blob: 849ad441cd76ba6b04ba2fa2aaac51471acdfd67 [file] [log] [blame]
Kent Overstreetcafe5632013-03-23 16:11:31 -07001/*
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"
Kent Overstreet65d45232013-12-20 17:22:05 -080012#include "extents.h"
Kent Overstreetcafe5632013-03-23 16:11:31 -070013#include "request.h"
Kent Overstreet279afba2013-06-05 06:21:07 -070014#include "writeback.h"
Kent Overstreetcafe5632013-03-23 16:11:31 -070015
Kent Overstreetc37511b2013-04-26 15:39:55 -070016#include <linux/blkdev.h>
Kent Overstreetcafe5632013-03-23 16:11:31 -070017#include <linux/buffer_head.h>
18#include <linux/debugfs.h>
19#include <linux/genhd.h>
Kent Overstreet28935ab2013-07-31 01:12:02 -070020#include <linux/idr.h>
Kent Overstreet79826c32013-07-10 18:31:58 -070021#include <linux/kthread.h>
Kent Overstreetcafe5632013-03-23 16:11:31 -070022#include <linux/module.h>
23#include <linux/random.h>
24#include <linux/reboot.h>
25#include <linux/sysfs.h>
26
27MODULE_LICENSE("GPL");
28MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
29
30static const char bcache_magic[] = {
31 0xc6, 0x85, 0x73, 0xf6, 0x4e, 0x1a, 0x45, 0xca,
32 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81
33};
34
35static 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 */
41const char * const bch_cache_modes[] = {
42 "default",
43 "writethrough",
44 "writeback",
45 "writearound",
46 "none",
47 NULL
48};
49
Kent Overstreetcafe5632013-03-23 16:11:31 -070050static struct kobject *bcache_kobj;
51struct mutex bch_register_lock;
52LIST_HEAD(bch_cache_sets);
53static LIST_HEAD(uncached_devices);
54
Kent Overstreet28935ab2013-07-31 01:12:02 -070055static int bcache_major;
56static DEFINE_IDA(bcache_minor);
Kent Overstreetcafe5632013-03-23 16:11:31 -070057static wait_queue_head_t unregister_wait;
58struct workqueue_struct *bcache_wq;
59
60#define BTREE_MAX_PAGES (256 * 1024 / PAGE_SIZE)
61
Kent Overstreetcafe5632013-03-23 16:11:31 -070062/* Superblock */
63
64static const char *read_super(struct cache_sb *sb, struct block_device *bdev,
65 struct page **res)
66{
67 const char *err;
68 struct cache_sb *s;
69 struct buffer_head *bh = __bread(bdev, 1, SB_SIZE);
70 unsigned i;
71
72 if (!bh)
73 return "IO error";
74
75 s = (struct cache_sb *) bh->b_data;
76
77 sb->offset = le64_to_cpu(s->offset);
78 sb->version = le64_to_cpu(s->version);
79
80 memcpy(sb->magic, s->magic, 16);
81 memcpy(sb->uuid, s->uuid, 16);
82 memcpy(sb->set_uuid, s->set_uuid, 16);
83 memcpy(sb->label, s->label, SB_LABEL_SIZE);
84
85 sb->flags = le64_to_cpu(s->flags);
86 sb->seq = le64_to_cpu(s->seq);
Kent Overstreetcafe5632013-03-23 16:11:31 -070087 sb->last_mount = le32_to_cpu(s->last_mount);
Kent Overstreetcafe5632013-03-23 16:11:31 -070088 sb->first_bucket = le16_to_cpu(s->first_bucket);
89 sb->keys = le16_to_cpu(s->keys);
90
91 for (i = 0; i < SB_JOURNAL_BUCKETS; i++)
92 sb->d[i] = le64_to_cpu(s->d[i]);
93
94 pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u",
95 sb->version, sb->flags, sb->seq, sb->keys);
96
97 err = "Not a bcache superblock";
98 if (sb->offset != SB_SECTOR)
99 goto err;
100
101 if (memcmp(sb->magic, bcache_magic, 16))
102 goto err;
103
104 err = "Too many journal buckets";
105 if (sb->keys > SB_JOURNAL_BUCKETS)
106 goto err;
107
108 err = "Bad checksum";
109 if (s->csum != csum_set(s))
110 goto err;
111
112 err = "Bad UUID";
Kent Overstreet169ef1c2013-03-28 12:50:55 -0600113 if (bch_is_zero(sb->uuid, 16))
Kent Overstreetcafe5632013-03-23 16:11:31 -0700114 goto err;
115
Kent Overstreet8abb2a52013-04-23 21:51:48 -0700116 sb->block_size = le16_to_cpu(s->block_size);
117
118 err = "Superblock block size smaller than device block size";
119 if (sb->block_size << 9 < bdev_logical_block_size(bdev))
120 goto err;
121
Kent Overstreet29033812013-04-11 15:14:35 -0700122 switch (sb->version) {
123 case BCACHE_SB_VERSION_BDEV:
Kent Overstreet29033812013-04-11 15:14:35 -0700124 sb->data_offset = BDEV_DATA_START_DEFAULT;
125 break;
126 case BCACHE_SB_VERSION_BDEV_WITH_OFFSET:
Kent Overstreet29033812013-04-11 15:14:35 -0700127 sb->data_offset = le64_to_cpu(s->data_offset);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700128
Kent Overstreet29033812013-04-11 15:14:35 -0700129 err = "Bad data offset";
130 if (sb->data_offset < BDEV_DATA_START_DEFAULT)
Kent Overstreetcafe5632013-03-23 16:11:31 -0700131 goto err;
132
Kent Overstreet29033812013-04-11 15:14:35 -0700133 break;
134 case BCACHE_SB_VERSION_CDEV:
135 case BCACHE_SB_VERSION_CDEV_WITH_UUID:
136 sb->nbuckets = le64_to_cpu(s->nbuckets);
Kent Overstreet29033812013-04-11 15:14:35 -0700137 sb->bucket_size = le16_to_cpu(s->bucket_size);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700138
Kent Overstreet29033812013-04-11 15:14:35 -0700139 sb->nr_in_set = le16_to_cpu(s->nr_in_set);
140 sb->nr_this_dev = le16_to_cpu(s->nr_this_dev);
141
142 err = "Too many buckets";
143 if (sb->nbuckets > LONG_MAX)
144 goto err;
145
146 err = "Not enough buckets";
147 if (sb->nbuckets < 1 << 7)
148 goto err;
149
150 err = "Bad block/bucket size";
151 if (!is_power_of_2(sb->block_size) ||
152 sb->block_size > PAGE_SECTORS ||
153 !is_power_of_2(sb->bucket_size) ||
154 sb->bucket_size < PAGE_SECTORS)
155 goto err;
156
157 err = "Invalid superblock: device too small";
158 if (get_capacity(bdev->bd_disk) < sb->bucket_size * sb->nbuckets)
159 goto err;
160
161 err = "Bad UUID";
162 if (bch_is_zero(sb->set_uuid, 16))
163 goto err;
164
165 err = "Bad cache device number in set";
166 if (!sb->nr_in_set ||
167 sb->nr_in_set <= sb->nr_this_dev ||
168 sb->nr_in_set > MAX_CACHES_PER_SET)
169 goto err;
170
171 err = "Journal buckets not sequential";
172 for (i = 0; i < sb->keys; i++)
173 if (sb->d[i] != sb->first_bucket + i)
174 goto err;
175
176 err = "Too many journal buckets";
177 if (sb->first_bucket + sb->keys > sb->nbuckets)
178 goto err;
179
180 err = "Invalid superblock: first bucket comes before end of super";
181 if (sb->first_bucket * sb->bucket_size < 16)
182 goto err;
183
184 break;
185 default:
186 err = "Unsupported superblock version";
Kent Overstreetcafe5632013-03-23 16:11:31 -0700187 goto err;
Kent Overstreet29033812013-04-11 15:14:35 -0700188 }
189
Kent Overstreetcafe5632013-03-23 16:11:31 -0700190 sb->last_mount = get_seconds();
191 err = NULL;
192
193 get_page(bh->b_page);
194 *res = bh->b_page;
195err:
196 put_bh(bh);
197 return err;
198}
199
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200200static void write_bdev_super_endio(struct bio *bio)
Kent Overstreetcafe5632013-03-23 16:11:31 -0700201{
202 struct cached_dev *dc = bio->bi_private;
203 /* XXX: error checking */
204
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800205 closure_put(&dc->sb_write);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700206}
207
208static void __write_super(struct cache_sb *sb, struct bio *bio)
209{
210 struct cache_sb *out = page_address(bio->bi_io_vec[0].bv_page);
211 unsigned i;
212
Kent Overstreet4f024f32013-10-11 15:44:27 -0700213 bio->bi_iter.bi_sector = SB_SECTOR;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700214 bio->bi_iter.bi_size = SB_SIZE;
Mike Christiead0d9e72016-06-05 14:32:05 -0500215 bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
Kent Overstreet169ef1c2013-03-28 12:50:55 -0600216 bch_bio_map(bio, NULL);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700217
218 out->offset = cpu_to_le64(sb->offset);
219 out->version = cpu_to_le64(sb->version);
220
221 memcpy(out->uuid, sb->uuid, 16);
222 memcpy(out->set_uuid, sb->set_uuid, 16);
223 memcpy(out->label, sb->label, SB_LABEL_SIZE);
224
225 out->flags = cpu_to_le64(sb->flags);
226 out->seq = cpu_to_le64(sb->seq);
227
228 out->last_mount = cpu_to_le32(sb->last_mount);
229 out->first_bucket = cpu_to_le16(sb->first_bucket);
230 out->keys = cpu_to_le16(sb->keys);
231
232 for (i = 0; i < sb->keys; i++)
233 out->d[i] = cpu_to_le64(sb->d[i]);
234
235 out->csum = csum_set(out);
236
237 pr_debug("ver %llu, flags %llu, seq %llu",
238 sb->version, sb->flags, sb->seq);
239
Mike Christie4e49ea42016-06-05 14:31:41 -0500240 submit_bio(bio);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700241}
242
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800243static void bch_write_bdev_super_unlock(struct closure *cl)
244{
245 struct cached_dev *dc = container_of(cl, struct cached_dev, sb_write);
246
247 up(&dc->sb_write_mutex);
248}
249
Kent Overstreetcafe5632013-03-23 16:11:31 -0700250void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent)
251{
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800252 struct closure *cl = &dc->sb_write;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700253 struct bio *bio = &dc->sb_bio;
254
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800255 down(&dc->sb_write_mutex);
256 closure_init(cl, parent);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700257
258 bio_reset(bio);
259 bio->bi_bdev = dc->bdev;
260 bio->bi_end_io = write_bdev_super_endio;
261 bio->bi_private = dc;
262
263 closure_get(cl);
264 __write_super(&dc->sb, bio);
265
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800266 closure_return_with_destructor(cl, bch_write_bdev_super_unlock);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700267}
268
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200269static void write_super_endio(struct bio *bio)
Kent Overstreetcafe5632013-03-23 16:11:31 -0700270{
271 struct cache *ca = bio->bi_private;
272
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200273 bch_count_io_errors(ca, bio->bi_error, "writing superblock");
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800274 closure_put(&ca->set->sb_write);
275}
276
277static void bcache_write_super_unlock(struct closure *cl)
278{
279 struct cache_set *c = container_of(cl, struct cache_set, sb_write);
280
281 up(&c->sb_write_mutex);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700282}
283
284void bcache_write_super(struct cache_set *c)
285{
