[PATCH] sysfs: Allow sysfs attribute files to be pollable

It works like this: Open the file Read all the contents. Call poll requesting POLLERR or POLLPRI (so select/exceptfds works) When poll returns, close the file and go to top of loop. or lseek to start of file and go back to the 'read'. Events are signaled by an object manager calling sysfs_notify(kobj, dir, attr); If the dir is non-NULL, it is used to find a subdirectory which contains the attribute (presumably created by sysfs_create_group). This has a cost of one int per attribute, one wait_queuehead per kobject, one int per open file. The name "sysfs_notify" may be confused with the inotify functionality. Maybe it would be nice to support inotify for sysfs attributes as well? This patch also uses sysfs_notify to allow /sys/block/md*/md/sync_action to be pollable Signed-off-by: Neil Brown <neilb@suse.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

[PATCH] sysfs: Allow sysfs attribute files to be pollable
It works like this: Open the file Read all the contents. Call poll requesting POLLERR or POLLPRI (so select/exceptfds works) When poll returns, close the file and go to top of loop. or lseek to start of file and go back to the 'read'. Events are signaled by an object manager calling sysfs_notify(kobj, dir, attr); If the dir is non-NULL, it is used to find a subdirectory which contains the attribute (presumably created by sysfs_create_group). This has a cost of one int per attribute, one wait_queuehead per kobject, one int per open file. The name "sysfs_notify" may be confused with the inotify functionality. Maybe it would be nice to support inotify for sysfs attributes as well? This patch also uses sysfs_notify to allow /sys/block/md*/md/sync_action to be pollable Signed-off-by: Neil Brown <neilb@suse.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
NeilBrown · Greg Kroah-Hartman
1 parent f043ca43c1
Showing 7 changed files with 88 additions and 0 deletions Side-by-side Diff
drivers/md/md.c
fs/sysfs/dir.c
fs/sysfs/file.c
fs/sysfs/sysfs.h
include/linux/kobject.h
include/linux/sysfs.h
lib/kobject.c
@@ -163,6 +163,7 @@
 {
 	atomic_inc(&md_event_count);
 	wake_up(&md_event_waiters);
+	sysfs_notify(&mddev->kobj, NULL, "sync_action");
 }
 EXPORT_SYMBOL_GPL(md_new_event);
  
@@ -43,6 +43,7 @@
  
 	memset(sd, 0, sizeof(*sd));
 	atomic_set(&sd->s_count, 1);
+	atomic_set(&sd->s_event, 0);
 	INIT_LIST_HEAD(&sd->s_children);
 	list_add(&sd->s_sibling, &parent_sd->s_children);
 	sd->s_element = element;
@@ -6,6 +6,7 @@
 #include <linux/fsnotify.h>
 #include <linux/kobject.h>
 #include <linux/namei.h>
+#include <linux/poll.h>
 #include <asm/uaccess.h>
 #include <asm/semaphore.h>
  
@@ -57,6 +58,7 @@
 	struct sysfs_ops	* ops;
 	struct semaphore	sem;
 	int			needs_read_fill;
+	int			event;
 };
  
  
@@ -72,6 +74,7 @@
  */
 static int fill_read_buffer(struct dentry * dentry, struct sysfs_buffer * buffer)
 {
+	struct sysfs_dirent * sd = dentry->d_fsdata;
 	struct attribute * attr = to_attr(dentry);
 	struct kobject * kobj = to_kobj(dentry->d_parent);
 	struct sysfs_ops * ops = buffer->ops;
@@ -83,6 +86,7 @@
 	if (!buffer->page)
 		return -ENOMEM;
  
+	buffer->event = atomic_read(&sd->s_event);
 	count = ops->show(kobj,attr,buffer->page);
 	buffer->needs_read_fill = 0;
 	BUG_ON(count > (ssize_t)PAGE_SIZE);
  
