xref: /linux/drivers/md/bcache/super.c (revision e58e871becec2d3b04ed91c0c16fe8deac9c9dfa)
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_minor);
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 #define BCACHE_MINORS		16 /* partition support */
62 
63 /* Superblock */
64 
65 static const char *read_super(struct cache_sb *sb, struct block_device *bdev,
66 			      struct page **res)
67 {
68 	const char *err;
69 	struct cache_sb *s;
70 	struct buffer_head *bh = __bread(bdev, 1, SB_SIZE);
71 	unsigned i;
72 
73 	if (!bh)
74 		return "IO error";
75 
76 	s = (struct cache_sb *) bh->b_data;
77 
78 	sb->offset		= le64_to_cpu(s->offset);
79 	sb->version		= le64_to_cpu(s->version);
80 
81 	memcpy(sb->magic,	s->magic, 16);
82 	memcpy(sb->uuid,	s->uuid, 16);
83 	memcpy(sb->set_uuid,	s->set_uuid, 16);
84 	memcpy(sb->label,	s->label, SB_LABEL_SIZE);
85 
86 	sb->flags		= le64_to_cpu(s->flags);
87 	sb->seq			= le64_to_cpu(s->seq);
88 	sb->last_mount		= le32_to_cpu(s->last_mount);
89 	sb->first_bucket	= le16_to_cpu(s->first_bucket);
90 	sb->keys		= le16_to_cpu(s->keys);
91 
92 	for (i = 0; i < SB_JOURNAL_BUCKETS; i++)
93 		sb->d[i] = le64_to_cpu(s->d[i]);
94 
95 	pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u",
96 		 sb->version, sb->flags, sb->seq, sb->keys);
97 
98 	err = "Not a bcache superblock";
99 	if (sb->offset != SB_SECTOR)
100 		goto err;
101 
102 	if (memcmp(sb->magic, bcache_magic, 16))
103 		goto err;
104 
105 	err = "Too many journal buckets";
106 	if (sb->keys > SB_JOURNAL_BUCKETS)
107 		goto err;
108 
109 	err = "Bad checksum";
110 	if (s->csum != csum_set(s))
111 		goto err;
112 
113 	err = "Bad UUID";
114 	if (bch_is_zero(sb->uuid, 16))
115 		goto err;
116 
117 	sb->block_size	= le16_to_cpu(s->block_size);
118 
119 	err = "Superblock block size smaller than device block size";
120 	if (sb->block_size << 9 < bdev_logical_block_size(bdev))
121 		goto err;
122 
123 	switch (sb->version) {
124 	case BCACHE_SB_VERSION_BDEV:
125 		sb->data_offset	= BDEV_DATA_START_DEFAULT;
126 		break;
127 	case BCACHE_SB_VERSION_BDEV_WITH_OFFSET:
128 		sb->data_offset	= le64_to_cpu(s->data_offset);
129 
130 		err = "Bad data offset";
131 		if (sb->data_offset < BDEV_DATA_START_DEFAULT)
132 			goto err;
133 
134 		break;
135 	case BCACHE_SB_VERSION_CDEV:
136 	case BCACHE_SB_VERSION_CDEV_WITH_UUID:
137 		sb->nbuckets	= le64_to_cpu(s->nbuckets);
138 		sb->bucket_size	= le16_to_cpu(s->bucket_size);
139 
140 		sb->nr_in_set	= le16_to_cpu(s->nr_in_set);
141 		sb->nr_this_dev	= le16_to_cpu(s->nr_this_dev);
142 
143 		err = "Too many buckets";
144 		if (sb->nbuckets > LONG_MAX)
145 			goto err;
146 
147 		err = "Not enough buckets";
148 		if (sb->nbuckets < 1 << 7)
149 			goto err;
150 
151 		err = "Bad block/bucket size";
152 		if (!is_power_of_2(sb->block_size) ||
153 		    sb->block_size > PAGE_SECTORS ||
154 		    !is_power_of_2(sb->bucket_size) ||
155 		    sb->bucket_size < PAGE_SECTORS)
156 			goto err;
157 
158 		err = "Invalid superblock: device too small";
159 		if (get_capacity(bdev->bd_disk) < sb->bucket_size * sb->nbuckets)
160 			goto err;
161 
162 		err = "Bad UUID";
163 		if (bch_is_zero(sb->set_uuid, 16))
164 			goto err;
165 
166 		err = "Bad cache device number in set";
167 		if (!sb->nr_in_set ||
168 		    sb->nr_in_set <= sb->nr_this_dev ||
169 		    sb->nr_in_set > MAX_CACHES_PER_SET)
170 			goto err;
171 
172 		err = "Journal buckets not sequential";
173 		for (i = 0; i < sb->keys; i++)
174 			if (sb->d[i] != sb->first_bucket + i)
175 				goto err;
176 
177 		err = "Too many journal buckets";
178 		if (sb->first_bucket + sb->keys > sb->nbuckets)
179 			goto err;
180 
181 		err = "Invalid superblock: first bucket comes before end of super";
182 		if (sb->first_bucket * sb->bucket_size < 16)
183 			goto err;
184 
185 		break;
186 	default:
187 		err = "Unsupported superblock version";
188 		goto err;
189 	}
190 
191 	sb->last_mount = get_seconds();
192 	err = NULL;
193 
194 	get_page(bh->b_page);
195 	*res = bh->b_page;
196 err:
197 	put_bh(bh);
198 	return err;
199 }
200 
201 static void write_bdev_super_endio(struct bio *bio)
202 {
203 	struct cached_dev *dc = bio->bi_private;
204 	/* XXX: error checking */
205 
206 	closure_put(&dc->sb_write);
207 }
208 
209 static void __write_super(struct cache_sb *sb, struct bio *bio)
210 {
211 	struct cache_sb *out = page_address(bio->bi_io_vec[0].bv_page);
212 	unsigned i;
213 
214 	bio->bi_iter.bi_sector	= SB_SECTOR;
215 	bio->bi_iter.bi_size	= SB_SIZE;
216 	bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
217 	bch_bio_map(bio, NULL);
218 
219 	out->offset		= cpu_to_le64(sb->offset);
220 	out->version		= cpu_to_le64(sb->version);
221 
222 	memcpy(out->uuid,	sb->uuid, 16);
223 	memcpy(out->set_uuid,	sb->set_uuid, 16);
224 	memcpy(out->label,	sb->label, SB_LABEL_SIZE);
225 
226 	out->flags		= cpu_to_le64(sb->flags);
227 	out->seq		= cpu_to_le64(sb->seq);
228 
229 	out->last_mount		= cpu_to_le32(sb->last_mount);
230 	out->first_bucket	= cpu_to_le16(sb->first_bucket);
231 	out->keys		= cpu_to_le16(sb->keys);
232 
233 	for (i = 0; i < sb->keys; i++)
234 		out->d[i] = cpu_to_le64(sb->d[i]);
235 
236 	out->csum = csum_set(out);
237 
238 	pr_debug("ver %llu, flags %llu, seq %llu",
239 		 sb->version, sb->flags, sb->seq);
240 
241 	submit_bio(bio);
242 }
243 
244 static void bch_write_bdev_super_unlock(struct closure *cl)
245 {
246 	struct cached_dev *dc = container_of(cl, struct cached_dev, sb_write);
247 
248 	up(&dc->sb_write_mutex);
249 }
250 
251 void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent)
252 {
253 	struct closure *cl = &dc->sb_write;
254 	struct bio *bio = &dc->sb_bio;
255 
256 	down(&dc->sb_write_mutex);
257 	closure_init(cl, parent);
258 
259 	bio_reset(bio);
260 	bio->bi_bdev	= dc->bdev;
261 	bio->bi_end_io	= write_bdev_super_endio;
262 	bio->bi_private = dc;
263 
264 	closure_get(cl);
265 	__write_super(&dc->sb, bio);
266 
267 	closure_return_with_destructor(cl, bch_write_bdev_super_unlock);
268 }
269 
270 static void write_super_endio(struct bio *bio)
271 {
272 	struct cache *ca = bio->bi_private;
273 
274 	bch_count_io_errors(ca, bio->bi_error, "writing superblock");
275 	closure_put(&ca->set->sb_write);
276 }
277 
278 static void bcache_write_super_unlock(struct closure *cl)
279 {
280 	struct cache_set *c = container_of(cl, struct cache_set, sb_write);
281 
282 	up(&c->sb_write_mutex);
283 }
284 
285 void bcache_write_super(struct cache_set *c)
286 {
287 	struct closure *cl = &c->sb_write;
288 	struct cache *ca;
289 	unsigned i;
290 
291 	down(&c->sb_write_mutex);
292 	closure_init(cl, &c->cl);
293 
294 	c->sb.seq++;
295 
296 	for_each_cache(ca, c, i) {
297 		struct bio *bio = &ca->sb_bio;
298 
299 		ca->sb.version		= BCACHE_SB_VERSION_CDEV_WITH_UUID;
300 		ca->sb.seq		= c->sb.seq;
301 		ca->sb.last_mount	= c->sb.last_mount;
302 
303 		SET_CACHE_SYNC(&ca->sb, CACHE_SYNC(&c->sb));
304 
305 		bio_reset(bio);
306 		bio->bi_bdev	= ca->bdev;
307 		bio->bi_end_io	= write_super_endio;
308 		bio->bi_private = ca;
309 
310 		closure_get(cl);
311 		__write_super(&ca->sb, bio);
312 	}
313 