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800286 struct closure *cl = &c->sb_write;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700287 struct cache *ca;
288 unsigned i;
289
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800290 down(&c->sb_write_mutex);
291 closure_init(cl, &c->cl);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700292
293 c->sb.seq++;
294
295 for_each_cache(ca, c, i) {
296 struct bio *bio = &ca->sb_bio;
297
Kent Overstreet29033812013-04-11 15:14:35 -0700298 ca->sb.version = BCACHE_SB_VERSION_CDEV_WITH_UUID;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700299 ca->sb.seq = c->sb.seq;
300 ca->sb.last_mount = c->sb.last_mount;
301
302 SET_CACHE_SYNC(&ca->sb, CACHE_SYNC(&c->sb));
303
304 bio_reset(bio);
305 bio->bi_bdev = ca->bdev;
306 bio->bi_end_io = write_super_endio;
307 bio->bi_private = ca;
308
309 closure_get(cl);
310 __write_super(&ca->sb, bio);
311 }
312
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800313 closure_return_with_destructor(cl, bcache_write_super_unlock);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700314}
315
316/* UUID io */
317
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200318static void uuid_endio(struct bio *bio)
Kent Overstreetcafe5632013-03-23 16:11:31 -0700319{
320 struct closure *cl = bio->bi_private;
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800321 struct cache_set *c = container_of(cl, struct cache_set, uuid_write);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700322
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200323 cache_set_err_on(bio->bi_error, c, "accessing uuids");
Kent Overstreetcafe5632013-03-23 16:11:31 -0700324 bch_bbio_free(bio, c);
325 closure_put(cl);
326}
327
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800328static void uuid_io_unlock(struct closure *cl)
329{
330 struct cache_set *c = container_of(cl, struct cache_set, uuid_write);
331
332 up(&c->uuid_write_mutex);
333}
334
Mike Christiead0d9e72016-06-05 14:32:05 -0500335static void uuid_io(struct cache_set *c, int op, unsigned long op_flags,
Kent Overstreetcafe5632013-03-23 16:11:31 -0700336 struct bkey *k, struct closure *parent)
337{
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800338 struct closure *cl = &c->uuid_write;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700339 struct uuid_entry *u;
340 unsigned i;
Kent Overstreet85b14922013-05-14 20:33:16 -0700341 char buf[80];
Kent Overstreetcafe5632013-03-23 16:11:31 -0700342
343 BUG_ON(!parent);
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800344 down(&c->uuid_write_mutex);
345 closure_init(cl, parent);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700346
347 for (i = 0; i < KEY_PTRS(k); i++) {
348 struct bio *bio = bch_bbio_alloc(c);
349
Jens Axboe1eff9d32016-08-05 15:35:16 -0600350 bio->bi_opf = REQ_SYNC | REQ_META | op_flags;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700351 bio->bi_iter.bi_size = KEY_SIZE(k) << 9;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700352
353 bio->bi_end_io = uuid_endio;
354 bio->bi_private = cl;
Mike Christiead0d9e72016-06-05 14:32:05 -0500355 bio_set_op_attrs(bio, op, REQ_SYNC|REQ_META|op_flags);
Kent Overstreet169ef1c2013-03-28 12:50:55 -0600356 bch_bio_map(bio, c->uuids);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700357
358 bch_submit_bbio(bio, c, k, i);
359
Mike Christiead0d9e72016-06-05 14:32:05 -0500360 if (op != REQ_OP_WRITE)
Kent Overstreetcafe5632013-03-23 16:11:31 -0700361 break;
362 }
363
Kent Overstreetdc9d98d2013-12-17 23:47:33 -0800364 bch_extent_to_text(buf, sizeof(buf), k);
Mike Christiead0d9e72016-06-05 14:32:05 -0500365 pr_debug("%s UUIDs at %s", op == REQ_OP_WRITE ? "wrote" : "read", buf);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700366
367 for (u = c->uuids; u < c->uuids + c->nr_uuids; u++)
Kent Overstreet169ef1c2013-03-28 12:50:55 -0600368 if (!bch_is_zero(u->uuid, 16))
Kent Overstreetcafe5632013-03-23 16:11:31 -0700369 pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u",
370 u - c->uuids, u->uuid, u->label,
371 u->first_reg, u->last_reg, u->invalidated);
372
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800373 closure_return_with_destructor(cl, uuid_io_unlock);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700374}
375
376static char *uuid_read(struct cache_set *c, struct jset *j, struct closure *cl)
377{
378 struct bkey *k = &j->uuid_bucket;
379
Kent Overstreet65d45232013-12-20 17:22:05 -0800380 if (__bch_btree_ptr_invalid(c, k))
Kent Overstreetcafe5632013-03-23 16:11:31 -0700381 return "bad uuid pointer";
382
383 bkey_copy(&c->uuid_bucket, k);
Mike Christiead0d9e72016-06-05 14:32:05 -0500384 uuid_io(c, REQ_OP_READ, READ_SYNC, k, cl);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700385
386 if (j->version < BCACHE_JSET_VERSION_UUIDv1) {
387 struct uuid_entry_v0 *u0 = (void *) c->uuids;
388 struct uuid_entry *u1 = (void *) c->uuids;
389 int i;
390
391 closure_sync(cl);
392
393 /*
394 * Since the new uuid entry is bigger than the old, we have to
395 * convert starting at the highest memory address and work down
396 * in order to do it in place
397 */
398
399 for (i = c->nr_uuids - 1;
400 i >= 0;
401 --i) {
402 memcpy(u1[i].uuid, u0[i].uuid, 16);
403 memcpy(u1[i].label, u0[i].label, 32);
404
405 u1[i].first_reg = u0[i].first_reg;
406 u1[i].last_reg = u0[i].last_reg;
407 u1[i].invalidated = u0[i].invalidated;
408
409 u1[i].flags = 0;
410 u1[i].sectors = 0;
411 }
412 }
413
414 return NULL;
415}
416
417static int __uuid_write(struct cache_set *c)
418{
419 BKEY_PADDED(key) k;
420 struct closure cl;
421 closure_init_stack(&cl);
422
423 lockdep_assert_held(&bch_register_lock);
424
Kent Overstreet78365412013-12-17 01:29:34 -0800425 if (bch_bucket_alloc_set(c, RESERVE_BTREE, &k.key, 1, true))
Kent Overstreetcafe5632013-03-23 16:11:31 -0700426 return 1;
427
428 SET_KEY_SIZE(&k.key, c->sb.bucket_size);
Mike Christiead0d9e72016-06-05 14:32:05 -0500429 uuid_io(c, REQ_OP_WRITE, 0, &k.key, &cl);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700430 closure_sync(&cl);
431
432 bkey_copy(&c->uuid_bucket, &k.key);
Kent Overstreet3a3b6a42013-07-24 16:46:42 -0700433 bkey_put(c, &k.key);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700434 return 0;
435}
436
437int bch_uuid_write(struct cache_set *c)
438{
439 int ret = __uuid_write(c);
440
441 if (!ret)
442 bch_journal_meta(c, NULL);
443
444 return ret;
445}
446
447static struct uuid_entry *uuid_find(struct cache_set *c, const char *uuid)
448{
449 struct uuid_entry *u;
450
451 for (u = c->uuids;
452 u < c->uuids + c->nr_uuids; u++)
453 if (!memcmp(u->uuid, uuid, 16))
454 return u;
455
456 return NULL;
457}
458
459static struct uuid_entry *uuid_find_empty(struct cache_set *c)
460{
461 static const char zero_uuid[16] = "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0";
462 return uuid_find(c, zero_uuid);
463}
464
465/*
466 * Bucket priorities/gens:
467 *
468 * For each bucket, we store on disk its
469 * 8 bit gen
470 * 16 bit priority
471 *
472 * See alloc.c for an explanation of the gen. The priority is used to implement
473 * lru (and in the future other) cache replacement policies; for most purposes
474 * it's just an opaque integer.
475 *
476 * The gens and the priorities don't have a whole lot to do with each other, and
477 * it's actually the gens that must be written out at specific times - it's no
478 * big deal if the priorities don't get written, if we lose them we just reuse
479 * buckets in suboptimal order.
480 *
481 * On disk they're stored in a packed array, and in as many buckets are required
482 * to fit them all. The buckets we use to store them form a list; the journal
483 * header points to the first bucket, the first bucket points to the second
484 * bucket, et cetera.
485 *
486 * This code is used by the allocation code; periodically (whenever it runs out
487 * of buckets to allocate from) the allocation code will invalidate some
488 * buckets, but it can't use those buckets until their new gens are safely on
489 * disk.
490 */
491
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200492static void prio_endio(struct bio *bio)
Kent Overstreetcafe5632013-03-23 16:11:31 -0700493{
494 struct cache *ca = bio->bi_private;
495
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200496 cache_set_err_on(bio->bi_error, ca->set, "accessing priorities");
Kent Overstreetcafe5632013-03-23 16:11:31 -0700497 bch_bbio_free(bio, ca->set);
498 closure_put(&ca->prio);
499}
500
Mike Christiead0d9e72016-06-05 14:32:05 -0500501static void prio_io(struct cache *ca, uint64_t bucket, int op,
502 unsigned long op_flags)
Kent Overstreetcafe5632013-03-23 16:11:31 -0700503{
504 struct closure *cl = &ca->prio;
505 struct bio *bio = bch_bbio_alloc(ca->set);
506
507 closure_init_stack(cl);
508
Kent Overstreet4f024f32013-10-11 15:44:27 -0700509 bio->bi_iter.bi_sector = bucket * ca->sb.bucket_size;
510 bio->bi_bdev = ca->bdev;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700511 bio->bi_iter.bi_size = bucket_bytes(ca);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700512
513 bio->bi_end_io = prio_endio;
514 bio->bi_private = ca;
Mike Christiead0d9e72016-06-05 14:32:05 -0500515 bio_set_op_attrs(bio, op, REQ_SYNC|REQ_META|op_flags);
Kent Overstreet169ef1c2013-03-28 12:50:55 -0600516 bch_bio_map(bio, ca->disk_buckets);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700517
Kent Overstreet749b61d2013-11-23 23:11:25 -0800518 closure_bio_submit(bio, &ca->prio);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700519 closure_sync(cl);
520}
521
Kent Overstreetcafe5632013-03-23 16:11:31 -0700522void bch_prio_write(struct cache *ca)
523{
524 int i;
525 struct bucket *b;
526 struct closure cl;
527
528 closure_init_stack(&cl);
529
530 lockdep_assert_held(&ca->set->bucket_lock);
531
Kent Overstreetcafe5632013-03-23 16:11:31 -0700532 ca->disk_buckets->seq++;
533
534 atomic_long_add(ca->sb.bucket_size * prio_buckets(ca),
535 &ca->meta_sectors_written);
536
