xref: /openbmc/linux/drivers/md/bcache/super.c (revision a06c488d)
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 
62 /* Superblock */
63 
64 static const char *read_super(struct cache_sb *sb, struct block_device *bdev,
65 			      struct page **res)
66 {
67 	const char *err;
68 	struct cache_sb *s;
69 	struct buffer_head *bh = __bread(bdev, 1, SB_SIZE);
70 	unsigned i;
71 
72 	if (!bh)
73 		return "IO error";
74 
75 	s = (struct cache_sb *) bh->b_data;
76 
77 	sb->offset		= le64_to_cpu(s->offset);
78 	sb->version		= le64_to_cpu(s->version);
79 
80 	memcpy(sb->magic,	s->magic, 16);
81 	memcpy(sb->uuid,	s->uuid, 16);
82 	memcpy(sb->set_uuid,	s->set_uuid, 16);
83 	memcpy(sb->label,	s->label, SB_LABEL_SIZE);
84 
85 	sb->flags		= le64_to_cpu(s->flags);
86 	sb->seq			= le64_to_cpu(s->seq);
87 	sb->last_mount		= le32_to_cpu(s->last_mount);
88 	sb->first_bucket	= le16_to_cpu(s->first_bucket);
89 	sb->keys		= le16_to_cpu(s->keys);
90 
91 	for (i = 0; i < SB_JOURNAL_BUCKETS; i++)
92 		sb->d[i] = le64_to_cpu(s->d[i]);
93 
94 	pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u",
95 		 sb->version, sb->flags, sb->seq, sb->keys);
96 
97 	err = "Not a bcache superblock";
98 	if (sb->offset != SB_SECTOR)
99 		goto err;
100 
101 	if (memcmp(sb->magic, bcache_magic, 16))
102 		goto err;
103 
104 	err = "Too many journal buckets";
105 	if (sb->keys > SB_JOURNAL_BUCKETS)
106 		goto err;
107 
108 	err = "Bad checksum";
109 	if (s->csum != csum_set(s))
110 		goto err;
111 
112 	err = "Bad UUID";
113 	if (bch_is_zero(sb->uuid, 16))
114 		goto err;
115 
116 	sb->block_size	= le16_to_cpu(s->block_size);
117 
118 	err = "Superblock block size smaller than device block size";
119 	if (sb->block_size << 9 < bdev_logical_block_size(bdev))
120 		goto err;
121 
122 	switch (sb->version) {
123 	case BCACHE_SB_VERSION_BDEV:
124 		sb->data_offset	= BDEV_DATA_START_DEFAULT;
125 		break;
126 	case BCACHE_SB_VERSION_BDEV_WITH_OFFSET:
127 		sb->data_offset	= le64_to_cpu(s->data_offset);
128 
129 		err = "Bad data offset";
130 		if (sb->data_offset < BDEV_DATA_START_DEFAULT)
131 			goto err;
132 
133 		break;
134 	case BCACHE_SB_VERSION_CDEV:
135 	case BCACHE_SB_VERSION_CDEV_WITH_UUID:
136 		sb->nbuckets	= le64_to_cpu(s->nbuckets);
137 		sb->block_size	= le16_to_cpu(s->block_size);
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_rw		= REQ_SYNC|REQ_META;
216 	bio->bi_iter.bi_size	= SB_SIZE;
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(REQ_WRITE, 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, unsigned long rw,
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_rw	= REQ_SYNC|REQ_META|rw;
352 		bio->bi_iter.bi_size = KEY_SIZE(k) << 9;
353 
354 		bio->bi_end_io	= uuid_endio;
355 		bio->bi_private = cl;
356 		bch_bio_map(bio, c->uuids);
357 
358 		bch_submit_bbio(bio, c, k, i);
359 
360 		if (!(rw & WRITE))
361 			break;
362 	}
363 
364 	bch_extent_to_text(buf, sizeof(buf), k);
365 	pr_debug("%s UUIDs at %s", rw & REQ_WRITE ? "wrote" : "read", buf);
366 
367 	for (u = c->uuids; u < c->uuids + c->nr_uuids; u++)
368 		if (!bch_is_zero(u->uuid, 16))
369 			pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u",
370 				 u - c->uuids, u->uuid, u->label,
371 				 u->first_reg, u->last_reg, u->invalidated);
372 
373 	closure_return_with_destructor(cl, uuid_io_unlock);
374 }
375 
376 static char *uuid_read(struct cache_set *c, struct jset *j, struct closure *cl)
377 {
378 	struct bkey *k = &j->uuid_bucket;
379 
380 	if (__bch_btree_ptr_invalid(c, k))
381 		return "bad uuid pointer";
382 
383 	bkey_copy(&c->uuid_bucket, k);
384 	uuid_io(c, READ_SYNC, k, cl);
385 
386 	if (j->version < BCACHE_JSET_VERSION_UUIDv1) {
387 		struct uuid_entry_v0	*u0 = (void *) c->uuids;
388 		struct uuid_entry	*u1 = (void *) c->uuids;
389 		int i;
390 
391 		closure_sync(cl);
392 
393 		/*
394 		 * Since the new uuid entry is bigger than the old, we have to
395 		 * convert starting at the highest memory address and work down
396 		 * in order to do it in place
397 		 */
398 
399 		for (i = c->nr_uuids - 1;
400 		     i >= 0;
401 		     --i) {
402 			memcpy(u1[i].uuid,	u0[i].uuid, 16);
403 			memcpy(u1[i].label,	u0[i].label, 32);
404 
405 			u1[i].first_reg		= u0[i].first_reg;
406 			u1[i].last_reg		= u0[i].last_reg;
407 			u1[i].invalidated	= u0[i].invalidated;
408 
409 			u1[i].flags	= 0;
410 			u1[i].sectors	= 0;
411 		}
412 	}
413 
414 	return NULL;
415 }
416 
417 static int __uuid_write(struct cache_set *c)
418 {
419 	BKEY_PADDED(key) k;
420 	struct closure cl;
421 	closure_init_stack(&cl);
422 
423 	lockdep_assert_held(&bch_register_lock);
424 
425 	if (bch_bucket_alloc_set(c, RESERVE_BTREE, &k.key, 1, true))
426 		return 1;
427 
428 	SET_KEY_SIZE(&k.key, c->sb.bucket_size);
429 	uuid_io(c, REQ_WRITE, &k.key, &cl);
430 	closure_sync(&cl);
431 
432 	bkey_copy(&c->uuid_bucket, &k.key);
433 	bkey_put(c, &k.key);
434 	return 0;
435 }
436 
437 int bch_uuid_write(struct cache_set *c)
438 {
439 	int ret = __uuid_write(c);
440 
441 	if (!ret)
442 		bch_journal_meta(c, NULL);
443 
444 	return ret;
445 }
446 
447 static struct uuid_entry *uuid_find(struct cache_set *c, const char *uuid)
448 {
449 	struct uuid_entry *u;
450 
451 	for (u = c->uuids;
452 	     u < c->uuids + c->nr_uuids; u++)
453 		if (!memcmp(u->uuid, uuid, 16))
454 			return u;
455 
456 	return NULL;
457 }
458 
459 static struct uuid_entry *uuid_find_empty(struct cache_set *c)
460 {
461 	static const char zero_uuid[16] = "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0";
462 	return uuid_find(c, zero_uuid);
463 }
464 
465 /*
466  * Bucket priorities/gens:
467  *
468  * For each bucket, we store on disk its
469    * 8 bit gen
470    * 16 bit priority
471  *
472  * See alloc.c for an explanation of the gen. The priority is used to implement
473  * lru (and in the future other) cache replacement policies; for most purposes
474  * it's just an opaque integer.
475  *
476  * The gens and the priorities don't have a whole lot to do with each other, and
477  * it's actually the gens that must be written out at specific times - it's no
478  * big deal if the priorities don't get written, if we lose them we just reuse
479  * buckets in suboptimal order.
480  *
481  * On disk they're stored in a packed array, and in as many buckets are required
482  * to fit them all. The buckets we use to store them form a list; the journal
483  * header points to the first bucket, the first bucket points to the second
484  * bucket, et cetera.
485  *
486  * This code is used by the allocation code; periodically (whenever it runs out
487  * of buckets to allocate from) the allocation code will invalidate some
488  * buckets, but it can't use those buckets until their new gens are safely on
489  * disk.