@@ -348,12 +352,84 @@
 	return 0;
 }
  
+/* Sysfs attribute files are pollable.  The idea is that you read
+ * the content and then you use 'poll' or 'select' to wait for
+ * the content to change.  When the content changes (assuming the
+ * manager for the kobject supports notification), poll will
+ * return POLLERR|POLLPRI, and select will return the fd whether
+ * it is waiting for read, write, or exceptions.
+ * Once poll/select indicates that the value has changed, you
+ * need to close and re-open the file, as simply seeking and reading
+ * again will not get new data, or reset the state of 'poll'.
+ * Reminder: this only works for attributes which actively support
+ * it, and it is not possible to test an attribute from userspace
+ * to see if it supports poll (Nether 'poll' or 'select' return
+ * an appropriate error code).  When in doubt, set a suitable timeout value.
+ */
+static unsigned int sysfs_poll(struct file *filp, poll_table *wait)
+{
+	struct sysfs_buffer * buffer = filp->private_data;
+	struct kobject * kobj = to_kobj(filp->f_dentry->d_parent);
+	struct sysfs_dirent * sd = filp->f_dentry->d_fsdata;
+	int res = 0;
+
+	poll_wait(filp, &kobj->poll, wait);
+
+	if (buffer->event != atomic_read(&sd->s_event)) {
+		res = POLLERR|POLLPRI;
+		buffer->needs_read_fill = 1;
+	}
+
+	return res;
+}
+
+
+static struct dentry *step_down(struct dentry *dir, const char * name)
+{
+	struct dentry * de;
+
+	if (dir == NULL || dir->d_inode == NULL)
+		return NULL;
+
+	mutex_lock(&dir->d_inode->i_mutex);
+	de = lookup_one_len(name, dir, strlen(name));
+	mutex_unlock(&dir->d_inode->i_mutex);
+	dput(dir);
+	if (IS_ERR(de))
+		return NULL;
+	if (de->d_inode == NULL) {
+		dput(de);
+		return NULL;
+	}
+	return de;
+}
+
+void sysfs_notify(struct kobject * k, char *dir, char *attr)
+{
+	struct dentry *de = k->dentry;
+	if (de)
+		dget(de);
+	if (de && dir)
+		de = step_down(de, dir);
+	if (de && attr)
+		de = step_down(de, attr);
+	if (de) {
+		struct sysfs_dirent * sd = de->d_fsdata;
+		if (sd)
+			atomic_inc(&sd->s_event);
+		wake_up_interruptible(&k->poll);
+		dput(de);
+	}
+}
+EXPORT_SYMBOL_GPL(sysfs_notify);
+
 const struct file_operations sysfs_file_operations = {
 	.read		= sysfs_read_file,
 	.write		= sysfs_write_file,
 	.llseek		= generic_file_llseek,
 	.open		= sysfs_open_file,
 	.release	= sysfs_release,
+	.poll		= sysfs_poll,
 };
  
  
@@ -11,6 +11,7 @@
  
 extern int sysfs_add_file(struct dentry *, const struct attribute *, int);
 extern void sysfs_hash_and_remove(struct dentry * dir, const char * name);
+extern struct sysfs_dirent *sysfs_find(struct sysfs_dirent *dir, const char * name);
  
 extern int sysfs_create_subdir(struct kobject *, const char *, struct dentry **);
 extern void sysfs_remove_subdir(struct dentry *);
@@ -24,6 +24,7 @@
 #include <linux/rwsem.h>
 #include <linux/kref.h>
 #include <linux/kernel.h>
+#include <linux/wait.h>
 #include <asm/atomic.h>
  
 #define KOBJ_NAME_LEN			20
@@ -56,6 +57,7 @@
 	struct kset		* kset;
 	struct kobj_type	* ktype;
 	struct dentry		* dentry;
+	wait_queue_head_t	poll;
 };
  
 extern int kobject_set_name(struct kobject *, const char *, ...)
@@ -74,6 +74,7 @@
 	umode_t			s_mode;
 	struct dentry		* s_dentry;
 	struct iattr		* s_iattr;
+	atomic_t		s_event;
 };
  
 #define SYSFS_ROOT		0x0001
@@ -117,6 +118,7 @@
  
 int sysfs_create_group(struct kobject *, const struct attribute_group *);
 void sysfs_remove_group(struct kobject *, const struct attribute_group *);
+void sysfs_notify(struct kobject * k, char *dir, char *attr);
  
 #else /* CONFIG_SYSFS */
  
@@ -183,6 +185,10 @@
 static inline void sysfs_remove_group(struct kobject * k, const struct attribute_group * g)
 {
 	;
+}
+
+static inline void sysfs_notify(struct kobject * k, char *dir, char *attr)
+{
 }
  
