xref: /openbmc/linux/drivers/md/bcache/super.c (revision e3b9f1e8)
1 /*
2  * bcache setup/teardown code, and some metadata io - read a superblock and
3  * figure out what to do with it.
4  *
5  * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
6  * Copyright 2012 Google, Inc.
7  */
8 
9 #include "bcache.h"
10 #include "btree.h"
11 #include "debug.h"
12 #include "extents.h"
13 #include "request.h"
14 #include "writeback.h"
15 
16 #include <linux/blkdev.h>
17 #include <linux/buffer_head.h>
18 #include <linux/debugfs.h>
19 #include <linux/genhd.h>
20 #include <linux/idr.h>
21 #include <linux/kthread.h>
22 #include <linux/module.h>
23 #include <linux/random.h>
24 #include <linux/reboot.h>
25 #include <linux/sysfs.h>
26 
27 MODULE_LICENSE("GPL");
28 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
29 
30 static const char bcache_magic[] = {
31 	0xc6, 0x85, 0x73, 0xf6, 0x4e, 0x1a, 0x45, 0xca,
32 	0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81
33 };
34 
35 static const char invalid_uuid[] = {
36 	0xa0, 0x3e, 0xf8, 0xed, 0x3e, 0xe1, 0xb8, 0x78,
37 	0xc8, 0x50, 0xfc, 0x5e, 0xcb, 0x16, 0xcd, 0x99
38 };
39 
40 /* Default is -1; we skip past it for struct cached_dev's cache mode */
41 const char * const bch_cache_modes[] = {
42 	"default",
43 	"writethrough",
44 	"writeback",
45 	"writearound",
46 	"none",
47 	NULL
48 };
49 
50 static struct kobject *bcache_kobj;
51 struct mutex bch_register_lock;
52 LIST_HEAD(bch_cache_sets);
53 static LIST_HEAD(uncached_devices);
54 
55 static int bcache_major;
56 static DEFINE_IDA(bcache_device_idx);
57 static wait_queue_head_t unregister_wait;
58 struct workqueue_struct *bcache_wq;
59 
60 #define BTREE_MAX_PAGES		(256 * 1024 / PAGE_SIZE)
61 /* limitation of partitions number on single bcache device */
62 #define BCACHE_MINORS		128
63 /* limitation of bcache devices number on single system */
64 #define BCACHE_DEVICE_IDX_MAX	((1U << MINORBITS)/BCACHE_MINORS)
65 
66 /* Superblock */
67 
68 static const char *read_super(struct cache_sb *sb, struct block_device *bdev,
69 			      struct page **res)
70 {
71 	const char *err;
72 	struct cache_sb *s;
73 	struct buffer_head *bh = __bread(bdev, 1, SB_SIZE);
74 	unsigned i;
75 
76 	if (!bh)
77 		return "IO error";
78 
79 	s = (struct cache_sb *) bh->b_data;
80 
81 	sb->offset		= le64_to_cpu(s->offset);
82 	sb->version		= le64_to_cpu(s->version);
83 
84 	memcpy(sb->magic,	s->magic, 16);
85 	memcpy(sb->uuid,	s->uuid, 16);
86 	memcpy(sb->set_uuid,	s->set_uuid, 16);
87 	memcpy(sb->label,	s->label, SB_LABEL_SIZE);
88 
89 	sb->flags		= le64_to_cpu(s->flags);
90 	sb->seq			= le64_to_cpu(s->seq);
91 	sb->last_mount		= le32_to_cpu(s->last_mount);
92 	sb->first_bucket	= le16_to_cpu(s->first_bucket);
93 	sb->keys		= le16_to_cpu(s->keys);
94 
95 	for (i = 0; i < SB_JOURNAL_BUCKETS; i++)
96 		sb->d[i] = le64_to_cpu(s->d[i]);
97 
98 	pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u",
99 		 sb->version, sb->flags, sb->seq, sb->keys);
100 
101 	err = "Not a bcache superblock";
102 	if (sb->offset != SB_SECTOR)
103 		goto err;
104 
105 	if (memcmp(sb->magic, bcache_magic, 16))
106 		goto err;
107 
108 	err = "Too many journal buckets";
109 	if (sb->keys > SB_JOURNAL_BUCKETS)
110 		goto err;
111 
112 	err = "Bad checksum";
113 	if (s->csum != csum_set(s))
114 		goto err;
115 
116 	err = "Bad UUID";
117 	if (bch_is_zero(sb->uuid, 16))
118 		goto err;
119 
120 	sb->block_size	= le16_to_cpu(s->block_size);
121 
122 	err = "Superblock block size smaller than device block size";
123 	if (sb->block_size << 9 < bdev_logical_block_size(bdev))
124 		goto err;
125 
126 	switch (sb->version) {
127 	case BCACHE_SB_VERSION_BDEV:
128 		sb->data_offset	= BDEV_DATA_START_DEFAULT;
129 		break;
130 	case BCACHE_SB_VERSION_BDEV_WITH_OFFSET:
131 		sb->data_offset	= le64_to_cpu(s->data_offset);
132 
133 		err = "Bad data offset";
134 		if (sb->data_offset < BDEV_DATA_START_DEFAULT)
135 			goto err;
136 
137 		break;
138 	case BCACHE_SB_VERSION_CDEV:
139 	case BCACHE_SB_VERSION_CDEV_WITH_UUID:
140 		sb->nbuckets	= le64_to_cpu(s->nbuckets);
141 		sb->bucket_size	= le16_to_cpu(s->bucket_size);
142 
143 		sb->nr_in_set	= le16_to_cpu(s->nr_in_set);
144 		sb->nr_this_dev	= le16_to_cpu(s->nr_this_dev);
145 
146 		err = "Too many buckets";
147 		if (sb->nbuckets > LONG_MAX)
148 			goto err;
149 
150 		err = "Not enough buckets";
151 		if (sb->nbuckets < 1 << 7)
152 			goto err;
153 
154 		err = "Bad block/bucket size";
155 		if (!is_power_of_2(sb->block_size) ||
156 		    sb->block_size > PAGE_SECTORS ||
157 		    !is_power_of_2(sb->bucket_size) ||
158 		    sb->bucket_size < PAGE_SECTORS)
159 			goto err;
160 
161 		err = "Invalid superblock: device too small";
162 		if (get_capacity(bdev->bd_disk) < sb->bucket_size * sb->nbuckets)
163 			goto err;
164 
165 		err = "Bad UUID";
166 		if (bch_is_zero(sb->set_uuid, 16))
167 			goto err;
168 
169 		err = "Bad cache device number in set";
170 		if (!sb->nr_in_set ||
171 		    sb->nr_in_set <= sb->nr_this_dev ||
172 		    sb->nr_in_set > MAX_CACHES_PER_SET)
173 			goto err;
174 
175 		err = "Journal buckets not sequential";
176 		for (i = 0; i < sb->keys; i++)
177 			if (sb->d[i] != sb->first_bucket + i)
178 				goto err;
179 
180 		err = "Too many journal buckets";
181 		if (sb->first_bucket + sb->keys > sb->nbuckets)
182 			goto err;
183 
184 		err = "Invalid superblock: first bucket comes before end of super";
185 		if (sb->first_bucket * sb->bucket_size < 16)
186 			goto err;
187 
188 		break;
189 	default:
190 		err = "Unsupported superblock version";
191 		goto err;
192 	}
193 
194 	sb->last_mount = get_seconds();
195 	err = NULL;
196 
197 	get_page(bh->b_page);
198 	*res = bh->b_page;
199 err:
200 	put_bh(bh);
201 	return err;
202 }
203 
204 static void write_bdev_super_endio(struct bio *bio)
205 {
206 	struct cached_dev *dc = bio->bi_private;
207 	/* XXX: error checking */
208 
209 	closure_put(&dc->sb_write);
210 }
211 
212 static void __write_super(struct cache_sb *sb, struct bio *bio)
213 {
214 	struct cache_sb *out = page_address(bio_first_page_all(bio));
215 	unsigned i;
216 
217 	bio->bi_iter.bi_sector	= SB_SECTOR;
218 	bio->bi_iter.bi_size	= SB_SIZE;
219 	bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
220 	bch_bio_map(bio, NULL);
221 
222 	out->offset		= cpu_to_le64(sb->offset);
223 	out->version		= cpu_to_le64(sb->version);
224 
225 	memcpy(out->uuid,	sb->uuid, 16);
226 	memcpy(out->set_uuid,	sb->set_uuid, 16);
227 	memcpy(out->label,	sb->label, SB_LABEL_SIZE);
228 
229 	out->flags		= cpu_to_le64(sb->flags);
230 	out->seq		= cpu_to_le64(sb->seq);
231 
232 	out->last_mount		= cpu_to_le32(sb->last_mount);
233 	out->first_bucket	= cpu_to_le16(sb->first_bucket);
234 	out->keys		= cpu_to_le16(sb->keys);
235 
236 	for (i = 0; i < sb->keys; i++)
237 		out->d[i] = cpu_to_le64(sb->d[i]);
238 
239 	out->csum = csum_set(out);
240 
241 	pr_debug("ver %llu, flags %llu, seq %llu",
242 		 sb->version, sb->flags, sb->seq);
243 
244 	submit_bio(bio);
245 }
246 
247 static void bch_write_bdev_super_unlock(struct closure *cl)
