev)
2350 {
2351 struct cache_set *c, *tc;
2352 struct cached_dev *dc, *t;
2353
2354 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
2355 list_for_each_entry_safe(dc, t, &c->cached_devs, list)
2356 if (dc->bdev == bdev)
2357 return true;
2358 list_for_each_entry_safe(dc, t, &uncached_devices, list)
2359 if (dc->bdev == bdev)
2360 return true;
2361 return false;
2362 }
2363
2364 static bool bch_is_open_cache(struct block_device *bdev)
2365 {
2366 struct cache_set *c, *tc;
2367 struct cache *ca;
2368 unsigned int i;
2369
2370 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
2371 for_each_cache(ca, c, i)
2372 if (ca->bdev == bdev)
2373 return true;
2374 return false;
2375 }
2376
2377 static bool bch_is_open(struct block_device *bdev)
2378 {
2379 return bch_is_open_cache(bdev) || bch_is_open_backing(bdev);
2380 }
2381
2382 static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr,
2383 const char *buffer, size_t size)
2384 {
2385 const char *err;
2386 char *path = NULL;
2387 struct cache_sb *sb;
2388 struct block_device *bdev = NULL;
2389 struct page *sb_page;
2390 ssize_t ret;
2391
2392 ret = -EBUSY;
2393 err = "failed to reference bcache module";
2394 if (!try_module_get(THIS_MODULE))
2395 goto out;
2396
2397 /* For latest state of bcache_is_reboot */
2398 smp_mb();
2399 err = "bcache is in reboot";
2400 if (bcache_is_reboot)
2401 goto out_module_put;
2402
2403 ret = -ENOMEM;
2404 err = "cannot allocate memory";
2405 path = kstrndup(buffer, size, GFP_KERNEL);
2406 if (!path)
2407 goto out_module_put;
2408
2409 sb = kmalloc(sizeof(struct cache_sb), GFP_KERNEL);
2410 if (!sb)
2411 goto out_free_path;
2412
2413 ret = -EINVAL;
2414 err = "failed to open device";
2415 bdev = blkdev_get_by_path(strim(path),
2416 FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2417 sb);
2418 if (IS_ERR(bdev)) {
2419 if (bdev == ERR_PTR(-EBUSY)) {
2420 bdev = lookup_bdev(strim(path));
2421 mutex_lock(&bch_register_lock);
2422 if (!IS_ERR(bdev) && bch_is_open(bdev))
2423 err = "device already registered";
2424 else
2425 err = "device busy";
2426 mutex_unlock(&bch_register_lock);
2427 if (!IS_ERR(bdev))
2428 bdput(bdev);
2429 if (attr == &ksysfs_register_quiet)
2430 goto done;
2431 }
2432 goto out_free_sb;
2433 }
2434
2435 err = "failed to set blocksize";
2436 if (set_blocksize(bdev, 4096))
2437 goto out_blkdev_put;
2438
2439 err = read_super(sb, bdev, &sb_page);
2440 if (err)
2441 goto out_blkdev_put;
2442
2443 err = "failed to register device";
2444 if (SB_IS_BDEV(sb)) {
2445 struct cached_dev *dc = kzalloc(sizeof(*dc), GFP_KERNEL);
2446
2447 if (!dc)
2448 goto out_put_sb_page;
2449
2450 mutex_lock(&bch_register_lock);
2451 ret = register_bdev(sb, sb_page, bdev, dc);
2452 mutex_unlock(&bch_register_lock);
2453 /* blkdev_put() will be called in cached_dev_free() */
2454 if (ret < 0) {
2455 bdev = NULL;
2456 goto out_put_sb_page;
2457 }
2458 } else {
2459 struct cache *ca = kzalloc(sizeof(*ca), GFP_KERNEL);
2460
2461 if (!ca)
2462 goto out_put_sb_page;
2463
2464 /* blkdev_put() will be called in bch_cache_release() */
2465 if (register_cache(sb, sb_page, bdev, ca) != 0) {
2466 bdev = NULL;
2467 goto out_put_sb_page;
2468 }
2469 }
2470
2471 put_page(sb_page);
2472 done:
2473 kfree(sb);
