blob: 08361b6aad5044f5f37a18e6eb162b034848578b [file] [log] [blame]
Andrew Morton3fcfab12006-10-19 23:28:16 -07001
2#include <linux/wait.h>
3#include <linux/backing-dev.h>
4#include <linux/fs.h>
5#include <linux/sched.h>
6#include <linux/module.h>
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -07007#include <linux/writeback.h>
8#include <linux/device.h>
9
10
11static struct class *bdi_class;
12
13static ssize_t read_ahead_kb_store(struct device *dev,
14 struct device_attribute *attr,
15 const char *buf, size_t count)
16{
17 struct backing_dev_info *bdi = dev_get_drvdata(dev);
18 char *end;
19 unsigned long read_ahead_kb;
20 ssize_t ret = -EINVAL;
21
22 read_ahead_kb = simple_strtoul(buf, &end, 10);
23 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
24 bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
25 ret = count;
26 }
27 return ret;
28}
29
30#define K(pages) ((pages) << (PAGE_SHIFT - 10))
31
32#define BDI_SHOW(name, expr) \
33static ssize_t name##_show(struct device *dev, \
34 struct device_attribute *attr, char *page) \
35{ \
36 struct backing_dev_info *bdi = dev_get_drvdata(dev); \
37 \
38 return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \
39}
40
41BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
42
43BDI_SHOW(reclaimable_kb, K(bdi_stat(bdi, BDI_RECLAIMABLE)))
44BDI_SHOW(writeback_kb, K(bdi_stat(bdi, BDI_WRITEBACK)))
45
46static inline unsigned long get_dirty(struct backing_dev_info *bdi, int i)
47{
48 unsigned long thresh[3];
49
50 get_dirty_limits(&thresh[0], &thresh[1], &thresh[2], bdi);
51
52 return thresh[i];
53}
54
55BDI_SHOW(dirty_kb, K(get_dirty(bdi, 1)))
56BDI_SHOW(bdi_dirty_kb, K(get_dirty(bdi, 2)))
57
Peter Zijlstra189d3c42008-04-30 00:54:35 -070058static ssize_t min_ratio_store(struct device *dev,
59 struct device_attribute *attr, const char *buf, size_t count)
60{
61 struct backing_dev_info *bdi = dev_get_drvdata(dev);
62 char *end;
63 unsigned int ratio;
64 ssize_t ret = -EINVAL;
65
66 ratio = simple_strtoul(buf, &end, 10);
67 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
68 ret = bdi_set_min_ratio(bdi, ratio);
69 if (!ret)
70 ret = count;
71 }
72 return ret;
73}
74BDI_SHOW(min_ratio, bdi->min_ratio)
75
Peter Zijlstraa42dde02008-04-30 00:54:36 -070076static ssize_t max_ratio_store(struct device *dev,
77 struct device_attribute *attr, const char *buf, size_t count)
78{
79 struct backing_dev_info *bdi = dev_get_drvdata(dev);
80 char *end;
81 unsigned int ratio;
82 ssize_t ret = -EINVAL;
83
84 ratio = simple_strtoul(buf, &end, 10);
85 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
86 ret = bdi_set_max_ratio(bdi, ratio);
87 if (!ret)
88 ret = count;
89 }
90 return ret;
91}
92BDI_SHOW(max_ratio, bdi->max_ratio)
93
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -070094#define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store)
95
96static struct device_attribute bdi_dev_attrs[] = {
97 __ATTR_RW(read_ahead_kb),
98 __ATTR_RO(reclaimable_kb),
99 __ATTR_RO(writeback_kb),
100 __ATTR_RO(dirty_kb),
101 __ATTR_RO(bdi_dirty_kb),
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700102 __ATTR_RW(min_ratio),
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700103 __ATTR_RW(max_ratio),
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700104 __ATTR_NULL,
105};
106
107static __init int bdi_class_init(void)
108{
109 bdi_class = class_create(THIS_MODULE, "bdi");
110 bdi_class->dev_attrs = bdi_dev_attrs;
111 return 0;
112}
113
114core_initcall(bdi_class_init);
115
116int bdi_register(struct backing_dev_info *bdi, struct device *parent,
117 const char *fmt, ...)