314 	closure_return_with_destructor(cl, bcache_write_super_unlock);
315 }
316 
317 /* UUID io */
318 
319 static void uuid_endio(struct bio *bio)
320 {
321 	struct closure *cl = bio->bi_private;
322 	struct cache_set *c = container_of(cl, struct cache_set, uuid_write);
323 
324 	cache_set_err_on(bio->bi_error, c, "accessing uuids");
325 	bch_bbio_free(bio, c);
326 	closure_put(cl);
327 }
328 
329 static void uuid_io_unlock(struct closure *cl)
330 {
331 	struct cache_set *c = container_of(cl, struct cache_set, uuid_write);
332 
333 	up(&c->uuid_write_mutex);
334 }
335 
336 static void uuid_io(struct cache_set *c, int op, unsigned long op_flags,
337 		    struct bkey *k, struct closure *parent)
338 {
339 	struct closure *cl = &c->uuid_write;
340 	struct uuid_entry *u;
341 	unsigned i;
342 	char buf[80];
343 
344 	BUG_ON(!parent);
345 	down(&c->uuid_write_mutex);
346 	closure_init(cl, parent);
347 
348 	for (i = 0; i < KEY_PTRS(k); i++) {
349 		struct bio *bio = bch_bbio_alloc(c);
350 
351 		bio->bi_opf = REQ_SYNC | REQ_META | op_flags;
352 		bio->bi_iter.bi_size = KEY_SIZE(k) << 9;
353 
354 		bio->bi_end_io	= uuid_endio;
355 		bio->bi_private = cl;
356 		bio_set_op_attrs(bio, op, REQ_SYNC|REQ_META|op_flags);
357 		bch_bio_map(bio, c->uuids);
358 
359 		bch_submit_bbio(bio, c, k, i);
360 
361 		if (op != REQ_OP_WRITE)
362 			break;
363 	}
364 
365 	bch_extent_to_text(buf, sizeof(buf), k);
366 	pr_debug("%s UUIDs at %s", op == REQ_OP_WRITE ? "wrote" : "read", buf);
367 
368 	for (u = c->uuids; u < c->uuids + c->nr_uuids; u++)
369 		if (!bch_is_zero(u->uuid, 16))
370 			pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u",
371 				 u - c->uuids, u->uuid, u->label,
372 				 u->first_reg, u->last_reg, u->invalidated);
373 
374 	closure_return_with_destructor(cl, uuid_io_unlock);
375 }
376 
377 static char *uuid_read(struct cache_set *c, struct jset *j, struct closure *cl)
378 {
379 	struct bkey *k = &j->uuid_bucket;
380 
381 	if (__bch_btree_ptr_invalid(c, k))
382 		return "bad uuid pointer";
383 
384 	bkey_copy(&c->uuid_bucket, k);
385 	uuid_io(c, REQ_OP_READ, 0, k, cl);
386 
387 	if (j->version < BCACHE_JSET_VERSION_UUIDv1) {
388 		struct uuid_entry_v0	*u0 = (void *) c->uuids;
389 		struct uuid_entry	*u1 = (void *) c->uuids;
390 		int i;
391 
392 		closure_sync(cl);
393 
394 		/*
395 		 * Since the new uuid entry is bigger than the old, we have to
396 		 * convert starting at the highest memory address and work down
397 		 * in order to do it in place
398 		 */
399 
400 		for (i = c->nr_uuids - 1;
401 		     i >= 0;
402 		     --i) {
403 			memcpy(u1[i].uuid,	u0[i].uuid, 16);
404 			memcpy(u1[i].label,	u0[i].label, 32);
405 
406 			u1[i].first_reg		= u0[i].first_reg;
407 			u1[i].last_reg		= u0[i].last_reg;
408 			u1[i].invalidated	= u0[i].invalidated;
409 
410 			u1[i].flags	= 0;
411 			u1[i].sectors	= 0;
412 		}
413 	}
414 
415 	return NULL;
416 }
417 
418 static int __uuid_write(struct cache_set *c)
419 {
420 	BKEY_PADDED(key) k;
421 	struct closure cl;
422 	closure_init_stack(&cl);
423 
424 	lockdep_assert_held(&bch_register_lock);
425 
426 	if (bch_bucket_alloc_set(c, RESERVE_BTREE, &k.key, 1, true))
427 		return 1;
428 
429 	SET_KEY_SIZE(&k.key, c->sb.bucket_size);
430 	uuid_io(c, REQ_OP_WRITE, 0, &k.key, &cl);
431 	closure_sync(&cl);
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_error, 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->bi_bdev		= 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(bio, &ca->prio);
520 	closure_sync(cl);
521 }
522 
523 void bch_prio_write(struct cache *ca)
524 {
525 	int i;
526 	struct bucket *b;
527 	struct closure cl;
528 
529 	closure_init_stack(&cl);
530 
531 	lockdep_assert_held(&ca->set->bucket_lock);
532 
533 	ca->disk_buckets->seq++;
534 
535 	atomic_long_add(ca->sb.bucket_size * prio_buckets(ca),
536 			&ca->meta_sectors_written);
537 
538 	//pr_debug("free %zu, free_inc %zu, unused %zu", fifo_used(&ca->free),
539 	//	 fifo_used(&ca->free_inc), fifo_used(&ca->unused));
540 
541 	for (i = prio_buckets(ca) - 1; i >= 0; --i) {
542 		long bucket;
543 		struct prio_set *p = ca->disk_buckets;
544 		struct bucket_disk *d = p->data;
545 		struct bucket_disk *end = d + prios_per_bucket(ca);
546 
547 		for (b = ca->buckets + i * prios_per_bucket(ca);
548 		     b < ca->buckets + ca->sb.nbuckets && d < end;
549 		     b++, d++) {
550 			d->prio = cpu_to_le16(b->prio);
551 			d->gen = b->gen;
552 		}
553 
554 		p->next_bucket	= ca->prio_buckets[i + 1];
555 		p->magic	= pset_magic(&ca->sb);
556 		p->csum		= bch_crc64(&p->magic, bucket_bytes(ca) - 8);
557 
558 		bucket = bch_bucket_alloc(ca, RESERVE_PRIO, true);
559 		BUG_ON(bucket == -1);
560 
561 		mutex_unlock(&ca->set->bucket_lock);
562 		prio_io(ca, bucket, REQ_OP_WRITE, 0);
563 		mutex_lock(&ca->set->bucket_lock);
564 
565 		ca->prio_buckets[i] = bucket;
566 		atomic_dec_bug(&ca->buckets[bucket].pin);
567 	}
568 
569 	mutex_unlock(&ca->set->bucket_lock);
570 
571 	bch_journal_meta(ca->set, &cl);
572 	closure_sync(&cl);
573 
574 	mutex_lock(&ca->set->bucket_lock);
575 
576 	/*
577 	 * Don't want the old priorities to get garbage collected until after we
578 	 * finish writing the new ones, and they're journalled
579 	 */
580 	for (i = 0; i < prio_buckets(ca); i++) {
581 		if (ca->prio_last_buckets[i])
582 			__bch_bucket_free(ca,
583 				&ca->buckets[ca->prio_last_buckets[i]]);
584 
585 		ca->prio_last_buckets[i] = ca->prio_buckets[i];
586 	}
587 }
588 
589 static void prio_read(struct cache *ca, uint64_t bucket)
590 {
591 	struct prio_set *p = ca->disk_buckets;
592 	struct bucket_disk *d = p->data + prios_per_bucket(ca), *end = d;
593 	struct bucket *b;
594 	unsigned bucket_nr = 0;
595 
596 	for (b = ca->buckets;
597 	     b < ca->buckets + ca->sb.nbuckets;
598 	     b++, d++) {
599 		if (d == end) {
600 			ca->prio_buckets[bucket_nr] = bucket;
601 			ca->prio_last_buckets[bucket_nr] = bucket;
602 			bucket_nr++;
603 
604 			prio_io(ca, bucket, REQ_OP_READ, 0);
605 
606 			if (p->csum != bch_crc64(&p->magic, bucket_bytes(ca) - 8))
607 				pr_warn("bad csum reading priorities");
608 
609 			if (p->magic != pset_magic(&ca->sb))
610 				pr_warn("bad magic reading priorities");
611 
612 			bucket = p->next_bucket;
613 			d = p->data;
614 		}
615 
616 		b->prio = le16_to_cpu(d->prio);
617 		b->gen = b->last_gc = d->gen;
618 	}
619 }
620 
621 /* Bcache device */
622 
623 static int open_dev(struct block_device *b, fmode_t mode)
624 {
625 	struct bcache_device *d = b->bd_disk->private_data;
626 	if (test_bit(BCACHE_DEV_CLOSING, &d->flags))
627 		return -ENXIO;
628 
629 	closure_get(&d->cl);
630 	return 0;
631 }
632 
633 static void release_dev(struct gendisk *b, fmode_t mode)
634 {
635 	struct bcache_device *d = b->private_data;
636 	closure_put(&d->cl);
637 }
638 
639 static int ioctl_dev(struct block_device *b, fmode_t mode,
640 		     unsigned int cmd, unsigned long arg)
641 {
642 	struct bcache_device *d = b->bd_disk->private_data;
643 	return d->ioctl(d, mode, cmd, arg);
644 }
645 
646 static const struct block_device_operations bcache_ops = {
647 	.open		= open_dev,
648 	.release	= release_dev,
