Re: 答复: [PATCH bpf-next] bpf: Don't redirect packets with invalid pkt_len

[sdf@google.com]

On 07/13, shaozhengchao wrote:


> -----邮件原件-----
> 发件人: Daniel Borkmann [mailto:daniel@iogearbox.net]
> 发送时间: 2022年7月13日 4:12
> 收件人: sdf@google.com; shaozhengchao <shaozhengchao@huawei.com>
> 抄送: bpf@vger.kernel.org; netdev@vger.kernel.org;  
> linux-kernel@vger.kernel.org; davem@davemloft.net; edumazet@google.com;  
> kuba@kernel.org; pabeni@redhat.com; hawk@kernel.org; ast@kernel.org;  
> andrii@kernel.org; martin.lau@linux.dev; song@kernel.org; yhs@fb.com;  
> john.fastabend@gmail.com; kpsingh@kernel.org; weiyongjun (A)  
> <weiyongjun1@huawei.com>; yuehaibing <yuehaibing@huawei.com>
> 主题: Re: [PATCH bpf-next] bpf: Don't redirect packets with invalid  
> pkt_len

> On 7/12/22 6:58 PM, sdf@google.com wrote:
> > On 07/12, Zhengchao Shao wrote:
> >> Syzbot found an issue [1]: fq_codel_drop() try to drop a flow whitout
> >> any skbs, that is, the flow->head is null.
> >> The root cause, as the [2] says, is because that
> >> bpf_prog_test_run_skb() run a bpf prog which redirects empty skbs.
> >> So we should determine whether the length of the packet modified by
> >> bpf prog or others like bpf_prog_test is valid before forwarding it  
> directly.
> >
> >> LINK: [1]
> >> https://syzkaller.appspot.com/bug?id=0b84da80c2917757915afa89f7738a9d
> >> 16ec96c5
> >> LINK: [2] https://www.spinics.net/lists/netdev/msg777503.html
> >
> >> Reported-by: syzbot+7a12909485b94426aceb@syzkaller.appspotmail.com
> >> Signed-off-by: Zhengchao Shao <shaozhengchao@huawei.com>
> >> ---
> >>   net/core/filter.c | 9 ++++++++-
> >>   1 file changed, 8 insertions(+), 1 deletion(-)
> >
> >> diff --git a/net/core/filter.c b/net/core/filter.c index
> >> 4ef77ec5255e..27801b314960 100644
> >> --- a/net/core/filter.c
> >> +++ b/net/core/filter.c
> >> @@ -2122,6 +2122,11 @@ static int __bpf_redirect_no_mac(struct
> >> sk_buff *skb, struct net_device *dev,
> >>   {
> >>       unsigned int mlen = skb_network_offset(skb);
> >
> >> +    if (unlikely(skb->len == 0)) {
> >> +        kfree_skb(skb);
> >> +        return -EINVAL;
> >> +    }
> >> +
> >>       if (mlen) {
> >>           __skb_pull(skb, mlen);
> >
> >> @@ -2143,7 +2148,9 @@ static int __bpf_redirect_common(struct sk_buff
> >> *skb, struct net_device *dev,
> >>                    u32 flags)
> >>   {
> >>       /* Verify that a link layer header is carried */
> >> -    if (unlikely(skb->mac_header >= skb->network_header)) {
> >> +    if (unlikely(skb->mac_header >= skb->network_header) ||
> >> +        (min_t(u32, skb_mac_header_len(skb), skb->len) <
> >> +         (u32)dev->min_header_len)) {
> >
> > Why check skb->len != 0 above but skb->len < dev->min_header_len here?
> > I guess it doesn't make sense in __bpf_redirect_no_mac because we know
> > that mac is empty, but why do we care in __bpf_redirect_common?
> > Why not put this check in the common __bpf_redirect?
> >
> > Also, it's still not clear to me whether we should bake it into the
> > core stack vs having some special checks from test_prog_run only. I'm
> > assuming the issue is that we can construct illegal skbs with that
> > test_prog_run interface, so maybe start by fixing that?

> Agree, ideally we can prevent it right at the source rather than adding  
> more tests into the fast-path.

> > Did you have a chance to look at the reproducer more closely? What
> > exactly is it doing?
> >
> >>           kfree_skb(skb);
> >>           return -ERANGE;
> >>       }
> >> --
> >> 2.17.1

> >


> Hi Daniel and sdf:
> 	Thank you for your reply. I read the poc code carefully, and I think the  
> current call stack is like:
> sys_bpf(BPF_PROG_TEST_RUN, &attr, sizeof(attr)) ->  
> bpf_prog_test_run->bpf_prog_test_run_skb.

> In function bpf_prog_test_run_skb, procedure will use build_skb to  
> generate a new skb. Poc code pass
> a 14Byte packet for direct. First ,skb->len = 14, but after trans eth  
> type, the len = 0; but is_l2 is false,
> so len=0 when run bpf_test_run. Is it possible to add check in  
> convert___skb_to_skb? When skb->len=0,
> we drop the packet.

Not sure it belongs in convert___skb_to_skb, but checking somewhere before
convert___skb_to_skb seems like a good way to go?

> But, if some other paths call bpf redirect with skb->len=0, this is not  
> effective, such as some driver call redirect fuction.
> I don't know if I'm thinking right.

I think the consensus so far that it's only bpf_prog_test_run that
generates these types of packets, so let's start with fixing that.