xref: /openbmc/linux/drivers/md/bcache/movinggc.c (revision 2c684d89)
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)
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 (bio->bi_error)
70 		io->op.error = bio->bi_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, bio->bi_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