649 	.ioctl		= ioctl_dev,
650 	.owner		= THIS_MODULE,
651 };
652 
653 void bcache_device_stop(struct bcache_device *d)
654 {
655 	if (!test_and_set_bit(BCACHE_DEV_CLOSING, &d->flags))
656 		closure_queue(&d->cl);
657 }
658 
659 static void bcache_device_unlink(struct bcache_device *d)
660 {
661 	lockdep_assert_held(&bch_register_lock);
662 
663 	if (d->c && !test_and_set_bit(BCACHE_DEV_UNLINK_DONE, &d->flags)) {
664 		unsigned i;
665 		struct cache *ca;
666 
667 		sysfs_remove_link(&d->c->kobj, d->name);
668 		sysfs_remove_link(&d->kobj, "cache");
669 
670 		for_each_cache(ca, d->c, i)
671 			bd_unlink_disk_holder(ca->bdev, d->disk);
672 	}
673 }
674 
675 static void bcache_device_link(struct bcache_device *d, struct cache_set *c,
676 			       const char *name)
677 {
678 	unsigned i;
679 	struct cache *ca;
680 
681 	for_each_cache(ca, d->c, i)
682 		bd_link_disk_holder(ca->bdev, d->disk);
683 
684 	snprintf(d->name, BCACHEDEVNAME_SIZE,
685 		 "%s%u", name, d->id);
686 
687 	WARN(sysfs_create_link(&d->kobj, &c->kobj, "cache") ||
688 	     sysfs_create_link(&c->kobj, &d->kobj, d->name),
689 	     "Couldn't create device <-> cache set symlinks");
690 
691 	clear_bit(BCACHE_DEV_UNLINK_DONE, &d->flags);
692 }
693 
694 static void bcache_device_detach(struct bcache_device *d)
695 {
696 	lockdep_assert_held(&bch_register_lock);
697 
698 	if (test_bit(BCACHE_DEV_DETACHING, &d->flags)) {
699 		struct uuid_entry *u = d->c->uuids + d->id;
700 
701 		SET_UUID_FLASH_ONLY(u, 0);
702 		memcpy(u->uuid, invalid_uuid, 16);
703 		u->invalidated = cpu_to_le32(get_seconds());
704 		bch_uuid_write(d->c);
705 	}
706 
707 	bcache_device_unlink(d);
708 
709 	d->c->devices[d->id] = NULL;
710 	closure_put(&d->c->caching);
711 	d->c = NULL;
712 }
713 
714 static void bcache_device_attach(struct bcache_device *d, struct cache_set *c,
715 				 unsigned id)
716 {
717 	d->id = id;
718 	d->c = c;
719 	c->devices[id] = d;
720 
721 	closure_get(&c->caching);
722 }
723 
724 static void bcache_device_free(struct bcache_device *d)
725 {
726 	lockdep_assert_held(&bch_register_lock);
727 
728 	pr_info("%s stopped", d->disk->disk_name);
729 
730 	if (d->c)
731 		bcache_device_detach(d);
732 	if (d->disk && d->disk->flags & GENHD_FL_UP)
733 		del_gendisk(d->disk);
734 	if (d->disk && d->disk->queue)
735 		blk_cleanup_queue(d->disk->queue);
736 	if (d->disk) {
737 		ida_simple_remove(&bcache_minor, d->disk->first_minor);
738 		put_disk(d->disk);
739 	}
740 
741 	if (d->bio_split)
742 		bioset_free(d->bio_split);
743 	kvfree(d->full_dirty_stripes);
744 	kvfree(d->stripe_sectors_dirty);
745 
746 	closure_debug_destroy(&d->cl);
747 }
748 
749 static int bcache_device_init(struct bcache_device *d, unsigned block_size,
750 			      sector_t sectors)
751 {
752 	struct request_queue *q;
753 	size_t n;
754 	int minor;
755 
756 	if (!d->stripe_size)
757 		d->stripe_size = 1 << 31;
758 
759 	d->nr_stripes = DIV_ROUND_UP_ULL(sectors, d->stripe_size);
760 
761 	if (!d->nr_stripes ||
762 	    d->nr_stripes > INT_MAX ||
763 	    d->nr_stripes > SIZE_MAX / sizeof(atomic_t)) {
764 		pr_err("nr_stripes too large or invalid: %u (start sector beyond end of disk?)",
765 			(unsigned)d->nr_stripes);
766 		return -ENOMEM;
767 	}
768 
769 	n = d->nr_stripes * sizeof(atomic_t);
770 	d->stripe_sectors_dirty = kvzalloc(n, GFP_KERNEL);
771 	if (!d->stripe_sectors_dirty)
772 		return -ENOMEM;
773 
774 	n = BITS_TO_LONGS(d->nr_stripes) * sizeof(unsigned long);
775 	d->full_dirty_stripes = kvzalloc(n, GFP_KERNEL);
776 	if (!d->full_dirty_stripes)
777 		return -ENOMEM;
778 
779 	minor = ida_simple_get(&bcache_minor, 0, MINORMASK + 1, GFP_KERNEL);
780 	if (minor < 0)
781 		return minor;
782 
783 	minor *= BCACHE_MINORS;
784 
785 	if (!(d->bio_split = bioset_create(4, offsetof(struct bbio, bio))) ||
786 	    !(d->disk = alloc_disk(BCACHE_MINORS))) {
787 		ida_simple_remove(&bcache_minor, minor);
788 		return -ENOMEM;
789 	}
790 
791 	set_capacity(d->disk, sectors);
792 	snprintf(d->disk->disk_name, DISK_NAME_LEN, "bcache%i", minor);
793 
794 	d->disk->major		= bcache_major;
795 	d->disk->first_minor	= minor;
796 	d->disk->fops		= &bcache_ops;
797 	d->disk->private_data	= d;
798 
799 	q = blk_alloc_queue(GFP_KERNEL);
800 	if (!q)
801 		return -ENOMEM;
802 
803 	blk_queue_make_request(q, NULL);
804 	d->disk->queue			= q;
805 	q->queuedata			= d;
806 	q->backing_dev_info->congested_data = d;
807 	q->limits.max_hw_sectors	= UINT_MAX;
808 	q->limits.max_sectors		= UINT_MAX;
809 	q->limits.max_segment_size	= UINT_MAX;
810 	q->limits.max_segments		= BIO_MAX_PAGES;
811 	blk_queue_max_discard_sectors(q, UINT_MAX);
812 	q->limits.discard_granularity	= 512;
813 	q->limits.io_min		= block_size;
814 	q->limits.logical_block_size	= block_size;
815 	q->limits.physical_block_size	= block_size;
816 	set_bit(QUEUE_FLAG_NONROT,	&d->disk->queue->queue_flags);
817 	clear_bit(QUEUE_FLAG_ADD_RANDOM, &d->disk->queue->queue_flags);
818 	set_bit(QUEUE_FLAG_DISCARD,	&d->disk->queue->queue_flags);
819 
820 	blk_queue_write_cache(q, true, true);
821 
822 	return 0;
823 }
824 
825 /* Cached device */
826 
827 static void calc_cached_dev_sectors(struct cache_set *c)
828 {
829 	uint64_t sectors = 0;
830 	struct cached_dev *dc;
831 
832 	list_for_each_entry(dc, &c->cached_devs, list)
833 		sectors += bdev_sectors(dc->bdev);
834 
835 	c->cached_dev_sectors = sectors;
836 }
837 
838 void bch_cached_dev_run(struct cached_dev *dc)
839 {
840 	struct bcache_device *d = &dc->disk;
841 	char buf[SB_LABEL_SIZE + 1];
842 	char *env[] = {
843 		"DRIVER=bcache",
844 		kasprintf(GFP_KERNEL, "CACHED_UUID=%pU", dc->sb.uuid),
845 		NULL,
846 		NULL,
847 	};
848 
849 	memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
850 	buf[SB_LABEL_SIZE] = '\0';
851 	env[2] = kasprintf(GFP_KERNEL, "CACHED_LABEL=%s", buf);
852 
853 	if (atomic_xchg(&dc->running, 1)) {
854 		kfree(env[1]);
855 		kfree(env[2]);
856 		return;
857 	}
858 
859 	if (!d->c &&
860 	    BDEV_STATE(&dc->sb) != BDEV_STATE_NONE) {
861 		struct closure cl;
862 		closure_init_stack(&cl);
863 
864 		SET_BDEV_STATE(&dc->sb, BDEV_STATE_STALE);
865 		bch_write_bdev_super(dc, &cl);
866 		closure_sync(&cl);
867 	}
868 
869 	add_disk(d->disk);
870 	bd_link_disk_holder(dc->bdev, dc->disk.disk);
871 	/* won't show up in the uevent file, use udevadm monitor -e instead
872 	 * only class / kset properties are persistent */
873 	kobject_uevent_env(&disk_to_dev(d->disk)->kobj, KOBJ_CHANGE, env);
874 	kfree(env[1]);
875 	kfree(env[2]);
876 
877 	if (sysfs_create_link(&d->kobj, &disk_to_dev(d->disk)->kobj, "dev") ||
878 	    sysfs_create_link(&disk_to_dev(d->disk)->kobj, &d->kobj, "bcache"))
879 		pr_debug("error creating sysfs link");
880 }
881 
882 static void cached_dev_detach_finish(struct work_struct *w)
883 {
884 	struct cached_dev *dc = container_of(w, struct cached_dev, detach);
885 	char buf[BDEVNAME_SIZE];
886 	struct closure cl;
887 	closure_init_stack(&cl);
888 
889 	BUG_ON(!test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags));
890 	BUG_ON(atomic_read(&dc->count));
891 
892 	mutex_lock(&bch_register_lock);
893 
