summaryrefslogtreecommitdiff
path: root/block/blk-ioc.c
blob: 2056412ba5e74c9ec9675838255cd90ca3ebec5b (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
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
/*
 * Functions related to io context handling
 */
#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/init.h>
#include <linux/bio.h>
#include <linux/blkdev.h>
#include <linux/bootmem.h>	/* for max_pfn/max_low_pfn */
#include <linux/slab.h>

#include "blk.h"

/*
 * For io context allocations
 */
static struct kmem_cache *iocontext_cachep;

#define IOC_RQ_FREE_TIMEOUT	(5 * HZ)

struct request *blk_ioc_get_cached_request(struct io_context *ioc)
{
	struct request *rq = NULL;

	/*
	 * See if there are available cached requests
	 */
	if (!list_empty_careful(&ioc->free_list)) {
		spin_lock_irq(&ioc->lock);
		if (!list_empty(&ioc->free_list)) {
			rq = list_entry(ioc->free_list.next, struct request,
					queuelist);
			list_del_init(&rq->queuelist);
		}
		spin_unlock_irq(&ioc->lock);
	}

	return rq;
}

/*
 * We abuse the stat_time of the request for knowing when this was
 * cached. This is completely safe, since the request is unused at
 * this point.
 */
bool blk_ioc_cache_request(struct request_queue *q, struct io_context *ioc,
			   struct request *rq)
{
	unsigned long now = jiffies;

	/*
	 * Don't cache something that came from the emergency pool
	 */
	if (rq->cmd_flags & REQ_MEMPOOL)
		return false;

	/*
	 * cache this request for reuse
	 */
	if ((rq->cmd_flags & REQ_ALLOCED) &&
	    (ioc->free_requests < 2 * q->nr_requests)) {
		rq->start_time = now + IOC_RQ_FREE_TIMEOUT;
		list_add(&rq->queuelist, &ioc->free_list);

		if (!timer_pending(&ioc->request_timer) &&
	    	    atomic_long_inc_not_zero(&ioc->refcount)) {
			mod_timer(&ioc->request_timer,
				  round_jiffies_up(now + IOC_RQ_FREE_TIMEOUT));
		}

		return true;
	}

	return false;
}

static void cfq_dtor(struct io_context *ioc)
{
	if (!hlist_empty(&ioc->cic_list)) {
		struct cfq_io_context *cic;

		cic = list_entry(ioc->cic_list.first, struct cfq_io_context,
								cic_list);
		cic->dtor(ioc);
	}
}

/*
 * IO Context helper functions. put_io_context() returns 1 if there are no
 * more users of this io context, 0 otherwise.
 */
int put_io_context(struct io_context *ioc)
{
	if (ioc == NULL)
		return 1;

	BUG_ON(atomic_long_read(&ioc->refcount) == 0);

	if (atomic_long_dec_and_test(&ioc->refcount)) {
		rcu_read_lock();
		cfq_dtor(ioc);
		rcu_read_unlock();

		BUG_ON(timer_pending(&ioc->request_timer));

		/*
		 * free cached requests, if any. this is safe without the
		 * lock, since we are the only ones that are still holding
		 * a reference to it and others cannot grab a ref to it now.
		 */
		while (!list_empty(&ioc->free_list)) {
			struct request *rq;

			rq = list_entry(ioc->free_list.next, struct request,
					queuelist);
			list_del(&rq->queuelist);
			blk_slab_free_request(rq);
		}

		kmem_cache_free(iocontext_cachep, ioc);
		return 1;
	}
	return 0;
}
EXPORT_SYMBOL(put_io_context);

static void cfq_exit(struct io_context *ioc)
{
	rcu_read_lock();

	if (!hlist_empty(&ioc->cic_list)) {
		struct cfq_io_context *cic;

		cic = list_entry(ioc->cic_list.first, struct cfq_io_context,
								cic_list);
		cic->exit(ioc);
	}
	rcu_read_unlock();
}

/*
 * Check if we need to prune requests from the cache
 */
static void ioc_request_timer(unsigned long data)
{
	struct io_context *ioc = (struct io_context *) data;
	unsigned long flags, now;
	bool put_ioc = false;

