xref: /linux/mm/backing-dev.c (revision f7511d5f66f01fc451747b24e79f3ada7a3af9af)
1 
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>
7 #include <linux/writeback.h>
8 #include <linux/device.h>
9 
10 
11 static struct class *bdi_class;
12 
13 #ifdef CONFIG_DEBUG_FS
14 #include <linux/debugfs.h>
15 #include <linux/seq_file.h>
16 
17 static struct dentry *bdi_debug_root;
18 
19 static void bdi_debug_init(void)
20 {
21 	bdi_debug_root = debugfs_create_dir("bdi", NULL);
22 }
23 
24 static int bdi_debug_stats_show(struct seq_file *m, void *v)
25 {
26 	struct backing_dev_info *bdi = m->private;
27 	long background_thresh;
28 	long dirty_thresh;
29 	long bdi_thresh;
30 
31 	get_dirty_limits(&background_thresh, &dirty_thresh, &bdi_thresh, bdi);
32 
33 #define K(x) ((x) << (PAGE_SHIFT - 10))
34 	seq_printf(m,
35 		   "BdiWriteback:     %8lu kB\n"
36 		   "BdiReclaimable:   %8lu kB\n"
37 		   "BdiDirtyThresh:   %8lu kB\n"
38 		   "DirtyThresh:      %8lu kB\n"
39 		   "BackgroundThresh: %8lu kB\n",
40 		   (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)),
41 		   (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)),
42 		   K(bdi_thresh),
43 		   K(dirty_thresh),
44 		   K(background_thresh));
45 #undef K
46 
47 	return 0;
48 }
49 
50 static int bdi_debug_stats_open(struct inode *inode, struct file *file)
51 {
52 	return single_open(file, bdi_debug_stats_show, inode->i_private);
53 }
54 
55 static const struct file_operations bdi_debug_stats_fops = {
56 	.open		= bdi_debug_stats_open,
57 	.read		= seq_read,
58 	.llseek		= seq_lseek,
59 	.release	= single_release,
60 };
61 
62 static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
63 {
64 	bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
65 	bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
66 					       bdi, &bdi_debug_stats_fops);
67 }
68 
69 static void bdi_debug_unregister(struct backing_dev_info *bdi)
70 {
71 	debugfs_remove(bdi->debug_stats);
72 	debugfs_remove(bdi->debug_dir);
73 }
74 #else
75 static inline void bdi_debug_init(void)
76 {
77 }
78 static inline void bdi_debug_register(struct backing_dev_info *bdi,
79 				      const char *name)
80 {
81 }
82 static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
83 {
84 }
85 #endif
86 
87 static ssize_t read_ahead_kb_store(struct device *dev,
88 				  struct device_attribute *attr,
89 				  const char *buf, size_t count)
90 {
91 	struct backing_dev_info *bdi = dev_get_drvdata(dev);
92 	char *end;
93 	unsigned long read_ahead_kb;
94 	ssize_t ret = -EINVAL;
95 
96 	read_ahead_kb = simple_strtoul(buf, &end, 10);
97 	if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
98 		bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
99 		ret = count;
100 	}
101 	return ret;
102 }
103 
104 #define K(pages) ((pages) << (PAGE_SHIFT - 10))
105 
106 #define BDI_SHOW(name, expr)						\
107 static ssize_t name##_show(struct device *dev,				\
108 			   struct device_attribute *attr, char *page)	\
109 {									\
110 	struct backing_dev_info *bdi = dev_get_drvdata(dev);		\
111 									\
112 	return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr);	\
113 }
114 
115 BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
116 
117 static ssize_t min_ratio_store(struct device *dev,
118 		struct device_attribute *attr, const char *buf, size_t count)
119 {
120 	struct backing_dev_info *bdi = dev_get_drvdata(dev);
121 	char *end;
122 	unsigned int ratio;
123 	ssize_t ret = -EINVAL;
124 
125 	ratio = simple_strtoul(buf, &end, 10);
126 	if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
127 		ret = bdi_set_min_ratio(bdi, ratio);
128 		if (!ret)
129 			ret = count;
130 	}
131 	return ret;
132 }
133 BDI_SHOW(min_ratio, bdi->min_ratio)
134 
135 static ssize_t max_ratio_store(struct device *dev,
136 		struct device_attribute *attr, const char *buf, size_t count)
137 {
138 	struct backing_dev_info *bdi = dev_get_drvdata(dev);
139 	char *end;
140 	unsigned int ratio;
141 	ssize_t ret = -EINVAL;
142 
143 	ratio = simple_strtoul(buf, &end, 10);
144 	if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
145 		ret = bdi_set_max_ratio(bdi, ratio);
146 		if (!ret)
147 			ret = count;
148 	}
149 	return ret;
150 }
151 BDI_SHOW(max_ratio, bdi->max_ratio)
152 
153 #define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store)
154 
155 static struct device_attribute bdi_dev_attrs[] = {
156 	__ATTR_RW(read_ahead_kb),
157 	__ATTR_RW(min_ratio),
158 	__ATTR_RW(max_ratio),
159 	__ATTR_NULL,
160 };
161 
162 static __init int bdi_class_init(void)
163 {
164 	bdi_class = class_create(THIS_MODULE, "bdi");
165 	bdi_class->dev_attrs = bdi_dev_attrs;
166 	bdi_debug_init();
167 	return 0;
168 }
169 
170 postcore_initcall(bdi_class_init);
171 
172 int bdi_register(struct backing_dev_info *bdi, struct device *parent,
173 		const char *fmt, ...)