2474 kfree(path);
2475 module_put(THIS_MODULE);
2476 return size;
2477
2478 out_put_sb_page:
2479 put_page(sb_page);
2480 out_blkdev_put:
2481 if (bdev)
2482 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2483 out_free_sb:
2484 kfree(sb);
2485 out_free_path:
2486 kfree(path);
2487 path = NULL;
2488 out_module_put:
2489 module_put(THIS_MODULE);
2490 out:
2491 pr_info("error %s: %s", path?path:"", err);
2492 return ret;
2493 }
2494
2495
2496 struct pdev {
2497 struct list_head list;
2498 struct cached_dev *dc;
2499 };
2500
2501 static ssize_t bch_pending_bdevs_cleanup(struct kobject *k,
2502 struct kobj_attribute *attr,
2503 const char *buffer,
2504 size_t size)
2505 {
2506 LIST_HEAD(pending_devs);
2507 ssize_t ret = size;
2508 struct cached_dev *dc, *tdc;
2509 struct pdev *pdev, *tpdev;
2510 struct cache_set *c, *tc;
2511
2512 mutex_lock(&bch_register_lock);
2513 list_for_each_entry_safe(dc, tdc, &uncached_devices, list) {
2514 pdev = kmalloc(sizeof(struct pdev), GFP_KERNEL);
2515 if (!pdev)
2516 break;
2517 pdev->dc = dc;
2518 list_add(&pdev->list, &pending_devs);
2519 }
2520
2521 list_for_each_entry_safe(pdev, tpdev, &pending_devs, list) {
2522 list_for_each_entry_safe(c, tc, &bch_cache_sets, list) {
2523 char *pdev_set_uuid = pdev->dc->sb.set_uuid;
2524 char *set_uuid = c->sb.uuid;
2525
2526 if (!memcmp(pdev_set_uuid, set_uuid, 16)) {
2527 list_del(&pdev->list);
2528 kfree(pdev);
2529 break;
2530 }
2531 }
2532 }
2533 mutex_unlock(&bch_register_lock);
2534
2535 list_for_each_entry_safe(pdev, tpdev, &pending_devs, list) {
2536 pr_info("delete pdev %p", pdev);
2537 list_del(&pdev->list);
2538 bcache_device_stop(&pdev->dc->disk);
2539 kfree(pdev);
2540 }
2541
2542 return ret;
2543 }
2544
2545 static int bcache_reboot(struct notifier_block *n, unsigned long code, void *x)
2546 {
2547 if (bcache_is_reboot)
2548 return NOTIFY_DONE;
2549
2550 if (code == SYS_DOWN ||
2551 code == SYS_HALT ||
2552 code == SYS_POWER_OFF) {
2553 DEFINE_WAIT(wait);
2554 unsigned long start = jiffies;
2555 bool stopped = false;
2556
2557 struct cache_set *c, *tc;
2558 struct cached_dev *dc, *tdc;
2559
2560 mutex_lock(&bch_register_lock);
2561
2562 if (bcache_is_reboot)
2563 goto out;
2564
2565 /* New registration is rejected since now */
2566 bcache_is_reboot = true;
2567 /*
2568 * Make registering caller (if there is) on other CPU
2569 * core know bcache_is_reboot set to true earlier
2570 */
2571 smp_mb();
2572
2573 if (list_empty(&bch_cache_sets) &&
2574 list_empty(&uncached_devices))
2575 goto out;
2576
2577 mutex_unlock(&bch_register_lock);
2578
2579 pr_info("Stopping all devices:");
2580
2581 /*
2582 * The reason bch_register_lock is not held to call
2583 * bch_cache_set_stop() and bcache_device_stop() is to
2584 * avoid potential deadlock during reboot, because cache
2585 * set or bcache device stopping process will acqurie
2586 * bch_register_lock too.
2587 *
2588 * We are safe here because bcache_is_reboot sets to
2589 * true already, register_bcache() will reject new
2590 * registration now. bcache_is_reboot also makes sure
2591 * bcache_reboot() won't be re-entered on by other thread,
2592 * so there is no race in following list iteration by
2593 * list_for_each_entry_safe().
