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 void bch_bi_idx_hack_endio(struct bio *bio, int error) |
15 | { | |
16 | struct bio *p = bio->bi_private; | |
17 | ||
18 | bio_endio(p, error); | |
19 | bio_put(bio); | |
20 | } | |
21 | ||
22 | static void bch_generic_make_request_hack(struct bio *bio) | |
23 | { | |
24 | if (bio->bi_idx) { | |
25 | struct bio *clone = bio_alloc(GFP_NOIO, bio_segments(bio)); | |
26 | ||
27 | memcpy(clone->bi_io_vec, | |
28 | bio_iovec(bio), | |
29 | bio_segments(bio) * sizeof(struct bio_vec)); | |
30 | ||
31 | clone->bi_sector = bio->bi_sector; | |
32 | clone->bi_bdev = bio->bi_bdev; | |
33 | clone->bi_rw = bio->bi_rw; | |
34 | clone->bi_vcnt = bio_segments(bio); | |
35 | clone->bi_size = bio->bi_size; | |
36 | ||
37 | clone->bi_private = bio; | |
38 | clone->bi_end_io = bch_bi_idx_hack_endio; | |
39 | ||
40 | bio = clone; | |
41 | } | |
42 | ||
bca97ada KO |
43 | /* |
44 | * Hack, since drivers that clone bios clone up to bi_max_vecs, but our | |
45 | * bios might have had more than that (before we split them per device | |
46 | * limitations). | |
47 | * | |
48 | * To be taken out once immutable bvec stuff is in. | |
49 | */ | |
50 | bio->bi_max_vecs = bio->bi_vcnt; | |
51 | ||
cafe5635 KO |
52 | generic_make_request(bio); |
53 | } | |
54 | ||
55 | /** | |
56 | * bch_bio_split - split a bio | |
57 | * @bio: bio to split | |
58 | * @sectors: number of sectors to split from the front of @bio | |
59 | * @gfp: gfp mask | |
60 | * @bs: bio set to allocate from | |
61 | * | |
62 | * Allocates and returns a new bio which represents @sectors from the start of | |
63 | * @bio, and updates @bio to represent the remaining sectors. | |
64 | * | |
65 | * If bio_sectors(@bio) was less than or equal to @sectors, returns @bio | |
66 | * unchanged. | |
67 | * | |
68 | * The newly allocated bio will point to @bio's bi_io_vec, if the split was on a | |
69 | * bvec boundry; it is the caller's responsibility to ensure that @bio is not | |
70 | * freed before the split. | |
cafe5635 KO |
71 | */ |
72 | struct bio *bch_bio_split(struct bio *bio, int sectors, | |
73 | gfp_t gfp, struct bio_set *bs) | |
74 | { | |
75 | unsigned idx = bio->bi_idx, vcnt = 0, nbytes = sectors << 9; | |
76 | struct bio_vec *bv; | |
77 | struct bio *ret = NULL; | |
78 | ||
79 | BUG_ON(sectors <= 0); | |
80 | ||
cafe5635 KO |
81 | if (sectors >= bio_sectors(bio)) |
82 | return bio; | |
83 | ||
84 | if (bio->bi_rw & REQ_DISCARD) { | |
85 | ret = bio_alloc_bioset(gfp, 1, bs); | |
5c694129 KAM |
86 | if (!ret) |
87 | return NULL; | |
cafe5635 KO |
88 | idx = 0; |
89 | goto out; | |
90 | } | |
91 | ||
92 | bio_for_each_segment(bv, bio, idx) { | |
93 | vcnt = idx - bio->bi_idx; | |
94 | ||
95 | if (!nbytes) { | |
96 | ret = bio_alloc_bioset(gfp, vcnt, bs); | |
97 | if (!ret) | |
98 | return NULL; | |
99 | ||
100 | memcpy(ret->bi_io_vec, bio_iovec(bio), | |
101 | sizeof(struct bio_vec) * vcnt); | |
