netfilter: conntrack: dccp: copy entire header to stack buffer, not just basic one
authorFlorian Westphal <fw@strlen.de>
Wed, 21 Jun 2023 15:56:53 +0000 (17:56 +0200)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 19 Jul 2023 14:21:13 +0000 (16:21 +0200)
commit5c618daa5038712c4a4ef8923905a2ea1b8836a1
tree104ac13ff8f894fe4aae1645d5c70c3b0713d503
parentcfcb9f0a499dafadcbfe43b1c587f20e4e953b19
netfilter: conntrack: dccp: copy entire header to stack buffer, not just basic one

[ Upstream commit ff0a3a7d52ff7282dbd183e7fc29a1fe386b0c30 ]

Eric Dumazet says:
  nf_conntrack_dccp_packet() has an unique:

  dh = skb_header_pointer(skb, dataoff, sizeof(_dh), &_dh);

  And nothing more is 'pulled' from the packet, depending on the content.
  dh->dccph_doff, and/or dh->dccph_x ...)
  So dccp_ack_seq() is happily reading stuff past the _dh buffer.

BUG: KASAN: stack-out-of-bounds in nf_conntrack_dccp_packet+0x1134/0x11c0
Read of size 4 at addr ffff000128f66e0c by task syz-executor.2/29371
[..]

Fix this by increasing the stack buffer to also include room for
the extra sequence numbers and all the known dccp packet type headers,
then pull again after the initial validation of the basic header.

While at it, mark packets invalid that lack 48bit sequence bit but
where RFC says the type MUST use them.

Compile tested only.

v2: first skb_header_pointer() now needs to adjust the size to
    only pull the generic header. (Eric)

Heads-up: I intend to remove dccp conntrack support later this year.

Fixes: 2bc780499aa3 ("[NETFILTER]: nf_conntrack: add DCCP protocol support")
Reported-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
net/netfilter/nf_conntrack_proto_dccp.c