490  */
491 
492 static void prio_endio(struct bio *bio)
493 {
494 	struct cache *ca = bio->bi_private;
495 
496 	cache_set_err_on(bio->bi_error, ca->set, "accessing priorities");
497 	bch_bbio_free(bio, ca->set);
498 	closure_put(&ca->prio);
499 }
500 
501 static void prio_io(struct cache *ca, uint64_t bucket, unsigned long rw)
502 {
503 	struct closure *cl = &ca->prio;
504 	struct bio *bio = bch_bbio_alloc(ca->set);
505 
506 	closure_init_stack(cl);
507 
508 	bio->bi_iter.bi_sector	= bucket * ca->sb.bucket_size;
509 	bio->bi_bdev		= ca->bdev;
510 	bio->bi_rw		= REQ_SYNC|REQ_META|rw;
511 	bio->bi_iter.bi_size	= bucket_bytes(ca);
512 
513 	bio->bi_end_io	= prio_endio;
514 	bio->bi_private = ca;
515 	bch_bio_map(bio, ca->disk_buckets);
516 
517 	closure_bio_submit(bio, &ca->prio);
518 	closure_sync(cl);
519 }
520 
521 void bch_prio_write(struct cache *ca)
522 {
523 	int i;
524 	struct bucket *b;
525 	struct closure cl;
526 
527 	closure_init_stack(&cl);
528 
529 	lockdep_assert_held(&ca->set->bucket_lock);
530 
531 	ca->disk_buckets->seq++;
532 
533 	atomic_long_add(ca->sb.bucket_size * prio_buckets(ca),
534 			&ca->meta_sectors_written);
535 
536 	//pr_debug("free %zu, free_inc %zu, unused %zu", fifo_used(&ca->free),
537 	//	 fifo_used(&ca->free_inc), fifo_used(&ca->unused));
538 
539 	for (i = prio_buckets(ca) - 1; i >= 0; --i) {
540 		long bucket;
541 		struct prio_set *p = ca->disk_buckets;
542 		struct bucket_disk *d = p->data;
543 		struct bucket_disk *end = d + prios_per_bucket(ca);
544 
545 		for (b = ca->buckets + i * prios_per_bucket(ca);
546 		     b < ca->buckets + ca->sb.nbuckets && d < end;
547 		     b++, d++) {
548 			d->prio = cpu_to_le16(b->prio);
549 			d->gen = b->gen;
550 		}
551 
552 		p->next_bucket	= ca->prio_buckets[i + 1];
553 		p->magic	= pset_magic(&ca->sb);
554 		p->csum		= bch_crc64(&p->magic, bucket_bytes(ca) - 8);
555 
556 		bucket = bch_bucket_alloc(ca, RESERVE_PRIO, true);
557 		BUG_ON(bucket == -1);
558 
559 		mutex_unlock(&ca->set->bucket_lock);
560 		prio_io(ca, bucket, REQ_WRITE);
561 		mutex_lock(&ca->set->bucket_lock);
562 
563 		ca->prio_buckets[i] = bucket;
564 		atomic_dec_bug(&ca->buckets[bucket].pin);
565 	}
566 
567 	mutex_unlock(&ca->set->bucket_lock);
568 
569 	bch_journal_meta(ca->set, &cl);
570 	closure_sync(&cl);
571 
572 	mutex_lock(&ca->set->bucket_lock);
573 
574 	/*
575 	 * Don't want the old priorities to get garbage collected until after we
576 	 * finish writing the new ones, and they're journalled
577 	 */
578 	for (i = 0; i < prio_buckets(ca); i++) {
579 		if (ca->prio_last_buckets[i])
580 			__bch_bucket_free(ca,
581 				&ca->buckets[ca->prio_last_buckets[i]]);
582 
583 		ca->prio_last_buckets[i] = ca->prio_buckets[i];
584 	}
585 }
586 
587 static void prio_read(struct cache *ca, uint64_t bucket)
588 {
589 	struct prio_set *p = ca->disk_buckets;
590 	struct bucket_disk *d = p->data + prios_per_bucket(ca), *end = d;
591 	struct bucket *b;
592 	unsigned bucket_nr = 0;
593 
594 	for (b = ca->buckets;
595 	     b < ca->buckets + ca->sb.nbuckets;
596 	     b++, d++) {
597 		if (d == end) {
598 			ca->prio_buckets[bucket_nr] = bucket;
599 			ca->prio_last_buckets[bucket_nr] = bucket;
600 			bucket_nr++;
601 
602 			prio_io(ca, bucket, READ_SYNC);
603 
604 			if (p->csum != bch_crc64(&p->magic, bucket_bytes(ca) - 8))
605 				pr_warn("bad csum reading priorities");
606 
607 			if (p->magic != pset_magic(&ca->sb))
608 				pr_warn("bad magic reading priorities");
609 
610 			bucket = p->next_bucket;
611 			d = p->data;
612 		}
613 
614 		b->prio = le16_to_cpu(d->prio);
615 		b->gen = b->last_gc = d->gen;
616 	}
617 }
618 
619 /* Bcache device */
620 
621 static int open_dev(struct block_device *b, fmode_t mode)
622 {
623 	struct bcache_device *d = b->bd_disk->private_data;
624 	if (test_bit(BCACHE_DEV_CLOSING, &d->flags))
625 		return -ENXIO;
626 
627 	closure_get(&d->cl);
628 	return 0;
629 }
630 
631 static void release_dev(struct gendisk *b, fmode_t mode)
632 {
633 	struct bcache_device *d = b->private_data;
634 	closure_put(&d->cl);
635 }
636 
637 static int ioctl_dev(struct block_device *b, fmode_t mode,
638 		     unsigned int cmd, unsigned long arg)
639 {
640 	struct bcache_device *d = b->bd_disk->private_data;
641 	return d->ioctl(d, mode, cmd, arg);
642 }
643 
644 static const struct block_device_operations bcache_ops = {
645 	.open		= open_dev,
646 	.release	= release_dev,
647 	.ioctl		= ioctl_dev,
648 	.owner		= THIS_MODULE,
649 };
650 
651 void bcache_device_stop(struct bcache_device *d)
652 {
653 	if (!test_and_set_bit(BCACHE_DEV_CLOSING, &d->flags))
654 		closure_queue(&d->cl);
655 }
656 
657 static void bcache_device_unlink(struct bcache_device *d)
658 {
659 	lockdep_assert_held(&bch_register_lock);
660 
661 	if (d->c && !test_and_set_bit(BCACHE_DEV_UNLINK_DONE, &d->flags)) {
662 		unsigned i;
663 		struct cache *ca;
664 
665 		sysfs_remove_link(&d->c->kobj, d->name);
666 		sysfs_remove_link(&d->kobj, "cache");
667 
668 		for_each_cache(ca, d->c, i)
669 			bd_unlink_disk_holder(ca->bdev, d->disk);
670 	}
671 }
672 
673 static void bcache_device_link(struct bcache_device *d, struct cache_set *c,
674 			       const char *name)
675 {
676 	unsigned i;
677 	struct cache *ca;
678 
679 	for_each_cache(ca, d->c, i)
680 		bd_link_disk_holder(ca->bdev, d->disk);
681 
682 	snprintf(d->name, BCACHEDEVNAME_SIZE,
683 		 "%s%u", name, d->id);
684 
685 	WARN(sysfs_create_link(&d->kobj, &c->kobj, "cache") ||
686 	     sysfs_create_link(&c->kobj, &d->kobj, d->name),
687 	     "Couldn't create device <-> cache set symlinks");
688 
689 	clear_bit(BCACHE_DEV_UNLINK_DONE, &d->flags);
690 }
691 
692 static void bcache_device_detach(struct bcache_device *d)
693 {
694 	lockdep_assert_held(&bch_register_lock);
695 
696 	if (test_bit(BCACHE_DEV_DETACHING, &d->flags)) {
697 		struct uuid_entry *u = d->c->uuids + d->id;
698 
699 		SET_UUID_FLASH_ONLY(u, 0);
700 		memcpy(u->uuid, invalid_uuid, 16);
701 		u->invalidated = cpu_to_le32(get_seconds());
702 		bch_uuid_write(d->c);
703 	}
704 
705 	bcache_device_unlink(d);
706 
707 	d->c->devices[d->id] = NULL;
708 	closure_put(&d->c->caching);
709 	d->c = NULL;
710 }
711 
712 static void bcache_device_attach(struct bcache_device *d, struct cache_set *c,
713 				 unsigned id)
714 {
715 	d->id = id;
716 	d->c = c;
717 	c->devices[id] = d;
718 
719 	closure_get(&c->caching);
720 }
721 
722 static void bcache_device_free(struct bcache_device *d)
723 {
724 	lockdep_assert_held(&bch_register_lock);
725 
726 	pr_info("%s stopped", d->disk->disk_name);
727 
728 	if (d->c)
729 		bcache_device_detach(d);
730 	if (d->disk && d->disk->flags & GENHD_FL_UP)
731 		del_gendisk(d->disk);
732 	if (d->disk && d->disk->queue)
733 		blk_cleanup_queue(d->disk->queue);
734 	if (d->disk) {
735 		ida_simple_remove(&bcache_minor, d->disk->first_minor);
736 		put_disk(d->disk);
737 	}
738 
739 	if (d->bio_split)
740 		bioset_free(d->bio_split);
741 	kvfree(d->full_dirty_stripes);