Kent Overstreet78365412013-12-17 01:29:34 -0800537 //pr_debug("free %zu, free_inc %zu, unused %zu", fifo_used(&ca->free),
538 // fifo_used(&ca->free_inc), fifo_used(&ca->unused));
Kent Overstreetcafe5632013-03-23 16:11:31 -0700539
540 for (i = prio_buckets(ca) - 1; i >= 0; --i) {
541 long bucket;
542 struct prio_set *p = ca->disk_buckets;
Kent Overstreetb1a67b02013-03-25 11:46:44 -0700543 struct bucket_disk *d = p->data;
544 struct bucket_disk *end = d + prios_per_bucket(ca);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700545
546 for (b = ca->buckets + i * prios_per_bucket(ca);
547 b < ca->buckets + ca->sb.nbuckets && d < end;
548 b++, d++) {
549 d->prio = cpu_to_le16(b->prio);
550 d->gen = b->gen;
551 }
552
553 p->next_bucket = ca->prio_buckets[i + 1];
Kent Overstreet81ab4192013-10-31 15:46:42 -0700554 p->magic = pset_magic(&ca->sb);
Kent Overstreet169ef1c2013-03-28 12:50:55 -0600555 p->csum = bch_crc64(&p->magic, bucket_bytes(ca) - 8);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700556
Kent Overstreet78365412013-12-17 01:29:34 -0800557 bucket = bch_bucket_alloc(ca, RESERVE_PRIO, true);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700558 BUG_ON(bucket == -1);
559
560 mutex_unlock(&ca->set->bucket_lock);
Mike Christiead0d9e72016-06-05 14:32:05 -0500561 prio_io(ca, bucket, REQ_OP_WRITE, 0);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700562 mutex_lock(&ca->set->bucket_lock);
563
564 ca->prio_buckets[i] = bucket;
565 atomic_dec_bug(&ca->buckets[bucket].pin);
566 }
567
568 mutex_unlock(&ca->set->bucket_lock);
569
570 bch_journal_meta(ca->set, &cl);
571 closure_sync(&cl);
572
573 mutex_lock(&ca->set->bucket_lock);
574
Kent Overstreetcafe5632013-03-23 16:11:31 -0700575 /*
576 * Don't want the old priorities to get garbage collected until after we
577 * finish writing the new ones, and they're journalled
578 */
Kent Overstreet2531d9ee2014-03-17 16:55:55 -0700579 for (i = 0; i < prio_buckets(ca); i++) {
580 if (ca->prio_last_buckets[i])
581 __bch_bucket_free(ca,
582 &ca->buckets[ca->prio_last_buckets[i]]);
583
Kent Overstreetcafe5632013-03-23 16:11:31 -0700584 ca->prio_last_buckets[i] = ca->prio_buckets[i];
Kent Overstreet2531d9ee2014-03-17 16:55:55 -0700585 }
Kent Overstreetcafe5632013-03-23 16:11:31 -0700586}
587
588static void prio_read(struct cache *ca, uint64_t bucket)
589{
590 struct prio_set *p = ca->disk_buckets;
591 struct bucket_disk *d = p->data + prios_per_bucket(ca), *end = d;
592 struct bucket *b;
593 unsigned bucket_nr = 0;
594
595 for (b = ca->buckets;
596 b < ca->buckets + ca->sb.nbuckets;
597 b++, d++) {
598 if (d == end) {
599 ca->prio_buckets[bucket_nr] = bucket;
600 ca->prio_last_buckets[bucket_nr] = bucket;
601 bucket_nr++;
602
Mike Christiead0d9e72016-06-05 14:32:05 -0500603 prio_io(ca, bucket, REQ_OP_READ, READ_SYNC);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700604
Kent Overstreet169ef1c2013-03-28 12:50:55 -0600605 if (p->csum != bch_crc64(&p->magic, bucket_bytes(ca) - 8))
Kent Overstreetcafe5632013-03-23 16:11:31 -0700606 pr_warn("bad csum reading priorities");
607
Kent Overstreet81ab4192013-10-31 15:46:42 -0700608 if (p->magic != pset_magic(&ca->sb))
Kent Overstreetcafe5632013-03-23 16:11:31 -0700609 pr_warn("bad magic reading priorities");
610
611 bucket = p->next_bucket;
612 d = p->data;
613 }
614
615 b->prio = le16_to_cpu(d->prio);
Kent Overstreet3a2fd9d2014-02-27 17:51:12 -0800616 b->gen = b->last_gc = d->gen;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700617 }
618}
619
620/* Bcache device */
621
622static int open_dev(struct block_device *b, fmode_t mode)
623{
624 struct bcache_device *d = b->bd_disk->private_data;
Kent Overstreetc4d951d2013-08-21 17:49:09 -0700625 if (test_bit(BCACHE_DEV_CLOSING, &d->flags))
Kent Overstreetcafe5632013-03-23 16:11:31 -0700626 return -ENXIO;
627
628 closure_get(&d->cl);
629 return 0;
630}
631
Emil Goode867e1162013-05-09 22:39:26 +0200632static void release_dev(struct gendisk *b, fmode_t mode)
Kent Overstreetcafe5632013-03-23 16:11:31 -0700633{
634 struct bcache_device *d = b->private_data;
635 closure_put(&d->cl);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700636}
637
638static int ioctl_dev(struct block_device *b, fmode_t mode,
639 unsigned int cmd, unsigned long arg)
640{
641 struct bcache_device *d = b->bd_disk->private_data;
642 return d->ioctl(d, mode, cmd, arg);
643}
644
645static const struct block_device_operations bcache_ops = {
646 .open = open_dev,
647 .release = release_dev,
648 .ioctl = ioctl_dev,
649 .owner = THIS_MODULE,
650};
651
652void bcache_device_stop(struct bcache_device *d)
653{
Kent Overstreetc4d951d2013-08-21 17:49:09 -0700654 if (!test_and_set_bit(BCACHE_DEV_CLOSING, &d->flags))
Kent Overstreetcafe5632013-03-23 16:11:31 -0700655 closure_queue(&d->cl);
656}
657
Kent Overstreetee668502013-02-01 07:29:41 -0800658static void bcache_device_unlink(struct bcache_device *d)
659{
Kent Overstreetc4d951d2013-08-21 17:49:09 -0700660 lockdep_assert_held(&bch_register_lock);
Kent Overstreetee668502013-02-01 07:29:41 -0800661
Kent Overstreetc4d951d2013-08-21 17:49:09 -0700662 if (d->c && !test_and_set_bit(BCACHE_DEV_UNLINK_DONE, &d->flags)) {
663 unsigned i;
664 struct cache *ca;
Kent Overstreetee668502013-02-01 07:29:41 -0800665
Kent Overstreetc4d951d2013-08-21 17:49:09 -0700666 sysfs_remove_link(&d->c->kobj, d->name);
667 sysfs_remove_link(&d->kobj, "cache");
668
669 for_each_cache(ca, d->c, i)
670 bd_unlink_disk_holder(ca->bdev, d->disk);
671 }
Kent Overstreetee668502013-02-01 07:29:41 -0800672}
673
674static void bcache_device_link(struct bcache_device *d, struct cache_set *c,
675 const char *name)
676{
677 unsigned i;
678 struct cache *ca;
679
680 for_each_cache(ca, d->c, i)
681 bd_link_disk_holder(ca->bdev, d->disk);
682
683 snprintf(d->name, BCACHEDEVNAME_SIZE,
684 "%s%u", name, d->id);
685
686 WARN(sysfs_create_link(&d->kobj, &c->kobj, "cache") ||
687 sysfs_create_link(&c->kobj, &d->kobj, d->name),
688 "Couldn't create device <-> cache set symlinks");
Zheng Liufecaee62015-11-29 17:19:32 -0800689
690 clear_bit(BCACHE_DEV_UNLINK_DONE, &d->flags);
Kent Overstreetee668502013-02-01 07:29:41 -0800691}
692
Kent Overstreetcafe5632013-03-23 16:11:31 -0700693static void bcache_device_detach(struct bcache_device *d)
694{
695 lockdep_assert_held(&bch_register_lock);
696
Kent Overstreetc4d951d2013-08-21 17:49:09 -0700697 if (test_bit(BCACHE_DEV_DETACHING, &d->flags)) {
Kent Overstreetcafe5632013-03-23 16:11:31 -0700698 struct uuid_entry *u = d->c->uuids + d->id;
699
700 SET_UUID_FLASH_ONLY(u, 0);
701 memcpy(u->uuid, invalid_uuid, 16);
702 u->invalidated = cpu_to_le32(get_seconds());
703 bch_uuid_write(d->c);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700704 }
705
Kent Overstreetc4d951d2013-08-21 17:49:09 -0700706 bcache_device_unlink(d);
Kent Overstreetee668502013-02-01 07:29:41 -0800707
Kent Overstreetcafe5632013-03-23 16:11:31 -0700708 d->c->devices[d->id] = NULL;
709 closure_put(&d->c->caching);
710 d->c = NULL;
711}
712
713static void bcache_device_attach(struct bcache_device *d, struct cache_set *c,
714 unsigned id)
715{
Kent Overstreetcafe5632013-03-23 16:11:31 -0700716 d->id = id;
717 d->c = c;
718 c->devices[id] = d;
719
720 closure_get(&c->caching);
721}
722
Kent Overstreetcafe5632013-03-23 16:11:31 -0700723static void bcache_device_free(struct bcache_device *d)
724{
725 lockdep_assert_held(&bch_register_lock);
726
727 pr_info("%s stopped", d->disk->disk_name);
728
729 if (d->c)
730 bcache_device_detach(d);
Kent Overstreetf59fce82013-05-15 00:11:26 -0700731 if (d->disk && d->disk->flags & GENHD_FL_UP)
Kent Overstreetcafe5632013-03-23 16:11:31 -0700732 del_gendisk(d->disk);
733 if (d->disk && d->disk->queue)
734 blk_cleanup_queue(d->disk->queue);
Kent Overstreet28935ab2013-07-31 01:12:02 -0700735 if (d->disk) {
736 ida_simple_remove(&bcache_minor, d->disk->first_minor);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700737 put_disk(d->disk);
Kent Overstreet28935ab2013-07-31 01:12:02 -0700738 }
Kent Overstreetcafe5632013-03-23 16:11:31 -0700739
Kent Overstreetcafe5632013-03-23 16:11:31 -0700740 if (d->bio_split)
741 bioset_free(d->bio_split);
Pekka Enberg958b4332015-06-30 14:59:30 -0700742 kvfree(d->full_dirty_stripes);
743 kvfree(d->stripe_sectors_dirty);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700744
745 closure_debug_destroy(&d->cl);
746}
747
Kent Overstreet279afba2013-06-05 06:21:07 -0700748static int bcache_device_init(struct bcache_device *d, unsigned block_size,
749 sector_t sectors)
Kent Overstreetcafe5632013-03-23 16:11:31 -0700750{
751 struct request_queue *q;
Kent Overstreet279afba2013-06-05 06:21:07 -0700752 size_t n;
Kent Overstreet28935ab2013-07-31 01:12:02 -0700753 int minor;
Kent Overstreet279afba2013-06-05 06:21:07 -0700754
Kent Overstreet2d679fc2013-08-17 02:13:15 -0700755 if (!d->stripe_size)
756 d->stripe_size = 1 << 31;
Kent Overstreet279afba2013-06-05 06:21:07 -0700757
Kent Overstreet2d679fc2013-08-17 02:13:15 -0700758 d->nr_stripes = DIV_ROUND_UP_ULL(sectors, d->stripe_size);
Kent Overstreet279afba2013-06-05 06:21:07 -0700759
Kent Overstreet48a915a2013-10-31 15:43:22 -0700760 if (!d->nr_stripes ||
761 d->nr_stripes > INT_MAX ||
762 d->nr_stripes > SIZE_MAX / sizeof(atomic_t)) {
Eric Wheeler90706092016-08-18 20:15:26 -0700763 pr_err("nr_stripes too large or invalid: %u (start sector beyond end of disk?)",
764 (unsigned)d->nr_stripes);
Kent Overstreet279afba2013-06-05 06:21:07 -0700765 return -ENOMEM;
Kent Overstreet48a915a2013-10-31 15:43:22 -0700766 }
Kent Overstreet279afba2013-06-05 06:21:07 -0700767
768 n = d->nr_stripes * sizeof(atomic_t);
769 d->stripe_sectors_dirty = n < PAGE_SIZE << 6
770 ? kzalloc(n, GFP_KERNEL)
771 : vzalloc(n);
772 if (!d->stripe_sectors_dirty)
773 return -ENOMEM;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700774
Kent Overstreet48a915a2013-10-31 15:43:22 -0700775 n = BITS_TO_LONGS(d->nr_stripes) * sizeof(unsigned long);
776 d->full_dirty_stripes = n < PAGE_SIZE << 6
777 ? kzalloc(n, GFP_KERNEL)
778 : vzalloc(n);
779 if (!d->full_dirty_stripes)
