tcp: fix marked lost packets not being retransmitted
diff mbox series

Message ID 1578993820-2114-1-git-send-email-yangpc@wangsu.com
State Accepted
Delegated to: David Miller
Headers show
Series
  • tcp: fix marked lost packets not being retransmitted
Related show

Commit Message

Pengcheng Yang Jan. 14, 2020, 9:23 a.m. UTC
When the packet pointed to by retransmit_skb_hint is unlinked by ACK,
retransmit_skb_hint will be set to NULL in tcp_clean_rtx_queue().
If packet loss is detected at this time, retransmit_skb_hint will be set
to point to the current packet loss in tcp_verify_retransmit_hint(),
then the packets that were previously marked lost but not retransmitted
due to the restriction of cwnd will be skipped and cannot be
retransmitted.

To fix this, when retransmit_skb_hint is NULL, retransmit_skb_hint can
be reset only after all marked lost packets are retransmitted
(retrans_out >= lost_out), otherwise we need to traverse from
tcp_rtx_queue_head in tcp_xmit_retransmit_queue().

Packetdrill to demonstrate:

// Disable RACK and set max_reordering to keep things simple
    0 `sysctl -q net.ipv4.tcp_recovery=0`
   +0 `sysctl -q net.ipv4.tcp_max_reordering=3`

// Establish a connection
   +0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
   +0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
   +0 bind(3, ..., ...) = 0
   +0 listen(3, 1) = 0

  +.1 < S 0:0(0) win 32792 <mss 1000,sackOK,nop,nop,nop,wscale 7>
   +0 > S. 0:0(0) ack 1 <...>
 +.01 < . 1:1(0) ack 1 win 257
   +0 accept(3, ..., ...) = 4

// Send 8 data segments
   +0 write(4, ..., 8000) = 8000
   +0 > P. 1:8001(8000) ack 1

// Enter recovery and 1:3001 is marked lost
 +.01 < . 1:1(0) ack 1 win 257 <sack 3001:4001,nop,nop>
   +0 < . 1:1(0) ack 1 win 257 <sack 5001:6001 3001:4001,nop,nop>
   +0 < . 1:1(0) ack 1 win 257 <sack 5001:7001 3001:4001,nop,nop>

// Retransmit 1:1001, now retransmit_skb_hint points to 1001:2001
   +0 > . 1:1001(1000) ack 1

// 1001:2001 was ACKed causing retransmit_skb_hint to be set to NULL
 +.01 < . 1:1(0) ack 2001 win 257 <sack 5001:8001 3001:4001,nop,nop>
// Now retransmit_skb_hint points to 4001:5001 which is now marked lost

// BUG: 2001:3001 was not retransmitted
   +0 > . 2001:3001(1000) ack 1

Signed-off-by: Pengcheng Yang <yangpc@wangsu.com>
---
 net/ipv4/tcp_input.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

Comments

Neal Cardwell Jan. 14, 2020, 3:26 p.m. UTC | #1
On Tue, Jan 14, 2020 at 4:24 AM Pengcheng Yang <yangpc@wangsu.com> wrote:
>
> When the packet pointed to by retransmit_skb_hint is unlinked by ACK,
> retransmit_skb_hint will be set to NULL in tcp_clean_rtx_queue().
> If packet loss is detected at this time, retransmit_skb_hint will be set
> to point to the current packet loss in tcp_verify_retransmit_hint(),
> then the packets that were previously marked lost but not retransmitted
> due to the restriction of cwnd will be skipped and cannot be
> retransmitted.
>
> To fix this, when retransmit_skb_hint is NULL, retransmit_skb_hint can
> be reset only after all marked lost packets are retransmitted
> (retrans_out >= lost_out), otherwise we need to traverse from
> tcp_rtx_queue_head in tcp_xmit_retransmit_queue().
...
> --- a/net/ipv4/tcp_input.c
> +++ b/net/ipv4/tcp_input.c
> @@ -915,9 +915,10 @@ static void tcp_check_sack_reordering(struct sock *sk, const u32 low_seq,
>  /* This must be called before lost_out is incremented */
>  static void tcp_verify_retransmit_hint(struct tcp_sock *tp, struct sk_buff *skb)
>  {
> -       if (!tp->retransmit_skb_hint ||
> -           before(TCP_SKB_CB(skb)->seq,
> -                  TCP_SKB_CB(tp->retransmit_skb_hint)->seq))
> +       if ((!tp->retransmit_skb_hint && tp->retrans_out >= tp->lost_out) ||
> +           (tp->retransmit_skb_hint &&
> +            before(TCP_SKB_CB(skb)->seq,
> +                   TCP_SKB_CB(tp->retransmit_skb_hint)->seq)))
>                 tp->retransmit_skb_hint = skb;
>  }

Thanks for finding and fixing this issue, and for providing the very
nice packetdrill test case! The fix looks good to me.