118{
119 char *name;
120 va_list args;
121 int ret = 0;
122 struct device *dev;
123
124 va_start(args, fmt);
125 name = kvasprintf(GFP_KERNEL, fmt, args);
126 va_end(args);
127
128 if (!name)
129 return -ENOMEM;
130
131 dev = device_create(bdi_class, parent, MKDEV(0, 0), name);
132 if (IS_ERR(dev)) {
133 ret = PTR_ERR(dev);
134 goto exit;
135 }
136
137 bdi->dev = dev;
138 dev_set_drvdata(bdi->dev, bdi);
139
140exit:
141 kfree(name);
142 return ret;
143}
144EXPORT_SYMBOL(bdi_register);
145
146int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
147{
148 return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
149}
150EXPORT_SYMBOL(bdi_register_dev);
151
152void bdi_unregister(struct backing_dev_info *bdi)
153{
154 if (bdi->dev) {
155 device_unregister(bdi->dev);
156 bdi->dev = NULL;
157 }
158}
159EXPORT_SYMBOL(bdi_unregister);
Andrew Morton3fcfab12006-10-19 23:28:16 -0700160
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700161int bdi_init(struct backing_dev_info *bdi)
162{
Denis Cheng4b01a0b2007-12-04 23:45:07 -0800163 int i;
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700164 int err;
165
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700166 bdi->dev = NULL;
167
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700168 bdi->min_ratio = 0;
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700169 bdi->max_ratio = 100;
170 bdi->max_prop_frac = PROP_FRAC_BASE;
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700171
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700172 for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
173 err = percpu_counter_init_irq(&bdi->bdi_stat[i], 0);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700174 if (err)
175 goto err;
176 }
177
178 bdi->dirty_exceeded = 0;
179 err = prop_local_init_percpu(&bdi->completions);
180
181 if (err) {
182err:
Denis Cheng4b01a0b2007-12-04 23:45:07 -0800183 while (i--)
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700184 percpu_counter_destroy(&bdi->bdi_stat[i]);
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700185 }
186
187 return err;
188}
189EXPORT_SYMBOL(bdi_init);
190
191void bdi_destroy(struct backing_dev_info *bdi)
192{
193 int i;
194
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700195 bdi_unregister(bdi);
196
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700197 for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
198 percpu_counter_destroy(&bdi->bdi_stat[i]);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700199
200 prop_local_destroy_percpu(&bdi->completions);
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700201}
202EXPORT_SYMBOL(bdi_destroy);
203
Andrew Morton3fcfab12006-10-19 23:28:16 -0700204static wait_queue_head_t congestion_wqh[2] = {
205 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
206 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
207 };
208
209
210void clear_bdi_congested(struct backing_dev_info *bdi, int rw)
211{
212 enum bdi_state bit;
213 wait_queue_head_t *wqh = &congestion_wqh[rw];
214
215 bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested;
216 clear_bit(bit, &bdi->state);
217 smp_mb__after_clear_bit();
218 if (waitqueue_active(wqh))
219 wake_up(wqh);
220}
221EXPORT_SYMBOL(clear_bdi_congested);
222
223void set_bdi_congested(struct backing_dev_info *bdi, int rw)
224{
225 enum bdi_state bit;
226
227 bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested;
228 set_bit(bit, &bdi->state);
229}
230EXPORT_SYMBOL(set_bdi_congested);
231
232/**
233 * congestion_wait - wait for a backing_dev to become uncongested
234 * @rw: READ or WRITE
235 * @timeout: timeout in jiffies
236 *
237 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
238 * write congestion. If no backing_devs are congested then just wait for the
239 * next write to be completed.
240 */
241long congestion_wait(int rw, long timeout)
242{
243 long ret;
244 DEFINE_WAIT(wait);
245 wait_queue_head_t *wqh = &congestion_wqh[rw];
246
247 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
248 ret = io_schedule_timeout(timeout);
249 finish_wait(wqh, &wait);
250 return ret;
251}
252EXPORT_SYMBOL(congestion_wait);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700253