]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - drivers/md/bcache/super.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
bcache: fix static checker warning in bcache_device_free()
[mirror_ubuntu-bionic-kernel.git] / drivers / md / bcache / super.c
1 /*
2 * bcache setup/teardown code, and some metadata io - read a superblock and
3 * figure out what to do with it.
4 *
5 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
6 * Copyright 2012 Google, Inc.
7 */
8
9 #include "bcache.h"
10 #include "btree.h"
11 #include "debug.h"
12 #include "extents.h"
13 #include "request.h"
14 #include "writeback.h"
15
16 #include <linux/blkdev.h>
17 #include <linux/buffer_head.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 MODULE_LICENSE("GPL");
28 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
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 LIST_HEAD(bch_cache_sets);
43 static LIST_HEAD(uncached_devices);
44
45 static int bcache_major;
46 static DEFINE_IDA(bcache_device_idx);
47 static wait_queue_head_t unregister_wait;
48 struct workqueue_struct *bcache_wq;
49
50 #define BTREE_MAX_PAGES (256 * 1024 / PAGE_SIZE)
51 /* limitation of partitions number on single bcache device */
52 #define BCACHE_MINORS 128
53 /* limitation of bcache devices number on single system */
54 #define BCACHE_DEVICE_IDX_MAX ((1U << MINORBITS)/BCACHE_MINORS)
55
56 /* Superblock */
57
58 static const char *read_super(struct cache_sb *sb, struct block_device *bdev,
59 struct page **res)
60 {
61 const char *err;
62 struct cache_sb *s;
63 struct buffer_head *bh = __bread(bdev, 1, SB_SIZE);
64 unsigned i;
65
66 if (!bh)
67 return "IO error";
68
69 s = (struct cache_sb *) bh->b_data;
70
71 sb->offset = le64_to_cpu(s->offset);
72 sb->version = le64_to_cpu(s->version);
73
74 memcpy(sb->magic, s->magic, 16);
75 memcpy(sb->uuid, s->uuid, 16);
76 memcpy(sb->set_uuid, s->set_uuid, 16);
77 memcpy(sb->label, s->label, SB_LABEL_SIZE);
78
79 sb->flags = le64_to_cpu(s->flags);
80 sb->seq = le64_to_cpu(s->seq);
81 sb->last_mount = le32_to_cpu(s->last_mount);
82 sb->first_bucket = le16_to_cpu(s->first_bucket);
83 sb->keys = le16_to_cpu(s->keys);
84
85 for (i = 0; i < SB_JOURNAL_BUCKETS; i++)
86 sb->d[i] = le64_to_cpu(s->d[i]);
87
88 pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u",
89 sb->version, sb->flags, sb->seq, sb->keys);
90
91 err = "Not a bcache superblock";
92 if (sb->offset != SB_SECTOR)
93 goto err;
94
95 if (memcmp(sb->magic, bcache_magic, 16))
96 goto err;
97
98 err = "Too many journal buckets";
99 if (sb->keys > SB_JOURNAL_BUCKETS)
100 goto err;
101
102 err = "Bad checksum";
103 if (s->csum != csum_set(s))
104 goto err;
105
106 err = "Bad UUID";
107 if (bch_is_zero(sb->uuid, 16))
108 goto err;
109
110 sb->block_size = le16_to_cpu(s->block_size);
111
112 err = "Superblock block size smaller than device block size";
113 if (sb->block_size << 9 < bdev_logical_block_size(bdev))
114 goto err;
115
116 switch (sb->version) {
117 case BCACHE_SB_VERSION_BDEV:
118 sb->data_offset = BDEV_DATA_START_DEFAULT;
119 break;
120 case BCACHE_SB_VERSION_BDEV_WITH_OFFSET:
121 sb->data_offset = le64_to_cpu(s->data_offset);
122
123 err = "Bad data offset";
124 if (sb->data_offset < BDEV_DATA_START_DEFAULT)
125 goto err;
126
127 break;
128 case BCACHE_SB_VERSION_CDEV:
129 case BCACHE_SB_VERSION_CDEV_WITH_UUID:
130 sb->nbuckets = le64_to_cpu(s->nbuckets);
131 sb->bucket_size = le16_to_cpu(s->bucket_size);
132
133 sb->nr_in_set = le16_to_cpu(s->nr_in_set);
134 sb->nr_this_dev = le16_to_cpu(s->nr_this_dev);
135
136 err = "Too many buckets";
137 if (sb->nbuckets > LONG_MAX)
138 goto err;
139
140 err = "Not enough buckets";
141 if (sb->nbuckets < 1 << 7)
142 goto err;
143
144 err = "Bad block/bucket size";
145 if (!is_power_of_2(sb->block_size) ||
146 sb->block_size > PAGE_SECTORS ||
147 !is_power_of_2(sb->bucket_size) ||
148 sb->bucket_size < PAGE_SECTORS)
149 goto err;
150
151 err = "Invalid superblock: device too small";
152 if (get_capacity(bdev->bd_disk) < sb->bucket_size * sb->nbuckets)
153 goto err;
154
155 err = "Bad UUID";
156 if (bch_is_zero(sb->set_uuid, 16))
157 goto err;
158
159 err = "Bad cache device number in set";
160 if (!sb->nr_in_set ||
161 sb->nr_in_set <= sb->nr_this_dev ||
162 sb->nr_in_set > MAX_CACHES_PER_SET)
163 goto err;
164
165 err = "Journal buckets not sequential";
166 for (i = 0; i < sb->keys; i++)
167 if (sb->d[i] != sb->first_bucket + i)
168 goto err;
169
170 err = "Too many journal buckets";
171 if (sb->first_bucket + sb->keys > sb->nbuckets)
172 goto err;
173
174 err = "Invalid superblock: first bucket comes before end of super";
175 if (sb->first_bucket * sb->bucket_size < 16)
176 goto err;
177
178 break;
179 default:
180 err = "Unsupported superblock version";
181 goto err;
182 }
183
184 sb->last_mount = get_seconds();
185 err = NULL;
186
187 get_page(bh->b_page);
188 *res = bh->b_page;
189 err:
190 put_bh(bh);
191 return err;
192 }
193
194 static void write_bdev_super_endio(struct bio *bio)
195 {
196 struct cached_dev *dc = bio->bi_private;
197 /* XXX: error checking */
198
199 closure_put(&dc->sb_write);
200 }
201
202 static void __write_super(struct cache_sb *sb, struct bio *bio)
203 {
204 struct cache_sb *out = page_address(bio->bi_io_vec[0].bv_page);
205 unsigned i;
206
207 bio->bi_iter.bi_sector = SB_SECTOR;
208 bio->bi_iter.bi_size = SB_SIZE;
209 bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
210 bch_bio_map(bio, NULL);
211
212 out->offset = cpu_to_le64(sb->offset);
213 out->version = cpu_to_le64(sb->version);
214
215 memcpy(out->uuid, sb->uuid, 16);
216 memcpy(out->set_uuid, sb->set_uuid, 16);
217 memcpy(out->label, sb->label, SB_LABEL_SIZE);
218
219 out->flags = cpu_to_le64(sb->flags);
220 out->seq = cpu_to_le64(sb->seq);
221
222 out->last_mount = cpu_to_le32(sb->last_mount);
223 out->first_bucket = cpu_to_le16(sb->first_bucket);
224 out->keys = cpu_to_le16(sb->keys);
225
226 for (i = 0; i < sb->keys; i++)
227 out->d[i] = cpu_to_le64(sb->d[i]);
228
229 out->csum = csum_set(out);
230
231 pr_debug("ver %llu, flags %llu, seq %llu",
232 sb->version, sb->flags, sb->seq);
233
234 submit_bio(bio);
235 }
236
237 static void bch_write_bdev_super_unlock(struct closure *cl)
238 {
239 struct cached_dev *dc = container_of(cl, struct cached_dev, sb_write);
240
241 up(&dc->sb_write_mutex);
242 }
243
244 void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent)
245 {
246 struct closure *cl = &dc->sb_write;
247 struct bio *bio = &dc->sb_bio;
248
249 down(&dc->sb_write_mutex);
250 closure_init(cl, parent);
251
252 bio_reset(bio);
253 bio_set_dev(bio, dc->bdev);
254 bio->bi_end_io = write_bdev_super_endio;
255 bio->bi_private = dc;
256
257 closure_get(cl);
258 /* I/O request sent to backing device */
259 __write_super(&dc->sb, bio);
260
261 closure_return_with_destructor(cl, bch_write_bdev_super_unlock);
262 }
263
264 static void write_super_endio(struct bio *bio)
265 {
266 struct cache *ca = bio->bi_private;
267
268 bch_count_io_errors(ca, bio->bi_status, "writing superblock");
269 closure_put(&ca->set->sb_write);
270 }
271
272 static void bcache_write_super_unlock(struct closure *cl)
273 {
274 struct cache_set *c = container_of(cl, struct cache_set, sb_write);
275
276 up(&c->sb_write_mutex);
277 }
278
279 void bcache_write_super(struct cache_set *c)
280 {
281 struct closure *cl = &c->sb_write;
282 struct cache *ca;
283 unsigned i;
284
285 down(&c->sb_write_mutex);
286 closure_init(cl, &c->cl);
287
288 c->sb.seq++;
289
290 for_each_cache(ca, c, i) {
291 struct bio *bio = &ca->sb_bio;
292
293 ca->sb.version = BCACHE_SB_VERSION_CDEV_WITH_UUID;
294 ca->sb.seq = c->sb.seq;
295 ca->sb.last_mount = c->sb.last_mount;
296
297 SET_CACHE_SYNC(&ca->sb, CACHE_SYNC(&c->sb));
298
299 bio_reset(bio);
300 bio_set_dev(bio, ca->bdev);
301 bio->bi_end_io = write_super_endio;
302 bio->bi_private = ca;
303
304 closure_get(cl);
305 __write_super(&ca->sb, bio);
306 }
307
308 closure_return_with_destructor(cl, bcache_write_super_unlock);
309 }
310
311 /* UUID io */
312
313 static void uuid_endio(struct bio *bio)
314 {
315 struct closure *cl = bio->bi_private;
316 struct cache_set *c = container_of(cl, struct cache_set, uuid_write);
317
318 cache_set_err_on(bio->bi_status, c, "accessing uuids");
319 bch_bbio_free(bio, c);
320 closure_put(cl);
321 }
322
323 static void uuid_io_unlock(struct closure *cl)
324 {
325 struct cache_set *c = container_of(cl, struct cache_set, uuid_write);
326
327 up(&c->uuid_write_mutex);
328 }
329
330 static void uuid_io(struct cache_set *c, int op, unsigned long op_flags,
331 struct bkey *k, struct closure *parent)
332 {
333 struct closure *cl = &c->uuid_write;
334 struct uuid_entry *u;
335 unsigned i;
336 char buf[80];
337
338 BUG_ON(!parent);
339 down(&c->uuid_write_mutex);
340 closure_init(cl, parent);
341
342 for (i = 0; i < KEY_PTRS(k); i++) {