I verified that the packetdrill test fails at the line notated "BUG"
without the patch applied:

fr-retrans-hint-skip-fix.pkt:33: error handling packet: live packet
field tcp_seq: expected: 2001 (0x7d1) vs actual: 4001 (0xfa1)
script packet:  0.137311 . 2001:3001(1000) ack 1
actual packet:  0.137307 . 4001:5001(1000) ack 1 win 256

Also verified that the test passes with the patch applied, and that
our internal SACK and fast recovery tests continue to pass with this
patch applied.

Acked-by: Neal Cardwell <ncardwell@google.com>
Tested-by: Neal Cardwell <ncardwell@google.com>

thanks,
neal
Eric Dumazet Jan. 14, 2020, 4:05 p.m. UTC | #2
On Tue, Jan 14, 2020 at 1:24 AM Pengcheng Yang <yangpc@wangsu.com> wrote:
>
> When the packet pointed to by retransmit_skb_hint is unlinked by ACK,
> retransmit_skb_hint will be set to NULL in tcp_clean_rtx_queue().
> If packet loss is detected at this time, retransmit_skb_hint will be set
> to point to the current packet loss in tcp_verify_retransmit_hint(),
> then the packets that were previously marked lost but not retransmitted
> due to the restriction of cwnd will be skipped and cannot be
> retransmitted.


"cannot be retransmittted"  sounds quite alarming.

You meant they will eventually be retransmitted, or that the flow is
completely frozen at this point ?

Thanks for the fix and test !

(Not sure why you CC all these people having little TCP expertise btw)

> To fix this, when retransmit_skb_hint is NULL, retransmit_skb_hint can
> be reset only after all marked lost packets are retransmitted
> (retrans_out >= lost_out), otherwise we need to traverse from
> tcp_rtx_queue_head in tcp_xmit_retransmit_queue().
>
> Packetdrill to demonstrate:
>
> // Disable RACK and set max_reordering to keep things simple
>     0 `sysctl -q net.ipv4.tcp_recovery=0`
>    +0 `sysctl -q net.ipv4.tcp_max_reordering=3`
>
> // Establish a connection
>    +0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
>    +0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
>    +0 bind(3, ..., ...) = 0
>    +0 listen(3, 1) = 0
>
>   +.1 < S 0:0(0) win 32792 <mss 1000,sackOK,nop,nop,nop,wscale 7>
>    +0 > S. 0:0(0) ack 1 <...>
>  +.01 < . 1:1(0) ack 1 win 257
>    +0 accept(3, ..., ...) = 4
>
> // Send 8 data segments
>    +0 write(4, ..., 8000) = 8000
>    +0 > P. 1:8001(8000) ack 1
>
> // Enter recovery and 1:3001 is marked lost
>  +.01 < . 1:1(0) ack 1 win 257 <sack 3001:4001,nop,nop>
>    +0 < . 1:1(0) ack 1 win 257 <sack 5001:6001 3001:4001,nop,nop>
>    +0 < . 1:1(0) ack 1 win 257 <sack 5001:7001 3001:4001,nop,nop>
>
> // Retransmit 1:1001, now retransmit_skb_hint points to 1001:2001
>    +0 > . 1:1001(1000) ack 1
>
> // 1001:2001 was ACKed causing retransmit_skb_hint to be set to NULL
>  +.01 < . 1:1(0) ack 2001 win 257 <sack 5001:8001 3001:4001,nop,nop>
> // Now retransmit_skb_hint points to 4001:5001 which is now marked lost
>
> // BUG: 2001:3001 was not retransmitted
>    +0 > . 2001:3001(1000) ack 1
>
> Signed-off-by: Pengcheng Yang <yangpc@wangsu.com>
> ---
>  net/ipv4/tcp_input.c | 7 ++++---
>  1 file changed, 4 insertions(+), 3 deletions(-)
>
> diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
> index 0238b55..5347ab2 100644
> --- a/net/ipv4/tcp_input.c
> +++ b/net/ipv4/tcp_input.c
> @@ -915,9 +915,10 @@ static void tcp_check_sack_reordering(struct sock *sk, const u32 low_seq,
>  /* This must be called before lost_out is incremented */
>  static void tcp_verify_retransmit_hint(struct tcp_sock *tp, struct sk_buff *skb)
>  {
> -       if (!tp->retransmit_skb_hint ||
> -           before(TCP_SKB_CB(skb)->seq,
> -                  TCP_SKB_CB(tp->retransmit_skb_hint)->seq))
> +       if ((!tp->retransmit_skb_hint && tp->retrans_out >= tp->lost_out) ||
> +           (tp->retransmit_skb_hint &&
> +            before(TCP_SKB_CB(skb)->seq,
> +                   TCP_SKB_CB(tp->retransmit_skb_hint)->seq)))
>                 tp->retransmit_skb_hint = skb;
>  }
>
> --
> 1.8.3.1
>
Yuchung Cheng Jan. 14, 2020, 6:39 p.m. UTC | #3
On Tue, Jan 14, 2020 at 8:06 AM Eric Dumazet <edumazet@google.com> wrote:
>
> On Tue, Jan 14, 2020 at 1:24 AM Pengcheng Yang <yangpc@wangsu.com> wrote:
> >
> > When the packet pointed to by retransmit_skb_hint is unlinked by ACK,
> > retransmit_skb_hint will be set to NULL in tcp_clean_rtx_queue().
> > If packet loss is detected at this time, retransmit_skb_hint will be set
> > to point to the current packet loss in tcp_verify_retransmit_hint(),
> > then the packets that were previously marked lost but not retransmitted
> > due to the restriction of cwnd will be skipped and cannot be
> > retransmitted.
>
>
> "cannot be retransmittted"  sounds quite alarming.
>
> You meant they will eventually be retransmitted, or that the flow is
> completely frozen at this point ?
He probably means those lost packets will be skipped until a timeout
that reset hint pointer. nice fix this would save some RTOs.

