xref: /openbmc/linux/drivers/md/bcache/movinggc.c (revision f35e839a)
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 struct moving_io {
13 	struct keybuf_key	*w;
14 	struct search		s;
15 	struct bbio		bio;
16 };
17 
18 static bool moving_pred(struct keybuf *buf, struct bkey *k)
19 {
20 	struct cache_set *c = container_of(buf, struct cache_set,
21 					   moving_gc_keys);
22 	unsigned i;
23 
24 	for (i = 0; i < KEY_PTRS(k); i++) {
25 		struct cache *ca = PTR_CACHE(c, k, i);
26 		struct bucket *g = PTR_BUCKET(c, k, i);
27 
28 		if (GC_SECTORS_USED(g) < ca->gc_move_threshold)
29 			return true;
30 	}
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, s.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, s.cl);
46 	struct bio *bio = &io->bio.bio;
47 	struct bio_vec *bv = bio_iovec_idx(bio, bio->bi_vcnt);
48 
49 	while (bv-- != bio->bi_io_vec)
50 		__free_page(bv->bv_page);
51 
52 	pr_debug("%s %s", io->s.op.insert_collision
53 		 ? "collision moving" : "moved",
54 		 pkey(&io->w->key));
55 
56 	bch_keybuf_del(&io->s.op.c->moving_gc_keys, io->w);
57 
58 	atomic_dec_bug(&io->s.op.c->in_flight);
59 	closure_wake_up(&io->s.op.c->moving_gc_wait);
60 
61 	closure_return_with_destructor(cl, moving_io_destructor);
62 }
63 
64 static void read_moving_endio(struct bio *bio, int error)
65 {
66 	struct moving_io *io = container_of(bio->bi_private,
67 					    struct moving_io, s.cl);
68 
69 	if (error)
70 		io->s.error = error;
71 
72 	bch_bbio_endio(io->s.op.c, bio, error, "reading data to move");
73 }
74 
75 static void moving_init(struct moving_io *io)
76 {
77 	struct bio *bio = &io->bio.bio;
78 
79 	bio_init(bio);
80 	bio_get(bio);
81 	bio_set_prio(bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
82 
83 	bio->bi_size		= KEY_SIZE(&io->w->key) << 9;
84 	bio->bi_max_vecs	= DIV_ROUND_UP(KEY_SIZE(&io->w->key),
85 					       PAGE_SECTORS);
86 	bio->bi_private		= &io->s.cl;
87 	bio->bi_io_vec		= bio->bi_inline_vecs;
88 	bch_bio_map(bio, NULL);
89 }
90 
91 static void write_moving(struct closure *cl)
92 {
93 	struct search *s = container_of(cl, struct search, cl);
94 	struct moving_io *io = container_of(s, struct moving_io, s);
95 
96 	if (!s->error) {
97 		trace_bcache_write_moving(&io->bio.bio);
98 
99 		moving_init(io);
100 
101 		io->bio.bio.bi_sector	= KEY_START(&io->w->key);
102 		s->op.lock		= -1;
103 		s->op.write_prio	= 1;
104 		s->op.cache_bio		= &io->bio.bio;
105 
106 		s->writeback		= KEY_DIRTY(&io->w->key);
107 		s->op.csum		= KEY_CSUM(&io->w->key);
108 
109 		s->op.type = BTREE_REPLACE;
110 		bkey_copy(&s->op.replace, &io->w->key);
111 
112 		closure_init(&s->op.cl, cl);
113 		bch_insert_data(&s->op.cl);
114 	}
115 
116 	continue_at(cl, write_moving_finish, NULL);
117 }
118 
119 static void read_moving_submit(struct closure *cl)
120 {
121 	struct search *s = container_of(cl, struct search, cl);
122 	struct moving_io *io = container_of(s, struct moving_io, s);
123 	struct bio *bio = &io->bio.bio;
124 
125 	trace_bcache_read_moving(bio);
126 	bch_submit_bbio(bio, s->op.c, &io->w->key, 0);
127 
128 	continue_at(cl, write_moving, bch_gc_wq);
129 }
130 