780 return -ENOMEM;
781
Kent Overstreet28935ab2013-07-31 01:12:02 -0700782 minor = ida_simple_get(&bcache_minor, 0, MINORMASK + 1, GFP_KERNEL);
783 if (minor < 0)
784 return minor;
785
Kent Overstreetcafe5632013-03-23 16:11:31 -0700786 if (!(d->bio_split = bioset_create(4, offsetof(struct bbio, bio))) ||
Kent Overstreet28935ab2013-07-31 01:12:02 -0700787 !(d->disk = alloc_disk(1))) {
788 ida_simple_remove(&bcache_minor, minor);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700789 return -ENOMEM;
Kent Overstreet28935ab2013-07-31 01:12:02 -0700790 }
Kent Overstreetcafe5632013-03-23 16:11:31 -0700791
Kent Overstreet279afba2013-06-05 06:21:07 -0700792 set_capacity(d->disk, sectors);
Kent Overstreet28935ab2013-07-31 01:12:02 -0700793 snprintf(d->disk->disk_name, DISK_NAME_LEN, "bcache%i", minor);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700794
795 d->disk->major = bcache_major;
Kent Overstreet28935ab2013-07-31 01:12:02 -0700796 d->disk->first_minor = minor;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700797 d->disk->fops = &bcache_ops;
798 d->disk->private_data = d;
799
Kent Overstreet28935ab2013-07-31 01:12:02 -0700800 q = blk_alloc_queue(GFP_KERNEL);
801 if (!q)
802 return -ENOMEM;
803
Kent Overstreetcafe5632013-03-23 16:11:31 -0700804 blk_queue_make_request(q, NULL);
805 d->disk->queue = q;
806 q->queuedata = d;
807 q->backing_dev_info.congested_data = d;
808 q->limits.max_hw_sectors = UINT_MAX;
809 q->limits.max_sectors = UINT_MAX;
810 q->limits.max_segment_size = UINT_MAX;
811 q->limits.max_segments = BIO_MAX_PAGES;
Jens Axboe2bb4cd52015-07-14 08:15:12 -0600812 blk_queue_max_discard_sectors(q, UINT_MAX);
Kent Overstreet90db6912014-02-10 17:26:40 -0800813 q->limits.discard_granularity = 512;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700814 q->limits.io_min = block_size;
815 q->limits.logical_block_size = block_size;
816 q->limits.physical_block_size = block_size;
817 set_bit(QUEUE_FLAG_NONROT, &d->disk->queue->queue_flags);
Mike Snitzerb277da02014-10-04 10:55:32 -0600818 clear_bit(QUEUE_FLAG_ADD_RANDOM, &d->disk->queue->queue_flags);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700819 set_bit(QUEUE_FLAG_DISCARD, &d->disk->queue->queue_flags);
820
Jens Axboe84b4ff92016-03-30 10:13:22 -0600821 blk_queue_write_cache(q, true, true);
Kent Overstreet54d12f22013-07-10 18:44:40 -0700822
Kent Overstreetcafe5632013-03-23 16:11:31 -0700823 return 0;
824}
825
826/* Cached device */
827
828static void calc_cached_dev_sectors(struct cache_set *c)
829{
830 uint64_t sectors = 0;
831 struct cached_dev *dc;
832
833 list_for_each_entry(dc, &c->cached_devs, list)
834 sectors += bdev_sectors(dc->bdev);
835
836 c->cached_dev_sectors = sectors;
837}
838
839void bch_cached_dev_run(struct cached_dev *dc)
840{
841 struct bcache_device *d = &dc->disk;
Gabriel de Perthuisab9e1402013-06-09 00:54:48 +0200842 char buf[SB_LABEL_SIZE + 1];
Gabriel de Perthuisa25c32b2013-06-07 23:27:01 +0200843 char *env[] = {
844 "DRIVER=bcache",
845 kasprintf(GFP_KERNEL, "CACHED_UUID=%pU", dc->sb.uuid),
Gabriel de Perthuisab9e1402013-06-09 00:54:48 +0200846 NULL,
847 NULL,
Gabriel de Perthuisa25c32b2013-06-07 23:27:01 +0200848 };
Kent Overstreetcafe5632013-03-23 16:11:31 -0700849
Gabriel de Perthuisab9e1402013-06-09 00:54:48 +0200850 memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
851 buf[SB_LABEL_SIZE] = '\0';
852 env[2] = kasprintf(GFP_KERNEL, "CACHED_LABEL=%s", buf);
853
Al Viro4d4d8572015-11-29 17:20:59 -0800854 if (atomic_xchg(&dc->running, 1)) {
855 kfree(env[1]);
856 kfree(env[2]);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700857 return;
Al Viro4d4d8572015-11-29 17:20:59 -0800858 }
Kent Overstreetcafe5632013-03-23 16:11:31 -0700859
860 if (!d->c &&
861 BDEV_STATE(&dc->sb) != BDEV_STATE_NONE) {
862 struct closure cl;
863 closure_init_stack(&cl);
864
865 SET_BDEV_STATE(&dc->sb, BDEV_STATE_STALE);
866 bch_write_bdev_super(dc, &cl);
867 closure_sync(&cl);
868 }
869
870 add_disk(d->disk);
Kent Overstreetee668502013-02-01 07:29:41 -0800871 bd_link_disk_holder(dc->bdev, dc->disk.disk);
Gabriel de Perthuisa25c32b2013-06-07 23:27:01 +0200872 /* won't show up in the uevent file, use udevadm monitor -e instead
873 * only class / kset properties are persistent */
Kent Overstreetcafe5632013-03-23 16:11:31 -0700874 kobject_uevent_env(&disk_to_dev(d->disk)->kobj, KOBJ_CHANGE, env);
Gabriel de Perthuisa25c32b2013-06-07 23:27:01 +0200875 kfree(env[1]);
Gabriel de Perthuisab9e1402013-06-09 00:54:48 +0200876 kfree(env[2]);
Gabriel de Perthuisa25c32b2013-06-07 23:27:01 +0200877
Kent Overstreetcafe5632013-03-23 16:11:31 -0700878 if (sysfs_create_link(&d->kobj, &disk_to_dev(d->disk)->kobj, "dev") ||
879 sysfs_create_link(&disk_to_dev(d->disk)->kobj, &d->kobj, "bcache"))
880 pr_debug("error creating sysfs link");
881}
882
883static void cached_dev_detach_finish(struct work_struct *w)
884{
885 struct cached_dev *dc = container_of(w, struct cached_dev, detach);
886 char buf[BDEVNAME_SIZE];
887 struct closure cl;
888 closure_init_stack(&cl);
889
Kent Overstreetc4d951d2013-08-21 17:49:09 -0700890 BUG_ON(!test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags));
Kent Overstreetcafe5632013-03-23 16:11:31 -0700891 BUG_ON(atomic_read(&dc->count));
892
Kent Overstreetcafe5632013-03-23 16:11:31 -0700893 mutex_lock(&bch_register_lock);
894
895 memset(&dc->sb.set_uuid, 0, 16);
896 SET_BDEV_STATE(&dc->sb, BDEV_STATE_NONE);
897
898 bch_write_bdev_super(dc, &cl);
899 closure_sync(&cl);
900
901 bcache_device_detach(&dc->disk);
902 list_move(&dc->list, &uncached_devices);
903
Kent Overstreetc4d951d2013-08-21 17:49:09 -0700904 clear_bit(BCACHE_DEV_DETACHING, &dc->disk.flags);
Kent Overstreet5b1016e2014-03-19 17:49:37 -0700905 clear_bit(BCACHE_DEV_UNLINK_DONE, &dc->disk.flags);
Kent Overstreetc4d951d2013-08-21 17:49:09 -0700906
Kent Overstreetcafe5632013-03-23 16:11:31 -0700907 mutex_unlock(&bch_register_lock);
908
909 pr_info("Caching disabled for %s", bdevname(dc->bdev, buf));
910
911 /* Drop ref we took in cached_dev_detach() */
912 closure_put(&dc->disk.cl);
913}
914
915void bch_cached_dev_detach(struct cached_dev *dc)
916{
917 lockdep_assert_held(&bch_register_lock);
918
Kent Overstreetc4d951d2013-08-21 17:49:09 -0700919 if (test_bit(BCACHE_DEV_CLOSING, &dc->disk.flags))
Kent Overstreetcafe5632013-03-23 16:11:31 -0700920 return;
921
Kent Overstreetc4d951d2013-08-21 17:49:09 -0700922 if (test_and_set_bit(BCACHE_DEV_DETACHING, &dc->disk.flags))
Kent Overstreetcafe5632013-03-23 16:11:31 -0700923 return;
924
925 /*
926 * Block the device from being closed and freed until we're finished
927 * detaching
928 */
929 closure_get(&dc->disk.cl);
930
931 bch_writeback_queue(dc);
932 cached_dev_put(dc);
933}
934
935int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c)
936{
937 uint32_t rtime = cpu_to_le32(get_seconds());
938 struct uuid_entry *u;
939 char buf[BDEVNAME_SIZE];
940
941 bdevname(dc->bdev, buf);
942
943 if (memcmp(dc->sb.set_uuid, c->sb.set_uuid, 16))
944 return -ENOENT;
945
946 if (dc->disk.c) {
947 pr_err("Can't attach %s: already attached", buf);
948 return -EINVAL;
949 }
950
951 if (test_bit(CACHE_SET_STOPPING, &c->flags)) {
952 pr_err("Can't attach %s: shutting down", buf);
953 return -EINVAL;
954 }
955
956 if (dc->sb.block_size < c->sb.block_size) {
957 /* Will die */
Kent Overstreetb1a67b02013-03-25 11:46:44 -0700958 pr_err("Couldn't attach %s: block size less than set's block size",
959 buf);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700960 return -EINVAL;
961 }
962
963 u = uuid_find(c, dc->sb.uuid);
964
965 if (u &&
966 (BDEV_STATE(&dc->sb) == BDEV_STATE_STALE ||
967 BDEV_STATE(&dc->sb) == BDEV_STATE_NONE)) {
968 memcpy(u->uuid, invalid_uuid, 16);
969 u->invalidated = cpu_to_le32(get_seconds());
970 u = NULL;
971 }
972
973 if (!u) {
974 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
975 pr_err("Couldn't find uuid for %s in set", buf);
976 return -ENOENT;
977 }
978
979 u = uuid_find_empty(c);
980 if (!u) {
981 pr_err("Not caching %s, no room for UUID", buf);
982 return -EINVAL;
983 }
984 }
985
986 /* Deadlocks since we're called via sysfs...
987 sysfs_remove_file(&dc->kobj, &sysfs_attach);
988 */
989
Kent Overstreet169ef1c2013-03-28 12:50:55 -0600990 if (bch_is_zero(u->uuid, 16)) {
Kent Overstreetcafe5632013-03-23 16:11:31 -0700991 struct closure cl;
992 closure_init_stack(&cl);
993
994 memcpy(u->uuid, dc->sb.uuid, 16);
995 memcpy(u->label, dc->sb.label, SB_LABEL_SIZE);
996 u->first_reg = u->last_reg = rtime;
997 bch_uuid_write(c);
998
999 memcpy(dc->sb.set_uuid, c->sb.set_uuid, 16);
1000 SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN);
1001
1002 bch_write_bdev_super(dc, &cl);
1003 closure_sync(&cl);
1004 } else {
1005 u->last_reg = rtime;
1006 bch_uuid_write(c);
1007 }
1008
1009 bcache_device_attach(&dc->disk, c, u - c->uuids);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001010 list_move(&dc->list, &c->cached_devs);
1011 calc_cached_dev_sectors(c);
1012
1013 smp_wmb();
1014 /*
1015 * dc->c must be set before dc->count != 0 - paired with the mb in
1016 * cached_dev_get()
1017 */
1018 atomic_set(&dc->count, 1);
1019
Eric Wheeler07cc6ef82016-02-26 14:39:06 -08001020 /* Block writeback thread, but spawn it */
1021 down_write(&dc->writeback_lock);
1022 if (bch_cached_dev_writeback_start(dc)) {
1023 up_write(&dc->writeback_lock);
Slava Pestov9e5c3532014-05-01 13:48:57 -07001024 return -ENOMEM;
Eric Wheeler07cc6ef82016-02-26 14:39:06 -08001025 }
Slava Pestov9e5c3532014-05-01 13:48:57 -07001026
Kent Overstreetcafe5632013-03-23 16:11:31 -07001027 if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
Kent Overstreet444fc0b2013-05-11 17:07:26 -07001028 bch_sectors_dirty_init(dc);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001029 atomic_set(&dc->has_dirty, 1);
1030 atomic_inc(&dc->count);
1031 bch_writeback_queue(dc);
1032 }
1033
1034 bch_cached_dev_run(dc);
Kent Overstreetee668502013-02-01 07:29:41 -08001035 bcache_device_link(&dc->disk, c, "bdev");