742 	kvfree(d->stripe_sectors_dirty);
743 
744 	closure_debug_destroy(&d->cl);
745 }
746 
747 static int bcache_device_init(struct bcache_device *d, unsigned block_size,
748 			      sector_t sectors)
749 {
750 	struct request_queue *q;
751 	size_t n;
752 	int minor;
753 
754 	if (!d->stripe_size)
755 		d->stripe_size = 1 << 31;
756 
757 	d->nr_stripes = DIV_ROUND_UP_ULL(sectors, d->stripe_size);
758 
759 	if (!d->nr_stripes ||
760 	    d->nr_stripes > INT_MAX ||
761 	    d->nr_stripes > SIZE_MAX / sizeof(atomic_t)) {
762 		pr_err("nr_stripes too large");
763 		return -ENOMEM;
764 	}
765 
766 	n = d->nr_stripes * sizeof(atomic_t);
767 	d->stripe_sectors_dirty = n < PAGE_SIZE << 6
768 		? kzalloc(n, GFP_KERNEL)
769 		: vzalloc(n);
770 	if (!d->stripe_sectors_dirty)
771 		return -ENOMEM;
772 
773 	n = BITS_TO_LONGS(d->nr_stripes) * sizeof(unsigned long);
774 	d->full_dirty_stripes = n < PAGE_SIZE << 6
775 		? kzalloc(n, GFP_KERNEL)
776 		: vzalloc(n);
777 	if (!d->full_dirty_stripes)
778 		return -ENOMEM;
779 
780 	minor = ida_simple_get(&bcache_minor, 0, MINORMASK + 1, GFP_KERNEL);
781 	if (minor < 0)
782 		return minor;
783 
784 	if (!(d->bio_split = bioset_create(4, offsetof(struct bbio, bio))) ||
785 	    !(d->disk = alloc_disk(1))) {
786 		ida_simple_remove(&bcache_minor, minor);
787 		return -ENOMEM;
788 	}
789 
790 	set_capacity(d->disk, sectors);
791 	snprintf(d->disk->disk_name, DISK_NAME_LEN, "bcache%i", minor);
792 
793 	d->disk->major		= bcache_major;
794 	d->disk->first_minor	= minor;
795 	d->disk->fops		= &bcache_ops;
796 	d->disk->private_data	= d;
797 
798 	q = blk_alloc_queue(GFP_KERNEL);
799 	if (!q)
800 		return -ENOMEM;
801 
802 	blk_queue_make_request(q, NULL);
803 	d->disk->queue			= q;
804 	q->queuedata			= d;
805 	q->backing_dev_info.congested_data = d;
806 	q->limits.max_hw_sectors	= UINT_MAX;
807 	q->limits.max_sectors		= UINT_MAX;
808 	q->limits.max_segment_size	= UINT_MAX;
809 	q->limits.max_segments		= BIO_MAX_PAGES;
810 	blk_queue_max_discard_sectors(q, UINT_MAX);
811 	q->limits.discard_granularity	= 512;
812 	q->limits.io_min		= block_size;
813 	q->limits.logical_block_size	= block_size;
814 	q->limits.physical_block_size	= block_size;
815 	set_bit(QUEUE_FLAG_NONROT,	&d->disk->queue->queue_flags);
816 	clear_bit(QUEUE_FLAG_ADD_RANDOM, &d->disk->queue->queue_flags);
817 	set_bit(QUEUE_FLAG_DISCARD,	&d->disk->queue->queue_flags);
818 
819 	blk_queue_flush(q, REQ_FLUSH|REQ_FUA);
820 
821 	return 0;
822 }
823 
824 /* Cached device */
825 
826 static void calc_cached_dev_sectors(struct cache_set *c)
827 {
828 	uint64_t sectors = 0;
829 	struct cached_dev *dc;
830 
831 	list_for_each_entry(dc, &c->cached_devs, list)
832 		sectors += bdev_sectors(dc->bdev);
833 
834 	c->cached_dev_sectors = sectors;
835 }
836 
837 void bch_cached_dev_run(struct cached_dev *dc)
838 {
839 	struct bcache_device *d = &dc->disk;
840 	char buf[SB_LABEL_SIZE + 1];
841 	char *env[] = {
842 		"DRIVER=bcache",
843 		kasprintf(GFP_KERNEL, "CACHED_UUID=%pU", dc->sb.uuid),
844 		NULL,
845 		NULL,
846 	};
847 
848 	memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
849 	buf[SB_LABEL_SIZE] = '\0';
850 	env[2] = kasprintf(GFP_KERNEL, "CACHED_LABEL=%s", buf);
851 
852 	if (atomic_xchg(&dc->running, 1)) {
853 		kfree(env[1]);
854 		kfree(env[2]);
855 		return;
856 	}
857 
858 	if (!d->c &&
859 	    BDEV_STATE(&dc->sb) != BDEV_STATE_NONE) {
860 		struct closure cl;
861 		closure_init_stack(&cl);
862 
863 		SET_BDEV_STATE(&dc->sb, BDEV_STATE_STALE);
864 		bch_write_bdev_super(dc, &cl);
865 		closure_sync(&cl);
866 	}
867 
868 	add_disk(d->disk);
869 	bd_link_disk_holder(dc->bdev, dc->disk.disk);
870 	/* won't show up in the uevent file, use udevadm monitor -e instead
871 	 * only class / kset properties are persistent */
872 	kobject_uevent_env(&disk_to_dev(d->disk)->kobj, KOBJ_CHANGE, env);
873 	kfree(env[1]);
874 	kfree(env[2]);
875 
876 	if (sysfs_create_link(&d->kobj, &disk_to_dev(d->disk)->kobj, "dev") ||
877 	    sysfs_create_link(&disk_to_dev(d->disk)->kobj, &d->kobj, "bcache"))
878 		pr_debug("error creating sysfs link");
879 }
880 
881 static void cached_dev_detach_finish(struct work_struct *w)
882 {
883 	struct cached_dev *dc = container_of(w, struct cached_dev, detach);
884 	char buf[BDEVNAME_SIZE];
885 	struct closure cl;
886 	closure_init_stack(&cl);
887 
888 	BUG_ON(!test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags));
889 	BUG_ON(atomic_read(&dc->count));
890 
891 	mutex_lock(&bch_register_lock);
892 
893 	memset(&dc->sb.set_uuid, 0, 16);
894 	SET_BDEV_STATE(&dc->sb, BDEV_STATE_NONE);
895 
896 	bch_write_bdev_super(dc, &cl);
897 	closure_sync(&cl);
898 
899 	bcache_device_detach(&dc->disk);
900 	list_move(&dc->list, &uncached_devices);
901 
902 	clear_bit(BCACHE_DEV_DETACHING, &dc->disk.flags);
903 	clear_bit(BCACHE_DEV_UNLINK_DONE, &dc->disk.flags);
904 
905 	mutex_unlock(&bch_register_lock);
906 
907 	pr_info("Caching disabled for %s", bdevname(dc->bdev, buf));
908 
909 	/* Drop ref we took in cached_dev_detach() */
910 	closure_put(&dc->disk.cl);
911 }
912 
913 void bch_cached_dev_detach(struct cached_dev *dc)
914 {
915 	lockdep_assert_held(&bch_register_lock);
916 
917 	if (test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags))
918 		return;
919 
920 	if (test_and_set_bit(BCACHE_DEV_DETACHING, &dc->disk.flags))
921 		return;
922 
923 	/*
924 	 * Block the device from being closed and freed until we're finished
925 	 * detaching
926 	 */
927 	closure_get(&dc->disk.cl);
928 
929 	bch_writeback_queue(dc);
930 	cached_dev_put(dc);
931 }
932 
933 int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c)
934 {
935 	uint32_t rtime = cpu_to_le32(get_seconds());
936 	struct uuid_entry *u;
937 	char buf[BDEVNAME_SIZE];
938 
939 	bdevname(dc->bdev, buf);
940 
941 	if (memcmp(dc->sb.set_uuid, c->sb.set_uuid, 16))
942 		return -ENOENT;
943 
944 	if (dc->disk.c) {
945 		pr_err("Can't attach %s: already attached", buf);
946 		return -EINVAL;
947 	}
948 
949 	if (test_bit(CACHE_SET_STOPPING, &c->flags)) {
950 		pr_err("Can't attach %s: shutting down", buf);
951 		return -EINVAL;
952 	}
953 
954 	if (dc->sb.block_size < c->sb.block_size) {
955 		/* Will die */
956 		pr_err("Couldn't attach %s: block size less than set's block size",
957 		       buf);
958 		return -EINVAL;
959 	}
960 
961 	u = uuid_find(c, dc->sb.uuid);
962 
963 	if (u &&
964 	    (BDEV_STATE(&dc->sb) == BDEV_STATE_STALE ||
965 	     BDEV_STATE(&dc->sb) == BDEV_STATE_NONE)) {
966 		memcpy(u->uuid, invalid_uuid, 16);
967 		u->invalidated = cpu_to_le32(get_seconds());
968 		u = NULL;
969 	}
970 
971 	if (!u) {
972 		if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
973 			pr_err("Couldn't find uuid for %s in set", buf);
974 			return -ENOENT;
975 		}
976 
977 		u = uuid_find_empty(c);
978 		if (!u) {
979 			pr_err("Not caching %s, no room for UUID", buf);
980 			return -EINVAL;
981 		}
982 	}
983 
984 	/* Deadlocks since we're called via sysfs...
