[bpf,v2,1/2] bpf, xdp: clean head/meta when expanding it

The device allocates an skb, it additionally allocates a prepad size
(usually equal to NET_SKB_PAD or XDP_PACKET_HEADROOM) but leaves it
uninitialized.

The bpf_xdp_adjust_head function moves skb->data forward, which allows
users to access data belonging to other programs, posing a security risk.

Reported-by: syzbot+0e6ddb1ef80986bdfe64@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/all/00000000000067f65105edbd295d@google.com/T/
Signed-off-by: Jiayuan Chen <jiayuan.chen@linux.dev>
---
 include/uapi/linux/bpf.h       | 8 +++++---
 net/core/filter.c              | 5 ++++-
 tools/include/uapi/linux/bpf.h | 6 ++++--
 3 files changed, 13 insertions(+), 6 deletions(-)

On Thu, Apr 3, 2025 at 7:32 AM Willem de Bruijn
<willemdebruijn.kernel@gmail.com> wrote:
>
> Alexei Starovoitov wrote:
> > On Sun, Mar 30, 2025 at 8:27 PM Jiayuan Chen <jiayuan.chen@linux.dev> wrote:
> > >
> > > The device allocates an skb, it additionally allocates a prepad size
> > > (usually equal to NET_SKB_PAD or XDP_PACKET_HEADROOM) but leaves it
> > > uninitialized.
> > >
> > > The bpf_xdp_adjust_head function moves skb->data forward, which allows
> > > users to access data belonging to other programs, posing a security risk.
> > >
> > > Reported-by: syzbot+0e6ddb1ef80986bdfe64@syzkaller.appspotmail.com
> > > Closes: https://lore.kernel.org/all/00000000000067f65105edbd295d@google.com/T/
> > > Signed-off-by: Jiayuan Chen <jiayuan.chen@linux.dev>
> > > ---
> > >  include/uapi/linux/bpf.h       | 8 +++++---
> > >  net/core/filter.c              | 5 ++++-
> > >  tools/include/uapi/linux/bpf.h | 6 ++++--
> > >  3 files changed, 13 insertions(+), 6 deletions(-)
> > >
> > > diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
> > > index defa5bb881f4..be01a848cbbf 100644
> > > --- a/include/uapi/linux/bpf.h
> > > +++ b/include/uapi/linux/bpf.h
> > > @@ -2760,8 +2760,9 @@ union bpf_attr {
> > >   *
> > >   * long bpf_xdp_adjust_head(struct xdp_buff *xdp_md, int delta)
> > >   *     Description
> > > - *             Adjust (move) *xdp_md*\ **->data** by *delta* bytes. Note that
> > > - *             it is possible to use a negative value for *delta*. This helper
> > > + *             Adjust (move) *xdp_md*\ **->data** by *delta* bytes. Note that
> > > + *             it is possible to use a negative value for *delta*. If *delta*
> > > + *             is negative, the new header will be memset to zero. This helper
> > >   *             can be used to prepare the packet for pushing or popping
> > >   *             headers.
> > >   *
> > > @@ -2989,7 +2990,8 @@ union bpf_attr {
> > >   * long bpf_xdp_adjust_meta(struct xdp_buff *xdp_md, int delta)
> > >   *     Description
> > >   *             Adjust the address pointed by *xdp_md*\ **->data_meta** by
> > > - *             *delta* (which can be positive or negative). Note that this
> > > + *             *delta* (which can be positive or negative). If *delta* is
> > > + *             negative, the new meta will be memset to zero. Note that this
> > >   *             operation modifies the address stored in *xdp_md*\ **->data**,
> > >   *             so the latter must be loaded only after the helper has been
> > >   *             called.
> > > diff --git a/net/core/filter.c b/net/core/filter.c
> > > index 46ae8eb7a03c..5f01d373b719 100644
> > > --- a/net/core/filter.c
> > > +++ b/net/core/filter.c
> > > @@ -3947,6 +3947,8 @@ BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
> > >         if (metalen)
> > >                 memmove(xdp->data_meta + offset,
> > >                         xdp->data_meta, metalen);
> > > +       if (offset < 0)
> > > +               memset(data, 0, -offset);
> > >         xdp->data_meta += offset;
> > >         xdp->data = data;
> > >
> > > @@ -4239,7 +4241,8 @@ BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
> > >                 return -EINVAL;
> > >         if (unlikely(xdp_metalen_invalid(metalen)))
> > >                 return -EACCES;
> > > -
> > > +       if (offset < 0)
> > > +               memset(meta, 0, -offset);
> >
> > Let's make everyone pay a performance penalty to silence
> > KMSAN warning?
> >
> > I don't think it's a good trade off.
> >
> > Soft nack.
>
> I also assumed that this was known when the feature was originally
> introduced and left as is for performance reasons.
>
> Might be good to have that explicit. And that it is deemed safe by
> virtue of XDP requiring superuser privileges anyway. Or at least I
> guess that was the thought process?

Correct. When prog extends the headroom it is suppose to write
something in there. Extending the packet just to capture
some garbage bytes from the previous packet is dumb,
but doesn't compromise the safety of the kernel.
There were proposals to ask the verifier to track that the headroom
is actually initialized by the program, but it's pointless.
Dumb prog can write garbage in there just as well.
bpf_probe_read_kernel( from_random_addr) and store into the headroom.