102 | ||
103 | break; | |
104 | } else if (nbytes < bv->bv_len) { | |
105 | ret = bio_alloc_bioset(gfp, ++vcnt, bs); | |
106 | if (!ret) | |
107 | return NULL; | |
108 | ||
109 | memcpy(ret->bi_io_vec, bio_iovec(bio), | |
110 | sizeof(struct bio_vec) * vcnt); | |
111 | ||
112 | ret->bi_io_vec[vcnt - 1].bv_len = nbytes; | |
113 | bv->bv_offset += nbytes; | |
114 | bv->bv_len -= nbytes; | |
115 | break; | |
116 | } | |
117 | ||
118 | nbytes -= bv->bv_len; | |
119 | } | |
120 | out: | |
121 | ret->bi_bdev = bio->bi_bdev; | |
122 | ret->bi_sector = bio->bi_sector; | |
123 | ret->bi_size = sectors << 9; | |
124 | ret->bi_rw = bio->bi_rw; | |
125 | ret->bi_vcnt = vcnt; | |
126 | ret->bi_max_vecs = vcnt; | |
127 | ||
128 | bio->bi_sector += sectors; | |
129 | bio->bi_size -= sectors << 9; | |
130 | bio->bi_idx = idx; | |
131 | ||
132 | if (bio_integrity(bio)) { | |
133 | if (bio_integrity_clone(ret, bio, gfp)) { | |
134 | bio_put(ret); | |
135 | return NULL; | |
136 | } | |
137 | ||
138 | bio_integrity_trim(ret, 0, bio_sectors(ret)); | |
139 | bio_integrity_trim(bio, bio_sectors(ret), bio_sectors(bio)); | |
140 | } | |
141 | ||
142 | return ret; | |
143 | } | |
144 | ||
145 | static unsigned bch_bio_max_sectors(struct bio *bio) | |
146 | { | |
147 | unsigned ret = bio_sectors(bio); | |
148 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); | |
1545f137 KO |
149 | unsigned max_segments = min_t(unsigned, BIO_MAX_PAGES, |
150 | queue_max_segments(q)); | |
cafe5635 | 151 | |
cafe5635 KO |
152 | if (bio->bi_rw & REQ_DISCARD) |
153 | return min(ret, q->limits.max_discard_sectors); | |
154 | ||
1545f137 | 155 | if (bio_segments(bio) > max_segments || |
cafe5635 | 156 | q->merge_bvec_fn) { |
8e51e414 KO |
157 | struct bio_vec *bv; |
158 | int i, seg = 0; | |
159 | ||
cafe5635 KO |
160 | ret = 0; |
161 | ||
8e51e414 | 162 | bio_for_each_segment(bv, bio, i) { |
a09ded8e KO |
163 | struct bvec_merge_data bvm = { |
164 | .bi_bdev = bio->bi_bdev, | |
165 | .bi_sector = bio->bi_sector, | |
166 | .bi_size = ret << 9, | |
167 | .bi_rw = bio->bi_rw, | |
168 | }; | |
169 | ||
8e51e414 KO |
170 | if (seg == max_segments) |
171 | break; | |
172 | ||
cafe5635 KO |
173 | if (q->merge_bvec_fn && |
174 | q->merge_bvec_fn(q, &bvm, bv) < (int) bv->bv_len) | |
175 | break; | |
176 | ||
8e51e414 | 177 | seg++; |
a09ded8e | 178 | ret += bv->bv_len >> 9; |
cafe5635 | 179 | } |
cafe5635 KO |
180 | } |
181 | ||
182 | ret = min(ret, queue_max_sectors(q)); | |
183 | ||
184 | WARN_ON(!ret); | |
185 | ret = max_t(int, ret, bio_iovec(bio)->bv_len >> 9); | |
186 | ||
187 | return ret; | |
188 | } | |
189 | ||
190 | static void bch_bio_submit_split_done(struct closure *cl) | |
191 | { | |
192 | struct bio_split_hook *s = container_of(cl, struct bio_split_hook, cl); | |
193 | ||
194 | s->bio->bi_end_io = s->bi_end_io; | |
195 | s->bio->bi_private = s->bi_private; | |
196 | bio_endio(s->bio, 0); | |
197 | ||
198 | closure_debug_destroy(&s->cl); | |