248 {
249 	struct cached_dev *dc = container_of(cl, struct cached_dev, sb_write);
250 
251 	up(&dc->sb_write_mutex);
252 }
253 
254 void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent)
255 {
256 	struct closure *cl = &dc->sb_write;
257 	struct bio *bio = &dc->sb_bio;
258 
259 	down(&dc->sb_write_mutex);
260 	closure_init(cl, parent);
261 
262 	bio_reset(bio);
263 	bio_set_dev(bio, dc->bdev);
264 	bio->bi_end_io	= write_bdev_super_endio;
265 	bio->bi_private = dc;
266 
267 	closure_get(cl);
268 	__write_super(&dc->sb, bio);
269 
270 	closure_return_with_destructor(cl, bch_write_bdev_super_unlock);
271 }
272 
273 static void write_super_endio(struct bio *bio)
274 {
275 	struct cache *ca = bio->bi_private;
276 
277 	/* is_read = 0 */
278 	bch_count_io_errors(ca, bio->bi_status, 0,
279 			    "writing superblock");
280 	closure_put(&ca->set->sb_write);
281 }
282 
283 static void bcache_write_super_unlock(struct closure *cl)
284 {
285 	struct cache_set *c = container_of(cl, struct cache_set, sb_write);
286 
287 	up(&c->sb_write_mutex);
288 }
289 
290 void bcache_write_super(struct cache_set *c)
291 {
292 	struct closure *cl = &c->sb_write;
293 	struct cache *ca;
294 	unsigned i;
295 
296 	down(&c->sb_write_mutex);
297 	closure_init(cl, &c->cl);
298 
299 	c->sb.seq++;
300 
301 	for_each_cache(ca, c, i) {
302 		struct bio *bio = &ca->sb_bio;
303 
304 		ca->sb.version		= BCACHE_SB_VERSION_CDEV_WITH_UUID;
305 		ca->sb.seq		= c->sb.seq;
306 		ca->sb.last_mount	= c->sb.last_mount;
307 
308 		SET_CACHE_SYNC(&ca->sb, CACHE_SYNC(&c->sb));
309 
310 		bio_reset(bio);
311 		bio_set_dev(bio, ca->bdev);
312 		bio->bi_end_io	= write_super_endio;
313 		bio->bi_private = ca;
314 
315 		closure_get(cl);
316 		__write_super(&ca->sb, bio);
317 	}
318 
319 	closure_return_with_destructor(cl, bcache_write_super_unlock);
320 }
321 
322 /* UUID io */
323 
324 static void uuid_endio(struct bio *bio)
325 {
326 	struct closure *cl = bio->bi_private;
327 	struct cache_set *c = container_of(cl, struct cache_set, uuid_write);
328 
329 	cache_set_err_on(bio->bi_status, c, "accessing uuids");
330 	bch_bbio_free(bio, c);
331 	closure_put(cl);
332 }
333 
334 static void uuid_io_unlock(struct closure *cl)
335 {
336 	struct cache_set *c = container_of(cl, struct cache_set, uuid_write);
337 
338 	up(&c->uuid_write_mutex);
339 }
340 
341 static void uuid_io(struct cache_set *c, int op, unsigned long op_flags,
342 		    struct bkey *k, struct closure *parent)
343 {
344 	struct closure *cl = &c->uuid_write;
345 	struct uuid_entry *u;
346 	unsigned i;
347 	char buf[80];
348 
349 	BUG_ON(!parent);
350 	down(&c->uuid_write_mutex);
351 	closure_init(cl, parent);
352 
353 	for (i = 0; i < KEY_PTRS(k); i++) {
354 		struct bio *bio = bch_bbio_alloc(c);
355 
356 		bio->bi_opf = REQ_SYNC | REQ_META | op_flags;
357 		bio->bi_iter.bi_size = KEY_SIZE(k) << 9;
358 
359 		bio->bi_end_io	= uuid_endio;
360 		bio->bi_private = cl;
361 		bio_set_op_attrs(bio, op, REQ_SYNC|REQ_META|op_flags);
362 		bch_bio_map(bio, c->uuids);
363 
364 		bch_submit_bbio(bio, c, k, i);
365 
366 		if (op != REQ_OP_WRITE)
367 			break;
368 	}
369 
370 	bch_extent_to_text(buf, sizeof(buf), k);
371 	pr_debug("%s UUIDs at %s", op == REQ_OP_WRITE ? "wrote" : "read", buf);
372 
373 	for (u = c->uuids; u < c->uuids + c->nr_uuids; u++)
374 		if (!bch_is_zero(u->uuid, 16))
375 			pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u",
376 				 u - c->uuids, u->uuid, u->label,
377 				 u->first_reg, u->last_reg, u->invalidated);
378 
379 	closure_return_with_destructor(cl, uuid_io_unlock);
380 }
381 
382 static char *uuid_read(struct cache_set *c, struct jset *j, struct closure *cl)
383 {
384 	struct bkey *k = &j->uuid_bucket;
385 
386 	if (__bch_btree_ptr_invalid(c, k))
387 		return "bad uuid pointer";
388 
389 	bkey_copy(&c->uuid_bucket, k);
390 	uuid_io(c, REQ_OP_READ, 0, k, cl);
391 
392 	if (j->version < BCACHE_JSET_VERSION_UUIDv1) {
393 		struct uuid_entry_v0	*u0 = (void *) c->uuids;
394 		struct uuid_entry	*u1 = (void *) c->uuids;
395 		int i;
396 
397 		closure_sync(cl);
398 
399 		/*
400 		 * Since the new uuid entry is bigger than the old, we have to
401 		 * convert starting at the highest memory address and work down
402 		 * in order to do it in place
403 		 */
404 
405 		for (i = c->nr_uuids - 1;
406 		     i >= 0;
407 		     --i) {
408 			memcpy(u1[i].uuid,	u0[i].uuid, 16);
409 			memcpy(u1[i].label,	u0[i].label, 32);
410 
411 			u1[i].first_reg		= u0[i].first_reg;
412 			u1[i].last_reg		= u0[i].last_reg;
413 			u1[i].invalidated	= u0[i].invalidated;
414 
415 			u1[i].flags	= 0;
416 			u1[i].sectors	= 0;
417 		}
418 	}
419 
420 	return NULL;
421 }
422 
423 static int __uuid_write(struct cache_set *c)
424 {
425 	BKEY_PADDED(key) k;
426 	struct closure cl;
427 	closure_init_stack(&cl);
428 
429 	lockdep_assert_held(&bch_register_lock);
430 
431 	if (bch_bucket_alloc_set(c, RESERVE_BTREE, &k.key, 1, true))
432 		return 1;
433 
434 	SET_KEY_SIZE(&k.key, c->sb.bucket_size);
435 	uuid_io(c, REQ_OP_WRITE, 0, &k.key, &cl);
436 	closure_sync(&cl);
437 
438 	bkey_copy(&c->uuid_bucket, &k.key);
439 	bkey_put(c, &k.key);
440 	return 0;
441 }
442 
443 int bch_uuid_write(struct cache_set *c)
444 {
445 	int ret = __uuid_write(c);
446 
447 	if (!ret)
448 		bch_journal_meta(c, NULL);
449 
450 	return ret;
451 }
452 
453 static struct uuid_entry *uuid_find(struct cache_set *c, const char *uuid)
454 {
455 	struct uuid_entry *u;
456 
457 	for (u = c->uuids;
458 	     u < c->uuids + c->nr_uuids; u++)
459 		if (!memcmp(u->uuid, uuid, 16))
460 			return u;
461 
462 	return NULL;
463 }
464 
465 static struct uuid_entry *uuid_find_empty(struct cache_set *c)
466 {
467 	static const char zero_uuid[16] = "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0";
468 	return uuid_find(c, zero_uuid);
469 }
470 
471 /*
472  * Bucket priorities/gens:
473  *
474  * For each bucket, we store on disk its
475    * 8 bit gen
476    * 16 bit priority
477  *
478  * See alloc.c for an explanation of the gen. The priority is used to implement
479  * lru (and in the future other) cache replacement policies; for most purposes
480  * it's just an opaque integer.
481  *
482  * The gens and the priorities don't have a whole lot to do with each other, and
483  * it's actually the gens that must be written out at specific times - it's no
484  * big deal if the priorities don't get written, if we lose them we just reuse
485  * buckets in suboptimal order.
486  *
487  * On disk they're stored in a packed array, and in as many buckets are required
488  * to fit them all. The buckets we use to store them form a list; the journal
489  * header points to the first bucket, the first bucket points to the second
490  * bucket, et cetera.
491  *
492  * This code is used by the allocation code; periodically (whenever it runs out
493  * of buckets to allocate from) the allocation code will invalidate some
494  * buckets, but it can't use those buckets until their new gens are safely on
495  * disk.
496  */
497 
498 static void prio_endio(struct bio *bio)
499 {
500 	struct cache *ca = bio->bi_private;
501 
502 	cache_set_err_on(bio->bi_status, ca->set, "accessing priorities");
503 	bch_bbio_free(bio, ca->set);
504 	closure_put(&ca->prio);
505 }
506 
507 static void prio_io(struct cache *ca, uint64_t bucket, int op,
508 		    unsigned long op_flags)
509 {
510 	struct closure *cl = &ca->prio;
511 	struct bio *bio = bch_bbio_alloc(ca->set);
512 
513 	closure_init_stack(cl);
514 
515 	bio->bi_iter.bi_sector	= bucket * ca->sb.bucket_size;
516 	bio_set_dev(bio, ca->bdev);
517 	bio->bi_iter.bi_size	= bucket_bytes(ca);
518 
519 	bio->bi_end_io	= prio_endio;
520 	bio->bi_private = ca;
521 	bio_set_op_attrs(bio, op, REQ_SYNC|REQ_META|op_flags);
522 	bch_bio_map(bio, ca->disk_buckets);
523 
524 	closure_bio_submit(bio, &ca->prio);
525 	closure_sync(cl);
526 }
527 
528 void bch_prio_write(struct cache *ca)
529 {
530 	int i;
531 	struct bucket *b;
532 	struct closure cl;
533 
534 	closure_init_stack(&cl);
535 
536 	lockdep_assert_held(&ca->set->bucket_lock);
537 
538 	ca->disk_buckets->seq++;
539 
540 	atomic_long_add(ca->sb.bucket_size * prio_buckets(ca),
541 			&ca->meta_sectors_written);
542 
543 	//pr_debug("free %zu, free_inc %zu, unused %zu", fifo_used(&ca->free),
544 	//	 fifo_used(&ca->free_inc), fifo_used(&ca->unused));
545 
546 	for (i = prio_buckets(ca) - 1; i >= 0; --i) {
547 		long bucket;
548 		struct prio_set *p = ca->disk_buckets;
549 		struct bucket_disk *d = p->data;
550 		struct bucket_disk *end = d + prios_per_bucket(ca);
551 
552 		for (b = ca->buckets + i * prios_per_bucket(ca);
553 		     b < ca->buckets + ca->sb.nbuckets && d < end;
554 		     b++, d++) {
555 			d->prio = cpu_to_le16(b->prio);
556 			d->gen = b->gen;
557 		}
558 
559 		p->next_bucket	= ca->prio_buckets[i + 1];
560 		p->magic	= pset_magic(&ca->sb);
561 		p->csum		= bch_crc64(&p->magic, bucket_bytes(ca) - 8);
562 
563 		bucket = bch_bucket_alloc(ca, RESERVE_PRIO, true);
564 		BUG_ON(bucket == -1);
565 
566 		mutex_unlock(&ca->set->bucket_lock);
567 		prio_io(ca, bucket, REQ_OP_WRITE, 0);
568 		mutex_lock(&ca->set->bucket_lock);
569 
570 		ca->prio_buckets[i] = bucket;
571 		atomic_dec_bug(&ca->buckets[bucket].pin);
572 	}
573 
574 	mutex_unlock(&ca->set->bucket_lock);
575 
576 	bch_journal_meta(ca->set, &cl);
577 	closure_sync(&cl);
578 
579 	mutex_lock(&ca->set->bucket_lock);
580 
581 	/*
582 	 * Don't want the old priorities to get garbage collected until after we
583 	 * finish writing the new ones, and they're journalled
584 	 */
585 	for (i = 0; i < prio_buckets(ca); i++) {
586 		if (ca->prio_last_buckets[i])
587 			__bch_bucket_free(ca,
588 				&ca->buckets[ca->prio_last_buckets[i]]);
589 
590 		ca->prio_last_buckets[i] = ca->prio_buckets[i];
591 	}
592 }
593 
594 static void prio_read(struct cache *ca, uint64_t bucket)
595 {
596 	struct prio_set *p = ca->disk_buckets;
597 	struct bucket_disk *d = p->data + prios_per_bucket(ca), *end = d;
598 	struct bucket *b;
599 	unsigned bucket_nr = 0;
600 
601 	for (b = ca->buckets;
602 	     b < ca->buckets + ca->sb.nbuckets;
603 	     b++, d++) {
604 		if (d == end) {
605 			ca->prio_buckets[bucket_nr] = bucket;
606 			ca->prio_last_buckets[bucket_nr] = bucket;
607 			bucket_nr++;
608 
609 			prio_io(ca, bucket, REQ_OP_READ, 0);
610 
611 			if (p->csum != bch_crc64(&p->magic, bucket_bytes(ca) - 8))
612 				pr_warn("bad csum reading priorities");
613 
614 			if (p->magic != pset_magic(&ca->sb))
615 				pr_warn("bad magic reading priorities");
616 
617 			bucket = p->next_bucket;
618 			d = p->data;
619 		}
620 
621 		b->prio = le16_to_cpu(d->prio);
622 		b->gen = b->last_gc = d->gen;
623 	}
624 }
625 
626 /* Bcache device */
627 
628 static int open_dev(struct block_device *b, fmode_t mode)
629 {
630 	struct bcache_device *d = b->bd_disk->private_data;
631 	if (test_bit(BCACHE_DEV_CLOSING, &d->flags))
632 		return -ENXIO;
633 
634 	closure_get(&d->cl);
635 	return 0;
636 }
637 
638 static void release_dev(struct gendisk *b, fmode_t mode)
639 {
640 	struct bcache_device *d = b->private_data;
641 	closure_put(&d->cl);
642 }
643 
644 static int ioctl_dev(struct block_device *b, fmode_t mode,
645 		     unsigned int cmd, unsigned long arg)
646 {
647 	struct bcache_device *d = b->bd_disk->private_data;
648 	return d->ioctl(d, mode, cmd, arg);
649 }
650 
651 static const struct block_device_operations bcache_ops = {
652 	.open		= open_dev,
653 	.release	= release_dev,
654 	.ioctl		= ioctl_dev,
655 	.owner		= THIS_MODULE,
656 };
657 
658 void bcache_device_stop(struct bcache_device *d)
659 {
660 	if (!test_and_set_bit(BCACHE_DEV_CLOSING, &d->flags))
661 		closure_queue(&d->cl);
662 }
663 
664 static void bcache_device_unlink(struct bcache_device *d)
665 {
666 	lockdep_assert_held(&bch_register_lock);
667 
668 	if (d->c && !test_and_set_bit(BCACHE_DEV_UNLINK_DONE, &d->flags)) {
669 		unsigned i;
670 		struct cache *ca;
671 
672 		sysfs_remove_link(&d->c->kobj, d->name);
673 		sysfs_remove_link(&d->kobj, "cache");
674 
675 		for_each_cache(ca, d->c, i)
676 			bd_unlink_disk_holder(ca->bdev, d->disk);
677 	}
678 }
679 
680 static void bcache_device_link(struct bcache_device *d, struct cache_set *c,
681 			       const char *name)
682 {
683 	unsigned i;
684 	struct cache *ca;
685 
686 	for_each_cache(ca, d->c, i)
687 		bd_link_disk_holder(ca->bdev, d->disk);
688 
689 	snprintf(d->name, BCACHEDEVNAME_SIZE,
690 		 "%s%u", name, d->id);
691 
692 	WARN(sysfs_create_link(&d->kobj, &c->kobj, "cache") ||
693 	     sysfs_create_link(&c->kobj, &d->kobj, d->name),
694 	     "Couldn't create device <-> cache set symlinks");
695 
696 	clear_bit(BCACHE_DEV_UNLINK_DONE, &d->flags);
697 }
698 
699 static void bcache_device_detach(struct bcache_device *d)
700 {
701 	lockdep_assert_held(&bch_register_lock);
702 
703 	if (test_bit(BCACHE_DEV_DETACHING, &d->flags)) {
704 		struct uuid_entry *u = d->c->uuids + d->id;
705 
706 		SET_UUID_FLASH_ONLY(u, 0);
707 		memcpy(u->uuid, invalid_uuid, 16);
708 		u->invalidated = cpu_to_le32(get_seconds());
709 		bch_uuid_write(d->c);
710 	}
711 
712 	bcache_device_unlink(d);
713 
714 	d->c->devices[d->id] = NULL;
715 	closure_put(&d->c->caching);
716 	d->c = NULL;
717 }
718 
719 static void bcache_device_attach(struct bcache_device *d, struct cache_set *c,
720 				 unsigned id)
721 {
722 	d->id = id;
723 	d->c = c;
724 	c->devices[id] = d;
725 
726 	if (id >= c->devices_max_used)
727 		c->devices_max_used = id + 1;
728 
729 	closure_get(&c->caching);
730 }
731 
732 static inline int first_minor_to_idx(int first_minor)
733 {
734 	return (first_minor/BCACHE_MINORS);
735 }
736 
737 static inline int idx_to_first_minor(int idx)
738 {
739 	return (idx * BCACHE_MINORS);
740 }
741 
742 static void bcache_device_free(struct bcache_device *d)
743 {
744 	lockdep_assert_held(&bch_register_lock);
745 
746 	pr_info("%s stopped", d->disk->disk_name);
747 
748 	if (d->c)
749 		bcache_device_detach(d);
750 	if (d->disk && d->disk->flags & GENHD_FL_UP)
751 		del_gendisk(d->disk);
752 	if (d->disk && d->disk->queue)
753 		blk_cleanup_queue(d->disk->queue);
754 	if (d->disk) {
755 		ida_simple_remove(&bcache_device_idx,
756 				  first_minor_to_idx(d->disk->first_minor));
757 		put_disk(d->disk);
758 	}
759 
760 	if (d->bio_split)
761 		bioset_free(d->bio_split);
762 	kvfree(d->full_dirty_stripes);
763 	kvfree(d->stripe_sectors_dirty);
764 
765 	closure_debug_destroy(&d->cl);
766 }
767 
768 static int bcache_device_init(struct bcache_device *d, unsigned block_size,
769 			      sector_t sectors)
770 {
771 	struct request_queue *q;
772 	size_t n;
773 	int idx;
774 
775 	if (!d->stripe_size)
776 		d->stripe_size = 1 << 31;
777 
778 	d->nr_stripes = DIV_ROUND_UP_ULL(sectors, d->stripe_size);
779 
780 	if (!d->nr_stripes ||
781 	    d->nr_stripes > INT_MAX ||
782 	    d->nr_stripes > SIZE_MAX / sizeof(atomic_t)) {