131 static void read_moving(struct closure *cl)
132 {
133 	struct cache_set *c = container_of(cl, struct cache_set, moving_gc);
134 	struct keybuf_key *w;
135 	struct moving_io *io;
136 	struct bio *bio;
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, &MAX_KEY);
142 		if (!w)
143 			break;
144 
145 		io = kzalloc(sizeof(struct moving_io) + sizeof(struct bio_vec)
146 			     * DIV_ROUND_UP(KEY_SIZE(&w->key), PAGE_SECTORS),
147 			     GFP_KERNEL);
148 		if (!io)
149 			goto err;
150 
151 		w->private	= io;
152 		io->w		= w;
153 		io->s.op.inode	= KEY_INODE(&w->key);
154 		io->s.op.c	= c;
155 
156 		moving_init(io);
157 		bio = &io->bio.bio;
158 
159 		bio->bi_rw	= READ;
160 		bio->bi_end_io	= read_moving_endio;
161 
162 		if (bch_bio_alloc_pages(bio, GFP_KERNEL))
163 			goto err;
164 
165 		pr_debug("%s", pkey(&w->key));
166 
167 		closure_call(&io->s.cl, read_moving_submit, NULL, &c->gc.cl);
168 
169 		if (atomic_inc_return(&c->in_flight) >= 64) {
170 			closure_wait_event(&c->moving_gc_wait, cl,
171 					   atomic_read(&c->in_flight) < 64);
172 			continue_at(cl, read_moving, bch_gc_wq);
173 		}
174 	}
175 
176 	if (0) {
177 err:		if (!IS_ERR_OR_NULL(w->private))
178 			kfree(w->private);
179 
180 		bch_keybuf_del(&c->moving_gc_keys, w);
181 	}
182 
183 	closure_return(cl);
184 }
185 
186 static bool bucket_cmp(struct bucket *l, struct bucket *r)
187 {
188 	return GC_SECTORS_USED(l) < GC_SECTORS_USED(r);
189 }
190 
191 static unsigned bucket_heap_top(struct cache *ca)
192 {
193 	return GC_SECTORS_USED(heap_peek(&ca->heap));
194 }
195 
196 void bch_moving_gc(struct closure *cl)
197 {
198 	struct cache_set *c = container_of(cl, struct cache_set, gc.cl);
199 	struct cache *ca;
200 	struct bucket *b;
201 	unsigned i;
202 
203 	if (!c->copy_gc_enabled)
204 		closure_return(cl);
205 
206 	mutex_lock(&c->bucket_lock);
207 
208 	for_each_cache(ca, c, i) {
209 		unsigned sectors_to_move = 0;
210 		unsigned reserve_sectors = ca->sb.bucket_size *
211 			min(fifo_used(&ca->free), ca->free.size / 2);
212 
213 		ca->heap.used = 0;
214 
215 		for_each_bucket(b, ca) {
216 			if (!GC_SECTORS_USED(b))
217 				continue;
218 
219 			if (!heap_full(&ca->heap)) {
220 				sectors_to_move += GC_SECTORS_USED(b);
221 				heap_add(&ca->heap, b, bucket_cmp);
222 			} else if (bucket_cmp(b, heap_peek(&ca->heap))) {
223 				sectors_to_move -= bucket_heap_top(ca);
224 				sectors_to_move += GC_SECTORS_USED(b);
225 
226 				ca->heap.data[0] = b;
227 				heap_sift(&ca->heap, 0, bucket_cmp);
228 			}
229 		}
230 
231 		while (sectors_to_move > reserve_sectors) {
232 			heap_pop(&ca->heap, b, bucket_cmp);
233 			sectors_to_move -= GC_SECTORS_USED(b);
234 		}
235 
236 		ca->gc_move_threshold = bucket_heap_top(ca);
237 
238 		pr_debug("threshold %u", ca->gc_move_threshold);
239 	}
240 
241 	mutex_unlock(&c->bucket_lock);
242 
243 	c->moving_gc_keys.last_scanned = ZERO_KEY;
244 
245 	closure_init(&c->moving_gc, cl);
246 	read_moving(&c->moving_gc);
247 
248 	closure_return(cl);
249 }
250 
251 void bch_moving_init_cache_set(struct cache_set *c)
252 {
253 	bch_keybuf_init(&c->moving_gc_keys, moving_pred);
254 }
255