1 /* 2 * Moving/copying garbage collector 3 * 4 * Copyright 2012 Google, Inc. 5 */ 6 7 #include "bcache.h" 8 #include "btree.h" 9 #include "debug.h" 10 #include "request.h" 11 12 #include <trace/events/bcache.h> 13 14 struct moving_io { 15 struct closure cl; 16 struct keybuf_key *w; 17 struct data_insert_op op; 18 struct bbio bio; 19 }; 20 21 static bool moving_pred(struct keybuf *buf, struct bkey *k) 22 { 23 struct cache_set *c = container_of(buf, struct cache_set, 24 moving_gc_keys); 25 unsigned i; 26 27 for (i = 0; i < KEY_PTRS(k); i++) 28 if (ptr_available(c, k, i) && 29 GC_MOVE(PTR_BUCKET(c, k, i))) 30 return true; 31 32 return false; 33 } 34 35 /* Moving GC - IO loop */ 36 37 static void moving_io_destructor(struct closure *cl) 38 { 39 struct moving_io *io = container_of(cl, struct moving_io, cl); 40 kfree(io); 41 } 42 43 static void write_moving_finish(struct closure *cl) 44 { 45 struct moving_io *io = container_of(cl, struct moving_io, cl); 46 struct bio *bio = &io->bio.bio; 47 struct bio_vec *bv; 48 int i; 49 50 bio_for_each_segment_all(bv, bio, i) 51 __free_page(bv->bv_page); 52 53 if (io->op.replace_collision) 54 trace_bcache_gc_copy_collision(&io->w->key); 55 56 bch_keybuf_del(&io->op.c->moving_gc_keys, io->w); 57 58 up(&io->op.c->moving_in_flight); 59 60 closure_return_with_destructor(cl, moving_io_destructor); 61 } 62 63 static void read_moving_endio(struct bio *bio, int error) 64 { 65 struct bbio *b = container_of(bio, struct bbio, bio); 66 struct moving_io *io = container_of(bio->bi_private, 67 struct moving_io, cl); 68 69 if (error) 70 io->op.error = error; 71 else if (!KEY_DIRTY(&b->key) && 72 ptr_stale(io->op.c, &b->key, 0)) { 73 io->op.error = -EINTR; 74 } 75 76 bch_bbio_endio(io->op.c, bio, error, "reading data to move"); 77 } 78 79 static void moving_init(struct moving_io *io) 80 { 81 struct bio *bio = &io->bio.bio; 82 83 bio_init(bio); 84 bio_get(bio); 85 bio_set_prio(bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0)); 86 87 bio->bi_iter.bi_size = KEY_SIZE(&io->w->key) << 9; 88 bio->bi_max_vecs = DIV_ROUND_UP(KEY_SIZE(&io->w->key), 89 PAGE_SECTORS); 90 bio->bi_private = &io->cl; 91 bio->bi_io_vec = bio->bi_inline_vecs; 92 bch_bio_map(bio, NULL); 93 } 94 95 static void write_moving(struct closure *cl) 96 { 97 struct moving_io *io = container_of(cl, struct moving_io, cl); 98 struct data_insert_op *op = &io->op; 99 100 if (!op->error) { 101 moving_init(io); 102 103 io->bio.bio.bi_iter.bi_sector = KEY_START(&io->w->key); 104 op->write_prio = 1; 105 op->bio = &io->bio.bio; 106 107 op->writeback = KEY_DIRTY(&io->w->key); 108 op->csum = KEY_CSUM(&io->w->key); 109 110 bkey_copy(&op->replace_key, &io->w->key); 111 op->replace = true; 112 113 closure_call(&op->cl, bch_data_insert, NULL, cl); 114 } 115 116 continue_at(cl, write_moving_finish, op->wq); 117 } 118 119 static void read_moving_submit(struct closure *cl) 120 { 121 struct moving_io *io = container_of(cl, struct moving_io, cl); 122 struct bio *bio = &io->bio.bio; 123 124 bch_submit_bbio(bio, io->op.c, &io->w->key, 0); 125 126 