?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
treewide: remove redundant IS_ERR() before error code check
[linux/fpc-iii.git] / drivers / md / bcache / super.c
blob3dea1d5acd5c395effae275a45b364ba70a16b04
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/debugfs.h>
19 #include <linux/genhd.h>
20 #include <linux/idr.h>
21 #include <linux/kthread.h>
22 #include <linux/module.h>
23 #include <linux/random.h>
24 #include <linux/reboot.h>
25 #include <linux/sysfs.h>
26
27 unsigned int bch_cutoff_writeback;
28 unsigned int bch_cutoff_writeback_sync;
29
30 static const char bcache_magic[] = {
31 0xc6, 0x85, 0x73, 0xf6, 0x4e, 0x1a, 0x45, 0xca,
32 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81
33 };
34
35 static 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 static struct kobject *bcache_kobj;
41 struct mutex bch_register_lock;
42 bool bcache_is_reboot;
43 LIST_HEAD(bch_cache_sets);
44 static LIST_HEAD(uncached_devices);
45
46 static int bcache_major;
47 static DEFINE_IDA(bcache_device_idx);
48 static wait_queue_head_t unregister_wait;
49 struct workqueue_struct *bcache_wq;
50 struct workqueue_struct *bch_journal_wq;
51
52
53 #define BTREE_MAX_PAGES (256 * 1024 / PAGE_SIZE)
54 /* limitation of partitions number on single bcache device */
55 #define BCACHE_MINORS 128
56 /* limitation of bcache devices number on single system */
57 #define BCACHE_DEVICE_IDX_MAX ((1U << MINORBITS)/BCACHE_MINORS)
58
59 /* Superblock */
60
61 static const char *read_super(struct cache_sb *sb, struct block_device *bdev,
62 struct cache_sb_disk **res)
63 {
64 const char *err;
65 struct cache_sb_disk *s;
66 struct page *page;
67 unsigned int i;
68
69 page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
70 SB_OFFSET >> PAGE_SHIFT, GFP_KERNEL);
71 if (IS_ERR(page))
72 return "IO error";
73 s = page_address(page) + offset_in_page(SB_OFFSET);
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 (bad offset)";
96 if (sb->offset != SB_SECTOR)
97 goto err;
98
99 err = "Not a bcache superblock (bad magic)";
100 if (memcmp(sb->magic, bcache_magic, 16))
101 goto err;
102
103 err = "Too many journal buckets";
104 if (sb->keys > SB_JOURNAL_BUCKETS)
105 goto err;
106
107 err = "Bad checksum";
108 if (s->csum != csum_set(s))
109 goto err;
110
111 err = "Bad UUID";
112 if (bch_is_zero(sb->uuid, 16))
113 goto err;
114
115 sb->block_size = le16_to_cpu(s->block_size);
116
117 err = "Superblock block size smaller than device block size";
118 if (sb->block_size << 9 < bdev_logical_block_size(bdev))
119 goto err;
120
121 switch (sb->version) {
122 case BCACHE_SB_VERSION_BDEV:
123 sb->data_offset = BDEV_DATA_START_DEFAULT;
124 break;
125 case BCACHE_SB_VERSION_BDEV_WITH_OFFSET:
126 sb->data_offset = le64_to_cpu(s->data_offset);
127
128 err = "Bad data offset";
129 if (sb->data_offset < BDEV_DATA_START_DEFAULT)
130 goto err;
131
132 break;
133 case BCACHE_SB_VERSION_CDEV:
134 case BCACHE_SB_VERSION_CDEV_WITH_UUID:
135 sb->nbuckets = le64_to_cpu(s->nbuckets);
136 sb->bucket_size = le16_to_cpu(s->bucket_size);
137
138 sb->nr_in_set = le16_to_cpu(s->nr_in_set);
139 sb->nr_this_dev = le16_to_cpu(s->nr_this_dev);
140
141 err = "Too many buckets";
142 if (sb->nbuckets > LONG_MAX)
143 goto err;
144
145 err = "Not enough buckets";
146 if (sb->nbuckets < 1 << 7)
147 goto err;
148
149 err = "Bad block/bucket size";
150 if (!is_power_of_2(sb->block_size) ||
151 sb->block_size > PAGE_SECTORS ||
152 !is_power_of_2(sb->bucket_size) ||
153 sb->bucket_size < PAGE_SECTORS)
154 goto err;
155
156 err = "Invalid superblock: device too small";
157 if (get_capacity(bdev->bd_disk) <
158 sb->bucket_size * sb->nbuckets)
159 goto err;
160
161 err = "Bad UUID";
162 if (bch_is_zero(sb->set_uuid, 16))
163 goto err;
164
165 err = "Bad cache device number in set";
166 if (!sb->nr_in_set ||
167 sb->nr_in_set <= sb->nr_this_dev ||
168 sb->nr_in_set > MAX_CACHES_PER_SET)
169 goto err;
170
171 err = "Journal buckets not sequential";
172 for (i = 0; i < sb->keys; i++)
173 if (sb->d[i] != sb->first_bucket + i)
174 goto err;
175
176 err = "Too many journal buckets";
177 if (sb->first_bucket + sb->keys > sb->nbuckets)
178 goto err;
179
180 err = "Invalid superblock: first bucket comes before end of super";
181 if (sb->first_bucket * sb->bucket_size < 16)
182 goto err;
183
184 break;
185 default:
186 err = "Unsupported superblock version";
187 goto err;
188 }
189
190 sb->last_mount = (u32)ktime_get_real_seconds();
191 *res = s;
192 return NULL;
193 err:
194 put_page(page);
195 return err;
196 }
197
198 static void write_bdev_super_endio(struct bio *bio)
199 {
200 struct cached_dev *dc = bio->bi_private;
201
202 if (bio->bi_status)
203 bch_count_backing_io_errors(dc, bio);
204
205 closure_put(&dc->sb_write);
206 }
207
208 static void __write_super(struct cache_sb *sb, struct cache_sb_disk *out,
209 struct bio *bio)
210 {
211 unsigned int i;
212
213 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META;
214 bio->bi_iter.bi_sector = SB_SECTOR;
215 __bio_add_page(bio, virt_to_page(out), SB_SIZE,
216 offset_in_page(out));
217
218 out->offset = cpu_to_le64(sb->offset);
219 out->version = cpu_to_le64(sb->version);
220
221 memcpy(out->uuid, sb->uuid, 16);
222 memcpy(out->set_uuid, sb->set_uuid, 16);
223 memcpy(out->label, sb->label, SB_LABEL_SIZE);
224
225 out->flags = cpu_to_le64(sb->flags);
226 out->seq = cpu_to_le64(sb->seq);
227
228 out->last_mount = cpu_to_le32(sb->last_mount);
229 out->first_bucket = cpu_to_le16(sb->first_bucket);
230 out->keys = cpu_to_le16(sb->keys);
231
232 for (i = 0; i < sb->keys; i++)
233 out->d[i] = cpu_to_le64(sb->d[i]);
234
235 out->csum = csum_set(out);
236
237 pr_debug("ver %llu, flags %llu, seq %llu",
238 sb->version, sb->flags, sb->seq);
239
240 submit_bio(bio);
241 }
242
243 static void bch_write_bdev_super_unlock(struct closure *cl)
244 {
245 struct cached_dev *dc = container_of(cl, struct cached_dev, sb_write);
246
247 up(&dc->sb_write_mutex);
248 }
249
250 void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent)
251 {
252 struct closure *cl = &dc->sb_write;
253 struct bio *bio = &dc->sb_bio;
254
255 down(&dc->sb_write_mutex);
256 closure_init(cl, parent);
257
258 bio_init(bio, dc->sb_bv, 1);
259 bio_set_dev(bio, dc->bdev);
260 bio->bi_end_io = write_bdev_super_endio;
261 bio->bi_private = dc;
262
263 closure_get(cl);
264 /* I/O request sent to backing device */
265 __write_super(&dc->sb, dc->sb_disk, bio);
266
267 closure_return_with_destructor(cl, bch_write_bdev_super_unlock);
268 }
269
270 static void write_super_endio(struct bio *bio)
271 {
272 struct cache *ca = bio->bi_private;
273
274 /* is_read = 0 */
275 bch_count_io_errors(ca, bio->bi_status, 0,
276 "writing superblock");
277 closure_put(&ca->set->sb_write);
278 }
279
280 static void bcache_write_super_unlock(struct closure *cl)
281 {
282 struct cache_set *c = container_of(cl, struct cache_set, sb_write);
283
284 up(&c->sb_write_mutex);
285 }
286
287 void bcache_write_super(struct cache_set *c)
288 {
289 struct closure *cl = &c->sb_write;
290 struct cache *ca;
291 unsigned int i;
292
293 down(&c->sb_write_mutex);
294 closure_init(cl, &c->cl);
295
296 c->sb.seq++;
297
298 for_each_cache(ca, c, i) {
299 struct bio *bio = &ca->sb_bio;
300
301 ca->sb.version = BCACHE_SB_VERSION_CDEV_WITH_UUID;
302 ca->sb.seq = c->sb.seq;
303 ca->sb.last_mount = c->sb.last_mount;
304
305 SET_CACHE_SYNC(&ca->sb, CACHE_SYNC(&c->sb));
306
307 bio_init(bio, ca->sb_bv, 1);
308 bio_set_dev(bio, ca->bdev);
309 bio->bi_end_io = write_super_endio;
310 bio->bi_private = ca;
311
312 closure_get(cl);
313 __write_super(&ca->sb, ca->sb_disk, bio);
314 }
315
316 closure_return_with_destructor(cl, bcache_write_super_unlock);
317 }
318
319 /* UUID io */
320
321 static void uuid_endio(struct bio *bio)
322 {
323 struct closure *cl = bio->bi_private;
324 struct cache_set *c = container_of(cl, struct cache_set, uuid_write);
325
326 cache_set_err_on(bio->bi_status, c, "accessing uuids");
327 bch_bbio_free(bio, c);
328 closure_put(cl);
329 }
330
331 static void uuid_io_unlock(struct closure *cl)
332 {
333 struct cache_set *c = container_of(cl, struct cache_set, uuid_write);
334
335 up(&c->uuid_write_mutex);
336 }
337
338 static void uuid_io(struct cache_set *c, int op, unsigned long op_flags,
339 struct bkey *k, struct closure *parent)
340 {
341 struct closure *cl = &c->uuid_write;
342 struct uuid_entry *u;
343 unsigned int i;
344 char buf[80];
345
346 BUG_ON(!parent);
347 down(&c->uuid_write_mutex);
348 closure_init(cl, parent);
349
350 for (i = 0; i < KEY_PTRS(k); i++) {
351 struct bio *bio = bch_bbio_alloc(c);
352
353 bio->bi_opf = REQ_SYNC | REQ_META | op_flags;
354 bio->bi_iter.bi_size = KEY_SIZE(k) << 9;
355
356 bio->bi_end_io = uuid_endio;
357 bio->bi_private = cl;
358 bio_set_op_attrs(bio, op, REQ_SYNC|REQ_META|op_flags);
359 bch_bio_map(bio, c->uuids);
360
361 bch_submit_bbio(bio, c, k, i);
362
363 if (op != REQ_OP_WRITE)
364 break;
365 }
366
367 bch_extent_to_text(buf, sizeof(buf), k);
368 pr_debug("%s UUIDs at %s", op == REQ_OP_WRITE ? "wrote" : "read", buf);
369
370 for (u = c->uuids; u < c->uuids + c->nr_uuids; u++)
371 if (!bch_is_zero(u->uuid, 16))
372 pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u",
373 u - c->uuids, u->uuid, u->label,
374 u->first_reg, u->last_reg, u->invalidated);
375
376 closure_return_with_destructor(cl, uuid_io_unlock);
377 }
378
379 static char *uuid_read(struct cache_set *c, struct jset *j, struct closure *cl)
380 {
381 struct bkey *k = &j->uuid_bucket;
382
383 if (__bch_btree_ptr_invalid(c, k))
384 return "bad uuid pointer";
385
386 bkey_copy(&c->uuid_bucket, k);
387 uuid_io(c, REQ_OP_READ, 0, k, cl);
388
389 if (j->version < BCACHE_JSET_VERSION_UUIDv1) {
390 struct uuid_entry_v0 *u0 = (void *) c->uuids;
391 struct uuid_entry *u1 = (void *) c->uuids;
392 int i;
393
394 closure_sync(cl);
395
396 /*
397 * Since the new uuid entry is bigger than the old, we have to
398 * convert starting at the highest memory address and work down
399 * in order to do it in place
400 */
401
402 for (i = c->nr_uuids - 1;
403 i >= 0;
404 --i) {
405 memcpy(u1[i].uuid, u0[i].uuid, 16);
406 memcpy(u1[i].label, u0[i].label, 32);
407
408 u1[i].first_reg = u0[i].first_reg;
409 u1[i].last_reg = u0[i].last_reg;
410 u1[i].invalidated = u0[i].invalidated;
411
412 u1[i].flags = 0;
413 u1[i].sectors = 0;
414 }
415 }
416
417 return NULL;
418 }
419
420 static int __uuid_write(struct cache_set *c)
421 {
422 BKEY_PADDED(key) k;
423 struct closure cl;
424 struct cache *ca;
425
426 closure_init_stack(&cl);
427 lockdep_assert_held(&bch_register_lock);
428
429 if (bch_bucket_alloc_set(c, RESERVE_BTREE, &k.key, 1, true))
430 return 1;
431
432 SET_KEY_SIZE(&k.key, c->sb.bucket_size);
433 uuid_io(c, REQ_OP_WRITE, 0, &k.key, &cl);
434 closure_sync(&cl);
435
436 /* Only one bucket used for uuid write */
437 ca = PTR_CACHE(c, &k.key, 0);
438 atomic_long_add(ca->sb.bucket_size, &ca->meta_sectors_written);
439
440 bkey_copy(&c->uuid_bucket, &k.key);
441 bkey_put(c, &k.key);
442 return 0;
443 }
444
445 int bch_uuid_write(struct cache_set *c)
446 {
447 int ret = __uuid_write(c);
448
449 if (!ret)
450 bch_journal_meta(c, NULL);
451
452 return ret;
453 }
454
455 static struct uuid_entry *uuid_find(struct cache_set *c, const char *uuid)
456 {
457 struct uuid_entry *u;
458
459 for (u = c->uuids;
460 u < c->uuids + c->nr_uuids; u++)
461 if (!memcmp(u->uuid, uuid, 16))
462 return u;
463
464 return NULL;
465 }
466
467 static struct uuid_entry *uuid_find_empty(struct cache_set *c)
468 {
469 static const char zero_uuid[16] = "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0";
470
471 return uuid_find(c, zero_uuid);
472 }
473
474 /*
475 * Bucket priorities/gens:
476 *
477 * For each bucket, we store on disk its
478 * 8 bit gen
479 * 16 bit priority
480 *
481 * See alloc.c for an explanation of the gen. The priority is used to implement
482 * lru (and in the future other) cache replacement policies; for most purposes
483 * it's just an opaque integer.