894 	memset(&dc->sb.set_uuid, 0, 16);
895 	SET_BDEV_STATE(&dc->sb, BDEV_STATE_NONE);
896 
897 	bch_write_bdev_super(dc, &cl);
898 	closure_sync(&cl);
899 
900 	bcache_device_detach(&dc->disk);
901 	list_move(&dc->list, &uncached_devices);
902 
903 	clear_bit(BCACHE_DEV_DETACHING, &dc->disk.flags);
904 	clear_bit(BCACHE_DEV_UNLINK_DONE, &dc->disk.flags);
905 
906 	mutex_unlock(&bch_register_lock);
907 
908 	pr_info("Caching disabled for %s", bdevname(dc->bdev, buf));
909 
910 	/* Drop ref we took in cached_dev_detach() */
911 	closure_put(&dc->disk.cl);
912 }
913 
914 void bch_cached_dev_detach(struct cached_dev *dc)
915 {
916 	lockdep_assert_held(&bch_register_lock);
917 
918 	if (test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags))
919 		return;
920 
921 	if (test_and_set_bit(BCACHE_DEV_DETACHING, &dc->disk.flags))
922 		return;
923 
924 	/*
925 	 * Block the device from being closed and freed until we're finished
926 	 * detaching
927 	 */
928 	closure_get(&dc->disk.cl);
929 
930 	bch_writeback_queue(dc);
931 	cached_dev_put(dc);
932 }
933 
934 int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c)
935 {
936 	uint32_t rtime = cpu_to_le32(get_seconds());
937 	struct uuid_entry *u;
938 	char buf[BDEVNAME_SIZE];
939 
940 	bdevname(dc->bdev, buf);
941 
942 	if (memcmp(dc->sb.set_uuid, c->sb.set_uuid, 16))
943 		return -ENOENT;
944 
945 	if (dc->disk.c) {
946 		pr_err("Can't attach %s: already attached", buf);
947 		return -EINVAL;
948 	}
949 
950 	if (test_bit(CACHE_SET_STOPPING, &c->flags)) {
951 		pr_err("Can't attach %s: shutting down", buf);
952 		return -EINVAL;
953 	}
954 
955 	if (dc->sb.block_size < c->sb.block_size) {
956 		/* Will die */
957 		pr_err("Couldn't attach %s: block size less than set's block size",
958 		       buf);
959 		return -EINVAL;
960 	}
961 
962 	u = uuid_find(c, dc->sb.uuid);
963 
964 	if (u &&
965 	    (BDEV_STATE(&dc->sb) == BDEV_STATE_STALE ||
966 	     BDEV_STATE(&dc->sb) == BDEV_STATE_NONE)) {
967 		memcpy(u->uuid, invalid_uuid, 16);
968 		u->invalidated = cpu_to_le32(get_seconds());
969 		u = NULL;
970 	}
971 
972 	if (!u) {
973 		if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
974 			pr_err("Couldn't find uuid for %s in set", buf);
975 			return -ENOENT;
976 		}
977 
978 		u = uuid_find_empty(c);
979 		if (!u) {
980 			pr_err("Not caching %s, no room for UUID", buf);
981 			return -EINVAL;
982 		}
983 	}
984 
985 	/* Deadlocks since we're called via sysfs...
986 	sysfs_remove_file(&dc->kobj, &sysfs_attach);
987 	 */
988 
989 	if (bch_is_zero(u->uuid, 16)) {
990 		struct closure cl;
991 		closure_init_stack(&cl);
992 
993 		memcpy(u->uuid, dc->sb.uuid, 16);
994 		memcpy(u->label, dc->sb.label, SB_LABEL_SIZE);
995 		u->first_reg = u->last_reg = rtime;
996 		bch_uuid_write(c);
997 
998 		memcpy(dc->sb.set_uuid, c->sb.set_uuid, 16);
999 		SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN);
1000 
1001 		bch_write_bdev_super(dc, &cl);
1002 		closure_sync(&cl);
1003 	} else {
1004 		u->last_reg = rtime;
1005 		bch_uuid_write(c);
1006 	}
1007 
1008 	bcache_device_attach(&dc->disk, c, u - c->uuids);
1009 	list_move(&dc->list, &c->cached_devs);
1010 	calc_cached_dev_sectors(c);
1011 
1012 	smp_wmb();
1013 	/*
1014 	 * dc->c must be set before dc->count != 0 - paired with the mb in
1015 	 * cached_dev_get()
1016 	 */
1017 	atomic_set(&dc->count, 1);
1018 
1019 	/* Block writeback thread, but spawn it */
1020 	down_write(&dc->writeback_lock);
1021 	if (bch_cached_dev_writeback_start(dc)) {
1022 		up_write(&dc->writeback_lock);
1023 		return -ENOMEM;
1024 	}
1025 
1026 	if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
1027 		bch_sectors_dirty_init(dc);
1028 		atomic_set(&dc->has_dirty, 1);
1029 		atomic_inc(&dc->count);
1030 		bch_writeback_queue(dc);
1031 	}
1032 
1033 	bch_cached_dev_run(dc);
1034 	bcache_device_link(&dc->disk, c, "bdev");
1035 
1036 	/* Allow the writeback thread to proceed */
1037 	up_write(&dc->writeback_lock);
1038 
1039 	pr_info("Caching %s as %s on set %pU",
1040 		bdevname(dc->bdev, buf), dc->disk.disk->disk_name,
1041 		dc->disk.c->sb.set_uuid);
1042 	return 0;
1043 }
1044 
1045 void bch_cached_dev_release(struct kobject *kobj)
1046 {
1047 	struct cached_dev *dc = container_of(kobj, struct cached_dev,
1048 					     disk.kobj);
1049 	kfree(dc);
1050 	module_put(THIS_MODULE);
1051 }
1052 
1053 static void cached_dev_free(struct closure *cl)
1054 {
1055 	struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1056 
1057 	cancel_delayed_work_sync(&dc->writeback_rate_update);
1058 	if (!IS_ERR_OR_NULL(dc->writeback_thread))
1059 		kthread_stop(dc->writeback_thread);
1060 
1061 	mutex_lock(&bch_register_lock);
1062 
1063 	if (atomic_read(&dc->running))
1064 		bd_unlink_disk_holder(dc->bdev, dc->disk.disk);
1065 	bcache_device_free(&dc->disk);
1066 	list_del(&dc->list);
1067 
1068 	mutex_unlock(&bch_register_lock);
1069 
1070 	if (!IS_ERR_OR_NULL(dc->bdev))
1071 		blkdev_put(dc->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1072 
1073 	wake_up(&unregister_wait);
1074 
1075 	kobject_put(&dc->disk.kobj);
1076 }
1077 
1078 static void cached_dev_flush(struct closure *cl)
1079 {
1080 	struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1081 	struct bcache_device *d = &dc->disk;
1082 
1083 	mutex_lock(&bch_register_lock);
1084 	bcache_device_unlink(d);
1085 	mutex_unlock(&bch_register_lock);
1086 
1087 	bch_cache_accounting_destroy(&dc->accounting);
1088 	kobject_del(&d->kobj);
1089 
1090 	continue_at(cl, cached_dev_free, system_wq);
1091 }
1092 
1093 static int cached_dev_init(struct cached_dev *dc, unsigned block_size)
1094 {
1095 	int ret;
1096 	struct io *io;
1097 	struct request_queue *q = bdev_get_queue(dc->bdev);
1098 
1099 	__module_get(THIS_MODULE);
1100 	INIT_LIST_HEAD(&dc->list);
1101 	closure_init(&dc->disk.cl, NULL);
1102 	set_closure_fn(&dc->disk.cl, cached_dev_flush, system_wq);
1103 	kobject_init(&dc->disk.kobj, &bch_cached_dev_ktype);
1104 	INIT_WORK(&dc->detach, cached_dev_detach_finish);
1105 	sema_init(&dc->sb_write_mutex, 1);
1106 	INIT_LIST_HEAD(&dc->io_lru);
1107 	spin_lock_init(&dc->io_lock);
1108 	bch_cache_accounting_init(&dc->accounting, &dc->disk.cl);
1109 
1110 	dc->sequential_cutoff		= 4 << 20;
1111 
1112 	for (io = dc->io; io < dc->io + RECENT_IO; io++) {
1113 		list_add(&io->lru, &dc->io_lru);
1114 		hlist_add_head(&io->hash, dc->io_hash + RECENT_IO);
1115 	}
1116 
1117 	dc->disk.stripe_size = q->limits.io_opt >> 9;
1118 
1119 	if (dc->disk.stripe_size)
1120 		dc->partial_stripes_expensive =
1121 			q->limits.raid_partial_stripes_expensive;
1122 
1123 	ret = bcache_device_init(&dc->disk, block_size,
1124 			 dc->bdev->bd_part->nr_sects - dc->sb.data_offset);
1125 	if (ret)
1126 		return ret;
1127 
1128 	set_capacity(dc->disk.disk,
1129 		     dc->bdev->bd_part->nr_sects - dc->sb.data_offset);
1130 
1131 	dc->disk.disk->queue->backing_dev_info->ra_pages =
1132 		max(dc->disk.disk->queue->backing_dev_info->ra_pages,
1133 		    q->backing_dev_info->ra_pages);
1134 
1135 	bch_cached_dev_request_init(dc);
1136 	bch_cached_dev_writeback_init(dc);
1137 	return 0;
1138 }