343 struct bio *bio = bch_bbio_alloc(c);
344
345 bio->bi_opf = REQ_SYNC | REQ_META | op_flags;
346 bio->bi_iter.bi_size = KEY_SIZE(k) << 9;
347
348 bio->bi_end_io = uuid_endio;
349 bio->bi_private = cl;
350 bio_set_op_attrs(bio, op, REQ_SYNC|REQ_META|op_flags);
351 bch_bio_map(bio, c->uuids);
352
353 bch_submit_bbio(bio, c, k, i);
354
355 if (op != REQ_OP_WRITE)
356 break;
357 }
358
359 bch_extent_to_text(buf, sizeof(buf), k);
360 pr_debug("%s UUIDs at %s", op == REQ_OP_WRITE ? "wrote" : "read", buf);
361
362 for (u = c->uuids; u < c->uuids + c->nr_uuids; u++)
363 if (!bch_is_zero(u->uuid, 16))
364 pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u",
365 u - c->uuids, u->uuid, u->label,
366 u->first_reg, u->last_reg, u->invalidated);
367
368 closure_return_with_destructor(cl, uuid_io_unlock);
369 }
370
371 static char *uuid_read(struct cache_set *c, struct jset *j, struct closure *cl)
372 {
373 struct bkey *k = &j->uuid_bucket;
374
375 if (__bch_btree_ptr_invalid(c, k))
376 return "bad uuid pointer";
377
378 bkey_copy(&c->uuid_bucket, k);
379 uuid_io(c, REQ_OP_READ, 0, k, cl);
380
381 if (j->version < BCACHE_JSET_VERSION_UUIDv1) {
382 struct uuid_entry_v0 *u0 = (void *) c->uuids;
383 struct uuid_entry *u1 = (void *) c->uuids;
384 int i;
385
386 closure_sync(cl);
387
388 /*
389 * Since the new uuid entry is bigger than the old, we have to
390 * convert starting at the highest memory address and work down
391 * in order to do it in place
392 */
393
394 for (i = c->nr_uuids - 1;
395 i >= 0;
396 --i) {
397 memcpy(u1[i].uuid, u0[i].uuid, 16);
398 memcpy(u1[i].label, u0[i].label, 32);
399
400 u1[i].first_reg = u0[i].first_reg;
401 u1[i].last_reg = u0[i].last_reg;
402 u1[i].invalidated = u0[i].invalidated;
403
404 u1[i].flags = 0;
405 u1[i].sectors = 0;
406 }
407 }
408
409 return NULL;
410 }
411
412 static int __uuid_write(struct cache_set *c)
413 {
414 BKEY_PADDED(key) k;
415 struct closure cl;
416 struct cache *ca;
417
418 closure_init_stack(&cl);
419
420 lockdep_assert_held(&bch_register_lock);
421
422 if (bch_bucket_alloc_set(c, RESERVE_BTREE, &k.key, 1, true))
423 return 1;
424
425 SET_KEY_SIZE(&k.key, c->sb.bucket_size);
426 uuid_io(c, REQ_OP_WRITE, 0, &k.key, &cl);
427 closure_sync(&cl);
428
429 /* Only one bucket used for uuid write */
430 ca = PTR_CACHE(c, &k.key, 0);
431 atomic_long_add(ca->sb.bucket_size, &ca->meta_sectors_written);
432
433 bkey_copy(&c->uuid_bucket, &k.key);
434 bkey_put(c, &k.key);
435 return 0;
436 }
437
438 int bch_uuid_write(struct cache_set *c)
439 {
440 int ret = __uuid_write(c);
441
442 if (!ret)
443 bch_journal_meta(c, NULL);
444
445 return ret;
446 }
447
448 static struct uuid_entry *uuid_find(struct cache_set *c, const char *uuid)
449 {
450 struct uuid_entry *u;
451
452 for (u = c->uuids;
453 u < c->uuids + c->nr_uuids; u++)
454 if (!memcmp(u->uuid, uuid, 16))
455 return u;
456
457 return NULL;
458 }
459
460 static struct uuid_entry *uuid_find_empty(struct cache_set *c)
461 {
462 static const char zero_uuid[16] = "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0";
463 return uuid_find(c, zero_uuid);
464 }
465
466 /*
467 * Bucket priorities/gens:
468 *
469 * For each bucket, we store on disk its
470 * 8 bit gen
471 * 16 bit priority
472 *
473 * See alloc.c for an explanation of the gen. The priority is used to implement
474 * lru (and in the future other) cache replacement policies; for most purposes
475 * it's just an opaque integer.
476 *
477 * The gens and the priorities don't have a whole lot to do with each other, and
478 * it's actually the gens that must be written out at specific times - it's no
479 * big deal if the priorities don't get written, if we lose them we just reuse
480 * buckets in suboptimal order.
481 *
482 * On disk they're stored in a packed array, and in as many buckets are required
483 * to fit them all. The buckets we use to store them form a list; the journal
484 * header points to the first bucket, the first bucket points to the second
485 * bucket, et cetera.
486 *
487 * This code is used by the allocation code; periodically (whenever it runs out
488 * of buckets to allocate from) the allocation code will invalidate some
489 * buckets, but it can't use those buckets until their new gens are safely on
490 * disk.
491 */
492
493 static void prio_endio(struct bio *bio)
494 {
495 struct cache *ca = bio->bi_private;
496
497 cache_set_err_on(bio->bi_status, ca->set, "accessing priorities");
498 bch_bbio_free(bio, ca->set);
499 closure_put(&ca->prio);
500 }
501
502 static void prio_io(struct cache *ca, uint64_t bucket, int op,
503 unsigned long op_flags)
504 {
505 struct closure *cl = &ca->prio;
506 struct bio *bio = bch_bbio_alloc(ca->set);
507
508 closure_init_stack(cl);
509
510 bio->bi_iter.bi_sector = bucket * ca->sb.bucket_size;
511 bio_set_dev(bio, ca->bdev);
512 bio->bi_iter.bi_size = bucket_bytes(ca);
513
514 bio->bi_end_io = prio_endio;
515 bio->bi_private = ca;
516 bio_set_op_attrs(bio, op, REQ_SYNC|REQ_META|op_flags);
517 bch_bio_map(bio, ca->disk_buckets);
518
519 closure_bio_submit(ca->set, bio, &ca->prio);
520 closure_sync(cl);
521 }
522
523 int bch_prio_write(struct cache *ca, bool wait)
524 {
525 int i;
526 struct bucket *b;
527 struct closure cl;
528
529 pr_debug("free_prio=%zu, free_none=%zu, free_inc=%zu",
530 fifo_used(&ca->free[RESERVE_PRIO]),
531 fifo_used(&ca->free[RESERVE_NONE]),
532 fifo_used(&ca->free_inc));
533
534 /*
535 * Pre-check if there are enough free buckets. In the non-blocking
536 * scenario it's better to fail early rather than starting to allocate
537 * buckets and do a cleanup later in case of failure.
538 */
539 if (!wait) {
540 size_t avail = fifo_used(&ca->free[RESERVE_PRIO]) +
541 fifo_used(&ca->free[RESERVE_NONE]);
542 if (prio_buckets(ca) > avail)
543 return -ENOMEM;
544 }
545
546 closure_init_stack(&cl);
547
548 lockdep_assert_held(&ca->set->bucket_lock);
549
550 ca->disk_buckets->seq++;
551
552 atomic_long_add(ca->sb.bucket_size * prio_buckets(ca),
553 &ca->meta_sectors_written);
554
555 for (i = prio_buckets(ca) - 1; i >= 0; --i) {
556 long bucket;
557 struct prio_set *p = ca->disk_buckets;
558 struct bucket_disk *d = p->data;
559 struct bucket_disk *end = d + prios_per_bucket(ca);
560
561 for (b = ca->buckets + i * prios_per_bucket(ca);
562 b < ca->buckets + ca->sb.nbuckets && d < end;
563 b++, d++) {
564 d->prio = cpu_to_le16(b->prio);
565 d->gen = b->gen;
566 }
567
568 p->next_bucket = ca->prio_buckets[i + 1];
569 p->magic = pset_magic(&ca->sb);
570 p->csum = bch_crc64(&p->magic, bucket_bytes(ca) - 8);
571
572 bucket = bch_bucket_alloc(ca, RESERVE_PRIO, wait);
573 BUG_ON(bucket == -1);
574
575 mutex_unlock(&ca->set->bucket_lock);
576 prio_io(ca, bucket, REQ_OP_WRITE, 0);
577 mutex_lock(&ca->set->bucket_lock);
578
579 ca->prio_buckets[i] = bucket;
580 atomic_dec_bug(&ca->buckets[bucket].pin);
581 }
582
583 mutex_unlock(&ca->set->bucket_lock);
584
585 bch_journal_meta(ca->set, &cl);
586 closure_sync(&cl);
587
588 mutex_lock(&ca->set->bucket_lock);
589
590 /*
591 * Don't want the old priorities to get garbage collected until after we
592 * finish writing the new ones, and they're journalled
593 */
594 for (i = 0; i < prio_buckets(ca); i++) {
595 if (ca->prio_last_buckets[i])
596 __bch_bucket_free(ca,
597 &ca->buckets[ca->prio_last_buckets[i]]);
598
599 ca->prio_last_buckets[i] = ca->prio_buckets[i];
600 }
601 return 0;
602 }
603
604 static void prio_read(struct cache *ca, uint64_t bucket)
605 {
606 struct prio_set *p = ca->disk_buckets;
607 struct bucket_disk *d = p->data + prios_per_bucket(ca), *end = d;
608 struct bucket *b;
609 unsigned bucket_nr = 0;
610
611 for (b = ca->buckets;
612 b < ca->buckets + ca->sb.nbuckets;
613 b++, d++) {
614 if (d == end) {
615 ca->prio_buckets[bucket_nr] = bucket;
616 ca->prio_last_buckets[bucket_nr] = bucket;
617 bucket_nr++;
618
619 prio_io(ca, bucket, REQ_OP_READ, 0);
620
621 if (p->csum != bch_crc64(&p->magic, bucket_bytes(ca) - 8))
622 pr_warn("bad csum reading priorities");
623
624 if (p->magic != pset_magic(&ca->sb))
625 pr_warn("bad magic reading priorities");
626
627 bucket = p->next_bucket;
628 d = p->data;
629 }
630
631 b->prio = le16_to_cpu(d->prio);
632 b->gen = b->last_gc = d->gen;
633 }
634 }
635
636 /* Bcache device */
637
638 static int open_dev(struct block_device *b, fmode_t mode)
639 {
640 struct bcache_device *d = b->bd_disk->private_data;
641 if (test_bit(BCACHE_DEV_CLOSING, &d->flags))
642 return -ENXIO;
643
644 closure_get(&d->cl);
645 return 0;
646 }
647
648 static void release_dev(struct gendisk *b, fmode_t mode)
649 {
650 struct bcache_device *d = b->private_data;
651 closure_put(&d->cl);
652 }
653
654 static int ioctl_dev(struct block_device *b, fmode_t mode,
655 unsigned int cmd, unsigned long arg)
656 {
657 struct bcache_device *d = b->bd_disk->private_data;
