xref: /openbmc/linux/drivers/md/bcache/super.c (revision 2831231d4c3f999d2d062b23dfbc8b0faa4bc6e0)
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 /* Default is -1; we skip past it for struct cached_dev's cache mode */
41 const char * const bch_cache_modes[] = {
42 	"default",
43 	"writethrough",
44 	"writeback",
45 	"writearound",
46 	"none",
47 	NULL
48 };
49 
50 static struct kobject *bcache_kobj;
51 struct mutex bch_register_lock;
52 LIST_HEAD(bch_cache_sets);
53 static LIST_HEAD(uncached_devices);
54 
55 static int bcache_major;
56 static DEFINE_IDA(bcache_device_idx);
57 static wait_queue_head_t unregister_wait;
58 struct workqueue_struct *bcache_wq;
59 
60 #define BTREE_MAX_PAGES		(256 * 1024 / PAGE_SIZE)
61 /* limitation of partitions number on single bcache device */
62 #define BCACHE_MINORS		128
63 /* limitation of bcache devices number on single system */
64 #define BCACHE_DEVICE_IDX_MAX	((1U << MINORBITS)/BCACHE_MINORS)
65 
66 /* Superblock */
67 
68 static const char *read_super(struct cache_sb *sb, struct block_device *bdev,
69 			      struct page **res)
70 {
71 	const char *err;
72 	struct cache_sb *s;
73 	struct buffer_head *bh = __bread(bdev, 1, SB_SIZE);
74 	unsigned i;
75 
76 	if (!bh)
77 		return "IO error";
78 
79 	s = (struct cache_sb *) bh->b_data;
80 
81 	sb->offset		= le64_to_cpu(s->offset);
82 	sb->version		= le64_to_cpu(s->version);
83 
84 	memcpy(sb->magic,	s->magic, 16);
85 	memcpy(sb->uuid,	s->uuid, 16);
86 	memcpy(sb->set_uuid,	s->set_uuid, 16);
87 	memcpy(sb->label,	s->label, SB_LABEL_SIZE);
88 
89 	sb->flags		= le64_to_cpu(s->flags);
90 	sb->seq			= le64_to_cpu(s->seq);
91 	sb->last_mount		= le32_to_cpu(s->last_mount);
92 	sb->first_bucket	= le16_to_cpu(s->first_bucket);
93 	sb->keys		= le16_to_cpu(s->keys);
94 
95 	for (i = 0; i < SB_JOURNAL_BUCKETS; i++)
96 		sb->d[i] = le64_to_cpu(s->d[i]);
97 
98 	pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u",
99 		 sb->version, sb->flags, sb->seq, sb->keys);
100 
101 	err = "Not a bcache superblock";
102 	if (sb->offset != SB_SECTOR)
103 		goto err;
104 
105 	if (memcmp(sb->magic, bcache_magic, 16))
106 		goto err;
107 
108 	err = "Too many journal buckets";
109 	if (sb->keys > SB_JOURNAL_BUCKETS)
110 		goto err;
111 
112 	err = "Bad checksum";
113 	if (s->csum != csum_set(s))
114 		goto err;
115 
116 	err = "Bad UUID";
117 	if (bch_is_zero(sb->uuid, 16))
118 		goto err;
119 
120 	sb->block_size	= le16_to_cpu(s->block_size);
121 
122 	err = "Superblock block size smaller than device block size";
123 	if (sb->block_size << 9 < bdev_logical_block_size(bdev))
124 		goto err;
125 
126 	switch (sb->version) {
127 	case BCACHE_SB_VERSION_BDEV:
128 		sb->data_offset	= BDEV_DATA_START_DEFAULT;
129 		break;
130 	case BCACHE_SB_VERSION_BDEV_WITH_OFFSET:
131 		sb->data_offset	= le64_to_cpu(s->data_offset);
132 
133 		err = "Bad data offset";
134 		if (sb->data_offset < BDEV_DATA_START_DEFAULT)
135 			goto err;
136 
137 		break;
138 	case BCACHE_SB_VERSION_CDEV:
139 	case BCACHE_SB_VERSION_CDEV_WITH_UUID:
140 		sb->nbuckets	= le64_to_cpu(s->nbuckets);
141 		sb->bucket_size	= le16_to_cpu(s->bucket_size);
142 
143 		sb->nr_in_set	= le16_to_cpu(s->nr_in_set);
144 		sb->nr_this_dev	= le16_to_cpu(s->nr_this_dev);
145 
146 		err = "Too many buckets";
147 		if (sb->nbuckets > LONG_MAX)
148 			goto err;
149 
150 		err = "Not enough buckets";
151 		if (sb->nbuckets < 1 << 7)
152 			goto err;
153 
154 		err = "Bad block/bucket size";
155 		if (!is_power_of_2(sb->block_size) ||
156 		    sb->block_size > PAGE_SECTORS ||
157 		    !is_power_of_2(sb->bucket_size) ||
158 		    sb->bucket_size < PAGE_SECTORS)
159 			goto err;
160 
161 		err = "Invalid superblock: device too small";
162 		if (get_capacity(bdev->bd_disk) < sb->bucket_size * sb->nbuckets)
163 			goto err;
164 
165 		err = "Bad UUID";
166 		if (bch_is_zero(sb->set_uuid, 16))
167 			goto err;
168 
169 		err = "Bad cache device number in set";
170 		if (!sb->nr_in_set ||
171 		    sb->nr_in_set <= sb->nr_this_dev ||
172 		    sb->nr_in_set > MAX_CACHES_PER_SET)
173 			goto err;
174 
175 		err = "Journal buckets not sequential";
176 		for (i = 0; i < sb->keys; i++)
177 			if (sb->d[i] != sb->first_bucket + i)
178 				goto err;
179 
180 		err = "Too many journal buckets";
181 		if (sb->first_bucket + sb->keys > sb->nbuckets)
182 			goto err;
183 
184 		err = "Invalid superblock: first bucket comes before end of super";
185 		if (sb->first_bucket * sb->bucket_size < 16)
186 			goto err;
187 
188 		break;
189 	default:
190 		err = "Unsupported superblock version";
191 		goto err;
192 	}
193 
194 	sb->last_mount = get_seconds();
195 	err = NULL;
196 
197 	get_page(bh->b_page);
198 	*res = bh->b_page;
199 err:
200 	put_bh(bh);
201 	return err;
202 }
203 
204 static void write_bdev_super_endio(struct bio *bio)
205 {
206 	struct cached_dev *dc = bio->bi_private;
207 	/* XXX: error checking */
208 
209 	closure_put(&dc->sb_write);
210 }
211 
212 static void __write_super(struct cache_sb *sb, struct bio *bio)
213 {
214 	struct cache_sb *out = page_address(bio_first_page_all(bio));
215 	unsigned i;
216 
217 	bio->bi_iter.bi_sector	= SB_SECTOR;
218 	bio->bi_iter.bi_size	= SB_SIZE;
219 	bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
220 	bch_bio_map(bio, NULL);
221 
222 	out->offset		= cpu_to_le64(sb->offset);
223 	out->version		= cpu_to_le64(sb->version);
224 
225 	memcpy(out->uuid,	sb->uuid, 16);
226 	memcpy(out->set_uuid,	sb->set_uuid, 16);
227 	memcpy(out->label,	sb->label, SB_LABEL_SIZE);
228 
229 	out->flags		= cpu_to_le64(sb->flags);
230 	out->seq		= cpu_to_le64(sb->seq);
231 
232 	out->last_mount		= cpu_to_le32(sb->last_mount);
233 	out->first_bucket	= cpu_to_le16(sb->first_bucket);
234 	out->keys		= cpu_to_le16(sb->keys);
235 
236 	for (i = 0; i < sb->keys; i++)
237 		out->d[i] = cpu_to_le64(sb->d[i]);
238 
239 	out->csum = csum_set(out);
240 
241 	pr_debug("ver %llu, flags %llu, seq %llu",
242 		 sb->version, sb->flags, sb->seq);
243 
244 	submit_bio(bio);
245 }
246 
247 static void bch_write_bdev_super_unlock(struct closure *cl)
248 {
249 	struct cached_dev *dc = container_of(cl, struct cached_dev, sb_write);
250 
251 	up(&dc->sb_write_mutex);
252 }
253 
254 void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent)
255 {
256 	struct closure *cl = &dc->sb_write;
257 	struct bio *bio = &dc->sb_bio;
258 
259 	down(&dc->sb_write_mutex);
260 	closure_init(cl, parent);
261 
262 	bio_reset(bio);
263 	bio_set_dev(bio, dc->bdev);
264 	bio->bi_end_io	= write_bdev_super_endio;
265 	bio->bi_private = dc;
266 
267 	closure_get(cl);
268 	__write_super(&dc->sb, bio);
269 
270 	closure_return_with_destructor(cl, bch_write_bdev_super_unlock);
271 }
272 
273 static void write_super_endio(struct bio *bio)
274 {
275 	struct cache *ca = bio->bi_private;
276 
277 	bch_count_io_errors(ca, bio->bi_status, "writing superblock");
278 	closure_put(&ca->set->sb_write);
279 }
280 
281 static void bcache_write_super_unlock(struct closure *cl)
282 {
283 	struct cache_set *c = container_of(cl, struct cache_set, sb_write);
284 
285 	up(&c->sb_write_mutex);
286 }
287 
288 void bcache_write_super(struct cache_set *c)
289 {
290 	struct closure *cl = &c->sb_write;
291 	struct cache *ca;
292 	unsigned i;
293 