985 	sysfs_remove_file(&dc->kobj, &sysfs_attach);
986 	 */
987 
988 	if (bch_is_zero(u->uuid, 16)) {
989 		struct closure cl;
990 		closure_init_stack(&cl);
991 
992 		memcpy(u->uuid, dc->sb.uuid, 16);
993 		memcpy(u->label, dc->sb.label, SB_LABEL_SIZE);
994 		u->first_reg = u->last_reg = rtime;
995 		bch_uuid_write(c);
996 
997 		memcpy(dc->sb.set_uuid, c->sb.set_uuid, 16);
998 		SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN);
999 
1000 		bch_write_bdev_super(dc, &cl);
1001 		closure_sync(&cl);
1002 	} else {
1003 		u->last_reg = rtime;
1004 		bch_uuid_write(c);
1005 	}
1006 
1007 	bcache_device_attach(&dc->disk, c, u - c->uuids);
1008 	list_move(&dc->list, &c->cached_devs);
1009 	calc_cached_dev_sectors(c);
1010 
1011 	smp_wmb();
1012 	/*
1013 	 * dc->c must be set before dc->count != 0 - paired with the mb in
1014 	 * cached_dev_get()
1015 	 */
1016 	atomic_set(&dc->count, 1);
1017 
1018 	if (bch_cached_dev_writeback_start(dc))
1019 		return -ENOMEM;
1020 
1021 	if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
1022 		bch_sectors_dirty_init(dc);
1023 		atomic_set(&dc->has_dirty, 1);
1024 		atomic_inc(&dc->count);
1025 		bch_writeback_queue(dc);
1026 	}
1027 
1028 	bch_cached_dev_run(dc);
1029 	bcache_device_link(&dc->disk, c, "bdev");
1030 
1031 	pr_info("Caching %s as %s on set %pU",
1032 		bdevname(dc->bdev, buf), dc->disk.disk->disk_name,
1033 		dc->disk.c->sb.set_uuid);
1034 	return 0;
1035 }
1036 
1037 void bch_cached_dev_release(struct kobject *kobj)
1038 {
1039 	struct cached_dev *dc = container_of(kobj, struct cached_dev,
1040 					     disk.kobj);
1041 	kfree(dc);
1042 	module_put(THIS_MODULE);
1043 }
1044 
1045 static void cached_dev_free(struct closure *cl)
1046 {
1047 	struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1048 
1049 	cancel_delayed_work_sync(&dc->writeback_rate_update);
1050 	if (!IS_ERR_OR_NULL(dc->writeback_thread))
1051 		kthread_stop(dc->writeback_thread);
1052 
1053 	mutex_lock(&bch_register_lock);
1054 
1055 	if (atomic_read(&dc->running))
1056 		bd_unlink_disk_holder(dc->bdev, dc->disk.disk);
1057 	bcache_device_free(&dc->disk);
1058 	list_del(&dc->list);
1059 
1060 	mutex_unlock(&bch_register_lock);
1061 
1062 	if (!IS_ERR_OR_NULL(dc->bdev))
1063 		blkdev_put(dc->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1064 
1065 	wake_up(&unregister_wait);
1066 
1067 	kobject_put(&dc->disk.kobj);
1068 }
1069 
1070 static void cached_dev_flush(struct closure *cl)
1071 {
1072 	struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1073 	struct bcache_device *d = &dc->disk;
1074 
1075 	mutex_lock(&bch_register_lock);
1076 	bcache_device_unlink(d);
1077 	mutex_unlock(&bch_register_lock);
1078 
1079 	bch_cache_accounting_destroy(&dc->accounting);
1080 	kobject_del(&d->kobj);
1081 
1082 	continue_at(cl, cached_dev_free, system_wq);
1083 }
1084 
1085 static int cached_dev_init(struct cached_dev *dc, unsigned block_size)
1086 {
1087 	int ret;
1088 	struct io *io;
1089 	struct request_queue *q = bdev_get_queue(dc->bdev);
1090 
1091 	__module_get(THIS_MODULE);
1092 	INIT_LIST_HEAD(&dc->list);
1093 	closure_init(&dc->disk.cl, NULL);
1094 	set_closure_fn(&dc->disk.cl, cached_dev_flush, system_wq);
1095 	kobject_init(&dc->disk.kobj, &bch_cached_dev_ktype);
1096 	INIT_WORK(&dc->detach, cached_dev_detach_finish);
1097 	sema_init(&dc->sb_write_mutex, 1);
1098 	INIT_LIST_HEAD(&dc->io_lru);
1099 	spin_lock_init(&dc->io_lock);
1100 	bch_cache_accounting_init(&dc->accounting, &dc->disk.cl);
1101 
1102 	dc->sequential_cutoff		= 4 << 20;
1103 
1104 	for (io = dc->io; io < dc->io + RECENT_IO; io++) {
1105 		list_add(&io->lru, &dc->io_lru);
1106 		hlist_add_head(&io->hash, dc->io_hash + RECENT_IO);
1107 	}
1108 
1109 	dc->disk.stripe_size = q->limits.io_opt >> 9;
1110 
1111 	if (dc->disk.stripe_size)
1112 		dc->partial_stripes_expensive =
1113 			q->limits.raid_partial_stripes_expensive;
1114 
1115 	ret = bcache_device_init(&dc->disk, block_size,
1116 			 dc->bdev->bd_part->nr_sects - dc->sb.data_offset);
1117 	if (ret)
1118 		return ret;
1119 
1120 	set_capacity(dc->disk.disk,
1121 		     dc->bdev->bd_part->nr_sects - dc->sb.data_offset);
1122 
1123 	dc->disk.disk->queue->backing_dev_info.ra_pages =
1124 		max(dc->disk.disk->queue->backing_dev_info.ra_pages,
1125 		    q->backing_dev_info.ra_pages);
1126 
1127 	bch_cached_dev_request_init(dc);
1128 	bch_cached_dev_writeback_init(dc);
1129 	return 0;
1130 }
1131 
1132 /* Cached device - bcache superblock */
1133 
1134 static void register_bdev(struct cache_sb *sb, struct page *sb_page,
1135 				 struct block_device *bdev,
1136 				 struct cached_dev *dc)
1137 {
1138 	char name[BDEVNAME_SIZE];
1139 	const char *err = "cannot allocate memory";
1140 	struct cache_set *c;
1141 
1142 	memcpy(&dc->sb, sb, sizeof(struct cache_sb));
1143 	dc->bdev = bdev;
1144 	dc->bdev->bd_holder = dc;
1145 
1146 	bio_init(&dc->sb_bio);
1147 	dc->sb_bio.bi_max_vecs	= 1;
1148 	dc->sb_bio.bi_io_vec	= dc->sb_bio.bi_inline_vecs;
1149 	dc->sb_bio.bi_io_vec[0].bv_page = sb_page;
1150 	get_page(sb_page);
1151 
1152 	if (cached_dev_init(dc, sb->block_size << 9))
1153 		goto err;
1154 
1155 	err = "error creating kobject";
1156 	if (kobject_add(&dc->disk.kobj, &part_to_dev(bdev->bd_part)->kobj,
1157 			"bcache"))
1158 		goto err;
1159 	if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj))
1160 		goto err;
1161 
1162 	pr_info("registered backing device %s", bdevname(bdev, name));
1163 
1164 	list_add(&dc->list, &uncached_devices);
1165 	list_for_each_entry(c, &bch_cache_sets, list)
1166 		bch_cached_dev_attach(dc, c);
1167 
1168 	if (BDEV_STATE(&dc->sb) == BDEV_STATE_NONE ||
1169 	    BDEV_STATE(&dc->sb) == BDEV_STATE_STALE)
1170 		bch_cached_dev_run(dc);
1171 
1172 	return;
1173 err:
1174 	pr_notice("error opening %s: %s", bdevname(bdev, name), err);
1175 	bcache_device_stop(&dc->disk);
1176 }
1177 
1178 /* Flash only volumes */
1179 
1180 void bch_flash_dev_release(struct kobject *kobj)
1181 {
1182 	struct bcache_device *d = container_of(kobj, struct bcache_device,
1183 					       kobj);
1184 	kfree(d);
1185 }
1186 
1187 static void flash_dev_free(struct closure *cl)
1188 {