658
659 return d->ioctl(d, mode, cmd, arg);
660 }
661
662 static const struct block_device_operations bcache_ops = {
663 .open = open_dev,
664 .release = release_dev,
665 .ioctl = ioctl_dev,
666 .owner = THIS_MODULE,
667 };
668
669 void bcache_device_stop(struct bcache_device *d)
670 {
671 if (!test_and_set_bit(BCACHE_DEV_CLOSING, &d->flags))
672 closure_queue(&d->cl);
673 }
674
675 static void bcache_device_unlink(struct bcache_device *d)
676 {
677 lockdep_assert_held(&bch_register_lock);
678
679 if (d->c && !test_and_set_bit(BCACHE_DEV_UNLINK_DONE, &d->flags)) {
680 unsigned i;
681 struct cache *ca;
682
683 sysfs_remove_link(&d->c->kobj, d->name);
684 sysfs_remove_link(&d->kobj, "cache");
685
686 for_each_cache(ca, d->c, i)
687 bd_unlink_disk_holder(ca->bdev, d->disk);
688 }
689 }
690
691 static void bcache_device_link(struct bcache_device *d, struct cache_set *c,
692 const char *name)
693 {
694 unsigned i;
695 struct cache *ca;
696
697 for_each_cache(ca, d->c, i)
698 bd_link_disk_holder(ca->bdev, d->disk);
699
700 snprintf(d->name, BCACHEDEVNAME_SIZE,
701 "%s%u", name, d->id);
702
703 WARN(sysfs_create_link(&d->kobj, &c->kobj, "cache") ||
704 sysfs_create_link(&c->kobj, &d->kobj, d->name),
705 "Couldn't create device <-> cache set symlinks");
706
707 clear_bit(BCACHE_DEV_UNLINK_DONE, &d->flags);
708 }
709
710 static void bcache_device_detach(struct bcache_device *d)
711 {
712 lockdep_assert_held(&bch_register_lock);
713
714 if (test_bit(BCACHE_DEV_DETACHING, &d->flags)) {
715 struct uuid_entry *u = d->c->uuids + d->id;
716
717 SET_UUID_FLASH_ONLY(u, 0);
718 memcpy(u->uuid, invalid_uuid, 16);
719 u->invalidated = cpu_to_le32(get_seconds());
720 bch_uuid_write(d->c);
721 }
722
723 bcache_device_unlink(d);
724
725 d->c->devices[d->id] = NULL;
726 closure_put(&d->c->caching);
727 d->c = NULL;
728 }
729
730 static void bcache_device_attach(struct bcache_device *d, struct cache_set *c,
731 unsigned id)
732 {
733 d->id = id;
734 d->c = c;
735 c->devices[id] = d;
736
737 closure_get(&c->caching);
738 }
739
740 static inline int first_minor_to_idx(int first_minor)
741 {
742 return (first_minor/BCACHE_MINORS);
743 }
744
745 static inline int idx_to_first_minor(int idx)
746 {
747 return (idx * BCACHE_MINORS);
748 }
749
750 static void bcache_device_free(struct bcache_device *d)
751 {
752 struct gendisk *disk = d->disk;
753
754 lockdep_assert_held(&bch_register_lock);
755
756 if (disk)
757 pr_info("%s stopped", disk->disk_name);
758 else
759 pr_err("bcache device (NULL gendisk) stopped");
760
761 if (d->c)
762 bcache_device_detach(d);
763
764 if (disk) {
765 if (disk->flags & GENHD_FL_UP)
766 del_gendisk(disk);
767
768 if (disk->queue)
769 blk_cleanup_queue(disk->queue);
770
771 ida_simple_remove(&bcache_device_idx,
772 first_minor_to_idx(disk->first_minor));
773 put_disk(disk);
774 }
775
776 if (d->bio_split)
777 bioset_free(d->bio_split);
778 kvfree(d->full_dirty_stripes);
779 kvfree(d->stripe_sectors_dirty);
780
781 closure_debug_destroy(&d->cl);
782 }
783
784 static int bcache_device_init(struct bcache_device *d, unsigned block_size,
785 sector_t sectors)
786 {
787 struct request_queue *q;
788 const size_t max_stripes = min_t(size_t, INT_MAX,
789 SIZE_MAX / sizeof(atomic_t));
790 size_t n;
791 int idx;
792
793 if (!d->stripe_size)
794 d->stripe_size = 1 << 31;
795
796 d->nr_stripes = DIV_ROUND_UP_ULL(sectors, d->stripe_size);
797
798 if (!d->nr_stripes || d->nr_stripes > max_stripes) {
799 pr_err("nr_stripes too large or invalid: %u (start sector beyond end of disk?)",
800 (unsigned)d->nr_stripes);
801 return -ENOMEM;
802 }
803
804 n = d->nr_stripes * sizeof(atomic_t);
805 d->stripe_sectors_dirty = kvzalloc(n, GFP_KERNEL);
806 if (!d->stripe_sectors_dirty)
807 return -ENOMEM;
808
809 n = BITS_TO_LONGS(d->nr_stripes) * sizeof(unsigned long);
810 d->full_dirty_stripes = kvzalloc(n, GFP_KERNEL);
811 if (!d->full_dirty_stripes)
812 return -ENOMEM;
813
814 idx = ida_simple_get(&bcache_device_idx, 0,
815 BCACHE_DEVICE_IDX_MAX, GFP_KERNEL);
816 if (idx < 0)
817 return idx;
818
819 if (!(d->bio_split = bioset_create(4, offsetof(struct bbio, bio),
820 BIOSET_NEED_BVECS |
821 BIOSET_NEED_RESCUER)) ||
822 !(d->disk = alloc_disk(BCACHE_MINORS))) {
823 ida_simple_remove(&bcache_device_idx, idx);
824 return -ENOMEM;
825 }
826
827 set_capacity(d->disk, sectors);
828 snprintf(d->disk->disk_name, DISK_NAME_LEN, "bcache%i", idx);
829
830 d->disk->major = bcache_major;
831 d->disk->first_minor = idx_to_first_minor(idx);
832 d->disk->fops = &bcache_ops;
833 d->disk->private_data = d;
834
835 q = blk_alloc_queue(GFP_KERNEL);
836 if (!q)
837 return -ENOMEM;
838
839 blk_queue_make_request(q, NULL);
840 d->disk->queue = q;
841 q->queuedata = d;
842 q->backing_dev_info->congested_data = d;
843 q->limits.max_hw_sectors = UINT_MAX;
844 q->limits.max_sectors = UINT_MAX;
845 q->limits.max_segment_size = UINT_MAX;
846 q->limits.max_segments = BIO_MAX_PAGES;
847 blk_queue_max_discard_sectors(q, UINT_MAX);
848 q->limits.discard_granularity = 512;
849 q->limits.io_min = block_size;
850 q->limits.logical_block_size = block_size;
851 q->limits.physical_block_size = block_size;
852 set_bit(QUEUE_FLAG_NONROT, &d->disk->queue->queue_flags);
853 clear_bit(QUEUE_FLAG_ADD_RANDOM, &d->disk->queue->queue_flags);
854 set_bit(QUEUE_FLAG_DISCARD, &d->disk->queue->queue_flags);
855
856 blk_queue_write_cache(q, true, true);
857
858 return 0;
859 }
860
861 /* Cached device */
862
863 static void calc_cached_dev_sectors(struct cache_set *c)
864 {
865 uint64_t sectors = 0;
866 struct cached_dev *dc;
867
868 list_for_each_entry(dc, &c->cached_devs, list)
869 sectors += bdev_sectors(dc->bdev);
870
871 c->cached_dev_sectors = sectors;
872 }
873
874 #define BACKING_DEV_OFFLINE_TIMEOUT 5
875 static int cached_dev_status_update(void *arg)
876 {
877 struct cached_dev *dc = arg;
878 struct request_queue *q;
879
880 /*
881 * If this delayed worker is stopping outside, directly quit here.
882 * dc->io_disable might be set via sysfs interface, so check it
883 * here too.
884 */
885 while (!kthread_should_stop() && !dc->io_disable) {
886 q = bdev_get_queue(dc->bdev);
887 if (blk_queue_dying(q))
888 dc->offline_seconds++;
889 else
890 dc->offline_seconds = 0;
891
892 if (dc->offline_seconds >= BACKING_DEV_OFFLINE_TIMEOUT) {
893 pr_err("%s: device offline for %d seconds",
894 dc->backing_dev_name,
895 BACKING_DEV_OFFLINE_TIMEOUT);
896 pr_err("%s: disable I/O request due to backing "
897 "device offline", dc->disk.name);
898 dc->io_disable = true;
899 /* let others know earlier that io_disable is true */
900 smp_mb();
901 bcache_device_stop(&dc->disk);
902 break;
903 }
904 schedule_timeout_interruptible(HZ);
905 }
906
907 wait_for_kthread_stop();
908 return 0;
909 }
910
911
912 void bch_cached_dev_emit_change(struct cached_dev *dc)
913 {
914 struct bcache_device *d = &dc->disk;
915 char buf[SB_LABEL_SIZE + 1];
916 char *env[] = {
917 "DRIVER=bcache",
918 kasprintf(GFP_KERNEL, "CACHED_UUID=%pU", dc->sb.uuid),
919 NULL,
920 NULL,
921 };
922
923 memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
924 buf[SB_LABEL_SIZE] = '\0';
925 env[2] = kasprintf(GFP_KERNEL, "CACHED_LABEL=%s", buf);
926
927 /* won't show up in the uevent file, use udevadm monitor -e instead
928 * only class / kset properties are persistent */
929 kobject_uevent_env(&disk_to_dev(d->disk)->kobj, KOBJ_CHANGE, env);
930 kfree(env[1]);
931 kfree(env[2]);
932
933 }
934
935 void bch_cached_dev_run(struct cached_dev *dc)
936 {
937 struct bcache_device *d = &dc->disk;
938 if (atomic_xchg(&dc->running, 1)) {
939 return;
940 }
941
942 if (!d->c &&
943 BDEV_STATE(&dc->sb) != BDEV_STATE_NONE) {
944 struct closure cl;
945 closure_init_stack(&cl);
946
947 SET_BDEV_STATE(&dc->sb, BDEV_STATE_STALE);
948 bch_write_bdev_super(dc, &cl);
949 closure_sync(&cl);
950 }
951
952 add_disk(d->disk);
953 bd_link_disk_holder(dc->bdev, dc->disk.disk);
954
955 /* emit change event */
956 bch_cached_dev_emit_change(dc);
957
958 if (sysfs_create_link(&d->kobj, &disk_to_dev(d->disk)->kobj, "dev") ||
959 sysfs_create_link(&disk_to_dev(d->disk)->kobj, &d->kobj, "bcache"))
960 pr_debug("error creating sysfs link");
961
962 dc->status_update_thread = kthread_run(cached_dev_status_update,
963 dc, "bcache_status_update");
964 if (IS_ERR(dc->status_update_thread)) {
965 pr_warn("failed to create bcache_status_update kthread, "
966 "continue to run without monitoring backing "
967 "device status");
968 }
969 }
970
971 /*
972 * If BCACHE_DEV_RATE_DW_RUNNING is set, it means routine of the delayed
973 * work dc->writeback_rate_update is running. Wait until the routine
974 * quits (BCACHE_DEV_RATE_DW_RUNNING is clear), then continue to
975 * cancel it. If BCACHE_DEV_RATE_DW_RUNNING is not clear after time_out
976 * seconds, give up waiting here and continue to cancel it too.
