~hc/RK356X_SDK_RELEASE.git

..	..	@@ -13,20 +13,21 @@
13	13	#include "extents.h"
14	14	#include "request.h"
15	15	#include "writeback.h"
	16	+#include "features.h"
16	17
17	18	#include <linux/blkdev.h>
18		-#include <linux/buffer_head.h>
19	19	#include <linux/debugfs.h>
20	20	#include <linux/genhd.h>
21	21	#include <linux/idr.h>
22	22	#include <linux/kthread.h>
	23	+#include <linux/workqueue.h>
23	24	#include <linux/module.h>
24	25	#include <linux/random.h>
25	26	#include <linux/reboot.h>
26	27	#include <linux/sysfs.h>
27	28
28		-MODULE_LICENSE("GPL");
29		-MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
	29	+unsigned int bch_cutoff_writeback;
	30	+unsigned int bch_cutoff_writeback_sync;
30	31
31	32	static const char bcache_magic[] = {
32	33	0xc6, 0x85, 0x73, 0xf6, 0x4e, 0x1a, 0x45, 0xca,
..	..	@@ -40,6 +41,7 @@
40	41
41	42	static struct kobject *bcache_kobj;
42	43	struct mutex bch_register_lock;
	44	+bool bcache_is_reboot;
43	45	LIST_HEAD(bch_cache_sets);
44	46	static LIST_HEAD(uncached_devices);
45	47
..	..	@@ -47,7 +49,9 @@
47	49	static DEFINE_IDA(bcache_device_idx);
48	50	static wait_queue_head_t unregister_wait;
49	51	struct workqueue_struct *bcache_wq;
	52	+struct workqueue_struct *bch_flush_wq;
50	53	struct workqueue_struct *bch_journal_wq;
	54	+
51	55
52	56	#define BTREE_MAX_PAGES (256 * 1024 / PAGE_SIZE)
53	57	/* limitation of partitions number on single bcache device */
..	..	@@ -57,18 +61,121 @@
57	61
58	62	/* Superblock */
59	63
60		-static const char read_super(struct cache_sb sb, struct block_device *bdev,
61		- struct page **res)
	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	+
	91	+static const char read_super_common(struct cache_sb sb, struct block_device *bdev,
	92	+ struct cache_sb_disk *s)
62	93	{
63	94	const char *err;
64		- struct cache_sb *s;
65		- struct buffer_head *bh = __bread(bdev, 1, SB_SIZE);
66	95	unsigned int i;
67	96
68		- if (!bh)
69		- return "IO error";
	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);
70	100
71		- s = (struct cache_sb *) bh->b_data;
	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	+
	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);
72	179
73	180	sb->offset = le64_to_cpu(s->offset);
74	181	sb->version = le64_to_cpu(s->version);
..	..	@@ -81,24 +188,20 @@
81	188	sb->flags = le64_to_cpu(s->flags);
82	189	sb->seq = le64_to_cpu(s->seq);
83	190	sb->last_mount = le32_to_cpu(s->last_mount);
84		- sb->first_bucket = le16_to_cpu(s->first_bucket);
85	191	sb->keys = le16_to_cpu(s->keys);
86	192
87	193	for (i = 0; i < SB_JOURNAL_BUCKETS; i++)
88	194	sb->d[i] = le64_to_cpu(s->d[i]);
89	195
90		- pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u",
	196	+ pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u\n",
91	197	sb->version, sb->flags, sb->seq, sb->keys);
92	198
93		- err = "Not a bcache superblock";
	199	+ err = "Not a bcache superblock (bad offset)";
94	200	if (sb->offset != SB_SECTOR)
95	201	goto err;
96	202
	203	+ err = "Not a bcache superblock (bad magic)";
97	204	if (memcmp(sb->magic, bcache_magic, 16))
98		- goto err;
99		-
100		- err = "Too many journal buckets";
101		- if (sb->keys > SB_JOURNAL_BUCKETS)
102	205	goto err;
103	206
104	207	err = "Bad checksum";
..	..	@@ -120,6 +223,7 @@
120	223	sb->data_offset = BDEV_DATA_START_DEFAULT;
121	224	break;
122	225	case BCACHE_SB_VERSION_BDEV_WITH_OFFSET:
	226	+ case BCACHE_SB_VERSION_BDEV_WITH_FEATURES:
123	227	sb->data_offset = le64_to_cpu(s->data_offset);
124	228
125	229	err = "Bad data offset";
..	..	@@ -129,55 +233,35 @@
129	233	break;
130	234	case BCACHE_SB_VERSION_CDEV:
131	235	case BCACHE_SB_VERSION_CDEV_WITH_UUID:
132		- sb->nbuckets = le64_to_cpu(s->nbuckets);
133		- sb->bucket_size = le16_to_cpu(s->bucket_size);
	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);
134	248
135		- sb->nr_in_set = le16_to_cpu(s->nr_in_set);
136		- sb->nr_this_dev = le16_to_cpu(s->nr_this_dev);
137		-
138		- err = "Too many buckets";
139		- if (sb->nbuckets > LONG_MAX)
	249	+ /* Check incompatible features */
	250	+ err = "Unsupported compatible feature found";
	251	+ if (bch_has_unknown_compat_features(sb))
140	252	goto err;
141	253
142		- err = "Not enough buckets";
143		- if (sb->nbuckets < 1 << 7)
	254	+ err = "Unsupported read-only compatible feature found";
	255	+ if (bch_has_unknown_ro_compat_features(sb))
144	256	goto err;
145	257
146		- err = "Bad block/bucket size";
147		- if (!is_power_of_2(sb->block_size) \|\|
148		- sb->block_size > PAGE_SECTORS \|\|
149		- !is_power_of_2(sb->bucket_size) \|\|
150		- sb->bucket_size < PAGE_SECTORS)
	258	+ err = "Unsupported incompatible feature found";
	259	+ if (bch_has_unknown_incompat_features(sb))
151	260	goto err;
152	261
153		- err = "Invalid superblock: device too small";
154		- if (get_capacity(bdev->bd_disk) <
155		- sb->bucket_size * sb->nbuckets)
	262	+ err = read_super_common(sb, bdev, s);
	263	+ if (err)
156	264	goto err;
157		-
158		- err = "Bad UUID";
159		- if (bch_is_zero(sb->set_uuid, 16))
160		- goto err;
161		-
162		- err = "Bad cache device number in set";
163		- if (!sb->nr_in_set \|\|
164		- sb->nr_in_set <= sb->nr_this_dev \|\|
165		- sb->nr_in_set > MAX_CACHES_PER_SET)
166		- goto err;
167		-
168		- err = "Journal buckets not sequential";
169		- for (i = 0; i < sb->keys; i++)
170		- if (sb->d[i] != sb->first_bucket + i)
171		- goto err;
172		-
173		- err = "Too many journal buckets";
174		- if (sb->first_bucket + sb->keys > sb->nbuckets)
175		- goto err;
176		-
177		- err = "Invalid superblock: first bucket comes before end of super";
178		- if (sb->first_bucket * sb->bucket_size < 16)
179		- goto err;
180		-
181	265	break;
182	266	default:
183	267	err = "Unsupported superblock version";
..	..	@@ -185,35 +269,34 @@
185	269	}
186	270
187	271	sb->last_mount = (u32)ktime_get_real_seconds();
188		- err = NULL;
189		-
190		- get_page(bh->b_page);
191		- *res = bh->b_page;
	272	+ *res = s;
	273	+ return NULL;
192	274	err:
193		- put_bh(bh);
	275	+ put_page(page);
194	276	return err;
195	277	}
196	278
197	279	static void write_bdev_super_endio(struct bio *bio)
198	280	{
199	281	struct cached_dev *dc = bio->bi_private;
200		- /* XXX: error checking */
	282	+
	283	+ if (bio->bi_status)
	284	+ bch_count_backing_io_errors(dc, bio);
201	285
202	286	closure_put(&dc->sb_write);
203	287	}
204	288
205		-static void __write_super(struct cache_sb sb, struct bio bio)
	289	+static void __write_super(struct cache_sb sb, struct cache_sb_disk out,
	290	+ struct bio *bio)
206	291	{
207		- struct cache_sb *out = page_address(bio_first_page_all(bio));
208	292	unsigned int i;
209	293
	294	+ bio->bi_opf = REQ_OP_WRITE \| REQ_SYNC \| REQ_META;
210	295	bio->bi_iter.bi_sector = SB_SECTOR;
211		- bio->bi_iter.bi_size = SB_SIZE;
212		- bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC\|REQ_META);
213		- bch_bio_map(bio, NULL);
	296	+ __bio_add_page(bio, virt_to_page(out), SB_SIZE,
	297	+ offset_in_page(out));
214	298
215	299	out->offset = cpu_to_le64(sb->offset);
216		- out->version = cpu_to_le64(sb->version);
217	300
218	301	memcpy(out->uuid, sb->uuid, 16);
219	302	memcpy(out->set_uuid, sb->set_uuid, 16);
..	..	@@ -229,9 +312,16 @@
229	312	for (i = 0; i < sb->keys; i++)
230	313	out->d[i] = cpu_to_le64(sb->d[i]);
231	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);
232	322	out->csum = csum_set(out);
233	323
234		- pr_debug("ver %llu, flags %llu, seq %llu",
	324	+ pr_debug("ver %llu, flags %llu, seq %llu\n",
235	325	sb->version, sb->flags, sb->seq);
236	326
237	327	submit_bio(bio);
..	..	@@ -252,14 +342,14 @@
252	342	down(&dc->sb_write_mutex);
253	343	closure_init(cl, parent);
254	344
255		- bio_reset(bio);
	345	+ bio_init(bio, dc->sb_bv, 1);
256	346	bio_set_dev(bio, dc->bdev);
257	347	bio->bi_end_io = write_bdev_super_endio;
258	348	bio->bi_private = dc;
259	349
260	350	closure_get(cl);
261	351	/* I/O request sent to backing device */
262		- __write_super(&dc->sb, bio);
	352	+ __write_super(&dc->sb, dc->sb_disk, bio);
263	353
264	354	closure_return_with_destructor(cl, bch_write_bdev_super_unlock);
265	355	}
..	..	@@ -284,31 +374,25 @@
284	374	void bcache_write_super(struct cache_set *c)
285	375	{
286	376	struct closure *cl = &c->sb_write;
287		- struct cache *ca;
288		- unsigned int i;
	377	+ struct cache *ca = c->cache;
	378	+ struct bio *bio = &ca->sb_bio;
	379	+ unsigned int version = BCACHE_SB_VERSION_CDEV_WITH_UUID;
289	380
290	381	down(&c->sb_write_mutex);
291	382	closure_init(cl, &c->cl);
292	383
293		- c->sb.seq++;
	384	+ ca->sb.seq++;
294	385
295		- for_each_cache(ca, c, i) {
296		- struct bio *bio = &ca->sb_bio;
	386	+ if (ca->sb.version < version)
	387	+ ca->sb.version = version;
297	388
298		- ca->sb.version = BCACHE_SB_VERSION_CDEV_WITH_UUID;
299		- ca->sb.seq = c->sb.seq;
300		- ca->sb.last_mount = c->sb.last_mount;
	389	+ bio_init(bio, ca->sb_bv, 1);
	390	+ bio_set_dev(bio, ca->bdev);
	391	+ bio->bi_end_io = write_super_endio;
	392	+ bio->bi_private = ca;
301	393
302		- SET_CACHE_SYNC(&ca->sb, CACHE_SYNC(&c->sb));
303		-
304		- bio_reset(bio);