continue_at(cl, write_moving, io->op.wq); 127 } 128 129 static void read_moving(struct cache_set *c) 130 { 131 struct keybuf_key *w; 132 struct moving_io *io; 133 struct bio *bio; 134 struct closure cl; 135 136 closure_init_stack(&cl); 137 138 /* XXX: if we error, background writeback could stall indefinitely */ 139 140 while (!test_bit(CACHE_SET_STOPPING, &c->flags)) { 141 w = bch_keybuf_next_rescan(c, &c->moving_gc_keys, 142 &MAX_KEY, moving_pred); 143 if (!w) 144 break; 145 146 if (ptr_stale(c, &w->key, 0)) { 147 bch_keybuf_del(&c->moving_gc_keys, w); 148 continue; 149 } 150 151 io = kzalloc(sizeof(struct moving_io) + sizeof(struct bio_vec) 152 * DIV_ROUND_UP(KEY_SIZE(&w->key), PAGE_SECTORS), 153 GFP_KERNEL); 154 if (!io) 155 goto err; 156 157 w->private = io; 158 io->w = w; 159 io->op.inode = KEY_INODE(&w->key); 160 io->op.c = c; 161 io->op.wq = c->moving_gc_wq; 162 163 moving_init(io); 164 bio = &io->bio.bio; 165 166 bio->bi_rw = READ; 167 bio->bi_end_io = read_moving_endio; 168 169 if (bio_alloc_pages(bio, GFP_KERNEL)) 170 goto err; 171 172 trace_bcache_gc_copy(&w->key); 173 174 down(&c->moving_in_flight); 175 closure_call(&io->cl, read_moving_submit, NULL, &cl); 176 } 177 178 if (0) { 179 err: if (!IS_ERR_OR_NULL(w->private)) 180 kfree(w->private); 181 182 bch_keybuf_del(&c->moving_gc_keys, w); 183 } 184 185 closure_sync(&cl); 186 } 187 188 static bool bucket_cmp(struct bucket *l, struct bucket *r) 189 { 190 return GC_SECTORS_USED(l) < GC_SECTORS_USED(r); 191 } 192 193 static unsigned bucket_heap_top(struct cache *ca) 194 { 195 struct bucket *b; 196 return (b = heap_peek(&ca->heap)) ? GC_SECTORS_USED(b) : 0; 197 } 198 199 void bch_moving_gc(struct cache_set *c) 200 { 201 struct cache *ca; 202 struct bucket *b; 203 unsigned i; 204 205 if (!c->copy_gc_enabled) 206 return; 207 208 mutex_lock(&c->bucket_lock); 209 210 for_each_cache(ca, c, i) { 211 unsigned sectors_to_move = 0; 212 unsigned reserve_sectors = ca->sb.bucket_size * 213 fifo_used(&ca->free[RESERVE_MOVINGGC]); 214 215 ca->heap.used = 0; 216 217 for_each_bucket(b, ca) { 218 if (GC_MARK(b) == GC_MARK_METADATA || 219 !GC_SECTORS_USED(b) || 220 GC_SECTORS_USED(b) == ca->sb.bucket_size || 221 atomic_read(&b->pin)) 222 continue; 223 224 if (!heap_full(&ca->heap)) { 225 sectors_to_move += GC_SECTORS_USED(b); 226 heap_add(&ca->heap, b, bucket_cmp); 227 } else if (bucket_cmp(b, heap_peek(&ca->heap))) { 228 sectors_to_move -= bucket_heap_top(ca); 229 sectors_to_move += GC_SECTORS_USED(b); 230 231 ca->heap.data[0] = b; 232 heap_sift(&ca->heap, 0, bucket_cmp); 233 } 234 } 235 236 while (sectors_to_move > reserve_sectors) { 237 heap_pop(&ca->heap, b, bucket_cmp); 238 sectors_to_move -= GC_SECTORS_USED(b); 239 } 240 241 while (heap_pop(&ca->heap, b, bucket_cmp)) 242 SET_GC_MOVE(b, 1); 243 } 244 245 mutex_unlock(&c->bucket_lock); 246 247 c->moving_gc_keys.last_scanned = ZERO_KEY; 248 249 read_moving(c); 250 } 251 252 void bch_moving_init_cache_set(struct cache_set *c) 253 { 254 bch_keybuf_init(&c->moving_gc_keys); 255 sema_init(&c->moving_in_flight, 64); 256 } 257