Re: [PATCH net 3/3] tcp: fix xmit timer to only be reset if data ACKed/SACKed

[Neal Cardwell]

On Sun, Aug 6, 2017 at 3:44 AM, Willy Tarreau  wrote:
>
> On Sun, Aug 06, 2017 at 07:39:57AM +, maowenan wrote:
> >
> > [Mao Wenan]It works well with these patches of v3.10, and the
> > retransmission packet is about 250ms after TLP probe.
> > Thank you very much for these patches under 3.10.

Great. Thanks for testing those patches!

> Thanks, I'll take them then.

And thanks for applying the patches to 3.10.y!

neal

Re: [PATCH net 3/3] tcp: fix xmit timer to only be reset if data ACKed/SACKed

[Neal Cardwell]

On Sun, Aug 6, 2017 at 3:44 AM, Willy Tarreau  wrote:
>
> On Sun, Aug 06, 2017 at 07:39:57AM +, maowenan wrote:
> >
> > [Mao Wenan]It works well with these patches of v3.10, and the
> > retransmission packet is about 250ms after TLP probe.
> > Thank you very much for these patches under 3.10.

Great. Thanks for testing those patches!

> Thanks, I'll take them then.

And thanks for applying the patches to 3.10.y!

neal

Re: [PATCH net 3/3] tcp: fix xmit timer to only be reset if data ACKed/SACKed

[Willy Tarreau]

On Sun, Aug 06, 2017 at 07:39:57AM +, maowenan wrote:
> 
> 
> > -Original Message-
> > From: Willy Tarreau [mailto:w...@1wt.eu]
> > Sent: Saturday, August 05, 2017 2:19 AM
> > To: Neal Cardwell
> > Cc: maowenan; David Miller; netdev@vger.kernel.org; Yuchung Cheng; Nandita
> > Dukkipati; Eric Dumazet
> > Subject: Re: [PATCH net 3/3] tcp: fix xmit timer to only be reset if data
> > ACKed/SACKed
> > 
> > On Fri, Aug 04, 2017 at 02:01:34PM -0400, Neal Cardwell wrote:
> > > On Fri, Aug 4, 2017 at 1:10 PM, Willy Tarreau  wrote:
> > > > Hi Neal,
> > > >
> > > > On Fri, Aug 04, 2017 at 12:59:51PM -0400, Neal Cardwell wrote:
> > > >> I have attached patches for this fix rebased on to v3.10.107, the
> > > >> latest stable release for 3.10. That's pretty far back in history,
> > > >> so there were substantial conflict resolutions and adjustments 
> > > >> required.
> > > >> :-) Hope that helps.
> > > >
> > > > At least it will help me :-)
> > > >
> > > > Do you suggest that I queue them for 3.10.108, that I wait for
> > > > Maowenan to test them more broadly first or anything else ?
> > >
> > > Let's wait for Maowenan to test them first.
> [Mao Wenan]It works well with these patches of v3.10, and the 
> retransmission packet is about 250ms after TLP probe.
> Thank you very much for these patches under 3.10.

Thanks, I'll take them then.

Willy

RE: [PATCH net 3/3] tcp: fix xmit timer to only be reset if data ACKed/SACKed

[maowenan]

> -Original Message-
> From: Willy Tarreau [mailto:w...@1wt.eu]
> Sent: Saturday, August 05, 2017 2:19 AM
> To: Neal Cardwell
> Cc: maowenan; David Miller; netdev@vger.kernel.org; Yuchung Cheng; Nandita
> Dukkipati; Eric Dumazet
> Subject: Re: [PATCH net 3/3] tcp: fix xmit timer to only be reset if data
> ACKed/SACKed
> 
> On Fri, Aug 04, 2017 at 02:01:34PM -0400, Neal Cardwell wrote:
> > On Fri, Aug 4, 2017 at 1:10 PM, Willy Tarreau  wrote:
> > > Hi Neal,
> > >
> > > On Fri, Aug 04, 2017 at 12:59:51PM -0400, Neal Cardwell wrote:
> > >> I have attached patches for this fix rebased on to v3.10.107, the
> > >> latest stable release for 3.10. That's pretty far back in history,
> > >> so there were substantial conflict resolutions and adjustments required.
> > >> :-) Hope that helps.
> > >
> > > At least it will help me :-)
> > >
> > > Do you suggest that I queue them for 3.10.108, that I wait for
> > > Maowenan to test them more broadly first or anything else ?
> >
> > Let's wait for Maowenan to test them first.
[Mao Wenan]It works well with these patches of v3.10, and the 
retransmission packet is about 250ms after TLP probe.
Thank you very much for these patches under 3.10.
> 
> Fine, thanks.
> Willy