305		- bio_set_dev(bio, 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		- }
	394	+ closure_get(cl);
	395	+ __write_super(&ca->sb, ca->sb_disk, bio);
312	396
313	397	closure_return_with_destructor(cl, bcache_write_super_unlock);
314	398	}
..	..	@@ -362,11 +446,11 @@
362	446	}
363	447
364	448	bch_extent_to_text(buf, sizeof(buf), k);
365		- pr_debug("%s UUIDs at %s", op == REQ_OP_WRITE ? "wrote" : "read", buf);
	449	+ pr_debug("%s UUIDs at %s\n", op == REQ_OP_WRITE ? "wrote" : "read", buf);
366	450
367	451	for (u = c->uuids; u < c->uuids + c->nr_uuids; u++)
368	452	if (!bch_is_zero(u->uuid, 16))
369		- pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u",
	453	+ pr_debug("Slot %zi: %pU: %s: 1st: %u last: %u inv: %u\n",
370	454	u - c->uuids, u->uuid, u->label,
371	455	u->first_reg, u->last_reg, u->invalidated);
372	456
..	..	@@ -418,20 +502,21 @@
418	502	{
419	503	BKEY_PADDED(key) k;
420	504	struct closure cl;
421		- struct cache *ca;
	505	+ struct cache *ca = c->cache;
	506	+ unsigned int size;
422	507
423	508	closure_init_stack(&cl);
424	509	lockdep_assert_held(&bch_register_lock);
425	510
426		- if (bch_bucket_alloc_set(c, RESERVE_BTREE, &k.key, 1, true))
	511	+ if (bch_bucket_alloc_set(c, RESERVE_BTREE, &k.key, true))
427	512	return 1;
428	513
429		- SET_KEY_SIZE(&k.key, c->sb.bucket_size);
	514	+ size = meta_bucket_pages(&ca->sb) * PAGE_SECTORS;
	515	+ SET_KEY_SIZE(&k.key, size);
430	516	uuid_io(c, REQ_OP_WRITE, 0, &k.key, &cl);
431	517	closure_sync(&cl);
432	518
433	519	/* Only one bucket used for uuid write */
434		- ca = PTR_CACHE(c, &k.key, 0);
435	520	atomic_long_add(ca->sb.bucket_size, &ca->meta_sectors_written);
436	521
437	522	bkey_copy(&c->uuid_bucket, &k.key);
..	..	@@ -514,7 +599,7 @@
514	599
515	600	bio->bi_iter.bi_sector = bucket * ca->sb.bucket_size;
516	601	bio_set_dev(bio, ca->bdev);
517		- bio->bi_iter.bi_size = bucket_bytes(ca);
	602	+ bio->bi_iter.bi_size = meta_bucket_bytes(&ca->sb);
518	603
519	604	bio->bi_end_io = prio_endio;
520	605	bio->bi_private = ca;
..	..	@@ -531,7 +616,7 @@
531	616	struct bucket *b;
532	617	struct closure cl;
533	618
534		- pr_debug("free_prio=%zu, free_none=%zu, free_inc=%zu",
	619	+ pr_debug("free_prio=%zu, free_none=%zu, free_inc=%zu\n",
535	620	fifo_used(&ca->free[RESERVE_PRIO]),
536	621	fifo_used(&ca->free[RESERVE_NONE]),
537	622	fifo_used(&ca->free_inc));
..	..	@@ -572,7 +657,7 @@
572	657
573	658	p->next_bucket = ca->prio_buckets[i + 1];
574	659	p->magic = pset_magic(&ca->sb);
575		- p->csum = bch_crc64(&p->magic, bucket_bytes(ca) - 8);
	660	+ p->csum = bch_crc64(&p->magic, meta_bucket_bytes(&ca->sb) - 8);
576	661
577	662	bucket = bch_bucket_alloc(ca, RESERVE_PRIO, wait);
578	663	BUG_ON(bucket == -1);
..	..	@@ -606,12 +691,13 @@
606	691	return 0;
607	692	}
608	693
609		-static void prio_read(struct cache *ca, uint64_t bucket)
	694	+static int prio_read(struct cache *ca, uint64_t bucket)
610	695	{
611	696	struct prio_set *p = ca->disk_buckets;
612	697	struct bucket_disk d = p->data + prios_per_bucket(ca), end = d;
613	698	struct bucket *b;
614	699	unsigned int bucket_nr = 0;
	700	+ int ret = -EIO;
615	701
616	702	for (b = ca->buckets;
617	703	b < ca->buckets + ca->sb.nbuckets;
..	..	@@ -624,11 +710,15 @@
624	710	prio_io(ca, bucket, REQ_OP_READ, 0);
625	711
626	712	if (p->csum !=
627		- bch_crc64(&p->magic, bucket_bytes(ca) - 8))
628		- pr_warn("bad csum reading priorities");
	713	+ bch_crc64(&p->magic, meta_bucket_bytes(&ca->sb) - 8)) {
	714	+ pr_warn("bad csum reading priorities\n");
	715	+ goto out;
	716	+ }
629	717
630		- if (p->magic != pset_magic(&ca->sb))
631		- pr_warn("bad magic reading priorities");
	718	+ if (p->magic != pset_magic(&ca->sb)) {
	719	+ pr_warn("bad magic reading priorities\n");
	720	+ goto out;
	721	+ }
632	722
633	723	bucket = p->next_bucket;
634	724	d = p->data;
..	..	@@ -637,6 +727,10 @@
637	727	b->prio = le16_to_cpu(d->prio);
638	728	b->gen = b->last_gc = d->gen;
639	729	}
	730	+
	731	+ ret = 0;
	732	+out:
	733	+ return ret;
640	734	}
641	735
642	736	/* Bcache device */
..	..	@@ -667,7 +761,16 @@
667	761	return d->ioctl(d, mode, cmd, arg);
668	762	}
669	763
670		-static const struct block_device_operations bcache_ops = {
	764	+static const struct block_device_operations bcache_cached_ops = {
	765	+ .submit_bio = cached_dev_submit_bio,
	766	+ .open = open_dev,
	767	+ .release = release_dev,
	768	+ .ioctl = ioctl_dev,
	769	+ .owner = THIS_MODULE,
	770	+};
	771	+
	772	+static const struct block_device_operations bcache_flash_ops = {
	773	+ .submit_bio = flash_dev_submit_bio,
671	774	.open = open_dev,
672	775	.release = release_dev,
673	776	.ioctl = ioctl_dev,
..	..	@@ -677,6 +780,11 @@
677	780	void bcache_device_stop(struct bcache_device *d)
678	781	{
679	782	if (!test_and_set_bit(BCACHE_DEV_CLOSING, &d->flags))
	783	+ /*
	784	+ * closure_fn set to
	785	+ * - cached device: cached_dev_flush()
	786	+ * - flash dev: flash_dev_flush()
	787	+ */
680	788	closure_queue(&d->cl);
681	789	}
682	790
..	..	@@ -685,32 +793,33 @@
685	793	lockdep_assert_held(&bch_register_lock);
686	794
687	795	if (d->c && !test_and_set_bit(BCACHE_DEV_UNLINK_DONE, &d->flags)) {
688		- unsigned int i;
689		- struct cache *ca;
	796	+ struct cache *ca = d->c->cache;
690	797
691	798	sysfs_remove_link(&d->c->kobj, d->name);
692	799	sysfs_remove_link(&d->kobj, "cache");
693	800
694		- for_each_cache(ca, d->c, i)
695		- bd_unlink_disk_holder(ca->bdev, d->disk);
	801	+ bd_unlink_disk_holder(ca->bdev, d->disk);
696	802	}
697	803	}
698	804
699	805	static void bcache_device_link(struct bcache_device d, struct cache_set c,
700	806	const char *name)
701	807	{
702		- unsigned int i;
703		- struct cache *ca;
	808	+ struct cache *ca = c->cache;
	809	+ int ret;
704	810
705		- for_each_cache(ca, d->c, i)
706		- bd_link_disk_holder(ca->bdev, d->disk);
	811	+ bd_link_disk_holder(ca->bdev, d->disk);
707	812
708	813	snprintf(d->name, BCACHEDEVNAME_SIZE,
709	814	"%s%u", name, d->id);
710	815
711		- WARN(sysfs_create_link(&d->kobj, &c->kobj, "cache") \|\|
712		- sysfs_create_link(&c->kobj, &d->kobj, d->name),
713		- "Couldn't create device <-> cache set symlinks");
	816	+ ret = sysfs_create_link(&d->kobj, &c->kobj, "cache");
	817	+ if (ret < 0)
	818	+ pr_err("Couldn't create device -> cache set symlink\n");
	819	+
	820	+ ret = sysfs_create_link(&c->kobj, &d->kobj, d->name);
	821	+ if (ret < 0)
	822	+ pr_err("Couldn't create cache set -> device symlink\n");
714	823
715	824	clear_bit(BCACHE_DEV_UNLINK_DONE, &d->flags);
716	825	}
..	..	@@ -767,9 +876,9 @@
767	876	lockdep_assert_held(&bch_register_lock);
768	877
769	878	if (disk)
770		- pr_info("%s stopped", disk->disk_name);
	879	+ pr_info("%s stopped\n", disk->disk_name);
771	880	else
772		- pr_err("bcache device (NULL gendisk) stopped");
	881	+ pr_err("bcache device (NULL gendisk) stopped\n");
773	882
774	883	if (d->c)
775	884	bcache_device_detach(d);
..	..	@@ -797,24 +906,25 @@
797	906	}
798	907
799	908	static int bcache_device_init(struct bcache_device *d, unsigned int block_size,
800		- sector_t sectors)
	909	+ sector_t sectors, struct block_device *cached_bdev,
	910	+ const struct block_device_operations *ops)
801	911	{
802	912	struct request_queue *q;
803	913	const size_t max_stripes = min_t(size_t, INT_MAX,
804	914	SIZE_MAX / sizeof(atomic_t));
805		- size_t n;
	915	+ uint64_t n;
806	916	int idx;
807	917
808	918	if (!d->stripe_size)
809	919	d->stripe_size = 1 << 31;
810	920
811		- d->nr_stripes = DIV_ROUND_UP_ULL(sectors, d->stripe_size);
812		-
813		- if (!d->nr_stripes \|\| d->nr_stripes > max_stripes) {
814		- pr_err("nr_stripes too large or invalid: %u (start sector beyond end of disk?)",
815		- (unsigned int)d->nr_stripes);
	921	+ n = DIV_ROUND_UP_ULL(sectors, d->stripe_size);
	922	+ if (!n \|\| n > max_stripes) {
	923	+ pr_err("nr_stripes too large or invalid: %llu (start sector beyond end of disk?)\n",
	924	+ n);
816	925	return -ENOMEM;
817	926	}
	927	+ d->nr_stripes = n;
818	928
819	929	n = d->nr_stripes * sizeof(atomic_t);
820	930	d->stripe_sectors_dirty = kvzalloc(n, GFP_KERNEL);
..	..	@@ -844,17 +954,14 @@
844	954
845	955	d->disk->major = bcache_major;
846	956	d->disk->first_minor = idx_to_first_minor(idx);
847		- d->disk->fops = &bcache_ops;
	957	+ d->disk->fops = ops;
848	958	d->disk->private_data = d;
849	959
850		- q = blk_alloc_queue(GFP_KERNEL);
	960	+ q = blk_alloc_queue(NUMA_NO_NODE);
851	961	if (!q)
852	962	return -ENOMEM;
853	963
854		- blk_queue_make_request(q, NULL);
855	964	d->disk->queue = q;
856		- q->queuedata = d;
857		- q->backing_dev_info->congested_data = d;
858	965	q->limits.max_hw_sectors = UINT_MAX;
859	966	q->limits.max_sectors = UINT_MAX;
860	967	q->limits.max_segment_size = UINT_MAX;
..	..	@@ -864,6 +971,20 @@
864	971	q->limits.io_min = block_size;
865	972	q->limits.logical_block_size = block_size;
866	973	q->limits.physical_block_size = block_size;
	974	+
	975	+ if (q->limits.logical_block_size > PAGE_SIZE && cached_bdev) {
	976	+ /*
	977	+ * This should only happen with BCACHE_SB_VERSION_BDEV.
	978	+ * Block/page size is checked for BCACHE_SB_VERSION_CDEV.