>
> Thanks for the fix and test !
>
> (Not sure why you CC all these people having little TCP expertise btw)
>
> > To fix this, when retransmit_skb_hint is NULL, retransmit_skb_hint can
> > be reset only after all marked lost packets are retransmitted
> > (retrans_out >= lost_out), otherwise we need to traverse from
> > tcp_rtx_queue_head in tcp_xmit_retransmit_queue().
> >
> > Packetdrill to demonstrate:
> >
> > // Disable RACK and set max_reordering to keep things simple
> >     0 `sysctl -q net.ipv4.tcp_recovery=0`
> >    +0 `sysctl -q net.ipv4.tcp_max_reordering=3`
> >
> > // Establish a connection
> >    +0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
> >    +0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
> >    +0 bind(3, ..., ...) = 0
> >    +0 listen(3, 1) = 0
> >
> >   +.1 < S 0:0(0) win 32792 <mss 1000,sackOK,nop,nop,nop,wscale 7>
> >    +0 > S. 0:0(0) ack 1 <...>
> >  +.01 < . 1:1(0) ack 1 win 257
> >    +0 accept(3, ..., ...) = 4
> >
> > // Send 8 data segments
> >    +0 write(4, ..., 8000) = 8000
> >    +0 > P. 1:8001(8000) ack 1
> >
> > // Enter recovery and 1:3001 is marked lost
> >  +.01 < . 1:1(0) ack 1 win 257 <sack 3001:4001,nop,nop>
> >    +0 < . 1:1(0) ack 1 win 257 <sack 5001:6001 3001:4001,nop,nop>
> >    +0 < . 1:1(0) ack 1 win 257 <sack 5001:7001 3001:4001,nop,nop>
> >
> > // Retransmit 1:1001, now retransmit_skb_hint points to 1001:2001
> >    +0 > . 1:1001(1000) ack 1
> >
> > // 1001:2001 was ACKed causing retransmit_skb_hint to be set to NULL
> >  +.01 < . 1:1(0) ack 2001 win 257 <sack 5001:8001 3001:4001,nop,nop>
> > // Now retransmit_skb_hint points to 4001:5001 which is now marked lost
> >
> > // BUG: 2001:3001 was not retransmitted
> >    +0 > . 2001:3001(1000) ack 1
> >
> > Signed-off-by: Pengcheng Yang <yangpc@wangsu.com>
> > ---
> >  net/ipv4/tcp_input.c | 7 ++++---
> >  1 file changed, 4 insertions(+), 3 deletions(-)
> >
> > diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
> > index 0238b55..5347ab2 100644
> > --- a/net/ipv4/tcp_input.c
> > +++ b/net/ipv4/tcp_input.c
> > @@ -915,9 +915,10 @@ static void tcp_check_sack_reordering(struct sock *sk, const u32 low_seq,
> >  /* This must be called before lost_out is incremented */
> >  static void tcp_verify_retransmit_hint(struct tcp_sock *tp, struct sk_buff *skb)
> >  {
> > -       if (!tp->retransmit_skb_hint ||
> > -           before(TCP_SKB_CB(skb)->seq,
> > -                  TCP_SKB_CB(tp->retransmit_skb_hint)->seq))
> > +       if ((!tp->retransmit_skb_hint && tp->retrans_out >= tp->lost_out) ||
> > +           (tp->retransmit_skb_hint &&
> > +            before(TCP_SKB_CB(skb)->seq,
> > +                   TCP_SKB_CB(tp->retransmit_skb_hint)->seq)))
> >                 tp->retransmit_skb_hint = skb;
> >  }
> >
> > --
> > 1.8.3.1
> >
Pengcheng Yang Jan. 15, 2020, 6:58 a.m. UTC | #4
On Wed, Jan 15, 2020 at 2:40 AM Yuchung Cheng <ycheng@google.com> wrote:
>
> On Tue, Jan 14, 2020 at 8:06 AM Eric Dumazet <edumazet@google.com>
> wrote:
> >
> > On Tue, Jan 14, 2020 at 1:24 AM Pengcheng Yang <yangpc@wangsu.com>
> wrote:
> > >
> > > When the packet pointed to by retransmit_skb_hint is unlinked by ACK,
> > > retransmit_skb_hint will be set to NULL in tcp_clean_rtx_queue().
> > > If packet loss is detected at this time, retransmit_skb_hint will be set
> > > to point to the current packet loss in tcp_verify_retransmit_hint(),
> > > then the packets that were previously marked lost but not retransmitted
> > > due to the restriction of cwnd will be skipped and cannot be
> > > retransmitted.
> >
> >
> > "cannot be retransmittted"  sounds quite alarming.
> >
> > You meant they will eventually be retransmitted, or that the flow is
> > completely frozen at this point ?
> He probably means those lost packets will be skipped until a timeout
> that reset hint pointer. nice fix this would save some RTOs.

