| Message ID | 1657525740-7585-1-git-send-email-liyonglong@chinatelecom.cn (mailing list archive) |
|---|---|
| State | Superseded |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | [v3] tcp: make retransmitted SKB fit into the send window | expand |
On Mon, Jul 11, 2022 at 9:56 AM Yonglong Li <liyonglong@chinatelecom.cn> wrote: > > current code of __tcp_retransmit_skb only check TCP_SKB_CB(skb)->seq > in send window, and TCP_SKB_CB(skb)->seq_end maybe out of send window. > If receiver has shrunk his window, and skb is out of new window, it > should retransmit a smaller portion of the payload. > > test packetdrill script: > 0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3 > +0 fcntl(3, F_GETFL) = 0x2 (flags O_RDWR) > +0 fcntl(3, F_SETFL, O_RDWR|O_NONBLOCK) = 0 > > +0 connect(3, ..., ...) = -1 EINPROGRESS (Operation now in progress) > +0 > S 0:0(0) win 65535 <mss 1460,sackOK,TS val 100 ecr 0,nop,wscale 8> > +.05 < S. 0:0(0) ack 1 win 6000 <mss 1000,nop,nop,sackOK> > +0 > . 1:1(0) ack 1 > > +0 write(3, ..., 10000) = 10000 > > +0 > . 1:2001(2000) ack 1 win 65535 > +0 > . 2001:4001(2000) ack 1 win 65535 > +0 > . 4001:6001(2000) ack 1 win 65535 > > +.05 < . 1:1(0) ack 4001 win 1001 > > and tcpdump show: > 192.168.226.67.55 > 192.0.2.1.8080: Flags [.], seq 1:2001, ack 1, win 65535, length 2000 > 192.168.226.67.55 > 192.0.2.1.8080: Flags [.], seq 2001:4001, ack 1, win 65535, length 2000 > 192.168.226.67.55 > 192.0.2.1.8080: Flags [P.], seq 4001:5001, ack 1, win 65535, length 1000 > 192.168.226.67.55 > 192.0.2.1.8080: Flags [.], seq 5001:6001, ack 1, win 65535, length 1000 > 192.0.2.1.8080 > 192.168.226.67.55: Flags [.], ack 4001, win 1001, length 0 > 192.168.226.67.55 > 192.0.2.1.8080: Flags [.], seq 5001:6001, ack 1, win 65535, length 1000 > 192.168.226.67.55 > 192.0.2.1.8080: Flags [P.], seq 4001:5001, ack 1, win 65535, length 1000 > > when cient retract window to 1001, send window is [4001,5002], > but TLP send 5001-6001 packet which is out of send window. > > Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2") > Signed-off-by: Yonglong Li <liyonglong@chinatelecom.cn> > --- > net/ipv4/tcp_output.c | 36 ++++++++++++++++++++++++------------ > 1 file changed, 24 insertions(+), 12 deletions(-) > > diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c > index 18c913a..efd0f05 100644 > --- a/net/ipv4/tcp_output.c > +++ b/net/ipv4/tcp_output.c > @@ -3100,7 +3100,6 @@ static bool tcp_can_collapse(const struct sock *sk, const struct sk_buff *skb) > static void tcp_retrans_try_collapse(struct sock *sk, struct sk_buff *to, > int space) > { > - struct tcp_sock *tp = tcp_sk(sk); > struct sk_buff *skb = to, *tmp; > bool first = true; > > @@ -3123,14 +3122,18 @@ static void tcp_retrans_try_collapse(struct sock *sk, struct sk_buff *to, > continue; > } > > - if (space < 0) > - break; > - > - if (after(TCP_SKB_CB(skb)->end_seq, tcp_wnd_end(tp))) > + if (space < 0) { > + if (unlikely(tcp_fragment(sk, TCP_FRAG_IN_RTX_QUEUE, > + skb, space + skb->len, > + tcp_current_mss(sk), GFP_ATOMIC))) What are you doing here ? This seems wrong. Can we please stick to the patch I sent earlier. If you want to amend it later, you can do this in a separate patch, with a clear explanation. > + break; > + tcp_collapse_retrans(sk, to); > break; > + } > > if (!tcp_collapse_retrans(sk, to)) > break; > + > } > } > > @@ -3144,7 +3147,7 @@ int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs) > struct tcp_sock *tp = tcp_sk(sk); > unsigned int cur_mss; > int diff, len, err; > - > + int avail_wnd; > > /* Inconclusive MTU probe */ > if (icsk->icsk_mtup.probe_size) > @@ -3166,17 +3169,25 @@ int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs) > return -EHOSTUNREACH; /* Routing failure or similar. */ > > cur_mss = tcp_current_mss(sk); > + avail_wnd = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq; > > /* If receiver has shrunk his window, and skb is out of > * new window, do not retransmit it. The exception is the > * case, when window is shrunk to zero. In this case > - * our retransmit serves as a zero window probe. > + * our retransmit of one segment serves as a zero window probe. > */ > - if (!before(TCP_SKB_CB(skb)->seq, tcp_wnd_end(tp)) && > - TCP_SKB_CB(skb)->seq != tp->snd_una) > - return -EAGAIN; > + if (avail_wnd <= 0) { > + if (TCP_SKB_CB(skb)->seq != tp->snd_una) > + return -EAGAIN; > + avail_wnd = cur_mss; > + } > > len = cur_mss * segs; > + if (len > avail_wnd) { > + len = rounddown(avail_wnd, cur_mss); > + if (!len) > + len = avail_wnd; > + } > if (skb->len > len) { > if (tcp_fragment(sk, TCP_FRAG_IN_RTX_QUEUE, skb, len, > cur_mss, GFP_ATOMIC)) > @@ -3190,8 +3201,9 @@ int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs) > diff -= tcp_skb_pcount(skb); > if (diff) > tcp_adjust_pcount(sk, skb, diff); > - if (skb->len < cur_mss) > - tcp_retrans_try_collapse(sk, skb, cur_mss); > + avail_wnd = min_t(int, avail_wnd, cur_mss); > + if (skb->len < avail_wnd) > + tcp_retrans_try_collapse(sk, skb, avail_wnd); > } > > /* RFC3168, section 6.1.1.1. ECN fallback */ > -- > 1.8.3.1 >
On 7/11/2022 5:06 PM, Eric Dumazet wrote: > On Mon, Jul 11, 2022 at 9:56 AM Yonglong Li <liyonglong@chinatelecom.cn> wrote: >> >> current code of __tcp_retransmit_skb only check TCP_SKB_CB(skb)->seq >> in send window, and TCP_SKB_CB(skb)->seq_end maybe out of send window. >> If receiver has shrunk his window, and skb is out of new window, it >> should retransmit a smaller portion of the payload. >> >> test packetdrill script: >> 0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3 >> +0 fcntl(3, F_GETFL) = 0x2 (flags O_RDWR) >> +0 fcntl(3, F_SETFL, O_RDWR|O_NONBLOCK) = 0 >> >> +0 connect(3, ..., ...) = -1 EINPROGRESS (Operation now in progress) >> +0 > S 0:0(0) win 65535 <mss 1460,sackOK,TS val 100 ecr 0,nop,wscale 8> >> +.05 < S. 0:0(0) ack 1 win 6000 <mss 1000,nop,nop,sackOK> >> +0 > . 1:1(0) ack 1 >> >> +0 write(3, ..., 10000) = 10000 >> >> +0 > . 1:2001(2000) ack 1 win 65535 >> +0 > . 2001:4001(2000) ack 1 win 65535 >> +0 > . 4001:6001(2000) ack 1 win 65535 >> >> +.05 < . 1:1(0) ack 4001 win 1001 >> >> and tcpdump show: >> 192.168.226.67.55 > 192.0.2.1.8080: Flags [.], seq 1:2001, ack 1, win 65535, length 2000 >> 192.168.226.67.55 > 192.0.2.1.8080: Flags [.], seq 2001:4001, ack 1, win 65535, length 2000 >> 192.168.226.67.55 > 192.0.2.1.8080: Flags [P.], seq 4001:5001, ack 1, win 65535, length 1000 >> 192.168.226.67.55 > 192.0.2.1.8080: Flags [.], seq 5001:6001, ack 1, win 65535, length 1000 >> 192.0.2.1.8080 > 192.168.226.67.55: Flags [.], ack 4001, win 1001, length 0 >> 192.168.226.67.55 > 192.0.2.1.8080: Flags [.], seq 5001:6001, ack 1, win 65535, length 1000 >> 192.168.226.67.55 > 192.0.2.1.8080: Flags [P.], seq 4001:5001, ack 1, win 65535, length 1000 >> >> when cient retract window to 1001, send window is [4001,5002], >> but TLP send 5001-6001 packet which is out of send window. >> >> Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2") >> Signed-off-by: Yonglong Li <liyonglong@chinatelecom.cn> >> --- >> net/ipv4/tcp_output.c | 36 ++++++++++++++++++++++++------------ >> 1 file changed, 24 insertions(+), 12 deletions(-) >> >> diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c >> index 18c913a..efd0f05 