1189 	struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1190 	mutex_lock(&bch_register_lock);
1191 	bcache_device_free(d);
1192 	mutex_unlock(&bch_register_lock);
1193 	kobject_put(&d->kobj);
1194 }
1195 
1196 static void flash_dev_flush(struct closure *cl)
1197 {
1198 	struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1199 
1200 	mutex_lock(&bch_register_lock);
1201 	bcache_device_unlink(d);
1202 	mutex_unlock(&bch_register_lock);
1203 	kobject_del(&d->kobj);
1204 	continue_at(cl, flash_dev_free, system_wq);
1205 }
1206 
1207 static int flash_dev_run(struct cache_set *c, struct uuid_entry *u)
1208 {
1209 	struct bcache_device *d = kzalloc(sizeof(struct bcache_device),
1210 					  GFP_KERNEL);
1211 	if (!d)
1212 		return -ENOMEM;
1213 
1214 	closure_init(&d->cl, NULL);
1215 	set_closure_fn(&d->cl, flash_dev_flush, system_wq);
1216 
1217 	kobject_init(&d->kobj, &bch_flash_dev_ktype);
1218 
1219 	if (bcache_device_init(d, block_bytes(c), u->sectors))
1220 		goto err;
1221 
1222 	bcache_device_attach(d, c, u - c->uuids);
1223 	bch_flash_dev_request_init(d);
1224 	add_disk(d->disk);
1225 
1226 	if (kobject_add(&d->kobj, &disk_to_dev(d->disk)->kobj, "bcache"))
1227 		goto err;
1228 
1229 	bcache_device_link(d, c, "volume");
1230 
1231 	return 0;
1232 err:
1233 	kobject_put(&d->kobj);
1234 	return -ENOMEM;
1235 }
1236 
1237 static int flash_devs_run(struct cache_set *c)
1238 {
1239 	int ret = 0;
1240 	struct uuid_entry *u;
1241 
1242 	for (u = c->uuids;
1243 	     u < c->uuids + c->nr_uuids && !ret;
1244 	     u++)
1245 		if (UUID_FLASH_ONLY(u))
1246 			ret = flash_dev_run(c, u);
1247 
1248 	return ret;
1249 }
1250 
1251 int bch_flash_dev_create(struct cache_set *c, uint64_t size)
1252 {
1253 	struct uuid_entry *u;
1254 
1255 	if (test_bit(CACHE_SET_STOPPING, &c->flags))
1256 		return -EINTR;
1257 
1258 	if (!test_bit(CACHE_SET_RUNNING, &c->flags))
1259 		return -EPERM;
1260 
1261 	u = uuid_find_empty(c);
1262 	if (!u) {
1263 		pr_err("Can't create volume, no room for UUID");
1264 		return -EINVAL;
1265 	}
1266 
1267 	get_random_bytes(u->uuid, 16);
1268 	memset(u->label, 0, 32);
1269 	u->first_reg = u->last_reg = cpu_to_le32(get_seconds());
1270 
1271 	SET_UUID_FLASH_ONLY(u, 1);
1272 	u->sectors = size >> 9;
1273 
1274 	bch_uuid_write(c);
1275 
1276 	return flash_dev_run(c, u);
1277 }
1278 
1279 /* Cache set */
1280 
1281 __printf(2, 3)
1282 bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...)
1283 {
1284 	va_list args;
1285 
1286 	if (c->on_error != ON_ERROR_PANIC &&
1287 	    test_bit(CACHE_SET_STOPPING, &c->flags))
1288 		return false;
1289 
1290 	/* XXX: we can be called from atomic context
1291 	acquire_console_sem();
1292 	*/
1293 
1294 	printk(KERN_ERR "bcache: error on %pU: ", c->sb.set_uuid);
1295 
1296 	va_start(args, fmt);
1297 	vprintk(fmt, args);
1298 	va_end(args);
1299 
1300 	printk(", disabling caching\n");
1301 
1302 	if (c->on_error == ON_ERROR_PANIC)
1303 		panic("panic forced after error\n");
1304 
1305 	bch_cache_set_unregister(c);
1306 	return true;
1307 }
1308 
1309 void bch_cache_set_release(struct kobject *kobj)
1310 {
1311 	struct cache_set *c = container_of(kobj, struct cache_set, kobj);
1312 	kfree(c);
1313 	module_put(THIS_MODULE);
1314 }
1315 
1316 static void cache_set_free(struct closure *cl)
1317 {
1318 	struct cache_set *c = container_of(cl, struct cache_set, cl);
1319 	struct cache *ca;
1320 	unsigned i;
1321 
1322 	if (!IS_ERR_OR_NULL(c->debug))
1323 		debugfs_remove(c->debug);
1324 
1325 	bch_open_buckets_free(c);
1326 	bch_btree_cache_free(c);
1327 	bch_journal_free(c);
1328 
1329 	for_each_cache(ca, c, i)
1330 		if (ca) {
1331 			ca->set = NULL;
1332 			c->cache[ca->sb.nr_this_dev] = NULL;
1333 			kobject_put(&ca->kobj);
1334 		}
1335 
1336 	bch_bset_sort_state_free(&c->sort);
1337 	free_pages((unsigned long) c->uuids, ilog2(bucket_pages(c)));
1338 
1339 	if (c->moving_gc_wq)
1340 		destroy_workqueue(c->moving_gc_wq);
1341 	if (c->bio_split)
1342 		bioset_free(c->bio_split);
1343 	if (c->fill_iter)
1344 		mempool_destroy(c->fill_iter);
1345 	if (c->bio_meta)
1346 		mempool_destroy(c->bio_meta);
1347 	if (c->search)
1348 		mempool_destroy(c->search);
1349 	kfree(c->devices);
1350 
1351 	mutex_lock(&bch_register_lock);
1352 	list_del(&c->list);
1353 	mutex_unlock(&bch_register_lock);
1354 
1355 	pr_info("Cache set %pU unregistered", c->sb.set_uuid);
1356 	wake_up(&unregister_wait);
1357 
1358 	closure_debug_destroy(&c->cl);
1359 	kobject_put(&c->kobj);
1360 }
1361 
1362 static void cache_set_flush(struct closure *cl)
1363 {
1364 	struct cache_set *c = container_of(cl, struct cache_set, caching);
1365 	struct cache *ca;
1366 	struct btree *b;
1367 	unsigned i;
1368 
1369 	bch_cache_accounting_destroy(&c->accounting);
1370 
1371 	kobject_put(&c->internal);
1372 	kobject_del(&c->kobj);
1373 
1374 	if (c->gc_thread)
1375 		kthread_stop(c->gc_thread);
1376 
1377 	if (!IS_ERR_OR_NULL(c->root))
1378 		list_add(&c->root->list, &c->btree_cache);
1379 
1380 	/* Should skip this if we're unregistering because of an error */
1381 	list_for_each_entry(b, &c->btree_cache, list) {
1382 		mutex_lock(&b->write_lock);
1383 		if (btree_node_dirty(b))
1384 			__bch_btree_node_write(b, NULL);
1385 		mutex_unlock(&b->write_lock);
1386 	}
1387 
1388 	for_each_cache(ca, c, i)
1389 		if (ca->alloc_thread)
1390 			kthread_stop(ca->alloc_thread);
1391 
1392 	if (c->journal.cur) {
1393 		cancel_delayed_work_sync(&c->journal.work);
1394 		/* flush last journal entry if needed */
1395 		c->journal.work.work.func(&c->journal.work.work);
1396 	}
1397 
1398 	closure_return(cl);
1399 }
1400 
1401 static void __cache_set_unregister(struct closure *cl)
1402 {
1403 	struct cache_set *c = container_of(cl, struct cache_set, caching);
1404 	struct cached_dev *dc;
1405 	size_t i;
1406 
1407 	mutex_lock(&bch_register_lock);
1408 
1409 	for (i = 0; i < c->nr_uuids; i++)
1410 		if (c->devices[i]) {
1411 			if (!UUID_FLASH_ONLY(&c->uuids[i]) &&
1412 			    test_bit(CACHE_SET_UNREGISTERING, &c->flags)) {
1413 				dc = container_of(c->devices[i],
1414 						  struct cached_dev, disk);
1415 				bch_cached_dev_detach(dc);
1416 			} else {
1417 				bcache_device_stop(c->devices[i]);
1418 			}
1419 		}
1420 