174 {
175 	char *name;
176 	va_list args;
177 	int ret = 0;
178 	struct device *dev;
179 
180 	va_start(args, fmt);
181 	name = kvasprintf(GFP_KERNEL, fmt, args);
182 	va_end(args);
183 
184 	if (!name)
185 		return -ENOMEM;
186 
187 	dev = device_create(bdi_class, parent, MKDEV(0, 0), name);
188 	if (IS_ERR(dev)) {
189 		ret = PTR_ERR(dev);
190 		goto exit;
191 	}
192 
193 	bdi->dev = dev;
194 	dev_set_drvdata(bdi->dev, bdi);
195 	bdi_debug_register(bdi, name);
196 
197 exit:
198 	kfree(name);
199 	return ret;
200 }
201 EXPORT_SYMBOL(bdi_register);
202 
203 int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
204 {
205 	return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
206 }
207 EXPORT_SYMBOL(bdi_register_dev);
208 
209 void bdi_unregister(struct backing_dev_info *bdi)
210 {
211 	if (bdi->dev) {
212 		bdi_debug_unregister(bdi);
213 		device_unregister(bdi->dev);
214 		bdi->dev = NULL;
215 	}
216 }
217 EXPORT_SYMBOL(bdi_unregister);
218 
219 int bdi_init(struct backing_dev_info *bdi)
220 {
221 	int i;
222 	int err;
223 
224 	bdi->dev = NULL;
225 
226 	bdi->min_ratio = 0;
227 	bdi->max_ratio = 100;
228 	bdi->max_prop_frac = PROP_FRAC_BASE;
229 
230 	for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
231 		err = percpu_counter_init_irq(&bdi->bdi_stat[i], 0);
232 		if (err)
233 			goto err;
234 	}
235 
236 	bdi->dirty_exceeded = 0;
237 	err = prop_local_init_percpu(&bdi->completions);
238 
239 	if (err) {
240 err:
241 		while (i--)
242 			percpu_counter_destroy(&bdi->bdi_stat[i]);
243 	}
244 
245 	return err;
246 }
247 EXPORT_SYMBOL(bdi_init);
248 
249 void bdi_destroy(struct backing_dev_info *bdi)
250 {
251 	int i;
252 
253 	bdi_unregister(bdi);
254 
255 	for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
256 		percpu_counter_destroy(&bdi->bdi_stat[i]);
257 
258 	prop_local_destroy_percpu(&bdi->completions);
259 }
260 EXPORT_SYMBOL(bdi_destroy);
261 
262 static wait_queue_head_t congestion_wqh[2] = {
263 		__WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
264 		__WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
265 	};
266 
267 
268 void clear_bdi_congested(struct backing_dev_info *bdi, int rw)
269 {
270 	enum bdi_state bit;
271 	wait_queue_head_t *wqh = &congestion_wqh[rw];
272 
273 	bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested;
274 	clear_bit(bit, &bdi->state);
275 	smp_mb__after_clear_bit();
276 	if (waitqueue_active(wqh))
277 		wake_up(wqh);
278 }
279 EXPORT_SYMBOL(clear_bdi_congested);
280 
281 void set_bdi_congested(struct backing_dev_info *bdi, int rw)
282 {
283 	enum bdi_state bit;
284 
285 	bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested;
286 	set_bit(bit, &bdi->state);
287 }
288 EXPORT_SYMBOL(set_bdi_congested);
289 
290 /**
291  * congestion_wait - wait for a backing_dev to become uncongested
292  * @rw: READ or WRITE
293  * @timeout: timeout in jiffies
294  *
295  * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
296  * write congestion.  If no backing_devs are congested then just wait for the
297  * next write to be completed.
298  */
299 long congestion_wait(int rw, long timeout)
300 {
301 	long ret;
302 	DEFINE_WAIT(wait);
303 	wait_queue_head_t *wqh = &congestion_wqh[rw];
304 
305 	prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
306 	ret = io_schedule_timeout(timeout);
307 	finish_wait(wqh, &wait);
308 	return ret;
309 }
310 EXPORT_SYMBOL(congestion_wait);
311 
312