977 */
978 static void cancel_writeback_rate_update_dwork(struct cached_dev *dc)
979 {
980 int time_out = WRITEBACK_RATE_UPDATE_SECS_MAX * HZ;
981
982 do {
983 if (!test_bit(BCACHE_DEV_RATE_DW_RUNNING,
984 &dc->disk.flags))
985 break;
986 time_out--;
987 schedule_timeout_interruptible(1);
988 } while (time_out > 0);
989
990 if (time_out == 0)
991 pr_warn("give up waiting for dc->writeback_write_update to quit");
992
993 cancel_delayed_work_sync(&dc->writeback_rate_update);
994 }
995
996 static void cached_dev_detach_finish(struct work_struct *w)
997 {
998 struct cached_dev *dc = container_of(w, struct cached_dev, detach);
999 struct closure cl;
1000 closure_init_stack(&cl);
1001
1002 BUG_ON(!test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags));
1003 BUG_ON(refcount_read(&dc->count));
1004
1005 mutex_lock(&bch_register_lock);
1006
1007 if (test_and_clear_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags))
1008 cancel_writeback_rate_update_dwork(dc);
1009
1010 if (!IS_ERR_OR_NULL(dc->writeback_thread)) {
1011 kthread_stop(dc->writeback_thread);
1012 dc->writeback_thread = NULL;
1013 }
1014
1015 memset(&dc->sb.set_uuid, 0, 16);
1016 SET_BDEV_STATE(&dc->sb, BDEV_STATE_NONE);
1017
1018 bch_write_bdev_super(dc, &cl);
1019 closure_sync(&cl);
1020
1021 calc_cached_dev_sectors(dc->disk.c);
1022 bcache_device_detach(&dc->disk);
1023 list_move(&dc->list, &uncached_devices);
1024
1025 clear_bit(BCACHE_DEV_DETACHING, &dc->disk.flags);
1026 clear_bit(BCACHE_DEV_UNLINK_DONE, &dc->disk.flags);
1027
1028 mutex_unlock(&bch_register_lock);
1029
1030 pr_info("Caching disabled for %s", dc->backing_dev_name);
1031
1032 /* Drop ref we took in cached_dev_detach() */
1033 closure_put(&dc->disk.cl);
1034 }
1035
1036 void bch_cached_dev_detach(struct cached_dev *dc)
1037 {
1038 lockdep_assert_held(&bch_register_lock);
1039
1040 if (test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags))
1041 return;
1042
1043 if (test_and_set_bit(BCACHE_DEV_DETACHING, &dc->disk.flags))
1044 return;
1045
1046 /*
1047 * Block the device from being closed and freed until we're finished
1048 * detaching
1049 */
1050 closure_get(&dc->disk.cl);
1051
1052 bch_writeback_queue(dc);
1053
1054 cached_dev_put(dc);
1055 }
1056
1057 int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c,
1058 uint8_t *set_uuid)
1059 {
1060 uint32_t rtime = cpu_to_le32(get_seconds());
1061 struct uuid_entry *u;
1062 struct cached_dev *exist_dc, *t;
1063
1064 if ((set_uuid && memcmp(set_uuid, c->sb.set_uuid, 16)) ||
1065 (!set_uuid && memcmp(dc->sb.set_uuid, c->sb.set_uuid, 16)))
1066 return -ENOENT;
1067
1068 if (dc->disk.c) {
1069 pr_err("Can't attach %s: already attached",
1070 dc->backing_dev_name);
1071 return -EINVAL;
1072 }
1073
1074 if (test_bit(CACHE_SET_STOPPING, &c->flags)) {
1075 pr_err("Can't attach %s: shutting down",
1076 dc->backing_dev_name);
1077 return -EINVAL;
1078 }
1079
1080 if (dc->sb.block_size < c->sb.block_size) {
1081 /* Will die */
1082 pr_err("Couldn't attach %s: block size less than set's block size",
1083 dc->backing_dev_name);
1084 return -EINVAL;
1085 }
1086
1087 /* Check whether already attached */
1088 list_for_each_entry_safe(exist_dc, t, &c->cached_devs, list) {
1089 if (!memcmp(dc->sb.uuid, exist_dc->sb.uuid, 16)) {
1090 pr_err("Tried to attach %s but duplicate UUID already attached",
1091 dc->backing_dev_name);
1092
1093 return -EINVAL;
1094 }
1095 }
1096
1097 u = uuid_find(c, dc->sb.uuid);
1098
1099 if (u &&
1100 (BDEV_STATE(&dc->sb) == BDEV_STATE_STALE ||
1101 BDEV_STATE(&dc->sb) == BDEV_STATE_NONE)) {
1102 memcpy(u->uuid, invalid_uuid, 16);
1103 u->invalidated = cpu_to_le32(get_seconds());
1104 u = NULL;
1105 }
1106
1107 if (!u) {
1108 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
1109 pr_err("Couldn't find uuid for %s in set",
1110 dc->backing_dev_name);
1111 return -ENOENT;
1112 }
1113
1114 u = uuid_find_empty(c);
1115 if (!u) {
1116 pr_err("Not caching %s, no room for UUID",
1117 dc->backing_dev_name);
1118 return -EINVAL;
1119 }
1120 }
1121
1122 /* Deadlocks since we're called via sysfs...
1123 sysfs_remove_file(&dc->kobj, &sysfs_attach);
1124 */
1125
1126 if (bch_is_zero(u->uuid, 16)) {
1127 struct closure cl;
1128 closure_init_stack(&cl);
1129
1130 memcpy(u->uuid, dc->sb.uuid, 16);
1131 memcpy(u->label, dc->sb.label, SB_LABEL_SIZE);
1132 u->first_reg = u->last_reg = rtime;
1133 bch_uuid_write(c);
1134
1135 memcpy(dc->sb.set_uuid, c->sb.set_uuid, 16);
1136 SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN);
1137
1138 bch_write_bdev_super(dc, &cl);
1139 closure_sync(&cl);
1140 } else {
1141 u->last_reg = rtime;
1142 bch_uuid_write(c);
1143 }
1144
1145 bcache_device_attach(&dc->disk, c, u - c->uuids);
1146 list_move(&dc->list, &c->cached_devs);
1147 calc_cached_dev_sectors(c);
1148
1149 smp_wmb();
1150 /*
1151 * dc->c must be set before dc->count != 0 - paired with the mb in
1152 * cached_dev_get()
1153 */
1154 refcount_set(&dc->count, 1);
1155
1156 /* Block writeback thread, but spawn it */
1157 down_write(&dc->writeback_lock);
1158 if (bch_cached_dev_writeback_start(dc)) {
1159 up_write(&dc->writeback_lock);
1160 return -ENOMEM;
1161 }
1162
1163 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
1164 atomic_set(&dc->has_dirty, 1);
1165 bch_writeback_queue(dc);
1166 }
1167
1168 bch_sectors_dirty_init(&dc->disk);
1169
1170 bch_cached_dev_run(dc);
1171 bcache_device_link(&dc->disk, c, "bdev");
1172
1173 /* Allow the writeback thread to proceed */
1174 up_write(&dc->writeback_lock);
1175
1176 pr_info("Caching %s as %s on set %pU",
1177 dc->backing_dev_name,
1178 dc->disk.disk->disk_name,
1179 dc->disk.c->sb.set_uuid);
1180 return 0;
1181 }
1182
1183 void bch_cached_dev_release(struct kobject *kobj)
1184 {
1185 struct cached_dev *dc = container_of(kobj, struct cached_dev,
1186 disk.kobj);
1187 kfree(dc);
1188 module_put(THIS_MODULE);
1189 }
1190
1191 static void cached_dev_free(struct closure *cl)
1192 {
1193 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1194
1195 if (test_and_clear_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags))
1196 cancel_writeback_rate_update_dwork(dc);
1197
1198 if (!IS_ERR_OR_NULL(dc->writeback_thread))
1199 kthread_stop(dc->writeback_thread);
1200 if (!IS_ERR_OR_NULL(dc->status_update_thread))
1201 kthread_stop(dc->status_update_thread);
1202
1203 mutex_lock(&bch_register_lock);
1204
1205 if (atomic_read(&dc->running))
1206 bd_unlink_disk_holder(dc->bdev, dc->disk.disk);
1207 bcache_device_free(&dc->disk);
1208 list_del(&dc->list);
1209
1210 mutex_unlock(&bch_register_lock);
1211
1212 if (!IS_ERR_OR_NULL(dc->bdev))
1213 blkdev_put(dc->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1214
1215 wake_up(&unregister_wait);
1216
1217 kobject_put(&dc->disk.kobj);
1218 }
1219
1220 static void cached_dev_flush(struct closure *cl)
1221 {
1222 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1223 struct bcache_device *d = &dc->disk;
1224
1225 mutex_lock(&bch_register_lock);
1226 bcache_device_unlink(d);
1227 mutex_unlock(&bch_register_lock);
1228
1229 bch_cache_accounting_destroy(&dc->accounting);
1230 kobject_del(&d->kobj);
1231
1232 continue_at(cl, cached_dev_free, system_wq);
1233 }
1234
1235 static int cached_dev_init(struct cached_dev *dc, unsigned block_size)
1236 {
1237 int ret;
1238 struct io *io;
1239 struct request_queue *q = bdev_get_queue(dc->bdev);
1240
1241 __module_get(THIS_MODULE);
1242 INIT_LIST_HEAD(&dc->list);
1243 closure_init(&dc->disk.cl, NULL);
1244 set_closure_fn(&dc->disk.cl, cached_dev_flush, system_wq);
1245 kobject_init(&dc->disk.kobj, &bch_cached_dev_ktype);
1246 INIT_WORK(&dc->detach, cached_dev_detach_finish);
1247 sema_init(&dc->sb_write_mutex, 1);
1248 INIT_LIST_HEAD(&dc->io_lru);
1249 spin_lock_init(&dc->io_lock);
1250 bch_cache_accounting_init(&dc->accounting, &dc->disk.cl);
1251
1252 dc->sequential_cutoff = 4 << 20;
1253
1254 for (io = dc->io; io < dc->io + RECENT_IO; io++) {
1255 list_add(&io->lru, &dc->io_lru);
1256 hlist_add_head(&io->hash, dc->io_hash + RECENT_IO);
1257 }
1258
1259 dc->disk.stripe_size = q->limits.io_opt >> 9;
1260
1261 if (dc->disk.stripe_size)
1262 dc->partial_stripes_expensive =
1263 q->limits.raid_partial_stripes_expensive;
1264
1265 ret = bcache_device_init(&dc->disk, block_size,
1266 dc->bdev->bd_part->nr_sects - dc->sb.data_offset);
1267 if (ret)
1268 return ret;
1269
1270 dc->disk.disk->queue->backing_dev_info->ra_pages =
1271 max(dc->disk.disk->queue->backing_dev_info->ra_pages,
1272 q->backing_dev_info->ra_pages);
1273
1274 atomic_set(&dc->io_errors, 0);
1275 dc->io_disable = false;