Re: [PATCH net 3/3] tcp: fix xmit timer to only be reset if data ACKed/SACKed

[Willy Tarreau]

On Fri, Aug 04, 2017 at 02:01:34PM -0400, Neal Cardwell wrote:
> On Fri, Aug 4, 2017 at 1:10 PM, Willy Tarreau  wrote:
> > Hi Neal,
> >
> > On Fri, Aug 04, 2017 at 12:59:51PM -0400, Neal Cardwell wrote:
> >> I have attached patches for this fix rebased on to v3.10.107, the
> >> latest stable release for 3.10. That's pretty far back in history, so
> >> there were substantial conflict resolutions and adjustments required.
> >> :-) Hope that helps.
> >
> > At least it will help me :-)
> >
> > Do you suggest that I queue them for 3.10.108, that I wait for Maowenan
> > to test them more broadly first or anything else ?
> 
> Let's wait for Maowenan to test them first.

Fine, thanks.
Willy

Re: [PATCH net 3/3] tcp: fix xmit timer to only be reset if data ACKed/SACKed

[Neal Cardwell]

On Fri, Aug 4, 2017 at 1:10 PM, Willy Tarreau  wrote:
> Hi Neal,
>
> On Fri, Aug 04, 2017 at 12:59:51PM -0400, Neal Cardwell wrote:
>> I have attached patches for this fix rebased on to v3.10.107, the
>> latest stable release for 3.10. That's pretty far back in history, so
>> there were substantial conflict resolutions and adjustments required.
>> :-) Hope that helps.
>
> At least it will help me :-)
>
> Do you suggest that I queue them for 3.10.108, that I wait for Maowenan
> to test them more broadly first or anything else ?

Let's wait for Maowenan to test them first.

Thanks!

neal

Re: [PATCH net 3/3] tcp: fix xmit timer to only be reset if data ACKed/SACKed

[Willy Tarreau]

Hi Neal,

On Fri, Aug 04, 2017 at 12:59:51PM -0400, Neal Cardwell wrote:
> I have attached patches for this fix rebased on to v3.10.107, the
> latest stable release for 3.10. That's pretty far back in history, so
> there were substantial conflict resolutions and adjustments required.
> :-) Hope that helps.

At least it will help me :-)

Do you suggest that I queue them for 3.10.108, that I wait for Maowenan
to test them more broadly first or anything else ?

I'm fine with any option.
Thanks!
Willy

Re: [PATCH net 3/3] tcp: fix xmit timer to only be reset if data ACKed/SACKed

[Neal Cardwell]

On Fri, Aug 4, 2017 at 3:33 AM, maowenan  wrote:
> [Mao Wenan]Follow previous mail, in lower version such as 3.10, I found
> there are many timer type, e.g:ICSK_TIME_EARLY_RETRANS, RTO,PTO
> are used. I'm not sure there exist some unknown problem if we don't check
> isck_pending here and below if branch? And could you please post one lower
> version patch such as 3.10?  Thanks a lot.
>
> #define ICSK_TIME_RETRANS   1   /* Retransmit timer */
> #define ICSK_TIME_DACK  2   /* Delayed ack timer */
> #define ICSK_TIME_PROBE03   /* Zero window probe timer */
> #define ICSK_TIME_EARLY_RETRANS 4   /* Early retransmit timer */
> #define ICSK_TIME_LOSS_PROBE5   /* Tail loss probe timer */

I think you'll find that it you audit how these patches interact with
those other timer types you'll find that the behavior either matches
the existing code, or is improved. We ran these patches through our
packetdrill test suite and found that was the case for all our
existing tests. But please let us know if you find a specific scenario
where that is not the case.

