Re: [PATCH v2] virtio_blk: Fix a slient kernel panic

From: Minfei Huang
Date: Mon Jul 18 2016 - 12:18:51 EST


On 07/18/16 at 05:21P, Cornelia Huck wrote:
> On Mon, 18 Jul 2016 22:01:29 +0800
> Minfei Huang <mnfhuang@xxxxxxxxx> wrote:
> > diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
> > index 42758b5..d920512 100644
> > --- a/drivers/block/virtio_blk.c
> > +++ b/drivers/block/virtio_blk.c
> > @@ -381,9 +381,9 @@ static int init_vq(struct virtio_blk *vblk)
> > {
> > int err = 0;
> > int i;
> > - vq_callback_t **callbacks;
> > - const char **names;
> > - struct virtqueue **vqs;
> > + vq_callback_t **callbacks = NULL;
> > + const char **names = NULL;
> > + struct virtqueue **vqs = NULL;
>
> If you init the variables to NULL anyway...

Hi, Cornelia.

Thanks for reviewing this patch.

Seems there is no need to init these variables to NULL. I will remove
them laster.

>
> > unsigned short num_vqs;
> > struct virtio_device *vdev = vblk->vdev;
> >
> > @@ -394,22 +394,16 @@ static int init_vq(struct virtio_blk *vblk)
> > num_vqs = 1;
> >
>
> ...just do
>
> err = -ENOMEM;
>
> here and...
>
> > vblk->vqs = kmalloc(sizeof(*vblk->vqs) * num_vqs, GFP_KERNEL);
> > - if (!vblk->vqs) {
> > - err = -ENOMEM;
> > - goto out;
> > - }
> > + if (!vblk->vqs)
> > + return -ENOMEM;
> >
> > names = kmalloc(sizeof(*names) * num_vqs, GFP_KERNEL);
> > - if (!names)
> > - goto err_names;
> > -
> > callbacks = kmalloc(sizeof(*callbacks) * num_vqs, GFP_KERNEL);
> > - if (!callbacks)
> > - goto err_callbacks;
> > -
> > vqs = kmalloc(sizeof(*vqs) * num_vqs, GFP_KERNEL);
> > - if (!vqs)
> > - goto err_vqs;
> > + if (!names || !callbacks || !vqs) {
> > + err = -ENOMEM;
> > + goto out;
> > + }
>
> ...you could use the
>
> foo = kmalloc(...);
> if (!foo)
> goto out;
>
> sequence in any case. This avoids trying again and again if e.g. the
> names allocation already failed.

For this implementation, I have referred others which calls
vdev->config->find_vqs as well. Yes, this continues trying to allocate
memory, although memory allocation failed before.

>
> Alternatively, you should be fine if you don't init the variables to
> NULL: The code is now either taking an early exit or setting all of the
> variables anyway.
>

It's a big change if we refactor the helper ->find_vqs, since other
devices also call it.

Thanks
Minfei