	/*
	 * We can wait, if someone else is holding the lock at the moment
	 */
	if (!spin_trylock_irqsave(&ioc->lock, flags))
		return;

	now = jiffies;

	while (!list_empty(&ioc->free_list)) {
		struct request *rq;

		rq = list_entry(ioc->free_list.prev, struct request,
				queuelist);
		if (!time_after_eq(rq->start_time, now))
			break;

		list_del(&rq->queuelist);
		blk_slab_free_request(rq);
	}

	/*
	 * We already have a ref to this ioc. Drop it, if there are no more
	 * requests left
	 */
	if (!list_empty(&ioc->free_list))
		mod_timer(&ioc->request_timer,
				round_jiffies_up(now + IOC_RQ_FREE_TIMEOUT));
	else
		put_ioc = true;

	spin_unlock_irqrestore(&ioc->lock, flags);

	if (put_ioc)
		put_io_context(ioc);
}

/* Called by the exiting task */
void exit_io_context(struct task_struct *task)
{
	struct io_context *ioc;

	task_lock(task);
	ioc = task->io_context;
	task->io_context = NULL;
	task_unlock(task);

	if (atomic_dec_and_test(&ioc->nr_tasks))
		cfq_exit(ioc);

	put_io_context(ioc);
}

struct io_context *alloc_io_context(gfp_t gfp_flags, int node)
{
	struct io_context *ret;

	ret = kmem_cache_alloc_node(iocontext_cachep, gfp_flags, node);
	if (ret) {
		atomic_long_set(&ret->refcount, 1);
		atomic_set(&ret->nr_tasks, 1);
		spin_lock_init(&ret->lock);
		ret->ioprio_changed = 0;
		ret->ioprio = 0;
		init_waitqueue_head(&ret->wait[BLK_RW_SYNC]);
		init_waitqueue_head(&ret->wait[BLK_RW_ASYNC]);
		INIT_LIST_HEAD(&ret->free_list);
		ret->free_requests = 0;
		ret->count[BLK_RW_SYNC] = ret->count[BLK_RW_ASYNC] = 0;
		setup_timer(&ret->request_timer, ioc_request_timer,
				(unsigned long) ret);
		init_timer(&ret->request_timer);
		INIT_RADIX_TREE(&ret->radix_root, GFP_ATOMIC | __GFP_HIGH);
		INIT_HLIST_HEAD(&ret->cic_list);
		ret->ioc_data = NULL;
	}

	return ret;
}

/*
 * If the current task has no IO context then create one and initialise it.
 * Otherwise, return its existing IO context.
 *
 * This returned IO context doesn't have a specifically elevated refcount,
 * but since the current task itself holds a reference, the context can be
 * used in general code, so long as it stays within `current` context.
 */
struct io_context *current_io_context(gfp_t gfp_flags, int node)
{
	struct task_struct *tsk = current;
	struct io_context *ret;

	ret = tsk->io_context;
	if (likely(ret))
		return ret;

	ret = alloc_io_context(gfp_flags, node);
	if (ret) {
		/* make sure set_task_ioprio() sees the settings above */
		smp_wmb();
		tsk->io_context = ret;
	}

	return ret;
}

/*
 * If the current task has no IO context then create one and initialise it.
 * If it does have a context, take a ref on it.
 *
 * This is always called in the context of the task which submitted the I/O.
 */
struct io_context *get_io_context(gfp_t gfp_flags, int node)
{
	struct io_context *ret = NULL;

	/*
	 * Check for unlikely race with exiting task. ioc ref count is
	 * zero when ioc is being detached.
	 */
	do {
		ret = current_io_context(gfp_flags, node);
		if (unlikely(!ret))
			break;
	} while (!atomic_long_inc_not_zero(&ret->refcount));

	return ret;
}
EXPORT_SYMBOL(get_io_context);

static int __init blk_ioc_init(void)
{
	iocontext_cachep = kmem_cache_create("blkdev_ioc",
			sizeof(struct io_context), 0, SLAB_PANIC, NULL);
	return 0;
}
subsys_initcall(blk_ioc_init);