Re: [PATCH net] udp: fix a skb extensions leak

[Florian Westphal <fw@strlen.de>]

Xin Long <lucien.xin@gmail.com> wrote:
> On Mon, Mar 30, 2020 at 4:29 PM Steffen Klassert
> <steffen.klassert@secunet.com> wrote:
> >
> > On Thu, Mar 26, 2020 at 05:06:25PM +0800, Xin Long wrote:
> > > On udp rx path udp_rcv_segment() may do segment where the frag skbs
> > > will get the header copied from the head skb in skb_segment_list()
> > > by calling __copy_skb_header(), which could overwrite the frag skbs'
> > > extensions by __skb_ext_copy() and cause a leak.
> > >
> > > This issue was found after loading esp_offload where a sec path ext
> > > is set in the skb.
> > >
> > > On udp tx gso path, it works well as the frag skbs' extensions are
> > > not set. So this issue should be fixed on udp's rx path only and
> > > release the frag skbs' extensions before going to do segment.
> >
> > Are you sure that this affects only the RX path? What if such
> > a packet is forwarded? Also, I think TCP has the same problem.
> You're right, just confirm it exists on the forwarded path.
> __copy_skb_header() is also called by skb_segment(), but
> I don't have tests to reproduce it on other protocols like TCP.

skb_segment() is fine, either nskb is a new allocation or a clone
with head state already discarded.

> > If a skb in the fraglist has a secpath, it is still valid.
> > So maybe instead of dropping it here and assign the one
> > from the head skb, we could just keep the secpath. But
> > I don't know about other extensions. I've CCed Florian,
> > he might know a bit more about other extensions. Also,
> > it might be good to check if the extensions of the GRO
> > packets are all the same before merging.
> >
> Not sure if we can improve __copy_skb_header() or add
> a new function to copy these members ONLY when nskb's
> are not set.

I don't see how.