[Xen-devel] Re: [PATCH ixgbe] Don't depend of skb->data for VMDq

[Steven Smith <steven.smith@xxxxxxxxxx>]

> Occasionally, vmq_alloc_skb() will give us an skb with a nonzero but
> bogus data pointer.  Dereferencing this pointer will then cause a
> panic.
That's disturbing.  The SKB needs to have a data area (because
otherwise there's nowhere to put the skb_shinfo, and hence no frags
list), so the data pointer really ought to be valid (i.e. NULL is also
bad here).

Is it possible that the entire SKB is bad?  If we're doing a VMQ
receive, we don't actually use very much of the skbuff structure apart
from the shinfo bit.  If the packet were getting kfree_skb()ed
somewhere then it might look like this.

I've applied the patch, but I think there's something fairly
fundamentally wrong here.

Steven.

Attachment: signature.asc
Description: Digital signature

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel