xref: /openbmc/linux/drivers/md/bcache/super.c (revision 6479b9f4)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * bcache setup/teardown code, and some metadata io - read a superblock and
4  * figure out what to do with it.
5  *
6  * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
7  * Copyright 2012 Google, Inc.
8  */
9 
10 #include "bcache.h"
11 #include "btree.h"
12 #include "debug.h"
13 #include "extents.h"
14 #include "request.h"
15 #include "writeback.h"
16 #include "features.h"
17 
18 #include <linux/blkdev.h>
19 #include <linux/pagemap.h>
20 #include <linux/debugfs.h>
21 #include <linux/idr.h>
22 #include <linux/kthread.h>
23 #include <linux/workqueue.h>
24 #include <linux/module.h>
25 #include <linux/random.h>
26 #include <linux/reboot.h>
27 #include <linux/sysfs.h>
28 
29 unsigned int bch_cutoff_writeback;
30 unsigned int bch_cutoff_writeback_sync;
31 
32 static const char bcache_magic[] = {
33 	0xc6, 0x85, 0x73, 0xf6, 0x4e, 0x1a, 0x45, 0xca,
34 	0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81
35 };
36 
37 static const char invalid_uuid[] = {
38 	0xa0, 0x3e, 0xf8, 0xed, 0x3e, 0xe1, 0xb8, 0x78,
39 	0xc8, 0x50, 0xfc, 0x5e, 0xcb, 0x16, 0xcd, 0x99
40 };
41 
42 static struct kobject *bcache_kobj;
43 struct mutex bch_register_lock;
44 bool bcache_is_reboot;
45 LIST_HEAD(bch_cache_sets);
46 static LIST_HEAD(uncached_devices);
47 
48 static int bcache_major;
49 static DEFINE_IDA(bcache_device_idx);
50 static wait_queue_head_t unregister_wait;
51 struct workqueue_struct *bcache_wq;
52 struct workqueue_struct *bch_flush_wq;
53 struct workqueue_struct *bch_journal_wq;
54 
55 
56 #define BTREE_MAX_PAGES		(256 * 1024 / PAGE_SIZE)
57 /* limitation of partitions number on single bcache device */
58 #define BCACHE_MINORS		128
59 /* limitation of bcache devices number on single system */
60 #define BCACHE_DEVICE_IDX_MAX	((1U << MINORBITS)/BCACHE_MINORS)
61 
62 /* Superblock */
63 
get_bucket_size(struct cache_sb * sb,struct cache_sb_disk * s)64 static unsigned int get_bucket_size(struct cache_sb *sb, struct cache_sb_disk *s)
65 {
66 	unsigned int bucket_size = le16_to_cpu(s->bucket_size);
67 
68 	if (sb->version >= BCACHE_SB_VERSION_CDEV_WITH_FEATURES) {
69 		if (bch_has_feature_large_bucket(sb)) {
70 			unsigned int max, order;
71 
72 			max = sizeof(unsigned int) * BITS_PER_BYTE - 1;
73 			order = le16_to_cpu(s->bucket_size);
74 			/*
75 			 * bcache tool will make sure the overflow won't
76 			 * happen, an error message here is enough.
77 			 */
78 			if (order > max)
79 				pr_err("Bucket size (1 << %u) overflows\n",
80 					order);
81 			bucket_size = 1 << order;
82 		} else if (bch_has_feature_obso_large_bucket(sb)) {
83 			bucket_size +=
84 				le16_to_cpu(s->obso_bucket_size_hi) << 16;
85 		}
86 	}
87 
88 	return bucket_size;
89 }
90 
read_super_common(struct cache_sb * sb,struct block_device * bdev,struct cache_sb_disk * s)91 static const char *read_super_common(struct cache_sb *sb,  struct block_device *bdev,
92 				     struct cache_sb_disk *s)
93 {
94 	const char *err;
95 	unsigned int i;
96 
97 	sb->first_bucket= le16_to_cpu(s->first_bucket);
98 	sb->nbuckets	= le64_to_cpu(s->nbuckets);
99 	sb->bucket_size	= get_bucket_size(sb, s);
100 
101 	sb->nr_in_set	= le16_to_cpu(s->nr_in_set);
102 	sb->nr_this_dev	= le16_to_cpu(s->nr_this_dev);
103 
104 	err = "Too many journal buckets";
105 	if (sb->keys > SB_JOURNAL_BUCKETS)
106 		goto err;
107 
108 	err = "Too many buckets";
109 	if (sb->nbuckets > LONG_MAX)
110 		goto err;
111 
112 	err = "Not enough buckets";
113 	if (sb->nbuckets < 1 << 7)
114 		goto err;
115 
116 	err = "Bad block size (not power of 2)";
117 	if (!is_power_of_2(sb->block_size))
118 		goto err;
119 
120 	err = "Bad block size (larger than page size)";
121 	if (sb->block_size > PAGE_SECTORS)
122 		goto err;
123 
124 	err = "Bad bucket size (not power of 2)";
125 	if (!is_power_of_2(sb->bucket_size))
126 		goto err;
127 
128 	err = "Bad bucket size (smaller than page size)";
129 	if (sb->bucket_size < PAGE_SECTORS)
130 		goto err;
131 
132 	err = "Invalid superblock: device too small";
133 	if (get_capacity(bdev->bd_disk) <
134 	    sb->bucket_size * sb->nbuckets)
135 		goto err;
136 
137 	err = "Bad UUID";
138 	if (bch_is_zero(sb->set_uuid, 16))
139 		goto err;
140 
141 	err = "Bad cache device number in set";
142 	if (!sb->nr_in_set ||
143 	    sb->nr_in_set <= sb->nr_this_dev ||
144 	    sb->nr_in_set > MAX_CACHES_PER_SET)
145 		goto err;
146 
147 	err = "Journal buckets not sequential";
148 	for (i = 0; i < sb->keys; i++)
149 		if (sb->d[i] != sb->first_bucket + i)
150 			goto err;
151 
152 	err = "Too many journal buckets";
153 	if (sb->first_bucket + sb->keys > sb->nbuckets)
154 		goto err;
155 
156 	err = "Invalid superblock: first bucket comes before end of super";
157 	if (sb->first_bucket * sb->bucket_size < 16)
158 		goto err;
159 
160 	err = NULL;
161 err:
162 	return err;
163 }
164 
165 
read_super(struct cache_sb * sb,struct block_device * bdev,struct cache_sb_disk ** res)166 static const char *read_super(struct cache_sb *sb, struct block_device *bdev,
167 			      struct cache_sb_disk **res)
168 {
169 	const char *err;
170 	struct cache_sb_disk *s;
171 	struct page *page;
172 	unsigned int i;
173 
174 	page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
175 				   SB_OFFSET >> PAGE_SHIFT, GFP_KERNEL);
176 	if (IS_ERR(page))
177 		return "IO error";
178 	s = page_address(page) + offset_in_page(SB_OFFSET);
179 
180 	sb->offset		= le64_to_cpu(s->offset);
181 	sb->version		= le64_to_cpu(s->version);
182 
183 	memcpy(sb->magic,	s->magic, 16);
184 	memcpy(sb->uuid,	s->uuid, 16);
185 	memcpy(sb->set_uuid,	s->set_uuid, 16);
186 	memcpy(sb->label,	s->label, SB_LABEL_SIZE);
187 
188 	sb->flags		= le64_to_cpu(s->flags);
189 	sb->seq			= le64_to_cpu(s->seq);
190 	sb->last_mount		= le32_to_cpu(s->last_mount);
191 	sb->keys		= le16_to_cpu(s->keys);
192 
193 	for (i = 0; i < SB_JOURNAL_BUCKETS; i++)
194 		sb->d[i] = le64_to_cpu(s->d[i]);
195 
196 	pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u\n",
197 		 sb->version, sb->flags, sb->seq, sb->keys);
198 
199 	err = "Not a bcache superblock (bad offset)";
200 	if (sb->offset != SB_SECTOR)
201 		goto err;
202 
203 	err = "Not a bcache superblock (bad magic)";
204 	if (memcmp(sb->magic, bcache_magic, 16))
205 		goto err;
206 
207 	err = "Bad checksum";
208 	if (s->csum != csum_set(s))
209 		goto err;
210 
211 	err = "Bad UUID";
212 	if (bch_is_zero(sb->uuid, 16))
213 		goto err;
214 
215 	sb->block_size	= le16_to_cpu(s->block_size);
216 
217 	err = "Superblock block size smaller than device block size";
218 	if (sb->block_size << 9 < bdev_logical_block_size(bdev))
219 		goto err;
220 
221 	switch (sb->version) {
222 	case BCACHE_SB_VERSION_BDEV:
223 		sb->data_offset	= BDEV_DATA_START_DEFAULT;
224 		break;
225 	case BCACHE_SB_VERSION_BDEV_WITH_OFFSET:
226 	case BCACHE_SB_VERSION_BDEV_WITH_FEATURES:
227 		sb->data_offset	= le64_to_cpu(s->data_offset);
228 
229 		err = "Bad data offset";
230 		if (sb->data_offset < BDEV_DATA_START_DEFAULT)
231 			goto err;
232 
233 		break;
234 	case BCACHE_SB_VERSION_CDEV:
235 	case BCACHE_SB_VERSION_CDEV_WITH_UUID:
236 		err = read_super_common(sb, bdev, s);
237 		if (err)
238 			goto err;
239 		break;
240 	case BCACHE_SB_VERSION_CDEV_WITH_FEATURES:
241 		/*
242 		 * Feature bits are needed in read_super_common(),
243 		 * convert them firstly.
244 		 */
245 		sb->feature_compat = le64_to_cpu(s->feature_compat);
246 		sb->feature_incompat = le64_to_cpu(s->feature_incompat);
247 		sb->feature_ro_compat = le64_to_cpu(s->feature_ro_compat);
248 
249 		/* Check incompatible features */
250 		err = "Unsupported compatible feature found";
251 		if (bch_has_unknown_compat_features(sb))
252 			goto err;
253 
254 		err = "Unsupported read-only compatible feature found";
255 		if (bch_has_unknown_ro_compat_features(sb))
256 			goto err;
257 
258 		err = "Unsupported incompatible feature found";
259 		if (bch_has_unknown_incompat_features(sb))
260 			goto err;
261 
262 		err = read_super_common(sb, bdev, s);
263 		if (err)
264 			goto err;
265 		break;
266 	default:
267 		err = "Unsupported superblock version";
268 		goto err;
269 	}
270 
271 	sb->last_mount = (u32)ktime_get_real_seconds();
272 	*res = s;
273 	return NULL;
274 err:
275 	put_page(page);
276 	return err;
277 }
278 
write_bdev_super_endio(struct bio * bio)279 static void write_bdev_super_endio(struct bio *bio)
280 {
281 	struct cached_dev *dc = bio->bi_private;
282 
283 	if (bio->bi_status)
284 		bch_count_backing_io_errors(dc, bio);
285 
286 	closure_put(&dc->sb_write);
287 }
288 
__write_super(struct cache_sb * sb,struct cache_sb_disk * out,struct bio * bio)289 static void __write_super(struct cache_sb *sb, struct cache_sb_disk *out,
290 		struct bio *bio)
291 {
292 	unsigned int i;
293 
294 	bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META;
295 	bio->bi_iter.bi_sector	= SB_SECTOR;
296 	__bio_add_page(bio, virt_to_page(out), SB_SIZE,
297 			offset_in_page(out));
298 
299 	out->offset		= cpu_to_le64(sb->offset);
300 
301 	memcpy(out->uuid,	sb->uuid, 16);
302 	memcpy(out->set_uuid,	sb->set_uuid, 16);
303 	memcpy(out->label,	sb->label, SB_LABEL_SIZE);
304 
305 	out->flags		= cpu_to_le64(sb->flags);
306 	out->seq		= cpu_to_le64(sb->seq);
307 
308 	out->last_mount		= cpu_to_le32(sb->last_mount);
309 	out->first_bucket	= cpu_to_le16(sb->first_bucket);
310 	out->keys		= cpu_to_le16(sb->keys);
311 
312 	for (i = 0; i < sb->keys; i++)
313 		out->d[i] = cpu_to_le64(sb->d[i]);
314 
315 	if (sb->version >= BCACHE_SB_VERSION_CDEV_WITH_FEATURES) {
316 		out->feature_compat    = cpu_to_le64(sb->feature_compat);
317 		out->feature_incompat  = cpu_to_le64(sb->feature_incompat);
318 		out->feature_ro_compat = cpu_to_le64(sb->feature_ro_compat);
319 	}
320 
321 	out->version		= cpu_to_le64(sb->version);
322 	out->csum = csum_set(out);
323 
324 	pr_debug("ver %llu, flags %llu, seq %llu\n",
325 		 sb->version, sb->flags, sb->seq);
326 
327 	submit_bio(bio);
328 }
329 
bch_write_bdev_super_unlock(struct closure * cl)330 static void bch_write_bdev_super_unlock(struct closure *cl)
331 {
332 	struct cached_dev *dc = container_of(cl, struct cached_dev, sb_write);
333 
334 	up(&dc->sb_write_mutex);
335 }
336 
bch_write_bdev_super(struct cached_dev * dc,struct closure * parent)337 void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent)
338 {
339 	struct closure *cl = &dc->sb_write;
340 	struct bio *bio = &dc->sb_bio;
341 
342 	down(&dc->sb_write_mutex);
343 	closure_init(cl, parent);
344 
345 	bio_init(bio, dc->bdev, dc->sb_bv, 1, 0);
346 	bio->bi_end_io	= write_bdev_super_endio;
347 	bio->bi_private = dc;
348 
349 	closure_get(cl);
350 	/* I/O request sent to backing device */
351 	__write_super(&dc->sb, dc->sb_disk, bio);
352 
353 	closure_return_with_destructor(cl, bch_write_bdev_super_unlock);
354 }
355 
write_super_endio(struct bio * bio)356 static void write_super_endio(struct bio *bio)
357 {
358 	struct cache *ca = bio->bi_private;
359 
360 	/* is_read = 0 */
361 	bch_count_io_errors(ca, bio->bi_status, 0,
362 			    "writing superblock");
363 	closure_put(&ca->set->sb_write);
364 }
365 
bcache_write_super_unlock(struct closure * cl)366 static void bcache_write_super_unlock(struct closure *cl)
367 {
368 	struct cache_set *c = container_of(cl, struct cache_set, sb_write);
369 
370 	up(&c->sb_write_mutex);
371 }
372 
bcache_write_super(struct cache_set * c)373 void bcache_write_super(struct cache_set *c)
374 {
375 	struct closure *cl = &c->sb_write;
376 	struct cache *ca = c->cache;
377 	struct bio *bio = &ca->sb_bio;
378 	unsigned int version = BCACHE_SB_VERSION_CDEV_WITH_UUID;
379 
380 	down(&c->sb_write_mutex);
381 	closure_init(cl, &c->cl);
382 
383 	ca->sb.seq++;
384 
385 	if (ca->sb.version < version)
386 		ca->sb.version = version;
387 
388 	bio_init(bio, ca->bdev, ca->sb_bv, 1, 0);
389 	bio->bi_end_io	= write_super_endio;
390 	bio->bi_private = ca;
391 
392 	closure_get(cl);
393 	__write_super(&ca->sb, ca->sb_disk, bio);
394 
395 	closure_return_with_destructor(cl, bcache_write_super_unlock);
396 }
397 
398 /* UUID io */
399 
uuid_endio(struct bio * bio)400 static void uuid_endio(struct bio *bio)
401 {
402 	struct closure *cl = bio->bi_private;
403 	struct cache_set *c = container_of(cl, struct cache_set, uuid_write);
404 
405 	cache_set_err_on(bio->bi_status, c, "accessing uuids");
406 	bch_bbio_free(bio, c);
407 	closure_put(cl);
408 }
409 
uuid_io_unlock(struct closure * cl)410 static void uuid_io_unlock(struct closure *cl)
411 {
412 	struct cache_set *c = container_of(cl, struct cache_set, uuid_write);
413 
414 	up(&c->uuid_write_mutex);
415 }
416 
uuid_io(struct cache_set * c,blk_opf_t opf,struct bkey * k,struct closure * parent)417 static void uuid_io(struct cache_set *c, blk_opf_t opf, struct bkey *k,
418 		    struct closure *parent)
419 {
420 	struct closure *cl = &c->uuid_write;
421 	struct uuid_entry *u;
422 	unsigned int i;
423 	char buf[80];
424 
425 	BUG_ON(!parent);
426 	down(&c->uuid_write_mutex);
427 	closure_init(cl, parent);
428 
429 	for (i = 0; i < KEY_PTRS(k); i++) {
430 		struct bio *bio = bch_bbio_alloc(c);
431 
432 		bio->bi_opf = opf | REQ_SYNC | REQ_META;
433 		bio->bi_iter.bi_size = KEY_SIZE(k) << 9;
434 
435 		bio->bi_end_io	= uuid_endio;
436 		bio->bi_private = cl;
437 		bch_bio_map(bio, c->uuids);
438 
439 		bch_submit_bbio(bio, c, k, i);
440 
441 		if ((opf & REQ_OP_MASK) != REQ_OP_WRITE)
442 			break;
443 	}
444 
445 	bch_extent_to_text(buf, sizeof(buf), k);
446 	pr_debug("%s UUIDs at %s\n", (opf & REQ_OP_MASK) == REQ_OP_WRITE ?
447 		 "wrote" : "read", buf);
448 
449 	for (u = c->uuids; u < c->uuids + c->nr_uuids; u++)
450 		if (!bch_is_zero(u->uuid, 16))
451 			pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u\n",
452 				 u - c->uuids, u->uuid, u->label,
453 				 u->first_reg, u->last_reg, u->invalidated);
454 
455 	closure_return_with_destructor(cl, uuid_io_unlock);
456 }
457 
uuid_read(struct cache_set * c,struct jset * j,struct closure * cl)458 static char *uuid_read(struct cache_set *c, struct jset *j, struct closure *cl)
459 {
460 	struct bkey *k = &j->uuid_bucket;
461 
462 	if (__bch_btree_ptr_invalid(c, k))
463 		return "bad uuid pointer";
464 
465 	bkey_copy(&c->uuid_bucket, k);
466 	uuid_io(c, REQ_OP_READ, k, cl);
467 
468 	if (j->version < BCACHE_JSET_VERSION_UUIDv1) {
469 		struct uuid_entry_v0	*u0 = (void *) c->uuids;
470 		struct uuid_entry	*u1 = (void *) c->uuids;
471 		int i;
472 
473 		closure_sync(cl);
474 
475 		/*
476 		 * Since the new uuid entry is bigger than the old, we have to
477 		 * convert starting at the highest memory address and work down
478 		 * in order to do it in place
479 		 */
480 
481 		for (i = c->nr_uuids - 1;
482 		     i >= 0;
483 		     --i) {
484 			memcpy(u1[i].uuid,	u0[i].uuid, 16);
485 			memcpy(u1[i].label,	u0[i].label, 32);
486 
487 			u1[i].first_reg		= u0[i].first_reg;
488 			u1[i].last_reg		= u0[i].last_reg;
489 			u1[i].invalidated	= u0[i].invalidated;
490 
491 			u1[i].flags	= 0;
492 			u1[i].sectors	= 0;
493 		}
494 	}
495 
496 	return NULL;
497 }
498 
__uuid_write(struct cache_set * c)499 static int __uuid_write(struct cache_set *c)
500 {
501 	BKEY_PADDED(key) k;
502 	struct closure cl;
503 	struct cache *ca = c->cache;
504 	unsigned int size;
505 
506 	closure_init_stack(&cl);
507 	lockdep_assert_held(&bch_register_lock);
508 
509 	if (bch_bucket_alloc_set(c, RESERVE_BTREE, &k.key, true))
510 		return 1;
511 
512 	size =  meta_bucket_pages(&ca->sb) * PAGE_SECTORS;
513 	SET_KEY_SIZE(&k.key, size);
514 	uuid_io(c, REQ_OP_WRITE, &k.key, &cl);
515 	closure_sync(&cl);
516 
517 	/* Only one bucket used for uuid write */
518 	atomic_long_add(ca->sb.bucket_size, &ca->meta_sectors_written);
519 
520 	bkey_copy(&c->uuid_bucket, &k.key);
521 	bkey_put(c, &k.key);
522 	return 0;
523 }
524 
bch_uuid_write(struct cache_set * c)525 int bch_uuid_write(struct cache_set *c)
526 {
527 	int ret = __uuid_write(c);
528 
529 	if (!ret)
530 		bch_journal_meta(c, NULL);
531 
532 	return ret;
533 }
534 
uuid_find(struct cache_set * c,const char * uuid)535 static struct uuid_entry *uuid_find(struct cache_set *c, const char *uuid)
536 {
537 	struct uuid_entry *u;
538 
539 	for (u = c->uuids;
540 	     u < c->uuids + c->nr_uuids; u++)
541 		if (!memcmp(u->uuid, uuid, 16))
542 			return u;
543 
544 	return NULL;
545 }
546 
uuid_find_empty(struct cache_set * c)547 static struct uuid_entry *uuid_find_empty(struct cache_set *c)
548 {
549 	static const char zero_uuid[16] = "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0";
550 
551 	return uuid_find(c, zero_uuid);
552 }
553 
554 /*
555  * Bucket priorities/gens:
556  *
557  * For each bucket, we store on disk its
558  *   8 bit gen
559  *  16 bit priority
560  *
561  * See alloc.c for an explanation of the gen. The priority is used to implement
562  * lru (and in the future other) cache replacement policies; for most purposes
563  * it's just an opaque integer.
564  *
565  * The gens and the priorities don't have a whole lot to do with each other, and
566  * it's actually the gens that must be written out at specific times - it's no
567  * big deal if the priorities don't get written, if we lose them we just reuse
568  * buckets in suboptimal order.
569  *
570  * On disk they're stored in a packed array, and in as many buckets are required
571  * to fit them all. The buckets we use to store them form a list; the journal
572  * header points to the first bucket, the first bucket points to the second
573  * bucket, et cetera.
574  *
575  * This code is used by the allocation code; periodically (whenever it runs out
576  * of buckets to allocate from) the allocation code will invalidate some
577  * buckets, but it can't use those buckets until their new gens are safely on
578  * disk.
579  */
580 
prio_endio(struct bio * bio)581 static void prio_endio(struct bio *bio)
582 {
583 	struct cache *ca = bio->bi_private;
584 
585 	cache_set_err_on(bio->bi_status, ca->set, "accessing priorities");
586 	bch_bbio_free(bio, ca->set);
587 	closure_put(&ca->prio);
588 }
589 
prio_io(struct cache * ca,uint64_t bucket,blk_opf_t opf)590 static void prio_io(struct cache *ca, uint64_t bucket, blk_opf_t opf)
591 {
592 	struct closure *cl = &ca->prio;
593 	struct bio *bio = bch_bbio_alloc(ca->set);
594 
595 	closure_init_stack(cl);
596 
597 	bio->bi_iter.bi_sector	= bucket * ca->sb.bucket_size;
598 	bio_set_dev(bio, ca->bdev);
599 	bio->bi_iter.bi_size	= meta_bucket_bytes(&ca->sb);
600 
601 	bio->bi_end_io	= prio_endio;
602 	bio->bi_private = ca;
603 	bio->bi_opf = opf | REQ_SYNC | REQ_META;
604 	bch_bio_map(bio, ca->disk_buckets);
605 
606 	closure_bio_submit(ca->set, bio, &ca->prio);
607 	closure_sync(cl);
608 }
609 
bch_prio_write(struct cache * ca,bool wait)610 int bch_prio_write(struct cache *ca, bool wait)
611 {
612 	int i;
613 	struct bucket *b;
614 	struct closure cl;
615 
616 	pr_debug("free_prio=%zu, free_none=%zu, free_inc=%zu\n",
617 		 fifo_used(&ca->free[RESERVE_PRIO]),
618 		 fifo_used(&ca->free[RESERVE_NONE]),
619 		 fifo_used(&ca->free_inc));
620 
621 	/*
622 	 * Pre-check if there are enough free buckets. In the non-blocking
623 	 * scenario it's better to fail early rather than starting to allocate
624 	 * buckets and do a cleanup later in case of failure.
625 	 */
626 	if (!wait) {
627 		size_t avail = fifo_used(&ca->free[RESERVE_PRIO]) +
628 			       fifo_used(&ca->free[RESERVE_NONE]);
629 		if (prio_buckets(ca) > avail)
630 			return -ENOMEM;
631 	}
632 
633 	closure_init_stack(&cl);
634 
635 	lockdep_assert_held(&ca->set->bucket_lock);
636 
637 	ca->disk_buckets->seq++;
638 
639 	atomic_long_add(ca->sb.bucket_size * prio_buckets(ca),
640 			&ca->meta_sectors_written);
641 
642 	for (i = prio_buckets(ca) - 1; i >= 0; --i) {
643 		long bucket;
644 		struct prio_set *p = ca->disk_buckets;
645 		struct bucket_disk *d = p->data;
646 		struct bucket_disk *end = d + prios_per_bucket(ca);
647 
648 		for (b = ca->buckets + i * prios_per_bucket(ca);
649 		     b < ca->buckets + ca->sb.nbuckets && d < end;
650 		     b++, d++) {
651 			d->prio = cpu_to_le16(b->prio);
652 			d->gen = b->gen;
653 		}
654 
655 		p->next_bucket	= ca->prio_buckets[i + 1];
656 		p->magic	= pset_magic(&ca->sb);
657 		p->csum		= bch_crc64(&p->magic, meta_bucket_bytes(&ca->sb) - 8);
658 
659 		bucket = bch_bucket_alloc(ca, RESERVE_PRIO, wait);
660 		BUG_ON(bucket == -1);
661 
662 		mutex_unlock(&ca->set->bucket_lock);
663 		prio_io(ca, bucket, REQ_OP_WRITE);
664 		mutex_lock(&ca->set->bucket_lock);
665 
666 		ca->prio_buckets[i] = bucket;
667 		atomic_dec_bug(&ca->buckets[bucket].pin);
668 	}
669 
670 	mutex_unlock(&ca->set->bucket_lock);
671 
672 	bch_journal_meta(ca->set, &cl);
673 	closure_sync(&cl);
674 
675 	mutex_lock(&ca->set->bucket_lock);
676 
677 	/*
678 	 * Don't want the old priorities to get garbage collected until after we
679 	 * finish writing the new ones, and they're journalled
680 	 */
681 	for (i = 0; i < prio_buckets(ca); i++) {
682 		if (ca->prio_last_buckets[i])
683 			__bch_bucket_free(ca,
684 				&ca->buckets[ca->prio_last_buckets[i]]);
685 
686 		ca->prio_last_buckets[i] = ca->prio_buckets[i];
687 	}
688 	return 0;
689 }
690 
prio_read(struct cache * ca,uint64_t bucket)691 static int prio_read(struct cache *ca, uint64_t bucket)
692 {
693 	struct prio_set *p = ca->disk_buckets;
694 	struct bucket_disk *d = p->data + prios_per_bucket(ca), *end = d;
695 	struct bucket *b;
696 	unsigned int bucket_nr = 0;
697 	int ret = -EIO;
698 
699 	for (b = ca->buckets;
700 	     b < ca->buckets + ca->sb.nbuckets;
701 	     b++, d++) {
702 		if (d == end) {
703 			ca->prio_buckets[bucket_nr] = bucket;
704 			ca->prio_last_buckets[bucket_nr] = bucket;
705 			bucket_nr++;
706 
707 			prio_io(ca, bucket, REQ_OP_READ);
708 
709 			if (p->csum !=
710 			    bch_crc64(&p->magic, meta_bucket_bytes(&ca->sb) - 8)) {
711 				pr_warn("bad csum reading priorities\n");
712 				goto out;
713 			}
714 
715 			if (p->magic != pset_magic(&ca->sb)) {
716 				pr_warn("bad magic reading priorities\n");
717 				goto out;
718 			}
719 
720 			bucket = p->next_bucket;
721 			d = p->data;
722 		}
723 
724 		b->prio = le16_to_cpu(d->prio);
725 		b->gen = b->last_gc = d->gen;
726 	}
727 
728 	ret = 0;
729 out:
730 	return ret;
731 }
732 
733 /* Bcache device */
734 
open_dev(struct gendisk * disk,blk_mode_t mode)735 static int open_dev(struct gendisk *disk, blk_mode_t mode)
736 {
737 	struct bcache_device *d = disk->private_data;
738 
739 	if (test_bit(BCACHE_DEV_CLOSING, &d->flags))
740 		return -ENXIO;
741 
742 	closure_get(&d->cl);
743 	return 0;
744 }
745 
release_dev(struct gendisk * b)746 static void release_dev(struct gendisk *b)
747 {
748 	struct bcache_device *d = b->private_data;
749 
750 	closure_put(&d->cl);
751 }
752 
ioctl_dev(struct block_device * b,blk_mode_t mode,unsigned int cmd,unsigned long arg)753 static int ioctl_dev(struct block_device *b, blk_mode_t mode,
754 		     unsigned int cmd, unsigned long arg)
755 {
756 	struct bcache_device *d = b->bd_disk->private_data;
757 
758 	return d->ioctl(d, mode, cmd, arg);
759 }
760 
761 static const struct block_device_operations bcache_cached_ops = {
762 	.submit_bio	= cached_dev_submit_bio,
763 	.open		= open_dev,
764 	.release	= release_dev,
765 	.ioctl		= ioctl_dev,
766 	.owner		= THIS_MODULE,
767 };
768 
769 static const struct block_device_operations bcache_flash_ops = {
770 	.submit_bio	= flash_dev_submit_bio,
771 	.open		= open_dev,
772 	.release	= release_dev,
773 	.ioctl		= ioctl_dev,
774 	.owner		= THIS_MODULE,
775 };
776 
bcache_device_stop(struct bcache_device * d)777 void bcache_device_stop(struct bcache_device *d)
778 {
779 	if (!test_and_set_bit(BCACHE_DEV_CLOSING, &d->flags))
780 		/*
781 		 * closure_fn set to
782 		 * - cached device: cached_dev_flush()
783 		 * - flash dev: flash_dev_flush()
784 		 */
785 		closure_queue(&d->cl);
786 }
787 
bcache_device_unlink(struct bcache_device * d)788 static void bcache_device_unlink(struct bcache_device *d)
789 {
790 	lockdep_assert_held(&bch_register_lock);
791 
792 	if (d->c && !test_and_set_bit(BCACHE_DEV_UNLINK_DONE, &d->flags)) {
793 		struct cache *ca = d->c->cache;
794 
795 		sysfs_remove_link(&d->c->kobj, d->name);
796 		sysfs_remove_link(&d->kobj, "cache");
797 
798 		bd_unlink_disk_holder(ca->bdev, d->disk);
799 	}
800 }
801 
bcache_device_link(struct bcache_device * d,struct cache_set * c,const char * name)802 static void bcache_device_link(struct bcache_device *d, struct cache_set *c,
803 			       const char *name)
804 {
805 	struct cache *ca = c->cache;
806 	int ret;
807 
808 	bd_link_disk_holder(ca->bdev, d->disk);
809 
810 	snprintf(d->name, BCACHEDEVNAME_SIZE,
811 		 "%s%u", name, d->id);
812 
813 	ret = sysfs_create_link(&d->kobj, &c->kobj, "cache");
814 	if (ret < 0)
815 		pr_err("Couldn't create device -> cache set symlink\n");
816 
817 	ret = sysfs_create_link(&c->kobj, &d->kobj, d->name);
818 	if (ret < 0)
819 		pr_err("Couldn't create cache set -> device symlink\n");
820 
821 	clear_bit(BCACHE_DEV_UNLINK_DONE, &d->flags);
822 }
823 
bcache_device_detach(struct bcache_device * d)824 static void bcache_device_detach(struct bcache_device *d)
825 {
826 	lockdep_assert_held(&bch_register_lock);
827 
828 	atomic_dec(&d->c->attached_dev_nr);
829 
830 	if (test_bit(BCACHE_DEV_DETACHING, &d->flags)) {
831 		struct uuid_entry *u = d->c->uuids + d->id;
832 
833 		SET_UUID_FLASH_ONLY(u, 0);
834 		memcpy(u->uuid, invalid_uuid, 16);
835 		u->invalidated = cpu_to_le32((u32)ktime_get_real_seconds());
836 		bch_uuid_write(d->c);
837 	}
838 
839 	bcache_device_unlink(d);
840 
841 	d->c->devices[d->id] = NULL;
842 	closure_put(&d->c->caching);
843 	d->c = NULL;
844 }
845 
bcache_device_attach(struct bcache_device * d,struct cache_set * c,unsigned int id)846 static void bcache_device_attach(struct bcache_device *d, struct cache_set *c,
847 				 unsigned int id)
848 {
849 	d->id = id;
850 	d->c = c;
851 	c->devices[id] = d;
852 
853 	if (id >= c->devices_max_used)
854 		c->devices_max_used = id + 1;
855 
856 	closure_get(&c->caching);
857 }
858 
first_minor_to_idx(int first_minor)859 static inline int first_minor_to_idx(int first_minor)
860 {
861 	return (first_minor/BCACHE_MINORS);
862 }
863 
idx_to_first_minor(int idx)864 static inline int idx_to_first_minor(int idx)
865 {
866 	return (idx * BCACHE_MINORS);
867 }
868 
bcache_device_free(struct bcache_device * d)869 static void bcache_device_free(struct bcache_device *d)
870 {
871 	struct gendisk *disk = d->disk;
872 
873 	lockdep_assert_held(&bch_register_lock);
874 
875 	if (disk)
876 		pr_info("%s stopped\n", disk->disk_name);
877 	else
878 		pr_err("bcache device (NULL gendisk) stopped\n");
879 
880 	if (d->c)
881 		bcache_device_detach(d);
882 
883 	if (disk) {
884 		ida_simple_remove(&bcache_device_idx,
885 				  first_minor_to_idx(disk->first_minor));
886 		put_disk(disk);
887 	}
888 
889 	bioset_exit(&d->bio_split);
890 	kvfree(d->full_dirty_stripes);
891 	kvfree(d->stripe_sectors_dirty);
892 
893 	closure_debug_destroy(&d->cl);
894 }
895 
bcache_device_init(struct bcache_device * d,unsigned int block_size,sector_t sectors,struct block_device * cached_bdev,const struct block_device_operations * ops)896 static int bcache_device_init(struct bcache_device *d, unsigned int block_size,
897 		sector_t sectors, struct block_device *cached_bdev,
898 		const struct block_device_operations *ops)
899 {
900 	struct request_queue *q;
901 	const size_t max_stripes = min_t(size_t, INT_MAX,
902 					 SIZE_MAX / sizeof(atomic_t));
903 	uint64_t n;
904 	int idx;
905 
906 	if (!d->stripe_size)
907 		d->stripe_size = 1 << 31;
908 	else if (d->stripe_size < BCH_MIN_STRIPE_SZ)
909 		d->stripe_size = roundup(BCH_MIN_STRIPE_SZ, d->stripe_size);
910 
911 	n = DIV_ROUND_UP_ULL(sectors, d->stripe_size);
912 	if (!n || n > max_stripes) {
913 		pr_err("nr_stripes too large or invalid: %llu (start sector beyond end of disk?)\n",
914 			n);
915 		return -ENOMEM;
916 	}
917 	d->nr_stripes = n;
918 
919 	n = d->nr_stripes * sizeof(atomic_t);
920 	d->stripe_sectors_dirty = kvzalloc(n, GFP_KERNEL);
921 	if (!d->stripe_sectors_dirty)
922 		return -ENOMEM;
923 
924 	n = BITS_TO_LONGS(d->nr_stripes) * sizeof(unsigned long);
925 	d->full_dirty_stripes = kvzalloc(n, GFP_KERNEL);
926 	if (!d->full_dirty_stripes)
927 		goto out_free_stripe_sectors_dirty;
928 
929 	idx = ida_simple_get(&bcache_device_idx, 0,
930 				BCACHE_DEVICE_IDX_MAX, GFP_KERNEL);
931 	if (idx < 0)
932 		goto out_free_full_dirty_stripes;
933 
934 	if (bioset_init(&d->bio_split, 4, offsetof(struct bbio, bio),
935 			BIOSET_NEED_BVECS|BIOSET_NEED_RESCUER))
936 		goto out_ida_remove;
937 
938 	d->disk = blk_alloc_disk(NUMA_NO_NODE);
939 	if (!d->disk)
940 		goto out_bioset_exit;
941 
942 	set_capacity(d->disk, sectors);
943 	snprintf(d->disk->disk_name, DISK_NAME_LEN, "bcache%i", idx);
944 
945 	d->disk->major		= bcache_major;
946 	d->disk->first_minor	= idx_to_first_minor(idx);
947 	d->disk->minors		= BCACHE_MINORS;
948 	d->disk->fops		= ops;
949 	d->disk->private_data	= d;
950 
951 	q = d->disk->queue;
952 	q->limits.max_hw_sectors	= UINT_MAX;
953 	q->limits.max_sectors		= UINT_MAX;
954 	q->limits.max_segment_size	= UINT_MAX;
955 	q->limits.max_segments		= BIO_MAX_VECS;
956 	blk_queue_max_discard_sectors(q, UINT_MAX);
957 	q->limits.discard_granularity	= 512;
958 	q->limits.io_min		= block_size;
959 	q->limits.logical_block_size	= block_size;
960 	q->limits.physical_block_size	= block_size;
961 
962 	if (q->limits.logical_block_size > PAGE_SIZE && cached_bdev) {
963 		/*
964 		 * This should only happen with BCACHE_SB_VERSION_BDEV.
965 		 * Block/page size is checked for BCACHE_SB_VERSION_CDEV.
966 		 */
967 		pr_info("%s: sb/logical block size (%u) greater than page size (%lu) falling back to device logical block size (%u)\n",
968 			d->disk->disk_name, q->limits.logical_block_size,
969 			PAGE_SIZE, bdev_logical_block_size(cached_bdev));
970 
971 		/* This also adjusts physical block size/min io size if needed */
972 		blk_queue_logical_block_size(q, bdev_logical_block_size(cached_bdev));
973 	}
974 
975 	blk_queue_flag_set(QUEUE_FLAG_NONROT, d->disk->queue);
976 
977 	blk_queue_write_cache(q, true, true);
978 
979 	return 0;
980 
981 out_bioset_exit:
982 	bioset_exit(&d->bio_split);
983 out_ida_remove:
984 	ida_simple_remove(&bcache_device_idx, idx);
985 out_free_full_dirty_stripes:
986 	kvfree(d->full_dirty_stripes);
987 out_free_stripe_sectors_dirty:
988 	kvfree(d->stripe_sectors_dirty);
989 	return -ENOMEM;
990 
991 }
992 
993 /* Cached device */
994 
calc_cached_dev_sectors(struct cache_set * c)995 static void calc_cached_dev_sectors(struct cache_set *c)
996 {
997 	uint64_t sectors = 0;
998 	struct cached_dev *dc;
999 
1000 	list_for_each_entry(dc, &c->cached_devs, list)
1001 		sectors += bdev_nr_sectors(dc->bdev);
1002 
1003 	c->cached_dev_sectors = sectors;
1004 }
1005 
1006 #define BACKING_DEV_OFFLINE_TIMEOUT 5
cached_dev_status_update(void * arg)1007 static int cached_dev_status_update(void *arg)
1008 {
1009 	struct cached_dev *dc = arg;
1010 	struct request_queue *q;
1011 
1012 	/*
1013 	 * If this delayed worker is stopping outside, directly quit here.
1014 	 * dc->io_disable might be set via sysfs interface, so check it
1015 	 * here too.
1016 	 */
1017 	while (!kthread_should_stop() && !dc->io_disable) {
1018 		q = bdev_get_queue(dc->bdev);
1019 		if (blk_queue_dying(q))
1020 			dc->offline_seconds++;
1021 		else
1022 			dc->offline_seconds = 0;
1023 
1024 		if (dc->offline_seconds >= BACKING_DEV_OFFLINE_TIMEOUT) {
1025 			pr_err("%pg: device offline for %d seconds\n",
1026 			       dc->bdev,
1027 			       BACKING_DEV_OFFLINE_TIMEOUT);
1028 			pr_err("%s: disable I/O request due to backing device offline\n",
1029 			       dc->disk.name);
1030 			dc->io_disable = true;
1031 			/* let others know earlier that io_disable is true */
1032 			smp_mb();
1033 			bcache_device_stop(&dc->disk);
1034 			break;
1035 		}
1036 		schedule_timeout_interruptible(HZ);
1037 	}
1038 
1039 	wait_for_kthread_stop();
1040 	return 0;
1041 }
1042 
1043 
bch_cached_dev_run(struct cached_dev * dc)1044 int bch_cached_dev_run(struct cached_dev *dc)
1045 {
1046 	int ret = 0;
1047 	struct bcache_device *d = &dc->disk;
1048 	char *buf = kmemdup_nul(dc->sb.label, SB_LABEL_SIZE, GFP_KERNEL);
1049 	char *env[] = {
1050 		"DRIVER=bcache",
1051 		kasprintf(GFP_KERNEL, "CACHED_UUID=%pU", dc->sb.uuid),
1052 		kasprintf(GFP_KERNEL, "CACHED_LABEL=%s", buf ? : ""),
1053 		NULL,
1054 	};
1055 
1056 	if (dc->io_disable) {
1057 		pr_err("I/O disabled on cached dev %pg\n", dc->bdev);
1058 		ret = -EIO;
1059 		goto out;
1060 	}
1061 
1062 	if (atomic_xchg(&dc->running, 1)) {
1063 		pr_info("cached dev %pg is running already\n", dc->bdev);
1064 		ret = -EBUSY;
1065 		goto out;
1066 	}
1067 
1068 	if (!d->c &&
1069 	    BDEV_STATE(&dc->sb) != BDEV_STATE_NONE) {
1070 		struct closure cl;
1071 
1072 		closure_init_stack(&cl);
1073 
1074 		SET_BDEV_STATE(&dc->sb, BDEV_STATE_STALE);
1075 		bch_write_bdev_super(dc, &cl);
1076 		closure_sync(&cl);
1077 	}
1078 
1079 	ret = add_disk(d->disk);
1080 	if (ret)
1081 		goto out;
1082 	bd_link_disk_holder(dc->bdev, dc->disk.disk);
1083 	/*
1084 	 * won't show up in the uevent file, use udevadm monitor -e instead
1085 	 * only class / kset properties are persistent
1086 	 */
1087 	kobject_uevent_env(&disk_to_dev(d->disk)->kobj, KOBJ_CHANGE, env);
1088 
1089 	if (sysfs_create_link(&d->kobj, &disk_to_dev(d->disk)->kobj, "dev") ||
1090 	    sysfs_create_link(&disk_to_dev(d->disk)->kobj,
1091 			      &d->kobj, "bcache")) {
1092 		pr_err("Couldn't create bcache dev <-> disk sysfs symlinks\n");
1093 		ret = -ENOMEM;
1094 		goto out;
1095 	}
1096 
1097 	dc->status_update_thread = kthread_run(cached_dev_status_update,
1098 					       dc, "bcache_status_update");
1099 	if (IS_ERR(dc->status_update_thread)) {
1100 		pr_warn("failed to create bcache_status_update kthread, continue to run without monitoring backing device status\n");
1101 	}
1102 
1103 out:
1104 	kfree(env[1]);
1105 	kfree(env[2]);
1106 	kfree(buf);
1107 	return ret;
1108 }
1109 
1110 /*
1111  * If BCACHE_DEV_RATE_DW_RUNNING is set, it means routine of the delayed
1112  * work dc->writeback_rate_update is running. Wait until the routine
1113  * quits (BCACHE_DEV_RATE_DW_RUNNING is clear), then continue to
1114  * cancel it. If BCACHE_DEV_RATE_DW_RUNNING is not clear after time_out
1115  * seconds, give up waiting here and continue to cancel it too.
1116  */
cancel_writeback_rate_update_dwork(struct cached_dev * dc)1117 static void cancel_writeback_rate_update_dwork(struct cached_dev *dc)
1118 {
1119 	int time_out = WRITEBACK_RATE_UPDATE_SECS_MAX * HZ;
1120 
1121 	do {
1122 		if (!test_bit(BCACHE_DEV_RATE_DW_RUNNING,
1123 			      &dc->disk.flags))
1124 			break;
1125 		time_out--;
1126 		schedule_timeout_interruptible(1);
1127 	} while (time_out > 0);
1128 
1129 	if (time_out == 0)
1130 		pr_warn("give up waiting for dc->writeback_write_update to quit\n");
1131 
1132 	cancel_delayed_work_sync(&dc->writeback_rate_update);
1133 }
1134 
cached_dev_detach_finish(struct work_struct * w)1135 static void cached_dev_detach_finish(struct work_struct *w)
1136 {
1137 	struct cached_dev *dc = container_of(w, struct cached_dev, detach);
1138 	struct cache_set *c = dc->disk.c;
1139 
1140 	BUG_ON(!test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags));
1141 	BUG_ON(refcount_read(&dc->count));
1142 
1143 
1144 	if (test_and_clear_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags))
1145 		cancel_writeback_rate_update_dwork(dc);
1146 
1147 	if (!IS_ERR_OR_NULL(dc->writeback_thread)) {
1148 		kthread_stop(dc->writeback_thread);
1149 		dc->writeback_thread = NULL;
1150 	}
1151 
1152 	mutex_lock(&bch_register_lock);
1153 
1154 	bcache_device_detach(&dc->disk);
1155 	list_move(&dc->list, &uncached_devices);
1156 	calc_cached_dev_sectors(c);
1157 
1158 	clear_bit(BCACHE_DEV_DETACHING, &dc->disk.flags);
1159 	clear_bit(BCACHE_DEV_UNLINK_DONE, &dc->disk.flags);
1160 
1161 	mutex_unlock(&bch_register_lock);
1162 
1163 	pr_info("Caching disabled for %pg\n", dc->bdev);
1164 
1165 	/* Drop ref we took in cached_dev_detach() */
1166 	closure_put(&dc->disk.cl);
1167 }
1168 
bch_cached_dev_detach(struct cached_dev * dc)1169 void bch_cached_dev_detach(struct cached_dev *dc)
1170 {
1171 	lockdep_assert_held(&bch_register_lock);
1172 
1173 	if (test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags))
1174 		return;
1175 
1176 	if (test_and_set_bit(BCACHE_DEV_DETACHING, &dc->disk.flags))
1177 		return;
1178 
1179 	/*
1180 	 * Block the device from being closed and freed until we're finished
1181 	 * detaching
1182 	 */
1183 	closure_get(&dc->disk.cl);
1184 
1185 	bch_writeback_queue(dc);
1186 
1187 	cached_dev_put(dc);
1188 }
1189 
bch_cached_dev_attach(struct cached_dev * dc,struct cache_set * c,uint8_t * set_uuid)1190 int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c,
1191 			  uint8_t *set_uuid)
1192 {
1193 	uint32_t rtime = cpu_to_le32((u32)ktime_get_real_seconds());
1194 	struct uuid_entry *u;
1195 	struct cached_dev *exist_dc, *t;
1196 	int ret = 0;
1197 
1198 	if ((set_uuid && memcmp(set_uuid, c->set_uuid, 16)) ||
1199 	    (!set_uuid && memcmp(dc->sb.set_uuid, c->set_uuid, 16)))
1200 		return -ENOENT;
1201 
1202 	if (dc->disk.c) {
1203 		pr_err("Can't attach %pg: already attached\n", dc->bdev);
1204 		return -EINVAL;
1205 	}
1206 
1207 	if (test_bit(CACHE_SET_STOPPING, &c->flags)) {
1208 		pr_err("Can't attach %pg: shutting down\n", dc->bdev);
1209 		return -EINVAL;
1210 	}
1211 
1212 	if (dc->sb.block_size < c->cache->sb.block_size) {
1213 		/* Will die */
1214 		pr_err("Couldn't attach %pg: block size less than set's block size\n",
1215 		       dc->bdev);
1216 		return -EINVAL;
1217 	}
1218 
1219 	/* Check whether already attached */
1220 	list_for_each_entry_safe(exist_dc, t, &c->cached_devs, list) {
1221 		if (!memcmp(dc->sb.uuid, exist_dc->sb.uuid, 16)) {
1222 			pr_err("Tried to attach %pg but duplicate UUID already attached\n",
1223 				dc->bdev);
1224 
1225 			return -EINVAL;
1226 		}
1227 	}
1228 
1229 	u = uuid_find(c, dc->sb.uuid);
1230 
1231 	if (u &&
1232 	    (BDEV_STATE(&dc->sb) == BDEV_STATE_STALE ||
1233 	     BDEV_STATE(&dc->sb) == BDEV_STATE_NONE)) {
1234 		memcpy(u->uuid, invalid_uuid, 16);
1235 		u->invalidated = cpu_to_le32((u32)ktime_get_real_seconds());
1236 		u = NULL;
1237 	}
1238 
1239 	if (!u) {
1240 		if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
1241 			pr_err("Couldn't find uuid for %pg in set\n", dc->bdev);
1242 			return -ENOENT;
1243 		}
1244 
1245 		u = uuid_find_empty(c);
1246 		if (!u) {
1247 			pr_err("Not caching %pg, no room for UUID\n", dc->bdev);
1248 			return -EINVAL;
1249 		}
1250 	}
1251 
1252 	/*
1253 	 * Deadlocks since we're called via sysfs...
1254 	 * sysfs_remove_file(&dc->kobj, &sysfs_attach);
1255 	 */
1256 
1257 	if (bch_is_zero(u->uuid, 16)) {
1258 		struct closure cl;
1259 
1260 		closure_init_stack(&cl);
1261 
1262 		memcpy(u->uuid, dc->sb.uuid, 16);
1263 		memcpy(u->label, dc->sb.label, SB_LABEL_SIZE);
1264 		u->first_reg = u->last_reg = rtime;
1265 		bch_uuid_write(c);
1266 
1267 		memcpy(dc->sb.set_uuid, c->set_uuid, 16);
1268 		SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN);
1269 
1270 		bch_write_bdev_super(dc, &cl);
1271 		closure_sync(&cl);
1272 	} else {
1273 		u->last_reg = rtime;
1274 		bch_uuid_write(c);
1275 	}
1276 
1277 	bcache_device_attach(&dc->disk, c, u - c->uuids);
1278 	list_move(&dc->list, &c->cached_devs);
1279 	calc_cached_dev_sectors(c);
1280 
1281 	/*
1282 	 * dc->c must be set before dc->count != 0 - paired with the mb in
1283 	 * cached_dev_get()
1284 	 */
1285 	smp_wmb();
1286 	refcount_set(&dc->count, 1);
1287 
1288 	/* Block writeback thread, but spawn it */
1289 	down_write(&dc->writeback_lock);
1290 	if (bch_cached_dev_writeback_start(dc)) {
1291 		up_write(&dc->writeback_lock);
1292 		pr_err("Couldn't start writeback facilities for %s\n",
1293 		       dc->disk.disk->disk_name);
1294 		return -ENOMEM;
1295 	}
1296 
1297 	if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
1298 		atomic_set(&dc->has_dirty, 1);
1299 		bch_writeback_queue(dc);
1300 	}
1301 
1302 	bch_sectors_dirty_init(&dc->disk);
1303 
1304 	ret = bch_cached_dev_run(dc);
1305 	if (ret && (ret != -EBUSY)) {
1306 		up_write(&dc->writeback_lock);
1307 		/*
1308 		 * bch_register_lock is held, bcache_device_stop() is not
1309 		 * able to be directly called. The kthread and kworker
1310 		 * created previously in bch_cached_dev_writeback_start()
1311 		 * have to be stopped manually here.
1312 		 */
1313 		kthread_stop(dc->writeback_thread);
1314 		cancel_writeback_rate_update_dwork(dc);
1315 		pr_err("Couldn't run cached device %pg\n", dc->bdev);
1316 		return ret;
1317 	}
1318 
1319 	bcache_device_link(&dc->disk, c, "bdev");
1320 	atomic_inc(&c->attached_dev_nr);
1321 
1322 	if (bch_has_feature_obso_large_bucket(&(c->cache->sb))) {
1323 		pr_err("The obsoleted large bucket layout is unsupported, set the bcache device into read-only\n");
1324 		pr_err("Please update to the latest bcache-tools to create the cache device\n");
1325 		set_disk_ro(dc->disk.disk, 1);
1326 	}
1327 
1328 	/* Allow the writeback thread to proceed */
1329 	up_write(&dc->writeback_lock);
1330 
1331 	pr_info("Caching %pg as %s on set %pU\n",
1332 		dc->bdev,
1333 		dc->disk.disk->disk_name,
1334 		dc->disk.c->set_uuid);
1335 	return 0;
1336 }
1337 
1338 /* when dc->disk.kobj released */
bch_cached_dev_release(struct kobject * kobj)1339 void bch_cached_dev_release(struct kobject *kobj)
1340 {
1341 	struct cached_dev *dc = container_of(kobj, struct cached_dev,
1342 					     disk.kobj);
1343 	kfree(dc);
1344 	module_put(THIS_MODULE);
1345 }
1346 
cached_dev_free(struct closure * cl)1347 static void cached_dev_free(struct closure *cl)
1348 {
1349 	struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1350 
1351 	if (test_and_clear_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags))
1352 		cancel_writeback_rate_update_dwork(dc);
1353 
1354 	if (!IS_ERR_OR_NULL(dc->writeback_thread))
1355 		kthread_stop(dc->writeback_thread);
1356 	if (!IS_ERR_OR_NULL(dc->status_update_thread))
1357 		kthread_stop(dc->status_update_thread);
1358 
1359 	mutex_lock(&bch_register_lock);
1360 
1361 	if (atomic_read(&dc->running)) {
1362 		bd_unlink_disk_holder(dc->bdev, dc->disk.disk);
1363 		del_gendisk(dc->disk.disk);
1364 	}
1365 	bcache_device_free(&dc->disk);
1366 	list_del(&dc->list);
1367 
1368 	mutex_unlock(&bch_register_lock);
1369 
1370 	if (dc->sb_disk)
1371 		put_page(virt_to_page(dc->sb_disk));
1372 
1373 	if (!IS_ERR_OR_NULL(dc->bdev))
1374 		blkdev_put(dc->bdev, dc);
1375 
1376 	wake_up(&unregister_wait);
1377 
1378 	kobject_put(&dc->disk.kobj);
1379 }
1380 
cached_dev_flush(struct closure * cl)1381 static void cached_dev_flush(struct closure *cl)
1382 {
1383 	struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1384 	struct bcache_device *d = &dc->disk;
1385 
1386 	mutex_lock(&bch_register_lock);
1387 	bcache_device_unlink(d);
1388 	mutex_unlock(&bch_register_lock);
1389 
1390 	bch_cache_accounting_destroy(&dc->accounting);
1391 	kobject_del(&d->kobj);
1392 
1393 	continue_at(cl, cached_dev_free, system_wq);
1394 }
1395 
cached_dev_init(struct cached_dev * dc,unsigned int block_size)1396 static int cached_dev_init(struct cached_dev *dc, unsigned int block_size)
1397 {
1398 	int ret;
1399 	struct io *io;
1400 	struct request_queue *q = bdev_get_queue(dc->bdev);
1401 
1402 	__module_get(THIS_MODULE);
1403 	INIT_LIST_HEAD(&dc->list);
1404 	closure_init(&dc->disk.cl, NULL);
1405 	set_closure_fn(&dc->disk.cl, cached_dev_flush, system_wq);
1406 	kobject_init(&dc->disk.kobj, &bch_cached_dev_ktype);
1407 	INIT_WORK(&dc->detach, cached_dev_detach_finish);
1408 	sema_init(&dc->sb_write_mutex, 1);
1409 	INIT_LIST_HEAD(&dc->io_lru);
1410 	spin_lock_init(&dc->io_lock);
1411 	bch_cache_accounting_init(&dc->accounting, &dc->disk.cl);
1412 
1413 	dc->sequential_cutoff		= 4 << 20;
1414 
1415 	for (io = dc->io; io < dc->io + RECENT_IO; io++) {
1416 		list_add(&io->lru, &dc->io_lru);
1417 		hlist_add_head(&io->hash, dc->io_hash + RECENT_IO);
1418 	}
1419 
1420 	dc->disk.stripe_size = q->limits.io_opt >> 9;
1421 
1422 	if (dc->disk.stripe_size)
1423 		dc->partial_stripes_expensive =
1424 			q->limits.raid_partial_stripes_expensive;
1425 
1426 	ret = bcache_device_init(&dc->disk, block_size,
1427 			 bdev_nr_sectors(dc->bdev) - dc->sb.data_offset,
1428 			 dc->bdev, &bcache_cached_ops);
1429 	if (ret)
1430 		return ret;
1431 
1432 	blk_queue_io_opt(dc->disk.disk->queue,
1433 		max(queue_io_opt(dc->disk.disk->queue), queue_io_opt(q)));
1434 
1435 	atomic_set(&dc->io_errors, 0);
1436 	dc->io_disable = false;
1437 	dc->error_limit = DEFAULT_CACHED_DEV_ERROR_LIMIT;
1438 	/* default to auto */
1439 	dc->stop_when_cache_set_failed = BCH_CACHED_DEV_STOP_AUTO;
1440 
1441 	bch_cached_dev_request_init(dc);
1442 	bch_cached_dev_writeback_init(dc);
1443 	return 0;
1444 }
1445 
1446 /* Cached device - bcache superblock */
1447 
register_bdev(struct cache_sb * sb,struct cache_sb_disk * sb_disk,struct block_device * bdev,struct cached_dev * dc)1448 static int register_bdev(struct cache_sb *sb, struct cache_sb_disk *sb_disk,
1449 				 struct block_device *bdev,
1450 				 struct cached_dev *dc)
1451 {
1452 	const char *err = "cannot allocate memory";
1453 	struct cache_set *c;
1454 	int ret = -ENOMEM;
1455 
1456 	memcpy(&dc->sb, sb, sizeof(struct cache_sb));
1457 	dc->bdev = bdev;
1458 	dc->sb_disk = sb_disk;
1459 
1460 	if (cached_dev_init(dc, sb->block_size << 9))
1461 		goto err;
1462 
1463 	err = "error creating kobject";
1464 	if (kobject_add(&dc->disk.kobj, bdev_kobj(bdev), "bcache"))
1465 		goto err;
1466 	if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj))
1467 		goto err;
1468 
1469 	pr_info("registered backing device %pg\n", dc->bdev);
1470 
1471 	list_add(&dc->list, &uncached_devices);
1472 	/* attach to a matched cache set if it exists */
1473 	list_for_each_entry(c, &bch_cache_sets, list)
1474 		bch_cached_dev_attach(dc, c, NULL);
1475 
1476 	if (BDEV_STATE(&dc->sb) == BDEV_STATE_NONE ||
1477 	    BDEV_STATE(&dc->sb) == BDEV_STATE_STALE) {
1478 		err = "failed to run cached device";
1479 		ret = bch_cached_dev_run(dc);
1480 		if (ret)
1481 			goto err;
1482 	}
1483 
1484 	return 0;
1485 err:
1486 	pr_notice("error %pg: %s\n", dc->bdev, err);
1487 	bcache_device_stop(&dc->disk);
1488 	return ret;
1489 }
1490 
1491 /* Flash only volumes */
1492 
1493 /* When d->kobj released */
bch_flash_dev_release(struct kobject * kobj)1494 void bch_flash_dev_release(struct kobject *kobj)
1495 {
1496 	struct bcache_device *d = container_of(kobj, struct bcache_device,
1497 					       kobj);
1498 	kfree(d);
1499 }
1500 
flash_dev_free(struct closure * cl)1501 static void flash_dev_free(struct closure *cl)
1502 {
1503 	struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1504 
1505 	mutex_lock(&bch_register_lock);
1506 	atomic_long_sub(bcache_dev_sectors_dirty(d),
1507 			&d->c->flash_dev_dirty_sectors);
1508 	del_gendisk(d->disk);
1509 	bcache_device_free(d);
1510 	mutex_unlock(&bch_register_lock);
1511 	kobject_put(&d->kobj);
1512 }
1513 
flash_dev_flush(struct closure * cl)1514 static void flash_dev_flush(struct closure *cl)
1515 {
1516 	struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1517 
1518 	mutex_lock(&bch_register_lock);
1519 	bcache_device_unlink(d);
1520 	mutex_unlock(&bch_register_lock);
1521 	kobject_del(&d->kobj);
1522 	continue_at(cl, flash_dev_free, system_wq);
1523 }
1524 
flash_dev_run(struct cache_set * c,struct uuid_entry * u)1525 static int flash_dev_run(struct cache_set *c, struct uuid_entry *u)
1526 {
1527 	int err = -ENOMEM;
1528 	struct bcache_device *d = kzalloc(sizeof(struct bcache_device),
1529 					  GFP_KERNEL);
1530 	if (!d)
1531 		goto err_ret;
1532 
1533 	closure_init(&d->cl, NULL);
1534 	set_closure_fn(&d->cl, flash_dev_flush, system_wq);
1535 
1536 	kobject_init(&d->kobj, &bch_flash_dev_ktype);
1537 
1538 	if (bcache_device_init(d, block_bytes(c->cache), u->sectors,
1539 			NULL, &bcache_flash_ops))
1540 		goto err;
1541 
1542 	bcache_device_attach(d, c, u - c->uuids);
1543 	bch_sectors_dirty_init(d);
1544 	bch_flash_dev_request_init(d);
1545 	err = add_disk(d->disk);
1546 	if (err)
1547 		goto err;
1548 
1549 	err = kobject_add(&d->kobj, &disk_to_dev(d->disk)->kobj, "bcache");
1550 	if (err)
1551 		goto err;
1552 
1553 	bcache_device_link(d, c, "volume");
1554 
1555 	if (bch_has_feature_obso_large_bucket(&c->cache->sb)) {
1556 		pr_err("The obsoleted large bucket layout is unsupported, set the bcache device into read-only\n");
1557 		pr_err("Please update to the latest bcache-tools to create the cache device\n");
1558 		set_disk_ro(d->disk, 1);
1559 	}
1560 
1561 	return 0;
1562 err:
1563 	kobject_put(&d->kobj);
1564 err_ret:
1565 	return err;
1566 }
1567 
flash_devs_run(struct cache_set * c)1568 static int flash_devs_run(struct cache_set *c)
1569 {
1570 	int ret = 0;
1571 	struct uuid_entry *u;
1572 
1573 	for (u = c->uuids;
1574 	     u < c->uuids + c->nr_uuids && !ret;
1575 	     u++)
1576 		if (UUID_FLASH_ONLY(u))
1577 			ret = flash_dev_run(c, u);
1578 
1579 	return ret;
1580 }
1581 
bch_flash_dev_create(struct cache_set * c,uint64_t size)1582 int bch_flash_dev_create(struct cache_set *c, uint64_t size)
1583 {
1584 	struct uuid_entry *u;
1585 
1586 	if (test_bit(CACHE_SET_STOPPING, &c->flags))
1587 		return -EINTR;
1588 
1589 	if (!test_bit(CACHE_SET_RUNNING, &c->flags))
1590 		return -EPERM;
1591 
1592 	u = uuid_find_empty(c);
1593 	if (!u) {
1594 		pr_err("Can't create volume, no room for UUID\n");
1595 		return -EINVAL;
1596 	}
1597 
1598 	get_random_bytes(u->uuid, 16);
1599 	memset(u->label, 0, 32);
1600 	u->first_reg = u->last_reg = cpu_to_le32((u32)ktime_get_real_seconds());
1601 
1602 	SET_UUID_FLASH_ONLY(u, 1);
1603 	u->sectors = size >> 9;
1604 
1605 	bch_uuid_write(c);
1606 
1607 	return flash_dev_run(c, u);
1608 }
1609 
bch_cached_dev_error(struct cached_dev * dc)1610 bool bch_cached_dev_error(struct cached_dev *dc)
1611 {
1612 	if (!dc || test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags))
1613 		return false;
1614 
1615 	dc->io_disable = true;
1616 	/* make others know io_disable is true earlier */
1617 	smp_mb();
1618 
1619 	pr_err("stop %s: too many IO errors on backing device %pg\n",
1620 	       dc->disk.disk->disk_name, dc->bdev);
1621 
1622 	bcache_device_stop(&dc->disk);
1623 	return true;
1624 }
1625 
1626 /* Cache set */
1627 
1628 __printf(2, 3)
bch_cache_set_error(struct cache_set * c,const char * fmt,...)1629 bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...)
1630 {
1631 	struct va_format vaf;
1632 	va_list args;
1633 
1634 	if (c->on_error != ON_ERROR_PANIC &&
1635 	    test_bit(CACHE_SET_STOPPING, &c->flags))
1636 		return false;
1637 
1638 	if (test_and_set_bit(CACHE_SET_IO_DISABLE, &c->flags))
1639 		pr_info("CACHE_SET_IO_DISABLE already set\n");
1640 
1641 	/*
1642 	 * XXX: we can be called from atomic context
1643 	 * acquire_console_sem();
1644 	 */
1645 
1646 	va_start(args, fmt);
1647 
1648 	vaf.fmt = fmt;
1649 	vaf.va = &args;
1650 
1651 	pr_err("error on %pU: %pV, disabling caching\n",
1652 	       c->set_uuid, &vaf);
1653 
1654 	va_end(args);
1655 
1656 	if (c->on_error == ON_ERROR_PANIC)
1657 		panic("panic forced after error\n");
1658 
1659 	bch_cache_set_unregister(c);
1660 	return true;
1661 }
1662 
1663 /* When c->kobj released */
bch_cache_set_release(struct kobject * kobj)1664 void bch_cache_set_release(struct kobject *kobj)
1665 {
1666 	struct cache_set *c = container_of(kobj, struct cache_set, kobj);
1667 
1668 	kfree(c);
1669 	module_put(THIS_MODULE);
1670 }
1671 
cache_set_free(struct closure * cl)1672 static void cache_set_free(struct closure *cl)
1673 {
1674 	struct cache_set *c = container_of(cl, struct cache_set, cl);
1675 	struct cache *ca;
1676 
1677 	debugfs_remove(c->debug);
1678 
1679 	bch_open_buckets_free(c);
1680 	bch_btree_cache_free(c);
1681 	bch_journal_free(c);
1682 
1683 	mutex_lock(&bch_register_lock);
1684 	bch_bset_sort_state_free(&c->sort);
1685 	free_pages((unsigned long) c->uuids, ilog2(meta_bucket_pages(&c->cache->sb)));
1686 
1687 	ca = c->cache;
1688 	if (ca) {
1689 		ca->set = NULL;
1690 		c->cache = NULL;
1691 		kobject_put(&ca->kobj);
1692 	}
1693 
1694 
1695 	if (c->moving_gc_wq)
1696 		destroy_workqueue(c->moving_gc_wq);
1697 	bioset_exit(&c->bio_split);
1698 	mempool_exit(&c->fill_iter);
1699 	mempool_exit(&c->bio_meta);
1700 	mempool_exit(&c->search);
1701 	kfree(c->devices);
1702 
1703 	list_del(&c->list);
1704 	mutex_unlock(&bch_register_lock);
1705 
1706 	pr_info("Cache set %pU unregistered\n", c->set_uuid);
1707 	wake_up(&unregister_wait);
1708 
1709 	closure_debug_destroy(&c->cl);
1710 	kobject_put(&c->kobj);
1711 }
1712 
cache_set_flush(struct closure * cl)1713 static void cache_set_flush(struct closure *cl)
1714 {
1715 	struct cache_set *c = container_of(cl, struct cache_set, caching);
1716 	struct cache *ca = c->cache;
1717 	struct btree *b;
1718 
1719 	bch_cache_accounting_destroy(&c->accounting);
1720 
1721 	kobject_put(&c->internal);
1722 	kobject_del(&c->kobj);
1723 
1724 	if (!IS_ERR_OR_NULL(c->gc_thread))
1725 		kthread_stop(c->gc_thread);
1726 
1727 	if (!IS_ERR(c->root))
1728 		list_add(&c->root->list, &c->btree_cache);
1729 
1730 	/*
1731 	 * Avoid flushing cached nodes if cache set is retiring
1732 	 * due to too many I/O errors detected.
1733 	 */
1734 	if (!test_bit(CACHE_SET_IO_DISABLE, &c->flags))
1735 		list_for_each_entry(b, &c->btree_cache, list) {
1736 			mutex_lock(&b->write_lock);
1737 			if (btree_node_dirty(b))
1738 				__bch_btree_node_write(b, NULL);
1739 			mutex_unlock(&b->write_lock);
1740 		}
1741 
1742 	if (ca->alloc_thread)
1743 		kthread_stop(ca->alloc_thread);
1744 
1745 	if (c->journal.cur) {
1746 		cancel_delayed_work_sync(&c->journal.work);
1747 		/* flush last journal entry if needed */
1748 		c->journal.work.work.func(&c->journal.work.work);
1749 	}
1750 
1751 	closure_return(cl);
1752 }
1753 
1754 /*
1755  * This function is only called when CACHE_SET_IO_DISABLE is set, which means
1756  * cache set is unregistering due to too many I/O errors. In this condition,
1757  * the bcache device might be stopped, it depends on stop_when_cache_set_failed
1758  * value and whether the broken cache has dirty data:
1759  *
1760  * dc->stop_when_cache_set_failed    dc->has_dirty   stop bcache device
1761  *  BCH_CACHED_STOP_AUTO               0               NO
1762  *  BCH_CACHED_STOP_AUTO               1               YES
1763  *  BCH_CACHED_DEV_STOP_ALWAYS         0               YES
1764  *  BCH_CACHED_DEV_STOP_ALWAYS         1               YES
1765  *
1766  * The expected behavior is, if stop_when_cache_set_failed is configured to
1767  * "auto" via sysfs interface, the bcache device will not be stopped if the
1768  * backing device is clean on the broken cache device.
1769  */
conditional_stop_bcache_device(struct cache_set * c,struct bcache_device * d,struct cached_dev * dc)1770 static void conditional_stop_bcache_device(struct cache_set *c,
1771 					   struct bcache_device *d,
1772 					   struct cached_dev *dc)
1773 {
1774 	if (dc->stop_when_cache_set_failed == BCH_CACHED_DEV_STOP_ALWAYS) {
1775 		pr_warn("stop_when_cache_set_failed of %s is \"always\", stop it for failed cache set %pU.\n",
1776 			d->disk->disk_name, c->set_uuid);
1777 		bcache_device_stop(d);
1778 	} else if (atomic_read(&dc->has_dirty)) {
1779 		/*
1780 		 * dc->stop_when_cache_set_failed == BCH_CACHED_STOP_AUTO
1781 		 * and dc->has_dirty == 1
1782 		 */
1783 		pr_warn("stop_when_cache_set_failed of %s is \"auto\" and cache is dirty, stop it to avoid potential data corruption.\n",
1784 			d->disk->disk_name);
1785 		/*
1786 		 * There might be a small time gap that cache set is
1787 		 * released but bcache device is not. Inside this time
1788 		 * gap, regular I/O requests will directly go into
1789 		 * backing device as no cache set attached to. This
1790 		 * behavior may also introduce potential inconsistence
1791 		 * data in writeback mode while cache is dirty.
1792 		 * Therefore before calling bcache_device_stop() due
1793 		 * to a broken cache device, dc->io_disable should be
1794 		 * explicitly set to true.
1795 		 */
1796 		dc->io_disable = true;
1797 		/* make others know io_disable is true earlier */
1798 		smp_mb();
1799 		bcache_device_stop(d);
1800 	} else {
1801 		/*
1802 		 * dc->stop_when_cache_set_failed == BCH_CACHED_STOP_AUTO
1803 		 * and dc->has_dirty == 0
1804 		 */
1805 		pr_warn("stop_when_cache_set_failed of %s is \"auto\" and cache is clean, keep it alive.\n",
1806 			d->disk->disk_name);
1807 	}
1808 }
1809 
__cache_set_unregister(struct closure * cl)1810 static void __cache_set_unregister(struct closure *cl)
1811 {
1812 	struct cache_set *c = container_of(cl, struct cache_set, caching);
1813 	struct cached_dev *dc;
1814 	struct bcache_device *d;
1815 	size_t i;
1816 
1817 	mutex_lock(&bch_register_lock);
1818 
1819 	for (i = 0; i < c->devices_max_used; i++) {
1820 		d = c->devices[i];
1821 		if (!d)
1822 			continue;
1823 
1824 		if (!UUID_FLASH_ONLY(&c->uuids[i]) &&
1825 		    test_bit(CACHE_SET_UNREGISTERING, &c->flags)) {
1826 			dc = container_of(d, struct cached_dev, disk);
1827 			bch_cached_dev_detach(dc);
1828 			if (test_bit(CACHE_SET_IO_DISABLE, &c->flags))
1829 				conditional_stop_bcache_device(c, d, dc);
1830 		} else {
1831 			bcache_device_stop(d);
1832 		}
1833 	}
1834 
1835 	mutex_unlock(&bch_register_lock);
1836 
1837 	continue_at(cl, cache_set_flush, system_wq);
1838 }
1839 
bch_cache_set_stop(struct cache_set * c)1840 void bch_cache_set_stop(struct cache_set *c)
1841 {
1842 	if (!test_and_set_bit(CACHE_SET_STOPPING, &c->flags))
1843 		/* closure_fn set to __cache_set_unregister() */
1844 		closure_queue(&c->caching);
1845 }
1846 
bch_cache_set_unregister(struct cache_set * c)1847 void bch_cache_set_unregister(struct cache_set *c)
1848 {
1849 	set_bit(CACHE_SET_UNREGISTERING, &c->flags);
1850 	bch_cache_set_stop(c);
1851 }
1852 
1853 #define alloc_meta_bucket_pages(gfp, sb)		\
1854 	((void *) __get_free_pages(__GFP_ZERO|__GFP_COMP|gfp, ilog2(meta_bucket_pages(sb))))
1855 
bch_cache_set_alloc(struct cache_sb * sb)1856 struct cache_set *bch_cache_set_alloc(struct cache_sb *sb)
1857 {
1858 	int iter_size;
1859 	struct cache *ca = container_of(sb, struct cache, sb);
1860 	struct cache_set *c = kzalloc(sizeof(struct cache_set), GFP_KERNEL);
1861 
1862 	if (!c)
1863 		return NULL;
1864 
1865 	__module_get(THIS_MODULE);
1866 	closure_init(&c->cl, NULL);
1867 	set_closure_fn(&c->cl, cache_set_free, system_wq);
1868 
1869 	closure_init(&c->caching, &c->cl);
1870 	set_closure_fn(&c->caching, __cache_set_unregister, system_wq);
1871 
1872 	/* Maybe create continue_at_noreturn() and use it here? */
1873 	closure_set_stopped(&c->cl);
1874 	closure_put(&c->cl);
1875 
1876 	kobject_init(&c->kobj, &bch_cache_set_ktype);
1877 	kobject_init(&c->internal, &bch_cache_set_internal_ktype);
1878 
1879 	bch_cache_accounting_init(&c->accounting, &c->cl);
1880 
1881 	memcpy(c->set_uuid, sb->set_uuid, 16);
1882 
1883 	c->cache		= ca;
1884 	c->cache->set		= c;
1885 	c->bucket_bits		= ilog2(sb->bucket_size);
1886 	c->block_bits		= ilog2(sb->block_size);
1887 	c->nr_uuids		= meta_bucket_bytes(sb) / sizeof(struct uuid_entry);
1888 	c->devices_max_used	= 0;
1889 	atomic_set(&c->attached_dev_nr, 0);
1890 	c->btree_pages		= meta_bucket_pages(sb);
1891 	if (c->btree_pages > BTREE_MAX_PAGES)
1892 		c->btree_pages = max_t(int, c->btree_pages / 4,
1893 				       BTREE_MAX_PAGES);
1894 
1895 	sema_init(&c->sb_write_mutex, 1);
1896 	mutex_init(&c->bucket_lock);
1897 	init_waitqueue_head(&c->btree_cache_wait);
1898 	spin_lock_init(&c->btree_cannibalize_lock);
1899 	init_waitqueue_head(&c->bucket_wait);
1900 	init_waitqueue_head(&c->gc_wait);
1901 	sema_init(&c->uuid_write_mutex, 1);
1902 
1903 	spin_lock_init(&c->btree_gc_time.lock);
1904 	spin_lock_init(&c->btree_split_time.lock);
1905 	spin_lock_init(&c->btree_read_time.lock);
1906 
1907 	bch_moving_init_cache_set(c);
1908 
1909 	INIT_LIST_HEAD(&c->list);
1910 	INIT_LIST_HEAD(&c->cached_devs);
1911 	INIT_LIST_HEAD(&c->btree_cache);
1912 	INIT_LIST_HEAD(&c->btree_cache_freeable);
1913 	INIT_LIST_HEAD(&c->btree_cache_freed);
1914 	INIT_LIST_HEAD(&c->data_buckets);
1915 
1916 	iter_size = sizeof(struct btree_iter) +
1917 		    ((meta_bucket_pages(sb) * PAGE_SECTORS) / sb->block_size) *
1918 			    sizeof(struct btree_iter_set);
1919 
1920 	c->devices = kcalloc(c->nr_uuids, sizeof(void *), GFP_KERNEL);
1921 	if (!c->devices)
1922 		goto err;
1923 
1924 	if (mempool_init_slab_pool(&c->search, 32, bch_search_cache))
1925 		goto err;
1926 
1927 	if (mempool_init_kmalloc_pool(&c->bio_meta, 2,
1928 			sizeof(struct bbio) +
1929 			sizeof(struct bio_vec) * meta_bucket_pages(sb)))
1930 		goto err;
1931 
1932 	if (mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size))
1933 		goto err;
1934 
1935 	if (bioset_init(&c->bio_split, 4, offsetof(struct bbio, bio),
1936 			BIOSET_NEED_RESCUER))
1937 		goto err;
1938 
1939 	c->uuids = alloc_meta_bucket_pages(GFP_KERNEL, sb);
1940 	if (!c->uuids)
1941 		goto err;
1942 
1943 	c->moving_gc_wq = alloc_workqueue("bcache_gc", WQ_MEM_RECLAIM, 0);
1944 	if (!c->moving_gc_wq)
1945 		goto err;
1946 
1947 	if (bch_journal_alloc(c))
1948 		goto err;
1949 
1950 	if (bch_btree_cache_alloc(c))
1951 		goto err;
1952 
1953 	if (bch_open_buckets_alloc(c))
1954 		goto err;
1955 
1956 	if (bch_bset_sort_state_init(&c->sort, ilog2(c->btree_pages)))
1957 		goto err;
1958 
1959 	c->congested_read_threshold_us	= 2000;
1960 	c->congested_write_threshold_us	= 20000;
1961 	c->error_limit	= DEFAULT_IO_ERROR_LIMIT;
1962 	c->idle_max_writeback_rate_enabled = 1;
1963 	WARN_ON(test_and_clear_bit(CACHE_SET_IO_DISABLE, &c->flags));
1964 
1965 	return c;
1966 err:
1967 	bch_cache_set_unregister(c);
1968 	return NULL;
1969 }
1970 
run_cache_set(struct cache_set * c)1971 static int run_cache_set(struct cache_set *c)
1972 {
1973 	const char *err = "cannot allocate memory";
1974 	struct cached_dev *dc, *t;
1975 	struct cache *ca = c->cache;
1976 	struct closure cl;
1977 	LIST_HEAD(journal);
1978 	struct journal_replay *l;
1979 
1980 	closure_init_stack(&cl);
1981 
1982 	c->nbuckets = ca->sb.nbuckets;
1983 	set_gc_sectors(c);
1984 
1985 	if (CACHE_SYNC(&c->cache->sb)) {
1986 		struct bkey *k;
1987 		struct jset *j;
1988 
1989 		err = "cannot allocate memory for journal";
1990 		if (bch_journal_read(c, &journal))
1991 			goto err;
1992 
1993 		pr_debug("btree_journal_read() done\n");
1994 
1995 		err = "no journal entries found";
1996 		if (list_empty(&journal))
1997 			goto err;
1998 
1999 		j = &list_entry(journal.prev, struct journal_replay, list)->j;
2000 
2001 		err = "IO error reading priorities";
2002 		if (prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev]))
2003 			goto err;
2004 
2005 		/*
2006 		 * If prio_read() fails it'll call cache_set_error and we'll
2007 		 * tear everything down right away, but if we perhaps checked
2008 		 * sooner we could avoid journal replay.
2009 		 */
2010 
2011 		k = &j->btree_root;
2012 
2013 		err = "bad btree root";
2014 		if (__bch_btree_ptr_invalid(c, k))
2015 			goto err;
2016 
2017 		err = "error reading btree root";
2018 		c->root = bch_btree_node_get(c, NULL, k,
2019 					     j->btree_level,
2020 					     true, NULL);
2021 		if (IS_ERR(c->root))
2022 			goto err;
2023 
2024 		list_del_init(&c->root->list);
2025 		rw_unlock(true, c->root);
2026 
2027 		err = uuid_read(c, j, &cl);
2028 		if (err)
2029 			goto err;
2030 
2031 		err = "error in recovery";
2032 		if (bch_btree_check(c))
2033 			goto err;
2034 
2035 		bch_journal_mark(c, &journal);
2036 		bch_initial_gc_finish(c);
2037 		pr_debug("btree_check() done\n");
2038 
2039 		/*
2040 		 * bcache_journal_next() can't happen sooner, or
2041 		 * btree_gc_finish() will give spurious errors about last_gc >
2042 		 * gc_gen - this is a hack but oh well.
2043 		 */
2044 		bch_journal_next(&c->journal);
2045 
2046 		err = "error starting allocator thread";
2047 		if (bch_cache_allocator_start(ca))
2048 			goto err;
2049 
2050 		/*
2051 		 * First place it's safe to allocate: btree_check() and
2052 		 * btree_gc_finish() have to run before we have buckets to
2053 		 * allocate, and bch_bucket_alloc_set() might cause a journal
2054 		 * entry to be written so bcache_journal_next() has to be called
2055 		 * first.
2056 		 *
2057 		 * If the uuids were in the old format we have to rewrite them
2058 		 * before the next journal entry is written:
2059 		 */
2060 		if (j->version < BCACHE_JSET_VERSION_UUID)
2061 			__uuid_write(c);
2062 
2063 		err = "bcache: replay journal failed";
2064 		if (bch_journal_replay(c, &journal))
2065 			goto err;
2066 	} else {
2067 		unsigned int j;
2068 
2069 		pr_notice("invalidating existing data\n");
2070 		ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7,
2071 					2, SB_JOURNAL_BUCKETS);
2072 
2073 		for (j = 0; j < ca->sb.keys; j++)
2074 			ca->sb.d[j] = ca->sb.first_bucket + j;
2075 
2076 		bch_initial_gc_finish(c);
2077 
2078 		err = "error starting allocator thread";
2079 		if (bch_cache_allocator_start(ca))
2080 			goto err;
2081 
2082 		mutex_lock(&c->bucket_lock);
2083 		bch_prio_write(ca, true);
2084 		mutex_unlock(&c->bucket_lock);
2085 
2086 		err = "cannot allocate new UUID bucket";
2087 		if (__uuid_write(c))
2088 			goto err;
2089 
2090 		err = "cannot allocate new btree root";
2091 		c->root = __bch_btree_node_alloc(c, NULL, 0, true, NULL);
2092 		if (IS_ERR(c->root))
2093 			goto err;
2094 
2095 		mutex_lock(&c->root->write_lock);
2096 		bkey_copy_key(&c->root->key, &MAX_KEY);
2097 		bch_btree_node_write(c->root, &cl);
2098 		mutex_unlock(&c->root->write_lock);
2099 
2100 		bch_btree_set_root(c->root);
2101 		rw_unlock(true, c->root);
2102 
2103 		/*
2104 		 * We don't want to write the first journal entry until
2105 		 * everything is set up - fortunately journal entries won't be
2106 		 * written until the SET_CACHE_SYNC() here:
2107 		 */
2108 		SET_CACHE_SYNC(&c->cache->sb, true);
2109 
2110 		bch_journal_next(&c->journal);
2111 		bch_journal_meta(c, &cl);
2112 	}
2113 
2114 	err = "error starting gc thread";
2115 	if (bch_gc_thread_start(c))
2116 		goto err;
2117 
2118 	closure_sync(&cl);
2119 	c->cache->sb.last_mount = (u32)ktime_get_real_seconds();
2120 	bcache_write_super(c);
2121 
2122 	if (bch_has_feature_obso_large_bucket(&c->cache->sb))
2123 		pr_err("Detect obsoleted large bucket layout, all attached bcache device will be read-only\n");
2124 
2125 	list_for_each_entry_safe(dc, t, &uncached_devices, list)
2126 		bch_cached_dev_attach(dc, c, NULL);
2127 
2128 	flash_devs_run(c);
2129 
2130 	bch_journal_space_reserve(&c->journal);
2131 	set_bit(CACHE_SET_RUNNING, &c->flags);
2132 	return 0;
2133 err:
2134 	while (!list_empty(&journal)) {
2135 		l = list_first_entry(&journal, struct journal_replay, list);
2136 		list_del(&l->list);
2137 		kfree(l);
2138 	}
2139 
2140 	closure_sync(&cl);
2141 
2142 	bch_cache_set_error(c, "%s", err);
2143 
2144 	return -EIO;
2145 }
2146 
register_cache_set(struct cache * ca)2147 static const char *register_cache_set(struct cache *ca)
2148 {
2149 	char buf[12];
2150 	const char *err = "cannot allocate memory";
2151 	struct cache_set *c;
2152 
2153 	list_for_each_entry(c, &bch_cache_sets, list)
2154 		if (!memcmp(c->set_uuid, ca->sb.set_uuid, 16)) {
2155 			if (c->cache)
2156 				return "duplicate cache set member";
2157 
2158 			goto found;
2159 		}
2160 
2161 	c = bch_cache_set_alloc(&ca->sb);
2162 	if (!c)
2163 		return err;
2164 
2165 	err = "error creating kobject";
2166 	if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->set_uuid) ||