484 *
485 * The gens and the priorities don't have a whole lot to do with each other, and
486 * it's actually the gens that must be written out at specific times - it's no
487 * big deal if the priorities don't get written, if we lose them we just reuse
488 * buckets in suboptimal order.
489 *
490 * On disk they're stored in a packed array, and in as many buckets are required
491 * to fit them all. The buckets we use to store them form a list; the journal
492 * header points to the first bucket, the first bucket points to the second
493 * bucket, et cetera.
494 *
495 * This code is used by the allocation code; periodically (whenever it runs out
496 * of buckets to allocate from) the allocation code will invalidate some
497 * buckets, but it can't use those buckets until their new gens are safely on
498 * disk.
499 */
500
501 static void prio_endio(struct bio *bio)
502 {
503 struct cache *ca = bio->bi_private;
504
505 cache_set_err_on(bio->bi_status, ca->set, "accessing priorities");
506 bch_bbio_free(bio, ca->set);
507 closure_put(&ca->prio);
508 }
509
510 static void prio_io(struct cache *ca, uint64_t bucket, int op,
511 unsigned long op_flags)
512 {
513 struct closure *cl = &ca->prio;
514 struct bio *bio = bch_bbio_alloc(ca->set);
515
516 closure_init_stack(cl);
517
518 bio->bi_iter.bi_sector = bucket * ca->sb.bucket_size;
519 bio_set_dev(bio, ca->bdev);
520 bio->bi_iter.bi_size = bucket_bytes(ca);
521
522 bio->bi_end_io = prio_endio;
523 bio->bi_private = ca;
524 bio_set_op_attrs(bio, op, REQ_SYNC|REQ_META|op_flags);
525 bch_bio_map(bio, ca->disk_buckets);
526
527 closure_bio_submit(ca->set, bio, &ca->prio);
528 closure_sync(cl);
529 }
530
531 int bch_prio_write(struct cache *ca, bool wait)
532 {
533 int i;
534 struct bucket *b;
535 struct closure cl;
536
537 pr_debug("free_prio=%zu, free_none=%zu, free_inc=%zu",
538 fifo_used(&ca->free[RESERVE_PRIO]),
539 fifo_used(&ca->free[RESERVE_NONE]),
540 fifo_used(&ca->free_inc));
541
542 /*
543 * Pre-check if there are enough free buckets. In the non-blocking
544 * scenario it's better to fail early rather than starting to allocate
545 * buckets and do a cleanup later in case of failure.
546 */
547 if (!wait) {
548 size_t avail = fifo_used(&ca->free[RESERVE_PRIO]) +
549 fifo_used(&ca->free[RESERVE_NONE]);
550 if (prio_buckets(ca) > avail)
551 return -ENOMEM;
552 }
553
554 closure_init_stack(&cl);
555
556 lockdep_assert_held(&ca->set->bucket_lock);
557
558 ca->disk_buckets->seq++;
559
560 atomic_long_add(ca->sb.bucket_size * prio_buckets(ca),
561 &ca->meta_sectors_written);
562
563 for (i = prio_buckets(ca) - 1; i >= 0; --i) {
564 long bucket;
565 struct prio_set *p = ca->disk_buckets;
566 struct bucket_disk *d = p->data;
567 struct bucket_disk *end = d + prios_per_bucket(ca);
568
569 for (b = ca->buckets + i * prios_per_bucket(ca);
570 b < ca->buckets + ca->sb.nbuckets && d < end;
571 b++, d++) {
572 d->prio = cpu_to_le16(b->prio);
573 d->gen = b->gen;
574 }
575
576 p->next_bucket = ca->prio_buckets[i + 1];
577 p->magic = pset_magic(&ca->sb);
578 p->csum = bch_crc64(&p->magic, bucket_bytes(ca) - 8);
579
580 bucket = bch_bucket_alloc(ca, RESERVE_PRIO, wait);
581 BUG_ON(bucket == -1);
582
583 mutex_unlock(&ca->set->bucket_lock);
584 prio_io(ca, bucket, REQ_OP_WRITE, 0);
585 mutex_lock(&ca->set->bucket_lock);
586
587 ca->prio_buckets[i] = bucket;
588 atomic_dec_bug(&ca->buckets[bucket].pin);
589 }
590
591 mutex_unlock(&ca->set->bucket_lock);
592
593 bch_journal_meta(ca->set, &cl);
594 closure_sync(&cl);
595
596 mutex_lock(&ca->set->bucket_lock);
597
598 /*
599 * Don't want the old priorities to get garbage collected until after we
600 * finish writing the new ones, and they're journalled
601 */
602 for (i = 0; i < prio_buckets(ca); i++) {
603 if (ca->prio_last_buckets[i])
604 __bch_bucket_free(ca,
605 &ca->buckets[ca->prio_last_buckets[i]]);
606
607 ca->prio_last_buckets[i] = ca->prio_buckets[i];
608 }
609 return 0;
610 }
611
612 static void prio_read(struct cache *ca, uint64_t bucket)
613 {
614 struct prio_set *p = ca->disk_buckets;
615 struct bucket_disk *d = p->data + prios_per_bucket(ca), *end = d;
616 struct bucket *b;
617 unsigned int bucket_nr = 0;
618
619 for (b = ca->buckets;
620 b < ca->buckets + ca->sb.nbuckets;
621 b++, d++) {
622 if (d == end) {
623 ca->prio_buckets[bucket_nr] = bucket;
624 ca->prio_last_buckets[bucket_nr] = bucket;
625 bucket_nr++;
626
627 prio_io(ca, bucket, REQ_OP_READ, 0);
628
629 if (p->csum !=
630 bch_crc64(&p->magic, bucket_bytes(ca) - 8))
631 pr_warn("bad csum reading priorities");
632
633 if (p->magic != pset_magic(&ca->sb))
634 pr_warn("bad magic reading priorities");
635
636 bucket = p->next_bucket;
637 d = p->data;
638 }
639
640 b->prio = le16_to_cpu(d->prio);
641 b->gen = b->last_gc = d->gen;
642 }
643 }
644
645 /* Bcache device */
646
647 static int open_dev(struct block_device *b, fmode_t mode)
648 {
649 struct bcache_device *d = b->bd_disk->private_data;
650
651 if (test_bit(BCACHE_DEV_CLOSING, &d->flags))
652 return -ENXIO;
653
654 closure_get(&d->cl);
655 return 0;
656 }
657
658 static void release_dev(struct gendisk *b, fmode_t mode)
659 {
660 struct bcache_device *d = b->private_data;
661
662 closure_put(&d->cl);
663 }
664
665 static int ioctl_dev(struct block_device *b, fmode_t mode,
666 unsigned int cmd, unsigned long arg)
667 {
668 struct bcache_device *d = b->bd_disk->private_data;
669
670 return d->ioctl(d, mode, cmd, arg);
671 }
672
673 static const struct block_device_operations bcache_ops = {
674 .open = open_dev,
675 .release = release_dev,
676 .ioctl = ioctl_dev,
677 .owner = THIS_MODULE,
678 };
679
680 void bcache_device_stop(struct bcache_device *d)
681 {
682 if (!test_and_set_bit(BCACHE_DEV_CLOSING, &d->flags))
683 /*
684 * closure_fn set to
685 * - cached device: cached_dev_flush()
686 * - flash dev: flash_dev_flush()
687 */
688 closure_queue(&d->cl);
689 }
690
691 static void bcache_device_unlink(struct bcache_device *d)
692 {
693 lockdep_assert_held(&bch_register_lock);
694
695 if (d->c && !test_and_set_bit(BCACHE_DEV_UNLINK_DONE, &d->flags)) {
696 unsigned int i;
697 struct cache *ca;
698
699 sysfs_remove_link(&d->c->kobj, d->name);
700 sysfs_remove_link(&d->kobj, "cache");
701
702 for_each_cache(ca, d->c, i)
703 bd_unlink_disk_holder(ca->bdev, d->disk);
704 }
705 }
706
707 static void bcache_device_link(struct bcache_device *d, struct cache_set *c,
708 const char *name)
709 {
710 unsigned int i;
711 struct cache *ca;
712 int ret;
713
714 for_each_cache(ca, d->c, i)
715 bd_link_disk_holder(ca->bdev, d->disk);
716
717 snprintf(d->name, BCACHEDEVNAME_SIZE,
718 "%s%u", name, d->id);
719
720 ret = sysfs_create_link(&d->kobj, &c->kobj, "cache");
721 if (ret < 0)
722 pr_err("Couldn't create device -> cache set symlink");
723
724 ret = sysfs_create_link(&c->kobj, &d->kobj, d->name);
725 if (ret < 0)
726 pr_err("Couldn't create cache set -> device symlink");
727
728 clear_bit(BCACHE_DEV_UNLINK_DONE, &d->flags);
729 }
730
731 static void bcache_device_detach(struct bcache_device *d)
732 {
733 lockdep_assert_held(&bch_register_lock);
734
735 atomic_dec(&d->c->attached_dev_nr);
736
737 if (test_bit(BCACHE_DEV_DETACHING, &d->flags)) {
738 struct uuid_entry *u = d->c->uuids + d->id;
739
740 SET_UUID_FLASH_ONLY(u, 0);
741 memcpy(u->uuid, invalid_uuid, 16);
742 u->invalidated = cpu_to_le32((u32)ktime_get_real_seconds());
743 bch_uuid_write(d->c);
744 }
745
746 bcache_device_unlink(d);
747
748 d->c->devices[d->id] = NULL;
749 closure_put(&d->c->caching);
750 d->c = NULL;
751 }
752
753 static void bcache_device_attach(struct bcache_device *d, struct cache_set *c,
754 unsigned int id)
755 {
756 d->id = id;
757 d->c = c;
758 c->devices[id] = d;
759
760 if (id >= c->devices_max_used)
761 c->devices_max_used = id + 1;
762
763 closure_get(&c->caching);
764 }
765
766 static inline int first_minor_to_idx(int first_minor)
767 {
768 return (first_minor/BCACHE_MINORS);
769 }
770
771 static inline int idx_to_first_minor(int idx)
772 {
773 return (idx * BCACHE_MINORS);
774 }
775
776 static void bcache_device_free(struct bcache_device *d)
777 {
778 struct gendisk *disk = d->disk;
779
780 lockdep_assert_held(&bch_register_lock);
781
782 if (disk)
783 pr_info("%s stopped", disk->disk_name);
784 else
785 pr_err("bcache device (NULL gendisk) stopped");
786
787 if (d->c)
788 bcache_device_detach(d);
789
790 if (disk) {
791 if (disk->flags & GENHD_FL_UP)
792 del_gendisk(disk);
793
794 if (disk->queue)
795 blk_cleanup_queue(disk->queue);
796
797 ida_simple_remove(&bcache_device_idx,
798 first_minor_to_idx(disk->first_minor));
799 put_disk(disk);
800 }
801
802 bioset_exit(&d->bio_split);
803 kvfree(d->full_dirty_stripes);
804 kvfree(d->stripe_sectors_dirty);
805
806 closure_debug_destroy(&d->cl);
807 }
808
809 static int bcache_device_init(struct bcache_device *d, unsigned int block_size,
810 sector_t sectors)
811 {
812 struct request_queue *q;
813 const size_t max_stripes = min_t(size_t, INT_MAX,
814 SIZE_MAX / sizeof(atomic_t));
815 size_t n;
816 int idx;
817
818 if (!d->stripe_size)
819 d->stripe_size = 1 << 31;
820
821 d->nr_stripes = DIV_ROUND_UP_ULL(sectors, d->stripe_size);
822
823 if (!d->nr_stripes || d->nr_stripes > max_stripes) {
824 pr_err("nr_stripes too large or invalid: %u (start sector beyond end of disk?)",
825 (unsigned int)d->nr_stripes);
826 return -ENOMEM;
827 }
828
829 n = d->nr_stripes * sizeof(atomic_t);
830 d->stripe_sectors_dirty = kvzalloc(n, GFP_KERNEL);
831 if (!d->stripe_sectors_dirty)
832 return -ENOMEM;
833
834 n = BITS_TO_LONGS(d->nr_stripes) * sizeof(unsigned long);
835 d->full_dirty_stripes = kvzalloc(n, GFP_KERNEL);
836 if (!d->full_dirty_stripes)
837 return -ENOMEM;
838
839 idx = ida_simple_get(&bcache_device_idx, 0,
840 BCACHE_DEVICE_IDX_MAX, GFP_KERNEL);
841 if (idx < 0)
842 return idx;
843
844 if (bioset_init(&d->bio_split, 4, offsetof(struct bbio, bio),
845 BIOSET_NEED_BVECS|BIOSET_NEED_RESCUER))
846 goto err;
847
848 d->disk = alloc_disk(BCACHE_MINORS);
849 if (!d->disk)
850 goto err;
851
852 set_capacity(d->disk, sectors);
853 snprintf(d->disk->disk_name, DISK_NAME_LEN, "bcache%i", idx);
854
855 d->disk->major = bcache_major;
856 d->disk->first_minor = idx_to_first_minor(idx);
857 d->disk->fops = &bcache_ops;
858 d->disk->private_data = d;
859
860 q = blk_alloc_queue(GFP_KERNEL);
861 if (!q)
862 return -ENOMEM;
863
864 blk_queue_make_request(q, NULL);
865 d->disk->queue = q;
866 q->queuedata = d;
867 q->backing_dev_info->congested_data = d;
868 q->limits.max_hw_sectors = UINT_MAX;
869 q->limits.max_sectors = UINT_MAX;
870 q->limits.max_segment_size = UINT_MAX;
871 q->limits.max_segments = BIO_MAX_PAGES;
872 blk_queue_max_discard_sectors(q, UINT_MAX);
873 q->limits.discard_granularity = 512;
874 q->limits.io_min = block_size;
875 q->limits.logical_block_size = block_size;
876 q->limits.physical_block_size = block_size;
877 blk_queue_flag_set(QUEUE_FLAG_NONROT, d->disk->queue);
878 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, d->disk->queue);
879 blk_queue_flag_set(QUEUE_FLAG_DISCARD, d->disk->queue);
880
881 blk_queue_write_cache(q, true, true);
882
883 return 0;
884
885 err:
886 ida_simple_remove(&bcache_device_idx, idx);
887 return -ENOMEM;
888
889 }
890
891 /* Cached device */
892
893 static void calc_cached_dev_sectors(struct cache_set *c)
894 {
895 uint64_t sectors = 0;
896 struct cached_dev *dc;
897
898 list_for_each_entry(dc, &c->cached_devs, list)
899 sectors += bdev_sectors(dc->bdev);
900
901 c->cached_dev_sectors = sectors;
902 }
903
904 #define BACKING_DEV_OFFLINE_TIMEOUT 5
905 static int cached_dev_status_update(void *arg)
906 {
907 struct cached_dev *dc = arg;
908 struct request_queue *q;
909
910 /*
911 * If this delayed worker is stopping outside, directly quit here.
912 * dc->io_disable might be set via sysfs interface, so check it
913 * here too.
914 */
915 while (!kthread_should_stop() && !dc->io_disable) {
916 q = bdev_get_queue(dc->bdev);
917 if (blk_queue_dying(q))
918 dc->offline_seconds++;
919 else
920 dc->offline_seconds = 0;
921
922 if (dc->offline_seconds >= BACKING_DEV_OFFLINE_TIMEOUT) {
923 pr_err("%s: device offline for %d seconds",
924 dc->backing_dev_name,
925 BACKING_DEV_OFFLINE_TIMEOUT);
926 pr_err("%s: disable I/O request due to backing "
927 "device offline", dc->disk.name);
928 dc->io_disable = true;
929 /* let others know earlier that io_disable is true */
930 smp_mb();
931 bcache_device_stop(&dc->disk);
932 break;
933 }
934 schedule_timeout_interruptible(HZ);
935 }
936
937 wait_for_kthread_stop();
938 return 0;
939 }
940
941
942 int bch_cached_dev_run(struct cached_dev *dc)
943 {
944 struct bcache_device *d = &dc->disk;
945 char *buf = kmemdup_nul(dc->sb.label, SB_LABEL_SIZE, GFP_KERNEL);
946 char *env[] = {
947 "DRIVER=bcache",
948 kasprintf(GFP_KERNEL, "CACHED_UUID=%pU", dc->sb.uuid),
949 kasprintf(GFP_KERNEL, "CACHED_LABEL=%s", buf ? : ""),
950 NULL,
951 };
952
953 if (dc->io_disable) {
954 pr_err("I/O disabled on cached dev %s",
955 dc->backing_dev_name);
956 kfree(env[1]);
957 kfree(env[2]);
958 kfree(buf);
959 return -EIO;
960 }
961
962 if (atomic_xchg(&dc->running, 1)) {
963 kfree(env[1]);
964 kfree(env[2]);
965 kfree(buf);
966 pr_info("cached dev %s is running already",
967 dc->backing_dev_name);
968 return -EBUSY;
969 }
970
971 if (!d->c &&
972 BDEV_STATE(&dc->sb) != BDEV_STATE_NONE) {
973 struct closure cl;
974
975 closure_init_stack(&cl);
976
977 SET_BDEV_STATE(&dc->sb, BDEV_STATE_STALE);
978 bch_write_bdev_super(dc, &cl);
979 closure_sync(&cl);
980 }
981
982 add_disk(d->disk);
983 bd_link_disk_holder(dc->bdev, dc->disk.disk);
984 /*
985 * won't show up in the uevent file, use udevadm monitor -e instead
986 * only class / kset properties are persistent
987 */
988 kobject_uevent_env(&disk_to_dev(d->disk)->kobj, KOBJ_CHANGE, env);
989 kfree(env[1]);
990 kfree(env[2]);
991 kfree(buf);
992
993 if (sysfs_create_link(&d->kobj, &disk_to_dev(d->disk)->kobj, "dev") ||
994 sysfs_create_link(&disk_to_dev(d->disk)->kobj,
995 &d->kobj, "bcache")) {
996 pr_err("Couldn't create bcache dev <-> disk sysfs symlinks");
997 return -ENOMEM;
998 }
999
1000 dc->status_update_thread = kthread_run(cached_dev_status_update,
1001 dc, "bcache_status_update");
1002 if (IS_ERR(dc->status_update_thread)) {
1003 pr_warn("failed to create bcache_status_update kthread, "
1004 "continue to run without monitoring backing "
1005 "device status");
1006 }
1007
1008 return 0;
1009 }
1010
1011 /*
1012 * If BCACHE_DEV_RATE_DW_RUNNING is set, it means routine of the delayed
1013 * work dc->writeback_rate_update is running. Wait until the routine
1014 * quits (BCACHE_DEV_RATE_DW_RUNNING is clear), then continue to
1015 * cancel it. If BCACHE_DEV_RATE_DW_RUNNING is not clear after time_out
1016 * seconds, give up waiting here and continue to cancel it too.