199 | mempool_free(s, s->p->bio_split_hook); | |
200 | } | |
201 | ||
202 | static void bch_bio_submit_split_endio(struct bio *bio, int error) | |
203 | { | |
204 | struct closure *cl = bio->bi_private; | |
205 | struct bio_split_hook *s = container_of(cl, struct bio_split_hook, cl); | |
206 | ||
207 | if (error) | |
208 | clear_bit(BIO_UPTODATE, &s->bio->bi_flags); | |
209 | ||
210 | bio_put(bio); | |
211 | closure_put(cl); | |
212 | } | |
213 | ||
cafe5635 KO |
214 | void bch_generic_make_request(struct bio *bio, struct bio_split_pool *p) |
215 | { | |
216 | struct bio_split_hook *s; | |
8e51e414 | 217 | struct bio *n; |
cafe5635 KO |
218 | |
219 | if (!bio_has_data(bio) && !(bio->bi_rw & REQ_DISCARD)) | |
220 | goto submit; | |
221 | ||
222 | if (bio_sectors(bio) <= bch_bio_max_sectors(bio)) | |
223 | goto submit; | |
224 | ||
225 | s = mempool_alloc(p->bio_split_hook, GFP_NOIO); | |
8e51e414 | 226 | closure_init(&s->cl, NULL); |
cafe5635 KO |
227 | |
228 | s->bio = bio; | |
229 | s->p = p; | |
230 | s->bi_end_io = bio->bi_end_io; | |
231 | s->bi_private = bio->bi_private; | |
232 | bio_get(bio); | |
233 | ||
8e51e414 KO |
234 | do { |
235 | n = bch_bio_split(bio, bch_bio_max_sectors(bio), | |
236 | GFP_NOIO, s->p->bio_split); | |
237 | ||
238 | n->bi_end_io = bch_bio_submit_split_endio; | |
239 | n->bi_private = &s->cl; | |
240 | ||
241 | closure_get(&s->cl); | |
242 | bch_generic_make_request_hack(n); | |
243 | } while (n != bio); | |
244 | ||
245 | continue_at(&s->cl, bch_bio_submit_split_done, NULL); | |
cafe5635 KO |
246 | submit: |
247 | bch_generic_make_request_hack(bio); | |
248 | } | |
249 | ||
250 | /* Bios with headers */ | |
251 | ||
252 | void bch_bbio_free(struct bio *bio, struct cache_set *c) | |
253 | { | |
254 | struct bbio *b = container_of(bio, struct bbio, bio); | |
255 | mempool_free(b, c->bio_meta); | |
256 | } | |
257 | ||
258 | struct bio *bch_bbio_alloc(struct cache_set *c) | |
259 | { | |
260 | struct bbio *b = mempool_alloc(c->bio_meta, GFP_NOIO); | |
261 | struct bio *bio = &b->bio; | |
262 | ||
263 | bio_init(bio); | |
264 | bio->bi_flags |= BIO_POOL_NONE << BIO_POOL_OFFSET; | |
265 | bio->bi_max_vecs = bucket_pages(c); | |
266 | bio->bi_io_vec = bio->bi_inline_vecs; | |
267 | ||
268 | return bio; | |
269 | } | |
270 | ||
271 | void __bch_submit_bbio(struct bio *bio, struct cache_set *c) | |
272 | { | |
273 | struct bbio *b = container_of(bio, struct bbio, bio); | |
274 | ||
275 | bio->bi_sector = PTR_OFFSET(&b->key, 0); | |
276 | bio->bi_bdev = PTR_CACHE(c, &b->key, 0)->bdev; | |
277 | ||
278 | b->submit_time_us = local_clock_us(); | |
279 | closure_bio_submit(bio, bio->bi_private, PTR_CACHE(c, &b->key, 0)); | |
280 | } | |
281 | ||
282 | void bch_submit_bbio(struct bio *bio, struct cache_set *c, | |
283 | struct bkey *k, unsigned ptr) | |
284 | { | |
285 | struct bbio *b = container_of(bio, struct bbio, bio); | |