294 	down(&c->sb_write_mutex);
295 	closure_init(cl, &c->cl);
296 
297 	c->sb.seq++;
298 
299 	for_each_cache(ca, c, i) {
300 		struct bio *bio = &ca->sb_bio;
301 
302 		ca->sb.version		= BCACHE_SB_VERSION_CDEV_WITH_UUID;
303 		ca->sb.seq		= c->sb.seq;
304 		ca->sb.last_mount	= c->sb.last_mount;
305 
306 		SET_CACHE_SYNC(&ca->sb, CACHE_SYNC(&c->sb));
307 
308 		bio_reset(bio);
309 		bio_set_dev(bio, ca->bdev);
310 		bio->bi_end_io	= write_super_endio;
311 		bio->bi_private = ca;
312 
313 		closure_get(cl);
314 		__write_super(&ca->sb, bio);
315 	}
316 
317 	closure_return_with_destructor(cl, bcache_write_super_unlock);
318 }
319 
320 /* UUID io */
321 
322 static void uuid_endio(struct bio *bio)
323 {
324 	struct closure *cl = bio->bi_private;
325 	struct cache_set *c = container_of(cl, struct cache_set, uuid_write);
326 
327 	cache_set_err_on(bio->bi_status, c, "accessing uuids");
328 	bch_bbio_free(bio, c);
329 	closure_put(cl);
330 }
331 
332 static void uuid_io_unlock(struct closure *cl)
333 {
334 	struct cache_set *c = container_of(cl, struct cache_set, uuid_write);
335 
336 	up(&c->uuid_write_mutex);
337 }
338 
339 static void uuid_io(struct cache_set *c, int op, unsigned long op_flags,
340 		    struct bkey *k, struct closure *parent)
341 {
342 	struct closure *cl = &c->uuid_write;
343 	struct uuid_entry *u;
344 	unsigned i;
345 	char buf[80];
346 
347 	BUG_ON(!parent);
348 	down(&c->uuid_write_mutex);
349 	closure_init(cl, parent);
350 
351 	for (i = 0; i < KEY_PTRS(k); i++) {
352 		struct bio *bio = bch_bbio_alloc(c);
353 
354 		bio->bi_opf = REQ_SYNC | REQ_META | op_flags;
355 		bio->bi_iter.bi_size = KEY_SIZE(k) << 9;
356 
357 		bio->bi_end_io	= uuid_endio;
358 		bio->bi_private = cl;
359 		bio_set_op_attrs(bio, op, REQ_SYNC|REQ_META|op_flags);
360 		bch_bio_map(bio, c->uuids);
361 
362 		bch_submit_bbio(bio, c, k, i);
363 
364 		if (op != REQ_OP_WRITE)
365 			break;
366 	}
367 
368 	bch_extent_to_text(buf, sizeof(buf), k);
369 	pr_debug("%s UUIDs at %s", op == REQ_OP_WRITE ? "wrote" : "read", buf);
370 
371 	for (u = c->uuids; u < c->uuids + c->nr_uuids; u++)
372 		if (!bch_is_zero(u->uuid, 16))
373 			pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u",
374 				 u - c->uuids, u->uuid, u->label,
375 				 u->first_reg, u->last_reg, u->invalidated);
376 
377 	closure_return_with_destructor(cl, uuid_io_unlock);
378 }
379 
380 static char *uuid_read(struct cache_set *c, struct jset *j, struct closure *cl)
381 {
382 	struct bkey *k = &j->uuid_bucket;
383 
384 	if (__bch_btree_ptr_invalid(c, k))
385 		return "bad uuid pointer";
386 
387 	bkey_copy(&c->uuid_bucket, k);
388 	uuid_io(c, REQ_OP_READ, 0, k, cl);
389 
390 	if (j->version < BCACHE_JSET_VERSION_UUIDv1) {
391 		struct uuid_entry_v0	*u0 = (void *) c->uuids;
392 		struct uuid_entry	*u1 = (void *) c->uuids;
393 		int i;
394 
395 		closure_sync(cl);
396 
397 		/*
398 		 * Since the new uuid entry is bigger than the old, we have to
399 		 * convert starting at the highest memory address and work down
400 		 * in order to do it in place
401 		 */
402 
403 		for (i = c->nr_uuids - 1;
404 		     i >= 0;
405 		     --i) {
406 			memcpy(u1[i].uuid,	u0[i].uuid, 16);
407 			memcpy(u1[i].label,	u0[i].label, 32);
408 
409 			u1[i].first_reg		= u0[i].first_reg;
410 			u1[i].last_reg		= u0[i].last_reg;
411 			u1[i].invalidated	= u0[i].invalidated;
412 
413 			u1[i].flags	= 0;
414 			u1[i].sectors	= 0;
415 		}
416 	}
417 
418 	return NULL;
419 }
420 
421 static int __uuid_write(struct cache_set *c)
422 {
423 	BKEY_PADDED(key) k;
424 	struct closure cl;
425 	closure_init_stack(&cl);
426 
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 	bkey_copy(&c->uuid_bucket, &k.key);
437 	bkey_put(c, &k.key);
438 	return 0;
439 }
440 
441 int bch_uuid_write(struct cache_set *c)
442 {
443 	int ret = __uuid_write(c);
444 
445 	if (!ret)
446 		bch_journal_meta(c, NULL);
447 
448 	return ret;
449 }
450 
451 static struct uuid_entry *uuid_find(struct cache_set *c, const char *uuid)
452 {
453 	struct uuid_entry *u;
454 
455 	for (u = c->uuids;
456 	     u < c->uuids + c->nr_uuids; u++)
457 		if (!memcmp(u->uuid, uuid, 16))
458 			return u;
459 
460 	return NULL;
461 }
462 
463 static struct uuid_entry *uuid_find_empty(struct cache_set *c)
464 {
465 	static const char zero_uuid[16] = "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0";
466 	return uuid_find(c, zero_uuid);
467 }
468 
469 /*
470  * Bucket priorities/gens:
471  *
472  * For each bucket, we store on disk its
473    * 8 bit gen
474    * 16 bit priority
475  *
476  * See alloc.c for an explanation of the gen. The priority is used to implement
477  * lru (and in the future other) cache replacement policies; for most purposes
478  * it's just an opaque integer.
479  *
480  * The gens and the priorities don't have a whole lot to do with each other, and
481  * it's actually the gens that must be written out at specific times - it's no
482  * big deal if the priorities don't get written, if we lose them we just reuse
483  * buckets in suboptimal order.
484  *
485  * On disk they're stored in a packed array, and in as many buckets are required
486  * to fit them all. The buckets we use to store them form a list; the journal
487  * header points to the first bucket, the first bucket points to the second
488  * bucket, et cetera.
489  *
490  * This code is used by the allocation code; periodically (whenever it runs out
491  * of buckets to allocate from) the allocation code will invalidate some
492  * buckets, but it can't use those buckets until their new gens are safely on
493  * disk.
494  */
495 
496 static void prio_endio(struct bio *bio)
497 {
498 	struct cache *ca = bio->bi_private;
499 
500 	cache_set_err_on(bio->bi_status, ca->set, "accessing priorities");
501 	bch_bbio_free(bio, ca->set);
502 	closure_put(&ca->prio);
503 }
504 
505 static void prio_io(struct cache *ca, uint64_t bucket, int op,
506 		    unsigned long op_flags)
507 {
508 	struct closure *cl = &ca->prio;
509 	struct bio *bio = bch_bbio_alloc(ca->set);
510 
511 	closure_init_stack(cl);
512 
513 	bio->bi_iter.bi_sector	= bucket * ca->sb.bucket_size;
514 	bio_set_dev(bio, ca->bdev);
515 	bio->bi_iter.bi_size	= bucket_bytes(ca);
516 
517 	bio->bi_end_io	= prio_endio;
518 	bio->bi_private = ca;
519 	bio_set_op_attrs(bio, op, REQ_SYNC|REQ_META|op_flags);
520 	bch_bio_map(bio, ca->disk_buckets);
521 
522 	closure_bio_submit(bio, &ca->prio);
523 	closure_sync(cl);
524 }
525 
526 void bch_prio_write(struct cache *ca)
527 {
528 	int i;
529 	struct bucket *b;
530 	struct closure cl;
531 
532 	closure_init_stack(&cl);
533 
534 	lockdep_assert_held(&ca->set->bucket_lock);
535 
536 	ca->disk_buckets->seq++;
537 
538 	atomic_long_add(ca->sb.bucket_size * prio_buckets(ca),
539 			&ca->meta_sectors_written);
540 
541 	//pr_debug("free %zu, free_inc %zu, unused %zu", fifo_used(&ca->free),
542 	//	 fifo_used(&ca->free_inc), fifo_used(&ca->unused));
543 
544 	for (i = prio_buckets(ca) - 1; i >= 0; --i) {
545 		long bucket;
546 		struct prio_set *p = ca->disk_buckets;
547 		struct bucket_disk *d = p->data;
548 		struct bucket_disk *end = d + prios_per_bucket(ca);
549 
550 		for (b = ca->buckets + i * prios_per_bucket(ca);
551 		     b < ca->buckets + ca->sb.nbuckets && d < end;