783 		pr_err("nr_stripes too large or invalid: %u (start sector beyond end of disk?)",
784 			(unsigned)d->nr_stripes);
785 		return -ENOMEM;
786 	}
787 
788 	n = d->nr_stripes * sizeof(atomic_t);
789 	d->stripe_sectors_dirty = kvzalloc(n, GFP_KERNEL);
790 	if (!d->stripe_sectors_dirty)
791 		return -ENOMEM;
792 
793 	n = BITS_TO_LONGS(d->nr_stripes) * sizeof(unsigned long);
794 	d->full_dirty_stripes = kvzalloc(n, GFP_KERNEL);
795 	if (!d->full_dirty_stripes)
796 		return -ENOMEM;
797 
798 	idx = ida_simple_get(&bcache_device_idx, 0,
799 				BCACHE_DEVICE_IDX_MAX, GFP_KERNEL);
800 	if (idx < 0)
801 		return idx;
802 
803 	if (!(d->bio_split = bioset_create(4, offsetof(struct bbio, bio),
804 					   BIOSET_NEED_BVECS |
805 					   BIOSET_NEED_RESCUER)) ||
806 	    !(d->disk = alloc_disk(BCACHE_MINORS))) {
807 		ida_simple_remove(&bcache_device_idx, idx);
808 		return -ENOMEM;
809 	}
810 
811 	set_capacity(d->disk, sectors);
812 	snprintf(d->disk->disk_name, DISK_NAME_LEN, "bcache%i", idx);
813 
814 	d->disk->major		= bcache_major;
815 	d->disk->first_minor	= idx_to_first_minor(idx);
816 	d->disk->fops		= &bcache_ops;
817 	d->disk->private_data	= d;
818 
819 	q = blk_alloc_queue(GFP_KERNEL);
820 	if (!q)
821 		return -ENOMEM;
822 
823 	blk_queue_make_request(q, NULL);
824 	d->disk->queue			= q;
825 	q->queuedata			= d;
826 	q->backing_dev_info->congested_data = d;
827 	q->limits.max_hw_sectors	= UINT_MAX;
828 	q->limits.max_sectors		= UINT_MAX;
829 	q->limits.max_segment_size	= UINT_MAX;
830 	q->limits.max_segments		= BIO_MAX_PAGES;
831 	blk_queue_max_discard_sectors(q, UINT_MAX);
832 	q->limits.discard_granularity	= 512;
833 	q->limits.io_min		= block_size;
834 	q->limits.logical_block_size	= block_size;
835 	q->limits.physical_block_size	= block_size;
836 	set_bit(QUEUE_FLAG_NONROT,	&d->disk->queue->queue_flags);
837 	clear_bit(QUEUE_FLAG_ADD_RANDOM, &d->disk->queue->queue_flags);
838 	set_bit(QUEUE_FLAG_DISCARD,	&d->disk->queue->queue_flags);
839 
840 	blk_queue_write_cache(q, true, true);
841 
842 	return 0;
843 }
844 
845 /* Cached device */
846 
847 static void calc_cached_dev_sectors(struct cache_set *c)
848 {
849 	uint64_t sectors = 0;
850 	struct cached_dev *dc;
851 
852 	list_for_each_entry(dc, &c->cached_devs, list)
853 		sectors += bdev_sectors(dc->bdev);
854 
855 	c->cached_dev_sectors = sectors;
856 }
857 
858 void bch_cached_dev_run(struct cached_dev *dc)
859 {
860 	struct bcache_device *d = &dc->disk;
861 	char buf[SB_LABEL_SIZE + 1];
862 	char *env[] = {
863 		"DRIVER=bcache",
864 		kasprintf(GFP_KERNEL, "CACHED_UUID=%pU", dc->sb.uuid),
865 		NULL,
866 		NULL,
867 	};
868 
869 	memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
870 	buf[SB_LABEL_SIZE] = '\0';
871 	env[2] = kasprintf(GFP_KERNEL, "CACHED_LABEL=%s", buf);
872 
873 	if (atomic_xchg(&dc->running, 1)) {
874 		kfree(env[1]);
875 		kfree(env[2]);
876 		return;
877 	}
878 
879 	if (!d->c &&
880 	    BDEV_STATE(&dc->sb) != BDEV_STATE_NONE) {
881 		struct closure cl;
882 		closure_init_stack(&cl);
883 
884 		SET_BDEV_STATE(&dc->sb, BDEV_STATE_STALE);
885 		bch_write_bdev_super(dc, &cl);
886 		closure_sync(&cl);
887 	}
888 
889 	add_disk(d->disk);
890 	bd_link_disk_holder(dc->bdev, dc->disk.disk);
891 	/* won't show up in the uevent file, use udevadm monitor -e instead
892 	 * only class / kset properties are persistent */
893 	kobject_uevent_env(&disk_to_dev(d->disk)->kobj, KOBJ_CHANGE, env);
894 	kfree(env[1]);
895 	kfree(env[2]);
896 
897 	if (sysfs_create_link(&d->kobj, &disk_to_dev(d->disk)->kobj, "dev") ||
898 	    sysfs_create_link(&disk_to_dev(d->disk)->kobj, &d->kobj, "bcache"))
899 		pr_debug("error creating sysfs link");
900 }
901 
902 static void cached_dev_detach_finish(struct work_struct *w)
903 {
904 	struct cached_dev *dc = container_of(w, struct cached_dev, detach);
905 	char buf[BDEVNAME_SIZE];
906 	struct closure cl;
907 	closure_init_stack(&cl);
908 
909 	BUG_ON(!test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags));
910 	BUG_ON(refcount_read(&dc->count));
911 
912 	mutex_lock(&bch_register_lock);
913 
914 	cancel_delayed_work_sync(&dc->writeback_rate_update);
915 	if (!IS_ERR_OR_NULL(dc->writeback_thread)) {
916 		kthread_stop(dc->writeback_thread);
917 		dc->writeback_thread = NULL;
918 	}
919 
920 	memset(&dc->sb.set_uuid, 0, 16);
921 	SET_BDEV_STATE(&dc->sb, BDEV_STATE_NONE);
922 
923 	bch_write_bdev_super(dc, &cl);
924 	closure_sync(&cl);
925 
926 	bcache_device_detach(&dc->disk);
927 	list_move(&dc->list, &uncached_devices);
928 
929 	clear_bit(BCACHE_DEV_DETACHING, &dc->disk.flags);
930 	clear_bit(BCACHE_DEV_UNLINK_DONE, &dc->disk.flags);
931 
932 	mutex_unlock(&bch_register_lock);
933 
934 	pr_info("Caching disabled for %s", bdevname(dc->bdev, buf));
935 
936 	/* Drop ref we took in cached_dev_detach() */
937 	closure_put(&dc->disk.cl);
938 }
939 
940 void bch_cached_dev_detach(struct cached_dev *dc)
941 {
942 	lockdep_assert_held(&bch_register_lock);
943 
944 	if (test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags))
945 		return;
946 
947 	if (test_and_set_bit(BCACHE_DEV_DETACHING, &dc->disk.flags))
948 		return;
949 
950 	/*
951 	 * Block the device from being closed and freed until we're finished
952 	 * detaching
953 	 */
954 	closure_get(&dc->disk.cl);
955 
956 	bch_writeback_queue(dc);
957 	cached_dev_put(dc);
958 }
959 
960 int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c,
961 			  uint8_t *set_uuid)
962 {
963 	uint32_t rtime = cpu_to_le32(get_seconds());
964 	struct uuid_entry *u;
965 	char buf[BDEVNAME_SIZE];
966 
967 	bdevname(dc->bdev, buf);
968 
969 	if ((set_uuid && memcmp(set_uuid, c->sb.set_uuid, 16)) ||
970 	    (!set_uuid && memcmp(dc->sb.set_uuid, c->sb.set_uuid, 16)))
971 		return -ENOENT;
972 
973 	if (dc->disk.c) {
974 		pr_err("Can't attach %s: already attached", buf);
975 		return -EINVAL;
976 	}
977 
978 	if (test_bit(CACHE_SET_STOPPING, &c->flags)) {
979 		pr_err("Can't attach %s: shutting down", buf);
980 		return -EINVAL;
981 	}
982 
983 	if (dc->sb.block_size < c->sb.block_size) {
984 		/* Will die */
985 		pr_err("Couldn't attach %s: block size less than set's block size",
986 		       buf);
987 		return -EINVAL;
988 	}
989 
990 	u = uuid_find(c, dc->sb.uuid);
991 
992 	if (u &&
993 	    (BDEV_STATE(&dc->sb) == BDEV_STATE_STALE ||
994 	     BDEV_STATE(&dc->sb) == BDEV_STATE_NONE)) {
995 		memcpy(u->uuid, invalid_uuid, 16);
996 		u->invalidated = cpu_to_le32(get_seconds());
997 		u = NULL;
998 	}
999 
1000 	if (!u) {
1001 		if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
1002 			pr_err("Couldn't find uuid for %s in set", buf);
1003 			return -ENOENT;
1004 		}
1005 
1006 		u = uuid_find_empty(c);
1007 		if (!u) {
1008 			pr_err("Not caching %s, no room for UUID", buf);
1009 			return -EINVAL;
1010 		}
1011 	}
1012 
1013 	/* Deadlocks since we're called via sysfs...
1014 	sysfs_remove_file(&dc->kobj, &sysfs_attach);
1015 	 */
1016 
1017 	if (bch_is_zero(u->uuid, 16)) {
1018 		struct closure cl;
1019 		closure_init_stack(&cl);
1020 
1021 		memcpy(u->uuid, dc->sb.uuid, 16);
1022 		memcpy(u->label, dc->sb.label, SB_LABEL_SIZE);
1023 		u->first_reg = u->last_reg = rtime;
1024 		bch_uuid_write(c);
1025 
1026 		memcpy(dc->sb.set_uuid, c->sb.set_uuid, 16);
1027 		SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN);
1028 
1029 		bch_write_bdev_super(dc, &cl);
1030 		closure_sync(&cl);
1031 	} else {
1032 		u->last_reg = rtime;
1033 		bch_uuid_write(c);
1034 	}
1035 
1036 	bcache_device_attach(&dc->disk, c, u - c->uuids);
1037 	list_move(&dc->list, &c->cached_devs);
1038 	calc_cached_dev_sectors(c);
1039 
1040 	smp_wmb();
1041 	/*
1042 	 * dc->c must be set before dc->count != 0 - paired with the mb in
1043 	 * cached_dev_get()
1044 	 */
1045 	refcount_set(&dc->count, 1);
1046 
1047 	/* Block writeback thread, but spawn it */