Kent Overstreetcafe5632013-03-23 16:11:31 -07001036
Eric Wheeler07cc6ef82016-02-26 14:39:06 -08001037 /* Allow the writeback thread to proceed */
1038 up_write(&dc->writeback_lock);
1039
Kent Overstreetcafe5632013-03-23 16:11:31 -07001040 pr_info("Caching %s as %s on set %pU",
1041 bdevname(dc->bdev, buf), dc->disk.disk->disk_name,
1042 dc->disk.c->sb.set_uuid);
1043 return 0;
1044}
1045
1046void bch_cached_dev_release(struct kobject *kobj)
1047{
1048 struct cached_dev *dc = container_of(kobj, struct cached_dev,
1049 disk.kobj);
1050 kfree(dc);
1051 module_put(THIS_MODULE);
1052}
1053
1054static void cached_dev_free(struct closure *cl)
1055{
1056 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1057
1058 cancel_delayed_work_sync(&dc->writeback_rate_update);
Slava Pestova664d0f2014-05-20 12:20:28 -07001059 if (!IS_ERR_OR_NULL(dc->writeback_thread))
1060 kthread_stop(dc->writeback_thread);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001061
1062 mutex_lock(&bch_register_lock);
1063
Kent Overstreetf59fce82013-05-15 00:11:26 -07001064 if (atomic_read(&dc->running))
1065 bd_unlink_disk_holder(dc->bdev, dc->disk.disk);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001066 bcache_device_free(&dc->disk);
1067 list_del(&dc->list);
1068
1069 mutex_unlock(&bch_register_lock);
1070
Kent Overstreet0781c872014-07-07 13:03:36 -07001071 if (!IS_ERR_OR_NULL(dc->bdev))
Kent Overstreetcafe5632013-03-23 16:11:31 -07001072 blkdev_put(dc->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001073
1074 wake_up(&unregister_wait);
1075
1076 kobject_put(&dc->disk.kobj);
1077}
1078
1079static void cached_dev_flush(struct closure *cl)
1080{
1081 struct cached_dev *dc = container_of(cl, struct cached_dev, disk.cl);
1082 struct bcache_device *d = &dc->disk;
1083
Kent Overstreetc9502ea2013-07-10 21:25:02 -07001084 mutex_lock(&bch_register_lock);
Kent Overstreetc4d951d2013-08-21 17:49:09 -07001085 bcache_device_unlink(d);
Kent Overstreetc9502ea2013-07-10 21:25:02 -07001086 mutex_unlock(&bch_register_lock);
1087
Kent Overstreetcafe5632013-03-23 16:11:31 -07001088 bch_cache_accounting_destroy(&dc->accounting);
1089 kobject_del(&d->kobj);
1090
1091 continue_at(cl, cached_dev_free, system_wq);
1092}
1093
1094static int cached_dev_init(struct cached_dev *dc, unsigned block_size)
1095{
Kent Overstreetf59fce82013-05-15 00:11:26 -07001096 int ret;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001097 struct io *io;
Kent Overstreetf59fce82013-05-15 00:11:26 -07001098 struct request_queue *q = bdev_get_queue(dc->bdev);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001099
1100 __module_get(THIS_MODULE);
1101 INIT_LIST_HEAD(&dc->list);
Kent Overstreetf59fce82013-05-15 00:11:26 -07001102 closure_init(&dc->disk.cl, NULL);
1103 set_closure_fn(&dc->disk.cl, cached_dev_flush, system_wq);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001104 kobject_init(&dc->disk.kobj, &bch_cached_dev_ktype);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001105 INIT_WORK(&dc->detach, cached_dev_detach_finish);
Kent Overstreetcb7a5832013-12-16 15:27:25 -08001106 sema_init(&dc->sb_write_mutex, 1);
Kent Overstreetf59fce82013-05-15 00:11:26 -07001107 INIT_LIST_HEAD(&dc->io_lru);
1108 spin_lock_init(&dc->io_lock);
1109 bch_cache_accounting_init(&dc->accounting, &dc->disk.cl);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001110
Kent Overstreetcafe5632013-03-23 16:11:31 -07001111 dc->sequential_cutoff = 4 << 20;
1112
Kent Overstreetcafe5632013-03-23 16:11:31 -07001113 for (io = dc->io; io < dc->io + RECENT_IO; io++) {
1114 list_add(&io->lru, &dc->io_lru);
1115 hlist_add_head(&io->hash, dc->io_hash + RECENT_IO);
1116 }
1117
Kent Overstreetc78afc62013-07-11 22:39:53 -07001118 dc->disk.stripe_size = q->limits.io_opt >> 9;
1119
1120 if (dc->disk.stripe_size)
1121 dc->partial_stripes_expensive =
1122 q->limits.raid_partial_stripes_expensive;
1123
Kent Overstreet279afba2013-06-05 06:21:07 -07001124 ret = bcache_device_init(&dc->disk, block_size,
1125 dc->bdev->bd_part->nr_sects - dc->sb.data_offset);
Kent Overstreetf59fce82013-05-15 00:11:26 -07001126 if (ret)
1127 return ret;
1128
1129 set_capacity(dc->disk.disk,
1130 dc->bdev->bd_part->nr_sects - dc->sb.data_offset);
1131
1132 dc->disk.disk->queue->backing_dev_info.ra_pages =
1133 max(dc->disk.disk->queue->backing_dev_info.ra_pages,
1134 q->backing_dev_info.ra_pages);
1135
1136 bch_cached_dev_request_init(dc);
1137 bch_cached_dev_writeback_init(dc);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001138 return 0;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001139}
1140
1141/* Cached device - bcache superblock */
1142
Kent Overstreetf59fce82013-05-15 00:11:26 -07001143static void register_bdev(struct cache_sb *sb, struct page *sb_page,
Kent Overstreetcafe5632013-03-23 16:11:31 -07001144 struct block_device *bdev,
1145 struct cached_dev *dc)
1146{
1147 char name[BDEVNAME_SIZE];
1148 const char *err = "cannot allocate memory";
Kent Overstreetcafe5632013-03-23 16:11:31 -07001149 struct cache_set *c;
1150
Kent Overstreetcafe5632013-03-23 16:11:31 -07001151 memcpy(&dc->sb, sb, sizeof(struct cache_sb));
Kent Overstreetcafe5632013-03-23 16:11:31 -07001152 dc->bdev = bdev;
1153 dc->bdev->bd_holder = dc;
1154
Kent Overstreetf59fce82013-05-15 00:11:26 -07001155 bio_init(&dc->sb_bio);
1156 dc->sb_bio.bi_max_vecs = 1;
1157 dc->sb_bio.bi_io_vec = dc->sb_bio.bi_inline_vecs;
1158 dc->sb_bio.bi_io_vec[0].bv_page = sb_page;
1159 get_page(sb_page);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001160
Kent Overstreetf59fce82013-05-15 00:11:26 -07001161 if (cached_dev_init(dc, sb->block_size << 9))
1162 goto err;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001163
1164 err = "error creating kobject";
1165 if (kobject_add(&dc->disk.kobj, &part_to_dev(bdev->bd_part)->kobj,
1166 "bcache"))
1167 goto err;
1168 if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj))
1169 goto err;
1170
Kent Overstreetf59fce82013-05-15 00:11:26 -07001171 pr_info("registered backing device %s", bdevname(bdev, name));
1172
Kent Overstreetcafe5632013-03-23 16:11:31 -07001173 list_add(&dc->list, &uncached_devices);
1174 list_for_each_entry(c, &bch_cache_sets, list)
1175 bch_cached_dev_attach(dc, c);
1176
1177 if (BDEV_STATE(&dc->sb) == BDEV_STATE_NONE ||
1178 BDEV_STATE(&dc->sb) == BDEV_STATE_STALE)
1179 bch_cached_dev_run(dc);
1180
Kent Overstreetf59fce82013-05-15 00:11:26 -07001181 return;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001182err:
Kent Overstreetcafe5632013-03-23 16:11:31 -07001183 pr_notice("error opening %s: %s", bdevname(bdev, name), err);
Kent Overstreetf59fce82013-05-15 00:11:26 -07001184 bcache_device_stop(&dc->disk);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001185}
1186
1187/* Flash only volumes */
1188
1189void bch_flash_dev_release(struct kobject *kobj)
1190{
1191 struct bcache_device *d = container_of(kobj, struct bcache_device,
1192 kobj);
1193 kfree(d);
1194}
1195
1196static void flash_dev_free(struct closure *cl)
1197{
1198 struct bcache_device *d = container_of(cl, struct bcache_device, cl);
Slava Pestove5112202014-04-29 15:39:27 -07001199 mutex_lock(&bch_register_lock);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001200 bcache_device_free(d);
Slava Pestove5112202014-04-29 15:39:27 -07001201 mutex_unlock(&bch_register_lock);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001202 kobject_put(&d->kobj);
1203}
1204
1205static void flash_dev_flush(struct closure *cl)
1206{
1207 struct bcache_device *d = container_of(cl, struct bcache_device, cl);
1208
Slava Pestove5112202014-04-29 15:39:27 -07001209 mutex_lock(&bch_register_lock);
Kent Overstreetee668502013-02-01 07:29:41 -08001210 bcache_device_unlink(d);
Slava Pestove5112202014-04-29 15:39:27 -07001211 mutex_unlock(&bch_register_lock);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001212 kobject_del(&d->kobj);
1213 continue_at(cl, flash_dev_free, system_wq);
1214}
1215
1216static int flash_dev_run(struct cache_set *c, struct uuid_entry *u)
1217{
1218 struct bcache_device *d = kzalloc(sizeof(struct bcache_device),
1219 GFP_KERNEL);
1220 if (!d)
1221 return -ENOMEM;
1222
1223 closure_init(&d->cl, NULL);
1224 set_closure_fn(&d->cl, flash_dev_flush, system_wq);
1225
1226 kobject_init(&d->kobj, &bch_flash_dev_ktype);
1227
Kent Overstreet279afba2013-06-05 06:21:07 -07001228 if (bcache_device_init(d, block_bytes(c), u->sectors))
Kent Overstreetcafe5632013-03-23 16:11:31 -07001229 goto err;
1230
1231 bcache_device_attach(d, c, u - c->uuids);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001232 bch_flash_dev_request_init(d);
1233 add_disk(d->disk);
1234
1235 if (kobject_add(&d->kobj, &disk_to_dev(d->disk)->kobj, "bcache"))
1236 goto err;
1237
1238 bcache_device_link(d, c, "volume");
1239
1240 return 0;
1241err:
1242 kobject_put(&d->kobj);
1243 return -ENOMEM;
1244}
1245
1246static int flash_devs_run(struct cache_set *c)
1247{
1248 int ret = 0;
1249 struct uuid_entry *u;
1250
1251 for (u = c->uuids;
1252 u < c->uuids + c->nr_uuids && !ret;
1253 u++)
1254 if (UUID_FLASH_ONLY(u))
1255 ret = flash_dev_run(c, u);
1256
1257 return ret;
1258}
1259
1260int bch_flash_dev_create(struct cache_set *c, uint64_t size)
1261{
1262 struct uuid_entry *u;
1263
1264 if (test_bit(CACHE_SET_STOPPING, &c->flags))
1265 return -EINTR;
1266
Slava Pestovbf0c55c2014-07-11 12:17:41 -07001267 if (!test_bit(CACHE_SET_RUNNING, &c->flags))
1268 return -EPERM;
1269
Kent Overstreetcafe5632013-03-23 16:11:31 -07001270 u = uuid_find_empty(c);
1271 if (!u) {
1272 pr_err("Can't create volume, no room for UUID");
1273 return -EINVAL;
1274 }
1275
1276 get_random_bytes(u->uuid, 16);
1277 memset(u->label, 0, 32);
1278 u->first_reg = u->last_reg = cpu_to_le32(get_seconds());
1279
1280 SET_UUID_FLASH_ONLY(u, 1);
1281 u->sectors = size >> 9;
1282
1283 bch_uuid_write(c);
1284
1285 return flash_dev_run(c, u);
1286}
1287
1288/* Cache set */
1289
1290__printf(2, 3)
1291bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...)