1421 	mutex_unlock(&bch_register_lock);
1422 
1423 	continue_at(cl, cache_set_flush, system_wq);
1424 }
1425 
1426 void bch_cache_set_stop(struct cache_set *c)
1427 {
1428 	if (!test_and_set_bit(CACHE_SET_STOPPING, &c->flags))
1429 		closure_queue(&c->caching);
1430 }
1431 
1432 void bch_cache_set_unregister(struct cache_set *c)
1433 {
1434 	set_bit(CACHE_SET_UNREGISTERING, &c->flags);
1435 	bch_cache_set_stop(c);
1436 }
1437 
1438 #define alloc_bucket_pages(gfp, c)			\
1439 	((void *) __get_free_pages(__GFP_ZERO|gfp, ilog2(bucket_pages(c))))
1440 
1441 struct cache_set *bch_cache_set_alloc(struct cache_sb *sb)
1442 {
1443 	int iter_size;
1444 	struct cache_set *c = kzalloc(sizeof(struct cache_set), GFP_KERNEL);
1445 	if (!c)
1446 		return NULL;
1447 
1448 	__module_get(THIS_MODULE);
1449 	closure_init(&c->cl, NULL);
1450 	set_closure_fn(&c->cl, cache_set_free, system_wq);
1451 
1452 	closure_init(&c->caching, &c->cl);
1453 	set_closure_fn(&c->caching, __cache_set_unregister, system_wq);
1454 
1455 	/* Maybe create continue_at_noreturn() and use it here? */
1456 	closure_set_stopped(&c->cl);
1457 	closure_put(&c->cl);
1458 
1459 	kobject_init(&c->kobj, &bch_cache_set_ktype);
1460 	kobject_init(&c->internal, &bch_cache_set_internal_ktype);
1461 
1462 	bch_cache_accounting_init(&c->accounting, &c->cl);
1463 
1464 	memcpy(c->sb.set_uuid, sb->set_uuid, 16);
1465 	c->sb.block_size	= sb->block_size;
1466 	c->sb.bucket_size	= sb->bucket_size;
1467 	c->sb.nr_in_set		= sb->nr_in_set;
1468 	c->sb.last_mount	= sb->last_mount;
1469 	c->bucket_bits		= ilog2(sb->bucket_size);
1470 	c->block_bits		= ilog2(sb->block_size);
1471 	c->nr_uuids		= bucket_bytes(c) / sizeof(struct uuid_entry);
1472 
1473 	c->btree_pages		= bucket_pages(c);
1474 	if (c->btree_pages > BTREE_MAX_PAGES)
1475 		c->btree_pages = max_t(int, c->btree_pages / 4,
1476 				       BTREE_MAX_PAGES);
1477 
1478 	sema_init(&c->sb_write_mutex, 1);
1479 	mutex_init(&c->bucket_lock);
1480 	init_waitqueue_head(&c->btree_cache_wait);
1481 	init_waitqueue_head(&c->bucket_wait);
1482 	sema_init(&c->uuid_write_mutex, 1);
1483 
1484 	spin_lock_init(&c->btree_gc_time.lock);
1485 	spin_lock_init(&c->btree_split_time.lock);
1486 	spin_lock_init(&c->btree_read_time.lock);
1487 
1488 	bch_moving_init_cache_set(c);
1489 
1490 	INIT_LIST_HEAD(&c->list);
1491 	INIT_LIST_HEAD(&c->cached_devs);
1492 	INIT_LIST_HEAD(&c->btree_cache);
1493 	INIT_LIST_HEAD(&c->btree_cache_freeable);
1494 	INIT_LIST_HEAD(&c->btree_cache_freed);
1495 	INIT_LIST_HEAD(&c->data_buckets);
1496 
1497 	c->search = mempool_create_slab_pool(32, bch_search_cache);
1498 	if (!c->search)
1499 		goto err;
1500 
1501 	iter_size = (sb->bucket_size / sb->block_size + 1) *
1502 		sizeof(struct btree_iter_set);
1503 
1504 	if (!(c->devices = kzalloc(c->nr_uuids * sizeof(void *), GFP_KERNEL)) ||
1505 	    !(c->bio_meta = mempool_create_kmalloc_pool(2,
1506 				sizeof(struct bbio) + sizeof(struct bio_vec) *
1507 				bucket_pages(c))) ||
1508 	    !(c->fill_iter = mempool_create_kmalloc_pool(1, iter_size)) ||
1509 	    !(c->bio_split = bioset_create(4, offsetof(struct bbio, bio))) ||
1510 	    !(c->uuids = alloc_bucket_pages(GFP_KERNEL, c)) ||
1511 	    !(c->moving_gc_wq = create_workqueue("bcache_gc")) ||
1512 	    bch_journal_alloc(c) ||
1513 	    bch_btree_cache_alloc(c) ||
1514 	    bch_open_buckets_alloc(c) ||
1515 	    bch_bset_sort_state_init(&c->sort, ilog2(c->btree_pages)))
1516 		goto err;
1517 
1518 	c->congested_read_threshold_us	= 2000;
1519 	c->congested_write_threshold_us	= 20000;
1520 	c->error_limit	= 8 << IO_ERROR_SHIFT;
1521 
1522 	return c;
1523 err:
1524 	bch_cache_set_unregister(c);
1525 	return NULL;
1526 }
1527 
1528 static void run_cache_set(struct cache_set *c)
1529 {
1530 	const char *err = "cannot allocate memory";
1531 	struct cached_dev *dc, *t;
1532 	struct cache *ca;
1533 	struct closure cl;
1534 	unsigned i;
1535 
1536 	closure_init_stack(&cl);
1537 
1538 	for_each_cache(ca, c, i)
1539 		c->nbuckets += ca->sb.nbuckets;
1540 
1541 	if (CACHE_SYNC(&c->sb)) {
1542 		LIST_HEAD(journal);
1543 		struct bkey *k;
1544 		struct jset *j;
1545 
1546 		err = "cannot allocate memory for journal";
1547 		if (bch_journal_read(c, &journal))
1548 			goto err;
1549 
1550 		pr_debug("btree_journal_read() done");
1551 
1552 		err = "no journal entries found";
1553 		if (list_empty(&journal))
1554 			goto err;
1555 
1556 		j = &list_entry(journal.prev, struct journal_replay, list)->j;
1557 
1558 		err = "IO error reading priorities";
1559 		for_each_cache(ca, c, i)
1560 			prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev]);
1561 
1562 		/*
1563 		 * If prio_read() fails it'll call cache_set_error and we'll
1564 		 * tear everything down right away, but if we perhaps checked
1565 		 * sooner we could avoid journal replay.
1566 		 */
1567 
1568 		k = &j->btree_root;
1569 
1570 		err = "bad btree root";
1571 		if (__bch_btree_ptr_invalid(c, k))
1572 			goto err;
1573 
1574 		err = "error reading btree root";
1575 		c->root = bch_btree_node_get(c, NULL, k, j->btree_level, true, NULL);
1576 		if (IS_ERR_OR_NULL(c->root))
1577 			goto err;
1578 
1579 		list_del_init(&c->root->list);
1580 		rw_unlock(true, c->root);
1581 
1582 		err = uuid_read(c, j, &cl);
1583 		if (err)
1584 			goto err;
1585 
1586 		err = "error in recovery";
1587 		if (bch_btree_check(c))
1588 			goto err;
1589 
1590 		bch_journal_mark(c, &journal);
1591 		bch_initial_gc_finish(c);
1592 		pr_debug("btree_check() done");
1593 
1594 		/*
1595 		 * bcache_journal_next() can't happen sooner, or
1596 		 * btree_gc_finish() will give spurious errors about last_gc >
1597 		 * gc_gen - this is a hack but oh well.
1598 		 */
1599 		bch_journal_next(&c->journal);
1600 
1601 		err = "error starting allocator thread";
1602 		for_each_cache(ca, c, i)
1603 			if (bch_cache_allocator_start(ca))
1604 				goto err;
1605 
1606 		/*
1607 		 * First place it's safe to allocate: btree_check() and
1608 		 * btree_gc_finish() have to run before we have buckets to
1609 		 * allocate, and bch_bucket_alloc_set() might cause a journal
1610 		 * entry to be written so bcache_journal_next() has to be called
1611 		 * first.