1048 	down_write(&dc->writeback_lock);
1049 	if (bch_cached_dev_writeback_start(dc)) {
1050 		up_write(&dc->writeback_lock);
1051 		return -ENOMEM;
1052 	}
1053 
1054 	if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
1055 		bch_sectors_dirty_init(&dc->disk);
1056 		atomic_set(&dc->has_dirty, 1);
1057 		refcount_inc(&dc->count);
1058 		bch_writeback_queue(dc);
1059 	}
1060 
1061 	bch_cached_dev_run(dc);
1062 	bcache_device_link(&dc->disk, c, "bdev");
1063 
1064 	/* Allow the writeback thread to proceed */
1065 	up_write(&dc->writeback_lock);
1066 
1067 	pr_info("Caching %s as %s on set %pU",
1068 		bdevname(dc->bdev, buf), dc->disk.disk->disk_name,
1069 		dc->disk.c->sb.set_uuid);
1070 	return 0;
1071 }
1072 
1073 void bch_cached_dev_release(struct kobject *kobj)
1074 {
1075 	struct cached_dev *dc = container_of(kobj, struct cached_dev,
1076 					     disk.kobj);
1077 	kfree(dc);
1078 	module_put(THIS_MODULE);
1079 }
1080 
1081 static void cached_dev_free(struct closure *cl)
1082 {
1083 	struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1084 
1085 	cancel_delayed_work_sync(&dc->writeback_rate_update);
1086 	if (!IS_ERR_OR_NULL(dc->writeback_thread))
1087 		kthread_stop(dc->writeback_thread);
1088 	if (dc->writeback_write_wq)
1089 		destroy_workqueue(dc->writeback_write_wq);
1090 
1091 	mutex_lock(&bch_register_lock);
1092 
1093 	if (atomic_read(&dc->running))
1094 		bd_unlink_disk_holder(dc->bdev, dc->disk.disk);
1095 	bcache_device_free(&dc->disk);
1096 	list_del(&dc->list);
1097 
1098 	mutex_unlock(&bch_register_lock);
1099 
1100 	if (!IS_ERR_OR_NULL(dc->bdev))
1101 		blkdev_put(dc->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1102 
1103 	wake_up(&unregister_wait);
1104 
1105 	kobject_put(&dc->disk.kobj);
1106 }
1107 
1108 static void cached_dev_flush(struct closure *cl)
1109 {
1110 	struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1111 	struct bcache_device *d = &dc->disk;
1112 
1113 	mutex_lock(&bch_register_lock);
1114 	bcache_device_unlink(d);
1115 	mutex_unlock(&bch_register_lock);
1116 
1117 	bch_cache_accounting_destroy(&dc->accounting);
1118 	kobject_del(&d->kobj);
1119 
1120 	continue_at(cl, cached_dev_free, system_wq);
1121 }
1122 
1123 static int cached_dev_init(struct cached_dev *dc, unsigned block_size)
1124 {
1125 	int ret;
1126 	struct io *io;
1127 	struct request_queue *q = bdev_get_queue(dc->bdev);
1128 
1129 	__module_get(THIS_MODULE);
1130 	INIT_LIST_HEAD(&dc->list);
1131 	closure_init(&dc->disk.cl, NULL);
1132 	set_closure_fn(&dc->disk.cl, cached_dev_flush, system_wq);
1133 	kobject_init(&dc->disk.kobj, &bch_cached_dev_ktype);
1134 	INIT_WORK(&dc->detach, cached_dev_detach_finish);
1135 	sema_init(&dc->sb_write_mutex, 1);
1136 	INIT_LIST_HEAD(&dc->io_lru);
1137 	spin_lock_init(&dc->io_lock);
1138 	bch_cache_accounting_init(&dc->accounting, &dc->disk.cl);
1139 
1140 	dc->sequential_cutoff		= 4 << 20;
1141 
1142 	for (io = dc->io; io < dc->io + RECENT_IO; io++) {
1143 		list_add(&io->lru, &dc->io_lru);
1144 		hlist_add_head(&io->hash, dc->io_hash + RECENT_IO);
1145 	}
1146 
1147 	dc->disk.stripe_size = q->limits.io_opt >> 9;
1148 
1149 	if (dc->disk.stripe_size)
1150 		dc->partial_stripes_expensive =
1151 			q->limits.raid_partial_stripes_expensive;
1152 
1153 	ret = bcache_device_init(&dc->disk, block_size,
1154 			 dc->bdev->bd_part->nr_sects - dc->sb.data_offset);
1155 	if (ret)
1156 		return ret;
1157 
1158 	dc->disk.disk->queue->backing_dev_info->ra_pages =
1159 		max(dc->disk.disk->queue->backing_dev_info->ra_pages,
1160 		    q->backing_dev_info->ra_pages);
1161 
1162 	bch_cached_dev_request_init(dc);
1163 	bch_cached_dev_writeback_init(dc);
1164 	return 0;
1165 }
1166 
1167 /* Cached device - bcache superblock */
1168 
1169 static void register_bdev(struct cache_sb *sb, struct page *sb_page,
1170 				 struct block_device *bdev,
1171 				 struct cached_dev *dc)
1172 {
1173 	char name[BDEVNAME_SIZE];
1174 	const char *err = "cannot allocate memory";
1175 	struct cache_set *c;
1176 
1177 	memcpy(&dc->sb, sb, sizeof(struct cache_sb));
1178 	dc->bdev = bdev;
1179 	dc->bdev->bd_holder = dc;
1180 
1181 	bio_init(&dc->sb_bio, dc->sb_bio.bi_inline_vecs, 1);
1182 	bio_first_bvec_all(&dc->sb_bio)->bv_page = sb_page;
1183 	get_page(sb_page);
1184 
1185 	if (cached_dev_init(dc, sb->block_size << 9))
1186 		goto err;
1187 
1188 	err = "error creating kobject";
1189 	if (kobject_add(&dc->disk.kobj, &part_to_dev(bdev->bd_part)->kobj,
1190 			"bcache"))
1191 		goto err;
1192 	if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj))
1193 		goto err;
1194 
1195 	pr_info("registered backing device %s", bdevname(bdev, name));
1196 
1197 	list_add(&dc->list, &uncached_devices);
1198 	list_for_each_entry(c, &bch_cache_sets, list)
1199 		bch_cached_dev_attach(dc, c, NULL);
1200 
1201 	if (BDEV_STATE(&dc->sb) == BDEV_STATE_NONE ||
1202 	    BDEV_STATE(&dc->sb) == BDEV_STATE_STALE)
1203 		bch_cached_dev_run(dc);
1204 
1205 	return;
1206 err:
1207 	pr_notice("error opening %s: %s", bdevname(bdev, name), err);
1208 	bcache_device_stop(&dc->disk);
1209 }
1210 
1211 /* Flash only volumes */
1212 
1213 void bch_flash_dev_release(struct kobject *kobj)
1214 {
1215 	struct bcache_device *d = container_of(kobj, struct bcache_device,
1216 					       kobj);
1217 	kfree(d);
1218 }
1219 
1220 static void flash_dev_free(struct closure *cl)
1221 {
1222 	struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1223 	mutex_lock(&bch_register_lock);
1224 	bcache_device_free(d);
1225 	mutex_unlock(&bch_register_lock);
1226 	kobject_put(&d->kobj);
1227 }
1228 
1229 static void flash_dev_flush(struct closure *cl)
1230 {
1231 	struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1232 
1233 	mutex_lock(&bch_register_lock);
1234 	bcache_device_unlink(d);
1235 	mutex_unlock(&bch_register_lock);
1236 	kobject_del(&d->kobj);
1237 	continue_at(cl, flash_dev_free, system_wq);
1238 }
1239 
1240 static int flash_dev_run(struct cache_set *c, struct uuid_entry *u)
1241 {
1242 	struct bcache_device *d = kzalloc(sizeof(struct bcache_device),
1243 					  GFP_KERNEL);
1244 	if (!d)
1245 		return -ENOMEM;
1246 
1247 	closure_init(&d->cl, NULL);
1248 	set_closure_fn(&d->cl, flash_dev_flush, system_wq);
1249 
1250 	kobject_init(&d->kobj, &bch_flash_dev_ktype);
1251 
1252 	if (bcache_device_init(d, block_bytes(c), u->sectors))
1253 		goto err;
1254 
1255 	bcache_device_attach(d, c, u - c->uuids);
1256 	bch_sectors_dirty_init(d);
1257 	bch_flash_dev_request_init(d);
1258 	add_disk(d->disk);
1259 
1260 	if (kobject_add(&d->kobj, &disk_to_dev(d->disk)->kobj, "bcache"))
1261 		goto err;
1262 
1263 	bcache_device_link(d, c, "volume");
1264 
1265 	return 0;
1266 err:
1267 	kobject_put(&d->kobj);
1268 	return -ENOMEM;
1269 }
1270 
1271 static int flash_devs_run(struct cache_set *c)
1272 {
1273 	int ret = 0;
1274 	struct uuid_entry *u;
1275 
1276 	for (u = c->uuids;
1277 	     u < c->uuids + c->nr_uuids && !ret;
1278 	     u++)
1279 		if (UUID_FLASH_ONLY(u))
1280 			ret = flash_dev_run(c, u);
1281 
1282 	return ret;
1283 }
1284 
1285 int bch_flash_dev_create(struct cache_set *c, uint64_t size)
1286 {
1287 	struct uuid_entry *u;
1288 
1289 	if (test_bit(CACHE_SET_STOPPING, &c->flags))
1290 		return -EINTR;
1291 
1292 	if (!test_bit(CACHE_SET_RUNNING, &c->flags))
1293 		return -EPERM;
1294 
1295 	u = uuid_find_empty(c);
1296 	if (!u) {
1297 		pr_err("Can't create volume, no room for UUID");
1298 		return -EINVAL;
1299 	}
1300 
1301 	get_random_bytes(u->uuid, 16);
1302 	memset(u->label, 0, 32);
1303 	u->first_reg = u->last_reg = cpu_to_le32(get_seconds());
1304 
1305 	SET_UUID_FLASH_ONLY(u, 1);
1306 	u->sectors = size >> 9;
1307 
1308 	bch_uuid_write(c);
1309 
1310 	return flash_dev_run(c, u);
1311 }
1312 
1313 /* Cache set */
1314 
1315 __printf(2, 3)
1316 bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...)