I have attached patches for this fix rebased on to v3.10.107, the
latest stable release for 3.10. That's pretty far back in history, so
there were substantial conflict resolutions and adjustments required.
:-) Hope that helps.

thanks,
neal


0001-tcp-introduce-tcp_rto_delta_us-helper-for-xmit-timer.patch
Description: Binary data


0002-tcp-enable-xmit-timer-fix-by-having-TLP-use-time-whe.patch
Description: Binary data


0003-tcp-fix-xmit-timer-to-only-be-reset-if-data-ACKed-SA.patch
Description: Binary data

Re: [PATCH net 3/3] tcp: fix xmit timer to only be reset if data ACKed/SACKed

[Neal Cardwell]

On Fri, Aug 4, 2017 at 3:12 AM, maowenan  wrote:
> > > --- a/net/ipv4/tcp_output.c
> > > +++ b/net/ipv4/tcp_output.c
> > > @@ -2380,21 +2380,12 @@ bool tcp_schedule_loss_probe(struct sock *sk)
> > > u32 rtt = usecs_to_jiffies(tp->srtt_us >> 3);
> > > u32 timeout, rto_delta_us;
> > >
> > > -   /* No consecutive loss probes. */
> > > -   if (WARN_ON(icsk->icsk_pending == ICSK_TIME_LOSS_PROBE)) {
> > > -   tcp_rearm_rto(sk);
> > > -   return false;
> > > -   }
> [Mao Wenan] I'm sorry I can't get why you delete this and below "if" branch?

We deleted those two "if" branches in tcp_schedule_loss_probe()
because they were assuming that TLP probes would only be scheduled in
a context where an RTO had already been scheduled. With the old
implementation that was true: on every ACK (tcp_ack()) or send of new
data (tcp_event_new_data_sent()) we would first schedule an RTO (by
calling tcp_rearm_rto()) and then schedule a TLP (by calling
tcp_schedule_loss_probe()). So the checks were the right ones for the
old implementation.

With the new implementation, we do not first rearm the RTO on every
incoming ACK. That means when we get to tcp_schedule_loss_probe() we
may find either an RTO or TLP is pending.

Hope that helps clear that up.

cheers,
neal

RE: [PATCH net 3/3] tcp: fix xmit timer to only be reset if data ACKed/SACKed

[maowenan]