	979	+ */
	980	+ pr_info("%s: sb/logical block size (%u) greater than page size (%lu) falling back to device logical block size (%u)\n",
	981	+ d->disk->disk_name, q->limits.logical_block_size,
	982	+ PAGE_SIZE, bdev_logical_block_size(cached_bdev));
	983	+
	984	+ /* This also adjusts physical block size/min io size if needed */
	985	+ blk_queue_logical_block_size(q, bdev_logical_block_size(cached_bdev));
	986	+ }
	987	+
867	988	blk_queue_flag_set(QUEUE_FLAG_NONROT, d->disk->queue);
868	989	blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, d->disk->queue);
869	990	blk_queue_flag_set(QUEUE_FLAG_DISCARD, d->disk->queue);
..	..	@@ -916,11 +1037,11 @@
916	1037	dc->offline_seconds = 0;
917	1038
918	1039	if (dc->offline_seconds >= BACKING_DEV_OFFLINE_TIMEOUT) {
919		- pr_err("%s: device offline for %d seconds",
	1040	+ pr_err("%s: device offline for %d seconds\n",
920	1041	dc->backing_dev_name,
921	1042	BACKING_DEV_OFFLINE_TIMEOUT);
922		- pr_err("%s: disable I/O request due to backing "
923		- "device offline", dc->disk.name);
	1043	+ pr_err("%s: disable I/O request due to backing device offline\n",
	1044	+ dc->disk.name);
924	1045	dc->io_disable = true;
925	1046	/* let others know earlier that io_disable is true */
926	1047	smp_mb();
..	..	@@ -935,25 +1056,33 @@
935	1056	}
936	1057
937	1058
938		-void bch_cached_dev_run(struct cached_dev *dc)
	1059	+int bch_cached_dev_run(struct cached_dev *dc)
939	1060	{
940	1061	struct bcache_device *d = &dc->disk;
941		- char buf[SB_LABEL_SIZE + 1];
	1062	+ char *buf = kmemdup_nul(dc->sb.label, SB_LABEL_SIZE, GFP_KERNEL);
942	1063	char *env[] = {
943	1064	"DRIVER=bcache",
944	1065	kasprintf(GFP_KERNEL, "CACHED_UUID=%pU", dc->sb.uuid),
945		- NULL,
	1066	+ kasprintf(GFP_KERNEL, "CACHED_LABEL=%s", buf ? : ""),
946	1067	NULL,
947	1068	};
948	1069
949		- memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
950		- buf[SB_LABEL_SIZE] = '\0';
951		- env[2] = kasprintf(GFP_KERNEL, "CACHED_LABEL=%s", buf);
	1070	+ if (dc->io_disable) {
	1071	+ pr_err("I/O disabled on cached dev %s\n",
	1072	+ dc->backing_dev_name);
	1073	+ kfree(env[1]);
	1074	+ kfree(env[2]);
	1075	+ kfree(buf);
	1076	+ return -EIO;
	1077	+ }
952	1078
953	1079	if (atomic_xchg(&dc->running, 1)) {
954	1080	kfree(env[1]);
955	1081	kfree(env[2]);
956		- return;
	1082	+ kfree(buf);
	1083	+ pr_info("cached dev %s is running already\n",
	1084	+ dc->backing_dev_name);
	1085	+ return -EBUSY;
957	1086	}
958	1087
959	1088	if (!d->c &&
..	..	@@ -976,18 +1105,22 @@
976	1105	kobject_uevent_env(&disk_to_dev(d->disk)->kobj, KOBJ_CHANGE, env);
977	1106	kfree(env[1]);
978	1107	kfree(env[2]);
	1108	+ kfree(buf);
979	1109
980	1110	if (sysfs_create_link(&d->kobj, &disk_to_dev(d->disk)->kobj, "dev") \|\|
981		- sysfs_create_link(&disk_to_dev(d->disk)->kobj, &d->kobj, "bcache"))
982		- pr_debug("error creating sysfs link");
	1111	+ sysfs_create_link(&disk_to_dev(d->disk)->kobj,
	1112	+ &d->kobj, "bcache")) {
	1113	+ pr_err("Couldn't create bcache dev <-> disk sysfs symlinks\n");
	1114	+ return -ENOMEM;
	1115	+ }
983	1116
984	1117	dc->status_update_thread = kthread_run(cached_dev_status_update,
985	1118	dc, "bcache_status_update");
986	1119	if (IS_ERR(dc->status_update_thread)) {
987		- pr_warn("failed to create bcache_status_update kthread, "
988		- "continue to run without monitoring backing "
989		- "device status");
	1120	+ pr_warn("failed to create bcache_status_update kthread, continue to run without monitoring backing device status\n");
990	1121	}
	1122	+
	1123	+ return 0;
991	1124	}
992	1125
993	1126	/*
..	..	@@ -1010,7 +1143,7 @@
1010	1143	} while (time_out > 0);
1011	1144
1012	1145	if (time_out == 0)
1013		- pr_warn("give up waiting for dc->writeback_write_update to quit");
	1146	+ pr_warn("give up waiting for dc->writeback_write_update to quit\n");
1014	1147
1015	1148	cancel_delayed_work_sync(&dc->writeback_rate_update);
1016	1149	}
..	..	@@ -1025,7 +1158,6 @@
1025	1158	BUG_ON(!test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags));
1026	1159	BUG_ON(refcount_read(&dc->count));
1027	1160
1028		- mutex_lock(&bch_register_lock);
1029	1161
1030	1162	if (test_and_clear_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags))
1031	1163	cancel_writeback_rate_update_dwork(dc);
..	..	@@ -1041,6 +1173,8 @@
1041	1173	bch_write_bdev_super(dc, &cl);
1042	1174	closure_sync(&cl);
1043	1175
	1176	+ mutex_lock(&bch_register_lock);
	1177	+
1044	1178	calc_cached_dev_sectors(dc->disk.c);
1045	1179	bcache_device_detach(&dc->disk);
1046	1180	list_move(&dc->list, &uncached_devices);
..	..	@@ -1050,7 +1184,7 @@
1050	1184
1051	1185	mutex_unlock(&bch_register_lock);
1052	1186
1053		- pr_info("Caching disabled for %s", dc->backing_dev_name);
	1187	+ pr_info("Caching disabled for %s\n", dc->backing_dev_name);
1054	1188
1055	1189	/* Drop ref we took in cached_dev_detach() */
1056	1190	closure_put(&dc->disk.cl);
..	..	@@ -1083,26 +1217,27 @@
1083	1217	uint32_t rtime = cpu_to_le32((u32)ktime_get_real_seconds());
1084	1218	struct uuid_entry *u;
1085	1219	struct cached_dev exist_dc, t;
	1220	+ int ret = 0;
1086	1221
1087		- if ((set_uuid && memcmp(set_uuid, c->sb.set_uuid, 16)) \|\|
1088		- (!set_uuid && memcmp(dc->sb.set_uuid, c->sb.set_uuid, 16)))
	1222	+ if ((set_uuid && memcmp(set_uuid, c->set_uuid, 16)) \|\|
	1223	+ (!set_uuid && memcmp(dc->sb.set_uuid, c->set_uuid, 16)))
1089	1224	return -ENOENT;
1090	1225
1091	1226	if (dc->disk.c) {
1092		- pr_err("Can't attach %s: already attached",
	1227	+ pr_err("Can't attach %s: already attached\n",
1093	1228	dc->backing_dev_name);
1094	1229	return -EINVAL;
1095	1230	}
1096	1231
1097	1232	if (test_bit(CACHE_SET_STOPPING, &c->flags)) {
1098		- pr_err("Can't attach %s: shutting down",
	1233	+ pr_err("Can't attach %s: shutting down\n",
1099	1234	dc->backing_dev_name);
1100	1235	return -EINVAL;
1101	1236	}
1102	1237
1103		- if (dc->sb.block_size < c->sb.block_size) {
	1238	+ if (dc->sb.block_size < c->cache->sb.block_size) {
1104	1239	/* Will die */
1105		- pr_err("Couldn't attach %s: block size less than set's block size",
	1240	+ pr_err("Couldn't attach %s: block size less than set's block size\n",
1106	1241	dc->backing_dev_name);
1107	1242	return -EINVAL;
1108	1243	}
..	..	@@ -1110,7 +1245,7 @@
1110	1245	/* Check whether already attached */
1111	1246	list_for_each_entry_safe(exist_dc, t, &c->cached_devs, list) {
1112	1247	if (!memcmp(dc->sb.uuid, exist_dc->sb.uuid, 16)) {
1113		- pr_err("Tried to attach %s but duplicate UUID already attached",
	1248	+ pr_err("Tried to attach %s but duplicate UUID already attached\n",
1114	1249	dc->backing_dev_name);
1115	1250
1116	1251	return -EINVAL;
..	..	@@ -1129,14 +1264,14 @@
1129	1264
1130	1265	if (!u) {
1131	1266	if (BDEV_STATE(&dc->sb) == BDEV_STATE_DIRTY) {
1132		- pr_err("Couldn't find uuid for %s in set",
	1267	+ pr_err("Couldn't find uuid for %s in set\n",
1133	1268	dc->backing_dev_name);
1134	1269	return -ENOENT;
1135	1270	}
1136	1271
1137	1272	u = uuid_find_empty(c);
1138	1273	if (!u) {
1139		- pr_err("Not caching %s, no room for UUID",
	1274	+ pr_err("Not caching %s, no room for UUID\n",
1140	1275	dc->backing_dev_name);
1141	1276	return -EINVAL;
1142	1277	}
..	..	@@ -1157,7 +1292,7 @@
1157	1292	u->first_reg = u->last_reg = rtime;
1158	1293	bch_uuid_write(c);
1159	1294
1160		- memcpy(dc->sb.set_uuid, c->sb.set_uuid, 16);
	1295	+ memcpy(dc->sb.set_uuid, c->set_uuid, 16);
1161	1296	SET_BDEV_STATE(&dc->sb, BDEV_STATE_CLEAN);
1162	1297
1163	1298	bch_write_bdev_super(dc, &cl);
..	..	@@ -1182,6 +1317,8 @@
1182	1317	down_write(&dc->writeback_lock);
1183	1318	if (bch_cached_dev_writeback_start(dc)) {
1184	1319	up_write(&dc->writeback_lock);
	1320	+ pr_err("Couldn't start writeback facilities for %s\n",
	1321	+ dc->disk.disk->disk_name);
1185	1322	return -ENOMEM;
1186	1323	}
1187	1324
..	..	@@ -1192,20 +1329,42 @@
1192	1329
1193	1330	bch_sectors_dirty_init(&dc->disk);
1194	1331
1195		- bch_cached_dev_run(dc);
	1332	+ ret = bch_cached_dev_run(dc);
	1333	+ if (ret && (ret != -EBUSY)) {
	1334	+ up_write(&dc->writeback_lock);
	1335	+ /*
	1336	+ * bch_register_lock is held, bcache_device_stop() is not
	1337	+ * able to be directly called. The kthread and kworker
	1338	+ * created previously in bch_cached_dev_writeback_start()
	1339	+ * have to be stopped manually here.