1292{
1293 va_list args;
1294
Kent Overstreet77c320e2013-07-11 19:42:51 -07001295 if (c->on_error != ON_ERROR_PANIC &&
1296 test_bit(CACHE_SET_STOPPING, &c->flags))
Kent Overstreetcafe5632013-03-23 16:11:31 -07001297 return false;
1298
1299 /* XXX: we can be called from atomic context
1300 acquire_console_sem();
1301 */
1302
1303 printk(KERN_ERR "bcache: error on %pU: ", c->sb.set_uuid);
1304
1305 va_start(args, fmt);
1306 vprintk(fmt, args);
1307 va_end(args);
1308
1309 printk(", disabling caching\n");
1310
Kent Overstreet77c320e2013-07-11 19:42:51 -07001311 if (c->on_error == ON_ERROR_PANIC)
1312 panic("panic forced after error\n");
1313
Kent Overstreetcafe5632013-03-23 16:11:31 -07001314 bch_cache_set_unregister(c);
1315 return true;
1316}
1317
1318void bch_cache_set_release(struct kobject *kobj)
1319{
1320 struct cache_set *c = container_of(kobj, struct cache_set, kobj);
1321 kfree(c);
1322 module_put(THIS_MODULE);
1323}
1324
1325static void cache_set_free(struct closure *cl)
1326{
1327 struct cache_set *c = container_of(cl, struct cache_set, cl);
1328 struct cache *ca;
1329 unsigned i;
1330
1331 if (!IS_ERR_OR_NULL(c->debug))
1332 debugfs_remove(c->debug);
1333
1334 bch_open_buckets_free(c);
1335 bch_btree_cache_free(c);
1336 bch_journal_free(c);
1337
1338 for_each_cache(ca, c, i)
Slava Pestovc9a78332014-06-19 15:05:59 -07001339 if (ca) {
1340 ca->set = NULL;
1341 c->cache[ca->sb.nr_this_dev] = NULL;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001342 kobject_put(&ca->kobj);
Slava Pestovc9a78332014-06-19 15:05:59 -07001343 }
Kent Overstreetcafe5632013-03-23 16:11:31 -07001344
Kent Overstreet67539e82013-09-10 22:53:34 -07001345 bch_bset_sort_state_free(&c->sort);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001346 free_pages((unsigned long) c->uuids, ilog2(bucket_pages(c)));
Kent Overstreetcafe5632013-03-23 16:11:31 -07001347
Nicholas Swensonda415a02014-01-09 16:03:04 -08001348 if (c->moving_gc_wq)
1349 destroy_workqueue(c->moving_gc_wq);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001350 if (c->bio_split)
1351 bioset_free(c->bio_split);
Kent Overstreet57943512013-04-25 13:58:35 -07001352 if (c->fill_iter)
1353 mempool_destroy(c->fill_iter);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001354 if (c->bio_meta)
1355 mempool_destroy(c->bio_meta);
1356 if (c->search)
1357 mempool_destroy(c->search);
1358 kfree(c->devices);
1359
1360 mutex_lock(&bch_register_lock);
1361 list_del(&c->list);
1362 mutex_unlock(&bch_register_lock);
1363
1364 pr_info("Cache set %pU unregistered", c->sb.set_uuid);
1365 wake_up(&unregister_wait);
1366
1367 closure_debug_destroy(&c->cl);
1368 kobject_put(&c->kobj);
1369}
1370
1371static void cache_set_flush(struct closure *cl)
1372{
1373 struct cache_set *c = container_of(cl, struct cache_set, caching);
Kent Overstreet79826c32013-07-10 18:31:58 -07001374 struct cache *ca;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001375 struct btree *b;
Kent Overstreet79826c32013-07-10 18:31:58 -07001376 unsigned i;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001377
Eric Wheelerf8b11262016-03-07 15:17:50 -08001378 if (!c)
1379 closure_return(cl);
1380
Kent Overstreetcafe5632013-03-23 16:11:31 -07001381 bch_cache_accounting_destroy(&c->accounting);
1382
1383 kobject_put(&c->internal);
1384 kobject_del(&c->kobj);
1385
Kent Overstreet72a44512013-10-24 17:19:26 -07001386 if (c->gc_thread)
1387 kthread_stop(c->gc_thread);
1388
Kent Overstreetcafe5632013-03-23 16:11:31 -07001389 if (!IS_ERR_OR_NULL(c->root))
1390 list_add(&c->root->list, &c->btree_cache);
1391
1392 /* Should skip this if we're unregistering because of an error */
Kent Overstreet2a285682014-03-04 16:42:42 -08001393 list_for_each_entry(b, &c->btree_cache, list) {
1394 mutex_lock(&b->write_lock);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001395 if (btree_node_dirty(b))
Kent Overstreet2a285682014-03-04 16:42:42 -08001396 __bch_btree_node_write(b, NULL);
1397 mutex_unlock(&b->write_lock);
1398 }
Kent Overstreetcafe5632013-03-23 16:11:31 -07001399
Kent Overstreet79826c32013-07-10 18:31:58 -07001400 for_each_cache(ca, c, i)
1401 if (ca->alloc_thread)
1402 kthread_stop(ca->alloc_thread);
1403
Kent Overstreet5b1016e2014-03-19 17:49:37 -07001404 if (c->journal.cur) {
1405 cancel_delayed_work_sync(&c->journal.work);
1406 /* flush last journal entry if needed */
1407 c->journal.work.work.func(&c->journal.work.work);
1408 }
Kent Overstreetdabb4432014-02-19 19:48:26 -08001409
Kent Overstreetcafe5632013-03-23 16:11:31 -07001410 closure_return(cl);
1411}
1412
1413static void __cache_set_unregister(struct closure *cl)
1414{
1415 struct cache_set *c = container_of(cl, struct cache_set, caching);
Kent Overstreet5caa52a2013-07-10 21:03:25 -07001416 struct cached_dev *dc;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001417 size_t i;
1418
1419 mutex_lock(&bch_register_lock);
1420
Kent Overstreetcafe5632013-03-23 16:11:31 -07001421 for (i = 0; i < c->nr_uuids; i++)
Kent Overstreet5caa52a2013-07-10 21:03:25 -07001422 if (c->devices[i]) {
1423 if (!UUID_FLASH_ONLY(&c->uuids[i]) &&
1424 test_bit(CACHE_SET_UNREGISTERING, &c->flags)) {
1425 dc = container_of(c->devices[i],
1426 struct cached_dev, disk);
1427 bch_cached_dev_detach(dc);
1428 } else {
1429 bcache_device_stop(c->devices[i]);
1430 }
1431 }
Kent Overstreetcafe5632013-03-23 16:11:31 -07001432
1433 mutex_unlock(&bch_register_lock);
1434
1435 continue_at(cl, cache_set_flush, system_wq);
1436}
1437
1438void bch_cache_set_stop(struct cache_set *c)
1439{
1440 if (!test_and_set_bit(CACHE_SET_STOPPING, &c->flags))
1441 closure_queue(&c->caching);
1442}
1443
1444void bch_cache_set_unregister(struct cache_set *c)
1445{
1446 set_bit(CACHE_SET_UNREGISTERING, &c->flags);
1447 bch_cache_set_stop(c);
1448}
1449
1450#define alloc_bucket_pages(gfp, c) \
1451 ((void *) __get_free_pages(__GFP_ZERO|gfp, ilog2(bucket_pages(c))))
1452
1453struct cache_set *bch_cache_set_alloc(struct cache_sb *sb)
1454{
1455 int iter_size;
1456 struct cache_set *c = kzalloc(sizeof(struct cache_set), GFP_KERNEL);
1457 if (!c)
1458 return NULL;
1459
1460 __module_get(THIS_MODULE);
1461 closure_init(&c->cl, NULL);
1462 set_closure_fn(&c->cl, cache_set_free, system_wq);
1463
1464 closure_init(&c->caching, &c->cl);
1465 set_closure_fn(&c->caching, __cache_set_unregister, system_wq);
1466
1467 /* Maybe create continue_at_noreturn() and use it here? */
1468 closure_set_stopped(&c->cl);
1469 closure_put(&c->cl);
1470
1471 kobject_init(&c->kobj, &bch_cache_set_ktype);
1472 kobject_init(&c->internal, &bch_cache_set_internal_ktype);
1473
1474 bch_cache_accounting_init(&c->accounting, &c->cl);
1475
1476 memcpy(c->sb.set_uuid, sb->set_uuid, 16);
1477 c->sb.block_size = sb->block_size;
1478 c->sb.bucket_size = sb->bucket_size;
1479 c->sb.nr_in_set = sb->nr_in_set;
1480 c->sb.last_mount = sb->last_mount;
1481 c->bucket_bits = ilog2(sb->bucket_size);
1482 c->block_bits = ilog2(sb->block_size);
1483 c->nr_uuids = bucket_bytes(c) / sizeof(struct uuid_entry);
1484
Kent Overstreetee811282013-12-17 23:49:49 -08001485 c->btree_pages = bucket_pages(c);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001486 if (c->btree_pages > BTREE_MAX_PAGES)
1487 c->btree_pages = max_t(int, c->btree_pages / 4,
1488 BTREE_MAX_PAGES);
1489
Kent Overstreetcb7a5832013-12-16 15:27:25 -08001490 sema_init(&c->sb_write_mutex, 1);
Kent Overstreete8e1d462013-07-24 17:27:07 -07001491 mutex_init(&c->bucket_lock);
Kent Overstreet0a63b662014-03-17 17:15:53 -07001492 init_waitqueue_head(&c->btree_cache_wait);
Kent Overstreet35fcd842013-07-24 17:29:09 -07001493 init_waitqueue_head(&c->bucket_wait);
Kent Overstreetcb7a5832013-12-16 15:27:25 -08001494 sema_init(&c->uuid_write_mutex, 1);
Kent Overstreet65d22e92013-07-31 00:03:54 -07001495
Kent Overstreet65d22e92013-07-31 00:03:54 -07001496 spin_lock_init(&c->btree_gc_time.lock);
1497 spin_lock_init(&c->btree_split_time.lock);
1498 spin_lock_init(&c->btree_read_time.lock);
Kent Overstreete8e1d462013-07-24 17:27:07 -07001499
Kent Overstreetcafe5632013-03-23 16:11:31 -07001500 bch_moving_init_cache_set(c);
1501
1502 INIT_LIST_HEAD(&c->list);
1503 INIT_LIST_HEAD(&c->cached_devs);
1504 INIT_LIST_HEAD(&c->btree_cache);
1505 INIT_LIST_HEAD(&c->btree_cache_freeable);
1506 INIT_LIST_HEAD(&c->btree_cache_freed);
1507 INIT_LIST_HEAD(&c->data_buckets);
1508
1509 c->search = mempool_create_slab_pool(32, bch_search_cache);
1510 if (!c->search)
1511 goto err;
1512
1513 iter_size = (sb->bucket_size / sb->block_size + 1) *
1514 sizeof(struct btree_iter_set);
1515
1516 if (!(c->devices = kzalloc(c->nr_uuids * sizeof(void *), GFP_KERNEL)) ||
1517 !(c->bio_meta = mempool_create_kmalloc_pool(2,
1518 sizeof(struct bbio) + sizeof(struct bio_vec) *
1519 bucket_pages(c))) ||
Kent Overstreet57943512013-04-25 13:58:35 -07001520 !(c->fill_iter = mempool_create_kmalloc_pool(1, iter_size)) ||
Kent Overstreetcafe5632013-03-23 16:11:31 -07001521 !(c->bio_split = bioset_create(4, offsetof(struct bbio, bio))) ||
Kent Overstreetcafe5632013-03-23 16:11:31 -07001522 !(c->uuids = alloc_bucket_pages(GFP_KERNEL, c)) ||
Bhaktipriya Shridhar81baf902016-06-08 01:57:19 +05301523 !(c->moving_gc_wq = alloc_workqueue("bcache_gc",
1524 WQ_MEM_RECLAIM, 0)) ||
Kent Overstreetcafe5632013-03-23 16:11:31 -07001525 bch_journal_alloc(c) ||
1526 bch_btree_cache_alloc(c) ||
Kent Overstreet67539e82013-09-10 22:53:34 -07001527 bch_open_buckets_alloc(c) ||
1528 bch_bset_sort_state_init(&c->sort, ilog2(c->btree_pages)))
Kent Overstreetcafe5632013-03-23 16:11:31 -07001529 goto err;
1530
Kent Overstreetcafe5632013-03-23 16:11:31 -07001531 c->congested_read_threshold_us = 2000;
1532 c->congested_write_threshold_us = 20000;
1533 c->error_limit = 8 << IO_ERROR_SHIFT;
1534
1535 return c;
1536err:
1537 bch_cache_set_unregister(c);
1538 return NULL;
1539}
1540
1541static void run_cache_set(struct cache_set *c)
1542{
1543 const char *err = "cannot allocate memory";
1544 struct cached_dev *dc, *t;
1545 struct cache *ca;
Kent Overstreetc18536a2013-07-24 17:44:17 -07001546 struct closure cl;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001547 unsigned i;
1548
Kent Overstreetc18536a2013-07-24 17:44:17 -07001549 closure_init_stack(&cl);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001550
1551 for_each_cache(ca, c, i)
1552 c->nbuckets += ca->sb.nbuckets;
1553
1554 if (CACHE_SYNC(&c->sb)) {
1555 LIST_HEAD(journal);
1556 struct bkey *k;
1557 struct jset *j;
1558
1559 err = "cannot allocate memory for journal";
Kent Overstreetc18536a2013-07-24 17:44:17 -07001560 if (bch_journal_read(c, &journal))
Kent Overstreetcafe5632013-03-23 16:11:31 -07001561 goto err;
1562
1563 pr_debug("btree_journal_read() done");
1564
1565 err = "no journal entries found";
1566 if (list_empty(&journal))
1567 goto err;
1568
1569 j = &list_entry(journal.prev, struct journal_replay, list)->j;
1570
1571 err = "IO error reading priorities";
1572 for_each_cache(ca, c, i)
1573 prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev]);
1574
1575 /*
1576 * If prio_read() fails it'll call cache_set_error and we'll
1577 * tear everything down right away, but if we perhaps checked
1578 * sooner we could avoid journal replay.
1579 */
1580
1581 k = &j->btree_root;
1582
1583 err = "bad btree root";
Kent Overstreet65d45232013-12-20 17:22:05 -08001584 if (__bch_btree_ptr_invalid(c, k))
Kent Overstreetcafe5632013-03-23 16:11:31 -07001585 goto err;
1586
1587 err = "error reading btree root";
Slava Pestov2452cc82014-07-12 00:22:53 -07001588 c->root = bch_btree_node_get(c, NULL, k, j->btree_level, true, NULL);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001589 if (IS_ERR_OR_NULL(c->root))
1590 goto err;
1591
1592 list_del_init(&c->root->list);
1593 rw_unlock(true, c->root);
1594
Kent Overstreetc18536a2013-07-24 17:44:17 -07001595 err = uuid_read(c, j, &cl);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001596 if (err)
1597 goto err;
1598
1599 err = "error in recovery";
Kent Overstreetc18536a2013-07-24 17:44:17 -07001600 if (bch_btree_check(c))
Kent Overstreetcafe5632013-03-23 16:11:31 -07001601 goto err;
1602
1603 bch_journal_mark(c, &journal);
Kent Overstreet2531d9ee2014-03-17 16:55:55 -07001604 bch_initial_gc_finish(c);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001605 pr_debug("btree_check() done");
1606
1607 /*
1608 * bcache_journal_next() can't happen sooner, or
1609 * btree_gc_finish() will give spurious errors about last_gc >
1610 * gc_gen - this is a hack but oh well.
1611 */
1612 bch_journal_next(&c->journal);
1613
Kent Overstreet119ba0f2013-04-24 19:01:12 -07001614 err = "error starting allocator thread";
Kent Overstreetcafe5632013-03-23 16:11:31 -07001615 for_each_cache(ca, c, i)
Kent Overstreet119ba0f2013-04-24 19:01:12 -07001616 if (bch_cache_allocator_start(ca))
1617 goto err;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001618
1619 /*
1620 * First place it's safe to allocate: btree_check() and
1621 * btree_gc_finish() have to run before we have buckets to
1622 * allocate, and bch_bucket_alloc_set() might cause a journal
1623 * entry to be written so bcache_journal_next() has to be called
1624 * first.
