| Message ID | 1414978173-6948-1-git-send-email-chenweilong@huawei.com |
|---|---|
| State | Changes Requested, archived |
| Delegated to: | David Miller |
| Headers | show |
On Mon, 2014-11-03 at 09:29 +0800, Chen Weilong wrote: > From: Weilong Chen <chenweilong@huawei.com> > > We got a problem like this: > There was a wrong ack packet coming during TCP handshake. The socket's state > was TCP_SYN_RECV, its rcv_mss was not initialize yet. So > tcp_send_dupack -> tcp_enter_quickack_mode got a divide 0 error. > This patch add a state check before tcp_enter_quickack_mode. > > Signed-off-by: Weilong Chen <chenweilong@huawei.com> > --- > net/ipv4/tcp_input.c | 3 ++- > 1 file changed, 2 insertions(+), 1 deletion(-) > > diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c > index 4e4617e..9eb56dc 100644 > --- a/net/ipv4/tcp_input.c > +++ b/net/ipv4/tcp_input.c > @@ -3986,7 +3986,8 @@ static void tcp_send_dupack(struct sock *sk, const struct sk_buff *skb) > if (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq && > before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) { > NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKLOST); > - tcp_enter_quickack_mode(sk); > + if (sk->sk_state != TCP_SYN_RECV) > + tcp_enter_quickack_mode(sk); > > if (tcp_is_sack(tp) && sysctl_tcp_dsack) { > u32 end_seq = TCP_SKB_CB(skb)->end_seq; Sorry I do not think this is the right fix. We have to not simply avoid the divide, but fix this issue by understanding the missing steps. -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On 2014/11/3 11:42, Eric Dumazet wrote: > On Mon, 2014-11-03 at 09:29 +0800, Chen Weilong wrote: >> From: Weilong Chen <chenweilong@huawei.com> >> >> We got a problem like this: > >> There was a wrong ack packet coming during TCP handshake. The socket's state >> was TCP_SYN_RECV, its rcv_mss was not initialize yet. So >> tcp_send_dupack -> tcp_enter_quickack_mode got a divide 0 error. >> This patch add a state check before tcp_enter_quickack_mode. >> >> Signed-off-by: Weilong Chen <chenweilong@huawei.com> >> --- >> net/ipv4/tcp_input.c | 3 ++- >> 1 file changed, 2 insertions(+), 1 deletion(-) >> >> diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c >> index 4e4617e..9eb56dc 100644 >> --- a/net/ipv4/tcp_input.c >> +++ b/net/ipv4/tcp_input.c >> @@ -3986,7 +3986,8 @@ static void tcp_send_dupack(struct sock *sk, const struct sk_buff *skb) >> if (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq && >> before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) { >> NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKLOST); >> - tcp_enter_quickack_mode(sk); >> + if (sk->sk_state != TCP_SYN_RECV) >> + tcp_enter_quickack_mode(sk); >> >> if (tcp_is_sack(tp) && sysctl_tcp_dsack) { >> u32 end_seq = TCP_SKB_CB(skb)->end_seq; > > > Sorry I do not think this is the right fix. > > We have to not simply avoid the divide, but fix this issue by > understanding the missing steps. > Hi Eric, I check the code and find that: 1.In function "tcp_rcv_state_process", the "tcp_initialize_rcv_mss" is called at "step 5: check the ACK field" when the sk->sk_state is TCP_SYN_RECV and there is a "tcp_validate_incoming" just before it. So when we call "tcp_validate_incoming", the rcv_mss may not been initialized. 2.In function "tcp_validate_incoming", the "Step 1: check sequence number", according to RFC793 page 69, If an incoming segment is not acceptable,an acknowledgment should be sent in reply (unless the RST bit is set, if so drop the segment and return). So we may call "tcp_send_dupack" while the rcv_mss hasn't been initialized. 3.In function "tcp_send_dupack", when the condition is suitable, it'll enter quick ack mode. Notice it only check the seq ! So I think add another state check should be OK. Any suggestion ? Thanks, Weilong > > > > . > -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Mon, 2014-11-03 at 13:31 +0800, chenweilong wrote: > Hi Eric, > > I check the code and find that: > > 1.In function "tcp_rcv_state_process", > the "tcp_initialize_rcv_mss" is called at "step 5: check the ACK field" when the sk->sk_state is TCP_SYN_RECV > and there is a "tcp_validate_incoming" just before it. > So when we call "tcp_validate_incoming", the rcv_mss may not been initialized. > > 2.In function "tcp_validate_incoming", > the "Step 1: check sequence number", according to RFC793 page 69, > If an incoming segment is not acceptable,an acknowledgment should be sent in reply (unless the RST > bit is set, if so drop the segment and return). > So we may call "tcp_send_dupack" while the rcv_mss hasn't been initialized. > > 3.In function "tcp_send_dupack", > when the condition is suitable, it'll enter quick ack mode. Notice it only check the seq ! > So I think add another state check should be OK. > > Any suggestion ? > You did find what immediate conditions for the crash (rcv_mss = 0, state = TCP_SYN_RCV) were. Your patch avoids the zero divide, but leaves other issues. rcv_mss = 0 here is a sign some logic is wrong in the stack. Given this potential zero divide had been there for years, I believe we should take the time for a more complete fix, instead of papering over the immediate problem. We have been working with Neal to reproduce the issue with packetdrill, we'll post our results when we manage to get our first crash ;) Thanks ! -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index 4e4617e..9eb56dc 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -3986,7 +3986,8 @@ static void tcp_send_dupack(struct sock *sk, const struct sk_buff *skb) if (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq && before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) { NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKLOST); - tcp_enter_quickack_mode(sk); + if (sk->sk_state != TCP_SYN_RECV) + tcp_enter_quickack_mode(sk); if (tcp_is_sack(tp) && sysctl_tcp_dsack) { u32 end_seq = TCP_SKB_CB(skb)->end_seq;