100644 >> --- a/net/ipv4/tcp_output.c >> +++ b/net/ipv4/tcp_output.c >> @@ -3100,7 +3100,6 @@ static bool tcp_can_collapse(const struct sock *sk, const struct sk_buff *skb) >> static void tcp_retrans_try_collapse(struct sock *sk, struct sk_buff *to, >> int space) >> { >> - struct tcp_sock *tp = tcp_sk(sk); >> struct sk_buff *skb = to, *tmp; >> bool first = true; >> >> @@ -3123,14 +3122,18 @@ static void tcp_retrans_try_collapse(struct sock *sk, struct sk_buff *to, >> continue; >> } >> >> - if (space < 0) >> - break; >> - >> - if (after(TCP_SKB_CB(skb)->end_seq, tcp_wnd_end(tp))) >> + if (space < 0) { >> + if (unlikely(tcp_fragment(sk, TCP_FRAG_IN_RTX_QUEUE, >> + skb, space + skb->len, >> + tcp_current_mss(sk), GFP_ATOMIC))) > > What are you doing here ? > > This seems wrong. I think we can collapse more data if skb->len < avail_wnd. If you think doing this is wrong I will remove it. > > Can we please stick to the patch I sent earlier. > > If you want to amend it later, you can do this in a separate patch, > with a clear explanation. > >> + break; >> + tcp_collapse_retrans(sk, to); >> break; >> + } >> >> if (!tcp_collapse_retrans(sk, to)) >> break; >> + >> } >> } >> >> @@ -3144,7 +3147,7 @@ int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs) >> struct tcp_sock *tp = tcp_sk(sk); >> unsigned int cur_mss; >> int diff, len, err; >> - >> + int avail_wnd; >> >> /* Inconclusive MTU probe */ >> if (icsk->icsk_mtup.probe_size) >> @@ -3166,17 +3169,25 @@ int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs) >> return -EHOSTUNREACH; /* Routing failure or similar. */ >> >> cur_mss = tcp_current_mss(sk); >> + avail_wnd = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq; >> >> /* If receiver has shrunk his window, and skb is out of >> * new window, do not retransmit it. The exception is the >> * case, when window is shrunk to zero. In this case >> - * our retransmit serves as a zero window probe. >> + * our retransmit of one segment serves as a zero window probe. >> */ >> - if (!before(TCP_SKB_CB(skb)->seq, tcp_wnd_end(tp)) && >> - TCP_SKB_CB(skb)->seq != tp->snd_una) >> - return -EAGAIN; >> + if (avail_wnd <= 0) { >> + if (TCP_SKB_CB(skb)->seq != tp->snd_una) >> + return -EAGAIN; >> + avail_wnd = cur_mss; >> + } >> >> len = cur_mss * segs; >> + if (len > avail_wnd) { >> + len = rounddown(avail_wnd, cur_mss); >> + if (!len) >> + len = avail_wnd; >> + } >> if (skb->len > len) { >> if (tcp_fragment(sk, TCP_FRAG_IN_RTX_QUEUE, skb, len, >> cur_mss, GFP_ATOMIC)) >> @@ -3190,8 +3201,9 @@ int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs) >> diff -= tcp_skb_pcount(skb); >> if (diff) >> tcp_adjust_pcount(sk, skb, diff); >> - if (skb->len < cur_mss) >> - tcp_retrans_try_collapse(sk, skb, cur_mss); >> + avail_wnd = min_t(int, avail_wnd, cur_mss); >> + if (skb->len < avail_wnd) >> + tcp_retrans_try_collapse(sk, skb, avail_wnd); >> } >> >> /* RFC3168, section 6.1.1.1. ECN fallback */ >> -- >> 1.8.3.1 >> >
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c index 18c913a..efd0f05 100644 --- a/net/ipv4/tcp_output.c +++ b/net/ipv4/tcp_output.c @@ -3100,7 +3100,6 @@ static bool tcp_can_collapse(const struct sock *sk, const struct sk_buff *skb) static void tcp_retrans_try_collapse(struct sock *sk, struct sk_buff *to, int space) { - struct tcp_sock *tp = tcp_sk(sk); struct sk_buff *skb = to, *tmp; bool first = true; @@ -3123,14 +3122,18 @@ static void tcp_retrans_try_collapse(struct sock *sk, struct sk_buff *to, continue; } - if (space < 0) - break; - - if (after(TCP_SKB_CB(skb)->end_seq, tcp_wnd_end(tp))) + if (space < 0) { + if (unlikely(tcp_fragment(sk, TCP_FRAG_IN_RTX_QUEUE, + skb, space + skb->len, + tcp_current_mss(sk), GFP_ATOMIC))) + break; + tcp_collapse_retrans(sk, to); break; + } if (!tcp_collapse_retrans(sk, to)) break; + } } @@ -3144,7 +3147,7 @@ int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs) struct tcp_sock *tp = tcp_sk(sk); unsigned int cur_mss; int diff, len, err; - + int avail_wnd; /* Inconclusive MTU probe */ if (icsk->icsk_mtup.probe_size) @@ -3166,17 +3169,25 @@ int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs) return -EHOSTUNREACH; /* Routing failure or similar. */ cur_mss = tcp_current_mss(sk); + avail_wnd = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq; /* If receiver has shrunk his window, and skb is out of * new window, do not retransmit it. The exception is the * case, when window is shrunk to zero. In this case - * our retransmit serves as a zero window probe. + * our retransmit of one segment serves as a zero window probe. */ - if (!before(TCP_SKB_CB(skb)->seq, tcp_wnd_end(tp)) && - TCP_SKB_CB(skb)->seq != tp->snd_una) - return -EAGAIN; + if (avail_wnd <= 0) { + if (TCP_SKB_CB(skb)->seq != tp->snd_una) + return -EAGAIN; + avail_wnd = cur_mss; + } len = cur_mss * segs; + if (len > avail_wnd) { + len = rounddown(avail_wnd, cur_mss); + if (!len) + len = avail_wnd; + } if (skb->len > len) { if (tcp_fragment(sk, TCP_FRAG_IN_RTX_QUEUE, skb, len, cur_mss, GFP_ATOMIC)) @@ -3190,8 +3201,9 @@ int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs) diff -= tcp_skb_pcount(skb); if (diff) tcp_adjust_pcount(sk, skb, diff); - if (skb->len < cur_mss) - tcp_retrans_try_collapse(sk, skb, cur_mss); + avail_wnd = min_t(int, avail_wnd, cur_mss); + if (skb->len < avail_wnd) + tcp_retrans_try_collapse(sk, skb, avail_wnd); } /* RFC3168, section 6.1.1.1. ECN fallback */
current code of __tcp_retransmit_skb only check TCP_SKB_CB(skb)->seq in send window, and TCP_SKB_CB(skb)->seq_end maybe out of send window. If receiver has shrunk his window, and skb is out of new window, it should retransmit a smaller portion of the payload. test packetdrill script: 0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3 +0 fcntl(3, F_GETFL) = 0x2 (flags O_RDWR) +0 fcntl(3, F_SETFL, O_RDWR|O_NONBLOCK) = 0 +0 connect(3, ..., ...) = -1 EINPROGRESS (Operation now in progress) +0 > S 0:0(0) win 65535 <mss 1460,sackOK,TS val 100 ecr 0,nop,wscale 8> +.05 < S. 0:0(0) ack 1 win 6000 <mss 1000,nop,nop,sackOK> +0 > . 1:1(0) ack 1 +0 write(3, ..., 10000) = 10000 +0 > . 1:2001(2000) ack 1 win 65535 +0 > . 2001:4001(2000) ack 1 win 65535 +0 > . 4001:6001(2000) ack 1 win 65535 +.05 < . 1:1(0) ack 4001 win 1001 and tcpdump show: 192.168.226.67.55 > 192.0.2.1.8080: Flags [.], seq 1:2001, ack 1, win 65535, length 2000 192.168.226.67.55 > 192.0.2.1.8080: Flags [.], seq 2001:4001, ack 1, win 65535, length 2000 192.168.226.67.55 > 192.0.2.1.8080: Flags [P.], seq 4001:5001, ack 1, win 65535, length 1000 192.168.226.67.55 > 192.0.2.1.8080: Flags [.], seq 5001:6001, ack 1, win 65535, length 1000 192.0.2.1.8080 > 192.168.226.67.55: Flags [.], ack 4001, win 1001, length 0 192.168.226.67.55 > 192.0.2.1.8080: Flags [.], seq 5001:6001, ack 1, win 65535, length 1000 192.168.226.67.55 > 192.0.2.1.8080: Flags [P.], seq 4001:5001, ack 1, win 65535, length 1000 when cient retract window to 1001, send window is [4001,5002], but TLP send 5001-6001 packet which is out of send window. Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2") Signed-off-by: Yonglong Li <liyonglong@chinatelecom.cn> --- net/ipv4/tcp_output.c | 36 ++++++++++++++++++++++++------------ 1 file changed, 24 insertions(+), 12 deletions(-)