Commit | Line | Data |
---|---|---|
cafe5635 KO |
1 | /* |
2 | * Some low level IO code, and hacks for various block layer limitations | |
3 | * | |
4 | * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com> | |
5 | * Copyright 2012 Google, Inc. | |
6 | */ | |
7 | ||
8 | #include "bcache.h" | |
9 | #include "bset.h" | |
10 | #include "debug.h" | |
11 | ||
c37511b8 KO |
12 | #include <linux/blkdev.h> |
13 | ||
cafe5635 KO |
14 | static unsigned bch_bio_max_sectors(struct bio *bio) |
15 | { | |
cafe5635 | 16 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
458b76ed KO |
17 | struct bio_vec bv; |
18 | struct bvec_iter iter; | |
19 | unsigned ret = 0, seg = 0; | |
cafe5635 | 20 | |
cafe5635 | 21 | if (bio->bi_rw & REQ_DISCARD) |
458b76ed | 22 | return min(bio_sectors(bio), q->limits.max_discard_sectors); |
cafe5635 | 23 | |
458b76ed KO |
24 | bio_for_each_segment(bv, bio, iter) { |
25 | struct bvec_merge_data bvm = { | |
26 | .bi_bdev = bio->bi_bdev, | |
27 | .bi_sector = bio->bi_iter.bi_sector, | |
28 | .bi_size = ret << 9, | |
29 | .bi_rw = bio->bi_rw, | |
30 | }; | |
31 | ||
32 | if (seg == min_t(unsigned, BIO_MAX_PAGES, | |
33 | queue_max_segments(q))) | |
34 | break; | |
8e51e414 | 35 | |
458b76ed KO |
36 | if (q->merge_bvec_fn && |
37 | q->merge_bvec_fn(q, &bvm, &bv) < (int) bv.bv_len) | |
38 | break; | |
cafe5635 | 39 | |
458b76ed KO |
40 | seg++; |
41 | ret += bv.bv_len >> 9; | |
cafe5635 KO |
42 | } |
43 | ||
44 | ret = min(ret, queue_max_sectors(q)); | |
45 | ||
46 | WARN_ON(!ret); | |
a4ad39b1 | 47 | ret = max_t(int, ret, bio_iovec(bio).bv_len >> 9); |
cafe5635 KO |
48 | |
49 | return ret; | |
50 | } | |
51 | ||
52 | static void bch_bio_submit_split_done(struct closure *cl) | |
53 | { | |
54 | struct bio_split_hook *s = container_of(cl, struct bio_split_hook, cl); | |
55 | ||
56 | s->bio->bi_end_io = s->bi_end_io; | |
57 | s->bio->bi_private = s->bi_private; | |
196d38bc | 58 | bio_endio_nodec(s->bio, 0); |
cafe5635 KO |
59 | |
60 | closure_debug_destroy(&s->cl); | |
61 | mempool_free(s, s->p->bio_split_hook); | |
62 | } | |
63 | ||
64 | static void bch_bio_submit_split_endio(struct bio *bio, int error) | |
65 | { | |
66 | struct closure *cl = bio->bi_private; | |
67 | struct bio_split_hook *s = container_of(cl, struct bio_split_hook, cl); | |
68 | ||
69 | if (error) | |
70 | clear_bit(BIO_UPTODATE, &s->bio->bi_flags); | |
71 | ||
72 | bio_put(bio); | |
73 | closure_put(cl); | |
74 | } | |
75 | ||
cafe5635 KO |
76 | void bch_generic_make_request(struct bio *bio, struct bio_split_pool *p) |
77 | { | |
78 | struct bio_split_hook *s; | |
8e51e414 | 79 | struct bio *n; |
cafe5635 KO |
80 | |
81 | if (!bio_has_data(bio) && !(bio->bi_rw & REQ_DISCARD)) | |
82 | goto submit; | |
83 | ||
84 | if (bio_sectors(bio) <= bch_bio_max_sectors(bio)) | |
85 | goto submit; | |
86 | ||
87 | s = mempool_alloc(p->bio_split_hook, GFP_NOIO); | |
8e51e414 | 88 | closure_init(&s->cl, NULL); |
cafe5635 KO |
89 | |
90 | s->bio = bio; | |
91 | s->p = p; | |
92 | s->bi_end_io = bio->bi_end_io; | |
93 | s->bi_private = bio->bi_private; | |
94 | bio_get(bio); | |
95 | ||
8e51e414 | 96 | do { |
20d0189b KO |
97 | n = bio_next_split(bio, bch_bio_max_sectors(bio), |
98 | GFP_NOIO, s->p->bio_split); | |
8e51e414 KO |
99 | |
100 | n->bi_end_io = bch_bio_submit_split_endio; | |
101 | n->bi_private = &s->cl; | |
102 | ||
103 | closure_get(&s->cl); | |
e90abc8e | 104 | generic_make_request(n); |
8e51e414 KO |
105 | } while (n != bio); |
106 | ||
107 | continue_at(&s->cl, bch_bio_submit_split_done, NULL); | |
cafe5635 | 108 | submit: |
e90abc8e | 109 | generic_make_request(bio); |
cafe5635 KO |
110 | } |
111 | ||
112 | /* Bios with headers */ | |
113 | ||
114 | void bch_bbio_free(struct bio *bio, struct cache_set *c) | |
115 | { | |
116 | struct bbio *b = container_of(bio, struct bbio, bio); | |
117 | mempool_free(b, c->bio_meta); | |
118 | } | |
119 | ||
120 | struct bio *bch_bbio_alloc(struct cache_set *c) | |