> -Original Message-
> From: maowenan
> Sent: Tuesday, August 01, 2017 8:20 PM
> To: 'Neal Cardwell'; David Miller
> Cc: netdev@vger.kernel.org; Yuchung Cheng; Nandita Dukkipati
> Subject: RE: [PATCH net 3/3] tcp: fix xmit timer to only be reset if data
> ACKed/SACKed
> 
> 
> 
> > -Original Message-
> > From: netdev-ow...@vger.kernel.org
> > [mailto:netdev-ow...@vger.kernel.org]
> > On Behalf Of Neal Cardwell
> > Sent: Tuesday, August 01, 2017 10:58 AM
> > To: David Miller
> > Cc: netdev@vger.kernel.org; Neal Cardwell; Yuchung Cheng; Nandita
> > Dukkipati
> > Subject: [PATCH net 3/3] tcp: fix xmit timer to only be reset if data
> > ACKed/SACKed
> >
> > Fix a TCP loss recovery performance bug raised recently on the netdev
> > list, in two threads:
> >
> > (i)  July 26, 2017: netdev thread "TCP fast retransmit issues"
> > (ii) July 26, 2017: netdev thread:
> >  "[PATCH V2 net-next] TLP: Don't reschedule PTO when there's one
> >  outstanding TLP retransmission"
> >
> > The basic problem is that incoming TCP packets that did not indicate
> > forward progress could cause the xmit timer (TLP or RTO) to be rearmed
> > and pushed back in time. In certain corner cases this could result in
> > the following problems noted in these threads:
> >
> >  - Repeated ACKs coming in with bogus SACKs corrupted by middleboxes
> >could cause TCP to repeatedly schedule TLPs forever. We kept
> >sending TLPs after every ~200ms, which elicited bogus SACKs, which
> >caused more TLPs, ad infinitum; we never fired an RTO to fill in
> >the holes.
> >
> >  - Incoming data segments could, in some cases, cause us to reschedule
> >our RTO or TLP timer further out in time, for no good reason. This
> >could cause repeated inbound data to result in stalls in outbound
> >data, in the presence of packet loss.
> >
> > This commit fixes these bugs by changing the TLP and RTO ACK processing to:
> >
> >  (a) Only reschedule the xmit timer once per ACK.
> >
> >  (b) Only reschedule the xmit timer if tcp_clean_rtx_queue() deems the
> >  ACK indicates sufficient forward progress (a packet was
> >  cumulatively ACKed, or we got a SACK for a packet that was sent
> >  before the most recent retransmit of the write queue head).
> >
> > This brings us back into closer compliance with the RFCs, since, as
> > the comment for tcp_rearm_rto() notes, we should only restart the RTO
> > timer after forward progress on the connection. Previously we were
> > restarting the xmit timer even in these cases where there was no forward
> progress.
> >
> > As a side benefit, this commit simplifies and speeds up the TCP timer
> > arming logic. We had been calling inet_csk_reset_xmit_timer() three
> > times on normal ACKs that cumulatively acknowledged some data:
> >
> > 1) Once near the top of tcp_ack() to switch from TLP timer to RTO:
> > if (icsk->icsk_pending == ICSK_TIME_LOSS_PROBE)
> >tcp_rearm_rto(sk);
> >
> > 2) Once in tcp_clean_rtx_queue(), to update the RTO:
> > if (flag & FLAG_ACKED) {
> >tcp_rearm_rto(sk);
> >
> > 3) Once in tcp_ack() after tcp_fastretrans_alert() to switch from RTO
> >to TLP:
> > if (icsk->icsk_pending == ICSK_TIME_RETRANS)
> >tcp_schedule_loss_probe(sk);
> >
> > This commit, by only rescheduling the xmit timer once per ACK,
> > simplifies the code and reduces CPU overhead.
> >
> > This commit was tested in an A/B test with Google web server traffic.
> > SNMP stats and request latency metrics were within noise levels,
> > substantiating that for normal web traffic patterns this is a rare
> > issue. This commit was also tested with packetdrill tests to verify
> > that it fixes the timer behavior in the corner cases discussed in the netdev
> threads mentioned above.
> >
> > This patch is a bug fix patch intended to be queued for -stable relases.
> >
> > Fixes: 6ba8a3b19e76 ("tcp: Tail loss probe (TLP)")
> > Reported-by: Klavs Klavsen 
> > Reported-by: Mao Wenan 
> > Signed-off-by: Neal Cardwell 
> > Signed-off-by: Yuchung Cheng 
> > Signed-off-by: Nandita Dukkipati 
> > ---
> >  net/ipv4/tcp_input.c  | 25 -
> > net/ipv4/tcp_output.c |  9 -
> >  2 files changed, 16 insertions(+), 18 deletions(-)
> >
> > diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index
> > 345febf0a46e..3e777cfbba56 100644
> > --- a/net/ipv4/tcp_input.c
> > +++ b/net/ipv4/tcp_input.c
> > @@ -107,6 +107,7 @@ int sysctl_tcp_invalid_ratelimit __read_mostly =
> HZ/2;
> >  #define FLAG_ORIG_SACK_ACKED   0x200 /* Never retransmitted data are
> > (s)acked*/
> >  #define FLAG_SND_UNA_ADVANCED  0x400 /* Snd_una was changed (!=
> > FLAG_DATA_ACKED) */
> >  #define FLAG_DSACKING_ACK  0x800 /* SACK blocks contained D-SACK info
> > */
> > +#define FLAG_SET_XMIT_TIMER0x1000 /* Set TLP or RTO 

RE: [PATCH net 3/3] tcp: fix xmit timer to only be reset if data ACKed/SACKed

[maowenan]

