Kent Overstreet | cafe563 | 2013-03-23 16:11:31 -0700 | [diff] [blame] | 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 | |
Kent Overstreet | c37511b | 2013-04-26 15:39:55 -0700 | [diff] [blame] | 12 | #include <linux/blkdev.h> |
| 13 | |
Kent Overstreet | cafe563 | 2013-03-23 16:11:31 -0700 | [diff] [blame] | 14 | /* Bios with headers */ |
| 15 | |
| 16 | void bch_bbio_free(struct bio *bio, struct cache_set *c) |
| 17 | { |
| 18 | struct bbio *b = container_of(bio, struct bbio, bio); |
| 19 | mempool_free(b, c->bio_meta); |
| 20 | } |
| 21 | |
| 22 | struct bio *bch_bbio_alloc(struct cache_set *c) |
| 23 | { |
| 24 | struct bbio *b = mempool_alloc(c->bio_meta, GFP_NOIO); |
| 25 | struct bio *bio = &b->bio; |
| 26 | |
| 27 | bio_init(bio); |
| 28 | bio->bi_flags |= BIO_POOL_NONE << BIO_POOL_OFFSET; |
| 29 | bio->bi_max_vecs = bucket_pages(c); |
| 30 | bio->bi_io_vec = bio->bi_inline_vecs; |
| 31 | |
| 32 | return bio; |
| 33 | } |
| 34 | |
| 35 | void __bch_submit_bbio(struct bio *bio, struct cache_set *c) |
| 36 | { |
| 37 | struct bbio *b = container_of(bio, struct bbio, bio); |
| 38 | |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 39 | bio->bi_iter.bi_sector = PTR_OFFSET(&b->key, 0); |
| 40 | bio->bi_bdev = PTR_CACHE(c, &b->key, 0)->bdev; |
Kent Overstreet | cafe563 | 2013-03-23 16:11:31 -0700 | [diff] [blame] | 41 | |
| 42 | b->submit_time_us = local_clock_us(); |
Kent Overstreet | 749b61d | 2013-11-23 23:11:25 -0800 | [diff] [blame] | 43 | closure_bio_submit(bio, bio->bi_private); |
Kent Overstreet | cafe563 | 2013-03-23 16:11:31 -0700 | [diff] [blame] | 44 | } |
| 45 | |
| 46 | void bch_submit_bbio(struct bio *bio, struct cache_set *c, |
| 47 | struct bkey *k, unsigned ptr) |
| 48 | { |
| 49 | struct bbio *b = container_of(bio, struct bbio, bio); |
| 50 | bch_bkey_copy_single_ptr(&b->key, k, ptr); |
| 51 | __bch_submit_bbio(bio, c); |
| 52 | } |
| 53 | |
| 54 | /* IO errors */ |
| 55 | |
| 56 | void bch_count_io_errors(struct cache *ca, int error, const char *m) |
| 57 | { |
| 58 | /* |
| 59 | * The halflife of an error is: |
| 60 | * log2(1/2)/log2(127/128) * refresh ~= 88 * refresh |
| 61 | */ |
| 62 | |
| 63 | if (ca->set->error_decay) { |
| 64 | unsigned count = atomic_inc_return(&ca->io_count); |
| 65 | |
| 66 | while (count > ca->set->error_decay) { |
| 67 | unsigned errors; |
| 68 | unsigned old = count; |
| 69 | unsigned new = count - ca->set->error_decay; |
| 70 | |
| 71 | /* |
| 72 | * First we subtract refresh from count; each time we |
| 73 | * succesfully do so, we rescale the errors once: |
| 74 | */ |
| 75 | |
| 76 | count = atomic_cmpxchg(&ca->io_count, old, new); |
| 77 | |
| 78 | if (count == old) { |
| 79 | count = new; |
| 80 | |
| 81 | errors = atomic_read(&ca->io_errors); |
| 82 | do { |
| 83 | old = errors; |
| 84 | new = ((uint64_t) errors * 127) / 128; |
| 85 | errors = atomic_cmpxchg(&ca->io_errors, |
| 86 | old, new); |
| 87 | } while (old != errors); |
| 88 | } |
| 89 | } |
| 90 | } |
| 91 | |
| 92 | if (error) { |
| 93 | char buf[BDEVNAME_SIZE]; |
| 94 | unsigned errors = atomic_add_return(1 << IO_ERROR_SHIFT, |
| 95 | &ca->io_errors); |
| 96 | errors >>= IO_ERROR_SHIFT; |
| 97 | |
| 98 | if (errors < ca->set->error_limit) |
| 99 | pr_err("%s: IO error on %s, recovering", |
| 100 | bdevname(ca->bdev, buf), m); |
| 101 | else |
| 102 | bch_cache_set_error(ca->set, |
| 103 | "%s: too many IO errors %s", |
| 104 | bdevname(ca->bdev, buf), m); |
| 105 | } |
| 106 | } |
| 107 | |
| 108 | void bch_bbio_count_io_errors(struct cache_set *c, struct bio *bio, |
| 109 | int error, const char *m) |
| 110 | { |
| 111 | struct bbio *b = container_of(bio, struct bbio, bio); |
| 112 | struct cache *ca = PTR_CACHE(c, &b->key, 0); |
| 113 | |
| 114 | unsigned threshold = bio->bi_rw & REQ_WRITE |
| 115 | ? c->congested_write_threshold_us |
| 116 | : c->congested_read_threshold_us; |
| 117 | |
| 118 | if (threshold) { |
| 119 | unsigned t = local_clock_us(); |
| 120 | |
| 121 | int us = t - b->submit_time_us; |
| 122 | int congested = atomic_read(&c->congested); |
| 123 | |
| 124 | if (us > (int) threshold) { |
| 125 | int ms = us / 1024; |
| 126 | c->congested_last_us = t; |
| 127 | |
| 128 | ms = min(ms, CONGESTED_MAX + congested); |
| 129 | atomic_sub(ms, &c->congested); |
| 130 | } else if (congested < 0) |
| 131 | atomic_inc(&c->congested); |
| 132 | } |
| 133 | |
| 134 | bch_count_io_errors(ca, error, m); |
| 135 | } |
| 136 | |
| 137 | void bch_bbio_endio(struct cache_set *c, struct bio *bio, |
| 138 | int error, const char *m) |
| 139 | { |
| 140 | struct closure *cl = bio->bi_private; |
| 141 | |
| 142 | bch_bbio_count_io_errors(c, bio, error, m); |
| 143 | bio_put(bio); |
| 144 | closure_put(cl); |
| 145 | } |