On 12/02, Martin KaFai Lau wrote:
> On 12/2/24 8:15 AM, Stanislav Fomichev wrote:
> > On 12/02, Shigeru Yoshida wrote:
> > > KMSAN reported a use-after-free issue in eth_skb_pkt_type()[1]. The
> > > cause of the issue was that eth_skb_pkt_type() accessed skb's data
> > > that didn't contain an Ethernet header. This occurs when
> > > bpf_prog_test_run_xdp() passes an invalid value as the user_data
> > > argument to bpf_test_init().
> > >
> > > Fix this by returning an error when user_data is less than ETH_HLEN in
> > > bpf_test_init().
> > >
> > > [1]
> > > BUG: KMSAN: use-after-free in eth_skb_pkt_type include/linux/etherdevice.h:627 [inline]
> > > BUG: KMSAN: use-after-free in eth_type_trans+0x4ee/0x980 net/ethernet/eth.c:165
> > > eth_skb_pkt_type include/linux/etherdevice.h:627 [inline]
> > > eth_type_trans+0x4ee/0x980 net/ethernet/eth.c:165
> > > __xdp_build_skb_from_frame+0x5a8/0xa50 net/core/xdp.c:635
> > > xdp_recv_frames net/bpf/test_run.c:272 [inline]
> > > xdp_test_run_batch net/bpf/test_run.c:361 [inline]
> > > bpf_test_run_xdp_live+0x2954/0x3330 net/bpf/test_run.c:390
> > > bpf_prog_test_run_xdp+0x148e/0x1b10 net/bpf/test_run.c:1318
> > > bpf_prog_test_run+0x5b7/0xa30 kernel/bpf/syscall.c:4371
> > > __sys_bpf+0x6a6/0xe20 kernel/bpf/syscall.c:5777
> > > __do_sys_bpf kernel/bpf/syscall.c:5866 [inline]
> > > __se_sys_bpf kernel/bpf/syscall.c:5864 [inline]
> > > __x64_sys_bpf+0xa4/0xf0 kernel/bpf/syscall.c:5864
> > > x64_sys_call+0x2ea0/0x3d90 arch/x86/include/generated/asm/syscalls_64.h:322
> > > do_syscall_x64 arch/x86/entry/common.c:52 [inline]
> > > do_syscall_64+0xd9/0x1d0 arch/x86/entry/common.c:83
> > > entry_SYSCALL_64_after_hwframe+0x77/0x7f
> > >
> > > Uninit was created at:
> > > free_pages_prepare mm/page_alloc.c:1056 [inline]
> > > free_unref_page+0x156/0x1320 mm/page_alloc.c:2657
> > > __free_pages+0xa3/0x1b0 mm/page_alloc.c:4838
> > > bpf_ringbuf_free kernel/bpf/ringbuf.c:226 [inline]
> > > ringbuf_map_free+0xff/0x1e0 kernel/bpf/ringbuf.c:235
> > > bpf_map_free kernel/bpf/syscall.c:838 [inline]
> > > bpf_map_free_deferred+0x17c/0x310 kernel/bpf/syscall.c:862
> > > process_one_work kernel/workqueue.c:3229 [inline]
> > > process_scheduled_works+0xa2b/0x1b60 kernel/workqueue.c:3310
> > > worker_thread+0xedf/0x1550 kernel/workqueue.c:3391
> > > kthread+0x535/0x6b0 kernel/kthread.c:389
> > > ret_from_fork+0x6e/0x90 arch/x86/kernel/process.c:147
> > > ret_from_fork_asm+0x1a/0x30 arch/x86/entry/entry_64.S:244
> > >
> > > CPU: 1 UID: 0 PID: 17276 Comm: syz.1.16450 Not tainted 6.12.0-05490-g9bb88c659673 #8
> > > Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.3-3.fc41 04/01/2014
> > >
> > > Fixes: be3d72a2896c ("bpf: move user_size out of bpf_test_init")
> > > Reported-by: syzkaller <syzkaller@xxxxxxxxxxxxxxxx>
> > > Signed-off-by: Shigeru Yoshida <syoshida@xxxxxxxxxx>
> > > ---
> > > net/bpf/test_run.c | 2 +-
> > > 1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > diff --git a/net/bpf/test_run.c b/net/bpf/test_run.c
> > > index 501ec4249fed..756250aa890f 100644
> > > --- a/net/bpf/test_run.c
> > > +++ b/net/bpf/test_run.c
> > > @@ -663,7 +663,7 @@ static void *bpf_test_init(const union bpf_attr *kattr, u32 user_size,
> > > if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom)
> > > return ERR_PTR(-EINVAL);
> > > - if (user_size > size)
> > > + if (user_size < ETH_HLEN || user_size > size)
> > > return ERR_PTR(-EMSGSIZE);
> > > size = SKB_DATA_ALIGN(size);
> > > --
> > > 2.47.0
> > >
> >
> > I wonder whether 'size < ETH_HLEN' above is needed after your patch.
> > Feels like 'user_size < ETH_HLEN' supersedes it.
>
> May be fixing it by replacing the existing "size" check with "user_size"
> check? Seems more intuitive that checking is needed on the "user_"size
> instead of the "size". The "if (user_size > size)" check looks useless also.
> Something like this?
>
> - if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom)
> + if (user_size < ETH_HLEN || user_size > PAGE_SIZE - headroom - tailroom)
> return ERR_PTR(-EINVAL);
>
> - if (user_size > size)
> - return ERR_PTR(-EMSGSIZE);
> -
>
Agreed, that should do it. IIUC, 'user_size > size' only makes sense
for the bpf_prog_test_run_xdp case and the caller handles this case
anyway (size > max_data_sz).