1017 */
1018 static void cancel_writeback_rate_update_dwork(struct cached_dev *dc)
1019 {
1020 int time_out = WRITEBACK_RATE_UPDATE_SECS_MAX * HZ;
1021
1022 do {
1023 if (!test_bit(BCACHE_DEV_RATE_DW_RUNNING,
1024 &dc->disk.flags))
1025 break;
1026 time_out--;
1027 schedule_timeout_interruptible(1);
1028 } while (time_out > 0);
1029
1030 if (time_out == 0)
1031 pr_warn("give up waiting for dc->writeback_write_update to quit");
1032
1033 cancel_delayed_work_sync(&dc->writeback_rate_update);
1034 }
1035
1036 static void cached_dev_detach_finish(struct work_struct *w)
1037 {
1038 struct cached_dev *dc = container_of(w, struct cached_dev, detach);
1039 struct closure cl;
1040
1041 closure_init_stack(&cl);
1042
1043 BUG_ON(!test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags));
1044 BUG_ON(refcount_read(&dc->count));
1045
1046
1047 if (test_and_clear_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags))
1048 cancel_writeback_rate_update_dwork(dc);
1049
1050 if (!IS_ERR_OR_NULL(dc->writeback_thread)) {
1051 kthread_stop(dc->writeback_thread);
1052 dc->writeback_thread = NULL;
1053 }
1054
1055 memset(&dc->sb.set_uuid, 0, 16);
1056 SET_BDEV_STATE(&dc->sb, BDEV_STATE_NONE);
1057
1058 bch_write_bdev_super(dc, &cl);
1059 closure_sync(&cl);
1060
1061 mutex_lock(&bch_register_lock);
1062
1063 calc_cached_dev_sectors(dc->disk.c);
1064 bcache_device_detach(&dc->disk);
1065 list_move(&dc->list, &uncached_devices);
1066
1067 clear_bit(BCACHE_DEV_DETACHING, &dc->disk.flags);
1068 clear_bit(BCACHE_DEV_UNLINK_DONE, &dc->disk.flags);
1069
1070 mutex_unlock(&bch_register_lock);
1071
1072 pr_info("Caching disabled for %s", dc->backing_dev_name);
1073
1074 /* Drop ref we took in cached_dev_detach() */
1075 closure_put(&dc->disk.cl);
1076 }
1077
1078 void bch_cached_dev_detach(struct cached_dev *dc)
1079 {
1080 lockdep_assert_held(&bch_register_lock);
1081
1082 if (test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags))
1083 return;
1084
1085 if (test_and_set_bit(BCACHE_DEV_DETACHING, &dc->disk.flags))
1086 return;
1087
1088 /*
1089 * Block the device from being closed and freed until we're finished
1090 * detaching
1091 */
1092 closure_get(&dc->disk.cl);
1093
1094 bch_writeback_queue(dc);
1095
1096 cached_dev_put(dc);
1097 }
1098
1099 int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c,
1100 uint8_t *set_uuid)
1101 {
1102 uint32_t rtime = cpu_to_le32((u32)ktime_get_real_seconds());
1103 struct uuid_entry *u;
1104 struct cached_dev *exist_dc, *t;
1105 int ret = 0;
1106
1107 if ((set_uuid && memcmp(set_uuid, c->sb.set_uuid, 16)) ||
1108 (!set_uuid && memcmp(dc->sb.set_uuid, c->sb.set_uuid, 16)))
1109 return -ENOENT;
1110
1111 if (dc->disk.c) {
1112 pr_err("Can't attach %s: already attached",
1113 dc->backing_dev_name);
1114 return -EINVAL;
1115 }
1116
1117 if (test_bit(CACHE_SET_STOPPING, &c->flags)) {
1118 pr_err("Can't attach %s: shutting down",
1119 dc->backing_dev_name);
1120 return -EINVAL;
1121 }
1122
1123 if (dc->sb.block_size < c->sb.block_size) {
1124 /* Will die */
1125 pr_err("Couldn't attach %s: block size less than set's block size",
1126 dc->backing_dev_name);
1127 return -EINVAL;
1128 }
1129
1130 /* Check whether already attached */
1131 list_for_each_entry_safe(exist_dc, t, &c->cached_devs, list) {
1132 if (!memcmp(dc->sb.uuid, exist_dc->sb.uuid, 16)) {
1133 pr_err("Tried to attach %s but duplicate UUID already attached",
1134 dc->backing_dev_name);
1135
1136 return -EINVAL;
1137 }
1138 }
1139
1140 u = uuid_find(c, dc->sb.uuid);
1141
1142 if (u &&
1143 (BDEV_STATE(&dc->sb) == BDEV_STATE_STALE ||
1144 BDEV_STATE(&dc->sb) == BDEV_STATE_NONE)) {
1145 memcpy(u->uuid, invalid_uuid, 16);
1146 u->invalidated = cpu_to_le32((u32)ktime_get_real_seconds());
1147 u = NULL;
1148 }
1149
1150 if (!u) {
1151 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
1152 pr_err("Couldn't find uuid for %s in set",
1153 dc->backing_dev_name);
1154 return -ENOENT;
1155 }
1156
1157 u = uuid_find_empty(c);
1158 if (!u) {
1159 pr_err("Not caching %s, no room for UUID",
1160 dc->backing_dev_name);
1161 return -EINVAL;
1162 }
1163 }
1164
1165 /*
1166 * Deadlocks since we're called via sysfs...
1167 * sysfs_remove_file(&dc->kobj, &sysfs_attach);
1168 */
1169
1170 if (bch_is_zero(u->uuid, 16)) {
1171 struct closure cl;
1172
1173 closure_init_stack(&cl);
1174
1175 memcpy(u->uuid, dc->sb.uuid, 16);
1176 memcpy(u->label, dc->sb.label, SB_LABEL_SIZE);
1177 u->first_reg = u->last_reg = rtime;
1178 bch_uuid_write(c);
1179
1180 memcpy(dc->sb.set_uuid, c->sb.set_uuid, 16);
1181 SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN);
1182
1183 bch_write_bdev_super(dc, &cl);
1184 closure_sync(&cl);
1185 } else {
1186 u->last_reg = rtime;
1187 bch_uuid_write(c);
1188 }
1189
1190 bcache_device_attach(&dc->disk, c, u - c->uuids);
1191 list_move(&dc->list, &c->cached_devs);
1192 calc_cached_dev_sectors(c);
1193
1194 /*
1195 * dc->c must be set before dc->count != 0 - paired with the mb in
1196 * cached_dev_get()
1197 */
1198 smp_wmb();
1199 refcount_set(&dc->count, 1);
1200
1201 /* Block writeback thread, but spawn it */
1202 down_write(&dc->writeback_lock);
1203 if (bch_cached_dev_writeback_start(dc)) {
1204 up_write(&dc->writeback_lock);
1205 pr_err("Couldn't start writeback facilities for %s",
1206 dc->disk.disk->disk_name);
1207 return -ENOMEM;
1208 }
1209
1210 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
1211 atomic_set(&dc->has_dirty, 1);
1212 bch_writeback_queue(dc);
1213 }
1214
1215 bch_sectors_dirty_init(&dc->disk);
1216
1217 ret = bch_cached_dev_run(dc);
1218 if (ret && (ret != -EBUSY)) {
1219 up_write(&dc->writeback_lock);
1220 /*
1221 * bch_register_lock is held, bcache_device_stop() is not
1222 * able to be directly called. The kthread and kworker
1223 * created previously in bch_cached_dev_writeback_start()
1224 * have to be stopped manually here.
1225 */
1226 kthread_stop(dc->writeback_thread);
1227 cancel_writeback_rate_update_dwork(dc);
1228 pr_err("Couldn't run cached device %s",
1229 dc->backing_dev_name);
1230 return ret;
1231 }
1232
1233 bcache_device_link(&dc->disk, c, "bdev");
1234 atomic_inc(&c->attached_dev_nr);
1235
1236 /* Allow the writeback thread to proceed */
1237 up_write(&dc->writeback_lock);
1238
1239 pr_info("Caching %s as %s on set %pU",
1240 dc->backing_dev_name,
1241 dc->disk.disk->disk_name,
1242 dc->disk.c->sb.set_uuid);
1243 return 0;
1244 }
1245
1246 /* when dc->disk.kobj released */
1247 void bch_cached_dev_release(struct kobject *kobj)
1248 {
1249 struct cached_dev *dc = container_of(kobj, struct cached_dev,
1250 disk.kobj);
1251 kfree(dc);
1252 module_put(THIS_MODULE);
1253 }
1254
1255 static void cached_dev_free(struct closure *cl)
1256 {
1257 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1258
1259 if (test_and_clear_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags))
1260 cancel_writeback_rate_update_dwork(dc);
1261
1262 if (!IS_ERR_OR_NULL(dc->writeback_thread))
1263 kthread_stop(dc->writeback_thread);
1264 if (!IS_ERR_OR_NULL(dc->status_update_thread))
1265 kthread_stop(dc->status_update_thread);
1266
1267 mutex_lock(&bch_register_lock);
1268
1269 if (atomic_read(&dc->running))
1270 bd_unlink_disk_holder(dc->bdev, dc->disk.disk);
1271 bcache_device_free(&dc->disk);
1272 list_del(&dc->list);
1273
1274 mutex_unlock(&bch_register_lock);
1275
1276 if (dc->sb_disk)
1277 put_page(virt_to_page(dc->sb_disk));
1278
1279 if (!IS_ERR_OR_NULL(dc->bdev))
1280 blkdev_put(dc->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1281
1282 wake_up(&unregister_wait);
1283
1284 kobject_put(&dc->disk.kobj);
1285 }
1286
1287 static void cached_dev_flush(struct closure *cl)
1288 {
1289 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1290 struct bcache_device *d = &dc->disk;
1291
1292 mutex_lock(&bch_register_lock);
1293 bcache_device_unlink(d);
1294 mutex_unlock(&bch_register_lock);
1295
1296 bch_cache_accounting_destroy(&dc->accounting);
1297 kobject_del(&d->kobj);
1298
1299 continue_at(cl, cached_dev_free, system_wq);
1300 }
1301
1302 static int cached_dev_init(struct cached_dev *dc, unsigned int block_size)
1303 {
1304 int ret;
1305 struct io *io;
1306 struct request_queue *q = bdev_get_queue(dc->bdev);
1307
1308 __module_get(THIS_MODULE);
1309 INIT_LIST_HEAD(&dc->list);
1310 closure_init(&dc->disk.cl, NULL);
1311 set_closure_fn(&dc->disk.cl, cached_dev_flush, system_wq);
1312 kobject_init(&dc->disk.kobj, &bch_cached_dev_ktype);
1313 INIT_WORK(&dc->detach, cached_dev_detach_finish);
1314 sema_init(&dc->sb_write_mutex, 1);
1315 INIT_LIST_HEAD(&dc->io_lru);
1316 spin_lock_init(&dc->io_lock);
1317 bch_cache_accounting_init(&dc->accounting, &dc->disk.cl);
1318
1319 dc->sequential_cutoff = 4 << 20;
1320
1321 for (io = dc->io; io < dc->io + RECENT_IO; io++) {
1322 list_add(&io->lru, &dc->io_lru);
1323 hlist_add_head(&io->hash, dc->io_hash + RECENT_IO);
1324 }
1325
1326 dc->disk.stripe_size = q->limits.io_opt >> 9;
1327
1328 if (dc->disk.stripe_size)
1329 dc->partial_stripes_expensive =
1330 q->limits.raid_partial_stripes_expensive;
1331
1332 ret = bcache_device_init(&dc->disk, block_size,
1333 dc->bdev->bd_part->nr_sects - dc->sb.data_offset);
1334 if (ret)
1335 return ret;
1336
1337 dc->disk.disk->queue->backing_dev_info->ra_pages =
1338 max(dc->disk.disk->queue->backing_dev_info->ra_pages,
1339 q->backing_dev_info->ra_pages);
1340
1341 atomic_set(&dc->io_errors, 0);
1342 dc->io_disable = false;
1343 dc->error_limit = DEFAULT_CACHED_DEV_ERROR_LIMIT;
1344 /* default to auto */
1345 dc->stop_when_cache_set_failed = BCH_CACHED_DEV_STOP_AUTO;
1346
1347 bch_cached_dev_request_init(dc);
1348 bch_cached_dev_writeback_init(dc);
1349 return 0;
1350 }
1351
1352 /* Cached device - bcache superblock */
1353