286 | bch_bkey_copy_single_ptr(&b->key, k, ptr); | |
287 | __bch_submit_bbio(bio, c); | |
288 | } | |
289 | ||
290 | /* IO errors */ | |
291 | ||
292 | void bch_count_io_errors(struct cache *ca, int error, const char *m) | |
293 | { | |
294 | /* | |
295 | * The halflife of an error is: | |
296 | * log2(1/2)/log2(127/128) * refresh ~= 88 * refresh | |
297 | */ | |
298 | ||
299 | if (ca->set->error_decay) { | |
300 | unsigned count = atomic_inc_return(&ca->io_count); | |
301 | ||
302 | while (count > ca->set->error_decay) { | |
303 | unsigned errors; | |
304 | unsigned old = count; | |
305 | unsigned new = count - ca->set->error_decay; | |
306 | ||
307 | /* | |
308 | * First we subtract refresh from count; each time we | |
309 | * succesfully do so, we rescale the errors once: | |
310 | */ | |
311 | ||
312 | count = atomic_cmpxchg(&ca->io_count, old, new); | |
313 | ||
314 | if (count == old) { | |
315 | count = new; | |
316 | ||
317 | errors = atomic_read(&ca->io_errors); | |
318 | do { | |
319 | old = errors; | |
320 | new = ((uint64_t) errors * 127) / 128; | |
321 | errors = atomic_cmpxchg(&ca->io_errors, | |
322 | old, new); | |
323 | } while (old != errors); | |
324 | } | |
325 | } | |
326 | } | |
327 | ||
328 | if (error) { | |
329 | char buf[BDEVNAME_SIZE]; | |
330 | unsigned errors = atomic_add_return(1 << IO_ERROR_SHIFT, | |
331 | &ca->io_errors); | |
332 | errors >>= IO_ERROR_SHIFT; | |
333 | ||
334 | if (errors < ca->set->error_limit) | |
335 | pr_err("%s: IO error on %s, recovering", | |
336 | bdevname(ca->bdev, buf), m); | |
337 | else | |
338 | bch_cache_set_error(ca->set, | |
339 | "%s: too many IO errors %s", | |
340 | bdevname(ca->bdev, buf), m); | |
341 | } | |
342 | } | |
343 | ||
344 | void bch_bbio_count_io_errors(struct cache_set *c, struct bio *bio, | |
345 | int error, const char *m) | |
346 | { | |
347 | struct bbio *b = container_of(bio, struct bbio, bio); | |
348 | struct cache *ca = PTR_CACHE(c, &b->key, 0); | |
349 | ||
350 | unsigned threshold = bio->bi_rw & REQ_WRITE | |
351 | ? c->congested_write_threshold_us | |
352 | : c->congested_read_threshold_us; | |
353 | ||
354 | if (threshold) { | |
355 | unsigned t = local_clock_us(); | |
356 | ||
357 | int us = t - b->submit_time_us; | |
358 | int congested = atomic_read(&c->congested); | |
359 | ||
360 | if (us > (int) threshold) { | |
361 | int ms = us / 1024; | |
362 | c->congested_last_us = t; | |
363 | ||
364 | ms = min(ms, CONGESTED_MAX + congested); | |
365 | atomic_sub(ms, &c->congested); | |
366 | } else if (congested < 0) | |
367 | atomic_inc(&c->congested); | |
368 | } | |
369 | ||
370 | bch_count_io_errors(ca, error, m); | |
371 | } | |
372 | ||
373 | void bch_bbio_endio(struct cache_set *c, struct bio *bio, | |
374 | int error, const char *m) | |
375 | { | |
376 | struct closure *cl = bio->bi_private; | |
377 | ||
378 | bch_bbio_count_io_errors(c, bio, error, m); | |
379 | bio_put(bio); | |
380 | closure_put(cl); | |
381 | } |