> -Original Message-
> From: maowenan
> Sent: Tuesday, August 01, 2017 8:20 PM
> To: 'Neal Cardwell'; David Miller
> Cc: netdev@vger.kernel.org; Yuchung Cheng; Nandita Dukkipati
> Subject: RE: [PATCH net 3/3] tcp: fix xmit timer to only be reset if data
> ACKed/SACKed
> 
> 
> 
> > -Original Message-
> > From: netdev-ow...@vger.kernel.org
> > [mailto:netdev-ow...@vger.kernel.org]
> > On Behalf Of Neal Cardwell
> > Sent: Tuesday, August 01, 2017 10:58 AM
> > To: David Miller
> > Cc: netdev@vger.kernel.org; Neal Cardwell; Yuchung Cheng; Nandita
> > Dukkipati
> > Subject: [PATCH net 3/3] tcp: fix xmit timer to only be reset if data
> > ACKed/SACKed
> >
> > Fix a TCP loss recovery performance bug raised recently on the netdev
> > list, in two threads:
> >
> > (i)  July 26, 2017: netdev thread "TCP fast retransmit issues"
> > (ii) July 26, 2017: netdev thread:
> >  "[PATCH V2 net-next] TLP: Don't reschedule PTO when there's one
> >  outstanding TLP retransmission"
> >
> > The basic problem is that incoming TCP packets that did not indicate
> > forward progress could cause the xmit timer (TLP or RTO) to be rearmed
> > and pushed back in time. In certain corner cases this could result in
> > the following problems noted in these threads:
> >
> >  - Repeated ACKs coming in with bogus SACKs corrupted by middleboxes
> >could cause TCP to repeatedly schedule TLPs forever. We kept
> >sending TLPs after every ~200ms, which elicited bogus SACKs, which
> >caused more TLPs, ad infinitum; we never fired an RTO to fill in
> >the holes.
> >
> >  - Incoming data segments could, in some cases, cause us to reschedule
> >our RTO or TLP timer further out in time, for no good reason. This
> >could cause repeated inbound data to result in stalls in outbound
> >data, in the presence of packet loss.
> >
> > This commit fixes these bugs by changing the TLP and RTO ACK processing to:
> >
> >  (a) Only reschedule the xmit timer once per ACK.
> >
> >  (b) Only reschedule the xmit timer if tcp_clean_rtx_queue() deems the
> >  ACK indicates sufficient forward progress (a packet was
> >  cumulatively ACKed, or we got a SACK for a packet that was sent
> >  before the most recent retransmit of the write queue head).
> >
> > This brings us back into closer compliance with the RFCs, since, as
> > the comment for tcp_rearm_rto() notes, we should only restart the RTO
> > timer after forward progress on the connection. Previously we were
> > restarting the xmit timer even in these cases where there was no forward
> progress.
> >
> > As a side benefit, this commit simplifies and speeds up the TCP timer
> > arming logic. We had been calling inet_csk_reset_xmit_timer() three
> > times on normal ACKs that cumulatively acknowledged some data:
> >
> > 1) Once near the top of tcp_ack() to switch from TLP timer to RTO:
> > if (icsk->icsk_pending == ICSK_TIME_LOSS_PROBE)
> >tcp_rearm_rto(sk);
> >
> > 2) Once in tcp_clean_rtx_queue(), to update the RTO:
> > if (flag & FLAG_ACKED) {
> >tcp_rearm_rto(sk);
> >
> > 3) Once in tcp_ack() after tcp_fastretrans_alert() to switch from RTO
> >to TLP:
> > if (icsk->icsk_pending == ICSK_TIME_RETRANS)
> >tcp_schedule_loss_probe(sk);
> >
> > This commit, by only rescheduling the xmit timer once per ACK,
> > simplifies the code and reduces CPU overhead.
> >
> > This commit was tested in an A/B test with Google web server traffic.
> > SNMP stats and request latency metrics were within noise levels,
> > substantiating that for normal web traffic patterns this is a rare
> > issue. This commit was also tested with packetdrill tests to verify
> > that it fixes the timer behavior in the corner cases discussed in the netdev
> threads mentioned above.
> >
> > This patch is a bug fix patch intended to be queued for -stable relases.
> >
> > Fixes: 6ba8a3b19e76 ("tcp: Tail loss probe (TLP)")
> > Reported-by: Klavs Klavsen 
> > Reported-by: Mao Wenan 
> > Signed-off-by: Neal Cardwell 
> > Signed-off-by: Yuchung Cheng 
> > Signed-off-by: Nandita Dukkipati 
> > ---
> >  net/ipv4/tcp_input.c  | 25 -
> > net/ipv4/tcp_output.c |  9 -
> >  2 files changed, 16 insertions(+), 18 deletions(-)
> >
> > diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index
> > 345febf0a46e..3e777cfbba56 100644
> > --- a/net/ipv4/tcp_input.c
> > +++ b/net/ipv4/tcp_input.c
> > @@ -107,6 +107,7 @@ int sysctl_tcp_invalid_ratelimit __read_mostly =
> HZ/2;
> >  #define FLAG_ORIG_SACK_ACKED   0x200 /* Never retransmitted data are
> > (s)acked*/
> >  #define FLAG_SND_UNA_ADVANCED  0x400 /* Snd_una was changed (!=
> > FLAG_DATA_ACKED) */
> >  #define FLAG_DSACKING_ACK  0x800 /* SACK blocks contained D-SACK info
> > */
> > +#define FLAG_SET_XMIT_TIMER0x1000 /* Set TLP or RTO 

Re: [PATCH net 3/3] tcp: fix xmit timer to only be reset if data ACKed/SACKed

[Neal Cardwell]

On Tue, Aug 1, 2017 at 8:20 AM, maowenan  wrote:
> > + /* If needed, reset TLP/RTO timer; RACK may later override this. */
> [Mao Wenan] I have question about RACK, if there is no RACK feature
> in lower version, who can clear this flag:FLAG_SET_XMIT_TIMER?

