On Mon, 2011-09-26 at 22:55 +0300, Michael S. Tsirkin wrote:
> On Mon, Sep 26, 2011 at 10:37:22PM +0300, Sasha Levin wrote:
> > On Mon, 2011-09-26 at 21:44 +0300, Michael S. Tsirkin wrote:
> > > On Mon, Sep 26, 2011 at 08:41:08PM +0300, Sasha Levin wrote:
> > > > This patch verifies that the length of a buffer stored in a linked list
> > > > of pages is small enough to fit into a skb.
> > > >
> > > > If the size is larger than a max size of a skb, it means that we shouldn't
> > > > go ahead building skbs anyway since we won't be able to send the buffer as
> > > > the user requested.
> > > >
> > > > Cc: Rusty Russell <rusty@xxxxxxxxxxxxxxx>
> > > > Cc: "Michael S. Tsirkin" <mst@xxxxxxxxxx>
> > > > Cc: virtualization@xxxxxxxxxxxxxxxxxxxxxxxxxx
> > > > Cc: netdev@xxxxxxxxxxxxxxx
> > > > Cc: kvm@xxxxxxxxxxxxxxx
> > > > Signed-off-by: Sasha Levin <levinsasha928@xxxxxxxxx>
> > >
> > > Interesting. This is a theoretical issue, correct?
> > > Not a crash you actually see.
> >
> > Actually it was an actual crash caused when our virtio-net driver in kvm
> > tools did funny things and passed '(u32)-1' length as a buffer length to
> > the guest kernel.
> >
> > > This crash would mean device is giving us packets
> > > that are way too large. Avoiding crashes even in the face of
> > > a misbehaved device is a good idea, but should
> > > we print a diagnostic to a system log?
> > > Maybe rate-limited or print once to avoid filling
> > > up the disk. Other places in driver print with pr_debug
> > > I'm not sure that's right but better than nothing.
> >
> > Yup, I'll add some debug info.
> >
> > > > ---
> > > > drivers/net/virtio_net.c | 3 +++
> > > > 1 files changed, 3 insertions(+), 0 deletions(-)
> > > >
> > > > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> > > > index 0c7321c..64e0717 100644
> > > > --- a/drivers/net/virtio_net.c
> > > > +++ b/drivers/net/virtio_net.c
> > > > @@ -165,6 +165,9 @@ static struct sk_buff *page_to_skb(struct virtnet_info *vi,
> > > > unsigned int copy, hdr_len, offset;
> > > > char *p;
> > > >
> > > > + if (len > MAX_SKB_FRAGS * PAGE_SIZE)
> > >
> > > unlikely()?
> > >
> > > Also, this seems too aggressive: at this point len includes the header
> > > and the linear part. The right place for this
> > > test is probably where we fill in the frags, just before
> > > while (len)
> > >
> > > The whole can only happen when mergeable buffers
> > > are disabled, right?
> >
> > >From what I understand it can happen whenever you're going to build a
> > skb longer than PAGE_SIZE.
>
> Hmm how exactly? With mergeable buffers this only gets
> the length of the 1st chunk which is up to 4K unless the driver
> is buggy ...
What about the case where TSO or ECN features are set? The code flow
suggests that mergeable would get ignored in that case:
/* If we can receive ANY GSO packets, we must allocate large ones. */
if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
vi->big_packets = true;
[...]
if (!vi->mergeable_rx_bufs && !vi->big_packets) {
skb = buf;
len -= sizeof(struct virtio_net_hdr);
skb_trim(skb, len);
} else {
page = buf;
skb = page_to_skb(vi, page, len);
...
I haven't actually tested it with mergeable buffers enabled, but could
do it later today.
--
Sasha.
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
