2007-12-12 02:08:44

by Daniel Walker

[permalink] [raw]
Subject: [PATCH -mm] media: video: usbvision: add mutex_unlock() to error paths

There are a few error paths which don't unlock the usbvision->lock.

So I've added mutex_unlock() calls to fix those paths.

Signed-off-by: Daniel Walker <[email protected]>

---
drivers/media/video/usbvision/usbvision-video.c | 2 ++
1 file changed, 2 insertions(+)

Index: linux-2.6.23/drivers/media/video/usbvision/usbvision-video.c
===================================================================
--- linux-2.6.23.orig/drivers/media/video/usbvision/usbvision-video.c
+++ linux-2.6.23/drivers/media/video/usbvision/usbvision-video.c
@@ -1290,6 +1290,7 @@ static int usbvision_radio_open(struct i
errCode = usbvision_set_alternate(usbvision);
if (errCode < 0) {
usbvision->last_error = errCode;
+ mutex_unlock(&usbvision->lock);
return -EBUSY;
}

@@ -1806,6 +1807,7 @@ static int __devinit usbvision_probe(str
usbvision->num_alt,GFP_KERNEL);
if (usbvision->alt_max_pkt_size == NULL) {
err("usbvision: out of memory!\n");
+ mutex_unlock(&usbvision->lock);
return -ENOMEM;
}

--

--


2007-12-12 04:54:21

by Andrew Morton

[permalink] [raw]
Subject: Re: [PATCH -mm] media: video: usbvision: add mutex_unlock() to error paths

On Tue, 11 Dec 2007 17:56:50 -0800 Daniel Walker <[email protected]> wrote:

> There are a few error paths which don't unlock the usbvision->lock.
>
> So I've added mutex_unlock() calls to fix those paths.
>
> Signed-off-by: Daniel Walker <[email protected]>
>
> ---
> drivers/media/video/usbvision/usbvision-video.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> Index: linux-2.6.23/drivers/media/video/usbvision/usbvision-video.c
> ===================================================================
> --- linux-2.6.23.orig/drivers/media/video/usbvision/usbvision-video.c
> +++ linux-2.6.23/drivers/media/video/usbvision/usbvision-video.c
> @@ -1290,6 +1290,7 @@ static int usbvision_radio_open(struct i
> errCode = usbvision_set_alternate(usbvision);
> if (errCode < 0) {
> usbvision->last_error = errCode;
> + mutex_unlock(&usbvision->lock);
> return -EBUSY;
> }
>
> @@ -1806,6 +1807,7 @@ static int __devinit usbvision_probe(str
> usbvision->num_alt,GFP_KERNEL);
> if (usbvision->alt_max_pkt_size == NULL) {
> err("usbvision: out of memory!\n");
> + mutex_unlock(&usbvision->lock);
> return -ENOMEM;
> }
>

Well yes. But the bug which you've found is a *direct* consequence of a
coding mistake in those functions: they have multiple deeply-nested `return'
statements. This is a common cause of locking errors and resource leaks and
is why we prefer the `goto place-which-unwinds' approach.

We can easily fix that in one case, at least.

One does wonder whether usbvision_radio_open() should be returning the
usbvision_set_alternate() result to the caller here, rather than
overwriting it with -EBUSY.


--- a/drivers/media/video/usbvision/usbvision-video.c~media-video-usbvision-add-mutex_unlock-to-error-paths-fix
+++ a/drivers/media/video/usbvision/usbvision-video.c
@@ -1290,8 +1290,8 @@ static int usbvision_radio_open(struct i
errCode = usbvision_set_alternate(usbvision);
if (errCode < 0) {
usbvision->last_error = errCode;
- mutex_unlock(&usbvision->lock);
- return -EBUSY;
+ errCode = -EBUSY;
+ goto out;
}

// If so far no errors then we shall start the radio
@@ -1308,6 +1308,7 @@ static int usbvision_radio_open(struct i
usbvision->initialized = 0;
}
}
+out:
mutex_unlock(&usbvision->lock);
return errCode;
}
_