In the comment, "this" is referring to the xmit timer. The comment is
referring to the fact that RACK (or similarly
tcp_pause_early_retransmit() in earlier kernels) can override the
value of the xmit timer set in tcp_set_xmit_timer(). The comment is
not claiming that RACK clears the FLAG_SET_XMIT_TIMER bit in "flag".

Note that "flag" is an argument to tcp_ack(). And while processing a
given incoming ACK, the callers of tcp_ack() pass in either 0 or one
or more constant bit flags. So FLAG_SET_XMIT_TIMER and all the other
bit flags are implicitly set to zero at the beginning of processing
each ACK.

neal

RE: [PATCH net 3/3] tcp: fix xmit timer to only be reset if data ACKed/SACKed

[maowenan]

> -Original Message-
> From: netdev-ow...@vger.kernel.org [mailto:netdev-ow...@vger.kernel.org]
> On Behalf Of Neal Cardwell
> Sent: Tuesday, August 01, 2017 10:58 AM
> To: David Miller
> Cc: netdev@vger.kernel.org; Neal Cardwell; Yuchung Cheng; Nandita Dukkipati
> Subject: [PATCH net 3/3] tcp: fix xmit timer to only be reset if data
> ACKed/SACKed
> 
> Fix a TCP loss recovery performance bug raised recently on the netdev list, in
> two threads:
> 
> (i)  July 26, 2017: netdev thread "TCP fast retransmit issues"
> (ii) July 26, 2017: netdev thread:
>  "[PATCH V2 net-next] TLP: Don't reschedule PTO when there's one
>  outstanding TLP retransmission"
> 
> The basic problem is that incoming TCP packets that did not indicate forward
> progress could cause the xmit timer (TLP or RTO) to be rearmed and pushed
> back in time. In certain corner cases this could result in the following 
> problems
> noted in these threads:
> 
>  - Repeated ACKs coming in with bogus SACKs corrupted by middleboxes
>could cause TCP to repeatedly schedule TLPs forever. We kept
>sending TLPs after every ~200ms, which elicited bogus SACKs, which
>caused more TLPs, ad infinitum; we never fired an RTO to fill in
>the holes.
> 
>  - Incoming data segments could, in some cases, cause us to reschedule
>our RTO or TLP timer further out in time, for no good reason. This
>could cause repeated inbound data to result in stalls in outbound
>data, in the presence of packet loss.
> 
> This commit fixes these bugs by changing the TLP and RTO ACK processing to:
> 
>  (a) Only reschedule the xmit timer once per ACK.
> 
>  (b) Only reschedule the xmit timer if tcp_clean_rtx_queue() deems the
>  ACK indicates sufficient forward progress (a packet was
>  cumulatively ACKed, or we got a SACK for a packet that was sent
>  before the most recent retransmit of the write queue head).
> 
> This brings us back into closer compliance with the RFCs, since, as the
> comment for tcp_rearm_rto() notes, we should only restart the RTO timer
> after forward progress on the connection. Previously we were restarting the
> xmit timer even in these cases where there was no forward progress.
> 
> As a side benefit, this commit simplifies and speeds up the TCP timer arming
> logic. We had been calling inet_csk_reset_xmit_timer() three times on normal
> ACKs that cumulatively acknowledged some data:
> 
> 1) Once near the top of tcp_ack() to switch from TLP timer to RTO:
> if (icsk->icsk_pending == ICSK_TIME_LOSS_PROBE)
>tcp_rearm_rto(sk);
> 
> 2) Once in tcp_clean_rtx_queue(), to update the RTO:
> if (flag & FLAG_ACKED) {
>tcp_rearm_rto(sk);
> 
> 3) Once in tcp_ack() after tcp_fastretrans_alert() to switch from RTO
>to TLP:
> if (icsk->icsk_pending == ICSK_TIME_RETRANS)
>tcp_schedule_loss_probe(sk);
> 
> This commit, by only rescheduling the xmit timer once per ACK, simplifies the
> code and reduces CPU overhead.
> 
> This commit was tested in an A/B test with Google web server traffic. SNMP
> stats and request latency metrics were within noise levels, substantiating 
> that
> for normal web traffic patterns this is a rare issue. This commit was also 
> tested
> with packetdrill tests to verify that it fixes the timer behavior in the 
> corner
> cases discussed in the netdev threads mentioned above.
> 
> This patch is a bug fix patch intended to be queued for -stable relases.
> 
> Fixes: 6ba8a3b19e76 ("tcp: Tail loss probe (TLP)")
> Reported-by: Klavs Klavsen 
> Reported-by: Mao Wenan 
> Signed-off-by: Neal Cardwell 
> Signed-off-by: Yuchung Cheng 
> Signed-off-by: Nandita Dukkipati 
> ---
>  net/ipv4/tcp_input.c  | 25 -
> net/ipv4/tcp_output.c |  9 -
>  2 files changed, 16 insertions(+), 18 deletions(-)
> 
> diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index
> 345febf0a46e..3e777cfbba56 100644
> --- a/net/ipv4/tcp_input.c
> +++ b/net/ipv4/tcp_input.c
> @@ -107,6 +107,7 @@ int sysctl_tcp_invalid_ratelimit __read_mostly = HZ/2;
>  #define FLAG_ORIG_SACK_ACKED 0x200 /* Never retransmitted data are
> (s)acked  */
>  #define FLAG_SND_UNA_ADVANCED0x400 /* Snd_una was changed (!=
> FLAG_DATA_ACKED) */
>  #define FLAG_DSACKING_ACK0x800 /* SACK blocks contained D-SACK info
> */
> +#define FLAG_SET_XMIT_TIMER  0x1000 /* Set TLP or RTO timer */
>  #define FLAG_SACK_RENEGING   0x2000 /* snd_una advanced to a sacked
> seq */
>  #define FLAG_UPDATE_TS_RECENT0x4000 /* tcp_replace_ts_recent() */
>  #define FLAG_NO_CHALLENGE_ACK0x8000 /* do not call
> tcp_send_challenge_ack()  */
> @@ -3016,6 +3017,13 @@ void tcp_rearm_rto(struct sock *sk)
>   }
>  }
> 
> +/* Try to schedule a loss probe; if that doesn't work, then schedule an
> +RTO. */ static void tcp_set_xmit_timer(struct sock 