552 		     b++, d++) {
553 			d->prio = cpu_to_le16(b->prio);
554 			d->gen = b->gen;
555 		}
556 
557 		p->next_bucket	= ca->prio_buckets[i + 1];
558 		p->magic	= pset_magic(&ca->sb);
559 		p->csum		= bch_crc64(&p->magic, bucket_bytes(ca) - 8);
560 
561 		bucket = bch_bucket_alloc(ca, RESERVE_PRIO, true);
562 		BUG_ON(bucket == -1);
563 
564 		mutex_unlock(&ca->set->bucket_lock);
565 		prio_io(ca, bucket, REQ_OP_WRITE, 0);
566 		mutex_lock(&ca->set->bucket_lock);
567 
568 		ca->prio_buckets[i] = bucket;
569 		atomic_dec_bug(&ca->buckets[bucket].pin);
570 	}
571 
572 	mutex_unlock(&ca->set->bucket_lock);
573 
574 	bch_journal_meta(ca->set, &cl);
575 	closure_sync(&cl);
576 
577 	mutex_lock(&ca->set->bucket_lock);
578 
579 	/*
580 	 * Don't want the old priorities to get garbage collected until after we
581 	 * finish writing the new ones, and they're journalled
582 	 */
583 	for (i = 0; i < prio_buckets(ca); i++) {
584 		if (ca->prio_last_buckets[i])
585 			__bch_bucket_free(ca,
586 				&ca->buckets[ca->prio_last_buckets[i]]);
587 
588 		ca->prio_last_buckets[i] = ca->prio_buckets[i];
589 	}
590 }
591 
592 static void prio_read(struct cache *ca, uint64_t bucket)
593 {
594 	struct prio_set *p = ca->disk_buckets;
595 	struct bucket_disk *d = p->data + prios_per_bucket(ca), *end = d;
596 	struct bucket *b;
597 	unsigned bucket_nr = 0;
598 
599 	for (b = ca->buckets;
600 	     b < ca->buckets + ca->sb.nbuckets;
601 	     b++, d++) {
602 		if (d == end) {
603 			ca->prio_buckets[bucket_nr] = bucket;
604 			ca->prio_last_buckets[bucket_nr] = bucket;
605 			bucket_nr++;
606 
607 			prio_io(ca, bucket, REQ_OP_READ, 0);
608 
609 			if (p->csum != bch_crc64(&p->magic, bucket_bytes(ca) - 8))
610 				pr_warn("bad csum reading priorities");
611 
612 			if (p->magic != pset_magic(&ca->sb))
613 				pr_warn("bad magic reading priorities");
614 
615 			bucket = p->next_bucket;
616 			d = p->data;
617 		}
618 
619 		b->prio = le16_to_cpu(d->prio);
620 		b->gen = b->last_gc = d->gen;
621 	}
622 }
623 
624 /* Bcache device */
625 
626 static int open_dev(struct block_device *b, fmode_t mode)
627 {
628 	struct bcache_device *d = b->bd_disk->private_data;
629 	if (test_bit(BCACHE_DEV_CLOSING, &d->flags))
630 		return -ENXIO;
631 
632 	closure_get(&d->cl);
633 	return 0;
634 }
635 
636 static void release_dev(struct gendisk *b, fmode_t mode)
637 {
638 	struct bcache_device *d = b->private_data;
639 	closure_put(&d->cl);
640 }
641 
642 static int ioctl_dev(struct block_device *b, fmode_t mode,
643 		     unsigned int cmd, unsigned long arg)
644 {
645 	struct bcache_device *d = b->bd_disk->private_data;
646 	return d->ioctl(d, mode, cmd, arg);
647 }
648 
649 static const struct block_device_operations bcache_ops = {
650 	.open		= open_dev,
651 	.release	= release_dev,
652 	.ioctl		= ioctl_dev,
653 	.owner		= THIS_MODULE,
654 };
655 
656 void bcache_device_stop(struct bcache_device *d)
657 {
658 	if (!test_and_set_bit(BCACHE_DEV_CLOSING, &d->flags))
659 		closure_queue(&d->cl);
660 }
661 
662 static void bcache_device_unlink(struct bcache_device *d)
663 {
664 	lockdep_assert_held(&bch_register_lock);
665 
666 	if (d->c && !test_and_set_bit(BCACHE_DEV_UNLINK_DONE, &d->flags)) {
667 		unsigned i;
668 		struct cache *ca;
669 
670 		sysfs_remove_link(&d->c->kobj, d->name);
671 		sysfs_remove_link(&d->kobj, "cache");
672 
673 		for_each_cache(ca, d->c, i)
674 			bd_unlink_disk_holder(ca->bdev, d->disk);
675 	}
676 }
677 
678 static void bcache_device_link(struct bcache_device *d, struct cache_set *c,
679 			       const char *name)
680 {
681 	unsigned i;
682 	struct cache *ca;
683 
684 	for_each_cache(ca, d->c, i)
685 		bd_link_disk_holder(ca->bdev, d->disk);
686 
687 	snprintf(d->name, BCACHEDEVNAME_SIZE,
688 		 "%s%u", name, d->id);
689 
690 	WARN(sysfs_create_link(&d->kobj, &c->kobj, "cache") ||
691 	     sysfs_create_link(&c->kobj, &d->kobj, d->name),
692 	     "Couldn't create device <-> cache set symlinks");
693 
694 	clear_bit(BCACHE_DEV_UNLINK_DONE, &d->flags);
695 }
696 
697 static void bcache_device_detach(struct bcache_device *d)
698 {
699 	lockdep_assert_held(&bch_register_lock);
700 
701 	if (test_bit(BCACHE_DEV_DETACHING, &d->flags)) {
702 		struct uuid_entry *u = d->c->uuids + d->id;
703 
704 		SET_UUID_FLASH_ONLY(u, 0);
705 		memcpy(u->uuid, invalid_uuid, 16);
706 		u->invalidated = cpu_to_le32(get_seconds());
707 		bch_uuid_write(d->c);
708 	}
709 
710 	bcache_device_unlink(d);
711 
712 	d->c->devices[d->id] = NULL;
713 	closure_put(&d->c->caching);
714 	d->c = NULL;
715 }
716 
717 static void bcache_device_attach(struct bcache_device *d, struct cache_set *c,
718 				 unsigned id)
719 {
720 	d->id = id;
721 	d->c = c;
722 	c->devices[id] = d;
723 
724 	if (id >= c->devices_max_used)
725 		c->devices_max_used = id + 1;
726 
727 	closure_get(&c->caching);
728 }
729 
730 static inline int first_minor_to_idx(int first_minor)
731 {
732 	return (first_minor/BCACHE_MINORS);
733 }
734 
735 static inline int idx_to_first_minor(int idx)
736 {
737 	return (idx * BCACHE_MINORS);
738 }
739 
740 static void bcache_device_free(struct bcache_device *d)
741 {
742 	lockdep_assert_held(&bch_register_lock);
743 
744 	pr_info("%s stopped", d->disk->disk_name);
745 
746 	if (d->c)
747 		bcache_device_detach(d);
748 	if (d->disk && d->disk->flags & GENHD_FL_UP)
749 		del_gendisk(d->disk);
750 	if (d->disk && d->disk->queue)
751 		blk_cleanup_queue(d->disk->queue);
752 	if (d->disk) {
753 		ida_simple_remove(&bcache_device_idx,
754 				  first_minor_to_idx(d->disk->first_minor));
755 		put_disk(d->disk);
756 	}
757 
758 	if (d->bio_split)
759 		bioset_free(d->bio_split);
760 	kvfree(d->full_dirty_stripes);
761 	kvfree(d->stripe_sectors_dirty);
762 
763 	closure_debug_destroy(&d->cl);
764 }
765 
766 static int bcache_device_init(struct bcache_device *d, unsigned block_size,
767 			      sector_t sectors)
768 {
769 	struct request_queue *q;
770 	size_t n;
771 	int idx;
772 
773 	if (!d->stripe_size)
774 		d->stripe_size = 1 << 31;
775 
776 	d->nr_stripes = DIV_ROUND_UP_ULL(sectors, d->stripe_size);
777 
778 	if (!d->nr_stripes ||
779 	    d->nr_stripes > INT_MAX ||
780 	    d->nr_stripes > SIZE_MAX / sizeof(atomic_t)) {
781 		pr_err("nr_stripes too large or invalid: %u (start sector beyond end of disk?)",
782 			(unsigned)d->nr_stripes);
783 		return -ENOMEM;
784 	}
785 
786 	n = d->nr_stripes * sizeof(atomic_t);
787 	d->stripe_sectors_dirty = kvzalloc(n, GFP_KERNEL);
788 	if (!d->stripe_sectors_dirty)
789 		return -ENOMEM;
790 
791 	n = BITS_TO_LONGS(d->nr_stripes) * sizeof(unsigned long);
792 	d->full_dirty_stripes = kvzalloc(n, GFP_KERNEL);
793 	if (!d->full_dirty_stripes)
794 		return -ENOMEM;
795 
796 	idx = ida_simple_get(&bcache_device_idx, 0,
797 				BCACHE_DEVICE_IDX_MAX, GFP_KERNEL);
798 	if (idx < 0)
799 		return idx;
800 
801 	if (!(d->bio_split = bioset_create(4, offsetof(struct bbio, bio),
802 					   BIOSET_NEED_BVECS |
803 					   BIOSET_NEED_RESCUER)) ||
804 	    !(d->disk = alloc_disk(BCACHE_MINORS))) {
805 		ida_simple_remove(&bcache_device_idx, idx);
806 		return -ENOMEM;
807 	}
808 
809 	set_capacity(d->disk, sectors);
810 	snprintf(d->disk->disk_name, DISK_NAME_LEN, "bcache%i", idx);
811 
812 	d->disk->major		= bcache_major;