1354 static int register_bdev(struct cache_sb *sb, struct cache_sb_disk *sb_disk,
1355 struct block_device *bdev,
1356 struct cached_dev *dc)
1357 {
1358 const char *err = "cannot allocate memory";
1359 struct cache_set *c;
1360 int ret = -ENOMEM;
1361
1362 bdevname(bdev, dc->backing_dev_name);
1363 memcpy(&dc->sb, sb, sizeof(struct cache_sb));
1364 dc->bdev = bdev;
1365 dc->bdev->bd_holder = dc;
1366 dc->sb_disk = sb_disk;
1367
1368 if (cached_dev_init(dc, sb->block_size << 9))
1369 goto err;
1370
1371 err = "error creating kobject";
1372 if (kobject_add(&dc->disk.kobj, &part_to_dev(bdev->bd_part)->kobj,
1373 "bcache"))
1374 goto err;
1375 if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj))
1376 goto err;
1377
1378 pr_info("registered backing device %s", dc->backing_dev_name);
1379
1380 list_add(&dc->list, &uncached_devices);
1381 /* attach to a matched cache set if it exists */
1382 list_for_each_entry(c, &bch_cache_sets, list)
1383 bch_cached_dev_attach(dc, c, NULL);
1384
1385 if (BDEV_STATE(&dc->sb) == BDEV_STATE_NONE ||
1386 BDEV_STATE(&dc->sb) == BDEV_STATE_STALE) {
1387 err = "failed to run cached device";
1388 ret = bch_cached_dev_run(dc);
1389 if (ret)
1390 goto err;
1391 }
1392
1393 return 0;
1394 err:
1395 pr_notice("error %s: %s", dc->backing_dev_name, err);
1396 bcache_device_stop(&dc->disk);
1397 return ret;
1398 }
1399
1400 /* Flash only volumes */
1401
1402 /* When d->kobj released */
1403 void bch_flash_dev_release(struct kobject *kobj)
1404 {
1405 struct bcache_device *d = container_of(kobj, struct bcache_device,
1406 kobj);
1407 kfree(d);
1408 }
1409
1410 static void flash_dev_free(struct closure *cl)
1411 {
1412 struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1413
1414 mutex_lock(&bch_register_lock);
1415 atomic_long_sub(bcache_dev_sectors_dirty(d),
1416 &d->c->flash_dev_dirty_sectors);
1417 bcache_device_free(d);
1418 mutex_unlock(&bch_register_lock);
1419 kobject_put(&d->kobj);
1420 }
1421
1422 static void flash_dev_flush(struct closure *cl)
1423 {
1424 struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1425
1426 mutex_lock(&bch_register_lock);
1427 bcache_device_unlink(d);
1428 mutex_unlock(&bch_register_lock);
1429 kobject_del(&d->kobj);
1430 continue_at(cl, flash_dev_free, system_wq);
1431 }
1432
1433 static int flash_dev_run(struct cache_set *c, struct uuid_entry *u)
1434 {
1435 struct bcache_device *d = kzalloc(sizeof(struct bcache_device),
1436 GFP_KERNEL);
1437 if (!d)
1438 return -ENOMEM;
1439
1440 closure_init(&d->cl, NULL);
1441 set_closure_fn(&d->cl, flash_dev_flush, system_wq);
1442
1443 kobject_init(&d->kobj, &bch_flash_dev_ktype);
1444
1445 if (bcache_device_init(d, block_bytes(c), u->sectors))
1446 goto err;
1447
1448 bcache_device_attach(d, c, u - c->uuids);
1449 bch_sectors_dirty_init(d);
1450 bch_flash_dev_request_init(d);
1451 add_disk(d->disk);
1452
1453 if (kobject_add(&d->kobj, &disk_to_dev(d->disk)->kobj, "bcache"))
1454 goto err;
1455
1456 bcache_device_link(d, c, "volume");
1457
1458 return 0;
1459 err:
1460 kobject_put(&d->kobj);
1461 return -ENOMEM;
1462 }
1463
1464 static int flash_devs_run(struct cache_set *c)
1465 {
1466 int ret = 0;
1467 struct uuid_entry *u;
1468
1469 for (u = c->uuids;
1470 u < c->uuids + c->nr_uuids && !ret;
1471 u++)
1472 if (UUID_FLASH_ONLY(u))
1473 ret = flash_dev_run(c, u);
1474
1475 return ret;
1476 }
1477
1478 int bch_flash_dev_create(struct cache_set *c, uint64_t size)
1479 {
1480 struct uuid_entry *u;
1481
1482 if (test_bit(CACHE_SET_STOPPING, &c->flags))
1483 return -EINTR;
1484
1485 if (!test_bit(CACHE_SET_RUNNING, &c->flags))
1486 return -EPERM;
1487
1488 u = uuid_find_empty(c);
1489 if (!u) {
1490 pr_err("Can't create volume, no room for UUID");
1491 return -EINVAL;
1492 }
1493
1494 get_random_bytes(u->uuid, 16);
1495 memset(u->label, 0, 32);
1496 u->first_reg = u->last_reg = cpu_to_le32((u32)ktime_get_real_seconds());
1497
1498 SET_UUID_FLASH_ONLY(u, 1);
1499 u->sectors = size >> 9;
1500
1501 bch_uuid_write(c);
1502
1503 return flash_dev_run(c, u);
1504 }
1505
1506 bool bch_cached_dev_error(struct cached_dev *dc)
1507 {
1508 if (!dc || test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags))
1509 return false;
1510
1511 dc->io_disable = true;
1512 /* make others know io_disable is true earlier */
1513 smp_mb();
1514
1515 pr_err("stop %s: too many IO errors on backing device %s\n",
1516 dc->disk.disk->disk_name, dc->backing_dev_name);
1517
1518 bcache_device_stop(&dc->disk);
1519 return true;
1520 }
1521
1522 /* Cache set */
1523
1524 __printf(2, 3)
1525 bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...)
1526 {
1527 va_list args;
1528
1529 if (c->on_error != ON_ERROR_PANIC &&
1530 test_bit(CACHE_SET_STOPPING, &c->flags))
1531 return false;
1532
1533 if (test_and_set_bit(CACHE_SET_IO_DISABLE, &c->flags))
1534 pr_info("CACHE_SET_IO_DISABLE already set");
1535
1536 /*
1537 * XXX: we can be called from atomic context
1538 * acquire_console_sem();
1539 */
1540
1541 pr_err("bcache: error on %pU: ", c->sb.set_uuid);
1542
1543 va_start(args, fmt);
1544 vprintk(fmt, args);
1545 va_end(args);
1546
1547 pr_err(", disabling caching\n");
1548
1549 if (c->on_error == ON_ERROR_PANIC)
1550 panic("panic forced after error\n");
1551
1552 bch_cache_set_unregister(c);
1553 return true;
1554 }
1555
1556 /* When c->kobj released */
1557 void bch_cache_set_release(struct kobject *kobj)
1558 {
1559 struct cache_set *c = container_of(kobj, struct cache_set, kobj);
1560
1561 kfree(c);
1562 module_put(THIS_MODULE);
1563 }
1564
1565 static void cache_set_free(struct closure *cl)
1566 {
1567 struct cache_set *c = container_of(cl, struct cache_set, cl);
1568 struct cache *ca;
1569 unsigned int i;
1570
1571 debugfs_remove(c->debug);
1572
1573 bch_open_buckets_free(c);
1574 bch_btree_cache_free(c);
1575 bch_journal_free(c);
1576
1577 mutex_lock(&bch_register_lock);
1578 for_each_cache(ca, c, i)
1579 if (ca) {
1580 ca->set = NULL;
1581 c->cache[ca->sb.nr_this_dev] = NULL;
1582 kobject_put(&ca->kobj);
1583 }
1584
1585 bch_bset_sort_state_free(&c->sort);
1586 free_pages((unsigned long) c->uuids, ilog2(bucket_pages(c)));
1587
1588 if (c->moving_gc_wq)
1589 destroy_workqueue(c->moving_gc_wq);
1590 bioset_exit(&c->bio_split);
1591 mempool_exit(&c->fill_iter);
1592 mempool_exit(&c->bio_meta);
1593 mempool_exit(&c->search);
1594 kfree(c->devices);
1595
1596 list_del(&c->list);
1597 mutex_unlock(&bch_register_lock);
1598
1599 pr_info("Cache set %pU unregistered", c->sb.set_uuid);
1600 wake_up(&unregister_wait);
1601
1602 closure_debug_destroy(&c->cl);
1603 kobject_put(&c->kobj);
1604 }
1605
1606 static void cache_set_flush(struct closure *cl)
1607 {
1608 struct cache_set *c = container_of(cl, struct cache_set, caching);
1609 struct cache *ca;
1610 struct btree *b;
1611 unsigned int i;
1612
1613 bch_cache_accounting_destroy(&c->accounting);
1614
1615 kobject_put(&c->internal);
1616 kobject_del(&c->kobj);
1617
1618 if (!IS_ERR_OR_NULL(c->gc_thread))
1619 kthread_stop(c->gc_thread);
1620
1621 if (!IS_ERR_OR_NULL(c->root))
1622 list_add(&c->root->list, &c->btree_cache);
1623
1624 /*
1625 * Avoid flushing cached nodes if cache set is retiring
1626 * due to too many I/O errors detected.
1627 */
1628 if (!test_bit(CACHE_SET_IO_DISABLE, &c->flags))
1629 list_for_each_entry(b, &c->btree_cache, list) {
1630 mutex_lock(&b->write_lock);
1631 if (btree_node_dirty(b))
1632 __bch_btree_node_write(b, NULL);
1633 mutex_unlock(&b->write_lock);
1634 }
1635
1636 for_each_cache(ca, c, i)
1637 if (ca->alloc_thread)
1638 kthread_stop(ca->alloc_thread);
1639
1640 if (c->journal.cur) {
1641 cancel_delayed_work_sync(&c->journal.work);
1642 /* flush last journal entry if needed */
1643 c->journal.work.work.func(&c->journal.work.work);
1644 }
1645
1646 closure_return(cl);
1647 }
1648
1649 /*
1650 * This function is only called when CACHE_SET_IO_DISABLE is set, which means
1651 * cache set is unregistering due to too many I/O errors. In this condition,
1652 * the bcache device might be stopped, it depends on stop_when_cache_set_failed
1653 * value and whether the broken cache has dirty data:
1654 *
1655 * dc->stop_when_cache_set_failed dc->has_dirty stop bcache device
1656 * BCH_CACHED_STOP_AUTO 0 NO
1657 * BCH_CACHED_STOP_AUTO 1 YES
1658 * BCH_CACHED_DEV_STOP_ALWAYS 0 YES
1659 * BCH_CACHED_DEV_STOP_ALWAYS 1 YES
1660 *
1661 * The expected behavior is, if stop_when_cache_set_failed is configured to
1662 * "auto" via sysfs interface, the bcache device will not be stopped if the
1663 * backing device is clean on the broken cache device.
1664 */
1665 static void conditional_stop_bcache_device(struct cache_set *c,
1666 struct bcache_device *d,
1667 struct cached_dev *dc)
1668 {
1669 if (dc->stop_when_cache_set_failed == BCH_CACHED_DEV_STOP_ALWAYS) {
1670 pr_warn("stop_when_cache_set_failed of %s is \"always\", stop it for failed cache set %pU.",
1671 d->disk->disk_name, c->sb.set_uuid);
1672 bcache_device_stop(d);
1673 } else if (atomic_read(&dc->has_dirty)) {
1674 /*
1675 * dc->stop_when_cache_set_failed == BCH_CACHED_STOP_AUTO
1676 * and dc->has_dirty == 1
1677 */
1678 pr_warn("stop_when_cache_set_failed of %s is \"auto\" and cache is dirty, stop it to avoid potential data corruption.",
1679 d->disk->disk_name);
1680 /*
1681 * There might be a small time gap that cache set is
1682 * released but bcache device is not. Inside this time
1683 * gap, regular I/O requests will directly go into
1684 * backing device as no cache set attached to. This
1685 * behavior may also introduce potential inconsistence
1686 * data in writeback mode while cache is dirty.
1687 * Therefore before calling bcache_device_stop() due
1688 * to a broken cache device, dc->io_disable should be
1689 * explicitly set to true.