Yes, I mean those packets will not be retransmitted until RTO.
I'm sorry I didn't describe it clearly.

> 
> >
> > Thanks for the fix and test !
> >
> > (Not sure why you CC all these people having little TCP expertise btw)

Maybe I should CC fewer people :)

> >
> > > To fix this, when retransmit_skb_hint is NULL, retransmit_skb_hint can
> > > be reset only after all marked lost packets are retransmitted
> > > (retrans_out >= lost_out), otherwise we need to traverse from
> > > tcp_rtx_queue_head in tcp_xmit_retransmit_queue().
> > >
> > > Packetdrill to demonstrate:
> > >
> > > // Disable RACK and set max_reordering to keep things simple
> > >     0 `sysctl -q net.ipv4.tcp_recovery=0`
> > >    +0 `sysctl -q net.ipv4.tcp_max_reordering=3`
> > >
> > > // Establish a connection
> > >    +0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
> > >    +0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
> > >    +0 bind(3, ..., ...) = 0
> > >    +0 listen(3, 1) = 0
> > >
> > >   +.1 < S 0:0(0) win 32792 <mss 1000,sackOK,nop,nop,nop,wscale 7>
> > >    +0 > S. 0:0(0) ack 1 <...>
> > >  +.01 < . 1:1(0) ack 1 win 257
> > >    +0 accept(3, ..., ...) = 4
> > >
> > > // Send 8 data segments
> > >    +0 write(4, ..., 8000) = 8000
> > >    +0 > P. 1:8001(8000) ack 1
> > >
> > > // Enter recovery and 1:3001 is marked lost
> > >  +.01 < . 1:1(0) ack 1 win 257 <sack 3001:4001,nop,nop>
> > >    +0 < . 1:1(0) ack 1 win 257 <sack 5001:6001 3001:4001,nop,nop>
> > >    +0 < . 1:1(0) ack 1 win 257 <sack 5001:7001 3001:4001,nop,nop>
> > >
> > > // Retransmit 1:1001, now retransmit_skb_hint points to 1001:2001
> > >    +0 > . 1:1001(1000) ack 1
> > >
> > > // 1001:2001 was ACKed causing retransmit_skb_hint to be set to NULL
> > >  +.01 < . 1:1(0) ack 2001 win 257 <sack 5001:8001 3001:4001,nop,nop>
> > > // Now retransmit_skb_hint points to 4001:5001 which is now marked lost
> > >
> > > // BUG: 2001:3001 was not retransmitted
> > >    +0 > . 2001:3001(1000) ack 1
> > >
> > > Signed-off-by: Pengcheng Yang <yangpc@wangsu.com>
> > > ---
> > >  net/ipv4/tcp_input.c | 7 ++++---
> > >  1 file changed, 4 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
> > > index 0238b55..5347ab2 100644
> > > --- a/net/ipv4/tcp_input.c
> > > +++ b/net/ipv4/tcp_input.c
> > > @@ -915,9 +915,10 @@ static void tcp_check_sack_reordering(struct sock
> *sk, const u32 low_seq,
> > >  /* This must be called before lost_out is incremented */
> > >  static void tcp_verify_retransmit_hint(struct tcp_sock *tp, struct sk_buff
> *skb)
> > >  {
> > > -       if (!tp->retransmit_skb_hint ||
> > > -           before(TCP_SKB_CB(skb)->seq,
> > > -                  TCP_SKB_CB(tp->retransmit_skb_hint)->seq))
> > > +       if ((!tp->retransmit_skb_hint && tp->retrans_out >= tp->lost_out)
> ||
> > > +           (tp->retransmit_skb_hint &&
> > > +            before(TCP_SKB_CB(skb)->seq,
> > > +                   TCP_SKB_CB(tp->retransmit_skb_hint)->seq)))
> > >                 tp->retransmit_skb_hint = skb;
> > >  }
> > >
> > > --
> > > 1.8.3.1
> > >
David Miller Jan. 15, 2020, 9:35 p.m. UTC | #5
From: Pengcheng Yang <yangpc@wangsu.com>
Date: Tue, 14 Jan 2020 17:23:40 +0800

> When the packet pointed to by retransmit_skb_hint is unlinked by ACK,
> retransmit_skb_hint will be set to NULL in tcp_clean_rtx_queue().
> If packet loss is detected at this time, retransmit_skb_hint will be set
> to point to the current packet loss in tcp_verify_retransmit_hint(),
> then the packets that were previously marked lost but not retransmitted
> due to the restriction of cwnd will be skipped and cannot be
> retransmitted.
> 
> To fix this, when retransmit_skb_hint is NULL, retransmit_skb_hint can
> be reset only after all marked lost packets are retransmitted
> (retrans_out >= lost_out), otherwise we need to traverse from
> tcp_rtx_queue_head in tcp_xmit_retransmit_queue().
> 
> Packetdrill to demonstrate:
 ...
> Signed-off-by: Pengcheng Yang <yangpc@wangsu.com>

Applied, thank you.

Patch
diff mbox series

diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 0238b55..5347ab2 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -915,9 +915,10 @@  static void tcp_check_sack_reordering(struct sock *sk, const u32 low_seq,
 /* This must be called before lost_out is incremented */
 static void tcp_verify_retransmit_hint(struct tcp_sock *tp, struct sk_buff *skb)
 {
-	if (!tp->retransmit_skb_hint ||
-	    before(TCP_SKB_CB(skb)->seq,
-		   TCP_SKB_CB(tp->retransmit_skb_hint)->seq))
+	if ((!tp->retransmit_skb_hint && tp->retrans_out >= tp->lost_out) ||
+	    (tp->retransmit_skb_hint &&
+	     before(TCP_SKB_CB(skb)->seq,
+		    TCP_SKB_CB(tp->retransmit_skb_hint)->seq)))
 		tp->retransmit_skb_hint = skb;
 }