813 	d->disk->first_minor	= idx_to_first_minor(idx);
814 	d->disk->fops		= &bcache_ops;
815 	d->disk->private_data	= d;
816 
817 	q = blk_alloc_queue(GFP_KERNEL);
818 	if (!q)
819 		return -ENOMEM;
820 
821 	blk_queue_make_request(q, NULL);
822 	d->disk->queue			= q;
823 	q->queuedata			= d;
824 	q->backing_dev_info->congested_data = d;
825 	q->limits.max_hw_sectors	= UINT_MAX;
826 	q->limits.max_sectors		= UINT_MAX;
827 	q->limits.max_segment_size	= UINT_MAX;
828 	q->limits.max_segments		= BIO_MAX_PAGES;
829 	blk_queue_max_discard_sectors(q, UINT_MAX);
830 	q->limits.discard_granularity	= 512;
831 	q->limits.io_min		= block_size;
832 	q->limits.logical_block_size	= block_size;
833 	q->limits.physical_block_size	= block_size;
834 	set_bit(QUEUE_FLAG_NONROT,	&d->disk->queue->queue_flags);
835 	clear_bit(QUEUE_FLAG_ADD_RANDOM, &d->disk->queue->queue_flags);
836 	set_bit(QUEUE_FLAG_DISCARD,	&d->disk->queue->queue_flags);
837 
838 	blk_queue_write_cache(q, true, true);
839 
840 	return 0;
841 }
842 
843 /* Cached device */
844 
845 static void calc_cached_dev_sectors(struct cache_set *c)
846 {
847 	uint64_t sectors = 0;
848 	struct cached_dev *dc;
849 
850 	list_for_each_entry(dc, &c->cached_devs, list)
851 		sectors += bdev_sectors(dc->bdev);
852 
853 	c->cached_dev_sectors = sectors;
854 }
855 
856 void bch_cached_dev_run(struct cached_dev *dc)
857 {
858 	struct bcache_device *d = &dc->disk;
859 	char buf[SB_LABEL_SIZE + 1];
860 	char *env[] = {
861 		"DRIVER=bcache",
862 		kasprintf(GFP_KERNEL, "CACHED_UUID=%pU", dc->sb.uuid),
863 		NULL,
864 		NULL,
865 	};
866 
867 	memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
868 	buf[SB_LABEL_SIZE] = '\0';
869 	env[2] = kasprintf(GFP_KERNEL, "CACHED_LABEL=%s", buf);
870 
871 	if (atomic_xchg(&dc->running, 1)) {
872 		kfree(env[1]);
873 		kfree(env[2]);
874 		return;
875 	}
876 
877 	if (!d->c &&
878 	    BDEV_STATE(&dc->sb) != BDEV_STATE_NONE) {
879 		struct closure cl;
880 		closure_init_stack(&cl);
881 
882 		SET_BDEV_STATE(&dc->sb, BDEV_STATE_STALE);
883 		bch_write_bdev_super(dc, &cl);
884 		closure_sync(&cl);
885 	}
886 
887 	add_disk(d->disk);
888 	bd_link_disk_holder(dc->bdev, dc->disk.disk);
889 	/* won't show up in the uevent file, use udevadm monitor -e instead
890 	 * only class / kset properties are persistent */
891 	kobject_uevent_env(&disk_to_dev(d->disk)->kobj, KOBJ_CHANGE, env);
892 	kfree(env[1]);
893 	kfree(env[2]);
894 
895 	if (sysfs_create_link(&d->kobj, &disk_to_dev(d->disk)->kobj, "dev") ||
896 	    sysfs_create_link(&disk_to_dev(d->disk)->kobj, &d->kobj, "bcache"))
897 		pr_debug("error creating sysfs link");
898 }
899 
900 static void cached_dev_detach_finish(struct work_struct *w)
901 {
902 	struct cached_dev *dc = container_of(w, struct cached_dev, detach);
903 	char buf[BDEVNAME_SIZE];
904 	struct closure cl;
905 	closure_init_stack(&cl);
906 
907 	BUG_ON(!test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags));
908 	BUG_ON(refcount_read(&dc->count));
909 
910 	mutex_lock(&bch_register_lock);
911 
912 	cancel_delayed_work_sync(&dc->writeback_rate_update);
913 	if (!IS_ERR_OR_NULL(dc->writeback_thread)) {
914 		kthread_stop(dc->writeback_thread);
915 		dc->writeback_thread = NULL;
916 	}
917 
918 	memset(&dc->sb.set_uuid, 0, 16);
919 	SET_BDEV_STATE(&dc->sb, BDEV_STATE_NONE);
920 
921 	bch_write_bdev_super(dc, &cl);
922 	closure_sync(&cl);
923 
924 	bcache_device_detach(&dc->disk);
925 	list_move(&dc->list, &uncached_devices);
926 
927 	clear_bit(BCACHE_DEV_DETACHING, &dc->disk.flags);
928 	clear_bit(BCACHE_DEV_UNLINK_DONE, &dc->disk.flags);
929 
930 	mutex_unlock(&bch_register_lock);
931 
932 	pr_info("Caching disabled for %s", bdevname(dc->bdev, buf));
933 
934 	/* Drop ref we took in cached_dev_detach() */
935 	closure_put(&dc->disk.cl);
936 }
937 
938 void bch_cached_dev_detach(struct cached_dev *dc)
939 {
940 	lockdep_assert_held(&bch_register_lock);
941 
942 	if (test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags))
943 		return;
944 
945 	if (test_and_set_bit(BCACHE_DEV_DETACHING, &dc->disk.flags))
946 		return;
947 
948 	/*
949 	 * Block the device from being closed and freed until we're finished
950 	 * detaching
951 	 */
952 	closure_get(&dc->disk.cl);
953 
954 	bch_writeback_queue(dc);
955 	cached_dev_put(dc);
956 }
957 
958 int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c)
959 {
960 	uint32_t rtime = cpu_to_le32(get_seconds());
961 	struct uuid_entry *u;
962 	char buf[BDEVNAME_SIZE];
963 
964 	bdevname(dc->bdev, buf);
965 
966 	if (memcmp(dc->sb.set_uuid, c->sb.set_uuid, 16))
967 		return -ENOENT;
968 
969 	if (dc->disk.c) {
970 		pr_err("Can't attach %s: already attached", buf);
971 		return -EINVAL;
972 	}
973 
974 	if (test_bit(CACHE_SET_STOPPING, &c->flags)) {
975 		pr_err("Can't attach %s: shutting down", buf);
976 		return -EINVAL;
977 	}
978 
979 	if (dc->sb.block_size < c->sb.block_size) {
980 		/* Will die */
981 		pr_err("Couldn't attach %s: block size less than set's block size",
982 		       buf);
983 		return -EINVAL;
984 	}
985 
986 	u = uuid_find(c, dc->sb.uuid);
987 
988 	if (u &&
989 	    (BDEV_STATE(&dc->sb) == BDEV_STATE_STALE ||
990 	     BDEV_STATE(&dc->sb) == BDEV_STATE_NONE)) {
991 		memcpy(u->uuid, invalid_uuid, 16);
992 		u->invalidated = cpu_to_le32(get_seconds());
993 		u = NULL;
994 	}
995 
996 	if (!u) {
997 		if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
998 			pr_err("Couldn't find uuid for %s in set", buf);
999 			return -ENOENT;
1000 		}
1001 
1002 		u = uuid_find_empty(c);
1003 		if (!u) {
1004 			pr_err("Not caching %s, no room for UUID", buf);
1005 			return -EINVAL;
1006 		}
1007 	}
1008 
1009 	/* Deadlocks since we're called via sysfs...
1010 	sysfs_remove_file(&dc->kobj, &sysfs_attach);
1011 	 */
1012 
1013 	if (bch_is_zero(u->uuid, 16)) {
1014 		struct closure cl;
1015 		closure_init_stack(&cl);
1016 
1017 		memcpy(u->uuid, dc->sb.uuid, 16);
1018 		memcpy(u->label, dc->sb.label, SB_LABEL_SIZE);
1019 		u->first_reg = u->last_reg = rtime;
1020 		bch_uuid_write(c);
1021 
1022 		memcpy(dc->sb.set_uuid, c->sb.set_uuid, 16);
1023 		SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN);
1024 
1025 		bch_write_bdev_super(dc, &cl);
1026 		closure_sync(&cl);
1027 	} else {
1028 		u->last_reg = rtime;
1029 		bch_uuid_write(c);
1030 	}
1031 
1032 	bcache_device_attach(&dc->disk, c, u - c->uuids);
1033 	list_move(&dc->list, &c->cached_devs);
1034 	calc_cached_dev_sectors(c);
1035 
1036 	smp_wmb();
1037 	/*
1038 	 * dc->c must be set before dc->count != 0 - paired with the mb in
1039 	 * cached_dev_get()
1040 	 */
1041 	refcount_set(&dc->count, 1);
1042 
1043 	/* Block writeback thread, but spawn it */
1044 	down_write(&dc->writeback_lock);
1045 	if (bch_cached_dev_writeback_start(dc)) {
1046 		up_write(&dc->writeback_lock);
1047 		return -ENOMEM;
1048 	}
1049 
1050 	if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
1051 		bch_sectors_dirty_init(&dc->disk);
1052 		atomic_set(&dc->has_dirty, 1);
1053 		refcount_inc(&dc->count);
1054 		bch_writeback_queue(dc);
1055 	}
1056 
1057 	bch_cached_dev_run(dc);
1058 	bcache_device_link(&dc->disk, c, "bdev");
1059 
1060 	/* Allow the writeback thread to proceed */
1061 	up_write(&dc->writeback_lock);
1062 
1063 	pr_info("Caching %s as %s on set %pU",