Re: [PATCH net 3/3] tcp: fix xmit timer to only be reset if data ACKed/SACKed

[Eric Dumazet]

On Mon, 2017-07-31 at 22:58 -0400, Neal Cardwell wrote:
> Fix a TCP loss recovery performance bug raised recently on the netdev
> list, in two threads:
> 
> (i)  July 26, 2017: netdev thread "TCP fast retransmit issues"
> (ii) July 26, 2017: netdev thread:
>  "[PATCH V2 net-next] TLP: Don't reschedule PTO when there's one
>  outstanding TLP retransmission"

Acked-by: Eric Dumazet 

[PATCH net 3/3] tcp: fix xmit timer to only be reset if data ACKed/SACKed

[Neal Cardwell]

Fix a TCP loss recovery performance bug raised recently on the netdev
list, in two threads:

(i)  July 26, 2017: netdev thread "TCP fast retransmit issues"
(ii) July 26, 2017: netdev thread:
 "[PATCH V2 net-next] TLP: Don't reschedule PTO when there's one
 outstanding TLP retransmission"

The basic problem is that incoming TCP packets that did not indicate
forward progress could cause the xmit timer (TLP or RTO) to be rearmed
and pushed back in time. In certain corner cases this could result in
the following problems noted in these threads:

 - Repeated ACKs coming in with bogus SACKs corrupted by middleboxes
   could cause TCP to repeatedly schedule TLPs forever. We kept
   sending TLPs after every ~200ms, which elicited bogus SACKs, which
   caused more TLPs, ad infinitum; we never fired an RTO to fill in
   the holes.

 - Incoming data segments could, in some cases, cause us to reschedule
   our RTO or TLP timer further out in time, for no good reason. This
   could cause repeated inbound data to result in stalls in outbound
   data, in the presence of packet loss.

