bpf, lwt: Fix crash when using bpf_skb_set_tunnel_key() from bpf_xmit lwt hook
xmit_check_hhlen() observes the dst for getting the device hard header
length to make sure a modified packet can fit. When a helper which changes
the dst - such as bpf_skb_set_tunnel_key() - is called as part of the
xmit program the accessed dst is no longer valid.
This leads to the following splat:
BUG: kernel NULL pointer dereference, address:
00000000000000de
#PF: supervisor read access in kernel mode
#PF: error_code(0x0000) - not-present page
PGD 0 P4D 0
Oops: 0000 [#1] PREEMPT SMP PTI
CPU: 0 PID: 798 Comm: ping Not tainted 5.18.0-rc2+ #103
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.14.0-2 04/01/2014
RIP: 0010:bpf_xmit+0xfb/0x17f
Code: c6 c0 4d cd 8e 48 c7 c7 7d 33 f0 8e e8 42 09 fb ff 48 8b 45 58 48 8b 95 c8 00 00 00 48 2b 95 c0 00 00 00 48 83 e0 fe 48 8b 00 <0f> b7 80 de 00 00 00 39 c2 73 22 29 d0 b9 20 0a 00 00 31 d2 48 89
RSP: 0018:
ffffb148c0bc7b98 EFLAGS:
00010282
RAX:
0000000000000000 RBX:
0000000000240008 RCX:
0000000000000000
RDX:
0000000000000010 RSI:
00000000ffffffea RDI:
00000000ffffffff
RBP:
ffff922a828a4e00 R08:
ffffffff8f1350e8 R09:
00000000ffffdfff
R10:
ffffffff8f055100 R11:
ffffffff8f105100 R12:
0000000000000000
R13:
ffff922a828a4e00 R14:
0000000000000040 R15:
0000000000000000
FS:
00007f414e8f0080(0000) GS:
ffff922afdc00000(0000) knlGS:
0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0:
0000000080050033
CR2:
00000000000000de CR3:
0000000002d80006 CR4:
0000000000370ef0
Call Trace:
<TASK>
lwtunnel_xmit.cold+0x71/0xc8
ip_finish_output2+0x279/0x520
? __ip_finish_output.part.0+0x21/0x130
Fix by fetching the device hard header length before running the BPF code.
Fixes: 3a0af8fd61f9 ("bpf: BPF for lightweight tunnel infrastructure")
Signed-off-by: Eyal Birger <eyal.birger@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20220420165219.1755407-1-eyal.birger@gmail.com