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