2594 */
2595 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
2596 bch_cache_set_stop(c);
2597
2598 list_for_each_entry_safe(dc, tdc, &uncached_devices, list)
2599 bcache_device_stop(&dc->disk);
2600
2601
2602 /*
2603 * Give an early chance for other kthreads and
2604 * kworkers to stop themselves
2605 */
2606 schedule();
2607
2608 /* What's a condition variable? */
2609 while (1) {
2610 long timeout = start + 10 * HZ - jiffies;
2611
2612 mutex_lock(&bch_register_lock);
2613 stopped = list_empty(&bch_cache_sets) &&
2614 list_empty(&uncached_devices);
2615
2616 if (timeout < 0 || stopped)
2617 break;
2618
2619 prepare_to_wait(&unregister_wait, &wait,
2620 TASK_UNINTERRUPTIBLE);
2621
2622 mutex_unlock(&bch_register_lock);
2623 schedule_timeout(timeout);
2624 }
2625
2626 finish_wait(&unregister_wait, &wait);
2627
2628 if (stopped)
2629 pr_info("All devices stopped");
2630 else
2631 pr_notice("Timeout waiting for devices to be closed");
2632 out:
2633 mutex_unlock(&bch_register_lock);
2634 }
2635
2636 return NOTIFY_DONE;
2637 }
2638
2639 static struct notifier_block reboot = {
2640 .notifier_call = bcache_reboot,
2641 .priority = INT_MAX, /* before any real devices */
2642 };
2643
2644 static void bcache_exit(void)
2645 {
2646 bch_debug_exit();
2647 bch_request_exit();
2648 if (bcache_kobj)
2649 kobject_put(bcache_kobj);
2650 if (bcache_wq)
2651 destroy_workqueue(bcache_wq);
2652 if (bch_journal_wq)
2653 destroy_workqueue(bch_journal_wq);
2654
2655 if (bcache_major)
2656 unregister_blkdev(bcache_major, "bcache");
2657 unregister_reboot_notifier(&reboot);
2658 mutex_destroy(&bch_register_lock);
2659 }
2660
2661 /* Check and fixup module parameters */
2662 static void check_module_parameters(void)
2663 {
2664 if (bch_cutoff_writeback_sync == 0)
2665 bch_cutoff_writeback_sync = CUTOFF_WRITEBACK_SYNC;
2666 else if (bch_cutoff_writeback_sync > CUTOFF_WRITEBACK_SYNC_MAX) {
2667 pr_warn("set bch_cutoff_writeback_sync (%u) to max value %u",
2668 bch_cutoff_writeback_sync, CUTOFF_WRITEBACK_SYNC_MAX);
2669 bch_cutoff_writeback_sync = CUTOFF_WRITEBACK_SYNC_MAX;
2670 }
2671
2672 if (bch_cutoff_writeback == 0)
2673 bch_cutoff_writeback = CUTOFF_WRITEBACK;
2674 else if (bch_cutoff_writeback > CUTOFF_WRITEBACK_MAX) {
2675 pr_warn("set bch_cutoff_writeback (%u) to max value %u",
2676 bch_cutoff_writeback, CUTOFF_WRITEBACK_MAX);
2677 bch_cutoff_writeback = CUTOFF_WRITEBACK_MAX;
2678 }
2679
2680 if (bch_cutoff_writeback > bch_cutoff_writeback_sync) {
2681 pr_warn("set bch_cutoff_writeback (%u) to %u",
2682 bch_cutoff_writeback, bch_cutoff_writeback_sync);
2683 bch_cutoff_writeback = bch_cutoff_writeback_sync;
2684 }
2685 }
2686
2687 static int __init bcache_init(void)
2688 {
2689 static const struct attribute *files[] = {
2690 &ksysfs_register.attr,
2691 &ksysfs_register_quiet.attr,
2692 &ksysfs_pendings_cleanup.attr,
2693 NULL
2694 };
2695
2696 check_module_parameters();
2697
2698 mutex_init(&bch_register_lock);
2699 init_waitqueue_head(&unregister_wait);
2700 register_reboot_notifier(&reboot);
2701
2702 bcache_major = register_blkdev(0, "bcache");
2703 if (bcache_major < 0) {
2704 unregister_reboot_notifier(&reboot);
2705 mutex_destroy(&bch_register_lock);
2706 return bcache_major;
2707 }
2708
2709 bcache_wq = alloc_workqueue("bcache", WQ_MEM_RECLAIM, 0);
2710 if (!bcache_wq)
2711 goto err;
2712
2713 bch_journal_wq = alloc_workqueue("bch_journal", WQ_MEM_RECLAIM, 0);
2714 if (!bch_journal_wq)
2715 goto err;
2716
2717 bcache_kobj = kobject_create_and_add("bcache", fs_kobj);
2718 if (!bcache_kobj)
2719 goto err;
2720
2721 if (bch_request_init() ||
2722 sysfs_create_files(bcache_kobj, files))
2723 goto err;
2724
2725 bch_debug_init();
2726 closure_debug_init();
2727
2728 bcache_is_reboot = false;
2729
2730 return 0;
2731 err:
2732 bcache_exit();
2733 return -ENOMEM;
2734 }
2735
2736 /*
2737 * Module hooks
2738 */
2739 module_exit(bcache_exit);
2740 module_init(bcache_init);
2741
2742 module_param(bch_cutoff_writeback, uint, 0);
2743 MODULE_PARM_DESC(bch_cutoff_writeback, "threshold to cutoff writeback");
2744
2745 module_param(bch_cutoff_writeback_sync, uint, 0);
2746 MODULE_PARM_DESC(bch_cutoff_writeback_sync, "hard threshold to cutoff writeback");
2747
2748 MODULE_DESCRIPTION("Bcache: a Linux block layer cache");
2749 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
2750 MODULE_LICENSE("GPL");
Linux with added FPC-III support