This commit fixes these bugs by changing the TLP and RTO ACK
processing to:

 (a) Only reschedule the xmit timer once per ACK.

 (b) Only reschedule the xmit timer if tcp_clean_rtx_queue() deems the
 ACK indicates sufficient forward progress (a packet was
 cumulatively ACKed, or we got a SACK for a packet that was sent
 before the most recent retransmit of the write queue head).

This brings us back into closer compliance with the RFCs, since, as
the comment for tcp_rearm_rto() notes, we should only restart the RTO
timer after forward progress on the connection. Previously we were
restarting the xmit timer even in these cases where there was no
forward progress.

As a side benefit, this commit simplifies and speeds up the TCP timer
arming logic. We had been calling inet_csk_reset_xmit_timer() three
times on normal ACKs that cumulatively acknowledged some data:

1) Once near the top of tcp_ack() to switch from TLP timer to RTO:
if (icsk->icsk_pending == ICSK_TIME_LOSS_PROBE)
   tcp_rearm_rto(sk);

2) Once in tcp_clean_rtx_queue(), to update the RTO:
if (flag & FLAG_ACKED) {
   tcp_rearm_rto(sk);

3) Once in tcp_ack() after tcp_fastretrans_alert() to switch from RTO
   to TLP:
if (icsk->icsk_pending == ICSK_TIME_RETRANS)
   tcp_schedule_loss_probe(sk);

This commit, by only rescheduling the xmit timer once per ACK,
simplifies the code and reduces CPU overhead.

This commit was tested in an A/B test with Google web server
traffic. SNMP stats and request latency metrics were within noise
levels, substantiating that for normal web traffic patterns this is a
rare issue. This commit was also tested with packetdrill tests to
verify that it fixes the timer behavior in the corner cases discussed
in the netdev threads mentioned above.

This patch is a bug fix patch intended to be queued for -stable
relases.

Fixes: 6ba8a3b19e76 ("tcp: Tail loss probe (TLP)")
Reported-by: Klavs Klavsen 
Reported-by: Mao Wenan 
Signed-off-by: Neal Cardwell 
Signed-off-by: Yuchung Cheng 
Signed-off-by: Nandita Dukkipati 
---
 net/ipv4/tcp_input.c  | 25 -
 net/ipv4/tcp_output.c |  9 -
 2 files changed, 16 insertions(+), 18 deletions(-)

diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 345febf0a46e..3e777cfbba56 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -107,6 +107,7 @@ int sysctl_tcp_invalid_ratelimit __read_mostly = HZ/2;
 #define FLAG_ORIG_SACK_ACKED   0x200 /* Never retransmitted data are (s)acked  
*/
 #define FLAG_SND_UNA_ADVANCED  0x400 /* Snd_una was changed (!= 
FLAG_DATA_ACKED) */
 #define FLAG_DSACKING_ACK  0x800 /* SACK blocks contained D-SACK info */
+#define FLAG_SET_XMIT_TIMER0x1000 /* Set TLP or RTO timer */
 #define FLAG_SACK_RENEGING 0x2000 /* snd_una advanced to a sacked seq */
 #define FLAG_UPDATE_TS_RECENT  0x4000 /* tcp_replace_ts_recent() */
 #define FLAG_NO_CHALLENGE_ACK  0x8000 /* do not call tcp_send_challenge_ack()  
*/
@@ -3016,6 +3017,13 @@ void tcp_rearm_rto(struct sock *sk)
}
 }
 
+/* Try to schedule a loss probe; if that doesn't work, then schedule an RTO. */
+static void tcp_set_xmit_timer(struct sock *sk)
+{
+   if (!tcp_schedule_loss_probe(sk))
+   tcp_rearm_rto(sk);
+}
+
 /* If we get here, the whole TSO packet has not been acked. */
 static u32 tcp_tso_acked(struct sock *sk, struct sk_buff *skb)
 {
@@ -3177,7 +3185,7 @@ static int tcp_clean_rtx_queue(struct sock *sk, int 
prior_fackets,
ca_rtt_us, sack->rate);
 
if (flag & FLAG_ACKED) {
-   tcp_rearm_rto(sk);
+   flag |= FLAG_SET_XMIT_TIMER;  /* set TLP or RTO timer */
if (unlikely(icsk->icsk_mtup.probe_size &&