1612 		 *
1613 		 * If the uuids were in the old format we have to rewrite them
1614 		 * before the next journal entry is written:
1615 		 */
1616 		if (j->version < BCACHE_JSET_VERSION_UUID)
1617 			__uuid_write(c);
1618 
1619 		bch_journal_replay(c, &journal);
1620 	} else {
1621 		pr_notice("invalidating existing data");
1622 
1623 		for_each_cache(ca, c, i) {
1624 			unsigned j;
1625 
1626 			ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7,
1627 					      2, SB_JOURNAL_BUCKETS);
1628 
1629 			for (j = 0; j < ca->sb.keys; j++)
1630 				ca->sb.d[j] = ca->sb.first_bucket + j;
1631 		}
1632 
1633 		bch_initial_gc_finish(c);
1634 
1635 		err = "error starting allocator thread";
1636 		for_each_cache(ca, c, i)
1637 			if (bch_cache_allocator_start(ca))
1638 				goto err;
1639 
1640 		mutex_lock(&c->bucket_lock);
1641 		for_each_cache(ca, c, i)
1642 			bch_prio_write(ca);
1643 		mutex_unlock(&c->bucket_lock);
1644 
1645 		err = "cannot allocate new UUID bucket";
1646 		if (__uuid_write(c))
1647 			goto err;
1648 
1649 		err = "cannot allocate new btree root";
1650 		c->root = __bch_btree_node_alloc(c, NULL, 0, true, NULL);
1651 		if (IS_ERR_OR_NULL(c->root))
1652 			goto err;
1653 
1654 		mutex_lock(&c->root->write_lock);
1655 		bkey_copy_key(&c->root->key, &MAX_KEY);
1656 		bch_btree_node_write(c->root, &cl);
1657 		mutex_unlock(&c->root->write_lock);
1658 
1659 		bch_btree_set_root(c->root);
1660 		rw_unlock(true, c->root);
1661 
1662 		/*
1663 		 * We don't want to write the first journal entry until
1664 		 * everything is set up - fortunately journal entries won't be
1665 		 * written until the SET_CACHE_SYNC() here:
1666 		 */
1667 		SET_CACHE_SYNC(&c->sb, true);
1668 
1669 		bch_journal_next(&c->journal);
1670 		bch_journal_meta(c, &cl);
1671 	}
1672 
1673 	err = "error starting gc thread";
1674 	if (bch_gc_thread_start(c))
1675 		goto err;
1676 
1677 	closure_sync(&cl);
1678 	c->sb.last_mount = get_seconds();
1679 	bcache_write_super(c);
1680 
1681 	list_for_each_entry_safe(dc, t, &uncached_devices, list)
1682 		bch_cached_dev_attach(dc, c);
1683 
1684 	flash_devs_run(c);
1685 
1686 	set_bit(CACHE_SET_RUNNING, &c->flags);
1687 	return;
1688 err:
1689 	closure_sync(&cl);
1690 	/* XXX: test this, it's broken */
1691 	bch_cache_set_error(c, "%s", err);
1692 }
1693 
1694 static bool can_attach_cache(struct cache *ca, struct cache_set *c)
1695 {
1696 	return ca->sb.block_size	== c->sb.block_size &&
1697 		ca->sb.bucket_size	== c->sb.bucket_size &&
1698 		ca->sb.nr_in_set	== c->sb.nr_in_set;
1699 }
1700 
1701 static const char *register_cache_set(struct cache *ca)
1702 {
1703 	char buf[12];
1704 	const char *err = "cannot allocate memory";
1705 	struct cache_set *c;
1706 
1707 	list_for_each_entry(c, &bch_cache_sets, list)
1708 		if (!memcmp(c->sb.set_uuid, ca->sb.set_uuid, 16)) {
1709 			if (c->cache[ca->sb.nr_this_dev])
1710 				return "duplicate cache set member";
1711 
1712 			if (!can_attach_cache(ca, c))
1713 				return "cache sb does not match set";
1714 
1715 			if (!CACHE_SYNC(&ca->sb))
1716 				SET_CACHE_SYNC(&c->sb, false);
1717 
1718 			goto found;
1719 		}
1720 
1721 	c = bch_cache_set_alloc(&ca->sb);
1722 	if (!c)
1723 		return err;
1724 
1725 	err = "error creating kobject";
1726 	if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->sb.set_uuid) ||
1727 	    kobject_add(&c->internal, &c->kobj, "internal"))
1728 		goto err;
1729 
1730 	if (bch_cache_accounting_add_kobjs(&c->accounting, &c->kobj))
1731 		goto err;
1732 
1733 	bch_debug_init_cache_set(c);
1734 
1735 	list_add(&c->list, &bch_cache_sets);
1736 found:
1737 	sprintf(buf, "cache%i", ca->sb.nr_this_dev);
1738 	if (sysfs_create_link(&ca->kobj, &c->kobj, "set") ||
1739 	    sysfs_create_link(&c->kobj, &ca->kobj, buf))
1740 		goto err;
1741 
1742 	if (ca->sb.seq > c->sb.seq) {
1743 		c->sb.version		= ca->sb.version;
1744 		memcpy(c->sb.set_uuid, ca->sb.set_uuid, 16);
1745 		c->sb.flags             = ca->sb.flags;
1746 		c->sb.seq		= ca->sb.seq;
1747 		pr_debug("set version = %llu", c->sb.version);
1748 	}
1749 
1750 	kobject_get(&ca->kobj);
1751 	ca->set = c;
1752 	ca->set->cache[ca->sb.nr_this_dev] = ca;
1753 	c->cache_by_alloc[c->caches_loaded++] = ca;
1754 
1755 	if (c->caches_loaded == c->sb.nr_in_set)
1756 		run_cache_set(c);
1757 
1758 	return NULL;
1759 err:
1760 	bch_cache_set_unregister(c);
1761 	return err;
1762 }
1763 
1764 /* Cache device */
1765 
1766 void bch_cache_release(struct kobject *kobj)
1767 {
1768 	struct cache *ca = container_of(kobj, struct cache, kobj);
1769 	unsigned i;
1770 
1771 	if (ca->set) {
1772 		BUG_ON(ca->set->cache[ca->sb.nr_this_dev] != ca);
1773 		ca->set->cache[ca->sb.nr_this_dev] = NULL;
1774 	}
1775 
1776 	free_pages((unsigned long) ca->disk_buckets, ilog2(bucket_pages(ca)));
1777 	kfree(ca->prio_buckets);
1778 	vfree(ca->buckets);
1779 
1780 	free_heap(&ca->heap);
1781 	free_fifo(&ca->free_inc);
1782 
1783 	for (i = 0; i < RESERVE_NR; i++)
1784 		free_fifo(&ca->free[i]);
1785 
1786 	if (ca->sb_bio.bi_inline_vecs[0].bv_page)
1787 		put_page(ca->sb_bio.bi_io_vec[0].bv_page);
1788 
1789 	if (!IS_ERR_OR_NULL(ca->bdev))
1790 		blkdev_put(ca->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1791 
1792 	kfree(ca);
1793 	module_put(THIS_MODULE);
1794 }
1795 
1796 static int cache_alloc(struct cache_sb *sb, struct cache *ca)
1797 {
1798 	size_t free;
1799 	struct bucket *b;
1800 
1801 	__module_get(THIS_MODULE);
1802 	kobject_init(&ca->kobj, &bch_cache_ktype);
1803 
1804 	bio_init(&ca->journal.bio);
1805 	ca->journal.bio.bi_max_vecs = 8;
1806 	ca->journal.bio.bi_io_vec = ca->journal.bio.bi_inline_vecs;
1807 
1808 	free = roundup_pow_of_two(ca->sb.nbuckets) >> 10;
1809 
1810 	if (!init_fifo(&ca->free[RESERVE_BTREE], 8, GFP_KERNEL) ||
1811 	    !init_fifo(&ca->free[RESERVE_PRIO], prio_buckets(ca), GFP_KERNEL) ||
1812 	    !init_fifo(&ca->free[RESERVE_MOVINGGC], free, GFP_KERNEL) ||
1813 	    !init_fifo(&ca->free[RESERVE_NONE], free, GFP_KERNEL) ||
1814 	    !init_fifo(&ca->free_inc,	free << 2, GFP_KERNEL) ||
1815 	    !init_heap(&ca->heap,	free << 3, GFP_KERNEL) ||
1816 	    !(ca->buckets	= vzalloc(sizeof(struct bucket) *
1817 					  ca->sb.nbuckets)) ||
1818 	    !(ca->prio_buckets	= kzalloc(sizeof(uint64_t) * prio_buckets(ca) *
1819 					  2, GFP_KERNEL)) ||
1820 	    !(ca->disk_buckets	= alloc_bucket_pages(GFP_KERNEL, ca)))
1821 		return -ENOMEM;
1822 
1823 	ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca);
1824 
1825 	for_each_bucket(b, ca)
1826 		atomic_set(&b->pin, 0);
1827 
1828 	return 0;
1829 }
1830 
1831 static void register_cache(struct cache_sb *sb, struct page *sb_page,
1832 				struct block_device *bdev, struct cache *ca)
1833 {
1834 	char name[BDEVNAME_SIZE];
1835 	const char *err = "cannot allocate memory";
1836 
1837 	memcpy(&ca->sb, sb, sizeof(struct cache_sb));
1838 	ca->bdev = bdev;
1839 	ca->bdev->bd_holder = ca;
1840 
1841 	bio_init(&ca->sb_bio);
1842 	ca->sb_bio.bi_max_vecs	= 1;
1843 	ca->sb_bio.bi_io_vec	= ca->sb_bio.bi_inline_vecs;