1276 dc->error_limit = DEFAULT_CACHED_DEV_ERROR_LIMIT;
1277 /* default to auto */
1278 dc->stop_when_cache_set_failed = BCH_CACHED_DEV_STOP_AUTO;
1279
1280 bch_cached_dev_request_init(dc);
1281 bch_cached_dev_writeback_init(dc);
1282 return 0;
1283 }
1284
1285 /* Cached device - bcache superblock */
1286
1287 static void register_bdev(struct cache_sb *sb, struct page *sb_page,
1288 struct block_device *bdev,
1289 struct cached_dev *dc)
1290 {
1291 const char *err = "cannot allocate memory";
1292 struct cache_set *c;
1293
1294 bdevname(bdev, dc->backing_dev_name);
1295 memcpy(&dc->sb, sb, sizeof(struct cache_sb));
1296 dc->bdev = bdev;
1297 dc->bdev->bd_holder = dc;
1298
1299 bio_init(&dc->sb_bio, dc->sb_bio.bi_inline_vecs, 1);
1300 dc->sb_bio.bi_io_vec[0].bv_page = sb_page;
1301 get_page(sb_page);
1302
1303
1304 if (cached_dev_init(dc, sb->block_size << 9))
1305 goto err;
1306
1307 err = "error creating kobject";
1308 if (kobject_add(&dc->disk.kobj, &part_to_dev(bdev->bd_part)->kobj,
1309 "bcache"))
1310 goto err;
1311 if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj))
1312 goto err;
1313
1314 pr_info("registered backing device %s", dc->backing_dev_name);
1315
1316 list_add(&dc->list, &uncached_devices);
1317 list_for_each_entry(c, &bch_cache_sets, list)
1318 bch_cached_dev_attach(dc, c, NULL);
1319
1320 if (BDEV_STATE(&dc->sb) == BDEV_STATE_NONE ||
1321 BDEV_STATE(&dc->sb) == BDEV_STATE_STALE)
1322 bch_cached_dev_run(dc);
1323
1324 return;
1325 err:
1326 pr_notice("error %s: %s", dc->backing_dev_name, err);
1327 bcache_device_stop(&dc->disk);
1328 }
1329
1330 /* Flash only volumes */
1331
1332 void bch_flash_dev_release(struct kobject *kobj)
1333 {
1334 struct bcache_device *d = container_of(kobj, struct bcache_device,
1335 kobj);
1336 kfree(d);
1337 }
1338
1339 static void flash_dev_free(struct closure *cl)
1340 {
1341 struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1342 mutex_lock(&bch_register_lock);
1343 bcache_device_free(d);
1344 mutex_unlock(&bch_register_lock);
1345 kobject_put(&d->kobj);
1346 }
1347
1348 static void flash_dev_flush(struct closure *cl)
1349 {
1350 struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1351
1352 mutex_lock(&bch_register_lock);
1353 bcache_device_unlink(d);
1354 mutex_unlock(&bch_register_lock);
1355 kobject_del(&d->kobj);
1356 continue_at(cl, flash_dev_free, system_wq);
1357 }
1358
1359 static int flash_dev_run(struct cache_set *c, struct uuid_entry *u)
1360 {
1361 struct bcache_device *d = kzalloc(sizeof(struct bcache_device),
1362 GFP_KERNEL);
1363 if (!d)
1364 return -ENOMEM;
1365
1366 closure_init(&d->cl, NULL);
1367 set_closure_fn(&d->cl, flash_dev_flush, system_wq);
1368
1369 kobject_init(&d->kobj, &bch_flash_dev_ktype);
1370
1371 if (bcache_device_init(d, block_bytes(c), u->sectors))
1372 goto err;
1373
1374 bcache_device_attach(d, c, u - c->uuids);
1375 bch_sectors_dirty_init(d);
1376 bch_flash_dev_request_init(d);
1377 add_disk(d->disk);
1378
1379 if (kobject_add(&d->kobj, &disk_to_dev(d->disk)->kobj, "bcache"))
1380 goto err;
1381
1382 bcache_device_link(d, c, "volume");
1383
1384 return 0;
1385 err:
1386 kobject_put(&d->kobj);
1387 return -ENOMEM;
1388 }
1389
1390 static int flash_devs_run(struct cache_set *c)
1391 {
1392 int ret = 0;
1393 struct uuid_entry *u;
1394
1395 for (u = c->uuids;
1396 u < c->uuids + c->nr_uuids && !ret;
1397 u++)
1398 if (UUID_FLASH_ONLY(u))
1399 ret = flash_dev_run(c, u);
1400
1401 return ret;
1402 }
1403
1404 int bch_flash_dev_create(struct cache_set *c, uint64_t size)
1405 {
1406 struct uuid_entry *u;
1407
1408 if (test_bit(CACHE_SET_STOPPING, &c->flags))
1409 return -EINTR;
1410
1411 if (!test_bit(CACHE_SET_RUNNING, &c->flags))
1412 return -EPERM;
1413
1414 u = uuid_find_empty(c);
1415 if (!u) {
1416 pr_err("Can't create volume, no room for UUID");
1417 return -EINVAL;
1418 }
1419
1420 get_random_bytes(u->uuid, 16);
1421 memset(u->label, 0, 32);
1422 u->first_reg = u->last_reg = cpu_to_le32(get_seconds());
1423
1424 SET_UUID_FLASH_ONLY(u, 1);
1425 u->sectors = size >> 9;
1426
1427 bch_uuid_write(c);
1428
1429 return flash_dev_run(c, u);
1430 }
1431
1432 bool bch_cached_dev_error(struct cached_dev *dc)
1433 {
1434 if (!dc || test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags))
1435 return false;
1436
1437 dc->io_disable = true;
1438 /* make others know io_disable is true earlier */
1439 smp_mb();
1440
1441 pr_err("stop %s: too many IO errors on backing device %s\n",
1442 dc->disk.disk->disk_name, dc->backing_dev_name);
1443
1444 bcache_device_stop(&dc->disk);
1445 return true;
1446 }
1447
1448 /* Cache set */
1449
1450 __printf(2, 3)
1451 bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...)
1452 {
1453 va_list args;
1454
1455 if (c->on_error != ON_ERROR_PANIC &&
1456 test_bit(CACHE_SET_STOPPING, &c->flags))
1457 return false;
1458
1459 if (test_and_set_bit(CACHE_SET_IO_DISABLE, &c->flags))
1460 pr_info("CACHE_SET_IO_DISABLE already set");
1461
1462 /* XXX: we can be called from atomic context
1463 acquire_console_sem();
1464 */
1465
1466 printk(KERN_ERR "bcache: error on %pU: ", c->sb.set_uuid);
1467
1468 va_start(args, fmt);
1469 vprintk(fmt, args);
1470 va_end(args);
1471
1472 printk(", disabling caching\n");
1473
1474 if (c->on_error == ON_ERROR_PANIC)
1475 panic("panic forced after error\n");
1476
1477 bch_cache_set_unregister(c);
1478 return true;
1479 }
1480
1481 void bch_cache_set_release(struct kobject *kobj)
1482 {
1483 struct cache_set *c = container_of(kobj, struct cache_set, kobj);
1484 kfree(c);
1485 module_put(THIS_MODULE);
1486 }
1487
1488 static void cache_set_free(struct closure *cl)
1489 {
1490 struct cache_set *c = container_of(cl, struct cache_set, cl);
1491 struct cache *ca;
1492 unsigned i;
1493
1494 debugfs_remove(c->debug);
1495
1496 bch_open_buckets_free(c);
1497 bch_btree_cache_free(c);
1498 bch_journal_free(c);
1499
1500 mutex_lock(&bch_register_lock);
1501 for_each_cache(ca, c, i)
1502 if (ca) {
1503 ca->set = NULL;
1504 c->cache[ca->sb.nr_this_dev] = NULL;
1505 kobject_put(&ca->kobj);
1506 }
1507
1508 bch_bset_sort_state_free(&c->sort);
1509 free_pages((unsigned long) c->uuids, ilog2(bucket_pages(c)));
1510
1511 if (c->moving_gc_wq)
1512 destroy_workqueue(c->moving_gc_wq);
1513 if (c->bio_split)
1514 bioset_free(c->bio_split);
1515 if (c->fill_iter)
1516 mempool_destroy(c->fill_iter);
1517 if (c->bio_meta)
1518 mempool_destroy(c->bio_meta);
1519 if (c->search)
1520 mempool_destroy(c->search);
1521 kfree(c->devices);
1522
1523 list_del(&c->list);
1524 mutex_unlock(&bch_register_lock);
1525
1526 pr_info("Cache set %pU unregistered", c->sb.set_uuid);
1527 wake_up(&unregister_wait);
1528
1529 closure_debug_destroy(&c->cl);
1530 kobject_put(&c->kobj);
1531 }
1532
1533 static void cache_set_flush(struct closure *cl)
1534 {
1535 struct cache_set *c = container_of(cl, struct cache_set, caching);
1536 struct cache *ca;
1537 struct btree *b;
1538 unsigned i;
1539
1540 bch_cache_accounting_destroy(&c->accounting);
1541
1542 kobject_put(&c->internal);
1543 kobject_del(&c->kobj);
1544
1545 if (!IS_ERR_OR_NULL(c->gc_thread))
1546 kthread_stop(c->gc_thread);
1547
1548 if (!IS_ERR_OR_NULL(c->root))
1549 list_add(&c->root->list, &c->btree_cache);
1550
1551 /* Should skip this if we're unregistering because of an error */
1552 list_for_each_entry(b, &c->btree_cache, list) {
1553 mutex_lock(&b->write_lock);
1554 if (btree_node_dirty(b))
1555 __bch_btree_node_write(b, NULL);
1556 mutex_unlock(&b->write_lock);
1557 }
1558
1559 for_each_cache(ca, c, i)
1560 if (ca->alloc_thread)
1561 kthread_stop(ca->alloc_thread);
1562
1563 if (c->journal.cur) {
1564 cancel_delayed_work_sync(&c->journal.work);
1565 /* flush last journal entry if needed */
1566 c->journal.work.work.func(&c->journal.work.work);
1567 }
1568
1569 closure_return(cl);
1570 }
1571
1572 /*
1573 * This function is only called when CACHE_SET_IO_DISABLE is set, which means
1574 * cache set is unregistering due to too many I/O errors. In this condition,
1575 * the bcache device might be stopped, it depends on stop_when_cache_set_failed
1576 * value and whether the broken cache has dirty data:
1577 *
1578 * dc->stop_when_cache_set_failed dc->has_dirty stop bcache device
1579 * BCH_CACHED_STOP_AUTO 0 NO
1580 * BCH_CACHED_STOP_AUTO 1 YES
1581 * BCH_CACHED_DEV_STOP_ALWAYS 0 YES
1582 * BCH_CACHED_DEV_STOP_ALWAYS 1 YES
1583 *
1584 * The expected behavior is, if stop_when_cache_set_failed is configured to
1585 * "auto" via sysfs interface, the bcache device will not be stopped if the
1586 * backing device is clean on the broken cache device.