	1340	+ */
	1341	+ kthread_stop(dc->writeback_thread);
	1342	+ cancel_writeback_rate_update_dwork(dc);
	1343	+ pr_err("Couldn't run cached device %s\n",
	1344	+ dc->backing_dev_name);
	1345	+ return ret;
	1346	+ }
	1347	+
1196	1348	bcache_device_link(&dc->disk, c, "bdev");
1197	1349	atomic_inc(&c->attached_dev_nr);
	1350	+
	1351	+ if (bch_has_feature_obso_large_bucket(&(c->cache->sb))) {
	1352	+ pr_err("The obsoleted large bucket layout is unsupported, set the bcache device into read-only\n");
	1353	+ pr_err("Please update to the latest bcache-tools to create the cache device\n");
	1354	+ set_disk_ro(dc->disk.disk, 1);
	1355	+ }
1198	1356
1199	1357	/* Allow the writeback thread to proceed */
1200	1358	up_write(&dc->writeback_lock);
1201	1359
1202		- pr_info("Caching %s as %s on set %pU",
	1360	+ pr_info("Caching %s as %s on set %pU\n",
1203	1361	dc->backing_dev_name,
1204	1362	dc->disk.disk->disk_name,
1205		- dc->disk.c->sb.set_uuid);
	1363	+ dc->disk.c->set_uuid);
1206	1364	return 0;
1207	1365	}
1208	1366
	1367	+/* when dc->disk.kobj released */
1209	1368	void bch_cached_dev_release(struct kobject *kobj)
1210	1369	{
1211	1370	struct cached_dev *dc = container_of(kobj, struct cached_dev,
..	..	@@ -1235,8 +1394,8 @@
1235	1394
1236	1395	mutex_unlock(&bch_register_lock);
1237	1396
1238		- if (dc->sb_bio.bi_inline_vecs[0].bv_page)
1239		- put_page(bio_first_page_all(&dc->sb_bio));
	1397	+ if (dc->sb_disk)
	1398	+ put_page(virt_to_page(dc->sb_disk));
1240	1399
1241	1400	if (!IS_ERR_OR_NULL(dc->bdev))
1242	1401	blkdev_put(dc->bdev, FMODE_READ\|FMODE_WRITE\|FMODE_EXCL);
..	..	@@ -1292,13 +1451,13 @@
1292	1451	q->limits.raid_partial_stripes_expensive;
1293	1452
1294	1453	ret = bcache_device_init(&dc->disk, block_size,
1295		- dc->bdev->bd_part->nr_sects - dc->sb.data_offset);
	1454	+ dc->bdev->bd_part->nr_sects - dc->sb.data_offset,
	1455	+ dc->bdev, &bcache_cached_ops);
1296	1456	if (ret)
1297	1457	return ret;
1298	1458
1299		- dc->disk.disk->queue->backing_dev_info->ra_pages =
1300		- max(dc->disk.disk->queue->backing_dev_info->ra_pages,
1301		- q->backing_dev_info->ra_pages);
	1459	+ blk_queue_io_opt(dc->disk.disk->queue,
	1460	+ max(queue_io_opt(dc->disk.disk->queue), queue_io_opt(q)));
1302	1461
1303	1462	atomic_set(&dc->io_errors, 0);
1304	1463	dc->io_disable = false;
..	..	@@ -1313,22 +1472,19 @@
1313	1472
1314	1473	/* Cached device - bcache superblock */
1315	1474
1316		-static void register_bdev(struct cache_sb sb, struct page sb_page,
	1475	+static int register_bdev(struct cache_sb sb, struct cache_sb_disk sb_disk,
1317	1476	struct block_device *bdev,
1318	1477	struct cached_dev *dc)
1319	1478	{
1320	1479	const char *err = "cannot allocate memory";
1321	1480	struct cache_set *c;
	1481	+ int ret = -ENOMEM;
1322	1482
1323	1483	bdevname(bdev, dc->backing_dev_name);
1324	1484	memcpy(&dc->sb, sb, sizeof(struct cache_sb));
1325	1485	dc->bdev = bdev;
1326	1486	dc->bdev->bd_holder = dc;
1327		-
1328		- bio_init(&dc->sb_bio, dc->sb_bio.bi_inline_vecs, 1);
1329		- bio_first_bvec_all(&dc->sb_bio)->bv_page = sb_page;
1330		- get_page(sb_page);
1331		-
	1487	+ dc->sb_disk = sb_disk;
1332	1488
1333	1489	if (cached_dev_init(dc, sb->block_size << 9))
1334	1490	goto err;
..	..	@@ -1340,7 +1496,7 @@
1340	1496	if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj))
1341	1497	goto err;
1342	1498
1343		- pr_info("registered backing device %s", dc->backing_dev_name);
	1499	+ pr_info("registered backing device %s\n", dc->backing_dev_name);
1344	1500
1345	1501	list_add(&dc->list, &uncached_devices);
1346	1502	/* attach to a matched cache set if it exists */
..	..	@@ -1348,17 +1504,23 @@
1348	1504	bch_cached_dev_attach(dc, c, NULL);
1349	1505
1350	1506	if (BDEV_STATE(&dc->sb) == BDEV_STATE_NONE \|\|
1351		- BDEV_STATE(&dc->sb) == BDEV_STATE_STALE)
1352		- bch_cached_dev_run(dc);
	1507	+ BDEV_STATE(&dc->sb) == BDEV_STATE_STALE) {
	1508	+ err = "failed to run cached device";
	1509	+ ret = bch_cached_dev_run(dc);
	1510	+ if (ret)
	1511	+ goto err;
	1512	+ }
1353	1513
1354		- return;
	1514	+ return 0;
1355	1515	err:
1356		- pr_notice("error %s: %s", dc->backing_dev_name, err);
	1516	+ pr_notice("error %s: %s\n", dc->backing_dev_name, err);
1357	1517	bcache_device_stop(&dc->disk);
	1518	+ return ret;
1358	1519	}
1359	1520
1360	1521	/* Flash only volumes */
1361	1522
	1523	+/* When d->kobj released */
1362	1524	void bch_flash_dev_release(struct kobject *kobj)
1363	1525	{
1364	1526	struct bcache_device *d = container_of(kobj, struct bcache_device,
..	..	@@ -1401,7 +1563,8 @@
1401	1563
1402	1564	kobject_init(&d->kobj, &bch_flash_dev_ktype);
1403	1565
1404		- if (bcache_device_init(d, block_bytes(c), u->sectors))
	1566	+ if (bcache_device_init(d, block_bytes(c->cache), u->sectors,
	1567	+ NULL, &bcache_flash_ops))
1405	1568	goto err;
1406	1569
1407	1570	bcache_device_attach(d, c, u - c->uuids);
..	..	@@ -1413,6 +1576,12 @@
1413	1576	goto err;
1414	1577
1415	1578	bcache_device_link(d, c, "volume");
	1579	+
	1580	+ if (bch_has_feature_obso_large_bucket(&c->cache->sb)) {
	1581	+ pr_err("The obsoleted large bucket layout is unsupported, set the bcache device into read-only\n");
	1582	+ pr_err("Please update to the latest bcache-tools to create the cache device\n");
	1583	+ set_disk_ro(d->disk, 1);
	1584	+ }
1416	1585
1417	1586	return 0;
1418	1587	err:
..	..	@@ -1446,7 +1615,7 @@
1446	1615
1447	1616	u = uuid_find_empty(c);
1448	1617	if (!u) {
1449		- pr_err("Can't create volume, no room for UUID");
	1618	+ pr_err("Can't create volume, no room for UUID\n");
1450	1619	return -EINVAL;
1451	1620	}
1452	1621
..	..	@@ -1472,7 +1641,7 @@
1472	1641	smp_mb();
1473	1642
1474	1643	pr_err("stop %s: too many IO errors on backing device %s\n",
1475		- dc->disk.disk->disk_name, dc->backing_dev_name);
	1644	+ dc->disk.disk->disk_name, dc->backing_dev_name);
1476	1645
1477	1646	bcache_device_stop(&dc->disk);
1478	1647	return true;
..	..	@@ -1483,6 +1652,7 @@
1483	1652	__printf(2, 3)
1484	1653	bool bch_cache_set_error(struct cache_set c, const char fmt, ...)
1485	1654	{
	1655	+ struct va_format vaf;
1486	1656	va_list args;
1487	1657
1488	1658	if (c->on_error != ON_ERROR_PANIC &&
..	..	@@ -1490,20 +1660,22 @@
1490	1660	return false;
1491	1661
1492	1662	if (test_and_set_bit(CACHE_SET_IO_DISABLE, &c->flags))
1493		- pr_info("CACHE_SET_IO_DISABLE already set");
	1663	+ pr_info("CACHE_SET_IO_DISABLE already set\n");
1494	1664
1495	1665	/*
1496	1666	* XXX: we can be called from atomic context
1497	1667	* acquire_console_sem();
1498	1668	*/
1499	1669
1500		- pr_err("bcache: error on %pU: ", c->sb.set_uuid);
1501		-
1502	1670	va_start(args, fmt);
1503		- vprintk(fmt, args);
1504		- va_end(args);
1505	1671
1506		- pr_err(", disabling caching\n");
	1672	+ vaf.fmt = fmt;
	1673	+ vaf.va = &args;
	1674	+
	1675	+ pr_err("error on %pU: %pV, disabling caching\n",
	1676	+ c->set_uuid, &vaf);
	1677	+
	1678	+ va_end(args);
1507	1679
1508	1680	if (c->on_error == ON_ERROR_PANIC)
1509	1681	panic("panic forced after error\n");
..	..	@@ -1512,6 +1684,7 @@
1512	1684	return true;
1513	1685	}
1514	1686
	1687	+/* When c->kobj released */
1515	1688	void bch_cache_set_release(struct kobject *kobj)
1516	1689	{
1517	1690	struct cache_set *c = container_of(kobj, struct cache_set, kobj);
..	..	@@ -1524,7 +1697,6 @@
1524	1697	{
1525	1698	struct cache_set *c = container_of(cl, struct cache_set, cl);
1526	1699	struct cache *ca;
1527		- unsigned int i;
1528	1700
1529	1701	debugfs_remove(c->debug);
1530	1702
..	..	@@ -1533,15 +1705,16 @@
1533	1705	bch_journal_free(c);
1534	1706
1535	1707	mutex_lock(&bch_register_lock);
1536		- for_each_cache(ca, c, i)
1537		- if (ca) {
1538		- ca->set = NULL;
1539		- c->cache[ca->sb.nr_this_dev] = NULL;
1540		- kobject_put(&ca->kobj);
1541		- }
1542		-
1543	1708	bch_bset_sort_state_free(&c->sort);
1544		- free_pages((unsigned long) c->uuids, ilog2(bucket_pages(c)));
	1709	+ free_pages((unsigned long) c->uuids, ilog2(meta_bucket_pages(&c->cache->sb)));
	1710	+
	1711	+ ca = c->cache;
	1712	+ if (ca) {
	1713	+ ca->set = NULL;
	1714	+ c->cache = NULL;
	1715	+ kobject_put(&ca->kobj);
	1716	+ }
	1717	+
1545	1718
1546	1719	if (c->moving_gc_wq)
1547	1720	destroy_workqueue(c->moving_gc_wq);
..	..	@@ -1554,7 +1727,7 @@
1554	1727	list_del(&c->list);
1555	1728	mutex_unlock(&bch_register_lock);
1556	1729
1557		- pr_info("Cache set %pU unregistered", c->sb.set_uuid);
	1730	+ pr_info("Cache set %pU unregistered\n", c->set_uuid);
1558	1731	wake_up(&unregister_wait);
1559	1732
1560	1733	closure_debug_destroy(&c->cl);
..	..	@@ -1564,9 +1737,8 @@
1564	1737	static void cache_set_flush(struct closure *cl)
1565	1738	{
1566	1739	struct cache_set *c = container_of(cl, struct cache_set, caching);
1567		- struct cache *ca;
	1740	+ struct cache *ca = c->cache;
1568	1741	struct btree *b;
1569		- unsigned int i;
1570	1742
1571	1743	bch_cache_accounting_destroy(&c->accounting);
1572	1744
..	..	@@ -1576,20 +1748,23 @@
1576	1748	if (!IS_ERR_OR_NULL(c->gc_thread))
1577	1749	kthread_stop(c->gc_thread);
1578	1750
1579		- if (!IS_ERR_OR_NULL(c->root))
	1751	+ if (!IS_ERR(c->root))
1580	1752	list_add(&c->root->list, &c->btree_cache);
1581	1753
1582		- /* Should skip this if we're unregistering because of an error */
1583		- list_for_each_entry(b, &c->btree_cache, list) {
1584		- mutex_lock(&b->write_lock);
1585		- if (btree_node_dirty(b))
1586		- __bch_btree_node_write(b, NULL);
1587		- mutex_unlock(&b->write_lock);
1588		- }
	1754	+ /*
	1755	+ * Avoid flushing cached nodes if cache set is retiring
	1756	+ * due to too many I/O errors detected.
	1757	+ */
	1758	+ if (!test_bit(CACHE_SET_IO_DISABLE, &c->flags))
	1759	+ list_for_each_entry(b, &c->btree_cache, list) {
	1760	+ mutex_lock(&b->write_lock);
	1761	+ if (btree_node_dirty(b))
	1762	+ __bch_btree_node_write(b, NULL);
	1763	+ mutex_unlock(&b->write_lock);
	1764	+ }
1589	1765
1590		- for_each_cache(ca, c, i)
1591		- if (ca->alloc_thread)
1592		- kthread_stop(ca->alloc_thread);
	1766	+ if (ca->alloc_thread)
	1767	+ kthread_stop(ca->alloc_thread);
1593	1768
1594	1769	if (c->journal.cur) {
1595	1770	cancel_delayed_work_sync(&c->journal.work);
..	..	@@ -1621,37 +1796,37 @@
1621	1796	struct cached_dev *dc)
1622	1797	{
1623	1798	if (dc->stop_when_cache_set_failed == BCH_CACHED_DEV_STOP_ALWAYS) {
1624		- pr_warn("stop_when_cache_set_failed of %s is \"always\", stop it for failed cache set %pU.",
1625		- d->disk->disk_name, c->sb.set_uuid);
	1799	+ pr_warn("stop_when_cache_set_failed of %s is \"always\", stop it for failed cache set %pU.\n",
	1800	+ d->disk->disk_name, c->set_uuid);
1626	1801	bcache_device_stop(d);
1627	1802	} else if (atomic_read(&dc->has_dirty)) {
1628	1803	/*
1629	1804	* dc->stop_when_cache_set_failed == BCH_CACHED_STOP_AUTO
1630	1805	* and dc->has_dirty == 1
1631	1806	*/
1632		- pr_warn("stop_when_cache_set_failed of %s is \"auto\" and cache is dirty, stop it to avoid potential data corruption.",
	1807	+ pr_warn("stop_when_cache_set_failed of %s is \"auto\" and cache is dirty, stop it to avoid potential data corruption.\n",
1633	1808	d->disk->disk_name);
1634		- /*
1635		- * There might be a small time gap that cache set is
1636		- * released but bcache device is not. Inside this time
1637		- * gap, regular I/O requests will directly go into
1638		- * backing device as no cache set attached to. This
1639		- * behavior may also introduce potential inconsistence
1640		- * data in writeback mode while cache is dirty.