1844 	ca->sb_bio.bi_io_vec[0].bv_page = sb_page;
1845 	get_page(sb_page);
1846 
1847 	if (blk_queue_discard(bdev_get_queue(ca->bdev)))
1848 		ca->discard = CACHE_DISCARD(&ca->sb);
1849 
1850 	if (cache_alloc(sb, ca) != 0)
1851 		goto err;
1852 
1853 	err = "error creating kobject";
1854 	if (kobject_add(&ca->kobj, &part_to_dev(bdev->bd_part)->kobj, "bcache"))
1855 		goto err;
1856 
1857 	mutex_lock(&bch_register_lock);
1858 	err = register_cache_set(ca);
1859 	mutex_unlock(&bch_register_lock);
1860 
1861 	if (err)
1862 		goto err;
1863 
1864 	pr_info("registered cache device %s", bdevname(bdev, name));
1865 out:
1866 	kobject_put(&ca->kobj);
1867 	return;
1868 err:
1869 	pr_notice("error opening %s: %s", bdevname(bdev, name), err);
1870 	goto out;
1871 }
1872 
1873 /* Global interfaces/init */
1874 
1875 static ssize_t register_bcache(struct kobject *, struct kobj_attribute *,
1876 			       const char *, size_t);
1877 
1878 kobj_attribute_write(register,		register_bcache);
1879 kobj_attribute_write(register_quiet,	register_bcache);
1880 
1881 static bool bch_is_open_backing(struct block_device *bdev) {
1882 	struct cache_set *c, *tc;
1883 	struct cached_dev *dc, *t;
1884 
1885 	list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
1886 		list_for_each_entry_safe(dc, t, &c->cached_devs, list)
1887 			if (dc->bdev == bdev)
1888 				return true;
1889 	list_for_each_entry_safe(dc, t, &uncached_devices, list)
1890 		if (dc->bdev == bdev)
1891 			return true;
1892 	return false;
1893 }
1894 
1895 static bool bch_is_open_cache(struct block_device *bdev) {
1896 	struct cache_set *c, *tc;
1897 	struct cache *ca;
1898 	unsigned i;
1899 
1900 	list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
1901 		for_each_cache(ca, c, i)
1902 			if (ca->bdev == bdev)
1903 				return true;
1904 	return false;
1905 }
1906 
1907 static bool bch_is_open(struct block_device *bdev) {
1908 	return bch_is_open_cache(bdev) || bch_is_open_backing(bdev);
1909 }
1910 
1911 static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr,
1912 			       const char *buffer, size_t size)
1913 {
1914 	ssize_t ret = size;
1915 	const char *err = "cannot allocate memory";
1916 	char *path = NULL;
1917 	struct cache_sb *sb = NULL;
1918 	struct block_device *bdev = NULL;
1919 	struct page *sb_page = NULL;
1920 
1921 	if (!try_module_get(THIS_MODULE))
1922 		return -EBUSY;
1923 
1924 	if (!(path = kstrndup(buffer, size, GFP_KERNEL)) ||
1925 	    !(sb = kmalloc(sizeof(struct cache_sb), GFP_KERNEL)))
1926 		goto err;
1927 
1928 	err = "failed to open device";
1929 	bdev = blkdev_get_by_path(strim(path),
1930 				  FMODE_READ|FMODE_WRITE|FMODE_EXCL,
1931 				  sb);
1932 	if (IS_ERR(bdev)) {
1933 		if (bdev == ERR_PTR(-EBUSY)) {
1934 			bdev = lookup_bdev(strim(path));
1935 			mutex_lock(&bch_register_lock);
1936 			if (!IS_ERR(bdev) && bch_is_open(bdev))
1937 				err = "device already registered";
1938 			else
1939 				err = "device busy";
1940 			mutex_unlock(&bch_register_lock);
1941 			if (attr == &ksysfs_register_quiet)
1942 				goto out;
1943 		}
1944 		goto err;
1945 	}
1946 
1947 	err = "failed to set blocksize";
1948 	if (set_blocksize(bdev, 4096))
1949 		goto err_close;
1950 
1951 	err = read_super(sb, bdev, &sb_page);
1952 	if (err)
1953 		goto err_close;
1954 
1955 	if (SB_IS_BDEV(sb)) {
1956 		struct cached_dev *dc = kzalloc(sizeof(*dc), GFP_KERNEL);
1957 		if (!dc)
1958 			goto err_close;
1959 
1960 		mutex_lock(&bch_register_lock);
1961 		register_bdev(sb, sb_page, bdev, dc);
1962 		mutex_unlock(&bch_register_lock);
1963 	} else {
1964 		struct cache *ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1965 		if (!ca)
1966 			goto err_close;
1967 
1968 		register_cache(sb, sb_page, bdev, ca);
1969 	}
1970 out:
1971 	if (sb_page)
1972 		put_page(sb_page);
1973 	kfree(sb);
1974 	kfree(path);
1975 	module_put(THIS_MODULE);
1976 	return ret;
1977 
1978 err_close:
1979 	blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1980 err:
1981 	pr_info("error opening %s: %s", path, err);
1982 	ret = -EINVAL;
1983 	goto out;
1984 }
1985 
1986 static int bcache_reboot(struct notifier_block *n, unsigned long code, void *x)
1987 {
1988 	if (code == SYS_DOWN ||
1989 	    code == SYS_HALT ||
1990 	    code == SYS_POWER_OFF) {
1991 		DEFINE_WAIT(wait);
1992 		unsigned long start = jiffies;
1993 		bool stopped = false;
1994 
1995 		struct cache_set *c, *tc;
1996 		struct cached_dev *dc, *tdc;
1997 
1998 		mutex_lock(&bch_register_lock);
1999 
2000 		if (list_empty(&bch_cache_sets) &&
2001 		    list_empty(&uncached_devices))
2002 			goto out;
2003 
2004 		pr_info("Stopping all devices:");
2005 
2006 		list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
2007 			bch_cache_set_stop(c);
2008 
2009 		list_for_each_entry_safe(dc, tdc, &uncached_devices, list)
2010 			bcache_device_stop(&dc->disk);
2011 
2012 		/* What's a condition variable? */
2013 		while (1) {
2014 			long timeout = start + 2 * HZ - jiffies;
2015 
2016 			stopped = list_empty(&bch_cache_sets) &&
2017 				list_empty(&uncached_devices);
2018 
2019 			if (timeout < 0 || stopped)
2020 				break;
2021 
2022 			prepare_to_wait(&unregister_wait, &wait,
2023 					TASK_UNINTERRUPTIBLE);
2024 
2025 			mutex_unlock(&bch_register_lock);
2026 			schedule_timeout(timeout);
2027 			mutex_lock(&bch_register_lock);
2028 		}
2029 
2030 		finish_wait(&unregister_wait, &wait);
2031 
2032 		if (stopped)
2033 			pr_info("All devices stopped");
2034 		else
2035 			pr_notice("Timeout waiting for devices to be closed");
2036 out:
2037 		mutex_unlock(&bch_register_lock);
2038 	}
2039 
2040 	return NOTIFY_DONE;
2041 }
2042 
2043 static struct notifier_block reboot = {
2044 	.notifier_call	= bcache_reboot,
2045 	.priority	= INT_MAX, /* before any real devices */
2046 };
2047 
2048 static void bcache_exit(void)
2049 {
2050 	bch_debug_exit();
2051 	bch_request_exit();
2052 	if (bcache_kobj)
2053 		kobject_put(bcache_kobj);
2054 	if (bcache_wq)
2055 		destroy_workqueue(bcache_wq);
2056 	if (bcache_major)
2057 		unregister_blkdev(bcache_major, "bcache");
2058 	unregister_reboot_notifier(&reboot);
2059 }
2060 
2061 static int __init bcache_init(void)
2062 {
2063 	static const struct attribute *files[] = {
2064 		&ksysfs_register.attr,
2065 		&ksysfs_register_quiet.attr,
2066 		NULL
2067 	};
2068 
2069 	mutex_init(&bch_register_lock);
2070 	init_waitqueue_head(&unregister_wait);
2071 	register_reboot_notifier(&reboot);
2072 	closure_debug_init();
2073 
2074 	bcache_major = register_blkdev(0, "bcache");
2075 	if (bcache_major < 0) {
2076 		unregister_reboot_notifier(&reboot);
2077 		return bcache_major;
2078 	}
2079 
2080 	if (!(bcache_wq = create_workqueue("bcache")) ||
2081 	    !(bcache_kobj = kobject_create_and_add("bcache", fs_kobj)) ||
2082 	    sysfs_create_files(bcache_kobj, files) ||
2083 	    bch_request_init() ||
2084 	    bch_debug_init(bcache_kobj))
2085 		goto err;
2086 
2087 	return 0;
2088 err:
2089 	bcache_exit();
2090 	return -ENOMEM;
2091 }
2092 
2093 module_exit(bcache_exit);
2094 module_init(bcache_init);
2095