1317 {
1318 	va_list args;
1319 
1320 	if (c->on_error != ON_ERROR_PANIC &&
1321 	    test_bit(CACHE_SET_STOPPING, &c->flags))
1322 		return false;
1323 
1324 	/* XXX: we can be called from atomic context
1325 	acquire_console_sem();
1326 	*/
1327 
1328 	printk(KERN_ERR "bcache: error on %pU: ", c->sb.set_uuid);
1329 
1330 	va_start(args, fmt);
1331 	vprintk(fmt, args);
1332 	va_end(args);
1333 
1334 	printk(", disabling caching\n");
1335 
1336 	if (c->on_error == ON_ERROR_PANIC)
1337 		panic("panic forced after error\n");
1338 
1339 	bch_cache_set_unregister(c);
1340 	return true;
1341 }
1342 
1343 void bch_cache_set_release(struct kobject *kobj)
1344 {
1345 	struct cache_set *c = container_of(kobj, struct cache_set, kobj);
1346 	kfree(c);
1347 	module_put(THIS_MODULE);
1348 }
1349 
1350 static void cache_set_free(struct closure *cl)
1351 {
1352 	struct cache_set *c = container_of(cl, struct cache_set, cl);
1353 	struct cache *ca;
1354 	unsigned i;
1355 
1356 	if (!IS_ERR_OR_NULL(c->debug))
1357 		debugfs_remove(c->debug);
1358 
1359 	bch_open_buckets_free(c);
1360 	bch_btree_cache_free(c);
1361 	bch_journal_free(c);
1362 
1363 	for_each_cache(ca, c, i)
1364 		if (ca) {
1365 			ca->set = NULL;
1366 			c->cache[ca->sb.nr_this_dev] = NULL;
1367 			kobject_put(&ca->kobj);
1368 		}
1369 
1370 	bch_bset_sort_state_free(&c->sort);
1371 	free_pages((unsigned long) c->uuids, ilog2(bucket_pages(c)));
1372 
1373 	if (c->moving_gc_wq)
1374 		destroy_workqueue(c->moving_gc_wq);
1375 	if (c->bio_split)
1376 		bioset_free(c->bio_split);
1377 	if (c->fill_iter)
1378 		mempool_destroy(c->fill_iter);
1379 	if (c->bio_meta)
1380 		mempool_destroy(c->bio_meta);
1381 	if (c->search)
1382 		mempool_destroy(c->search);
1383 	kfree(c->devices);
1384 
1385 	mutex_lock(&bch_register_lock);
1386 	list_del(&c->list);
1387 	mutex_unlock(&bch_register_lock);
1388 
1389 	pr_info("Cache set %pU unregistered", c->sb.set_uuid);
1390 	wake_up(&unregister_wait);
1391 
1392 	closure_debug_destroy(&c->cl);
1393 	kobject_put(&c->kobj);
1394 }
1395 
1396 static void cache_set_flush(struct closure *cl)
1397 {
1398 	struct cache_set *c = container_of(cl, struct cache_set, caching);
1399 	struct cache *ca;
1400 	struct btree *b;
1401 	unsigned i;
1402 
1403 	bch_cache_accounting_destroy(&c->accounting);
1404 
1405 	kobject_put(&c->internal);
1406 	kobject_del(&c->kobj);
1407 
1408 	if (c->gc_thread)
1409 		kthread_stop(c->gc_thread);
1410 
1411 	if (!IS_ERR_OR_NULL(c->root))
1412 		list_add(&c->root->list, &c->btree_cache);
1413 
1414 	/* Should skip this if we're unregistering because of an error */
1415 	list_for_each_entry(b, &c->btree_cache, list) {
1416 		mutex_lock(&b->write_lock);
1417 		if (btree_node_dirty(b))
1418 			__bch_btree_node_write(b, NULL);
1419 		mutex_unlock(&b->write_lock);
1420 	}
1421 
1422 	for_each_cache(ca, c, i)
1423 		if (ca->alloc_thread)
1424 			kthread_stop(ca->alloc_thread);
1425 
1426 	if (c->journal.cur) {
1427 		cancel_delayed_work_sync(&c->journal.work);
1428 		/* flush last journal entry if needed */
1429 		c->journal.work.work.func(&c->journal.work.work);
1430 	}
1431 
1432 	closure_return(cl);
1433 }
1434 
1435 static void __cache_set_unregister(struct closure *cl)
1436 {
1437 	struct cache_set *c = container_of(cl, struct cache_set, caching);
1438 	struct cached_dev *dc;
1439 	size_t i;
1440 
1441 	mutex_lock(&bch_register_lock);
1442 
1443 	for (i = 0; i < c->devices_max_used; i++)
1444 		if (c->devices[i]) {
1445 			if (!UUID_FLASH_ONLY(&c->uuids[i]) &&
1446 			    test_bit(CACHE_SET_UNREGISTERING, &c->flags)) {
1447 				dc = container_of(c->devices[i],
1448 						  struct cached_dev, disk);
1449 				bch_cached_dev_detach(dc);
1450 			} else {
1451 				bcache_device_stop(c->devices[i]);
1452 			}
1453 		}
1454 
1455 	mutex_unlock(&bch_register_lock);
1456 
1457 	continue_at(cl, cache_set_flush, system_wq);
1458 }
1459 
1460 void bch_cache_set_stop(struct cache_set *c)
1461 {
1462 	if (!test_and_set_bit(CACHE_SET_STOPPING, &c->flags))
1463 		closure_queue(&c->caching);
1464 }
1465 
1466 void bch_cache_set_unregister(struct cache_set *c)
1467 {
1468 	set_bit(CACHE_SET_UNREGISTERING, &c->flags);
1469 	bch_cache_set_stop(c);
1470 }
1471 
1472 #define alloc_bucket_pages(gfp, c)			\
1473 	((void *) __get_free_pages(__GFP_ZERO|gfp, ilog2(bucket_pages(c))))
1474 
1475 struct cache_set *bch_cache_set_alloc(struct cache_sb *sb)
1476 {
1477 	int iter_size;
1478 	struct cache_set *c = kzalloc(sizeof(struct cache_set), GFP_KERNEL);
1479 	if (!c)
1480 		return NULL;
1481 
1482 	__module_get(THIS_MODULE);
1483 	closure_init(&c->cl, NULL);
1484 	set_closure_fn(&c->cl, cache_set_free, system_wq);
1485 
1486 	closure_init(&c->caching, &c->cl);
1487 	set_closure_fn(&c->caching, __cache_set_unregister, system_wq);
1488 
1489 	/* Maybe create continue_at_noreturn() and use it here? */
1490 	closure_set_stopped(&c->cl);
1491 	closure_put(&c->cl);
1492 
1493 	kobject_init(&c->kobj, &bch_cache_set_ktype);
1494 	kobject_init(&c->internal, &bch_cache_set_internal_ktype);
1495 
1496 	bch_cache_accounting_init(&c->accounting, &c->cl);
1497 
1498 	memcpy(c->sb.set_uuid, sb->set_uuid, 16);
1499 	c->sb.block_size	= sb->block_size;
1500 	c->sb.bucket_size	= sb->bucket_size;
1501 	c->sb.nr_in_set		= sb->nr_in_set;
1502 	c->sb.last_mount	= sb->last_mount;
1503 	c->bucket_bits		= ilog2(sb->bucket_size);
1504 	c->block_bits		= ilog2(sb->block_size);
1505 	c->nr_uuids		= bucket_bytes(c) / sizeof(struct uuid_entry);
1506 	c->devices_max_used	= 0;
1507 	c->btree_pages		= bucket_pages(c);
1508 	if (c->btree_pages > BTREE_MAX_PAGES)
1509 		c->btree_pages = max_t(int, c->btree_pages / 4,
1510 				       BTREE_MAX_PAGES);
1511 
1512 	sema_init(&c->sb_write_mutex, 1);
1513 	mutex_init(&c->bucket_lock);
1514 	init_waitqueue_head(&c->btree_cache_wait);
1515 	init_waitqueue_head(&c->bucket_wait);
1516 	init_waitqueue_head(&c->gc_wait);
1517 	sema_init(&c->uuid_write_mutex, 1);
1518 
1519 	spin_lock_init(&c->btree_gc_time.lock);
1520 	spin_lock_init(&c->btree_split_time.lock);
1521 	spin_lock_init(&c->btree_read_time.lock);
1522 
1523 	bch_moving_init_cache_set(c);
1524 
1525 	INIT_LIST_HEAD(&c->list);
1526 	INIT_LIST_HEAD(&c->cached_devs);
1527 	INIT_LIST_HEAD(&c->btree_cache);
1528 	INIT_LIST_HEAD(&c->btree_cache_freeable);
1529 	INIT_LIST_HEAD(&c->btree_cache_freed);
1530 	INIT_LIST_HEAD(&c->data_buckets);
1531 
1532 	c->search = mempool_create_slab_pool(32, bch_search_cache);
1533 	if (!c->search)
1534 		goto err;
1535 
1536 	iter_size = (sb->bucket_size / sb->block_size + 1) *
1537 		sizeof(struct btree_iter_set);
1538 
1539 	if (!(c->devices = kzalloc(c->nr_uuids * sizeof(void *), GFP_KERNEL)) ||
1540 	    !(c->bio_meta = mempool_create_kmalloc_pool(2,
1541 				sizeof(struct bbio) + sizeof(struct bio_vec) *
1542 				bucket_pages(c))) ||
1543 	    !(c->fill_iter = mempool_create_kmalloc_pool(1, iter_size)) ||
1544 	    !(c->bio_split = bioset_create(4, offsetof(struct bbio, bio),
1545 					   BIOSET_NEED_BVECS |
1546 					   BIOSET_NEED_RESCUER)) ||
1547 	    !(c->uuids = alloc_bucket_pages(GFP_KERNEL, c)) ||
1548 	    !(c->moving_gc_wq = alloc_workqueue("bcache_gc",
1549 						WQ_MEM_RECLAIM, 0)) ||
1550 	    bch_journal_alloc(c) ||
1551 	    bch_btree_cache_alloc(c) ||
1552 	    bch_open_buckets_alloc(c) ||
1553 	    bch_bset_sort_state_init(&c->sort, ilog2(c->btree_pages)))
1554 		goto err;
1555 
1556 	c->congested_read_threshold_us	= 2000;
1557 	c->congested_write_threshold_us	= 20000;
1558 	c->error_limit	= DEFAULT_IO_ERROR_LIMIT;
1559 
1560 	return c;
1561 err:
1562 	bch_cache_set_unregister(c);
1563 	return NULL;
1564 }
1565 
1566 static void run_cache_set(struct cache_set *c)
1567 {
1568 	const char *err = "cannot allocate memory";
1569 	struct cached_dev *dc, *t;
1570 	struct cache *ca;
1571 	struct closure cl;
1572 	unsigned i;
1573 
1574 	closure_init_stack(&cl);
1575 
1576 	for_each_cache(ca, c, i)
1577 		c->nbuckets += ca->sb.nbuckets;
1578 	set_gc_sectors(c);
1579 
1580 	if (CACHE_SYNC(&c->sb)) {
1581 		LIST_HEAD(journal);
1582 		struct bkey *k;
1583 		struct jset *j;
1584 
1585 		err = "cannot allocate memory for journal";
1586 		if (bch_journal_read(c, &journal))
1587 			goto err;
1588 
1589 		pr_debug("btree_journal_read() done");
1590 
1591 		err = "no journal entries found";
1592 		if (list_empty(&journal))
1593 			goto err;
1594 
1595 		j = &list_entry(journal.prev, struct journal_replay, list)->j;
1596 
1597 		err = "IO error reading priorities";
1598 		for_each_cache(ca, c, i)
1599 			prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev]);
1600 
1601 		/*
1602 		 * If prio_read() fails it'll call cache_set_error and we'll
1603 		 * tear everything down right away, but if we perhaps checked
1604 		 * sooner we could avoid journal replay.