1587 */
1588 static void conditional_stop_bcache_device(struct cache_set *c,
1589 struct bcache_device *d,
1590 struct cached_dev *dc)
1591 {
1592 if (dc->stop_when_cache_set_failed == BCH_CACHED_DEV_STOP_ALWAYS) {
1593 pr_warn("stop_when_cache_set_failed of %s is \"always\", stop it for failed cache set %pU.",
1594 d->disk->disk_name, c->sb.set_uuid);
1595 bcache_device_stop(d);
1596 } else if (atomic_read(&dc->has_dirty)) {
1597 /*
1598 * dc->stop_when_cache_set_failed == BCH_CACHED_STOP_AUTO
1599 * and dc->has_dirty == 1
1600 */
1601 pr_warn("stop_when_cache_set_failed of %s is \"auto\" and cache is dirty, stop it to avoid potential data corruption.",
1602 d->disk->disk_name);
1603 /*
1604 * There might be a small time gap that cache set is
1605 * released but bcache device is not. Inside this time
1606 * gap, regular I/O requests will directly go into
1607 * backing device as no cache set attached to. This
1608 * behavior may also introduce potential inconsistence
1609 * data in writeback mode while cache is dirty.
1610 * Therefore before calling bcache_device_stop() due
1611 * to a broken cache device, dc->io_disable should be
1612 * explicitly set to true.
1613 */
1614 dc->io_disable = true;
1615 /* make others know io_disable is true earlier */
1616 smp_mb();
1617 bcache_device_stop(d);
1618 } else {
1619 /*
1620 * dc->stop_when_cache_set_failed == BCH_CACHED_STOP_AUTO
1621 * and dc->has_dirty == 0
1622 */
1623 pr_warn("stop_when_cache_set_failed of %s is \"auto\" and cache is clean, keep it alive.",
1624 d->disk->disk_name);
1625 }
1626 }
1627
1628 static void __cache_set_unregister(struct closure *cl)
1629 {
1630 struct cache_set *c = container_of(cl, struct cache_set, caching);
1631 struct cached_dev *dc;
1632 struct bcache_device *d;
1633 size_t i;
1634
1635 mutex_lock(&bch_register_lock);
1636
1637 for (i = 0; i < c->nr_uuids; i++) {
1638 d = c->devices[i];
1639 if (!d)
1640 continue;
1641
1642 if (!UUID_FLASH_ONLY(&c->uuids[i]) &&
1643 test_bit(CACHE_SET_UNREGISTERING, &c->flags)) {
1644 dc = container_of(d, struct cached_dev, disk);
1645 bch_cached_dev_detach(dc);
1646 if (test_bit(CACHE_SET_IO_DISABLE, &c->flags))
1647 conditional_stop_bcache_device(c, d, dc);
1648 } else {
1649 bcache_device_stop(d);
1650 }
1651 }
1652
1653 mutex_unlock(&bch_register_lock);
1654
1655 continue_at(cl, cache_set_flush, system_wq);
1656 }
1657
1658 void bch_cache_set_stop(struct cache_set *c)
1659 {
1660 if (!test_and_set_bit(CACHE_SET_STOPPING, &c->flags))
1661 closure_queue(&c->caching);
1662 }
1663
1664 void bch_cache_set_unregister(struct cache_set *c)
1665 {
1666 set_bit(CACHE_SET_UNREGISTERING, &c->flags);
1667 bch_cache_set_stop(c);
1668 }
1669
1670 #define alloc_bucket_pages(gfp, c) \
1671 ((void *) __get_free_pages(__GFP_ZERO|gfp, ilog2(bucket_pages(c))))
1672
1673 struct cache_set *bch_cache_set_alloc(struct cache_sb *sb)
1674 {
1675 int iter_size;
1676 struct cache_set *c = kzalloc(sizeof(struct cache_set), GFP_KERNEL);
1677 if (!c)
1678 return NULL;
1679
1680 __module_get(THIS_MODULE);
1681 closure_init(&c->cl, NULL);
1682 set_closure_fn(&c->cl, cache_set_free, system_wq);
1683
1684 closure_init(&c->caching, &c->cl);
1685 set_closure_fn(&c->caching, __cache_set_unregister, system_wq);
1686
1687 /* Maybe create continue_at_noreturn() and use it here? */
1688 closure_set_stopped(&c->cl);
1689 closure_put(&c->cl);
1690
1691 kobject_init(&c->kobj, &bch_cache_set_ktype);
1692 kobject_init(&c->internal, &bch_cache_set_internal_ktype);
1693
1694 bch_cache_accounting_init(&c->accounting, &c->cl);
1695
1696 memcpy(c->sb.set_uuid, sb->set_uuid, 16);
1697 c->sb.block_size = sb->block_size;
1698 c->sb.bucket_size = sb->bucket_size;
1699 c->sb.nr_in_set = sb->nr_in_set;
1700 c->sb.last_mount = sb->last_mount;
1701 c->bucket_bits = ilog2(sb->bucket_size);
1702 c->block_bits = ilog2(sb->block_size);
1703 c->nr_uuids = bucket_bytes(c) / sizeof(struct uuid_entry);
1704
1705 c->btree_pages = bucket_pages(c);
1706 if (c->btree_pages > BTREE_MAX_PAGES)
1707 c->btree_pages = max_t(int, c->btree_pages / 4,
1708 BTREE_MAX_PAGES);
1709
1710 sema_init(&c->sb_write_mutex, 1);
1711 mutex_init(&c->bucket_lock);
1712 init_waitqueue_head(&c->btree_cache_wait);
1713 init_waitqueue_head(&c->bucket_wait);
1714 init_waitqueue_head(&c->gc_wait);
1715 sema_init(&c->uuid_write_mutex, 1);
1716
1717 spin_lock_init(&c->btree_gc_time.lock);
1718 spin_lock_init(&c->btree_split_time.lock);
1719 spin_lock_init(&c->btree_read_time.lock);
1720
1721 bch_moving_init_cache_set(c);
1722
1723 INIT_LIST_HEAD(&c->list);
1724 INIT_LIST_HEAD(&c->cached_devs);
1725 INIT_LIST_HEAD(&c->btree_cache);
1726 INIT_LIST_HEAD(&c->btree_cache_freeable);
1727 INIT_LIST_HEAD(&c->btree_cache_freed);
1728 INIT_LIST_HEAD(&c->data_buckets);
1729
1730 c->search = mempool_create_slab_pool(32, bch_search_cache);
1731 if (!c->search)
1732 goto err;
1733
1734 iter_size = (sb->bucket_size / sb->block_size + 1) *
1735 sizeof(struct btree_iter_set);
1736
1737 if (!(c->devices = kzalloc(c->nr_uuids * sizeof(void *), GFP_KERNEL)) ||
1738 !(c->bio_meta = mempool_create_kmalloc_pool(2,
1739 sizeof(struct bbio) + sizeof(struct bio_vec) *
1740 bucket_pages(c))) ||
1741 !(c->fill_iter = mempool_create_kmalloc_pool(1, iter_size)) ||
1742 !(c->bio_split = bioset_create(4, offsetof(struct bbio, bio),
1743 BIOSET_NEED_BVECS |
1744 BIOSET_NEED_RESCUER)) ||
1745 !(c->uuids = alloc_bucket_pages(GFP_KERNEL, c)) ||
1746 !(c->moving_gc_wq = alloc_workqueue("bcache_gc",
1747 WQ_MEM_RECLAIM, 0)) ||
1748 bch_journal_alloc(c) ||
1749 bch_btree_cache_alloc(c) ||
1750 bch_open_buckets_alloc(c) ||
1751 bch_bset_sort_state_init(&c->sort, ilog2(c->btree_pages)))
1752 goto err;
1753
1754 c->congested_read_threshold_us = 2000;
1755 c->congested_write_threshold_us = 20000;
1756 c->error_limit = 8 << IO_ERROR_SHIFT;
1757 WARN_ON(test_and_clear_bit(CACHE_SET_IO_DISABLE, &c->flags));
1758
1759 return c;
1760 err:
1761 bch_cache_set_unregister(c);
1762 return NULL;
1763 }
1764
1765 static int run_cache_set(struct cache_set *c)
1766 {
1767 const char *err = "cannot allocate memory";
1768 struct cached_dev *dc, *t;
1769 struct cache *ca;
1770 struct closure cl;
1771 unsigned i;
1772 LIST_HEAD(journal);
1773 struct journal_replay *l;
1774
1775 closure_init_stack(&cl);
1776
1777 for_each_cache(ca, c, i)
1778 c->nbuckets += ca->sb.nbuckets;
1779 set_gc_sectors(c);
1780
1781 if (CACHE_SYNC(&c->sb)) {
1782 struct bkey *k;
1783 struct jset *j;
1784
1785 err = "cannot allocate memory for journal";
1786 if (bch_journal_read(c, &journal))
1787 goto err;
1788
1789 pr_debug("btree_journal_read() done");
1790
1791 err = "no journal entries found";
1792 if (list_empty(&journal))
1793 goto err;
1794
1795 j = &list_entry(journal.prev, struct journal_replay, list)->j;
1796
1797 err = "IO error reading priorities";
1798 for_each_cache(ca, c, i)
1799 prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev]);
1800
1801 /*
1802 * If prio_read() fails it'll call cache_set_error and we'll
1803 * tear everything down right away, but if we perhaps checked
1804 * sooner we could avoid journal replay.