1064 		bdevname(dc->bdev, buf), dc->disk.disk->disk_name,
1065 		dc->disk.c->sb.set_uuid);
1066 	return 0;
1067 }
1068 
1069 void bch_cached_dev_release(struct kobject *kobj)
1070 {
1071 	struct cached_dev *dc = container_of(kobj, struct cached_dev,
1072 					     disk.kobj);
1073 	kfree(dc);
1074 	module_put(THIS_MODULE);
1075 }
1076 
1077 static void cached_dev_free(struct closure *cl)
1078 {
1079 	struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1080 
1081 	cancel_delayed_work_sync(&dc->writeback_rate_update);
1082 	if (!IS_ERR_OR_NULL(dc->writeback_thread))
1083 		kthread_stop(dc->writeback_thread);
1084 	if (dc->writeback_write_wq)
1085 		destroy_workqueue(dc->writeback_write_wq);
1086 
1087 	mutex_lock(&bch_register_lock);
1088 
1089 	if (atomic_read(&dc->running))
1090 		bd_unlink_disk_holder(dc->bdev, dc->disk.disk);
1091 	bcache_device_free(&dc->disk);
1092 	list_del(&dc->list);
1093 
1094 	mutex_unlock(&bch_register_lock);
1095 
1096 	if (!IS_ERR_OR_NULL(dc->bdev))
1097 		blkdev_put(dc->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1098 
1099 	wake_up(&unregister_wait);
1100 
1101 	kobject_put(&dc->disk.kobj);
1102 }
1103 
1104 static void cached_dev_flush(struct closure *cl)
1105 {
1106 	struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1107 	struct bcache_device *d = &dc->disk;
1108 
1109 	mutex_lock(&bch_register_lock);
1110 	bcache_device_unlink(d);
1111 	mutex_unlock(&bch_register_lock);
1112 
1113 	bch_cache_accounting_destroy(&dc->accounting);
1114 	kobject_del(&d->kobj);
1115 
1116 	continue_at(cl, cached_dev_free, system_wq);
1117 }
1118 
1119 static int cached_dev_init(struct cached_dev *dc, unsigned block_size)
1120 {
1121 	int ret;
1122 	struct io *io;
1123 	struct request_queue *q = bdev_get_queue(dc->bdev);
1124 
1125 	__module_get(THIS_MODULE);
1126 	INIT_LIST_HEAD(&dc->list);
1127 	closure_init(&dc->disk.cl, NULL);
1128 	set_closure_fn(&dc->disk.cl, cached_dev_flush, system_wq);
1129 	kobject_init(&dc->disk.kobj, &bch_cached_dev_ktype);
1130 	INIT_WORK(&dc->detach, cached_dev_detach_finish);
1131 	sema_init(&dc->sb_write_mutex, 1);
1132 	INIT_LIST_HEAD(&dc->io_lru);
1133 	spin_lock_init(&dc->io_lock);
1134 	bch_cache_accounting_init(&dc->accounting, &dc->disk.cl);
1135 
1136 	dc->sequential_cutoff		= 4 << 20;
1137 
1138 	for (io = dc->io; io < dc->io + RECENT_IO; io++) {
1139 		list_add(&io->lru, &dc->io_lru);
1140 		hlist_add_head(&io->hash, dc->io_hash + RECENT_IO);
1141 	}
1142 
1143 	dc->disk.stripe_size = q->limits.io_opt >> 9;
1144 
1145 	if (dc->disk.stripe_size)
1146 		dc->partial_stripes_expensive =
1147 			q->limits.raid_partial_stripes_expensive;
1148 
1149 	ret = bcache_device_init(&dc->disk, block_size,
1150 			 dc->bdev->bd_part->nr_sects - dc->sb.data_offset);
1151 	if (ret)
1152 		return ret;
1153 
1154 	dc->disk.disk->queue->backing_dev_info->ra_pages =
1155 		max(dc->disk.disk->queue->backing_dev_info->ra_pages,
1156 		    q->backing_dev_info->ra_pages);
1157 
1158 	bch_cached_dev_request_init(dc);
1159 	bch_cached_dev_writeback_init(dc);
1160 	return 0;
1161 }
1162 
1163 /* Cached device - bcache superblock */
1164 
1165 static void register_bdev(struct cache_sb *sb, struct page *sb_page,
1166 				 struct block_device *bdev,
1167 				 struct cached_dev *dc)
1168 {
1169 	char name[BDEVNAME_SIZE];
1170 	const char *err = "cannot allocate memory";
1171 	struct cache_set *c;
1172 
1173 	memcpy(&dc->sb, sb, sizeof(struct cache_sb));
1174 	dc->bdev = bdev;
1175 	dc->bdev->bd_holder = dc;
1176 
1177 	bio_init(&dc->sb_bio, dc->sb_bio.bi_inline_vecs, 1);
1178 	bio_first_bvec_all(&dc->sb_bio)->bv_page = sb_page;
1179 	get_page(sb_page);
1180 
1181 	if (cached_dev_init(dc, sb->block_size << 9))
1182 		goto err;
1183 
1184 	err = "error creating kobject";
1185 	if (kobject_add(&dc->disk.kobj, &part_to_dev(bdev->bd_part)->kobj,
1186 			"bcache"))
1187 		goto err;
1188 	if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj))
1189 		goto err;
1190 
1191 	pr_info("registered backing device %s", bdevname(bdev, name));
1192 
1193 	list_add(&dc->list, &uncached_devices);
1194 	list_for_each_entry(c, &bch_cache_sets, list)
1195 		bch_cached_dev_attach(dc, c);
1196 
1197 	if (BDEV_STATE(&dc->sb) == BDEV_STATE_NONE ||
1198 	    BDEV_STATE(&dc->sb) == BDEV_STATE_STALE)
1199 		bch_cached_dev_run(dc);
1200 
1201 	return;
1202 err:
1203 	pr_notice("error opening %s: %s", bdevname(bdev, name), err);
1204 	bcache_device_stop(&dc->disk);
1205 }
1206 
1207 /* Flash only volumes */
1208 
1209 void bch_flash_dev_release(struct kobject *kobj)
1210 {
1211 	struct bcache_device *d = container_of(kobj, struct bcache_device,
1212 					       kobj);
1213 	kfree(d);
1214 }
1215 
1216 static void flash_dev_free(struct closure *cl)
1217 {
1218 	struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1219 	mutex_lock(&bch_register_lock);
1220 	bcache_device_free(d);
1221 	mutex_unlock(&bch_register_lock);
1222 	kobject_put(&d->kobj);
1223 }
1224 
1225 static void flash_dev_flush(struct closure *cl)
1226 {
1227 	struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1228 
1229 	mutex_lock(&bch_register_lock);
1230 	bcache_device_unlink(d);
1231 	mutex_unlock(&bch_register_lock);
1232 	kobject_del(&d->kobj);
1233 	continue_at(cl, flash_dev_free, system_wq);
1234 }
1235 
1236 static int flash_dev_run(struct cache_set *c, struct uuid_entry *u)
1237 {
1238 	struct bcache_device *d = kzalloc(sizeof(struct bcache_device),
1239 					  GFP_KERNEL);
1240 	if (!d)
1241 		return -ENOMEM;
1242 
1243 	closure_init(&d->cl, NULL);
1244 	set_closure_fn(&d->cl, flash_dev_flush, system_wq);
1245 
1246 	kobject_init(&d->kobj, &bch_flash_dev_ktype);
1247 
1248 	if (bcache_device_init(d, block_bytes(c), u->sectors))
1249 		goto err;
1250 
1251 	bcache_device_attach(d, c, u - c->uuids);
1252 	bch_sectors_dirty_init(d);
1253 	bch_flash_dev_request_init(d);
1254 	add_disk(d->disk);
1255 
1256 	if (kobject_add(&d->kobj, &disk_to_dev(d->disk)->kobj, "bcache"))
1257 		goto err;
1258 
1259 	bcache_device_link(d, c, "volume");
1260 
1261 	return 0;
1262 err:
1263 	kobject_put(&d->kobj);
1264 	return -ENOMEM;
1265 }
1266 
1267 static int flash_devs_run(struct cache_set *c)
1268 {
1269 	int ret = 0;
1270 	struct uuid_entry *u;
1271 
1272 	for (u = c->uuids;
1273 	     u < c->uuids + c->devices_max_used && !ret;
1274 	     u++)
1275 		if (UUID_FLASH_ONLY(u))
1276 			ret = flash_dev_run(c, u);
1277 
1278 	return ret;
1279 }
1280 
1281 int bch_flash_dev_create(struct cache_set *c, uint64_t size)
1282 {
1283 	struct uuid_entry *u;
1284 
1285 	if (test_bit(CACHE_SET_STOPPING, &c->flags))
1286 		return -EINTR;
1287 
1288 	if (!test_bit(CACHE_SET_RUNNING, &c->flags))
1289 		return -EPERM;
1290 
1291 	u = uuid_find_empty(c);
1292 	if (!u) {
1293 		pr_err("Can't create volume, no room for UUID");
1294 		return -EINVAL;
1295 	}
1296 
1297 	get_random_bytes(u->uuid, 16);
1298 	memset(u->label, 0, 32);
1299 	u->first_reg = u->last_reg = cpu_to_le32(get_seconds());
1300 
1301 	SET_UUID_FLASH_ONLY(u, 1);
1302 	u->sectors = size >> 9;
1303 
1304 	bch_uuid_write(c);
1305 
1306 	return flash_dev_run(c, u);
1307 }
1308 
1309 /* Cache set */
1310 
1311 __printf(2, 3)
1312 bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...)