 #endif /* CONFIG_SYSFS */
@@ -128,6 +128,7 @@
 {
 	kref_init(&kobj->kref);
 	INIT_LIST_HEAD(&kobj->entry);
+	init_waitqueue_head(&kobj->poll);
 	kobj->kset = kset_get(kobj->kset);
 }
...	...	@@ -163,6 +163,7 @@
163	163	{
164	164	atomic_inc(&md_event_count);
165	165	wake_up(&md_event_waiters);
	166	+ sysfs_notify(&mddev->kobj, NULL, "sync_action");
166	167	}
167	168	EXPORT_SYMBOL_GPL(md_new_event);
168	169
...	...	@@ -43,6 +43,7 @@
43	43
44	44	memset(sd, 0, sizeof(*sd));
45	45	atomic_set(&sd->s_count, 1);
	46	+ atomic_set(&sd->s_event, 0);
46	47	INIT_LIST_HEAD(&sd->s_children);
47	48	list_add(&sd->s_sibling, &parent_sd->s_children);
48	49	sd->s_element = element;
...	...	@@ -6,6 +6,7 @@
6	6	#include <linux/fsnotify.h>
7	7	#include <linux/kobject.h>
8	8	#include <linux/namei.h>
	9	+#include <linux/poll.h>
9	10	#include <asm/uaccess.h>
10	11	#include <asm/semaphore.h>
11	12
...	...	@@ -57,6 +58,7 @@
57	58	struct sysfs_ops * ops;
58	59	struct semaphore sem;
59	60	int needs_read_fill;
	61	+ int event;
60	62	};
61	63
62	64
...	...	@@ -72,6 +74,7 @@
72	74	*/
73	75	static int fill_read_buffer(struct dentry * dentry, struct sysfs_buffer * buffer)
74	76	{
	77	+ struct sysfs_dirent * sd = dentry->d_fsdata;
75	78	struct attribute * attr = to_attr(dentry);
76	79	struct kobject * kobj = to_kobj(dentry->d_parent);
77	80	struct sysfs_ops * ops = buffer->ops;
...	...	@@ -83,6 +86,7 @@
83	86	if (!buffer->page)
84	87	return -ENOMEM;
85	88
	89	+ buffer->event = atomic_read(&sd->s_event);
86	90	count = ops->show(kobj,attr,buffer->page);
87	91	buffer->needs_read_fill = 0;
88	92	BUG_ON(count > (ssize_t)PAGE_SIZE);
89	93
...	...	@@ -348,12 +352,84 @@
348	352	return 0;
349	353	}
350	354
	355	+/* Sysfs attribute files are pollable. The idea is that you read
	356	+ * the content and then you use 'poll' or 'select' to wait for
	357	+ * the content to change. When the content changes (assuming the
	358	+ * manager for the kobject supports notification), poll will
	359	+ * return POLLERR\|POLLPRI, and select will return the fd whether
	360	+ * it is waiting for read, write, or exceptions.
	361	+ * Once poll/select indicates that the value has changed, you
	362	+ * need to close and re-open the file, as simply seeking and reading
	363	+ * again will not get new data, or reset the state of 'poll'.
	364	+ * Reminder: this only works for attributes which actively support
	365	+ * it, and it is not possible to test an attribute from userspace
	366	+ * to see if it supports poll (Nether 'poll' or 'select' return
	367	+ * an appropriate error code). When in doubt, set a suitable timeout value.
	368	+ */
	369	+static unsigned int sysfs_poll(struct file filp, poll_table wait)
	370	+{
	371	+ struct sysfs_buffer * buffer = filp->private_data;
	372	+ struct kobject * kobj = to_kobj(filp->f_dentry->d_parent);
	373	+ struct sysfs_dirent * sd = filp->f_dentry->d_fsdata;
	374	+ int res = 0;
	375	+
	376	+ poll_wait(filp, &kobj->poll, wait);
	377	+
	378	+ if (buffer->event != atomic_read(&sd->s_event)) {
	379	+ res = POLLERR\|POLLPRI;
	380	+ buffer->needs_read_fill = 1;
	381	+ }
	382	+
	383	+ return res;
	384	+}
	385	+
	386	+