1641		- * Therefore before calling bcache_device_stop() due
1642		- * to a broken cache device, dc->io_disable should be
1643		- * explicitly set to true.
1644		- */
1645		- dc->io_disable = true;
1646		- /* make others know io_disable is true earlier */
1647		- smp_mb();
1648		- bcache_device_stop(d);
	1809	+ /*
	1810	+ * There might be a small time gap that cache set is
	1811	+ * released but bcache device is not. Inside this time
	1812	+ * gap, regular I/O requests will directly go into
	1813	+ * backing device as no cache set attached to. This
	1814	+ * behavior may also introduce potential inconsistence
	1815	+ * data in writeback mode while cache is dirty.
	1816	+ * Therefore before calling bcache_device_stop() due
	1817	+ * to a broken cache device, dc->io_disable should be
	1818	+ * explicitly set to true.
	1819	+ */
	1820	+ dc->io_disable = true;
	1821	+ /* make others know io_disable is true earlier */
	1822	+ smp_mb();
	1823	+ bcache_device_stop(d);
1649	1824	} else {
1650	1825	/*
1651	1826	* dc->stop_when_cache_set_failed == BCH_CACHED_STOP_AUTO
1652	1827	* and dc->has_dirty == 0
1653	1828	*/
1654		- pr_warn("stop_when_cache_set_failed of %s is \"auto\" and cache is clean, keep it alive.",
	1829	+ pr_warn("stop_when_cache_set_failed of %s is \"auto\" and cache is clean, keep it alive.\n",
1655	1830	d->disk->disk_name);
1656	1831	}
1657	1832	}
..	..	@@ -1689,6 +1864,7 @@
1689	1864	void bch_cache_set_stop(struct cache_set *c)
1690	1865	{
1691	1866	if (!test_and_set_bit(CACHE_SET_STOPPING, &c->flags))
	1867	+ /* closure_fn set to __cache_set_unregister() */
1692	1868	closure_queue(&c->caching);
1693	1869	}
1694	1870
..	..	@@ -1698,12 +1874,13 @@
1698	1874	bch_cache_set_stop(c);
1699	1875	}
1700	1876
1701		-#define alloc_bucket_pages(gfp, c) \
1702		- ((void *) __get_free_pages(__GFP_ZERO\|__GFP_COMP\|gfp, ilog2(bucket_pages(c))))
	1877	+#define alloc_meta_bucket_pages(gfp, sb) \
	1878	+ ((void *) __get_free_pages(__GFP_ZERO\|__GFP_COMP\|gfp, ilog2(meta_bucket_pages(sb))))
1703	1879
1704	1880	struct cache_set bch_cache_set_alloc(struct cache_sb sb)
1705	1881	{
1706	1882	int iter_size;
	1883	+ struct cache *ca = container_of(sb, struct cache, sb);
1707	1884	struct cache_set *c = kzalloc(sizeof(struct cache_set), GFP_KERNEL);
1708	1885
1709	1886	if (!c)
..	..	@@ -1725,17 +1902,16 @@
1725	1902
1726	1903	bch_cache_accounting_init(&c->accounting, &c->cl);
1727	1904
1728		- memcpy(c->sb.set_uuid, sb->set_uuid, 16);
1729		- c->sb.block_size = sb->block_size;
1730		- c->sb.bucket_size = sb->bucket_size;
1731		- c->sb.nr_in_set = sb->nr_in_set;
1732		- c->sb.last_mount = sb->last_mount;
	1905	+ memcpy(c->set_uuid, sb->set_uuid, 16);
	1906	+
	1907	+ c->cache = ca;
	1908	+ c->cache->set = c;
1733	1909	c->bucket_bits = ilog2(sb->bucket_size);
1734	1910	c->block_bits = ilog2(sb->block_size);
1735		- c->nr_uuids = bucket_bytes(c) / sizeof(struct uuid_entry);
	1911	+ c->nr_uuids = meta_bucket_bytes(sb) / sizeof(struct uuid_entry);
1736	1912	c->devices_max_used = 0;
1737	1913	atomic_set(&c->attached_dev_nr, 0);
1738		- c->btree_pages = bucket_pages(c);
	1914	+ c->btree_pages = meta_bucket_pages(sb);
1739	1915	if (c->btree_pages > BTREE_MAX_PAGES)
1740	1916	c->btree_pages = max_t(int, c->btree_pages / 4,
1741	1917	BTREE_MAX_PAGES);
..	..	@@ -1761,29 +1937,52 @@
1761	1937	INIT_LIST_HEAD(&c->btree_cache_freed);
1762	1938	INIT_LIST_HEAD(&c->data_buckets);
1763	1939
1764		- iter_size = (sb->bucket_size / sb->block_size + 1) *
	1940	+ iter_size = ((meta_bucket_pages(sb) * PAGE_SECTORS) / sb->block_size + 1) *
1765	1941	sizeof(struct btree_iter_set);
1766	1942
1767		- if (!(c->devices = kcalloc(c->nr_uuids, sizeof(void *), GFP_KERNEL)) \|\|
1768		- mempool_init_slab_pool(&c->search, 32, bch_search_cache) \|\|
1769		- mempool_init_kmalloc_pool(&c->bio_meta, 2,
1770		- sizeof(struct bbio) + sizeof(struct bio_vec) *
1771		- bucket_pages(c)) \|\|
1772		- mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) \|\|
1773		- bioset_init(&c->bio_split, 4, offsetof(struct bbio, bio),
1774		- BIOSET_NEED_BVECS\|BIOSET_NEED_RESCUER) \|\|
1775		- !(c->uuids = alloc_bucket_pages(GFP_KERNEL, c)) \|\|
1776		- !(c->moving_gc_wq = alloc_workqueue("bcache_gc",
1777		- WQ_MEM_RECLAIM, 0)) \|\|
1778		- bch_journal_alloc(c) \|\|
1779		- bch_btree_cache_alloc(c) \|\|
1780		- bch_open_buckets_alloc(c) \|\|
1781		- bch_bset_sort_state_init(&c->sort, ilog2(c->btree_pages)))
	1943	+ c->devices = kcalloc(c->nr_uuids, sizeof(void *), GFP_KERNEL);
	1944	+ if (!c->devices)
	1945	+ goto err;
	1946	+
	1947	+ if (mempool_init_slab_pool(&c->search, 32, bch_search_cache))
	1948	+ goto err;
	1949	+
	1950	+ if (mempool_init_kmalloc_pool(&c->bio_meta, 2,
	1951	+ sizeof(struct bbio) +
	1952	+ sizeof(struct bio_vec) * meta_bucket_pages(sb)))
	1953	+ goto err;
	1954	+
	1955	+ if (mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size))
	1956	+ goto err;
	1957	+
	1958	+ if (bioset_init(&c->bio_split, 4, offsetof(struct bbio, bio),
	1959	+ BIOSET_NEED_BVECS\|BIOSET_NEED_RESCUER))
	1960	+ goto err;
	1961	+
	1962	+ c->uuids = alloc_meta_bucket_pages(GFP_KERNEL, sb);
	1963	+ if (!c->uuids)
	1964	+ goto err;
	1965	+
	1966	+ c->moving_gc_wq = alloc_workqueue("bcache_gc", WQ_MEM_RECLAIM, 0);
	1967	+ if (!c->moving_gc_wq)
	1968	+ goto err;
	1969	+
	1970	+ if (bch_journal_alloc(c))
	1971	+ goto err;
	1972	+
	1973	+ if (bch_btree_cache_alloc(c))
	1974	+ goto err;
	1975	+
	1976	+ if (bch_open_buckets_alloc(c))
	1977	+ goto err;
	1978	+
	1979	+ if (bch_bset_sort_state_init(&c->sort, ilog2(c->btree_pages)))
1782	1980	goto err;
1783	1981
1784	1982	c->congested_read_threshold_us = 2000;
1785	1983	c->congested_write_threshold_us = 20000;
1786	1984	c->error_limit = DEFAULT_IO_ERROR_LIMIT;
	1985	+ c->idle_max_writeback_rate_enabled = 1;
1787	1986	WARN_ON(test_and_clear_bit(CACHE_SET_IO_DISABLE, &c->flags));
1788	1987
1789	1988	return c;
..	..	@@ -1796,19 +1995,17 @@
1796	1995	{
1797	1996	const char *err = "cannot allocate memory";
1798	1997	struct cached_dev dc, t;
1799		- struct cache *ca;
	1998	+ struct cache *ca = c->cache;
1800	1999	struct closure cl;
1801		- unsigned int i;
1802	2000	LIST_HEAD(journal);
1803	2001	struct journal_replay *l;
1804	2002
1805	2003	closure_init_stack(&cl);
1806	2004
1807		- for_each_cache(ca, c, i)
1808		- c->nbuckets += ca->sb.nbuckets;
	2005	+ c->nbuckets = ca->sb.nbuckets;
1809	2006	set_gc_sectors(c);
1810	2007
1811		- if (CACHE_SYNC(&c->sb)) {
	2008	+ if (CACHE_SYNC(&c->cache->sb)) {
1812	2009	struct bkey *k;
1813	2010	struct jset *j;
1814	2011
..	..	@@ -1816,7 +2013,7 @@
1816	2013	if (bch_journal_read(c, &journal))
1817	2014	goto err;
1818	2015
1819		- pr_debug("btree_journal_read() done");
	2016	+ pr_debug("btree_journal_read() done\n");
1820	2017
1821	2018	err = "no journal entries found";
1822	2019	if (list_empty(&journal))
..	..	@@ -1825,8 +2022,8 @@
1825	2022	j = &list_entry(journal.prev, struct journal_replay, list)->j;
1826	2023
1827	2024	err = "IO error reading priorities";
1828		- for_each_cache(ca, c, i)
1829		- prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev]);
	2025	+ if (prio_read(ca, j->prio_bucket[ca->sb.nr_this_dev]))
	2026	+ goto err;
1830	2027
1831	2028	/*
1832	2029	* If prio_read() fails it'll call cache_set_error and we'll
..	..	@@ -1860,7 +2057,7 @@
1860	2057
1861	2058	bch_journal_mark(c, &journal);
1862	2059	bch_initial_gc_finish(c);
1863		- pr_debug("btree_check() done");
	2060	+ pr_debug("btree_check() done\n");
1864	2061
1865	2062	/*
1866	2063	* bcache_journal_next() can't happen sooner, or
..	..	@@ -1870,9 +2067,8 @@
1870	2067	bch_journal_next(&c->journal);
1871	2068
1872	2069	err = "error starting allocator thread";
1873		- for_each_cache(ca, c, i)
1874		- if (bch_cache_allocator_start(ca))
1875		- goto err;
	2070	+ if (bch_cache_allocator_start(ca))
	2071	+ goto err;
1876	2072
1877	2073	/*
1878	2074	* First place it's safe to allocate: btree_check() and
..	..	@@ -1891,28 +2087,23 @@
1891	2087	if (bch_journal_replay(c, &journal))
1892	2088	goto err;
1893	2089	} else {
1894		- pr_notice("invalidating existing data");
	2090	+ unsigned int j;
1895	2091
1896		- for_each_cache(ca, c, i) {
1897		- unsigned int j;
	2092	+ pr_notice("invalidating existing data\n");
	2093	+ ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7,
	2094	+ 2, SB_JOURNAL_BUCKETS);
1898	2095
1899		- ca->sb.keys = clamp_t(int, ca->sb.nbuckets >> 7,
1900		- 2, SB_JOURNAL_BUCKETS);
1901		-
1902		- for (j = 0; j < ca->sb.keys; j++)
1903		- ca->sb.d[j] = ca->sb.first_bucket + j;
1904		- }
	2096	+ for (j = 0; j < ca->sb.keys; j++)
	2097	+ ca->sb.d[j] = ca->sb.first_bucket + j;
1905	2098
1906	2099	bch_initial_gc_finish(c);
1907	2100
1908	2101	err = "error starting allocator thread";
1909		- for_each_cache(ca, c, i)
1910		- if (bch_cache_allocator_start(ca))
1911		- goto err;
	2102	+ if (bch_cache_allocator_start(ca))
	2103	+ goto err;
1912	2104
1913	2105	mutex_lock(&c->bucket_lock);
1914		- for_each_cache(ca, c, i)
1915		- bch_prio_write(ca, true);
	2106	+ bch_prio_write(ca, true);
1916	2107	mutex_unlock(&c->bucket_lock);
1917	2108
1918	2109	err = "cannot allocate new UUID bucket";
..	..	@@ -1921,7 +2112,7 @@
1921	2112
1922	2113	err = "cannot allocate new btree root";
1923	2114	c->root = __bch_btree_node_alloc(c, NULL, 0, true, NULL);
1924		- if (IS_ERR_OR_NULL(c->root))
	2115	+ if (IS_ERR(c->root))
1925	2116	goto err;
1926	2117
1927	2118	mutex_lock(&c->root->write_lock);
..	..	@@ -1937,7 +2128,7 @@
1937	2128	* everything is set up - fortunately journal entries won't be
1938	2129	* written until the SET_CACHE_SYNC() here:
1939	2130	*/
1940		- SET_CACHE_SYNC(&c->sb, true);
	2131	+ SET_CACHE_SYNC(&c->cache->sb, true);