1690 */
1691 dc->io_disable = true;
1692 /* make others know io_disable is true earlier */
1693 smp_mb();
1694 bcache_device_stop(d);
1695 } else {
1696 /*
1697 * dc->stop_when_cache_set_failed == BCH_CACHED_STOP_AUTO
1698 * and dc->has_dirty == 0
1699 */
1700 pr_warn("stop_when_cache_set_failed of %s is \"auto\" and cache is clean, keep it alive.",
1701 d->disk->disk_name);
1702 }
1703 }
1704
1705 static void __cache_set_unregister(struct closure *cl)
1706 {
1707 struct cache_set *c = container_of(cl, struct cache_set, caching);
1708 struct cached_dev *dc;
1709 struct bcache_device *d;
1710 size_t i;
1711
1712 mutex_lock(&bch_register_lock);
1713
1714 for (i = 0; i < c->devices_max_used; i++) {
1715 d = c->devices[i];
1716 if (!d)
1717 continue;
1718
1719 if (!UUID_FLASH_ONLY(&c->uuids[i]) &&
1720 test_bit(CACHE_SET_UNREGISTERING, &c->flags)) {
1721 dc = container_of(d, struct cached_dev, disk);
1722 bch_cached_dev_detach(dc);
1723 if (test_bit(CACHE_SET_IO_DISABLE, &c->flags))
1724 conditional_stop_bcache_device(c, d, dc);
1725 } else {
1726 bcache_device_stop(d);
1727 }
1728 }
1729
1730 mutex_unlock(&bch_register_lock);
1731
1732 continue_at(cl, cache_set_flush, system_wq);
1733 }
1734
1735 void bch_cache_set_stop(struct cache_set *c)
1736 {
1737 if (!test_and_set_bit(CACHE_SET_STOPPING, &c->flags))
1738 /* closure_fn set to __cache_set_unregister() */
1739 closure_queue(&c->caching);
1740 }
1741
1742 void bch_cache_set_unregister(struct cache_set *c)
1743 {
1744 set_bit(CACHE_SET_UNREGISTERING, &c->flags);
1745 bch_cache_set_stop(c);
1746 }
1747
1748 #define alloc_bucket_pages(gfp, c) \
1749 ((void *) __get_free_pages(__GFP_ZERO|gfp, ilog2(bucket_pages(c))))
1750
1751 struct cache_set *bch_cache_set_alloc(struct cache_sb *sb)
1752 {
1753 int iter_size;
1754 struct cache_set *c = kzalloc(sizeof(struct cache_set), GFP_KERNEL);
1755
1756 if (!c)
1757 return NULL;
1758
1759 __module_get(THIS_MODULE);
1760 closure_init(&c->cl, NULL);
1761 set_closure_fn(&c->cl, cache_set_free, system_wq);
1762
1763 closure_init(&c->caching, &c->cl);
1764 set_closure_fn(&c->caching, __cache_set_unregister, system_wq);
1765
1766 /* Maybe create continue_at_noreturn() and use it here? */
1767 closure_set_stopped(&c->cl);
1768 closure_put(&c->cl);
1769
1770 kobject_init(&c->kobj, &bch_cache_set_ktype);
1771 kobject_init(&c->internal, &bch_cache_set_internal_ktype);
1772
1773 bch_cache_accounting_init(&c->accounting, &c->cl);
1774
1775 memcpy(c->sb.set_uuid, sb->set_uuid, 16);
1776 c->sb.block_size = sb->block_size;
1777 c->sb.bucket_size = sb->bucket_size;
1778 c->sb.nr_in_set = sb->nr_in_set;
1779 c->sb.last_mount = sb->last_mount;
1780 c->bucket_bits = ilog2(sb->bucket_size);
1781 c->block_bits = ilog2(sb->block_size);
1782 c->nr_uuids = bucket_bytes(c) / sizeof(struct uuid_entry);
1783 c->devices_max_used = 0;
1784 atomic_set(&c->attached_dev_nr, 0);
1785 c->btree_pages = bucket_pages(c);
1786 if (c->btree_pages > BTREE_MAX_PAGES)
1787 c->btree_pages = max_t(int, c->btree_pages / 4,
1788 BTREE_MAX_PAGES);
1789
1790 sema_init(&c->sb_write_mutex, 1);
1791 mutex_init(&c->bucket_lock);
1792 init_waitqueue_head(&c->btree_cache_wait);
1793 spin_lock_init(&c->btree_cannibalize_lock);
1794 init_waitqueue_head(&c->bucket_wait);
1795 init_waitqueue_head(&c->gc_wait);
1796 sema_init(&c->uuid_write_mutex, 1);
1797
1798 spin_lock_init(&c->btree_gc_time.lock);
1799 spin_lock_init(&c->btree_split_time.lock);
1800 spin_lock_init(&c->btree_read_time.lock);
1801
1802 bch_moving_init_cache_set(c);
1803
1804 INIT_LIST_HEAD(&c->list);
1805 INIT_LIST_HEAD(&c->cached_devs);
1806 INIT_LIST_HEAD(&c->btree_cache);
1807 INIT_LIST_HEAD(&c->btree_cache_freeable);
1808 INIT_LIST_HEAD(&c->btree_cache_freed);
1809 INIT_LIST_HEAD(&c->data_buckets);
1810
1811 iter_size = (sb->bucket_size / sb->block_size + 1) *
1812 sizeof(struct btree_iter_set);
1813
1814 if (!(c->devices = kcalloc(c->nr_uuids, sizeof(void *), GFP_KERNEL)) ||
1815 mempool_init_slab_pool(&c->search, 32, bch_search_cache) ||
1816 mempool_init_kmalloc_pool(&c->bio_meta, 2,
1817 sizeof(struct bbio) + sizeof(struct bio_vec) *
1818 bucket_pages(c)) ||
1819 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
1820 bioset_init(&c->bio_split, 4, offsetof(struct bbio, bio),
1821 BIOSET_NEED_BVECS|BIOSET_NEED_RESCUER) ||
1822 !(c->uuids = alloc_bucket_pages(GFP_KERNEL, c)) ||
1823 !(c->moving_gc_wq = alloc_workqueue("bcache_gc",
1824 WQ_MEM_RECLAIM, 0)) ||
1825 bch_journal_alloc(c) ||
1826 bch_btree_cache_alloc(c) ||
1827 bch_open_buckets_alloc(c) ||
1828 bch_bset_sort_state_init(&c->sort, ilog2(c->btree_pages)))
1829 goto err;
1830
1831 c->congested_read_threshold_us = 2000;
1832 c->congested_write_threshold_us = 20000;
1833 c->error_limit = DEFAULT_IO_ERROR_LIMIT;
1834 c->idle_max_writeback_rate_enabled = 1;
1835 WARN_ON(test_and_clear_bit(CACHE_SET_IO_DISABLE, &c->flags));
1836
1837 return c;
1838 err:
1839 bch_cache_set_unregister(c);
1840 return NULL;
1841 }
1842
1843 static int run_cache_set(struct cache_set *c)
1844 {
1845 const char *err = "cannot allocate memory";
1846 struct cached_dev *dc, *t;
1847 struct cache *ca;
1848 struct closure cl;
1849 unsigned int i;
1850 LIST_HEAD(journal);
1851 struct journal_replay *l;
1852
1853 closure_init_stack(&cl);
1854
1855 for_each_cache(ca, c, i)
1856 c->nbuckets += ca->sb.nbuckets;
1857 set_gc_sectors(c);
1858
1859 if (CACHE_SYNC(&c->sb)) {
1860 struct bkey *k;
1861 struct jset *j;
1862
1863 err = "cannot allocate memory for journal";
1864 if (bch_journal_read(c, &journal))
1865 goto err;
1866
1867 pr_debug("btree_journal_read() done");
1868
1869 err = "no journal entries found";
1870 if (list_empty(&journal))
1871 goto err;
1872
1873 j = &list_entry(journal.prev, struct journal_replay, list)->j;
1874
1875 err = "IO error reading priorities";
1876 for_each_cache(ca, c, i)
1877 prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev]);
1878
1879 /*
1880 * If prio_read() fails it'll call cache_set_error and we'll
1881 * tear everything down right away, but if we perhaps checked
1882 * sooner we could avoid journal replay.
1883 */
1884
1885 k = &j->btree_root;
1886
1887 err = "bad btree root";
1888 if (__bch_btree_ptr_invalid(c, k))
1889 goto err;
1890
1891 err = "error reading btree root";
1892 c->root = bch_btree_node_get(c, NULL, k,
1893 j->btree_level,
1894 true, NULL);
1895 if (IS_ERR_OR_NULL(c->root))
1896 goto err;
1897
1898 list_del_init(&c->root->list);
1899 rw_unlock(true, c->root);
1900
1901 err = uuid_read(c, j, &cl);
1902 if (err)
1903 goto err;
1904
1905 err = "error in recovery";
1906 if (bch_btree_check(c))
1907 goto err;
1908
1909 /*
1910 * bch_btree_check() may occupy too much system memory which
1911 * has negative effects to user space application (e.g. data
1912 * base) performance. Shrink the mca cache memory proactively
1913 * here to avoid competing memory with user space workloads..
1914 */
1915 if (!c->shrinker_disabled) {
1916 struct shrink_control sc;
1917
1918 sc.gfp_mask = GFP_KERNEL;
1919 sc.nr_to_scan = c->btree_cache_used * c->btree_pages;
1920 /* first run to clear b->accessed tag */
1921 c->shrink.scan_objects(&c->shrink, &sc);
1922 /* second run to reap non-accessed nodes */
1923 c->shrink.scan_objects(&c->shrink, &sc);
1924 }
1925
1926 bch_journal_mark(c, &journal);
1927 bch_initial_gc_finish(c);
1928 pr_debug("btree_check() done");
1929
1930 /*
1931 * bcache_journal_next() can't happen sooner, or
1932 * btree_gc_finish() will give spurious errors about last_gc >
1933 * gc_gen - this is a hack but oh well.
1934 */
1935 bch_journal_next(&c->journal);
1936
1937 err = "error starting allocator thread";
1938 for_each_cache(ca, c, i)
1939 if (bch_cache_allocator_start(ca))
1940 goto err;
1941
1942 /*
1943 * First place it's safe to allocate: btree_check() and
1944 * btree_gc_finish() have to run before we have buckets to
1945 * allocate, and bch_bucket_alloc_set() might cause a journal
1946 * entry to be written so bcache_journal_next() has to be called
1947 * first.