1805 */
1806
1807 k = &j->btree_root;
1808
1809 err = "bad btree root";
1810 if (__bch_btree_ptr_invalid(c, k))
1811 goto err;
1812
1813 err = "error reading btree root";
1814 c->root = bch_btree_node_get(c, NULL, k, j->btree_level, true, NULL);
1815 if (IS_ERR_OR_NULL(c->root))
1816 goto err;
1817
1818 list_del_init(&c->root->list);
1819 rw_unlock(true, c->root);
1820
1821 err = uuid_read(c, j, &cl);
1822 if (err)
1823 goto err;
1824
1825 err = "error in recovery";
1826 if (bch_btree_check(c))
1827 goto err;
1828
1829 bch_journal_mark(c, &journal);
1830 bch_initial_gc_finish(c);
1831 pr_debug("btree_check() done");
1832
1833 /*
1834 * bcache_journal_next() can't happen sooner, or
1835 * btree_gc_finish() will give spurious errors about last_gc >
1836 * gc_gen - this is a hack but oh well.
1837 */
1838 bch_journal_next(&c->journal);
1839
1840 err = "error starting allocator thread";
1841 for_each_cache(ca, c, i)
1842 if (bch_cache_allocator_start(ca))
1843 goto err;
1844
1845 /*
1846 * First place it's safe to allocate: btree_check() and
1847 * btree_gc_finish() have to run before we have buckets to
1848 * allocate, and bch_bucket_alloc_set() might cause a journal
1849 * entry to be written so bcache_journal_next() has to be called
1850 * first.
1851 *
1852 * If the uuids were in the old format we have to rewrite them
1853 * before the next journal entry is written:
1854 */
1855 if (j->version < BCACHE_JSET_VERSION_UUID)
1856 __uuid_write(c);
1857
1858 err = "bcache: replay journal failed";
1859 if (bch_journal_replay(c, &journal))
1860 goto err;
1861 } else {
1862 pr_notice("invalidating existing data");
1863
1864 for_each_cache(ca, c, i) {
1865 unsigned j;
1866
1867 ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7,
1868 2, SB_JOURNAL_BUCKETS);
1869
1870 for (j = 0; j < ca->sb.keys; j++)
1871 ca->sb.d[j] = ca->sb.first_bucket + j;
1872 }
1873
1874 bch_initial_gc_finish(c);
1875
1876 err = "error starting allocator thread";
1877 for_each_cache(ca, c, i)
1878 if (bch_cache_allocator_start(ca))
1879 goto err;
1880
1881 mutex_lock(&c->bucket_lock);
1882 for_each_cache(ca, c, i)
1883 bch_prio_write(ca, true);
1884 mutex_unlock(&c->bucket_lock);
1885
1886 err = "cannot allocate new UUID bucket";
1887 if (__uuid_write(c))
1888 goto err;
1889
1890 err = "cannot allocate new btree root";
1891 c->root = __bch_btree_node_alloc(c, NULL, 0, true, NULL);
1892 if (IS_ERR_OR_NULL(c->root))
1893 goto err;
1894
1895 mutex_lock(&c->root->write_lock);
1896 bkey_copy_key(&c->root->key, &MAX_KEY);
1897 bch_btree_node_write(c->root, &cl);
1898 mutex_unlock(&c->root->write_lock);
1899
1900 bch_btree_set_root(c->root);
1901 rw_unlock(true, c->root);
1902
1903 /*
1904 * We don't want to write the first journal entry until
1905 * everything is set up - fortunately journal entries won't be
1906 * written until the SET_CACHE_SYNC() here:
1907 */
1908 SET_CACHE_SYNC(&c->sb, true);
1909
1910 bch_journal_next(&c->journal);
1911 bch_journal_meta(c, &cl);
1912 }
1913
1914 err = "error starting gc thread";
1915 if (bch_gc_thread_start(c))
1916 goto err;
1917
1918 closure_sync(&cl);
1919 c->sb.last_mount = get_seconds();
1920 bcache_write_super(c);
1921
1922 list_for_each_entry_safe(dc, t, &uncached_devices, list)
1923 bch_cached_dev_attach(dc, c, NULL);
1924
1925 flash_devs_run(c);
1926
1927 set_bit(CACHE_SET_RUNNING, &c->flags);
1928 return 0;
1929 err:
1930 while (!list_empty(&journal)) {
1931 l = list_first_entry(&journal, struct journal_replay, list);
1932 list_del(&l->list);
1933 kfree(l);
1934 }
1935
1936 closure_sync(&cl);
1937 /* XXX: test this, it's broken */
1938 bch_cache_set_error(c, "%s", err);
1939
1940 return -EIO;
1941 }
1942
1943 static bool can_attach_cache(struct cache *ca, struct cache_set *c)
1944 {
1945 return ca->sb.block_size == c->sb.block_size &&
1946 ca->sb.bucket_size == c->sb.bucket_size &&
1947 ca->sb.nr_in_set == c->sb.nr_in_set;
1948 }
1949
1950 static const char *register_cache_set(struct cache *ca)
1951 {
1952 char buf[12];
1953 const char *err = "cannot allocate memory";
1954 struct cache_set *c;
1955
1956 list_for_each_entry(c, &bch_cache_sets, list)
1957 if (!memcmp(c->sb.set_uuid, ca->sb.set_uuid, 16)) {
1958 if (c->cache[ca->sb.nr_this_dev])
1959 return "duplicate cache set member";
1960
1961 if (!can_attach_cache(ca, c))
1962 return "cache sb does not match set";
1963
1964 if (!CACHE_SYNC(&ca->sb))
1965 SET_CACHE_SYNC(&c->sb, false);
1966
1967 goto found;
1968 }
1969
1970 c = bch_cache_set_alloc(&ca->sb);
1971 if (!c)
1972 return err;
1973
1974 err = "error creating kobject";
1975 if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->sb.set_uuid) ||
1976 kobject_add(&c->internal, &c->kobj, "internal"))
1977 goto err;
1978
1979 if (bch_cache_accounting_add_kobjs(&c->accounting, &c->kobj))
1980 goto err;
1981
1982 bch_debug_init_cache_set(c);
1983
1984 list_add(&c->list, &bch_cache_sets);
1985 found:
1986 sprintf(buf, "cache%i", ca->sb.nr_this_dev);
1987 if (sysfs_create_link(&ca->kobj, &c->kobj, "set") ||
1988 sysfs_create_link(&c->kobj, &ca->kobj, buf))
1989 goto err;
1990
1991 if (ca->sb.seq > c->sb.seq) {
1992 c->sb.version = ca->sb.version;
1993 memcpy(c->sb.set_uuid, ca->sb.set_uuid, 16);
1994 c->sb.flags = ca->sb.flags;
1995 c->sb.seq = ca->sb.seq;
1996 pr_debug("set version = %llu", c->sb.version);
1997 }
1998
1999 kobject_get(&ca->kobj);
2000 ca->set = c;
2001 ca->set->cache[ca->sb.nr_this_dev] = ca;
2002 c->cache_by_alloc[c->caches_loaded++] = ca;
2003
2004 if (c->caches_loaded == c->sb.nr_in_set) {
2005 err = "failed to run cache set";
2006 if (run_cache_set(c) < 0)
2007 goto err;
2008 }
2009
2010 return NULL;
2011 err:
2012 bch_cache_set_unregister(c);
2013 return err;
2014 }
2015
2016 /* Cache device */
2017
2018 void bch_cache_release(struct kobject *kobj)
2019 {
2020 struct cache *ca = container_of(kobj, struct cache, kobj);
2021 unsigned i;
2022
2023 if (ca->set) {
2024 BUG_ON(ca->set->cache[ca->sb.nr_this_dev] != ca);
2025 ca->set->cache[ca->sb.nr_this_dev] = NULL;
2026 }
2027
2028 free_pages((unsigned long) ca->disk_buckets, ilog2(bucket_pages(ca)));
2029 kfree(ca->prio_buckets);
2030 vfree(ca->buckets);
2031
2032 free_heap(&ca->heap);
2033 free_fifo(&ca->free_inc);
2034
2035 for (i = 0; i < RESERVE_NR; i++)
2036 free_fifo(&ca->free[i]);
2037
2038 if (ca->sb_bio.bi_inline_vecs[0].bv_page)
2039 put_page(ca->sb_bio.bi_io_vec[0].bv_page);
2040
2041 if (!IS_ERR_OR_NULL(ca->bdev))
2042 blkdev_put(ca->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2043
2044 kfree(ca);
2045 module_put(THIS_MODULE);
2046 }
2047
2048 static int cache_alloc(struct cache *ca)
2049 {
2050 size_t free;
2051 size_t btree_buckets;
2052 struct bucket *b;
2053
2054 __module_get(THIS_MODULE);
2055 kobject_init(&ca->kobj, &bch_cache_ktype);
2056
2057 bio_init(&ca->journal.bio, ca->journal.bio.bi_inline_vecs, 8);
2058
2059 /*
2060 * when ca->sb.njournal_buckets is not zero, journal exists,
2061 * and in bch_journal_replay(), tree node may split,
2062 * so bucket of RESERVE_BTREE type is needed,
2063 * the worst situation is all journal buckets are valid journal,
2064 * and all the keys need to replay,
2065 * so the number of RESERVE_BTREE type buckets should be as much
2066 * as journal buckets
2067 */
2068 btree_buckets = ca->sb.njournal_buckets ?: 8;
2069 free = roundup_pow_of_two(ca->sb.nbuckets) >> 10;
2070
2071 if (!init_fifo(&ca->free[RESERVE_BTREE], btree_buckets, GFP_KERNEL) ||
2072 !init_fifo_exact(&ca->free[RESERVE_PRIO], prio_buckets(ca), GFP_KERNEL) ||
2073 !init_fifo(&ca->free[RESERVE_MOVINGGC], free, GFP_KERNEL) ||
2074 !init_fifo(&ca->free[RESERVE_NONE], free, GFP_KERNEL) ||
2075 !init_fifo(&ca->free_inc, free << 2, GFP_KERNEL) ||
2076 !init_heap(&ca->heap, free << 3, GFP_KERNEL) ||
2077 !(ca->buckets = vzalloc(sizeof(struct bucket) *
2078 ca->sb.nbuckets)) ||
2079 !(ca->prio_buckets = kzalloc(sizeof(uint64_t) * prio_buckets(ca) *
2080 2, GFP_KERNEL)) ||
2081 !(ca->disk_buckets = alloc_bucket_pages(GFP_KERNEL, ca)))
2082 return -ENOMEM;
2083
2084 ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca);
2085
2086 for_each_bucket(b, ca)
2087 atomic_set(&b->pin, 0);
2088
2089 return 0;
2090 }
2091
2092 static int register_cache(struct cache_sb *sb, struct page *sb_page,
2093 struct block_device *bdev, struct cache *ca)
2094 {
2095 const char *err = NULL; /* must be set for any error case */