1313 {
1314 	va_list args;
1315 
1316 	if (c->on_error != ON_ERROR_PANIC &&
1317 	    test_bit(CACHE_SET_STOPPING, &c->flags))
1318 		return false;
1319 
1320 	/* XXX: we can be called from atomic context
1321 	acquire_console_sem();
1322 	*/
1323 
1324 	printk(KERN_ERR "bcache: error on %pU: ", c->sb.set_uuid);
1325 
1326 	va_start(args, fmt);
1327 	vprintk(fmt, args);
1328 	va_end(args);
1329 
1330 	printk(", disabling caching\n");
1331 
1332 	if (c->on_error == ON_ERROR_PANIC)
1333 		panic("panic forced after error\n");
1334 
1335 	bch_cache_set_unregister(c);
1336 	return true;
1337 }
1338 
1339 void bch_cache_set_release(struct kobject *kobj)
1340 {
1341 	struct cache_set *c = container_of(kobj, struct cache_set, kobj);
1342 	kfree(c);
1343 	module_put(THIS_MODULE);
1344 }
1345 
1346 static void cache_set_free(struct closure *cl)
1347 {
1348 	struct cache_set *c = container_of(cl, struct cache_set, cl);
1349 	struct cache *ca;
1350 	unsigned i;
1351 
1352 	if (!IS_ERR_OR_NULL(c->debug))
1353 		debugfs_remove(c->debug);
1354 
1355 	bch_open_buckets_free(c);
1356 	bch_btree_cache_free(c);
1357 	bch_journal_free(c);
1358 
1359 	for_each_cache(ca, c, i)
1360 		if (ca) {
1361 			ca->set = NULL;
1362 			c->cache[ca->sb.nr_this_dev] = NULL;
1363 			kobject_put(&ca->kobj);
1364 		}
1365 
1366 	bch_bset_sort_state_free(&c->sort);
1367 	free_pages((unsigned long) c->uuids, ilog2(bucket_pages(c)));
1368 
1369 	if (c->moving_gc_wq)
1370 		destroy_workqueue(c->moving_gc_wq);
1371 	if (c->bio_split)
1372 		bioset_free(c->bio_split);
1373 	if (c->fill_iter)
1374 		mempool_destroy(c->fill_iter);
1375 	if (c->bio_meta)
1376 		mempool_destroy(c->bio_meta);
1377 	if (c->search)
1378 		mempool_destroy(c->search);
1379 	kfree(c->devices);
1380 
1381 	mutex_lock(&bch_register_lock);
1382 	list_del(&c->list);
1383 	mutex_unlock(&bch_register_lock);
1384 
1385 	pr_info("Cache set %pU unregistered", c->sb.set_uuid);
1386 	wake_up(&unregister_wait);
1387 
1388 	closure_debug_destroy(&c->cl);
1389 	kobject_put(&c->kobj);
1390 }
1391 
1392 static void cache_set_flush(struct closure *cl)
1393 {
1394 	struct cache_set *c = container_of(cl, struct cache_set, caching);
1395 	struct cache *ca;
1396 	struct btree *b;
1397 	unsigned i;
1398 
1399 	bch_cache_accounting_destroy(&c->accounting);
1400 
1401 	kobject_put(&c->internal);
1402 	kobject_del(&c->kobj);
1403 
1404 	if (c->gc_thread)
1405 		kthread_stop(c->gc_thread);
1406 
1407 	if (!IS_ERR_OR_NULL(c->root))
1408 		list_add(&c->root->list, &c->btree_cache);
1409 
1410 	/* Should skip this if we're unregistering because of an error */
1411 	list_for_each_entry(b, &c->btree_cache, list) {
1412 		mutex_lock(&b->write_lock);
1413 		if (btree_node_dirty(b))
1414 			__bch_btree_node_write(b, NULL);
1415 		mutex_unlock(&b->write_lock);
1416 	}
1417 
1418 	for_each_cache(ca, c, i)
1419 		if (ca->alloc_thread)
1420 			kthread_stop(ca->alloc_thread);
1421 
1422 	if (c->journal.cur) {
1423 		cancel_delayed_work_sync(&c->journal.work);
1424 		/* flush last journal entry if needed */
1425 		c->journal.work.work.func(&c->journal.work.work);
1426 	}
1427 
1428 	closure_return(cl);
1429 }
1430 
1431 static void __cache_set_unregister(struct closure *cl)
1432 {
1433 	struct cache_set *c = container_of(cl, struct cache_set, caching);
1434 	struct cached_dev *dc;
1435 	size_t i;
1436 
1437 	mutex_lock(&bch_register_lock);
1438 
1439 	for (i = 0; i < c->devices_max_used; i++)
1440 		if (c->devices[i]) {
1441 			if (!UUID_FLASH_ONLY(&c->uuids[i]) &&
1442 			    test_bit(CACHE_SET_UNREGISTERING, &c->flags)) {
1443 				dc = container_of(c->devices[i],
1444 						  struct cached_dev, disk);
1445 				bch_cached_dev_detach(dc);
1446 			} else {
1447 				bcache_device_stop(c->devices[i]);
1448 			}
1449 		}
1450 
1451 	mutex_unlock(&bch_register_lock);
1452 
1453 	continue_at(cl, cache_set_flush, system_wq);
1454 }
1455 
1456 void bch_cache_set_stop(struct cache_set *c)
1457 {
1458 	if (!test_and_set_bit(CACHE_SET_STOPPING, &c->flags))
1459 		closure_queue(&c->caching);
1460 }
1461 
1462 void bch_cache_set_unregister(struct cache_set *c)
1463 {
1464 	set_bit(CACHE_SET_UNREGISTERING, &c->flags);
1465 	bch_cache_set_stop(c);
1466 }
1467 
1468 #define alloc_bucket_pages(gfp, c)			\
1469 	((void *) __get_free_pages(__GFP_ZERO|gfp, ilog2(bucket_pages(c))))
1470 
1471 struct cache_set *bch_cache_set_alloc(struct cache_sb *sb)
1472 {
1473 	int iter_size;
1474 	struct cache_set *c = kzalloc(sizeof(struct cache_set), GFP_KERNEL);
1475 	if (!c)
1476 		return NULL;
1477 
1478 	__module_get(THIS_MODULE);
1479 	closure_init(&c->cl, NULL);
1480 	set_closure_fn(&c->cl, cache_set_free, system_wq);
1481 
1482 	closure_init(&c->caching, &c->cl);
1483 	set_closure_fn(&c->caching, __cache_set_unregister, system_wq);
1484 
1485 	/* Maybe create continue_at_noreturn() and use it here? */
1486 	closure_set_stopped(&c->cl);
1487 	closure_put(&c->cl);
1488 
1489 	kobject_init(&c->kobj, &bch_cache_set_ktype);
1490 	kobject_init(&c->internal, &bch_cache_set_internal_ktype);
1491 
1492 	bch_cache_accounting_init(&c->accounting, &c->cl);
1493 
1494 	memcpy(c->sb.set_uuid, sb->set_uuid, 16);
1495 	c->sb.block_size	= sb->block_size;
1496 	c->sb.bucket_size	= sb->bucket_size;
1497 	c->sb.nr_in_set		= sb->nr_in_set;
1498 	c->sb.last_mount	= sb->last_mount;
1499 	c->bucket_bits		= ilog2(sb->bucket_size);
1500 	c->block_bits		= ilog2(sb->block_size);
1501 	c->nr_uuids		= bucket_bytes(c) / sizeof(struct uuid_entry);
1502 	c->devices_max_used	= 0;
1503 	c->btree_pages		= bucket_pages(c);
1504 	if (c->btree_pages > BTREE_MAX_PAGES)
1505 		c->btree_pages = max_t(int, c->btree_pages / 4,
1506 				       BTREE_MAX_PAGES);
1507 
1508 	sema_init(&c->sb_write_mutex, 1);
1509 	mutex_init(&c->bucket_lock);
1510 	init_waitqueue_head(&c->btree_cache_wait);
1511 	init_waitqueue_head(&c->bucket_wait);
1512 	init_waitqueue_head(&c->gc_wait);
1513 	sema_init(&c->uuid_write_mutex, 1);
1514 
1515 	spin_lock_init(&c->btree_gc_time.lock);
1516 	spin_lock_init(&c->btree_split_time.lock);
1517 	spin_lock_init(&c->btree_read_time.lock);
1518 
1519 	bch_moving_init_cache_set(c);
1520 
1521 	INIT_LIST_HEAD(&c->list);
1522 	INIT_LIST_HEAD(&c->cached_devs);
1523 	INIT_LIST_HEAD(&c->btree_cache);
1524 	INIT_LIST_HEAD(&c->btree_cache_freeable);
1525 	INIT_LIST_HEAD(&c->btree_cache_freed);
1526 	INIT_LIST_HEAD(&c->data_buckets);
1527 
1528 	c->search = mempool_create_slab_pool(32, bch_search_cache);
1529 	if (!c->search)
1530 		goto err;
1531 
1532 	iter_size = (sb->bucket_size / sb->block_size + 1) *
1533 		sizeof(struct btree_iter_set);
1534 
1535 	if (!(c->devices = kzalloc(c->nr_uuids * sizeof(void *), GFP_KERNEL)) ||
1536 	    !(c->bio_meta = mempool_create_kmalloc_pool(2,
1537 				sizeof(struct bbio) + sizeof(struct bio_vec) *
1538 				bucket_pages(c))) ||
1539 	    !(c->fill_iter = mempool_create_kmalloc_pool(1, iter_size)) ||
1540 	    !(c->bio_split = bioset_create(4, offsetof(struct bbio, bio),
1541 					   BIOSET_NEED_BVECS |
1542 					   BIOSET_NEED_RESCUER)) ||
1543 	    !(c->uuids = alloc_bucket_pages(GFP_KERNEL, c)) ||
1544 	    !(c->moving_gc_wq = alloc_workqueue("bcache_gc",
1545 						WQ_MEM_RECLAIM, 0)) ||
1546 	    bch_journal_alloc(c) ||
1547 	    bch_btree_cache_alloc(c) ||
1548 	    bch_open_buckets_alloc(c) ||
1549 	    bch_bset_sort_state_init(&c->sort, ilog2(c->btree_pages)))
1550 		goto err;
1551 
1552 	c->congested_read_threshold_us	= 2000;
1553 	c->congested_write_threshold_us	= 20000;
1554 	c->error_limit	= 8 << IO_ERROR_SHIFT;
1555 
1556 	return c;
1557 err:
1558 	bch_cache_set_unregister(c);
1559 	return NULL;
1560 }
1561 
1562 static void run_cache_set(struct cache_set *c)
1563 {
1564 	const char *err = "cannot allocate memory";
1565 	struct cached_dev *dc, *t;
1566 	struct cache *ca;
1567 	struct closure cl;
1568 	unsigned i;
1569 
1570 	closure_init_stack(&cl);
1571 
1572 	for_each_cache(ca, c, i)
1573 		c->nbuckets += ca->sb.nbuckets;
1574 	set_gc_sectors(c);
1575 
1576 	if (CACHE_SYNC(&c->sb)) {
1577 		LIST_HEAD(journal);
1578 		struct bkey *k;
1579 		struct jset *j;
1580 
1581 		err = "cannot allocate memory for journal";
1582 		if (bch_journal_read(c, &journal))
1583 			goto err;
1584 
1585 		pr_debug("btree_journal_read() done");
1586 
1587 		err = "no journal entries found";
1588 		if (list_empty(&journal))
1589 			goto err;
1590 
1591 		j = &list_entry(journal.prev, struct journal_replay, list)->j;
1592 
1593 		err = "IO error reading priorities";
1594 		for_each_cache(ca, c, i)
1595 			prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev]);
1596 
1597 		/*
1598 		 * If prio_read() fails it'll call cache_set_error and we'll
1599 		 * tear everything down right away, but if we perhaps checked
1600 		 * sooner we could avoid journal replay.
1601 		 */
1602 
1603 		k = &j->btree_root;
1604 
1605 		err = "bad btree root";
1606 		if (__bch_btree_ptr_invalid(c, k))
1607 			goto err;
1608 
1609 		err = "error reading btree root";
1610 		c->root = bch_btree_node_get(c, NULL, k, j->btree_level, true, NULL);
1611 		if (IS_ERR_OR_NULL(c->root))
1612 			goto err;
1613 
1614 		list_del_init(&c->root->list);
1615 		rw_unlock(true, c->root);
1616 
1617 		err = uuid_read(c, j, &cl);
1618 		if (err)
1619 			goto err;
1620 
1621 		err = "error in recovery";
1622 		if (bch_btree_check(c))
1623 			goto err;
1624 
1625 		bch_journal_mark(c, &journal);
1626 		bch_initial_gc_finish(c);
1627 		pr_debug("btree_check() done");
1628 
1629 		/*
1630 		 * bcache_journal_next() can't happen sooner, or
1631 		 * btree_gc_finish() will give spurious errors about last_gc >
1632 		 * gc_gen - this is a hack but oh well.