1605 		 */
1606 
1607 		k = &j->btree_root;
1608 
1609 		err = "bad btree root";
1610 		if (__bch_btree_ptr_invalid(c, k))
1611 			goto err;
1612 
1613 		err = "error reading btree root";
1614 		c->root = bch_btree_node_get(c, NULL, k, j->btree_level, true, NULL);
1615 		if (IS_ERR_OR_NULL(c->root))
1616 			goto err;
1617 
1618 		list_del_init(&c->root->list);
1619 		rw_unlock(true, c->root);
1620 
1621 		err = uuid_read(c, j, &cl);
1622 		if (err)
1623 			goto err;
1624 
1625 		err = "error in recovery";
1626 		if (bch_btree_check(c))
1627 			goto err;
1628 
1629 		bch_journal_mark(c, &journal);
1630 		bch_initial_gc_finish(c);
1631 		pr_debug("btree_check() done");
1632 
1633 		/*
1634 		 * bcache_journal_next() can't happen sooner, or
1635 		 * btree_gc_finish() will give spurious errors about last_gc >
1636 		 * gc_gen - this is a hack but oh well.
1637 		 */
1638 		bch_journal_next(&c->journal);
1639 
1640 		err = "error starting allocator thread";
1641 		for_each_cache(ca, c, i)
1642 			if (bch_cache_allocator_start(ca))
1643 				goto err;
1644 
1645 		/*
1646 		 * First place it's safe to allocate: btree_check() and
1647 		 * btree_gc_finish() have to run before we have buckets to
1648 		 * allocate, and bch_bucket_alloc_set() might cause a journal
1649 		 * entry to be written so bcache_journal_next() has to be called
1650 		 * first.
1651 		 *
1652 		 * If the uuids were in the old format we have to rewrite them
1653 		 * before the next journal entry is written:
1654 		 */
1655 		if (j->version < BCACHE_JSET_VERSION_UUID)
1656 			__uuid_write(c);
1657 
1658 		bch_journal_replay(c, &journal);
1659 	} else {
1660 		pr_notice("invalidating existing data");
1661 
1662 		for_each_cache(ca, c, i) {
1663 			unsigned j;
1664 
1665 			ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7,
1666 					      2, SB_JOURNAL_BUCKETS);
1667 
1668 			for (j = 0; j < ca->sb.keys; j++)
1669 				ca->sb.d[j] = ca->sb.first_bucket + j;
1670 		}
1671 
1672 		bch_initial_gc_finish(c);
1673 
1674 		err = "error starting allocator thread";
1675 		for_each_cache(ca, c, i)
1676 			if (bch_cache_allocator_start(ca))
1677 				goto err;
1678 
1679 		mutex_lock(&c->bucket_lock);
1680 		for_each_cache(ca, c, i)
1681 			bch_prio_write(ca);
1682 		mutex_unlock(&c->bucket_lock);
1683 
1684 		err = "cannot allocate new UUID bucket";
1685 		if (__uuid_write(c))
1686 			goto err;
1687 
1688 		err = "cannot allocate new btree root";
1689 		c->root = __bch_btree_node_alloc(c, NULL, 0, true, NULL);
1690 		if (IS_ERR_OR_NULL(c->root))
1691 			goto err;
1692 
1693 		mutex_lock(&c->root->write_lock);
1694 		bkey_copy_key(&c->root->key, &MAX_KEY);
1695 		bch_btree_node_write(c->root, &cl);
1696 		mutex_unlock(&c->root->write_lock);
1697 
1698 		bch_btree_set_root(c->root);
1699 		rw_unlock(true, c->root);
1700 
1701 		/*
1702 		 * We don't want to write the first journal entry until
1703 		 * everything is set up - fortunately journal entries won't be
1704 		 * written until the SET_CACHE_SYNC() here:
1705 		 */
1706 		SET_CACHE_SYNC(&c->sb, true);
1707 
1708 		bch_journal_next(&c->journal);
1709 		bch_journal_meta(c, &cl);
1710 	}
1711 
1712 	err = "error starting gc thread";
1713 	if (bch_gc_thread_start(c))
1714 		goto err;
1715 
1716 	closure_sync(&cl);
1717 	c->sb.last_mount = get_seconds();
1718 	bcache_write_super(c);
1719 
1720 	list_for_each_entry_safe(dc, t, &uncached_devices, list)
1721 		bch_cached_dev_attach(dc, c, NULL);
1722 
1723 	flash_devs_run(c);
1724 
1725 	set_bit(CACHE_SET_RUNNING, &c->flags);
1726 	return;
1727 err:
1728 	closure_sync(&cl);
1729 	/* XXX: test this, it's broken */
1730 	bch_cache_set_error(c, "%s", err);
1731 }
1732 
1733 static bool can_attach_cache(struct cache *ca, struct cache_set *c)
1734 {
1735 	return ca->sb.block_size	== c->sb.block_size &&
1736 		ca->sb.bucket_size	== c->sb.bucket_size &&
1737 		ca->sb.nr_in_set	== c->sb.nr_in_set;
1738 }
1739 
1740 static const char *register_cache_set(struct cache *ca)
1741 {
1742 	char buf[12];
1743 	const char *err = "cannot allocate memory";
1744 	struct cache_set *c;
1745 
1746 	list_for_each_entry(c, &bch_cache_sets, list)
1747 		if (!memcmp(c->sb.set_uuid, ca->sb.set_uuid, 16)) {
1748 			if (c->cache[ca->sb.nr_this_dev])
1749 				return "duplicate cache set member";
1750 
1751 			if (!can_attach_cache(ca, c))
1752 				return "cache sb does not match set";
1753 
1754 			if (!CACHE_SYNC(&ca->sb))
1755 				SET_CACHE_SYNC(&c->sb, false);
1756 
1757 			goto found;
1758 		}
1759 
1760 	c = bch_cache_set_alloc(&ca->sb);
1761 	if (!c)
1762 		return err;
1763 
1764 	err = "error creating kobject";
1765 	if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->sb.set_uuid) ||
1766 	    kobject_add(&c->internal, &c->kobj, "internal"))
1767 		goto err;
1768 
1769 	if (bch_cache_accounting_add_kobjs(&c->accounting, &c->kobj))
1770 		goto err;
1771 
1772 	bch_debug_init_cache_set(c);
1773 
1774 	list_add(&c->list, &bch_cache_sets);
1775 found:
1776 	sprintf(buf, "cache%i", ca->sb.nr_this_dev);
1777 	if (sysfs_create_link(&ca->kobj, &c->kobj, "set") ||
1778 	    sysfs_create_link(&c->kobj, &ca->kobj, buf))
1779 		goto err;
1780 
1781 	if (ca->sb.seq > c->sb.seq) {
1782 		c->sb.version		= ca->sb.version;
1783 		memcpy(c->sb.set_uuid, ca->sb.set_uuid, 16);
1784 		c->sb.flags             = ca->sb.flags;
1785 		c->sb.seq		= ca->sb.seq;
1786 		pr_debug("set version = %llu", c->sb.version);
1787 	}
1788 
1789 	kobject_get(&ca->kobj);
1790 	ca->set = c;
1791 	ca->set->cache[ca->sb.nr_this_dev] = ca;
1792 	c->cache_by_alloc[c->caches_loaded++] = ca;
1793 
1794 	if (c->caches_loaded == c->sb.nr_in_set)
1795 		run_cache_set(c);
1796 
1797 	return NULL;
1798 err:
1799 	bch_cache_set_unregister(c);
1800 	return err;
1801 }
1802 
1803 /* Cache device */
1804 
1805 void bch_cache_release(struct kobject *kobj)
1806 {
1807 	struct cache *ca = container_of(kobj, struct cache, kobj);
1808 	unsigned i;
1809 
1810 	if (ca->set) {
1811 		BUG_ON(ca->set->cache[ca->sb.nr_this_dev] != ca);
1812 		ca->set->cache[ca->sb.nr_this_dev] = NULL;
1813 	}
1814 
1815 	free_pages((unsigned long) ca->disk_buckets, ilog2(bucket_pages(ca)));
1816 	kfree(ca->prio_buckets);
1817 	vfree(ca->buckets);
1818 
1819 	free_heap(&ca->heap);
1820 	free_fifo(&ca->free_inc);
1821 
1822 	for (i = 0; i < RESERVE_NR; i++)
1823 		free_fifo(&ca->free[i]);
1824 
1825 	if (ca->sb_bio.bi_inline_vecs[0].bv_page)
1826 		put_page(bio_first_page_all(&ca->sb_bio));
1827 
1828 	if (!IS_ERR_OR_NULL(ca->bdev))
1829 		blkdev_put(ca->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1830 
1831 	kfree(ca);
1832 	module_put(THIS_MODULE);
1833 }
1834 
1835 static int cache_alloc(struct cache *ca)
1836 {
1837 	size_t free;
1838 	size_t btree_buckets;
1839 	struct bucket *b;
1840 
1841 	__module_get(THIS_MODULE);
1842 	kobject_init(&ca->kobj, &bch_cache_ktype);
1843 
1844 	bio_init(&ca->journal.bio, ca->journal.bio.bi_inline_vecs, 8);
1845 
1846 	/*
1847 	 * when ca->sb.njournal_buckets is not zero, journal exists,
1848 	 * and in bch_journal_replay(), tree node may split,
1849 	 * so bucket of RESERVE_BTREE type is needed,
1850 	 * the worst situation is all journal buckets are valid journal,
1851 	 * and all the keys need to replay,
1852 	 * so the number of  RESERVE_BTREE type buckets should be as much
1853 	 * as journal buckets
1854 	 */
1855 	btree_buckets = ca->sb.njournal_buckets ?: 8;
1856 	free = roundup_pow_of_two(ca->sb.nbuckets) >> 10;
1857 
1858 	if (!init_fifo(&ca->free[RESERVE_BTREE], btree_buckets, GFP_KERNEL) ||
1859 	    !init_fifo_exact(&ca->free[RESERVE_PRIO], prio_buckets(ca), GFP_KERNEL) ||
1860 	    !init_fifo(&ca->free[RESERVE_MOVINGGC], free, GFP_KERNEL) ||
1861 	    !init_fifo(&ca->free[RESERVE_NONE], free, GFP_KERNEL) ||
1862 	    !init_fifo(&ca->free_inc,	free << 2, GFP_KERNEL) ||
1863 	    !init_heap(&ca->heap,	free << 3, GFP_KERNEL) ||
1864 	    !(ca->buckets	= vzalloc(sizeof(struct bucket) *
1865 					  ca->sb.nbuckets)) ||
1866 	    !(ca->prio_buckets	= kzalloc(sizeof(uint64_t) * prio_buckets(ca) *
1867 					  2, GFP_KERNEL)) ||
1868 	    !(ca->disk_buckets	= alloc_bucket_pages(GFP_KERNEL, ca)))