1625 *
1626 * If the uuids were in the old format we have to rewrite them
1627 * before the next journal entry is written:
1628 */
1629 if (j->version < BCACHE_JSET_VERSION_UUID)
1630 __uuid_write(c);
1631
Kent Overstreetc18536a2013-07-24 17:44:17 -07001632 bch_journal_replay(c, &journal);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001633 } else {
1634 pr_notice("invalidating existing data");
Kent Overstreetcafe5632013-03-23 16:11:31 -07001635
1636 for_each_cache(ca, c, i) {
1637 unsigned j;
1638
1639 ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7,
1640 2, SB_JOURNAL_BUCKETS);
1641
1642 for (j = 0; j < ca->sb.keys; j++)
1643 ca->sb.d[j] = ca->sb.first_bucket + j;
1644 }
1645
Kent Overstreet2531d9ee2014-03-17 16:55:55 -07001646 bch_initial_gc_finish(c);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001647
Kent Overstreet119ba0f2013-04-24 19:01:12 -07001648 err = "error starting allocator thread";
Kent Overstreetcafe5632013-03-23 16:11:31 -07001649 for_each_cache(ca, c, i)
Kent Overstreet119ba0f2013-04-24 19:01:12 -07001650 if (bch_cache_allocator_start(ca))
1651 goto err;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001652
1653 mutex_lock(&c->bucket_lock);
1654 for_each_cache(ca, c, i)
1655 bch_prio_write(ca);
1656 mutex_unlock(&c->bucket_lock);
1657
Kent Overstreetcafe5632013-03-23 16:11:31 -07001658 err = "cannot allocate new UUID bucket";
1659 if (__uuid_write(c))
Kent Overstreet72a44512013-10-24 17:19:26 -07001660 goto err;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001661
1662 err = "cannot allocate new btree root";
Slava Pestov2452cc82014-07-12 00:22:53 -07001663 c->root = __bch_btree_node_alloc(c, NULL, 0, true, NULL);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001664 if (IS_ERR_OR_NULL(c->root))
Kent Overstreet72a44512013-10-24 17:19:26 -07001665 goto err;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001666
Kent Overstreet2a285682014-03-04 16:42:42 -08001667 mutex_lock(&c->root->write_lock);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001668 bkey_copy_key(&c->root->key, &MAX_KEY);
Kent Overstreetc18536a2013-07-24 17:44:17 -07001669 bch_btree_node_write(c->root, &cl);
Kent Overstreet2a285682014-03-04 16:42:42 -08001670 mutex_unlock(&c->root->write_lock);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001671
1672 bch_btree_set_root(c->root);
1673 rw_unlock(true, c->root);
1674
1675 /*
1676 * We don't want to write the first journal entry until
1677 * everything is set up - fortunately journal entries won't be
1678 * written until the SET_CACHE_SYNC() here:
1679 */
1680 SET_CACHE_SYNC(&c->sb, true);
1681
1682 bch_journal_next(&c->journal);
Kent Overstreetc18536a2013-07-24 17:44:17 -07001683 bch_journal_meta(c, &cl);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001684 }
1685
Kent Overstreet72a44512013-10-24 17:19:26 -07001686 err = "error starting gc thread";
1687 if (bch_gc_thread_start(c))
1688 goto err;
1689
Kent Overstreetc18536a2013-07-24 17:44:17 -07001690 closure_sync(&cl);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001691 c->sb.last_mount = get_seconds();
1692 bcache_write_super(c);
1693
1694 list_for_each_entry_safe(dc, t, &uncached_devices, list)
1695 bch_cached_dev_attach(dc, c);
1696
1697 flash_devs_run(c);
1698
Slava Pestovbf0c55c2014-07-11 12:17:41 -07001699 set_bit(CACHE_SET_RUNNING, &c->flags);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001700 return;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001701err:
Kent Overstreetc18536a2013-07-24 17:44:17 -07001702 closure_sync(&cl);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001703 /* XXX: test this, it's broken */
Kees Cookc8694942013-09-10 21:41:34 -07001704 bch_cache_set_error(c, "%s", err);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001705}
1706
1707static bool can_attach_cache(struct cache *ca, struct cache_set *c)
1708{
1709 return ca->sb.block_size == c->sb.block_size &&
Nicholas Swenson9eb8ebe2013-10-22 13:19:23 -07001710 ca->sb.bucket_size == c->sb.bucket_size &&
Kent Overstreetcafe5632013-03-23 16:11:31 -07001711 ca->sb.nr_in_set == c->sb.nr_in_set;
1712}
1713
1714static const char *register_cache_set(struct cache *ca)
1715{
1716 char buf[12];
1717 const char *err = "cannot allocate memory";
1718 struct cache_set *c;
1719
1720 list_for_each_entry(c, &bch_cache_sets, list)
1721 if (!memcmp(c->sb.set_uuid, ca->sb.set_uuid, 16)) {
1722 if (c->cache[ca->sb.nr_this_dev])
1723 return "duplicate cache set member";
1724
1725 if (!can_attach_cache(ca, c))
1726 return "cache sb does not match set";
1727
1728 if (!CACHE_SYNC(&ca->sb))
1729 SET_CACHE_SYNC(&c->sb, false);
1730
1731 goto found;
1732 }
1733
1734 c = bch_cache_set_alloc(&ca->sb);
1735 if (!c)
1736 return err;
1737
1738 err = "error creating kobject";
1739 if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->sb.set_uuid) ||
1740 kobject_add(&c->internal, &c->kobj, "internal"))
1741 goto err;
1742
1743 if (bch_cache_accounting_add_kobjs(&c->accounting, &c->kobj))
1744 goto err;
1745
1746 bch_debug_init_cache_set(c);
1747
1748 list_add(&c->list, &bch_cache_sets);
1749found:
1750 sprintf(buf, "cache%i", ca->sb.nr_this_dev);
1751 if (sysfs_create_link(&ca->kobj, &c->kobj, "set") ||
1752 sysfs_create_link(&c->kobj, &ca->kobj, buf))
1753 goto err;
1754
1755 if (ca->sb.seq > c->sb.seq) {
1756 c->sb.version = ca->sb.version;
1757 memcpy(c->sb.set_uuid, ca->sb.set_uuid, 16);
1758 c->sb.flags = ca->sb.flags;
1759 c->sb.seq = ca->sb.seq;
1760 pr_debug("set version = %llu", c->sb.version);
1761 }
1762
Kent Overstreetd83353b2014-06-11 19:44:49 -07001763 kobject_get(&ca->kobj);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001764 ca->set = c;
1765 ca->set->cache[ca->sb.nr_this_dev] = ca;
1766 c->cache_by_alloc[c->caches_loaded++] = ca;
1767
1768 if (c->caches_loaded == c->sb.nr_in_set)
1769 run_cache_set(c);
1770
1771 return NULL;
1772err:
1773 bch_cache_set_unregister(c);
1774 return err;
1775}
1776
1777/* Cache device */
1778
1779void bch_cache_release(struct kobject *kobj)
1780{
1781 struct cache *ca = container_of(kobj, struct cache, kobj);
Kent Overstreet78365412013-12-17 01:29:34 -08001782 unsigned i;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001783
Slava Pestovc9a78332014-06-19 15:05:59 -07001784 if (ca->set) {
1785 BUG_ON(ca->set->cache[ca->sb.nr_this_dev] != ca);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001786 ca->set->cache[ca->sb.nr_this_dev] = NULL;
Slava Pestovc9a78332014-06-19 15:05:59 -07001787 }
Kent Overstreetcafe5632013-03-23 16:11:31 -07001788
Kent Overstreetcafe5632013-03-23 16:11:31 -07001789 free_pages((unsigned long) ca->disk_buckets, ilog2(bucket_pages(ca)));
1790 kfree(ca->prio_buckets);
1791 vfree(ca->buckets);
1792
1793 free_heap(&ca->heap);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001794 free_fifo(&ca->free_inc);
Kent Overstreet78365412013-12-17 01:29:34 -08001795
1796 for (i = 0; i < RESERVE_NR; i++)
1797 free_fifo(&ca->free[i]);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001798
1799 if (ca->sb_bio.bi_inline_vecs[0].bv_page)
1800 put_page(ca->sb_bio.bi_io_vec[0].bv_page);
1801
Kent Overstreet0781c872014-07-07 13:03:36 -07001802 if (!IS_ERR_OR_NULL(ca->bdev))
Kent Overstreetcafe5632013-03-23 16:11:31 -07001803 blkdev_put(ca->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001804
1805 kfree(ca);
1806 module_put(THIS_MODULE);
1807}
1808
Yijing Wangc50d4d52016-07-04 09:23:25 +08001809static int cache_alloc(struct cache *ca)
Kent Overstreetcafe5632013-03-23 16:11:31 -07001810{
1811 size_t free;
1812 struct bucket *b;
1813
Kent Overstreetcafe5632013-03-23 16:11:31 -07001814 __module_get(THIS_MODULE);
1815 kobject_init(&ca->kobj, &bch_cache_ktype);
1816
Kent Overstreetcafe5632013-03-23 16:11:31 -07001817 bio_init(&ca->journal.bio);
1818 ca->journal.bio.bi_max_vecs = 8;
1819 ca->journal.bio.bi_io_vec = ca->journal.bio.bi_inline_vecs;
1820
Kent Overstreet78365412013-12-17 01:29:34 -08001821 free = roundup_pow_of_two(ca->sb.nbuckets) >> 10;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001822
Kent Overstreet78365412013-12-17 01:29:34 -08001823 if (!init_fifo(&ca->free[RESERVE_BTREE], 8, GFP_KERNEL) ||
Kent Overstreetacc9cf82016-08-17 18:21:24 -07001824 !init_fifo_exact(&ca->free[RESERVE_PRIO], prio_buckets(ca), GFP_KERNEL) ||
Kent Overstreet78365412013-12-17 01:29:34 -08001825 !init_fifo(&ca->free[RESERVE_MOVINGGC], free, GFP_KERNEL) ||
1826 !init_fifo(&ca->free[RESERVE_NONE], free, GFP_KERNEL) ||
Kent Overstreetcafe5632013-03-23 16:11:31 -07001827 !init_fifo(&ca->free_inc, free << 2, GFP_KERNEL) ||
Kent Overstreetcafe5632013-03-23 16:11:31 -07001828 !init_heap(&ca->heap, free << 3, GFP_KERNEL) ||
Kent Overstreetf59fce82013-05-15 00:11:26 -07001829 !(ca->buckets = vzalloc(sizeof(struct bucket) *
Kent Overstreetcafe5632013-03-23 16:11:31 -07001830 ca->sb.nbuckets)) ||
1831 !(ca->prio_buckets = kzalloc(sizeof(uint64_t) * prio_buckets(ca) *
1832 2, GFP_KERNEL)) ||
Kent Overstreet749b61d2013-11-23 23:11:25 -08001833 !(ca->disk_buckets = alloc_bucket_pages(GFP_KERNEL, ca)))
Kent Overstreetf59fce82013-05-15 00:11:26 -07001834 return -ENOMEM;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001835
1836 ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca);
1837
Kent Overstreetcafe5632013-03-23 16:11:31 -07001838 for_each_bucket(b, ca)
1839 atomic_set(&b->pin, 0);
1840
Kent Overstreetcafe5632013-03-23 16:11:31 -07001841 return 0;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001842}
1843
Eric Wheeler9b299722016-02-26 14:33:56 -08001844static int register_cache(struct cache_sb *sb, struct page *sb_page,
Slava Pestovc9a78332014-06-19 15:05:59 -07001845 struct block_device *bdev, struct cache *ca)
Kent Overstreetcafe5632013-03-23 16:11:31 -07001846{
1847 char name[BDEVNAME_SIZE];
Eric Wheelerd9dc1702016-06-17 15:01:54 -07001848 const char *err = NULL; /* must be set for any error case */
Eric Wheeler9b299722016-02-26 14:33:56 -08001849 int ret = 0;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001850
Kent Overstreetf59fce82013-05-15 00:11:26 -07001851 memcpy(&ca->sb, sb, sizeof(struct cache_sb));
Kent Overstreetcafe5632013-03-23 16:11:31 -07001852 ca->bdev = bdev;