1633 		 */
1634 		bch_journal_next(&c->journal);
1635 
1636 		err = "error starting allocator thread";
1637 		for_each_cache(ca, c, i)
1638 			if (bch_cache_allocator_start(ca))
1639 				goto err;
1640 
1641 		/*
1642 		 * First place it's safe to allocate: btree_check() and
1643 		 * btree_gc_finish() have to run before we have buckets to
1644 		 * allocate, and bch_bucket_alloc_set() might cause a journal
1645 		 * entry to be written so bcache_journal_next() has to be called
1646 		 * first.
1647 		 *
1648 		 * If the uuids were in the old format we have to rewrite them
1649 		 * before the next journal entry is written:
1650 		 */
1651 		if (j->version < BCACHE_JSET_VERSION_UUID)
1652 			__uuid_write(c);
1653 
1654 		bch_journal_replay(c, &journal);
1655 	} else {
1656 		pr_notice("invalidating existing data");
1657 
1658 		for_each_cache(ca, c, i) {
1659 			unsigned j;
1660 
1661 			ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7,
1662 					      2, SB_JOURNAL_BUCKETS);
1663 
1664 			for (j = 0; j < ca->sb.keys; j++)
1665 				ca->sb.d[j] = ca->sb.first_bucket + j;
1666 		}
1667 
1668 		bch_initial_gc_finish(c);
1669 
1670 		err = "error starting allocator thread";
1671 		for_each_cache(ca, c, i)
1672 			if (bch_cache_allocator_start(ca))
1673 				goto err;
1674 
1675 		mutex_lock(&c->bucket_lock);
1676 		for_each_cache(ca, c, i)
1677 			bch_prio_write(ca);
1678 		mutex_unlock(&c->bucket_lock);
1679 
1680 		err = "cannot allocate new UUID bucket";
1681 		if (__uuid_write(c))
1682 			goto err;
1683 
1684 		err = "cannot allocate new btree root";
1685 		c->root = __bch_btree_node_alloc(c, NULL, 0, true, NULL);
1686 		if (IS_ERR_OR_NULL(c->root))
1687 			goto err;
1688 
1689 		mutex_lock(&c->root->write_lock);
1690 		bkey_copy_key(&c->root->key, &MAX_KEY);
1691 		bch_btree_node_write(c->root, &cl);
1692 		mutex_unlock(&c->root->write_lock);
1693 
1694 		bch_btree_set_root(c->root);
1695 		rw_unlock(true, c->root);
1696 
1697 		/*
1698 		 * We don't want to write the first journal entry until
1699 		 * everything is set up - fortunately journal entries won't be
1700 		 * written until the SET_CACHE_SYNC() here:
1701 		 */
1702 		SET_CACHE_SYNC(&c->sb, true);
1703 
1704 		bch_journal_next(&c->journal);
1705 		bch_journal_meta(c, &cl);
1706 	}
1707 
1708 	err = "error starting gc thread";
1709 	if (bch_gc_thread_start(c))
1710 		goto err;
1711 
1712 	closure_sync(&cl);
1713 	c->sb.last_mount = get_seconds();
1714 	bcache_write_super(c);
1715 
1716 	list_for_each_entry_safe(dc, t, &uncached_devices, list)
1717 		bch_cached_dev_attach(dc, c);
1718 
1719 	flash_devs_run(c);
1720 
1721 	set_bit(CACHE_SET_RUNNING, &c->flags);
1722 	return;
1723 err:
1724 	closure_sync(&cl);
1725 	/* XXX: test this, it's broken */
1726 	bch_cache_set_error(c, "%s", err);
1727 }
1728 
1729 static bool can_attach_cache(struct cache *ca, struct cache_set *c)
1730 {
1731 	return ca->sb.block_size	== c->sb.block_size &&
1732 		ca->sb.bucket_size	== c->sb.bucket_size &&
1733 		ca->sb.nr_in_set	== c->sb.nr_in_set;
1734 }
1735 
1736 static const char *register_cache_set(struct cache *ca)
1737 {
1738 	char buf[12];
1739 	const char *err = "cannot allocate memory";
1740 	struct cache_set *c;
1741 
1742 	list_for_each_entry(c, &bch_cache_sets, list)
1743 		if (!memcmp(c->sb.set_uuid, ca->sb.set_uuid, 16)) {
1744 			if (c->cache[ca->sb.nr_this_dev])
1745 				return "duplicate cache set member";
1746 
1747 			if (!can_attach_cache(ca, c))
1748 				return "cache sb does not match set";
1749 
1750 			if (!CACHE_SYNC(&ca->sb))
1751 				SET_CACHE_SYNC(&c->sb, false);
1752 
1753 			goto found;
1754 		}
1755 
1756 	c = bch_cache_set_alloc(&ca->sb);
1757 	if (!c)
1758 		return err;
1759 
1760 	err = "error creating kobject";
1761 	if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->sb.set_uuid) ||
1762 	    kobject_add(&c->internal, &c->kobj, "internal"))
1763 		goto err;
1764 
1765 	if (bch_cache_accounting_add_kobjs(&c->accounting, &c->kobj))
1766 		goto err;
1767 
1768 	bch_debug_init_cache_set(c);
1769 
1770 	list_add(&c->list, &bch_cache_sets);
1771 found:
1772 	sprintf(buf, "cache%i", ca->sb.nr_this_dev);
1773 	if (sysfs_create_link(&ca->kobj, &c->kobj, "set") ||
1774 	    sysfs_create_link(&c->kobj, &ca->kobj, buf))
1775 		goto err;
1776 
1777 	if (ca->sb.seq > c->sb.seq) {
1778 		c->sb.version		= ca->sb.version;
1779 		memcpy(c->sb.set_uuid, ca->sb.set_uuid, 16);
1780 		c->sb.flags             = ca->sb.flags;
1781 		c->sb.seq		= ca->sb.seq;
1782 		pr_debug("set version = %llu", c->sb.version);
1783 	}
1784 
1785 	kobject_get(&ca->kobj);
1786 	ca->set = c;
1787 	ca->set->cache[ca->sb.nr_this_dev] = ca;
1788 	c->cache_by_alloc[c->caches_loaded++] = ca;
1789 
1790 	if (c->caches_loaded == c->sb.nr_in_set)
1791 		run_cache_set(c);
1792 
1793 	return NULL;
1794 err:
1795 	bch_cache_set_unregister(c);
1796 	return err;
1797 }
1798 
1799 /* Cache device */
1800 
1801 void bch_cache_release(struct kobject *kobj)
1802 {
1803 	struct cache *ca = container_of(kobj, struct cache, kobj);
1804 	unsigned i;
1805 
1806 	if (ca->set) {
1807 		BUG_ON(ca->set->cache[ca->sb.nr_this_dev] != ca);
1808 		ca->set->cache[ca->sb.nr_this_dev] = NULL;
1809 	}
1810 
1811 	free_pages((unsigned long) ca->disk_buckets, ilog2(bucket_pages(ca)));
1812 	kfree(ca->prio_buckets);
1813 	vfree(ca->buckets);
1814 
1815 	free_heap(&ca->heap);
1816 	free_fifo(&ca->free_inc);
1817 
1818 	for (i = 0; i < RESERVE_NR; i++)
1819 		free_fifo(&ca->free[i]);
1820 
1821 	if (ca->sb_bio.bi_inline_vecs[0].bv_page)
1822 		put_page(bio_first_page_all(&ca->sb_bio));
1823 
1824 	if (!IS_ERR_OR_NULL(ca->bdev))
1825 		blkdev_put(ca->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1826 
1827 	kfree(ca);
1828 	module_put(THIS_MODULE);
1829 }
1830 
1831 static int cache_alloc(struct cache *ca)
1832 {
1833 	size_t free;
1834 	struct bucket *b;
1835 
1836 	__module_get(THIS_MODULE);
1837 	kobject_init(&ca->kobj, &bch_cache_ktype);
1838 
1839 	bio_init(&ca->journal.bio, ca->journal.bio.bi_inline_vecs, 8);
1840 
1841 	free = roundup_pow_of_two(ca->sb.nbuckets) >> 10;
1842 
1843 	if (!init_fifo(&ca->free[RESERVE_BTREE], 8, GFP_KERNEL) ||
1844 	    !init_fifo_exact(&ca->free[RESERVE_PRIO], prio_buckets(ca), GFP_KERNEL) ||
1845 	    !init_fifo(&ca->free[RESERVE_MOVINGGC], free, GFP_KERNEL) ||
1846 	    !init_fifo(&ca->free[RESERVE_NONE], free, GFP_KERNEL) ||
1847 	    !init_fifo(&ca->free_inc,	free << 2, GFP_KERNEL) ||
1848 	    !init_heap(&ca->heap,	free << 3, GFP_KERNEL) ||
1849 	    !(ca->buckets	= vzalloc(sizeof(struct bucket) *
1850 					  ca->sb.nbuckets)) ||
1851 	    !(ca->prio_buckets	= kzalloc(sizeof(uint64_t) * prio_buckets(ca) *
1852 					  2, GFP_KERNEL)) ||
1853 	    !(ca->disk_buckets	= alloc_bucket_pages(GFP_KERNEL, ca)))
1854 		return -ENOMEM;
1855 
1856 	ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca);
1857 
1858 	for_each_bucket(b, ca)
1859 		atomic_set(&b->pin, 0);
1860 
1861 	return 0;
1862 }
1863 
1864 static int register_cache(struct cache_sb *sb, struct page *sb_page,
1865 				struct block_device *bdev, struct cache *ca)
1866 {
1867 	char name[BDEVNAME_SIZE];
1868 	const char *err = NULL; /* must be set for any error case */
1869 	int ret = 0;
1870 
1871 	memcpy(&ca->sb, sb, sizeof(struct cache_sb));
1872 	ca->bdev = bdev;
1873 	ca->bdev->bd_holder = ca;
1874 
1875 	bio_init(&ca->sb_bio, ca->sb_bio.bi_inline_vecs, 1);
1876 	bio_first_bvec_all(&ca->sb_bio)->bv_page = sb_page;