1941	2132
1942	2133	bch_journal_next(&c->journal);
1943	2134	bch_journal_meta(c, &cl);
..	..	@@ -1948,14 +2139,18 @@
1948	2139	goto err;
1949	2140
1950	2141	closure_sync(&cl);
1951		- c->sb.last_mount = (u32)ktime_get_real_seconds();
	2142	+ c->cache->sb.last_mount = (u32)ktime_get_real_seconds();
1952	2143	bcache_write_super(c);
	2144	+
	2145	+ if (bch_has_feature_obso_large_bucket(&c->cache->sb))
	2146	+ pr_err("Detect obsoleted large bucket layout, all attached bcache device will be read-only\n");
1953	2147
1954	2148	list_for_each_entry_safe(dc, t, &uncached_devices, list)
1955	2149	bch_cached_dev_attach(dc, c, NULL);
1956	2150
1957	2151	flash_devs_run(c);
1958	2152
	2153	+ bch_journal_space_reserve(&c->journal);
1959	2154	set_bit(CACHE_SET_RUNNING, &c->flags);
1960	2155	return 0;
1961	2156	err:
..	..	@@ -1966,17 +2161,10 @@
1966	2161	}
1967	2162
1968	2163	closure_sync(&cl);
1969		- /* XXX: test this, it's broken */
	2164	+
1970	2165	bch_cache_set_error(c, "%s", err);
1971	2166
1972	2167	return -EIO;
1973		-}
1974		-
1975		-static bool can_attach_cache(struct cache ca, struct cache_set c)
1976		-{
1977		- return ca->sb.block_size == c->sb.block_size &&
1978		- ca->sb.bucket_size == c->sb.bucket_size &&
1979		- ca->sb.nr_in_set == c->sb.nr_in_set;
1980	2168	}
1981	2169
1982	2170	static const char register_cache_set(struct cache ca)
..	..	@@ -1986,15 +2174,9 @@
1986	2174	struct cache_set *c;
1987	2175
1988	2176	list_for_each_entry(c, &bch_cache_sets, list)
1989		- if (!memcmp(c->sb.set_uuid, ca->sb.set_uuid, 16)) {
1990		- if (c->cache[ca->sb.nr_this_dev])
	2177	+ if (!memcmp(c->set_uuid, ca->sb.set_uuid, 16)) {
	2178	+ if (c->cache)
1991	2179	return "duplicate cache set member";
1992		-
1993		- if (!can_attach_cache(ca, c))
1994		- return "cache sb does not match set";
1995		-
1996		- if (!CACHE_SYNC(&ca->sb))
1997		- SET_CACHE_SYNC(&c->sb, false);
1998	2180
1999	2181	goto found;
2000	2182	}
..	..	@@ -2004,7 +2186,7 @@
2004	2186	return err;
2005	2187
2006	2188	err = "error creating kobject";
2007		- if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->sb.set_uuid) \|\|
	2189	+ if (kobject_add(&c->kobj, bcache_kobj, "%pU", c->set_uuid) \|\|
2008	2190	kobject_add(&c->internal, &c->kobj, "internal"))
2009	2191	goto err;
2010	2192
..	..	@@ -2020,31 +2202,13 @@
2020	2202	sysfs_create_link(&c->kobj, &ca->kobj, buf))
2021	2203	goto err;
2022	2204
2023		- /*
2024		- * A special case is both ca->sb.seq and c->sb.seq are 0,
2025		- * such condition happens on a new created cache device whose
2026		- * super block is never flushed yet. In this case c->sb.version
2027		- * and other members should be updated too, otherwise we will
2028		- * have a mistaken super block version in cache set.
2029		- */
2030		- if (ca->sb.seq > c->sb.seq \|\| c->sb.seq == 0) {
2031		- c->sb.version = ca->sb.version;
2032		- memcpy(c->sb.set_uuid, ca->sb.set_uuid, 16);
2033		- c->sb.flags = ca->sb.flags;
2034		- c->sb.seq = ca->sb.seq;
2035		- pr_debug("set version = %llu", c->sb.version);
2036		- }
2037		-
2038	2205	kobject_get(&ca->kobj);
2039	2206	ca->set = c;
2040		- ca->set->cache[ca->sb.nr_this_dev] = ca;
2041		- c->cache_by_alloc[c->caches_loaded++] = ca;
	2207	+ ca->set->cache = ca;
2042	2208
2043		- if (c->caches_loaded == c->sb.nr_in_set) {
2044		- err = "failed to run cache set";
2045		- if (run_cache_set(c) < 0)
2046		- goto err;
2047		- }
	2209	+ err = "failed to run cache set";
	2210	+ if (run_cache_set(c) < 0)
	2211	+ goto err;
2048	2212
2049	2213	return NULL;
2050	2214	err:
..	..	@@ -2054,17 +2218,18 @@
2054	2218
2055	2219	/* Cache device */
2056	2220
	2221	+/* When ca->kobj released */
2057	2222	void bch_cache_release(struct kobject *kobj)
2058	2223	{
2059	2224	struct cache *ca = container_of(kobj, struct cache, kobj);
2060	2225	unsigned int i;
2061	2226
2062	2227	if (ca->set) {
2063		- BUG_ON(ca->set->cache[ca->sb.nr_this_dev] != ca);
2064		- ca->set->cache[ca->sb.nr_this_dev] = NULL;
	2228	+ BUG_ON(ca->set->cache != ca);
	2229	+ ca->set->cache = NULL;
2065	2230	}
2066	2231
2067		- free_pages((unsigned long) ca->disk_buckets, ilog2(bucket_pages(ca)));
	2232	+ free_pages((unsigned long) ca->disk_buckets, ilog2(meta_bucket_pages(&ca->sb)));
2068	2233	kfree(ca->prio_buckets);
2069	2234	vfree(ca->buckets);
2070	2235
..	..	@@ -2074,8 +2239,8 @@
2074	2239	for (i = 0; i < RESERVE_NR; i++)
2075	2240	free_fifo(&ca->free[i]);
2076	2241
2077		- if (ca->sb_bio.bi_inline_vecs[0].bv_page)
2078		- put_page(bio_first_page_all(&ca->sb_bio));
	2242	+ if (ca->sb_disk)
	2243	+ put_page(virt_to_page(ca->sb_disk));
2079	2244
2080	2245	if (!IS_ERR_OR_NULL(ca->bdev))
2081	2246	blkdev_put(ca->bdev, FMODE_READ\|FMODE_WRITE\|FMODE_EXCL);
..	..	@@ -2089,6 +2254,8 @@
2089	2254	size_t free;
2090	2255	size_t btree_buckets;
2091	2256	struct bucket *b;
	2257	+ int ret = -ENOMEM;
	2258	+ const char *err = NULL;
2092	2259
2093	2260	__module_get(THIS_MODULE);
2094	2261	kobject_init(&ca->kobj, &bch_cache_ktype);
..	..	@@ -2106,30 +2273,96 @@
2106	2273	*/
2107	2274	btree_buckets = ca->sb.njournal_buckets ?: 8;
2108	2275	free = roundup_pow_of_two(ca->sb.nbuckets) >> 10;
	2276	+ if (!free) {
	2277	+ ret = -EPERM;
	2278	+ err = "ca->sb.nbuckets is too small";
	2279	+ goto err_free;
	2280	+ }
2109	2281
2110		- if (!init_fifo(&ca->free[RESERVE_BTREE], btree_buckets, GFP_KERNEL) \|\|
2111		- !init_fifo_exact(&ca->free[RESERVE_PRIO], prio_buckets(ca), GFP_KERNEL) \|\|
2112		- !init_fifo(&ca->free[RESERVE_MOVINGGC], free, GFP_KERNEL) \|\|
2113		- !init_fifo(&ca->free[RESERVE_NONE], free, GFP_KERNEL) \|\|
2114		- !init_fifo(&ca->free_inc, free << 2, GFP_KERNEL) \|\|
2115		- !init_heap(&ca->heap, free << 3, GFP_KERNEL) \|\|
2116		- !(ca->buckets = vzalloc(array_size(sizeof(struct bucket),
2117		- ca->sb.nbuckets))) \|\|
2118		- !(ca->prio_buckets = kzalloc(array3_size(sizeof(uint64_t),
2119		- prio_buckets(ca), 2),
2120		- GFP_KERNEL)) \|\|
2121		- !(ca->disk_buckets = alloc_bucket_pages(GFP_KERNEL, ca)))
2122		- return -ENOMEM;
	2282	+ if (!init_fifo(&ca->free[RESERVE_BTREE], btree_buckets,
	2283	+ GFP_KERNEL)) {
	2284	+ err = "ca->free[RESERVE_BTREE] alloc failed";
	2285	+ goto err_btree_alloc;
	2286	+ }
	2287	+
	2288	+ if (!init_fifo_exact(&ca->free[RESERVE_PRIO], prio_buckets(ca),
	2289	+ GFP_KERNEL)) {
	2290	+ err = "ca->free[RESERVE_PRIO] alloc failed";
	2291	+ goto err_prio_alloc;
	2292	+ }
	2293	+
	2294	+ if (!init_fifo(&ca->free[RESERVE_MOVINGGC], free, GFP_KERNEL)) {
	2295	+ err = "ca->free[RESERVE_MOVINGGC] alloc failed";
	2296	+ goto err_movinggc_alloc;
	2297	+ }
	2298	+
	2299	+ if (!init_fifo(&ca->free[RESERVE_NONE], free, GFP_KERNEL)) {
	2300	+ err = "ca->free[RESERVE_NONE] alloc failed";
	2301	+ goto err_none_alloc;
	2302	+ }
	2303	+
	2304	+ if (!init_fifo(&ca->free_inc, free << 2, GFP_KERNEL)) {
	2305	+ err = "ca->free_inc alloc failed";
	2306	+ goto err_free_inc_alloc;
	2307	+ }
	2308	+
	2309	+ if (!init_heap(&ca->heap, free << 3, GFP_KERNEL)) {
	2310	+ err = "ca->heap alloc failed";
	2311	+ goto err_heap_alloc;
	2312	+ }
	2313	+
	2314	+ ca->buckets = vzalloc(array_size(sizeof(struct bucket),
	2315	+ ca->sb.nbuckets));
	2316	+ if (!ca->buckets) {
	2317	+ err = "ca->buckets alloc failed";
	2318	+ goto err_buckets_alloc;
	2319	+ }
	2320	+
	2321	+ ca->prio_buckets = kzalloc(array3_size(sizeof(uint64_t),
	2322	+ prio_buckets(ca), 2),
	2323	+ GFP_KERNEL);
	2324	+ if (!ca->prio_buckets) {
	2325	+ err = "ca->prio_buckets alloc failed";
	2326	+ goto err_prio_buckets_alloc;
	2327	+ }
	2328	+
	2329	+ ca->disk_buckets = alloc_meta_bucket_pages(GFP_KERNEL, &ca->sb);
	2330	+ if (!ca->disk_buckets) {
	2331	+ err = "ca->disk_buckets alloc failed";
	2332	+ goto err_disk_buckets_alloc;
	2333	+ }
2123	2334
2124	2335	ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca);
2125	2336
2126	2337	for_each_bucket(b, ca)
2127	2338	atomic_set(&b->pin, 0);
2128		-
2129	2339	return 0;
	2340	+
	2341	+err_disk_buckets_alloc:
	2342	+ kfree(ca->prio_buckets);
	2343	+err_prio_buckets_alloc:
	2344	+ vfree(ca->buckets);
	2345	+err_buckets_alloc:
	2346	+ free_heap(&ca->heap);
	2347	+err_heap_alloc:
	2348	+ free_fifo(&ca->free_inc);
	2349	+err_free_inc_alloc:
	2350	+ free_fifo(&ca->free[RESERVE_NONE]);
	2351	+err_none_alloc:
	2352	+ free_fifo(&ca->free[RESERVE_MOVINGGC]);
	2353	+err_movinggc_alloc:
	2354	+ free_fifo(&ca->free[RESERVE_PRIO]);
	2355	+err_prio_alloc:
	2356	+ free_fifo(&ca->free[RESERVE_BTREE]);
	2357	+err_btree_alloc:
	2358	+err_free:
	2359	+ module_put(THIS_MODULE);
	2360	+ if (err)
	2361	+ pr_notice("error %s: %s\n", ca->cache_dev_name, err);
	2362	+ return ret;
2130	2363	}
2131	2364
2132		-static int register_cache(struct cache_sb sb, struct page sb_page,
	2365	+static int register_cache(struct cache_sb sb, struct cache_sb_disk sb_disk,
2133	2366	struct block_device bdev, struct cache ca)
2134	2367	{
2135	2368	const char err = NULL; / must be set for any error case */
..	..	@@ -2139,19 +2372,24 @@
2139	2372	memcpy(&ca->sb, sb, sizeof(struct cache_sb));
2140	2373	ca->bdev = bdev;
2141	2374	ca->bdev->bd_holder = ca;
2142		-
2143		- bio_init(&ca->sb_bio, ca->sb_bio.bi_inline_vecs, 1);
2144		- bio_first_bvec_all(&ca->sb_bio)->bv_page = sb_page;
2145		- get_page(sb_page);
	2375	+ ca->sb_disk = sb_disk;
2146	2376
2147	2377	if (blk_queue_discard(bdev_get_queue(bdev)))
2148	2378	ca->discard = CACHE_DISCARD(&ca->sb);
2149	2379
2150	2380	ret = cache_alloc(ca);
2151	2381	if (ret != 0) {
	2382	+ /*
	2383	+ * If we failed here, it means ca->kobj is not initialized yet,
	2384	+ * kobject_put() won't be called and there is no chance to
	2385	+ * call blkdev_put() to bdev in bch_cache_release(). So we
	2386	+ * explicitly call blkdev_put() here.