1948 *
1949 * If the uuids were in the old format we have to rewrite them
1950 * before the next journal entry is written:
1951 */
1952 if (j->version < BCACHE_JSET_VERSION_UUID)
1953 __uuid_write(c);
1954
1955 err = "bcache: replay journal failed";
1956 if (bch_journal_replay(c, &journal))
1957 goto err;
1958 } else {
1959 pr_notice("invalidating existing data");
1960
1961 for_each_cache(ca, c, i) {
1962 unsigned int j;
1963
1964 ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7,
1965 2, SB_JOURNAL_BUCKETS);
1966
1967 for (j = 0; j < ca->sb.keys; j++)
1968 ca->sb.d[j] = ca->sb.first_bucket + j;
1969 }
1970
1971 bch_initial_gc_finish(c);
1972
1973 err = "error starting allocator thread";
1974 for_each_cache(ca, c, i)
1975 if (bch_cache_allocator_start(ca))
1976 goto err;
1977
1978 mutex_lock(&c->bucket_lock);
1979 for_each_cache(ca, c, i)
1980 bch_prio_write(ca, true);
1981 mutex_unlock(&c->bucket_lock);
1982
1983 err = "cannot allocate new UUID bucket";
1984 if (__uuid_write(c))
1985 goto err;
1986
1987 err = "cannot allocate new btree root";
1988 c->root = __bch_btree_node_alloc(c, NULL, 0, true, NULL);
1989 if (IS_ERR_OR_NULL(c->root))
1990 goto err;
1991
1992 mutex_lock(&c->root->write_lock);
1993 bkey_copy_key(&c->root->key, &MAX_KEY);
1994 bch_btree_node_write(c->root, &cl);
1995 mutex_unlock(&c->root->write_lock);
1996
1997 bch_btree_set_root(c->root);
1998 rw_unlock(true, c->root);
1999
2000 /*
2001 * We don't want to write the first journal entry until
2002 * everything is set up - fortunately journal entries won't be
2003 * written until the SET_CACHE_SYNC() here:
2004 */
2005 SET_CACHE_SYNC(&c->sb, true);
2006
2007 bch_journal_next(&c->journal);
2008 bch_journal_meta(c, &cl);
2009 }
2010
2011 err = "error starting gc thread";
2012 if (bch_gc_thread_start(c))
2013 goto err;
2014
2015 closure_sync(&cl);
2016 c->sb.last_mount = (u32)ktime_get_real_seconds();
2017 bcache_write_super(c);
2018
2019 list_for_each_entry_safe(dc, t, &uncached_devices, list)
2020 bch_cached_dev_attach(dc, c, NULL);
2021
2022 flash_devs_run(c);
2023
2024 set_bit(CACHE_SET_RUNNING, &c->flags);
2025 return 0;
2026 err:
2027 while (!list_empty(&journal)) {
2028 l = list_first_entry(&journal, struct journal_replay, list);
2029 list_del(&l->list);
2030 kfree(l);
2031 }
2032
2033 closure_sync(&cl);
2034
2035 bch_cache_set_error(c, "%s", err);
2036
2037 return -EIO;
2038 }
2039
2040 static bool can_attach_cache(struct cache *ca, struct cache_set *c)
2041 {
2042 return ca->sb.block_size == c->sb.block_size &&
2043 ca->sb.bucket_size == c->sb.bucket_size &&
2044 ca->sb.nr_in_set == c->sb.nr_in_set;
2045 }
2046
2047 static const char *register_cache_set(struct cache *ca)
2048 {
2049 char buf[12];
2050 const char *err = "cannot allocate memory";
2051 struct cache_set *c;
2052
2053 list_for_each_entry(c, &bch_cache_sets, list)
2054 if (!memcmp(c->sb.set_uuid, ca->sb.set_uuid, 16)) {
2055 if (c->cache[ca->sb.nr_this_dev])
2056 return "duplicate cache set member";
2057
2058 if (!can_attach_cache(ca, c))
2059 return "cache sb does not match set";
2060
2061 if (!CACHE_SYNC(&ca->sb))
2062 SET_CACHE_SYNC(&c->sb, false);
2063
2064 goto found;
2065 }
2066
2067 c = bch_cache_set_alloc(&ca->sb);
2068 if (!c)
2069 return err;
2070
2071 err = "error creating kobject";
2072 if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->sb.set_uuid) ||
2073 kobject_add(&c->internal, &c->kobj, "internal"))
2074 goto err;
2075
2076 if (bch_cache_accounting_add_kobjs(&c->accounting, &c->kobj))
2077 goto err;
2078
2079 bch_debug_init_cache_set(c);
2080
2081 list_add(&c->list, &bch_cache_sets);
2082 found:
2083 sprintf(buf, "cache%i", ca->sb.nr_this_dev);
2084 if (sysfs_create_link(&ca->kobj, &c->kobj, "set") ||
2085 sysfs_create_link(&c->kobj, &ca->kobj, buf))
2086 goto err;
2087
2088 if (ca->sb.seq > c->sb.seq) {
2089 c->sb.version = ca->sb.version;
2090 memcpy(c->sb.set_uuid, ca->sb.set_uuid, 16);
2091 c->sb.flags = ca->sb.flags;
2092 c->sb.seq = ca->sb.seq;
2093 pr_debug("set version = %llu", c->sb.version);
2094 }
2095
2096 kobject_get(&ca->kobj);
2097 ca->set = c;
2098 ca->set->cache[ca->sb.nr_this_dev] = ca;
2099 c->cache_by_alloc[c->caches_loaded++] = ca;
2100
2101 if (c->caches_loaded == c->sb.nr_in_set) {
2102 err = "failed to run cache set";
2103 if (run_cache_set(c) < 0)
2104 goto err;
2105 }
2106
2107 return NULL;
2108 err:
2109 bch_cache_set_unregister(c);
2110 return err;
2111 }
2112
2113 /* Cache device */
2114
2115 /* When ca->kobj released */
2116 void bch_cache_release(struct kobject *kobj)
2117 {
2118 struct cache *ca = container_of(kobj, struct cache, kobj);
2119 unsigned int i;
2120
2121 if (ca->set) {
2122 BUG_ON(ca->set->cache[ca->sb.nr_this_dev] != ca);
2123 ca->set->cache[ca->sb.nr_this_dev] = NULL;
2124 }
2125
2126 free_pages((unsigned long) ca->disk_buckets, ilog2(bucket_pages(ca)));
2127 kfree(ca->prio_buckets);
2128 vfree(ca->buckets);
2129
2130 free_heap(&ca->heap);
2131 free_fifo(&ca->free_inc);
2132
2133 for (i = 0; i < RESERVE_NR; i++)
2134 free_fifo(&ca->free[i]);
2135
2136 if (ca->sb_disk)
2137 put_page(virt_to_page(ca->sb_disk));
2138
2139 if (!IS_ERR_OR_NULL(ca->bdev))
2140 blkdev_put(ca->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2141
2142 kfree(ca);
2143 module_put(THIS_MODULE);
2144 }
2145
2146 static int cache_alloc(struct cache *ca)
2147 {
2148 size_t free;
2149 size_t btree_buckets;
2150 struct bucket *b;
2151 int ret = -ENOMEM;
2152 const char *err = NULL;
2153
2154 __module_get(THIS_MODULE);
2155 kobject_init(&ca->kobj, &bch_cache_ktype);
2156
2157 bio_init(&ca->journal.bio, ca->journal.bio.bi_inline_vecs, 8);
2158
2159 /*
2160 * when ca->sb.njournal_buckets is not zero, journal exists,
2161 * and in bch_journal_replay(), tree node may split,
2162 * so bucket of RESERVE_BTREE type is needed,
2163 * the worst situation is all journal buckets are valid journal,
2164 * and all the keys need to replay,
2165 * so the number of RESERVE_BTREE type buckets should be as much
2166 * as journal buckets
2167 */
2168 btree_buckets = ca->sb.njournal_buckets ?: 8;
2169 free = roundup_pow_of_two(ca->sb.nbuckets) >> 10;
2170 if (!free) {
2171 ret = -EPERM;
2172 err = "ca->sb.nbuckets is too small";
2173 goto err_free;
2174 }
2175
2176 if (!init_fifo(&ca->free[RESERVE_BTREE], btree_buckets,
2177 GFP_KERNEL)) {
2178 err = "ca->free[RESERVE_BTREE] alloc failed";
2179 goto err_btree_alloc;
2180 }
2181
2182 if (!init_fifo_exact(&ca->free[RESERVE_PRIO], prio_buckets(ca),
2183 GFP_KERNEL)) {
2184 err = "ca->free[RESERVE_PRIO] alloc failed";
2185 goto err_prio_alloc;
2186 }
2187
2188 if (!init_fifo(&ca->free[RESERVE_MOVINGGC], free, GFP_KERNEL)) {
2189 err = "ca->free[RESERVE_MOVINGGC] alloc failed";
2190 goto err_movinggc_alloc;
2191 }
2192
2193 if (!init_fifo(&ca->free[RESERVE_NONE], free, GFP_KERNEL)) {
2194 err = "ca->free[RESERVE_NONE] alloc failed";
2195 goto err_none_alloc;
2196 }
2197
2198 if (!init_fifo(&ca->free_inc, free << 2, GFP_KERNEL)) {
2199 err = "ca->free_inc alloc failed";
2200 goto err_free_inc_alloc;
2201 }
2202
2203 if (!init_heap(&ca->heap, free << 3, GFP_KERNEL)) {
2204 err = "ca->heap alloc failed";
2205 goto err_heap_alloc;
2206 }
2207
2208 ca->buckets = vzalloc(array_size(sizeof(struct bucket),
2209 ca->sb.nbuckets));
2210 if (!ca->buckets) {
2211 err = "ca->buckets alloc failed";
2212 goto err_buckets_alloc;
2213 }
2214
2215 ca->prio_buckets = kzalloc(array3_size(sizeof(uint64_t),
2216 prio_buckets(ca), 2),
2217 GFP_KERNEL);
2218 if (!ca->prio_buckets) {
2219 err = "ca->prio_buckets alloc failed";
2220 goto err_prio_buckets_alloc;
2221 }
2222
2223 ca->disk_buckets = alloc_bucket_pages(GFP_KERNEL, ca);
2224 if (!ca->disk_buckets) {
2225 err = "ca->disk_buckets alloc failed";
2226 goto err_disk_buckets_alloc;
2227 }
2228
2229 ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca);
2230
2231 for_each_bucket(b, ca)
2232 atomic_set(&b->pin, 0);
2233 return 0;
2234
2235 err_disk_buckets_alloc:
2236 kfree(ca->prio_buckets);
2237 err_prio_buckets_alloc:
2238 vfree(ca->buckets);
2239 err_buckets_alloc:
2240 free_heap(&ca->heap);
2241 err_heap_alloc:
2242 free_fifo(&ca->free_inc);
2243 err_free_inc_alloc:
2244 free_fifo(&ca->free[RESERVE_NONE]);
2245 err_none_alloc:
2246 free_fifo(&ca->free[RESERVE_MOVINGGC]);
2247 err_movinggc_alloc:
2248 free_fifo(&ca->free[RESERVE_PRIO]);
2249 err_prio_alloc:
2250 free_fifo(&ca->free[RESERVE_BTREE]);
2251 err_btree_alloc:
2252 err_free:
2253 module_put(THIS_MODULE);
2254 if (err)
2255 pr_notice("error %s: %s", ca->cache_dev_name, err);
2256 return ret;
2257 }
2258
2259 static int register_cache(struct cache_sb *sb, struct cache_sb_disk *sb_disk,
2260 struct block_device *bdev, struct cache *ca)
2261 {
2262 const char *err = NULL; /* must be set for any error case */
2263 int ret = 0;
2264
2265 bdevname(bdev, ca->cache_dev_name);
2266 memcpy(&ca->sb, sb, sizeof(struct cache_sb));
2267 ca->bdev = bdev;
2268 ca->bdev->bd_holder = ca;
2269 ca->sb_disk = sb_disk;
2270
2271 if (blk_queue_discard(bdev_get_queue(bdev)))
2272 ca->discard = CACHE_DISCARD(&ca->sb);
2273
2274 ret = cache_alloc(ca);
2275 if (ret != 0) {
2276 /*
2277 * If we failed here, it means ca->kobj is not initialized yet,
2278 * kobject_put() won't be called and there is no chance to
2279 * call blkdev_put() to bdev in bch_cache_release(). So we
2280 * explicitly call blkdev_put() here.
