2012-11-12 13:50:42

by Wei Yongjun

Subject: [PATCH -next] virtio: balloon: fix missing unlock on error in fill_balloon()

From: Wei Yongjun <[email protected]>

Add the missing unlock before return from function fill_balloon()
in the error handling case.

Introduced by 9864a8(virtio_balloon: introduce migration primitives
to balloon pages)

dpatch engine is used to auto generate this patch.
(https://github.com/weiyj/dpatch)

Signed-off-by: Wei Yongjun <[email protected]>
---
drivers/virtio/virtio_balloon.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/virtio/virtio_balloon.c b/drivers/virtio/virtio_balloon.c
index f70151b..72e8dcb 100644
--- a/drivers/virtio/virtio_balloon.c
+++ b/drivers/virtio/virtio_balloon.c
@@ -152,8 +152,10 @@ static void fill_balloon(struct virtio_balloon *vb, size_t num)
}

/* Didn't get any? Oh well. */
- if (vb->num_pfns == 0)
+ if (vb->num_pfns == 0) {
+ mutex_unlock(&vb->balloon_lock);
return;
+ }

tell_host(vb, vb->inflate_vq);
mutex_unlock(&vb->balloon_lock);

2012-11-12 14:17:11

by Rafael Aquini

Subject: Re: [PATCH -next] virtio: balloon: fix missing unlock on error in fill_balloon()

On Mon, Nov 12, 2012 at 09:50:40PM +0800, Wei Yongjun wrote:
> From: Wei Yongjun <[email protected]>
>
> Add the missing unlock before return from function fill_balloon()
> in the error handling case.
>
> Introduced by 9864a8(virtio_balloon: introduce migration primitives
> to balloon pages)
>
> dpatch engine is used to auto generate this patch.
> (https://github.com/weiyj/dpatch)
>
> Signed-off-by: Wei Yongjun <[email protected]>
> ---
> drivers/virtio/virtio_balloon.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/virtio/virtio_balloon.c b/drivers/virtio/virtio_balloon.c
> index f70151b..72e8dcb 100644
> --- a/drivers/virtio/virtio_balloon.c
> +++ b/drivers/virtio/virtio_balloon.c
> @@ -152,8 +152,10 @@ static void fill_balloon(struct virtio_balloon *vb, size_t num)
> }
>
> /* Didn't get any? Oh well. */
> - if (vb->num_pfns == 0)
> + if (vb->num_pfns == 0) {
> + mutex_unlock(&vb->balloon_lock);
> return;
> + }
>
> tell_host(vb, vb->inflate_vq);
> mutex_unlock(&vb->balloon_lock);
>

I missed that one when rewriting v10 to v11. :(

Good catch, Wei! thanks!

Andrew, we need this pick for the virtio_balloon v12 patch, as well.

-- Rafael

2012-11-12 23:34:56

by Andrew Morton

Subject: Re: [PATCH -next] virtio: balloon: fix missing unlock on error in fill_balloon()

On Mon, 12 Nov 2012 21:50:40 +0800
Wei Yongjun <[email protected]> wrote:

> From: Wei Yongjun <[email protected]>
>
> Add the missing unlock before return from function fill_balloon()
> in the error handling case.
>
> Introduced by 9864a8(virtio_balloon: introduce migration primitives
> to balloon pages)
>
> dpatch engine is used to auto generate this patch.
> (https://github.com/weiyj/dpatch)
>
> Signed-off-by: Wei Yongjun <[email protected]>
> ---
> drivers/virtio/virtio_balloon.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/virtio/virtio_balloon.c b/drivers/virtio/virtio_balloon.c
> index f70151b..72e8dcb 100644
> --- a/drivers/virtio/virtio_balloon.c
> +++ b/drivers/virtio/virtio_balloon.c
> @@ -152,8 +152,10 @@ static void fill_balloon(struct virtio_balloon *vb, size_t num)
> }
>
> /* Didn't get any? Oh well. */
> - if (vb->num_pfns == 0)
> + if (vb->num_pfns == 0) {
> + mutex_unlock(&vb->balloon_lock);
> return;
> + }
>
> tell_host(vb, vb->inflate_vq);
> mutex_unlock(&vb->balloon_lock);

Well. Why did this bug occur in the first place? Because
fill_balloon() has multiple return points and one of those was
overlooked when adding new locking.