1139 
1140 /* Cached device - bcache superblock */
1141 
1142 static void register_bdev(struct cache_sb *sb, struct page *sb_page,
1143 				 struct block_device *bdev,
1144 				 struct cached_dev *dc)
1145 {
1146 	char name[BDEVNAME_SIZE];
1147 	const char *err = "cannot allocate memory";
1148 	struct cache_set *c;
1149 
1150 	memcpy(&dc->sb, sb, sizeof(struct cache_sb));
1151 	dc->bdev = bdev;
1152 	dc->bdev->bd_holder = dc;
1153 
1154 	bio_init(&dc->sb_bio, dc->sb_bio.bi_inline_vecs, 1);
1155 	dc->sb_bio.bi_io_vec[0].bv_page = sb_page;
1156 	get_page(sb_page);
1157 
1158 	if (cached_dev_init(dc, sb->block_size << 9))
1159 		goto err;
1160 
1161 	err = "error creating kobject";
1162 	if (kobject_add(&dc->disk.kobj, &part_to_dev(bdev->bd_part)->kobj,
1163 			"bcache"))
1164 		goto err;
1165 	if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj))
1166 		goto err;
1167 
1168 	pr_info("registered backing device %s", bdevname(bdev, name));
1169 
1170 	list_add(&dc->list, &uncached_devices);
1171 	list_for_each_entry(c, &bch_cache_sets, list)
1172 		bch_cached_dev_attach(dc, c);
1173 
1174 	if (BDEV_STATE(&dc->sb) == BDEV_STATE_NONE ||
1175 	    BDEV_STATE(&dc->sb) == BDEV_STATE_STALE)
1176 		bch_cached_dev_run(dc);
1177 
1178 	return;
1179 err:
1180 	pr_notice("error opening %s: %s", bdevname(bdev, name), err);
1181 	bcache_device_stop(&dc->disk);
1182 }
1183 
1184 /* Flash only volumes */
1185 
1186 void bch_flash_dev_release(struct kobject *kobj)
1187 {
1188 	struct bcache_device *d = container_of(kobj, struct bcache_device,
1189 					       kobj);
1190 	kfree(d);
1191 }
1192 
1193 static void flash_dev_free(struct closure *cl)
1194 {
1195 	struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1196 	mutex_lock(&bch_register_lock);
1197 	bcache_device_free(d);
1198 	mutex_unlock(&bch_register_lock);
1199 	kobject_put(&d->kobj);
1200 }
1201 
1202 static void flash_dev_flush(struct closure *cl)
1203 {
1204 	struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1205 
1206 	mutex_lock(&bch_register_lock);
1207 	bcache_device_unlink(d);
1208 	mutex_unlock(&bch_register_lock);
1209 	kobject_del(&d->kobj);
1210 	continue_at(cl, flash_dev_free, system_wq);
1211 }
1212 
1213 static int flash_dev_run(struct cache_set *c, struct uuid_entry *u)
1214 {
1215 	struct bcache_device *d = kzalloc(sizeof(struct bcache_device),
1216 					  GFP_KERNEL);
1217 	if (!d)
1218 		return -ENOMEM;
1219 
1220 	closure_init(&d->cl, NULL);
1221 	set_closure_fn(&d->cl, flash_dev_flush, system_wq);
1222 
1223 	kobject_init(&d->kobj, &bch_flash_dev_ktype);
1224 
1225 	if (bcache_device_init(d, block_bytes(c), u->sectors))
1226 		goto err;
1227 
1228 	bcache_device_attach(d, c, u - c->uuids);
1229 	bch_flash_dev_request_init(d);
1230 	add_disk(d->disk);
1231 
1232 	if (kobject_add(&d->kobj, &disk_to_dev(d->disk)->kobj, "bcache"))
1233 		goto err;
1234 
1235 	bcache_device_link(d, c, "volume");
1236 
1237 	return 0;
1238 err:
1239 	kobject_put(&d->kobj);
1240 	return -ENOMEM;
1241 }
1242 
1243 static int flash_devs_run(struct cache_set *c)
1244 {
1245 	int ret = 0;
1246 	struct uuid_entry *u;
1247 
1248 	for (u = c->uuids;
1249 	     u < c->uuids + c->nr_uuids && !ret;
1250 	     u++)
1251 		if (UUID_FLASH_ONLY(u))
1252 			ret = flash_dev_run(c, u);
1253 
1254 	return ret;
1255 }
1256 
1257 int bch_flash_dev_create(struct cache_set *c, uint64_t size)
1258 {
1259 	struct uuid_entry *u;
1260 
1261 	if (test_bit(CACHE_SET_STOPPING, &c->flags))
1262 		return -EINTR;
1263 
1264 	if (!test_bit(CACHE_SET_RUNNING, &c->flags))
1265 		return -EPERM;
1266 
1267 	u = uuid_find_empty(c);
1268 	if (!u) {
1269 		pr_err("Can't create volume, no room for UUID");
1270 		return -EINVAL;
1271 	}
1272 
1273 	get_random_bytes(u->uuid, 16);
1274 	memset(u->label, 0, 32);
1275 	u->first_reg = u->last_reg = cpu_to_le32(get_seconds());
1276 
1277 	SET_UUID_FLASH_ONLY(u, 1);
1278 	u->sectors = size >> 9;
1279 
1280 	bch_uuid_write(c);
1281 
1282 	return flash_dev_run(c, u);
1283 }
1284 
1285 /* Cache set */
1286 
1287 __printf(2, 3)
1288 bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...)
1289 {
1290 	va_list args;
1291 
1292 	if (c->on_error != ON_ERROR_PANIC &&
1293 	    test_bit(CACHE_SET_STOPPING, &c->flags))
1294 		return false;
1295 
1296 	/* XXX: we can be called from atomic context
1297 	acquire_console_sem();
1298 	*/
1299 
1300 	printk(KERN_ERR "bcache: error on %pU: ", c->sb.set_uuid);
1301 
1302 	va_start(args, fmt);
1303 	vprintk(fmt, args);
1304 	va_end(args);
1305 
1306 	printk(", disabling caching\n");
1307 
1308 	if (c->on_error == ON_ERROR_PANIC)
1309 		panic("panic forced after error\n");
1310 
1311 	bch_cache_set_unregister(c);
1312 	return true;
1313 }
1314 
1315 void bch_cache_set_release(struct kobject *kobj)
1316 {
1317 	struct cache_set *c = container_of(kobj, struct cache_set, kobj);
1318 	kfree(c);
1319 	module_put(THIS_MODULE);
1320 }
1321 
1322 static void cache_set_free(struct closure *cl)
1323 {
1324 	struct cache_set *c = container_of(cl, struct cache_set, cl);
1325 	struct cache *ca;
1326 	unsigned i;
1327 
1328 	if (!IS_ERR_OR_NULL(c->debug))
1329 		debugfs_remove(c->debug);
1330 
1331 	bch_open_buckets_free(c);
1332 	bch_btree_cache_free(c);
1333 	bch_journal_free(c);
1334 
1335 	for_each_cache(ca, c, i)
1336 		if (ca) {
1337 			ca->set = NULL;
1338 			c->cache[ca->sb.nr_this_dev] = NULL;
1339 			kobject_put(&ca->kobj);
1340 		}
1341 
1342 	bch_bset_sort_state_free(&c->sort);
1343 	free_pages((unsigned long) c->uuids, ilog2(bucket_pages(c)));
1344 
1345 	if (c->moving_gc_wq)
1346 		destroy_workqueue(c->moving_gc_wq);
1347 	if (c->bio_split)
1348 		bioset_free(c->bio_split);
1349 	if (c->fill_iter)
1350 		mempool_destroy(c->fill_iter);
1351 	if (c->bio_meta)
1352 		mempool_destroy(c->bio_meta);
1353 	if (c->search)
1354 		mempool_destroy(c->search);
1355 	kfree(c->devices);
1356 
1357 	mutex_lock(&bch_register_lock);
1358 	list_del(&c->list);
1359 	mutex_unlock(&bch_register_lock);
1360 
1361 	pr_info("Cache set %pU unregistered", c->sb.set_uuid);
1362 	wake_up(&unregister_wait);
1363 
1364 	closure_debug_destroy(&c->cl);
1365 	kobject_put(&c->kobj);
1366 }
1367 
1368 static void cache_set_flush(struct closure *cl)
1369 {
1370 	struct cache_set *c = container_of(cl, struct cache_set, caching);
1371 	struct cache *ca;
1372 	struct btree *b;
1373 	unsigned i;
1374 
1375 	if (!c)
1376 		closure_return(cl);
1377 
1378 	bch_cache_accounting_destroy(&c->accounting);
1379 
1380 	kobject_put(&c->internal);
1381 	kobject_del(&c->kobj);
1382 
1383 	if (c->gc_thread)
1384 		kthread_stop(c->gc_thread);
1385 
1386 	if (!IS_ERR_OR_NULL(c->root))
1387 		list_add(&c->root->list, &c->btree_cache);
1388 
1389 	/* Should skip this if we're unregistering because of an error */
1390 	list_for_each_entry(b, &c->btree_cache, list) {
1391 		mutex_lock(&b->write_lock);
1392 		if (btree_node_dirty(b))
1393 			__bch_btree_node_write(b, NULL);
1394 		mutex_unlock(&b->write_lock);
1395 	}
1396 
1397 	for_each_cache(ca, c, i)
1398 		if (ca->alloc_thread)
1399 			kthread_stop(ca->alloc_thread);
1400 
1401 	if (c->journal.cur) {