2167 	    kobject_add(&c->internal, &c->kobj, "internal"))
2168 		goto err;
2169 
2170 	if (bch_cache_accounting_add_kobjs(&c->accounting, &c->kobj))
2171 		goto err;
2172 
2173 	bch_debug_init_cache_set(c);
2174 
2175 	list_add(&c->list, &bch_cache_sets);
2176 found:
2177 	sprintf(buf, "cache%i", ca->sb.nr_this_dev);
2178 	if (sysfs_create_link(&ca->kobj, &c->kobj, "set") ||
2179 	    sysfs_create_link(&c->kobj, &ca->kobj, buf))
2180 		goto err;
2181 
2182 	kobject_get(&ca->kobj);
2183 	ca->set = c;
2184 	ca->set->cache = ca;
2185 
2186 	err = "failed to run cache set";
2187 	if (run_cache_set(c) < 0)
2188 		goto err;
2189 
2190 	return NULL;
2191 err:
2192 	bch_cache_set_unregister(c);
2193 	return err;
2194 }
2195 
2196 /* Cache device */
2197 
2198 /* When ca->kobj released */
bch_cache_release(struct kobject * kobj)2199 void bch_cache_release(struct kobject *kobj)
2200 {
2201 	struct cache *ca = container_of(kobj, struct cache, kobj);
2202 	unsigned int i;
2203 
2204 	if (ca->set) {
2205 		BUG_ON(ca->set->cache != ca);
2206 		ca->set->cache = NULL;
2207 	}
2208 
2209 	free_pages((unsigned long) ca->disk_buckets, ilog2(meta_bucket_pages(&ca->sb)));
2210 	kfree(ca->prio_buckets);
2211 	vfree(ca->buckets);
2212 
2213 	free_heap(&ca->heap);
2214 	free_fifo(&ca->free_inc);
2215 
2216 	for (i = 0; i < RESERVE_NR; i++)
2217 		free_fifo(&ca->free[i]);
2218 
2219 	if (ca->sb_disk)
2220 		put_page(virt_to_page(ca->sb_disk));
2221 
2222 	if (!IS_ERR_OR_NULL(ca->bdev))
2223 		blkdev_put(ca->bdev, ca);
2224 
2225 	kfree(ca);
2226 	module_put(THIS_MODULE);
2227 }
2228 
cache_alloc(struct cache * ca)2229 static int cache_alloc(struct cache *ca)
2230 {
2231 	size_t free;
2232 	size_t btree_buckets;
2233 	struct bucket *b;
2234 	int ret = -ENOMEM;
2235 	const char *err = NULL;
2236 
2237 	__module_get(THIS_MODULE);
2238 	kobject_init(&ca->kobj, &bch_cache_ktype);
2239 
2240 	bio_init(&ca->journal.bio, NULL, ca->journal.bio.bi_inline_vecs, 8, 0);
2241 
2242 	/*
2243 	 * when ca->sb.njournal_buckets is not zero, journal exists,
2244 	 * and in bch_journal_replay(), tree node may split,
2245 	 * so bucket of RESERVE_BTREE type is needed,
2246 	 * the worst situation is all journal buckets are valid journal,
2247 	 * and all the keys need to replay,
2248 	 * so the number of  RESERVE_BTREE type buckets should be as much
2249 	 * as journal buckets
2250 	 */
2251 	btree_buckets = ca->sb.njournal_buckets ?: 8;
2252 	free = roundup_pow_of_two(ca->sb.nbuckets) >> 10;
2253 	if (!free) {
2254 		ret = -EPERM;
2255 		err = "ca->sb.nbuckets is too small";
2256 		goto err_free;
2257 	}
2258 
2259 	if (!init_fifo(&ca->free[RESERVE_BTREE], btree_buckets,
2260 						GFP_KERNEL)) {
2261 		err = "ca->free[RESERVE_BTREE] alloc failed";
2262 		goto err_btree_alloc;
2263 	}
2264 
2265 	if (!init_fifo_exact(&ca->free[RESERVE_PRIO], prio_buckets(ca),
2266 							GFP_KERNEL)) {
2267 		err = "ca->free[RESERVE_PRIO] alloc failed";
2268 		goto err_prio_alloc;
2269 	}
2270 
2271 	if (!init_fifo(&ca->free[RESERVE_MOVINGGC], free, GFP_KERNEL)) {
2272 		err = "ca->free[RESERVE_MOVINGGC] alloc failed";
2273 		goto err_movinggc_alloc;
2274 	}
2275 
2276 	if (!init_fifo(&ca->free[RESERVE_NONE], free, GFP_KERNEL)) {
2277 		err = "ca->free[RESERVE_NONE] alloc failed";
2278 		goto err_none_alloc;
2279 	}
2280 
2281 	if (!init_fifo(&ca->free_inc, free << 2, GFP_KERNEL)) {
2282 		err = "ca->free_inc alloc failed";
2283 		goto err_free_inc_alloc;
2284 	}
2285 
2286 	if (!init_heap(&ca->heap, free << 3, GFP_KERNEL)) {
2287 		err = "ca->heap alloc failed";
2288 		goto err_heap_alloc;
2289 	}
2290 
2291 	ca->buckets = vzalloc(array_size(sizeof(struct bucket),
2292 			      ca->sb.nbuckets));
2293 	if (!ca->buckets) {
2294 		err = "ca->buckets alloc failed";
2295 		goto err_buckets_alloc;
2296 	}
2297 
2298 	ca->prio_buckets = kzalloc(array3_size(sizeof(uint64_t),
2299 				   prio_buckets(ca), 2),
2300 				   GFP_KERNEL);
2301 	if (!ca->prio_buckets) {
2302 		err = "ca->prio_buckets alloc failed";
2303 		goto err_prio_buckets_alloc;
2304 	}
2305 
2306 	ca->disk_buckets = alloc_meta_bucket_pages(GFP_KERNEL, &ca->sb);
2307 	if (!ca->disk_buckets) {
2308 		err = "ca->disk_buckets alloc failed";
2309 		goto err_disk_buckets_alloc;
2310 	}
2311 
2312 	ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca);
2313 
2314 	for_each_bucket(b, ca)
2315 		atomic_set(&b->pin, 0);
2316 	return 0;
2317 
2318 err_disk_buckets_alloc:
2319 	kfree(ca->prio_buckets);
2320 err_prio_buckets_alloc:
2321 	vfree(ca->buckets);
2322 err_buckets_alloc:
2323 	free_heap(&ca->heap);
2324 err_heap_alloc:
2325 	free_fifo(&ca->free_inc);
2326 err_free_inc_alloc:
2327 	free_fifo(&ca->free[RESERVE_NONE]);
2328 err_none_alloc:
2329 	free_fifo(&ca->free[RESERVE_MOVINGGC]);
2330 err_movinggc_alloc:
2331 	free_fifo(&ca->free[RESERVE_PRIO]);
2332 err_prio_alloc:
2333 	free_fifo(&ca->free[RESERVE_BTREE]);
2334 err_btree_alloc:
2335 err_free:
2336 	module_put(THIS_MODULE);
2337 	if (err)
2338 		pr_notice("error %pg: %s\n", ca->bdev, err);
2339 	return ret;
2340 }
2341 
register_cache(struct cache_sb * sb,struct cache_sb_disk * sb_disk,struct block_device * bdev,struct cache * ca)2342 static int register_cache(struct cache_sb *sb, struct cache_sb_disk *sb_disk,
2343 				struct block_device *bdev, struct cache *ca)
2344 {
2345 	const char *err = NULL; /* must be set for any error case */
2346 	int ret = 0;
2347 
2348 	memcpy(&ca->sb, sb, sizeof(struct cache_sb));
2349 	ca->bdev = bdev;
2350 	ca->sb_disk = sb_disk;
2351 
2352 	if (bdev_max_discard_sectors((bdev)))
2353 		ca->discard = CACHE_DISCARD(&ca->sb);
2354 
2355 	ret = cache_alloc(ca);
2356 	if (ret != 0) {
2357 		/*
2358 		 * If we failed here, it means ca->kobj is not initialized yet,
2359 		 * kobject_put() won't be called and there is no chance to
2360 		 * call blkdev_put() to bdev in bch_cache_release(). So we
2361 		 * explicitly call blkdev_put() here.
2362 		 */
2363 		blkdev_put(bdev, ca);
2364 		if (ret == -ENOMEM)
2365 			err = "cache_alloc(): -ENOMEM";
2366 		else if (ret == -EPERM)
2367 			err = "cache_alloc(): cache device is too small";
2368 		else
2369 			err = "cache_alloc(): unknown error";
2370 		goto err;
2371 	}
2372 
2373 	if (kobject_add(&ca->kobj, bdev_kobj(bdev), "bcache")) {
2374 		err = "error calling kobject_add";
2375 		ret = -ENOMEM;
2376 		goto out;
2377 	}
2378 
2379 	mutex_lock(&bch_register_lock);
2380 	err = register_cache_set(ca);
2381 	mutex_unlock(&bch_register_lock);
2382 
2383 	if (err) {
2384 		ret = -ENODEV;
2385 		goto out;
2386 	}
2387 
2388 	pr_info("registered cache device %pg\n", ca->bdev);
2389 
2390 out:
2391 	kobject_put(&ca->kobj);
2392 
2393 err:
2394 	if (err)
2395 		pr_notice("error %pg: %s\n", ca->bdev, err);
2396 
2397 	return ret;
2398 }
2399 
2400 /* Global interfaces/init */
2401 
2402 static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr,
2403 			       const char *buffer, size_t size);
2404 static ssize_t bch_pending_bdevs_cleanup(struct kobject *k,
2405 					 struct kobj_attribute *attr,
2406 					 const char *buffer, size_t size);
2407 
2408 kobj_attribute_write(register,		register_bcache);
2409 kobj_attribute_write(register_quiet,	register_bcache);
2410 kobj_attribute_write(pendings_cleanup,	bch_pending_bdevs_cleanup);
2411 
bch_is_open_backing(dev_t dev)2412 static bool bch_is_open_backing(dev_t dev)
2413 {
2414 	struct cache_set *c, *tc;
2415 	struct cached_dev *dc, *t;
2416 
2417 	list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
2418 		list_for_each_entry_safe(dc, t, &c->cached_devs, list)
2419 			if (dc->bdev->bd_dev == dev)
2420 				return true;
2421 	list_for_each_entry_safe(dc, t, &uncached_devices, list)
2422 		if (dc->bdev->bd_dev == dev)
2423 			return true;
2424 	return false;
2425 }
2426 
bch_is_open_cache(dev_t dev)2427 static bool bch_is_open_cache(dev_t dev)
2428 {
2429 	struct cache_set *c, *tc;
2430 
2431 	list_for_each_entry_safe(c, tc, &bch_cache_sets, list) {
2432 		struct cache *ca = c->cache;
2433 
2434 		if (ca->bdev->bd_dev == dev)
2435 			return true;
2436 	}
2437 
2438 	return false;
2439 }
2440 
bch_is_open(dev_t dev)2441 static bool bch_is_open(dev_t dev)
2442 {
2443 	return bch_is_open_cache(dev) || bch_is_open_backing(dev);
2444 }
2445 
2446 struct async_reg_args {
2447 	struct delayed_work reg_work;
2448 	char *path;
2449 	struct cache_sb *sb;
2450 	struct cache_sb_disk *sb_disk;
2451 	struct block_device *bdev;
2452 	void *holder;
2453 };
2454 
register_bdev_worker(struct work_struct * work)2455 static void register_bdev_worker(struct work_struct *work)
2456 {
2457 	int fail = false;
2458 	struct async_reg_args *args =
2459 		container_of(work, struct async_reg_args, reg_work.work);
2460 
2461 	mutex_lock(&bch_register_lock);
2462 	if (register_bdev(args->sb, args->sb_disk, args->bdev, args->holder)
2463 	    < 0)
2464 		fail = true;
2465 	mutex_unlock(&bch_register_lock);
2466 
2467 	if (fail)
2468 		pr_info("error %s: fail to register backing device\n",
2469 			args->path);
2470 	kfree(args->sb);
2471 	kfree(args->path);
2472 	kfree(args);
2473 	module_put(THIS_MODULE);
2474 }
2475 
register_cache_worker(struct work_struct * work)2476 static void register_cache_worker(struct work_struct *work)
2477 {
2478 	int fail = false;
2479 	struct async_reg_args *args =
2480 		container_of(work, struct async_reg_args, reg_work.work);
2481 
2482 	/* blkdev_put() will be called in bch_cache_release() */
2483 	if (register_cache(args->sb, args->sb_disk, args->bdev, args->holder))
2484 		fail = true;
2485 
2486 	if (fail)
2487 		pr_info("error %s: fail to register cache device\n",
2488 			args->path);
2489 	kfree(args->sb);
2490 	kfree(args->path);
2491 	kfree(args);
2492 	module_put(THIS_MODULE);
2493 }
2494 
register_device_async(struct async_reg_args * args)2495 static void register_device_async(struct async_reg_args *args)
2496 {
2497 	if (SB_IS_BDEV(args->sb))
2498 		INIT_DELAYED_WORK(&args->reg_work, register_bdev_worker);
2499 	else
2500 		INIT_DELAYED_WORK(&args->reg_work, register_cache_worker);
2501 
2502 	/* 10 jiffies is enough for a delay */
2503 	queue_delayed_work(system_wq, &args->reg_work, 10);
2504 }
2505 
alloc_holder_object(struct cache_sb * sb)2506 static void *alloc_holder_object(struct cache_sb *sb)
2507 {
2508 	if (SB_IS_BDEV(sb))
2509 		return kzalloc(sizeof(struct cached_dev), GFP_KERNEL);
2510 	return kzalloc(sizeof(struct cache), GFP_KERNEL);
2511 }
2512 
register_bcache(struct kobject * k,struct kobj_attribute * attr,const char * buffer,size_t size)2513 static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr,
2514 			       const char *buffer, size_t size)
2515 {
2516 	const char *err;
2517 	char *path = NULL;
2518 	struct cache_sb *sb;
2519 	struct cache_sb_disk *sb_disk;
2520 	struct block_device *bdev, *bdev2;
2521 	void *holder = NULL;
2522 	ssize_t ret;
2523 	bool async_registration = false;
2524 	bool quiet = false;
2525 
2526 #ifdef CONFIG_BCACHE_ASYNC_REGISTRATION
2527 	async_registration = true;
2528 #endif
2529 
2530 	ret = -EBUSY;
2531 	err = "failed to reference bcache module";
2532 	if (!try_module_get(THIS_MODULE))
2533 		goto out;
2534 
2535 	/* For latest state of bcache_is_reboot */
2536 	smp_mb();
2537 	err = "bcache is in reboot";
2538 	if (bcache_is_reboot)
2539 		goto out_module_put;
2540 
2541 	ret = -ENOMEM;
2542 	err = "cannot allocate memory";
2543 	path = kstrndup(buffer, size, GFP_KERNEL);
2544 	if (!path)
2545 		goto out_module_put;
