sysfs: release mutex when kmalloc() failed in sysfs_open_file().

sysfs: release mutex when kmalloc() failed in sysfs_open_file().

[YOSHIFUJI Hideaki / 吉藤英明]

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>

--
diff --git a/fs/sysfs/file.c b/fs/sysfs/file.c
index b502c71..1f64ce5 100644
--- a/fs/sysfs/file.c
+++ b/fs/sysfs/file.c
@@ -283,6 +283,7 @@ static int sysfs_open_file(struct inode *inode, struct file *file)
 	mutex_lock(&inode->i_mutex);
 	if (!(set = inode->i_private)) {
 		if (!(set = inode->i_private = kmalloc(sizeof(struct sysfs_buffer_collection), GFP_KERNEL))) {
+			mutex_unlock(&inode->i_mutex);
 			error = -ENOMEM;
 			goto Done;
 		} else {

-- 
YOSHIFUJI Hideaki @ USAGI Project  <yoshfuji@linux-ipv6.org>
GPG-FP  : 9022 65EB 1ECF 3AD1 0BDF  80D8 4807 F894 E062 0EEA

Re: sysfs: release mutex when kmalloc() failed in sysfs_open_file().

[Christoph Hellwig]

On Tue, Jul 10, 2007 at 01:56:46PM +0900, YOSHIFUJI Hideaki / ?$B5HF#1QL@ wrote:
> @@ -283,6 +283,7 @@ static int sysfs_open_file(struct inode *inode, struct file *file)
>  	mutex_lock(&inode->i_mutex);
>  	if (!(set = inode->i_private)) {
>  		if (!(set = inode->i_private = kmalloc(sizeof(struct sysfs_buffer_collection), GFP_KERNEL))) {
> +			mutex_unlock(&inode->i_mutex);
>  			error = -ENOMEM;
>  			goto Done;
>  		} else {

Not related to your code, but whoever wrote this code originally deserves
a big slap in the face for writing such obsfucated code.  This should be
something like:

 	mutex_lock(&inode->i_mutex);
 	set = inode->i_private;
	if (!set) {
 		set = kmalloc(sizeof(struct sysfs_buffer_collection),
				GFP_KERNEL))) {
		if (!set)
			mutex_unlock(&inode->i_mutex);
 			error = -ENOMEM;
 			goto Done;
 		}
		INIT_LIST_HEAD(&set->associates);
		inode->i_private = set;
	}
	mutex_unlock(&inode->i_mutex);

Re: sysfs: release mutex when kmalloc() failed in sysfs_open_file().

[Andrew Morton]

On Tue, 10 Jul 2007 13:56:46 +0900 (JST) YOSHIFUJI Hideaki / 吉藤英明  <yoshfuji@linux-ipv6.org> wrote:

> diff --git a/fs/sysfs/file.c b/fs/sysfs/file.c
> index b502c71..1f64ce5 100644
> --- a/fs/sysfs/file.c
> +++ b/fs/sysfs/file.c
> @@ -283,6 +283,7 @@ static int sysfs_open_file(struct inode *inode, struct file *file)
>  	mutex_lock(&inode->i_mutex);
>  	if (!(set = inode->i_private)) {
>  		if (!(set = inode->i_private = kmalloc(sizeof(struct sysfs_buffer_collection), GFP_KERNEL))) {
> +			mutex_unlock(&inode->i_mutex);
>  			error = -ENOMEM;
>  			goto Done;
>  		} else {

All this code is totally different in Greg's development tree, so this
fix really is only applicable to 2.6.22.x.

I suppose we could apply this to -stable, but given GFP_KERNEL's present behaviour
for small allocations it's pretty theoretical.

Unless you've enabled fault-injection in the page allocator, or if you get
oom-killed in that allocation.  In which case your kernel is dead.

So hm, perhaps you'd best send this to stable@kernel.org.