1402 		cancel_delayed_work_sync(&c->journal.work);
1403 		/* flush last journal entry if needed */
1404 		c->journal.work.work.func(&c->journal.work.work);
1405 	}
1406 
1407 	closure_return(cl);
1408 }
1409 
1410 static void __cache_set_unregister(struct closure *cl)
1411 {
1412 	struct cache_set *c = container_of(cl, struct cache_set, caching);
1413 	struct cached_dev *dc;
1414 	size_t i;
1415 
1416 	mutex_lock(&bch_register_lock);
1417 
1418 	for (i = 0; i < c->nr_uuids; i++)
1419 		if (c->devices[i]) {
1420 			if (!UUID_FLASH_ONLY(&c->uuids[i]) &&
1421 			    test_bit(CACHE_SET_UNREGISTERING, &c->flags)) {
1422 				dc = container_of(c->devices[i],
1423 						  struct cached_dev, disk);
1424 				bch_cached_dev_detach(dc);
1425 			} else {
1426 				bcache_device_stop(c->devices[i]);
1427 			}
1428 		}
1429 
1430 	mutex_unlock(&bch_register_lock);
1431 
1432 	continue_at(cl, cache_set_flush, system_wq);
1433 }
1434 
1435 void bch_cache_set_stop(struct cache_set *c)
1436 {
1437 	if (!test_and_set_bit(CACHE_SET_STOPPING, &c->flags))
1438 		closure_queue(&c->caching);
1439 }
1440 
1441 void bch_cache_set_unregister(struct cache_set *c)
1442 {
1443 	set_bit(CACHE_SET_UNREGISTERING, &c->flags);
1444 	bch_cache_set_stop(c);
1445 }
1446 
1447 #define alloc_bucket_pages(gfp, c)			\
1448 	((void *) __get_free_pages(__GFP_ZERO|gfp, ilog2(bucket_pages(c))))
1449 
1450 struct cache_set *bch_cache_set_alloc(struct cache_sb *sb)
1451 {
1452 	int iter_size;
1453 	struct cache_set *c = kzalloc(sizeof(struct cache_set), GFP_KERNEL);
1454 	if (!c)
1455 		return NULL;
1456 
1457 	__module_get(THIS_MODULE);
1458 	closure_init(&c->cl, NULL);
1459 	set_closure_fn(&c->cl, cache_set_free, system_wq);
1460 
1461 	closure_init(&c->caching, &c->cl);
1462 	set_closure_fn(&c->caching, __cache_set_unregister, system_wq);
1463 
1464 	/* Maybe create continue_at_noreturn() and use it here? */
1465 	closure_set_stopped(&c->cl);
1466 	closure_put(&c->cl);
1467 
1468 	kobject_init(&c->kobj, &bch_cache_set_ktype);
1469 	kobject_init(&c->internal, &bch_cache_set_internal_ktype);
1470 
1471 	bch_cache_accounting_init(&c->accounting, &c->cl);
1472 
1473 	memcpy(c->sb.set_uuid, sb->set_uuid, 16);
1474 	c->sb.block_size	= sb->block_size;
1475 	c->sb.bucket_size	= sb->bucket_size;
1476 	c->sb.nr_in_set		= sb->nr_in_set;
1477 	c->sb.last_mount	= sb->last_mount;
1478 	c->bucket_bits		= ilog2(sb->bucket_size);
1479 	c->block_bits		= ilog2(sb->block_size);
1480 	c->nr_uuids		= bucket_bytes(c) / sizeof(struct uuid_entry);
1481 
1482 	c->btree_pages		= bucket_pages(c);
1483 	if (c->btree_pages > BTREE_MAX_PAGES)
1484 		c->btree_pages = max_t(int, c->btree_pages / 4,
1485 				       BTREE_MAX_PAGES);
1486 
1487 	sema_init(&c->sb_write_mutex, 1);
1488 	mutex_init(&c->bucket_lock);
1489 	init_waitqueue_head(&c->btree_cache_wait);
1490 	init_waitqueue_head(&c->bucket_wait);
1491 	init_waitqueue_head(&c->gc_wait);
1492 	sema_init(&c->uuid_write_mutex, 1);
1493 
1494 	spin_lock_init(&c->btree_gc_time.lock);
1495 	spin_lock_init(&c->btree_split_time.lock);
1496 	spin_lock_init(&c->btree_read_time.lock);
1497 
1498 	bch_moving_init_cache_set(c);
1499 
1500 	INIT_LIST_HEAD(&c->list);
1501 	INIT_LIST_HEAD(&c->cached_devs);
1502 	INIT_LIST_HEAD(&c->btree_cache);
1503 	INIT_LIST_HEAD(&c->btree_cache_freeable);
1504 	INIT_LIST_HEAD(&c->btree_cache_freed);
1505 	INIT_LIST_HEAD(&c->data_buckets);
1506 
1507 	c->search = mempool_create_slab_pool(32, bch_search_cache);
1508 	if (!c->search)
1509 		goto err;
1510 
1511 	iter_size = (sb->bucket_size / sb->block_size + 1) *
1512 		sizeof(struct btree_iter_set);
1513 
1514 	if (!(c->devices = kzalloc(c->nr_uuids * sizeof(void *), GFP_KERNEL)) ||
1515 	    !(c->bio_meta = mempool_create_kmalloc_pool(2,
1516 				sizeof(struct bbio) + sizeof(struct bio_vec) *
1517 				bucket_pages(c))) ||
1518 	    !(c->fill_iter = mempool_create_kmalloc_pool(1, iter_size)) ||
1519 	    !(c->bio_split = bioset_create(4, offsetof(struct bbio, bio))) ||
1520 	    !(c->uuids = alloc_bucket_pages(GFP_KERNEL, c)) ||
1521 	    !(c->moving_gc_wq = alloc_workqueue("bcache_gc",
1522 						WQ_MEM_RECLAIM, 0)) ||
1523 	    bch_journal_alloc(c) ||
1524 	    bch_btree_cache_alloc(c) ||
1525 	    bch_open_buckets_alloc(c) ||
1526 	    bch_bset_sort_state_init(&c->sort, ilog2(c->btree_pages)))
1527 		goto err;
1528 
1529 	c->congested_read_threshold_us	= 2000;
1530 	c->congested_write_threshold_us	= 20000;
1531 	c->error_limit	= 8 << IO_ERROR_SHIFT;
1532 
1533 	return c;
1534 err:
1535 	bch_cache_set_unregister(c);
1536 	return NULL;
1537 }
1538 
1539 static void run_cache_set(struct cache_set *c)
1540 {
1541 	const char *err = "cannot allocate memory";
1542 	struct cached_dev *dc, *t;
1543 	struct cache *ca;
1544 	struct closure cl;
1545 	unsigned i;
1546 
1547 	closure_init_stack(&cl);
1548 
1549 	for_each_cache(ca, c, i)
1550 		c->nbuckets += ca->sb.nbuckets;
1551 	set_gc_sectors(c);
1552 
1553 	if (CACHE_SYNC(&c->sb)) {
1554 		LIST_HEAD(journal);
1555 		struct bkey *k;
1556 		struct jset *j;
1557 
1558 		err = "cannot allocate memory for journal";
1559 		if (bch_journal_read(c, &journal))
1560 			goto err;
1561 
1562 		pr_debug("btree_journal_read() done");
1563 
1564 		err = "no journal entries found";
1565 		if (list_empty(&journal))
1566 			goto err;
1567 
1568 		j = &list_entry(journal.prev, struct journal_replay, list)->j;
1569 
1570 		err = "IO error reading priorities";
1571 		for_each_cache(ca, c, i)
1572 			prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev]);
1573 
1574 		/*
1575 		 * If prio_read() fails it'll call cache_set_error and we'll
1576 		 * tear everything down right away, but if we perhaps checked
1577 		 * sooner we could avoid journal replay.
1578 		 */
1579 
1580 		k = &j->btree_root;
1581 
1582 		err = "bad btree root";
1583 		if (__bch_btree_ptr_invalid(c, k))
1584 			goto err;
1585 
1586 		err = "error reading btree root";
1587 		c->root = bch_btree_node_get(c, NULL, k, j->btree_level, true, NULL);
1588 		if (IS_ERR_OR_NULL(c->root))
1589 			goto err;
1590 
1591 		list_del_init(&c->root->list);
1592 		rw_unlock(true, c->root);
1593 
1594 		err = uuid_read(c, j, &cl);
1595 		if (err)
1596 			goto err;
1597 
1598 		err = "error in recovery";
1599 		if (bch_btree_check(c))
1600 			goto err;
1601 
1602 		bch_journal_mark(c, &journal);
1603 		bch_initial_gc_finish(c);
1604 		pr_debug("btree_check() done");
1605 
1606 		/*
1607 		 * bcache_journal_next() can't happen sooner, or
1608 		 * btree_gc_finish() will give spurious errors about last_gc >
1609 		 * gc_gen - this is a hack but oh well.
1610 		 */
1611 		bch_journal_next(&c->journal);
1612 
1613 		err = "error starting allocator thread";
1614 		for_each_cache(ca, c, i)
1615 			if (bch_cache_allocator_start(ca))
1616 				goto err;
1617 
1618 		/*
1619 		 * First place it's safe to allocate: btree_check() and
1620 		 * btree_gc_finish() have to run before we have buckets to
1621 		 * allocate, and bch_bucket_alloc_set() might cause a journal
1622 		 * entry to be written so bcache_journal_next() has to be called
1623 		 * first.