121 | { | |
122 | struct bbio *b = mempool_alloc(c->bio_meta, GFP_NOIO); | |
123 | struct bio *bio = &b->bio; | |
124 | ||
125 | bio_init(bio); | |
126 | bio->bi_flags |= BIO_POOL_NONE << BIO_POOL_OFFSET; | |
127 | bio->bi_max_vecs = bucket_pages(c); | |
128 | bio->bi_io_vec = bio->bi_inline_vecs; | |
129 | ||
130 | return bio; | |
131 | } | |
132 | ||
133 | void __bch_submit_bbio(struct bio *bio, struct cache_set *c) | |
134 | { | |
135 | struct bbio *b = container_of(bio, struct bbio, bio); | |
136 | ||
4f024f37 KO |
137 | bio->bi_iter.bi_sector = PTR_OFFSET(&b->key, 0); |
138 | bio->bi_bdev = PTR_CACHE(c, &b->key, 0)->bdev; | |
cafe5635 KO |
139 | |
140 | b->submit_time_us = local_clock_us(); | |
141 | closure_bio_submit(bio, bio->bi_private, PTR_CACHE(c, &b->key, 0)); | |
142 | } | |
143 | ||
144 | void bch_submit_bbio(struct bio *bio, struct cache_set *c, | |
145 | struct bkey *k, unsigned ptr) | |
146 | { | |
147 | struct bbio *b = container_of(bio, struct bbio, bio); | |
148 | bch_bkey_copy_single_ptr(&b->key, k, ptr); | |
149 | __bch_submit_bbio(bio, c); | |
150 | } | |
151 | ||
152 | /* IO errors */ | |
153 | ||
154 | void bch_count_io_errors(struct cache *ca, int error, const char *m) | |
155 | { | |
156 | /* | |
157 | * The halflife of an error is: | |
158 | * log2(1/2)/log2(127/128) * refresh ~= 88 * refresh | |
159 | */ | |
160 | ||
161 | if (ca->set->error_decay) { | |
162 | unsigned count = atomic_inc_return(&ca->io_count); | |
163 | ||
164 | while (count > ca->set->error_decay) { | |
165 | unsigned errors; | |
166 | unsigned old = count; | |
167 | unsigned new = count - ca->set->error_decay; | |
168 | ||
169 | /* | |
170 | * First we subtract refresh from count; each time we | |
171 | * succesfully do so, we rescale the errors once: | |
172 | */ | |
173 | ||
174 | count = atomic_cmpxchg(&ca->io_count, old, new); | |
175 | ||
176 | if (count == old) { | |
177 | count = new; | |
178 | ||
179 | errors = atomic_read(&ca->io_errors); | |
180 | do { | |
181 | old = errors; | |
182 | new = ((uint64_t) errors * 127) / 128; | |
183 | errors = atomic_cmpxchg(&ca->io_errors, | |
184 | old, new); | |
185 | } while (old != errors); | |
186 | } | |
187 | } | |
188 | } | |
189 | ||
190 | if (error) { | |
191 | char buf[BDEVNAME_SIZE]; | |
192 | unsigned errors = atomic_add_return(1 << IO_ERROR_SHIFT, | |
193 | &ca->io_errors); | |
194 | errors >>= IO_ERROR_SHIFT; | |
195 | ||
196 | if (errors < ca->set->error_limit) | |
197 | pr_err("%s: IO error on %s, recovering", | |
198 | bdevname(ca->bdev, buf), m); | |
199 | else | |
200 | bch_cache_set_error(ca->set, | |
201 | "%s: too many IO errors %s", | |
202 | bdevname(ca->bdev, buf), m); | |
203 | } | |
204 | } | |
205 | ||
206 | void bch_bbio_count_io_errors(struct cache_set *c, struct bio *bio, | |
207 | int error, const char *m) | |
208 | { | |
209 | struct bbio *b = container_of(bio, struct bbio, bio); | |
210 | struct cache *ca = PTR_CACHE(c, &b->key, 0); | |
211 | ||
212 | unsigned threshold = bio->bi_rw & REQ_WRITE | |
213 | ? c->congested_write_threshold_us | |
214 | : c->congested_read_threshold_us; | |
215 | ||
216 | if (threshold) { | |
217 | unsigned t = local_clock_us(); | |
218 | ||
219 | int us = t - b->submit_time_us; | |
220 | int congested = atomic_read(&c->congested); | |
221 | ||
222 | if (us > (int) threshold) { | |
223 | int ms = us / 1024; | |
224 | c->congested_last_us = t; | |
225 | ||
226 | ms = min(ms, CONGESTED_MAX + congested); | |
227 | atomic_sub(ms, &c->congested); | |
228 | } else if (congested < 0) | |
229 | atomic_inc(&c->congested); | |
230 | } | |
231 | ||
232 | bch_count_io_errors(ca, error, m); | |
233 | } | |
234 | ||
235 | void bch_bbio_endio(struct cache_set *c, struct bio *bio, | |
236 | int error, const char *m) | |
237 | { | |
238 | struct closure *cl = bio->bi_private; | |
239 | ||
240 | bch_bbio_count_io_errors(c, bio, error, m); | |
241 | bio_put(bio); | |
242 | closure_put(cl); | |
243 | } |