2281 */
2282 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2283 if (ret == -ENOMEM)
2284 err = "cache_alloc(): -ENOMEM";
2285 else if (ret == -EPERM)
2286 err = "cache_alloc(): cache device is too small";
2287 else
2288 err = "cache_alloc(): unknown error";
2289 goto err;
2290 }
2291
2292 if (kobject_add(&ca->kobj,
2293 &part_to_dev(bdev->bd_part)->kobj,
2294 "bcache")) {
2295 err = "error calling kobject_add";
2296 ret = -ENOMEM;
2297 goto out;
2298 }
2299
2300 mutex_lock(&bch_register_lock);
2301 err = register_cache_set(ca);
2302 mutex_unlock(&bch_register_lock);
2303
2304 if (err) {
2305 ret = -ENODEV;
2306 goto out;
2307 }
2308
2309 pr_info("registered cache device %s", ca->cache_dev_name);
2310
2311 out:
2312 kobject_put(&ca->kobj);
2313
2314 err:
2315 if (err)
2316 pr_notice("error %s: %s", ca->cache_dev_name, err);
2317
2318 return ret;
2319 }
2320
2321 /* Global interfaces/init */
2322
2323 static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr,
2324 const char *buffer, size_t size);
2325 static ssize_t bch_pending_bdevs_cleanup(struct kobject *k,
2326 struct kobj_attribute *attr,
2327 const char *buffer, size_t size);
2328
2329 kobj_attribute_write(register, register_bcache);
2330 kobj_attribute_write(register_quiet, register_bcache);
2331 kobj_attribute_write(pendings_cleanup, bch_pending_bdevs_cleanup);
2332
2333 static bool bch_is_open_backing(struct block_device *bdev)
2334 {
2335 struct cache_set *c, *tc;
2336 struct cached_dev *dc, *t;
2337
2338 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
2339 list_for_each_entry_safe(dc, t, &c->cached_devs, list)
2340 if (dc->bdev == bdev)
2341 return true;
2342 list_for_each_entry_safe(dc, t, &uncached_devices, list)
2343 if (dc->bdev == bdev)
2344 return true;
2345 return false;
2346 }
2347
2348 static bool bch_is_open_cache(struct block_device *bdev)
2349 {
2350 struct cache_set *c, *tc;
2351 struct cache *ca;
2352 unsigned int i;
2353
2354 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
2355 for_each_cache(ca, c, i)
2356 if (ca->bdev == bdev)
2357 return true;
2358 return false;
2359 }
2360
2361 static bool bch_is_open(struct block_device *bdev)
2362 {
2363 return bch_is_open_cache(bdev) || bch_is_open_backing(bdev);
2364 }
2365
2366 static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr,
2367 const char *buffer, size_t size)
2368 {
2369 const char *err;
2370 char *path = NULL;
2371 struct cache_sb *sb;
2372 struct cache_sb_disk *sb_disk;
2373 struct block_device *bdev;
2374 ssize_t ret;
2375
2376 ret = -EBUSY;
2377 err = "failed to reference bcache module";
2378 if (!try_module_get(THIS_MODULE))
2379 goto out;
2380
2381 /* For latest state of bcache_is_reboot */
2382 smp_mb();
2383 err = "bcache is in reboot";
2384 if (bcache_is_reboot)
2385 goto out_module_put;
2386
2387 ret = -ENOMEM;
2388 err = "cannot allocate memory";
2389 path = kstrndup(buffer, size, GFP_KERNEL);
2390 if (!path)
2391 goto out_module_put;
2392
2393 sb = kmalloc(sizeof(struct cache_sb), GFP_KERNEL);
2394 if (!sb)
2395 goto out_free_path;
2396
2397 ret = -EINVAL;
2398 err = "failed to open device";
2399 bdev = blkdev_get_by_path(strim(path),
2400 FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2401 sb);
2402 if (IS_ERR(bdev)) {
2403 if (bdev == ERR_PTR(-EBUSY)) {
2404 bdev = lookup_bdev(strim(path));
2405 mutex_lock(&bch_register_lock);
2406 if (!IS_ERR(bdev) && bch_is_open(bdev))
2407 err = "device already registered";
2408 else
2409 err = "device busy";
2410 mutex_unlock(&bch_register_lock);
2411 if (!IS_ERR(bdev))
2412 bdput(bdev);
2413 if (attr == &ksysfs_register_quiet)
2414 goto done;
2415 }
2416 goto out_free_sb;
2417 }
2418
2419 err = "failed to set blocksize";
2420 if (set_blocksize(bdev, 4096))
2421 goto out_blkdev_put;
2422
2423 err = read_super(sb, bdev, &sb_disk);
2424 if (err)
2425 goto out_blkdev_put;
2426
2427 err = "failed to register device";
2428 if (SB_IS_BDEV(sb)) {
2429 struct cached_dev *dc = kzalloc(sizeof(*dc), GFP_KERNEL);
2430
2431 if (!dc)
2432 goto out_put_sb_page;
2433
2434 mutex_lock(&bch_register_lock);
2435 ret = register_bdev(sb, sb_disk, bdev, dc);
2436 mutex_unlock(&bch_register_lock);
2437 /* blkdev_put() will be called in cached_dev_free() */
2438 if (ret < 0)
2439 goto out_free_sb;
2440 } else {
2441 struct cache *ca = kzalloc(sizeof(*ca), GFP_KERNEL);
2442
2443 if (!ca)
2444 goto out_put_sb_page;
2445
2446 /* blkdev_put() will be called in bch_cache_release() */
2447 if (register_cache(sb, sb_disk, bdev, ca) != 0)
2448 goto out_free_sb;
2449 }
2450
2451 done:
2452 kfree(sb);
2453 kfree(path);
2454 module_put(THIS_MODULE);
2455 return size;
2456
2457 out_put_sb_page:
2458 put_page(virt_to_page(sb_disk));
2459 out_blkdev_put:
2460 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2461 out_free_sb:
2462 kfree(sb);
2463 out_free_path:
2464 kfree(path);
2465 path = NULL;
2466 out_module_put:
2467 module_put(THIS_MODULE);
2468 out:
2469 pr_info("error %s: %s", path?path:"", err);
2470 return ret;
2471 }
2472
2473
2474 struct pdev {
2475 struct list_head list;
2476 struct cached_dev *dc;
2477 };
2478
2479 static ssize_t bch_pending_bdevs_cleanup(struct kobject *k,
2480 struct kobj_attribute *attr,
2481 const char *buffer,
2482 size_t size)
2483 {
2484 LIST_HEAD(pending_devs);
2485 ssize_t ret = size;
2486 struct cached_dev *dc, *tdc;
2487 struct pdev *pdev, *tpdev;
2488 struct cache_set *c, *tc;
2489
2490 mutex_lock(&bch_register_lock);
2491 list_for_each_entry_safe(dc, tdc, &uncached_devices, list) {
2492 pdev = kmalloc(sizeof(struct pdev), GFP_KERNEL);
2493 if (!pdev)
2494 break;
2495 pdev->dc = dc;
2496 list_add(&pdev->list, &pending_devs);
2497 }
2498
2499 list_for_each_entry_safe(pdev, tpdev, &pending_devs, list) {
2500 list_for_each_entry_safe(c, tc, &bch_cache_sets, list) {
2501 char *pdev_set_uuid = pdev->dc->sb.set_uuid;
2502 char *set_uuid = c->sb.uuid;
2503
2504 if (!memcmp(pdev_set_uuid, set_uuid, 16)) {
2505 list_del(&pdev->list);
2506 kfree(pdev);
2507 break;
2508 }
2509 }
2510 }
2511 mutex_unlock(&bch_register_lock);
2512
2513 list_for_each_entry_safe(pdev, tpdev, &pending_devs, list) {
2514 pr_info("delete pdev %p", pdev);
2515 list_del(&pdev->list);
2516 bcache_device_stop(&pdev->dc->disk);
2517 kfree(pdev);
2518 }
2519
2520 return ret;
2521 }
2522
2523 static int bcache_reboot(struct notifier_block *n, unsigned long code, void *x)
2524 {
2525 if (bcache_is_reboot)
2526 return NOTIFY_DONE;
2527
2528 if (code == SYS_DOWN ||
2529 code == SYS_HALT ||
2530 code == SYS_POWER_OFF) {
2531 DEFINE_WAIT(wait);
2532 unsigned long start = jiffies;
2533 bool stopped = false;
2534
2535 struct cache_set *c, *tc;
2536 struct cached_dev *dc, *tdc;
2537
2538 mutex_lock(&bch_register_lock);
2539
2540 if (bcache_is_reboot)
2541 goto out;
2542
2543 /* New registration is rejected since now */
2544 bcache_is_reboot = true;
2545 /*
2546 * Make registering caller (if there is) on other CPU
2547 * core know bcache_is_reboot set to true earlier
2548 */
2549 smp_mb();
2550
2551 if (list_empty(&bch_cache_sets) &&
2552 list_empty(&uncached_devices))
2553 goto out;
2554
2555 mutex_unlock(&bch_register_lock);
2556
2557 pr_info("Stopping all devices:");
2558
2559 /*
2560 * The reason bch_register_lock is not held to call
2561 * bch_cache_set_stop() and bcache_device_stop() is to
2562 * avoid potential deadlock during reboot, because cache
2563 * set or bcache device stopping process will acqurie
2564 * bch_register_lock too.
2565 *
2566 * We are safe here because bcache_is_reboot sets to
2567 * true already, register_bcache() will reject new
2568 * registration now. bcache_is_reboot also makes sure
2569 * bcache_reboot() won't be re-entered on by other thread,
2570 * so there is no race in following list iteration by
2571 * list_for_each_entry_safe().
2572 */
2573 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
2574 bch_cache_set_stop(c);
2575
2576 list_for_each_entry_safe(dc, tdc, &uncached_devices, list)
2577 bcache_device_stop(&dc->disk);
2578
2579
2580 /*
2581 * Give an early chance for other kthreads and
2582 * kworkers to stop themselves
2583 */
2584 schedule();
2585
2586 /* What's a condition variable? */
2587 while (1) {
2588 long timeout = start + 10 * HZ - jiffies;
2589
2590 mutex_lock(&bch_register_lock);
2591 stopped = list_empty(&bch_cache_sets) &&
2592 list_empty(&uncached_devices);
2593
2594 if (timeout < 0 || stopped)
2595 break;
2596
2597 prepare_to_wait(&unregister_wait, &wait,
2598 TASK_UNINTERRUPTIBLE);
2599
2600 mutex_unlock(&bch_register_lock);
2601 schedule_timeout(timeout);
2602 }
2603
2604 finish_wait(&unregister_wait, &wait);
2605
2606 if (stopped)
2607 pr_info("All devices stopped");
2608 else
2609 pr_notice("Timeout waiting for devices to be closed");
2610 out:
2611 mutex_unlock(&bch_register_lock);
2612 }
2613
2614 return NOTIFY_DONE;
2615 }
2616
2617 static struct notifier_block reboot = {
2618 .notifier_call = bcache_reboot,
2619 .priority = INT_MAX, /* before any real devices */
2620 };
2621
2622 static void bcache_exit(void)
2623 {
2624 bch_debug_exit();
2625 bch_request_exit();
2626 if (bcache_kobj)
2627 kobject_put(bcache_kobj);
2628 if (bcache_wq)
2629 destroy_workqueue(bcache_wq);
2630 if (bch_journal_wq)
2631 destroy_workqueue(bch_journal_wq);
2632
2633 if (bcache_major)
2634 unregister_blkdev(bcache_major, "bcache");
2635 unregister_reboot_notifier(&reboot);
2636 mutex_destroy(&bch_register_lock);
2637 }
2638
2639 /* Check and fixup module parameters */
2640 static void check_module_parameters(void)
2641 {
2642 if (bch_cutoff_writeback_sync == 0)
2643 bch_cutoff_writeback_sync = CUTOFF_WRITEBACK_SYNC;
2644 else if (bch_cutoff_writeback_sync > CUTOFF_WRITEBACK_SYNC_MAX) {
2645 pr_warn("set bch_cutoff_writeback_sync (%u) to max value %u",
2646 bch_cutoff_writeback_sync, CUTOFF_WRITEBACK_SYNC_MAX);
2647 bch_cutoff_writeback_sync = CUTOFF_WRITEBACK_SYNC_MAX;
2648 }
2649
2650 if (bch_cutoff_writeback == 0)
2651 bch_cutoff_writeback = CUTOFF_WRITEBACK;
2652 else if (bch_cutoff_writeback > CUTOFF_WRITEBACK_MAX) {
2653 pr_warn("set bch_cutoff_writeback (%u) to max value %u",
2654 bch_cutoff_writeback, CUTOFF_WRITEBACK_MAX);
2655 bch_cutoff_writeback = CUTOFF_WRITEBACK_MAX;
2656 }
2657
2658 if (bch_cutoff_writeback > bch_cutoff_writeback_sync) {
2659 pr_warn("set bch_cutoff_writeback (%u) to %u",
2660 bch_cutoff_writeback, bch_cutoff_writeback_sync);
2661 bch_cutoff_writeback = bch_cutoff_writeback_sync;
2662 }
2663 }
2664
2665 static int __init bcache_init(void)
2666 {
2667 static const struct attribute *files[] = {
2668 &ksysfs_register.attr,
2669 &ksysfs_register_quiet.attr,
2670 &ksysfs_pendings_cleanup.attr,
2671 NULL
2672 };
2673
2674 check_module_parameters();
2675
2676 mutex_init(&bch_register_lock);
2677 init_waitqueue_head(&unregister_wait);
2678 register_reboot_notifier(&reboot);
2679
2680 bcache_major = register_blkdev(0, "bcache");
2681 if (bcache_major < 0) {
2682 unregister_reboot_notifier(&reboot);
2683 mutex_destroy(&bch_register_lock);
2684 return bcache_major;
2685 }
2686
2687 bcache_wq = alloc_workqueue("bcache", WQ_MEM_RECLAIM, 0);
2688 if (!bcache_wq)
2689 goto err;
2690
2691 bch_journal_wq = alloc_workqueue("bch_journal", WQ_MEM_RECLAIM, 0);
2692 if (!bch_journal_wq)
2693 goto err;
2694
2695 bcache_kobj = kobject_create_and_add("bcache", fs_kobj);
2696 if (!bcache_kobj)
2697 goto err;
2698
2699 if (bch_request_init() ||
2700 sysfs_create_files(bcache_kobj, files))
2701 goto err;
2702
2703 bch_debug_init();
2704 closure_debug_init();
2705
2706 bcache_is_reboot = false;
2707
2708 return 0;
2709 err:
2710 bcache_exit();
2711 return -ENOMEM;
2712 }
2713
2714 /*
2715 * Module hooks
2716 */
2717 module_exit(bcache_exit);
2718 module_init(bcache_init);
2719
2720 module_param(bch_cutoff_writeback, uint, 0);
2721 MODULE_PARM_DESC(bch_cutoff_writeback, "threshold to cutoff writeback");
2722
2723 module_param(bch_cutoff_writeback_sync, uint, 0);
2724 MODULE_PARM_DESC(bch_cutoff_writeback_sync, "hard threshold to cutoff writeback");
2725
2726 MODULE_DESCRIPTION("Bcache: a Linux block layer cache");
2727 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
2728 MODULE_LICENSE("GPL");
Linux with added FPC-III support