1624 		 *
1625 		 * If the uuids were in the old format we have to rewrite them
1626 		 * before the next journal entry is written:
1627 		 */
1628 		if (j->version < BCACHE_JSET_VERSION_UUID)
1629 			__uuid_write(c);
1630 
1631 		bch_journal_replay(c, &journal);
1632 	} else {
1633 		pr_notice("invalidating existing data");
1634 
1635 		for_each_cache(ca, c, i) {
1636 			unsigned j;
1637 
1638 			ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7,
1639 					      2, SB_JOURNAL_BUCKETS);
1640 
1641 			for (j = 0; j < ca->sb.keys; j++)
1642 				ca->sb.d[j] = ca->sb.first_bucket + j;
1643 		}
1644 
1645 		bch_initial_gc_finish(c);
1646 
1647 		err = "error starting allocator thread";
1648 		for_each_cache(ca, c, i)
1649 			if (bch_cache_allocator_start(ca))
1650 				goto err;
1651 
1652 		mutex_lock(&c->bucket_lock);
1653 		for_each_cache(ca, c, i)
1654 			bch_prio_write(ca);
1655 		mutex_unlock(&c->bucket_lock);
1656 
1657 		err = "cannot allocate new UUID bucket";
1658 		if (__uuid_write(c))
1659 			goto err;
1660 
1661 		err = "cannot allocate new btree root";
1662 		c->root = __bch_btree_node_alloc(c, NULL, 0, true, NULL);
1663 		if (IS_ERR_OR_NULL(c->root))
1664 			goto err;
1665 
1666 		mutex_lock(&c->root->write_lock);
1667 		bkey_copy_key(&c->root->key, &MAX_KEY);
1668 		bch_btree_node_write(c->root, &cl);
1669 		mutex_unlock(&c->root->write_lock);
1670 
1671 		bch_btree_set_root(c->root);
1672 		rw_unlock(true, c->root);
1673 
1674 		/*
1675 		 * We don't want to write the first journal entry until
1676 		 * everything is set up - fortunately journal entries won't be
1677 		 * written until the SET_CACHE_SYNC() here:
1678 		 */
1679 		SET_CACHE_SYNC(&c->sb, true);
1680 
1681 		bch_journal_next(&c->journal);
1682 		bch_journal_meta(c, &cl);
1683 	}
1684 
1685 	err = "error starting gc thread";
1686 	if (bch_gc_thread_start(c))
1687 		goto err;
1688 
1689 	closure_sync(&cl);
1690 	c->sb.last_mount = get_seconds();
1691 	bcache_write_super(c);
1692 
1693 	list_for_each_entry_safe(dc, t, &uncached_devices, list)
1694 		bch_cached_dev_attach(dc, c);
1695 
1696 	flash_devs_run(c);
1697 
1698 	set_bit(CACHE_SET_RUNNING, &c->flags);
1699 	return;
1700 err:
1701 	closure_sync(&cl);
1702 	/* XXX: test this, it's broken */
1703 	bch_cache_set_error(c, "%s", err);
1704 }
1705 
1706 static bool can_attach_cache(struct cache *ca, struct cache_set *c)
1707 {
1708 	return ca->sb.block_size	== c->sb.block_size &&
1709 		ca->sb.bucket_size	== c->sb.bucket_size &&
1710 		ca->sb.nr_in_set	== c->sb.nr_in_set;
1711 }
1712 
1713 static const char *register_cache_set(struct cache *ca)
1714 {
1715 	char buf[12];
1716 	const char *err = "cannot allocate memory";
1717 	struct cache_set *c;
1718 
1719 	list_for_each_entry(c, &bch_cache_sets, list)
1720 		if (!memcmp(c->sb.set_uuid, ca->sb.set_uuid, 16)) {
1721 			if (c->cache[ca->sb.nr_this_dev])
1722 				return "duplicate cache set member";
1723 
1724 			if (!can_attach_cache(ca, c))
1725 				return "cache sb does not match set";
1726 
1727 			if (!CACHE_SYNC(&ca->sb))
1728 				SET_CACHE_SYNC(&c->sb, false);
1729 
1730 			goto found;
1731 		}
1732 
1733 	c = bch_cache_set_alloc(&ca->sb);
1734 	if (!c)
1735 		return err;
1736 
1737 	err = "error creating kobject";
1738 	if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->sb.set_uuid) ||
1739 	    kobject_add(&c->internal, &c->kobj, "internal"))
1740 		goto err;
1741 
1742 	if (bch_cache_accounting_add_kobjs(&c->accounting, &c->kobj))
1743 		goto err;
1744 
1745 	bch_debug_init_cache_set(c);
1746 
1747 	list_add(&c->list, &bch_cache_sets);
1748 found:
1749 	sprintf(buf, "cache%i", ca->sb.nr_this_dev);
1750 	if (sysfs_create_link(&ca->kobj, &c->kobj, "set") ||
1751 	    sysfs_create_link(&c->kobj, &ca->kobj, buf))
1752 		goto err;
1753 
1754 	if (ca->sb.seq > c->sb.seq) {
1755 		c->sb.version		= ca->sb.version;
1756 		memcpy(c->sb.set_uuid, ca->sb.set_uuid, 16);
1757 		c->sb.flags             = ca->sb.flags;
1758 		c->sb.seq		= ca->sb.seq;
1759 		pr_debug("set version = %llu", c->sb.version);
1760 	}
1761 
1762 	kobject_get(&ca->kobj);
1763 	ca->set = c;
1764 	ca->set->cache[ca->sb.nr_this_dev] = ca;
1765 	c->cache_by_alloc[c->caches_loaded++] = ca;
1766 
1767 	if (c->caches_loaded == c->sb.nr_in_set)
1768 		run_cache_set(c);
1769 
1770 	return NULL;
1771 err:
1772 	bch_cache_set_unregister(c);
1773 	return err;
1774 }
1775 
1776 /* Cache device */
1777 
1778 void bch_cache_release(struct kobject *kobj)
1779 {
1780 	struct cache *ca = container_of(kobj, struct cache, kobj);
1781 	unsigned i;
1782 
1783 	if (ca->set) {
1784 		BUG_ON(ca->set->cache[ca->sb.nr_this_dev] != ca);
1785 		ca->set->cache[ca->sb.nr_this_dev] = NULL;
1786 	}
1787 
1788 	free_pages((unsigned long) ca->disk_buckets, ilog2(bucket_pages(ca)));
1789 	kfree(ca->prio_buckets);
1790 	vfree(ca->buckets);
1791 
1792 	free_heap(&ca->heap);
1793 	free_fifo(&ca->free_inc);
1794 
1795 	for (i = 0; i < RESERVE_NR; i++)
1796 		free_fifo(&ca->free[i]);
1797 
1798 	if (ca->sb_bio.bi_inline_vecs[0].bv_page)
1799 		put_page(ca->sb_bio.bi_io_vec[0].bv_page);
1800 
1801 	if (!IS_ERR_OR_NULL(ca->bdev))
1802 		blkdev_put(ca->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1803 
1804 	kfree(ca);
1805 	module_put(THIS_MODULE);
1806 }
1807 
1808 static int cache_alloc(struct cache *ca)
1809 {
1810 	size_t free;
1811 	struct bucket *b;
1812 
1813 	__module_get(THIS_MODULE);
1814 	kobject_init(&ca->kobj, &bch_cache_ktype);
1815 
1816 	bio_init(&ca->journal.bio, ca->journal.bio.bi_inline_vecs, 8);
1817 
1818 	free = roundup_pow_of_two(ca->sb.nbuckets) >> 10;
1819 
1820 	if (!init_fifo(&ca->free[RESERVE_BTREE], 8, GFP_KERNEL) ||
1821 	    !init_fifo_exact(&ca->free[RESERVE_PRIO], prio_buckets(ca), GFP_KERNEL) ||
1822 	    !init_fifo(&ca->free[RESERVE_MOVINGGC], free, GFP_KERNEL) ||
1823 	    !init_fifo(&ca->free[RESERVE_NONE], free, GFP_KERNEL) ||
1824 	    !init_fifo(&ca->free_inc,	free << 2, GFP_KERNEL) ||
1825 	    !init_heap(&ca->heap,	free << 3, GFP_KERNEL) ||
1826 	    !(ca->buckets	= vzalloc(sizeof(struct bucket) *
1827 					  ca->sb.nbuckets)) ||
1828 	    !(ca->prio_buckets	= kzalloc(sizeof(uint64_t) * prio_buckets(ca) *
1829 					  2, GFP_KERNEL)) ||
1830 	    !(ca->disk_buckets	= alloc_bucket_pages(GFP_KERNEL, ca)))
1831 		return -ENOMEM;
1832 
1833 	ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca);
1834 
1835 	for_each_bucket(b, ca)
1836 		atomic_set(&b->pin, 0);
1837 
1838 	return 0;
1839 }
1840 
1841 static int register_cache(struct cache_sb *sb, struct page *sb_page,
1842 				struct block_device *bdev, struct cache *ca)
1843 {
1844 	char name[BDEVNAME_SIZE];
1845 	const char *err = NULL; /* must be set for any error case */
1846 	int ret = 0;
1847 
1848 	memcpy(&ca->sb, sb, sizeof(struct cache_sb));
1849 	ca->bdev = bdev;
1850 	ca->bdev->bd_holder = ca;
1851 
1852 	bio_init(&ca->sb_bio, ca->sb_bio.bi_inline_vecs, 1);
1853 	ca->sb_bio.bi_io_vec[0].bv_page = sb_page;
1854 	get_page(sb_page);
1855 
1856 	if (blk_queue_discard(bdev_get_queue(ca->bdev)))
1857 		ca->discard = CACHE_DISCARD(&ca->sb);
1858 
1859 	ret = cache_alloc(ca);