2546 
2547 	sb = kmalloc(sizeof(struct cache_sb), GFP_KERNEL);
2548 	if (!sb)
2549 		goto out_free_path;
2550 
2551 	ret = -EINVAL;
2552 	err = "failed to open device";
2553 	bdev = blkdev_get_by_path(strim(path), BLK_OPEN_READ, NULL, NULL);
2554 	if (IS_ERR(bdev))
2555 		goto out_free_sb;
2556 
2557 	err = "failed to set blocksize";
2558 	if (set_blocksize(bdev, 4096))
2559 		goto out_blkdev_put;
2560 
2561 	err = read_super(sb, bdev, &sb_disk);
2562 	if (err)
2563 		goto out_blkdev_put;
2564 
2565 	holder = alloc_holder_object(sb);
2566 	if (!holder) {
2567 		ret = -ENOMEM;
2568 		err = "cannot allocate memory";
2569 		goto out_put_sb_page;
2570 	}
2571 
2572 	/* Now reopen in exclusive mode with proper holder */
2573 	bdev2 = blkdev_get_by_dev(bdev->bd_dev, BLK_OPEN_READ | BLK_OPEN_WRITE,
2574 				  holder, NULL);
2575 	blkdev_put(bdev, NULL);
2576 	bdev = bdev2;
2577 	if (IS_ERR(bdev)) {
2578 		ret = PTR_ERR(bdev);
2579 		bdev = NULL;
2580 		if (ret == -EBUSY) {
2581 			dev_t dev;
2582 
2583 			mutex_lock(&bch_register_lock);
2584 			if (lookup_bdev(strim(path), &dev) == 0 &&
2585 			    bch_is_open(dev))
2586 				err = "device already registered";
2587 			else
2588 				err = "device busy";
2589 			mutex_unlock(&bch_register_lock);
2590 			if (attr == &ksysfs_register_quiet) {
2591 				quiet = true;
2592 				ret = size;
2593 			}
2594 		}
2595 		goto out_free_holder;
2596 	}
2597 
2598 	err = "failed to register device";
2599 
2600 	if (async_registration) {
2601 		/* register in asynchronous way */
2602 		struct async_reg_args *args =
2603 			kzalloc(sizeof(struct async_reg_args), GFP_KERNEL);
2604 
2605 		if (!args) {
2606 			ret = -ENOMEM;
2607 			err = "cannot allocate memory";
2608 			goto out_free_holder;
2609 		}
2610 
2611 		args->path	= path;
2612 		args->sb	= sb;
2613 		args->sb_disk	= sb_disk;
2614 		args->bdev	= bdev;
2615 		args->holder	= holder;
2616 		register_device_async(args);
2617 		/* No wait and returns to user space */
2618 		goto async_done;
2619 	}
2620 
2621 	if (SB_IS_BDEV(sb)) {
2622 		mutex_lock(&bch_register_lock);
2623 		ret = register_bdev(sb, sb_disk, bdev, holder);
2624 		mutex_unlock(&bch_register_lock);
2625 		/* blkdev_put() will be called in cached_dev_free() */
2626 		if (ret < 0)
2627 			goto out_free_sb;
2628 	} else {
2629 		/* blkdev_put() will be called in bch_cache_release() */
2630 		ret = register_cache(sb, sb_disk, bdev, holder);
2631 		if (ret)
2632 			goto out_free_sb;
2633 	}
2634 
2635 	kfree(sb);
2636 	kfree(path);
2637 	module_put(THIS_MODULE);
2638 async_done:
2639 	return size;
2640 
2641 out_free_holder:
2642 	kfree(holder);
2643 out_put_sb_page:
2644 	put_page(virt_to_page(sb_disk));
2645 out_blkdev_put:
2646 	if (bdev)
2647 		blkdev_put(bdev, holder);
2648 out_free_sb:
2649 	kfree(sb);
2650 out_free_path:
2651 	kfree(path);
2652 	path = NULL;
2653 out_module_put:
2654 	module_put(THIS_MODULE);
2655 out:
2656 	if (!quiet)
2657 		pr_info("error %s: %s\n", path?path:"", err);
2658 	return ret;
2659 }
2660 
2661 
2662 struct pdev {
2663 	struct list_head list;
2664 	struct cached_dev *dc;
2665 };
2666 
bch_pending_bdevs_cleanup(struct kobject * k,struct kobj_attribute * attr,const char * buffer,size_t size)2667 static ssize_t bch_pending_bdevs_cleanup(struct kobject *k,
2668 					 struct kobj_attribute *attr,
2669 					 const char *buffer,
2670 					 size_t size)
2671 {
2672 	LIST_HEAD(pending_devs);
2673 	ssize_t ret = size;
2674 	struct cached_dev *dc, *tdc;
2675 	struct pdev *pdev, *tpdev;
2676 	struct cache_set *c, *tc;
2677 
2678 	mutex_lock(&bch_register_lock);
2679 	list_for_each_entry_safe(dc, tdc, &uncached_devices, list) {
2680 		pdev = kmalloc(sizeof(struct pdev), GFP_KERNEL);
2681 		if (!pdev)
2682 			break;
2683 		pdev->dc = dc;
2684 		list_add(&pdev->list, &pending_devs);
2685 	}
2686 
2687 	list_for_each_entry_safe(pdev, tpdev, &pending_devs, list) {
2688 		char *pdev_set_uuid = pdev->dc->sb.set_uuid;
2689 		list_for_each_entry_safe(c, tc, &bch_cache_sets, list) {
2690 			char *set_uuid = c->set_uuid;
2691 
2692 			if (!memcmp(pdev_set_uuid, set_uuid, 16)) {
2693 				list_del(&pdev->list);
2694 				kfree(pdev);
2695 				break;
2696 			}
2697 		}
2698 	}
2699 	mutex_unlock(&bch_register_lock);
2700 
2701 	list_for_each_entry_safe(pdev, tpdev, &pending_devs, list) {
2702 		pr_info("delete pdev %p\n", pdev);
2703 		list_del(&pdev->list);
2704 		bcache_device_stop(&pdev->dc->disk);
2705 		kfree(pdev);
2706 	}
2707 
2708 	return ret;
2709 }
2710 
bcache_reboot(struct notifier_block * n,unsigned long code,void * x)2711 static int bcache_reboot(struct notifier_block *n, unsigned long code, void *x)
2712 {
2713 	if (bcache_is_reboot)
2714 		return NOTIFY_DONE;
2715 
2716 	if (code == SYS_DOWN ||
2717 	    code == SYS_HALT ||
2718 	    code == SYS_POWER_OFF) {
2719 		DEFINE_WAIT(wait);
2720 		unsigned long start = jiffies;
2721 		bool stopped = false;
2722 
2723 		struct cache_set *c, *tc;
2724 		struct cached_dev *dc, *tdc;
2725 
2726 		mutex_lock(&bch_register_lock);
2727 
2728 		if (bcache_is_reboot)
2729 			goto out;
2730 
2731 		/* New registration is rejected since now */
2732 		bcache_is_reboot = true;
2733 		/*
2734 		 * Make registering caller (if there is) on other CPU
2735 		 * core know bcache_is_reboot set to true earlier
2736 		 */
2737 		smp_mb();
2738 
2739 		if (list_empty(&bch_cache_sets) &&
2740 		    list_empty(&uncached_devices))
2741 			goto out;
2742 
2743 		mutex_unlock(&bch_register_lock);
2744 
2745 		pr_info("Stopping all devices:\n");
2746 
2747 		/*
2748 		 * The reason bch_register_lock is not held to call
2749 		 * bch_cache_set_stop() and bcache_device_stop() is to
2750 		 * avoid potential deadlock during reboot, because cache
2751 		 * set or bcache device stopping process will acquire
2752 		 * bch_register_lock too.
2753 		 *
2754 		 * We are safe here because bcache_is_reboot sets to
2755 		 * true already, register_bcache() will reject new
2756 		 * registration now. bcache_is_reboot also makes sure
2757 		 * bcache_reboot() won't be re-entered on by other thread,
2758 		 * so there is no race in following list iteration by
2759 		 * list_for_each_entry_safe().
2760 		 */
2761 		list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
2762 			bch_cache_set_stop(c);
2763 
2764 		list_for_each_entry_safe(dc, tdc, &uncached_devices, list)
2765 			bcache_device_stop(&dc->disk);
2766 
2767 
2768 		/*
2769 		 * Give an early chance for other kthreads and
2770 		 * kworkers to stop themselves
2771 		 */
2772 		schedule();
2773 
2774 		/* What's a condition variable? */
2775 		while (1) {
2776 			long timeout = start + 10 * HZ - jiffies;
2777 
2778 			mutex_lock(&bch_register_lock);
2779 			stopped = list_empty(&bch_cache_sets) &&
2780 				list_empty(&uncached_devices);
2781 
2782 			if (timeout < 0 || stopped)
2783 				break;
2784 
2785 			prepare_to_wait(&unregister_wait, &wait,
2786 					TASK_UNINTERRUPTIBLE);
2787 
2788 			mutex_unlock(&bch_register_lock);
2789 			schedule_timeout(timeout);
2790 		}
2791 
2792 		finish_wait(&unregister_wait, &wait);
2793 
2794 		if (stopped)
2795 			pr_info("All devices stopped\n");
2796 		else
2797 			pr_notice("Timeout waiting for devices to be closed\n");
2798 out:
2799 		mutex_unlock(&bch_register_lock);
2800 	}
2801 
2802 	return NOTIFY_DONE;
2803 }
2804 
2805 static struct notifier_block reboot = {
2806 	.notifier_call	= bcache_reboot,
2807 	.priority	= INT_MAX, /* before any real devices */
2808 };
2809 
bcache_exit(void)2810 static void bcache_exit(void)
2811 {
2812 	bch_debug_exit();
2813 	bch_request_exit();
2814 	if (bcache_kobj)
2815 		kobject_put(bcache_kobj);
2816 	if (bcache_wq)
2817 		destroy_workqueue(bcache_wq);
2818 	if (bch_journal_wq)
2819 		destroy_workqueue(bch_journal_wq);
2820 	if (bch_flush_wq)
2821 		destroy_workqueue(bch_flush_wq);
2822 	bch_btree_exit();
2823 
2824 	if (bcache_major)
2825 		unregister_blkdev(bcache_major, "bcache");
2826 	unregister_reboot_notifier(&reboot);
2827 	mutex_destroy(&bch_register_lock);
2828 }
2829 
2830 /* Check and fixup module parameters */
check_module_parameters(void)2831 static void check_module_parameters(void)
2832 {
2833 	if (bch_cutoff_writeback_sync == 0)
2834 		bch_cutoff_writeback_sync = CUTOFF_WRITEBACK_SYNC;
2835 	else if (bch_cutoff_writeback_sync > CUTOFF_WRITEBACK_SYNC_MAX) {
2836 		pr_warn("set bch_cutoff_writeback_sync (%u) to max value %u\n",
2837 			bch_cutoff_writeback_sync, CUTOFF_WRITEBACK_SYNC_MAX);
2838 		bch_cutoff_writeback_sync = CUTOFF_WRITEBACK_SYNC_MAX;
2839 	}
2840 
2841 	if (bch_cutoff_writeback == 0)
2842 		bch_cutoff_writeback = CUTOFF_WRITEBACK;
2843 	else if (bch_cutoff_writeback > CUTOFF_WRITEBACK_MAX) {
2844 		pr_warn("set bch_cutoff_writeback (%u) to max value %u\n",
2845 			bch_cutoff_writeback, CUTOFF_WRITEBACK_MAX);
2846 		bch_cutoff_writeback = CUTOFF_WRITEBACK_MAX;
2847 	}
2848 
2849 	if (bch_cutoff_writeback > bch_cutoff_writeback_sync) {
2850 		pr_warn("set bch_cutoff_writeback (%u) to %u\n",
2851 			bch_cutoff_writeback, bch_cutoff_writeback_sync);
2852 		bch_cutoff_writeback = bch_cutoff_writeback_sync;
2853 	}
2854 }
2855 
bcache_init(void)2856 static int __init bcache_init(void)
2857 {
2858 	static const struct attribute *files[] = {
2859 		&ksysfs_register.attr,
2860 		&ksysfs_register_quiet.attr,
2861 		&ksysfs_pendings_cleanup.attr,
2862 		NULL
2863 	};
2864 
2865 	check_module_parameters();
2866 
2867 	mutex_init(&bch_register_lock);
2868 	init_waitqueue_head(&unregister_wait);
2869 	register_reboot_notifier(&reboot);
2870 
2871 	bcache_major = register_blkdev(0, "bcache");
2872 	if (bcache_major < 0) {
2873 		unregister_reboot_notifier(&reboot);
2874 		mutex_destroy(&bch_register_lock);
2875 		return bcache_major;
2876 	}
2877 
2878 	if (bch_btree_init())
2879 		goto err;
2880 
2881 	bcache_wq = alloc_workqueue("bcache", WQ_MEM_RECLAIM, 0);
2882 	if (!bcache_wq)
2883 		goto err;
2884 
2885 	/*
2886 	 * Let's not make this `WQ_MEM_RECLAIM` for the following reasons:
2887 	 *
2888 	 * 1. It used `system_wq` before which also does no memory reclaim.
2889 	 * 2. With `WQ_MEM_RECLAIM` desktop stalls, increased boot times, and
2890 	 *    reduced throughput can be observed.
2891 	 *
2892 	 * We still want to user our own queue to not congest the `system_wq`.
2893 	 */
2894 	bch_flush_wq = alloc_workqueue("bch_flush", 0, 0);
2895 	if (!bch_flush_wq)
2896 		goto err;
2897 
2898 	bch_journal_wq = alloc_workqueue("bch_journal", WQ_MEM_RECLAIM, 0);
2899 	if (!bch_journal_wq)
2900 		goto err;
2901 
2902 	bcache_kobj = kobject_create_and_add("bcache", fs_kobj);
2903 	if (!bcache_kobj)
2904 		goto err;
2905 
2906 	if (bch_request_init() ||
2907 	    sysfs_create_files(bcache_kobj, files))
2908 		goto err;
2909 
2910 	bch_debug_init();
2911 	closure_debug_init();
2912 
2913 	bcache_is_reboot = false;
2914 
2915 	return 0;
2916 err:
2917 	bcache_exit();
2918 	return -ENOMEM;
2919 }
2920 
2921 /*
2922  * Module hooks
2923  */
2924 module_exit(bcache_exit);
2925 module_init(bcache_init);
2926 
2927 module_param(bch_cutoff_writeback, uint, 0);
2928 MODULE_PARM_DESC(bch_cutoff_writeback, "threshold to cutoff writeback");
2929 
2930 module_param(bch_cutoff_writeback_sync, uint, 0);
2931 MODULE_PARM_DESC(bch_cutoff_writeback_sync, "hard threshold to cutoff writeback");
2932 
2933 MODULE_DESCRIPTION("Bcache: a Linux block layer cache");
2934 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
2935 MODULE_LICENSE("GPL");
2936