This (and the addition of memory leaks) is quite common. It is part of
the reason why so much kernel code uses goto's to avoid multiple return
points. A maintainability thing.

And we can fix this shortcoming in fill_balloon() without even needing a
goto:

--- a/drivers/virtio/virtio_balloon.c~virtio_balloon-introduce-migration-primitives-to-balloon-pages-fix-fix
+++ a/drivers/virtio/virtio_balloon.c
@@ -151,13 +151,9 @@ static void fill_balloon(struct virtio_b
totalram_pages--;
}

- /* Didn't get any? Oh well. */
- if (vb->num_pfns == 0) {
- mutex_unlock(&vb->balloon_lock);
- return;
- }
-
- tell_host(vb, vb->inflate_vq);
+ /* Did we get any? */
+ if (vb->num_pfns != 0)
+ tell_host(vb, vb->inflate_vq);
mutex_unlock(&vb->balloon_lock);
}

_

2012-11-13 12:48:08

by Rafael Aquini

Subject: Re: [PATCH -next] virtio: balloon: fix missing unlock on error in fill_balloon()

On Mon, Nov 12, 2012 at 03:34:53PM -0800, Andrew Morton wrote:
> On Mon, 12 Nov 2012 21:50:40 +0800
> Wei Yongjun <[email protected]> wrote:
>
> > From: Wei Yongjun <[email protected]>
> >
> > Add the missing unlock before return from function fill_balloon()
> > in the error handling case.
> >
> > Introduced by 9864a8(virtio_balloon: introduce migration primitives
> > to balloon pages)
> >
> > dpatch engine is used to auto generate this patch.
> > (https://github.com/weiyj/dpatch)
> >
> > Signed-off-by: Wei Yongjun <[email protected]>
> > ---
> > drivers/virtio/virtio_balloon.c | 4 +++-
> > 1 file changed, 3 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/virtio/virtio_balloon.c b/drivers/virtio/virtio_balloon.c
> > index f70151b..72e8dcb 100644
> > --- a/drivers/virtio/virtio_balloon.c
> > +++ b/drivers/virtio/virtio_balloon.c
> > @@ -152,8 +152,10 @@ static void fill_balloon(struct virtio_balloon *vb, size_t num)
> > }
> >
> > /* Didn't get any? Oh well. */
> > - if (vb->num_pfns == 0)
> > + if (vb->num_pfns == 0) {
> > + mutex_unlock(&vb->balloon_lock);
> > return;
> > + }
> >
> > tell_host(vb, vb->inflate_vq);
> > mutex_unlock(&vb->balloon_lock);
>
> Well. Why did this bug occur in the first place? Because
> fill_balloon() has multiple return points and one of those was
> overlooked when adding new locking.
>
> This (and the addition of memory leaks) is quite common. It is part of
> the reason why so much kernel code uses goto's to avoid multiple return
> points. A maintainability thing.
>
> And we can fix this shortcoming in fill_balloon() without even needing a
> goto:
>
> --- a/drivers/virtio/virtio_balloon.c~virtio_balloon-introduce-migration-primitives-to-balloon-pages-fix-fix
> +++ a/drivers/virtio/virtio_balloon.c
> @@ -151,13 +151,9 @@ static void fill_balloon(struct virtio_b
> totalram_pages--;
> }
>
> - /* Didn't get any? Oh well. */
> - if (vb->num_pfns == 0) {
> - mutex_unlock(&vb->balloon_lock);
> - return;
> - }
> -
> - tell_host(vb, vb->inflate_vq);
> + /* Did we get any? */
> + if (vb->num_pfns != 0)
> + tell_host(vb, vb->inflate_vq);
> mutex_unlock(&vb->balloon_lock);
> }
>

Your fix is far better, as it simplifies the code without loosing that test --
which only evaluates as true, if we fail to allocate a page right after getting
into the loop -- /me thinks driver folks place it there to not bother
tell_host() and the virt_queues if no pageframe is present at that pfns array.