	387	+static struct dentry step_down(struct dentry dir, const char * name)
	388	+{
	389	+ struct dentry * de;
	390	+
	391	+ if (dir == NULL \|\| dir->d_inode == NULL)
	392	+ return NULL;
	393	+
	394	+ mutex_lock(&dir->d_inode->i_mutex);
	395	+ de = lookup_one_len(name, dir, strlen(name));
	396	+ mutex_unlock(&dir->d_inode->i_mutex);
	397	+ dput(dir);
	398	+ if (IS_ERR(de))
	399	+ return NULL;
	400	+ if (de->d_inode == NULL) {
	401	+ dput(de);
	402	+ return NULL;
	403	+ }
	404	+ return de;
	405	+}
	406	+
	407	+void sysfs_notify(struct kobject * k, char dir, char attr)
	408	+{
	409	+ struct dentry *de = k->dentry;
	410	+ if (de)
	411	+ dget(de);
	412	+ if (de && dir)
	413	+ de = step_down(de, dir);
	414	+ if (de && attr)
	415	+ de = step_down(de, attr);
	416	+ if (de) {
	417	+ struct sysfs_dirent * sd = de->d_fsdata;
	418	+ if (sd)
	419	+ atomic_inc(&sd->s_event);
	420	+ wake_up_interruptible(&k->poll);
	421	+ dput(de);
	422	+ }
	423	+}
	424	+EXPORT_SYMBOL_GPL(sysfs_notify);
	425	+
351	426	const struct file_operations sysfs_file_operations = {
352	427	.read = sysfs_read_file,
353	428	.write = sysfs_write_file,
354	429	.llseek = generic_file_llseek,
355	430	.open = sysfs_open_file,
356	431	.release = sysfs_release,
	432	+ .poll = sysfs_poll,
357	433	};
358	434
359	435
...	...	@@ -11,6 +11,7 @@
11	11
12	12	extern int sysfs_add_file(struct dentry , const struct attribute , int);
13	13	extern void sysfs_hash_and_remove(struct dentry * dir, const char * name);
	14	+extern struct sysfs_dirent sysfs_find(struct sysfs_dirent dir, const char * name);
14	15
15	16	extern int sysfs_create_subdir(struct kobject , const char , struct dentry **);
16	17	extern void sysfs_remove_subdir(struct dentry *);
...	...	@@ -24,6 +24,7 @@
24	24	#include <linux/rwsem.h>
25	25	#include <linux/kref.h>
26	26	#include <linux/kernel.h>
	27	+#include <linux/wait.h>
27	28	#include <asm/atomic.h>
28	29
29	30	#define KOBJ_NAME_LEN 20
...	...	@@ -56,6 +57,7 @@
56	57	struct kset * kset;
57	58	struct kobj_type * ktype;
58	59	struct dentry * dentry;
	60	+ wait_queue_head_t poll;
59	61	};
60	62
61	63	extern int kobject_set_name(struct kobject , const char , ...)
...	...	@@ -74,6 +74,7 @@
74	74	umode_t s_mode;
75	75	struct dentry * s_dentry;
76	76	struct iattr * s_iattr;
	77	+ atomic_t s_event;
77	78	};
78	79
79	80	#define SYSFS_ROOT 0x0001
...	...	@@ -117,6 +118,7 @@
117	118
118	119	int sysfs_create_group(struct kobject , const struct attribute_group );
119	120	void sysfs_remove_group(struct kobject , const struct attribute_group );
	121	+void sysfs_notify(struct kobject * k, char dir, char attr);
120	122
121	123	#else /* CONFIG_SYSFS */
122	124
...	...	@@ -183,6 +185,10 @@
183	185	static inline void sysfs_remove_group(struct kobject * k, const struct attribute_group * g)
184	186	{
185	187	;
	188	+}
	189	+
	190	+static inline void sysfs_notify(struct kobject * k, char dir, char attr)
	191	+{
186	192	}
187	193
188	194	#endif /* CONFIG_SYSFS */
...	...	@@ -128,6 +128,7 @@
128	128	{
129	129	kref_init(&kobj->kref);
130	130	INIT_LIST_HEAD(&kobj->entry);
	131	+ init_waitqueue_head(&kobj->poll);
131	132	kobj->kset = kset_get(kobj->kset);
132	133	}
133	134