	2387	+ */
2152	2388	blkdev_put(bdev, FMODE_READ\|FMODE_WRITE\|FMODE_EXCL);
2153	2389	if (ret == -ENOMEM)
2154	2390	err = "cache_alloc(): -ENOMEM";
	2391	+ else if (ret == -EPERM)
	2392	+ err = "cache_alloc(): cache device is too small";
2155	2393	else
2156	2394	err = "cache_alloc(): unknown error";
2157	2395	goto err;
..	..	@@ -2174,14 +2412,14 @@
2174	2412	goto out;
2175	2413	}
2176	2414
2177		- pr_info("registered cache device %s", ca->cache_dev_name);
	2415	+ pr_info("registered cache device %s\n", ca->cache_dev_name);
2178	2416
2179	2417	out:
2180	2418	kobject_put(&ca->kobj);
2181	2419
2182	2420	err:
2183	2421	if (err)
2184		- pr_notice("error %s: %s", ca->cache_dev_name, err);
	2422	+ pr_notice("error %s: %s\n", ca->cache_dev_name, err);
2185	2423
2186	2424	return ret;
2187	2425	}
..	..	@@ -2190,9 +2428,13 @@
2190	2428
2191	2429	static ssize_t register_bcache(struct kobject k, struct kobj_attribute attr,
2192	2430	const char *buffer, size_t size);
	2431	+static ssize_t bch_pending_bdevs_cleanup(struct kobject *k,
	2432	+ struct kobj_attribute *attr,
	2433	+ const char *buffer, size_t size);
2193	2434
2194	2435	kobj_attribute_write(register, register_bcache);
2195	2436	kobj_attribute_write(register_quiet, register_bcache);
	2437	+kobj_attribute_write(pendings_cleanup, bch_pending_bdevs_cleanup);
2196	2438
2197	2439	static bool bch_is_open_backing(struct block_device *bdev)
2198	2440	{
..	..	@@ -2212,13 +2454,14 @@
2212	2454	static bool bch_is_open_cache(struct block_device *bdev)
2213	2455	{
2214	2456	struct cache_set c, tc;
2215		- struct cache *ca;
2216		- unsigned int i;
2217	2457
2218		- list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
2219		- for_each_cache(ca, c, i)
2220		- if (ca->bdev == bdev)
2221		- return true;
	2458	+ list_for_each_entry_safe(c, tc, &bch_cache_sets, list) {
	2459	+ struct cache *ca = c->cache;
	2460	+
	2461	+ if (ca->bdev == bdev)
	2462	+ return true;
	2463	+ }
	2464	+
2222	2465	return false;
2223	2466	}
2224	2467
..	..	@@ -2227,27 +2470,121 @@
2227	2470	return bch_is_open_cache(bdev) \|\| bch_is_open_backing(bdev);
2228	2471	}
2229	2472
	2473	+struct async_reg_args {
	2474	+ struct delayed_work reg_work;
	2475	+ char *path;
	2476	+ struct cache_sb *sb;
	2477	+ struct cache_sb_disk *sb_disk;
	2478	+ struct block_device *bdev;
	2479	+};
	2480	+
	2481	+static void register_bdev_worker(struct work_struct *work)
	2482	+{
	2483	+ int fail = false;
	2484	+ struct async_reg_args *args =
	2485	+ container_of(work, struct async_reg_args, reg_work.work);
	2486	+ struct cached_dev *dc;
	2487	+
	2488	+ dc = kzalloc(sizeof(*dc), GFP_KERNEL);
	2489	+ if (!dc) {
	2490	+ fail = true;
	2491	+ put_page(virt_to_page(args->sb_disk));
	2492	+ blkdev_put(args->bdev, FMODE_READ \| FMODE_WRITE \| FMODE_EXCL);
	2493	+ goto out;
	2494	+ }
	2495	+
	2496	+ mutex_lock(&bch_register_lock);
	2497	+ if (register_bdev(args->sb, args->sb_disk, args->bdev, dc) < 0)
	2498	+ fail = true;
	2499	+ mutex_unlock(&bch_register_lock);
	2500	+
	2501	+out:
	2502	+ if (fail)
	2503	+ pr_info("error %s: fail to register backing device\n",
	2504	+ args->path);
	2505	+ kfree(args->sb);
	2506	+ kfree(args->path);
	2507	+ kfree(args);
	2508	+ module_put(THIS_MODULE);
	2509	+}
	2510	+
	2511	+static void register_cache_worker(struct work_struct *work)
	2512	+{
	2513	+ int fail = false;
	2514	+ struct async_reg_args *args =
	2515	+ container_of(work, struct async_reg_args, reg_work.work);
	2516	+ struct cache *ca;
	2517	+
	2518	+ ca = kzalloc(sizeof(*ca), GFP_KERNEL);
	2519	+ if (!ca) {
	2520	+ fail = true;
	2521	+ put_page(virt_to_page(args->sb_disk));
	2522	+ blkdev_put(args->bdev, FMODE_READ \| FMODE_WRITE \| FMODE_EXCL);
	2523	+ goto out;
	2524	+ }
	2525	+
	2526	+ /* blkdev_put() will be called in bch_cache_release() */
	2527	+ if (register_cache(args->sb, args->sb_disk, args->bdev, ca) != 0)
	2528	+ fail = true;
	2529	+
	2530	+out:
	2531	+ if (fail)
	2532	+ pr_info("error %s: fail to register cache device\n",
	2533	+ args->path);
	2534	+ kfree(args->sb);
	2535	+ kfree(args->path);
	2536	+ kfree(args);
	2537	+ module_put(THIS_MODULE);
	2538	+}
	2539	+
	2540	+static void register_device_aync(struct async_reg_args *args)
	2541	+{
	2542	+ if (SB_IS_BDEV(args->sb))
	2543	+ INIT_DELAYED_WORK(&args->reg_work, register_bdev_worker);
	2544	+ else
	2545	+ INIT_DELAYED_WORK(&args->reg_work, register_cache_worker);
	2546	+
	2547	+ /* 10 jiffies is enough for a delay */
	2548	+ queue_delayed_work(system_wq, &args->reg_work, 10);
	2549	+}
	2550	+
2230	2551	static ssize_t register_bcache(struct kobject k, struct kobj_attribute attr,
2231	2552	const char *buffer, size_t size)
2232	2553	{
2233		- ssize_t ret = size;
2234		- const char *err = "cannot allocate memory";
	2554	+ const char *err;
2235	2555	char *path = NULL;
2236		- struct cache_sb *sb = NULL;
2237		- struct block_device *bdev = NULL;
2238		- struct page *sb_page = NULL;
	2556	+ struct cache_sb *sb;
	2557	+ struct cache_sb_disk *sb_disk;
	2558	+ struct block_device *bdev;
	2559	+ ssize_t ret;
	2560	+ bool async_registration = false;
2239	2561
	2562	+#ifdef CONFIG_BCACHE_ASYNC_REGISTRATION
	2563	+ async_registration = true;
	2564	+#endif
	2565	+
	2566	+ ret = -EBUSY;
	2567	+ err = "failed to reference bcache module";
2240	2568	if (!try_module_get(THIS_MODULE))
2241		- return -EBUSY;
	2569	+ goto out;
2242	2570
	2571	+ /* For latest state of bcache_is_reboot */
	2572	+ smp_mb();
	2573	+ err = "bcache is in reboot";
	2574	+ if (bcache_is_reboot)
	2575	+ goto out_module_put;
	2576	+
	2577	+ ret = -ENOMEM;
	2578	+ err = "cannot allocate memory";
2243	2579	path = kstrndup(buffer, size, GFP_KERNEL);
2244	2580	if (!path)
2245		- goto err;
	2581	+ goto out_module_put;
2246	2582
2247	2583	sb = kmalloc(sizeof(struct cache_sb), GFP_KERNEL);
2248	2584	if (!sb)
2249		- goto err;
	2585	+ goto out_free_path;
2250	2586
	2587	+ ret = -EINVAL;
2251	2588	err = "failed to open device";
2252	2589	bdev = blkdev_get_by_path(strim(path),
2253	2590	FMODE_READ\|FMODE_WRITE\|FMODE_EXCL,
..	..	@@ -2264,56 +2601,142 @@
2264	2601	if (!IS_ERR(bdev))
2265	2602	bdput(bdev);
2266	2603	if (attr == &ksysfs_register_quiet)
2267		- goto out;
	2604	+ goto done;
2268	2605	}
2269		- goto err;
	2606	+ goto out_free_sb;
2270	2607	}
2271	2608
2272	2609	err = "failed to set blocksize";
2273	2610	if (set_blocksize(bdev, 4096))
2274		- goto err_close;
	2611	+ goto out_blkdev_put;
2275	2612
2276		- err = read_super(sb, bdev, &sb_page);
	2613	+ err = read_super(sb, bdev, &sb_disk);
2277	2614	if (err)
2278		- goto err_close;
	2615	+ goto out_blkdev_put;
2279	2616
2280	2617	err = "failed to register device";
	2618	+
	2619	+ if (async_registration) {
	2620	+ /* register in asynchronous way */
	2621	+ struct async_reg_args *args =
	2622	+ kzalloc(sizeof(struct async_reg_args), GFP_KERNEL);
	2623	+
	2624	+ if (!args) {
	2625	+ ret = -ENOMEM;
	2626	+ err = "cannot allocate memory";
	2627	+ goto out_put_sb_page;
	2628	+ }
	2629	+
	2630	+ args->path = path;
	2631	+ args->sb = sb;
	2632	+ args->sb_disk = sb_disk;
	2633	+ args->bdev = bdev;
	2634	+ register_device_aync(args);
	2635	+ /* No wait and returns to user space */
	2636	+ goto async_done;
	2637	+ }
	2638	+
2281	2639	if (SB_IS_BDEV(sb)) {
2282	2640	struct cached_dev dc = kzalloc(sizeof(dc), GFP_KERNEL);
2283	2641
2284	2642	if (!dc)
2285		- goto err_close;
	2643	+ goto out_put_sb_page;
2286	2644
2287	2645	mutex_lock(&bch_register_lock);
2288		- register_bdev(sb, sb_page, bdev, dc);
	2646	+ ret = register_bdev(sb, sb_disk, bdev, dc);
2289	2647	mutex_unlock(&bch_register_lock);
	2648	+ /* blkdev_put() will be called in cached_dev_free() */
	2649	+ if (ret < 0)
	2650	+ goto out_free_sb;
2290	2651	} else {
2291	2652	struct cache ca = kzalloc(sizeof(ca), GFP_KERNEL);
2292	2653
2293	2654	if (!ca)
2294		- goto err_close;
	2655	+ goto out_put_sb_page;
2295	2656
2296		- if (register_cache(sb, sb_page, bdev, ca) != 0)
2297		- goto err;
	2657	+ /* blkdev_put() will be called in bch_cache_release() */
	2658	+ if (register_cache(sb, sb_disk, bdev, ca) != 0)
	2659	+ goto out_free_sb;
2298	2660	}
2299		-out:
2300		- if (sb_page)
2301		- put_page(sb_page);
	2661	+
	2662	+done:
2302	2663	kfree(sb);
2303	2664	kfree(path);
2304	2665	module_put(THIS_MODULE);
2305		- return ret;
	2666	+async_done:
	2667	+ return size;
2306	2668
2307		-err_close:
2308		- blkdev_put(bdev, FMODE_READ\|FMODE_WRITE\|FMODE_EXCL);
2309		-err:
2310		- pr_info("error %s: %s", path, err);