2096 int ret = 0;
2097
2098 bdevname(bdev, ca->cache_dev_name);
2099 memcpy(&ca->sb, sb, sizeof(struct cache_sb));
2100 ca->bdev = bdev;
2101 ca->bdev->bd_holder = ca;
2102
2103 bio_init(&ca->sb_bio, ca->sb_bio.bi_inline_vecs, 1);
2104 ca->sb_bio.bi_io_vec[0].bv_page = sb_page;
2105 get_page(sb_page);
2106
2107 if (blk_queue_discard(bdev_get_queue(bdev)))
2108 ca->discard = CACHE_DISCARD(&ca->sb);
2109
2110 ret = cache_alloc(ca);
2111 if (ret != 0) {
2112 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2113 if (ret == -ENOMEM)
2114 err = "cache_alloc(): -ENOMEM";
2115 else
2116 err = "cache_alloc(): unknown error";
2117 goto err;
2118 }
2119
2120 if (kobject_add(&ca->kobj, &part_to_dev(bdev->bd_part)->kobj, "bcache")) {
2121 err = "error calling kobject_add";
2122 ret = -ENOMEM;
2123 goto out;
2124 }
2125
2126 mutex_lock(&bch_register_lock);
2127 err = register_cache_set(ca);
2128 mutex_unlock(&bch_register_lock);
2129
2130 if (err) {
2131 ret = -ENODEV;
2132 goto out;
2133 }
2134
2135 pr_info("registered cache device %s", ca->cache_dev_name);
2136
2137 out:
2138 kobject_put(&ca->kobj);
2139
2140 err:
2141 if (err)
2142 pr_notice("error %s: %s", ca->cache_dev_name, err);
2143
2144 return ret;
2145 }
2146
2147 /* Global interfaces/init */
2148
2149 static ssize_t register_bcache(struct kobject *, struct kobj_attribute *,
2150 const char *, size_t);
2151
2152 kobj_attribute_write(register, register_bcache);
2153 kobj_attribute_write(register_quiet, register_bcache);
2154
2155 static bool bch_is_open_backing(struct block_device *bdev) {
2156 struct cache_set *c, *tc;
2157 struct cached_dev *dc, *t;
2158
2159 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
2160 list_for_each_entry_safe(dc, t, &c->cached_devs, list)
2161 if (dc->bdev == bdev)
2162 return true;
2163 list_for_each_entry_safe(dc, t, &uncached_devices, list)
2164 if (dc->bdev == bdev)
2165 return true;
2166 return false;
2167 }
2168
2169 static struct cached_dev *bch_find_cached_dev(struct block_device *bdev) {
2170 struct cache_set *c, *tc;
2171 struct cached_dev *dc, *t;
2172
2173 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
2174 list_for_each_entry_safe(dc, t, &c->cached_devs, list)
2175 if (dc->bdev == bdev)
2176 return dc;
2177 list_for_each_entry_safe(dc, t, &uncached_devices, list)
2178 if (dc->bdev == bdev)
2179 return dc;
2180
2181 return NULL;
2182 }
2183
2184 static bool bch_is_open_cache(struct block_device *bdev) {
2185 struct cache_set *c, *tc;
2186 struct cache *ca;
2187 unsigned i;
2188
2189 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
2190 for_each_cache(ca, c, i)
2191 if (ca->bdev == bdev)
2192 return true;
2193 return false;
2194 }
2195
2196 static bool bch_is_open(struct block_device *bdev) {
2197 return bch_is_open_cache(bdev) || bch_is_open_backing(bdev);
2198 }
2199
2200 static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr,
2201 const char *buffer, size_t size)
2202 {
2203 ssize_t ret = size;
2204 const char *err = "cannot allocate memory";
2205 char *path = NULL;
2206 struct cache_sb *sb = NULL;
2207 struct block_device *bdev = NULL;
2208 struct page *sb_page = NULL;
2209 struct cached_dev *dc = NULL;
2210
2211 if (!try_module_get(THIS_MODULE))
2212 return -EBUSY;
2213
2214 if (!(path = kstrndup(buffer, size, GFP_KERNEL)) ||
2215 !(sb = kmalloc(sizeof(struct cache_sb), GFP_KERNEL)))
2216 goto err;
2217
2218 err = "failed to open device";
2219 bdev = blkdev_get_by_path(strim(path),
2220 FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2221 sb);
2222 if (IS_ERR(bdev)) {
2223 if (bdev == ERR_PTR(-EBUSY)) {
2224 bdev = lookup_bdev(strim(path), 0);
2225 mutex_lock(&bch_register_lock);
2226 if (!IS_ERR(bdev) && bch_is_open(bdev)) {
2227 err = "device already registered";
2228 /* emit CHANGE event for backing devices to export
2229 * CACHED_{UUID/LABEL} values to udev */
2230 if (bch_is_open_backing(bdev)) {
2231 dc = bch_find_cached_dev(bdev);
2232 if (dc) {
2233 bch_cached_dev_emit_change(dc);
2234 err = "device already registered (emitting change event)";
2235 }
2236 }
2237 } else {
2238 err = "device busy";
2239 }
2240 mutex_unlock(&bch_register_lock);
2241 if (!IS_ERR(bdev))
2242 bdput(bdev);
2243 if (attr == &ksysfs_register_quiet)
2244 goto out;
2245 }
2246 goto err;
2247 }
2248
2249 err = "failed to set blocksize";
2250 if (set_blocksize(bdev, 4096))
2251 goto err_close;
2252
2253 err = read_super(sb, bdev, &sb_page);
2254 if (err)
2255 goto err_close;
2256
2257 err = "failed to register device";
2258 if (SB_IS_BDEV(sb)) {
2259 struct cached_dev *dc = kzalloc(sizeof(*dc), GFP_KERNEL);
2260 if (!dc)
2261 goto err_close;
2262
2263 mutex_lock(&bch_register_lock);
2264 register_bdev(sb, sb_page, bdev, dc);
2265 mutex_unlock(&bch_register_lock);
2266 } else {
2267 struct cache *ca = kzalloc(sizeof(*ca), GFP_KERNEL);
2268 if (!ca)
2269 goto err_close;
2270
2271 if (register_cache(sb, sb_page, bdev, ca) != 0)
2272 goto err;
2273 }
2274 out:
2275 if (sb_page)
2276 put_page(sb_page);
2277 kfree(sb);
2278 kfree(path);
2279 module_put(THIS_MODULE);
2280 return ret;
2281
2282 err_close:
2283 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2284 err:
2285 pr_info("error %s: %s", path, err);
2286 ret = -EINVAL;
2287 goto out;
2288 }
2289
2290 static int bcache_reboot(struct notifier_block *n, unsigned long code, void *x)
2291 {
2292 if (code == SYS_DOWN ||
2293 code == SYS_HALT ||
2294 code == SYS_POWER_OFF) {
2295 DEFINE_WAIT(wait);
2296 unsigned long start = jiffies;
2297 bool stopped = false;
2298
2299 struct cache_set *c, *tc;
2300 struct cached_dev *dc, *tdc;
2301
2302 mutex_lock(&bch_register_lock);
2303
2304 if (list_empty(&bch_cache_sets) &&
2305 list_empty(&uncached_devices))
2306 goto out;
2307
2308 pr_info("Stopping all devices:");
2309
2310 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
2311 bch_cache_set_stop(c);
2312
2313 list_for_each_entry_safe(dc, tdc, &uncached_devices, list)
2314 bcache_device_stop(&dc->disk);
2315
2316 mutex_unlock(&bch_register_lock);
2317
2318 /*
2319 * Give an early chance for other kthreads and
2320 * kworkers to stop themselves
2321 */
2322 schedule();
2323
2324 /* What's a condition variable? */
2325 while (1) {
2326 long timeout = start + 10 * HZ - jiffies;
2327
2328 mutex_lock(&bch_register_lock);
2329 stopped = list_empty(&bch_cache_sets) &&
2330 list_empty(&uncached_devices);
2331
2332 if (timeout < 0 || stopped)
2333 break;
2334
2335 prepare_to_wait(&unregister_wait, &wait,
2336 TASK_UNINTERRUPTIBLE);
2337
2338 mutex_unlock(&bch_register_lock);
2339 schedule_timeout(timeout);
2340 }
2341
2342 finish_wait(&unregister_wait, &wait);
2343
2344 if (stopped)
2345 pr_info("All devices stopped");
2346 else
2347 pr_notice("Timeout waiting for devices to be closed");
2348 out:
2349 mutex_unlock(&bch_register_lock);
2350 }
2351
2352 return NOTIFY_DONE;
2353 }
2354
2355 static struct notifier_block reboot = {
2356 .notifier_call = bcache_reboot,
2357 .priority = INT_MAX, /* before any real devices */
2358 };
2359
2360 static void bcache_exit(void)
2361 {
2362 bch_debug_exit();
2363 bch_request_exit();
2364 if (bcache_kobj)
2365 kobject_put(bcache_kobj);
2366 if (bcache_wq)
2367 destroy_workqueue(bcache_wq);
2368 if (bcache_major)
2369 unregister_blkdev(bcache_major, "bcache");
2370 unregister_reboot_notifier(&reboot);
2371 mutex_destroy(&bch_register_lock);
2372 }
2373
2374 static int __init bcache_init(void)
2375 {
2376 static const struct attribute *files[] = {
2377 &ksysfs_register.attr,
2378 &ksysfs_register_quiet.attr,
2379 NULL
2380 };
2381
2382 mutex_init(&bch_register_lock);
2383 init_waitqueue_head(&unregister_wait);
2384 register_reboot_notifier(&reboot);
2385 closure_debug_init();
2386
2387 bcache_major = register_blkdev(0, "bcache");
2388 if (bcache_major < 0) {
2389 unregister_reboot_notifier(&reboot);
2390 mutex_destroy(&bch_register_lock);
2391 return bcache_major;
2392 }
2393
2394 if (!(bcache_wq = alloc_workqueue("bcache", WQ_MEM_RECLAIM, 0)) ||
2395 !(bcache_kobj = kobject_create_and_add("bcache", fs_kobj)) ||
2396 bch_request_init() ||
2397 bch_debug_init(bcache_kobj) ||
2398 sysfs_create_files(bcache_kobj, files))
2399 goto err;
2400
2401 return 0;
2402 err:
2403 bcache_exit();
2404 return -ENOMEM;
2405 }
2406
2407 module_exit(bcache_exit);
2408 module_init(bcache_init);
ubuntu-bionic-kernel mirror