1860 	if (ret != 0) {
1861 		if (ret == -ENOMEM)
1862 			err = "cache_alloc(): -ENOMEM";
1863 		else
1864 			err = "cache_alloc(): unknown error";
1865 		goto err;
1866 	}
1867 
1868 	if (kobject_add(&ca->kobj, &part_to_dev(bdev->bd_part)->kobj, "bcache")) {
1869 		err = "error calling kobject_add";
1870 		ret = -ENOMEM;
1871 		goto out;
1872 	}
1873 
1874 	mutex_lock(&bch_register_lock);
1875 	err = register_cache_set(ca);
1876 	mutex_unlock(&bch_register_lock);
1877 
1878 	if (err) {
1879 		ret = -ENODEV;
1880 		goto out;
1881 	}
1882 
1883 	pr_info("registered cache device %s", bdevname(bdev, name));
1884 
1885 out:
1886 	kobject_put(&ca->kobj);
1887 
1888 err:
1889 	if (err)
1890 		pr_notice("error opening %s: %s", bdevname(bdev, name), err);
1891 
1892 	return ret;
1893 }
1894 
1895 /* Global interfaces/init */
1896 
1897 static ssize_t register_bcache(struct kobject *, struct kobj_attribute *,
1898 			       const char *, size_t);
1899 
1900 kobj_attribute_write(register,		register_bcache);
1901 kobj_attribute_write(register_quiet,	register_bcache);
1902 
1903 static bool bch_is_open_backing(struct block_device *bdev) {
1904 	struct cache_set *c, *tc;
1905 	struct cached_dev *dc, *t;
1906 
1907 	list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
1908 		list_for_each_entry_safe(dc, t, &c->cached_devs, list)
1909 			if (dc->bdev == bdev)
1910 				return true;
1911 	list_for_each_entry_safe(dc, t, &uncached_devices, list)
1912 		if (dc->bdev == bdev)
1913 			return true;
1914 	return false;
1915 }
1916 
1917 static bool bch_is_open_cache(struct block_device *bdev) {
1918 	struct cache_set *c, *tc;
1919 	struct cache *ca;
1920 	unsigned i;
1921 
1922 	list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
1923 		for_each_cache(ca, c, i)
1924 			if (ca->bdev == bdev)
1925 				return true;
1926 	return false;
1927 }
1928 
1929 static bool bch_is_open(struct block_device *bdev) {
1930 	return bch_is_open_cache(bdev) || bch_is_open_backing(bdev);
1931 }
1932 
1933 static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr,
1934 			       const char *buffer, size_t size)
1935 {
1936 	ssize_t ret = size;
1937 	const char *err = "cannot allocate memory";
1938 	char *path = NULL;
1939 	struct cache_sb *sb = NULL;
1940 	struct block_device *bdev = NULL;
1941 	struct page *sb_page = NULL;
1942 
1943 	if (!try_module_get(THIS_MODULE))
1944 		return -EBUSY;
1945 
1946 	if (!(path = kstrndup(buffer, size, GFP_KERNEL)) ||
1947 	    !(sb = kmalloc(sizeof(struct cache_sb), GFP_KERNEL)))
1948 		goto err;
1949 
1950 	err = "failed to open device";
1951 	bdev = blkdev_get_by_path(strim(path),
1952 				  FMODE_READ|FMODE_WRITE|FMODE_EXCL,
1953 				  sb);
1954 	if (IS_ERR(bdev)) {
1955 		if (bdev == ERR_PTR(-EBUSY)) {
1956 			bdev = lookup_bdev(strim(path));
1957 			mutex_lock(&bch_register_lock);
1958 			if (!IS_ERR(bdev) && bch_is_open(bdev))
1959 				err = "device already registered";
1960 			else
1961 				err = "device busy";
1962 			mutex_unlock(&bch_register_lock);
1963 			if (attr == &ksysfs_register_quiet)
1964 				goto out;
1965 		}
1966 		goto err;
1967 	}
1968 
1969 	err = "failed to set blocksize";
1970 	if (set_blocksize(bdev, 4096))
1971 		goto err_close;
1972 
1973 	err = read_super(sb, bdev, &sb_page);
1974 	if (err)
1975 		goto err_close;
1976 
1977 	if (SB_IS_BDEV(sb)) {
1978 		struct cached_dev *dc = kzalloc(sizeof(*dc), GFP_KERNEL);
1979 		if (!dc)
1980 			goto err_close;
1981 
1982 		mutex_lock(&bch_register_lock);
1983 		register_bdev(sb, sb_page, bdev, dc);
1984 		mutex_unlock(&bch_register_lock);
1985 	} else {
1986 		struct cache *ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1987 		if (!ca)
1988 			goto err_close;
1989 
1990 		if (register_cache(sb, sb_page, bdev, ca) != 0)
1991 			goto err_close;
1992 	}
1993 out:
1994 	if (sb_page)
1995 		put_page(sb_page);
1996 	kfree(sb);
1997 	kfree(path);
1998 	module_put(THIS_MODULE);
1999 	return ret;
2000 
2001 err_close:
2002 	blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2003 err:
2004 	pr_info("error opening %s: %s", path, err);
2005 	ret = -EINVAL;
2006 	goto out;
2007 }
2008 
2009 static int bcache_reboot(struct notifier_block *n, unsigned long code, void *x)
2010 {
2011 	if (code == SYS_DOWN ||
2012 	    code == SYS_HALT ||
2013 	    code == SYS_POWER_OFF) {
2014 		DEFINE_WAIT(wait);
2015 		unsigned long start = jiffies;
2016 		bool stopped = false;
2017 
2018 		struct cache_set *c, *tc;
2019 		struct cached_dev *dc, *tdc;
2020 
2021 		mutex_lock(&bch_register_lock);
2022 
2023 		if (list_empty(&bch_cache_sets) &&
2024 		    list_empty(&uncached_devices))
2025 			goto out;
2026 
2027 		pr_info("Stopping all devices:");
2028 
2029 		list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
2030 			bch_cache_set_stop(c);
2031 
2032 		list_for_each_entry_safe(dc, tdc, &uncached_devices, list)
2033 			bcache_device_stop(&dc->disk);
2034 
2035 		/* What's a condition variable? */
2036 		while (1) {
2037 			long timeout = start + 2 * HZ - jiffies;
2038 
2039 			stopped = list_empty(&bch_cache_sets) &&
2040 				list_empty(&uncached_devices);
2041 
2042 			if (timeout < 0 || stopped)
2043 				break;
2044 
2045 			prepare_to_wait(&unregister_wait, &wait,
2046 					TASK_UNINTERRUPTIBLE);
2047 
2048 			mutex_unlock(&bch_register_lock);
2049 			schedule_timeout(timeout);
2050 			mutex_lock(&bch_register_lock);
2051 		}
2052 
2053 		finish_wait(&unregister_wait, &wait);
2054 
2055 		if (stopped)
2056 			pr_info("All devices stopped");
2057 		else
2058 			pr_notice("Timeout waiting for devices to be closed");
2059 out:
2060 		mutex_unlock(&bch_register_lock);
2061 	}
2062 
2063 	return NOTIFY_DONE;
2064 }
2065 
2066 static struct notifier_block reboot = {
2067 	.notifier_call	= bcache_reboot,
2068 	.priority	= INT_MAX, /* before any real devices */
2069 };
2070 
2071 static void bcache_exit(void)
2072 {
2073 	bch_debug_exit();
2074 	bch_request_exit();
2075 	if (bcache_kobj)
2076 		kobject_put(bcache_kobj);
2077 	if (bcache_wq)
2078 		destroy_workqueue(bcache_wq);
2079 	if (bcache_major)
2080 		unregister_blkdev(bcache_major, "bcache");
2081 	unregister_reboot_notifier(&reboot);
2082 }
2083 
2084 static int __init bcache_init(void)
2085 {
2086 	static const struct attribute *files[] = {
2087 		&ksysfs_register.attr,
2088 		&ksysfs_register_quiet.attr,
2089 		NULL
2090 	};
2091 
2092 	mutex_init(&bch_register_lock);
2093 	init_waitqueue_head(&unregister_wait);
2094 	register_reboot_notifier(&reboot);
2095 	closure_debug_init();
2096 
2097 	bcache_major = register_blkdev(0, "bcache");
2098 	if (bcache_major < 0) {
2099 		unregister_reboot_notifier(&reboot);
2100 		return bcache_major;
2101 	}
2102 
2103 	if (!(bcache_wq = alloc_workqueue("bcache", WQ_MEM_RECLAIM, 0)) ||
2104 	    !(bcache_kobj = kobject_create_and_add("bcache", fs_kobj)) ||
2105 	    sysfs_create_files(bcache_kobj, files) ||
2106 	    bch_request_init() ||
2107 	    bch_debug_init(bcache_kobj))
2108 		goto err;
2109 
2110 	return 0;
2111 err:
2112 	bcache_exit();
2113 	return -ENOMEM;
2114 }
2115 
2116 module_exit(bcache_exit);
2117 module_init(bcache_init);
2118