2311		- ret = -EINVAL;
2312		- goto out;
	2669	+out_put_sb_page:
	2670	+ put_page(virt_to_page(sb_disk));
	2671	+out_blkdev_put:
	2672	+ blkdev_put(bdev, FMODE_READ \| FMODE_WRITE \| FMODE_EXCL);
	2673	+out_free_sb:
	2674	+ kfree(sb);
	2675	+out_free_path:
	2676	+ kfree(path);
	2677	+ path = NULL;
	2678	+out_module_put:
	2679	+ module_put(THIS_MODULE);
	2680	+out:
	2681	+ pr_info("error %s: %s\n", path?path:"", err);
	2682	+ return ret;
	2683	+}
	2684	+
	2685	+
	2686	+struct pdev {
	2687	+ struct list_head list;
	2688	+ struct cached_dev *dc;
	2689	+};
	2690	+
	2691	+static ssize_t bch_pending_bdevs_cleanup(struct kobject *k,
	2692	+ struct kobj_attribute *attr,
	2693	+ const char *buffer,
	2694	+ size_t size)
	2695	+{
	2696	+ LIST_HEAD(pending_devs);
	2697	+ ssize_t ret = size;
	2698	+ struct cached_dev dc, tdc;
	2699	+ struct pdev pdev, tpdev;
	2700	+ struct cache_set c, tc;
	2701	+
	2702	+ mutex_lock(&bch_register_lock);
	2703	+ list_for_each_entry_safe(dc, tdc, &uncached_devices, list) {
	2704	+ pdev = kmalloc(sizeof(struct pdev), GFP_KERNEL);
	2705	+ if (!pdev)
	2706	+ break;
	2707	+ pdev->dc = dc;
	2708	+ list_add(&pdev->list, &pending_devs);
	2709	+ }
	2710	+
	2711	+ list_for_each_entry_safe(pdev, tpdev, &pending_devs, list) {
	2712	+ list_for_each_entry_safe(c, tc, &bch_cache_sets, list) {
	2713	+ char *pdev_set_uuid = pdev->dc->sb.set_uuid;
	2714	+ char *set_uuid = c->set_uuid;
	2715	+
	2716	+ if (!memcmp(pdev_set_uuid, set_uuid, 16)) {
	2717	+ list_del(&pdev->list);
	2718	+ kfree(pdev);
	2719	+ break;
	2720	+ }
	2721	+ }
	2722	+ }
	2723	+ mutex_unlock(&bch_register_lock);
	2724	+
	2725	+ list_for_each_entry_safe(pdev, tpdev, &pending_devs, list) {
	2726	+ pr_info("delete pdev %p\n", pdev);
	2727	+ list_del(&pdev->list);
	2728	+ bcache_device_stop(&pdev->dc->disk);
	2729	+ kfree(pdev);
	2730	+ }
	2731	+
	2732	+ return ret;
2313	2733	}
2314	2734
2315	2735	static int bcache_reboot(struct notifier_block n, unsigned long code, void x)
2316	2736	{
	2737	+ if (bcache_is_reboot)
	2738	+ return NOTIFY_DONE;
	2739	+
2317	2740	if (code == SYS_DOWN \|\|
2318	2741	code == SYS_HALT \|\|
2319	2742	code == SYS_POWER_OFF) {
..	..	@@ -2326,22 +2749,57 @@
2326	2749
2327	2750	mutex_lock(&bch_register_lock);
2328	2751
	2752	+ if (bcache_is_reboot)
	2753	+ goto out;
	2754	+
	2755	+ /* New registration is rejected since now */
	2756	+ bcache_is_reboot = true;
	2757	+ /*
	2758	+ * Make registering caller (if there is) on other CPU
	2759	+ * core know bcache_is_reboot set to true earlier
	2760	+ */
	2761	+ smp_mb();
	2762	+
2329	2763	if (list_empty(&bch_cache_sets) &&
2330	2764	list_empty(&uncached_devices))
2331	2765	goto out;
2332	2766
2333		- pr_info("Stopping all devices:");
	2767	+ mutex_unlock(&bch_register_lock);
2334	2768
	2769	+ pr_info("Stopping all devices:\n");
	2770	+
	2771	+ /*
	2772	+ * The reason bch_register_lock is not held to call
	2773	+ * bch_cache_set_stop() and bcache_device_stop() is to
	2774	+ * avoid potential deadlock during reboot, because cache
	2775	+ * set or bcache device stopping process will acqurie
	2776	+ * bch_register_lock too.
	2777	+ *
	2778	+ * We are safe here because bcache_is_reboot sets to
	2779	+ * true already, register_bcache() will reject new
	2780	+ * registration now. bcache_is_reboot also makes sure
	2781	+ * bcache_reboot() won't be re-entered on by other thread,
	2782	+ * so there is no race in following list iteration by
	2783	+ * list_for_each_entry_safe().
	2784	+ */
2335	2785	list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
2336	2786	bch_cache_set_stop(c);
2337	2787
2338	2788	list_for_each_entry_safe(dc, tdc, &uncached_devices, list)
2339	2789	bcache_device_stop(&dc->disk);
2340	2790
	2791	+
	2792	+ /*
	2793	+ * Give an early chance for other kthreads and
	2794	+ * kworkers to stop themselves
	2795	+ */
	2796	+ schedule();
	2797	+
2341	2798	/* What's a condition variable? */
2342	2799	while (1) {
2343		- long timeout = start + 2 * HZ - jiffies;
	2800	+ long timeout = start + 10 * HZ - jiffies;
2344	2801
	2802	+ mutex_lock(&bch_register_lock);
2345	2803	stopped = list_empty(&bch_cache_sets) &&
2346	2804	list_empty(&uncached_devices);
2347	2805
..	..	@@ -2353,15 +2811,14 @@
2353	2811
2354	2812	mutex_unlock(&bch_register_lock);
2355	2813	schedule_timeout(timeout);
2356		- mutex_lock(&bch_register_lock);
2357	2814	}
2358	2815
2359	2816	finish_wait(&unregister_wait, &wait);
2360	2817
2361	2818	if (stopped)
2362		- pr_info("All devices stopped");
	2819	+ pr_info("All devices stopped\n");
2363	2820	else
2364		- pr_notice("Timeout waiting for devices to be closed");
	2821	+ pr_notice("Timeout waiting for devices to be closed\n");
2365	2822	out:
2366	2823	mutex_unlock(&bch_register_lock);
2367	2824	}
..	..	@@ -2384,6 +2841,9 @@
2384	2841	destroy_workqueue(bcache_wq);
2385	2842	if (bch_journal_wq)
2386	2843	destroy_workqueue(bch_journal_wq);
	2844	+ if (bch_flush_wq)
	2845	+ destroy_workqueue(bch_flush_wq);
	2846	+ bch_btree_exit();
2387	2847
2388	2848	if (bcache_major)
2389	2849	unregister_blkdev(bcache_major, "bcache");
..	..	@@ -2391,13 +2851,42 @@
2391	2851	mutex_destroy(&bch_register_lock);
2392	2852	}
2393	2853
	2854	+/* Check and fixup module parameters */
	2855	+static void check_module_parameters(void)
	2856	+{
	2857	+ if (bch_cutoff_writeback_sync == 0)
	2858	+ bch_cutoff_writeback_sync = CUTOFF_WRITEBACK_SYNC;
	2859	+ else if (bch_cutoff_writeback_sync > CUTOFF_WRITEBACK_SYNC_MAX) {
	2860	+ pr_warn("set bch_cutoff_writeback_sync (%u) to max value %u\n",
	2861	+ bch_cutoff_writeback_sync, CUTOFF_WRITEBACK_SYNC_MAX);
	2862	+ bch_cutoff_writeback_sync = CUTOFF_WRITEBACK_SYNC_MAX;
	2863	+ }
	2864	+
	2865	+ if (bch_cutoff_writeback == 0)
	2866	+ bch_cutoff_writeback = CUTOFF_WRITEBACK;
	2867	+ else if (bch_cutoff_writeback > CUTOFF_WRITEBACK_MAX) {
	2868	+ pr_warn("set bch_cutoff_writeback (%u) to max value %u\n",
	2869	+ bch_cutoff_writeback, CUTOFF_WRITEBACK_MAX);
	2870	+ bch_cutoff_writeback = CUTOFF_WRITEBACK_MAX;
	2871	+ }
	2872	+
	2873	+ if (bch_cutoff_writeback > bch_cutoff_writeback_sync) {
	2874	+ pr_warn("set bch_cutoff_writeback (%u) to %u\n",
	2875	+ bch_cutoff_writeback, bch_cutoff_writeback_sync);
	2876	+ bch_cutoff_writeback = bch_cutoff_writeback_sync;
	2877	+ }
	2878	+}
	2879	+
2394	2880	static int __init bcache_init(void)
2395	2881	{
2396	2882	static const struct attribute *files[] = {
2397	2883	&ksysfs_register.attr,
2398	2884	&ksysfs_register_quiet.attr,
	2885	+ &ksysfs_pendings_cleanup.attr,
2399	2886	NULL
2400	2887	};
	2888	+
	2889	+ check_module_parameters();
2401	2890
2402	2891	mutex_init(&bch_register_lock);
2403	2892	init_waitqueue_head(&unregister_wait);
..	..	@@ -2410,8 +2899,24 @@
2410	2899	return bcache_major;
2411	2900	}
2412	2901
	2902	+ if (bch_btree_init())
	2903	+ goto err;
	2904	+
2413	2905	bcache_wq = alloc_workqueue("bcache", WQ_MEM_RECLAIM, 0);
2414	2906	if (!bcache_wq)
	2907	+ goto err;
	2908	+
	2909	+ /*
	2910	+ * Let's not make this `WQ_MEM_RECLAIM` for the following reasons:
	2911	+ *
	2912	+ * 1. It used `system_wq` before which also does no memory reclaim.
	2913	+ * 2. With `WQ_MEM_RECLAIM` desktop stalls, increased boot times, and
	2914	+ * reduced throughput can be observed.
	2915	+ *
	2916	+ * We still want to user our own queue to not congest the `system_wq`.
	2917	+ */
	2918	+ bch_flush_wq = alloc_workqueue("bch_flush", 0, 0);
	2919	+ if (!bch_flush_wq)
2415	2920	goto err;
2416	2921
2417	2922	bch_journal_wq = alloc_workqueue("bch_journal", WQ_MEM_RECLAIM, 0);
..	..	@@ -2426,8 +2931,10 @@
2426	2931	sysfs_create_files(bcache_kobj, files))
2427	2932	goto err;
2428	2933
2429		- bch_debug_init(bcache_kobj);
	2934	+ bch_debug_init();
2430	2935	closure_debug_init();
	2936	+
	2937	+ bcache_is_reboot = false;
2431	2938
2432	2939	return 0;
2433	2940	err:
..	..	@@ -2435,5 +2942,18 @@
2435	2942	return -ENOMEM;
2436	2943	}
2437	2944
	2945	+/*
	2946	+ * Module hooks
	2947	+ */
2438	2948	module_exit(bcache_exit);
2439	2949	module_init(bcache_init);
	2950	+
	2951	+module_param(bch_cutoff_writeback, uint, 0);
	2952	+MODULE_PARM_DESC(bch_cutoff_writeback, "threshold to cutoff writeback");
	2953	+
	2954	+module_param(bch_cutoff_writeback_sync, uint, 0);
	2955	+MODULE_PARM_DESC(bch_cutoff_writeback_sync, "hard threshold to cutoff writeback");
	2956	+
	2957	+MODULE_DESCRIPTION("Bcache: a Linux block layer cache");
	2958	+MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
	2959	+MODULE_LICENSE("GPL");