1877 	get_page(sb_page);
1878 
1879 	if (blk_queue_discard(bdev_get_queue(ca->bdev)))
1880 		ca->discard = CACHE_DISCARD(&ca->sb);
1881 
1882 	ret = cache_alloc(ca);
1883 	if (ret != 0) {
1884 		if (ret == -ENOMEM)
1885 			err = "cache_alloc(): -ENOMEM";
1886 		else
1887 			err = "cache_alloc(): unknown error";
1888 		goto err;
1889 	}
1890 
1891 	if (kobject_add(&ca->kobj, &part_to_dev(bdev->bd_part)->kobj, "bcache")) {
1892 		err = "error calling kobject_add";
1893 		ret = -ENOMEM;
1894 		goto out;
1895 	}
1896 
1897 	mutex_lock(&bch_register_lock);
1898 	err = register_cache_set(ca);
1899 	mutex_unlock(&bch_register_lock);
1900 
1901 	if (err) {
1902 		ret = -ENODEV;
1903 		goto out;
1904 	}
1905 
1906 	pr_info("registered cache device %s", bdevname(bdev, name));
1907 
1908 out:
1909 	kobject_put(&ca->kobj);
1910 
1911 err:
1912 	if (err)
1913 		pr_notice("error opening %s: %s", bdevname(bdev, name), err);
1914 
1915 	return ret;
1916 }
1917 
1918 /* Global interfaces/init */
1919 
1920 static ssize_t register_bcache(struct kobject *, struct kobj_attribute *,
1921 			       const char *, size_t);
1922 
1923 kobj_attribute_write(register,		register_bcache);
1924 kobj_attribute_write(register_quiet,	register_bcache);
1925 
1926 static bool bch_is_open_backing(struct block_device *bdev) {
1927 	struct cache_set *c, *tc;
1928 	struct cached_dev *dc, *t;
1929 
1930 	list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
1931 		list_for_each_entry_safe(dc, t, &c->cached_devs, list)
1932 			if (dc->bdev == bdev)
1933 				return true;
1934 	list_for_each_entry_safe(dc, t, &uncached_devices, list)
1935 		if (dc->bdev == bdev)
1936 			return true;
1937 	return false;
1938 }
1939 
1940 static bool bch_is_open_cache(struct block_device *bdev) {
1941 	struct cache_set *c, *tc;
1942 	struct cache *ca;
1943 	unsigned i;
1944 
1945 	list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
1946 		for_each_cache(ca, c, i)
1947 			if (ca->bdev == bdev)
1948 				return true;
1949 	return false;
1950 }
1951 
1952 static bool bch_is_open(struct block_device *bdev) {
1953 	return bch_is_open_cache(bdev) || bch_is_open_backing(bdev);
1954 }
1955 
1956 static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr,
1957 			       const char *buffer, size_t size)
1958 {
1959 	ssize_t ret = size;
1960 	const char *err = "cannot allocate memory";
1961 	char *path = NULL;
1962 	struct cache_sb *sb = NULL;
1963 	struct block_device *bdev = NULL;
1964 	struct page *sb_page = NULL;
1965 
1966 	if (!try_module_get(THIS_MODULE))
1967 		return -EBUSY;
1968 
1969 	if (!(path = kstrndup(buffer, size, GFP_KERNEL)) ||
1970 	    !(sb = kmalloc(sizeof(struct cache_sb), GFP_KERNEL)))
1971 		goto err;
1972 
1973 	err = "failed to open device";
1974 	bdev = blkdev_get_by_path(strim(path),
1975 				  FMODE_READ|FMODE_WRITE|FMODE_EXCL,
1976 				  sb);
1977 	if (IS_ERR(bdev)) {
1978 		if (bdev == ERR_PTR(-EBUSY)) {
1979 			bdev = lookup_bdev(strim(path));
1980 			mutex_lock(&bch_register_lock);
1981 			if (!IS_ERR(bdev) && bch_is_open(bdev))
1982 				err = "device already registered";
1983 			else
1984 				err = "device busy";
1985 			mutex_unlock(&bch_register_lock);
1986 			if (!IS_ERR(bdev))
1987 				bdput(bdev);
1988 			if (attr == &ksysfs_register_quiet)
1989 				goto out;
1990 		}
1991 		goto err;
1992 	}
1993 
1994 	err = "failed to set blocksize";
1995 	if (set_blocksize(bdev, 4096))
1996 		goto err_close;
1997 
1998 	err = read_super(sb, bdev, &sb_page);
1999 	if (err)
2000 		goto err_close;
2001 
2002 	if (SB_IS_BDEV(sb)) {
2003 		struct cached_dev *dc = kzalloc(sizeof(*dc), GFP_KERNEL);
2004 		if (!dc)
2005 			goto err_close;
2006 
2007 		mutex_lock(&bch_register_lock);
2008 		register_bdev(sb, sb_page, bdev, dc);
2009 		mutex_unlock(&bch_register_lock);
2010 	} else {
2011 		struct cache *ca = kzalloc(sizeof(*ca), GFP_KERNEL);
2012 		if (!ca)
2013 			goto err_close;
2014 
2015 		if (register_cache(sb, sb_page, bdev, ca) != 0)
2016 			goto err_close;
2017 	}
2018 out:
2019 	if (sb_page)
2020 		put_page(sb_page);
2021 	kfree(sb);
2022 	kfree(path);
2023 	module_put(THIS_MODULE);
2024 	return ret;
2025 
2026 err_close:
2027 	blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2028 err:
2029 	pr_info("error opening %s: %s", path, err);
2030 	ret = -EINVAL;
2031 	goto out;
2032 }
2033 
2034 static int bcache_reboot(struct notifier_block *n, unsigned long code, void *x)
2035 {
2036 	if (code == SYS_DOWN ||
2037 	    code == SYS_HALT ||
2038 	    code == SYS_POWER_OFF) {
2039 		DEFINE_WAIT(wait);
2040 		unsigned long start = jiffies;
2041 		bool stopped = false;
2042 
2043 		struct cache_set *c, *tc;
2044 		struct cached_dev *dc, *tdc;
2045 
2046 		mutex_lock(&bch_register_lock);
2047 
2048 		if (list_empty(&bch_cache_sets) &&
2049 		    list_empty(&uncached_devices))
2050 			goto out;
2051 
2052 		pr_info("Stopping all devices:");
2053 
2054 		list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
2055 			bch_cache_set_stop(c);
2056 
2057 		list_for_each_entry_safe(dc, tdc, &uncached_devices, list)
2058 			bcache_device_stop(&dc->disk);
2059 
2060 		/* What's a condition variable? */
2061 		while (1) {
2062 			long timeout = start + 2 * HZ - jiffies;
2063 
2064 			stopped = list_empty(&bch_cache_sets) &&
2065 				list_empty(&uncached_devices);
2066 
2067 			if (timeout < 0 || stopped)
2068 				break;
2069 
2070 			prepare_to_wait(&unregister_wait, &wait,
2071 					TASK_UNINTERRUPTIBLE);
2072 
2073 			mutex_unlock(&bch_register_lock);
2074 			schedule_timeout(timeout);
2075 			mutex_lock(&bch_register_lock);
2076 		}
2077 
2078 		finish_wait(&unregister_wait, &wait);
2079 
2080 		if (stopped)
2081 			pr_info("All devices stopped");
2082 		else
2083 			pr_notice("Timeout waiting for devices to be closed");
2084 out:
2085 		mutex_unlock(&bch_register_lock);
2086 	}
2087 
2088 	return NOTIFY_DONE;
2089 }
2090 
2091 static struct notifier_block reboot = {
2092 	.notifier_call	= bcache_reboot,
2093 	.priority	= INT_MAX, /* before any real devices */
2094 };
2095 
2096 static void bcache_exit(void)
2097 {
2098 	bch_debug_exit();
2099 	bch_request_exit();
2100 	if (bcache_kobj)
2101 		kobject_put(bcache_kobj);
2102 	if (bcache_wq)
2103 		destroy_workqueue(bcache_wq);
2104 	if (bcache_major)
2105 		unregister_blkdev(bcache_major, "bcache");
2106 	unregister_reboot_notifier(&reboot);
2107 	mutex_destroy(&bch_register_lock);
2108 }
2109 
2110 static int __init bcache_init(void)
2111 {
2112 	static const struct attribute *files[] = {
2113 		&ksysfs_register.attr,
2114 		&ksysfs_register_quiet.attr,
2115 		NULL
2116 	};
2117 
2118 	mutex_init(&bch_register_lock);
2119 	init_waitqueue_head(&unregister_wait);
2120 	register_reboot_notifier(&reboot);
2121 	closure_debug_init();
2122 
2123 	bcache_major = register_blkdev(0, "bcache");
2124 	if (bcache_major < 0) {
2125 		unregister_reboot_notifier(&reboot);
2126 		mutex_destroy(&bch_register_lock);
2127 		return bcache_major;
2128 	}
2129 
2130 	if (!(bcache_wq = alloc_workqueue("bcache", WQ_MEM_RECLAIM, 0)) ||
2131 	    !(bcache_kobj = kobject_create_and_add("bcache", fs_kobj)) ||
2132 	    bch_request_init() ||
2133 	    bch_debug_init(bcache_kobj) ||
2134 	    sysfs_create_files(bcache_kobj, files))
2135 		goto err;
2136 
2137 	return 0;
2138 err:
2139 	bcache_exit();
2140 	return -ENOMEM;
2141 }
2142 
2143 module_exit(bcache_exit);
2144 module_init(bcache_init);
2145