1869 		return -ENOMEM;
1870 
1871 	ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca);
1872 
1873 	for_each_bucket(b, ca)
1874 		atomic_set(&b->pin, 0);
1875 
1876 	return 0;
1877 }
1878 
1879 static int register_cache(struct cache_sb *sb, struct page *sb_page,
1880 				struct block_device *bdev, struct cache *ca)
1881 {
1882 	char name[BDEVNAME_SIZE];
1883 	const char *err = NULL; /* must be set for any error case */
1884 	int ret = 0;
1885 
1886 	memcpy(&ca->sb, sb, sizeof(struct cache_sb));
1887 	ca->bdev = bdev;
1888 	ca->bdev->bd_holder = ca;
1889 
1890 	bio_init(&ca->sb_bio, ca->sb_bio.bi_inline_vecs, 1);
1891 	bio_first_bvec_all(&ca->sb_bio)->bv_page = sb_page;
1892 	get_page(sb_page);
1893 
1894 	if (blk_queue_discard(bdev_get_queue(ca->bdev)))
1895 		ca->discard = CACHE_DISCARD(&ca->sb);
1896 
1897 	ret = cache_alloc(ca);
1898 	if (ret != 0) {
1899 		if (ret == -ENOMEM)
1900 			err = "cache_alloc(): -ENOMEM";
1901 		else
1902 			err = "cache_alloc(): unknown error";
1903 		goto err;
1904 	}
1905 
1906 	if (kobject_add(&ca->kobj, &part_to_dev(bdev->bd_part)->kobj, "bcache")) {
1907 		err = "error calling kobject_add";
1908 		ret = -ENOMEM;
1909 		goto out;
1910 	}
1911 
1912 	mutex_lock(&bch_register_lock);
1913 	err = register_cache_set(ca);
1914 	mutex_unlock(&bch_register_lock);
1915 
1916 	if (err) {
1917 		ret = -ENODEV;
1918 		goto out;
1919 	}
1920 
1921 	pr_info("registered cache device %s", bdevname(bdev, name));
1922 
1923 out:
1924 	kobject_put(&ca->kobj);
1925 
1926 err:
1927 	if (err)
1928 		pr_notice("error opening %s: %s", bdevname(bdev, name), err);
1929 
1930 	return ret;
1931 }
1932 
1933 /* Global interfaces/init */
1934 
1935 static ssize_t register_bcache(struct kobject *, struct kobj_attribute *,
1936 			       const char *, size_t);
1937 
1938 kobj_attribute_write(register,		register_bcache);
1939 kobj_attribute_write(register_quiet,	register_bcache);
1940 
1941 static bool bch_is_open_backing(struct block_device *bdev) {
1942 	struct cache_set *c, *tc;
1943 	struct cached_dev *dc, *t;
1944 
1945 	list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
1946 		list_for_each_entry_safe(dc, t, &c->cached_devs, list)
1947 			if (dc->bdev == bdev)
1948 				return true;
1949 	list_for_each_entry_safe(dc, t, &uncached_devices, list)
1950 		if (dc->bdev == bdev)
1951 			return true;
1952 	return false;
1953 }
1954 
1955 static bool bch_is_open_cache(struct block_device *bdev) {
1956 	struct cache_set *c, *tc;
1957 	struct cache *ca;
1958 	unsigned i;
1959 
1960 	list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
1961 		for_each_cache(ca, c, i)
1962 			if (ca->bdev == bdev)
1963 				return true;
1964 	return false;
1965 }
1966 
1967 static bool bch_is_open(struct block_device *bdev) {
1968 	return bch_is_open_cache(bdev) || bch_is_open_backing(bdev);
1969 }
1970 
1971 static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr,
1972 			       const char *buffer, size_t size)
1973 {
1974 	ssize_t ret = size;
1975 	const char *err = "cannot allocate memory";
1976 	char *path = NULL;
1977 	struct cache_sb *sb = NULL;
1978 	struct block_device *bdev = NULL;
1979 	struct page *sb_page = NULL;
1980 
1981 	if (!try_module_get(THIS_MODULE))
1982 		return -EBUSY;
1983 
1984 	if (!(path = kstrndup(buffer, size, GFP_KERNEL)) ||
1985 	    !(sb = kmalloc(sizeof(struct cache_sb), GFP_KERNEL)))
1986 		goto err;
1987 
1988 	err = "failed to open device";
1989 	bdev = blkdev_get_by_path(strim(path),
1990 				  FMODE_READ|FMODE_WRITE|FMODE_EXCL,
1991 				  sb);
1992 	if (IS_ERR(bdev)) {
1993 		if (bdev == ERR_PTR(-EBUSY)) {
1994 			bdev = lookup_bdev(strim(path));
1995 			mutex_lock(&bch_register_lock);
1996 			if (!IS_ERR(bdev) && bch_is_open(bdev))
1997 				err = "device already registered";
1998 			else
1999 				err = "device busy";
2000 			mutex_unlock(&bch_register_lock);
2001 			if (!IS_ERR(bdev))
2002 				bdput(bdev);
2003 			if (attr == &ksysfs_register_quiet)
2004 				goto out;
2005 		}
2006 		goto err;
2007 	}
2008 
2009 	err = "failed to set blocksize";
2010 	if (set_blocksize(bdev, 4096))
2011 		goto err_close;
2012 
2013 	err = read_super(sb, bdev, &sb_page);
2014 	if (err)
2015 		goto err_close;
2016 
2017 	if (SB_IS_BDEV(sb)) {
2018 		struct cached_dev *dc = kzalloc(sizeof(*dc), GFP_KERNEL);
2019 		if (!dc)
2020 			goto err_close;
2021 
2022 		mutex_lock(&bch_register_lock);
2023 		register_bdev(sb, sb_page, bdev, dc);
2024 		mutex_unlock(&bch_register_lock);
2025 	} else {
2026 		struct cache *ca = kzalloc(sizeof(*ca), GFP_KERNEL);
2027 		if (!ca)
2028 			goto err_close;
2029 
2030 		if (register_cache(sb, sb_page, bdev, ca) != 0)
2031 			goto err_close;
2032 	}
2033 out:
2034 	if (sb_page)
2035 		put_page(sb_page);
2036 	kfree(sb);
2037 	kfree(path);
2038 	module_put(THIS_MODULE);
2039 	return ret;
2040 
2041 err_close:
2042 	blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2043 err:
2044 	pr_info("error opening %s: %s", path, err);
2045 	ret = -EINVAL;
2046 	goto out;
2047 }
2048 
2049 static int bcache_reboot(struct notifier_block *n, unsigned long code, void *x)
2050 {
2051 	if (code == SYS_DOWN ||
2052 	    code == SYS_HALT ||
2053 	    code == SYS_POWER_OFF) {
2054 		DEFINE_WAIT(wait);
2055 		unsigned long start = jiffies;
2056 		bool stopped = false;
2057 
2058 		struct cache_set *c, *tc;
2059 		struct cached_dev *dc, *tdc;
2060 
2061 		mutex_lock(&bch_register_lock);
2062 
2063 		if (list_empty(&bch_cache_sets) &&
2064 		    list_empty(&uncached_devices))
2065 			goto out;
2066 
2067 		pr_info("Stopping all devices:");
2068 
2069 		list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
2070 			bch_cache_set_stop(c);
2071 
2072 		list_for_each_entry_safe(dc, tdc, &uncached_devices, list)
2073 			bcache_device_stop(&dc->disk);
2074 
2075 		/* What's a condition variable? */
2076 		while (1) {
2077 			long timeout = start + 2 * HZ - jiffies;
2078 
2079 			stopped = list_empty(&bch_cache_sets) &&
2080 				list_empty(&uncached_devices);
2081 
2082 			if (timeout < 0 || stopped)
2083 				break;
2084 
2085 			prepare_to_wait(&unregister_wait, &wait,
2086 					TASK_UNINTERRUPTIBLE);
2087 
2088 			mutex_unlock(&bch_register_lock);
2089 			schedule_timeout(timeout);
2090 			mutex_lock(&bch_register_lock);
2091 		}
2092 
2093 		finish_wait(&unregister_wait, &wait);
2094 
2095 		if (stopped)
2096 			pr_info("All devices stopped");
2097 		else
2098 			pr_notice("Timeout waiting for devices to be closed");
2099 out:
2100 		mutex_unlock(&bch_register_lock);
2101 	}
2102 
2103 	return NOTIFY_DONE;
2104 }
2105 
2106 static struct notifier_block reboot = {
2107 	.notifier_call	= bcache_reboot,
2108 	.priority	= INT_MAX, /* before any real devices */
2109 };
2110 
2111 static void bcache_exit(void)
2112 {
2113 	bch_debug_exit();
2114 	bch_request_exit();
2115 	if (bcache_kobj)
2116 		kobject_put(bcache_kobj);
2117 	if (bcache_wq)
2118 		destroy_workqueue(bcache_wq);
2119 	if (bcache_major)
2120 		unregister_blkdev(bcache_major, "bcache");
2121 	unregister_reboot_notifier(&reboot);
2122 	mutex_destroy(&bch_register_lock);
2123 }
2124 
2125 static int __init bcache_init(void)
2126 {
2127 	static const struct attribute *files[] = {
2128 		&ksysfs_register.attr,
2129 		&ksysfs_register_quiet.attr,
2130 		NULL
2131 	};
2132 
2133 	mutex_init(&bch_register_lock);
2134 	init_waitqueue_head(&unregister_wait);
2135 	register_reboot_notifier(&reboot);
2136 	closure_debug_init();
2137 
2138 	bcache_major = register_blkdev(0, "bcache");
2139 	if (bcache_major < 0) {
2140 		unregister_reboot_notifier(&reboot);
2141 		mutex_destroy(&bch_register_lock);
2142 		return bcache_major;
2143 	}
2144 
2145 	if (!(bcache_wq = alloc_workqueue("bcache", WQ_MEM_RECLAIM, 0)) ||
2146 	    !(bcache_kobj = kobject_create_and_add("bcache", fs_kobj)) ||
2147 	    bch_request_init() ||
2148 	    bch_debug_init(bcache_kobj) ||
2149 	    sysfs_create_files(bcache_kobj, files))
2150 		goto err;
2151 
2152 	return 0;
2153 err:
2154 	bcache_exit();
2155 	return -ENOMEM;
2156 }
2157 
2158 module_exit(bcache_exit);
2159 module_init(bcache_init);
2160