1853 ca->bdev->bd_holder = ca;
1854
Kent Overstreetf59fce82013-05-15 00:11:26 -07001855 bio_init(&ca->sb_bio);
1856 ca->sb_bio.bi_max_vecs = 1;
1857 ca->sb_bio.bi_io_vec = ca->sb_bio.bi_inline_vecs;
1858 ca->sb_bio.bi_io_vec[0].bv_page = sb_page;
1859 get_page(sb_page);
1860
Kent Overstreetcafe5632013-03-23 16:11:31 -07001861 if (blk_queue_discard(bdev_get_queue(ca->bdev)))
1862 ca->discard = CACHE_DISCARD(&ca->sb);
1863
Yijing Wangc50d4d52016-07-04 09:23:25 +08001864 ret = cache_alloc(ca);
Eric Wheelerd9dc1702016-06-17 15:01:54 -07001865 if (ret != 0) {
1866 if (ret == -ENOMEM)
1867 err = "cache_alloc(): -ENOMEM";
1868 else
1869 err = "cache_alloc(): unknown error";
Kent Overstreetf59fce82013-05-15 00:11:26 -07001870 goto err;
Eric Wheelerd9dc1702016-06-17 15:01:54 -07001871 }
Kent Overstreetf59fce82013-05-15 00:11:26 -07001872
Eric Wheeler9b299722016-02-26 14:33:56 -08001873 if (kobject_add(&ca->kobj, &part_to_dev(bdev->bd_part)->kobj, "bcache")) {
1874 err = "error calling kobject_add";
1875 ret = -ENOMEM;
1876 goto out;
1877 }
Kent Overstreetcafe5632013-03-23 16:11:31 -07001878
Kent Overstreet4fa03402014-03-17 18:58:55 -07001879 mutex_lock(&bch_register_lock);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001880 err = register_cache_set(ca);
Kent Overstreet4fa03402014-03-17 18:58:55 -07001881 mutex_unlock(&bch_register_lock);
1882
Eric Wheeler9b299722016-02-26 14:33:56 -08001883 if (err) {
1884 ret = -ENODEV;
1885 goto out;
1886 }
Kent Overstreetcafe5632013-03-23 16:11:31 -07001887
1888 pr_info("registered cache device %s", bdevname(bdev, name));
Eric Wheeler9b299722016-02-26 14:33:56 -08001889
Kent Overstreetd83353b2014-06-11 19:44:49 -07001890out:
1891 kobject_put(&ca->kobj);
Eric Wheeler9b299722016-02-26 14:33:56 -08001892
Kent Overstreetcafe5632013-03-23 16:11:31 -07001893err:
Eric Wheeler9b299722016-02-26 14:33:56 -08001894 if (err)
1895 pr_notice("error opening %s: %s", bdevname(bdev, name), err);
1896
1897 return ret;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001898}
1899
1900/* Global interfaces/init */
1901
1902static ssize_t register_bcache(struct kobject *, struct kobj_attribute *,
1903 const char *, size_t);
1904
1905kobj_attribute_write(register, register_bcache);
1906kobj_attribute_write(register_quiet, register_bcache);
1907
Gabriel de Perthuisa9dd53a2013-05-04 12:19:41 +02001908static bool bch_is_open_backing(struct block_device *bdev) {
1909 struct cache_set *c, *tc;
1910 struct cached_dev *dc, *t;
1911
1912 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
1913 list_for_each_entry_safe(dc, t, &c->cached_devs, list)
1914 if (dc->bdev == bdev)
1915 return true;
1916 list_for_each_entry_safe(dc, t, &uncached_devices, list)
1917 if (dc->bdev == bdev)
1918 return true;
1919 return false;
1920}
1921
1922static bool bch_is_open_cache(struct block_device *bdev) {
1923 struct cache_set *c, *tc;
1924 struct cache *ca;
1925 unsigned i;
1926
1927 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
1928 for_each_cache(ca, c, i)
1929 if (ca->bdev == bdev)
1930 return true;
1931 return false;
1932}
1933
1934static bool bch_is_open(struct block_device *bdev) {
1935 return bch_is_open_cache(bdev) || bch_is_open_backing(bdev);
1936}
1937
Kent Overstreetcafe5632013-03-23 16:11:31 -07001938static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr,
1939 const char *buffer, size_t size)
1940{
1941 ssize_t ret = size;
1942 const char *err = "cannot allocate memory";
1943 char *path = NULL;
1944 struct cache_sb *sb = NULL;
1945 struct block_device *bdev = NULL;
1946 struct page *sb_page = NULL;
1947
1948 if (!try_module_get(THIS_MODULE))
1949 return -EBUSY;
1950
Kent Overstreetcafe5632013-03-23 16:11:31 -07001951 if (!(path = kstrndup(buffer, size, GFP_KERNEL)) ||
1952 !(sb = kmalloc(sizeof(struct cache_sb), GFP_KERNEL)))
1953 goto err;
1954
1955 err = "failed to open device";
1956 bdev = blkdev_get_by_path(strim(path),
1957 FMODE_READ|FMODE_WRITE|FMODE_EXCL,
1958 sb);
Kent Overstreetf59fce82013-05-15 00:11:26 -07001959 if (IS_ERR(bdev)) {
Gabriel de Perthuisa9dd53a2013-05-04 12:19:41 +02001960 if (bdev == ERR_PTR(-EBUSY)) {
1961 bdev = lookup_bdev(strim(path));
Jianjian Huo789d21d2014-07-13 09:08:59 -07001962 mutex_lock(&bch_register_lock);
Gabriel de Perthuisa9dd53a2013-05-04 12:19:41 +02001963 if (!IS_ERR(bdev) && bch_is_open(bdev))
1964 err = "device already registered";
1965 else
1966 err = "device busy";
Jianjian Huo789d21d2014-07-13 09:08:59 -07001967 mutex_unlock(&bch_register_lock);
Gabriel de Perthuisd7076f22015-11-29 18:40:23 -08001968 if (attr == &ksysfs_register_quiet)
1969 goto out;
Gabriel de Perthuisa9dd53a2013-05-04 12:19:41 +02001970 }
Kent Overstreetcafe5632013-03-23 16:11:31 -07001971 goto err;
Kent Overstreetf59fce82013-05-15 00:11:26 -07001972 }
1973
1974 err = "failed to set blocksize";
1975 if (set_blocksize(bdev, 4096))
1976 goto err_close;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001977
1978 err = read_super(sb, bdev, &sb_page);
1979 if (err)
1980 goto err_close;
1981
Kent Overstreet29033812013-04-11 15:14:35 -07001982 if (SB_IS_BDEV(sb)) {
Kent Overstreetcafe5632013-03-23 16:11:31 -07001983 struct cached_dev *dc = kzalloc(sizeof(*dc), GFP_KERNEL);
Kent Overstreetf59fce82013-05-15 00:11:26 -07001984 if (!dc)
1985 goto err_close;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001986
Kent Overstreet4fa03402014-03-17 18:58:55 -07001987 mutex_lock(&bch_register_lock);
Kent Overstreetf59fce82013-05-15 00:11:26 -07001988 register_bdev(sb, sb_page, bdev, dc);
Kent Overstreet4fa03402014-03-17 18:58:55 -07001989 mutex_unlock(&bch_register_lock);
Kent Overstreetcafe5632013-03-23 16:11:31 -07001990 } else {
1991 struct cache *ca = kzalloc(sizeof(*ca), GFP_KERNEL);
Kent Overstreetf59fce82013-05-15 00:11:26 -07001992 if (!ca)
1993 goto err_close;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001994
Eric Wheeler9b299722016-02-26 14:33:56 -08001995 if (register_cache(sb, sb_page, bdev, ca) != 0)
1996 goto err_close;
Kent Overstreetcafe5632013-03-23 16:11:31 -07001997 }
Kent Overstreetf59fce82013-05-15 00:11:26 -07001998out:
1999 if (sb_page)
Kent Overstreetcafe5632013-03-23 16:11:31 -07002000 put_page(sb_page);
Kent Overstreetcafe5632013-03-23 16:11:31 -07002001 kfree(sb);
2002 kfree(path);
Kent Overstreetcafe5632013-03-23 16:11:31 -07002003 module_put(THIS_MODULE);
2004 return ret;
Kent Overstreetf59fce82013-05-15 00:11:26 -07002005
2006err_close:
2007 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2008err:
Gabriel de Perthuisd7076f22015-11-29 18:40:23 -08002009 pr_info("error opening %s: %s", path, err);
Kent Overstreetf59fce82013-05-15 00:11:26 -07002010 ret = -EINVAL;
2011 goto out;
Kent Overstreetcafe5632013-03-23 16:11:31 -07002012}
2013
2014static int bcache_reboot(struct notifier_block *n, unsigned long code, void *x)
2015{
2016 if (code == SYS_DOWN ||
2017 code == SYS_HALT ||
2018 code == SYS_POWER_OFF) {
2019 DEFINE_WAIT(wait);
2020 unsigned long start = jiffies;
2021 bool stopped = false;
2022
2023 struct cache_set *c, *tc;
2024 struct cached_dev *dc, *tdc;
2025
2026 mutex_lock(&bch_register_lock);
2027
2028 if (list_empty(&bch_cache_sets) &&
2029 list_empty(&uncached_devices))
2030 goto out;
2031
2032 pr_info("Stopping all devices:");
2033
2034 list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
2035 bch_cache_set_stop(c);
2036
2037 list_for_each_entry_safe(dc, tdc, &uncached_devices, list)
2038 bcache_device_stop(&dc->disk);
2039
2040 /* What's a condition variable? */
2041 while (1) {
2042 long timeout = start + 2 * HZ - jiffies;
2043
2044 stopped = list_empty(&bch_cache_sets) &&
2045 list_empty(&uncached_devices);
2046
2047 if (timeout < 0 || stopped)
2048 break;
2049
2050 prepare_to_wait(&unregister_wait, &wait,
2051 TASK_UNINTERRUPTIBLE);
2052
2053 mutex_unlock(&bch_register_lock);
2054 schedule_timeout(timeout);
2055 mutex_lock(&bch_register_lock);
2056 }
2057
2058 finish_wait(&unregister_wait, &wait);
2059
2060 if (stopped)
2061 pr_info("All devices stopped");
2062 else
2063 pr_notice("Timeout waiting for devices to be closed");
2064out:
2065 mutex_unlock(&bch_register_lock);
2066 }
2067
2068 return NOTIFY_DONE;
2069}
2070
2071static struct notifier_block reboot = {
2072 .notifier_call = bcache_reboot,
2073 .priority = INT_MAX, /* before any real devices */
2074};
2075
2076static void bcache_exit(void)
2077{
2078 bch_debug_exit();
Kent Overstreetcafe5632013-03-23 16:11:31 -07002079 bch_request_exit();
Kent Overstreetcafe5632013-03-23 16:11:31 -07002080 if (bcache_kobj)
2081 kobject_put(bcache_kobj);
2082 if (bcache_wq)
2083 destroy_workqueue(bcache_wq);
Kent Overstreet5c41c8a2013-07-08 17:53:26 -07002084 if (bcache_major)
2085 unregister_blkdev(bcache_major, "bcache");
Kent Overstreetcafe5632013-03-23 16:11:31 -07002086 unregister_reboot_notifier(&reboot);
2087}
2088
2089static int __init bcache_init(void)
2090{
2091 static const struct attribute *files[] = {
2092 &ksysfs_register.attr,
2093 &ksysfs_register_quiet.attr,
2094 NULL
2095 };
2096
2097 mutex_init(&bch_register_lock);
2098 init_waitqueue_head(&unregister_wait);
2099 register_reboot_notifier(&reboot);
Kent Overstreet07e86cc2013-03-25 11:46:43 -07002100 closure_debug_init();
Kent Overstreetcafe5632013-03-23 16:11:31 -07002101
2102 bcache_major = register_blkdev(0, "bcache");
Zheng Liu2ecf0cd2015-11-29 17:21:57 -08002103 if (bcache_major < 0) {
2104 unregister_reboot_notifier(&reboot);
Kent Overstreetcafe5632013-03-23 16:11:31 -07002105 return bcache_major;
Zheng Liu2ecf0cd2015-11-29 17:21:57 -08002106 }
Kent Overstreetcafe5632013-03-23 16:11:31 -07002107
Bhaktipriya Shridhar81baf902016-06-08 01:57:19 +05302108 if (!(bcache_wq = alloc_workqueue("bcache", WQ_MEM_RECLAIM, 0)) ||
Kent Overstreetcafe5632013-03-23 16:11:31 -07002109 !(bcache_kobj = kobject_create_and_add("bcache", fs_kobj)) ||
2110 sysfs_create_files(bcache_kobj, files) ||
Kent Overstreetcafe5632013-03-23 16:11:31 -07002111 bch_request_init() ||
Kent Overstreetcafe5632013-03-23 16:11:31 -07002112 bch_debug_init(bcache_kobj))
2113 goto err;
2114
2115 return 0;
2116err:
2117 bcache_exit();
2118 return -ENOMEM;
2119}
2120
2121module_exit(bcache_exit);
2122module_init(bcache_init);