summaryrefslogtreecommitdiff
path: root/libbcache/tier.c
blob: 4686459433e9d12725b2dde1dd79150f96ca6243 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247

#include "bcache.h"
#include "alloc.h"
#include "btree_iter.h"
#include "buckets.h"
#include "clock.h"
#include "extents.h"
#include "io.h"
#include "keylist.h"
#include "move.h"
#include "super-io.h"
#include "tier.h"

#include <linux/freezer.h>
#include <linux/kthread.h>
#include <trace/events/bcache.h>

struct tiering_state {
	struct cache_group	*tier;
	unsigned		tier_idx;
	unsigned		sectors;
	unsigned		stripe_size;
	unsigned		dev_idx;
	struct cache		*ca;
};

static bool tiering_pred(struct cache_set *c,
			 struct tiering_state *s,
			 struct bkey_s_c k)
{
	if (bkey_extent_is_data(k.k)) {
		struct bkey_s_c_extent e = bkey_s_c_to_extent(k);
		const struct bch_extent_ptr *ptr;
		struct cache_member_rcu *mi;
		unsigned replicas = 0;

		/* Make sure we have room to add a new pointer: */
		if (bkey_val_u64s(e.k) + BKEY_EXTENT_PTR_U64s_MAX >
		    BKEY_EXTENT_VAL_U64s_MAX)
			return false;

		mi = cache_member_info_get(c);
		extent_for_each_ptr(e, ptr)
			if (ptr->dev < mi->nr_devices &&
			    mi->m[ptr->dev].tier >= s->tier_idx)
				replicas++;
		cache_member_info_put();

		return replicas < c->opts.data_replicas;
	}

	return false;
}

static void tier_put_device(struct tiering_state *s)
{
	if (s->ca)
		percpu_ref_put(&s->ca->ref);
	s->ca = NULL;
}

/**
 * refill_next - move on to refilling the next cache's tiering keylist
 */
static void tier_next_device(struct cache_set *c, struct tiering_state *s)
{
	if (!s->ca || s->sectors > s->stripe_size) {
		tier_put_device(s);
		s->sectors = 0;
		s->dev_idx++;

		spin_lock(&s->tier->lock);
		if (s->dev_idx >= s->tier->nr_devices)
			s->dev_idx = 0;

		if (s->tier->nr_devices) {
			s->ca = s->tier->d[s->dev_idx].dev;
			percpu_ref_get(&s->ca->ref);
		}
		spin_unlock(&s->tier->lock);
	}
}

static int issue_tiering_move(struct cache_set *c,
			      struct tiering_state *s,
			      struct moving_context *ctxt,
			      struct bkey_s_c k)
{
	int ret;

	ret = bch_data_move(c, ctxt, &s->ca->tiering_write_point, k, NULL);
	if (!ret) {
		trace_bcache_tiering_copy(k.k);
		s->sectors += k.k->size;
	} else {
		trace_bcache_tiering_alloc_fail(c, k.k->size);
	}

	return ret;
}

/**
 * tiering_next_cache - issue a move to write an extent to the next cache
 * device in round robin order
 */
static s64 read_tiering(struct cache_set *c, struct cache_group *tier)
{
	struct moving_context ctxt;
	struct tiering_state s;
	struct btree_iter iter;
	struct bkey_s_c k;
	unsigned nr_devices = READ_ONCE(tier->nr_devices);
	int ret;

	if (!nr_devices)
		return 0;

	trace_bcache_tiering_start(c);

	memset(&s, 0, sizeof(s));
	s.tier		= tier;
	s.tier_idx	= tier - c->cache_tiers;
	s.stripe_size	= 2048; /* 1 mb for now */

	bch_move_ctxt_init(&ctxt, &c->tiering_pd.rate,
			   nr_devices * SECTORS_IN_FLIGHT_PER_DEVICE);
	bch_btree_iter_init(&iter, c, BTREE_ID_EXTENTS, POS_MIN);

	while (!kthread_should_stop() &&
	       !bch_move_ctxt_wait(&ctxt) &&
	       (k = bch_btree_iter_peek(&iter)).k &&
	       !btree_iter_err(k)) {
		if (!tiering_pred(c, &s, k))
			goto next;

		tier_next_device(c, &s);
		if (!s.ca)
			break;

		ret = issue_tiering_move(c, &s, &ctxt, k);
		if (ret) {
			bch_btree_iter_unlock(&iter);

			/* memory allocation failure, wait for some IO to finish */
			bch_move_ctxt_wait_for_io(&ctxt);
			continue;
		}
next:
		bch_btree_iter_advance_pos(&iter);
		//bch_btree_iter_cond_resched(&iter);

		/* unlock before calling moving_context_wait() */
		bch_btree_iter_unlock(&iter);
		cond_resched();
	}

	bch_btree_iter_unlock(&iter);
	tier_put_device(&s);
	bch_move_ctxt_exit(&ctxt);
	trace_bcache_tiering_end(c, ctxt.sectors_moved, ctxt.keys_moved);

	return ctxt.sectors_moved;
}

static int bch_tiering_thread(void *arg)
{
	struct cache_set *c = arg;
	struct cache_group *tier = &c->cache_tiers[1];
	struct io_clock *clock = &c->io_clock[WRITE];
	struct cache *ca;
	u64 tier_capacity, available_sectors;
	unsigned long last;
	unsigned i;

	set_freezable();

	while (!kthread_should_stop()) {
		if (kthread_wait_freezable(c->tiering_enabled &&
					   tier->nr_devices))
			break;

		while (1) {
			struct cache_group *faster_tier;

			last = atomic_long_read(&clock->now);

			tier_capacity = available_sectors = 0;
			rcu_read_lock();
			for (faster_tier = c->cache_tiers;
			     faster_tier != tier;
			     faster_tier++) {
				group_for_each_cache_rcu(ca, faster_tier, i) {
					tier_capacity +=
						(ca->mi.nbuckets -
						 ca->mi.first_bucket) << ca->bucket_bits;
					available_sectors +=
						buckets_available_cache(ca) << ca->bucket_bits;
				}
			}
			rcu_read_unlock();

			if (available_sectors < (tier_capacity >> 1))
				break;

			bch_kthread_io_clock_wait(clock,
						  last +
						  available_sectors -
						  (tier_capacity >> 1));
			if (kthread_should_stop())
				return 0;
		}

		read_tiering(c, tier);
	}

	return 0;
}

void bch_tiering_init_cache_set(struct cache_set *c)
{
	bch_pd_controller_init(&c->tiering_pd);
}

int bch_tiering_read_start(struct cache_set *c)
{
	struct task_struct *t;

	if (c->opts.nochanges)
		return 0;

	t = kthread_create(bch_tiering_thread, c, "bch_tier_read");
	if (IS_ERR(t))
		return PTR_ERR(t);

	c->tiering_read = t;
	wake_up_process(c->tiering_read);

	return 0;
}

void bch_tiering_read_stop(struct cache_set *c)
{
	if (!IS_ERR_OR_NULL(c->tiering_read)) {
		kthread_stop(c->tiering_read);
		c->tiering_read = NULL;
	}
}